diff --git a/PG-PuReMD/src/cuda_environment.cu b/PG-PuReMD/src/cuda_environment.cu
index 24898ec57ea754e0658d63193b264ea78c406c14..be6101a21af4af5d6f1688ad023a5aaa330263a9 100644
--- a/PG-PuReMD/src/cuda_environment.cu
+++ b/PG-PuReMD/src/cuda_environment.cu
@@ -1,13 +1,15 @@
#include "cuda_environment.h"
+
#include "cuda_utils.h"
-extern "C" void Setup_Cuda_Environment (int rank, int nprocs, int gpus_per_node)
+
+extern "C" void Setup_Cuda_Environment(int rank, int nprocs, int gpus_per_node)
{
int deviceCount;
cudaError_t flag;
- flag = cudaGetDeviceCount (&deviceCount);
+ flag = cudaGetDeviceCount(&deviceCount);
if ( flag != cudaSuccess )
{
@@ -18,32 +20,21 @@ extern "C" void Setup_Cuda_Environment (int rank, int nprocs, int gpus_per_node)
//Calculate the # of GPUs per processor
//and assign the GPU for each process
//TODO: handle condition where # CPU procs > # GPUs
- cudaSetDevice ( (rank % (deviceCount)) );
+ cudaSetDevice( (rank % (deviceCount)) );
#if defined(__CUDA_DEBUG__)
fprintf( stderr, "p:%d is using GPU: %d \n", rank, (rank % deviceCount));
#endif
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- // CHANGE ORIGINAL/////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- //cudaDeviceSetLimit ( cudaLimitStackSize, 8192 );
- //cudaDeviceSetCacheConfig ( cudaFuncCachePreferL1 );
- //cudaCheckError ();
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
- ///////////////////////////////////////////////
+ //CHANGE ORIGINAL
+ //cudaDeviceSetLimit( cudaLimitStackSize, 8192 );
+ //cudaDeviceSetCacheConfig( cudaFuncCachePreferL1 );
+ //cudaCheckError();
}
-extern "C" void Cleanup_Cuda_Environment ()
+extern "C" void Cleanup_Cuda_Environment()
{
- cudaDeviceReset ();
- cudaDeviceSynchronize ();
+ cudaDeviceReset();
+ cudaDeviceSynchronize();
}
diff --git a/PG-PuReMD/src/reax_types.h b/PG-PuReMD/src/reax_types.h
index debe20aac39b241c5064a55b80603a3a3886952f..8c353a7ca600b369ad69b2bdabd8c8e883915d21 100644
--- a/PG-PuReMD/src/reax_types.h
+++ b/PG-PuReMD/src/reax_types.h
@@ -22,30 +22,26 @@
#if !(defined(__REAX_TYPES_H_) || defined(__CUDA_REAX_TYPES_H_))
#ifdef __CUDACC__
-
-#ifndef __CUDA_REAX_TYPES_H_
-#define __CUDA_REAX_TYPES_H_
-#define CUDA_HOST __host__
-#define CUDA_DEVICE __device__
-#define CUDA_GLOBAL __global__
-#define CUDA_HOST_DEVICE __host__ __device__
-#endif
-
+ #ifndef __CUDA_REAX_TYPES_H_
+ #define __CUDA_REAX_TYPES_H_
+ #define CUDA_HOST __host__
+ #define CUDA_DEVICE __device__
+ #define CUDA_GLOBAL __global__
+ #define CUDA_HOST_DEVICE __host__ __device__
+ #endif
#else
-
-#ifndef __REAX_TYPES_H_
-#define __REAX_TYPES_H_
-#define CUDA_HOST
-#define CUDA_DEVICE
-#define CUDA_GLOBAL
-#define CUDA_HOST_DEVICE
-#endif
-
+ #ifndef __REAX_TYPES_H_
+ #define __REAX_TYPES_H_
+ #define CUDA_HOST
+ #define CUDA_DEVICE
+ #define CUDA_GLOBAL
+ #define CUDA_HOST_DEVICE
+ #endif
#endif
#if (defined(HAVE_CONFIG_H) && !defined(__CONFIG_H_))
-#define __CONFIG_H_
-#include "config.h"
+ #define __CONFIG_H_
+ #include "config.h"
#endif
#include <ctype.h>
@@ -57,13 +53,13 @@
#include <sys/time.h>
#include <time.h>
#include <zlib.h>
-#define HOST_SCRATCH_SIZE (1024 * 1024 * 20)
+#define HOST_SCRATCH_SIZE (1024 * 1024 * 20)
#ifdef HAVE_CUDA
-#include <cuda.h>
+ #include <cuda.h>
#endif
#if defined(__IBMC__)
-#define inline __inline__
+ #define inline __inline__
#endif /*IBMC*/
#define PURE_REAX
diff --git a/PuReMD-GPU/Makefile.am b/PuReMD-GPU/Makefile.am
index d433237070eb800b0142d0f81be5f10b42553035..016114db2dfaf3b9af623069b2f5f3ffc677907a 100644
--- a/PuReMD-GPU/Makefile.am
+++ b/PuReMD-GPU/Makefile.am
@@ -15,32 +15,51 @@ AM_LDFLAGS =
NVCCFLAGS += -use_fast_math
NVCCFLAGS += -gencode arch=compute_35,code=sm_35
NVCCFLAGS += --compiler-options "$(DEFS) -D__SM_35__ -O3 -funroll-loops -fstrict-aliasing"
+#NVCCFLAGS += -Xcompiler -fPIC -dc
#NVCCFLAGS += --ptxas-options -v
bin_PROGRAMS = bin/puremd-gpu
-bin_puremd_gpu_SOURCES = src/analyze.c src/print_utils.c src/reset_utils.c src/param.c src/pdb_tools.c \
- src/GMRES.cu src/QEq.cu src/allocate.cu src/bond_orders.cu \
- src/box.cu src/forces.cu src/four_body_interactions.cu \
- src/grid.cu src/init_md.cu src/integrate.cu src/list.cu \
- src/lookup.cu src/neighbors.cu \
- src/restart.cu src/single_body_interactions.cu \
- src/system_props.cu src/three_body_interactions.cu \
- src/traj.cu src/two_body_interactions.cu src/vector.cu \
- src/testmd.cu \
- src/cuda_utils.cu src/cuda_copy.cu src/cuda_init.cu src/reduction.cu \
- src/center_mass.cu src/helpers.cu src/validation.cu src/matvec.cu
-
+bin_puremd_gpu_SOURCES = src/analyze.c src/print_utils.c \
+ src/restart.c src/param.c src/pdb_tools.c src/box.c \
+ src/lin_alg.c src/QEq.c src/allocate.c src/bond_orders.c \
+ src/forces.c src/four_body_interactions.c \
+ src/grid.c src/init_md.c src/integrate.c src/list.c \
+ src/lookup.c src/neighbors.c \
+ src/reset_utils.c src/single_body_interactions.c \
+ src/system_props.c src/three_body_interactions.c \
+ src/traj.c src/two_body_interactions.c src/vector.c \
+ src/testmd.c \
+ src/cuda_utils.cu src/cuda_copy.cu src/cuda_init.cu src/cuda_reduction.cu \
+ src/cuda_center_mass.cu src/cuda_box.cu src/validation.cu \
+ src/cuda_allocate.cu src/cuda_bond_orders.cu \
+ src/cuda_lin_alg.cu src/cuda_QEq.cu \
+ src/cuda_forces.cu src/cuda_four_body_interactions.cu \
+ src/cuda_grid.cu src/cuda_init_md.cu src/cuda_integrate.cu src/cuda_list.cu \
+ src/cuda_lookup.cu src/cuda_neighbors.cu \
+ src/cuda_reset_utils.cu src/cuda_single_body_interactions.cu \
+ src/cuda_system_props.cu src/cuda_three_body_interactions.cu \
+ src/cuda_two_body_interactions.cu src/cuda_environment.cu \
+ src/cuda_post_evolve.cu
include_HEADERS = src/mytypes.h src/analyze.h src/print_utils.h \
- src/reset_utils.h src/param.h src/pdb_tools.h \
- src/GMRES.h src/QEq.h src/allocate.h src/bond_orders.h \
- src/box.h src/forces.h src/four_body_interactions.h \
+ src/restart.h src/param.h src/pdb_tools.h src/box.h \
+ src/lin_alg.h src/QEq.h src/allocate.h src/bond_orders.h \
+ src/forces.h src/four_body_interactions.h \
src/grid.h src/init_md.h src/integrate.h src/list.h \
src/lookup.h src/neighbors.h \
- src/restart.h src/single_body_interactions.h \
+ src/reset_utils.h src/single_body_interactions.h \
src/system_props.h src/three_body_interactions.h \
src/traj.h src/two_body_interactions.h src/vector.h \
- src/cuda_utils.h src/cuda_copy.h src/cuda_init.h src/reduction.h \
- src/center_mass.h src/helpers.h src/validation.h src/matvec.h
+ src/cuda_utils.h src/cuda_copy.h src/cuda_init.h src/cuda_reduction.h \
+ src/cuda_center_mass.h src/cuda_box.h src/validation.h \
+ src/cuda_allocate.h src/cuda_bond_orders.h \
+ src/cuda_lin_alg.h src/cuda_QEq.h \
+ src/cuda_forces.h src/cuda_four_body_interactions.h \
+ src/cuda_grid.h src/cuda_init_md.h src/cuda_integrate.h src/cuda_list.h \
+ src/cuda_lookup.h src/cuda_neighbors.h \
+ src/cuda_reset_utils.h src/cuda_single_body_interactions.h \
+ src/cuda_system_props.h src/cuda_three_body_interactions.h \
+ src/cuda_two_body_interactions.h src/cuda_environment.h \
+ src/cuda_post_evolve.h
# dummy source to cause C linking
nodist_EXTRA_bin_puremd_gpu_SOURCES = src/dummy.c
diff --git a/PuReMD-GPU/configure.ac b/PuReMD-GPU/configure.ac
index c947ed021a0544653895f01c59dce33e4a07a902..30e7e0bff6bc3115193c75d3b2d0dcbe29a5cf4b 100644
--- a/PuReMD-GPU/configure.ac
+++ b/PuReMD-GPU/configure.ac
@@ -42,19 +42,6 @@ AC_SEARCH_LIBS([gzeof], [z])
AC_SEARCH_LIBS([gzgets], [z])
AC_SEARCH_LIBS([gzseek], [z])
AC_SEARCH_LIBS([gzclose, [z]])
-AC_SEARCH_LIBS([cublasCheckError], [cublas])
-AC_SEARCH_LIBS([cublasDnrm2], [cublas])
-AC_SEARCH_LIBS([cublasDaxpy], [cublas])
-AC_SEARCH_LIBS([cublasDscal], [cublas])
-AC_SEARCH_LIBS([cublasDdot], [cublas])
-AC_SEARCH_LIBS([cudaThreadSynchronize], [cuda])
-AC_SEARCH_LIBS([cudaCheckError], [cuda])
-# FIXME: Replace `main' with a function in `-lcudart':
-#AC_CHECK_LIB([cudart], [main])
-AC_SEARCH_LIBS([cusparseCheckError], [cusparse])
-AC_SEARCH_LIBS([cusparseCreateMatDescr], [cusparse])
-AC_SEARCH_LIBS([cusparseSetMatType], [cusparse])
-AC_SEARCH_LIBS([cusparseSetMatIndexBase], [cusparse])
# Checks for typedefs, structures, and compiler characteristics.
AC_CHECK_HEADER_STDBOOL
@@ -78,10 +65,26 @@ then
fi
AC_DEFINE([HAVE_CUDA], [1], [Define to 1 if you have CUDA support enabled.])
-if test "BUILD_PROF" = "true"
-then
- NVCCFLAGS+=" --compiler-options ${gprof_flags}"
-fi
+AC_SEARCH_LIBS([cublasDnrm2], [cublas])
+AC_SEARCH_LIBS([cublasDaxpy], [cublas])
+AC_SEARCH_LIBS([cublasDscal], [cublas])
+AC_SEARCH_LIBS([cublasDdot], [cublas])
+AC_SEARCH_LIBS([cudaThreadSynchronize], [cudart])
+AC_SEARCH_LIBS([cudaGetLastError], [cudart])
+AC_CHECK_LIB([cudart], [cudaMalloc])
+AC_SEARCH_LIBS([cusparseCreateMatDescr], [cusparse])
+AC_SEARCH_LIBS([cusparseSetMatType], [cusparse])
+AC_SEARCH_LIBS([cusparseSetMatIndexBase], [cusparse])
+
+AC_SEARCH_LIBS([cublasDnrm2], [cublas],
+ [CUBLAS_FOUND_LIBS="yes"], [CUBLAS_FOUND_LIBS="no"], [-lcublas])
+AS_IF([test "x${CUBLAS_FOUND_LIBS}" != "xyes"],
+ [AC_MSG_ERROR([Unable to find CUBLAS library.])])
+
+AC_SEARCH_LIBS([cusparseSetMatType], [cusparse],
+ [CUSPARSE_FOUND_LIBS="yes"], [CUSPARSE_FOUND_LIBS="no"], [-lcusparse])
+AS_IF([test "x${CUSPARSE_FOUND_LIBS}" != "xyes"],
+ [AC_MSG_ERROR([Unable to find CUSPARSE library.])])
AC_CHECK_TYPES([cublasHandle_t], [],
[AC_MSG_FAILURE([cublasHandle_t type not found in cublas.h], [1])], [#include<cublas_v2.h>])
@@ -89,10 +92,12 @@ AC_CHECK_TYPES([cusparseHandle_t], [],
[AC_MSG_FAILURE([cusparseHandle_t type not found in cusparse.h], [1])], [#include<cusparse_v2.h>])
AC_CHECK_TYPES([cusparseMatDescr_t], [],
[AC_MSG_FAILURE([cusparseMatDescr_t type not found in cusparse.h], [1])], [#include<cusparse_v2.h>])
-#AC_CHECK_TYPES([CUSPARSE_MATRIX_TYPE_GENERAL], [],
-# [AC_MSG_FAILURE([CUSPARSE_MATRIX_TYPE_GENERAL type not found in cusparse.h], [1])], [#include<cusparse_v2.h>])
-#AC_CHECK_TYPES([CUSPARSE_INDEX_BASE_ZERO], [],
-# [AC_MSG_FAILURE([CUSPARSE_INDEX_BASE_ZERO type not found in cusparse.h], [1])], [#include<cusparse_v2.h>])
+
+if test "BUILD_PROF" = "true"
+then
+ NVCCFLAGS+=" --compiler-options ${gprof_flags}"
+fi
+
AC_CONFIG_FILES([Makefile])
diff --git a/PuReMD-GPU/src/GMRES.cu b/PuReMD-GPU/src/GMRES.cu
deleted file mode 100644
index d00100e9ced86d2b1b7a8d1f37b67648576cdca1..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/GMRES.cu
+++ /dev/null
@@ -1,1138 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "GMRES.h"
-#include "list.h"
-#include "vector.h"
-#include "index_utils.h"
-
-#include "cuda_copy.h"
-#include "cuda_utils.h"
-#include "reduction.h"
-#include "matvec.h"
-#include "system_props.h"
-
-#include "cublas_v2.h"
-#include "cusparse_v2.h"
-
-void Sparse_MatVec( sparse_matrix *A, real *x, real *b )
-{
- int i, j, k, n, si, ei;
- real H;
-
- n = A->n;
- for( i = 0; i < n; ++i )
- b[i] = 0;
-
- for( i = 0; i < n; ++i ) {
- si = A->start[i];
- ei = A->start[i+1]-1;
-
- for( k = si; k < ei; ++k ) {
- j = A->entries[k].j;
- H = A->entries[k].val;
- b[j] += H * x[i];
- b[i] += H * x[j];
- }
-
- // the diagonal entry is the last one in
- b[i] += A->entries[k].val * x[i];
- }
-}
-
-
-void Forward_Subs( sparse_matrix *L, real *b, real *y )
-{
- int i, pj, j, si, ei;
- real val;
-
- for( i = 0; i < L->n; ++i ) {
- y[i] = b[i];
- si = L->start[i];
- ei = L->start[i+1];
- for( pj = si; pj < ei-1; ++pj ){
- j = L->entries[pj].j;
- val = L->entries[pj].val;
- y[i] -= val * y[j];
- }
- y[i] /= L->entries[pj].val;
- }
-}
-
-
-void Backward_Subs( sparse_matrix *U, real *y, real *x )
-{
- int i, pj, j, si, ei;
- real val;
-
- for( i = U->n-1; i >= 0; --i ) {
- x[i] = y[i];
- si = U->start[i];
- ei = U->start[i+1];
- for( pj = si+1; pj < ei; ++pj ){
- j = U->entries[pj].j;
- val = U->entries[pj].val;
- x[i] -= val * x[j];
- }
- x[i] /= U->entries[si].val;
- }
-}
-
-
-int GMRES( static_storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, FILE *fout, reax_system* system)
-{
- int i, j, k, itr, N;
- real cc, tmp1, tmp2, temp, bnorm;
-
- N = H->n;
- bnorm = Norm( b, N );
-
- /* apply the diagonal pre-conditioner to rhs */
- for( i = 0; i < N; ++i )
- workspace->b_prc[i] = b[i] * workspace->Hdia_inv[i];
-
- /* GMRES outer-loop */
- for( itr = 0; itr < MAX_ITR; ++itr ) {
- /* calculate r0 */
- Sparse_MatVec( H, x, workspace->b_prm );
-
- for( i = 0; i < N; ++i )
- workspace->b_prm[i] *= workspace->Hdia_inv[i]; /* pre-conditioner */
-
-
- Vector_Sum(&workspace->v[ index_wkspace_sys (0,0,system) ], 1.,workspace->b_prc, -1., workspace->b_prm, N);
- workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system)], N );
- Vector_Scale( &workspace->v[ index_wkspace_sys (0,0,system) ], 1.0/workspace->g[0], &workspace->v[index_wkspace_sys(0,0,system)], N );
-
- /* GMRES inner-loop */
- for( j = 0; j < RESTART && fabs(workspace->g[j]) / bnorm > tol; j++ ) {
- /* matvec */
- Sparse_MatVec( H, &workspace->v[index_wkspace_sys(j,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)] );
-
- for( k = 0; k < N; ++k )
- workspace->v[ index_wkspace_sys (j+1,k,system)] *= workspace->Hdia_inv[k]; /*pre-conditioner*/
-
- /* apply modified Gram-Schmidt to orthogonalize the new residual */
- for( i = 0; i <= j; i++ ) {
- workspace->h[ index_wkspace_res (i,j) ] = Dot( &workspace->v[index_wkspace_sys(i,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- Vector_Add( &workspace->v[index_wkspace_sys(j+1,0,system)],
- -workspace->h[index_wkspace_res (i,j) ], &workspace->v[index_wkspace_sys(i,0,system)], N );
- }
-
-
- workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- Vector_Scale( &workspace->v[index_wkspace_sys(j+1,0,system)],
- 1. / workspace->h[ index_wkspace_res (j+1,j) ], &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- // fprintf( stderr, "%d-%d: orthogonalization completed.\n", itr, j );
-
-
- /* Givens rotations on the upper-Hessenberg matrix to make it U */
- for( i = 0; i <= j; i++ ) {
- if( i == j ) {
- cc = SQRT( SQR(workspace->h[ index_wkspace_res (j,j) ])+SQR(workspace->h[ index_wkspace_res (j+1,j) ]) );
- workspace->hc[j] = workspace->h[ index_wkspace_res (j,j) ] / cc;
- workspace->hs[j] = workspace->h[ index_wkspace_res (j+1,j) ] / cc;
- }
-
- tmp1 = workspace->hc[i] * workspace->h[ index_wkspace_res (i,j) ] +
- workspace->hs[i] * workspace->h[ index_wkspace_res (i+1,j) ];
- tmp2 = -workspace->hs[i] * workspace->h[ index_wkspace_res (i,j) ] +
- workspace->hc[i] * workspace->h[ index_wkspace_res (i+1,j) ];
-
- workspace->h[ index_wkspace_res (i,j) ] = tmp1;
- workspace->h[ index_wkspace_res (i+1,j) ] = tmp2;
- }
-
- /* apply Givens rotations to the rhs as well */
- tmp1 = workspace->hc[j] * workspace->g[j];
- tmp2 = -workspace->hs[j] * workspace->g[j];
- workspace->g[j] = tmp1;
- workspace->g[j+1] = tmp2;
-
- // fprintf( stderr, "h: " );
- // for( i = 0; i <= j+1; ++i )
- // fprintf( stderr, "%.6f ", workspace->h[i][j] );
- // fprintf( stderr, "\n" );
- //fprintf( stderr, "res: %.15e\n", workspace->g[j+1] );
- }
-
-
- /* solve Hy = g.
- H is now upper-triangular, do back-substitution */
- for( i = j-1; i >= 0; i-- ) {
- temp = workspace->g[i];
- for( k = j-1; k > i; k-- )
- temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
-
- workspace->y[i] = temp / workspace->h[ index_wkspace_res (i,i) ];
- }
-
-
- /* update x = x_0 + Vy */
- for( i = 0; i < j; i++ )
- Vector_Add( x, workspace->y[i], &workspace->v[index_wkspace_sys(i,0,system)], N );
-
- /* stopping condition */
- if( fabs(workspace->g[j]) / bnorm <= tol )
- break;
- }
-
- // Sparse_MatVec( H, x, workspace->b_prm );
- // for( i = 0; i < N; ++i )
- // workspace->b_prm[i] *= workspace->Hdia_inv[i];
- // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
- // for( i = 0; i < N; ++i )
- // fprintf( fout, "%10.5f%15.12f%15.12f\n",
- // workspace->b_prc[i], workspace->b_prm[i], x[i] );*/
-
- // fprintf(fout,"GMRES outer:%d, inner:%d iters - residual norm: %25.20f\n",
- // itr, j, fabs( workspace->g[j] ) / bnorm );
- // data->timing.matvec += itr * RESTART + j;
-
- if( itr >= MAX_ITR ) {
- fprintf( stderr, "GMRES convergence failed\n" );
- // return -1;
- return itr * (RESTART+1) + j + 1;
- }
-
- return itr * (RESTART+1) + j + 1;
-}
-
-
-/////////////////////////////////////////////////////////////////
-//Cuda Functions for GMRES implementation
-/////////////////////////////////////////////////////////////////
-
-GLOBAL void GMRES_Diagonal_Preconditioner (real *b_proc, real *b, real *Hdia_inv, int entries)
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= entries) return;
-
- b_proc [i] = b[i] * Hdia_inv[i];
-}
-
-GLOBAL void GMRES_Givens_Rotation (int j, real *h, real *hc, real *hs, real g_j, real *output)
-{
- real tmp1, tmp2, cc;
-
- for( int i = 0; i <= j; i++ ) {
- if( i == j ) {
- cc = SQRT( SQR(h[ index_wkspace_res (j,j) ])+SQR(h[ index_wkspace_res (j+1,j) ]) );
- hc[j] = h[ index_wkspace_res (j,j) ] / cc;
- hs[j] = h[ index_wkspace_res (j+1,j) ] / cc;
- }
-
- tmp1 = hc[i] * h[ index_wkspace_res (i,j) ] + hs[i] * h[ index_wkspace_res (i+1,j) ];
- tmp2 = -hs[i] * h[ index_wkspace_res (i,j) ] + hc[i] * h[ index_wkspace_res (i+1,j) ];
-
- h[ index_wkspace_res (i,j) ] = tmp1;
- h[ index_wkspace_res (i+1,j) ] = tmp2;
- }
-
- /* apply Givens rotations to the rhs as well */
- tmp1 = hc[j] * g_j;
- tmp2 = -hs[j] * g_j;
-
- output[0] = tmp1;
- output[1] = tmp2;
-}
-
-GLOBAL void GMRES_BackSubstitution (int j, real *g, real *h, real *y)
-{
- real temp;
- for( int i = j-1; i >= 0; i-- ) {
- temp = g[i];
- for( int k = j-1; k > i; k-- )
- temp -= h[ index_wkspace_res (i,k) ] * y[k];
-
- y[i] = temp / h[ index_wkspace_res (i,i) ];
- }
-}
-
-
-int Cuda_GMRES( static_storage *workspace, real *b, real tol, real *x )
-{
- int i, j, k, itr, N;
- real cc, tmp1, tmp2, temp, bnorm;
- real v_add_tmp;
- sparse_matrix *H = &workspace->H;
-
- real t_start, t_elapsed;
-
- real *spad = (real *)scratch;
- real *g = (real *) calloc ((RESTART+1), REAL_SIZE);
-
- N = H->n;
-
- cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
-
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (b, spad, H->n, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, spad + BLOCKS_POW_2, BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device ( &bnorm, spad + BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Norm of the array is %e \n", bnorm );
-#endif
-
- /* apply the diagonal pre-conditioner to rhs */
- GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
- (workspace->b_prc, b, workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* GMRES outer-loop */
- for( itr = 0; itr < MAX_ITR; ++itr ) {
- /* calculate r0 */
- //Sparse_MatVec( H, x, workspace->b_prm );
- Cuda_Matvec_csr <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>> ( *H, x, workspace->b_prm, N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- GMRES_Diagonal_Preconditioner <<< BLOCKS, BLOCK_SIZE >>>
- (workspace->b_prm, workspace->b_prm, workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Vector_Sum <<< BLOCKS, BLOCK_SIZE >>>
- (&workspace->v[ index_wkspace_sys (0,0,N) ], 1.,workspace->b_prc, -1., workspace->b_prm, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system)], N );
- {
- cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
-
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (&workspace->v [index_wkspace_sys (0, 0, N)], spad, N, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->g[0], BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device( g, workspace->g, REAL_SIZE, cudaMemcpyDeviceToHost, RES_STORAGE_G);
- }
-
- Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[ index_wkspace_sys (0,0,N) ], 1.0/g[0], &workspace->v[index_wkspace_sys(0,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* GMRES inner-loop */
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Inner loop inputs bnorm : %f , tol : %f g[j] : %f \n", bnorm, tol, g[0] );
-#endif
- for( j = 0; j < RESTART && fabs(g[j]) / bnorm > tol; j++ ) {
- /* matvec */
- //Sparse_MatVec( H, &workspace->v[index_wkspace_sys(j,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)] );
- Cuda_Matvec_csr
- <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>>
- ( *H, &workspace->v[ index_wkspace_sys (j, 0, N)], &workspace->v[ index_wkspace_sys (j+1, 0, N) ], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
- (&workspace->v[ index_wkspace_sys (j+1,0,N) ], &workspace->v[ index_wkspace_sys (j+1,0,N) ], workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- /* apply modified Gram-Schmidt to orthogonalize the new residual */
- for( i = 0; i <= j; i++ ) {
- Cuda_Dot <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (&workspace->v[index_wkspace_sys(i,0,N)], &workspace->v[index_wkspace_sys(j+1,0,N)], spad, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (i,j) ], BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
- Cuda_Vector_Add <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[index_wkspace_sys(j+1,0,N)],
- -v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
-
-
- //workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- cuda_memset (spad, 0, REAL_SIZE * N * 2, RES_SCRATCH );
-
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (&workspace->v[index_wkspace_sys(j+1,0,N)], spad, N, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (j+1,j) ], BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
- Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[index_wkspace_sys(j+1,0,N)],
- 1. / v_add_tmp, &workspace->v[index_wkspace_sys(j+1,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* Givens rotations on the upper-Hessenberg matrix to make it U */
- GMRES_Givens_Rotation <<<1, 1>>>
- (j, workspace->h, workspace->hc, workspace->hs, g[j], spad);
- cudaThreadSynchronize ();
- cudaCheckError ();
- copy_host_device (&g[j], spad, 2 * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- }
-
- copy_host_device (g, workspace->g, (RESTART+1)*REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
-
- /* solve Hy = g.
- H is now upper-triangular, do back-substitution */
- copy_host_device (g, spad, (RESTART+1) * REAL_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_G);
- GMRES_BackSubstitution <<<1, 1>>>
- (j, spad, workspace->h, workspace->y);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* update x = x_0 + Vy */
- for( i = 0; i < j; i++ )
- {
- copy_host_device (&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- Cuda_Vector_Add <<<BLOCKS, BLOCK_SIZE>>>
- ( x, v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
-
- /* stopping condition */
- if( fabs(g[j]) / bnorm <= tol )
- break;
- }
-
- if( itr >= MAX_ITR ) {
- fprintf( stderr, "GMRES convergence failed\n" );
- return itr * (RESTART+1) + j + 1;
- }
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " GPU values itr : %d, RESTART: %d, j: %d \n", itr, RESTART, j);
-#endif
- return itr * (RESTART+1) + j + 1;
-}
-
-
-int Cublas_GMRES(reax_system *system, static_storage *workspace, real *b, real tol, real *x )
-{
-
- real CSR_ALPHA = 1, CSR_BETA = 0;
-
- int i, j, k, itr, N;
- real cc, tmp1, tmp2, temp, bnorm;
- real v_add_tmp;
- sparse_matrix *H = &workspace->H;
-
- real t_start, t_elapsed;
-
- real *spad = (real *)scratch;
- real *g = (real *) calloc ((RESTART+1), REAL_SIZE);
-
- N = H->n;
-
- cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
-
- /*
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (b, spad, H->n, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, spad + BLOCKS_POW_2, BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device ( &bnorm, spad + BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- */
-
- cublasCheckError (cublasDnrm2 ( cublasHandle, N, b, 1, &bnorm ));
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Norm of the array is %e \n", bnorm );
-#endif
-
- /* apply the diagonal pre-conditioner to rhs */
- GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
- (workspace->b_prc, b, workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* GMRES outer-loop */
- for( itr = 0; itr < MAX_ITR; ++itr ) {
- /* calculate r0 */
- //Sparse_MatVec( H, x, workspace->b_prm );
- Cuda_Matvec_csr <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>> ( *H, x, workspace->b_prm, N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- GMRES_Diagonal_Preconditioner <<< BLOCKS, BLOCK_SIZE >>>
- (workspace->b_prm, workspace->b_prm, workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /*
- Cuda_Vector_Sum <<< BLOCKS, BLOCK_SIZE >>>
- (&workspace->v[ index_wkspace_sys (0,0,N) ], 1.,workspace->b_prc, -1., workspace->b_prm, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
- cuda_memset (workspace->v, 0, REAL_SIZE * (RESTART+1) * N, RES_STORAGE_V);
-
- double D_ONE = 1.;
- double D_MINUS_ONE = -1.;
- cublasCheckError (cublasDaxpy (cublasHandle, N, &D_ONE, workspace->b_prc, 1, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
- cublasCheckError (cublasDaxpy (cublasHandle, N, &D_MINUS_ONE, workspace->b_prm, 1, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
-
- //workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system)], N );
- {
- /*
- cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
-
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (&workspace->v [index_wkspace_sys (0, 0, N)], spad, N, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->g[0], BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device( g, workspace->g, REAL_SIZE, cudaMemcpyDeviceToHost, RES_STORAGE_G);
- */
-
- cublasCheckError (cublasDnrm2 ( cublasHandle, N, &workspace->v [index_wkspace_sys (0, 0, N)], 1, g ));
- copy_host_device( g, workspace->g, REAL_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_G);
- }
-
- /*
- Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[ index_wkspace_sys (0,0,N) ], 1.0/g[0], &workspace->v[index_wkspace_sys(0,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
-
- double D_SCALE = 1.0 / g[0];
- cublasCheckError (cublasDscal (cublasHandle, N, &D_SCALE, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
-
-
- /* GMRES inner-loop */
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Inner loop inputs bnorm : %f , tol : %f g[j] : %f \n", bnorm, tol, g[0] );
-#endif
- for( j = 0; j < RESTART && fabs(g[j]) / bnorm > tol; j++ ) {
- /* matvec */
- Cuda_Matvec_csr
- <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>>
- ( *H, &workspace->v[ index_wkspace_sys (j, 0, N)], &workspace->v[ index_wkspace_sys (j+1, 0, N) ], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
- (&workspace->v[ index_wkspace_sys (j+1,0,N) ], &workspace->v[ index_wkspace_sys (j+1,0,N) ], workspace->Hdia_inv, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- /* apply modified Gram-Schmidt to orthogonalize the new residual */
- for( i = 0; i <= j; i++ ) {
-
- /*
- Cuda_Dot <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (&workspace->v[index_wkspace_sys(i,0,N)], &workspace->v[index_wkspace_sys(j+1,0,N)], spad, N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (i,j) ], BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- */
-
- cublasCheckError (cublasDdot (cublasHandle, N, &workspace->v[index_wkspace_sys(i,0,N)], 1,
- &workspace->v[index_wkspace_sys(j+1,0,N)], 1,
- &v_add_tmp));
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
-
- /*
- Cuda_Vector_Add <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[index_wkspace_sys(j+1,0,N)],
- -v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
-
- double NEG_V_ADD_TMP = -v_add_tmp;
- cublasCheckError (cublasDaxpy (cublasHandle, N, &NEG_V_ADD_TMP, &workspace->v[index_wkspace_sys(i,0,N)], 1,
- &workspace->v[index_wkspace_sys(j+1,0,N)], 1 ));
- }
-
-
- //workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- /*
- cuda_memset (spad, 0, REAL_SIZE * N * 2, RES_SCRATCH );
-
- Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (&workspace->v[index_wkspace_sys(j+1,0,N)], spad, N, INITIAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (j+1,j) ], BLOCKS_POW_2, FINAL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- */
- cublasCheckError (cublasDnrm2 ( cublasHandle, N, &workspace->v [index_wkspace_sys (j+1, 0, N)], 1, &v_add_tmp ));
- copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
-
-
- /*
- Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
- ( &workspace->v[index_wkspace_sys(j+1,0,N)],
- 1. / v_add_tmp, &workspace->v[index_wkspace_sys(j+1,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
- double REC_V_ADD_TMP = 1. / v_add_tmp;
- cublasCheckError (cublasDscal (cublasHandle, N, &REC_V_ADD_TMP, &workspace->v[index_wkspace_sys(j+1,0,N)], 1));
-
-
-
- /* Givens rotations on the upper-Hessenberg matrix to make it U */
- GMRES_Givens_Rotation <<<1, 1>>>
- (j, workspace->h, workspace->hc, workspace->hs, g[j], spad);
- cudaThreadSynchronize ();
- cudaCheckError ();
- copy_host_device (&g[j], spad, 2 * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- }
-
- copy_host_device (g, workspace->g, (RESTART+1)*REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
-
- /* solve Hy = g.
- H is now upper-triangular, do back-substitution */
- copy_host_device (g, spad, (RESTART+1) * REAL_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_G);
- GMRES_BackSubstitution <<<1, 1>>>
- (j, spad, workspace->h, workspace->y);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* update x = x_0 + Vy */
- for( i = 0; i < j; i++ )
- {
- /*
- copy_host_device (&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- Cuda_Vector_Add <<<BLOCKS, BLOCK_SIZE>>>
- ( x, v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
-
- copy_host_device (&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- cublasCheckError (cublasDaxpy (cublasHandle, N, &v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], 1,
- x, 1));
- }
-
- /* stopping condition */
- if( fabs(g[j]) / bnorm <= tol )
- break;
- }
-
- if( itr >= MAX_ITR ) {
- fprintf( stderr, "GMRES convergence failed\n" );
- return itr * (RESTART+1) + j + 1;
- }
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " GPU values itr : %d, RESTART: %d, j: %d \n", itr, RESTART, j);
-#endif
- return itr * (RESTART+1) + j + 1;
-}
-
-int GMRES_HouseHolder( static_storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, FILE *fout, reax_system *system)
-{
- int i, j, k, itr, N;
- real cc, tmp1, tmp2, temp, bnorm;
- real v[10000], z[RESTART+2][10000], w[RESTART+2];
- real u[RESTART+2][10000];
-
- N = H->n;
- bnorm = Norm( b, N );
-
- /* apply the diagonal pre-conditioner to rhs */
- for( i = 0; i < N; ++i )
- workspace->b_prc[i] = b[i] * workspace->Hdia_inv[i];
-
- // memset( x, 0, sizeof(real) * N );
-
- /* GMRES outer-loop */
- for( itr = 0; itr < MAX_ITR; ++itr ) {
- /* compute z = r0 */
- Sparse_MatVec( H, x, workspace->b_prm );
- for( i = 0; i < N; ++i )
- workspace->b_prm[i] *= workspace->Hdia_inv[i]; /* pre-conditioner */
- Vector_Sum( z[0], 1., workspace->b_prc, -1., workspace->b_prm, N );
-
- Vector_MakeZero( w, RESTART+1 );
- w[0] = Norm( z[0], N );
-
- Vector_Copy( u[0], z[0], N );
- u[0][0] += ( u[0][0] < 0.0 ? -1 : 1 ) * w[0];
- Vector_Scale( u[0], 1 / Norm( u[0], N ), u[0], N );
-
- w[0] *= ( u[0][0] < 0.0 ? 1 :-1 );
- // fprintf( stderr, "\n\n%12.6f\n", w[0] );
-
- /* GMRES inner-loop */
- for( j = 0; j < RESTART && fabs( w[j] ) / bnorm > tol; j++ ) {
- /* compute v_j */
- Vector_Scale( z[j], -2 * u[j][j], u[j], N );
- z[j][j] += 1.; /* due to e_j */
-
- for( i = j-1; i >= 0; --i )
- Vector_Add( z[j]+i, -2 * Dot( u[i]+i, z[j]+i, N-i ), u[i]+i, N-i );
-
-
- /* matvec */
- Sparse_MatVec( H, z[j], v );
-
- for( k = 0; k < N; ++k )
- v[k] *= workspace->Hdia_inv[k]; /* pre-conditioner */
-
- for( i = 0; i <= j; ++i )
- Vector_Add( v+i, -2 * Dot( u[i]+i, v+i, N-i ), u[i]+i, N-i );
-
-
- if( !Vector_isZero( v + (j+1), N - (j+1) ) ) {
- /* compute the HouseHolder unit vector u_j+1 */
- for( i = 0; i <= j; ++i )
- u[j+1][i] = 0;
-
- Vector_Copy( u[j+1] + (j+1), v + (j+1), N - (j+1) );
-
- u[j+1][j+1] += ( v[j+1]<0.0 ? -1:1 ) * Norm( v+(j+1), N-(j+1) );
-
- Vector_Scale( u[j+1], 1 / Norm( u[j+1], N ), u[j+1], N );
-
- /* overwrite v with P_m+1 * v */
- v[j+1] -= 2 * Dot( u[j+1]+(j+1), v+(j+1), N-(j+1) ) * u[j+1][j+1];
- Vector_MakeZero( v + (j+2), N - (j+2) );
- // Vector_Add( v, -2 * Dot( u[j+1], v, N ), u[j+1], N );
- }
-
-
- /* prev Givens rots on the upper-Hessenberg matrix to make it U */
- for( i = 0; i < j; i++ ) {
- tmp1 = workspace->hc[i] * v[i] + workspace->hs[i] * v[i+1];
- tmp2 = -workspace->hs[i] * v[i] + workspace->hc[i] * v[i+1];
-
- v[i] = tmp1;
- v[i+1] = tmp2;
- }
-
- /* apply the new Givens rotation to H and right-hand side */
- if( fabs(v[j+1]) >= ALMOST_ZERO ) {
- cc = SQRT( SQR( v[j] ) + SQR( v[j+1] ) );
- workspace->hc[j] = v[j] / cc;
- workspace->hs[j] = v[j+1] / cc;
-
- tmp1 = workspace->hc[j] * v[j] + workspace->hs[j] * v[j+1];
- tmp2 = -workspace->hs[j] * v[j] + workspace->hc[j] * v[j+1];
-
- v[j] = tmp1;
- v[j+1] = tmp2;
-
- /* Givens rotations to rhs */
- tmp1 = workspace->hc[j] * w[j];
- tmp2 = -workspace->hs[j] * w[j];
- w[j] = tmp1;
- w[j+1] = tmp2;
- }
-
- /* extend R */
- for( i = 0; i <= j; ++i )
- workspace->h[ index_wkspace_res (i,j) ] = v[i];
-
-
- // fprintf( stderr, "h:" );
- // for( i = 0; i <= j+1 ; ++i )
- // fprintf( stderr, "%.6f ", h[i][j] );
- // fprintf( stderr, "\n" );
- // fprintf( stderr, "%12.6f\n", w[j+1] );
- }
-
-
- /* solve Hy = w.
- H is now upper-triangular, do back-substitution */
- for( i = j-1; i >= 0; i-- ) {
- temp = w[i];
- for( k = j-1; k > i; k-- )
- temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
-
- workspace->y[i] = temp / workspace->h[ index_wkspace_res (i,i) ];
- }
-
- // fprintf( stderr, "y: " );
- // for( i = 0; i < RESTART+1; ++i )
- // fprintf( stderr, "%8.3f ", workspace->y[i] );
-
-
- /* update x = x_0 + Vy */
- // memset( z, 0, sizeof(real) * N );
- // for( i = j-1; i >= 0; i-- )
- // {
- // Vector_Copy( v, z, N );
- // v[i] += workspace->y[i];
- //
- // Vector_Sum( z, 1., v, -2 * Dot( u[i], v, N ), u[i], N );
- // }
- //
- // fprintf( stderr, "\nz: " );
- // for( k = 0; k < N; ++k )
- // fprintf( stderr, "%6.2f ", z[k] );
-
- // fprintf( stderr, "\nx_bef: " );
- // for( i = 0; i < N; ++i )
- // fprintf( stderr, "%6.2f ", x[i] );
-
- // Vector_Add( x, 1, z, N );
- for( i = j-1; i >= 0; i-- )
- Vector_Add( x, workspace->y[i], z[i], N );
-
- // fprintf( stderr, "\nx_aft: " );
- // for( i = 0; i < N; ++i )
- // fprintf( stderr, "%6.2f ", x[i] );
-
- /* stopping condition */
- if( fabs( w[j] ) / bnorm <= tol )
- break;
- }
-
- // Sparse_MatVec( H, x, workspace->b_prm );
- // for( i = 0; i < N; ++i )
- // workspace->b_prm[i] *= workspace->Hdia_inv[i];
-
- // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
- // for( i = 0; i < N; ++i )
- // fprintf( fout, "%10.5f%15.12f%15.12f\n",
- // workspace->b_prc[i], workspace->b_prm[i], x[i] );
-
- //fprintf( fout,"GMRES outer:%d, inner:%d iters - residual norm: %15.10f\n",
- // itr, j, fabs( workspace->g[j] ) / bnorm );
-
- if( itr >= MAX_ITR ) {
- fprintf( stderr, "GMRES convergence failed\n" );
- // return -1;
- return itr * (RESTART+1) + j + 1;
- }
-
- return itr * (RESTART+1) + j + 1;
-}
-
-
-int PGMRES( static_storage *workspace, sparse_matrix *H, real *b, real tol,
- sparse_matrix *L, sparse_matrix *U, real *x, FILE *fout, reax_system *system )
-{
- int i, j, k, itr, N;
- real cc, tmp1, tmp2, temp, bnorm;
-
- N = H->n;
- bnorm = Norm( b, N );
-
- /* GMRES outer-loop */
- for( itr = 0; itr < MAX_ITR; ++itr ) {
- /* calculate r0 */
- Sparse_MatVec( H, x, workspace->b_prm );
- Vector_Sum( &workspace->v[index_wkspace_sys(0,0,system)], 1., b, -1., workspace->b_prm, N );
- Forward_Subs( L, &workspace->v[index_wkspace_sys(0,0,system)], &workspace->v[index_wkspace_sys(0,0,system)] );
- Backward_Subs( U, &workspace->v[index_wkspace_sys(0,0,system)], &workspace->v[index_wkspace_sys(0,0,system)] );
- workspace->g[0] = Norm( &workspace->v[index_wkspace_sys(0,0,system)], N );
- Vector_Scale( &workspace->v[index_wkspace_sys(0,0,system)], 1. / workspace->g[0], &workspace->v[index_wkspace_sys (0,0,system)], N );
- //fprintf( stderr, "res: %.15e\n", workspace->g[0] );
-
- /* GMRES inner-loop */
- for( j = 0; j < RESTART && fabs(workspace->g[j]) / bnorm > tol; j++ ) {
- /* matvec */
- Sparse_MatVec( H, &workspace->v[index_wkspace_sys (j,0,system)], &workspace->v[index_wkspace_sys (j+1,0,system)] );
- Forward_Subs( L, &workspace->v[index_wkspace_sys(j+1,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)] );
- Backward_Subs( U, &workspace->v[index_wkspace_sys(j+1,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)] );
-
- /* apply modified Gram-Schmidt to orthogonalize the new residual */
- for( i = 0; i < j-1; i++ ) workspace->h[ index_wkspace_res (i,j)] = 0;
-
- //for( i = 0; i <= j; i++ ) {
- for( i = MAX(j-1,0); i <= j; i++ ) {
- workspace->h[index_wkspace_res (i,j)] = Dot( &workspace->v[index_wkspace_sys (i,0,system)], &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- Vector_Add( &workspace->v[index_wkspace_sys(j+1,0,system)],-workspace->h[ index_wkspace_res (i,j) ], &workspace->v[index_wkspace_sys(i,0,system)], N );
- }
-
- workspace->h[index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys (j+1,0,system)], N );
- Vector_Scale( &workspace->v[index_wkspace_sys(j+1,0,system)],
- 1. / workspace->h[ index_wkspace_res (j+1,j)], &workspace->v[index_wkspace_sys(j+1,0,system)], N );
- // fprintf( stderr, "%d-%d: orthogonalization completed.\n", itr, j );
-
- /* Givens rotations on the upper-Hessenberg matrix to make it U */
- for( i = MAX(j-1,0); i <= j; i++ ) {
- if( i == j ) {
- cc = SQRT( SQR(workspace->h[ index_wkspace_res (j,j) ])+SQR(workspace->h[ index_wkspace_res (j+1,j) ]) );
- workspace->hc[j] = workspace->h[ index_wkspace_res (j,j) ] / cc;
- workspace->hs[j] = workspace->h[ index_wkspace_res (j+1,j) ] / cc;
- }
-
- tmp1 = workspace->hc[i] * workspace->h[ index_wkspace_res (i,j) ] +
- workspace->hs[i] * workspace->h[index_wkspace_res (i+1,j) ];
- tmp2 = -workspace->hs[i] * workspace->h[index_wkspace_res (i,j)] +
- workspace->hc[i] * workspace->h[index_wkspace_res (i+1,j) ];
-
- workspace->h[ index_wkspace_res (i,j) ] = tmp1;
- workspace->h[ index_wkspace_res (i+1,j) ] = tmp2;
- }
-
- /* apply Givens rotations to the rhs as well */
- tmp1 = workspace->hc[j] * workspace->g[j];
- tmp2 = -workspace->hs[j] * workspace->g[j];
- workspace->g[j] = tmp1;
- workspace->g[j+1] = tmp2;
-
- //fprintf( stderr, "h: " );
- //for( i = 0; i <= j+1; ++i )
- //fprintf( stderr, "%.6f ", workspace->h[i][j] );
- //fprintf( stderr, "\n" );
- //fprintf( stderr, "res: %.15e\n", workspace->g[j+1] );
- }
-
-
- /* solve Hy = g: H is now upper-triangular, do back-substitution */
- for( i = j-1; i >= 0; i-- ) {
- temp = workspace->g[i];
- for( k = j-1; k > i; k-- )
- temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
-
- workspace->y[i] = temp / workspace->h[index_wkspace_res (i,i)];
- }
-
- /* update x = x_0 + Vy */
- Vector_MakeZero( workspace->p, N );
- for( i = 0; i < j; i++ )
- Vector_Add( workspace->p, workspace->y[i], &workspace->v[index_wkspace_sys(i,0,system)], N );
- //Backward_Subs( U, workspace->p, workspace->p );
- //Forward_Subs( L, workspace->p, workspace->p );
- Vector_Add( x, 1., workspace->p, N );
-
- /* stopping condition */
- if( fabs(workspace->g[j]) / bnorm <= tol )
- break;
- }
-
- // Sparse_MatVec( H, x, workspace->b_prm );
- // for( i = 0; i < N; ++i )
- // workspace->b_prm[i] *= workspace->Hdia_inv[i];
- // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
- // for( i = 0; i < N; ++i )
- // fprintf( fout, "%10.5f%15.12f%15.12f\n",
- // workspace->b_prc[i], workspace->b_prm[i], x[i] );*/
-
- // fprintf(fout,"GMRES outer:%d, inner:%d iters - residual norm: %25.20f\n",
- // itr, j, fabs( workspace->g[j] ) / bnorm );
- // data->timing.matvec += itr * RESTART + j;
-
- if( itr >= MAX_ITR ) {
- fprintf( stderr, "GMRES convergence failed\n" );
- // return -1;
- return itr * (RESTART+1) + j + 1;
- }
-
- return itr * (RESTART+1) + j + 1;
- }
-
-
-
- int PCG( static_storage *workspace, sparse_matrix *A, real *b, real tol,
- sparse_matrix *L, sparse_matrix *U, real *x, FILE *fout, reax_system* system )
- {
- int i, N;
- real tmp, alpha, beta, b_norm, r_norm;
- real sig0, sig_old, sig_new;
-
- N = A->n;
- b_norm = Norm( b, N );
- //fprintf( stderr, "b_norm: %.15e\n", b_norm );
-
- Sparse_MatVec( A, x, workspace->q );
- Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
- r_norm = Norm(workspace->r, N);
- //Print_Soln( workspace, x, q, b, N );
- //fprintf( stderr, "res: %.15e\n", r_norm );
-
- Forward_Subs( L, workspace->r, workspace->d );
- Backward_Subs( U, workspace->d, workspace->p );
- sig_new = Dot( workspace->r, workspace->p, N );
- sig0 = sig_new;
-
- for( i = 0; i < 200 && r_norm/b_norm > tol; ++i ) {
- //for( i = 0; i < 200 && sig_new > SQR(tol) * sig0; ++i ) {
- Sparse_MatVec( A, workspace->p, workspace->q );
- tmp = Dot( workspace->q, workspace->p, N );
- alpha = sig_new / tmp;
- Vector_Add( x, alpha, workspace->p, N );
- //fprintf( stderr, "iter%d: |p|=%.15e |q|=%.15e tmp=%.15e\n",
- // i+1, Norm(workspace->p,N), Norm(workspace->q,N), tmp );
-
- Vector_Add( workspace->r, -alpha, workspace->q, N );
- r_norm = Norm(workspace->r, N);
- //fprintf( stderr, "res: %.15e\n", r_norm );
-
- Forward_Subs( L, workspace->r, workspace->d );
- Backward_Subs( U, workspace->d, workspace->d );
- sig_old = sig_new;
- sig_new = Dot( workspace->r, workspace->d, N );
- beta = sig_new / sig_old;
- Vector_Sum( workspace->p, 1., workspace->d, beta, workspace->p, N );
- }
-
- //fprintf( fout, "CG took %d iterations\n", i );
- if( i >= 200 ) {
- fprintf( stderr, "CG convergence failed!\n" );
- return i;
- }
-
- return i;
- }
-
-
- int CG( static_storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, FILE *fout, reax_system *system)
- {
- int i, j, N;
- real tmp, alpha, beta, b_norm;
- real sig_old, sig_new, sig0;
-
- N = H->n;
- b_norm = Norm( b, N );
- //fprintf( stderr, "b_norm: %10.6f\n", b_norm );
-
- Sparse_MatVec( H, x, workspace->q );
- Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
- for( j = 0; j < N; ++j )
- workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
-
- sig_new = Dot( workspace->r, workspace->d, N );
- sig0 = sig_new;
- //Print_Soln( workspace, x, q, b, N );
- //fprintf( stderr, "sig_new: %24.15e, d_norm:%24.15e, q_norm:%24.15e\n",
- // sqrt(sig_new), Norm(workspace->d,N), Norm(workspace->q,N) );
- //fprintf( stderr, "sig_new: %f\n", sig_new );
-
- for( i = 0; i < 300 && SQRT(sig_new) / b_norm > tol; ++i ) {
- //for( i = 0; i < 300 && sig_new > SQR(tol)*sig0; ++i ) {
- Sparse_MatVec( H, workspace->d, workspace->q );
- tmp = Dot( workspace->d, workspace->q, N );
- //fprintf( stderr, "tmp: %f\n", tmp );
- alpha = sig_new / tmp;
- Vector_Add( x, alpha, workspace->d, N );
- //fprintf( stderr, "d_norm:%24.15e, q_norm:%24.15e, tmp:%24.15e\n",
- // Norm(workspace->d,N), Norm(workspace->q,N), tmp );
-
- Vector_Add( workspace->r, -alpha, workspace->q, N );
- for( j = 0; j < N; ++j )
- workspace->p[j] = workspace->r[j] * workspace->Hdia_inv[j];
-
- sig_old = sig_new;
- sig_new = Dot( workspace->r, workspace->p, N );
- beta = sig_new / sig_old;
- Vector_Sum( workspace->d, 1., workspace->p, beta, workspace->d, N );
- //fprintf( stderr, "sig_new: %f\n", sig_new );
- }
-
- fprintf( stderr, "CG took %d iterations\n", i );
-
- if( i >= 300 ) {
- fprintf( stderr, "CG convergence failed!\n" );
- return i;
- }
-
- return i;
- }
-
-
-
- /* Steepest Descent */
- int SDM( static_storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, FILE *fout )
- {
- int i, j, N;
- real tmp, alpha, beta, b_norm;
- real sig0, sig;
-
- N = H->n;
- b_norm = Norm( b, N );
- //fprintf( stderr, "b_norm: %10.6f\n", b_norm );
-
- Sparse_MatVec( H, x, workspace->q );
- Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
- for( j = 0; j < N; ++j )
- workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
-
- sig = Dot( workspace->r, workspace->d, N );
- sig0 = sig;
-
- for( i = 0; i < 300 && SQRT(sig) / b_norm > tol; ++i ) {
- Sparse_MatVec( H, workspace->d, workspace->q );
-
- sig = Dot( workspace->r, workspace->d, N );
- tmp = Dot( workspace->d, workspace->q, N );
- alpha = sig / tmp;
-
- Vector_Add( x, alpha, workspace->d, N );
- Vector_Add( workspace->r, -alpha, workspace->q, N );
- for( j = 0; j < N; ++j )
- workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
-
- //fprintf( stderr, "d_norm:%24.15e, q_norm:%24.15e, tmp:%24.15e\n",
- // Norm(workspace->d,N), Norm(workspace->q,N), tmp );
- }
-
- fprintf( stderr, "SDM took %d iterations\n", i );
-
- if( i >= 300 ) {
- fprintf( stderr, "SDM convergence failed!\n" );
- return i;
- }
-
- return i;
- }
-
diff --git a/PuReMD-GPU/src/QEq.c b/PuReMD-GPU/src/QEq.c
new file mode 100644
index 0000000000000000000000000000000000000000..8cc638ea90dcc25f86d33f275b162c8e531d82bb
--- /dev/null
+++ b/PuReMD-GPU/src/QEq.c
@@ -0,0 +1,396 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "QEq.h"
+
+#include "allocate.h"
+#include "lin_alg.h"
+#include "list.h"
+#include "print_utils.h"
+#include "index_utils.h"
+#include "system_props.h"
+
+#include "sort.h"
+
+
+int compare_matrix_entry(const void *v1, const void *v2)
+{
+ return ((sparse_matrix_entry *)v1)->j - ((sparse_matrix_entry *)v2)->j;
+}
+
+
+void Sort_Matrix_Rows( sparse_matrix *A )
+{
+ int i, si, ei;
+
+ for( i = 0; i < A->n; ++i ) {
+ si = A->start[i];
+ ei = A->start[i+1];
+ qsort( &(A->entries[si]), ei - si,
+ sizeof(sparse_matrix_entry), compare_matrix_entry );
+ }
+}
+
+
+void Calculate_Droptol( sparse_matrix *A, real *droptol, real dtol )
+{
+ int i, j, k;
+ real val;
+
+ /* init droptol to 0 */
+ for( i = 0; i < A->n; ++i )
+ droptol[i] = 0;
+
+ /* calculate sqaure of the norm of each row */
+ for( i = 0; i < A->n; ++i ) {
+ for( k = A->start[i]; k < A->start[i+1]-1; ++k ) {
+ j = A->entries[k].j;
+ val = A->entries[k].val;
+
+ droptol[i] += val*val;
+ droptol[j] += val*val;
+ }
+
+ val = A->entries[k].val; // diagonal entry
+ droptol[i] += val*val;
+ }
+
+ /* calculate local droptol for each row */
+ //fprintf( stderr, "droptol: " );
+ for( i = 0; i < A->n; ++i ) {
+ //fprintf( stderr, "%f-->", droptol[i] );
+ droptol[i] = SQRT( droptol[i] ) * dtol;
+ //fprintf( stderr, "%f ", droptol[i] );
+ }
+ //fprintf( stderr, "\n" );
+}
+
+
+int Estimate_LU_Fill( sparse_matrix *A, real *droptol )
+{
+ int i, j, pj;
+ int fillin;
+ real val;
+
+ fillin = 0;
+
+ //fprintf( stderr, "n: %d\n", A->n );
+ for( i = 0; i < A->n; ++i )
+ for( pj = A->start[i]; pj < A->start[i+1]-1; ++pj ){
+ j = A->entries[pj].j;
+ val = A->entries[pj].val;
+ //fprintf( stderr, "i: %d, j: %d", i, j );
+
+ if( fabs(val) > droptol[i] )
+ ++fillin;
+ }
+
+ return fillin + A->n;
+}
+
+
+void ICHOLT( sparse_matrix *A, real *droptol,
+ sparse_matrix *L, sparse_matrix *U )
+{
+ sparse_matrix_entry tmp[1000];
+ int i, j, pj, k1, k2, tmptop, Ltop;
+ real val;
+ int *Utop;
+
+ Utop = (int*) malloc((A->n+1) * sizeof(int));
+
+ // clear variables
+ Ltop = 0;
+ tmptop = 0;
+ for( i = 0; i <= A->n; ++i )
+ L->start[i] = U->start[i] = 0;
+
+ for( i = 0; i < A->n; ++i )
+ Utop[i] = 0;
+
+ //fprintf( stderr, "n: %d\n", A->n );
+ for( i = 0; i < A->n; ++i ){
+ L->start[i] = Ltop;
+ tmptop = 0;
+
+ for( pj = A->start[i]; pj < A->start[i+1]-1; ++pj ){
+ j = A->entries[pj].j;
+ val = A->entries[pj].val;
+ //fprintf( stderr, "i: %d, j: %d", i, j );
+
+ if( fabs(val) > droptol[i] ){
+ k1 = 0;
+ k2 = L->start[j];
+ while( k1 < tmptop && k2 < L->start[j+1] ){
+ if( tmp[k1].j < L->entries[k2].j )
+ ++k1;
+ else if( tmp[k1].j > L->entries[k2].j )
+ ++k2;
+ else
+ val -= (tmp[k1++].val * L->entries[k2++].val);
+ }
+
+ // L matrix is lower triangular,
+ // so right before the start of next row comes jth diagonal
+ val /= L->entries[L->start[j+1]-1].val;
+
+ tmp[tmptop].j = j;
+ tmp[tmptop].val = val;
+ ++tmptop;
+ }
+ //fprintf( stderr, " -- done\n" );
+ }
+
+ // compute the ith diagonal in L
+ // sanity check
+ if( A->entries[pj].j != i ) {
+ fprintf( stderr, "i=%d, badly built A matrix!\n", i );
+ exit(999);
+ }
+
+ val = A->entries[pj].val;
+ for( k1 = 0; k1 < tmptop; ++k1 )
+ val -= (tmp[k1].val * tmp[k1].val);
+
+ tmp[tmptop].j = i;
+ tmp[tmptop].val = SQRT(val);
+
+ // apply the dropping rule once again
+ //fprintf( stderr, "row%d: tmptop: %d\n", i, tmptop );
+ //for( k1 = 0; k1<= tmptop; ++k1 )
+ // fprintf( stderr, "%d(%f) ", tmp[k1].j, tmp[k1].val );
+ //fprintf( stderr, "\n" );
+ //fprintf( stderr, "row(%d): droptol=%.4f\n", i+1, droptol[i] );
+ for( k1 = 0; k1 < tmptop; ++k1 )
+ if( fabs(tmp[k1].val) > droptol[i] / tmp[tmptop].val ){
+ L->entries[Ltop].j = tmp[k1].j;
+ L->entries[Ltop].val = tmp[k1].val;
+ U->start[tmp[k1].j+1]++;
+ ++Ltop;
+ //fprintf( stderr, "%d(%.4f) ", tmp[k1].j+1, tmp[k1].val );
+ }
+ // keep the diagonal in any case
+ L->entries[Ltop].j = tmp[k1].j;
+ L->entries[Ltop].val = tmp[k1].val;
+ ++Ltop;
+ //fprintf( stderr, "%d(%.4f)\n", tmp[k1].j+1, tmp[k1].val );
+ }
+
+ L->start[i] = Ltop;
+ //fprintf( stderr, "nnz(L): %d, max: %d\n", Ltop, L->n * 50 );
+
+ for( i = 1; i <= U->n; ++i )
+ Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
+
+ for( i = 0; i < L->n; ++i )
+ for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
+ j = L->entries[pj].j;
+ U->entries[Utop[j]].j = i;
+ U->entries[Utop[j]].val = L->entries[pj].val;
+ Utop[j]++;
+ }
+
+ //fprintf( stderr, "nnz(U): %d, max: %d\n", Utop[U->n], U->n * 50 );
+}
+
+
+void Init_MatVec( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list *far_nbrs )
+{
+ int i, fillin;
+ real s_tmp, t_tmp;
+ //char fname[100];
+
+ if(control->refactor > 0 &&
+ ((data->step-data->prev_steps)%control->refactor==0 || workspace->L.entries==NULL))
+ {
+ //Print_Linear_System( system, control, workspace, data->step );
+ Sort_Matrix_Rows( &workspace->H );
+
+ //fprintf( stderr, "H matrix sorted\n" );
+
+ Calculate_Droptol( &workspace->H, workspace->droptol, control->droptol );
+ //fprintf( stderr, "drop tolerances calculated\n" );
+
+ if( workspace->L.entries == NULL )
+ {
+ fillin = Estimate_LU_Fill( &workspace->H, workspace->droptol );
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "fillin = %d\n", fillin );
+#endif
+
+ if( Allocate_Matrix( &(workspace->L), far_nbrs->n, fillin ) == 0 ||
+ Allocate_Matrix( &(workspace->U), far_nbrs->n, fillin ) == 0 )
+ {
+ fprintf( stderr, "not enough memory for LU matrices. terminating.\n" );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "fillin = %d\n", fillin );
+ fprintf( stderr, "allocated memory: L = U = %ldMB\n",
+ fillin * sizeof(sparse_matrix_entry) / (1024*1024) );
+#endif
+ }
+
+ ICHOLT( &workspace->H, workspace->droptol, &workspace->L, &workspace->U );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "icholt-" );
+ //sprintf( fname, "%s.L%d.out", control->sim_name, data->step );
+ //Print_Sparse_Matrix2( workspace->L, fname );
+ //Print_Sparse_Matrix( U );
+#endif
+ }
+
+ /* extrapolation for s & t */
+ for( i = 0; i < system->N; ++i ) {
+ // no extrapolation
+ //s_tmp = workspace->s[0][i];
+ //t_tmp = workspace->t[0][i];
+
+ // linear
+ //s_tmp = 2 * workspace->s[0][i] - workspace->s[1][i];
+ //t_tmp = 2 * workspace->t[0][i] - workspace->t[1][i];
+
+ // quadratic
+ //s_tmp = workspace->s[2][i] + 3 * (workspace->s[0][i]-workspace->s[1][i]);
+ t_tmp = workspace->t[index_wkspace_sys(2,i,system->N)] + 3*(workspace->t[index_wkspace_sys(0,i,system->N)]-workspace->t[index_wkspace_sys(1,i,system->N)]);
+
+ // cubic
+ s_tmp = 4 * (workspace->s[index_wkspace_sys(0,i,system->N)] + workspace->s[index_wkspace_sys(2,i,system->N)]) -
+ (6 * workspace->s[index_wkspace_sys(1,i,system->N)] + workspace->s[index_wkspace_sys(3,i,system->N)] );
+ //t_tmp = 4 * (workspace->t[0][i] + workspace->t[2][i]) -
+ // (6 * workspace->t[1][i] + workspace->t[3][i] );
+
+ // 4th order
+ //s_tmp = 5 * (workspace->s[0][i] - workspace->s[3][i]) +
+ // 10 * (-workspace->s[1][i] + workspace->s[2][i] ) + workspace->s[4][i];
+ //t_tmp = 5 * (workspace->t[0][i] - workspace->t[3][i]) +
+ // 10 * (-workspace->t[1][i] + workspace->t[2][i] ) + workspace->t[4][i];
+
+ workspace->s[index_wkspace_sys(4,i,system->N)] = workspace->s[index_wkspace_sys(3,i,system->N)];
+ workspace->s[index_wkspace_sys(3,i,system->N)] = workspace->s[index_wkspace_sys(2,i,system->N)];
+ workspace->s[index_wkspace_sys(2,i,system->N)] = workspace->s[index_wkspace_sys(1,i,system->N)];
+ workspace->s[index_wkspace_sys(1,i,system->N)] = workspace->s[index_wkspace_sys(0,i,system->N)];
+ workspace->s[index_wkspace_sys(0,i,system->N)] = s_tmp;
+
+ workspace->t[index_wkspace_sys(4,i,system->N)] = workspace->t[index_wkspace_sys(3,i,system->N)];
+ workspace->t[index_wkspace_sys(3,i,system->N)] = workspace->t[index_wkspace_sys(2,i,system->N)];
+ workspace->t[index_wkspace_sys(2,i,system->N)] = workspace->t[index_wkspace_sys(1,i,system->N)];
+ workspace->t[index_wkspace_sys(1,i,system->N)] = workspace->t[index_wkspace_sys(0,i,system->N)];
+ workspace->t[index_wkspace_sys(0,i,system->N)] = t_tmp;
+ }
+}
+
+
+void Calculate_Charges( reax_system *system, static_storage *workspace )
+{
+ int i;
+ real u, s_sum, t_sum;
+
+ s_sum = t_sum = 0.;
+ for( i = 0; i < system->N; ++i ) {
+ s_sum += workspace->s[index_wkspace_sys(0,i,system->N)];
+ t_sum += workspace->t[index_wkspace_sys(0,i,system->N)];
+ }
+
+ u = s_sum / t_sum;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Host --->s %13.2f, t %13.f, u %13.2f \n", s_sum, t_sum, u );
+#endif
+
+ for( i = 0; i < system->N; ++i )
+ {
+ system->atoms[i].q = workspace->s[index_wkspace_sys(0,i,system->N)] - u * workspace->t[index_wkspace_sys(0,i,system->N)];
+ }
+}
+
+
+void QEq( reax_system *system, control_params *control, simulation_data *data,
+ static_storage *workspace, list *far_nbrs,
+ output_controls *out_control )
+{
+ int matvecs;
+
+ //real t_start, t_elapsed;
+
+ //t_start = Get_Time ();
+ Init_MatVec( system, control, data, workspace, far_nbrs );
+ //t_elapsed = Get_Timing_Info ( t_start );
+
+ //fprintf (stderr, " CPU Init_MatVec timing ----> %f \n", t_elapsed );
+
+ //if( data->step % 10 == 0 )
+ // Print_Linear_System( system, control, workspace, far_nbrs, data->step );
+
+ //t_start = Get_Time ( );
+ matvecs = GMRES( workspace, &workspace->H,
+ workspace->b_s, control->q_err, &workspace->s[0], out_control->log, system );
+ matvecs += GMRES( workspace, &workspace->H,
+ workspace->b_t, control->q_err, &workspace->t[0], out_control->log, system );
+ //t_elapsed = Get_Timing_Info ( t_start );
+
+ //fprintf (stderr, " CPU GMRES timing ---> %f \n", t_elapsed );
+
+ //matvecs = GMRES_HouseHolder( workspace, workspace->H,
+ // workspace->b_s, control->q_err, workspace->s[0], out_control->log );
+ //matvecs += GMRES_HouseHolder( workspace, workspace->H,
+ // workspace->b_t, control->q_err, workspace->t[0], out_control->log );
+
+ //matvecs = PGMRES( workspace, &workspace->H, workspace->b_s, control->q_err,
+ // &workspace->L, &workspace->U, &workspace->s[index_wkspace_sys(0,0,system->N)], out_control->log, system );
+ //matvecs += PGMRES( workspace, &workspace->H, workspace->b_t, control->q_err,
+ // &workspace->L, &workspace->U, &workspace->t[index_wkspace_sys(0,0,system->N)], out_control->log, system );
+
+ //matvecs=PCG( workspace, workspace->H, workspace->b_s, control->q_err,
+ // workspace->L, workspace->U, workspace->s[0], out_control->log ) + 1;
+ ///matvecs+=PCG( workspace, workspace->H, workspace->b_t, control->q_err,
+ // workspace->L, workspace->U, workspace->t[0], out_control->log ) + 1;
+
+ //matvecs = CG( workspace, workspace->H,
+ // workspace->b_s, control->q_err, workspace->s[0], out_control->log ) + 1;
+ //matvecs += CG( workspace, workspace->H,
+ // workspace->b_t, control->q_err, workspace->t[0], out_control->log ) + 1;
+
+ //matvecs = SDM( workspace, workspace->H,
+ // workspace->b_s, control->q_err, workspace->s[0], out_control->log ) + 1;
+ //matvecs += SDM( workspace, workspace->H,
+ // workspace->b_t, control->q_err, workspace->t[0], out_control->log ) + 1;
+
+ //fprintf (stderr, " GMRES done with iterations %d \n", matvecs );
+
+ data->timing.matvecs += matvecs;
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "linsolve-" );
+#endif
+
+ Calculate_Charges( system, workspace );
+ //fprintf( stderr, "%d %.9f %.9f %.9f %.9f %.9f %.9f\n",
+ // data->step,
+ // workspace->s[0][0], workspace->t[0][0],
+ // workspace->s[0][1], workspace->t[0][1],
+ // workspace->s[0][2], workspace->t[0][2] );
+ // if( data->step == control->nsteps )
+ //Print_Charges( system, control, workspace, data->step );
+}
diff --git a/PuReMD-GPU/src/QEq.cu b/PuReMD-GPU/src/QEq.cu
deleted file mode 100644
index 5d849b261b2e8396ec2243f6985f12e335c80430..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/QEq.cu
+++ /dev/null
@@ -1,1073 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "QEq.h"
-#include "allocate.h"
-#include "GMRES.h"
-#include "list.h"
-#include "print_utils.h"
-#include "index_utils.h"
-
-#include "cuda_utils.h"
-#include "cuda_init.h"
-#include "cuda_copy.h"
-#include "sort.h"
-#include "validation.h"
-#include "reduction.h"
-
-#include "system_props.h"
-
-HOST_DEVICE void swap(sparse_matrix_entry *array, int index1, int index2)
-{
- sparse_matrix_entry temp = array[index1];
- array[index1] = array[index2];
- array[index2] = temp;
-}
-
-HOST_DEVICE void quick_sort(sparse_matrix_entry *array, int start, int end)
-{
- int i = start;
- int k = end;
-
- if (end - start >= 1)
- {
- int pivot = array[start].j;
-
- while (k > i)
- {
- while ((array[i].j <= pivot) && (i <= end) && (k > i)) i++;
- while ((array[k].j > pivot) && (k >= start) && (k >= i)) k--;
- if (k > i) swap(array, i, k);
- }
- swap(array, start, k);
- quick_sort(array, start, k - 1);
- quick_sort(array, k + 1, end);
- }
-}
-
-int compare_matrix_entry(const void *v1, const void *v2)
-{
- return ((sparse_matrix_entry *)v1)->j - ((sparse_matrix_entry *)v2)->j;
-}
-
-
-void Sort_Matrix_Rows( sparse_matrix *A )
-{
- int i, si, ei;
-
- for( i = 0; i < A->n; ++i ) {
- si = A->start[i];
- ei = A->start[i+1];
- qsort( &(A->entries[si]), ei - si,
- sizeof(sparse_matrix_entry), compare_matrix_entry );
- }
-}
-
-GLOBAL void Cuda_Sort_Matrix_Rows ( sparse_matrix A )
-{
- int i;
- int si, ei;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( i >= A.n ) return;
-
- si = A.start[i];
- ei = A.end [i];
-
- quick_sort( A.entries + si, 0, ei-si-1 );
-}
-
-
-void Calculate_Droptol( sparse_matrix *A, real *droptol, real dtol )
-{
- int i, j, k;
- real val;
-
- /* init droptol to 0 */
- for( i = 0; i < A->n; ++i )
- droptol[i] = 0;
-
- /* calculate sqaure of the norm of each row */
- for( i = 0; i < A->n; ++i ) {
- for( k = A->start[i]; k < A->start[i+1]-1; ++k ) {
- j = A->entries[k].j;
- val = A->entries[k].val;
-
- droptol[i] += val*val;
- droptol[j] += val*val;
- }
-
- val = A->entries[k].val; // diagonal entry
- droptol[i] += val*val;
- }
-
- /* calculate local droptol for each row */
- //fprintf( stderr, "droptol: " );
- for( i = 0; i < A->n; ++i ) {
- //fprintf( stderr, "%f-->", droptol[i] );
- droptol[i] = SQRT( droptol[i] ) * dtol;
- //fprintf( stderr, "%f ", droptol[i] );
- }
- //fprintf( stderr, "\n" );
-}
-
-GLOBAL void Cuda_Calculate_Droptol ( sparse_matrix p_A, real *droptol, real dtol )
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- int k, j, offset, x, diagnol;
- real val;
- sparse_matrix *A = &p_A;
-
- if ( i < A->n ) {
- droptol [i] = 0;
-
- for (k = A->start[i]; k < A->end[i]; ++k ) {
- val = A->entries[k].val;
- droptol [i] += val*val;
- }
- }
-
- __syncthreads ();
- if ( i < A->n ) {
- droptol [i] = SQRT (droptol[i]) * dtol;
- }
-
-}
-
-GLOBAL void Cuda_Calculate_Droptol_js ( sparse_matrix p_A, real *droptol, real dtol )
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- int k, j, offset, x, diagnol;
- real val;
- sparse_matrix *A = &p_A;
-
- for (x = 0; x < A->n; x ++)
- {
- if (i < (A->end[i]-1 - A->start[i])) {
- offset = A->start [i] + i;
- j = A->entries[offset].j;
- val = A->entries[offset].val;
- droptol [j] += val * val;
- }
- __syncthreads ();
- }
-}
-
-GLOBAL void Cuda_Calculate_Droptol_diagnol ( sparse_matrix p_A, real *droptol, real dtol )
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- int k, j, offset, x, diagnol;
- real val;
- sparse_matrix *A = &p_A;
-
- if ( i < A->n ) {
- //diagnol element
- diagnol = A->end[i]-1;
- val = A->entries [diagnol].val;
- droptol [i] += val*val;
- }
-
- /*calculate local droptol for each row*/
- if ( i < A->n )
- droptol [i] = SQRT (droptol[i]) * dtol;
-}
-
-
-int Estimate_LU_Fill( sparse_matrix *A, real *droptol )
-{
- int i, j, pj;
- int fillin;
- real val;
-
- fillin = 0;
-
- //fprintf( stderr, "n: %d\n", A->n );
- for( i = 0; i < A->n; ++i )
- for( pj = A->start[i]; pj < A->start[i+1]-1; ++pj ){
- j = A->entries[pj].j;
- val = A->entries[pj].val;
- //fprintf( stderr, "i: %d, j: %d", i, j );
-
- if( fabs(val) > droptol[i] )
- ++fillin;
- }
-
- return fillin + A->n;
-}
-
-GLOBAL void Cuda_Estimate_LU_Fill ( sparse_matrix p_A, real *droptol, int *fillin)
-{
- int i, j, pj;
- real val;
- sparse_matrix *A = &p_A;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= A->n) return;
-
- fillin [i] = 0;
-
- for (pj = A->start[i]; pj < A->end[i]-1; ++pj)
- {
- j = A->entries [pj].j;
- val = A->entries[pj].val;
-
- if (fabs (val) > droptol [i]) ++fillin [i];
- }
-}
-
-void ICHOLT( sparse_matrix *A, real *droptol,
- sparse_matrix *L, sparse_matrix *U )
-{
- sparse_matrix_entry tmp[1000];
- int i, j, pj, k1, k2, tmptop, Ltop;
- real val;
- int *Utop;
-
- Utop = (int*) malloc((A->n+1) * sizeof(int));
-
- // clear variables
- Ltop = 0;
- tmptop = 0;
- for( i = 0; i <= A->n; ++i )
- L->start[i] = U->start[i] = 0;
-
- for( i = 0; i < A->n; ++i )
- Utop[i] = 0;
-
- //fprintf( stderr, "n: %d\n", A->n );
- for( i = 0; i < A->n; ++i ){
- L->start[i] = Ltop;
- tmptop = 0;
-
- for( pj = A->start[i]; pj < A->start[i+1]-1; ++pj ){
- j = A->entries[pj].j;
- val = A->entries[pj].val;
- //fprintf( stderr, "i: %d, j: %d", i, j );
-
- if( fabs(val) > droptol[i] ){
- k1 = 0;
- k2 = L->start[j];
- while( k1 < tmptop && k2 < L->start[j+1] ){
- if( tmp[k1].j < L->entries[k2].j )
- ++k1;
- else if( tmp[k1].j > L->entries[k2].j )
- ++k2;
- else
- val -= (tmp[k1++].val * L->entries[k2++].val);
- }
-
- // L matrix is lower triangular,
- // so right before the start of next row comes jth diagonal
- val /= L->entries[L->start[j+1]-1].val;
-
- tmp[tmptop].j = j;
- tmp[tmptop].val = val;
- ++tmptop;
- }
- //fprintf( stderr, " -- done\n" );
- }
-
- // compute the ith diagonal in L
- // sanity check
- if( A->entries[pj].j != i ) {
- fprintf( stderr, "i=%d, badly built A matrix!\n", i );
- exit(999);
- }
-
- val = A->entries[pj].val;
- for( k1 = 0; k1 < tmptop; ++k1 )
- val -= (tmp[k1].val * tmp[k1].val);
-
- tmp[tmptop].j = i;
- tmp[tmptop].val = SQRT(val);
-
- // apply the dropping rule once again
- //fprintf( stderr, "row%d: tmptop: %d\n", i, tmptop );
- //for( k1 = 0; k1<= tmptop; ++k1 )
- // fprintf( stderr, "%d(%f) ", tmp[k1].j, tmp[k1].val );
- //fprintf( stderr, "\n" );
- //fprintf( stderr, "row(%d): droptol=%.4f\n", i+1, droptol[i] );
- for( k1 = 0; k1 < tmptop; ++k1 )
- if( fabs(tmp[k1].val) > droptol[i] / tmp[tmptop].val ){
- L->entries[Ltop].j = tmp[k1].j;
- L->entries[Ltop].val = tmp[k1].val;
- U->start[tmp[k1].j+1]++;
- ++Ltop;
- //fprintf( stderr, "%d(%.4f) ", tmp[k1].j+1, tmp[k1].val );
- }
- // keep the diagonal in any case
- L->entries[Ltop].j = tmp[k1].j;
- L->entries[Ltop].val = tmp[k1].val;
- ++Ltop;
- //fprintf( stderr, "%d(%.4f)\n", tmp[k1].j+1, tmp[k1].val );
- }
-
- L->start[i] = Ltop;
- //fprintf( stderr, "nnz(L): %d, max: %d\n", Ltop, L->n * 50 );
-
- for( i = 1; i <= U->n; ++i )
- Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
-
- for( i = 0; i < L->n; ++i )
- for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
- j = L->entries[pj].j;
- U->entries[Utop[j]].j = i;
- U->entries[Utop[j]].val = L->entries[pj].val;
- Utop[j]++;
- }
-
- //fprintf( stderr, "nnz(U): %d, max: %d\n", Utop[U->n], U->n * 50 );
-}
-
-
-
-void Cuda_ICHOLT( sparse_matrix *A, real *droptol,
- sparse_matrix *L, sparse_matrix *U )
-{
- sparse_matrix_entry tmp[1000];
- int i, j, pj, k1, k2, tmptop, Ltop;
- real val;
- int *Utop;
-
- Utop = (int*) malloc((A->n+1) * sizeof(int));
-
- // clear variables
- Ltop = 0;
- tmptop = 0;
- for( i = 0; i <= A->n; ++i )
- L->start[i] = U->start[i] = 0;
-
- for( i = 0; i < A->n; ++i )
- Utop[i] = 0;
-
- //fprintf( stderr, "n: %d\n", A->n );
- for( i = 0; i < A->n; ++i ){
- L->start[i] = Ltop;
- tmptop = 0;
-
- for( pj = A->start[i]; pj < A->end[i]-1; ++pj ){
- j = A->entries[pj].j;
- val = A->entries[pj].val;
- //fprintf( stderr, "i: %d, j: %d", i, j );
-
- //CHANGE ORIGINAL
- if (j >= i) break;
- //CHANGE ORIGINAL
-
- if( fabs(val) > droptol[i] ){
- k1 = 0;
- k2 = L->start[j];
- while( k1 < tmptop && k2 < L->start[j+1] ){
- if( tmp[k1].j < L->entries[k2].j )
- ++k1;
- else if( tmp[k1].j > L->entries[k2].j )
- ++k2;
- else
- val -= (tmp[k1++].val * L->entries[k2++].val);
- }
-
- // L matrix is lower triangular,
- // so right before the start of next row comes jth diagonal
- val /= L->entries[L->start[j+1]-1].val;
-
- tmp[tmptop].j = j;
- tmp[tmptop].val = val;
- ++tmptop;
- }
-
- //fprintf( stderr, " -- done\n" );
- }
-
- // compute the ith diagonal in L
- // sanity check
- if( A->entries[pj].j != i ) {
- fprintf( stderr, "i=%d, badly built A matrix!\n", i );
- exit(999);
- }
-
- val = A->entries[pj].val;
- for( k1 = 0; k1 < tmptop; ++k1 )
- val -= (tmp[k1].val * tmp[k1].val);
-
- tmp[tmptop].j = i;
- tmp[tmptop].val = SQRT(val);
-
- // apply the dropping rule once again
- //fprintf( stderr, "row%d: tmptop: %d\n", i, tmptop );
- //for( k1 = 0; k1<= tmptop; ++k1 )
- // fprintf( stderr, "%d(%f) ", tmp[k1].j, tmp[k1].val );
- //fprintf( stderr, "\n" );
- //fprintf( stderr, "row(%d): droptol=%.4f\n", i+1, droptol[i] );
- for( k1 = 0; k1 < tmptop; ++k1 )
- if( fabs(tmp[k1].val) > droptol[i] / tmp[tmptop].val ){
- L->entries[Ltop].j = tmp[k1].j;
- L->entries[Ltop].val = tmp[k1].val;
- U->start[tmp[k1].j+1]++;
- ++Ltop;
- //fprintf( stderr, "%d(%.4f) ", tmp[k1].j+1, tmp[k1].val );
- }
- // keep the diagonal in any case
- L->entries[Ltop].j = tmp[k1].j;
- L->entries[Ltop].val = tmp[k1].val;
- ++Ltop;
- //fprintf( stderr, "%d(%.4f)\n", tmp[k1].j+1, tmp[k1].val );
- }
-
- L->start[i] = Ltop;
- //fprintf( stderr, "nnz(L): %d, max: %d\n", Ltop, L->n * 50 );
-
- for( i = 1; i <= U->n; ++i )
- Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
-
- for( i = 0; i < L->n; ++i )
- for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
- j = L->entries[pj].j;
- U->entries[Utop[j]].j = i;
- U->entries[Utop[j]].val = L->entries[pj].val;
- Utop[j]++;
- }
-
- //fprintf( stderr, "nnz(U): %d, max: %d\n", Utop[U->n], U->n * 50 );
-}
-
-
-
-/*
-//Parallel for each row
-//Each kernel will run for 6540 number of times.
-GLOBAL void Cuda_ICHOLT( reax_system *system, sparse_matrix p_A, real *droptol,
-sparse_matrix p_L, sparse_matrix p_U )
-{
-int start, end, count;
-real tempvalue, val;
-int i,pj,tmptop, offset;
-int j, k1, k2;
-
-sparse_matrix *A, *L, *U;
-sparse_matrix_entry *tmp;
-
-A = &p_A;
-L = &p_L;
-U = &p_U;
-
-real *null_val;
-null_val = 0;
-
-extern __shared__ real tmp_val[];
-extern __shared__ sparse_matrix_entry sh_tmp[];
-
-int kid = blockIdx.x * blockDim.x + threadIdx.x;
-tmp = (sparse_matrix_entry *) (tmp_val + blockDim.x);
-
-offset = 0;
-for( i = 0; i < 10; ++i )
-{
-//if (kid == 0) L->start[i] = i * system->max_sparse_matrix_entries;
-if (kid == 0) L->start[i] = offset;
-tmptop = 0;
-
-start = A->start[i];
-end = A->end[i]-1; //inclusive
-count = end - start; //inclusive
-tmp_val [kid] = 0;
-
-if (kid < count) //diagnol not included
-{
-pj = start + kid;
-
-j = A->entries[pj].j;
-val = A->entries[pj].val;
-
-if( fabs(val) > droptol[i] )
-{
-k1 = 0;
-k2 = L->start[j];
-while( k1 < tmptop && k2 < L->end[j] ){
-if( tmp[k1].j < L->entries[k2].j )
-++k1;
-else if( tmp[k1].j > L->entries[k2].j )
-++k2;
-else
-tmp_val[kid] = (tmp[k1++].val * L->entries[k2++].val);
-}
-
-//here read the shared memory of all the kernels
-if (kid == 0)
-{
-for (i = 0; i < count; i++)
-tempvalue += tmp_val [i];
-
-val -= tempvalue;
-
-// L matrix is lower triangular,
-// so right before the start of next row comes jth diagonal
-val /= L->entries[L->end[j]-1].val;
-
-tmp[tmptop].j = j;
-tmp[tmptop].val = val;
-++tmptop;
-}
-}
-}
-__syncthreads ();
-
-
-// compute the ith diagonal in L
-// sanity check
-if (kid == 0)
-{
- if( A->entries[end].j != i ) {
- //intentional core dump here for sanity sake
- *null_val = 1;
- }
-}
-
-//diagnol element
-//val = A->entries[pj].val;
-//for( k1 = 0; k1 < tmptop; ++k1 )
-if (kid < count)
- tmp_val[kid] = (tmp[kid].val * tmp[kid].val);
-
- __syncthreads ();
-
-if (kid == 0)
-{
- val = A->entries [end].val;
- for (i = 0; i < count; i++)
- tempvalue += tmp_val [i];
-
- val -= tempvalue;
- tmp[tmptop].j = i;
- tmp[tmptop].val = SQRT(val);
-}
-__syncthreads ();
-
-//Fill in the LU entries
-//for( k1 = 0; k1 < count; ++k1 )
-if (kid < count )
-{
- if( fabs(tmp[kid].val) > droptol[i] / tmp[tmptop].val ){
- L->entries[offset + kid].j = tmp[kid].j;
- L->entries[offset + kid].val = tmp[kid].val;
- U->start[tmp[kid].j+1]++;
- }
-}
-__syncthreads ();
-
-if (kid == 0) {
- // keep the diagonal in any case
- offset += count;
- L->entries[offset].j = tmp[count].j;
- L->entries[offset].val = tmp[count].val;
- ++offset;
- L->end [i] = offset;
-}
-__syncthreads ();
-} // end of main for loop
-}
-
-void Cuda_Fill_U ( sparse_matrix *A, real *droptol,
- sparse_matrix *L, sparse_matrix *U )
-{
- int i, pj, j;
-
- for( i = 1; i <= U->n; ++i )
- Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
-
- for( i = 0; i < L->n; ++i )
- for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
- j = L->entries[pj].j;
- U->entries[Utop[j]].j = i;
- U->entries[Utop[j]].val = L->entries[pj].val;
- Utop[j]++;
- }
-}
-*/
-
-
-void Init_MatVec( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list *far_nbrs )
-{
- int i, fillin;
- real s_tmp, t_tmp;
- //char fname[100];
-
- if(control->refactor > 0 &&
- ((data->step-data->prev_steps)%control->refactor==0 || workspace->L.entries==NULL)){
- //Print_Linear_System( system, control, workspace, data->step );
- Sort_Matrix_Rows( &workspace->H );
-
- //fprintf( stderr, "H matrix sorted\n" );
-
- Calculate_Droptol( &workspace->H, workspace->droptol, control->droptol );
- //fprintf( stderr, "drop tolerances calculated\n" );
-
-
- if( workspace->L.entries == NULL ) {
- fillin = Estimate_LU_Fill( &workspace->H, workspace->droptol );
-#ifdef __DEBUG_CUDA__
- fprintf( stderr, "fillin = %d\n", fillin );
-#endif
- if( Allocate_Matrix( &(workspace->L), far_nbrs->n, fillin ) == 0 ||
- Allocate_Matrix( &(workspace->U), far_nbrs->n, fillin ) == 0 ){
- fprintf( stderr, "not enough memory for LU matrices. terminating.\n" );
- exit(INSUFFICIENT_SPACE);
- }
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "fillin = %d\n", fillin );
- fprintf( stderr, "allocated memory: L = U = %ldMB\n",
- fillin * sizeof(sparse_matrix_entry) / (1024*1024) );
-#endif
- }
-
- ICHOLT( &workspace->H, workspace->droptol, &workspace->L, &workspace->U );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "icholt-" );
- //sprintf( fname, "%s.L%d.out", control->sim_name, data->step );
- //Print_Sparse_Matrix2( workspace->L, fname );
- //Print_Sparse_Matrix( U );
-#endif
- }
-
- /* extrapolation for s & t */
- for( i = 0; i < system->N; ++i ) {
- // no extrapolation
- //s_tmp = workspace->s[0][i];
- //t_tmp = workspace->t[0][i];
-
- // linear
- //s_tmp = 2 * workspace->s[0][i] - workspace->s[1][i];
- //t_tmp = 2 * workspace->t[0][i] - workspace->t[1][i];
-
- // quadratic
- //s_tmp = workspace->s[2][i] + 3 * (workspace->s[0][i]-workspace->s[1][i]);
- t_tmp = workspace->t[index_wkspace_sys(2,i,system)] + 3*(workspace->t[index_wkspace_sys(0,i,system)]-workspace->t[index_wkspace_sys(1,i,system)]);
-
- // cubic
- s_tmp = 4 * (workspace->s[index_wkspace_sys(0,i,system)] + workspace->s[index_wkspace_sys(2,i,system)]) -
- (6 * workspace->s[index_wkspace_sys(1,i,system)] + workspace->s[index_wkspace_sys(3,i,system)] );
- //t_tmp = 4 * (workspace->t[0][i] + workspace->t[2][i]) -
- // (6 * workspace->t[1][i] + workspace->t[3][i] );
-
- // 4th order
- //s_tmp = 5 * (workspace->s[0][i] - workspace->s[3][i]) +
- // 10 * (-workspace->s[1][i] + workspace->s[2][i] ) + workspace->s[4][i];
- //t_tmp = 5 * (workspace->t[0][i] - workspace->t[3][i]) +
- // 10 * (-workspace->t[1][i] + workspace->t[2][i] ) + workspace->t[4][i];
-
- workspace->s[index_wkspace_sys(4,i,system)] = workspace->s[index_wkspace_sys(3,i,system)];
- workspace->s[index_wkspace_sys(3,i,system)] = workspace->s[index_wkspace_sys(2,i,system)];
- workspace->s[index_wkspace_sys(2,i,system)] = workspace->s[index_wkspace_sys(1,i,system)];
- workspace->s[index_wkspace_sys(1,i,system)] = workspace->s[index_wkspace_sys(0,i,system)];
- workspace->s[index_wkspace_sys(0,i,system)] = s_tmp;
-
- workspace->t[index_wkspace_sys(4,i,system)] = workspace->t[index_wkspace_sys(3,i,system)];
- workspace->t[index_wkspace_sys(3,i,system)] = workspace->t[index_wkspace_sys(2,i,system)];
- workspace->t[index_wkspace_sys(2,i,system)] = workspace->t[index_wkspace_sys(1,i,system)];
- workspace->t[index_wkspace_sys(1,i,system)] = workspace->t[index_wkspace_sys(0,i,system)];
- workspace->t[index_wkspace_sys(0,i,system)] = t_tmp;
- }
-}
-
-void Cuda_Init_MatVec( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list *far_nbrs )
-{
- int i, fillin;
- real s_tmp, t_tmp;
- int *spad = (int *)scratch;
- real start = 0, end = 0;
-
- if(control->refactor > 0 &&
- ((data->step-data->prev_steps)%control->refactor==0 || dev_workspace->L.entries==NULL)){
-
- Cuda_Sort_Matrix_Rows <<< BLOCKS, BLOCK_SIZE >>>
- ( dev_workspace->H );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Sorting done... \n");
-#endif
-
- Cuda_Calculate_Droptol <<<BLOCKS, BLOCK_SIZE >>>
- ( dev_workspace->H, dev_workspace->droptol, control->droptol );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Droptol done... \n");
-#endif
-
- if( dev_workspace->L.entries == NULL ) {
-
- cuda_memset ( spad, 0, 2 * INT_SIZE * system->N, RES_SCRATCH );
- Cuda_Estimate_LU_Fill <<< BLOCKS, BLOCK_SIZE >>>
- ( dev_workspace->H, dev_workspace->droptol, spad );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for fill in
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, INT_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, INT_SIZE * BLOCKS_POW_2>>>
- (spad + system->N, spad + system->N + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&fillin, spad + system->N + BLOCKS_POW_2, INT_SIZE, cudaMemcpyDeviceToHost, RES_SCRATCH );
- fillin += dev_workspace->H.n;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Calculated value of the fill in is --> %d \n ", fillin );
-#endif
-
- dev_workspace->L.n = far_nbrs->n;
- dev_workspace->L.m = fillin;
- Cuda_Init_Sparse_Matrix( &dev_workspace->L, fillin, far_nbrs->n );
-
- dev_workspace->U.n = far_nbrs->n;
- dev_workspace->U.m = fillin;
- Cuda_Init_Sparse_Matrix( &dev_workspace->U, fillin, far_nbrs->n );
- }
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "LU matrix done...\n");
-#endif
-
- //TODO -- This is the ILU Factorization of the H Matrix.
- //This is present in the CUDA 5.0 compilation which is not working currently.
- //Fix this when CUDA 5.0 is correctly setup.
- //TODO
- //shared memory is per block
- // here we have only one block -
- /*
- fprintf (stderr, "max sparse matrix entries %d \n", system->max_sparse_matrix_entries );
- Cuda_ICHOLT <<<1, system->max_sparse_matrix_entries,
- system->max_sparse_matrix_entries *(REAL_SIZE + SPARSE_MATRIX_ENTRY_SIZE) >>>
- ( system, dev_workspace->H,
- dev_workspace->droptol,
- dev_workspace->L,
- dev_workspace->U );
- cudaThreadSynchronize ();
- fprintf (stderr, "Cuda_ICHOLT .. done ...-> %d\n ", cudaGetLastError ());
- */
-
- //1. copy the H matrix from device to host
- //2. Allocate the L/U matrices on the host and device.
- //3. Compute the L/U on the host
- //4. copy the results to the device
- //5. Continue the computation.
- sparse_matrix t_H, t_L, t_U;
- real *t_droptol;
-
- t_droptol = (real *) malloc (REAL_SIZE * system->N);
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Allocation temp matrices count %d entries %d \n", dev_workspace->H.n, dev_workspace->H.m );
-#endif
- start = Get_Time ();
- if (!Allocate_Matrix (&t_H, dev_workspace->H.n, dev_workspace->H.m)) { fprintf (stderr, "No space for H matrix \n"); exit (0);}
- if (!Allocate_Matrix (&t_L, far_nbrs->n, dev_workspace->L.m)) { fprintf (stderr, "No space for L matrix \n"); exit (0); }
- if (!Allocate_Matrix (&t_U, far_nbrs->n, dev_workspace->U.m)) { fprintf (stderr, "No space for U matrix \n"); exit (0); }
-
- copy_host_device ( t_H.start, dev_workspace->H.start, INT_SIZE * (dev_workspace->H.n + 1), cudaMemcpyDeviceToHost, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( t_H.end, dev_workspace->H.end, INT_SIZE * (dev_workspace->H.n + 1), cudaMemcpyDeviceToHost, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( t_H.entries, dev_workspace->H.entries, SPARSE_MATRIX_ENTRY_SIZE * dev_workspace->H.m, cudaMemcpyDeviceToHost, RES_SPARSE_MATRIX_ENTRY );
-
- copy_host_device ( t_droptol, dev_workspace->droptol, REAL_SIZE * system->N, cudaMemcpyDeviceToHost, RES_STORAGE_DROPTOL );
-
- //fprintf (stderr, " Done copying LUH .. \n");
- Cuda_ICHOLT (&t_H, t_droptol, &t_L, &t_U);
-
- Sync_Host_Device (&t_L, &t_U, cudaMemcpyHostToDevice);
- end += Get_Timing_Info (start);
-
- /*
- fprintf (stderr, "Done syncing .... \n");
- free (t_droptol);
- fprintf (stderr, "Freed droptol ... \n");
- Deallocate_Matrix (&t_H);
- fprintf (stderr, "Freed H ... \n");
- Deallocate_Matrix (&t_L);
- fprintf (stderr, "Freed l ... \n");
- Deallocate_Matrix (&t_U);
- fprintf (stderr, "Freed u ... \n");
- */
-
- //#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Done copying the L/U matrices to the device ---> %f \n", end);
- //#endif
-
- //#ifdef __BUILD_DEBUG__
- // validate_lu (workspace);
- //#endif
- }
-}
-
-GLOBAL void Init_MatVec_Postprocess (static_storage p_workspace, int N )
-{
-
- static_storage *workspace = &p_workspace;
- real s_tmp, t_tmp;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= N) return;
- // no extrapolation
- //s_tmp = workspace->s[0][i];
- //t_tmp = workspace->t[0][i];
-
- // linear
- //s_tmp = 2 * workspace->s[0][i] - workspace->s[1][i];
- //t_tmp = 2 * workspace->t[0][i] - workspace->t[1][i];
-
- // quadratic
- //s_tmp = workspace->s[2][i] + 3 * (workspace->s[0][i]-workspace->s[1][i]);
- t_tmp = workspace->t[index_wkspace_sys(2,i,N)] + 3*(workspace->t[index_wkspace_sys(0,i,N)]-workspace->t[index_wkspace_sys(1,i,N)]);
-
- // cubic
- s_tmp = 4 * (workspace->s[index_wkspace_sys(0,i,N)] + workspace->s[index_wkspace_sys(2,i,N)]) -
- (6 * workspace->s[index_wkspace_sys(1,i,N)] + workspace->s[index_wkspace_sys(3,i,N)] );
- //t_tmp = 4 * (workspace->t[0][i] + workspace->t[2][i]) -
- // (6 * workspace->t[1][i] + workspace->t[3][i] );
-
- // 4th order
- //s_tmp = 5 * (workspace->s[0][i] - workspace->s[3][i]) +
- // 10 * (-workspace->s[1][i] + workspace->s[2][i] ) + workspace->s[4][i];
- //t_tmp = 5 * (workspace->t[0][i] - workspace->t[3][i]) +
- // 10 * (-workspace->t[1][i] + workspace->t[2][i] ) + workspace->t[4][i];
-
- workspace->s[index_wkspace_sys(4,i,N)] = workspace->s[index_wkspace_sys(3,i,N)];
- workspace->s[index_wkspace_sys(3,i,N)] = workspace->s[index_wkspace_sys(2,i,N)];
- workspace->s[index_wkspace_sys(2,i,N)] = workspace->s[index_wkspace_sys(1,i,N)];
- workspace->s[index_wkspace_sys(1,i,N)] = workspace->s[index_wkspace_sys(0,i,N)];
- workspace->s[index_wkspace_sys(0,i,N)] = s_tmp;
-
- workspace->t[index_wkspace_sys(4,i,N)] = workspace->t[index_wkspace_sys(3,i,N)];
- workspace->t[index_wkspace_sys(3,i,N)] = workspace->t[index_wkspace_sys(2,i,N)];
- workspace->t[index_wkspace_sys(2,i,N)] = workspace->t[index_wkspace_sys(1,i,N)];
- workspace->t[index_wkspace_sys(1,i,N)] = workspace->t[index_wkspace_sys(0,i,N)];
- workspace->t[index_wkspace_sys(0,i,N)] = t_tmp;
-}
-
-void Calculate_Charges( reax_system *system, static_storage *workspace )
-{
- int i;
- real u, s_sum, t_sum;
-
- s_sum = t_sum = 0.;
- for( i = 0; i < system->N; ++i ) {
- s_sum += workspace->s[index_wkspace_sys(0,i,system)];
- t_sum += workspace->t[index_wkspace_sys(0,i,system)];
- }
-
- u = s_sum / t_sum;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Host --->s %13.2f, t %13.f, u %13.2f \n", s_sum, t_sum, u );
-#endif
-
- for( i = 0; i < system->N; ++i )
- system->atoms[i].q = workspace->s[index_wkspace_sys(0,i,system)] - u * workspace->t[index_wkspace_sys(0,i,system)];
-}
-
-GLOBAL void Cuda_Update_Atoms_q ( reax_atom *atoms, real *s, real u, real *t, int N)
-{
- int i = blockIdx.x*blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- atoms[i].q = s[index_wkspace_sys(0,i,N)] - u * t[index_wkspace_sys(0,i,N)];
-}
-
-void Cuda_Calculate_Charges (reax_system *system, static_storage *workspace)
-{
- real *spad = (real *) scratch;
- real u, s_sum, t_sum;
-
- cuda_memset (spad, 0, (BLOCKS_POW_2 * 2 * REAL_SIZE), RES_SCRATCH );
-
- //s_sum
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (&dev_workspace->s [index_wkspace_sys (0, 0,system->N)], spad, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
- (spad, spad+BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&s_sum, spad+BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
- //t_sum
- cuda_memset (spad, 0, (BLOCKS_POW_2 * 2 * REAL_SIZE), RES_SCRATCH );
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (&dev_workspace->t [index_wkspace_sys (0, 0,system->N)], spad, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
- (spad, spad+BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&t_sum, spad+BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
- //fraction here
- u = s_sum / t_sum;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "DEVICE ---> s %13.2f, t %13.f, u %13.2f \n", s_sum, t_sum, u );
-#endif
-
- Cuda_Update_Atoms_q <<< BLOCKS, BLOCK_SIZE >>>
- ( (reax_atom *)system->d_atoms, dev_workspace->s, u, dev_workspace->t, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
-
-
-void QEq( reax_system *system, control_params *control, simulation_data *data,
- static_storage *workspace, list *far_nbrs,
- output_controls *out_control )
-{
- int matvecs;
-
- //real t_start, t_elapsed;
-
- //t_start = Get_Time ();
- Init_MatVec( system, control, data, workspace, far_nbrs );
- //t_elapsed = Get_Timing_Info ( t_start );
-
- //fprintf (stderr, " CPU Init_MatVec timing ----> %f \n", t_elapsed );
-
- //if( data->step % 10 == 0 )
- // Print_Linear_System( system, control, workspace, far_nbrs, data->step );
-
- //t_start = Get_Time ( );
- matvecs = GMRES( workspace, &workspace->H,
- workspace->b_s, control->q_err, &workspace->s[0], out_control->log, system );
- matvecs += GMRES( workspace, &workspace->H,
- workspace->b_t, control->q_err, &workspace->t[0], out_control->log, system );
- //t_elapsed = Get_Timing_Info ( t_start );
-
- //fprintf (stderr, " CPU GMRES timing ---> %f \n", t_elapsed );
-
- //matvecs = GMRES_HouseHolder( workspace, workspace->H,
- // workspace->b_s, control->q_err, workspace->s[0], out_control->log );
- //matvecs += GMRES_HouseHolder( workspace, workspace->H,
- // workspace->b_t, control->q_err, workspace->t[0], out_control->log );
-
- //matvecs = PGMRES( workspace, &workspace->H, workspace->b_s, control->q_err,
- // &workspace->L, &workspace->U, &workspace->s[index_wkspace_sys(0,0,system)], out_control->log, system );
- //matvecs += PGMRES( workspace, &workspace->H, workspace->b_t, control->q_err,
- // &workspace->L, &workspace->U, &workspace->t[index_wkspace_sys(0,0,system)], out_control->log, system );
-
- //matvecs=PCG( workspace, workspace->H, workspace->b_s, control->q_err,
- // workspace->L, workspace->U, workspace->s[0], out_control->log ) + 1;
- ///matvecs+=PCG( workspace, workspace->H, workspace->b_t, control->q_err,
- // workspace->L, workspace->U, workspace->t[0], out_control->log ) + 1;
-
- //matvecs = CG( workspace, workspace->H,
- // workspace->b_s, control->q_err, workspace->s[0], out_control->log ) + 1;
- //matvecs += CG( workspace, workspace->H,
- // workspace->b_t, control->q_err, workspace->t[0], out_control->log ) + 1;
-
- //matvecs = SDM( workspace, workspace->H,
- // workspace->b_s, control->q_err, workspace->s[0], out_control->log ) + 1;
- //matvecs += SDM( workspace, workspace->H,
- // workspace->b_t, control->q_err, workspace->t[0], out_control->log ) + 1;
-
- //fprintf (stderr, " GMRES done with iterations %d \n", matvecs );
-
- data->timing.matvecs += matvecs;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "linsolve-" );
-#endif
-
- Calculate_Charges( system, workspace );
- //fprintf( stderr, "%d %.9f %.9f %.9f %.9f %.9f %.9f\n",
- // data->step,
- // workspace->s[0][0], workspace->t[0][0],
- // workspace->s[0][1], workspace->t[0][1],
- // workspace->s[0][2], workspace->t[0][2] );
- // if( data->step == control->nsteps )
- //Print_Charges( system, control, workspace, data->step );
-}
-
-void Cuda_QEq( reax_system *system, control_params *control, simulation_data *data,
- static_storage *workspace, list *far_nbrs,
- output_controls *out_control )
-{
- int matvecs = 0;
- real t_start, t_elapsed;
-
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time ();
-#endif
-
- /*
- //Cuda_Init_MatVec( system, control, data, workspace, far_nbrs );
-
- Cuda_Sort_Matrix_Rows <<< BLOCKS, BLOCK_SIZE >>>
- ( dev_workspace->H );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- t_elapsed = Get_Timing_Info (t_start);
- fprintf (stderr, "Sorting done...tming --> %f \n", t_elapsed);
- */
- Init_MatVec_Postprocess <<< BLOCKS, BLOCK_SIZE >>>
- (*dev_workspace, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info (t_start);
- fprintf (stderr, "Done with post processing of init_matvec --> %d with time ---> %f \n", cudaGetLastError (), t_elapsed);
-#endif
-
- //Here goes the GMRES part of the program ()
- //#ifdef __DEBUG_CUDA__
- t_start = Get_Time ();
- //#endif
-
- //matvecs = Cuda_GMRES( dev_workspace, dev_workspace->b_s, control->q_err, dev_workspace->s );
- //matvecs += Cuda_GMRES( dev_workspace, dev_workspace->b_t, control->q_err, dev_workspace->t );
-
- matvecs = Cublas_GMRES( system, dev_workspace, dev_workspace->b_s, control->q_err, dev_workspace->s );
- matvecs += Cublas_GMRES( system, dev_workspace, dev_workspace->b_t, control->q_err, dev_workspace->t );
-
- d_timing.matvecs += matvecs;
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info ( t_start );
- fprintf (stderr, " Cuda_GMRES done with iterations %d with timing ---> %f \n", matvecs, t_elapsed );
-#endif
-
- //Here cuda calculate charges
- Cuda_Calculate_Charges (system, workspace);
-}
diff --git a/PuReMD-GPU/src/QEq.h b/PuReMD-GPU/src/QEq.h
index 27eceb9764b3a412728a13f37f6735f086afa546..31dfbf61ba05ec79d32313c3ab648eb259f183f2 100644
--- a/PuReMD-GPU/src/QEq.h
+++ b/PuReMD-GPU/src/QEq.h
@@ -23,10 +23,39 @@
#include "mytypes.h"
+
void QEq( reax_system*, control_params*, simulation_data*, static_storage*,
- list*, output_controls* );
+ list*, output_controls* );
+
+
+static inline HOST_DEVICE void swap(sparse_matrix_entry *array, int index1, int index2)
+{
+ sparse_matrix_entry temp = array[index1];
+ array[index1] = array[index2];
+ array[index2] = temp;
+}
+
+
+static inline HOST_DEVICE void quick_sort(sparse_matrix_entry *array, int start, int end)
+{
+ int i = start;
+ int k = end;
+
+ if (end - start >= 1)
+ {
+ int pivot = array[start].j;
+
+ while (k > i)
+ {
+ while ((array[i].j <= pivot) && (i <= end) && (k > i)) i++;
+ while ((array[k].j > pivot) && (k >= start) && (k >= i)) k--;
+ if (k > i) swap(array, i, k);
+ }
+ swap(array, start, k);
+ quick_sort(array, start, k - 1);
+ quick_sort(array, k + 1, end);
+ }
+}
-void Cuda_QEq( reax_system*, control_params*, simulation_data*, static_storage*,
- list*, output_controls* );
#endif
diff --git a/PuReMD-GPU/src/allocate.c b/PuReMD-GPU/src/allocate.c
new file mode 100644
index 0000000000000000000000000000000000000000..65f0eb2a872673259d508f17fc0da43530a7426f
--- /dev/null
+++ b/PuReMD-GPU/src/allocate.c
@@ -0,0 +1,281 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "allocate.h"
+
+#include "list.h"
+
+
+void Reallocate_Neighbor_List( list *far_nbrs, int n, int num_intrs )
+{
+ Delete_List( far_nbrs );
+ if(!Make_List( n, num_intrs, TYP_FAR_NEIGHBOR, far_nbrs ))
+ {
+ fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
+ exit( INIT_ERR );
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "num_far = %d, far_nbrs = %d -> reallocating!\n",
+ num_intrs, far_nbrs->num_intrs );
+ fprintf( stderr, "memory allocated: far_nbrs = %ldMB\n",
+ num_intrs * sizeof(far_neighbor_data) / (1024*1024) );
+#endif
+}
+
+
+HOST int Allocate_Matrix( sparse_matrix *H, int n, int m )
+{
+ H->n = n;
+ H->m = m;
+ if( (H->start = (int*) malloc(sizeof(int) * n+1)) == NULL )
+ return 0;
+
+ if( (H->end = (int*) malloc(sizeof(int) * n+1)) == NULL )
+ return 0;
+
+ if( (H->entries =
+ (sparse_matrix_entry*) malloc(sizeof(sparse_matrix_entry)*m)) == NULL )
+ return 0;
+
+ return 1;
+}
+
+
+void Deallocate_Matrix( sparse_matrix *H )
+{
+ free(H->start);
+ free(H->entries);
+ free(H->end);
+}
+
+
+int Reallocate_Matrix( sparse_matrix *H, int n, int m, char *name )
+{
+ Deallocate_Matrix( H );
+ if( !Allocate_Matrix( H, n, m ) ) {
+ fprintf(stderr, "not enough space for %s matrix. terminating!\n", name);
+ exit( 1 );
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "reallocating %s matrix, n = %d, m = %d\n",
+ name, n, m );
+ fprintf( stderr, "memory allocated: %s = %ldMB\n",
+ name, m * sizeof(sparse_matrix_entry) / (1024*1024) );
+#endif
+ return 1;
+}
+
+
+int Allocate_HBond_List( int n, int num_h, int *h_index, int *hb_top,
+ list *hbonds )
+{
+ int i, num_hbonds;
+
+ num_hbonds = 0;
+ /* find starting indexes for each H and the total number of hbonds */
+ for( i = 1; i < n; ++i )
+ hb_top[i] += hb_top[i-1];
+ num_hbonds = hb_top[n-1];
+
+ if( !Make_List(num_h, num_hbonds, TYP_HBOND, hbonds ) )
+ {
+ fprintf( stderr, "not enough space for hbonds list. terminating!\n" );
+ exit( INIT_ERR );
+ }
+
+ for( i = 0; i < n; ++i )
+ if( h_index[i] == 0 ){
+ Set_Start_Index( 0, 0, hbonds );
+ Set_End_Index( 0, 0, hbonds );
+ }
+ else if( h_index[i] > 0 ){
+ Set_Start_Index( h_index[i], hb_top[i-1], hbonds );
+ Set_End_Index( h_index[i], hb_top[i-1], hbonds );
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "allocating hbonds - num_hbonds: %d\n", num_hbonds );
+ fprintf( stderr, "memory allocated: hbonds = %ldMB\n",
+ num_hbonds * sizeof(hbond_data) / (1024*1024) );
+#endif
+ return 1;
+}
+
+
+int Reallocate_HBonds_List( int n, int num_h, int *h_index, list *hbonds )
+{
+ int i;
+ int *hb_top;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "reallocating hbonds\n" );
+#endif
+ hb_top = (int *)calloc( n, sizeof(int) );
+ for( i = 0; i < n; ++i )
+ if( h_index[i] >= 0 )
+ hb_top[i] = MAX(Num_Entries(h_index[i],hbonds)*SAFE_HBONDS, MIN_HBONDS);
+
+ Delete_List( hbonds );
+
+ Allocate_HBond_List( n, num_h, h_index, hb_top, hbonds );
+
+ free( hb_top );
+
+ return 1;
+}
+
+
+int Allocate_Bond_List( int n, int *bond_top, list *bonds )
+{
+ int i, num_bonds;
+
+ num_bonds = 0;
+ /* find starting indexes for each atom and the total number of bonds */
+ for( i = 1; i < n; ++i )
+ bond_top[i] += bond_top[i-1];
+ num_bonds = bond_top[n-1];
+
+ if( !Make_List(n, num_bonds, TYP_BOND, bonds ) )
+ {
+ fprintf( stderr, "not enough space for bonds list. terminating!\n" );
+ exit( INIT_ERR );
+ }
+
+ Set_Start_Index( 0, 0, bonds );
+ Set_End_Index( 0, 0, bonds );
+ for( i = 1; i < n; ++i ) {
+ Set_Start_Index( i, bond_top[i-1], bonds );
+ Set_End_Index( i, bond_top[i-1], bonds );
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "allocating bonds - num_bonds: %d\n", num_bonds );
+ fprintf( stderr, "memory allocated: bonds = %ldMB\n",
+ num_bonds * sizeof(bond_data) / (1024*1024) );
+#endif
+ return 1;
+}
+
+
+int Reallocate_Bonds_List( int n, list *bonds, int *num_bonds, int *est_3body )
+{
+ int i;
+ int *bond_top;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "reallocating bonds\n" );
+#endif
+ bond_top = (int *)calloc( n, sizeof(int) );
+ *est_3body = 0;
+ for( i = 0; i < n; ++i ){
+ *est_3body += SQR( Num_Entries( i, bonds ) );
+ bond_top[i] = MAX( Num_Entries( i, bonds ) * 2, MIN_BONDS );
+ }
+
+ Delete_List( bonds );
+
+ Allocate_Bond_List( n, bond_top, bonds );
+ *num_bonds = bond_top[n-1];
+
+ free( bond_top );
+
+ return 1;
+}
+
+
+void Reallocate( reax_system *system, static_storage *workspace, list **lists,
+ int nbr_flag )
+{
+ int num_bonds, est_3body;
+ reallocate_data *realloc;
+ grid *g;
+
+ realloc = &(workspace->realloc);
+ g = &(system->g);
+
+ if( realloc->num_far > 0 && nbr_flag ) {
+ fprintf (stderr, " Reallocating neighbors \n");
+ Reallocate_Neighbor_List( (*lists)+FAR_NBRS,
+ system->N, realloc->num_far * SAFE_ZONE );
+ realloc->num_far = -1;
+ }
+
+ if( realloc->Htop > 0 ){
+ fprintf (stderr, " Reallocating Matrix \n");
+ Reallocate_Matrix(&(workspace->H), system->N, realloc->Htop*SAFE_ZONE,"H");
+ realloc->Htop = -1;
+
+ Deallocate_Matrix( &workspace->L );
+ Deallocate_Matrix( &workspace->U );
+ }
+
+ if( realloc->hbonds > 0 ){
+ fprintf (stderr, " Reallocating hbonds \n");
+ Reallocate_HBonds_List(system->N, workspace->num_H, workspace->hbond_index,
+ (*lists)+HBONDS );
+ realloc->hbonds = -1;
+ }
+
+ num_bonds = est_3body = -1;
+ if( realloc->bonds > 0 ){
+ fprintf (stderr, " Reallocating bonds \n");
+ Reallocate_Bonds_List( system->N, (*lists)+BONDS, &num_bonds, &est_3body );
+ realloc->bonds = -1;
+ realloc->num_3body = MAX( realloc->num_3body, est_3body );
+ }
+
+ if( realloc->num_3body > 0 ) {
+ fprintf (stderr, " Reallocating 3Body \n");
+ Delete_List( (*lists)+THREE_BODIES );
+
+ if( num_bonds == -1 )
+ num_bonds = ((*lists)+BONDS)->num_intrs;
+ realloc->num_3body *= SAFE_ZONE;
+
+ if( !Make_List( num_bonds, realloc->num_3body,
+ TYP_THREE_BODY, (*lists)+THREE_BODIES ) )
+ {
+ fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+ realloc->num_3body = -1;
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "reallocating 3 bodies\n" );
+ fprintf( stderr, "reallocated - num_bonds: %d\n", num_bonds );
+ fprintf( stderr, "reallocated - num_3body: %d\n", realloc->num_3body );
+ fprintf( stderr, "reallocated 3body memory: %ldMB\n",
+ realloc->num_3body*sizeof(three_body_interaction_data)/
+ (1024*1024) );
+#endif
+ }
+
+ if( realloc->gcell_atoms > -1 ){
+#if defined(DEBUG_FOCUS)
+ fprintf(stderr, "reallocating gcell: g->max_atoms: %d\n", g->max_atoms);
+#endif
+
+ free (g->atoms);
+ g->atoms = (int *) calloc ( g->ncell[0]*g->ncell[1]*g->ncell[2],
+ sizeof (int) * workspace->realloc.gcell_atoms);
+ realloc->gcell_atoms = -1;
+ }
+}
diff --git a/PuReMD-GPU/src/allocate.h b/PuReMD-GPU/src/allocate.h
index 7ab146bbb763073377b2ffb16905b6694a85765d..b03ed80b34f153b9929ccaa80bc5c27fbf6ce540 100644
--- a/PuReMD-GPU/src/allocate.h
+++ b/PuReMD-GPU/src/allocate.h
@@ -23,6 +23,11 @@
#include "mytypes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Reallocate( reax_system*, static_storage*, list**, int );
int Allocate_Matrix( sparse_matrix*, int, int );
@@ -32,13 +37,9 @@ int Allocate_HBond_List( int, int, int*, int*, list* );
int Allocate_Bond_List( int, int*, list* );
-//Cuda Functions
-int Cuda_Allocate_Matrix( sparse_matrix*, int, int );
-int Cuda_Allocate_HBond_List( int, int, int*, int*, list* );
-int Cuda_Allocate_Bond_List( int, int*, list* );
-void Cuda_Reallocate( reax_system*, static_storage*, list*, int, int );
+#ifdef __cplusplus
+}
+#endif
-GLOBAL void Init_HBond_Indexes ( int *, int *, list , int );
-GLOBAL void Init_Bond_Indexes ( int *, list , int );
#endif
diff --git a/PuReMD-GPU/src/bond_orders.cu b/PuReMD-GPU/src/bond_orders.c
similarity index 57%
rename from PuReMD-GPU/src/bond_orders.cu
rename to PuReMD-GPU/src/bond_orders.c
index 57f5baacb761a34cd8012f28b6ecb3e7c63d6081..49eaed6532449ef34ee3dac26a71462b2edbdd36 100644
--- a/PuReMD-GPU/src/bond_orders.cu
+++ b/PuReMD-GPU/src/bond_orders.c
@@ -19,23 +19,21 @@
----------------------------------------------------------------------*/
#include "bond_orders.h"
+
+#include "index_utils.h"
#include "list.h"
#include "lookup.h"
#include "print_utils.h"
#include "vector.h"
-#include "index_utils.h"
-#include "cuda_utils.h"
-#include "cuda_helpers.h"
-
-
inline real Cf45( real p1, real p2 )
{
return -EXP(-p2 / 2) /
( SQR( EXP(-p1 / 2) + EXP(p1 / 2) ) * (EXP(-p2 / 2) + EXP(p2 / 2)) );
}
+
#ifdef TEST_FORCES
void Get_dBO( reax_system *system, list **lists,
int i, int pj, real C, rvec *v )
@@ -66,7 +64,8 @@ void Get_dBOpinpi2( reax_system *system, list **lists,
start_pj = Start_Index(pj, dBOs);
end_pj = End_Index(pj, dBOs);
- for( k = start_pj; k < end_pj; ++k ) {
+ for( k = start_pj; k < end_pj; ++k )
+ {
dbo_k = &(dBOs->select.dbo_list[k]);
rvec_Scale( vpi[dbo_k->wrt], Cpi, dbo_k->dBOpi );
rvec_Scale( vpi2[dbo_k->wrt], Cpi2, dbo_k->dBOpi2 );
@@ -179,7 +178,6 @@ void Add_dDelta_to_Forces( reax_system *system, list **lists, int i, real C )
}
-
HOST_DEVICE void Calculate_dBO( int i, int pj, static_storage p_workspace,
list p_bonds, list p_dBOs, int *top )
{
@@ -367,7 +365,6 @@ HOST_DEVICE void Calculate_dBO( int i, int pj, static_storage p_workspace,
#endif
-
void Add_dBond_to_Forces_NPT( int i, int pj, reax_system *system,
simulation_data *data, static_storage *workspace,
list **lists )
@@ -521,134 +518,7 @@ void Add_dBond_to_Forces_NPT( int i, int pj, reax_system *system,
temp[0], temp[1], temp[2] ); */
}
-/////////////////////////////////////////////////////////////
-//Cuda Functions
-/////////////////////////////////////////////////////////////
-
-HOST_DEVICE void Cuda_Add_dBond_to_Forces_NPT( int i, int pj, reax_atom *atoms,
- simulation_data *data, static_storage *workspace,
- list *bonds )
-{
- bond_data *nbr_j, *nbr_k;
- bond_order_data *bo_ij, *bo_ji;
- dbond_coefficients coef;
- rvec temp, ext_press;
- ivec rel_box;
- int pk, k, j;
-
- /* Initializations */
- nbr_j = &(bonds->select.bond_list[pj]);
- j = nbr_j->nbr;
- bo_ij = &(nbr_j->bo_data);
- bo_ji = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
-
- coef.C1dbo = bo_ij->C1dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
- coef.C2dbo = bo_ij->C2dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
- coef.C3dbo = bo_ij->C3dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
-
- coef.C1dbopi = bo_ij->C1dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C2dbopi = bo_ij->C2dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C3dbopi = bo_ij->C3dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C4dbopi = bo_ij->C4dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
-
- coef.C1dbopi2 = bo_ij->C1dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C2dbopi2 = bo_ij->C2dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C3dbopi2 = bo_ij->C3dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C4dbopi2 = bo_ij->C4dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
-
- coef.C1dDelta = bo_ij->C1dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
- coef.C2dDelta = bo_ij->C2dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
- coef.C3dDelta = bo_ij->C3dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
-
- /************************************
- * forces related to atom i *
- * first neighbors of atom i *
- ************************************/
- for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- k = nbr_k->nbr;
-
- rvec_Scale( temp, -coef.C2dbo, nbr_k->bo_data.dBOp ); /*2nd,dBO*/
- rvec_ScaledAdd( temp, -coef.C2dDelta, nbr_k->bo_data.dBOp );/*dDelta*/
- rvec_ScaledAdd( temp, -coef.C3dbopi, nbr_k->bo_data.dBOp ); /*3rd,dBOpi*/
- rvec_ScaledAdd( temp, -coef.C3dbopi2, nbr_k->bo_data.dBOp );/*3rd,dBOpi2*/
-
- /* force */
- rvec_Add( atoms[k].f, temp );
- /* pressure */
- rvec_iMultiply( ext_press, nbr_k->rel_box, temp );
- rvec_Add( data->ext_press, ext_press );
- }
-
- /* then atom i itself */
- rvec_Scale( temp, coef.C1dbo, bo_ij->dBOp ); /*1st, dBO*/
- rvec_ScaledAdd( temp, coef.C2dbo, workspace->dDeltap_self[i] ); /*2nd, dBO*/
-
- rvec_ScaledAdd( temp, coef.C1dDelta, bo_ij->dBOp ); /*1st, dBO*/
- rvec_ScaledAdd( temp, coef.C2dDelta, workspace->dDeltap_self[i] );/*2nd, dBO*/
-
- rvec_ScaledAdd( temp, coef.C1dbopi, bo_ij->dln_BOp_pi ); /*1st,dBOpi*/
- rvec_ScaledAdd( temp, coef.C2dbopi, bo_ij->dBOp ); /*2nd,dBOpi*/
- rvec_ScaledAdd( temp, coef.C3dbopi, workspace->dDeltap_self[i] );/*3rd,dBOpi*/
-
- rvec_ScaledAdd(temp, coef.C1dbopi2, bo_ij->dln_BOp_pi2) ; /*1st,dBO_pi2*/
- rvec_ScaledAdd(temp, coef.C2dbopi2, bo_ij->dBOp); /*2nd,dBO_pi2*/
- rvec_ScaledAdd(temp, coef.C3dbopi2, workspace->dDeltap_self[i]);/*3rd,dBO_pi2*/
-
- /* force */
- rvec_Add( atoms[i].f, temp );
- /* ext pressure due to i dropped, counting force on j only will be enough */
-
-
- /****************************************************************************
- * forces and pressure related to atom j *
- * first neighbors of atom j *
- ***************************************************************************/
- for( pk = Start_Index(j, bonds); pk < End_Index(j, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- k = nbr_k->nbr;
-
- rvec_Scale( temp, -coef.C3dbo, nbr_k->bo_data.dBOp ); /*3rd,dBO*/
- rvec_ScaledAdd( temp, -coef.C3dDelta, nbr_k->bo_data.dBOp );/*dDelta*/
- rvec_ScaledAdd( temp, -coef.C4dbopi, nbr_k->bo_data.dBOp ); /*4th,dBOpi*/
- rvec_ScaledAdd( temp, -coef.C4dbopi2, nbr_k->bo_data.dBOp );/*4th,dBOpi2*/
-
- /* force */
- rvec_Add( atoms[k].f, temp );
- /* pressure */
- if( k != i ) {
- ivec_Sum(rel_box, nbr_k->rel_box, nbr_j->rel_box);//k's rel_box wrt i
- rvec_iMultiply( ext_press, rel_box, temp );
- rvec_Add( data->ext_press, ext_press );
- }
- }
-
- /* then atom j itself */
- rvec_Scale( temp, -coef.C1dbo, bo_ij->dBOp ); /*1st, dBO*/
- rvec_ScaledAdd( temp, coef.C3dbo, workspace->dDeltap_self[j] ); /*2nd, dBO*/
-
- rvec_ScaledAdd( temp, -coef.C1dDelta, bo_ij->dBOp ); /*1st, dBO*/
- rvec_ScaledAdd( temp, coef.C3dDelta, workspace->dDeltap_self[j] );/*2nd, dBO*/
-
- rvec_ScaledAdd( temp, -coef.C1dbopi, bo_ij->dln_BOp_pi ); /*1st,dBOpi*/
- rvec_ScaledAdd( temp, -coef.C2dbopi, bo_ij->dBOp ); /*2nd,dBOpi*/
- rvec_ScaledAdd( temp, coef.C4dbopi, workspace->dDeltap_self[j] );/*3rd,dBOpi*/
-
- rvec_ScaledAdd(temp, -coef.C1dbopi2, bo_ij->dln_BOp_pi2); /*1st,dBOpi2*/
- rvec_ScaledAdd(temp, -coef.C2dbopi2, bo_ij->dBOp); /*2nd,dBOpi2*/
- rvec_ScaledAdd(temp, coef.C4dbopi2, workspace->dDeltap_self[j]);/*3rd,dBOpi2*/
-
- /* force */
- rvec_Add( atoms[j].f, temp );
- /* pressure */
- rvec_iMultiply( ext_press, nbr_j->rel_box, temp );
- rvec_Add( data->ext_press, ext_press );
-}
-
-/////////////////////////////////////////////////////////////
-//Cuda Functions
-/////////////////////////////////////////////////////////////
void Add_dBond_to_Forces( int i, int pj, reax_system *system,
simulation_data *data, static_storage *workspace,
list **lists )
@@ -761,154 +631,6 @@ void Add_dBond_to_Forces( int i, int pj, reax_system *system,
/*3rd, dBOpi2*/
}
-HOST_DEVICE void Cuda_Add_dBond_to_Forces ( int i, int pj, reax_atom *atoms,
- static_storage *workspace, list *bonds )
-{
- bond_data *nbr_j, *nbr_k;
- bond_order_data *bo_ij, *bo_ji;
- dbond_coefficients coef;
- int pk, k, j;
- rvec t_f;
-
- /* Initializations */
- nbr_j = &(bonds->select.bond_list[pj]);
- j = nbr_j->nbr;
-
- if (i < j)
- {
- bo_ij = &(nbr_j->bo_data);
- bo_ji = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
- } else {
- bo_ji = &(nbr_j->bo_data);
- bo_ij = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
- }
-
- coef.C1dbo = bo_ij->C1dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
- coef.C2dbo = bo_ij->C2dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
- coef.C3dbo = bo_ij->C3dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
-
- coef.C1dbopi = bo_ij->C1dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C2dbopi = bo_ij->C2dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C3dbopi = bo_ij->C3dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
- coef.C4dbopi = bo_ij->C4dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
-
- coef.C1dbopi2 = bo_ij->C1dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C2dbopi2 = bo_ij->C2dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C3dbopi2 = bo_ij->C3dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
- coef.C4dbopi2 = bo_ij->C4dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
-
- coef.C1dDelta = bo_ij->C1dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
- coef.C2dDelta = bo_ij->C2dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
- coef.C3dDelta = bo_ij->C3dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
-
- if ( i < j) {
- for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- k = nbr_k->nbr;
- rvec_MakeZero (t_f);
-
- rvec_ScaledAdd( t_f, -coef.C2dbo, nbr_k->bo_data.dBOp );
- /*2nd, dBO*/
- rvec_ScaledAdd( t_f, -coef.C2dDelta, nbr_k->bo_data.dBOp );
- /*dDelta*/
- rvec_ScaledAdd( t_f, -coef.C3dbopi, nbr_k->bo_data.dBOp );
- /*3rd, dBOpi*/
- rvec_ScaledAdd( t_f, -coef.C3dbopi2, nbr_k->bo_data.dBOp );
- /*3rd, dBOpi2*/
-
- //Store in the temp place
- rvec_Add (nbr_k->t_f, t_f);
- }
-
- rvec_ScaledAdd( atoms[i].f, coef.C1dbo, bo_ij->dBOp );
- /*1st, dBO*/
- rvec_ScaledAdd( atoms[i].f, coef.C2dbo, workspace->dDeltap_self[i] );
- /*2nd, dBO*/
-
- rvec_ScaledAdd(atoms[i].f, coef.C1dDelta, bo_ij->dBOp);
- /*1st, dBO*/
- rvec_ScaledAdd(atoms[i].f, coef.C2dDelta, workspace->dDeltap_self[i]);
- /*2nd, dBO*/
-
- rvec_ScaledAdd( atoms[i].f, coef.C1dbopi, bo_ij->dln_BOp_pi );
- /*1st, dBOpi*/
- rvec_ScaledAdd( atoms[i].f, coef.C2dbopi, bo_ij->dBOp );
- /*2nd, dBOpi*/
- rvec_ScaledAdd( atoms[i].f, coef.C3dbopi, workspace->dDeltap_self[i]);
- /*3rd, dBOpi*/
-
- rvec_ScaledAdd( atoms[i].f, coef.C1dbopi2, bo_ij->dln_BOp_pi2 );
- /*1st, dBO_pi2*/
- rvec_ScaledAdd( atoms[i].f, coef.C2dbopi2, bo_ij->dBOp );
- /*2nd, dBO_pi2*/
- rvec_ScaledAdd( atoms[i].f, coef.C3dbopi2, workspace->dDeltap_self[i]);
- /*3rd, dBO_pi2*/
- }
- else
- {
- for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- k = nbr_k->nbr;
- rvec_MakeZero (t_f);
-
- rvec_ScaledAdd( t_f, -coef.C3dbo, nbr_k->bo_data.dBOp );
- /*3rd, dBO*/
- rvec_ScaledAdd( t_f, -coef.C3dDelta, nbr_k->bo_data.dBOp );
- /*dDelta*/
- rvec_ScaledAdd( t_f, -coef.C4dbopi, nbr_k->bo_data.dBOp );
- /*4th, dBOpi*/
- rvec_ScaledAdd( t_f, -coef.C4dbopi2, nbr_k->bo_data.dBOp );
- /*4th, dBOpi2*/
-
- //Store in the temp place
- rvec_Add (nbr_k->t_f, t_f);
- }
-
- rvec_ScaledAdd( atoms[i].f, -coef.C1dbo, bo_ij->dBOp );
- /*1st, dBO*/
- rvec_ScaledAdd( atoms[i].f, coef.C3dbo, workspace->dDeltap_self[i] );
- /*2nd, dBO*/
-
- rvec_ScaledAdd( atoms[i].f, -coef.C1dDelta, bo_ij->dBOp );
- /*1st, dBO*/
- rvec_ScaledAdd(atoms[i].f, coef.C3dDelta, workspace->dDeltap_self[i]);
- /*2nd, dBO*/
-
- rvec_ScaledAdd( atoms[i].f, -coef.C1dbopi, bo_ij->dln_BOp_pi );
- /*1st, dBOpi*/
- rvec_ScaledAdd( atoms[i].f, -coef.C2dbopi, bo_ij->dBOp );
- /*2nd, dBOpi*/
- rvec_ScaledAdd(atoms[i].f, coef.C4dbopi, workspace->dDeltap_self[i]);
- /*3rd, dBOpi*/
-
- rvec_ScaledAdd( atoms[i].f, -coef.C1dbopi2, bo_ij->dln_BOp_pi2 );
- /*1st, dBOpi2*/
- rvec_ScaledAdd( atoms[i].f, -coef.C2dbopi2, bo_ij->dBOp );
- /*2nd, dBOpi2*/
- rvec_ScaledAdd(atoms[i].f, coef.C4dbopi2, workspace->dDeltap_self[i]);
- /*3rd, dBOpi2*/
- }
-}
-
-HOST_DEVICE void Cuda_dbond_to_Forces_postprocess (int i, reax_atom *atoms, list *bonds)
-{
- int pk;
- bond_data *nbr_k, *nbr_k_sym;
-
- /*
- for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- rvec_Add (atoms[i].f, nbr_k->t_f);
- }
- */
-
- for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
- nbr_k = &(bonds->select.bond_list[pk]);
- nbr_k_sym = &( bonds->select.bond_list [nbr_k->sym_index] );
-
- rvec_Add (atoms[i].f, nbr_k_sym->t_f);
- }
-}
/* Locate j on i's list.
This function assumes that j is there for sure!
@@ -1031,7 +753,7 @@ void Calculate_Bond_Orders( reax_system *system, control_params *control,
//fprintf( stderr, "\tj:%d - ubo: %8.3f\n", j+1, bo_ij->BO );
if( i < j ) {
- twbp = &( system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ] );
+ twbp = &( system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ] );
#ifdef TEST_FORCES
Set_Start_Index( pj, top_dbo, dBOs );
/* fprintf( stderr, "%6d%6d%23.15e%23.15e%23.15e\n",
@@ -1348,557 +1070,3 @@ bo_ij->C3dbopi2, bo_ij->C4dbopi2 ); */
Print_Bond_Orders( system, control, data, workspace, lists, out_control );
#endif
}
-
-
-//Cuda Functions
-GLOBAL void Cuda_Calculate_Bond_Orders_Init ( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
- static_storage workspace, int num_atom_types, int N )
-{
- int i, type_i;
- real p_boc1, p_boc2;
- single_body_parameters *sbp_i;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- p_boc1 = g_params.l[0];
- p_boc2 = g_params.l[1];
-
- /* Calculate Deltaprime, Deltaprime_boc values */
- type_i = atoms[i].type;
- sbp_i = &(sbp[type_i]);
- workspace.Deltap[i] = workspace.total_bond_order[i] - sbp_i->valency;
- workspace.Deltap_boc[i] =
- workspace.total_bond_order[i] - sbp_i->valency_val;
- workspace.total_bond_order[i] = 0;
-}
-
-
-/* A very important and crucial assumption here is that each segment
- belonging to a different atom in nbrhoods->nbr_list is sorted in its own.
- This can either be done in the general coordinator function or here */
-
-GLOBAL void Cuda_Calculate_Bond_Orders ( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
- two_body_parameters *tbp, static_storage workspace, list bonds,
- list dDeltas, list dBOs, int num_atom_types, int N )
-{
- int i, j, pj, type_i, type_j;
- int start_i, end_i;
- int num_bonds, sym_index;
- real p_boc1, p_boc2;
- real val_i, Deltap_i, Deltap_boc_i;
- real val_j, Deltap_j, Deltap_boc_j;
- real temp, f1, f2, f3, f4, f5, f4f5, exp_f4, exp_f5;
- real exp_p1i, exp_p2i, exp_p1j, exp_p2j;
- real u1_ij, u1_ji, Cf1A_ij, Cf1B_ij, Cf1_ij, Cf1_ji;
- real Cf45_ij, Cf45_ji, p_lp1;
- real A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji;
- real explp1;
- two_body_parameters *twbp;
- bond_order_data *bo_ij, *bo_ji;
- single_body_parameters *sbp_i, *sbp_j;
-
-
-#if defined(TEST_FORCES)
- int k, pk, start_j, end_j;
- int top_dbo=0, top_dDelta=0;
- dbond_data *pdbo;
- dDelta_data *ptop_dDelta;
-
-#endif
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- num_bonds = 0;
- p_boc1 = g_params.l[0];
- p_boc2 = g_params.l[1];
-
- /* Calculate Deltaprime, Deltaprime_boc values */
- //for( i = 0; i < system->N; ++i ) {
- /*
- if (i < N) {
- type_i = atoms[i].type;
- sbp_i = &(sbp[type_i]);
- workspace.Deltap[i] = workspace.total_bond_order[i] - sbp_i->valency;
- workspace.Deltap_boc[i] =
- workspace.total_bond_order[i] - sbp_i->valency_val;
- workspace.total_bond_order[i] = 0;
-
- }
-
- __syncthreads ();
- */
-
-
- // fprintf( stderr, "done with uncorrected bond orders\n" );
-
-
- /* Corrected Bond Order calculations */
- //for( i = 0; i < system->N; ++i ) {
- type_i = atoms[i].type;
- sbp_i = &(sbp[type_i]);
- val_i = sbp_i->valency;
- Deltap_i = workspace.Deltap[i];
- Deltap_boc_i = workspace.Deltap_boc[i];
- start_i = Start_Index(i, &bonds);
- end_i = End_Index(i, &bonds);
- //fprintf( stderr, "i:%d Dp:%g Dbocp:%g s:%d e:%d\n",
- // i+1, Deltap_i, Deltap_boc_i, start_i, end_i );
-
- for( pj = start_i; pj < end_i; ++pj ) {
- j = bonds.select.bond_list[pj].nbr;
- type_j = atoms[j].type;
- bo_ij = &( bonds.select.bond_list[pj].bo_data );
- //fprintf( stderr, "\tj:%d - ubo: %8.3f\n", j+1, bo_ij->BO );
-
- if( i < j ) {
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ] );
-#ifdef TEST_FORCES
- Set_Start_Index( pj, top_dbo, &dBOs );
- /* fprintf( stderr, "%6d%6d%23.15e%23.15e%23.15e\n",
- workspace->reverse_map[i], workspace->reverse_map[j],
- twbp->ovc, twbp->v13cor, bo_ij->BO ); */
-#endif
- if( twbp->ovc < 0.001 && twbp->v13cor < 0.001 ) {
- /* There is no correction to bond orders nor to derivatives of
- bond order prime! So we leave bond orders unchanged and
- set derivative of bond order coefficients s.t.
- dBO = dBOp & dBOxx = dBOxxp in Add_dBO_to_Forces */
- bo_ij->C1dbo = 1.000000;
- bo_ij->C2dbo = 0.000000;
- bo_ij->C3dbo = 0.000000;
-
- bo_ij->C1dbopi = bo_ij->BO_pi;
- bo_ij->C2dbopi = 0.000000;
- bo_ij->C3dbopi = 0.000000;
- bo_ij->C4dbopi = 0.000000;
-
- bo_ij->C1dbopi2 = bo_ij->BO_pi2;
- bo_ij->C2dbopi2 = 0.000000;
- bo_ij->C3dbopi2 = 0.000000;
- bo_ij->C4dbopi2 = 0.000000;
-
-#ifdef TEST_FORCES
- pdbo = &(dBOs.select.dbo_list[ top_dbo ]);
-
- // compute dBO_ij/dr_i
- pdbo->wrt = i;
- rvec_Copy( pdbo->dBO, bo_ij->dBOp );
- rvec_Scale( pdbo->dBOpi, bo_ij->BO_pi, bo_ij->dln_BOp_pi );
- rvec_Scale( pdbo->dBOpi2, bo_ij->BO_pi2, bo_ij->dln_BOp_pi2 );
-
- // compute dBO_ij/dr_j
- pdbo++;
- pdbo->wrt = j;
- rvec_Scale( pdbo->dBO,-1.0,bo_ij->dBOp );
- rvec_Scale( pdbo->dBOpi,-bo_ij->BO_pi,bo_ij->dln_BOp_pi );
- rvec_Scale( pdbo->dBOpi2,-bo_ij->BO_pi2,bo_ij->dln_BOp_pi2 );
-
- top_dbo += 2;
-#endif
- }
- else {
- val_j = sbp[type_j].valency;
- Deltap_j = workspace.Deltap[j];
- Deltap_boc_j = workspace.Deltap_boc[j];
-
- /* on page 1 */
- if( twbp->ovc >= 0.001 ) {
- /* Correction for overcoordination */
- exp_p1i = EXP( -p_boc1 * Deltap_i );
- exp_p2i = EXP( -p_boc2 * Deltap_i );
- exp_p1j = EXP( -p_boc1 * Deltap_j );
- exp_p2j = EXP( -p_boc2 * Deltap_j );
-
- f2 = exp_p1i + exp_p1j;
- f3 = -1.0 / p_boc2 * log( 0.5 * ( exp_p2i + exp_p2j ) );
- f1 = 0.5 * ( ( val_i + f2 )/( val_i + f2 + f3 ) +
- ( val_j + f2 )/( val_j + f2 + f3 ) );
-
- /*fprintf( stderr,"%6d%6d\t%g %g j:%g %g p_boc:%g %g\n",
- i+1, j+1, val_i, Deltap_i, val_j, Deltap_j, p_boc1, p_boc2 );
- fprintf( stderr,"\tf:%g %g %g, exp:%g %g %g %g\n",
- f1, f2, f3, exp_p1i, exp_p2i, exp_p1j, exp_p2j );*/
-
- /* Now come the derivates */
- /* Bond Order pages 5-7, derivative of f1 */
- temp = f2 + f3;
- u1_ij = val_i + temp;
- u1_ji = val_j + temp;
- Cf1A_ij = 0.5 * f3 * (1.0 / SQR( u1_ij ) + 1.0 / SQR( u1_ji ));
- Cf1B_ij = -0.5 * (( u1_ij - f3 ) / SQR( u1_ij ) +
- ( u1_ji - f3 ) / SQR( u1_ji ));
-
- //Cf1_ij = -Cf1A_ij * p_boc1 * exp_p1i +
- // Cf1B_ij * exp_p2i / ( exp_p2i + exp_p2j );
- Cf1_ij = 0.50 * ( -p_boc1 * exp_p1i / u1_ij -
- ((val_i+f2) / SQR(u1_ij)) *
- ( -p_boc1 * exp_p1i +
- exp_p2i / ( exp_p2i + exp_p2j ) ) +
- -p_boc1 * exp_p1i / u1_ji -
- ((val_j+f2)/SQR(u1_ji)) * ( -p_boc1*exp_p1i +
- exp_p2i / ( exp_p2i + exp_p2j ) ));
-
- Cf1_ji = -Cf1A_ij * p_boc1 * exp_p1j +
- Cf1B_ij * exp_p2j / ( exp_p2i + exp_p2j );
- //fprintf( stderr, "\tCf1:%g %g\n", Cf1_ij, Cf1_ji );
- }
- else {
- /* No overcoordination correction! */
- f1 = 1.0;
- Cf1_ij = Cf1_ji = 0.0;
- }
-
- if( twbp->v13cor >= 0.001 ) {
- /* Correction for 1-3 bond orders */
- exp_f4 =EXP(-(twbp->p_boc4 * SQR( bo_ij->BO ) -
- Deltap_boc_i) * twbp->p_boc3 + twbp->p_boc5);
- exp_f5 =EXP(-(twbp->p_boc4 * SQR( bo_ij->BO ) -
- Deltap_boc_j) * twbp->p_boc3 + twbp->p_boc5);
-
- f4 = 1. / (1. + exp_f4);
- f5 = 1. / (1. + exp_f5);
- f4f5 = f4 * f5;
-
- /* Bond Order pages 8-9, derivative of f4 and f5 */
- /*temp = twbp->p_boc5 -
- twbp->p_boc3 * twbp->p_boc4 * SQR( bo_ij->BO );
- u_ij = temp + twbp->p_boc3 * Deltap_boc_i;
- u_ji = temp + twbp->p_boc3 * Deltap_boc_j;
- Cf45_ij = Cf45( u_ij, u_ji ) / f4f5;
- Cf45_ji = Cf45( u_ji, u_ij ) / f4f5;*/
- Cf45_ij = -f4 * exp_f4;
- Cf45_ji = -f5 * exp_f5;
- }
- else {
- f4 = f5 = f4f5 = 1.0;
- Cf45_ij = Cf45_ji = 0.0;
- }
-
- /* Bond Order page 10, derivative of total bond order */
- A0_ij = f1 * f4f5;
- A1_ij = -2 * twbp->p_boc3 * twbp->p_boc4 * bo_ij->BO *
- (Cf45_ij + Cf45_ji);
- A2_ij = Cf1_ij / f1 + twbp->p_boc3 * Cf45_ij;
- A2_ji = Cf1_ji / f1 + twbp->p_boc3 * Cf45_ji;
- A3_ij = A2_ij + Cf1_ij / f1;
- A3_ji = A2_ji + Cf1_ji / f1;
-
- /*fprintf( stderr, "\tBO: %f, A0: %f, A1: %f, A2_ij: %f
-A2_ji: %f, A3_ij: %f, A3_ji: %f\n",
-bo_ij->BO, A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji );*/
-
- /* find corrected bond order values and their deriv coefs */
- bo_ij->BO = bo_ij->BO * A0_ij;
- bo_ij->BO_pi = bo_ij->BO_pi * A0_ij *f1;
- bo_ij->BO_pi2= bo_ij->BO_pi2* A0_ij *f1;
- bo_ij->BO_s = bo_ij->BO - ( bo_ij->BO_pi + bo_ij->BO_pi2 );
-
- bo_ij->C1dbo = A0_ij + bo_ij->BO * A1_ij;
- bo_ij->C2dbo = bo_ij->BO * A2_ij;
- bo_ij->C3dbo = bo_ij->BO * A2_ji;
-
- bo_ij->C1dbopi = f1*f1*f4*f5;
- bo_ij->C2dbopi = bo_ij->BO_pi * A1_ij;
- bo_ij->C3dbopi = bo_ij->BO_pi * A3_ij;
- bo_ij->C4dbopi = bo_ij->BO_pi * A3_ji;
-
- bo_ij->C1dbopi2 = f1*f1*f4*f5;
- bo_ij->C2dbopi2 = bo_ij->BO_pi2 * A1_ij;
- bo_ij->C3dbopi2 = bo_ij->BO_pi2 * A3_ij;
-
-#ifdef TEST_FORCES
- /*fprintf( stderr, "%6d%6d%13.6f%13.6f%13.6f%13.6f\n",
- i+1, j+1, bo_ij->BO, bo_ij->C1dbo, Cf45_ij, Cf45_ji );*/
-
- /* fprintf( stderr, "%6d%6d%13.6f%13.6f%13.6f%13.6f\n",
- //"%6d%6d%10.6f%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f%10.6f\n\n",
- workspace->orig_id[i], workspace->orig_id[j]
- A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji
- bo_ij->BO, bo_ij->BO_pi, bo_ij->BO_pi2, bo_ij->BO_s,
- bo_ij->C1dbo, bo_ij->C2dbo, bo_ij->C3dbo,
- bo_ij->C1dbopi,bo_ij->C2dbopi,bo_ij->C3dbopi,bo_ij->C4dbopi,
- bo_ij->C1dbopi2,bo_ij->C2dbopi2,bo_ij->C3dbopi2,bo_ij->C4dbopi2
- ); */
-
- Calculate_dBO( i, pj, workspace, lists, &top_dbo );
-#endif
- }
-
- /* neglect bonds that are < 1e-10 */
- if( bo_ij->BO < 1e-10 )
- bo_ij->BO = 0.0;
- if( bo_ij->BO_s < 1e-10 )
- bo_ij->BO_s = 0.0;
- if( bo_ij->BO_pi < 1e-10 )
- bo_ij->BO_pi = 0.0;
- if( bo_ij->BO_pi2 < 1e-10 )
- bo_ij->BO_pi2 = 0.0;
-
- workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
-
-
- /* fprintf( stderr, "%d %d\t%g %g %g %g\n
-Cdbo:\t%g %g %g\n
-Cdbopi:\t%g %g %g %g\n
-Cdbopi2:%g %g %g %g\n\n",
-i+1, j+1, bonds->select.bond_list[ pj ].d,
-bo_ij->BO,bo_ij->BO_pi, bo_ij->BO_pi2,
-bo_ij->C1dbo, bo_ij->C2dbo, bo_ij->C3dbo,
-bo_ij->C1dbopi, bo_ij->C2dbopi, bo_ij->C3dbopi, bo_ij->C4dbopi,
-bo_ij->C1dbopi2, bo_ij->C2dbopi2,
-bo_ij->C3dbopi2, bo_ij->C4dbopi2 ); */
-
- /* fprintf( stderr, "%d %d, BO:%f BO_s:%f BO_pi:%f BO_pi2:%f\n",
- i+1,j+1,bo_ij->BO,bo_ij->BO_s,bo_ij->BO_pi,bo_ij->BO_pi2 ); */
-
-#ifdef TEST_FORCES
- Set_End_Index( pj, top_dbo, &dBOs );
- //Add_dBO( system, lists, i, pj, 1.0, workspace->dDelta );
-#endif
- }
- /*
- else {
- // We only need to update bond orders from bo_ji
- // everything else is set in uncorrected_bo calculations
- sym_index = bonds.select.bond_list[pj].sym_index;
- bo_ji = &(bonds.select.bond_list[ sym_index ].bo_data);
- bo_ij->BO = bo_ji->BO;
- bo_ij->BO_s = bo_ji->BO_s;
- bo_ij->BO_pi = bo_ji->BO_pi;
- bo_ij->BO_pi2 = bo_ji->BO_pi2;
-
- workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
-
-#ifdef TEST_FORCES
- //Add_dBO( system, lists, j, sym_index, 1.0, workspace.dDelta );
-#endif
-}
- */
-}
-
-#ifdef TEST_FORCES
-// fprintf( stderr, "dDelta computations\nj:" );
-Set_Start_Index( i, top_dDelta, &dDeltas );
-ptop_dDelta = &( dDeltas.select.dDelta_list[top_dDelta] );
-
-for( pj = start_i; pj < end_i; ++pj ) {
- j = bonds.select.bond_list[pj].nbr;
- // fprintf( stderr, "%d ", j );
-
- if( !rvec_isZero( workspace.dDelta[j] ) ) {
- ptop_dDelta->wrt = j;
- rvec_Copy( ptop_dDelta->dVal, workspace.dDelta[j] );
- rvec_MakeZero( workspace.dDelta[j] );
- ++top_dDelta, ++ptop_dDelta;
- }
-
- start_j = Start_Index(j, &bonds);
- end_j = End_Index(j, &bonds);
- for( pk = start_j; pk < end_j; ++pk ) {
- k = bonds.select.bond_list[pk].nbr;
- if( !rvec_isZero( workspace.dDelta[k] ) ) {
- ptop_dDelta->wrt = k;
- rvec_Copy( ptop_dDelta->dVal, workspace.dDelta[k] );
- rvec_MakeZero( workspace.dDelta[k] );
- ++top_dDelta, ++ptop_dDelta;
- }
- }
-}
-
-Set_End_Index( i, top_dDelta, &dDeltas );
-
-/*for( pj=Start_Index(i,dDeltas); pj<End_Index(i,dDeltas); ++pj )
- fprintf( stdout, "dDel: %d %d [%g %g %g]\n",
- i+1, dDeltas->select.dDelta_list[pj].wrt+1,
- dDeltas->select.dDelta_list[pj].dVal[0],
- dDeltas->select.dDelta_list[pj].dVal[1],
- dDeltas->select.dDelta_list[pj].dVal[2] );*/
-#endif
-//}
-
-/*fprintf(stderr,"\tCalculated actual bond orders ...\n" );
- fprintf(stderr,"%6s%8s%8s%8s%8s%8s%8s%8s\n",
- "atom", "Delta", "Delta_e", "Delta_boc", "nlp",
- "Delta_lp", "Clp", "dDelta_lp" );*/
-
-/*
- p_lp1 = g_params.l[15];
-
-//get the kernel ID for the following computation
-j = i;
-
-// Calculate some helper variables that are used at many places
-// throughout force calculations
-//for( j = 0; j < system->N; ++j ) {
-type_j = atoms[j].type;
-sbp_j = &(sbp[ type_j ]);
-
-workspace.Delta[j] = workspace.total_bond_order[j] - sbp_j->valency;
-workspace.Delta_e[j] = workspace.total_bond_order[j] - sbp_j->valency_e;
-workspace.Delta_boc[j] = workspace.total_bond_order[j] -
-sbp_j->valency_boc;
-
-workspace.vlpex[j] = workspace.Delta_e[j] -
-2.0 * (int)(workspace.Delta_e[j]/2.0);
-explp1 = EXP(-p_lp1 * SQR(2.0 + workspace.vlpex[j]));
-workspace.nlp[j] = explp1 - (int)(workspace.Delta_e[j] / 2.0);
-workspace.Delta_lp[j] = sbp_j->nlp_opt - workspace.nlp[j];
-workspace.Clp[j] = 2.0 * p_lp1 * explp1 * (2.0 + workspace.vlpex[j]);
-// Adri uses different dDelta_lp values than the ones in notes... //
-workspace.dDelta_lp[j] = workspace.Clp[j];
-//workspace->dDelta_lp[j] = workspace->Clp[j] + (0.5-workspace->Clp[j]) *
-//((fabs(workspace->Delta_e[j]/2.0 -
-// (int)(workspace->Delta_e[j]/2.0)) < 0.1) ? 1 : 0 );
-
-if( sbp_j->mass > 21.0 ) {
-workspace.nlp_temp[j] = 0.5 * (sbp_j->valency_e - sbp_j->valency);
-workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
-workspace.dDelta_lp_temp[j] = 0.;
-}
-else {
-workspace.nlp_temp[j] = workspace.nlp[j];
-workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
-workspace.dDelta_lp_temp[j] = workspace.Clp[j];
-}
-
-//fprintf( stderr, "%d\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\n",
-//j, workspace->Delta[j], workspace->Delta_e[j], workspace->Delta_boc[j],
-//workspace->nlp[j], system->reaxprm.sbp[type_j].nlp_opt,
-//workspace->Delta_lp[j], workspace->Clp[j], workspace->dDelta_lp[j] );
-//}
- */
-
-//Print_Bonds( system, bonds, "sbonds.out" );
-
-#if defined(DEBUG)
-//fprintf( stderr, "Number of bonds: %d\n", num_bonds );
-//Print_Bond_Orders( system, control, data, workspace, lists, out_control );
-#endif
-}
-
-GLOBAL void Cuda_Update_Uncorrected_BO ( static_storage workspace, list bonds, int N )
-{
- int i, j, pj;
- int start_i, end_i;
- int sym_index;
-
- bond_order_data *bo_ij, *bo_ji;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- start_i = Start_Index(i, &bonds);
- end_i = End_Index(i, &bonds);
-
- for( pj = start_i; pj < end_i; ++pj ) {
-
- j = bonds.select.bond_list[pj].nbr;
- bo_ij = &( bonds.select.bond_list[pj].bo_data );
-
- if( i >= j ) {
- // We only need to update bond orders from bo_ji
- // everything else is set in uncorrected_bo calculations
- sym_index = bonds.select.bond_list[pj].sym_index;
- bo_ji = &(bonds.select.bond_list[ sym_index ].bo_data);
- bo_ij->BO = bo_ji->BO;
- bo_ij->BO_s = bo_ji->BO_s;
- bo_ij->BO_pi = bo_ji->BO_pi;
- bo_ij->BO_pi2 = bo_ji->BO_pi2;
-
- workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
- }
- }
-}
-
-GLOBAL void Cuda_Update_Workspace_After_Bond_Orders( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
- static_storage workspace, int N )
-{
- int j, type_j;
- real explp1;
- real p_lp1;
- single_body_parameters *sbp_i, *sbp_j;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
- if (j >= N) return;
-
- p_lp1 = g_params.l[15];
-
- /* Calculate some helper variables that are used at many places
- throughout force calculations */
- //for( j = 0; j < system->N; ++j ) {
- type_j = atoms[j].type;
- sbp_j = &(sbp[ type_j ]);
-
- workspace.Delta[j] = workspace.total_bond_order[j] - sbp_j->valency;
- workspace.Delta_e[j] = workspace.total_bond_order[j] - sbp_j->valency_e;
- workspace.Delta_boc[j] = workspace.total_bond_order[j] -
- sbp_j->valency_boc;
-
- workspace.vlpex[j] = workspace.Delta_e[j] -
- 2.0 * (int)(workspace.Delta_e[j]/2.0);
- explp1 = EXP(-p_lp1 * SQR(2.0 + workspace.vlpex[j]));
- workspace.nlp[j] = explp1 - (int)(workspace.Delta_e[j] / 2.0);
- workspace.Delta_lp[j] = sbp_j->nlp_opt - workspace.nlp[j];
- workspace.Clp[j] = 2.0 * p_lp1 * explp1 * (2.0 + workspace.vlpex[j]);
- /* Adri uses different dDelta_lp values than the ones in notes... */
- workspace.dDelta_lp[j] = workspace.Clp[j];
- //workspace->dDelta_lp[j] = workspace->Clp[j] + (0.5-workspace->Clp[j]) *
- //((fabs(workspace->Delta_e[j]/2.0 -
- // (int)(workspace->Delta_e[j]/2.0)) < 0.1) ? 1 : 0 );
-
- if( sbp_j->mass > 21.0 ) {
- workspace.nlp_temp[j] = 0.5 * (sbp_j->valency_e - sbp_j->valency);
- workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
- workspace.dDelta_lp_temp[j] = 0.;
- }
- else {
- workspace.nlp_temp[j] = workspace.nlp[j];
- workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
- workspace.dDelta_lp_temp[j] = workspace.Clp[j];
- }
-
- //fprintf( stderr, "%d\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\n",
- //j, workspace->Delta[j], workspace->Delta_e[j], workspace->Delta_boc[j],
- //workspace->nlp[j], system->reaxprm.sbp[type_j].nlp_opt,
- //workspace->Delta_lp[j], workspace->Clp[j], workspace->dDelta_lp[j] );
- //}
-
-}
-
-//Import from the forces file.
-
-GLOBAL void Cuda_Compute_Total_Force (reax_atom *atoms, simulation_data *data,
- static_storage workspace, list p_bonds, int ensemble, int N)
-{
- int i, pj;
- list *bonds = &p_bonds;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i < N)
- {
- for (pj = Start_Index (i, bonds); pj < End_Index (i, bonds); ++pj)
- {
- //int j = bonds->select.bond_list[pj].nbr;
- if (ensemble == NVE || ensemble == NVT || ensemble == bNVT)
- Cuda_Add_dBond_to_Forces (i, pj, atoms, &workspace, bonds );
- else
- Cuda_Add_dBond_to_Forces_NPT (i, pj, atoms, data, &workspace, bonds );
- }
- }
-}
-
-GLOBAL void Cuda_Compute_Total_Force_PostProcess (reax_atom *atoms, simulation_data *data,
- static_storage workspace, list p_bonds, int ensemble, int N)
-{
- int i, pj;
- list *bonds = &p_bonds;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i < N)
- {
- if (ensemble == NVE || ensemble == NVT || ensemble == bNVT)
- Cuda_dbond_to_Forces_postprocess (i, atoms, bonds );
- }
-}
diff --git a/PuReMD-GPU/src/bond_orders.h b/PuReMD-GPU/src/bond_orders.h
index 19fea911f129715cec305f0a4f1f0ea433d412d4..476bacfb0704c97d4b3b44b15e76883be4cad9b9 100644
--- a/PuReMD-GPU/src/bond_orders.h
+++ b/PuReMD-GPU/src/bond_orders.h
@@ -21,8 +21,10 @@
#ifndef __BOND_ORDERS_H_
#define __BOND_ORDERS_H_
+
#include "mytypes.h"
+
typedef struct
{
real C1dbo, C2dbo, C3dbo;
@@ -31,6 +33,7 @@ typedef struct
real C1dDelta, C2dDelta, C3dDelta;
} dbond_coefficients;
+
#ifdef TEST_FORCES
void Get_dBO( reax_system*, list**, int, int, real, rvec* );
void Get_dBOpinpi2( reax_system*, list**, int, int, real, real, rvec*, rvec* );
@@ -52,16 +55,4 @@ void Add_dBond_to_Forces_NPT( int, int, reax_system*, simulation_data*,
void Calculate_Bond_Orders( reax_system*, control_params*, simulation_data*,
static_storage*, list**, output_controls* );
-//CUDA Functions
-GLOBAL void Cuda_Calculate_Bond_Orders_Init ( reax_atom *, global_parameters , single_body_parameters *,
- static_storage , int , int );
-GLOBAL void Cuda_Calculate_Bond_Orders ( reax_atom *, global_parameters , single_body_parameters *,
- two_body_parameters *, static_storage , list , list , list , int , int );
-GLOBAL void Cuda_Update_Uncorrected_BO ( static_storage , list , int );
-GLOBAL void Cuda_Update_Workspace_After_Bond_Orders( reax_atom *, global_parameters , single_body_parameters *,
- static_storage , int );
-GLOBAL void Cuda_Compute_Total_Force (reax_atom *, simulation_data *, static_storage , list , int , int );
-GLOBAL void Cuda_Compute_Total_Force_PostProcess (reax_atom *, simulation_data *, static_storage , list , int , int );
-//HOST_DEVICE void Cuda_Add_dBond_to_Forces( int, int, reax_atom *, static_storage*, list* );
-//HOST_DEVICE void Cuda_Add_dBond_to_Forces_NPT( int, int, reax_atom *, simulation_data*, static_storage*, list* );
#endif
diff --git a/PuReMD-GPU/src/box.cu b/PuReMD-GPU/src/box.c
similarity index 100%
rename from PuReMD-GPU/src/box.cu
rename to PuReMD-GPU/src/box.c
diff --git a/PuReMD-GPU/src/box.h b/PuReMD-GPU/src/box.h
index ed8cc9f4e1900a71f5a6b319f98e7d425400f928..418aa6208a81fb05ee56ff09afc6ff76751f75c9 100644
--- a/PuReMD-GPU/src/box.h
+++ b/PuReMD-GPU/src/box.h
@@ -21,11 +21,13 @@
#ifndef __BOX_H__
#define __BOX_H__
+
#include "mytypes.h"
+
/* Initializes box from CRYST1 line of PDB */
void Init_Box_From_CRYST(real, real, real, real, real, real,
- simulation_box*/*, int*/);
+ simulation_box*/*, int*/);
/* Initializes box from box rtensor */
void Update_Box(rtensor, simulation_box* /*, int*/);
@@ -43,13 +45,11 @@ void Transform( rvec, simulation_box*, char, rvec );
void Transform_to_UnitBox( rvec, simulation_box*, char, rvec );
void Get_NonPeriodic_Far_Neighbors( rvec, rvec, simulation_box*,
- control_params*, far_neighbor_data*, int* );
+ control_params*, far_neighbor_data*, int* );
void Get_Periodic_Far_Neighbors_Big_Box( rvec, rvec, simulation_box*,
- control_params*, far_neighbor_data*,
- int* );
+ control_params*, far_neighbor_data*, int* );
void Get_Periodic_Far_Neighbors_Small_Box( rvec, rvec, simulation_box*,
- control_params*, far_neighbor_data*,
- int* );
+ control_params*, far_neighbor_data*, int* );
void Distance_on_T3_Gen( rvec, rvec, simulation_box*, rvec );
void Inc_on_T3_Gen( rvec, rvec, simulation_box* );
@@ -61,7 +61,9 @@ void Inc_Nbr_Box_Press( simulation_box*, int, int, int, rvec );*/
/* this function returns cartesian norm but triclinic distance vector */
real Metric_Product( rvec, rvec, simulation_box* );
-HOST_DEVICE inline real Sq_Distance_on_T3( rvec x1, rvec x2, simulation_box* box, rvec r)
+void Print_Box_Information( simulation_box*, FILE* );
+
+static inline HOST_DEVICE real Sq_Distance_on_T3( rvec x1, rvec x2, simulation_box* box, rvec r)
{
real norm = 0.0;
@@ -94,12 +96,8 @@ HOST_DEVICE inline real Sq_Distance_on_T3( rvec x1, rvec x2, simulation_box* box
}
-void Print_Box_Information( simulation_box*, FILE* );
-
-//CUDA Device Functions
-//HOST_DEVICE inline void Inc_on_T3( rvec, rvec, simulation_box* );
-HOST_DEVICE inline void Inc_on_T3( rvec x, rvec dx, simulation_box *box )
+static inline HOST_DEVICE void Inc_on_T3( rvec x, rvec dx, simulation_box *box )
{
int i;
real tmp;
@@ -115,4 +113,5 @@ HOST_DEVICE inline void Inc_on_T3( rvec x, rvec dx, simulation_box *box )
}
}
+
#endif
diff --git a/PuReMD-GPU/src/center_mass.h b/PuReMD-GPU/src/center_mass.h
deleted file mode 100644
index 4048511e2e4b311f9de74cef94c2a661d51b4b39..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/center_mass.h
+++ /dev/null
@@ -1,48 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#ifndef __CENTER_MASS_H__
-#define __CENTER_MASS_H__
-
-#include "mytypes.h"
-
-GLOBAL void center_of_mass_blocks (single_body_parameters *, reax_atom *,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n);
-
-GLOBAL void center_of_mass (rvec *xcm,
- rvec *vcm,
- rvec *amcm,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n);
-
-GLOBAL void compute_center_mass (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n);
-
-GLOBAL void compute_center_mass (real *input, real *output, size_t n);
-
-#endif
diff --git a/PuReMD-GPU/src/cuda_QEq.cu b/PuReMD-GPU/src/cuda_QEq.cu
new file mode 100644
index 0000000000000000000000000000000000000000..033945338aa76aa4c02909f90d2ca3eb1dacad58
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_QEq.cu
@@ -0,0 +1,724 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_QEq.h"
+
+#include "QEq.h"
+#include "allocate.h"
+#include "lin_alg.h"
+#include "list.h"
+#include "print_utils.h"
+#include "index_utils.h"
+#include "system_props.h"
+
+#include "cuda_copy.h"
+#include "cuda_init.h"
+#include "cuda_utils.h"
+#include "cuda_lin_alg.h"
+#include "cuda_reduction.h"
+
+#include "sort.h"
+#include "validation.h"
+
+
+GLOBAL void Cuda_Sort_Matrix_Rows( sparse_matrix A )
+{
+ int i;
+ int si, ei;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( i >= A.n ) return;
+
+ si = A.start[i];
+ ei = A.end [i];
+
+ quick_sort( A.entries + si, 0, ei-si-1 );
+}
+
+
+GLOBAL void Cuda_Calculate_Droptol( sparse_matrix p_A, real *droptol, real dtol )
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ int k, j, offset, x, diagonal;
+ real val;
+ sparse_matrix *A = &p_A;
+
+ if ( i < A->n ) {
+ droptol [i] = 0;
+
+ for (k = A->start[i]; k < A->end[i]; ++k ) {
+ val = A->entries[k].val;
+ droptol [i] += val*val;
+ }
+ }
+
+ __syncthreads ();
+ if ( i < A->n ) {
+ droptol [i] = SQRT (droptol[i]) * dtol;
+ }
+
+}
+
+
+GLOBAL void Cuda_Calculate_Droptol_js( sparse_matrix p_A, real *droptol, real dtol )
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ int k, j, offset, x, diagonal;
+ real val;
+ sparse_matrix *A = &p_A;
+
+ for (x = 0; x < A->n; x ++)
+ {
+ if (i < (A->end[i]-1 - A->start[i])) {
+ offset = A->start [i] + i;
+ j = A->entries[offset].j;
+ val = A->entries[offset].val;
+ droptol [j] += val * val;
+ }
+ __syncthreads ();
+ }
+}
+
+
+GLOBAL void Cuda_Calculate_Droptol_diagonal( sparse_matrix p_A, real *droptol, real dtol )
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ int k, j, offset, x, diagonal;
+ real val;
+ sparse_matrix *A = &p_A;
+
+ if ( i < A->n ) {
+ //diagonal element
+ diagonal = A->end[i]-1;
+ val = A->entries [diagonal].val;
+ droptol [i] += val*val;
+ }
+
+ /*calculate local droptol for each row*/
+ if ( i < A->n )
+ droptol [i] = SQRT (droptol[i]) * dtol;
+}
+
+
+GLOBAL void Cuda_Estimate_LU_Fill( sparse_matrix p_A, real *droptol, int *fillin )
+{
+ int i, j, pj;
+ real val;
+ sparse_matrix *A = &p_A;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= A->n) return;
+
+ fillin [i] = 0;
+
+ for (pj = A->start[i]; pj < A->end[i]-1; ++pj)
+ {
+ j = A->entries [pj].j;
+ val = A->entries[pj].val;
+
+ if (fabs (val) > droptol [i]) ++fillin [i];
+ }
+}
+
+
+void Cuda_ICHOLT( sparse_matrix *A, real *droptol,
+ sparse_matrix *L, sparse_matrix *U )
+{
+ sparse_matrix_entry tmp[1000];
+ int i, j, pj, k1, k2, tmptop, Ltop;
+ real val;
+ int *Utop;
+
+ Utop = (int*) malloc((A->n+1) * sizeof(int));
+
+ // clear variables
+ Ltop = 0;
+ tmptop = 0;
+ for( i = 0; i <= A->n; ++i )
+ L->start[i] = U->start[i] = 0;
+
+ for( i = 0; i < A->n; ++i )
+ Utop[i] = 0;
+
+ //fprintf( stderr, "n: %d\n", A->n );
+ for( i = 0; i < A->n; ++i ){
+ L->start[i] = Ltop;
+ tmptop = 0;
+
+ for( pj = A->start[i]; pj < A->end[i]-1; ++pj ){
+ j = A->entries[pj].j;
+ val = A->entries[pj].val;
+ //fprintf( stderr, "i: %d, j: %d", i, j );
+
+ //CHANGE ORIGINAL
+ if (j >= i) break;
+ //CHANGE ORIGINAL
+
+ if( fabs(val) > droptol[i] ){
+ k1 = 0;
+ k2 = L->start[j];
+ while( k1 < tmptop && k2 < L->start[j+1] ){
+ if( tmp[k1].j < L->entries[k2].j )
+ ++k1;
+ else if( tmp[k1].j > L->entries[k2].j )
+ ++k2;
+ else
+ val -= (tmp[k1++].val * L->entries[k2++].val);
+ }
+
+ // L matrix is lower triangular,
+ // so right before the start of next row comes jth diagonal
+ val /= L->entries[L->start[j+1]-1].val;
+
+ tmp[tmptop].j = j;
+ tmp[tmptop].val = val;
+ ++tmptop;
+ }
+
+ //fprintf( stderr, " -- done\n" );
+ }
+
+ // compute the ith diagonal in L
+ // sanity check
+ if( A->entries[pj].j != i ) {
+ fprintf( stderr, "i=%d, badly built A matrix!\n", i );
+ exit(999);
+ }
+
+ val = A->entries[pj].val;
+ for( k1 = 0; k1 < tmptop; ++k1 )
+ val -= (tmp[k1].val * tmp[k1].val);
+
+ tmp[tmptop].j = i;
+ tmp[tmptop].val = SQRT(val);
+
+ // apply the dropping rule once again
+ //fprintf( stderr, "row%d: tmptop: %d\n", i, tmptop );
+ //for( k1 = 0; k1<= tmptop; ++k1 )
+ // fprintf( stderr, "%d(%f) ", tmp[k1].j, tmp[k1].val );
+ //fprintf( stderr, "\n" );
+ //fprintf( stderr, "row(%d): droptol=%.4f\n", i+1, droptol[i] );
+ for( k1 = 0; k1 < tmptop; ++k1 )
+ if( fabs(tmp[k1].val) > droptol[i] / tmp[tmptop].val ){
+ L->entries[Ltop].j = tmp[k1].j;
+ L->entries[Ltop].val = tmp[k1].val;
+ U->start[tmp[k1].j+1]++;
+ ++Ltop;
+ //fprintf( stderr, "%d(%.4f) ", tmp[k1].j+1, tmp[k1].val );
+ }
+ // keep the diagonal in any case
+ L->entries[Ltop].j = tmp[k1].j;
+ L->entries[Ltop].val = tmp[k1].val;
+ ++Ltop;
+ //fprintf( stderr, "%d(%.4f)\n", tmp[k1].j+1, tmp[k1].val );
+ }
+
+ L->start[i] = Ltop;
+ //fprintf( stderr, "nnz(L): %d, max: %d\n", Ltop, L->n * 50 );
+
+ for( i = 1; i <= U->n; ++i )
+ Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
+
+ for( i = 0; i < L->n; ++i )
+ for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
+ j = L->entries[pj].j;
+ U->entries[Utop[j]].j = i;
+ U->entries[Utop[j]].val = L->entries[pj].val;
+ Utop[j]++;
+ }
+
+ //fprintf( stderr, "nnz(U): %d, max: %d\n", Utop[U->n], U->n * 50 );
+}
+
+
+/*
+//Parallel for each row
+//Each kernel will run for 6540 number of times.
+GLOBAL void Cuda_ICHOLT( reax_system *system, sparse_matrix p_A, real *droptol,
+sparse_matrix p_L, sparse_matrix p_U )
+{
+int start, end, count;
+real tempvalue, val;
+int i,pj,tmptop, offset;
+int j, k1, k2;
+
+sparse_matrix *A, *L, *U;
+sparse_matrix_entry *tmp;
+
+A = &p_A;
+L = &p_L;
+U = &p_U;
+
+real *null_val;
+null_val = 0;
+
+extern __shared__ real tmp_val[];
+extern __shared__ sparse_matrix_entry sh_tmp[];
+
+int kid = blockIdx.x * blockDim.x + threadIdx.x;
+tmp = (sparse_matrix_entry *) (tmp_val + blockDim.x);
+
+offset = 0;
+for( i = 0; i < 10; ++i )
+{
+//if (kid == 0) L->start[i] = i * system->max_sparse_matrix_entries;
+if (kid == 0) L->start[i] = offset;
+tmptop = 0;
+
+start = A->start[i];
+end = A->end[i]-1; //inclusive
+count = end - start; //inclusive
+tmp_val [kid] = 0;
+
+if (kid < count) //diagonal not included
+{
+pj = start + kid;
+
+j = A->entries[pj].j;
+val = A->entries[pj].val;
+
+if( fabs(val) > droptol[i] )
+{
+k1 = 0;
+k2 = L->start[j];
+while( k1 < tmptop && k2 < L->end[j] ){
+if( tmp[k1].j < L->entries[k2].j )
+++k1;
+else if( tmp[k1].j > L->entries[k2].j )
+++k2;
+else
+tmp_val[kid] = (tmp[k1++].val * L->entries[k2++].val);
+}
+
+//here read the shared memory of all the kernels
+if (kid == 0)
+{
+for (i = 0; i < count; i++)
+tempvalue += tmp_val [i];
+
+val -= tempvalue;
+
+// L matrix is lower triangular,
+// so right before the start of next row comes jth diagonal
+val /= L->entries[L->end[j]-1].val;
+
+tmp[tmptop].j = j;
+tmp[tmptop].val = val;
+++tmptop;
+}
+}
+}
+__syncthreads ();
+
+
+// compute the ith diagonal in L
+// sanity check
+if (kid == 0)
+{
+ if( A->entries[end].j != i ) {
+ //intentional core dump here for sanity sake
+ *null_val = 1;
+ }
+}
+
+//diagonal element
+//val = A->entries[pj].val;
+//for( k1 = 0; k1 < tmptop; ++k1 )
+if (kid < count)
+ tmp_val[kid] = (tmp[kid].val * tmp[kid].val);
+
+ __syncthreads ();
+
+if (kid == 0)
+{
+ val = A->entries [end].val;
+ for (i = 0; i < count; i++)
+ tempvalue += tmp_val [i];
+
+ val -= tempvalue;
+ tmp[tmptop].j = i;
+ tmp[tmptop].val = SQRT(val);
+}
+__syncthreads ();
+
+//Fill in the LU entries
+//for( k1 = 0; k1 < count; ++k1 )
+if (kid < count )
+{
+ if( fabs(tmp[kid].val) > droptol[i] / tmp[tmptop].val ){
+ L->entries[offset + kid].j = tmp[kid].j;
+ L->entries[offset + kid].val = tmp[kid].val;
+ U->start[tmp[kid].j+1]++;
+ }
+}
+__syncthreads ();
+
+if (kid == 0) {
+ // keep the diagonal in any case
+ offset += count;
+ L->entries[offset].j = tmp[count].j;
+ L->entries[offset].val = tmp[count].val;
+ ++offset;
+ L->end [i] = offset;
+}
+__syncthreads ();
+} // end of main for loop
+}
+
+void Cuda_Fill_U ( sparse_matrix *A, real *droptol,
+ sparse_matrix *L, sparse_matrix *U )
+{
+ int i, pj, j;
+
+ for( i = 1; i <= U->n; ++i )
+ Utop[i] = U->start[i] = U->start[i] + U->start[i-1] + 1;
+
+ for( i = 0; i < L->n; ++i )
+ for( pj = L->start[i]; pj < L->start[i+1]; ++pj ){
+ j = L->entries[pj].j;
+ U->entries[Utop[j]].j = i;
+ U->entries[Utop[j]].val = L->entries[pj].val;
+ Utop[j]++;
+ }
+}
+*/
+
+
+void Cuda_Init_MatVec( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace, list *far_nbrs )
+{
+ int i, fillin;
+ real s_tmp, t_tmp;
+ int *spad = (int *)scratch;
+ real start = 0, end = 0;
+
+ if( control->refactor > 0 &&
+ ((data->step-data->prev_steps)%control->refactor==0 ||
+ dev_workspace->L.entries==NULL) )
+ {
+ Cuda_Sort_Matrix_Rows<<< BLOCKS, BLOCK_SIZE >>>
+ ( dev_workspace->H );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Sorting done... \n");
+#endif
+
+ Cuda_Calculate_Droptol<<<BLOCKS, BLOCK_SIZE >>>
+ ( dev_workspace->H, dev_workspace->droptol, control->droptol );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Droptol done... \n");
+#endif
+
+ if( dev_workspace->L.entries == NULL )
+ {
+ cuda_memset( spad, 0, 2 * INT_SIZE * system->N, RES_SCRATCH );
+ Cuda_Estimate_LU_Fill <<< BLOCKS, BLOCK_SIZE >>>
+ ( dev_workspace->H, dev_workspace->droptol, spad );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //Reduction for fill in
+ Cuda_reduction_int<<<BLOCKS_POW_2, BLOCK_SIZE, INT_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_reduction_int<<<1, BLOCKS_POW_2, INT_SIZE * BLOCKS_POW_2>>>
+ (spad + system->N, spad + system->N + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( &fillin, spad + system->N + BLOCKS_POW_2, INT_SIZE, cudaMemcpyDeviceToHost, RES_SCRATCH );
+ fillin += dev_workspace->H.n;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Calculated value of the fill in is --> %d \n ", fillin );
+#endif
+
+ dev_workspace->L.n = far_nbrs->n;
+ dev_workspace->L.m = fillin;
+ Cuda_Init_Sparse_Matrix( &dev_workspace->L, fillin, far_nbrs->n );
+
+ dev_workspace->U.n = far_nbrs->n;
+ dev_workspace->U.m = fillin;
+ Cuda_Init_Sparse_Matrix( &dev_workspace->U, fillin, far_nbrs->n );
+ }
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "LU matrix done...\n");
+#endif
+
+ //TODO -- This is the ILU Factorization of the H Matrix.
+ //This is present in the CUDA 5.0 compilation which is not working currently.
+ //Fix this when CUDA 5.0 is correctly setup.
+ //TODO
+ //shared memory is per block
+ // here we have only one block -
+ /*
+ fprintf (stderr, "max sparse matrix entries %d \n", system->max_sparse_matrix_entries );
+ Cuda_ICHOLT <<<1, system->max_sparse_matrix_entries,
+ system->max_sparse_matrix_entries *(REAL_SIZE + SPARSE_MATRIX_ENTRY_SIZE) >>>
+ ( system, dev_workspace->H,
+ dev_workspace->droptol,
+ dev_workspace->L,
+ dev_workspace->U );
+ cudaThreadSynchronize ();
+ fprintf (stderr, "Cuda_ICHOLT .. done ...-> %d\n ", cudaGetLastError ());
+ */
+
+ //1. copy the H matrix from device to host
+ //2. Allocate the L/U matrices on the host and device.
+ //3. Compute the L/U on the host
+ //4. copy the results to the device
+ //5. Continue the computation.
+ sparse_matrix t_H, t_L, t_U;
+ real *t_droptol;
+
+ t_droptol = (real *) malloc( REAL_SIZE * system->N );
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, " Allocation temp matrices count %d entries %d \n", dev_workspace->H.n, dev_workspace->H.m );
+#endif
+
+ start = Get_Time( );
+ if( !Allocate_Matrix(&t_H, dev_workspace->H.n, dev_workspace->H.m) )
+ {
+ fprintf(stderr, "No space for H matrix \n");
+ exit( 0 );
+ }
+ if( !Allocate_Matrix(&t_L, far_nbrs->n, dev_workspace->L.m) )
+ {
+ fprintf( stderr, "No space for L matrix \n" );
+ exit( 0 );
+ }
+ if( !Allocate_Matrix(&t_U, far_nbrs->n, dev_workspace->U.m) )
+ {
+ fprintf( stderr, "No space for U matrix \n" );
+ exit( 0 );
+ }
+
+ copy_host_device( t_H.start, dev_workspace->H.start, INT_SIZE *
+ (dev_workspace->H.n + 1), cudaMemcpyDeviceToHost,
+ RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( t_H.end, dev_workspace->H.end, INT_SIZE *
+ (dev_workspace->H.n + 1), cudaMemcpyDeviceToHost,
+ RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( t_H.entries, dev_workspace->H.entries,
+ SPARSE_MATRIX_ENTRY_SIZE * dev_workspace->H.m,
+ cudaMemcpyDeviceToHost, RES_SPARSE_MATRIX_ENTRY );
+
+ copy_host_device( t_droptol, dev_workspace->droptol, REAL_SIZE *
+ system->N, cudaMemcpyDeviceToHost, RES_STORAGE_DROPTOL );
+
+ //fprintf (stderr, " Done copying LUH .. \n");
+ Cuda_ICHOLT( &t_H, t_droptol, &t_L, &t_U );
+
+ Sync_Host_Device_Mat( &t_L, &t_U, cudaMemcpyHostToDevice );
+ end += Get_Timing_Info( start );
+
+ /*
+ fprintf (stderr, "Done syncing .... \n");
+ free (t_droptol);
+ fprintf (stderr, "Freed droptol ... \n");
+ Deallocate_Matrix (&t_H);
+ fprintf (stderr, "Freed H ... \n");
+ Deallocate_Matrix (&t_L);
+ fprintf (stderr, "Freed l ... \n");
+ Deallocate_Matrix (&t_U);
+ fprintf (stderr, "Freed u ... \n");
+ */
+
+ //#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Done copying the L/U matrices to the device ---> %f \n", end );
+ //#endif
+
+ //#ifdef __BUILD_DEBUG__
+ // validate_lu (workspace);
+ //#endif
+ }
+}
+
+
+GLOBAL void Init_MatVec_Postprocess( static_storage p_workspace, int N )
+{
+
+ static_storage *workspace = &p_workspace;
+ real s_tmp, t_tmp;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= N) return;
+ // no extrapolation
+ //s_tmp = workspace->s[0][i];
+ //t_tmp = workspace->t[0][i];
+
+ // linear
+ //s_tmp = 2 * workspace->s[0][i] - workspace->s[1][i];
+ //t_tmp = 2 * workspace->t[0][i] - workspace->t[1][i];
+
+ // quadratic
+ //s_tmp = workspace->s[2][i] + 3 * (workspace->s[0][i]-workspace->s[1][i]);
+ t_tmp = workspace->t[index_wkspace_sys(2,i,N)] + 3*(workspace->t[index_wkspace_sys(0,i,N)]-workspace->t[index_wkspace_sys(1,i,N)]);
+
+ // cubic
+ s_tmp = 4 * (workspace->s[index_wkspace_sys(0,i,N)] + workspace->s[index_wkspace_sys(2,i,N)]) -
+ (6 * workspace->s[index_wkspace_sys(1,i,N)] + workspace->s[index_wkspace_sys(3,i,N)] );
+ //t_tmp = 4 * (workspace->t[0][i] + workspace->t[2][i]) -
+ // (6 * workspace->t[1][i] + workspace->t[3][i] );
+
+ // 4th order
+ //s_tmp = 5 * (workspace->s[0][i] - workspace->s[3][i]) +
+ // 10 * (-workspace->s[1][i] + workspace->s[2][i] ) + workspace->s[4][i];
+ //t_tmp = 5 * (workspace->t[0][i] - workspace->t[3][i]) +
+ // 10 * (-workspace->t[1][i] + workspace->t[2][i] ) + workspace->t[4][i];
+
+ workspace->s[index_wkspace_sys(4,i,N)] = workspace->s[index_wkspace_sys(3,i,N)];
+ workspace->s[index_wkspace_sys(3,i,N)] = workspace->s[index_wkspace_sys(2,i,N)];
+ workspace->s[index_wkspace_sys(2,i,N)] = workspace->s[index_wkspace_sys(1,i,N)];
+ workspace->s[index_wkspace_sys(1,i,N)] = workspace->s[index_wkspace_sys(0,i,N)];
+ workspace->s[index_wkspace_sys(0,i,N)] = s_tmp;
+
+ workspace->t[index_wkspace_sys(4,i,N)] = workspace->t[index_wkspace_sys(3,i,N)];
+ workspace->t[index_wkspace_sys(3,i,N)] = workspace->t[index_wkspace_sys(2,i,N)];
+ workspace->t[index_wkspace_sys(2,i,N)] = workspace->t[index_wkspace_sys(1,i,N)];
+ workspace->t[index_wkspace_sys(1,i,N)] = workspace->t[index_wkspace_sys(0,i,N)];
+ workspace->t[index_wkspace_sys(0,i,N)] = t_tmp;
+}
+
+
+GLOBAL void Cuda_Update_Atoms_q( reax_atom *atoms, real *s, real u, real *t, int N )
+{
+ int i = blockIdx.x*blockDim.x + threadIdx.x;
+
+ if (i >= N)
+ {
+ return;
+ }
+
+ atoms[i].q = s[index_wkspace_sys(0,i,N)] - u * t[index_wkspace_sys(0,i,N)];
+}
+
+
+void Cuda_Calculate_Charges( reax_system *system, static_storage *workspace )
+{
+ real *spad = (real *) scratch;
+ real u, s_sum, t_sum;
+
+ cuda_memset( spad, 0, (BLOCKS_POW_2 * 2 * REAL_SIZE), RES_SCRATCH );
+
+ //s_sum
+ Cuda_reduction<<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (&dev_workspace->s [index_wkspace_sys (0, 0,system->N)], spad, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_reduction<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, spad+BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( &s_sum, spad+BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ //t_sum
+ cuda_memset( spad, 0, (BLOCKS_POW_2 * 2 * REAL_SIZE), RES_SCRATCH );
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (&dev_workspace->t [index_wkspace_sys (0, 0,system->N)], spad, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_reduction<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, spad+BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( &t_sum, spad+BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ //fraction here
+ u = s_sum / t_sum;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "DEVICE ---> s %13.2f, t %13.f, u %13.2f \n", s_sum, t_sum, u );
+#endif
+
+ Cuda_Update_Atoms_q<<< BLOCKS, BLOCK_SIZE >>>
+ ( (reax_atom *)system->d_atoms, dev_workspace->s, u, dev_workspace->t, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+void Cuda_QEq( reax_system *system, control_params *control, simulation_data *data,
+ static_storage *workspace, list *far_nbrs,
+ output_controls *out_control )
+{
+ int matvecs = 0;
+ real t_start, t_elapsed;
+
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+
+ /*
+ //Cuda_Init_MatVec( system, control, data, workspace, far_nbrs );
+
+ Cuda_Sort_Matrix_Rows<<< BLOCKS, BLOCK_SIZE >>>
+ ( dev_workspace->H );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ t_elapsed = Get_Timing_Info (t_start);
+ fprintf (stderr, "Sorting done...tming --> %f \n", t_elapsed);
+ */
+
+ Init_MatVec_Postprocess<<< BLOCKS, BLOCK_SIZE >>>
+ (*dev_workspace, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf( stderr, "Done with post processing of init_matvec --> %d with time ---> %f \n", cudaGetLastError (), t_elapsed );
+#endif
+
+ //Here goes the GMRES part of the program ()
+ //#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+ //#endif
+
+ //matvecs = Cuda_GMRES( dev_workspace, dev_workspace->b_s, control->q_err, dev_workspace->s );
+ //matvecs += Cuda_GMRES( dev_workspace, dev_workspace->b_t, control->q_err, dev_workspace->t );
+
+ matvecs = Cublas_GMRES( system, dev_workspace, dev_workspace->b_s, control->q_err, dev_workspace->s );
+ matvecs += Cublas_GMRES( system, dev_workspace, dev_workspace->b_t, control->q_err, dev_workspace->t );
+
+ d_timing.matvecs += matvecs;
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf( stderr, " Cuda_GMRES done with iterations %d with timing ---> %f \n", matvecs, t_elapsed );
+#endif
+
+ Cuda_Calculate_Charges( system, workspace );
+}
diff --git a/PuReMD-GPU/src/cuda_QEq.h b/PuReMD-GPU/src/cuda_QEq.h
new file mode 100644
index 0000000000000000000000000000000000000000..f62ab1157e3812837cd3a8a65e07856a78b22bf2
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_QEq.h
@@ -0,0 +1,39 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_QEq_H_
+#define __CUDA_QEq_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_QEq( reax_system*, control_params*, simulation_data*, static_storage*,
+ list*, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/allocate.cu b/PuReMD-GPU/src/cuda_allocate.cu
similarity index 64%
rename from PuReMD-GPU/src/allocate.cu
rename to PuReMD-GPU/src/cuda_allocate.cu
index 37b80693217700dc20bbffdb515cfe5e1f6eb986..d0e1f22b6f970038820f82c504dc3472d50d1d7a 100644
--- a/PuReMD-GPU/src/allocate.cu
+++ b/PuReMD-GPU/src/cuda_allocate.cu
@@ -18,32 +18,24 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "allocate.h"
-#include "list.h"
+#include "cuda_allocate.h"
#include "cuda_utils.h"
-#include "reduction.h"
+#include "cuda_list.h"
+#include "cuda_reduction.h"
-void Reallocate_Neighbor_List( list *far_nbrs, int n, int num_intrs )
-{
- Delete_List( far_nbrs );
- if(!Make_List( n, num_intrs, TYP_FAR_NEIGHBOR, far_nbrs )){
- fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
- exit( INIT_ERR );
- }
+#include "list.h"
+
+
+GLOBAL void Init_HBond_Indexes ( int *, int *, list , int );
+GLOBAL void Init_Bond_Indexes ( int *, list , int );
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "num_far = %d, far_nbrs = %d -> reallocating!\n",
- num_intrs, far_nbrs->num_intrs );
- fprintf( stderr, "memory allocated: far_nbrs = %ldMB\n",
- num_intrs * sizeof(far_neighbor_data) / (1024*1024) );
-#endif
-}
void Cuda_Reallocate_Neighbor_List( list *far_nbrs, int n, int num_intrs )
{
- Delete_List( far_nbrs, TYP_DEVICE );
- if(!Make_List( n, num_intrs, TYP_FAR_NEIGHBOR, far_nbrs, TYP_DEVICE )){
+ Cuda_Delete_List( far_nbrs );
+ if(!Cuda_Make_List( n, num_intrs, TYP_FAR_NEIGHBOR, far_nbrs ))
+ {
fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
exit( INIT_ERR );
}
@@ -57,23 +49,6 @@ void Cuda_Reallocate_Neighbor_List( list *far_nbrs, int n, int num_intrs )
}
-int Allocate_Matrix( sparse_matrix *H, int n, int m )
-{
- H->n = n;
- H->m = m;
- if( (H->start = (int*) malloc(sizeof(int) * n+1)) == NULL )
- return 0;
-
- if( (H->end = (int*) malloc(sizeof(int) * n+1)) == NULL )
- return 0;
-
- if( (H->entries =
- (sparse_matrix_entry*) malloc(sizeof(sparse_matrix_entry)*m)) == NULL )
- return 0;
-
- return 1;
-}
-
int Cuda_Allocate_Matrix( sparse_matrix *H, int n, int m )
{
H->n = n;
@@ -87,13 +62,6 @@ int Cuda_Allocate_Matrix( sparse_matrix *H, int n, int m )
}
-void Deallocate_Matrix( sparse_matrix *H )
-{
- free(H->start);
- free(H->entries);
- free(H->end);
-}
-
void Cuda_Deallocate_Matrix( sparse_matrix *H )
{
cuda_free(H->start, RES_SPARSE_MATRIX_INDEX);
@@ -106,23 +74,6 @@ void Cuda_Deallocate_Matrix( sparse_matrix *H )
}
-int Reallocate_Matrix( sparse_matrix *H, int n, int m, char *name )
-{
- Deallocate_Matrix( H );
- if( !Allocate_Matrix( H, n, m ) ) {
- fprintf(stderr, "not enough space for %s matrix. terminating!\n", name);
- exit( 1 );
- }
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "reallocating %s matrix, n = %d, m = %d\n",
- name, n, m );
- fprintf( stderr, "memory allocated: %s = %ldMB\n",
- name, m * sizeof(sparse_matrix_entry) / (1024*1024) );
-#endif
- return 1;
-}
-
int Cuda_Reallocate_Matrix( sparse_matrix *H, int n, int m, char *name )
{
Cuda_Deallocate_Matrix( H );
@@ -142,56 +93,6 @@ int Cuda_Reallocate_Matrix( sparse_matrix *H, int n, int m, char *name )
}
-int Allocate_HBond_List( int n, int num_h, int *h_index, int *hb_top,
- list *hbonds )
-{
- int i, num_hbonds;
-
- num_hbonds = 0;
- /* find starting indexes for each H and the total number of hbonds */
- for( i = 1; i < n; ++i )
- hb_top[i] += hb_top[i-1];
- num_hbonds = hb_top[n-1];
-
- if( !Make_List(num_h, num_hbonds, TYP_HBOND, hbonds ) ) {
- fprintf( stderr, "not enough space for hbonds list. terminating!\n" );
- exit( INIT_ERR );
- }
-
- for( i = 0; i < n; ++i )
- if( h_index[i] == 0 ){
- Set_Start_Index( 0, 0, hbonds );
- Set_End_Index( 0, 0, hbonds );
- }
- else if( h_index[i] > 0 ){
- Set_Start_Index( h_index[i], hb_top[i-1], hbonds );
- Set_End_Index( h_index[i], hb_top[i-1], hbonds );
- }
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "allocating hbonds - num_hbonds: %d\n", num_hbonds );
- fprintf( stderr, "memory allocated: hbonds = %ldMB\n",
- num_hbonds * sizeof(hbond_data) / (1024*1024) );
-#endif
- return 1;
-}
-
-GLOBAL void Init_HBond_Indexes ( int *h_index, int *hb_top, list hbonds, int N )
-{
- int index = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (index >= N) return;
-
- if( h_index[index] == 0 ){
- Set_Start_Index( 0, 0, &hbonds );
- Set_End_Index( 0, 0, &hbonds );
- }
- else if( h_index[index] > 0 ){
- Set_Start_Index( h_index[index], hb_top[index-1], &hbonds );
- Set_End_Index( h_index[index], hb_top[index-1], &hbonds );
- }
-}
-
int Cuda_Allocate_HBond_List( int n, int num_h, int *h_index, int *hb_top, list *hbonds )
{
int i, num_hbonds;
@@ -204,7 +105,7 @@ int Cuda_Allocate_HBond_List( int n, int num_h, int *h_index, int *hb_top, list
hb_top[i] += hb_top[i-1];
num_hbonds = hb_top[n-1];
- if( !Make_List(num_h, num_hbonds, TYP_HBOND, hbonds , TYP_DEVICE) ) {
+ if( !Cuda_Make_List(num_h, num_hbonds, TYP_HBOND, hbonds ) ) {
fprintf( stderr, "not enough space for hbonds list. terminating!\n" );
exit( INIT_ERR );
}
@@ -225,27 +126,6 @@ int Cuda_Allocate_HBond_List( int n, int num_h, int *h_index, int *hb_top, list
return 1;
}
-int Reallocate_HBonds_List( int n, int num_h, int *h_index, list *hbonds )
-{
- int i;
- int *hb_top;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "reallocating hbonds\n" );
-#endif
- hb_top = (int *)calloc( n, sizeof(int) );
- for( i = 0; i < n; ++i )
- if( h_index[i] >= 0 )
- hb_top[i] = MAX(Num_Entries(h_index[i],hbonds)*SAFE_HBONDS, MIN_HBONDS);
-
- Delete_List( hbonds );
-
- Allocate_HBond_List( n, num_h, h_index, hb_top, hbonds );
-
- free( hb_top );
-
- return 1;
-}
int Cuda_Reallocate_HBonds_List( int n, int num_h, int *h_index, list *hbonds )
{
@@ -270,7 +150,7 @@ int Cuda_Reallocate_HBonds_List( int n, int num_h, int *h_index, list *hbonds )
//if( h_index[i] >= 0 )
hb_top[i] = MAX((hb_end [i] - hb_start[i])*SAFE_HBONDS, MIN_HBONDS);
- Delete_List( hbonds, TYP_DEVICE );
+ Cuda_Delete_List( hbonds );
Cuda_Allocate_HBond_List( n, num_h, h_index, hb_top, hbonds );
@@ -281,21 +161,6 @@ int Cuda_Reallocate_HBonds_List( int n, int num_h, int *h_index, list *hbonds )
return 1;
}
-GLOBAL void Init_Bond_Indexes ( int *b_top, list bonds, int N )
-{
- int index = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (index >= N) return;
-
- if( index == 0 ){
- Set_Start_Index( 0, 0, &bonds );
- Set_End_Index( 0, 0, &bonds );
- }
- else if( index > 0 ){
- Set_Start_Index( index, b_top[index-1], &bonds );
- Set_End_Index( index, b_top[index-1], &bonds );
- }
-}
int Cuda_Allocate_Bond_List( int num_b, int *b_top, list *bonds )
{
@@ -308,7 +173,7 @@ int Cuda_Allocate_Bond_List( int num_b, int *b_top, list *bonds )
b_top[i] += b_top[i-1];
num_bonds = b_top[num_b-1];
- if( !Make_List(num_b, num_bonds, TYP_BOND, bonds, TYP_DEVICE) ) {
+ if( !Cuda_Make_List(num_b, num_bonds, TYP_BOND, bonds ) ) {
fprintf( stderr, "not enough space for bonds list. terminating!\n" );
exit( INIT_ERR );
}
@@ -326,93 +191,6 @@ int Cuda_Allocate_Bond_List( int num_b, int *b_top, list *bonds )
}
-int Allocate_Bond_List( int n, int *bond_top, list *bonds )
-{
- int i, num_bonds;
-
- num_bonds = 0;
- /* find starting indexes for each atom and the total number of bonds */
- for( i = 1; i < n; ++i )
- bond_top[i] += bond_top[i-1];
- num_bonds = bond_top[n-1];
-
- if( !Make_List(n, num_bonds, TYP_BOND, bonds ) ) {
- fprintf( stderr, "not enough space for bonds list. terminating!\n" );
- exit( INIT_ERR );
- }
-
- Set_Start_Index( 0, 0, bonds );
- Set_End_Index( 0, 0, bonds );
- for( i = 1; i < n; ++i ) {
- Set_Start_Index( i, bond_top[i-1], bonds );
- Set_End_Index( i, bond_top[i-1], bonds );
- }
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "allocating bonds - num_bonds: %d\n", num_bonds );
- fprintf( stderr, "memory allocated: bonds = %ldMB\n",
- num_bonds * sizeof(bond_data) / (1024*1024) );
-#endif
- return 1;
-}
-
-
-int Reallocate_Bonds_List( int n, list *bonds, int *num_bonds, int *est_3body )
-{
- int i;
- int *bond_top;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "reallocating bonds\n" );
-#endif
- bond_top = (int *)calloc( n, sizeof(int) );
- *est_3body = 0;
- for( i = 0; i < n; ++i ){
- *est_3body += SQR( Num_Entries( i, bonds ) );
- bond_top[i] = MAX( Num_Entries( i, bonds ) * 2, MIN_BONDS );
- }
-
- Delete_List( bonds );
-
- Allocate_Bond_List( n, bond_top, bonds );
- *num_bonds = bond_top[n-1];
-
- free( bond_top );
-
- return 1;
-}
-
-void GLOBAL Calculate_Bond_Indexes (int *bond_top, list bonds, int *per_block_results, int n)
-{
- extern __shared__ int sh_input[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real x = 0;
-
- if(i < n)
- {
- x = SQR (Num_Entries( i, &bonds ) );
- bond_top[i] = MAX( Num_Entries( i, &bonds ) * 2, MIN_BONDS );
- }
- sh_input[threadIdx.x] = x;
- __syncthreads();
-
- for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
- {
- if(threadIdx.x < offset)
- {
- sh_input[threadIdx.x] += sh_input[threadIdx.x + offset];
- }
-
- __syncthreads();
- }
-
- if(threadIdx.x == 0)
- {
- per_block_results[blockIdx.x] = sh_input[0];
- }
-}
-
-
int Cuda_Reallocate_Bonds_List( int n, list *bonds, int *num_3body )
{
int i;
@@ -437,7 +215,7 @@ int Cuda_Reallocate_Bonds_List( int n, list *bonds, int *num_3body )
b_top[i] = MAX((b_end [i] - b_start[i])*2, MIN_BONDS);
}
- Delete_List( bonds, TYP_DEVICE );
+ Cuda_Delete_List( bonds );
Cuda_Allocate_Bond_List(n, b_top, bonds );
@@ -450,6 +228,7 @@ int Cuda_Reallocate_Bonds_List( int n, list *bonds, int *num_3body )
return i;
}
+
int Cuda_Reallocate_ThreeBody_List ( list *thblist, int count )
{
int i;
@@ -479,10 +258,10 @@ int Cuda_Reallocate_ThreeBody_List ( list *thblist, int count )
new_total = thb_total;
new_count = count;
- Delete_List( thblist, TYP_DEVICE );
+ Cuda_Delete_List( thblist );
/*Allocate the list */
- if(!Make_List( new_count, new_total, TYP_THREE_BODY, thblist, TYP_DEVICE )){
+ if(!Cuda_Make_List( new_count, new_total, TYP_THREE_BODY, thblist )){
fprintf(stderr, "Problem in reallocating three-body list. Terminating!\n");
exit( INIT_ERR );
}
@@ -523,14 +302,14 @@ cuda_memset (d_bond_top, 0, (n+BLOCKS_POW_2+1) * INT_SIZE, RES_SCRATCH );
cudaThreadSynchronize ();
cudaCheckError ();
- Cuda_reduction <<<1, BLOCKS_POW_2, INT_SIZE * BLOCKS_POW_2>>>
+ Cuda_reduction_int<<<1, BLOCKS_POW_2, INT_SIZE * BLOCKS_POW_2>>>
(d_bond_top + n, d_bond_top + n + BLOCKS_POW_2, BLOCKS_POW_2);
cudaThreadSynchronize ();
copy_host_device (bond_top, d_bond_top, n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
copy_host_device (est_3body, d_bond_top + n + BLOCKS_POW_2, INT_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- Delete_List( bonds, TYP_DEVICE );
+ Cuda_Delete_List( bonds );
Cuda_Allocate_Bond_List( n, bond_top, bonds );
*num_bonds = bond_top[n-1];
@@ -542,83 +321,6 @@ cuda_memset (d_bond_top, 0, (n+BLOCKS_POW_2+1) * INT_SIZE, RES_SCRATCH );
*/
-void Reallocate( reax_system *system, static_storage *workspace, list **lists,
- int nbr_flag )
-{
- int num_bonds, est_3body;
- reallocate_data *realloc;
- grid *g;
-
- realloc = &(workspace->realloc);
- g = &(system->g);
-
- if( realloc->num_far > 0 && nbr_flag ) {
- fprintf (stderr, " Reallocating neighbors \n");
- Reallocate_Neighbor_List( (*lists)+FAR_NBRS,
- system->N, realloc->num_far * SAFE_ZONE );
- realloc->num_far = -1;
- }
-
- if( realloc->Htop > 0 ){
- fprintf (stderr, " Reallocating Matrix \n");
- Reallocate_Matrix(&(workspace->H), system->N, realloc->Htop*SAFE_ZONE,"H");
- realloc->Htop = -1;
-
- Deallocate_Matrix( &workspace->L );
- Deallocate_Matrix( &workspace->U );
- }
-
- if( realloc->hbonds > 0 ){
- fprintf (stderr, " Reallocating hbonds \n");
- Reallocate_HBonds_List(system->N, workspace->num_H, workspace->hbond_index,
- (*lists)+HBONDS );
- realloc->hbonds = -1;
- }
-
- num_bonds = est_3body = -1;
- if( realloc->bonds > 0 ){
- fprintf (stderr, " Reallocating bonds \n");
- Reallocate_Bonds_List( system->N, (*lists)+BONDS, &num_bonds, &est_3body );
- realloc->bonds = -1;
- realloc->num_3body = MAX( realloc->num_3body, est_3body );
- }
-
- if( realloc->num_3body > 0 ) {
- fprintf (stderr, " Reallocating 3Body \n");
- Delete_List( (*lists)+THREE_BODIES );
-
- if( num_bonds == -1 )
- num_bonds = ((*lists)+BONDS)->num_intrs;
- realloc->num_3body *= SAFE_ZONE;
-
- if( !Make_List( num_bonds, realloc->num_3body,
- TYP_THREE_BODY, (*lists)+THREE_BODIES ) ) {
- fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
- exit( INIT_ERR );
- }
- realloc->num_3body = -1;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "reallocating 3 bodies\n" );
- fprintf( stderr, "reallocated - num_bonds: %d\n", num_bonds );
- fprintf( stderr, "reallocated - num_3body: %d\n", realloc->num_3body );
- fprintf( stderr, "reallocated 3body memory: %ldMB\n",
- realloc->num_3body*sizeof(three_body_interaction_data)/
- (1024*1024) );
-#endif
- }
-
- if( realloc->gcell_atoms > -1 ){
-#if defined(DEBUG_FOCUS)
- fprintf(stderr, "reallocating gcell: g->max_atoms: %d\n", g->max_atoms);
-#endif
-
- free (g->atoms);
- g->atoms = (int *) calloc ( g->ncell[0]*g->ncell[1]*g->ncell[2],
- sizeof (int) * workspace->realloc.gcell_atoms);
- realloc->gcell_atoms = -1;
- }
-}
-
void Cuda_Reallocate( reax_system *system, static_storage *workspace, list *lists,
int nbr_flag, int step )
{
@@ -724,3 +426,68 @@ void Cuda_Reallocate( reax_system *system, static_storage *workspace, list *list
realloc->gcell_atoms = -1;
}
}
+
+
+GLOBAL void Init_HBond_Indexes ( int *h_index, int *hb_top, list hbonds, int N )
+{
+ int index = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (index >= N) return;
+
+ if( h_index[index] == 0 ){
+ Set_Start_Index( 0, 0, &hbonds );
+ Set_End_Index( 0, 0, &hbonds );
+ }
+ else if( h_index[index] > 0 ){
+ Set_Start_Index( h_index[index], hb_top[index-1], &hbonds );
+ Set_End_Index( h_index[index], hb_top[index-1], &hbonds );
+ }
+}
+
+
+GLOBAL void Init_Bond_Indexes ( int *b_top, list bonds, int N )
+{
+ int index = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (index >= N) return;
+
+ if( index == 0 ){
+ Set_Start_Index( 0, 0, &bonds );
+ Set_End_Index( 0, 0, &bonds );
+ }
+ else if( index > 0 ){
+ Set_Start_Index( index, b_top[index-1], &bonds );
+ Set_End_Index( index, b_top[index-1], &bonds );
+ }
+}
+
+
+void GLOBAL Calculate_Bond_Indexes (int *bond_top, list bonds, int *per_block_results, int n)
+{
+ extern __shared__ int sh_input[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real x = 0;
+
+ if(i < n)
+ {
+ x = SQR (Num_Entries( i, &bonds ) );
+ bond_top[i] = MAX( Num_Entries( i, &bonds ) * 2, MIN_BONDS );
+ }
+ sh_input[threadIdx.x] = x;
+ __syncthreads();
+
+ for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
+ {
+ if(threadIdx.x < offset)
+ {
+ sh_input[threadIdx.x] += sh_input[threadIdx.x + offset];
+ }
+
+ __syncthreads();
+ }
+
+ if(threadIdx.x == 0)
+ {
+ per_block_results[blockIdx.x] = sh_input[0];
+ }
+}
diff --git a/PuReMD-GPU/src/cuda_allocate.h b/PuReMD-GPU/src/cuda_allocate.h
new file mode 100644
index 0000000000000000000000000000000000000000..dc672d3bd76c1afa7a468aa2fdda3dd0ca3d3ec9
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_allocate.h
@@ -0,0 +1,41 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_ALLOCATE_H_
+#define __CUDA_ALLOCATE_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+int Cuda_Allocate_Matrix( sparse_matrix*, int, int );
+int Cuda_Allocate_HBond_List( int, int, int*, int*, list* );
+int Cuda_Allocate_Bond_List( int, int*, list* );
+void Cuda_Reallocate( reax_system*, static_storage*, list*, int, int );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_bond_orders.cu b/PuReMD-GPU/src/cuda_bond_orders.cu
new file mode 100644
index 0000000000000000000000000000000000000000..81a7462033a3da94d8ae601288efb83a93e387f2
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_bond_orders.cu
@@ -0,0 +1,857 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_bond_orders.h"
+
+#include "bond_orders.h"
+#include "list.h"
+#include "lookup.h"
+#include "print_utils.h"
+#include "vector.h"
+#include "index_utils.h"
+
+#include "cuda_utils.h"
+#include "cuda_helpers.h"
+
+
+HOST_DEVICE void Cuda_Add_dBond_to_Forces_NPT( int i, int pj, reax_atom *atoms,
+ simulation_data *data, static_storage *workspace,
+ list *bonds )
+{
+ bond_data *nbr_j, *nbr_k;
+ bond_order_data *bo_ij, *bo_ji;
+ dbond_coefficients coef;
+ rvec temp, ext_press;
+ ivec rel_box;
+ int pk, k, j;
+
+ /* Initializations */
+ nbr_j = &(bonds->select.bond_list[pj]);
+ j = nbr_j->nbr;
+ bo_ij = &(nbr_j->bo_data);
+ bo_ji = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
+
+ coef.C1dbo = bo_ij->C1dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+ coef.C2dbo = bo_ij->C2dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+ coef.C3dbo = bo_ij->C3dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+
+ coef.C1dbopi = bo_ij->C1dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C2dbopi = bo_ij->C2dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C3dbopi = bo_ij->C3dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C4dbopi = bo_ij->C4dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+
+ coef.C1dbopi2 = bo_ij->C1dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C2dbopi2 = bo_ij->C2dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C3dbopi2 = bo_ij->C3dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C4dbopi2 = bo_ij->C4dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+
+ coef.C1dDelta = bo_ij->C1dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+ coef.C2dDelta = bo_ij->C2dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+ coef.C3dDelta = bo_ij->C3dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+
+
+ /************************************
+ * forces related to atom i *
+ * first neighbors of atom i *
+ ************************************/
+ for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ k = nbr_k->nbr;
+
+ rvec_Scale( temp, -coef.C2dbo, nbr_k->bo_data.dBOp ); /*2nd,dBO*/
+ rvec_ScaledAdd( temp, -coef.C2dDelta, nbr_k->bo_data.dBOp );/*dDelta*/
+ rvec_ScaledAdd( temp, -coef.C3dbopi, nbr_k->bo_data.dBOp ); /*3rd,dBOpi*/
+ rvec_ScaledAdd( temp, -coef.C3dbopi2, nbr_k->bo_data.dBOp );/*3rd,dBOpi2*/
+
+ /* force */
+ rvec_Add( atoms[k].f, temp );
+ /* pressure */
+ rvec_iMultiply( ext_press, nbr_k->rel_box, temp );
+ rvec_Add( data->ext_press, ext_press );
+ }
+
+ /* then atom i itself */
+ rvec_Scale( temp, coef.C1dbo, bo_ij->dBOp ); /*1st, dBO*/
+ rvec_ScaledAdd( temp, coef.C2dbo, workspace->dDeltap_self[i] ); /*2nd, dBO*/
+
+ rvec_ScaledAdd( temp, coef.C1dDelta, bo_ij->dBOp ); /*1st, dBO*/
+ rvec_ScaledAdd( temp, coef.C2dDelta, workspace->dDeltap_self[i] );/*2nd, dBO*/
+
+ rvec_ScaledAdd( temp, coef.C1dbopi, bo_ij->dln_BOp_pi ); /*1st,dBOpi*/
+ rvec_ScaledAdd( temp, coef.C2dbopi, bo_ij->dBOp ); /*2nd,dBOpi*/
+ rvec_ScaledAdd( temp, coef.C3dbopi, workspace->dDeltap_self[i] );/*3rd,dBOpi*/
+
+ rvec_ScaledAdd(temp, coef.C1dbopi2, bo_ij->dln_BOp_pi2) ; /*1st,dBO_pi2*/
+ rvec_ScaledAdd(temp, coef.C2dbopi2, bo_ij->dBOp); /*2nd,dBO_pi2*/
+ rvec_ScaledAdd(temp, coef.C3dbopi2, workspace->dDeltap_self[i]);/*3rd,dBO_pi2*/
+
+ /* force */
+ rvec_Add( atoms[i].f, temp );
+ /* ext pressure due to i dropped, counting force on j only will be enough */
+
+
+ /****************************************************************************
+ * forces and pressure related to atom j *
+ * first neighbors of atom j *
+ ***************************************************************************/
+ for( pk = Start_Index(j, bonds); pk < End_Index(j, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ k = nbr_k->nbr;
+
+ rvec_Scale( temp, -coef.C3dbo, nbr_k->bo_data.dBOp ); /*3rd,dBO*/
+ rvec_ScaledAdd( temp, -coef.C3dDelta, nbr_k->bo_data.dBOp );/*dDelta*/
+ rvec_ScaledAdd( temp, -coef.C4dbopi, nbr_k->bo_data.dBOp ); /*4th,dBOpi*/
+ rvec_ScaledAdd( temp, -coef.C4dbopi2, nbr_k->bo_data.dBOp );/*4th,dBOpi2*/
+
+ /* force */
+ rvec_Add( atoms[k].f, temp );
+ /* pressure */
+ if( k != i ) {
+ ivec_Sum(rel_box, nbr_k->rel_box, nbr_j->rel_box);//k's rel_box wrt i
+ rvec_iMultiply( ext_press, rel_box, temp );
+ rvec_Add( data->ext_press, ext_press );
+ }
+ }
+
+ /* then atom j itself */
+ rvec_Scale( temp, -coef.C1dbo, bo_ij->dBOp ); /*1st, dBO*/
+ rvec_ScaledAdd( temp, coef.C3dbo, workspace->dDeltap_self[j] ); /*2nd, dBO*/
+
+ rvec_ScaledAdd( temp, -coef.C1dDelta, bo_ij->dBOp ); /*1st, dBO*/
+ rvec_ScaledAdd( temp, coef.C3dDelta, workspace->dDeltap_self[j] );/*2nd, dBO*/
+
+ rvec_ScaledAdd( temp, -coef.C1dbopi, bo_ij->dln_BOp_pi ); /*1st,dBOpi*/
+ rvec_ScaledAdd( temp, -coef.C2dbopi, bo_ij->dBOp ); /*2nd,dBOpi*/
+ rvec_ScaledAdd( temp, coef.C4dbopi, workspace->dDeltap_self[j] );/*3rd,dBOpi*/
+
+ rvec_ScaledAdd(temp, -coef.C1dbopi2, bo_ij->dln_BOp_pi2); /*1st,dBOpi2*/
+ rvec_ScaledAdd(temp, -coef.C2dbopi2, bo_ij->dBOp); /*2nd,dBOpi2*/
+ rvec_ScaledAdd(temp, coef.C4dbopi2, workspace->dDeltap_self[j]);/*3rd,dBOpi2*/
+
+ /* force */
+ rvec_Add( atoms[j].f, temp );
+ /* pressure */
+ rvec_iMultiply( ext_press, nbr_j->rel_box, temp );
+ rvec_Add( data->ext_press, ext_press );
+}
+
+
+HOST_DEVICE void Cuda_Add_dBond_to_Forces ( int i, int pj, reax_atom *atoms,
+ static_storage *workspace, list *bonds )
+{
+ bond_data *nbr_j, *nbr_k;
+ bond_order_data *bo_ij, *bo_ji;
+ dbond_coefficients coef;
+ int pk, k, j;
+ rvec t_f;
+
+ /* Initializations */
+ nbr_j = &(bonds->select.bond_list[pj]);
+ j = nbr_j->nbr;
+
+ if (i < j)
+ {
+ bo_ij = &(nbr_j->bo_data);
+ bo_ji = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
+ } else {
+ bo_ji = &(nbr_j->bo_data);
+ bo_ij = &(bonds->select.bond_list[ nbr_j->sym_index ].bo_data);
+ }
+
+ coef.C1dbo = bo_ij->C1dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+ coef.C2dbo = bo_ij->C2dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+ coef.C3dbo = bo_ij->C3dbo * (bo_ij->Cdbo + bo_ji->Cdbo);
+
+ coef.C1dbopi = bo_ij->C1dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C2dbopi = bo_ij->C2dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C3dbopi = bo_ij->C3dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+ coef.C4dbopi = bo_ij->C4dbopi * (bo_ij->Cdbopi + bo_ji->Cdbopi);
+
+ coef.C1dbopi2 = bo_ij->C1dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C2dbopi2 = bo_ij->C2dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C3dbopi2 = bo_ij->C3dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+ coef.C4dbopi2 = bo_ij->C4dbopi2 * (bo_ij->Cdbopi2 + bo_ji->Cdbopi2);
+
+ coef.C1dDelta = bo_ij->C1dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+ coef.C2dDelta = bo_ij->C2dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+ coef.C3dDelta = bo_ij->C3dbo * (workspace->CdDelta[i]+workspace->CdDelta[j]);
+
+ if ( i < j) {
+ for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ k = nbr_k->nbr;
+ rvec_MakeZero (t_f);
+
+ rvec_ScaledAdd( t_f, -coef.C2dbo, nbr_k->bo_data.dBOp );
+ /*2nd, dBO*/
+ rvec_ScaledAdd( t_f, -coef.C2dDelta, nbr_k->bo_data.dBOp );
+ /*dDelta*/
+ rvec_ScaledAdd( t_f, -coef.C3dbopi, nbr_k->bo_data.dBOp );
+ /*3rd, dBOpi*/
+ rvec_ScaledAdd( t_f, -coef.C3dbopi2, nbr_k->bo_data.dBOp );
+ /*3rd, dBOpi2*/
+
+ //Store in the temp place
+ rvec_Add (nbr_k->t_f, t_f);
+ }
+
+ rvec_ScaledAdd( atoms[i].f, coef.C1dbo, bo_ij->dBOp );
+ /*1st, dBO*/
+ rvec_ScaledAdd( atoms[i].f, coef.C2dbo, workspace->dDeltap_self[i] );
+ /*2nd, dBO*/
+
+ rvec_ScaledAdd(atoms[i].f, coef.C1dDelta, bo_ij->dBOp);
+ /*1st, dBO*/
+ rvec_ScaledAdd(atoms[i].f, coef.C2dDelta, workspace->dDeltap_self[i]);
+ /*2nd, dBO*/
+
+ rvec_ScaledAdd( atoms[i].f, coef.C1dbopi, bo_ij->dln_BOp_pi );
+ /*1st, dBOpi*/
+ rvec_ScaledAdd( atoms[i].f, coef.C2dbopi, bo_ij->dBOp );
+ /*2nd, dBOpi*/
+ rvec_ScaledAdd( atoms[i].f, coef.C3dbopi, workspace->dDeltap_self[i]);
+ /*3rd, dBOpi*/
+
+ rvec_ScaledAdd( atoms[i].f, coef.C1dbopi2, bo_ij->dln_BOp_pi2 );
+ /*1st, dBO_pi2*/
+ rvec_ScaledAdd( atoms[i].f, coef.C2dbopi2, bo_ij->dBOp );
+ /*2nd, dBO_pi2*/
+ rvec_ScaledAdd( atoms[i].f, coef.C3dbopi2, workspace->dDeltap_self[i]);
+ /*3rd, dBO_pi2*/
+ }
+ else
+ {
+ for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ k = nbr_k->nbr;
+ rvec_MakeZero (t_f);
+
+ rvec_ScaledAdd( t_f, -coef.C3dbo, nbr_k->bo_data.dBOp );
+ /*3rd, dBO*/
+ rvec_ScaledAdd( t_f, -coef.C3dDelta, nbr_k->bo_data.dBOp );
+ /*dDelta*/
+ rvec_ScaledAdd( t_f, -coef.C4dbopi, nbr_k->bo_data.dBOp );
+ /*4th, dBOpi*/
+ rvec_ScaledAdd( t_f, -coef.C4dbopi2, nbr_k->bo_data.dBOp );
+ /*4th, dBOpi2*/
+
+ //Store in the temp place
+ rvec_Add (nbr_k->t_f, t_f);
+ }
+
+ rvec_ScaledAdd( atoms[i].f, -coef.C1dbo, bo_ij->dBOp );
+ /*1st, dBO*/
+ rvec_ScaledAdd( atoms[i].f, coef.C3dbo, workspace->dDeltap_self[i] );
+ /*2nd, dBO*/
+
+ rvec_ScaledAdd( atoms[i].f, -coef.C1dDelta, bo_ij->dBOp );
+ /*1st, dBO*/
+ rvec_ScaledAdd(atoms[i].f, coef.C3dDelta, workspace->dDeltap_self[i]);
+ /*2nd, dBO*/
+
+ rvec_ScaledAdd( atoms[i].f, -coef.C1dbopi, bo_ij->dln_BOp_pi );
+ /*1st, dBOpi*/
+ rvec_ScaledAdd( atoms[i].f, -coef.C2dbopi, bo_ij->dBOp );
+ /*2nd, dBOpi*/
+ rvec_ScaledAdd(atoms[i].f, coef.C4dbopi, workspace->dDeltap_self[i]);
+ /*3rd, dBOpi*/
+
+ rvec_ScaledAdd( atoms[i].f, -coef.C1dbopi2, bo_ij->dln_BOp_pi2 );
+ /*1st, dBOpi2*/
+ rvec_ScaledAdd( atoms[i].f, -coef.C2dbopi2, bo_ij->dBOp );
+ /*2nd, dBOpi2*/
+ rvec_ScaledAdd(atoms[i].f, coef.C4dbopi2, workspace->dDeltap_self[i]);
+ /*3rd, dBOpi2*/
+ }
+}
+
+
+HOST_DEVICE void Cuda_dbond_to_Forces_postprocess (int i, reax_atom *atoms, list *bonds)
+{
+ int pk;
+ bond_data *nbr_k, *nbr_k_sym;
+
+ /*
+ for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ rvec_Add (atoms[i].f, nbr_k->t_f);
+ }
+ */
+
+ for( pk = Start_Index(i, bonds); pk < End_Index(i, bonds); ++pk ) {
+ nbr_k = &(bonds->select.bond_list[pk]);
+ nbr_k_sym = &( bonds->select.bond_list [nbr_k->sym_index] );
+
+ rvec_Add (atoms[i].f, nbr_k_sym->t_f);
+ }
+}
+
+
+GLOBAL void Cuda_Calculate_Bond_Orders_Init ( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
+ static_storage workspace, int num_atom_types, int N )
+{
+ int i, type_i;
+ real p_boc1, p_boc2;
+ single_body_parameters *sbp_i;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ p_boc1 = g_params.l[0];
+ p_boc2 = g_params.l[1];
+
+ /* Calculate Deltaprime, Deltaprime_boc values */
+ type_i = atoms[i].type;
+ sbp_i = &(sbp[type_i]);
+ workspace.Deltap[i] = workspace.total_bond_order[i] - sbp_i->valency;
+ workspace.Deltap_boc[i] =
+ workspace.total_bond_order[i] - sbp_i->valency_val;
+ workspace.total_bond_order[i] = 0;
+}
+
+
+/* A very important and crucial assumption here is that each segment
+ belonging to a different atom in nbrhoods->nbr_list is sorted in its own.
+ This can either be done in the general coordinator function or here */
+GLOBAL void Cuda_Calculate_Bond_Orders ( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
+ two_body_parameters *tbp, static_storage workspace, list bonds,
+ list dDeltas, list dBOs, int num_atom_types, int N )
+{
+ int i, j, pj, type_i, type_j;
+ int start_i, end_i;
+ int num_bonds, sym_index;
+ real p_boc1, p_boc2;
+ real val_i, Deltap_i, Deltap_boc_i;
+ real val_j, Deltap_j, Deltap_boc_j;
+ real temp, f1, f2, f3, f4, f5, f4f5, exp_f4, exp_f5;
+ real exp_p1i, exp_p2i, exp_p1j, exp_p2j;
+ real u1_ij, u1_ji, Cf1A_ij, Cf1B_ij, Cf1_ij, Cf1_ji;
+ real Cf45_ij, Cf45_ji, p_lp1;
+ real A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji;
+ real explp1;
+ two_body_parameters *twbp;
+ bond_order_data *bo_ij, *bo_ji;
+ single_body_parameters *sbp_i, *sbp_j;
+
+#if defined(TEST_FORCES)
+ int k, pk, start_j, end_j;
+ int top_dbo=0, top_dDelta=0;
+ dbond_data *pdbo;
+ dDelta_data *ptop_dDelta;
+
+#endif
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ num_bonds = 0;
+ p_boc1 = g_params.l[0];
+ p_boc2 = g_params.l[1];
+
+ /* Calculate Deltaprime, Deltaprime_boc values */
+ //for( i = 0; i < system->N; ++i ) {
+ /*
+ if (i < N) {
+ type_i = atoms[i].type;
+ sbp_i = &(sbp[type_i]);
+ workspace.Deltap[i] = workspace.total_bond_order[i] - sbp_i->valency;
+ workspace.Deltap_boc[i] =
+ workspace.total_bond_order[i] - sbp_i->valency_val;
+ workspace.total_bond_order[i] = 0;
+
+ }
+
+ __syncthreads ();
+ */
+
+
+ // fprintf( stderr, "done with uncorrected bond orders\n" );
+
+ /* Corrected Bond Order calculations */
+ //for( i = 0; i < system->N; ++i ) {
+ type_i = atoms[i].type;
+ sbp_i = &(sbp[type_i]);
+ val_i = sbp_i->valency;
+ Deltap_i = workspace.Deltap[i];
+ Deltap_boc_i = workspace.Deltap_boc[i];
+ start_i = Start_Index(i, &bonds);
+ end_i = End_Index(i, &bonds);
+ //fprintf( stderr, "i:%d Dp:%g Dbocp:%g s:%d e:%d\n",
+ // i+1, Deltap_i, Deltap_boc_i, start_i, end_i );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ j = bonds.select.bond_list[pj].nbr;
+ type_j = atoms[j].type;
+ bo_ij = &( bonds.select.bond_list[pj].bo_data );
+ //fprintf( stderr, "\tj:%d - ubo: %8.3f\n", j+1, bo_ij->BO );
+
+ if( i < j ) {
+ twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ] );
+#ifdef TEST_FORCES
+ Set_Start_Index( pj, top_dbo, &dBOs );
+ /* fprintf( stderr, "%6d%6d%23.15e%23.15e%23.15e\n",
+ workspace->reverse_map[i], workspace->reverse_map[j],
+ twbp->ovc, twbp->v13cor, bo_ij->BO ); */
+#endif
+ if( twbp->ovc < 0.001 && twbp->v13cor < 0.001 ) {
+ /* There is no correction to bond orders nor to derivatives of
+ bond order prime! So we leave bond orders unchanged and
+ set derivative of bond order coefficients s.t.
+ dBO = dBOp & dBOxx = dBOxxp in Add_dBO_to_Forces */
+ bo_ij->C1dbo = 1.000000;
+ bo_ij->C2dbo = 0.000000;
+ bo_ij->C3dbo = 0.000000;
+
+ bo_ij->C1dbopi = bo_ij->BO_pi;
+ bo_ij->C2dbopi = 0.000000;
+ bo_ij->C3dbopi = 0.000000;
+ bo_ij->C4dbopi = 0.000000;
+
+ bo_ij->C1dbopi2 = bo_ij->BO_pi2;
+ bo_ij->C2dbopi2 = 0.000000;
+ bo_ij->C3dbopi2 = 0.000000;
+ bo_ij->C4dbopi2 = 0.000000;
+
+#ifdef TEST_FORCES
+ pdbo = &(dBOs.select.dbo_list[ top_dbo ]);
+
+ // compute dBO_ij/dr_i
+ pdbo->wrt = i;
+ rvec_Copy( pdbo->dBO, bo_ij->dBOp );
+ rvec_Scale( pdbo->dBOpi, bo_ij->BO_pi, bo_ij->dln_BOp_pi );
+ rvec_Scale( pdbo->dBOpi2, bo_ij->BO_pi2, bo_ij->dln_BOp_pi2 );
+
+ // compute dBO_ij/dr_j
+ pdbo++;
+ pdbo->wrt = j;
+ rvec_Scale( pdbo->dBO,-1.0,bo_ij->dBOp );
+ rvec_Scale( pdbo->dBOpi,-bo_ij->BO_pi,bo_ij->dln_BOp_pi );
+ rvec_Scale( pdbo->dBOpi2,-bo_ij->BO_pi2,bo_ij->dln_BOp_pi2 );
+
+ top_dbo += 2;
+#endif
+ }
+ else {
+ val_j = sbp[type_j].valency;
+ Deltap_j = workspace.Deltap[j];
+ Deltap_boc_j = workspace.Deltap_boc[j];
+
+ /* on page 1 */
+ if( twbp->ovc >= 0.001 ) {
+ /* Correction for overcoordination */
+ exp_p1i = EXP( -p_boc1 * Deltap_i );
+ exp_p2i = EXP( -p_boc2 * Deltap_i );
+ exp_p1j = EXP( -p_boc1 * Deltap_j );
+ exp_p2j = EXP( -p_boc2 * Deltap_j );
+
+ f2 = exp_p1i + exp_p1j;
+ f3 = -1.0 / p_boc2 * log( 0.5 * ( exp_p2i + exp_p2j ) );
+ f1 = 0.5 * ( ( val_i + f2 )/( val_i + f2 + f3 ) +
+ ( val_j + f2 )/( val_j + f2 + f3 ) );
+
+ /*fprintf( stderr,"%6d%6d\t%g %g j:%g %g p_boc:%g %g\n",
+ i+1, j+1, val_i, Deltap_i, val_j, Deltap_j, p_boc1, p_boc2 );
+ fprintf( stderr,"\tf:%g %g %g, exp:%g %g %g %g\n",
+ f1, f2, f3, exp_p1i, exp_p2i, exp_p1j, exp_p2j );*/
+
+ /* Now come the derivates */
+ /* Bond Order pages 5-7, derivative of f1 */
+ temp = f2 + f3;
+ u1_ij = val_i + temp;
+ u1_ji = val_j + temp;
+ Cf1A_ij = 0.5 * f3 * (1.0 / SQR( u1_ij ) + 1.0 / SQR( u1_ji ));
+ Cf1B_ij = -0.5 * (( u1_ij - f3 ) / SQR( u1_ij ) +
+ ( u1_ji - f3 ) / SQR( u1_ji ));
+
+ //Cf1_ij = -Cf1A_ij * p_boc1 * exp_p1i +
+ // Cf1B_ij * exp_p2i / ( exp_p2i + exp_p2j );
+ Cf1_ij = 0.50 * ( -p_boc1 * exp_p1i / u1_ij -
+ ((val_i+f2) / SQR(u1_ij)) *
+ ( -p_boc1 * exp_p1i +
+ exp_p2i / ( exp_p2i + exp_p2j ) ) +
+ -p_boc1 * exp_p1i / u1_ji -
+ ((val_j+f2)/SQR(u1_ji)) * ( -p_boc1*exp_p1i +
+ exp_p2i / ( exp_p2i + exp_p2j ) ));
+
+ Cf1_ji = -Cf1A_ij * p_boc1 * exp_p1j +
+ Cf1B_ij * exp_p2j / ( exp_p2i + exp_p2j );
+ //fprintf( stderr, "\tCf1:%g %g\n", Cf1_ij, Cf1_ji );
+ }
+ else {
+ /* No overcoordination correction! */
+ f1 = 1.0;
+ Cf1_ij = Cf1_ji = 0.0;
+ }
+
+ if( twbp->v13cor >= 0.001 ) {
+ /* Correction for 1-3 bond orders */
+ exp_f4 =EXP(-(twbp->p_boc4 * SQR( bo_ij->BO ) -
+ Deltap_boc_i) * twbp->p_boc3 + twbp->p_boc5);
+ exp_f5 =EXP(-(twbp->p_boc4 * SQR( bo_ij->BO ) -
+ Deltap_boc_j) * twbp->p_boc3 + twbp->p_boc5);
+
+ f4 = 1. / (1. + exp_f4);
+ f5 = 1. / (1. + exp_f5);
+ f4f5 = f4 * f5;
+
+ /* Bond Order pages 8-9, derivative of f4 and f5 */
+ /*temp = twbp->p_boc5 -
+ twbp->p_boc3 * twbp->p_boc4 * SQR( bo_ij->BO );
+ u_ij = temp + twbp->p_boc3 * Deltap_boc_i;
+ u_ji = temp + twbp->p_boc3 * Deltap_boc_j;
+ Cf45_ij = Cf45( u_ij, u_ji ) / f4f5;
+ Cf45_ji = Cf45( u_ji, u_ij ) / f4f5;*/
+ Cf45_ij = -f4 * exp_f4;
+ Cf45_ji = -f5 * exp_f5;
+ }
+ else {
+ f4 = f5 = f4f5 = 1.0;
+ Cf45_ij = Cf45_ji = 0.0;
+ }
+
+ /* Bond Order page 10, derivative of total bond order */
+ A0_ij = f1 * f4f5;
+ A1_ij = -2 * twbp->p_boc3 * twbp->p_boc4 * bo_ij->BO *
+ (Cf45_ij + Cf45_ji);
+ A2_ij = Cf1_ij / f1 + twbp->p_boc3 * Cf45_ij;
+ A2_ji = Cf1_ji / f1 + twbp->p_boc3 * Cf45_ji;
+ A3_ij = A2_ij + Cf1_ij / f1;
+ A3_ji = A2_ji + Cf1_ji / f1;
+
+ /*fprintf( stderr, "\tBO: %f, A0: %f, A1: %f, A2_ij: %f
+A2_ji: %f, A3_ij: %f, A3_ji: %f\n",
+bo_ij->BO, A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji );*/
+
+ /* find corrected bond order values and their deriv coefs */
+ bo_ij->BO = bo_ij->BO * A0_ij;
+ bo_ij->BO_pi = bo_ij->BO_pi * A0_ij *f1;
+ bo_ij->BO_pi2= bo_ij->BO_pi2* A0_ij *f1;
+ bo_ij->BO_s = bo_ij->BO - ( bo_ij->BO_pi + bo_ij->BO_pi2 );
+
+ bo_ij->C1dbo = A0_ij + bo_ij->BO * A1_ij;
+ bo_ij->C2dbo = bo_ij->BO * A2_ij;
+ bo_ij->C3dbo = bo_ij->BO * A2_ji;
+
+ bo_ij->C1dbopi = f1*f1*f4*f5;
+ bo_ij->C2dbopi = bo_ij->BO_pi * A1_ij;
+ bo_ij->C3dbopi = bo_ij->BO_pi * A3_ij;
+ bo_ij->C4dbopi = bo_ij->BO_pi * A3_ji;
+
+ bo_ij->C1dbopi2 = f1*f1*f4*f5;
+ bo_ij->C2dbopi2 = bo_ij->BO_pi2 * A1_ij;
+ bo_ij->C3dbopi2 = bo_ij->BO_pi2 * A3_ij;
+
+#ifdef TEST_FORCES
+ /*fprintf( stderr, "%6d%6d%13.6f%13.6f%13.6f%13.6f\n",
+ i+1, j+1, bo_ij->BO, bo_ij->C1dbo, Cf45_ij, Cf45_ji );*/
+
+ /* fprintf( stderr, "%6d%6d%13.6f%13.6f%13.6f%13.6f\n",
+ //"%6d%6d%10.6f%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f%10.6f\n%10.6f%10.6f%10.6f%10.6f\n\n",
+ workspace->orig_id[i], workspace->orig_id[j]
+ A0_ij, A1_ij, A2_ij, A2_ji, A3_ij, A3_ji
+ bo_ij->BO, bo_ij->BO_pi, bo_ij->BO_pi2, bo_ij->BO_s,
+ bo_ij->C1dbo, bo_ij->C2dbo, bo_ij->C3dbo,
+ bo_ij->C1dbopi,bo_ij->C2dbopi,bo_ij->C3dbopi,bo_ij->C4dbopi,
+ bo_ij->C1dbopi2,bo_ij->C2dbopi2,bo_ij->C3dbopi2,bo_ij->C4dbopi2
+ ); */
+
+ Calculate_dBO( i, pj, workspace, lists, &top_dbo );
+#endif
+ }
+
+ /* neglect bonds that are < 1e-10 */
+ if( bo_ij->BO < 1e-10 )
+ bo_ij->BO = 0.0;
+ if( bo_ij->BO_s < 1e-10 )
+ bo_ij->BO_s = 0.0;
+ if( bo_ij->BO_pi < 1e-10 )
+ bo_ij->BO_pi = 0.0;
+ if( bo_ij->BO_pi2 < 1e-10 )
+ bo_ij->BO_pi2 = 0.0;
+
+ workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
+
+
+ /* fprintf( stderr, "%d %d\t%g %g %g %g\n
+Cdbo:\t%g %g %g\n
+Cdbopi:\t%g %g %g %g\n
+Cdbopi2:%g %g %g %g\n\n",
+i+1, j+1, bonds->select.bond_list[ pj ].d,
+bo_ij->BO,bo_ij->BO_pi, bo_ij->BO_pi2,
+bo_ij->C1dbo, bo_ij->C2dbo, bo_ij->C3dbo,
+bo_ij->C1dbopi, bo_ij->C2dbopi, bo_ij->C3dbopi, bo_ij->C4dbopi,
+bo_ij->C1dbopi2, bo_ij->C2dbopi2,
+bo_ij->C3dbopi2, bo_ij->C4dbopi2 ); */
+
+ /* fprintf( stderr, "%d %d, BO:%f BO_s:%f BO_pi:%f BO_pi2:%f\n",
+ i+1,j+1,bo_ij->BO,bo_ij->BO_s,bo_ij->BO_pi,bo_ij->BO_pi2 ); */
+
+#ifdef TEST_FORCES
+ Set_End_Index( pj, top_dbo, &dBOs );
+ //Add_dBO( system, lists, i, pj, 1.0, workspace->dDelta );
+#endif
+ }
+ /*
+ else {
+ // We only need to update bond orders from bo_ji
+ // everything else is set in uncorrected_bo calculations
+ sym_index = bonds.select.bond_list[pj].sym_index;
+ bo_ji = &(bonds.select.bond_list[ sym_index ].bo_data);
+ bo_ij->BO = bo_ji->BO;
+ bo_ij->BO_s = bo_ji->BO_s;
+ bo_ij->BO_pi = bo_ji->BO_pi;
+ bo_ij->BO_pi2 = bo_ji->BO_pi2;
+
+ workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
+
+#ifdef TEST_FORCES
+ //Add_dBO( system, lists, j, sym_index, 1.0, workspace.dDelta );
+#endif
+}
+ */
+ }
+
+#ifdef TEST_FORCES
+ // fprintf( stderr, "dDelta computations\nj:" );
+ Set_Start_Index( i, top_dDelta, &dDeltas );
+ ptop_dDelta = &( dDeltas.select.dDelta_list[top_dDelta] );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ j = bonds.select.bond_list[pj].nbr;
+ // fprintf( stderr, "%d ", j );
+
+ if( !rvec_isZero( workspace.dDelta[j] ) ) {
+ ptop_dDelta->wrt = j;
+ rvec_Copy( ptop_dDelta->dVal, workspace.dDelta[j] );
+ rvec_MakeZero( workspace.dDelta[j] );
+ ++top_dDelta, ++ptop_dDelta;
+ }
+
+ start_j = Start_Index(j, &bonds);
+ end_j = End_Index(j, &bonds);
+ for( pk = start_j; pk < end_j; ++pk ) {
+ k = bonds.select.bond_list[pk].nbr;
+ if( !rvec_isZero( workspace.dDelta[k] ) ) {
+ ptop_dDelta->wrt = k;
+ rvec_Copy( ptop_dDelta->dVal, workspace.dDelta[k] );
+ rvec_MakeZero( workspace.dDelta[k] );
+ ++top_dDelta, ++ptop_dDelta;
+ }
+ }
+ }
+
+ Set_End_Index( i, top_dDelta, &dDeltas );
+
+ /*for( pj=Start_Index(i,dDeltas); pj<End_Index(i,dDeltas); ++pj )
+ fprintf( stdout, "dDel: %d %d [%g %g %g]\n",
+ i+1, dDeltas->select.dDelta_list[pj].wrt+1,
+ dDeltas->select.dDelta_list[pj].dVal[0],
+ dDeltas->select.dDelta_list[pj].dVal[1],
+ dDeltas->select.dDelta_list[pj].dVal[2] );*/
+#endif
+ //}
+
+ /*fprintf(stderr,"\tCalculated actual bond orders ...\n" );
+ fprintf(stderr,"%6s%8s%8s%8s%8s%8s%8s%8s\n",
+ "atom", "Delta", "Delta_e", "Delta_boc", "nlp",
+ "Delta_lp", "Clp", "dDelta_lp" );*/
+
+ /*
+ p_lp1 = g_params.l[15];
+
+ //get the kernel ID for the following computation
+ j = i;
+
+ // Calculate some helper variables that are used at many places
+ // throughout force calculations
+ //for( j = 0; j < system->N; ++j ) {
+ type_j = atoms[j].type;
+ sbp_j = &(sbp[ type_j ]);
+
+ workspace.Delta[j] = workspace.total_bond_order[j] - sbp_j->valency;
+ workspace.Delta_e[j] = workspace.total_bond_order[j] - sbp_j->valency_e;
+ workspace.Delta_boc[j] = workspace.total_bond_order[j] -
+ sbp_j->valency_boc;
+
+ workspace.vlpex[j] = workspace.Delta_e[j] -
+ 2.0 * (int)(workspace.Delta_e[j]/2.0);
+ explp1 = EXP(-p_lp1 * SQR(2.0 + workspace.vlpex[j]));
+ workspace.nlp[j] = explp1 - (int)(workspace.Delta_e[j] / 2.0);
+ workspace.Delta_lp[j] = sbp_j->nlp_opt - workspace.nlp[j];
+ workspace.Clp[j] = 2.0 * p_lp1 * explp1 * (2.0 + workspace.vlpex[j]);
+ // Adri uses different dDelta_lp values than the ones in notes... //
+ workspace.dDelta_lp[j] = workspace.Clp[j];
+ //workspace->dDelta_lp[j] = workspace->Clp[j] + (0.5-workspace->Clp[j]) *
+ //((fabs(workspace->Delta_e[j]/2.0 -
+ // (int)(workspace->Delta_e[j]/2.0)) < 0.1) ? 1 : 0 );
+
+ if( sbp_j->mass > 21.0 ) {
+ workspace.nlp_temp[j] = 0.5 * (sbp_j->valency_e - sbp_j->valency);
+ workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
+ workspace.dDelta_lp_temp[j] = 0.;
+ }
+ else {
+ workspace.nlp_temp[j] = workspace.nlp[j];
+ workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
+ workspace.dDelta_lp_temp[j] = workspace.Clp[j];
+ }
+
+ //fprintf( stderr, "%d\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\n",
+ //j, workspace->Delta[j], workspace->Delta_e[j], workspace->Delta_boc[j],
+ //workspace->nlp[j], system->reaxprm.sbp[type_j].nlp_opt,
+ //workspace->Delta_lp[j], workspace->Clp[j], workspace->dDelta_lp[j] );
+ //}
+ */
+
+ //Print_Bonds( system, bonds, "sbonds.out" );
+
+#if defined(DEBUG)
+ //fprintf( stderr, "Number of bonds: %d\n", num_bonds );
+ //Print_Bond_Orders( system, control, data, workspace, lists, out_control );
+#endif
+}
+
+
+GLOBAL void Cuda_Update_Uncorrected_BO ( static_storage workspace, list bonds, int N )
+{
+ int i, j, pj;
+ int start_i, end_i;
+ int sym_index;
+
+ bond_order_data *bo_ij, *bo_ji;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ start_i = Start_Index(i, &bonds);
+ end_i = End_Index(i, &bonds);
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+
+ j = bonds.select.bond_list[pj].nbr;
+ bo_ij = &( bonds.select.bond_list[pj].bo_data );
+
+ if( i >= j ) {
+ // We only need to update bond orders from bo_ji
+ // everything else is set in uncorrected_bo calculations
+ sym_index = bonds.select.bond_list[pj].sym_index;
+ bo_ji = &(bonds.select.bond_list[ sym_index ].bo_data);
+ bo_ij->BO = bo_ji->BO;
+ bo_ij->BO_s = bo_ji->BO_s;
+ bo_ij->BO_pi = bo_ji->BO_pi;
+ bo_ij->BO_pi2 = bo_ji->BO_pi2;
+
+ workspace.total_bond_order[i] += bo_ij->BO; // now keeps total_BO
+ }
+ }
+}
+
+
+GLOBAL void Cuda_Update_Workspace_After_Bond_Orders( reax_atom *atoms, global_parameters g_params, single_body_parameters *sbp,
+ static_storage workspace, int N )
+{
+ int j, type_j;
+ real explp1;
+ real p_lp1;
+ single_body_parameters *sbp_i, *sbp_j;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+ if (j >= N) return;
+
+ p_lp1 = g_params.l[15];
+
+ /* Calculate some helper variables that are used at many places
+ throughout force calculations */
+ //for( j = 0; j < system->N; ++j ) {
+ type_j = atoms[j].type;
+ sbp_j = &(sbp[ type_j ]);
+
+ workspace.Delta[j] = workspace.total_bond_order[j] - sbp_j->valency;
+ workspace.Delta_e[j] = workspace.total_bond_order[j] - sbp_j->valency_e;
+ workspace.Delta_boc[j] = workspace.total_bond_order[j] -
+ sbp_j->valency_boc;
+
+ workspace.vlpex[j] = workspace.Delta_e[j] -
+ 2.0 * (int)(workspace.Delta_e[j]/2.0);
+ explp1 = EXP(-p_lp1 * SQR(2.0 + workspace.vlpex[j]));
+ workspace.nlp[j] = explp1 - (int)(workspace.Delta_e[j] / 2.0);
+ workspace.Delta_lp[j] = sbp_j->nlp_opt - workspace.nlp[j];
+ workspace.Clp[j] = 2.0 * p_lp1 * explp1 * (2.0 + workspace.vlpex[j]);
+ /* Adri uses different dDelta_lp values than the ones in notes... */
+ workspace.dDelta_lp[j] = workspace.Clp[j];
+ //workspace->dDelta_lp[j] = workspace->Clp[j] + (0.5-workspace->Clp[j]) *
+ //((fabs(workspace->Delta_e[j]/2.0 -
+ // (int)(workspace->Delta_e[j]/2.0)) < 0.1) ? 1 : 0 );
+
+ if( sbp_j->mass > 21.0 ) {
+ workspace.nlp_temp[j] = 0.5 * (sbp_j->valency_e - sbp_j->valency);
+ workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
+ workspace.dDelta_lp_temp[j] = 0.;
+ }
+ else {
+ workspace.nlp_temp[j] = workspace.nlp[j];
+ workspace.Delta_lp_temp[j] = sbp_j->nlp_opt - workspace.nlp_temp[j];
+ workspace.dDelta_lp_temp[j] = workspace.Clp[j];
+ }
+
+ //fprintf( stderr, "%d\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\n",
+ //j, workspace->Delta[j], workspace->Delta_e[j], workspace->Delta_boc[j],
+ //workspace->nlp[j], system->reaxprm.sbp[type_j].nlp_opt,
+ //workspace->Delta_lp[j], workspace->Clp[j], workspace->dDelta_lp[j] );
+ //}
+
+}
+
+
+//Import from the forces file.
+GLOBAL void Cuda_Compute_Total_Force (reax_atom *atoms, simulation_data *data,
+ static_storage workspace, list p_bonds, int ensemble, int N)
+{
+ int i, pj;
+ list *bonds = &p_bonds;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i < N)
+ {
+ for (pj = Start_Index (i, bonds); pj < End_Index (i, bonds); ++pj)
+ {
+ //int j = bonds->select.bond_list[pj].nbr;
+ if (ensemble == NVE || ensemble == NVT || ensemble == bNVT)
+ Cuda_Add_dBond_to_Forces (i, pj, atoms, &workspace, bonds );
+ else
+ Cuda_Add_dBond_to_Forces_NPT (i, pj, atoms, data, &workspace, bonds );
+ }
+ }
+}
+
+
+GLOBAL void Cuda_Compute_Total_Force_PostProcess (reax_atom *atoms, simulation_data *data,
+ static_storage workspace, list p_bonds, int ensemble, int N)
+{
+ int i, pj;
+ list *bonds = &p_bonds;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i < N)
+ {
+ if (ensemble == NVE || ensemble == NVT || ensemble == bNVT)
+ Cuda_dbond_to_Forces_postprocess (i, atoms, bonds );
+ }
+}
diff --git a/PuReMD-GPU/src/cuda_bond_orders.h b/PuReMD-GPU/src/cuda_bond_orders.h
new file mode 100644
index 0000000000000000000000000000000000000000..4015b9fa34aa1e66843efc8dc1a49f1c75da749c
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_bond_orders.h
@@ -0,0 +1,48 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_BOND_ORDERS_H_
+#define __CUDA_BOND_ORDERS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void Cuda_Calculate_Bond_Orders_Init ( reax_atom *, global_parameters , single_body_parameters *,
+ static_storage , int , int );
+GLOBAL void Cuda_Calculate_Bond_Orders ( reax_atom *, global_parameters , single_body_parameters *,
+ two_body_parameters *, static_storage , list , list , list , int , int );
+GLOBAL void Cuda_Update_Uncorrected_BO ( static_storage , list , int );
+GLOBAL void Cuda_Update_Workspace_After_Bond_Orders( reax_atom *, global_parameters , single_body_parameters *,
+ static_storage , int );
+GLOBAL void Cuda_Compute_Total_Force (reax_atom *, simulation_data *, static_storage , list , int , int );
+GLOBAL void Cuda_Compute_Total_Force_PostProcess (reax_atom *, simulation_data *, static_storage , list , int , int );
+//HOST_DEVICE void Cuda_Add_dBond_to_Forces( int, int, reax_atom *, static_storage*, list* );
+//HOST_DEVICE void Cuda_Add_dBond_to_Forces_NPT( int, int, reax_atom *, simulation_data*, static_storage*, list* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/helpers.cu b/PuReMD-GPU/src/cuda_box.cu
similarity index 87%
rename from PuReMD-GPU/src/helpers.cu
rename to PuReMD-GPU/src/cuda_box.cu
index 29ae31e305598009053100ae239ac6dbeed59d44..a9eb16faac9970f6bc89eeec4a42241cd30f0724 100644
--- a/PuReMD-GPU/src/helpers.cu
+++ b/PuReMD-GPU/src/cuda_box.cu
@@ -18,11 +18,13 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "cuda_helpers.h"
-#include "helpers.h"
#include "box.h"
-GLOBAL void compute_Inc_on_T3 (reax_atom *atoms, unsigned int N, simulation_box *box, real d1, real d2, real d3)
+
+GLOBAL void k_compute_Inc_on_T3(reax_atom *atoms, unsigned int N,
+ simulation_box *box, real d1, real d2, real d3)
{
int index = blockIdx.x * blockDim.x + threadIdx.x;
rvec dx;
@@ -31,5 +33,7 @@ GLOBAL void compute_Inc_on_T3 (reax_atom *atoms, unsigned int N, simulation_box
dx[2] = d3;
if (index < N )
+ {
Inc_on_T3( atoms[index].x, dx, box );
+ }
}
diff --git a/PuReMD-GPU/src/matvec.h b/PuReMD-GPU/src/cuda_box.h
similarity index 83%
rename from PuReMD-GPU/src/matvec.h
rename to PuReMD-GPU/src/cuda_box.h
index e032febd4ef79968a7334e43c07bfb1db5e6f49c..913abc15867cf5ec9d8e87dad0fa56198de9b1ff 100644
--- a/PuReMD-GPU/src/matvec.h
+++ b/PuReMD-GPU/src/cuda_box.h
@@ -18,13 +18,14 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
-#ifndef __MATVEC__H_
-#define __MATVEC__H_
+#ifndef __CUDA_BOX_H__
+#define __CUDA_BOX_H__
#include "mytypes.h"
-GLOBAL void Cuda_Matvec (sparse_matrix , real *, real *, int );
-GLOBAL void Cuda_Matvec_csr (sparse_matrix , real *, real *, int );
+
+GLOBAL void k_compute_Inc_on_T3 (reax_atom *atoms, unsigned int N,
+ simulation_box *box, real d1, real d2, real d3);
+
#endif
diff --git a/PuReMD-GPU/src/center_mass.cu b/PuReMD-GPU/src/cuda_center_mass.cu
similarity index 91%
rename from PuReMD-GPU/src/center_mass.cu
rename to PuReMD-GPU/src/cuda_center_mass.cu
index ea8f799846b0d8c794007762ba18869bc9686787..158d3a16489f20362bd854312eb39aa0dcec57e8 100644
--- a/PuReMD-GPU/src/center_mass.cu
+++ b/PuReMD-GPU/src/cuda_center_mass.cu
@@ -18,17 +18,13 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "cuda_center_mass.h"
-
-
-#include "center_mass.h"
#include "vector.h"
-GLOBAL void center_of_mass_blocks (single_body_parameters *sbp, reax_atom *atoms,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n)
+
+GLOBAL void k_center_of_mass_blocks( single_body_parameters *sbp, reax_atom *atoms,
+ rvec *res_xcm, rvec *res_vcm, rvec *res_amcm, size_t n )
{
extern __shared__ rvec xcm[];
extern __shared__ rvec vcm[];
@@ -76,13 +72,9 @@ GLOBAL void center_of_mass_blocks (single_body_parameters *sbp, reax_atom *atoms
}
}
-GLOBAL void center_of_mass (rvec *xcm,
- rvec *vcm,
- rvec *amcm,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n)
+
+GLOBAL void k_center_of_mass( rvec *xcm, rvec *vcm, rvec *amcm,
+ rvec *res_xcm, rvec *res_vcm, rvec *res_amcm, size_t n )
{
extern __shared__ rvec sh_xcm[];
extern __shared__ rvec sh_vcm[];
@@ -131,11 +123,9 @@ GLOBAL void center_of_mass (rvec *xcm,
}
}
-GLOBAL void compute_center_mass (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n)
+
+GLOBAL void k_compute_center_mass_sbp( single_body_parameters *sbp, reax_atom *atoms,
+ real *results, real xcm0, real xcm1, real xcm2, size_t n )
{
extern __shared__ real xx[];
extern __shared__ real xy[];
@@ -160,11 +150,11 @@ GLOBAL void compute_center_mass (single_body_parameters *sbp,
xcm[1] = xcm1;
xcm[2] = xcm2;
-
xx[xx_i] = xy [xy_i + threadIdx.x] = xz[xz_i + threadIdx.x] =
yy[yy_i + threadIdx.x] = yz[yz_i + threadIdx.x] = zz[zz_i + threadIdx.x] = 0;
- if (i < n){
+ if (i < n)
+ {
m = sbp[ atoms[i].type ].mass;
rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
xx[ xx_i ] = diff[0] * diff[0] * m;
@@ -176,8 +166,10 @@ GLOBAL void compute_center_mass (single_body_parameters *sbp,
}
__syncthreads ();
- for (int offset = blockDim.x / 2; offset > 0; offset >>= 1){
- if (threadIdx.x < offset){
+ for (int offset = blockDim.x / 2; offset > 0; offset >>= 1)
+ {
+ if (threadIdx.x < offset)
+ {
index = threadIdx.x + offset;
xx[ threadIdx.x ] += xx[ index ];
xy[ xy_i + threadIdx.x ] += xy [ xy_i + index ];
@@ -189,7 +181,8 @@ GLOBAL void compute_center_mass (single_body_parameters *sbp,
__syncthreads ();
}
- if (threadIdx.x == 0) {
+ if (threadIdx.x == 0)
+ {
results [ blockIdx.x*6 ] = xx [ 0 ];
results [ blockIdx.x*6 + 1 ] = xy [ xy_i + 0 ];
results [ blockIdx.x*6 + 2 ] = xz [ xz_i + 0 ];
@@ -199,7 +192,8 @@ GLOBAL void compute_center_mass (single_body_parameters *sbp,
}
}
-GLOBAL void compute_center_mass (real *input, real *output, size_t n)
+
+GLOBAL void k_compute_center_mass( real *input, real *output, size_t n )
{
extern __shared__ real xx[];
extern __shared__ real xy[];
diff --git a/PuReMD-GPU/src/cuda_center_mass.h b/PuReMD-GPU/src/cuda_center_mass.h
new file mode 100644
index 0000000000000000000000000000000000000000..0c1d76ec63defe3cb6a0738a2843b10d02104bb5
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_center_mass.h
@@ -0,0 +1,44 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_CENTER_MASS_H__
+#define __CUDA_CENTER_MASS_H__
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void k_center_of_mass_blocks( single_body_parameters *, reax_atom *,
+ rvec *res_xcm, rvec *res_vcm, rvec *res_amcm, size_t n );
+GLOBAL void k_center_of_mass( rvec *xcm,
+ rvec *vcm, rvec *amcm, rvec *res_xcm, rvec *res_vcm, rvec *res_amcm, size_t n );
+GLOBAL void k_compute_center_mass_sbp( single_body_parameters *sbp,
+ reax_atom *atoms, real *results, real xcm0, real xcm1, real xcm2, size_t n );
+GLOBAL void k_compute_center_mass( real *input, real *output, size_t n );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_copy.cu b/PuReMD-GPU/src/cuda_copy.cu
index 2db79e3718e0c5ea3c54913e29147c71735552f9..1f50dbf3c74fae4d9ab73149d02c51dbec8ca10e 100644
--- a/PuReMD-GPU/src/cuda_copy.cu
+++ b/PuReMD-GPU/src/cuda_copy.cu
@@ -18,91 +18,96 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "cuda_copy.h"
+#include "cuda_list.h"
-
-#include "cuda_copy.h"
#include "vector.h"
-void Sync_Host_Device (grid *host, grid *dev, enum cudaMemcpyKind dir)
+
+void Sync_Host_Device_Grid( grid *host, grid *dev, enum cudaMemcpyKind dir )
{
- copy_host_device (host->top, dev->top,
- INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_TOP);
+ copy_host_device( host->top, dev->top,
+ INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_TOP );
- copy_host_device (host->mark, dev->mark,
- INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_MARK);
+ copy_host_device( host->mark, dev->mark,
+ INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_MARK );
- copy_host_device (host->start, dev->start,
- INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_START);
+ copy_host_device( host->start, dev->start,
+ INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_START );
- copy_host_device (host->end, dev->end,
- INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_END);
+ copy_host_device( host->end, dev->end,
+ INT_SIZE * host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_END );
- copy_host_device (host->atoms, dev->atoms,
- INT_SIZE * host->max_atoms*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_ATOMS);
+ copy_host_device( host->atoms, dev->atoms,
+ INT_SIZE * host->max_atoms*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_ATOMS );
- copy_host_device (host->nbrs, dev->nbrs,
- IVEC_SIZE * host->max_nbrs*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_NBRS);
+ copy_host_device( host->nbrs, dev->nbrs,
+ IVEC_SIZE * host->max_nbrs*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_NBRS );
- copy_host_device (host->nbrs_cp, dev->nbrs_cp,
- RVEC_SIZE * host->max_nbrs*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_NBRS_CP);
+ copy_host_device( host->nbrs_cp, dev->nbrs_cp,
+ RVEC_SIZE * host->max_nbrs*host->ncell[0]*host->ncell[1]*host->ncell[2], dir, RES_GRID_NBRS_CP );
}
-void Sync_Host_Device (reax_system *sys, enum cudaMemcpyKind dir)
+void Sync_Host_Device_Sys( reax_system *sys, enum cudaMemcpyKind dir )
{
- copy_host_device (sys->atoms, sys->d_atoms,
- REAX_ATOM_SIZE * sys->N, dir, RES_SYSTEM_ATOMS);
+ copy_host_device( sys->atoms, sys->d_atoms,
+ REAX_ATOM_SIZE * sys->N, dir, RES_SYSTEM_ATOMS );
- copy_host_device (&(sys->box), sys->d_box, SIMULATION_BOX_SIZE, dir, RES_SYSTEM_SIMULATION_BOX );
+ copy_host_device( &(sys->box), sys->d_box, SIMULATION_BOX_SIZE, dir, RES_SYSTEM_SIMULATION_BOX );
//synch bonds here.
- copy_host_device (sys->reaxprm.sbp, sys->reaxprm.d_sbp, SBP_SIZE * sys->reaxprm.num_atom_types,
+ copy_host_device( sys->reaxprm.sbp, sys->reaxprm.d_sbp, SBP_SIZE * sys->reaxprm.num_atom_types,
dir, RES_REAX_INT_SBP );
- copy_host_device (sys->reaxprm.tbp, sys->reaxprm.d_tbp, TBP_SIZE * pow (sys->reaxprm.num_atom_types, 2),
+ copy_host_device( sys->reaxprm.tbp, sys->reaxprm.d_tbp, TBP_SIZE * pow (sys->reaxprm.num_atom_types, 2),
dir, RES_REAX_INT_TBP );
- copy_host_device (sys->reaxprm.thbp, sys->reaxprm.d_thbp, THBP_SIZE * pow (sys->reaxprm.num_atom_types, 3),
+ copy_host_device( sys->reaxprm.thbp, sys->reaxprm.d_thbp, THBP_SIZE * pow (sys->reaxprm.num_atom_types, 3),
dir, RES_REAX_INT_THBP );
- copy_host_device (sys->reaxprm.hbp, sys->reaxprm.d_hbp, HBP_SIZE * pow (sys->reaxprm.num_atom_types, 3),
+ copy_host_device( sys->reaxprm.hbp, sys->reaxprm.d_hbp, HBP_SIZE * pow (sys->reaxprm.num_atom_types, 3),
dir, RES_REAX_INT_HBP );
- copy_host_device (sys->reaxprm.fbp, sys->reaxprm.d_fbp, FBP_SIZE * pow (sys->reaxprm.num_atom_types, 4),
+ copy_host_device( sys->reaxprm.fbp, sys->reaxprm.d_fbp, FBP_SIZE * pow (sys->reaxprm.num_atom_types, 4),
dir, RES_REAX_INT_FBP );
- copy_host_device (sys->reaxprm.gp.l, sys->reaxprm.d_gp.l, REAL_SIZE * sys->reaxprm.gp.n_global,
+ copy_host_device( sys->reaxprm.gp.l, sys->reaxprm.d_gp.l, REAL_SIZE * sys->reaxprm.gp.n_global,
dir, RES_GLOBAL_PARAMS );
sys->reaxprm.d_gp.n_global = sys->reaxprm.gp.n_global;
sys->reaxprm.d_gp.vdw_type = sys->reaxprm.gp.vdw_type;
}
-void Sync_Host_Device (simulation_data *host, simulation_data *dev, enum cudaMemcpyKind dir)
+
+void Sync_Host_Device_Data( simulation_data *host, simulation_data *dev, enum cudaMemcpyKind dir )
{
- copy_host_device (host, dev, SIMULATION_DATA_SIZE, dir, RES_SIMULATION_DATA );
+ copy_host_device( host, dev, SIMULATION_DATA_SIZE, dir, RES_SIMULATION_DATA );
}
-void Sync_Host_Device (sparse_matrix *L, sparse_matrix *U, enum cudaMemcpyKind dir )
+
+void Sync_Host_Device_Mat( sparse_matrix *L, sparse_matrix *U, enum cudaMemcpyKind dir )
{
- copy_host_device ( L->start, dev_workspace->L.start, INT_SIZE * (L->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( L->end, dev_workspace->L.end, INT_SIZE * (L->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( L->entries, dev_workspace->L.entries, SPARSE_MATRIX_ENTRY_SIZE * L->m, dir, RES_SPARSE_MATRIX_ENTRY );
+ copy_host_device( L->start, dev_workspace->L.start, INT_SIZE * (L->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( L->end, dev_workspace->L.end, INT_SIZE * (L->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( L->entries, dev_workspace->L.entries, SPARSE_MATRIX_ENTRY_SIZE * L->m, dir, RES_SPARSE_MATRIX_ENTRY );
- copy_host_device ( U->start, dev_workspace->U.start, INT_SIZE * (U->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( U->end, dev_workspace->U.end, INT_SIZE * (U->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
- copy_host_device ( U->entries, dev_workspace->U.entries, SPARSE_MATRIX_ENTRY_SIZE * U->m, dir, RES_SPARSE_MATRIX_ENTRY );
+ copy_host_device( U->start, dev_workspace->U.start, INT_SIZE * (U->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( U->end, dev_workspace->U.end, INT_SIZE * (U->n + 1), dir, RES_SPARSE_MATRIX_INDEX );
+ copy_host_device( U->entries, dev_workspace->U.entries, SPARSE_MATRIX_ENTRY_SIZE * U->m, dir, RES_SPARSE_MATRIX_ENTRY );
}
-void Sync_Host_Device (output_controls *, control_params *, enum cudaMemcpyKind)
+
+void Sync_Host_Device_Control( output_controls *, control_params *, enum cudaMemcpyKind )
{
}
-void Sync_Host_Device (control_params *host, control_params *device, enum cudaMemcpyKind)
+
+void Sync_Host_Device_Params( control_params *host, control_params *device, enum cudaMemcpyKind )
{
- copy_host_device (host, device, CONTROL_PARAMS_SIZE, cudaMemcpyHostToDevice, RES_CONTROL_PARAMS );
+ copy_host_device( host, device, CONTROL_PARAMS_SIZE, cudaMemcpyHostToDevice, RES_CONTROL_PARAMS );
}
-void Prep_Device_For_Output (reax_system *system, simulation_data *data )
+void Prep_Device_For_Output( reax_system *system, simulation_data *data )
{
//int size = sizeof (simulation_data) - (2*sizeof (reax_timing) + sizeof (void *));
//unsigned long start_address = (unsigned long)data->d_simulation_data + (unsigned long) (2 * INT_SIZE + REAL_SIZE);
@@ -112,7 +117,7 @@ void Prep_Device_For_Output (reax_system *system, simulation_data *data )
//fprintf (stderr, "size to copy --> %d \n", size );
//copy_host_device (data, (simulation_data *)data->d_simulation_data, size, cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
- //Sync_Host_Device (data, (simulation_data *)data->d_simulation_data, cudaMemcpyDeviceToHost );
+ //Sync_Host_Device_Data( data, (simulation_data *)data->d_simulation_data, cudaMemcpyDeviceToHost );
/*
copy_host_device (&data->E_BE, &((simulation_data *)data->d_simulation_data)->E_BE,
REAL_SIZE * 13, cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
@@ -126,7 +131,7 @@ void Prep_Device_For_Output (reax_system *system, simulation_data *data )
*/
simulation_data local_data;
- copy_host_device (&local_data, (simulation_data *)data->d_simulation_data,
+ copy_host_device( &local_data, (simulation_data *)data->d_simulation_data,
SIMULATION_DATA_SIZE, cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
data->E_BE = local_data.E_BE;
data->E_Ov = local_data.E_Ov;
@@ -141,43 +146,46 @@ void Prep_Device_For_Output (reax_system *system, simulation_data *data )
data->E_vdW = local_data.E_vdW;
data->E_Ele = local_data.E_Ele;
data->E_Kin = local_data.E_Kin;
- rvec_Copy (data->int_press, local_data.int_press);
- rvec_Copy (data->ext_press, local_data.ext_press);
+ rvec_Copy( data->int_press, local_data.int_press);
+ rvec_Copy( data->ext_press, local_data.ext_press);
data->kin_press = local_data.kin_press;
data->therm.T = local_data.therm.T;
- //Sync_Host_Device (&system.g, &system.d_g, cudaMemcpyDeviceToHost );
- Sync_Host_Device (system, cudaMemcpyDeviceToHost );
+ //Sync_Host_Device_Sys( &system.g, &system.d_g, cudaMemcpyDeviceToHost );
+ Sync_Host_Device_Sys( system, cudaMemcpyDeviceToHost );
}
-void Sync_Host_Device (list *host, list *device, int type)
+
+void Sync_Host_Device_List( list *host, list *device, int type )
{
//list is already allocated -- discard it first
if (host->n > 0)
- Delete_List (host, TYP_HOST);
+ {
+ Cuda_Delete_List( host );
+ }
//memory is allocated on the host
- Make_List(device->n, device->num_intrs, type, host, TYP_HOST );
+ Cuda_Make_List( device->n, device->num_intrs, type, host );
//memcpy the entries from device to host
- copy_host_device (host->index, device->index, INT_SIZE * device->n, cudaMemcpyDeviceToHost, LIST_INDEX );
- copy_host_device (host->end_index, device->end_index, INT_SIZE * device->n, cudaMemcpyDeviceToHost, LIST_END_INDEX );
+ copy_host_device( host->index, device->index, INT_SIZE * device->n, cudaMemcpyDeviceToHost, LIST_INDEX );
+ copy_host_device( host->end_index, device->end_index, INT_SIZE * device->n, cudaMemcpyDeviceToHost, LIST_END_INDEX );
switch (type)
{
case TYP_BOND:
- copy_host_device (host->select.bond_list, device->select.bond_list,
+ copy_host_device( host->select.bond_list, device->select.bond_list,
BOND_DATA_SIZE * device->num_intrs, cudaMemcpyDeviceToHost, LIST_BOND_DATA );
break;
case TYP_THREE_BODY:
- copy_host_device (host->select.three_body_list, device->select.three_body_list,
- sizeof (three_body_interaction_data )* device->num_intrs, cudaMemcpyDeviceToHost, LIST_THREE_BODY_DATA );
+ copy_host_device( host->select.three_body_list, device->select.three_body_list,
+ sizeof( three_body_interaction_data ) * device->num_intrs, cudaMemcpyDeviceToHost, LIST_THREE_BODY_DATA );
break;
default:
- fprintf (stderr, "Unknown list synching from device to host ---- > %d \n", type );
- exit (1);
+ fprintf( stderr, "Unknown list synching from device to host ---- > %d \n", type );
+ exit( 1 );
break;
}
}
diff --git a/PuReMD-GPU/src/cuda_copy.h b/PuReMD-GPU/src/cuda_copy.h
index 561a49fbcce0fbb7b4ff03a38874b0fcadb5d7e8..6b6f38ae3d12d7d00d8cd38eef996d61cb765bd5 100644
--- a/PuReMD-GPU/src/cuda_copy.h
+++ b/PuReMD-GPU/src/cuda_copy.h
@@ -18,24 +18,32 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
-
#ifndef __CUDA_COPY_H_
#define __CUDA_COPY_H_
#include "cuda_utils.h"
-#include "cuda.h"
+
#include "mytypes.h"
#include "list.h"
-void Sync_Host_Device (grid *, grid *, enum cudaMemcpyKind);
-void Sync_Host_Device (reax_system *, enum cudaMemcpyKind);
-void Sync_Host_Device (control_params *, control_params *, enum cudaMemcpyKind);
-void Sync_Host_Device (simulation_data *, simulation_data *, enum cudaMemcpyKind);
-void Sync_Host_Device (sparse_matrix *, sparse_matrix *, enum cudaMemcpyKind);
-void Sync_Host_Device (output_controls *, enum cudaMemcpyKind);
-void Prep_Device_For_Output (reax_system *, simulation_data *);
-void Sync_Host_Device (list *host, list *device, int type);
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Sync_Host_Device_Grid( grid *, grid *, enum cudaMemcpyKind );
+void Sync_Host_Device_Sys( reax_system *, enum cudaMemcpyKind );
+void Sync_Host_Device_Params( control_params *, control_params *, enum cudaMemcpyKind );
+void Sync_Host_Device_Data( simulation_data *, simulation_data *, enum cudaMemcpyKind );
+void Sync_Host_Device_Mat( sparse_matrix *, sparse_matrix *, enum cudaMemcpyKind );
+void Sync_Host_Device_Control( output_controls *, enum cudaMemcpyKind );
+
+void Prep_Device_For_Output( reax_system *, simulation_data * );
+void Sync_Host_Device_List( list *host, list *device, int type );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PuReMD-GPU/src/cuda_environment.cu b/PuReMD-GPU/src/cuda_environment.cu
new file mode 100644
index 0000000000000000000000000000000000000000..cd6ae50d82b4716aef1d16f2f957dc5f4976cca7
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_environment.cu
@@ -0,0 +1,71 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_environment.h"
+
+#include "cuda_utils.h"
+
+
+void Setup_Cuda_Environment( int rank, int nprocs, int gpus_per_node )
+{
+
+ int deviceCount = 0;
+ cudaError_t flag;
+ cublasHandle_t cublasHandle;
+ cusparseHandle_t cusparseHandle;
+ cusparseMatDescr_t matdescriptor;
+
+ flag = cudaGetDeviceCount( &deviceCount );
+
+ if ( flag != cudaSuccess || deviceCount < 1 )
+ {
+ fprintf( stderr, "ERROR: no CUDA capable device(s) found. Terminating...\n" );
+ exit( 1 );
+ }
+
+ //Calculate the # of GPUs per processor
+ //and assign the GPU for each process
+ //TODO: handle condition where # CPU procs > # GPUs
+ cudaSetDevice( rank % deviceCount );
+
+#if defined(__CUDA_DEBUG__)
+ fprintf( stderr, "p:%d is using GPU: %d \n", rank, rank % deviceCount );
+#endif
+
+ //CHANGE ORIGINAL
+ //cudaDeviceSetLimit( cudaLimitStackSize, 8192 );
+ //cudaDeviceSetCacheConfig( cudaFuncCachePreferL1 );
+ //cudaCheckError( );
+
+ cublasCheckError( cublasCreate(&cublasHandle) );
+
+ cusparseCheckError( cusparseCreate(&cusparseHandle) );
+ cusparseCheckError( cusparseCreateMatDescr(&matdescriptor) );
+ cusparseSetMatType( matdescriptor, CUSPARSE_MATRIX_TYPE_GENERAL );
+ cusparseSetMatIndexBase( matdescriptor, CUSPARSE_INDEX_BASE_ZERO );
+
+}
+
+
+void Cleanup_Cuda_Environment( )
+{
+ cudaDeviceReset( );
+ cudaDeviceSynchronize( );
+}
diff --git a/PuReMD-GPU/src/cuda_environment.h b/PuReMD-GPU/src/cuda_environment.h
new file mode 100644
index 0000000000000000000000000000000000000000..61f811db3354628a3067a3423442f875f16e6871
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_environment.h
@@ -0,0 +1,39 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_ENVIRONMENT_H__
+#define __CUDA_ENVIRONMENT_H__
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Setup_Cuda_Environment( int, int, int );
+void Cleanup_Cuda_Environment( );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_forces.cu b/PuReMD-GPU/src/cuda_forces.cu
new file mode 100644
index 0000000000000000000000000000000000000000..bf277b391ce0df0c5336ea0a0653b6863ca14fec
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_forces.cu
@@ -0,0 +1,2002 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_forces.h"
+
+#include "box.h"
+#include "forces.h"
+#include "index_utils.h"
+#include "list.h"
+#include "print_utils.h"
+#include "system_props.h"
+#include "vector.h"
+
+#include "cuda_utils.h"
+#include "cuda_init.h"
+#include "cuda_bond_orders.h"
+#include "cuda_single_body_interactions.h"
+#include "cuda_two_body_interactions.h"
+#include "cuda_three_body_interactions.h"
+#include "cuda_four_body_interactions.h"
+#include "cuda_list.h"
+#include "cuda_QEq.h"
+#include "cuda_reduction.h"
+#include "cuda_system_props.h"
+#include "validation.h"
+
+#include "cudaProfiler.h"
+
+
+void Cuda_Compute_Bonded_Forces( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ real t_start, t_elapsed;
+ real *spad = (real *)scratch;
+ rvec *rvec_spad;
+
+ //Compute the bonded for interaction here.
+ //Step 1.
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+ fprintf (stderr, " Begin Bonded Forces ... %d x %d\n", BLOCKS, BLOCK_SIZE);
+#endif
+
+ Cuda_Calculate_Bond_Orders_Init<<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
+ *dev_workspace, system->reaxprm.num_atom_types, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_Calculate_Bond_Orders<<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
+ system->reaxprm.d_tbp, *dev_workspace,
+ *(dev_lists + BONDS), *(dev_lists + DDELTA), *(dev_lists + DBO),
+ system->reaxprm.num_atom_types, system->N );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_Update_Uncorrected_BO<<<BLOCKS, BLOCK_SIZE>>>
+ (*dev_workspace, *(dev_lists + BONDS), system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_Update_Workspace_After_Bond_Orders<<<BLOCKS, BLOCK_SIZE>>>
+ (system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
+ *dev_workspace, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf( stderr, "Bond Orders... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
+ fprintf( stderr, "Cuda_Calculate_Bond_Orders Done... \n" );
+#endif
+
+ //Step 2.
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+ //cuda_memset (spad, 0, system->N * ( 2 * REAL_SIZE + system->N * REAL_SIZE + 16 * REAL_SIZE), RES_SCRATCH );
+ cuda_memset (spad, 0, system->N * ( 2 * REAL_SIZE ) , RES_SCRATCH );
+
+ Cuda_Bond_Energy <<< BLOCKS, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ (simulation_data *)data->d_simulation_data, *dev_workspace, *(dev_lists + BONDS),
+ system->N, system->reaxprm.num_atom_types, spad );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_BE
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ //(spad + system->N, spad + system->N + 16, 16);
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_BE, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf (stderr, "Cuda_Bond_Energy ... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
+ fprintf (stderr, "Cuda_Bond_Energy Done... \n");
+#endif
+
+ //Step 3.
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+ cuda_memset (spad, 0, ( 6 * REAL_SIZE * system->N ), RES_SCRATCH );
+
+ test_LonePair_OverUnder_Coordination_Energy_LP <<<BLOCKS, BLOCK_SIZE>>>( system->d_atoms, system->reaxprm.d_gp,
+ system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ *dev_workspace, (simulation_data *)data->d_simulation_data,
+ *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types,
+ spad, spad + 2 * system->N, spad + 4*system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ test_LonePair_OverUnder_Coordination_Energy <<<BLOCKS, BLOCK_SIZE>>>( system->d_atoms, system->reaxprm.d_gp,
+ system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ *dev_workspace, (simulation_data *)data->d_simulation_data,
+ *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types,
+ spad, spad + 2 * system->N, spad + 4*system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ test_LonePair_Postprocess <<<BLOCKS, BLOCK_SIZE, 0>>>( system->d_atoms, system->reaxprm.d_gp,
+ system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ *dev_workspace, (simulation_data *)data->d_simulation_data,
+ *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+
+ //Reduction for E_Lp
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Lp, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_Ov
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad + 2*system->N, spad + 3*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Ov, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_Un
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad + 4*system->N, spad + 5*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + 5*system->N, &((simulation_data *)data->d_simulation_data)->E_Un, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf (stderr, "test_LonePair_postprocess ... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
+ fprintf (stderr, "test_LonePair_postprocess Done... \n");
+#endif
+
+ //Step 4.
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+
+ cuda_memset(spad, 0, (dev_lists + BONDS)->num_intrs * sizeof (int), RES_SCRATCH);
+ k_Three_Body_Estimate<<<BLOCKS, BLOCK_SIZE>>>
+ (system->d_atoms,
+ (control_params *)control->d_control,
+ *(dev_lists + BONDS),
+ system->N, (int *)spad);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf (stderr, "Three_Body_Estimate... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
+#endif
+
+ int *thbody = (int *) malloc (sizeof (int) * (dev_lists + BONDS)->num_intrs);
+ memset (thbody, 0, sizeof (int) * (dev_lists + BONDS)->num_intrs);
+ copy_host_device (thbody, spad, (dev_lists + BONDS)->num_intrs * sizeof (int), cudaMemcpyDeviceToHost, RES_SCRATCH);
+
+ int total_3body = thbody [0] * SAFE_ZONE;
+ for (int x = 1; x < (dev_lists + BONDS)->num_intrs; x++) {
+ total_3body += thbody [x]*SAFE_ZONE;
+ thbody [x] += thbody [x-1];
+ }
+ system->num_thbodies = thbody [(dev_lists+BONDS)->num_intrs-1];
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Total Three body estimate is %d (bonds: %d) \n", total_3body, (dev_lists+BONDS)->num_intrs);
+#endif
+
+ if (!system->init_thblist)
+ {
+ system->init_thblist = TRUE;
+ if(!Cuda_Make_List( (dev_lists+BONDS)->num_intrs, total_3body, TYP_THREE_BODY, dev_lists + THREE_BODIES )) {
+ fprintf( stderr, "Problem in initializing three-body list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+#ifdef __CUDA_MEM__
+ fprintf (stderr, "Device memory allocated: three body list = %d MB\n",
+ sizeof (three_body_interaction_data) * total_3body / (1024*1024));
+#endif
+ } else {
+ if ((dev_workspace->realloc.bonds > 0) || (system->num_thbodies > (dev_lists+THREE_BODIES)->num_intrs )) {
+ int size = MAX (dev_workspace->realloc.num_bonds, (dev_lists+BONDS)->num_intrs);
+
+ /*Delete Three-body list*/
+ Cuda_Delete_List( dev_lists + THREE_BODIES );
+
+#ifdef __CUDA_MEM__
+ fprintf (stderr, "Reallocating Three-body list: step: %d n - %d num_intrs - %d used: %d \n",
+ data->step, dev_workspace->realloc.num_bonds, total_3body, system->num_thbodies);
+#endif
+ /*Recreate Three-body list */
+ if(!Cuda_Make_List( size, total_3body, TYP_THREE_BODY, dev_lists + THREE_BODIES )) {
+ fprintf( stderr, "Problem in initializing three-body list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+ }
+ }
+
+ //copy the indexes into the thb list;
+ copy_host_device (thbody, ((dev_lists + THREE_BODIES)->index + 1), sizeof (int) * ((dev_lists+BONDS)->num_intrs - 1),
+ cudaMemcpyHostToDevice, LIST_INDEX);
+ copy_host_device (thbody, ((dev_lists + THREE_BODIES)->end_index + 1), sizeof (int) * ((dev_lists+BONDS)->num_intrs - 1),
+ cudaMemcpyHostToDevice, LIST_END_INDEX);
+
+ free (thbody );
+
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+
+ cuda_memset (spad, 0, ( 6 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2), RES_SCRATCH );
+
+ k_Three_Body_Interactions <<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms,
+ system->reaxprm.d_sbp, system->reaxprm.d_thbp, system->reaxprm.d_gp,
+ (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data,
+ *dev_workspace,
+ *(dev_lists + BONDS), *(dev_lists + THREE_BODIES),
+ system->N, system->reaxprm.num_atom_types,
+ spad, spad + 2*system->N, spad + 4*system->N, (rvec *)(spad + 6*system->N));
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Not necessary to validate three-body list anymore,
+ // Estimate is already done at the beginning which makes sure that
+ // we have sufficient size for this list
+ //Cuda_Threebody_List( system, workspace, dev_lists + THREE_BODIES, data->step );
+
+ //Reduction for E_Ang
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Ang, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_Pen
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad + 2*system->N, spad + 3*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Pen, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_Coa
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad + 4*system->N, spad + 5*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + 5*system->N, &((simulation_data *)data->d_simulation_data)->E_Coa, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //Reduction for ext_pres
+ rvec_spad = (rvec *) (spad + 6*system->N);
+ Cuda_reduction_rvec<<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
+ (rvec_spad, rvec_spad + system->N, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ Cuda_reduction_rvec<<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
+ (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ real t_1, t_2;
+ t_1 = Get_Time( );
+ //Sum up the f vector for each atom and collect the CdDelta from all the bonds
+ k_Three_Body_Interactions_results <<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, (control_params *)control->d_control,
+ *dev_workspace, *(dev_lists + BONDS), system->N );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+ t_2 = Get_Timing_Info( t_1 );
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf( stderr, "Three_Body_Interactions post process Timing %lf \n", t_2 );
+ fprintf( stderr, "Three_Body_Interactions ... Timing %lf \n", t_elapsed );
+ fprintf( stderr, "Three_Body_Interactions Done... \n" );
+#endif
+
+ //Step 5.
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+
+ cuda_memset( spad, 0, ( 4 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2), RES_SCRATCH );
+ //k_Four_Body_Interactions<<< system->N, 32, 32*( 2*REAL_SIZE + RVEC_SIZE)>>>
+ k_Four_Body_Interactions<<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_fbp,
+ (control_params *)control->d_control, *(dev_lists + BONDS), *(dev_lists + THREE_BODIES),
+ (simulation_box *)system->d_box, (simulation_data *)data->d_simulation_data,
+ *dev_workspace, system->N, system->reaxprm.num_atom_types,
+ spad, spad + 2*system->N, (rvec *) (spad + 4*system->N) );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //Reduction for E_Tor
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Tor, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for E_Con
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad + 2*system->N, spad + 3*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Con, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Reduction for ext_pres
+ rvec_spad = (rvec *) (spad + 4*system->N);
+ Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
+ (rvec_spad, rvec_spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
+ (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Post process here
+ k_Four_Body_Postprocess<<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms, *dev_workspace, *(dev_lists + BONDS),
+ system->N );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+#ifdef __DEBUG_CUDA__
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf (stderr, "Four_Body_post process return value --> %d --- Four body Timing %lf \n", cudaGetLastError (), t_elapsed );
+ fprintf (stderr, " Four_Body_ Done... \n");
+#endif
+
+ //Step 6.
+ if (control->hb_cut > 0) {
+#ifdef __DEBUG_CUDA__
+ t_start = Get_Time( );
+#endif
+ cuda_memset (spad, 0, ( 2 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2 ), RES_SCRATCH );
+
+ /*
+ k_Hydrogen_Bonds <<< BLOCKS, BLOCK_SIZE, BLOCK_SIZE *( REAL_SIZE + RVEC_SIZE) >>>
+ ( system->d_atoms,
+ system->reaxprm.d_sbp,
+ system->reaxprm.d_hbp,
+ (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data,
+ *dev_workspace,
+ *(dev_lists + BONDS), *(dev_lists + HBONDS),
+ system->N, system->reaxprm.num_atom_types,
+ spad, (rvec *) (spad + 2*system->N), NULL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+
+#ifdef __DEBUG_CUDA__
+ real test1,test2;
+ test1 = Get_Time ();
+#endif
+
+ int hbs = (system->N * HBONDS_THREADS_PER_ATOM/ HBONDS_BLOCK_SIZE) +
+ (((system->N * HBONDS_THREADS_PER_ATOM) % HBONDS_BLOCK_SIZE) == 0 ? 0 : 1);
+ k_Hydrogen_Bonds_HB <<< hbs, HBONDS_BLOCK_SIZE, HBONDS_BLOCK_SIZE * ( 2 * REAL_SIZE + 2 * RVEC_SIZE ) >>>
+ ( system->d_atoms,
+ system->reaxprm.d_sbp,
+ system->reaxprm.d_hbp,
+ (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data,
+ *dev_workspace,
+ *(dev_lists + BONDS), *(dev_lists + HBONDS),
+ system->N, system->reaxprm.num_atom_types,
+ spad, (rvec *) (spad + 2*system->N), NULL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+#ifdef __DEBUG_CUDA__
+ test2 = Get_Timing_Info (test1);
+ fprintf (stderr, "Timing for the hb and forces ---> %f \n", test2);
+#endif
+
+ //Reduction for E_HB
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_HB, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+
+ //Reduction for ext_pres
+ rvec_spad = (rvec *) (spad + 2*system->N);
+ Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
+ (rvec_spad, rvec_spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
+ (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //Post process here
+#ifdef __DEBUG_CUDA__
+ real t_1, t_2;
+ t_1 = Get_Time ();
+#endif
+
+ k_Hydrogen_Bonds_Postprocess <<< BLOCKS, BLOCK_SIZE, BLOCK_SIZE * RVEC_SIZE >>>
+ ( system->d_atoms,
+ system->reaxprm.d_sbp,
+ *dev_workspace,
+ *(dev_lists + BONDS),
+ *(dev_lists + HBONDS),
+ *(dev_lists + FAR_NBRS),
+ system->N,
+ spad); //this is for the fix to use the shared memory
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+#ifdef __DEBUG_CUDA__
+ t_2 = Get_Timing_Info ( t_1 );
+ fprintf (stderr, " Hydrogen Bonds post process -----%f \n", t_2);
+ t_1 = Get_Time ();
+#endif
+
+ //k_Hydrogen_Bonds_Far_Nbrs <<< system->N, 32, 32 * RVEC_SIZE>>>
+ k_Hydrogen_Bonds_HNbrs <<< system->N, 32, 32 * RVEC_SIZE>>>
+ ( system->d_atoms,
+ system->reaxprm.d_sbp,
+ *dev_workspace,
+ *(dev_lists + BONDS),
+ *(dev_lists + HBONDS),
+ *(dev_lists + FAR_NBRS),
+ system->N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ t_2 = Get_Timing_Info ( t_1 );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Hydrogen Bonds post process -----%f \n", t_2);
+ t_elapsed = Get_Timing_Info( t_start );
+ fprintf (stderr, "Hydrogen bonds post process return value --> %d --- HydrogenBonds Timing %lf \n", cudaGetLastError (), t_elapsed );
+ fprintf (stderr, "Hydrogen_Bond Done... \n");
+#endif
+ }
+ return;
+}
+
+
+void Cuda_Compute_NonBonded_Forces( reax_system *system, control_params *control,
+ simulation_data *data,static_storage *workspace,
+ list** lists, output_controls *out_control )
+{
+ real t_start, t_elapsed;
+ real t1 = 0, t2 = 0;
+ real *spad = (real *) scratch;
+ rvec *rvec_spad;
+ int cblks;
+
+ t_start = Get_Time( );
+ Cuda_QEq( system, control, data, workspace, lists[FAR_NBRS], out_control );
+ t_elapsed = Get_Timing_Info( t_start );
+ d_timing.QEq += t_elapsed;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Cuda_QEq done with timing %lf \n", t_elapsed );
+#endif
+
+ cuda_memset (spad, 0, system->N * ( 4 * REAL_SIZE + 2 * RVEC_SIZE), RES_SCRATCH );
+
+ t_start = Get_Time ();
+ if ( control->tabulate == 0)
+ {
+ cblks = (system->N * VDW_THREADS_PER_ATOM / VDW_BLOCK_SIZE) +
+ ((system->N * VDW_THREADS_PER_ATOM/VDW_BLOCK_SIZE) == 0 ? 0 : 1);
+ Cuda_vdW_Coulomb_Energy <<< cblks, VDW_BLOCK_SIZE, VDW_BLOCK_SIZE * ( 2*REAL_SIZE + RVEC_SIZE) >>>
+ ( system->d_atoms,
+ system->reaxprm.d_tbp,
+ system->reaxprm.d_gp,
+ (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data,
+ *(dev_lists + FAR_NBRS),
+ spad , spad + 2 * system->N, (rvec *) (spad + system->N * 4),
+ system->reaxprm.num_atom_types,
+ system->N ) ;
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ }
+ else
+ {
+ cblks = (system->N * VDW_THREADS_PER_ATOM / VDW_BLOCK_SIZE) +
+ ((system->N * VDW_THREADS_PER_ATOM/VDW_BLOCK_SIZE) == 0 ? 0 : 1);
+ Cuda_Tabulated_vdW_Coulomb_Energy <<< cblks, VDW_BLOCK_SIZE, VDW_BLOCK_SIZE* (2*REAL_SIZE + RVEC_SIZE)>>>
+ ( (reax_atom *)system->d_atoms,
+ (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data,
+ *(dev_lists + FAR_NBRS),
+ spad , spad + 2 * system->N, (rvec *) (spad + system->N * 4),
+ d_LR,
+ system->reaxprm.num_atom_types,
+ out_control->energy_update_freq,
+ system->N ) ;
+
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ }
+
+ t_elapsed = Get_Timing_Info (t_start );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Cuda_Tabulated_vdW_Coulomb_Energy done... %lf \n", (t_elapsed - t2));
+ fprintf (stderr, "Cuda_Tabulated_vdW_Coulomb_Energy done... %lf \n", (t_elapsed));
+#endif
+
+ //Reduction on E_vdW
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (spad, spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_vdW, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ //reduction on E_Ele
+ Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (spad + 2*system->N, spad + 3*system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Ele, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ rvec_spad = (rvec *) (spad + 4*system->N);
+
+ //reduction on ext_press
+ Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE>>>
+ (rvec_spad, rvec_spad + system->N, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2>>>
+ (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
+
+
+void Cuda_Validate_Lists( reax_system *system, static_storage *workspace, list **lists, int step, int n,
+ int num_bonds, int num_hbonds )
+{
+ int i, flag;
+ list *bonds, *hbonds, *thblist;
+ int *bonds_start, *bonds_end;
+ int *hbonds_start, *hbonds_end;
+ int *mat_start, *mat_end;
+ int max_sparse_entries = 0;
+
+ bonds = *lists + BONDS;
+ hbonds = *lists + HBONDS;
+
+ bonds_start = (int *) calloc (bonds->n, INT_SIZE);
+ bonds_end = (int *) calloc (bonds->n, INT_SIZE);
+
+ hbonds_start = (int *) calloc (hbonds->n, INT_SIZE );
+ hbonds_end = (int *) calloc (hbonds->n, INT_SIZE );
+
+ mat_start = (int *) calloc (workspace->H.n, INT_SIZE );
+ mat_end = (int *) calloc (workspace->H.n, INT_SIZE );
+
+ copy_host_device (bonds_start, bonds->index, bonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+ copy_host_device (bonds_end, bonds->end_index, bonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ copy_host_device (hbonds_start, hbonds->index, hbonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+ copy_host_device (hbonds_end, hbonds->end_index, hbonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ copy_host_device (mat_start, workspace->H.start, workspace->H.n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+ copy_host_device (mat_end, workspace->H.end, workspace->H.n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ /* Sparse Matrix entries */
+
+#ifdef __CUDA_TEST__
+ /*
+ workspace->realloc.Htop = 0;
+ for (i = 0; i < workspace->H.n-1; i++) {
+ if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i])){
+ workspace->realloc.Htop = mat_end[i] - mat_start[i];
+ }
+ }
+ */
+#endif
+
+ flag = -1;
+ workspace->realloc.Htop = 0;
+ for ( i = 0; i < n-1; i ++){
+
+ if( (mat_end[i] - mat_start[i]) >
+ (system->max_sparse_matrix_entries * DANGER_ZONE )) {
+ //fprintf (stderr, "step %d, Reached the water mark for sparse matrix for index: %d (%d %d) \n",
+ // step, i, mat_start[i], mat_end[i]);
+ if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i]))
+ workspace->realloc.Htop = (mat_end[i] - mat_start[i]) ;
+ }
+
+ if ( (mat_end[i] > mat_start[i+1]) ){
+ fprintf( stderr, "step%d-matcheck failed: i=%d end(i)=%d start(i+1)=%d\n",
+ step, flag, mat_end[i], mat_start[i+1]);
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+ if( (mat_end[i] - mat_start[i]) > system->max_sparse_matrix_entries * DANGER_ZONE ) {
+ if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i]))
+ workspace->realloc.Htop = (mat_end[i] - mat_start[i]) ;
+ //fprintf (stderr, "step %d, Reached the water mark for sparse matrix for index %d (%d %d) -- %d \n",
+ // step, i, mat_start[i], mat_end[i],
+ // (int) (system->max_sparse_matrix_entries * DANGER_ZONE));
+
+ if( mat_end[i] > system->N * system->max_sparse_matrix_entries ) {
+ fprintf( stderr, "step%d-matchk failed: i=%d end(i)=%d mat_end=%d\n",
+ step, flag, mat_end[i], system->N * system->max_sparse_matrix_entries);
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+ /* bond list */
+#ifdef __CUDA_TEST__
+ //workspace->realloc.bonds = 1;
+#endif
+ flag = -1;
+ workspace->realloc.num_bonds = 0;
+ for( i = 0; i < n-1; ++i ) {
+ workspace->realloc.num_bonds += MAX((bonds_end [i] - bonds_start[i]) * 2, MIN_BONDS );
+ if( bonds_end[i] >= bonds_start[i+1]-2 ) {
+ workspace->realloc.bonds = 1;
+ //fprintf (stderr, "step: %d, reached the water mark for bonds for atom: %d (%d %d) \n",
+ // step, i, bonds_start [i], bonds_end[i]);
+ if( bonds_end[i] > bonds_start[i+1] )
+ flag = i;
+ }
+ }
+
+ if( flag > -1 ) {
+ fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
+ step, flag, bonds_end[flag], bonds_start[flag+1] );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+ workspace->realloc.num_bonds += MAX((bonds_end [i] - bonds_start[i]) * 2, MIN_BONDS );
+ if( bonds_end[i] >= bonds->num_intrs-2 ) {
+ workspace->realloc.bonds = 1;
+ //fprintf (stderr, "step: %d, reached the water mark for bonds for atom: %d (%d %d) \n",
+ // step, i, bonds_start [i], bonds_end[i]);
+
+ if( bonds_end[i] > bonds->num_intrs ) {
+ fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d bond_end=%d\n",
+ step, flag, bonds_end[i], bonds->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+ //fprintf (stderr, "step:%d Total bonds: %d \n", step, workspace->realloc.num_bonds);
+
+ /* hbonds list */
+ if( workspace->num_H > 0 ) {
+#ifdef __CUDA_TEST__
+ //workspace->realloc.hbonds = 1;
+#endif
+ flag = -1;
+ workspace->realloc.num_hbonds = 0;
+ for( i = 0; i < workspace->num_H-1; ++i ) {
+ workspace->realloc.num_hbonds += MAX( (hbonds_end[i] - hbonds_start[i]) * SAFE_HBONDS, MIN_HBONDS );
+
+ if( (hbonds_end[i] - hbonds_start[i]) >=
+ (hbonds_start[i+1] - hbonds_start[i]) * DANGER_ZONE ) {
+ workspace->realloc.hbonds = 1;
+ //fprintf (stderr, "step: %d, reached the water mark for hbonds for atom: %d (%d %d) \n",
+ // step, i, hbonds_start [i], hbonds_end[i]);
+ if( hbonds_end[i] > hbonds_start[i+1] )
+ flag = i;
+ }
+ }
+
+ if( flag > -1 ) {
+ fprintf( stderr, "step%d-hbondchk failed: i=%d start(i)=%d,end(i)=%d str(i+1)=%d\n",
+ step, flag, hbonds_start[(flag)],hbonds_end[(flag)], hbonds_start[(flag+1)] );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+ workspace->realloc.num_hbonds += MAX( (hbonds_end[i] - hbonds_start[i]) * SAFE_HBONDS, MIN_HBONDS );
+ if( (hbonds_end[i] - hbonds_start[i]) >=
+ (hbonds->num_intrs - hbonds_start[i]) * DANGER_ZONE ) {
+ workspace->realloc.hbonds = 1;
+ //fprintf (stderr, "step: %d, reached the water mark for hbonds for atom: %d (%d %d) \n",
+ // step, i, hbonds_start [i], hbonds_end[i]);
+
+ if( hbonds_end[i] > hbonds->num_intrs ) {
+ fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d hbondend=%d\n",
+ step, flag, hbonds_end[i], hbonds->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+ }
+
+ //fprintf (stderr, "step:%d Total Hbonds: %d \n", step, workspace->realloc.num_hbonds);
+
+ free (bonds_start);
+ free (bonds_end );
+
+ free (hbonds_start );
+ free (hbonds_end );
+
+ free (mat_start );
+ free (mat_end );
+}
+
+
+void Cuda_Threebody_List( reax_system *system, static_storage *workspace, list *thblist, int step )
+{
+ int *thb_start, *thb_end;
+ int i, flag;
+
+ thb_start = (int *) calloc (thblist->n, INT_SIZE);
+ thb_end = (int *) calloc (thblist->n, INT_SIZE );
+
+ copy_host_device (thb_start, thblist->index, thblist->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+ copy_host_device (thb_end, thblist->end_index, thblist->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ /*three_body list*/
+ flag = -1;
+ workspace->realloc.num_3body = 0;
+ for( i = 0; i < thblist->n-1; ++i ){
+ if( (thb_end[i] - thb_start[i]) >= (thb_start[i+1] - thb_start[i])*DANGER_ZONE ) {
+ workspace->realloc.thbody = 1;
+ if( thb_end[i] > thb_end[i+1] || thb_end[i] > thblist->num_intrs ) {
+ flag = i;
+ break;
+ }
+ }
+ }
+
+ if( flag > -1 ) {
+ //fprintf( stderr, "step%d-thbchk failed: i=%d end(i)=%d str(i+1)=%d\n",
+ // step, flag, thb_end[flag], thb_start[flag+1] );
+ fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
+ step, flag-1, thb_start[flag-1], thb_end[flag-1], thblist->num_intrs );
+ fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
+ step, flag, thb_start[flag], thb_end[flag], thblist->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+ if( (thb_end[i]-thb_start[i]) >= (thblist->num_intrs - thb_start[i])*DANGER_ZONE ) {
+ workspace->realloc.thbody = 1;
+
+ if( thb_end[i] > thblist->num_intrs ) {
+ fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
+ step, i-1, thb_start[i-1], thb_end[i-1], thblist->num_intrs );
+ fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
+ step, i, thb_start[i], thb_end[i], thblist->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+ free (thb_start);
+ free (thb_end);
+}
+
+
+GLOBAL void k_Estimate_Sparse_Matrix_Entries ( reax_atom *atoms, control_params *control,
+ simulation_data *data, simulation_box *box, list far_nbrs, int N, int *indices ) {
+
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int Htop;
+ int flag;
+ far_neighbor_data *nbr_pj;
+ reax_atom *atom_i, *atom_j;
+
+ int temp;
+
+ Htop = 0;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ atom_i = &(atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, &far_nbrs);
+ end_i = End_Index(i, &far_nbrs);
+ indices[i] = Htop;
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(atoms[j]);
+
+ //CHANGE ORIGINAL
+ //if (i < j) continue;
+ //CHANGE ORIGINAL
+
+ flag = 0;
+ if((data->step-data->prev_steps) % control->reneighbor == 0) {
+ if( nbr_pj->d <= control->r_cut)
+ flag = 1;
+ else flag = 0;
+ }
+ else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec)) <=
+ SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+
+ if( flag ){
+ ++Htop;
+ }
+ }
+
+ ++Htop;
+
+ // mark the end of j list
+ indices[i] = Htop;
+}
+
+
+GLOBAL void k_Init_Forces( reax_atom *atoms, global_parameters g_params, control_params *control,
+ single_body_parameters *sbp, two_body_parameters *tbp,
+ simulation_data *data, simulation_box *box, static_storage workspace,
+ list far_nbrs, list bonds, list hbonds,
+ int N, int max_sparse_entries, int num_atom_types )
+{
+
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int Htop, btop_i, btop_j, num_bonds, num_hbonds;
+ int ihb, jhb, ihb_top, jhb_top;
+ int flag;
+ real r_ij, r2, self_coef;
+ real dr3gamij_1, dr3gamij_3, Tap;
+ //real val, dif, base;
+ real C12, C34, C56;
+ real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ sparse_matrix *H;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ //LR_lookup_table *t;
+ reax_atom *atom_i, *atom_j;
+ bond_data *ibond, *jbond;
+ bond_order_data *bo_ij, *bo_ji;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ H = &( workspace.H );
+ //CHANGE ORIGINAL
+ //Htop = 0;
+ Htop = i * max_sparse_entries;
+ //CHANGE ORIGINAL
+ num_bonds = 0;
+ num_hbonds = 0;
+ btop_i = btop_j = 0;
+ p_boc1 = g_params.l[0];
+ p_boc2 = g_params.l[1];
+
+ //for( i = 0; i < system->N; ++i )
+ atom_i = &(atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, &far_nbrs);
+ end_i = End_Index(i, &far_nbrs);
+
+ H->start[i] = Htop;
+ H->end[i] = Htop;
+
+ btop_i = End_Index( i, &bonds );
+ sbp_i = &(sbp[type_i]);
+ ihb = ihb_top = -1;
+
+ ihb = sbp_i->p_hbond;
+
+ if( control->hb_cut > 0 && (ihb==1 || ihb == 2))
+ ihb_top = End_Index( workspace.hbond_index[i], &hbonds );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(atoms[j]);
+
+ flag = 0;
+ if((data->step-data->prev_steps) % control->reneighbor == 0) {
+ if( nbr_pj->d <= control->r_cut)
+ flag = 1;
+ else flag = 0;
+ }
+ else if (i > j) {
+ if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+ } else if (i < j) {
+ if((nbr_pj->d=Sq_Distance_on_T3(atom_j->x,atom_i->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+ }
+
+ if( flag ){
+
+ type_j = atoms[j].type;
+ r_ij = nbr_pj->d;
+ sbp_j = &(sbp[type_j]);
+ twbp = &(tbp[ index_tbp (type_i,type_j, num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0;
+
+ /* H matrix entry */
+
+ //CHANGE ORIGINAL
+ //if (i > j) {
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ H->entries[Htop].j = j;
+ H->entries[Htop].val = self_coef * Tap * EV_to_KCALpMOL / dr3gamij_3;
+
+ ++Htop;
+ //}
+ //CHANGE ORIGINAL
+
+ /* hydrogen bond lists */
+ if( control->hb_cut > 0 && (ihb==1 || ihb == 2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ // fprintf( stderr, "%d %d\n", atom1, atom2 );
+ jhb = sbp_j->p_hbond;
+
+ if (ihb == 1 && jhb == 2) {
+ if (i > j) {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = 1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+
+ //Auxilary data structures
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ } else {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = -1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+
+ //Auxilary data structures
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ } else if (ihb == 2 && jhb == 1) {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = 1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+ //TODO
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ }
+
+ /* uncorrected bond orders */
+ if( far_nbrs.select.far_nbr_list[pj].d <= control->nbr_cut ) {
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+
+ if( BO >= control->bo_cut ) {
+ //CHANGE ORIGINAL
+ num_bonds += 1;
+ //CHANGE ORIGINAL
+
+ /****** bonds i-j and j-i ******/
+
+ /* Bond Order page2-3, derivative of total bond order prime */
+ Cln_BOp_s = twbp->p_bo2 * C12 / r2;
+ Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
+ Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
+
+
+ if (i > j)
+ {
+ ibond = &( bonds.select.bond_list[btop_i] );
+ ibond->nbr = j;
+ ibond->d = r_ij;
+ rvec_Copy( ibond->dvec, nbr_pj->dvec );
+ ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
+
+ //ibond->dbond_index = btop_i;
+ //ibond->sym_index = btop_j;
+ ++btop_i;
+
+ bo_ij = &( ibond->bo_data );
+ bo_ij->BO = BO;
+ bo_ij->BO_s = BO_s;
+ bo_ij->BO_pi = BO_pi;
+ bo_ij->BO_pi2 = BO_pi2;
+
+ //Auxilary data structures
+ ibond->scratch = 0;
+ ibond->CdDelta_ij = 0;
+ rvec_MakeZero (ibond->f);
+
+ ibond->l = -1;
+ ibond->CdDelta_jk = 0;
+ ibond->Cdbo_kl = 0;
+ rvec_MakeZero (ibond->i_f);
+ rvec_MakeZero (ibond->k_f);
+
+ rvec_MakeZero (ibond->h_f);
+
+ rvec_MakeZero (ibond->t_f);
+
+ // Only dln_BOp_xx wrt. dr_i is stored here, note that
+ // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
+ rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi2,
+ -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
+
+ // Only dBOp wrt. dr_i is stored here, note that
+ // dBOp/dr_i = -dBOp/dr_j and all others are 0
+ rvec_Scale( bo_ij->dBOp,
+ -(bo_ij->BO_s * Cln_BOp_s +
+ bo_ij->BO_pi * Cln_BOp_pi +
+ bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
+
+ rvec_Add( workspace.dDeltap_self[i], bo_ij->dBOp );
+
+ bo_ij->BO_s -= control->bo_cut;
+ bo_ij->BO -= control->bo_cut;
+ workspace.total_bond_order[i] += bo_ij->BO; //currently total_BOp
+
+ bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
+
+
+ } else if ( i < j )
+ {
+ rvec dln_BOp_s, dln_BOp_pi, dln_BOp_pi2;
+ rvec dBOp;
+
+ btop_j = btop_i;
+
+ jbond = &(bonds.select.bond_list[btop_j]);
+ jbond->nbr = j;
+ jbond->d = r_ij;
+ rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
+ ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
+
+ btop_i ++;
+ //jbond->dbond_index = btop_i;
+ //jbond->sym_index = btop_i;
+
+ bo_ji = &( jbond->bo_data );
+ bo_ji->BO = BO;
+ bo_ji->BO_s = BO_s;
+ bo_ji->BO_pi = BO_pi;
+ bo_ji->BO_pi2 = BO_pi2;
+
+ //Auxilary data structures
+ jbond->scratch = 0;
+ jbond->CdDelta_ij = 0;
+ rvec_MakeZero (jbond->f);
+
+ jbond->l = -1;
+ jbond->CdDelta_jk = 0;
+ jbond->Cdbo_kl = 0;
+ rvec_MakeZero (jbond->i_f);
+ rvec_MakeZero (jbond->k_f);
+
+ rvec_MakeZero (jbond->h_f);
+
+ rvec_MakeZero (jbond->t_f);
+
+ // Only dln_BOp_xx wrt. dr_i is stored here, note that
+ // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
+ rvec_Scale(dln_BOp_s,-BO_s*Cln_BOp_s,nbr_pj->dvec);
+ rvec_Scale(dln_BOp_pi,-BO_pi*Cln_BOp_pi,nbr_pj->dvec);
+ rvec_Scale(dln_BOp_pi2,
+ -BO_pi2*Cln_BOp_pi2,nbr_pj->dvec);
+
+ rvec_Scale(bo_ji->dln_BOp_s, -1., dln_BOp_s);
+ rvec_Scale(bo_ji->dln_BOp_pi, -1., dln_BOp_pi );
+ rvec_Scale(bo_ji->dln_BOp_pi2, -1., dln_BOp_pi2 );
+
+ // Only dBOp wrt. dr_i is stored here, note that
+ // dBOp/dr_i = -dBOp/dr_j and all others are 0
+ rvec_Scale( dBOp,
+ -(BO_s * Cln_BOp_s +
+ BO_pi * Cln_BOp_pi +
+ BO_pi2 * Cln_BOp_pi2), nbr_pj->dvec );
+ rvec_Scale( bo_ji->dBOp, -1., dBOp );
+
+ rvec_Add( workspace.dDeltap_self[i] , bo_ji->dBOp );
+
+ bo_ji->BO_s -= control->bo_cut;
+ bo_ji->BO -= control->bo_cut;
+ workspace.total_bond_order[i] += bo_ji->BO; //currently total_BOp
+
+ bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
+
+ }
+ }
+ }
+ }
+ }
+
+ H->entries[Htop].j = i;
+ H->entries[Htop].val = sbp[type_i].eta;
+ ++Htop;
+
+ H->end[i] = Htop;
+
+ Set_End_Index( i, btop_i, &bonds );
+ if( ihb == 1 || ihb == 2)
+ Set_End_Index( workspace.hbond_index[i], ihb_top, &hbonds );
+
+ //fprintf( stderr, "%d bonds start: %d, end: %d\n",
+ // i, Start_Index( i, bonds ), End_Index( i, bonds ) );
+ //}
+
+ // mark the end of j list
+ //H->start[i] = Htop;
+ /* validate lists - decide if reallocation is required! */
+ //Validate_Lists( workspace, lists,
+ // data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
+}
+
+
+GLOBAL void k_Init_Forces_Tab ( reax_atom *atoms, global_parameters g_params, control_params *control,
+ single_body_parameters *sbp, two_body_parameters *tbp,
+ simulation_data *data, simulation_box *box, static_storage workspace,
+ list far_nbrs, list bonds, list hbonds,
+ int N, int max_sparse_entries, int num_atom_types,
+ LR_lookup_table *d_LR)
+{
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int Htop, btop_i, btop_j, num_bonds, num_hbonds;
+ int tmin, tmax, r;
+ int ihb, jhb, ihb_top, jhb_top;
+ int flag;
+ real r_ij, r2, self_coef;
+ real val, dif, base;
+ real C12, C34, C56;
+ real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ sparse_matrix *H;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ LR_lookup_table *t;
+ reax_atom *atom_i, *atom_j;
+ bond_data *ibond, *jbond;
+ bond_order_data *bo_ij, *bo_ji;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ H = &(workspace.H);
+ //CHANGE ORIGINAL
+ Htop = i * max_sparse_entries;
+ //CHANGE ORIGINAL
+ num_bonds = 0;
+ num_hbonds = 0;
+ btop_i = btop_j = 0;
+ p_boc1 = g_params.l[0];
+ p_boc2 = g_params.l[1];
+
+ //for( i = 0; i < system->N; ++i )
+ atom_i = &(atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, &far_nbrs);
+ end_i = End_Index(i, &far_nbrs);
+ H->start[i] = Htop;
+ H->end[i] = Htop;
+ btop_i = End_Index( i, &bonds );
+ sbp_i = &(sbp[type_i]);
+ ihb = ihb_top = -1;
+
+ ihb = sbp_i->p_hbond;
+
+ if( control->hb_cut > 0 && (ihb==1 || ihb == 2))
+ ihb_top = End_Index( workspace.hbond_index[i], &hbonds );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(atoms[j]);
+
+ flag = 0;
+ if((data->step-data->prev_steps) % control->reneighbor == 0) {
+ if(nbr_pj->d <= control->r_cut)
+ flag = 1;
+ else flag = 0;
+ }
+ else if (i > j) {
+ if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+ }
+ else if ( i < j) {
+ if((nbr_pj->d=Sq_Distance_on_T3(atom_j->x,atom_i->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+ }
+
+ if( flag ){
+ type_j = atoms[j].type;
+ r_ij = nbr_pj->d;
+ sbp_j = &(sbp[type_j]);
+ twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( d_LR[ index_lr (tmin, tmax, num_atom_types) ]);
+
+ /* cubic spline interpolation */
+ //CHANGE ORIGINAL
+ //if (i > j) {
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+ val = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
+ t->ele[r].a;
+ val *= EV_to_KCALpMOL / C_ele;
+
+ H->entries[Htop].j = j;
+ H->entries[Htop].val = self_coef * val;
+ //H->j [Htop] = j;
+ //H->val [Htop] = self_coef * val;
+ ++Htop;
+ //}
+ //CHANGE ORIGINAL
+
+ /* hydrogen bond lists */
+ if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ // fprintf( stderr, "%d %d\n", atom1, atom2 );
+ jhb = sbp_j->p_hbond;
+
+ if ( ihb == 1 && jhb == 2 ) {
+ if (i > j) {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = 1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+
+ //Auxilary data structures
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ } else {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = -1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+
+ //Auxilary data structures
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ } else if (ihb == 2 && jhb == 1) {
+ hbonds.select.hbond_list[ihb_top].nbr = j;
+ hbonds.select.hbond_list[ihb_top].scl = 1;
+ hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
+
+ //Auxilary data structures
+ rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
+ hbonds.select.hbond_list[ihb_top].sym_index= -1;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ }
+
+ /* uncorrected bond orders */
+ if( far_nbrs.select.far_nbr_list[pj].d <= control->nbr_cut ) {
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+ if( BO >= control->bo_cut ) {
+
+ //CHANGE ORIGINAL
+ num_bonds += 1;
+ //CHANGE ORIGINAL
+
+ /****** bonds i-j and j-i ******/
+ if ( i > j )
+ {
+ ibond = &( bonds.select.bond_list[btop_i] );
+ ibond->nbr = j;
+ ibond->d = r_ij;
+
+ rvec_Copy( ibond->dvec, nbr_pj->dvec );
+ ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
+
+ //ibond->dbond_index = btop_i;
+ //ibond->sym_index = btop_j;
+
+ ++btop_i;
+
+ bo_ij = &( ibond->bo_data );
+ bo_ij->BO = BO;
+ bo_ij->BO_s = BO_s;
+ bo_ij->BO_pi = BO_pi;
+ bo_ij->BO_pi2 = BO_pi2;
+
+ //Auxilary data strucutres to resolve dependencies
+ ibond->scratch = 0;
+ ibond->CdDelta_ij = 0;
+ rvec_MakeZero (ibond->f);
+
+ ibond->l = -1;
+ ibond->CdDelta_jk = 0;
+ ibond->Cdbo_kl = 0;
+ rvec_MakeZero (ibond->i_f);
+ rvec_MakeZero (ibond->k_f);
+
+ rvec_MakeZero (ibond->h_f);
+
+ rvec_MakeZero (ibond->t_f);
+
+ /* Bond Order page2-3, derivative of total bond order prime */
+ Cln_BOp_s = twbp->p_bo2 * C12 / r2;
+ Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
+ Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
+
+ /* Only dln_BOp_xx wrt. dr_i is stored here, note that
+ dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
+ rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi2,
+ -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
+
+ /* Only dBOp wrt. dr_i is stored here, note that
+ dBOp/dr_i = -dBOp/dr_j and all others are 0 */
+ rvec_Scale( bo_ij->dBOp,
+ -(bo_ij->BO_s * Cln_BOp_s +
+ bo_ij->BO_pi * Cln_BOp_pi +
+ bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
+
+ rvec_Add( workspace.dDeltap_self[i], bo_ij->dBOp );
+
+ bo_ij->BO_s -= control->bo_cut;
+ bo_ij->BO -= control->bo_cut;
+
+ workspace.total_bond_order[i] += bo_ij->BO; //currently total_BOp
+
+ bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
+ }
+ else {
+ rvec dln_BOp_s, dln_BOp_pi, dln_BOp_pi2;
+ rvec dBOp;
+
+ btop_j = btop_i;
+
+ jbond = &( bonds.select.bond_list[btop_j] );
+ jbond->nbr = j;
+ jbond->d = r_ij;
+
+ rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
+ ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
+
+ //jbond->dbond_index = btop_i;
+ //jbond->sym_index = btop_i;
+
+ ++btop_i;
+
+ bo_ji = &( jbond->bo_data );
+
+ bo_ji->BO = BO;
+ bo_ji->BO_s = BO_s;
+ bo_ji->BO_pi = BO_pi;
+ bo_ji->BO_pi2 = BO_pi2;
+
+ // Auxilary data structures to resolve dependencies
+ jbond->scratch = 0;
+ jbond->CdDelta_ij = 0;
+ rvec_MakeZero (jbond->f);
+
+ jbond->l = -1;
+ jbond->CdDelta_jk = 0;
+ jbond->Cdbo_kl = 0;
+ rvec_MakeZero (jbond->i_f);
+ rvec_MakeZero (jbond->k_f);
+
+ rvec_MakeZero (jbond->h_f);
+
+ rvec_MakeZero (jbond->t_f);
+
+ // Bond Order page2-3, derivative of total bond order prime
+ Cln_BOp_s = twbp->p_bo2 * C12 / r2;
+ Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
+ Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
+
+ // Only dln_BOp_xx wrt. dr_i is stored here, note that
+ // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
+
+ rvec_Scale(dln_BOp_s,-BO_s*Cln_BOp_s,nbr_pj->dvec);
+ rvec_Scale(dln_BOp_pi,-BO_pi*Cln_BOp_pi,nbr_pj->dvec);
+ rvec_Scale(dln_BOp_pi2, -BO_pi2*Cln_BOp_pi2,nbr_pj->dvec);
+
+ rvec_Scale(bo_ji->dln_BOp_s, -1., dln_BOp_s);
+ rvec_Scale(bo_ji->dln_BOp_pi, -1., dln_BOp_pi );
+ rvec_Scale(bo_ji->dln_BOp_pi2, -1., dln_BOp_pi2 );
+
+ // Only dBOp wrt. dr_i is stored here, note that
+ // dBOp/dr_i = -dBOp/dr_j and all others are 0
+ //CHANGE ORIGINAL
+ rvec_Scale( dBOp,
+ -(BO_s * Cln_BOp_s +
+ BO_pi * Cln_BOp_pi +
+ BO_pi2 * Cln_BOp_pi2), nbr_pj->dvec);
+ rvec_Scale( bo_ji->dBOp, -1., dBOp);
+ //CHANGE ORIGINAL
+
+ rvec_Add( workspace.dDeltap_self[i], bo_ji->dBOp );
+
+ bo_ji->BO_s -= control->bo_cut;
+ bo_ji->BO -= control->bo_cut;
+
+ workspace.total_bond_order[i] += bo_ji->BO; //currently total_BOp
+
+ bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
+ }
+ }
+ }
+ }
+ }
+
+ H->entries[Htop].j = i;
+ H->entries[Htop].val = sbp[type_i].eta;
+
+ //H->j [Htop] = i;
+ //H->val [Htop] = sbp[type_i].eta;
+
+ ++Htop;
+
+ H->end[i] = Htop;
+ Set_End_Index( i, btop_i, &bonds );
+ if( ihb == 1 || ihb == 2)
+ Set_End_Index( workspace.hbond_index[i], ihb_top, &hbonds );
+}
+
+
+GLOBAL void k_fix_sym_dbond_indices (list pbonds, int N)
+{
+ int i, nbr;
+ bond_data *ibond, *jbond;
+ int atom_j;
+
+ list *bonds = &pbonds;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
+ {
+ ibond = &( bonds->select.bond_list [j] );
+ nbr = ibond->nbr;
+
+ for (int k = Start_Index (nbr, bonds); k < End_Index (nbr, bonds); k ++)
+ {
+ jbond = &( bonds->select.bond_list[ k ] );
+ atom_j = jbond->nbr;
+
+ if ( (atom_j == i) )
+ {
+ if (i > nbr) {
+ ibond->dbond_index = j;
+ jbond->dbond_index = j;
+
+ ibond->sym_index = k;
+ jbond->sym_index = j;
+ }
+ }
+ }
+ }
+}
+
+
+GLOBAL void k_fix_sym_hbond_indices (static_storage p_workspace, list hbonds, int N)
+{
+ static_storage *workspace = &p_workspace;
+ hbond_data *ihbond, *jhbond;
+ int nbr;
+
+ //int i = (blockIdx.x * blockDim.x + threadIdx.x) >> 4;
+ int i = (blockIdx.x);
+ int start = Start_Index (workspace->hbond_index[i], &hbonds);
+ int end = End_Index (workspace->hbond_index[i], &hbonds);
+ //int j = start + threadIdx.x;
+ //int j = start + (threadIdx.x % 16);
+
+ //for (int j = Start_Index (workspace->hbond_index[i], &hbonds);
+ // j < End_Index (workspace->hbond_index[i], &hbonds); j++)
+ int j = start + threadIdx.x;
+ while (j < end)
+ //for (int j = start; j < end; j++)
+ {
+ ihbond = &( hbonds.select.hbond_list [j] );
+ nbr = ihbond->nbr;
+
+ int nbrstart = Start_Index (workspace->hbond_index[nbr], &hbonds);
+ int nbrend = End_Index (workspace->hbond_index[nbr], &hbonds);
+
+ for (int k = nbrstart; k < nbrend; k++)
+ //k = nbrstart + threadIdx.x;
+ //while (k < nbrend)
+ {
+ jhbond = &( hbonds.select.hbond_list [k] );
+
+ if (jhbond->nbr == i){
+ ihbond->sym_index = k;
+ jhbond->sym_index = j;
+ break;
+ }
+
+ //k += blockDim.x;
+ }
+
+ j += 32;
+ }
+}
+
+
+GLOBAL void k_New_fix_sym_hbond_indices (static_storage p_workspace, list hbonds, int N )
+{
+
+ static_storage *workspace = &p_workspace;
+ hbond_data *ihbond, *jhbond;
+
+ int __THREADS_PER_ATOM__ = HBONDS_SYM_THREADS_PER_ATOM;
+ int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
+ int warp_id = thread_id / __THREADS_PER_ATOM__;
+ int lane_id = thread_id & (__THREADS_PER_ATOM__ - 1);
+ int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
+
+ if (warp_id >= N) return;
+
+ int i = warp_id;
+ int nbr;
+ int k;
+ int start = Start_Index (workspace->hbond_index[i], &hbonds);
+ int end = End_Index (workspace->hbond_index[i], &hbonds);
+ int j = start + lane_id;
+ //for (int j = start; j < end; j++)
+ while (j < end)
+ {
+ ihbond = &( hbonds.select.hbond_list [j] );
+ nbr = ihbond->nbr;
+
+ int nbrstart = Start_Index (workspace->hbond_index[nbr], &hbonds);
+ int nbrend = End_Index (workspace->hbond_index[nbr], &hbonds);
+
+ //k = nbrstart + lane_id;
+ //if (lane_id == 0) found [my_bucket] = 0;
+ //while (k < nbrend)
+ for (k = nbrstart; k < nbrend; k++)
+ {
+ jhbond = &( hbonds.select.hbond_list [k] );
+
+ if (jhbond->nbr == i){
+ ihbond->sym_index = k;
+ jhbond->sym_index = j;
+ break;
+ }
+ }
+
+ j += __THREADS_PER_ATOM__;
+ }
+}
+
+
+GLOBAL void k_Estimate_Storage_Sizes(reax_atom *atoms,
+ int N, single_body_parameters *sbp,
+ two_body_parameters *tbp,
+ global_parameters gp,
+ control_params *control,
+ list far_nbrs,
+ int num_atom_types, int *results)
+{
+ int *Htop = &results[0];
+ int *num_3body = &results[1];
+ int *hb_top = &results [ 2 ];
+ int *bond_top = &results [ 2 + N ];
+
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int ihb, jhb;
+ real r_ij, r2;
+ real C12, C34, C56;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ reax_atom *atom_i, *atom_j;
+
+ p_boc1 = gp.l[0];
+ p_boc2 = gp.l[1];
+
+ //for( i = 0; i < N; ++i ) {
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= N ) return ;
+
+ atom_i = &(atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, &far_nbrs);
+ end_i = End_Index(i, &far_nbrs);
+ sbp_i = &(sbp[type_i]);
+ ihb = sbp_i->p_hbond;
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &( atoms[j] );
+ type_j = atom_j->type;
+ sbp_j = &( sbp[type_j] );
+ twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ] );
+
+
+ if( nbr_pj->d <= control->r_cut ) {
+ //++(*Htop);
+ atomicAdd(Htop, 1);
+
+ /* hydrogen bond lists */
+ //TODO - CHANGE ORIGINAL
+ if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ jhb = sbp_j->p_hbond;
+ if( ihb == 1 && jhb == 2 )
+ //++hb_top[i];
+ atomicAdd(&hb_top[i], 1);
+ else if( ihb == 2 && jhb == 1 )
+ //++hb_top[j];
+ //atomicAdd(&hb_top[j], 1);
+ atomicAdd(&hb_top[i], 1);
+ }
+ //TODO -- CHANGE ORIGINAL
+
+ //CHANGE ORIGINAL
+ if (i < j) continue;
+ //CHANGE ORIGINAL
+
+
+ /* uncorrected bond orders */
+ if( nbr_pj->d <= control->nbr_cut ) {
+ r_ij = nbr_pj->d;
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+ if( BO >= control->bo_cut ) {
+ //++bond_top[i];
+ //++bond_top[j];
+ atomicAdd(&bond_top[i], 1);
+ atomicAdd(&bond_top[j], 1);
+ }
+ }
+ }
+ }
+ //}
+}
+
+
+void Cuda_Estimate_Storage_Sizes (reax_system *system, control_params *control, int *output)
+{
+ int *Htop, *num_3body, input_size;
+ int *hb_top, *bond_top;
+ int *input = (int *) scratch;
+ int max_3body = 0;
+
+ Htop = 0;
+ num_3body = 0;
+ input_size = INT_SIZE * (2 * system->N + 1 + 1);
+
+ //cuda_malloc ((void **) &input, input_size, 1, __LINE__);
+ cuda_memset (input, 0, input_size, RES_SCRATCH );
+
+ k_Estimate_Storage_Sizes <<<BLOCKS_POW_2, BLOCK_SIZE>>>
+ (system->d_atoms, system->N, system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ system->reaxprm.d_gp, (control_params *)control->d_control, *(dev_lists + FAR_NBRS),
+ system->reaxprm.num_atom_types, input);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ copy_host_device (output, input, input_size, cudaMemcpyDeviceToHost, __LINE__ );
+
+ Htop = &output[0];
+ num_3body = &output[1];
+ hb_top = &output[ 2 ];
+ bond_top = &output[ 2 + system->N ];
+
+ *Htop += system->N;
+ *Htop *= SAFE_ZONE;
+
+ for( int i = 0; i < system->N; ++i ) {
+ hb_top[i] = MAX( hb_top[i] * SAFE_HBONDS, MIN_HBONDS );
+
+ if (max_3body <= SQR (bond_top[i]))
+ max_3body = SQR (bond_top[i]);
+
+ *num_3body += SQR(bond_top[i]);
+ bond_top[i] = MAX( bond_top[i] * 2, MIN_BONDS );
+ }
+
+ *num_3body = max_3body * SAFE_ZONE;
+}
+
+
+void Cuda_Compute_Forces( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list** lists, output_controls *out_control )
+{
+ real t_start, t_elapsed;
+ real t_1, t_2;
+ int *indices;
+ int *Htop;
+ int max_sparse_entries = 0;
+ list *far_nbrs = dev_lists + FAR_NBRS;
+ int hblocks;
+
+ t_start = Get_Time ();
+ if ( !control->tabulate ) {
+ k_Init_Forces <<<BLOCKS, BLOCK_SIZE>>>
+ (system->d_atoms, system->reaxprm.d_gp, (control_params *)control->d_control,
+ system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box, *dev_workspace,
+ *(dev_lists + FAR_NBRS), *(dev_lists + BONDS), *(dev_lists + HBONDS),
+ system->N, system->max_sparse_matrix_entries, system->reaxprm.num_atom_types );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ }
+ else
+ {
+ k_Init_Forces_Tab <<< BLOCKS, BLOCK_SIZE >>>
+ ( system->d_atoms, system->reaxprm.d_gp, (control_params *)control->d_control,
+ system->reaxprm.d_sbp, system->reaxprm.d_tbp,
+ (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box, *dev_workspace,
+ *(dev_lists + FAR_NBRS), *(dev_lists + BONDS), *(dev_lists + HBONDS),
+ system->N, system->max_sparse_matrix_entries, system->reaxprm.num_atom_types,
+ d_LR );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ }
+
+ /*This is for bonds processing to fix dbond and sym_indexes */
+ t_1 = Get_Time ();
+ k_fix_sym_dbond_indices <<<BLOCKS, BLOCK_SIZE>>> (*(dev_lists + BONDS), system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ t_2 = Get_Timing_Info ( t_1 );
+
+ //FIX -1 HYDROGEN BOND fix for cases where there are no hbonds.
+ if ((control->hb_cut > 0) && (dev_workspace->num_H > 0))
+ {
+
+ hblocks = (system->N * HBONDS_SYM_THREADS_PER_ATOM / HBONDS_SYM_BLOCK_SIZE) +
+ ((system->N * HBONDS_SYM_THREADS_PER_ATOM % HBONDS_SYM_BLOCK_SIZE) == 0 ? 0 : 1);
+ t_1 = Get_Time ();
+ /*
+ int bs = system->N;
+ int ss = 32;
+ fix_sym_hbond_indices <<<bs, ss>>> (*dev_workspace, *(dev_lists + HBONDS), system->N);
+ */
+ k_New_fix_sym_hbond_indices <<<hblocks, HBONDS_SYM_BLOCK_SIZE>>> (*dev_workspace, *(dev_lists + HBONDS), system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ }
+ t_2 = Get_Timing_Info ( t_1 );
+
+ t_elapsed = Get_Timing_Info (t_start);
+ d_timing.init_forces+= t_elapsed;
+
+ Cuda_Validate_Lists( system, dev_workspace, &dev_lists, data->step, system->N,
+ system->num_bonds, system->num_hbonds );
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Done with Cuda List Validation \n");
+#endif
+
+ //Bonded Force Calculations here.
+ t_start = Get_Time ();
+ Cuda_Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
+ t_elapsed = Get_Timing_Info (t_start);
+ d_timing.bonded += t_elapsed;
+
+ //Compute the Non Bonded Forces here.
+ t_start = Get_Time ();
+ Cuda_Compute_NonBonded_Forces( system, control, data, workspace, lists, out_control );
+ t_elapsed = Get_Timing_Info (t_start);
+ d_timing.nonb += t_elapsed;
+
+ //Compute Total Forces here
+ Cuda_Compute_Total_Force<<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, (simulation_data *)data->d_simulation_data, *dev_workspace,
+ *(dev_lists + BONDS), control->ensemble, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_Compute_Total_Force_PostProcess<<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, (simulation_data *)data->d_simulation_data, *dev_workspace,
+ *(dev_lists + BONDS), control->ensemble, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
+
+
+int validate_device (reax_system *system, simulation_data *data, static_storage *workspace, list **lists )
+{
+ int retval = FALSE;
+
+#ifdef __BUILD_DEBUG__
+
+ retval |= validate_neighbors (system, lists);
+ retval |= validate_sym_dbond_indices (system, workspace, lists);
+ retval |= validate_bonds (system, workspace, lists);
+ retval |= validate_sparse_matrix (system, workspace);
+ retval |= validate_three_bodies (system, workspace, lists );
+ retval |= validate_hbonds (system, workspace, lists);
+ retval |= validate_workspace (system, workspace, lists);
+ retval |= validate_data (system, data);
+ retval |= validate_atoms (system, lists);
+ //analyze_hbonds (system, workspace, lists);
+
+ if (!retval) {
+ fprintf (stderr, "Results *DOES NOT* mattch between device and host \n");
+ }
+#endif
+
+ return retval;
+}
diff --git a/PuReMD-GPU/src/cuda_forces.h b/PuReMD-GPU/src/cuda_forces.h
new file mode 100644
index 0000000000000000000000000000000000000000..b017e63ebeee45c03d8926a79f1a9a0dd7a4771d
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_forces.h
@@ -0,0 +1,48 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_FORCES_H_
+#define __CUDA_FORCES_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void k_Estimate_Sparse_Matrix_Entries ( reax_atom *, control_params *,
+ simulation_data *, simulation_box *, list, int, int * );
+
+void Cuda_Compute_Forces( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls* );
+
+void Cuda_Estimate_Storage_Sizes (reax_system *, control_params *, int *);
+
+void Cuda_Threebody_List( reax_system *, static_storage *, list *, int );
+
+int validate_device (reax_system *, simulation_data *, static_storage *, list **);
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/four_body_interactions.cu b/PuReMD-GPU/src/cuda_four_body_interactions.cu
similarity index 53%
rename from PuReMD-GPU/src/four_body_interactions.cu
rename to PuReMD-GPU/src/cuda_four_body_interactions.cu
index d7bf757eff65253989cfe58d1ac4dfabd63d602a..60d9973482ddd61a34d874301f1ed360443625b9 100644
--- a/PuReMD-GPU/src/four_body_interactions.cu
+++ b/PuReMD-GPU/src/cuda_four_body_interactions.cu
@@ -18,20 +18,19 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "four_body_interactions.h"
-#include "bond_orders.h"
+#include "cuda_four_body_interactions.h"
+
#include "box.h"
+#include "index_utils.h"
#include "list.h"
-#include "lookup.h"
#include "vector.h"
-#include "math.h"
-#include "index_utils.h"
#include "cuda_helpers.h"
#define MIN_SINE 1e-10
-HOST_DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_jk,
+
+DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_jk,
rvec dvec_kl, real r_kl, rvec dvec_li, real r_li,
three_body_interaction_data *p_ijk,
three_body_interaction_data *p_jkl,
@@ -72,7 +71,6 @@ HOST_DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_
hnhd = r_ij * r_kl * cos_ijk * sin_jkl;
hnhe = r_ij * r_kl * sin_ijk * cos_jkl;
-
poem = 2.0 * r_ij * r_kl * sin_ijk * sin_jkl;
if( poem < 1e-20 ) poem = 1e-20;
@@ -81,9 +79,14 @@ HOST_DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_
r_jk * r_kl * cos_jkl );
arg = tel / poem;
- if( arg > 1.0 ) arg = 1.0;
- if( arg < -1.0 ) arg = -1.0;
-
+ if( arg > 1.0 )
+ {
+ arg = 1.0;
+ }
+ if( arg < -1.0 )
+ {
+ arg = -1.0;
+ }
/*fprintf( out_control->etor,
"%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
@@ -111,10 +114,22 @@ HOST_DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_
-p_jkl->dcos_dk[1]/sin_jkl,
-p_jkl->dcos_dk[2]/sin_jkl );*/
- if( sin_ijk >= 0 && sin_ijk <= MIN_SINE ) sin_ijk = MIN_SINE;
- else if( sin_ijk <= 0 && sin_ijk >= -MIN_SINE ) sin_ijk = -MIN_SINE;
- if( sin_jkl >= 0 && sin_jkl <= MIN_SINE ) sin_jkl = MIN_SINE;
- else if( sin_jkl <= 0 && sin_jkl >= -MIN_SINE ) sin_jkl = -MIN_SINE;
+ if( sin_ijk >= 0 && sin_ijk <= MIN_SINE )
+ {
+ sin_ijk = MIN_SINE;
+ }
+ else if( sin_ijk <= 0 && sin_ijk >= -MIN_SINE )
+ {
+ sin_ijk = -MIN_SINE;
+ }
+ if( sin_jkl >= 0 && sin_jkl <= MIN_SINE )
+ {
+ sin_jkl = MIN_SINE;
+ }
+ else if( sin_jkl <= 0 && sin_jkl >= -MIN_SINE )
+ {
+ sin_jkl = -MIN_SINE;
+ }
// dcos_omega_di
rvec_ScaledSum( dcos_omega_di, (htra-arg*hnra)/r_ij, dvec_ij, -1., dvec_li );
@@ -145,532 +160,7 @@ HOST_DEVICE real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_
}
-
-
-
-void Four_Body_Interactions( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, k, l, pi, pj, pk, pl, pij, plk;
- int type_i, type_j, type_k, type_l;
- int start_j, end_j, start_k, end_k;
- int start_pj, end_pj, start_pk, end_pk;
- int num_frb_intrs = 0;
-
- real Delta_j, Delta_k;
- real r_ij, r_jk, r_kl, r_li;
- real BOA_ij, BOA_jk, BOA_kl;
-
- real exp_tor2_ij, exp_tor2_jk, exp_tor2_kl;
- real exp_tor1, exp_tor3_DjDk, exp_tor4_DjDk, exp_tor34_inv;
- real exp_cot2_jk, exp_cot2_ij, exp_cot2_kl;
- real fn10, f11_DjDk, dfn11, fn12;
-
- real theta_ijk, theta_jkl;
- real sin_ijk, sin_jkl;
- real cos_ijk, cos_jkl;
- real tan_ijk_i, tan_jkl_i;
-
- real omega, cos_omega, cos2omega, cos3omega;
- rvec dcos_omega_di, dcos_omega_dj, dcos_omega_dk, dcos_omega_dl;
-
- real CV, cmn, CEtors1, CEtors2, CEtors3, CEtors4;
- real CEtors5, CEtors6, CEtors7, CEtors8, CEtors9;
- real Cconj, CEconj1, CEconj2, CEconj3;
- real CEconj4, CEconj5, CEconj6;
-
- real e_tor, e_con;
- rvec dvec_li;
- rvec force, ext_press;
- ivec rel_box_jl;
- // rtensor total_rtensor, temp_rtensor;
-
- four_body_header *fbh;
- four_body_parameters *fbp;
- bond_data *pbond_ij, *pbond_jk, *pbond_kl;
- bond_order_data *bo_ij, *bo_jk, *bo_kl;
- three_body_interaction_data *p_ijk, *p_jkl;
-
- real p_tor2 = system->reaxprm.gp.l[23];
- real p_tor3 = system->reaxprm.gp.l[24];
- real p_tor4 = system->reaxprm.gp.l[25];
- real p_cot2 = system->reaxprm.gp.l[27];
-
- list *bonds = (*lists) + BONDS;
- list *thb_intrs = (*lists) + THREE_BODIES;
-
-
- for( j = 0; j < system->N; ++j ) {
- type_j = system->atoms[j].type;
- Delta_j = workspace->Delta_boc[j];
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
-
-
- for( pk = start_j; pk < end_j; ++pk ) {
- pbond_jk = &( bonds->select.bond_list[pk] );
- k = pbond_jk->nbr;
- bo_jk = &( pbond_jk->bo_data );
- BOA_jk = bo_jk->BO - control->thb_cut;
-
- /* see if there are any 3-body interactions involving j&k
- where j is the central atom. Otherwise there is no point in
- trying to form a 4-body interaction out of this neighborhood */
- if( j < k && bo_jk->BO > control->thb_cut/*0*/ &&
- Num_Entries(pk, thb_intrs) ) {
- start_k = Start_Index(k, bonds);
- end_k = End_Index(k, bonds);
- pj = pbond_jk->sym_index; // pj points to j on k's list
-
- /* do the same check as above: are there any 3-body interactions
- involving k&j where k is the central atom */
- if( Num_Entries(pj, thb_intrs) ) {
- type_k = system->atoms[k].type;
- Delta_k = workspace->Delta_boc[k];
- r_jk = pbond_jk->d;
-
- start_pk = Start_Index(pk, thb_intrs );
- end_pk = End_Index(pk, thb_intrs );
- start_pj = Start_Index(pj, thb_intrs );
- end_pj = End_Index(pj, thb_intrs );
-
- exp_tor2_jk = EXP( -p_tor2 * BOA_jk );
- exp_cot2_jk = EXP( -p_cot2 * SQR(BOA_jk - 1.5) );
- exp_tor3_DjDk = EXP( -p_tor3 * (Delta_j + Delta_k) );
- exp_tor4_DjDk = EXP( p_tor4 * (Delta_j + Delta_k) );
- exp_tor34_inv = 1.0 / (1.0 + exp_tor3_DjDk + exp_tor4_DjDk);
- f11_DjDk = (2.0 + exp_tor3_DjDk) * exp_tor34_inv;
-
-
- /* pick i up from j-k interaction where j is the centre atom */
- for( pi = start_pk; pi < end_pk; ++pi ) {
- p_ijk = &( thb_intrs->select.three_body_list[pi] );
- pij = p_ijk->pthb; // pij is pointer to i on j's bond_list
- pbond_ij = &( bonds->select.bond_list[pij] );
- bo_ij = &( pbond_ij->bo_data );
-
-
- if( bo_ij->BO > control->thb_cut/*0*/ ) {
- i = p_ijk->thb;
- type_i = system->atoms[i].type;
- r_ij = pbond_ij->d;
- BOA_ij = bo_ij->BO - control->thb_cut;
-
- theta_ijk = p_ijk->theta;
- sin_ijk = SIN( theta_ijk );
- cos_ijk = COS( theta_ijk );
- //tan_ijk_i = 1. / TAN( theta_ijk );
- if( sin_ijk >= 0 && sin_ijk <= MIN_SINE )
- tan_ijk_i = cos_ijk / MIN_SINE;
- else if( sin_ijk <= 0 && sin_ijk >= -MIN_SINE )
- tan_ijk_i = cos_ijk / -MIN_SINE;
- else tan_ijk_i = cos_ijk / sin_ijk;
-
- exp_tor2_ij = EXP( -p_tor2 * BOA_ij );
- exp_cot2_ij = EXP( -p_cot2 * SQR(BOA_ij -1.5) );
-
- /* pick l up from j-k intr. where k is the centre */
- for( pl = start_pj; pl < end_pj; ++pl ) {
- p_jkl = &( thb_intrs->select.three_body_list[pl] );
- l = p_jkl->thb;
- plk = p_jkl->pthb; //pointer to l on k's bond_list!
- pbond_kl = &( bonds->select.bond_list[plk] );
- bo_kl = &( pbond_kl->bo_data );
- type_l = system->atoms[l].type;
- fbh = &(system->reaxprm.fbp[ index_fbp (type_i,type_j,type_k,type_l,&system->reaxprm ) ]);
- fbp = &(system->reaxprm.fbp[ index_fbp (type_i,type_j,type_k,type_l,&system->reaxprm )].prm[0]);
-
- if( i != l && fbh->cnt && bo_kl->BO > control->thb_cut/*0*/ &&
- bo_ij->BO * bo_jk->BO * bo_kl->BO > control->thb_cut/*0*/ ){
- ++num_frb_intrs;
- r_kl = pbond_kl->d;
- BOA_kl = bo_kl->BO - control->thb_cut;
-
- theta_jkl = p_jkl->theta;
- sin_jkl = SIN( theta_jkl );
- cos_jkl = COS( theta_jkl );
- //tan_jkl_i = 1. / TAN( theta_jkl );
- if( sin_jkl >= 0 && sin_jkl <= MIN_SINE )
- tan_jkl_i = cos_jkl / MIN_SINE;
- else if( sin_jkl <= 0 && sin_jkl >= -MIN_SINE )
- tan_jkl_i = cos_jkl / -MIN_SINE;
- else tan_jkl_i = cos_jkl /sin_jkl;
-
- Sq_Distance_on_T3( system->atoms[l].x, system->atoms[i].x,
- &(system->box), dvec_li );
- r_li = rvec_Norm( dvec_li );
-
-
- /* omega and its derivative */
- //cos_omega=Calculate_Omega(pbond_ij->dvec,r_ij,pbond_jk->dvec,
- omega = Calculate_Omega(pbond_ij->dvec, r_ij, pbond_jk->dvec,
- r_jk, pbond_kl->dvec, r_kl,
- dvec_li, r_li, p_ijk, p_jkl,
- dcos_omega_di, dcos_omega_dj,
- dcos_omega_dk, dcos_omega_dl,
- out_control);
- cos_omega = COS( omega );
- cos2omega = COS( 2. * omega );
- cos3omega = COS( 3. * omega );
- /* end omega calculations */
-
- /* torsion energy */
- exp_tor1 = EXP(fbp->p_tor1 * SQR(2.-bo_jk->BO_pi-f11_DjDk));
- exp_tor2_kl = EXP( -p_tor2 * BOA_kl );
- exp_cot2_kl = EXP( -p_cot2 * SQR(BOA_kl-1.5) );
- fn10 = (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_jk) *
- (1.0 - exp_tor2_kl);
-
- CV = 0.5 * ( fbp->V1 * (1.0 + cos_omega) +
- fbp->V2 * exp_tor1 * (1.0 - cos2omega) +
- fbp->V3 * (1.0 + cos3omega) );
- //CV = 0.5 * fbp->V1 * (1.0 + cos_omega) +
- // fbp->V2 * exp_tor1 * (1.0 - SQR(cos_omega)) +
- // fbp->V3 * (0.5 + 2.0*CUBE(cos_omega) - 1.5 * cos_omega);
-
- data->E_Tor += e_tor = fn10 * sin_ijk * sin_jkl * CV;
-
- dfn11 = (-p_tor3 * exp_tor3_DjDk +
- (p_tor3 * exp_tor3_DjDk - p_tor4 * exp_tor4_DjDk) *
- (2.+exp_tor3_DjDk) * exp_tor34_inv) * exp_tor34_inv;
-
- CEtors1 = sin_ijk * sin_jkl * CV;
-
- CEtors2 = -fn10 * 2.0 * fbp->p_tor1 * fbp->V2 * exp_tor1 *
- (2.0 - bo_jk->BO_pi - f11_DjDk) * (1.0 - SQR(cos_omega)) *
- sin_ijk * sin_jkl;
-
- CEtors3 = CEtors2 * dfn11;
-
- CEtors4 = CEtors1 * p_tor2 * exp_tor2_ij *
- (1.0 - exp_tor2_jk) * (1.0 - exp_tor2_kl);
-
- CEtors5 = CEtors1 * p_tor2 * exp_tor2_jk *
- (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_kl);
-
- CEtors6 = CEtors1 * p_tor2 * exp_tor2_kl *
- (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_jk);
-
- cmn = -fn10 * CV;
- CEtors7 = cmn * sin_jkl * tan_ijk_i;
- CEtors8 = cmn * sin_ijk * tan_jkl_i;
- CEtors9 = fn10 * sin_ijk * sin_jkl *
- (0.5 * fbp->V1 - 2.0 * fbp->V2 * exp_tor1 * cos_omega +
- 1.5 * fbp->V3 * (cos2omega + 2. * SQR(cos_omega)));
- //cmn = -fn10 * CV;
- //CEtors7 = cmn * sin_jkl * cos_ijk;
- //CEtors8 = cmn * sin_ijk * cos_jkl;
- //CEtors9 = fn10 * sin_ijk * sin_jkl *
- // (0.5 * fbp->V1 - 2.0 * fbp->V2 * exp_tor1 * cos_omega +
- // fbp->V3 * (6*SQR(cos_omega) - 1.50));
- /* end of torsion energy */
-
-
- /* 4-body conjugation energy */
- fn12 = exp_cot2_ij * exp_cot2_jk * exp_cot2_kl;
- data->E_Con += e_con = fbp->p_cot1 * fn12 *
- (1. + (SQR(cos_omega)-1.) * sin_ijk*sin_jkl);
-
- Cconj = -2.0 * fn12 * fbp->p_cot1 * p_cot2 *
- (1. + (SQR(cos_omega)-1.) * sin_ijk*sin_jkl);
-
- CEconj1 = Cconj * (BOA_ij - 1.5e0);
- CEconj2 = Cconj * (BOA_jk - 1.5e0);
- CEconj3 = Cconj * (BOA_kl - 1.5e0);
-
- CEconj4 = -fbp->p_cot1 * fn12 *
- (SQR(cos_omega) - 1.0) * sin_jkl * tan_ijk_i;
- CEconj5 = -fbp->p_cot1 * fn12 *
- (SQR(cos_omega) - 1.0) * sin_ijk * tan_jkl_i;
- //CEconj4 = -fbp->p_cot1 * fn12 *
- // (SQR(cos_omega) - 1.0) * sin_jkl * cos_ijk;
- //CEconj5 = -fbp->p_cot1 * fn12 *
- // (SQR(cos_omega) - 1.0) * sin_ijk * cos_jkl;
- CEconj6 = 2.0 * fbp->p_cot1 * fn12 *
- cos_omega * sin_ijk * sin_jkl;
- /* end 4-body conjugation energy */
-
- //fprintf(stdout, "%6d %6d %6d %6d %7.3f %7.3f %7.3f %7.3f ",
- // workspace->orig_id[i], workspace->orig_id[j],
- // workspace->orig_id[k], workspace->orig_id[l],
- // omega, cos_omega, cos2omega, cos3omega );
- //fprintf(stdout,
- // "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- // CEtors2, CEtors3, CEtors4, CEtors5,
- // CEtors6, CEtors7, CEtors8, CEtors9 );
- //fprintf(stdout, "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- // theta_ijk, theta_jkl, sin_ijk,
- // sin_jkl, cos_jkl, tan_jkl_i );
-
- /* forces */
- bo_jk->Cdbopi += CEtors2;
- workspace->CdDelta[j] += CEtors3;
- workspace->CdDelta[k] += CEtors3;
- bo_ij->Cdbo += (CEtors4 + CEconj1);
- bo_jk->Cdbo += (CEtors5 + CEconj2);
-
- bo_kl->Cdbo += (CEtors6 + CEconj3);
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- /* dcos_theta_ijk */
- rvec_ScaledAdd( system->atoms[i].f,
- CEtors7 + CEconj4, p_ijk->dcos_dk );
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors7 + CEconj4, p_ijk->dcos_dj );
- rvec_ScaledAdd( system->atoms[k].f,
- CEtors7 + CEconj4, p_ijk->dcos_di );
-
- /* dcos_theta_jkl */
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors8 + CEconj5, p_jkl->dcos_di );
- rvec_ScaledAdd( system->atoms[k].f,
- CEtors8 + CEconj5, p_jkl->dcos_dj );
- rvec_ScaledAdd( system->atoms[l].f,
- CEtors8 + CEconj5, p_jkl->dcos_dk );
-
- /* dcos_omega */
- rvec_ScaledAdd( system->atoms[i].f,
- CEtors9 + CEconj6, dcos_omega_di );
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors9 + CEconj6, dcos_omega_dj );
- rvec_ScaledAdd( system->atoms[k].f,
- CEtors9 + CEconj6, dcos_omega_dk );
- rvec_ScaledAdd( system->atoms[l].f,
- CEtors9 + CEconj6, dcos_omega_dl );
- }
- else {
- ivec_Sum(rel_box_jl, pbond_jk->rel_box, pbond_kl->rel_box);
-
- /* dcos_theta_ijk */
- rvec_Scale( force, CEtors7 + CEconj4, p_ijk->dcos_dk );
- rvec_Add( system->atoms[i].f, force );
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors7 + CEconj4, p_ijk->dcos_dj );
-
- rvec_Scale( force, CEtors7 + CEconj4, p_ijk->dcos_di );
- rvec_Add( system->atoms[k].f, force );
- rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
-
- /* dcos_theta_jkl */
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors8 + CEconj5, p_jkl->dcos_di );
-
- rvec_Scale( force, CEtors8 + CEconj5, p_jkl->dcos_dj );
- rvec_Add( system->atoms[k].f, force );
- rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
- rvec_Scale( force, CEtors8 + CEconj5, p_jkl->dcos_dk );
- rvec_Add( system->atoms[l].f, force );
- rvec_iMultiply( ext_press, rel_box_jl, force );
- rvec_Add( data->ext_press, ext_press );
-
-
- /* dcos_omega */
- rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_di );
- rvec_Add( system->atoms[i].f, force );
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
- rvec_ScaledAdd( system->atoms[j].f,
- CEtors9 + CEconj6, dcos_omega_dj );
-
- rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_dk );
- rvec_Add( system->atoms[k].f, force );
- rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
- rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_dl );
- rvec_Add( system->atoms[l].f, force );
- rvec_iMultiply( ext_press, rel_box_jl, force );
- rvec_Add( data->ext_press, ext_press );
-
-
- /* This part is intended for a fully-flexible box */
- /* rvec_ScaledSum( temp_rvec,
- CEtors7 + CEconj4, p_ijk->dcos_dk, // i
- CEtors9 + CEconj6, dcos_omega_di );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[i].x );
- rtensor_Copy( total_rtensor, temp_rtensor );
-
- rvec_ScaledSum( temp_rvec,
- CEtors7 + CEconj4, p_ijk->dcos_dj, // j
- CEtors8 + CEconj5, p_jkl->dcos_di );
- rvec_ScaledAdd( temp_rvec,
- CEtors9 + CEconj6, dcos_omega_dj );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[j].x );
- rtensor_Add( total_rtensor, temp_rtensor );
-
- rvec_ScaledSum( temp_rvec,
- CEtors7 + CEconj4, p_ijk->dcos_di, // k
- CEtors8 + CEconj5, p_jkl->dcos_dj );
- rvec_ScaledAdd( temp_rvec,
- CEtors9 + CEconj6, dcos_omega_dk );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[k].x );
- rtensor_Add( total_rtensor, temp_rtensor );
-
- rvec_ScaledSum( temp_rvec,
- CEtors8 + CEconj5, p_jkl->dcos_dk, // l
- CEtors9 + CEconj6, dcos_omega_dl );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[l].x );
- rtensor_Copy( total_rtensor, temp_rtensor );
-
- if( pbond_ij->imaginary || pbond_jk->imaginary ||
- pbond_kl->imaginary )
- rtensor_ScaledAdd( data->flex_bar.P, -1., total_rtensor );
- else
- rtensor_Add( data->flex_bar.P, total_rtensor ); */
- }
-
-#ifdef TEST_ENERGY
- /*fprintf( out_control->etor,
- //"%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- //r_ij, r_jk, r_kl,
- "%12.8f%12.8f%12.8f%12.8f\n",
- cos_ijk, cos_jkl, sin_ijk, sin_jkl );*/
- // fprintf( out_control->etor, "%12.8f\n", dfn11 );
- fprintf( out_control->etor, "%12.8f%12.8f%12.8f\n",
- fn10, cos_omega, CV );
-
- fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- CEtors2, CEtors3, CEtors4, CEtors5,
- CEtors6, CEtors7, CEtors8, CEtors9 );
-
- /* fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- htra, htrb, htrc, hthd, hthe, hnra, hnrc, hnhd, hnhe ); */
-
- fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
- CEconj1, CEconj2, CEconj3, CEconj4, CEconj5, CEconj6 );
- /* fprintf(out_control->etor,"%23.15e%23.15e%23.15e%23.15e\n",
- fbp->V1, fbp->V2, fbp->V3, fbp->p_tor1 );*/
-
- fprintf( out_control->etor,
- //"%6d%6d%6d%6d%23.15e%23.15e%23.15e%23.15e\n",
- "%6d%6d%6d%6d%12.8f%12.8f\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], workspace->orig_id[l],
- e_tor, e_con );
- //RAD2DEG(omega), BOA_jk, e_tor, data->E_Tor );
-
- fprintf( out_control->econ,
- "%6d%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], workspace->orig_id[l],
- RAD2DEG(omega), BOA_ij, BOA_jk, BOA_kl,
- e_con,data->E_Con );
-
- /* fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
- (CEtors7 + CEconj4)*p_ijk->dcos_dk[0],
- (CEtors7 + CEconj4)*p_ijk->dcos_dk[1],
- (CEtors7 + CEconj4)*p_ijk->dcos_dk[2],
- (CEtors7 + CEconj4)*p_ijk->dcos_dj[0],
- (CEtors7 + CEconj4)*p_ijk->dcos_dj[1],
- (CEtors7 + CEconj4)*p_ijk->dcos_dj[2],
- (CEtors7 + CEconj4)*p_ijk->dcos_di[0],
- (CEtors7 + CEconj4)*p_ijk->dcos_di[1],
- (CEtors7 + CEconj4)*p_ijk->dcos_di[2] ); */
-
-
- /* fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
- (CEtors8 + CEconj5)*p_jkl->dcos_di[0],
- (CEtors8 + CEconj5)*p_jkl->dcos_di[1],
- (CEtors8 + CEconj5)*p_jkl->dcos_di[2],
- (CEtors8 + CEconj5)*p_jkl->dcos_dj[0],
- (CEtors8 + CEconj5)*p_jkl->dcos_dj[1],
- (CEtors8 + CEconj5)*p_jkl->dcos_dj[2],
- (CEtors8 + CEconj5)*p_jkl->dcos_dk[0],
- (CEtors8 + CEconj5)*p_jkl->dcos_dk[1],
- (CEtors8 + CEconj5)*p_jkl->dcos_dk[2] ); */
-
- fprintf( out_control->etor,
- "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
- dcos_omega_di[0], dcos_omega_di[1], dcos_omega_di[2],
- dcos_omega_dj[0], dcos_omega_dj[1], dcos_omega_dj[2],
- dcos_omega_dk[0], dcos_omega_dk[1], dcos_omega_dk[2],
- dcos_omega_dl[0], dcos_omega_dl[1], dcos_omega_dl[2] );
-#endif
-
-#ifdef TEST_FORCES
- // Torsion Forces
- Add_dBOpinpi2(system, lists, j, pk, CEtors2, 0.,
- workspace->f_tor, workspace->f_tor);
- Add_dDelta( system, lists, j, CEtors3, workspace->f_tor );
- Add_dDelta( system, lists, k, CEtors3, workspace->f_tor );
- Add_dBO( system, lists, j, pij, CEtors4, workspace->f_tor );
- Add_dBO( system, lists, j, pk, CEtors5, workspace->f_tor );
- Add_dBO( system, lists, k, plk, CEtors6, workspace->f_tor );
-
- rvec_ScaledAdd(workspace->f_tor[i], CEtors7, p_ijk->dcos_dk);
- rvec_ScaledAdd(workspace->f_tor[j], CEtors7, p_ijk->dcos_dj);
- rvec_ScaledAdd(workspace->f_tor[k], CEtors7, p_ijk->dcos_di);
-
- rvec_ScaledAdd(workspace->f_tor[j], CEtors8, p_jkl->dcos_di);
- rvec_ScaledAdd(workspace->f_tor[k], CEtors8, p_jkl->dcos_dj);
- rvec_ScaledAdd(workspace->f_tor[l], CEtors8, p_jkl->dcos_dk);
-
- rvec_ScaledAdd( workspace->f_tor[i], CEtors9, dcos_omega_di );
- rvec_ScaledAdd( workspace->f_tor[j], CEtors9, dcos_omega_dj );
- rvec_ScaledAdd( workspace->f_tor[k], CEtors9, dcos_omega_dk );
- rvec_ScaledAdd( workspace->f_tor[l], CEtors9, dcos_omega_dl );
-
- // Conjugation Forces
- Add_dBO( system, lists, j, pij, CEconj1, workspace->f_con );
- Add_dBO( system, lists, j, pk, CEconj2, workspace->f_con );
- Add_dBO( system, lists, k, plk, CEconj3, workspace->f_con );
-
- rvec_ScaledAdd(workspace->f_con[i], CEconj4, p_ijk->dcos_dk);
- rvec_ScaledAdd(workspace->f_con[j], CEconj4, p_ijk->dcos_dj);
- rvec_ScaledAdd(workspace->f_con[k], CEconj4, p_ijk->dcos_di);
-
- rvec_ScaledAdd(workspace->f_con[j], CEconj5, p_jkl->dcos_di);
- rvec_ScaledAdd(workspace->f_con[k], CEconj5, p_jkl->dcos_dj);
- rvec_ScaledAdd(workspace->f_con[l], CEconj5, p_jkl->dcos_dk);
-
- rvec_ScaledAdd( workspace->f_con[i], CEconj6, dcos_omega_di );
- rvec_ScaledAdd( workspace->f_con[j], CEconj6, dcos_omega_dj );
- rvec_ScaledAdd( workspace->f_con[k], CEconj6, dcos_omega_dk );
- rvec_ScaledAdd( workspace->f_con[l], CEconj6, dcos_omega_dl );
-#endif
- } // pl check ends
- } // pl loop ends
- } // pi check ends
- } // pi loop ends
- } // k-j neighbor check ends
- } // j<k && j-k neighbor check ends
- } // pk loop ends
- } // j loop
-
- /* fprintf( stderr, "4body: ext_press (%23.15e %23.15e %23.15e)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] );*/
-
-#ifdef TEST_FORCES
- fprintf( stderr, "Number of torsion angles: %d\n", num_frb_intrs );
- fprintf( stderr, "Torsion Energy: %g\t Conjugation Energy: %g\n",
- data->E_Tor, data->E_Con );
-#endif
-}
-
-
-////////////////////////////////////////////////////////////////////////
-//Cuda Functions
-////////////////////////////////////////////////////////////////////////
-
-GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
+GLOBAL void k_Four_Body_Interactions ( reax_atom *atoms,
global_parameters g_params,
four_body_header *d_fbp,
control_params *control,
@@ -741,7 +231,6 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
list *thb_intrs = &p_thb_intrs;
static_storage *workspace = &p_workspace;
-
//for( j = 0; j < system->N; ++j ) {
type_j = atoms[j].type;
Delta_j = workspace->Delta_boc[j];
@@ -836,8 +325,8 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
pbond_kl = &( bonds->select.bond_list[plk] );
bo_kl = &( pbond_kl->bo_data );
type_l = atoms[l].type;
- fbh = &(d_fbp[ index_fbp (type_i,type_j,type_k,type_l,num_atom_types) ]);
- fbp = &(d_fbp[ index_fbp (type_i,type_j,type_k,type_l,num_atom_types)].prm[0]);
+ fbh = &(d_fbp[ index_fbp(type_i,type_j,type_k,type_l,num_atom_types) ]);
+ fbp = &(d_fbp[ index_fbp(type_i,type_j,type_k,type_l,num_atom_types)].prm[0]);
if( i != l && fbh->cnt && bo_kl->BO > control->thb_cut/*0*/ &&
bo_ij->BO * bo_jk->BO * bo_kl->BO > control->thb_cut/*0*/ ){
@@ -889,7 +378,7 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
//PERFORMANCE IMPACT
e_tor = fn10 * sin_ijk * sin_jkl * CV;
- //atomicAdd (&data->E_Tor ,e_tor );
+ //MYATOMICADD(&data->E_Tor ,e_tor );
E_Tor [j] += e_tor;
//sh_tor [threadIdx.x] += e_tor;
@@ -933,7 +422,7 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
fn12 = exp_cot2_ij * exp_cot2_jk * exp_cot2_kl;
//PERFORMANCE IMPACT
e_con = fbp->p_cot1 * fn12 * (1. + (SQR(cos_omega)-1.) * sin_ijk*sin_jkl);
- //atomicAdd (&data->E_Con ,e_con );
+ //MYATOMICADD(&data->E_Con ,e_con );
E_Con [j] += e_con ;
//sh_con [threadIdx.x] += e_con;
@@ -971,12 +460,12 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
/* forces */
//PERFORMANCE IMPACT
/*
- atomicAdd ( &bo_jk->Cdbopi, CEtors2 );
- atomicAdd ( &workspace->CdDelta[j], CEtors3 );
- atomicAdd ( &workspace->CdDelta[k], CEtors3 );
- atomicAdd ( &bo_ij->Cdbo, (CEtors4 + CEconj1) );
- atomicAdd ( &bo_jk->Cdbo, (CEtors5 + CEconj2) );
- atomicAdd ( &bo_kl->Cdbo, (CEtors6 + CEconj3) );
+ MYATOMICADD( &bo_jk->Cdbopi, CEtors2 );
+ MYATOMICADD( &workspace->CdDelta[j], CEtors3 );
+ MYATOMICADD( &workspace->CdDelta[k], CEtors3 );
+ MYATOMICADD( &bo_ij->Cdbo, (CEtors4 + CEconj1) );
+ MYATOMICADD( &bo_jk->Cdbo, (CEtors5 + CEconj2) );
+ MYATOMICADD( &bo_kl->Cdbo, (CEtors6 + CEconj3) );
*/
//PERFORMANCE IMPACT
@@ -987,39 +476,29 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
bo_jk->Cdbo += CEtors5 + CEconj2;
//TODO REMOVE THIS ATOMIC OPERATION IF POSSIBLE
- atomicAdd (&pbond_kl->Cdbo_kl, CEtors6 + CEconj3 );
+ MYATOMICADD(&pbond_kl->Cdbo_kl, CEtors6 + CEconj3 );
//TODO REMOVE THIS ATOMIC OPERATION IF POSSIBLE
if( control->ensemble == NVE || control->ensemble == NVT ||control->ensemble == bNVT) {
/* dcos_theta_ijk */
//PERFORMANCE IMPACT
- atomic_rvecScaledAdd (pbond_ij->i_f,
- CEtors7 + CEconj4, p_ijk->dcos_dk );
- rvec_ScaledAdd( atoms[j].f,
- CEtors7 + CEconj4, p_ijk->dcos_dj );
- atomic_rvecScaledAdd( pbond_jk->k_f,
- CEtors7 + CEconj4, p_ijk->dcos_di );
+ atomic_rvecScaledAdd( pbond_ij->i_f, CEtors7 + CEconj4, p_ijk->dcos_dk );
+ rvec_ScaledAdd( atoms[j].f, CEtors7 + CEconj4, p_ijk->dcos_dj );
+ atomic_rvecScaledAdd( pbond_jk->k_f, CEtors7 + CEconj4, p_ijk->dcos_di );
/* dcos_theta_jkl */
//PERFORMANCE IMPACT
- rvec_ScaledAdd( atoms[j].f,
- CEtors8 + CEconj5, p_jkl->dcos_di );
- atomic_rvecScaledAdd( pbond_jk->i_f,
- CEtors8 + CEconj5, p_jkl->dcos_dj );
- atomic_rvecScaledAdd( pbond_kl->k_f,
- CEtors8 + CEconj5, p_jkl->dcos_dk );
+ rvec_ScaledAdd( atoms[j].f, CEtors8 + CEconj5, p_jkl->dcos_di );
+ atomic_rvecScaledAdd( pbond_jk->i_f, CEtors8 + CEconj5, p_jkl->dcos_dj );
+ atomic_rvecScaledAdd( pbond_kl->k_f, CEtors8 + CEconj5, p_jkl->dcos_dk );
/* dcos_omega */
//PERFORMANCE IMPACT
- atomic_rvecScaledAdd( pbond_ij->i_f,
- CEtors9 + CEconj6, dcos_omega_di );
- rvec_ScaledAdd( atoms[j].f,
- CEtors9 + CEconj6, dcos_omega_dj );
- atomic_rvecScaledAdd( pbond_jk->i_f,
- CEtors9 + CEconj6, dcos_omega_dk );
- atomic_rvecScaledAdd( pbond_kl->k_f,
- CEtors9 + CEconj6, dcos_omega_dl );
+ atomic_rvecScaledAdd( pbond_ij->i_f, CEtors9 + CEconj6, dcos_omega_di );
+ rvec_ScaledAdd( atoms[j].f, CEtors9 + CEconj6, dcos_omega_dj );
+ atomic_rvecScaledAdd( pbond_jk->i_f, CEtors9 + CEconj6, dcos_omega_dk );
+ atomic_rvecScaledAdd( pbond_kl->k_f, CEtors9 + CEconj6, dcos_omega_dl );
}
else {
ivec_Sum(rel_box_jl, pbond_jk->rel_box, pbond_kl->rel_box);
@@ -1033,8 +512,7 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
//rvec_Add (sh_press [threadIdx.x], ext_press);
//PERFORMANCE IMPACT
- rvec_ScaledAdd( atoms[j].f,
- CEtors7 + CEconj4, p_ijk->dcos_dj );
+ rvec_ScaledAdd( atoms[j].f, CEtors7 + CEconj4, p_ijk->dcos_dj );
rvec_Scale( force, CEtors7 + CEconj4, p_ijk->dcos_di );
//PERFORMANCE IMPACT
@@ -1047,8 +525,7 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
/* dcos_theta_jkl */
//PERFORMANCE IMPACT
- rvec_ScaledAdd( atoms[j].f,
- CEtors8 + CEconj5, p_jkl->dcos_di );
+ rvec_ScaledAdd( atoms[j].f, CEtors8 + CEconj5, p_jkl->dcos_di );
rvec_Scale( force, CEtors8 + CEconj5, p_jkl->dcos_dj );
//PERFORMANCE IMPACT
@@ -1327,7 +804,7 @@ GLOBAL void Four_Body_Interactions ( reax_atom *atoms,
}
-GLOBAL void Four_Body_Postprocess ( reax_atom *atoms,
+GLOBAL void k_Four_Body_Postprocess( reax_atom *atoms,
static_storage p_workspace,
list p_bonds, int N )
{
diff --git a/PuReMD-GPU/src/cuda_four_body_interactions.h b/PuReMD-GPU/src/cuda_four_body_interactions.h
new file mode 100644
index 0000000000000000000000000000000000000000..088e24f4f15a1e83405b83427ba56d1e5d4e67ba
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_four_body_interactions.h
@@ -0,0 +1,42 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_FOUR_BODY_INTERACTIONS_H_
+#define __CUDA_FOUR_BODY_INTERACTIONS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void k_Four_Body_Interactions( reax_atom *, global_parameters ,
+ four_body_header *, control_params *, list , list , simulation_box *,
+ simulation_data *, static_storage , int , int , real *, real *, rvec * );
+
+GLOBAL void k_Four_Body_Postprocess( reax_atom *, static_storage, list , int );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_grid.cu b/PuReMD-GPU/src/cuda_grid.cu
new file mode 100644
index 0000000000000000000000000000000000000000..ca9556d6b43ffcac8a828d90f87e743d2647152c
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_grid.cu
@@ -0,0 +1,48 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_grid.h"
+
+#include "grid.h"
+#include "index_utils.h"
+#include "vector.h"
+
+#include "cuda_utils.h"
+#include "cuda_reset_utils.h"
+
+
+void Cuda_Bin_Atoms (reax_system *system, static_storage *workspace )
+{
+ Cuda_Reset_Grid ( &system->d_g);
+
+ Bin_Atoms ( system, workspace );
+
+ dev_workspace->realloc.gcell_atoms = workspace->realloc.gcell_atoms;
+}
+
+
+void Cuda_Bin_Atoms_Sync (reax_system *system)
+{
+ copy_host_device (system->g.top, system->d_g.top,
+ INT_SIZE * system->g.ncell[0]*system->g.ncell[1]*system->g.ncell[2], cudaMemcpyHostToDevice, RES_GRID_TOP);
+
+ copy_host_device (system->g.atoms, system->d_g.atoms,
+ INT_SIZE * system->g.max_atoms*system->g.ncell[0]*system->g.ncell[1]*system->g.ncell[2], cudaMemcpyHostToDevice, RES_GRID_ATOMS);
+}
diff --git a/PuReMD-GPU/src/cuda_grid.h b/PuReMD-GPU/src/cuda_grid.h
new file mode 100644
index 0000000000000000000000000000000000000000..28a20797a56bda9682354f967634eeb6301e44ae
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_grid.h
@@ -0,0 +1,39 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_GRID_H_
+#define __CUDA_GRID_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Bin_Atoms( reax_system*, static_storage* );
+void Cuda_Bin_Atoms_Sync (reax_system *);
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_helpers.h b/PuReMD-GPU/src/cuda_helpers.h
index e021cf84a99b22713fa2fdc4a00af81db0ef5672..e306c673f8aec585bbda7b92250e2f0ec91eb2e3 100644
--- a/PuReMD-GPU/src/cuda_helpers.h
+++ b/PuReMD-GPU/src/cuda_helpers.h
@@ -21,9 +21,15 @@
#ifndef __CUDA_HELPERS__
#define __CUDA_HELPERS__
+
#include "mytypes.h"
-DEVICE inline int cuda_strcmp (char *a, char *b, int len)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static inline DEVICE int cuda_strcmp(char *a, char *b, int len)
{
char *src, *dst;
@@ -32,20 +38,25 @@ DEVICE inline int cuda_strcmp (char *a, char *b, int len)
for (int i = 0; i < len; i++)
{
-
if (*dst == '\0')
+ {
return 0;
+ }
- if (*src != *dst) return 1;
+ if (*src != *dst)
+ {
+ return 1;
+ }
- src ++;
- dst ++;
+ src++;
+ dst++;
}
return 0;
}
-DEVICE inline real atomicAdd(real* address, real val)
+
+static inline DEVICE double myAtomicAdd(double* address, double val)
{
unsigned long long int* address_as_ull =
(unsigned long long int*)address;
@@ -54,24 +65,31 @@ DEVICE inline real atomicAdd(real* address, real val)
{
assumed = old;
old = atomicCAS(address_as_ull, assumed,
- __double_as_longlong(val + __longlong_as_double(assumed)));
+ __double_as_longlong(val + __longlong_as_double(assumed)));
}
while (assumed != old);
return __longlong_as_double(old);
}
-DEVICE inline void atomic_rvecAdd( rvec ret, rvec v )
+
+static inline DEVICE void atomic_rvecAdd( rvec ret, rvec v )
{
- atomicAdd ( &ret[0], v[0] );
- atomicAdd ( &ret[1], v[1] );
- atomicAdd ( &ret[2], v[2] );
+ MYATOMICADD( (double*)&ret[0], (double)v[0] );
+ MYATOMICADD( (double*)&ret[1], (double)v[1] );
+ MYATOMICADD( (double*)&ret[2], (double)v[2] );
}
-DEVICE inline void atomic_rvecScaledAdd( rvec ret, real c, rvec v )
+
+static inline DEVICE void atomic_rvecScaledAdd( rvec ret, real c, rvec v )
{
- atomicAdd ( &ret[0], c * v[0] );
- atomicAdd ( &ret[1], c * v[1] );
- atomicAdd ( &ret[2], c * v[2] );
+ MYATOMICADD( (double*)&ret[0], (double)(c * v[0]) );
+ MYATOMICADD( (double*)&ret[1], (double)(c * v[1]) );
+ MYATOMICADD( (double*)&ret[2], (double)(c * v[2]) );
+}
+
+#ifdef __cplusplus
}
+#endif
+
#endif
diff --git a/PuReMD-GPU/src/cuda_init.cu b/PuReMD-GPU/src/cuda_init.cu
index 09515038daa71b15ef2c5796f54f3c79c29063df..4ca4bac18d08052311d0bdcd83290b5617b66b55 100644
--- a/PuReMD-GPU/src/cuda_init.cu
+++ b/PuReMD-GPU/src/cuda_init.cu
@@ -18,59 +18,65 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
-
-
#include "cuda_init.h"
+
#include "cuda_utils.h"
#include "cuda_copy.h"
+#include "cuda_reset_utils.h"
+
#include "vector.h"
-#include "reset_utils.h"
-void Cuda_Init_System ( reax_system *system)
+
+void Cuda_Init_System( reax_system *system)
{
- cuda_malloc ( (void **) &system->d_atoms, system->N * REAX_ATOM_SIZE, 1, RES_SYSTEM_ATOMS );
+ cuda_malloc( (void **) &system->d_atoms, system->N * REAX_ATOM_SIZE, 1, RES_SYSTEM_ATOMS );
- cuda_malloc ( (void **) &system->d_box, sizeof (simulation_box), 1, RES_SYSTEM_SIMULATION_BOX );
+ cuda_malloc( (void **) &system->d_box, sizeof (simulation_box), 1, RES_SYSTEM_SIMULATION_BOX );
//interaction parameters
- cuda_malloc ((void **) &system->reaxprm.d_sbp, system->reaxprm.num_atom_types * SBP_SIZE,
+ cuda_malloc((void **) &system->reaxprm.d_sbp, system->reaxprm.num_atom_types * SBP_SIZE,
1, RES_REAX_INT_SBP );
- cuda_malloc ((void **) &system->reaxprm.d_tbp, pow (system->reaxprm.num_atom_types, 2) * TBP_SIZE,
+ cuda_malloc((void **) &system->reaxprm.d_tbp, pow (system->reaxprm.num_atom_types, 2) * TBP_SIZE,
1, RES_REAX_INT_TBP );
- cuda_malloc ((void **) &system->reaxprm.d_thbp, pow (system->reaxprm.num_atom_types, 3) * THBP_SIZE,
+ cuda_malloc((void **) &system->reaxprm.d_thbp, pow (system->reaxprm.num_atom_types, 3) * THBP_SIZE,
1, RES_REAX_INT_THBP );
- cuda_malloc ((void **) &system->reaxprm.d_hbp, pow (system->reaxprm.num_atom_types, 3) * HBP_SIZE,
+ cuda_malloc((void **) &system->reaxprm.d_hbp, pow (system->reaxprm.num_atom_types, 3) * HBP_SIZE,
1, RES_REAX_INT_HBP );
- cuda_malloc ((void **) &system->reaxprm.d_fbp, pow (system->reaxprm.num_atom_types, 4) * FBP_SIZE,
+ cuda_malloc((void **) &system->reaxprm.d_fbp, pow (system->reaxprm.num_atom_types, 4) * FBP_SIZE,
1, RES_REAX_INT_FBP );
- cuda_malloc ((void **) &system->reaxprm.d_gp.l, REAL_SIZE * system->reaxprm.gp.n_global, 1, RES_GLOBAL_PARAMS );
+ cuda_malloc((void **) &system->reaxprm.d_gp.l, REAL_SIZE * system->reaxprm.gp.n_global, 1, RES_GLOBAL_PARAMS );
system->reaxprm.d_gp.n_global = 0;
system->reaxprm.d_gp.vdw_type = 0;
}
-void Cuda_Init_Control (control_params *control)
+
+void Cuda_Init_Control(control_params *control)
{
- cuda_malloc ((void **)&control->d_control, CONTROL_PARAMS_SIZE, 1, RES_CONTROL_PARAMS );
- copy_host_device (control, control->d_control, CONTROL_PARAMS_SIZE, cudaMemcpyHostToDevice, RES_CONTROL_PARAMS );
+ cuda_malloc((void **)&control->d_control, CONTROL_PARAMS_SIZE, 1, RES_CONTROL_PARAMS );
+ copy_host_device(control, control->d_control, CONTROL_PARAMS_SIZE, cudaMemcpyHostToDevice, RES_CONTROL_PARAMS );
}
+
void Cuda_Init_Simulation_Data (simulation_data *data)
{
- cuda_malloc ((void **) &(data->d_simulation_data), SIMULATION_DATA_SIZE, 1, RES_SIMULATION_DATA );
+ cuda_malloc((void **) &(data->d_simulation_data), SIMULATION_DATA_SIZE, 1, RES_SIMULATION_DATA );
}
-GLOBAL void Initialize_Grid (ivec *nbrs, rvec *nbrs_cp, int N)
+
+GLOBAL void Initialize_Grid(ivec *nbrs, rvec *nbrs_cp, int N)
{
int index = blockIdx.x * blockDim.x + threadIdx.x;
- if (index >= N) return;
+ if (index >= N)
+ {
+ return;
+ }
nbrs[index][0] = -1;
nbrs[index][1] = -1;
@@ -80,6 +86,7 @@ GLOBAL void Initialize_Grid (ivec *nbrs, rvec *nbrs_cp, int N)
nbrs_cp[index][2] = -1;
}
+
void Cuda_Init_Grid (grid *host, grid *dev)
{
int total = host->ncell[0] * host->ncell[1] * host->ncell[2];
@@ -89,30 +96,31 @@ void Cuda_Init_Grid (grid *host, grid *dev)
dev->max_cuda_nbrs = host->max_cuda_nbrs;
dev->cell_size = host->cell_size;
- ivec_Copy (dev->spread, host->spread);
- ivec_Copy (dev->ncell, host->ncell);
- rvec_Copy (dev->len, host->len);
- rvec_Copy (dev->inv_len, host->inv_len);
+ ivec_Copy( dev->spread, host->spread );
+ ivec_Copy( dev->ncell, host->ncell );
+ rvec_Copy( dev->len, host->len );
+ rvec_Copy( dev->inv_len, host->inv_len );
- cuda_malloc ((void **) &dev->top, INT_SIZE * total , 1, RES_GRID_TOP );
- cuda_malloc ((void **) &dev->mark, INT_SIZE * total , 1, RES_GRID_MARK );
- cuda_malloc ((void **) &dev->start, INT_SIZE * total , 1, RES_GRID_START );
- cuda_malloc ((void **) &dev->end, INT_SIZE * total , 1, RES_GRID_END );
+ cuda_malloc((void **) &dev->top, INT_SIZE * total , 1, RES_GRID_TOP );
+ cuda_malloc((void **) &dev->mark, INT_SIZE * total , 1, RES_GRID_MARK );
+ cuda_malloc((void **) &dev->start, INT_SIZE * total , 1, RES_GRID_START );
+ cuda_malloc((void **) &dev->end, INT_SIZE * total , 1, RES_GRID_END );
- cuda_malloc ((void **) &dev->atoms, INT_SIZE * total * host->max_atoms, 1, RES_GRID_ATOMS );
- cuda_malloc ((void **) &dev->nbrs, IVEC_SIZE * total * host->max_nbrs, 0, RES_GRID_NBRS );
- cuda_malloc ((void **) &dev->nbrs_cp, RVEC_SIZE * total * host->max_nbrs, 0, RES_GRID_NBRS_CP );
+ cuda_malloc((void **) &dev->atoms, INT_SIZE * total * host->max_atoms, 1, RES_GRID_ATOMS );
+ cuda_malloc((void **) &dev->nbrs, IVEC_SIZE * total * host->max_nbrs, 0, RES_GRID_NBRS );
+ cuda_malloc((void **) &dev->nbrs_cp, RVEC_SIZE * total * host->max_nbrs, 0, RES_GRID_NBRS_CP );
int block_size = 512;
int blocks = (total*dev->max_nbrs) / block_size + ((total*dev->max_nbrs) % block_size == 0 ? 0 : 1);
- Initialize_Grid <<<blocks, block_size>>>
- (dev->nbrs, dev->nbrs_cp, total * host->max_nbrs );
- cudaThreadSynchronize ();
- cudaCheckError ();
+ Initialize_Grid<<<blocks, block_size>>>
+ ( dev->nbrs, dev->nbrs_cp, total * host->max_nbrs );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
}
-GLOBAL void Init_Workspace_Arrays (single_body_parameters *sbp, reax_atom *atoms,
+
+GLOBAL void Init_Workspace_Arrays(single_body_parameters *sbp, reax_atom *atoms,
static_storage workspace, int N)
{
@@ -127,6 +135,7 @@ GLOBAL void Init_Workspace_Arrays (single_body_parameters *sbp, reax_atom *atoms
workspace.b[i+N] = -1.0;
}
+
GLOBAL void Init_Map_Serials (int *input, int N)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -135,6 +144,7 @@ GLOBAL void Init_Map_Serials (int *input, int N)
input[i] = -1;
}
+
void Cuda_Init_Workspace_System (reax_system *system, static_storage *workspace )
{
int blocks, block_size = BLOCK_SIZE;
@@ -262,6 +272,7 @@ void Cuda_Init_Workspace( reax_system *system, control_params *control,
Cuda_Reset_Workspace( system, workspace );
}
+
void Cuda_Init_Workspace_Device ( static_storage *workspace )
{
workspace->realloc.estimate_nbrs = -1;
@@ -273,6 +284,7 @@ void Cuda_Init_Workspace_Device ( static_storage *workspace )
workspace->realloc.gcell_atoms = -1;
}
+
void Cuda_Init_Sparse_Matrix (sparse_matrix *matrix, int entries, int N)
{
cuda_malloc ((void **) &matrix->start, INT_SIZE * (N + 1), 1, RES_SPARSE_MATRIX_INDEX );
@@ -284,7 +296,8 @@ void Cuda_Init_Sparse_Matrix (sparse_matrix *matrix, int entries, int N)
}
-void Cuda_Init_Scratch ()
+
+void Cuda_Init_Scratch()
{
cuda_malloc ((void **) &scratch, SCRATCH_SIZE, 0, RES_SCRATCH );
diff --git a/PuReMD-GPU/src/cuda_init.h b/PuReMD-GPU/src/cuda_init.h
index 233761691fe56bc8b1290c8972c4b413cc876f54..cd9c568130730d298fbaf781f4f6b18b0f02f65a 100644
--- a/PuReMD-GPU/src/cuda_init.h
+++ b/PuReMD-GPU/src/cuda_init.h
@@ -18,22 +18,31 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
#ifndef __CUDA_INIT_H__
#define __CUDA_INIT_H__
#include "mytypes.h"
-void Cuda_Init_System ( reax_system* );
-void Cuda_Init_Simulation_Data (simulation_data *);
-void Cuda_Init_Workspace_System ( reax_system *, static_storage *);
-void Cuda_Init_Workspace ( reax_system *, control_params *, static_storage *);
-void Cuda_Init_Workspace_Device ( static_storage *);
-void Cuda_Init_Control (control_params *);
-void Cuda_Init_Grid (grid *, grid *);
-void Cuda_Init_Sparse_Matrix (sparse_matrix *, int, int);
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Init_System( reax_system* );
+void Cuda_Init_Simulation_Data( simulation_data * );
+void Cuda_Init_Workspace_System( reax_system *, static_storage * );
+void Cuda_Init_Workspace( reax_system *, control_params *, static_storage * );
+void Cuda_Init_Workspace_Device( static_storage * );
+void Cuda_Init_Control( control_params * );
+void Cuda_Init_Grid( grid *, grid * );
+
+void Cuda_Init_Sparse_Matrix( sparse_matrix *, int, int );
+
+void Cuda_Init_Scratch( );
+
+#ifdef __cplusplus
+}
+#endif
-void Cuda_Init_Scratch ();
#endif
diff --git a/PuReMD-GPU/src/cuda_init_md.cu b/PuReMD-GPU/src/cuda_init_md.cu
new file mode 100644
index 0000000000000000000000000000000000000000..1a205506e4c5ff767e02398a3859f838818c1e1a
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_init_md.cu
@@ -0,0 +1,586 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_init_md.h"
+
+#include "allocate.h"
+#include "box.h"
+#include "forces.h"
+#include "grid.h"
+#include "index_utils.h"
+#include "init_md.h"
+#include "integrate.h"
+#include "lookup.h"
+#include "print_utils.h"
+#include "reset_utils.h"
+#include "system_props.h"
+#include "traj.h"
+#include "vector.h"
+
+#include "cuda_allocate.h"
+#include "cuda_utils.h"
+#include "cuda_init.h"
+#include "cuda_copy.h"
+#include "cuda_box.h"
+#include "cuda_forces.h"
+#include "cuda_grid.h"
+#include "cuda_integrate.h"
+#include "cuda_lin_alg.h"
+#include "cuda_list.h"
+#include "cuda_lookup.h"
+#include "cuda_neighbors.h"
+#include "cuda_reduction.h"
+#include "cuda_reset_utils.h"
+#include "cuda_system_props.h"
+#include "validation.h"
+
+
+void Cuda_Init_System( reax_system *system, control_params *control,
+ simulation_data *data )
+{
+ int i;
+ rvec dx;
+
+ if( !control->restart )
+ {
+ Cuda_Reset_Atoms( system );
+ }
+
+ Cuda_Compute_Total_Mass( system, data );
+
+ Cuda_Compute_Center_of_Mass( system, data, stderr );
+
+ /* reposition atoms */
+ // just fit the atoms to the periodic box
+ if( control->reposition_atoms == 0 )
+ {
+ rvec_MakeZero( dx );
+ }
+ // put the center of mass to the center of the box
+ else if( control->reposition_atoms == 1 )
+ {
+ rvec_Scale( dx, 0.5, system->box.box_norms );
+ rvec_ScaledAdd( dx, -1., data->xcm );
+ }
+ // put the center of mass to the origin
+ else if( control->reposition_atoms == 2 )
+ {
+ rvec_Scale( dx, -1., data->xcm );
+ }
+ else
+ {
+ fprintf( stderr, "UNKNOWN OPTION: reposition_atoms. Terminating...\n" );
+ exit( UNKNOWN_OPTION );
+ }
+
+ k_compute_Inc_on_T3<<<BLOCKS_POW_2, BLOCK_SIZE>>>
+ (system->d_atoms, system->N, system->d_box, dx[0], dx[1], dx[2]);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //copy back the atoms from device to the host
+ copy_host_device( system->atoms, system->d_atoms, REAX_ATOM_SIZE * system->N ,
+ cudaMemcpyDeviceToHost, RES_SYSTEM_ATOMS );
+
+ /* Initialize velocities so that desired init T can be attained */
+ if( !control->restart || (control->restart && control->random_vel) ) {
+ Generate_Initial_Velocities( system, control->T_init );
+ }
+
+ Setup_Grid( system );
+}
+
+
+void Cuda_Init_Simulation_Data( reax_system *system, control_params *control,
+ simulation_data *data, output_controls *out_control,
+ evolve_function *Evolve )
+{
+
+ Reset_Simulation_Data( data );
+
+ if( !control->restart )
+ data->step = data->prev_steps = 0;
+
+ switch( control->ensemble ) {
+ case NVE:
+ data->N_f = 3 * system->N;
+ *Evolve = Cuda_Velocity_Verlet_NVE;
+ break;
+
+
+ case NVT:
+ data->N_f = 3 * system->N + 1;
+ //control->Tau_T = 100 * data->N_f * K_B * control->T_final;
+ if( !control->restart || (control->restart && control->random_vel) ) {
+ data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
+ data->N_f * K_B * control->T );
+ data->therm.v_xi = data->therm.G_xi * control->dt;
+ data->therm.v_xi_old = 0;
+ data->therm.xi = 0;
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "init_md: G_xi=%f Tau_T=%f E_kin=%f N_f=%f v_xi=%f\n",
+ data->therm.G_xi, control->Tau_T, data->E_Kin,
+ data->N_f, data->therm.v_xi );
+#endif
+ }
+
+ *Evolve = Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein;
+ break;
+
+
+ case NPT: // Anisotropic NPT
+ fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
+ exit( UNKNOWN_OPTION );
+ data->N_f = 3 * system->N + 9;
+ if( !control->restart ) {
+ data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
+ data->N_f * K_B * control->T );
+ data->therm.v_xi = data->therm.G_xi * control->dt;
+ data->iso_bar.eps = 0.33333 * log(system->box.volume);
+ //data->inv_W = 1. / (data->N_f*K_B*control->T*SQR(control->Tau_P));
+ //Compute_Pressure( system, data, workspace );
+ }
+ *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
+ break;
+
+
+ case sNPT: // Semi-Isotropic NPT
+ fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
+ exit( UNKNOWN_OPTION );
+ data->N_f = 3 * system->N + 4;
+ *Evolve = Velocity_Verlet_Berendsen_SemiIsotropic_NPT;
+ break;
+
+
+ case iNPT: // Isotropic NPT
+ fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
+ exit( UNKNOWN_OPTION );
+ data->N_f = 3 * system->N + 2;
+ *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
+ break;
+
+ case bNVT: //berendensen NVT
+ data->N_f = 3 * system->N + 1;
+ *Evolve = Cuda_Velocity_Verlet_Berendsen_NVT;
+ break;
+
+ default:
+ break;
+ }
+
+ Cuda_Compute_Kinetic_Energy( system, data );
+
+#ifdef __BUILD_DEBUG__
+ real t_E_Kin = 0;
+ t_E_Kin = data->E_Kin;
+#endif
+
+ copy_host_device( &data->E_Kin, &((simulation_data *)data->d_simulation_data)->E_Kin,
+ REAL_SIZE, cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
+ data->therm.T = (2. * data->E_Kin) / (data->N_f * K_B);
+ if( fabs(data->therm.T) < ALMOST_ZERO ) // avoid T being an absolute zero!
+ data->therm.T = ALMOST_ZERO;
+
+#ifdef __BUILD_DEBUG__
+ if (check_zero( t_E_Kin, data->E_Kin)){
+ fprintf( stderr, "SimulationData:E_Kin does not match between host and device (%f %f) \n", t_E_Kin, data->E_Kin );
+ exit( 1 );
+ }
+ //validate_data ( system, data );
+#endif
+
+ /* init timing info for the host*/
+ data->timing.start = Get_Time( );
+ data->timing.total = data->timing.start;
+ data->timing.nbrs = 0;
+ data->timing.init_forces = 0;
+ data->timing.bonded = 0;
+ data->timing.nonb = 0;
+ data->timing.QEq = 0;
+ data->timing.matvecs = 0;
+
+ /* init timing info for the device */
+ d_timing.start = Get_Time( );
+ d_timing.total = data->timing.start;
+ d_timing.nbrs = 0;
+ d_timing.init_forces = 0;
+ d_timing.bonded = 0;
+ d_timing.nonb = 0;
+ d_timing.QEq = 0;
+ d_timing.matvecs = 0;
+}
+
+
+int Estimate_Device_Matrix( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int *indices, *Htop;
+ list *far_nbrs = dev_lists + FAR_NBRS;
+ int max_sparse_entries = 0;
+ real t1, t2;
+
+ indices = (int *) scratch;
+ cuda_memset( indices, 0, INT_SIZE * system->N, RES_SCRATCH );
+
+ t1 = Get_Time( );
+
+ k_Estimate_Sparse_Matrix_Entries<<<BLOCKS, BLOCK_SIZE>>>
+ ( system->d_atoms, (control_params *)control->d_control,
+ (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box,
+ *far_nbrs, system->N, indices );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ t2 = Get_Timing_Info( t1 );
+
+ //fprintf (stderr, " Time to estimate sparse matrix entries --- > %f \n", t2 );
+
+ Htop = (int *) malloc( INT_SIZE * (system->N + 1) );
+ memset( Htop, 0, INT_SIZE * (system->N + 1) );
+ copy_host_device( Htop, indices, system->N * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
+
+ for (int i = 0; i < system->N; i++)
+ {
+ if (max_sparse_entries < Htop[i]) {
+ max_sparse_entries = Htop[i];
+ }
+ }
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr,
+ " Max sparse entries for this run are ---> %d \n", max_sparse_entries );
+#endif
+
+ return max_sparse_entries * SAFE_ZONE;
+ //return max_sparse_entries;
+}
+
+
+void Allocate_Device_Matrix (reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+
+ //Allocate space for the sparse Matrix entries here.
+ system->max_sparse_matrix_entries =
+ Estimate_Device_Matrix( system, control, data, workspace, lists, out_control );
+ dev_workspace->H.n = system->N ;
+ dev_workspace->H.m = system->N * system->max_sparse_matrix_entries;
+ Cuda_Init_Sparse_Matrix( &dev_workspace->H, system->max_sparse_matrix_entries * system->N, system->N );
+
+#ifdef __CUDA_MEM__
+ fprintf( stderr, "Device memory allocated: sparse matrix= %ld (MB)\n",
+ system->max_sparse_matrix_entries * system->N * sizeof(sparse_matrix_entry) / (1024*1024) );
+#endif
+}
+
+
+void Cuda_Init_Lists( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, num_nbrs, num_hbonds, num_bonds, num_3body, Htop;
+ int *hb_top, *bond_top;
+
+ real t_start, t_elapsed;
+
+ grid *g = &( system->g );
+ int *d_indices = (int *) scratch;
+ int total = g->ncell[0] * g->ncell[1] * g->ncell[2];
+
+ cuda_memset( d_indices, 0, INT_SIZE * system->N, RES_SCRATCH );
+
+#ifdef __BUILD_DEBUG__
+ for (int i = 0; i < g->max_nbrs; i ++)
+ {
+ if ((g->nbrs[i][0] >= g->ncell[0]) ||
+ (g->nbrs[i][1] >= g->ncell[1]) ||
+ (g->nbrs[i][2] >= g->ncell[2]) )
+ {
+ fprintf( stderr, " Grid Incorrectly built.... \n" );
+ exit( 1 );
+ }
+
+ }
+#endif
+
+ dim3 blockspergrid( system->g.ncell[0], system->g.ncell[1], system->g.ncell[2] );
+ dim3 threadsperblock( system->g.max_atoms );
+
+#ifdef __BUILD_DEBUG__
+ fprintf( stderr, "Blocks per grid (%d %d %d)\n", system->g.ncell[0], system->g.ncell[1], system->g.ncell[2] );
+ fprintf( stderr, "Estimate Num Neighbors with threads per block as %d \n", system->d_g.max_atoms );
+ fprintf( stderr, "Max nbrs %d \n", system->d_g.max_nbrs );
+#endif
+
+ //First Bin atoms and they sync the host and the device for the grid.
+ //This will copy the atoms from host to device.
+ Cuda_Bin_Atoms( system, workspace );
+ Sync_Host_Device_Grid( &system->g, &system->d_g, cudaMemcpyHostToDevice );
+
+ k_Estimate_NumNeighbors<<<blockspergrid, threadsperblock >>>
+ (system->d_atoms, system->d_g, system->d_box,
+ (control_params *)control->d_control, d_indices);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ int *nbrs_indices = (int *) malloc( INT_SIZE * (system->N+1) );
+ memset( nbrs_indices , 0, INT_SIZE * (system->N + 1) );
+
+ nbrs_indices [0] = 0;
+ copy_host_device( &nbrs_indices [1], d_indices, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__ );
+
+ for (int i = 1; i <= system->N; i++)
+ {
+ nbrs_indices [i] += nbrs_indices [i-1];
+ }
+
+ num_nbrs = nbrs_indices [system->N] ;
+ system->num_nbrs = num_nbrs;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Total neighbors %d \n", nbrs_indices[system->N]);
+ fprintf (stderr, "Corrected Total neighbors %d \n", num_nbrs);
+#endif
+
+ list *far_nbrs = (dev_lists + FAR_NBRS);
+ if( !Cuda_Make_List(system->N, num_nbrs, TYP_FAR_NEIGHBOR, far_nbrs) ) {
+ fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
+ exit( INIT_ERR );
+ }
+
+#ifdef __CUDA_MEM__
+ fprintf( stderr, "Device memory allocated: far_nbrs = %ld (MB)\n",
+ num_nbrs * sizeof(far_neighbor_data) / (1024*1024) );
+#endif
+
+ copy_host_device( nbrs_indices, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyHostToDevice, __LINE__ );
+ copy_host_device( nbrs_indices, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyHostToDevice, __LINE__ );
+ Cuda_Generate_Neighbor_Lists( system, workspace, control, FALSE );
+
+#ifdef __BUILD_DEBUG__
+
+ int *end = (int *)malloc( sizeof (int) * system->N );
+ int *start = (int *) malloc( sizeof (int) * system->N );
+
+ copy_host_device( start, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, 0 );
+ copy_host_device( end, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, 0 );
+
+ far_neighbor_data *far_data = (far_neighbor_data *)
+ malloc( FAR_NEIGHBOR_SIZE * num_nbrs );
+ copy_host_device( far_data, far_nbrs->select.far_nbr_list,
+ FAR_NEIGHBOR_SIZE * num_nbrs, cudaMemcpyDeviceToHost, 0 );
+
+ compare_far_neighbors( nbrs_indices, start, end, far_data, *lists + FAR_NBRS, system->N );
+
+ free( start );
+ free( end );
+#endif
+
+ int *output, size;
+ size = INT_SIZE * 2 * system->N + 2;
+ output = (int *) malloc (size);
+ Cuda_Estimate_Storage_Sizes( system, control, output );
+
+ Htop = output[0];
+ num_3body = output[1];
+ hb_top = &output[ 2 ];
+ bond_top = &output[ 2 + system->N ];
+
+#ifdef __DEBUG_CUDA__
+ int max_hbonds = 0;
+ int min_hbonds = 1000;
+ int max_bonds = 0;
+ int min_bonds = 1000;
+
+ for (int i = 0; i < system->N; i++)
+ {
+ if ( max_hbonds < hb_top[i])
+ {
+ max_hbonds = hb_top[i];
+ }
+ if (min_hbonds > hb_top[i])
+ {
+ min_hbonds = hb_top[i];
+ }
+
+ if (max_bonds < bond_top [i])
+ {
+ max_bonds = bond_top[i];
+ }
+ if (min_bonds > bond_top[i])
+ {
+ min_bonds = bond_top[i];
+ }
+ }
+
+ fprintf( stderr, "Max Hbonds %d min Hbonds %d \n", max_hbonds, min_hbonds );
+ fprintf( stderr, "Max bonds %d min bonds %d \n", max_bonds, min_bonds );
+ fprintf( stderr, "Device HTop --> %d and num_3body --> %d \n", Htop, num_3body );
+#endif
+
+ Allocate_Device_Matrix( system, control, data, workspace, lists, out_control );
+
+ dev_workspace->num_H = 0;
+
+ if( control->hb_cut > 0 )
+ {
+
+ int *hbond_index = (int *) malloc ( INT_SIZE * system->N );
+ // init H indexes
+ num_hbonds = 0;
+ for( i = 0; i < system->N; ++i )
+ {
+ if( system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 1 ||
+ system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 2 ) // H atom
+ {
+ //hbond_index[i] = workspace->num_H++;
+ hbond_index[i] = num_hbonds ++;
+ }
+ else
+ {
+ hbond_index[i] = -1;
+ }
+ }
+
+ copy_host_device( hbond_index, dev_workspace->hbond_index,
+ system->N * INT_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_HBOND_INDEX );
+ dev_workspace->num_H = num_hbonds;
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Device num_H --> %d \n", dev_workspace->num_H );
+#endif
+
+ Cuda_Allocate_HBond_List( system->N, dev_workspace->num_H, dev_workspace->hbond_index,
+ hb_top, (dev_lists+HBONDS) );
+ num_hbonds = hb_top[system->N-1];
+ system->num_hbonds = num_hbonds;
+
+#ifdef __CUDA_MEM__
+ fprintf( stderr, "Device memory allocated: Hydrogen Bonds list: %ld (MB) \n",
+ sizeof (hbond_data) * num_hbonds / (1024*1024) );
+#endif
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Device Total number of HBonds --> %d \n", num_hbonds );
+#endif
+
+ free( hbond_index );
+ }
+
+ // bonds list
+ Cuda_Allocate_Bond_List( system->N, bond_top, dev_lists+BONDS );
+ num_bonds = bond_top[system->N-1];
+ system->num_bonds = num_bonds;
+
+#ifdef __CUDA_MEM__
+ fprintf( stderr, "Device memory allocated: Bonds list: %ld (MB) \n",
+ sizeof (bond_data) * num_bonds / (1024*1024));
+#endif
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Device Total Bonds --> %d \n", num_bonds );
+#endif
+
+ // system->max_thb_intrs = num_3body;
+ // 3bodies list
+ //if(!Cuda_Make_List(num_bonds, num_bonds * MAX_THREE_BODIES, TYP_THREE_BODY, dev_lists + THREE_BODIES)) {
+ // fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
+ // exit( INIT_ERR );
+ //}
+
+ //fprintf( stderr, "***memory allocated: three_body = %ldMB\n",
+ // num_bonds * MAX_THREE_BODIES *sizeof(three_body_interaction_data) / (1024*1024) );
+ //fprintf (stderr, "size of (three_body_interaction_data) : %d \n", sizeof (three_body_interaction_data));
+
+ free( output );
+ free( nbrs_indices );
+}
+
+
+void Cuda_Initialize( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace, list **lists,
+ output_controls *out_control, evolve_function *Evolve )
+{
+ compute_blocks( &BLOCKS, &BLOCK_SIZE, system->N );
+ compute_nearest_pow_2( BLOCKS, &BLOCKS_POW_2 );
+
+ //MATVEC_BLOCKS = system.N;
+ //MATVEC_BLOCK_SIZE = 32;
+
+ MATVEC_BLOCKS = (system->N * MATVEC_THREADS_PER_ROW / MATVEC_BLOCK_SIZE) +
+ ((system->N * MATVEC_THREADS_PER_ROW / MATVEC_BLOCK_SIZE) == 0 ? 0 : 1);
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, " MATVEC Blocks : %d, Block_Size : %d \n", MATVEC_BLOCKS, MATVEC_BLOCK_SIZE );
+ fprintf( stderr, " Blocks : %d, Blocks_Pow_2 : %d, Block_Size : %d \n", BLOCKS, BLOCKS_POW_2, BLOCK_SIZE );
+ fprintf( stderr, " Size of far neighbor data %d \n", sizeof (far_neighbor_data) );
+ fprintf( stderr, " Size of reax_atom %d \n", sizeof (reax_atom) );
+ fprintf( stderr, " size of sparse matrix entry %d \n", sizeof (sparse_matrix_entry) );
+ fprintf( stderr, " TOTAL NUMBER OF ATOMS IN THE SYSTEM --> %d \n", system.N );
+#endif
+
+ Randomize( );
+
+ Cuda_Init_Scratch( );
+
+ //System
+ Cuda_Init_System( system );
+ Sync_Host_Device_Sys( system, cudaMemcpyHostToDevice );
+ Cuda_Init_System( system, control, data );
+
+ //Simulation Data
+ copy_host_device( system->atoms, system->d_atoms, REAX_ATOM_SIZE * system->N ,
+ cudaMemcpyHostToDevice, RES_SYSTEM_ATOMS );
+ Cuda_Init_Simulation_Data( data );
+ //Sync_Host_Device_Data( data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice );
+ Cuda_Init_Simulation_Data( system, control, data, out_control, Evolve );
+ Sync_Host_Device_Data( data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice );
+
+ //static storage
+ Cuda_Init_Workspace_System( system, dev_workspace );
+ Cuda_Init_Workspace( system, control, dev_workspace );
+ Cuda_Init_Workspace_Device( workspace );
+
+ //control
+ Cuda_Init_Control( control );
+
+ //Grid
+ Cuda_Init_Grid( &system->g, &system->d_g );
+
+ //lists
+ Cuda_Init_Lists( system, control, data, workspace, lists, out_control );
+
+ Init_Out_Controls( system, control, workspace, out_control );
+
+ if( control->tabulate )
+ {
+ real start, end;
+ start = Get_Time( );
+ Make_LR_Lookup_Table( system, control );
+ copy_LR_table_to_device( system, control );
+ end = Get_Timing_Info( start );
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Done copying the LR table to the device ---> %f \n", end );
+#endif
+ }
+}
diff --git a/PuReMD-GPU/src/cuda_init_md.h b/PuReMD-GPU/src/cuda_init_md.h
new file mode 100644
index 0000000000000000000000000000000000000000..0a6544b017d31ca0a82651840682c3aae49d1b11
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_init_md.h
@@ -0,0 +1,40 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_INIT_MD_H_
+#define __CUDA_INIT_MD_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Initialize( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls*, evolve_function* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
+
diff --git a/PuReMD-GPU/src/cuda_integrate.cu b/PuReMD-GPU/src/cuda_integrate.cu
new file mode 100644
index 0000000000000000000000000000000000000000..cba0b79c39b4f9b66e5b506d11dcffb81adc488d
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_integrate.cu
@@ -0,0 +1,517 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_integrate.h"
+
+#include "allocate.h"
+#include "box.h"
+#include "forces.h"
+#include "grid.h"
+#include "print_utils.h"
+#include "reset_utils.h"
+#include "system_props.h"
+#include "vector.h"
+#include "list.h"
+
+#include "cuda_utils.h"
+#include "cuda_reduction.h"
+#include "cuda_allocate.h"
+#include "cuda_forces.h"
+#include "cuda_grid.h"
+#include "cuda_neighbors.h"
+#include "cuda_QEq.h"
+#include "cuda_reset_utils.h"
+#include "cuda_system_props.h"
+#include "validation.h"
+
+
+GLOBAL void Cuda_Velocity_Verlet_NVE_atoms1 (reax_atom *atoms,
+ single_body_parameters *sbp,
+ simulation_box *box,
+ int N, real dt)
+{
+ real inv_m, dt_sqr;
+ rvec dx;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ dt_sqr = SQR(dt);
+ //for( i = 0; i < system->N; i++ ) {
+ inv_m = 1.0 / sbp[atoms[i].type].mass;
+
+ rvec_ScaledSum( dx, dt, atoms[i].v,
+ 0.5 * dt_sqr * -F_CONV * inv_m, atoms[i].f );
+ Inc_on_T3( atoms[i].x, dx, box );
+
+ rvec_ScaledAdd( atoms[i].v,
+ 0.5 * dt * -F_CONV * inv_m, atoms[i].f );
+ //}
+}
+
+
+GLOBAL void Cuda_Velocity_Verlet_NVE_atoms2 (reax_atom *atoms, single_body_parameters *sbp, int N, real dt)
+{
+ real inv_m;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ //for( i = 0; i < system->N; i++ ) {
+ inv_m = 1.0 / sbp[atoms[i].type].mass;
+ rvec_ScaledAdd( atoms[i].v,
+ 0.5 * dt * -F_CONV * inv_m, atoms[i].f );
+ //}
+}
+
+
+void Cuda_Velocity_Verlet_NVE(reax_system* system, control_params* control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, steps, renbr;
+ real inv_m, dt, dt_sqr;
+ rvec dx;
+ int blocks, block_size;
+
+ dt = control->dt;
+ dt_sqr = SQR(dt);
+ steps = data->step - data->prev_steps;
+ renbr = (steps % control->reneighbor == 0);
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d: ", data->step );
+#endif
+
+ compute_blocks (&blocks, &block_size, system->N);
+ Cuda_Velocity_Verlet_NVE_atoms1 <<<blocks, block_size>>>
+ (system->d_atoms, system->reaxprm.d_sbp,
+ (simulation_box *)system->d_box, system->N, dt);
+ cudaThreadSynchronize ();
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "verlet1 - ");
+#endif
+
+ Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
+ Cuda_Reset( system, control, data, workspace, lists );
+
+ if( renbr ) {
+ Cuda_Generate_Neighbor_Lists (system, dev_workspace, control, TRUE);
+ }
+
+ Cuda_Compute_Forces( system, control, data, workspace, lists, out_control );
+
+ Cuda_Velocity_Verlet_NVE_atoms2<<<blocks, block_size>>>
+ (system->d_atoms, system->reaxprm.d_sbp, system->N, dt);
+ cudaThreadSynchronize ();
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "verlet2\n");
+#endif
+}
+
+
+GLOBAL void Compute_X_t_dt (real dt, real dt_sqr, thermostat p_therm,
+ reax_atom *atoms, single_body_parameters *sbp,
+ simulation_box *box,
+ static_storage p_workspace, int N)
+{
+
+ real inv_m;
+ rvec dx;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= N) return;
+
+ static_storage *workspace = &p_workspace;
+ thermostat *therm = &p_therm;
+
+ /* Compute x(t + dt) and copy old forces */
+ //for (i=0; i < system->N; i++) {
+ inv_m = 1.0 / sbp[atoms[i].type].mass;
+
+ rvec_ScaledSum( dx, dt - 0.5 * dt_sqr * therm->v_xi, atoms[i].v,
+ 0.5 * dt_sqr * inv_m * -F_CONV, atoms[i].f );
+
+ Inc_on_T3( atoms[i].x, dx, box );
+
+ rvec_Copy( workspace->f_old[i], atoms[i].f );
+ //}
+
+}
+
+
+GLOBAL void Update_Velocity (reax_atom *atoms, single_body_parameters *sbp,
+ static_storage p_workspace, real dt, thermostat p_therm,
+ int N)
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ real inv_m;
+ static_storage *workspace = &p_workspace;
+ thermostat *therm = &p_therm;
+
+ //for( i = 0; i < system->N; ++i ) {
+ inv_m = 1.0 / sbp[atoms[i].type].mass;
+
+ rvec_Scale( workspace->v_const[i],
+ 1.0 - 0.5 * dt * therm->v_xi, atoms[i].v );
+ rvec_ScaledAdd( workspace->v_const[i],
+ 0.5 * dt * inv_m * -F_CONV, workspace->f_old[i] );
+ rvec_ScaledAdd( workspace->v_const[i],
+ 0.5 * dt * inv_m * -F_CONV, atoms[i].f );
+ //}
+}
+
+
+GLOBAL void E_Kin_Reduction (reax_atom *atoms, static_storage p_workspace,
+ single_body_parameters *sbp,
+ real *per_block_results, real coef_v, const size_t n)
+{
+ extern __shared__ real sdata[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real x = 0;
+ static_storage *workspace = &p_workspace;
+
+ if(i < n)
+ {
+ rvec_Scale( atoms[i].v, coef_v, workspace->v_const[i] );
+ x = ( 0.5 * sbp[atoms[i].type].mass *
+ rvec_Dot( atoms[i].v, atoms[i].v ) );
+ }
+ sdata[threadIdx.x] = x;
+ __syncthreads();
+
+ for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
+ {
+ if(threadIdx.x < offset)
+ {
+ sdata[threadIdx.x] += sdata[threadIdx.x + offset];
+ }
+
+ __syncthreads();
+ }
+
+ if(threadIdx.x == 0)
+ {
+ per_block_results[blockIdx.x] = sdata[0];
+ }
+}
+
+
+void Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
+ control_params* control,
+ simulation_data *data,
+ static_storage *workspace,
+ list **lists,
+ output_controls *out_control )
+{
+ int i, itr, steps, renbr;
+ real inv_m, coef_v, dt, dt_sqr;
+ real E_kin_new, G_xi_new, v_xi_new, v_xi_old;
+ rvec dx;
+ thermostat *therm;
+
+ real *results = (real *)scratch;
+
+ dt = control->dt;
+ dt_sqr = SQR( dt );
+ therm = &( data->therm );
+ steps = data->step - data->prev_steps;
+ renbr = (steps % control->reneighbor == 0);
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Device: Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein --> coef to update velocity --> %6.10f\n", therm->v_xi_old);
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d: ", data->step );
+#endif
+
+ Compute_X_t_dt <<< BLOCKS, BLOCK_SIZE >>>
+ (dt, dt_sqr, data->therm, system->d_atoms,
+ system->reaxprm.d_sbp, system->d_box, *dev_workspace, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ /* Compute xi(t + dt) */
+ therm->xi += ( therm->v_xi * dt + 0.5 * dt_sqr * therm->G_xi );
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "verlet1 - " );
+#endif
+
+ Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
+ Cuda_Reset( system, control, data, workspace, lists );
+
+ if( renbr )
+ {
+ //generate_neighbor_lists here
+ Cuda_Generate_Neighbor_Lists (system, dev_workspace, control, TRUE);
+ }
+
+ /* Calculate Forces at time (t + dt) */
+ Cuda_Compute_Forces( system,control,data, workspace, lists, out_control );
+
+ /* Compute iteration constants for each atom's velocity */
+ Update_Velocity <<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, system->reaxprm.d_sbp, *dev_workspace,
+ dt, *therm, system->N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ v_xi_new = therm->v_xi_old + 2.0 * dt * therm->G_xi;
+ E_kin_new = G_xi_new = v_xi_old = 0;
+ itr = 0;
+ do {
+ itr++;
+
+ /* new values become old in this iteration */
+ v_xi_old = v_xi_new;
+ coef_v = 1.0 / (1.0 + 0.5 * dt * v_xi_old);
+ E_kin_new = 0;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Device: coef to update velocity --> %6.10f, %6.10f, %6.10f\n", coef_v, dt, therm->v_xi_old);
+#endif
+
+ /*reduction for the E_Kin_new here*/
+ cuda_memset (results, 0, 2 * BLOCK_SIZE * REAL_SIZE, RES_SCRATCH );
+ E_Kin_Reduction <<< BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ (system->d_atoms, *dev_workspace, system->reaxprm.d_sbp,
+ results, coef_v, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ (results, results + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ copy_host_device (&E_kin_new, results + BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, RES_SCRATCH );
+
+ G_xi_new = control->Tau_T * ( 2.0 * E_kin_new -
+ data->N_f * K_B * control->T );
+ v_xi_new = therm->v_xi + 0.5 * dt * ( therm->G_xi + G_xi_new );
+
+#if defined(DEBUG)
+ fprintf( stderr, "itr%d: G_xi_new = %f, v_xi_new = %f, v_xi_old = %f\n",
+ itr, G_xi_new, v_xi_new, v_xi_old );
+#endif
+ }
+ while( fabs(v_xi_new - v_xi_old ) > 1e-5 );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Iteration Count in NVE --> %d \n", itr );
+#endif
+
+ therm->v_xi_old = therm->v_xi;
+ therm->v_xi = v_xi_new;
+ therm->G_xi = G_xi_new;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr,"vel scale\n" );
+#endif
+}
+
+
+GLOBAL void ker_update_velocity_1 (reax_atom *atoms,
+ single_body_parameters *sbp,
+ real dt,
+ simulation_box *box,
+ int N)
+{
+ real inv_m;
+ rvec dx;
+ reax_atom *atom;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if ( i >= N ) return;
+
+ /* velocity verlet, 1st part */
+ //for( i = 0; i < system->n; i++ ) {
+ atom = &(atoms[i]);
+ inv_m = 1.0 / sbp[atom->type].mass;
+ /* Compute x(t + dt) */
+ rvec_ScaledSum( dx, dt, atom->v, 0.5 * -F_CONV * inv_m * SQR(dt), atom->f );
+ rvec_Add( atom->x, dx );
+
+ /* Metin's suggestion to rebox the atoms */
+ /* bNVT fix */
+ Inc_on_T3( atoms[i].x, dx, box );
+ /* bNVT fix */
+
+ /* Compute v(t + dt/2) */
+ rvec_ScaledAdd( atom->v, 0.5 * -F_CONV * inv_m * dt, atom->f );
+ //}
+}
+
+
+void bNVT_update_velocity_part1 (reax_system *system, simulation_box *box, real dt)
+{
+ ker_update_velocity_1 <<< BLOCKS, BLOCK_SIZE>>>
+ (system->d_atoms, system->reaxprm.d_sbp, dt, box, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
+
+
+GLOBAL void ker_update_velocity_2 (reax_atom *atoms,
+ single_body_parameters *sbp,
+ real dt,
+ int N)
+{
+ reax_atom *atom;
+ real inv_m;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if ( i >= N ) return;
+
+ /* velocity verlet, 2nd part */
+ //for( i = 0; i < system->n; i++ ) {
+ atom = &(atoms[i]);
+ inv_m = 1.0 / sbp[atom->type].mass;
+ /* Compute v(t + dt) */
+ rvec_ScaledAdd( atom->v, 0.5 * dt * -F_CONV * inv_m, atom->f );
+ //}
+}
+
+
+void bNVT_update_velocity_part2 (reax_system *system, real dt)
+{
+ ker_update_velocity_2 <<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, system->reaxprm.d_sbp, dt, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
+
+
+GLOBAL void ker_scale_velocities (reax_atom *atoms, real lambda, int N)
+{
+ reax_atom *atom;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if ( i >= N ) return;
+
+ /* Scale velocities and positions at t+dt */
+ //for( i = 0; i < system->n; ++i ) {
+ atom = &(atoms[i]);
+ rvec_Scale( atom->v, lambda, atom->v );
+ //}
+}
+
+
+void bNVT_scale_velocities (reax_system *system, real lambda)
+{
+ ker_scale_velocities <<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, lambda, system->N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
+
+
+void Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* system,
+ control_params* control,
+ simulation_data *data,
+ static_storage *workspace,
+ list **lists,
+ output_controls *out_control
+ )
+{
+ int i, steps, renbr;
+ real inv_m, dt, lambda;
+ rvec dx;
+ reax_atom *atom;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d\n", data->step );
+#endif
+ dt = control->dt;
+ steps = data->step - data->prev_steps;
+ renbr = (steps % control->reneighbor == 0);
+
+ /* velocity verlet, 1st part
+ for( i = 0; i < system->N; i++ ) {
+ atom = &(system->atoms[i]);
+ inv_m = 1.0 / system->reaxprm.sbp[atom->type].mass;
+ // Compute x(t + dt)
+ rvec_ScaledSum( dx, dt, atom->v, 0.5 * -F_CONV * inv_m * SQR(dt), atom->f );
+ rvec_Add( atom->x, dx );
+ // Compute v(t + dt/2)
+ rvec_ScaledAdd( atom->v, 0.5 * -F_CONV * inv_m * dt, atom->f );
+ }
+ */
+ bNVT_update_velocity_part1 (system, (simulation_box *) system->d_box, dt);
+
+#if defined(DEBUG_FOCUS)
+ fprintf(stderr, "step%d: verlet1 done\n", data->step);
+#endif
+
+ Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
+ Cuda_Reset( system, control, data, workspace, lists );
+
+ if( renbr ) {
+ Cuda_Generate_Neighbor_Lists( system, workspace, control, TRUE);
+ }
+
+ Cuda_Compute_Forces( system, control, data, workspace,
+ lists, out_control );
+
+ /* velocity verlet, 2nd part
+ for( i = 0; i < system->N; i++ ) {
+ atom = &(system->atoms[i]);
+ inv_m = 1.0 / system->reaxprm.sbp[atom->type].mass;
+ // Compute v(t + dt)
+ rvec_ScaledAdd( atom->v, 0.5 * dt * -F_CONV * inv_m, atom->f );
+ }
+ */
+ bNVT_update_velocity_part2 (system, dt);
+#if defined(DEBUG_FOCUS)
+ fprintf(stderr, "step%d: verlet2 done\n", data->step);
+#endif
+
+ /* temperature scaler */
+ Cuda_Compute_Kinetic_Energy( system, data );
+ //get the latest temperature from the device to the host.
+ copy_host_device (&data->therm, &((simulation_data *)data->d_simulation_data)->therm,
+ sizeof (thermostat), cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
+
+ lambda = 1.0 + (dt / control->Tau_T) * (control->T / data->therm.T - 1.0);
+ if( lambda < MIN_dT )
+ lambda = MIN_dT;
+ else if (lambda > MAX_dT )
+ lambda = MAX_dT;
+ lambda = SQRT( lambda );
+
+ //fprintf (stderr, "step:%d lambda -> %f \n", data->step, lambda);
+
+ /* Scale velocities and positions at t+dt
+ for( i = 0; i < system->N; ++i ) {
+ atom = &(system->atoms[i]);
+ rvec_Scale( atom->v, lambda, atom->v );
+ }
+ */
+ bNVT_scale_velocities (system, lambda);
+ Cuda_Compute_Kinetic_Energy( system, data );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d: scaled velocities\n",
+ data->step );
+#endif
+
+}
+
+
diff --git a/PuReMD-GPU/src/cuda_integrate.h b/PuReMD-GPU/src/cuda_integrate.h
new file mode 100644
index 0000000000000000000000000000000000000000..959b6684fef618b86252a5606b6feb4a6d093f15
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_integrate.h
@@ -0,0 +1,46 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_INTEGRATE_H_
+#define __CUDA_INTEGRATE_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Velocity_Verlet_NVE( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls* );
+void Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein( reax_system*, control_params*,
+ simulation_data*, static_storage*,
+ list**, output_controls* );
+void Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* , control_params* ,
+ simulation_data *, static_storage *,
+ list **, output_controls * );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
+
diff --git a/PuReMD-GPU/src/cuda_lin_alg.cu b/PuReMD-GPU/src/cuda_lin_alg.cu
new file mode 100644
index 0000000000000000000000000000000000000000..5dc9eb3518ed8b46174838a498cfdba62d34e989
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_lin_alg.cu
@@ -0,0 +1,589 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_lin_alg.h"
+
+#include "list.h"
+#include "vector.h"
+#include "index_utils.h"
+
+#include "cuda_copy.h"
+#include "cuda_utils.h"
+#include "cuda_reduction.h"
+#include "system_props.h"
+
+#include "cublas_v2.h"
+#include "cusparse_v2.h"
+
+
+//one thread per row
+GLOBAL void Cuda_Matvec (sparse_matrix H, real *vec, real *results, int rows)
+{
+ real results_row = 0;
+ int col;
+ real val;
+
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if ( i >= rows) return;
+
+ for (int c = H.start[i]; c < H.end[i]; c++)
+ {
+ col = H.entries [c].j;
+ val = H.entries[c].val;
+
+ results_row += val * vec [col];
+ }
+
+ results [i] = results_row;
+}
+
+
+//32 thread warp per matrix row.
+//invoked as follows
+// <<< system->N, 32 >>>
+GLOBAL void Cuda_Matvec_csr (sparse_matrix H, real *vec, real *results, int num_rows)
+{
+ extern __shared__ real vals [];
+ int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
+ int warp_id = thread_id / 32;
+ int lane = thread_id & (32 - 1);
+
+ int row_start;
+ int row_end;
+
+ // one warp per row
+ //int row = warp_id;
+ int row = warp_id;
+ //if (row < num_rows)
+ {
+ vals[threadIdx.x] = 0;
+
+ if (row < num_rows) {
+ row_start = H.start[row];
+ row_end = H.end[row];
+
+ // compute running sum per thread
+ for(int jj = row_start + lane; jj < row_end; jj += 32)
+ vals[threadIdx.x] += H.entries[jj].val * vec [ H.entries[jj].j ];
+ //vals[threadIdx.x] += H.val[jj] * vec [ H.j[jj] ];
+ }
+
+ __syncthreads ();
+
+ // parallel reduction in shared memory
+ //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
+ if (lane < 16) vals[threadIdx.x] += vals[threadIdx.x + 16]; __syncthreads();
+ if (lane < 8) vals[threadIdx.x] += vals[threadIdx.x + 8]; __syncthreads ();
+ if (lane < 4) vals[threadIdx.x] += vals[threadIdx.x + 4]; __syncthreads ();
+ if (lane < 2) vals[threadIdx.x] += vals[threadIdx.x + 2]; __syncthreads ();
+ if (lane < 1) vals[threadIdx.x] += vals[threadIdx.x + 1]; __syncthreads ();
+
+ // first thread writes the result
+ if (lane == 0 && row < num_rows)
+ results[row] = vals[threadIdx.x];
+ }
+}
+
+
+GLOBAL void GMRES_Diagonal_Preconditioner (real *b_proc, real *b, real *Hdia_inv, int entries)
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= entries) return;
+
+ b_proc [i] = b[i] * Hdia_inv[i];
+}
+
+
+GLOBAL void GMRES_Givens_Rotation (int j, real *h, real *hc, real *hs, real g_j, real *output)
+{
+ real tmp1, tmp2, cc;
+
+ for( int i = 0; i <= j; i++ ) {
+ if( i == j ) {
+ cc = SQRT( SQR(h[ index_wkspace_res (j,j) ])+SQR(h[ index_wkspace_res (j+1,j) ]) );
+ hc[j] = h[ index_wkspace_res (j,j) ] / cc;
+ hs[j] = h[ index_wkspace_res (j+1,j) ] / cc;
+ }
+
+ tmp1 = hc[i] * h[ index_wkspace_res (i,j) ] + hs[i] * h[ index_wkspace_res (i+1,j) ];
+ tmp2 = -hs[i] * h[ index_wkspace_res (i,j) ] + hc[i] * h[ index_wkspace_res (i+1,j) ];
+
+ h[ index_wkspace_res (i,j) ] = tmp1;
+ h[ index_wkspace_res (i+1,j) ] = tmp2;
+ }
+
+ /* apply Givens rotations to the rhs as well */
+ tmp1 = hc[j] * g_j;
+ tmp2 = -hs[j] * g_j;
+
+ output[0] = tmp1;
+ output[1] = tmp2;
+}
+
+
+GLOBAL void GMRES_BackSubstitution (int j, real *g, real *h, real *y)
+{
+ real temp;
+ for( int i = j-1; i >= 0; i-- ) {
+ temp = g[i];
+ for( int k = j-1; k > i; k-- )
+ temp -= h[ index_wkspace_res (i,k) ] * y[k];
+
+ y[i] = temp / h[ index_wkspace_res (i,i) ];
+ }
+}
+
+
+int Cuda_GMRES( static_storage *workspace, real *b, real tol, real *x )
+{
+ int i, j, k, itr, N;
+ real cc, tmp1, tmp2, temp, bnorm;
+ real v_add_tmp;
+ sparse_matrix *H = &workspace->H;
+ real t_start, t_elapsed;
+ real *spad = (real *)scratch;
+ real *g = (real *) calloc ((RESTART+1), REAL_SIZE);
+
+ N = H->n;
+
+ cuda_memset(spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
+
+ Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (b, spad, H->n, INITIAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, spad + BLOCKS_POW_2, BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ copy_host_device( &bnorm, spad + BLOCKS_POW_2, REAL_SIZE,
+ cudaMemcpyDeviceToHost, __LINE__);
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Norm of the array is %e \n", bnorm );
+#endif
+
+ /* apply the diagonal pre-conditioner to rhs */
+ GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
+ (workspace->b_prc, b, workspace->Hdia_inv, N);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ /* GMRES outer-loop */
+ for( itr = 0; itr < MAX_ITR; ++itr ) {
+ /* calculate r0 */
+ //Sparse_MatVec( H, x, workspace->b_prm );
+ Cuda_Matvec_csr <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>>
+ ( *H, x, workspace->b_prm, N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ GMRES_Diagonal_Preconditioner <<< BLOCKS, BLOCK_SIZE >>>
+ (workspace->b_prm, workspace->b_prm, workspace->Hdia_inv, N);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ Cuda_Vector_Sum <<< BLOCKS, BLOCK_SIZE >>>
+ (&workspace->v[ index_wkspace_sys (0,0,N) ], 1.,
+ workspace->b_prc, -1., workspace->b_prm, N);
+ cudaThreadSynchronize();
+ cudaCheckError ();
+
+ //workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system->N)], N );
+ {
+ cuda_memset( spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
+
+ Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (&workspace->v [index_wkspace_sys (0, 0, N)], spad, N, INITIAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, &workspace->g[0], BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ copy_host_device( g, workspace->g, REAL_SIZE,
+ cudaMemcpyDeviceToHost, RES_STORAGE_G);
+ }
+
+ Cuda_Vector_Scale<<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[ index_wkspace_sys (0,0,N) ], 1.0/g[0],
+ &workspace->v[index_wkspace_sys(0,0,N)], N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ /* GMRES inner-loop */
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr,
+ " Inner loop inputs bnorm : %f , tol : %f g[j] : %f \n", bnorm,
+ tol, g[0] );
+#endif
+
+ for( j = 0; j < RESTART && fabs(g[j]) / bnorm > tol; j++ ) {
+ /* matvec */
+ //Sparse_MatVec( H, &workspace->v[index_wkspace_sys(j,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)] );
+ Cuda_Matvec_csr<<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>>
+ ( *H, &workspace->v[ index_wkspace_sys (j, 0, N)],
+ &workspace->v[ index_wkspace_sys (j+1, 0, N) ], N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ GMRES_Diagonal_Preconditioner<<<BLOCKS, BLOCK_SIZE>>>
+ (&workspace->v[ index_wkspace_sys (j+1,0,N) ],
+ &workspace->v[ index_wkspace_sys( j+1,0,N) ],
+ workspace->Hdia_inv, N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+
+ /* apply modified Gram-Schmidt to orthogonalize the new residual */
+ for( i = 0; i <= j; i++ )
+ {
+ Cuda_Dot <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (&workspace->v[index_wkspace_sys(i,0,N)],
+ &workspace->v[index_wkspace_sys(j+1,0,N)], spad, N);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ Cuda_reduction<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, &workspace->h[ index_wkspace_res (i,j) ], BLOCKS_POW_2);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+
+ Cuda_Vector_Add<<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[index_wkspace_sys(j+1,0,N)],
+ -v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+ }
+
+ //workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ cuda_memset(spad, 0, REAL_SIZE * N * 2, RES_SCRATCH );
+
+ Cuda_Norm<<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (&workspace->v[index_wkspace_sys(j+1,0,N)], spad, N, INITIAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ Cuda_Norm<<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
+ (spad, &workspace->h[ index_wkspace_res (j+1,j) ], BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ copy_host_device(&v_add_tmp,
+ &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+
+ Cuda_Vector_Scale<<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[index_wkspace_sys(j+1,0,N)],
+ 1. / v_add_tmp, &workspace->v[index_wkspace_sys(j+1,0,N)], N );
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ /* Givens rotations on the upper-Hessenberg matrix to make it U */
+ GMRES_Givens_Rotation<<<1, 1>>>
+ (j, workspace->h, workspace->hc, workspace->hs, g[j], spad);
+ cudaThreadSynchronize();
+ cudaCheckError();
+ copy_host_device(&g[j], spad, 2 * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ }
+
+ copy_host_device(g, workspace->g, (RESTART+1)*REAL_SIZE,
+ cudaMemcpyHostToDevice, __LINE__);
+
+ /* solve Hy = g.
+ H is now upper-triangular, do back-substitution */
+ copy_host_device(g, spad, (RESTART+1) * REAL_SIZE,
+ cudaMemcpyHostToDevice, RES_STORAGE_G);
+ GMRES_BackSubstitution<<<1, 1>>>
+ (j, spad, workspace->h, workspace->y);
+ cudaThreadSynchronize();
+ cudaCheckError();
+
+ /* update x = x_0 + Vy */
+ for( i = 0; i < j; i++ )
+ {
+ copy_host_device(&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ Cuda_Vector_Add <<<BLOCKS, BLOCK_SIZE>>>
+ ( x, v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
+ cudaThreadSynchronize ();
+ cudaCheckError();
+ }
+
+ /* stopping condition */
+ if( fabs(g[j]) / bnorm <= tol )
+ {
+ break;
+ }
+ }
+
+ if( itr >= MAX_ITR ) {
+ fprintf( stderr, "GMRES convergence failed\n" );
+ return itr * (RESTART+1) + j + 1;
+ }
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " GPU values itr : %d, RESTART: %d, j: %d \n", itr, RESTART, j);
+#endif
+
+ return itr * (RESTART+1) + j + 1;
+}
+
+
+int Cublas_GMRES(reax_system *system, static_storage *workspace, real *b, real tol, real *x )
+{
+
+ real CSR_ALPHA = 1, CSR_BETA = 0;
+
+ int i, j, k, itr, N;
+ real cc, tmp1, tmp2, temp, bnorm;
+ real v_add_tmp;
+ sparse_matrix *H = &workspace->H;
+ real t_start, t_elapsed;
+ real *spad = (real *)scratch;
+ real *g = (real *) calloc ((RESTART+1), REAL_SIZE);
+ cublasHandle_t cublasHandle;
+
+ N = H->n;
+
+ cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
+
+ /*
+ Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (b, spad, H->n, INITIAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, spad + BLOCKS_POW_2, BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ copy_host_device ( &bnorm, spad + BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ */
+
+ cublasCheckError (cublasDnrm2 ( cublasHandle, N, b, 1, &bnorm ));
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Norm of the array is %e \n", bnorm );
+#endif
+
+ /* apply the diagonal pre-conditioner to rhs */
+ GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
+ (workspace->b_prc, b, workspace->Hdia_inv, N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ /* GMRES outer-loop */
+ for( itr = 0; itr < MAX_ITR; ++itr ) {
+ /* calculate r0 */
+ //Sparse_MatVec( H, x, workspace->b_prm );
+ Cuda_Matvec_csr <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>> ( *H, x, workspace->b_prm, N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ GMRES_Diagonal_Preconditioner <<< BLOCKS, BLOCK_SIZE >>>
+ (workspace->b_prm, workspace->b_prm, workspace->Hdia_inv, N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ /*
+ Cuda_Vector_Sum <<< BLOCKS, BLOCK_SIZE >>>
+ (&workspace->v[ index_wkspace_sys (0,0,N) ], 1.,workspace->b_prc, -1., workspace->b_prm, N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+ cuda_memset (workspace->v, 0, REAL_SIZE * (RESTART+1) * N, RES_STORAGE_V);
+
+ double D_ONE = 1.;
+ double D_MINUS_ONE = -1.;
+ cublasCheckError (cublasDaxpy (cublasHandle, N, &D_ONE, workspace->b_prc, 1, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
+ cublasCheckError (cublasDaxpy (cublasHandle, N, &D_MINUS_ONE, workspace->b_prm, 1, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
+
+ //workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system->N)], N );
+ {
+ /*
+ cuda_memset (spad, 0, REAL_SIZE * H->n * 2, RES_SCRATCH );
+
+ Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (&workspace->v [index_wkspace_sys (0, 0, N)], spad, N, INITIAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->g[0], BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ copy_host_device( g, workspace->g, REAL_SIZE, cudaMemcpyDeviceToHost, RES_STORAGE_G);
+ */
+
+ cublasCheckError (cublasDnrm2 ( cublasHandle, N, &workspace->v [index_wkspace_sys (0, 0, N)], 1, g ));
+ copy_host_device( g, workspace->g, REAL_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_G);
+ }
+
+ /*
+ Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[ index_wkspace_sys (0,0,N) ], 1.0/g[0], &workspace->v[index_wkspace_sys(0,0,N)], N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+
+ double D_SCALE = 1.0 / g[0];
+ cublasCheckError (cublasDscal (cublasHandle, N, &D_SCALE, &workspace->v[ index_wkspace_sys (0,0,N) ], 1));
+
+
+ /* GMRES inner-loop */
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Inner loop inputs bnorm : %f , tol : %f g[j] : %f \n", bnorm, tol, g[0] );
+#endif
+ for( j = 0; j < RESTART && fabs(g[j]) / bnorm > tol; j++ ) {
+ /* matvec */
+ Cuda_Matvec_csr
+ <<<MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, REAL_SIZE * MATVEC_BLOCK_SIZE>>>
+ ( *H, &workspace->v[ index_wkspace_sys (j, 0, N)], &workspace->v[ index_wkspace_sys (j+1, 0, N) ], N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ GMRES_Diagonal_Preconditioner <<<BLOCKS, BLOCK_SIZE>>>
+ (&workspace->v[ index_wkspace_sys (j+1,0,N) ], &workspace->v[ index_wkspace_sys (j+1,0,N) ], workspace->Hdia_inv, N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+
+ /* apply modified Gram-Schmidt to orthogonalize the new residual */
+ for( i = 0; i <= j; i++ ) {
+
+ /*
+ Cuda_Dot <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (&workspace->v[index_wkspace_sys(i,0,N)], &workspace->v[index_wkspace_sys(j+1,0,N)], spad, N);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (i,j) ], BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ */
+
+ cublasCheckError (cublasDdot (cublasHandle, N, &workspace->v[index_wkspace_sys(i,0,N)], 1,
+ &workspace->v[index_wkspace_sys(j+1,0,N)], 1,
+ &v_add_tmp));
+ copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (i,j)], REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
+
+ /*
+ Cuda_Vector_Add <<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[index_wkspace_sys(j+1,0,N)],
+ -v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+
+ double NEG_V_ADD_TMP = -v_add_tmp;
+ cublasCheckError (cublasDaxpy (cublasHandle, N, &NEG_V_ADD_TMP, &workspace->v[index_wkspace_sys(i,0,N)], 1,
+ &workspace->v[index_wkspace_sys(j+1,0,N)], 1 ));
+ }
+
+
+ //workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ /*
+ cuda_memset (spad, 0, REAL_SIZE * N * 2, RES_SCRATCH );
+
+ Cuda_Norm <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>> (&workspace->v[index_wkspace_sys(j+1,0,N)], spad, N, INITIAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ Cuda_Norm <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>> (spad, &workspace->h[ index_wkspace_res (j+1,j) ], BLOCKS_POW_2, FINAL);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ */
+ cublasCheckError (cublasDnrm2 ( cublasHandle, N, &workspace->v [index_wkspace_sys (j+1, 0, N)], 1, &v_add_tmp ));
+ copy_host_device (&v_add_tmp, &workspace->h[ index_wkspace_res (j+1,j) ], REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
+
+
+ /*
+ Cuda_Vector_Scale <<< BLOCKS, BLOCK_SIZE >>>
+ ( &workspace->v[index_wkspace_sys(j+1,0,N)],
+ 1. / v_add_tmp, &workspace->v[index_wkspace_sys(j+1,0,N)], N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+ double REC_V_ADD_TMP = 1. / v_add_tmp;
+ cublasCheckError (cublasDscal (cublasHandle, N, &REC_V_ADD_TMP, &workspace->v[index_wkspace_sys(j+1,0,N)], 1));
+
+
+
+ /* Givens rotations on the upper-Hessenberg matrix to make it U */
+ GMRES_Givens_Rotation <<<1, 1>>>
+ (j, workspace->h, workspace->hc, workspace->hs, g[j], spad);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ copy_host_device (&g[j], spad, 2 * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ }
+
+ copy_host_device (g, workspace->g, (RESTART+1)*REAL_SIZE, cudaMemcpyHostToDevice, __LINE__);
+
+ /* solve Hy = g.
+ H is now upper-triangular, do back-substitution */
+ copy_host_device (g, spad, (RESTART+1) * REAL_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_G);
+ GMRES_BackSubstitution <<<1, 1>>>
+ (j, spad, workspace->h, workspace->y);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ /* update x = x_0 + Vy */
+ for( i = 0; i < j; i++ )
+ {
+ /*
+ copy_host_device (&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ Cuda_Vector_Add <<<BLOCKS, BLOCK_SIZE>>>
+ ( x, v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+
+ copy_host_device (&v_add_tmp, &workspace->y[i], REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ cublasCheckError (cublasDaxpy (cublasHandle, N, &v_add_tmp, &workspace->v[index_wkspace_sys(i,0,N)], 1,
+ x, 1));
+ }
+
+ /* stopping condition */
+ if( fabs(g[j]) / bnorm <= tol )
+ break;
+ }
+
+ if( itr >= MAX_ITR ) {
+ fprintf( stderr, "GMRES convergence failed\n" );
+ return itr * (RESTART+1) + j + 1;
+ }
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " GPU values itr : %d, RESTART: %d, j: %d \n", itr, RESTART, j);
+#endif
+
+ return itr * (RESTART+1) + j + 1;
+}
+
+
diff --git a/PuReMD-GPU/src/cuda_lin_alg.h b/PuReMD-GPU/src/cuda_lin_alg.h
new file mode 100644
index 0000000000000000000000000000000000000000..6b464152280f3590c1c70761c1b3ef206779cf5b
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_lin_alg.h
@@ -0,0 +1,43 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_LIN_ALG_H_
+#define __CUDA_LIN_ALG_H_
+
+#define SIGN(x) (x < 0.0 ? -1 : 1);
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void Cuda_Matvec (sparse_matrix , real *, real *, int );
+GLOBAL void Cuda_Matvec_csr (sparse_matrix , real *, real *, int );
+int Cuda_GMRES( static_storage *, real *b, real tol, real *x );
+int Cublas_GMRES( reax_system *, static_storage *, real *b, real tol, real *x );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_list.cu b/PuReMD-GPU/src/cuda_list.cu
new file mode 100644
index 0000000000000000000000000000000000000000..5375297c8da15000d673cc70ef5cad3a91dae541
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_list.cu
@@ -0,0 +1,114 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_list.h"
+
+#include "cuda_utils.h"
+
+
+char Cuda_Make_List(int n, int num_intrs, int type, list* l)
+{
+ char success=1;
+
+ l->n = n;
+ l->num_intrs = num_intrs;
+
+ cuda_malloc ((void **)&l->index, n * sizeof(int), 1, LIST_INDEX );
+ cuda_malloc ((void **)&l->end_index, n * sizeof(int), 1, LIST_END_INDEX );
+
+ switch(type)
+ {
+ case TYP_FAR_NEIGHBOR:
+ cuda_malloc ((void **) &l->select.far_nbr_list,
+ l->num_intrs*sizeof(far_neighbor_data),
+ 1, LIST_FAR_NEIGHBOR_DATA);
+ /*
+ cudaHostAlloc ((void **) &l->select.far_nbr_list,
+ l->num_intrs*sizeof(far_neighbor_data),
+ cudaHostAllocMapped);
+
+ cudaHostGetDevicePointer ( (void **) &l->select.far_nbr_list,
+ (void *)l->select.far_nbr_list, 0);
+ */
+ break;
+
+ case TYP_HBOND:
+ cuda_malloc ((void **) &l->select.hbond_list,
+ l->num_intrs * sizeof(hbond_data),
+ 1, LIST_HBOND_DATA );
+ break;
+
+ case TYP_BOND:
+ cuda_malloc ((void **) &l->select.bond_list,
+ l->num_intrs * sizeof(bond_data),
+ 1, LIST_BOND_DATA );
+ break;
+
+ case TYP_THREE_BODY:
+ cuda_malloc ( (void **) &l->select.three_body_list,
+ l->num_intrs * sizeof(three_body_interaction_data),
+ 1, LIST_THREE_BODY_DATA );
+ break;
+
+ default:
+ fprintf (stderr, "Unknown list creation \n" );
+ exit (1);
+ }
+
+ return success;
+}
+
+
+void Cuda_Delete_List(list* l)
+{
+ if (l->index != NULL)
+ cuda_free (l->index, LIST_INDEX );
+ if (l->end_index != NULL)
+ cuda_free (l->end_index, LIST_END_INDEX );
+
+ switch(l->type)
+ {
+ case TYP_FAR_NEIGHBOR:
+ if (l->select.far_nbr_list != NULL)
+ cuda_free (l->select.far_nbr_list, LIST_FAR_NEIGHBOR_DATA);
+ break;
+
+ case TYP_HBOND:
+ if (l->select.hbond_list != NULL)
+ cuda_free (l->select.hbond_list, LIST_HBOND_DATA );
+ break;
+
+ case TYP_BOND:
+ if (l->select.bond_list != NULL)
+ cuda_free (l->select.bond_list, LIST_BOND_DATA );
+ break;
+
+ case TYP_THREE_BODY:
+ if (l->select.three_body_list != NULL)
+ cuda_free ( l->select.three_body_list, LIST_THREE_BODY_DATA );
+ break;
+
+ default:
+ fprintf (stderr, "Unknown list deletion \n" );
+ exit (1);
+ }
+}
+
+
diff --git a/PuReMD-GPU/src/helpers.h b/PuReMD-GPU/src/cuda_list.h
similarity index 82%
rename from PuReMD-GPU/src/helpers.h
rename to PuReMD-GPU/src/cuda_list.h
index aa1bb62080ee5cdd9b92718c21124a387e248bad..cafc85743b9eb6b31da5b35b063361ef97fddf54 100644
--- a/PuReMD-GPU/src/helpers.h
+++ b/PuReMD-GPU/src/cuda_list.h
@@ -18,12 +18,22 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __HELPERS_H__
-#define __HELPERS_H__
+#ifndef __CUDA_LIST_H_
+#define __CUDA_LIST_H_
#include "mytypes.h"
-GLOBAL void compute_Inc_on_T3 (reax_atom *atoms, unsigned int N,
- simulation_box *box, real d1, real d2, real d3);
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+char Cuda_Make_List( int, int, int, list* );
+void Cuda_Delete_List( list* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PuReMD-GPU/src/lookup.cu b/PuReMD-GPU/src/cuda_lookup.cu
similarity index 60%
rename from PuReMD-GPU/src/lookup.cu
rename to PuReMD-GPU/src/cuda_lookup.cu
index c6cc23cfcb431d3fbb2921068fa958816b5287fc..9804ad8b6afee5b0c7e04ea37f46d9c206c5ba68 100644
--- a/PuReMD-GPU/src/lookup.cu
+++ b/PuReMD-GPU/src/cuda_lookup.cu
@@ -18,37 +18,16 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "lookup.h"
-#include "two_body_interactions.h"
+#include "cuda_lookup.h"
-#include "cuda_utils.h"
#include "index_utils.h"
-void Make_Lookup_Table(real xmin, real xmax, int n,
- lookup_function f, lookup_table* t)
-{
- int i;
-
- t->xmin = xmin;
- t->xmax = xmax;
- t->n = n;
- t->dx = (xmax - xmin)/(n-1);
- t->inv_dx = 1.0 / t->dx;
- t->a = (n-1)/(xmax-xmin);
- t->y = (real*) malloc(n*sizeof(real));
-
- for(i=0; i < n; i++)
- t->y[i] = f(i*t->dx + t->xmin);
-
- // //fprintf(stdout,"dx = %lf\n",t->dx);
- // for(i=0; i < n; i++)
- // //fprintf( stdout,"%d %lf %lf %lf\n",
- // i, i/t->a+t->xmin, t->y[i], exp(i/t->a+t->xmin) );
-}
+#include "cuda_utils.h"
+#include "cuda_two_body_interactions.h"
/* Fills solution into x. Warning: will modify c and d! */
-HOST_DEVICE void Tridiagonal_Solve( const real *a, const real *b,
+DEVICE void Tridiagonal_Solve( const real *a, const real *b,
real *c, real *d, real *x, unsigned int n){
int i;
real id;
@@ -68,6 +47,7 @@ HOST_DEVICE void Tridiagonal_Solve( const real *a, const real *b,
x[i] = d[i] - c[i] * x[i + 1];
}
+
GLOBAL void Cuda_Tridiagonal_Solve (const real *a, const real *b,
real *c, real *d, real *x, unsigned int n)
{
@@ -75,65 +55,6 @@ GLOBAL void Cuda_Tridiagonal_Solve (const real *a, const real *b,
}
-
-
-
-
-
-
-
-
-void Natural_Cubic_Spline( const real *h, const real *f,
- cubic_spline_coef *coef, unsigned int n )
-{
- int i;
- real *a, *b, *c, *d, *v;
-
- /* allocate space for the linear system */
- a = (real*) malloc( n * sizeof(real) );
- b = (real*) malloc( n * sizeof(real) );
- c = (real*) malloc( n * sizeof(real) );
- d = (real*) malloc( n * sizeof(real) );
- v = (real*) malloc( n * sizeof(real) );
-
- /* build the linear system */
- a[0] = a[1] = a[n-1] = 0;
- for( i = 2; i < n-1; ++i )
- a[i] = h[i-1];
-
- b[0] = b[n-1] = 0;
- for( i = 1; i < n-1; ++i )
- b[i] = 2 * (h[i-1] + h[i]);
-
- c[0] = c[n-2] = c[n-1] = 0;
- for( i = 1; i < n-2; ++i )
- c[i] = h[i];
-
- d[0] = d[n-1] = 0;
- for( i = 1; i < n-1; ++i )
- d[i] = 6 * ((f[i+1]-f[i])/h[i] - (f[i]-f[i-1])/h[i-1]);
-
- /*//fprintf( stderr, "i a b c d\n" );
- for( i = 0; i < n; ++i )
- //fprintf( stderr, "%d %f %f %f %f\n", i, a[i], b[i], c[i], d[i] );*/
- v[0] = 0;
- v[n-1] = 0;
- Tridiagonal_Solve( &(a[1]), &(b[1]), &(c[1]), &(d[1]), &(v[1]), n-2 );
-
- for( i = 1; i < n; ++i ){
- coef[i-1].d = (v[i] - v[i-1]) / (6*h[i-1]);
- coef[i-1].c = v[i]/2;
- coef[i-1].b = (f[i]-f[i-1])/h[i-1] + h[i-1]*(2*v[i] + v[i-1])/6;
- coef[i-1].a = f[i];
- }
-
- /*//fprintf( stderr, "i v coef\n" );
- for( i = 0; i < n; ++i )
- //fprintf( stderr, "%d %f %f %f %f %f\n",
- i, v[i], coef[i].a, coef[i].b, coef[i].c, coef[i].d ); */
-}
-
-
GLOBAL void cubic_spline_init_a ( real *a, const real *h, int n )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -146,6 +67,7 @@ GLOBAL void cubic_spline_init_a ( real *a, const real *h, int n )
}
}
+
GLOBAL void cubic_spline_init_b (real *b, const real *h, int n )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -158,6 +80,7 @@ GLOBAL void cubic_spline_init_b (real *b, const real *h, int n )
}
}
+
GLOBAL void cubic_spline_init_c (real *c, const real *h, int n )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -170,6 +93,7 @@ GLOBAL void cubic_spline_init_c (real *c, const real *h, int n )
}
}
+
GLOBAL void cubic_spline_init_d (real *d, const real *f, const real *h, int n )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -182,6 +106,7 @@ GLOBAL void cubic_spline_init_d (real *d, const real *f, const real *h, int n )
}
}
+
GLOBAL void calculate_cubic_spline_coef ( const real *f, real *v, const real *h, LR_lookup_table *data, int offset, int n )
{
cubic_spline_coef *coef;
@@ -270,66 +195,6 @@ void Cuda_Natural_Cubic_Spline( const real *h, const real *f,
}
-
-
-
-
-
-
-
-
-
-void Complete_Cubic_Spline( const real *h, const real *f, real v0, real vlast,
- cubic_spline_coef *coef, unsigned int n )
-{
- int i;
- real *a, *b, *c, *d, *v;
-
- /* allocate space for the linear system */
- a = (real*) malloc( n * sizeof(real) );
- b = (real*) malloc( n * sizeof(real) );
- c = (real*) malloc( n * sizeof(real) );
- d = (real*) malloc( n * sizeof(real) );
- v = (real*) malloc( n * sizeof(real) );
-
- /* build the linear system */
- a[0] = 0;
- for( i = 1; i < n; ++i )
- a[i] = h[i-1];
-
- b[0] = 2*h[0];
- for( i = 1; i < n; ++i )
- b[i] = 2 * (h[i-1] + h[i]);
-
- c[n-1] = 0;
- for( i = 0; i < n-1; ++i )
- c[i] = h[i];
-
- d[0] = 6 * (f[1]-f[0])/h[0] - 6 * v0;
- d[n-1] = 6 * vlast - 6 * (f[n-1]-f[n-2]/h[n-2]);
- for( i = 1; i < n-1; ++i )
- d[i] = 6 * ((f[i+1]-f[i])/h[i] - (f[i]-f[i-1])/h[i-1]);
-
- /*//fprintf( stderr, "i a b c d\n" );
- for( i = 0; i < n; ++i )
- //fprintf( stderr, "%d %f %f %f %f\n", i, a[i], b[i], c[i], d[i] );*/
- Tridiagonal_Solve( &(a[0]), &(b[0]), &(c[0]), &(d[0]), &(v[0]), n );
- // Tridiagonal_Solve( &(a[1]), &(b[1]), &(c[1]), &(d[1]), &(v[1]), n-2 );
-
- for( i = 1; i < n; ++i ){
- coef[i-1].d = (v[i] - v[i-1]) / (6*h[i-1]);
- coef[i-1].c = v[i]/2;
- coef[i-1].b = (f[i]-f[i-1])/h[i-1] + h[i-1]*(2*v[i] + v[i-1])/6;
- coef[i-1].a = f[i];
- }
-
- /*//fprintf( stderr, "i v coef\n" );
- for( i = 0; i < n; ++i )
- //fprintf( stderr, "%d %f %f %f %f %f\n",
- i, v[i], coef[i].a, coef[i].b, coef[i].c, coef[i].d ); */
-}
-
-
GLOBAL void complete_cubic_spline_init_a (real *a, const real *h, int n)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -341,6 +206,7 @@ GLOBAL void complete_cubic_spline_init_a (real *a, const real *h, int n)
}
}
+
GLOBAL void complete_cubic_spline_init_b (real *b, const real *h, int n)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -364,6 +230,7 @@ GLOBAL void complete_cubic_spline_init_c (real *c, const real *h, int n )
}
}
+
GLOBAL void complete_cubic_spline_init_d (real *d, const real *f, const real *h, int v0_r, int vlast_r, int n)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -383,6 +250,7 @@ GLOBAL void complete_cubic_spline_init_d (real *d, const real *f, const real *h,
d[i] = 6 * ((f[i+1]-f[i])/h[i] - (f[i]-f[i-1])/h[i-1]);
}
+
GLOBAL void calculate_complete_cubic_spline_coef (LR_lookup_table *data, int offset, real *v, const real *h, const real *f, int n)
{
@@ -409,6 +277,7 @@ GLOBAL void calculate_complete_cubic_spline_coef (LR_lookup_table *data, int off
coef[i-1].a = f[i];
}
+
void Cuda_Complete_Cubic_Spline( const real *h, const real *f, int v0_r, int vlast_r,
LR_lookup_table *data, int offset, unsigned int n )
{
@@ -471,206 +340,6 @@ void Cuda_Complete_Cubic_Spline( const real *h, const real *f, int v0_r, int vla
}
-
-
-void LR_Lookup( LR_lookup_table *t, real r, LR_data *y )
-{
- int i;
- real base, dif;
-
- i = (int)(r * t->inv_dx);
- if( i == 0 ) ++i;
- base = (real)(i+1) * t->dx;
- dif = r - base;
- ////fprintf( stderr, "r: %f, i: %d, base: %f, dif: %f\n", r, i, base, dif );
-
- y->e_vdW = ((t->vdW[i].d*dif + t->vdW[i].c)*dif + t->vdW[i].b)*dif +
- t->vdW[i].a;
- y->CEvd = ((t->CEvd[i].d*dif + t->CEvd[i].c)*dif +
- t->CEvd[i].b)*dif + t->CEvd[i].a;
- //y->CEvd = (3*t->vdW[i].d*dif + 2*t->vdW[i].c)*dif + t->vdW[i].b;
-
- y->e_ele = ((t->ele[i].d*dif + t->ele[i].c)*dif + t->ele[i].b)*dif +
- t->ele[i].a;
- y->CEclmb = ((t->CEclmb[i].d*dif + t->CEclmb[i].c)*dif + t->CEclmb[i].b)*dif +
- t->CEclmb[i].a;
-
- y->H = y->e_ele * EV_to_KCALpMOL / C_ele;
- //y->H = ((t->H[i].d*dif + t->H[i].c)*dif + t->H[i].b)*dif + t->H[i].a;
-}
-
-
-void Make_LR_Lookup_Table( reax_system *system, control_params *control )
-{
- int i, j, r;
- int num_atom_types;
- int existing_types[MAX_ATOM_TYPES];
- real dr;
- real *h, *fh, *fvdw, *fele, *fCEvd, *fCEclmb;
- real v0_vdw, v0_ele, vlast_vdw, vlast_ele;
- /* real rand_dist;
- real evdw_abserr, evdw_relerr, fvdw_abserr, fvdw_relerr;
- real eele_abserr, eele_relerr, fele_abserr, fele_relerr;
- real evdw_maxerr, eele_maxerr;
- LR_data y, y_spline; */
-
- /* initializations */
- vlast_ele = 0;
- vlast_vdw = 0;
- v0_ele = 0;
- v0_vdw = 0;
-
- num_atom_types = system->reaxprm.num_atom_types;
- dr = control->r_cut / control->tabulate;
- h = (real*) malloc( (control->tabulate+1) * sizeof(real) );
- fh = (real*) malloc( (control->tabulate+1) * sizeof(real) );
- fvdw = (real*) malloc( (control->tabulate+1) * sizeof(real) );
- fCEvd = (real*) malloc( (control->tabulate+1) * sizeof(real) );
- fele = (real*) malloc( (control->tabulate+1) * sizeof(real) );
- fCEclmb = (real*) malloc( (control->tabulate+1) * sizeof(real) );
-
- /* allocate Long-Range LookUp Table space based on
- number of atom types in the ffield file */
- //LR = (LR_lookup_table**) malloc( num_atom_types * sizeof(LR_lookup_table*) );
- //for( i = 0; i < num_atom_types; ++i )
- // LR[i] = (LR_lookup_table*) malloc(num_atom_types * sizeof(LR_lookup_table));
-
- LR = (LR_lookup_table*) malloc(num_atom_types * num_atom_types * sizeof(LR_lookup_table));
-
- /* most atom types in ffield file will not exist in the current
- simulation. to avoid unnecessary lookup table space, determine
- the atom types that exist in the current simulation */
- for( i = 0; i < MAX_ATOM_TYPES; ++i )
- existing_types[i] = 0;
- for( i = 0; i < system->N; ++i )
- existing_types[ system->atoms[i].type ] = 1;
-
- /* fill in the lookup table entries for existing atom types.
- only lower half should be enough. */
- for( i = 0; i < num_atom_types; ++i )
- if( existing_types[i] )
- for( j = i; j < num_atom_types; ++j )
- if( existing_types[j] ) {
- LR[ index_lr (i,j,num_atom_types) ].xmin = 0;
- LR[ index_lr (i,j,num_atom_types) ].xmax = control->r_cut;
- LR[ index_lr (i,j,num_atom_types) ].n = control->tabulate + 1;
- LR[ index_lr (i,j,num_atom_types) ].dx = dr;
- LR[ index_lr (i,j,num_atom_types) ].inv_dx = control->tabulate / control->r_cut;
- LR[ index_lr (i,j,num_atom_types) ].y = (LR_data*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(LR_data));
- LR[ index_lr (i,j,num_atom_types) ].H = (cubic_spline_coef*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
- LR[ index_lr (i,j,num_atom_types) ].vdW = (cubic_spline_coef*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
- LR[ index_lr (i,j,num_atom_types) ].CEvd = (cubic_spline_coef*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
- LR[ index_lr (i,j,num_atom_types) ].ele = (cubic_spline_coef*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
- LR[ index_lr (i,j,num_atom_types) ].CEclmb = (cubic_spline_coef*)
- malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
-
- for( r = 1; r <= control->tabulate; ++r ) {
- LR_vdW_Coulomb( system, control, i, j, r * dr, &(LR[ index_lr (i,j,num_atom_types) ].y[r]) );
- h[r] = LR[ index_lr (i,j,num_atom_types) ].dx;
- fh[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].H;
- fvdw[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].e_vdW;
- fCEvd[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
- fele[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].e_ele;
- fCEclmb[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
-
- if( r == 1 ){
- v0_vdw = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
- v0_ele = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
- }
- else if( r == control->tabulate ){
- vlast_vdw = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
- vlast_ele = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
- }
- }
-
- /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fh" );
- for( r = 1; r <= control->tabulate; ++r )
- //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fh[r] ); */
- Natural_Cubic_Spline( &h[1], &fh[1],
- &(LR[ index_lr (i,j,num_atom_types) ].H[1]), control->tabulate+1 );
-
- /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fvdw" );
- for( r = 1; r <= control->tabulate; ++r )
- //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fvdw[r] );
- //fprintf( stderr, "v0_vdw: %f, vlast_vdw: %f\n", v0_vdw, vlast_vdw );
- */
- Complete_Cubic_Spline( &h[1], &fvdw[1], v0_vdw, vlast_vdw,
- &(LR[ index_lr (i,j,num_atom_types) ].vdW[1]), control->tabulate+1 );
- Natural_Cubic_Spline( &h[1], &fCEvd[1],
- &(LR[ index_lr (i,j,num_atom_types) ].CEvd[1]), control->tabulate+1 );
-
- /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fele" );
- for( r = 1; r <= control->tabulate; ++r )
- //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fele[r] );
- //fprintf( stderr, "v0_ele: %f, vlast_ele: %f\n", v0_ele, vlast_ele );
- */
- Complete_Cubic_Spline( &h[1], &fele[1], v0_ele, vlast_ele,
- &(LR[ index_lr (i,j,num_atom_types) ].ele[1]), control->tabulate+1 );
- Natural_Cubic_Spline( &h[1], &fCEclmb[1],
- &(LR[ index_lr (i,j,num_atom_types) ].CEclmb[1]), control->tabulate+1 );
- }
-
- /***** //test LR-Lookup table
- evdw_maxerr = 0;
- eele_maxerr = 0;
- for( i = 0; i < num_atom_types; ++i )
- if( existing_types[i] )
- for( j = i; j < num_atom_types; ++j )
- if( existing_types[j] ) {
- for( r = 1; r <= 100; ++r ) {
- rand_dist = (real)rand()/RAND_MAX * control->r_cut;
- LR_vdW_Coulomb( system, control, i, j, rand_dist, &y );
- LR_Lookup( &(LR[i][j]), rand_dist, &y_spline );
-
- evdw_abserr = fabs(y.e_vdW - y_spline.e_vdW);
- evdw_relerr = fabs(evdw_abserr / y.e_vdW);
- fvdw_abserr = fabs(y.CEvd - y_spline.CEvd);
- fvdw_relerr = fabs(fvdw_abserr / y.CEvd);
- eele_abserr = fabs(y.e_ele - y_spline.e_ele);
- eele_relerr = fabs(eele_abserr / y.e_ele);
- fele_abserr = fabs(y.CEclmb - y_spline.CEclmb);
- fele_relerr = fabs(fele_abserr / y.CEclmb);
-
- if( evdw_relerr > 1e-10 || eele_relerr > 1e-10 ){
-//fprintf( stderr, "rand_dist = %24.15e\n", rand_dist );
-//fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
-y.H, y_spline.H,
-fabs(y.H-y_spline.H), fabs((y.H-y_spline.H)/y.H) );
-
-//fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
-y.e_vdW, y_spline.e_vdW, evdw_abserr, evdw_relerr );
-//fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
-y.CEvd, y_spline.CEvd, fvdw_abserr, fvdw_relerr );
-
-//fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
-y.e_ele, y_spline.e_ele, eele_abserr, eele_relerr );
-//fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
-y.CEclmb, y_spline.CEclmb, fele_abserr, fele_relerr );
-}
-
-if( evdw_relerr > evdw_maxerr )
-evdw_maxerr = evdw_relerr;
-if( eele_relerr > eele_maxerr )
-eele_maxerr = eele_relerr;
-}
-}
-//fprintf( stderr, "evdw_maxerr: %24.15e\n", evdw_maxerr );
-//fprintf( stderr, "eele_maxerr: %24.15e\n", eele_maxerr );
- *******/
-
-free(h);
-free(fh);
-free(fvdw);
-free(fCEvd);
-free(fele);
-free(fCEclmb);
-}
-
void copy_LR_table_to_device (reax_system *system, control_params *control)
{
int i, j, r;
@@ -728,30 +397,6 @@ void copy_LR_table_to_device (reax_system *system, control_params *control)
}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-//////////////////////////////////////////////////////////////////////////
-// CUDA Functions for Lookup Table
-//////////////////////////////////////////////////////////////////////////
-
GLOBAL void calculate_LR_Values ( LR_lookup_table *d_LR, real *h, real *fh, real *fvdw, real *fCEvd, real *fele, real *fCEclmb,
global_parameters g_params, two_body_parameters *tbp,
control_params *control, int i,
@@ -760,17 +405,18 @@ GLOBAL void calculate_LR_Values ( LR_lookup_table *d_LR, real *h, real *fh, real
int r = blockIdx.x * blockDim.x + threadIdx.x;
if ( r == 0 || r > count ) return;
- LR_vdW_Coulomb ( g_params, tbp, control, i, j, r * dr, &data[r], num_atom_types );
+ d_LR_vdW_Coulomb( g_params, tbp, control, i, j, r * dr, &data[r], num_atom_types );
- h[r] = d_LR[ index_lr (i, j, num_atom_types) ].dx;
- fh[r] = d_LR[ index_lr (i, j, num_atom_types) ].y[r].H;
- fvdw[r] = d_LR[ index_lr (i, j, num_atom_types) ].y[r].e_vdW;
- fCEvd[r] = d_LR[ index_lr (i, j, num_atom_types) ].y[r].CEvd;
- fele[r] = d_LR[ index_lr (i, j, num_atom_types) ].y[r].e_ele;
- fCEclmb[r] = d_LR[ index_lr (i, j, num_atom_types) ].y[r].CEclmb;
+ h[r] = d_LR[ index_lr(i, j, num_atom_types) ].dx;
+ fh[r] = d_LR[ index_lr(i, j, num_atom_types) ].y[r].H;
+ fvdw[r] = d_LR[ index_lr(i, j, num_atom_types) ].y[r].e_vdW;
+ fCEvd[r] = d_LR[ index_lr(i, j, num_atom_types) ].y[r].CEvd;
+ fele[r] = d_LR[ index_lr(i, j, num_atom_types) ].y[r].e_ele;
+ fCEclmb[r] = d_LR[ index_lr(i, j, num_atom_types) ].y[r].CEclmb;
}
-GLOBAL void init_LR_values ( LR_lookup_table *d_LR, control_params *control, real dr, int i, int j, int num_atom_types )
+
+GLOBAL void init_LR_values( LR_lookup_table *d_LR, control_params *control, real dr, int i, int j, int num_atom_types )
{
d_LR[ index_lr (i, j, num_atom_types) ].xmin = 0;
d_LR[ index_lr (i, j, num_atom_types) ].xmax = control->r_cut;
@@ -779,6 +425,7 @@ GLOBAL void init_LR_values ( LR_lookup_table *d_LR, control_params *control, rea
d_LR[ index_lr (i, j, num_atom_types) ].inv_dx = control->tabulate / control->r_cut;
}
+
void Cuda_Make_LR_Lookup_Table( reax_system *system, control_params *control )
{
int i, j, r;
@@ -907,54 +554,3 @@ void Cuda_Make_LR_Lookup_Table( reax_system *system, control_params *control )
cuda_free(fele, RES_LR_LOOKUP_ELE);
cuda_free(fCEclmb, RES_LR_LOOKUP_CECLMB);
}
-
-
-
-
-
-
-
-
-//////////////////////////////////////////////////////////////////////////
-// CUDA Functions for Lookup Table
-//////////////////////////////////////////////////////////////////////////
-
-
-
-int Lookup_Index_Of( real x, lookup_table* t )
-{
- return (int)( t->a * ( x - t->xmin ) );
-}
-
-
-real Lookup( real x, lookup_table* t )
-{
- real x1, x2;
- real b;
- int i;
-
- /* if ( x < t->xmin)
- {
- //fprintf(stderr,"Domain check %lf > %lf\n",t->xmin,x);
- exit(0);
- }
- if ( x > t->xmax)
- {
- //fprintf(stderr,"Domain check %lf < %lf\n",t->xmax,x);
- exit(0);
- } */
-
- i = Lookup_Index_Of( x, t );
- x1 = i * t->dx + t->xmin;
- x2 = (i+1) * t->dx + t->xmin;
-
- b = ( x2 * t->y[i] - x1 * t->y[i+1] ) * t->inv_dx;
- // //fprintf( stdout,"SLookup_Entry: %d, %lf, %lf, %lf, %lf: %lf, %lf\n",
- // i,x1,x2,x,b,t->one_over_dx*(t->y[i+1]-t->y[i])*x+b,exp(x));
-
- return t->inv_dx * ( t->y[i+1] - t->y[i] ) * x + b;
-}
-
-
-
-
diff --git a/PuReMD-GPU/src/cuda_lookup.h b/PuReMD-GPU/src/cuda_lookup.h
new file mode 100644
index 0000000000000000000000000000000000000000..a0e05e0c092f1634a1d21fd902f198f714ac2391
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_lookup.h
@@ -0,0 +1,40 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_LOOKUP_H_
+#define __CUDA_LOOKUP_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Make_LR_Lookup_Table( reax_system*, control_params* );
+void copy_LR_table_to_device ( reax_system*, control_params* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
+
diff --git a/PuReMD-GPU/src/cuda_neighbors.cu b/PuReMD-GPU/src/cuda_neighbors.cu
new file mode 100644
index 0000000000000000000000000000000000000000..876b6b9913e4d825e0cc8be5a2fc1d092c56d9f8
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_neighbors.cu
@@ -0,0 +1,764 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_neighbors.h"
+
+#include "box.h"
+#include "grid.h"
+#include "list.h"
+#include "neighbors.h"
+#include "reset_utils.h"
+#include "system_props.h"
+#include "vector.h"
+#include "index_utils.h"
+
+#include "cuda_utils.h"
+#include "cuda_grid.h"
+
+
+extern inline DEVICE int index_grid (int blocksize)
+{
+ return blockIdx.x * gridDim.y * gridDim.z * blocksize +
+ blockIdx.y * gridDim.z * blocksize +
+ blockIdx.z * blocksize ;
+}
+
+
+DEVICE int d_Are_Far_Neighbors( rvec x1, rvec x2, simulation_box *box,
+ real cutoff, far_neighbor_data *data )
+{
+ real norm_sqr, d, tmp;
+ int i;
+
+ norm_sqr = 0;
+
+ for( i = 0; i < 3; i++ ) {
+ d = x2[i] - x1[i];
+ tmp = SQR(d);
+
+ if( tmp >= SQR( box->box_norms[i] / 2.0 ) ) {
+ if( x2[i] > x1[i] ) {
+ d -= box->box_norms[i];
+ data->rel_box[i] = -1;
+ }
+ else {
+ d += box->box_norms[i];
+ data->rel_box[i] = +1;
+ }
+
+ data->dvec[i] = d;
+ norm_sqr += SQR(d);
+ }
+ else {
+ data->dvec[i] = d;
+ norm_sqr += tmp;
+ data->rel_box[i] = 0;
+ }
+ }
+
+ if( norm_sqr <= SQR(cutoff) ){
+ data->d = sqrt(norm_sqr);
+ return 1;
+ }
+
+ return 0;
+}
+
+
+GLOBAL void k_Estimate_NumNeighbors( reax_atom *sys_atoms,
+ grid g, simulation_box *box, control_params *control, int *indices )
+{
+ int *atoms = g.atoms;
+ int *top = g.top;
+ ivec *nbrs = g.nbrs;
+ rvec *nbrs_cp = g.nbrs_cp;
+
+ int *nbr_atoms;
+ int atom1, atom2, l, iter, max, m, num_far;
+ far_neighbor_data nbr_data;
+ int x, y, z, i;
+
+ if (threadIdx.x >= *(top + index_grid(1))){
+ return;
+ }
+
+ nbrs = nbrs + index_grid (g.max_nbrs);
+ nbrs_cp = nbrs_cp + index_grid (g.max_nbrs);
+ atom1 = atoms [ index_grid (g.max_atoms) + threadIdx.x];
+
+ num_far = 0;
+ iter = 0;
+
+ while (nbrs[iter][0] >= 0) {
+ x = nbrs[iter][0];
+ y = nbrs[iter][1];
+ z = nbrs[iter][2];
+
+ //condition check for cutoff here
+ if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
+ SQR (control->vlist_cut))
+ {
+ nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
+ max = top [index_grid_3d(x, y, z, &g)];
+ for (m = 0; m < max; m++) {
+ atom2 = nbr_atoms[m];
+
+ //CHANGE ORIGINAL
+ /*
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &nbr_data)){
+ ++num_far;
+ }
+ }
+ */
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &nbr_data)){
+ ++num_far;
+ }
+ }
+ else if (atom1 < atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
+ control->vlist_cut, &nbr_data)){
+ ++num_far;
+ }
+ }
+ //CHANGE ORIGINAL
+ }
+ }
+ ++iter;
+ }
+
+ //indices[ atom1 ] = num_far;// * SAFE_ZONE;
+ indices[ atom1 ] = num_far * SAFE_ZONE;
+}
+
+
+/*One thread per atom Implementation */
+GLOBAL void k_New_Estimate_NumNeighbors( reax_atom *sys_atoms,
+ grid g, simulation_box *box, control_params* control, int N, int *indices )
+{
+ int *atoms = g.atoms;
+ int *top = g.top;
+ ivec *nbrs = g.nbrs;
+ rvec *nbrs_cp = g.nbrs_cp;
+
+ int *nbr_atoms;
+ int atom1, atom2, iter, max, m, num_far;
+ int x, y, z, i;
+ int atom_x, atom_y, atom_z;
+ far_neighbor_data temp;
+ rvec atom1_x;
+
+ int index = blockIdx.x * blockDim.x + threadIdx.x;
+ if (index > N) return;
+
+ atom_x = (int)(sys_atoms[index].x[0] * g.inv_len[0]);
+ atom_y = (int)(sys_atoms[index].x[1] * g.inv_len[1]);
+ atom_z = (int)(sys_atoms[index].x[2] * g.inv_len[2]);
+
+#ifdef __BNVT_FIX__
+ if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
+ if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
+ if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
+#endif
+
+ nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+ nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+ atom1 = index;
+
+ rvec_Copy (atom1_x, sys_atoms [atom1].x );
+
+ num_far = 0;
+ iter = 0;
+
+ while (nbrs[iter][0] >= 0) {
+ x = nbrs[iter][0];
+ y = nbrs[iter][1];
+ z = nbrs[iter][2];
+
+ if (DistSqr_to_CP (nbrs_cp[iter], atom1_x) <=
+ SQR (control->vlist_cut))
+ {
+ nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
+ max = top [index_grid_3d(x, y, z, &g)];
+
+ for (m = 0; m < max; m++)
+ {
+ atom2 = nbr_atoms[m];
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (atom1_x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &temp)){
+ num_far++;
+ }
+ }
+ else if (atom1 < atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom2].x, atom1_x, box,
+ control->vlist_cut, &temp)){
+ num_far ++;
+ }
+ }
+ }
+ }
+ ++iter;
+ }
+ indices [atom1] = num_far * SAFE_ZONE;
+}
+
+
+/*One thread per entry in the gcell implementation */
+GLOBAL void k_Generate_Neighbor_Lists ( reax_atom *sys_atoms,
+ grid g, simulation_box *box, control_params* control,
+ list far_nbrs )
+{
+ int *atoms = g.atoms;
+ int *top = g.top;
+ ivec *nbrs = g.nbrs;
+ rvec *nbrs_cp = g.nbrs_cp;
+
+ int *nbr_atoms;
+ int atom1, atom2, l, iter, max, m, num_far;
+ int x, y, z, i;
+ far_neighbor_data *nbr_data;
+ far_neighbor_data temp;
+
+ if (threadIdx.x >= *(top + index_grid(1))){
+ return;
+ }
+
+ nbrs = nbrs + index_grid (g.max_nbrs);
+ nbrs_cp = nbrs_cp + index_grid (g.max_nbrs);
+ atom1 = atoms [ index_grid (g.max_atoms) + threadIdx.x];
+
+ num_far = Start_Index (atom1, &far_nbrs);
+ //Set_Start_Index (atom1, 0, &far_nbrs);
+ //num_far = 0;
+ iter = 0;
+
+ while (nbrs[iter][0] >= 0) {
+ x = nbrs[iter][0];
+ y = nbrs[iter][1];
+ z = nbrs[iter][2];
+
+ //condition check for cutoff here
+ if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
+ SQR (control->vlist_cut))
+ {
+ nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
+ max = top [index_grid_3d(x, y, z, &g)];
+
+ for (m = 0; m < max; m++) {
+ atom2 = nbr_atoms[m];
+
+ //nbr_data = & ( far_nbrs.select.far_nbr_list[atom1 * g.max_cuda_nbrs + num_far] );
+
+ //CHANGE ORIGINAL
+ /*
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &temp)){
+
+ nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+ ++num_far;
+ }
+ }
+ */
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &temp)){
+ nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+ ++num_far;
+ }
+ }
+ else if (atom1 < atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
+ control->vlist_cut, &temp)){
+ nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+ ++num_far;
+ }
+ }
+ //CHANGE ORIGINAL
+ }
+ }
+ ++iter;
+ }
+
+ //end the far_neighbor list here
+ Set_End_Index (atom1, num_far, &far_nbrs);
+}
+
+
+/*One thread per atom Implementation */
+GLOBAL void k_New_Generate_Neighbor_Lists( reax_atom *sys_atoms,
+ grid g, simulation_box *box, control_params* control,
+ list far_nbrs, int N )
+{
+ int *atoms = g.atoms;
+ int *top = g.top;
+ ivec *nbrs = g.nbrs;
+ rvec *nbrs_cp = g.nbrs_cp;
+
+ int *nbr_atoms;
+ int atom1, atom2, l, iter, max, m, num_far;
+ int x, y, z, i;
+ far_neighbor_data *nbr_data, *my_start;
+ far_neighbor_data temp;
+ int atom_x, atom_y, atom_z;
+ rvec atom1_x;
+
+ int index = blockIdx.x * blockDim.x + threadIdx.x;
+ if (index > N) return;
+
+ atom_x = (int)(sys_atoms[index].x[0] * g.inv_len[0]);
+ atom_y = (int)(sys_atoms[index].x[1] * g.inv_len[1]);
+ atom_z = (int)(sys_atoms[index].x[2] * g.inv_len[2]);
+
+#ifdef __BNVT_FIX__
+ if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
+ if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
+ if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
+#endif
+
+ nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+ nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+ atom1 = index;
+
+ rvec_Copy (atom1_x, sys_atoms [atom1].x );
+
+ num_far = Start_Index (atom1, &far_nbrs);
+ my_start = & (far_nbrs.select.far_nbr_list [num_far] );
+
+ //Set_Start_Index (atom1, 0, &far_nbrs);
+ //num_far = 0;
+ iter = 0;
+
+ while (nbrs[iter][0] >= 0) {
+ x = nbrs[iter][0];
+ y = nbrs[iter][1];
+ z = nbrs[iter][2];
+
+ //condition check for cutoff here
+ //if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
+ if (DistSqr_to_CP (nbrs_cp[iter], atom1_x) <=
+ SQR (control->vlist_cut))
+ {
+ nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
+ max = top [index_grid_3d(x, y, z, &g)];
+
+ for (m = 0; m < max; m++)
+ {
+ atom2 = nbr_atoms[m];
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (atom1_x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &temp)){
+ //nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
+ nbr_data = my_start;
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+ num_far++;
+ my_start ++;
+ }
+ }
+ else if (atom1 < atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom2].x, atom1_x, box,
+ control->vlist_cut, &temp)){
+ //nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
+ nbr_data = my_start;
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+ num_far ++;
+ my_start ++;
+ }
+ }
+ //CHANGE ORIGINAL
+ }
+ }
+ ++iter;
+ }
+
+ //end the far_neighbor list here
+ Set_End_Index (atom1, num_far, &far_nbrs);
+}
+
+
+/*Multiple threads per atom Implementation */
+GLOBAL void Test_Generate_Neighbor_Lists( reax_atom *sys_atoms,
+ grid g, simulation_box *box, control_params* control,
+ list far_nbrs, int N )
+{
+
+ extern __shared__ int __nbr[];
+ extern __shared__ int __sofar [];
+ int nbrgen;
+
+ int __THREADS_PER_ATOM__ = NBRS_THREADS_PER_ATOM;
+
+ int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
+ int warp_id = thread_id / __THREADS_PER_ATOM__;
+ int lane_id = thread_id & (__THREADS_PER_ATOM__ -1);
+ int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
+
+ if (warp_id >= N ) return;
+
+ int *tnbr = __nbr;
+ //int *nbrssofar = __nbr + __THREADS_PER_ATOM__;
+ int *nbrssofar = __nbr + blockDim.x;
+
+ int *atoms = g.atoms;
+ int *top = g.top;
+ ivec *nbrs = g.nbrs;
+ rvec *nbrs_cp = g.nbrs_cp;
+
+ int *nbr_atoms;
+ int atom1, atom2, l, iter, max, m, num_far;
+ int leader = -10;
+ int x, y, z, i;
+ far_neighbor_data *nbr_data, *my_start;
+ far_neighbor_data temp;
+ int atom_x, atom_y, atom_z;
+
+
+ atom1 = warp_id;
+ atom_x = (int)(sys_atoms[atom1].x[0] * g.inv_len[0]);
+ atom_y = (int)(sys_atoms[atom1].x[1] * g.inv_len[1]);
+ atom_z = (int)(sys_atoms[atom1].x[2] * g.inv_len[2]);
+
+#ifdef __BNVT_FIX__
+ if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
+ if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
+ if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
+#endif
+
+ nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+ nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
+
+ num_far = Start_Index (atom1, &far_nbrs);
+ my_start = & (far_nbrs.select.far_nbr_list [num_far] );
+
+ iter = 0;
+ tnbr[threadIdx.x] = 0;
+
+ if (lane_id == 0) {
+ //nbrssofar [threadIdx.x /__THREADS_PER_ATOM__] = 0;
+ nbrssofar [my_bucket] = 0;
+ }
+
+ __syncthreads ();
+
+ while ((nbrs[iter][0] >= 0)) {
+ x = nbrs[iter][0];
+ y = nbrs[iter][1];
+ z = nbrs[iter][2];
+
+ tnbr[threadIdx.x] = 0;
+ nbrgen = FALSE;
+
+ if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms [atom1].x) <=
+ SQR (control->vlist_cut))
+ {
+ nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
+ max = top [index_grid_3d(x, y, z, &g)];
+
+ tnbr[threadIdx.x] = 0;
+ nbrgen = FALSE;
+ m = lane_id ; //0-31
+ int loopcount = max / __THREADS_PER_ATOM__ + ((max % __THREADS_PER_ATOM__) == 0 ? 0 : 1);
+ int iterations = 0;
+ //while (m < max)
+ while (iterations < loopcount)
+ {
+ tnbr [threadIdx.x] = 0;
+ nbrgen = FALSE;
+
+ if (m < max) {
+ atom2 = nbr_atoms[m];
+ if (atom1 > atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
+ control->vlist_cut, &temp))
+ {
+ tnbr [threadIdx.x] = 1;
+ nbrgen = TRUE;
+ }
+ }
+ else if (atom1 < atom2) {
+ if (d_Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
+ control->vlist_cut, &temp)){
+ tnbr [threadIdx.x] = 1;
+ nbrgen = TRUE;
+ }
+ }
+ }
+
+ if (nbrgen)
+ {
+ //do leader selection here
+ leader = -1;
+ //for (l = threadIdx.x / __THREADS_PER_ATOM__; l < threadIdx.x / __THREADS_PER_ATOM__ + __THREADS_PER_ATOM__; l++)
+ for (l = my_bucket *__THREADS_PER_ATOM__; l < (my_bucket)*__THREADS_PER_ATOM__ + __THREADS_PER_ATOM__; l++)
+ if (tnbr[l]){
+ leader = l;
+ break;
+ }
+
+ //do the reduction;
+ if (threadIdx.x == leader)
+ for (l = 1; l < __THREADS_PER_ATOM__; l++)
+ //tnbr [(threadIdx.x / __THREADS_PER_ATOM__) * __THREADS_PER_ATOM__ + l] += tnbr [(threadIdx.x / __THREADS_PER_ATOM__) * __THREADS_PER_ATOM__ + (l-1)];
+ tnbr [my_bucket * __THREADS_PER_ATOM__ + l] += tnbr [my_bucket * __THREADS_PER_ATOM__ + (l-1)];
+ }
+
+ //__syncthreads ();
+ //MYATOMICADD( &warp_sync [threadIdx.x / __THREADS_PER_ATOM__ ], 1);
+ //while ( warp_sync [threadIdx.x / __THREADS_PER_ATOM__ ] < __THREADS_PER_ATOM__ ) ;
+
+ if (nbrgen)
+ {
+ //got the indices
+ //nbr_data = my_start + nbrssofar[threadIdx.x / __THREADS_PER_ATOM__] + tnbr [threadIdx.x] - 1;
+ nbr_data = my_start + nbrssofar[my_bucket] + tnbr [threadIdx.x] - 1;
+ nbr_data->nbr = atom2;
+ nbr_data->rel_box[0] = temp.rel_box[0];
+ nbr_data->rel_box[1] = temp.rel_box[1];
+ nbr_data->rel_box[2] = temp.rel_box[2];
+
+ nbr_data->d = temp.d;
+ nbr_data->dvec[0] = temp.dvec[0];
+ nbr_data->dvec[1] = temp.dvec[1];
+ nbr_data->dvec[2] = temp.dvec[2];
+
+ if (threadIdx.x == leader)
+ //nbrssofar[threadIdx.x / __THREADS_PER_ATOM__] += tnbr[(threadIdx.x / __THREADS_PER_ATOM__)*__THREADS_PER_ATOM__ + (__THREADS_PER_ATOM__ - 1)];
+ nbrssofar[my_bucket] += tnbr[my_bucket *__THREADS_PER_ATOM__ + (__THREADS_PER_ATOM__ - 1)];
+ }
+
+ m += __THREADS_PER_ATOM__;
+ iterations ++;
+
+ //cleanup
+ nbrgen = FALSE;
+ tnbr [threadIdx.x] = 0;
+ }
+ }
+ ++iter;
+ }
+
+ __syncthreads ();
+
+ //end the far_neighbor list here
+ if (lane_id == 0)
+ Set_End_Index (atom1, num_far + nbrssofar[my_bucket], &far_nbrs);
+ //Set_End_Index (atom1, num_far + tnbr[63], &far_nbrs);
+}
+
+
+void Cuda_Generate_Neighbor_Lists (reax_system *system, static_storage *workspace, control_params *control, int estimate)
+{
+ real t_start, t_elapsed;
+ real t_1, t_2;
+
+ list *far_nbrs = dev_lists + FAR_NBRS;
+
+ int *d_indices = (int *) scratch;
+ int *nbrs_start, *nbrs_end;
+ int i, max_nbrs = 0;
+ int nbs;
+
+ t_start = Get_Time ();
+
+ Cuda_Bin_Atoms (system, workspace);
+ Cuda_Bin_Atoms_Sync ( system );
+
+ if (dev_workspace->realloc.estimate_nbrs > -1) {
+
+ /*reset the re-neighbor condition */
+ dev_workspace->realloc.estimate_nbrs = -1;
+
+ //#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Recomputing the neighbors estimate.... \n");
+ //#endif
+ cuda_memset (d_indices, 0, INT_SIZE * system->N, RES_SCRATCH );
+ /*
+ dim3 blockspergrid (system->g.ncell[0], system->g.ncell[1], system->g.ncell[2]);
+ dim3 threadsperblock (system->g.max_atoms);
+
+ k_Estimate_NumNeighbors <<<blockspergrid, threadsperblock >>>
+ (system->d_atoms, system->d_g, system->d_box,
+ (control_params *)control->d_control, d_indices);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+ nbs = (system->N / NBRS_BLOCK_SIZE) + (((system->N) % NBRS_BLOCK_SIZE) == 0 ? 0 : 1);
+ k_New_Estimate_NumNeighbors <<<nbs, NBRS_BLOCK_SIZE>>>
+ ( system->d_atoms, system->d_g,
+ system->d_box, (control_params *)control->d_control,
+ system->N, d_indices);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+
+ int *nbrs_indices = NULL;
+ nbrs_indices = (int *) malloc( INT_SIZE * (system->N+1) );
+ if (nbrs_indices == NULL)
+ {
+ fprintf (stderr, "Malloc failed for nbrs indices .... \n");
+ exit (1);
+ }
+ memset (nbrs_indices , 0, INT_SIZE * (system->N+1) );
+
+ copy_host_device (nbrs_indices+1, d_indices, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ for (int i = 1; i <= system->N; i++)
+ nbrs_indices [i] += nbrs_indices [i-1];
+
+ copy_host_device (nbrs_indices, (far_nbrs->index), INT_SIZE * (system->N), cudaMemcpyHostToDevice, __LINE__ );
+ copy_host_device (nbrs_indices, (far_nbrs->end_index), INT_SIZE * (system->N), cudaMemcpyHostToDevice, __LINE__ );
+
+ free (nbrs_indices);
+ }
+
+ /*
+ One thread per atom Implementation
+ Generate_Neighbor_Lists <<<blockspergrid, threadsperblock >>>
+ (system->d_atoms, system->d_g, system->d_box,
+ (control_params *)control->d_control, *far_nbrs);
+ */
+ nbs = (system->N * NBRS_THREADS_PER_ATOM/ NBRS_BLOCK_SIZE) +
+ (((system->N *NBRS_THREADS_PER_ATOM) % NBRS_BLOCK_SIZE) == 0 ? 0 : 1);
+
+ /* Multiple threads per atom Implementation */
+ Test_Generate_Neighbor_Lists <<<nbs, NBRS_BLOCK_SIZE,
+ INT_SIZE * (NBRS_BLOCK_SIZE+ NBRS_BLOCK_SIZE/NBRS_THREADS_PER_ATOM) >>>
+ (system->d_atoms, system->d_g, system->d_box,
+ (control_params *)control->d_control, *far_nbrs, system->N );
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+
+ t_elapsed = Get_Timing_Info (t_start);
+ d_timing.nbrs += t_elapsed;
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Done with neighbor generation ---> %f \n", t_elapsed);
+#endif
+
+ /*validate neighbors list*/
+ nbrs_start = (int *) calloc (system->N, INT_SIZE);
+ nbrs_end = (int *) calloc (system->N, INT_SIZE);
+
+ copy_host_device (nbrs_start, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__ );
+ copy_host_device (nbrs_end, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__ );
+
+ int device_nbrs = 0;
+ for(i = 0; i < system->N; i++)
+ {
+ if ((nbrs_end[i] - nbrs_start[i]) > max_nbrs)
+ max_nbrs = nbrs_end[i] - nbrs_start[i];
+
+ device_nbrs += nbrs_end[i] - nbrs_start[i];
+ }
+
+#ifdef __CUDA_TEST__
+ //fprintf (stderr, " New Device count is : %d \n", device_nbrs);
+ //dev_workspace->realloc.num_far = device_nbrs;
+#endif
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Max neighbors is ---> %d \n", max_nbrs );
+ fprintf (stderr, "DEVICE NEIGHBORS ---> %d \n", device_nbrs);
+#endif
+
+ //validate check here
+ //get the num_far from the list here
+ for (i = 0; i < system->N-1; i++)
+ {
+ if ((nbrs_end[i] - nbrs_start[i]) > (nbrs_start[i+1] - nbrs_start[i]) * DANGER_ZONE )
+ {
+ dev_workspace->realloc.num_far = device_nbrs;
+ //#ifdef __CUDA_MEM__
+ //fprintf (stderr, "Need to reallocate the neighbors ----> %d \n", dev_workspace->realloc.num_far);
+ //fprintf (stderr, "Reaching the limits of neighbors for index ----> %d (%d %d %d) \n",
+ // i, nbrs_start[i], nbrs_end[i], nbrs_start[i+1]);
+ //#endif
+ }
+
+ if (nbrs_end[i] > nbrs_start[i+1]) {
+ fprintf( stderr, "**ran out of space on far_nbrs: start[i] = %d, end[i]=%d, start[i+1]=%d, end[i+1] = %d",
+ nbrs_start[i], nbrs_end[i], nbrs_start[i+1], nbrs_end[i+1]);
+ exit( INSUFFICIENT_SPACE );
+ }
+ }
+
+ if ((nbrs_end[i] - nbrs_start[i]) > (far_nbrs->num_intrs - nbrs_start[i]) * DANGER_ZONE ) {
+ dev_workspace->realloc.num_far = device_nbrs;
+ //#ifdef __CUDA_MEM__
+ //fprintf (stderr, "Need to reallocate the neighbors ----> %d \n", dev_workspace->realloc.num_far);
+ //fprintf (stderr, "Reaching the limits of neighbors for index ----> %d start: %d, end: %d, count: %d\n"
+ // , i, nbrs_start[i], nbrs_end[i], far_nbrs->num_intrs);
+ //#endif
+ }
+ if (nbrs_end[i] > far_nbrs->num_intrs) {
+ fprintf( stderr, "**ran out of space on far_nbrs: top=%d, max=%d",
+ nbrs_end[i], far_nbrs->num_intrs );
+ exit( INSUFFICIENT_SPACE );
+ }
+
+ free (nbrs_start);
+ free (nbrs_end);
+}
diff --git a/PuReMD-GPU/src/cuda_neighbors.h b/PuReMD-GPU/src/cuda_neighbors.h
new file mode 100644
index 0000000000000000000000000000000000000000..13656f62f12b53509fad82f535e974e45cc45805
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_neighbors.h
@@ -0,0 +1,44 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_NEIGHBORS_H_
+#define __CUDA_NEIGHBORS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void k_Estimate_NumNeighbors( reax_atom *, grid, simulation_box *,
+ control_params *, int * );
+
+void Cuda_Generate_Neighbor_Lists (reax_system *system,
+ static_storage *workspace, control_params *control, int);
+
+DEVICE int d_Are_Far_Neighbors( rvec, rvec, simulation_box*, real, far_neighbor_data* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_post_evolve.cu b/PuReMD-GPU/src/cuda_post_evolve.cu
new file mode 100644
index 0000000000000000000000000000000000000000..f5dbad825f2ac9f6e120e6cf7147258c89868c71
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_post_evolve.cu
@@ -0,0 +1,148 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_post_evolve.h"
+
+#include "vector.h"
+
+#include "cuda_utils.h"
+#include "cuda_copy.h"
+#include "cuda_system_props.h"
+
+
+void Cuda_Setup_Evolve( reax_system* system, control_params* control,
+ simulation_data* data, static_storage* workspace,
+ list** lists, output_controls *out_control )
+{
+ //fprintf (stderr, "Begin ... \n");
+ //to Sync step to the device.
+ //Sync_Host_Device_Data( &data, (simulation_data *)data.d_simulation_data, cudaMemcpyHostToDevice );
+ copy_host_device( &data->step, &((simulation_data *)data->d_simulation_data)->step,
+ INT_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
+
+}
+
+
+void Cuda_Setup_Output( reax_system* system, simulation_data* data )
+{
+ // Here sync the simulation data, because it has been changed.
+ Prep_Device_For_Output( system, data );
+}
+
+
+void Cuda_Sync_Temp( control_params* control )
+{
+ Sync_Host_Device_Params( control, (control_params*)control->d_control, cudaMemcpyHostToDevice );
+}
+
+
+GLOBAL void Update_Atoms_Post_Evolve (reax_atom *atoms, simulation_data *data, int N)
+{
+ rvec diff, cross;
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+ if (i >= N) return;
+
+ //for( i = 0; i < system->N; i++ ) {
+ // remove translational
+ rvec_ScaledAdd( atoms[i].v, -1., data->vcm );
+
+ // remove rotational
+ rvec_ScaledSum( diff, 1., atoms[i].x, -1., data->xcm );
+ rvec_Cross( cross, data->avcm, diff );
+ rvec_ScaledAdd( atoms[i].v, -1., cross );
+ //}
+}
+
+
+void Cuda_Post_Evolve( reax_system* system, control_params* control,
+ simulation_data* data, static_storage* workspace,
+ list** lists, output_controls *out_control )
+{
+ int i;
+ rvec diff, cross;
+
+ /* compute kinetic energy of the system */
+ /*
+ real *results = (real *) scratch;
+ cuda_memset (results, 0, REAL_SIZE * BLOCKS_POW_2, RES_SCRATCH);
+ Compute_Kinetic_Energy <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (system->reaxprm.d_sbp, system->d_atoms, system->N,
+ (simulation_data *)data->d_simulation_data, (real *) results);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+ */
+
+ //fprintf (stderr, "Cuda_Post_Evolve: Begin\n");
+ Cuda_Compute_Kinetic_Energy( system, data );
+ //fprintf (stderr, " Cuda_Compute_Kinetic_Energy done.... \n");
+
+ /* remove rotational and translational velocity of the center of mass */
+ if( control->ensemble != NVE &&
+ control->remove_CoM_vel &&
+ data->step && data->step % control->remove_CoM_vel == 0 ) {
+
+ /*
+ rvec t_xcm, t_vcm, t_avcm;
+ rvec_MakeZero (t_xcm);
+ rvec_MakeZero (t_vcm);
+ rvec_MakeZero (t_avcm);
+
+ rvec_Copy (t_xcm, data->xcm);
+ rvec_Copy (t_vcm, data->vcm);
+ rvec_Copy (t_avcm, data->avcm);
+ */
+
+ /* compute velocity of the center of mass */
+ Cuda_Compute_Center_of_Mass( system, data, out_control->prs );
+ //fprintf (stderr, "Cuda_Compute_Center_of_Mass done... \n");
+ /*
+ fprintf (stderr, "center of mass done on the device \n");
+
+ fprintf (stderr, "xcm --> %4.10f %4.10f \n", t_xcm, data->xcm );
+ fprintf (stderr, "vcm --> %4.10f %4.10f \n", t_vcm, data->vcm );
+ fprintf (stderr, "avcm --> %4.10f %4.10f \n", t_avcm, data->avcm );
+
+ if (check_zero (t_xcm, data->xcm) ||
+ check_zero (t_vcm, data->vcm) ||
+ check_zero (t_avcm, data->avcm)){
+ fprintf (stderr, "SimulationData (xcm, vcm, avcm) does not match between device and host \n");
+ exit (0);
+ }
+ */
+
+ //xcm, avcm,
+ copy_host_device( data->vcm,
+ ((simulation_data *)data->d_simulation_data)->vcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
+ copy_host_device( data->xcm,
+ ((simulation_data *)data->d_simulation_data)->xcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
+ copy_host_device( data->avcm,
+ ((simulation_data *)data->d_simulation_data)->avcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
+
+ //fprintf (stderr, "data copied.... \n");
+
+ Update_Atoms_Post_Evolve<<< BLOCKS, BLOCK_SIZE >>>
+ (system->d_atoms, (simulation_data *)data->d_simulation_data, system->N);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //fprintf (stderr, " Cuda_Post_Evolve:End \n");
+
+ }
+}
diff --git a/PuReMD-GPU/src/cuda_post_evolve.h b/PuReMD-GPU/src/cuda_post_evolve.h
new file mode 100644
index 0000000000000000000000000000000000000000..1d8fdc270edd28e0f387d53d9b9837c3a4879542
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_post_evolve.h
@@ -0,0 +1,48 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_POST_EVOLVE_H__
+#define __CUDA_POST_EVOLVE_H__
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Setup_Evolve( reax_system *, control_params *,
+ simulation_data *, static_storage *,
+ list **, output_controls * );
+
+void Cuda_Setup_Output( reax_system *, simulation_data * );
+
+void Cuda_Sync_Temp( control_params * );
+
+void Cuda_Post_Evolve( reax_system *, control_params *,
+ simulation_data *, static_storage *,
+ list **, output_controls * );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/reduction.cu b/PuReMD-GPU/src/cuda_reduction.cu
similarity index 97%
rename from PuReMD-GPU/src/reduction.cu
rename to PuReMD-GPU/src/cuda_reduction.cu
index 48fb5efc473adda2540a3dac4ffb2664ad9ca2a9..e22f9ad8474830ebdbe11089cdfcc986ff8dd8f3 100644
--- a/PuReMD-GPU/src/reduction.cu
+++ b/PuReMD-GPU/src/cuda_reduction.cu
@@ -18,9 +18,9 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reduction.h"
+#include "cuda_reduction.h"
+
#include "vector.h"
-#include "mytypes.h"
GLOBAL void Cuda_reduction(const real *input, real *per_block_results, const size_t n)
@@ -52,6 +52,7 @@ GLOBAL void Cuda_reduction(const real *input, real *per_block_results, const siz
}
}
+
GLOBAL void Cuda_Norm (const real *input, real *per_block_results, const size_t n, int pass)
{
extern __shared__ real sdata[];
@@ -87,6 +88,7 @@ GLOBAL void Cuda_Norm (const real *input, real *per_block_results, const size_t
}
}
+
GLOBAL void Cuda_Dot (const real *a, const real *b, real *per_block_results, const size_t n )
{
extern __shared__ real sdata[];
@@ -116,6 +118,7 @@ GLOBAL void Cuda_Dot (const real *a, const real *b, real *per_block_results, con
}
}
+
GLOBAL void Cuda_matrix_col_reduction(const real *input, real *per_block_results, const size_t n)
{
extern __shared__ real sdata[];
@@ -146,11 +149,7 @@ GLOBAL void Cuda_matrix_col_reduction(const real *input, real *per_block_results
}
-
-
-
-
-GLOBAL void Cuda_reduction(const int *input, int *per_block_results, const size_t n)
+GLOBAL void Cuda_reduction_int(const int *input, int *per_block_results, const size_t n)
{
extern __shared__ int sh_input[];
unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -213,10 +212,10 @@ GLOBAL void Cuda_reduction_rvec (rvec *input, rvec *results, size_t n)
}
}
+
//////////////////////////////////////////////////
//vector functions
//////////////////////////////////////////////////
-
GLOBAL void Cuda_Vector_Sum( real* dest, real c, real* v, real d, real* y, int k )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -225,6 +224,7 @@ GLOBAL void Cuda_Vector_Sum( real* dest, real c, real* v, real d, real* y, int k
dest[i] = c * v[i] + d * y[i];
}
+
GLOBAL void Cuda_Vector_Scale( real* dest, real c, real* v, int k )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -233,6 +233,7 @@ GLOBAL void Cuda_Vector_Scale( real* dest, real c, real* v, int k )
dest[i] = c * v[i];
}
+
GLOBAL void Cuda_Vector_Add( real* dest, real c, real* v, int k )
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
diff --git a/PuReMD-GPU/src/reduction.h b/PuReMD-GPU/src/cuda_reduction.h
similarity index 66%
rename from PuReMD-GPU/src/reduction.h
rename to PuReMD-GPU/src/cuda_reduction.h
index fefbe4c2de5e7674c2d53c73c2aba698b7c93f2a..5b9baf1df69f2a4ab2bdab0d3bff26e390634951 100644
--- a/PuReMD-GPU/src/reduction.h
+++ b/PuReMD-GPU/src/cuda_reduction.h
@@ -18,22 +18,32 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __REDUCTION_H__
-#define __REDUCTION_H__
+#ifndef __CUDA_REDUCTION_H__
+#define __CUDA_REDUCTION_H__
#include "mytypes.h"
#define INITIAL 0
#define FINAL 1
-GLOBAL void Cuda_reduction (const real *input, real *per_block_results, const size_t n);
-GLOBAL void Cuda_Norm (const real *input, real *per_block_results, const size_t n, int pass);
-GLOBAL void Cuda_Dot (const real *a, const real *b, real *per_block_results, const size_t n);
-GLOBAL void Cuda_reduction (const int *input, int *per_block_results, const size_t n);
-GLOBAL void Cuda_reduction_rvec (rvec *, rvec *, size_t n);
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void Cuda_reduction( const real *input, real *per_block_results, const size_t n );
+GLOBAL void Cuda_Norm( const real *input, real *per_block_results, const size_t n, int pass );
+GLOBAL void Cuda_Dot( const real *a, const real *b, real *per_block_results, const size_t n );
+GLOBAL void Cuda_reduction_int( const int *input, int *per_block_results, const size_t n );
+GLOBAL void Cuda_reduction_rvec( rvec *, rvec *, size_t n );
GLOBAL void Cuda_Vector_Sum( real* , real , real* , real , real* , int ) ;
GLOBAL void Cuda_Vector_Scale( real* , real , real* , int ) ;
GLOBAL void Cuda_Vector_Add( real* , real , real* , int );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PuReMD-GPU/src/reset_utils.cu b/PuReMD-GPU/src/cuda_reset_utils.cu
similarity index 55%
rename from PuReMD-GPU/src/reset_utils.cu
rename to PuReMD-GPU/src/cuda_reset_utils.cu
index 0c6f852bdf295ec9ad48af63f9195737adc387c0..d18d9f749a9a57a269f731aecae7754bdb8925a4 100644
--- a/PuReMD-GPU/src/reset_utils.cu
+++ b/PuReMD-GPU/src/cuda_reset_utils.cu
@@ -18,13 +18,16 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reset_utils.h"
+#include "cuda_reset_utils.h"
+
#include "list.h"
+#include "reset_utils.h"
#include "vector.h"
#include "cuda_utils.h"
#include "cuda_copy.h"
+
GLOBAL void Reset_Atoms (reax_atom *atoms, int N)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -35,6 +38,7 @@ GLOBAL void Reset_Atoms (reax_atom *atoms, int N)
atoms[i].f[2] = 0.0;
}
+
void Cuda_Reset_Atoms (reax_system *system )
{
Reset_Atoms <<<BLOCKS, BLOCK_SIZE>>>
@@ -43,42 +47,6 @@ void Cuda_Reset_Atoms (reax_system *system )
cudaCheckError ();
}
-void Reset_Atoms( reax_system* system )
-{
- int i;
-
- for( i = 0; i < system->N; ++i )
- memset( system->atoms[i].f, 0.0, RVEC_SIZE );
-}
-
-
-void Reset_Pressures( simulation_data *data )
-{
- rtensor_MakeZero( data->flex_bar.P );
- data->iso_bar.P = 0;
- rvec_MakeZero( data->int_press );
- rvec_MakeZero( data->ext_press );
- /* fprintf( stderr, "reset: ext_press (%12.6f %12.6f %12.6f)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] ); */
-}
-
-
-void Reset_Simulation_Data( simulation_data* data )
-{
- data->E_BE = 0;
- data->E_Ov = 0;
- data->E_Un = 0;
- data->E_Lp = 0;
- data->E_Ang = 0;
- data->E_Pen = 0;
- data->E_Coa = 0;
- data->E_HB = 0;
- data->E_Tor = 0;
- data->E_Con = 0;
- data->E_vdW = 0;
- data->E_Ele = 0;
- data->E_Kin = 0;
-}
void Cuda_Sync_Simulation_Data (simulation_data *data)
{
@@ -93,40 +61,6 @@ void Cuda_Sync_Simulation_Data (simulation_data *data)
}
-#ifdef TEST_FORCES
-void Reset_Test_Forces( reax_system *system, static_storage *workspace )
-{
- memset( workspace->f_ele, 0, system->N * sizeof(rvec) );
- memset( workspace->f_vdw, 0, system->N * sizeof(rvec) );
- memset( workspace->f_bo, 0, system->N * sizeof(rvec) );
- memset( workspace->f_be, 0, system->N * sizeof(rvec) );
- memset( workspace->f_lp, 0, system->N * sizeof(rvec) );
- memset( workspace->f_ov, 0, system->N * sizeof(rvec) );
- memset( workspace->f_un, 0, system->N * sizeof(rvec) );
- memset( workspace->f_ang, 0, system->N * sizeof(rvec) );
- memset( workspace->f_coa, 0, system->N * sizeof(rvec) );
- memset( workspace->f_pen, 0, system->N * sizeof(rvec) );
- memset( workspace->f_hb, 0, system->N * sizeof(rvec) );
- memset( workspace->f_tor, 0, system->N * sizeof(rvec) );
- memset( workspace->f_con, 0, system->N * sizeof(rvec) );
-}
-#endif
-
-
-void Reset_Workspace( reax_system *system, static_storage *workspace )
-{
- memset( workspace->total_bond_order, 0, system->N * sizeof( real ) );
- memset( workspace->dDeltap_self, 0, system->N * sizeof( rvec ) );
-
- memset( workspace->CdDelta, 0, system->N * sizeof( real ) );
- //memset( workspace->virial_forces, 0, system->N * sizeof( rvec ) );
-
-#ifdef TEST_FORCES
- memset( workspace->dDelta, 0, sizeof(rvec) * system->N );
- Reset_Test_Forces( system, workspace );
-#endif
-}
-
void Cuda_Reset_Workspace( reax_system *system, static_storage *workspace )
{
cuda_memset( workspace->total_bond_order, 0, system->N * REAL_SIZE, RES_STORAGE_TOTAL_BOND_ORDER );
@@ -157,6 +91,7 @@ GLOBAL void Reset_Neighbor_Lists (single_body_parameters *sbp, reax_atom *atoms,
}
}
+
void Cuda_Reset_Neighbor_Lists (reax_system *system, control_params *control,
static_storage *workspace, list **lists )
{
@@ -172,6 +107,7 @@ void Cuda_Reset_Neighbor_Lists (reax_system *system, control_params *control,
cuda_memset (bonds->select.bond_list, 0, BOND_DATA_SIZE * bonds->num_intrs, LIST_BOND_DATA );
}
+
GLOBAL void Reset_Far_Neighbors_List (list far_nbrs, int N)
{
int tmp;
@@ -183,6 +119,7 @@ GLOBAL void Reset_Far_Neighbors_List (list far_nbrs, int N)
Set_End_Index (index, tmp, &far_nbrs);
}
+
void Cuda_Reset_Far_Neighbors_List ( reax_system *system )
{
Reset_Far_Neighbors_List <<<BLOCKS, BLOCK_SIZE>>>
@@ -191,52 +128,6 @@ void Cuda_Reset_Far_Neighbors_List ( reax_system *system )
cudaCheckError ();
}
-void Reset_Neighbor_Lists( reax_system *system, control_params *control,
- static_storage *workspace, list **lists )
-{
- int i, tmp;
- list *bonds = (*lists) + BONDS;
- list *hbonds = (*lists) + HBONDS;
-
- for( i = 0; i < system->N; ++i ) {
- tmp = Start_Index( i, bonds );
- Set_End_Index( i, tmp, bonds );
- }
-
- //TODO check if this is needed
- memset (bonds->select.bond_list, 0, BOND_DATA_SIZE * bonds->num_intrs );
-
- if( control->hb_cut > 0 )
- for( i = 0; i < system->N; ++i )
- if( system->reaxprm.sbp[system->atoms[i].type].p_hbond == 1) {
- tmp = Start_Index( workspace->hbond_index[i], hbonds );
- Set_End_Index( workspace->hbond_index[i], tmp, hbonds );
- /* fprintf( stderr, "i:%d, hbond: %d-%d\n",
- i, Start_Index( workspace->hbond_index[i], hbonds ),
- End_Index( workspace->hbond_index[i], hbonds ) );*/
- }
-}
-
-
-void Reset( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace, list **lists )
-{
- Reset_Atoms( system );
-
- Reset_Simulation_Data( data );
-
- if( control->ensemble == NPT || control->ensemble == sNPT ||
- control->ensemble == iNPT )
- Reset_Pressures( data );
-
- Reset_Workspace( system, workspace );
-
- Reset_Neighbor_Lists( system, control, workspace, lists );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "reset - ");
-#endif
-}
void Cuda_Reset_Sparse_Matrix (reax_system *system, static_storage *workspace)
{
@@ -244,6 +135,7 @@ void Cuda_Reset_Sparse_Matrix (reax_system *system, static_storage *workspace)
cuda_memset (workspace->H.val, 0, (system->N * system->max_sparse_matrix_entries) * INT_SIZE, RES_SPARSE_MATRIX_INDEX );
}
+
void Cuda_Reset( reax_system *system, control_params *control,
simulation_data *data, static_storage *workspace, list **lists )
{
@@ -251,7 +143,7 @@ void Cuda_Reset( reax_system *system, control_params *control,
//Reset_Simulation_Data( data );
Cuda_Sync_Simulation_Data ( data );
- //Sync_Host_Device (data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice);
+ //Sync_Host_Device_Data( data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice );
if( control->ensemble == NPT || control->ensemble == sNPT ||
control->ensemble == iNPT )
@@ -268,23 +160,7 @@ void Cuda_Reset( reax_system *system, control_params *control,
}
-void Reset_Grid( grid *g )
-{
- memset (g->top, 0, INT_SIZE * g->ncell[0]*g->ncell[1]*g->ncell[2]);
-}
-
void Cuda_Reset_Grid (grid *g)
{
cuda_memset (g->top, 0, INT_SIZE * g->ncell[0]*g->ncell[1]*g->ncell[2], RES_GRID_TOP);
}
-
-
-void Reset_Marks( grid *g, ivec *grid_stack, int grid_top )
-{
- int i;
-
- for( i = 0; i < grid_top; ++i )
- g->mark[grid_stack[i][0] * g->ncell[1]*g->ncell[2] +
- grid_stack[i][1] * g->ncell[2] +
- grid_stack[i][2]] = 0;
-}
diff --git a/PuReMD-GPU/src/cuda_reset_utils.h b/PuReMD-GPU/src/cuda_reset_utils.h
new file mode 100644
index 0000000000000000000000000000000000000000..bf730b935cef4857eb8e0aa97d614ef122bfca0e
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_reset_utils.h
@@ -0,0 +1,45 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_RESET_UTILS_H_
+#define __CUDA_RESET_UTILS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Reset_Grid( grid* );
+
+void Cuda_Reset_Workspace (reax_system *, static_storage *);
+
+void Cuda_Reset( reax_system*, control_params*, simulation_data*,
+ static_storage*, list** );
+
+void Cuda_Reset_Atoms (reax_system *);
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/single_body_interactions.cu b/PuReMD-GPU/src/cuda_single_body_interactions.cu
similarity index 63%
rename from PuReMD-GPU/src/single_body_interactions.cu
rename to PuReMD-GPU/src/cuda_single_body_interactions.cu
index 3c6c08822fb056c18c78bf4db5f7aa334496ea34..530e63d7071135b143d0b432bf1350070d46667c 100644
--- a/PuReMD-GPU/src/single_body_interactions.cu
+++ b/PuReMD-GPU/src/cuda_single_body_interactions.cu
@@ -18,7 +18,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "single_body_interactions.h"
+#include "cuda_single_body_interactions.h"
+
#include "bond_orders.h"
#include "list.h"
#include "lookup.h"
@@ -28,301 +29,6 @@
#include "cuda_helpers.h"
-void LonePair_OverUnder_Coordination_Energy( reax_system *system,
- control_params *control,
- simulation_data *data,
- static_storage *workspace,
- list **lists,
- output_controls *out_control )
-{
- int i, j, pj, type_i, type_j;
- real Delta_lpcorr, dfvl;
- real e_lp, expvd2, inv_expvd2, dElp, CElp, DlpVi;
- real e_lph, Di, vov3, deahu2dbo, deahu2dsbo;
- real e_ov, CEover1, CEover2, CEover3, CEover4;
- real exp_ovun1, exp_ovun2, sum_ovun1, sum_ovun2;
- real exp_ovun2n, exp_ovun6, exp_ovun8;
- real inv_exp_ovun1, inv_exp_ovun2, inv_exp_ovun2n, inv_exp_ovun8;
- real e_un, CEunder1, CEunder2, CEunder3, CEunder4;
- real p_lp1, p_lp2, p_lp3;
- real p_ovun2, p_ovun3, p_ovun4, p_ovun5, p_ovun6, p_ovun7, p_ovun8;
-
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- bond_data *pbond;
- bond_order_data *bo_ij;
- list *bonds = (*lists) + BONDS;
-
- /* Initialize parameters */
- p_lp1 = system->reaxprm.gp.l[15];
- p_lp3 = system->reaxprm.gp.l[5];
- p_ovun3 = system->reaxprm.gp.l[32];
- p_ovun4 = system->reaxprm.gp.l[31];
- p_ovun6 = system->reaxprm.gp.l[6];
- p_ovun7 = system->reaxprm.gp.l[8];
- p_ovun8 = system->reaxprm.gp.l[9];
-
- for( i = 0; i < system->N; ++i ) {
- /* set the parameter pointer */
- type_i = system->atoms[i].type;
- sbp_i = &(system->reaxprm.sbp[ type_i ]);
-
- /* lone-pair Energy */
- p_lp2 = sbp_i->p_lp2;
- expvd2 = EXP( -75 * workspace->Delta_lp[i] );
- inv_expvd2 = 1. / (1. + expvd2 );
-
- /* calculate the energy */
- data->E_Lp += e_lp =
- p_lp2 * workspace->Delta_lp[i] * inv_expvd2;
-
- dElp = p_lp2 * inv_expvd2 +
- 75 * p_lp2 * workspace->Delta_lp[i] * expvd2 * SQR(inv_expvd2);
- CElp = dElp * workspace->dDelta_lp[i];
-
- workspace->CdDelta[i] += CElp; // lp - 1st term
-
-#ifdef TEST_ENERGY
- fprintf( out_control->elp, "%23.15e%23.15e%23.15e%23.15e\n",
- p_lp2, workspace->Delta_lp_temp[i], expvd2, dElp );
- fprintf( out_control->elp, "%6d%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i]+1, workspace->nlp[i], e_lp, data->E_Lp );
-#endif
-#ifdef TEST_FORCES
- Add_dDelta( system, lists, i, CElp, workspace->f_lp ); // lp - 1st term
-#endif
-
- /* correction for C2 */
- if( system->reaxprm.gp.l[5] > 0.001 &&
- !strcmp( system->reaxprm.sbp[type_i].name, "C" ) )
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj )
- if( i < bonds->select.bond_list[pj].nbr ) {
- j = bonds->select.bond_list[pj].nbr;
- type_j = system->atoms[j].type;
-
- if( !strcmp( system->reaxprm.sbp[type_j].name, "C" ) ) {
- twbp = &( system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
- bo_ij = &( bonds->select.bond_list[pj].bo_data );
- Di = workspace->Delta[i];
- vov3 = bo_ij->BO - Di - 0.040*POW(Di, 4.);
-
- if( vov3 > 3. ) {
- data->E_Lp += e_lph = p_lp3 * SQR(vov3-3.0);
- //estrain(i) += e_lph;
-
- deahu2dbo = 2.*p_lp3*(vov3 - 3.);
- deahu2dsbo = 2.*p_lp3*(vov3 - 3.)*(-1. - 0.16*POW(Di, 3.));
-
- bo_ij->Cdbo += deahu2dbo;
- workspace->CdDelta[i] += deahu2dsbo;
-#ifdef TEST_ENERGY
- fprintf(out_control->elp,"C2cor%6d%6d%23.15e%23.15e%23.15e\n",
- // workspace->orig_id[i], workspace->orig_id[j],
- i+1, j+1, e_lph, deahu2dbo, deahu2dsbo );
-#endif
-#ifdef TEST_FORCES
- Add_dBO(system, lists, i, pj, deahu2dbo, workspace->f_lp);
- Add_dDelta(system, lists, i, deahu2dsbo, workspace->f_lp);
-#endif
- }
- }
-
- }
- }
-
-
- for( i = 0; i < system->N; ++i ) {
- type_i = system->atoms[i].type;
- sbp_i = &(system->reaxprm.sbp[ type_i ]);
-
- /* over-coordination energy */
- if( sbp_i->mass > 21.0 )
- dfvl = 0.0;
- else dfvl = 1.0; // only for 1st-row elements
-
- p_ovun2 = sbp_i->p_ovun2;
- sum_ovun1 = 0;
- sum_ovun2 = 0;
-
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ) {
- j = bonds->select.bond_list[pj].nbr;
- type_j = system->atoms[j].type;
- bo_ij = &(bonds->select.bond_list[pj].bo_data);
- sbp_j = &(system->reaxprm.sbp[ type_j ]);
- twbp = &(system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
-
- sum_ovun1 += twbp->p_ovun1 * twbp->De_s * bo_ij->BO;
- sum_ovun2 += (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j])*
- ( bo_ij->BO_pi + bo_ij->BO_pi2 );
-
- /*fprintf( stdout, "%4d%4d%23.15e%23.15e%23.15e\n",
- i+1, j+1,
- dfvl * workspace->Delta_lp_temp[j],
- sbp_j->nlp_opt,
- workspace->nlp_temp[j] );*/
- }
-
- exp_ovun1 = p_ovun3 * EXP( p_ovun4 * sum_ovun2 );
- inv_exp_ovun1 = 1.0 / (1 + exp_ovun1);
- Delta_lpcorr = workspace->Delta[i] -
- (dfvl*workspace->Delta_lp_temp[i]) * inv_exp_ovun1;
-
- exp_ovun2 = EXP( p_ovun2 * Delta_lpcorr );
- inv_exp_ovun2 = 1.0 / (1.0 + exp_ovun2);
-
- DlpVi = 1.0 / (Delta_lpcorr + sbp_i->valency + 1e-8 );
- CEover1 = Delta_lpcorr * DlpVi * inv_exp_ovun2;
-
- data->E_Ov += e_ov = sum_ovun1 * CEover1;
-
- CEover2 = sum_ovun1 * DlpVi * inv_exp_ovun2 *
- ( 1.0 - Delta_lpcorr*( DlpVi + p_ovun2 * exp_ovun2 * inv_exp_ovun2 ) );
-
- CEover3 = CEover2 * (1.0 - dfvl*workspace->dDelta_lp[i]*inv_exp_ovun1 );
-
- CEover4 = CEover2 * (dfvl*workspace->Delta_lp_temp[i]) *
- p_ovun4 * exp_ovun1 * SQR(inv_exp_ovun1);
-
-
- /* under-coordination potential */
- p_ovun2 = sbp_i->p_ovun2;
- p_ovun5 = sbp_i->p_ovun5;
-
- exp_ovun2n = 1.0 / exp_ovun2;
- exp_ovun6 = EXP( p_ovun6 * Delta_lpcorr );
- exp_ovun8 = p_ovun7 * EXP(p_ovun8 * sum_ovun2);
- inv_exp_ovun2n = 1.0 / (1.0 + exp_ovun2n);
- inv_exp_ovun8 = 1.0 / (1.0 + exp_ovun8);
-
- data->E_Un += e_un =
- -p_ovun5 * (1.0 - exp_ovun6) * inv_exp_ovun2n * inv_exp_ovun8;
-
- CEunder1 = inv_exp_ovun2n * ( p_ovun5*p_ovun6*exp_ovun6*inv_exp_ovun8 +
- p_ovun2 * e_un * exp_ovun2n);
- CEunder2 = -e_un * p_ovun8 * exp_ovun8 * inv_exp_ovun8;
- CEunder3 = CEunder1 * (1.0 - dfvl*workspace->dDelta_lp[i]*inv_exp_ovun1);
- CEunder4 = CEunder1 * (dfvl*workspace->Delta_lp_temp[i]) *
- p_ovun4 * exp_ovun1 * SQR(inv_exp_ovun1) + CEunder2;
-
- //fprintf( stdout, "%6d%23.15e%23.15e%23.15e\n",
- // i+1, sum_ovun2, e_ov, e_un );
-
- /* forces */
- workspace->CdDelta[i] += CEover3; // OvCoor - 2nd term
- workspace->CdDelta[i] += CEunder3; // UnCoor - 1st term
-
-#ifdef TEST_FORCES
- Add_dDelta( system, lists, i, CEover3, workspace->f_ov ); // OvCoor - 2nd
- Add_dDelta( system, lists, i, CEunder3, workspace->f_un ); // UnCoor - 1st
-#endif
-
-
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
- pbond = &(bonds->select.bond_list[pj]);
- j = pbond->nbr;
- type_j = system->atoms[j].type;
- bo_ij = &(pbond->bo_data);
- twbp = &(system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
-
-
- bo_ij->Cdbo += CEover1 * twbp->p_ovun1 * twbp->De_s; // OvCoor - 1st
- workspace->CdDelta[j] += CEover4*(1.0 - dfvl*workspace->dDelta_lp[j])*
- (bo_ij->BO_pi + bo_ij->BO_pi2); // OvCoor - 3a
- bo_ij->Cdbopi += CEover4 *
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//OvCoor-3b
- bo_ij->Cdbopi2 += CEover4 *
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//OvCoor-3b
-
-
- workspace->CdDelta[j] += CEunder4*(1.0-dfvl*workspace->dDelta_lp[j]) *
- (bo_ij->BO_pi + bo_ij->BO_pi2); // UnCoor - 2a
- bo_ij->Cdbopi += CEunder4 *
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//UnCoor-2b
- bo_ij->Cdbopi2 += CEunder4 *
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//UnCoor-2b
-
-
-#ifdef TEST_ENERGY
- /* fprintf( out_control->eov, "%6d%23.15e%23.15e"
- workspace->orig_id[j]+1,
- //twbp->p_ovun1,twbp->De_s,Delta_lpcorr*DlpVi*inv_exp_ovun2,
- CEover1*twbp->p_ovun1*twbp->De_s, CEover3 ); */
-
- /*fprintf( out_control->eov, "%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[j]+1,
- CEover4,
- CEover4*
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
- CEover4 * (bo_ij->BO_pi + bo_ij->BO_pi2),
- (1.0 - dfvl*workspace->dDelta_lp[j]),
- CEover4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
- (bo_ij->BO_pi + bo_ij->BO_pi2) );*/
-
- /* fprintf( out_control->eun, "%6d%23.15e\n",
- workspace->orig_id[j]+1, CEunder3 ); */
-
- /*fprintf( out_control->eun, "%6d%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[j]+1,
- CEunder4,
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
- CEunder4*
- (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
- CEunder4*(1.0 - dfvl*workspace->dDelta_lp[j])*
- (bo_ij->BO_pi + bo_ij->BO_pi2) );*/
-#endif
-
-#ifdef TEST_FORCES
- Add_dBO( system, lists, i, pj, CEover1 * twbp->p_ovun1 * twbp->De_s,
- workspace->f_ov ); // OvCoor - 1st term
-
- Add_dDelta( system, lists, j,
- CEover4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
- (bo_ij->BO_pi+bo_ij->BO_pi2), workspace->f_ov );//OvCoor3a
-
- Add_dBOpinpi2( system, lists, i, pj,
- CEover4 * (workspace->Delta[j] -
- dfvl * workspace->Delta_lp_temp[j]),
- CEover4 * (workspace->Delta[j] -
- dfvl * workspace->Delta_lp_temp[j]),
- workspace->f_ov, workspace->f_ov ); // OvCoor - 3b
-
- Add_dDelta( system, lists, j,
- CEunder4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
- (bo_ij->BO_pi + bo_ij->BO_pi2),
- workspace->f_un ); // UnCoor - 2a
-
- Add_dBOpinpi2( system, lists, i, pj,
- CEunder4 * (workspace->Delta[j] -
- dfvl * workspace->Delta_lp_temp[j]),
- CEunder4 * (workspace->Delta[j] -
- dfvl * workspace->Delta_lp_temp[j]),
- workspace->f_un, workspace->f_un ); // UnCoor - 2b
-#endif
- }
-
-#ifdef TEST_ENERGY
-
- fprintf( out_control->eov, "%6d%15.8f%15.8f%15.8f\n",
- i+1, DlpVi, Delta_lpcorr, sbp_i->valency );
-
- fprintf( out_control->eov, "%6d%15.8f%15.8f\n",
- i+1/*workspace->orig_id[i]+1*/, e_ov, data->E_Ov + data->E_Un );
-
- fprintf( out_control->eov, "%6d%15.8f%15.8f\n",
- i+1/*workspace->orig_id[i]+1*/, e_un, data->E_Ov + data->E_Un );
-#endif
- }
-}
-
-
-
-
-
-
-
-
-
-//CUDA Functions
GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, global_parameters g_params,
single_body_parameters *sbp, two_body_parameters *tbp,
static_storage p_workspace, simulation_data *data,
@@ -374,7 +80,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
e_lp = p_lp2 * workspace->Delta_lp[i] * inv_expvd2;
//PERFORMANCE IMPACT
- atomicAdd (&data->E_Lp, e_lp);
+ MYATOMICADD(&data->E_Lp, e_lp);
dElp = p_lp2 * inv_expvd2 +
75 * p_lp2 * workspace->Delta_lp[i] * expvd2 * SQR(inv_expvd2);
@@ -382,7 +88,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
//workspace->CdDelta[i] += CElp; // lp - 1st term
- atomicAdd (&workspace->CdDelta[i], CElp);
+ MYATOMICADD(&workspace->CdDelta[i], CElp);
#ifdef TEST_ENERGY
@@ -407,7 +113,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
type_j = atoms[j].type;
if( !cuda_strcmp( sbp[type_j].name, "C", 15 ) ) {
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &( tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
bo_ij = &( bonds->select.bond_list[pj].bo_data );
Di = workspace->Delta[i];
vov3 = bo_ij->BO - Di - 0.040*POW(Di, 4.);
@@ -416,7 +122,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
e_lph = p_lp3 * SQR(vov3-3.0);
- atomicAdd (&data->E_Lp, e_lph );
+ MYATOMICADD(&data->E_Lp, e_lph );
//estrain(i) += e_lph;
deahu2dbo = 2.*p_lp3*(vov3 - 3.);
@@ -426,7 +132,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
- atomicAdd (&workspace->CdDelta[i], deahu2dsbo);
+ MYATOMICADD(&workspace->CdDelta[i], deahu2dsbo);
#ifdef TEST_ENERGY
//TODO
//fprintf(out_control->elp,"C2cor%6d%6d%23.15e%23.15e%23.15e\n",
@@ -469,7 +175,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
type_j = atoms[j].type;
bo_ij = &(bonds->select.bond_list[pj].bo_data);
sbp_j = &(sbp[ type_j ]);
- twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &(tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
sum_ovun1 += twbp->p_ovun1 * twbp->De_s * bo_ij->BO;
sum_ovun2 += (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j])*
@@ -500,7 +206,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
//data->E_Ov += e_ov = sum_ovun1 * CEover1;
e_ov = sum_ovun1 * CEover1;
- atomicAdd (&data->E_Ov, e_ov );
+ MYATOMICADD(&data->E_Ov, e_ov );
CEover2 = sum_ovun1 * DlpVi * inv_exp_ovun2 *
( 1.0 - Delta_lpcorr*( DlpVi + p_ovun2 * exp_ovun2 * inv_exp_ovun2 ) );
@@ -523,7 +229,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
e_un = -p_ovun5 * (1.0 - exp_ovun6) * inv_exp_ovun2n * inv_exp_ovun8;
- atomicAdd (&data->E_Un, e_un );
+ MYATOMICADD(&data->E_Un, e_un );
CEunder1 = inv_exp_ovun2n * ( p_ovun5*p_ovun6*exp_ovun6*inv_exp_ovun8 +
p_ovun2 * e_un * exp_ovun2n);
@@ -537,8 +243,8 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
// forces
//PERFORMANCE IMPACT
- atomicAdd (&workspace->CdDelta[i] , CEover3); // OvCoor - 2nd term
- atomicAdd (&workspace->CdDelta[i], CEunder3); // UnCoor - 1st term
+ MYATOMICADD(&workspace->CdDelta[i] , CEover3); // OvCoor - 2nd term
+ MYATOMICADD(&workspace->CdDelta[i], CEunder3); // UnCoor - 1st term
#ifdef TEST_FORCES
//TODO
@@ -553,13 +259,13 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
j = pbond->nbr;
type_j = atoms[j].type;
bo_ij = &(pbond->bo_data);
- twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &(tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
bo_ij->Cdbo += CEover1 * twbp->p_ovun1 * twbp->De_s; // OvCoor - 1st
//PERFORMANCE IMPACT
- atomicAdd (&workspace->CdDelta[j], CEover4*(1.0 - dfvl*workspace->dDelta_lp[j])* (bo_ij->BO_pi + bo_ij->BO_pi2)); // OvCoor - 3a
+ MYATOMICADD(&workspace->CdDelta[j], CEover4*(1.0 - dfvl*workspace->dDelta_lp[j])* (bo_ij->BO_pi + bo_ij->BO_pi2)); // OvCoor - 3a
bo_ij->Cdbopi += CEover4 *
(workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//OvCoor-3b
@@ -568,7 +274,7 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
//PERFORMANCE IMPACT
- atomicAdd (&workspace->CdDelta[j], CEunder4*(1.0-dfvl*workspace->dDelta_lp[j]) * (bo_ij->BO_pi + bo_ij->BO_pi2) ); // UnCoor - 2a
+ MYATOMICADD(&workspace->CdDelta[j], CEunder4*(1.0-dfvl*workspace->dDelta_lp[j]) * (bo_ij->BO_pi + bo_ij->BO_pi2) ); // UnCoor - 2a
bo_ij->Cdbopi += CEunder4 *
(workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//UnCoor-2b
@@ -647,14 +353,11 @@ GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
}
-
////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////
//TEST ONLY CODE -- See if this is working.
////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////
-
-//CUDA Functions
GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, global_parameters g_params,
single_body_parameters *sbp, two_body_parameters *tbp,
static_storage p_workspace, simulation_data *data,
@@ -705,7 +408,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
// calculate the energy
e_lp = p_lp2 * workspace->Delta_lp[i] * inv_expvd2;
- //atomicAdd (&data->E_Lp, e_lp );
+ //MYATOMICADD(&data->E_Lp, e_lp );
E_Lp [ i ] = e_lp;
dElp = p_lp2 * inv_expvd2 +
@@ -723,7 +426,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
type_j = atoms[j].type;
if( !cuda_strcmp( sbp[type_j].name, "C", 15 ) ) {
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &( tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
bo_ij = &( bonds->select.bond_list[pj].bo_data );
Di = workspace->Delta[i];
vov3 = bo_ij->BO - Di - 0.040*POW(Di, 4.);
@@ -732,7 +435,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
e_lph = p_lp3 * SQR(vov3-3.0);
E_Lp [i] += e_lph;
- //atomicAdd (&data->E_Lp, e_lph );
+ //MYATOMICADD(&data->E_Lp, e_lph );
//estrain(i) += e_lph;
deahu2dbo = 2.*p_lp3*(vov3 - 3.);
@@ -769,7 +472,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
type_j = atoms[j].type;
bo_ij = &(bonds->select.bond_list[pj].bo_data);
sbp_j = &(sbp[ type_j ]);
- twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &(tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
sum_ovun1 += twbp->p_ovun1 * twbp->De_s * bo_ij->BO;
sum_ovun2 += (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j])*
@@ -790,7 +493,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
e_ov = sum_ovun1 * CEover1;
E_Ov [ i ] = e_ov;
- //atomicAdd ( &data->E_Ov, e_ov );
+ //MYATOMICADD( &data->E_Ov, e_ov );
CEover2 = sum_ovun1 * DlpVi * inv_exp_ovun2 *
( 1.0 - Delta_lpcorr*( DlpVi + p_ovun2 * exp_ovun2 * inv_exp_ovun2 ) );
@@ -813,7 +516,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
e_un = -p_ovun5 * (1.0 - exp_ovun6) * inv_exp_ovun2n * inv_exp_ovun8;
E_Un [i] = e_un;
- //atomicAdd ( &data->E_Un, e_un );
+ //MYATOMICADD( &data->E_Un, e_un );
CEunder1 = inv_exp_ovun2n * ( p_ovun5*p_ovun6*exp_ovun6*inv_exp_ovun8 +
p_ovun2 * e_un * exp_ovun2n);
@@ -831,7 +534,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
j = pbond->nbr;
type_j = atoms[j].type;
bo_ij = &(pbond->bo_data);
- twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &(tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
bo_ij->Cdbo += CEover1 * twbp->p_ovun1 * twbp->De_s; // OvCoor - 1st
@@ -854,7 +557,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *atoms, glob
}
}
-///////////////////////////////////////////////////////////
+
GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *atoms, global_parameters g_params,
single_body_parameters *sbp, two_body_parameters *tbp,
static_storage p_workspace, simulation_data *data,
@@ -903,7 +606,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *atoms, g
// calculate the energy
e_lp = p_lp2 * workspace->Delta_lp[i] * inv_expvd2;
- //atomicAdd (&data->E_Lp, e_lp );
+ //MYATOMICADD(&data->E_Lp, e_lp );
E_Lp [ i ] = e_lp;
dElp = p_lp2 * inv_expvd2 +
@@ -921,7 +624,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *atoms, g
type_j = atoms[j].type;
if( !cuda_strcmp( sbp[type_j].name, "C", 15 ) ) {
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
+ twbp = &( tbp[ index_tbp(type_i,type_j,num_atom_types) ]);
bo_ij = &( bonds->select.bond_list[pj].bo_data );
Di = workspace->Delta[i];
vov3 = bo_ij->BO - Di - 0.040*POW(Di, 4.);
@@ -930,7 +633,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *atoms, g
e_lph = p_lp3 * SQR(vov3-3.0);
E_Lp [i] += e_lph;
- //atomicAdd (&data->E_Lp, e_lph );
+ //MYATOMICADD(&data->E_Lp, e_lph );
//estrain(i) += e_lph;
deahu2dbo = 2.*p_lp3*(vov3 - 3.);
@@ -943,7 +646,7 @@ GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *atoms, g
}
}
}
-///////////////////////////////////////////////////////////
+
GLOBAL void test_LonePair_Postprocess ( reax_atom *atoms, global_parameters g_params,
single_body_parameters *sbp, two_body_parameters *tbp,
diff --git a/PuReMD-GPU/src/cuda_single_body_interactions.h b/PuReMD-GPU/src/cuda_single_body_interactions.h
new file mode 100644
index 0000000000000000000000000000000000000000..3ecd4b9a68b174624ac1dcf56aae4cec360750be
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_single_body_interactions.h
@@ -0,0 +1,59 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_SINGLE_BODY_INTERACTIONS_H_
+#define __CUDA_SINGLE_BODY_INTERACTIONS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *, global_parameters ,
+ single_body_parameters *, two_body_parameters *,
+ static_storage , simulation_data *,
+ list , int , int );
+
+GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *, global_parameters ,
+ single_body_parameters *, two_body_parameters *,
+ static_storage , simulation_data *,
+ list , int , int,
+ real *, real *, real *);
+
+GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *, global_parameters ,
+ single_body_parameters *, two_body_parameters *,
+ static_storage , simulation_data *,
+ list , int , int,
+ real *, real *, real *);
+
+GLOBAL void test_LonePair_Postprocess ( reax_atom *, global_parameters ,
+ single_body_parameters *, two_body_parameters *,
+ static_storage , simulation_data *,
+ list , int , int );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
+
diff --git a/PuReMD-GPU/src/system_props.cu b/PuReMD-GPU/src/cuda_system_props.cu
similarity index 52%
rename from PuReMD-GPU/src/system_props.cu
rename to PuReMD-GPU/src/cuda_system_props.cu
index 3ec39134a3fdd647f8bb81b292d3184e51a6a2b0..7ff5c11fe6d644e6878cd5e25e67b82de39ad0f5 100644
--- a/PuReMD-GPU/src/system_props.cu
+++ b/PuReMD-GPU/src/cuda_system_props.cu
@@ -18,80 +18,29 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "system_props.h"
+#include "cuda_system_props.h"
+
#include "box.h"
#include "vector.h"
-#include "cuda_utils.h"
+#include "cuda_center_mass.h"
#include "cuda_copy.h"
-#include "reduction.h"
-#include "center_mass.h"
-#include "validation.h"
-
-
-real Get_Time( )
-{
- struct timeval tim;
-
- gettimeofday(&tim, NULL );
- return( tim.tv_sec + (tim.tv_usec / 1000000.0) );
-}
-
-
-real Get_Timing_Info( real t_start )
-{
- struct timeval tim;
- real t_end;
+#include "cuda_utils.h"
+#include "cuda_reduction.h"
- gettimeofday(&tim, NULL );
- t_end = tim.tv_sec + (tim.tv_usec / 1000000.0);
- return (t_end - t_start);
-}
+GLOBAL void k_Compute_Total_Mass(single_body_parameters *, reax_atom *, real *, size_t );
+GLOBAL void k_Compute_Kinetic_Energy(single_body_parameters *, reax_atom *, unsigned int , real *);
+GLOBAL void k_Kinetic_Energy_Reduction(simulation_data *, real *, int);
-void Temperature_Control( control_params *control, simulation_data *data,
- output_controls *out_control )
-{
- real tmp;
-
- if( control->T_mode == 1 ) { // step-wise temperature control
- if( (data->step - data->prev_steps) %
- ((int)(control->T_freq / control->dt)) == 0 ) {
- if( fabs( control->T - control->T_final ) >= fabs( control->T_rate ) )
- control->T += control->T_rate;
- else control->T = control->T_final;
- }
- }
- else if( control->T_mode == 2 ) { // constant slope control
- tmp = control->T_rate * control->dt / control->T_freq;
-
- if( fabs( control->T - control->T_final ) >= fabs( tmp ) )
- control->T += tmp;
- }
-}
void prep_dev_system (reax_system *system)
{
//copy the system atoms to the device
- Sync_Host_Device ( system, cudaMemcpyHostToDevice );
+ Sync_Host_Device_Sys( system, cudaMemcpyHostToDevice );
}
-void Compute_Total_Mass( reax_system *system, simulation_data *data )
-{
- int i;
- int blocks;
- int block_size;
- real *partial_sums = 0;
-
- data->M = 0;
-
- for( i = 0; i < system->N; i++ )
- data->M += system->reaxprm.sbp[ system->atoms[i].type ].mass;
-
- data->inv_M = 1. / data->M;
-}
-
void Cuda_Compute_Total_Mass( reax_system *system, simulation_data *data )
{
real *partial_sums = (real *) scratch;
@@ -100,7 +49,7 @@ void Cuda_Compute_Total_Mass( reax_system *system, simulation_data *data )
//cuda_malloc ((void **)&partial_sums, sizeof (real) * (blocks + 1), 1, 0);
cuda_memset (partial_sums, 0, REAL_SIZE * (BLOCKS_POW_2 + 1), RES_SCRATCH );
- Compute_Total_Mass <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
+ k_Compute_Total_Mass <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
(system->reaxprm.d_sbp, system->d_atoms, partial_sums, system->N);
cudaThreadSynchronize ();
cudaCheckError ();
@@ -133,158 +82,6 @@ void Cuda_Compute_Total_Mass( reax_system *system, simulation_data *data )
}
-GLOBAL void Compute_Total_Mass (single_body_parameters *sbp, reax_atom *atoms, real *per_block_results, size_t n)
-{
- extern __shared__ real sdata[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real x = 0;
-
- if(i < n)
- x = sbp [ atoms[ i ].type ].mass;
-
- sdata[threadIdx.x] = x;
- __syncthreads();
-
- for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
- {
- if(threadIdx.x < offset)
- {
- sdata[threadIdx.x] += sdata[threadIdx.x + offset];
- }
- __syncthreads();
- }
-
- if(threadIdx.x == 0)
- {
- per_block_results[blockIdx.x] = sdata[0];
- }
-}
-
-
-void Compute_Center_of_Mass( reax_system *system, simulation_data *data,
- FILE *fout )
-{
- int i;
- real m, xx, xy, xz, yy, yz, zz, det;
- rvec tvec, diff;
- rtensor mat, inv;
-
- int blocks;
- int block_size;
- rvec *l_xcm, *l_vcm, *l_amcm;
- real t_start, t_end;
-
- rvec_MakeZero( data->xcm ); // position of CoM
- rvec_MakeZero( data->vcm ); // velocity of CoM
- rvec_MakeZero( data->amcm ); // angular momentum of CoM
- rvec_MakeZero( data->avcm ); // angular velocity of CoM
-
- /* Compute the position, velocity and angular momentum about the CoM */
- for( i = 0; i < system->N; ++i ) {
- m = system->reaxprm.sbp[ system->atoms[i].type ].mass;
-
- rvec_ScaledAdd( data->xcm, m, system->atoms[i].x );
- rvec_ScaledAdd( data->vcm, m, system->atoms[i].v );
-
- rvec_Cross( tvec, system->atoms[i].x, system->atoms[i].v );
- rvec_ScaledAdd( data->amcm, m, tvec );
-
- /*fprintf( fout,"%3d %g %g %g\n",
- i+1,
- system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2] );
- fprintf( fout, "vcm: %g %g %g\n",
- data->vcm[0], data->vcm[1], data->vcm[2] );
- */
- }
-
- rvec_Scale( data->xcm, data->inv_M, data->xcm );
- rvec_Scale( data->vcm, data->inv_M, data->vcm );
-
- rvec_Cross( tvec, data->xcm, data->vcm );
- rvec_ScaledAdd( data->amcm, -data->M, tvec );
-
- data->etran_cm = 0.5 * data->M * rvec_Norm_Sqr( data->vcm );
-
- /* Calculate and then invert the inertial tensor */
- xx = xy = xz = yy = yz = zz = 0;
-
- for( i = 0; i < system->N; ++i ) {
- m = system->reaxprm.sbp[ system->atoms[i].type ].mass;
-
- rvec_ScaledSum( diff, 1., system->atoms[i].x, -1., data->xcm );
- xx += diff[0] * diff[0] * m;
- xy += diff[0] * diff[1] * m;
- xz += diff[0] * diff[2] * m;
- yy += diff[1] * diff[1] * m;
- yz += diff[1] * diff[2] * m;
- zz += diff[2] * diff[2] * m;
- }
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " xx: %f \n", xx);
- fprintf (stderr, " xy: %f \n", xy);
- fprintf (stderr, " xz: %f \n", xz);
- fprintf (stderr, " yy: %f \n", yy);
- fprintf (stderr, " yz: %f \n", yz);
- fprintf (stderr, " zz: %f \n", zz);
-#endif
-
- mat[0][0] = yy + zz;
- mat[0][1] = mat[1][0] = -xy;
- mat[0][2] = mat[2][0] = -xz;
- mat[1][1] = xx + zz;
- mat[2][1] = mat[1][2] = -yz;
- mat[2][2] = xx + yy;
-
- /* invert the inertial tensor */
- det = ( mat[0][0] * mat[1][1] * mat[2][2] +
- mat[0][1] * mat[1][2] * mat[2][0] +
- mat[0][2] * mat[1][0] * mat[2][1] ) -
- ( mat[0][0] * mat[1][2] * mat[2][1] +
- mat[0][1] * mat[1][0] * mat[2][2] +
- mat[0][2] * mat[1][1] * mat[2][0] );
-
- inv[0][0] = mat[1][1] * mat[2][2] - mat[1][2] * mat[2][1];
- inv[0][1] = mat[0][2] * mat[2][1] - mat[0][1] * mat[2][2];
- inv[0][2] = mat[0][1] * mat[1][2] - mat[0][2] * mat[1][1];
- inv[1][0] = mat[1][2] * mat[2][0] - mat[1][0] * mat[2][2];
- inv[1][1] = mat[0][0] * mat[2][2] - mat[0][2] * mat[2][0];
- inv[1][2] = mat[0][2] * mat[1][0] - mat[0][0] * mat[1][2];
- inv[2][0] = mat[1][0] * mat[2][1] - mat[2][0] * mat[1][1];
- inv[2][1] = mat[2][0] * mat[0][1] - mat[0][0] * mat[2][1];
- inv[2][2] = mat[0][0] * mat[1][1] - mat[1][0] * mat[0][1];
-
- if( fabs(det) > ALMOST_ZERO )
- rtensor_Scale( inv, 1./det, inv );
- else
- rtensor_MakeZero( inv );
-
- /* Compute the angular velocity about the centre of mass */
- rtensor_MatVec( data->avcm, inv, data->amcm );
- data->erot_cm = 0.5 * E_CONV * rvec_Dot( data->avcm, data->amcm );
-
-#if defined(DEBUG)
- fprintf( stderr, "xcm: %24.15e %24.15e %24.15e\n",
- data->xcm[0], data->xcm[1], data->xcm[2] );
- fprintf( stderr, "vcm: %24.15e %24.15e %24.15e\n",
- data->vcm[0], data->vcm[1], data->vcm[2] );
- fprintf( stderr, "amcm: %24.15e %24.15e %24.15e\n",
- data->amcm[0], data->amcm[1], data->amcm[2] );
- /* fprintf( fout, "mat: %f %f %f\n %f %f %f\n %f %f %f\n",
- mat[0][0], mat[0][1], mat[0][2],
- mat[1][0], mat[1][1], mat[1][2],
- mat[2][0], mat[2][1], mat[2][2] );
- fprintf( fout, "inv: %g %g %g\n %g %g %g\n %g %g %g\n",
- inv[0][0], inv[0][1], inv[0][2],
- inv[1][0], inv[1][1], inv[1][2],
- inv[2][0], inv[2][1], inv[2][2] );
- fflush( fout ); */
- fprintf( stderr, "avcm: %24.15e %24.15e %24.15e\n",
- data->avcm[0], data->avcm[1], data->avcm[2] );
-#endif
-}
-
-
void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
FILE *fout )
{
@@ -316,12 +113,12 @@ void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
l_vcm = r_scratch + (BLOCKS_POW_2 + 1);
l_amcm = r_scratch + 2 * (BLOCKS_POW_2 + 1);
- center_of_mass_blocks <<<BLOCKS_POW_2, BLOCK_SIZE, 3 * (RVEC_SIZE * BLOCK_SIZE) >>>
+ k_center_of_mass_blocks<<<BLOCKS_POW_2, BLOCK_SIZE, 3 * (RVEC_SIZE * BLOCK_SIZE) >>>
(system->reaxprm.d_sbp, system->d_atoms, l_xcm, l_vcm, l_amcm, system->N);
cudaThreadSynchronize ();
cudaCheckError ();
- center_of_mass <<<1, BLOCKS_POW_2, 3 * (RVEC_SIZE * BLOCKS_POW_2) >>>
+ k_center_of_mass<<<1, BLOCKS_POW_2, 3 * (RVEC_SIZE * BLOCKS_POW_2) >>>
(l_xcm, l_vcm, l_amcm,
l_xcm + BLOCKS_POW_2,
l_vcm + BLOCKS_POW_2,
@@ -391,18 +188,18 @@ void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
cuda_memset (partial_results, 0, REAL_SIZE * 6 * (BLOCKS_POW_2 + 1), RES_SCRATCH );
local_results = (real *) malloc (REAL_SIZE * 6 *(BLOCKS_POW_2+ 1));
- compute_center_mass <<<BLOCKS_POW_2, BLOCK_SIZE, 6 * (REAL_SIZE * BLOCK_SIZE) >>>
+ k_compute_center_mass_sbp<<<BLOCKS_POW_2, BLOCK_SIZE, 6 * (REAL_SIZE * BLOCK_SIZE) >>>
(system->reaxprm.d_sbp, system->d_atoms, partial_results,
data->xcm[0], data->xcm[1], data->xcm[2], system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
+ cudaThreadSynchronize( );
+ cudaCheckError( );
- compute_center_mass <<<1, BLOCKS_POW_2, 6 * (REAL_SIZE * BLOCKS_POW_2) >>>
+ k_compute_center_mass<<<1, BLOCKS_POW_2, 6 * (REAL_SIZE * BLOCKS_POW_2) >>>
(partial_results, partial_results + (BLOCKS_POW_2 * 6), BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
+ cudaThreadSynchronize( );
+ cudaCheckError( );
- copy_host_device (local_results, partial_results + 6 * BLOCKS_POW_2, REAL_SIZE * 6, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device( local_results, partial_results + 6 * BLOCKS_POW_2, REAL_SIZE * 6, cudaMemcpyDeviceToHost, __LINE__ );
#ifdef __BUILD_DEBUG__
if (check_zero (local_results[0],xx) ||
@@ -456,16 +253,19 @@ void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
inv[2][2] = mat[0][0] * mat[1][1] - mat[1][0] * mat[0][1];
if( fabs(det) > ALMOST_ZERO )
+ {
rtensor_Scale( inv, 1./det, inv );
+ }
else
+ {
rtensor_MakeZero( inv );
+ }
/* Compute the angular velocity about the centre of mass */
rtensor_MatVec( data->avcm, inv, data->amcm );
data->erot_cm = 0.5 * E_CONV * rvec_Dot( data->avcm, data->amcm );
- //free the resources
- free (local_results);
+ free( local_results );
#if defined(DEBUG)
fprintf( stderr, "xcm: %24.15e %24.15e %24.15e\n",
@@ -489,34 +289,51 @@ void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
}
-
-void Compute_Kinetic_Energy( reax_system* system, simulation_data* data )
+void Cuda_Compute_Kinetic_Energy( reax_system *system, simulation_data *data )
{
- int i;
- rvec p;
- real m;
+ real *results = (real *) scratch;
+ cuda_memset (results, 0, REAL_SIZE * BLOCKS_POW_2, RES_SCRATCH);
+ k_Compute_Kinetic_Energy <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
+ (system->reaxprm.d_sbp, system->d_atoms, system->N, (real *) results);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
- data->E_Kin = 0.0;
+ k_Kinetic_Energy_Reduction <<< 1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
+ ((simulation_data *)data->d_simulation_data, results, BLOCKS_POW_2);
+ cudaThreadSynchronize ();
+ cudaCheckError ();
+}
- for (i=0; i < system->N; i++) {
- m = system->reaxprm.sbp[system->atoms[i].type].mass;
- rvec_Scale( p, m, system->atoms[i].v );
- data->E_Kin += 0.5 * rvec_Dot( p, system->atoms[i].v );
+GLOBAL void k_Compute_Total_Mass (single_body_parameters *sbp, reax_atom *atoms, real *per_block_results, size_t n)
+{
+ extern __shared__ real sdata[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real x = 0;
- /* fprintf(stderr,"%d, %lf, %lf, %lf %lf\n",
- i,system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2],
- system->reaxprm.sbp[system->atoms[i].type].mass); */
- }
+ if(i < n)
+ x = sbp [ atoms[ i ].type ].mass;
- data->therm.T = (2. * data->E_Kin) / (data->N_f * K_B);
+ sdata[threadIdx.x] = x;
+ __syncthreads();
+
+ for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
+ {
+ if(threadIdx.x < offset)
+ {
+ sdata[threadIdx.x] += sdata[threadIdx.x + offset];
+ }
+ __syncthreads();
+ }
- if ( fabs(data->therm.T) < ALMOST_ZERO ) /* avoid T being an absolute zero! */
- data->therm.T = ALMOST_ZERO;
+ if(threadIdx.x == 0)
+ {
+ per_block_results[blockIdx.x] = sdata[0];
+ }
}
-GLOBAL void Compute_Kinetic_Energy( single_body_parameters* sbp, reax_atom* atoms,
+GLOBAL void k_Compute_Kinetic_Energy( single_body_parameters* sbp, reax_atom* atoms,
unsigned int N, real *output)
{
extern __shared__ real sh_ekin[];
@@ -547,8 +364,8 @@ GLOBAL void Compute_Kinetic_Energy( single_body_parameters* sbp, reax_atom* atom
}
}
-GLOBAL void Kinetic_Energy_Reduction (simulation_data *data,
- real *input, int n)
+
+GLOBAL void k_Kinetic_Energy_Reduction (simulation_data *data, real *input, int n)
{
extern __shared__ real sdata[];
unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
@@ -582,20 +399,6 @@ GLOBAL void Kinetic_Energy_Reduction (simulation_data *data,
}
}
-void Cuda_Compute_Kinetic_Energy (reax_system *system, simulation_data *data)
-{
- real *results = (real *) scratch;
- cuda_memset (results, 0, REAL_SIZE * BLOCKS_POW_2, RES_SCRATCH);
- Compute_Kinetic_Energy <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (system->reaxprm.d_sbp, system->d_atoms, system->N, (real *) results);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Kinetic_Energy_Reduction <<< 1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- ((simulation_data *)data->d_simulation_data, results, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
/*
GLOBAL void Compute_Kinetic_Energy( single_body_parameters* sbp, reax_atom* atoms,
@@ -658,119 +461,3 @@ data->therm.T = ALMOST_ZERO;
}
}
*/
-
-
-/* IMPORTANT: This function assumes that current kinetic energy and
- * the center of mass of the system is already computed before.
- *
- * IMPORTANT: In Klein's paper, it is stated that a dU/dV term needs
- * to be added when there are long-range interactions or long-range
- * corrections to short-range interactions present.
- * We may want to add that for more accuracy.
- */
-void Compute_Pressure_Isotropic( reax_system* system, control_params *control,
- simulation_data* data,
- output_controls *out_control )
-{
- int i;
- reax_atom *p_atom;
- rvec tx;
- rvec tmp;
- simulation_box *box = &(system->box);
-
- /* Calculate internal pressure */
- rvec_MakeZero( data->int_press );
-
- // 0: both int and ext, 1: ext only, 2: int only
- if( control->press_mode == 0 || control->press_mode == 2 ) {
- for( i = 0; i < system->N; ++i ) {
- p_atom = &( system->atoms[i] );
-
- /* transform x into unitbox coordinates */
- Transform_to_UnitBox( p_atom->x, box, 1, tx );
-
- /* this atom's contribution to internal pressure */
- rvec_Multiply( tmp, p_atom->f, tx );
- rvec_Add( data->int_press, tmp );
-
- if( out_control->debug_level > 0 ) {
- fprintf( out_control->prs, "%-8d%8.2f%8.2f%8.2f",
- i+1, p_atom->x[0], p_atom->x[1], p_atom->x[2] );
- fprintf( out_control->prs, "%8.2f%8.2f%8.2f",
- p_atom->f[0], p_atom->f[1], p_atom->f[2] );
- fprintf( out_control->prs, "%8.2f%8.2f%8.2f\n",
- data->int_press[0],data->int_press[1],data->int_press[2]);
- }
- }
- }
-
- /* kinetic contribution */
- data->kin_press = 2. * (E_CONV * data->E_Kin) / ( 3. * box->volume * P_CONV );
-
- /* Calculate total pressure in each direction */
- data->tot_press[0] = data->kin_press -
- ((data->int_press[0] + data->ext_press[0]) /
- (box->box_norms[1] * box->box_norms[2] * P_CONV));
-
- data->tot_press[1] = data->kin_press -
- ((data->int_press[1] + data->ext_press[1])/
- (box->box_norms[0] * box->box_norms[2] * P_CONV));
-
- data->tot_press[2] = data->kin_press -
- ((data->int_press[2] + data->ext_press[2])/
- (box->box_norms[0] * box->box_norms[1] * P_CONV));
-
- /* Average pressure for the whole box */
- data->iso_bar.P=(data->tot_press[0]+data->tot_press[1]+data->tot_press[2])/3;
-}
-
-
-void Compute_Pressure_Isotropic_Klein( reax_system* system,
- simulation_data* data )
-{
- int i;
- reax_atom *p_atom;
- rvec dx;
-
- // IMPORTANT: This function assumes that current kinetic energy and
- // the center of mass of the system is already computed before.
- data->iso_bar.P = 2.0 * data->E_Kin;
-
- for( i = 0; i < system->N; ++i )
- {
- p_atom = &( system->atoms[i] );
- rvec_ScaledSum(dx,1.0,p_atom->x,-1.0,data->xcm);
- data->iso_bar.P += ( -F_CONV * rvec_Dot(p_atom->f, dx) );
- }
-
- data->iso_bar.P /= (3.0 * system->box.volume);
-
- // IMPORTANT: In Klein's paper, it is stated that a dU/dV term needs
- // to be added when there are long-range interactions or long-range
- // corrections to short-range interactions present.
- // We may want to add that for more accuracy.
-}
-
-
-void Compute_Pressure( reax_system* system, simulation_data* data,
- static_storage *workspace )
-{
- int i;
- reax_atom *p_atom;
- rtensor temp;
-
- rtensor_MakeZero( data->flex_bar.P );
-
- for( i = 0; i < system->N; ++i ) {
- p_atom = &( system->atoms[i] );
- // Distance_on_T3_Gen( data->rcm, p_atom->x, &(system->box), &dx );
- rvec_OuterProduct( temp, p_atom->v, p_atom->v );
- rtensor_ScaledAdd( data->flex_bar.P,
- system->reaxprm.sbp[ p_atom->type ].mass, temp );
- // rvec_OuterProduct(temp, workspace->virial_forces[i], p_atom->x );
- rtensor_ScaledAdd( data->flex_bar.P, -F_CONV, temp );
- }
-
- rtensor_Scale( data->flex_bar.P, 1.0 / system->box.volume, data->flex_bar.P );
- data->iso_bar.P = rtensor_Trace( data->flex_bar.P ) / 3.0;
-}
diff --git a/PuReMD-GPU/src/cuda_system_props.h b/PuReMD-GPU/src/cuda_system_props.h
new file mode 100644
index 0000000000000000000000000000000000000000..026999e9e7bbc6ecc70b0945e7a30bd853a6cbe0
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_system_props.h
@@ -0,0 +1,42 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_SYSTEM_PROP_H_
+#define __CUDA_SYSTEM_PROP_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void prep_dev_system (reax_system *system);
+
+void Cuda_Compute_Total_Mass( reax_system*, simulation_data* );
+void Cuda_Compute_Center_of_Mass( reax_system*, simulation_data*, FILE* );
+void Cuda_Compute_Kinetic_Energy( reax_system*, simulation_data* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_three_body_interactions.cu b/PuReMD-GPU/src/cuda_three_body_interactions.cu
new file mode 100644
index 0000000000000000000000000000000000000000..038b88402b49e516ad4594de520a09464b4c7c8e
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_three_body_interactions.cu
@@ -0,0 +1,1636 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_three_body_interactions.h"
+
+#include "bond_orders.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+#include "index_utils.h"
+
+#include "cuda_helpers.h"
+
+
+/* calculates the theta angle between i-j-k */
+DEVICE void d_Calculate_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
+ real *theta, real *cos_theta )
+{
+ (*cos_theta) = Dot( dvec_ji, dvec_jk, 3 ) / ( d_ji * d_jk );
+ if( *cos_theta > 1. ) *cos_theta = 1.0;
+ if( *cos_theta < -1. ) *cos_theta = -1.0;
+
+ (*theta) = ACOS( *cos_theta );
+}
+
+
+/* calculates the derivative of the cosine of the angle between i-j-k */
+DEVICE void d_Calculate_dCos_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
+ rvec* dcos_theta_di, rvec* dcos_theta_dj,
+ rvec* dcos_theta_dk )
+{
+ int t;
+ real sqr_d_ji = SQR(d_ji);
+ real sqr_d_jk = SQR(d_jk);
+ real inv_dists = 1.0 / (d_ji * d_jk);
+ real inv_dists3 = POW( inv_dists, 3 );
+ real dot_dvecs = Dot( dvec_ji, dvec_jk, 3 );
+ real Cdot_inv3 = dot_dvecs * inv_dists3;
+
+ for( t = 0; t < 3; ++t ) {
+ (*dcos_theta_di)[t] = dvec_jk[t] * inv_dists -
+ Cdot_inv3 * sqr_d_jk * dvec_ji[t];
+
+ (*dcos_theta_dj)[t] = -(dvec_jk[t] + dvec_ji[t]) * inv_dists +
+ Cdot_inv3 * ( sqr_d_jk * dvec_ji[t] + sqr_d_ji * dvec_jk[t] );
+
+ (*dcos_theta_dk)[t] = dvec_ji[t] * inv_dists -
+ Cdot_inv3 * sqr_d_ji * dvec_jk[t];
+ }
+
+ /*fprintf( stderr,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ dvec_jk[t] * inv_dists*/
+}
+
+
+/* this is a 3-body interaction in which the main role is
+ played by j which sits in the middle of the other two. */
+GLOBAL void k_Three_Body_Interactions( reax_atom *atoms,
+ single_body_parameters *sbp,
+ three_body_header *d_thbp,
+ global_parameters g_params,
+ control_params *control,
+ simulation_data *data,
+ static_storage p_workspace,
+ list p_bonds, list p_thb_intrs,
+ int N, int num_atom_types,
+ real *E_Ang, real *E_Pen, real *E_Coa, rvec *aux_ext_press )
+{
+ int i, j, pi, k, pk, t;
+ int type_i, type_j, type_k;
+ int start_j, end_j, start_pk, end_pk;
+ int flag, cnt, num_thb_intrs;
+
+ real temp, temp_bo_jt, pBOjt7;
+ real p_val1, p_val2, p_val3, p_val4, p_val5;
+ real p_val6, p_val7, p_val8, p_val9, p_val10;
+ real p_pen1, p_pen2, p_pen3, p_pen4;
+ real p_coa1, p_coa2, p_coa3, p_coa4;
+ real trm8, expval6, expval7, expval2theta, expval12theta, exp3ij, exp3jk;
+ real exp_pen2ij, exp_pen2jk, exp_pen3, exp_pen4, trm_pen34, exp_coa2;
+ real dSBO1, dSBO2, SBO, SBO2, CSBO2, SBOp, prod_SBO;
+ real CEval1, CEval2, CEval3, CEval4, CEval5, CEval6, CEval7, CEval8;
+ real CEpen1, CEpen2, CEpen3;
+ real e_ang, e_coa, e_pen;
+ real CEcoa1, CEcoa2, CEcoa3, CEcoa4, CEcoa5;
+ real Cf7ij, Cf7jk, Cf8j, Cf9j;
+ real f7_ij, f7_jk, f8_Dj, f9_Dj;
+ real Ctheta_0, theta_0, theta_00, theta, cos_theta, sin_theta;
+ real r_ij, r_jk;
+ real BOA_ij, BOA_jk;
+ real vlpadj;
+ rvec force, ext_press;
+ // rtensor temp_rtensor, total_rtensor;
+ real *total_bo;
+ three_body_header *thbh;
+ three_body_parameters *thbp;
+ three_body_interaction_data *p_ijk, *p_kji;
+ bond_data *pbond_ij, *pbond_jk, *pbond_jt;
+ bond_order_data *bo_ij, *bo_jk, *bo_jt;
+ list *bonds, *thb_intrs;
+ bond_data *bond_list;
+ three_body_interaction_data *thb_list;
+ static_storage *workspace = &p_workspace;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+ if (j >= N) return;
+
+
+ total_bo = workspace->total_bond_order;
+ bonds = &p_bonds;
+ bond_list = bonds->select.bond_list;
+ thb_intrs = &p_thb_intrs;
+ thb_list = thb_intrs->select.three_body_list;
+
+ /* global parameters used in these calculations */
+ p_val6 = g_params.l[14];
+ p_val8 = g_params.l[33];
+ p_val9 = g_params.l[16];
+ p_val10 = g_params.l[17];
+
+ //TODO check this, initially this was zero,
+ // I am changing it to the starting index for this atom.
+ //num_thb_intrs = j * MAX_TH_BODY;
+
+ //for( j = 0; j < system->N; ++j ) {
+ // fprintf( out_control->eval, "j: %d\n", j );
+ type_j = atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+
+ p_val3 = sbp[ type_j ].p_val3;
+ p_val5 = sbp[ type_j ].p_val5;
+
+ SBOp = 0, prod_SBO = 1;
+ for( t = start_j; t < end_j; ++t ) {
+ bo_jt = &(bond_list[t].bo_data);
+ SBOp += (bo_jt->BO_pi + bo_jt->BO_pi2);
+ temp = SQR( bo_jt->BO );
+ temp *= temp;
+ temp *= temp;
+ prod_SBO *= EXP( -temp );
+ }
+
+ /* modifications to match Adri's code - 09/01/09 */
+ if( workspace->vlpex[j] >= 0 ){
+ vlpadj = 0;
+ dSBO2 = prod_SBO - 1;
+ }
+ else{
+ vlpadj = workspace->nlp[j];
+ dSBO2 = (prod_SBO - 1) * (1 - p_val8 * workspace->dDelta_lp[j]);
+ }
+
+ SBO = SBOp + (1 - prod_SBO) * (-workspace->Delta_boc[j] - p_val8 * vlpadj);
+ dSBO1 = -8 * prod_SBO * ( workspace->Delta_boc[j] + p_val8 * vlpadj );
+
+ if( SBO <= 0 )
+ SBO2 = 0, CSBO2 = 0;
+ else if( SBO > 0 && SBO <= 1 ) {
+ SBO2 = POW( SBO, p_val9 );
+ CSBO2 = p_val9 * POW( SBO, p_val9 - 1 );
+ }
+ else if( SBO > 1 && SBO < 2 ) {
+ SBO2 = 2 - POW( 2-SBO, p_val9 );
+ CSBO2 = p_val9 * POW( 2 - SBO, p_val9 - 1 );
+ }
+ else
+ SBO2 = 2, CSBO2 = 0;
+
+ expval6 = EXP( p_val6 * workspace->Delta_boc[j] );
+
+ /* unlike 2-body intrs where we enforce i<j, we cannot put any such
+ restrictions here. such a restriction would prevent us from producing
+ all 4-body intrs correctly */
+ for( pi = start_j; pi < end_j; ++pi ) {
+
+ //TODO
+ //num_thb_intrs = pi * MAX_THREE_BODIES;
+ //TODO
+
+ //Set_Start_Index( pi, num_thb_intrs, thb_intrs );
+ num_thb_intrs = Start_Index (pi, thb_intrs);
+
+ pbond_ij = &(bond_list[pi]);
+ bo_ij = &(pbond_ij->bo_data);
+ BOA_ij = bo_ij->BO - control->thb_cut;
+
+
+ if( BOA_ij/*bo_ij->BO*/ > 0.0 ) {
+ i = pbond_ij->nbr;
+ r_ij = pbond_ij->d;
+ type_i = atoms[i].type;
+ // fprintf( out_control->eval, "i: %d\n", i );
+
+
+ /* first copy 3-body intrs from previously computed ones where i>k.
+IMPORTANT: if it is less costly to compute theta and its
+derivative, we should definitely re-compute them,
+instead of copying!
+in the second for-loop below, we compute only new 3-body intrs
+where i < k */
+ for( pk = start_j; pk < pi; ++pk ) {
+ // fprintf( out_control->eval, "pk: %d\n", pk );
+ start_pk = Start_Index( pk, thb_intrs );
+ end_pk = End_Index( pk, thb_intrs );
+
+ for( t = start_pk; t < end_pk; ++t )
+ if( thb_list[t].thb == i ) {
+ p_ijk = &(thb_list[num_thb_intrs]);
+ p_kji = &(thb_list[t]);
+
+ p_ijk->thb = bond_list[pk].nbr;
+ p_ijk->pthb = pk;
+ p_ijk->theta = p_kji->theta;
+ rvec_Copy( p_ijk->dcos_di, p_kji->dcos_dk );
+ rvec_Copy( p_ijk->dcos_dj, p_kji->dcos_dj );
+ rvec_Copy( p_ijk->dcos_dk, p_kji->dcos_di );
+
+ ++num_thb_intrs;
+ break;
+ }
+ }
+
+
+ /* and this is the second for loop mentioned above */
+ for( pk = pi+1; pk < end_j; ++pk ) {
+ pbond_jk = &(bond_list[pk]);
+ bo_jk = &(pbond_jk->bo_data);
+ BOA_jk = bo_jk->BO - control->thb_cut;
+ k = pbond_jk->nbr;
+ type_k = atoms[k].type;
+ p_ijk = &( thb_list[num_thb_intrs] );
+
+ //CHANGE ORIGINAL
+ if (BOA_jk <= 0) continue;
+ //CHANGE ORIGINAL
+
+ d_Calculate_Theta( pbond_ij->dvec, pbond_ij->d,
+ pbond_jk->dvec, pbond_jk->d,
+ &theta, &cos_theta );
+
+ d_Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d,
+ pbond_jk->dvec, pbond_jk->d,
+ &(p_ijk->dcos_di), &(p_ijk->dcos_dj),
+ &(p_ijk->dcos_dk) );
+
+ p_ijk->thb = k;
+ p_ijk->pthb = pk;
+ p_ijk->theta = theta;
+
+ sin_theta = SIN( theta );
+ if( sin_theta < 1.0e-5 )
+ sin_theta = 1.0e-5;
+
+ ++num_thb_intrs;
+
+
+ if( BOA_jk > 0.0 &&
+ (bo_ij->BO * bo_jk->BO) > SQR(control->thb_cut)/*0*/) {
+ r_jk = pbond_jk->d;
+ thbh = &( d_thbp[ index_thbp(type_i,type_j,type_k,num_atom_types) ] );
+ flag = 0;
+
+ /* if( workspace->orig_id[i] < workspace->orig_id[k] )
+ fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], bo_ij->BO, bo_jk->BO, p_ijk->theta );
+ else
+ fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
+ workspace->orig_id[k], workspace->orig_id[j],
+ workspace->orig_id[i], bo_jk->BO, bo_ij->BO, p_ijk->theta ); */
+
+ //TODO:
+ //pbond_jk->scratch = thbh->cnt;
+
+ for( cnt = 0; cnt < thbh->cnt; ++cnt ) {
+ // fprintf( out_control->eval,
+ // "%6d%6d%6d -- exists in thbp\n", i+1, j+1, k+1 );
+
+ if( fabs(thbh->prm[cnt].p_val1) > 0.001 ) {
+ thbp = &( thbh->prm[cnt] );
+
+ /* ANGLE ENERGY */
+ p_val1 = thbp->p_val1;
+ p_val2 = thbp->p_val2;
+ p_val4 = thbp->p_val4;
+ p_val7 = thbp->p_val7;
+ theta_00 = thbp->theta_00;
+
+ exp3ij = EXP( -p_val3 * POW( BOA_ij, p_val4 ) );
+ f7_ij = 1.0 - exp3ij;
+ Cf7ij = p_val3 * p_val4 *
+ POW( BOA_ij, p_val4 - 1.0 ) * exp3ij;
+
+ exp3jk = EXP( -p_val3 * POW( BOA_jk, p_val4 ) );
+ f7_jk = 1.0 - exp3jk;
+ Cf7jk = p_val3 * p_val4 *
+ POW( BOA_jk, p_val4 - 1.0 ) * exp3jk;
+
+ expval7 = EXP( -p_val7 * workspace->Delta_boc[j] );
+ trm8 = 1.0 + expval6 + expval7;
+ f8_Dj = p_val5 - ( (p_val5 - 1.0) * (2.0 + expval6) / trm8 );
+ Cf8j = ( (1.0 - p_val5) / SQR(trm8) ) *
+ (p_val6 * expval6 * trm8 -
+ (2.0 + expval6) * ( p_val6 * expval6 - p_val7 * expval7 ));
+
+ theta_0 = 180.0 -
+ theta_00 * (1.0 - EXP(-p_val10 * (2.0 - SBO2)));
+ theta_0 = DEG2RAD( theta_0 );
+
+ expval2theta = EXP(-p_val2 * SQR(theta_0-theta));
+ if( p_val1 >= 0 )
+ expval12theta = p_val1 * (1.0 - expval2theta);
+ else // To avoid linear Me-H-Me angles (6/6/06)
+ expval12theta = p_val1 * -expval2theta;
+
+ CEval1 = Cf7ij * f7_jk * f8_Dj * expval12theta;
+ CEval2 = Cf7jk * f7_ij * f8_Dj * expval12theta;
+ CEval3 = Cf8j * f7_ij * f7_jk * expval12theta;
+ CEval4 = -2.0 * p_val1 * p_val2 * f7_ij * f7_jk * f8_Dj *
+ expval2theta * (theta_0 - theta);
+
+ Ctheta_0 = p_val10 * DEG2RAD(theta_00) *
+ exp( -p_val10 * (2.0 - SBO2) );
+
+ CEval5 = -CEval4 * Ctheta_0 * CSBO2;
+ CEval6 = CEval5 * dSBO1;
+ CEval7 = CEval5 * dSBO2;
+ CEval8 = -CEval4 / sin_theta;
+
+ e_ang = f7_ij * f7_jk * f8_Dj * expval12theta;
+ //PERFORMANCE IMPACT
+ //MYATOMICADD(&data->E_Ang, e_ang);
+ E_Ang [j] += e_ang;
+ /* END ANGLE ENERGY*/
+
+
+ /* PENALTY ENERGY */
+ p_pen1 = thbp->p_pen1;
+ p_pen2 = g_params.l[19];
+ p_pen3 = g_params.l[20];
+ p_pen4 = g_params.l[21];
+
+ exp_pen2ij = EXP( -p_pen2 * SQR( BOA_ij - 2.0 ) );
+ exp_pen2jk = EXP( -p_pen2 * SQR( BOA_jk - 2.0 ) );
+ exp_pen3 = EXP( -p_pen3 * workspace->Delta[j] );
+ exp_pen4 = EXP( p_pen4 * workspace->Delta[j] );
+ trm_pen34 = 1.0 + exp_pen3 + exp_pen4;
+ f9_Dj = ( 2.0 + exp_pen3 ) / trm_pen34;
+ Cf9j = (-p_pen3 * exp_pen3 * trm_pen34 -
+ (2.0 + exp_pen3) * ( -p_pen3 * exp_pen3 +
+ p_pen4 * exp_pen4 )) /
+ SQR( trm_pen34 );
+
+ e_pen = p_pen1 * f9_Dj * exp_pen2ij * exp_pen2jk;
+ //PERFORMANCE IMPACT
+ //MYATOMICADD(&data->E_Pen, e_pen);
+ E_Pen [j] += e_pen;
+
+
+ CEpen1 = e_pen * Cf9j / f9_Dj;
+ temp = -2.0 * p_pen2 * e_pen;
+ CEpen2 = temp * (BOA_ij - 2.0);
+ CEpen3 = temp * (BOA_jk - 2.0);
+ /* END PENALTY ENERGY */
+
+
+ /* COALITION ENERGY */
+ p_coa1 = thbp->p_coa1;
+ p_coa2 = g_params.l[2];
+ p_coa3 = g_params.l[38];
+ p_coa4 = g_params.l[30];
+
+ exp_coa2 = EXP( p_coa2 * workspace->Delta_boc[j] );
+ e_coa =
+ p_coa1 / (1. + exp_coa2) *
+ EXP( -p_coa3 * SQR(total_bo[i] - BOA_ij) ) *
+ EXP( -p_coa3 * SQR(total_bo[k] - BOA_jk) ) *
+ EXP( -p_coa4 * SQR(BOA_ij - 1.5) ) *
+ EXP( -p_coa4 * SQR(BOA_jk - 1.5) );
+
+ //PERFORMANCE IMPACT
+ //MYATOMICADD(&data->E_Coa, e_coa);
+ E_Coa [j] += e_coa;
+
+ CEcoa1 = -2 * p_coa4 * (BOA_ij - 1.5) * e_coa;
+ CEcoa2 = -2 * p_coa4 * (BOA_jk - 1.5) * e_coa;
+ CEcoa3 = -p_coa2 * exp_coa2 * e_coa / (1+exp_coa2);
+ CEcoa4 = -2*p_coa3 * (total_bo[i]-BOA_ij) * e_coa;
+ CEcoa5 = -2*p_coa3 * (total_bo[k]-BOA_jk) * e_coa;
+ /* END COALITION ENERGY */
+
+ /* FORCES */
+ /*
+ MYATOMICADD(&bo_ij->Cdbo, (CEval1 + CEpen2 + (CEcoa1-CEcoa4)) );
+ MYATOMICADD(&bo_jk->Cdbo, (CEval2 + CEpen3 + (CEcoa2-CEcoa5)) );
+ MYATOMICADD(&workspace->CdDelta[j], ((CEval3 + CEval7) + CEpen1 + CEcoa3) );
+ MYATOMICADD(&workspace->CdDelta[i], CEcoa4 );
+ MYATOMICADD(&workspace->CdDelta[k], CEcoa5 );
+ */
+
+ bo_ij->Cdbo += (CEval1 + CEpen2 + (CEcoa1-CEcoa4)) ;
+ bo_jk->Cdbo += (CEval2 + CEpen3 + (CEcoa2-CEcoa5)) ;
+ workspace->CdDelta[j] += ((CEval3 + CEval7) + CEpen1 + CEcoa3) ;
+ //MYATOMICADD(&workspace->CdDelta[i], CEcoa4 );
+ pbond_ij->CdDelta_ij += CEcoa4 ;
+ //MYATOMICADD(&workspace->CdDelta[k], CEcoa5 );
+ pbond_jk->CdDelta_ij += CEcoa5;
+
+ for( t = start_j; t < end_j; ++t ) {
+ pbond_jt = &( bond_list[t] );
+ bo_jt = &(pbond_jt->bo_data);
+ temp_bo_jt = bo_jt->BO;
+ temp = CUBE( temp_bo_jt );
+ pBOjt7 = temp * temp * temp_bo_jt;
+
+ // fprintf( out_control->eval, "%6d%12.8f\n",
+ // workspace->orig_id[ bond_list[t].nbr ],
+ // (CEval6 * pBOjt7) );
+
+ /*
+ MYATOMICADD(&bo_jt->Cdbo, (CEval6 * pBOjt7) );
+ MYATOMICADD(&bo_jt->Cdbopi, CEval5 );
+ MYATOMICADD(&bo_jt->Cdbopi2, CEval5 );
+ */
+ bo_jt->Cdbo += (CEval6 * pBOjt7) ;
+ bo_jt->Cdbopi += CEval5 ;
+ bo_jt->Cdbopi2 += CEval5 ;
+ }
+
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ /*
+ atomic_rvecScaledAdd( atoms[i].f, CEval8, p_ijk->dcos_di );
+ atomic_rvecScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
+ atomic_rvecScaledAdd( atoms[k].f, CEval8, p_ijk->dcos_dk );
+ */
+ rvec_ScaledAdd( pbond_ij->f, CEval8, p_ijk->dcos_di );
+ rvec_ScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
+ rvec_ScaledAdd( pbond_jk->f, CEval8, p_ijk->dcos_dk );
+
+
+ }
+ else {
+ /* terms not related to bond order derivatives
+ are added directly into
+ forces and pressure vector/tensor */
+ rvec_Scale( force, CEval8, p_ijk->dcos_di );
+ //atomic_rvecAdd( atoms[i].f, force );
+ rvec_Add( pbond_ij->f, force );
+
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ //atomic_rvecAdd( data->ext_press, ext_press );
+ rvec_Add( aux_ext_press [j], ext_press );
+
+ //atomic_rvecScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
+ rvec_ScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
+
+ rvec_Scale( force, CEval8, p_ijk->dcos_dk );
+ //atomic_rvecAdd( atoms[k].f, force );
+ rvec_Add( pbond_jk->f, force );
+ rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
+ //atomic_rvecAdd( data->ext_press, ext_press );
+ rvec_Add( aux_ext_press [j], ext_press );
+
+
+ /* This part is for a fully-flexible box */
+ /* rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_di, system->atoms[i].x );
+ rtensor_Scale( total_rtensor, +CEval8, temp_rtensor );
+
+ rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_dj, system->atoms[j].x );
+ rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
+
+ rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_dk, system->atoms[k].x );
+ rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
+
+ if( pbond_ij->imaginary || pbond_jk->imaginary )
+ rtensor_ScaledAdd( data->flex_bar.P,
+ -1.0, total_rtensor );
+ else
+ rtensor_Add( data->flex_bar.P, total_rtensor ); */
+ }
+
+#ifdef TEST_ENERGY
+ //TODO -- check this
+ // fprintf( out_control->eval,
+ //"%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e",
+ // "%6d%6d%6d%23.15e%23.15e%23.15e\n",
+ // i+1, j+1, k+1,
+ //workspace->orig_id[i]+1,
+ //workspace->orig_id[j]+1,
+ //workspace->orig_id[k]+1,
+ //workspace->Delta_boc[j],
+ // RAD2DEG(theta), /*BOA_ij, BOA_jk, */
+ // e_ang, data->E_Ang );
+
+ /*fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e",
+ p_val3, p_val4, BOA_ij, BOA_jk );
+ fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e",
+ f7_ij, f7_jk, f8_Dj, expval12theta );
+ fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ CEval1, CEval2, CEval3, CEval4, CEval5
+ //CEval6, CEval7, CEval8 );*/
+
+ /*fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ -p_ijk->dcos_di[0]/sin_theta,
+ -p_ijk->dcos_di[1]/sin_theta,
+ -p_ijk->dcos_di[2]/sin_theta,
+ -p_ijk->dcos_dj[0]/sin_theta,
+ -p_ijk->dcos_dj[1]/sin_theta,
+ -p_ijk->dcos_dj[2]/sin_theta,
+ -p_ijk->dcos_dk[0]/sin_theta,
+ -p_ijk->dcos_dk[1]/sin_theta,
+ -p_ijk->dcos_dk[2]/sin_theta );*/
+
+ /* fprintf( out_control->epen,
+ "%23.15e%23.15e%23.15e\n",
+ CEpen1, CEpen2, CEpen3 );
+ fprintf( out_control->epen,
+ "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], RAD2DEG(theta),
+ BOA_ij, BOA_jk, e_pen, data->E_Pen ); */
+
+ // fprintf( out_control->ecoa,
+ // "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ // workspace->orig_id[i],
+ // workspace->orig_id[j],
+ // workspace->orig_id[k],
+ // RAD2DEG(theta), BOA_ij, BOA_jk,
+ // e_coa, data->E_Coa );
+#endif
+
+#ifdef TEST_FORCES /* angle forces */
+ //TODO -- check this
+ /*
+ Add_dBO( system, lists, j, pi, CEval1, workspace->f_ang );
+ Add_dBO( system, lists, j, pk, CEval2, workspace->f_ang );
+ Add_dDelta( system, lists,
+ j, CEval3 + CEval7, workspace->f_ang );
+
+ for( t = start_j; t < end_j; ++t ) {
+ pbond_jt = &( bond_list[t] );
+ bo_jt = &(pbond_jt->bo_data);
+ temp_bo_jt = bo_jt->BO;
+ temp = CUBE( temp_bo_jt );
+ pBOjt7 = temp * temp * temp_bo_jt;
+
+ Add_dBO( system, lists, j, t, pBOjt7 * CEval6,
+ workspace->f_ang );
+ Add_dBOpinpi2( system, lists, j, t,
+ CEval5, CEval5,
+ workspace->f_ang, workspace->f_ang );
+ }
+
+ rvec_ScaledAdd( workspace->f_ang[i], CEval8, p_ijk->dcos_di );
+ rvec_ScaledAdd( workspace->f_ang[j], CEval8, p_ijk->dcos_dj );
+ rvec_ScaledAdd( workspace->f_ang[k], CEval8, p_ijk->dcos_dk );
+ // end angle forces
+
+ // penalty forces
+ Add_dDelta( system, lists, j, CEpen1, workspace->f_pen );
+ Add_dBO( system, lists, j, pi, CEpen2, workspace->f_pen );
+ Add_dBO( system, lists, j, pk, CEpen3, workspace->f_pen );
+ // end penalty forces
+
+ // coalition forces
+ Add_dBO( system, lists,
+ j, pi, CEcoa1-CEcoa4, workspace->f_coa );
+ Add_dBO( system, lists,
+ j, pk, CEcoa2-CEcoa5, workspace->f_coa );
+ Add_dDelta( system, lists, j, CEcoa3, workspace->f_coa );
+ Add_dDelta( system, lists, i, CEcoa4, workspace->f_coa );
+ Add_dDelta( system, lists, k, CEcoa5, workspace->f_coa );
+ // end coalition forces
+
+ */
+#endif
+ }
+ }
+ }
+ }
+ }
+
+ Set_End_Index(pi, num_thb_intrs, thb_intrs );
+ }
+ // } // end of the main for loop here
+
+
+ //TODO - to be done on the CPU
+ /*
+
+ if( num_thb_intrs >= thb_intrs->num_intrs * DANGER_ZONE ) {
+ workspace->realloc.num_3body = num_thb_intrs;
+ if( num_thb_intrs > thb_intrs->num_intrs ) {
+ fprintf( stderr, "step%d-ran out of space on angle_list: top=%d, max=%d",
+ data->step, num_thb_intrs, thb_intrs->num_intrs );
+ exit( INSUFFICIENT_SPACE );
+ }
+ }
+ */
+
+ //fprintf( stderr,"%d: Number of angle interactions: %d\n",
+ // data->step, num_thb_intrs );
+
+#ifdef TEST_ENERGY
+ /*
+ fprintf( stderr,"Number of angle interactions: %d\n", num_thb_intrs );
+
+ fprintf( stderr,"Angle Energy:%g\t Penalty Energy:%g\t Coalition Energy:%g\n",
+ data->E_Ang, data->E_Pen, data->E_Coa );
+
+ fprintf( stderr,"3body: ext_press (%23.15e %23.15e %23.15e)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] );
+ */
+#endif
+}
+
+
+GLOBAL void k_Three_Body_Interactions_results ( reax_atom *atoms, control_params *control,
+ static_storage p_workspace,
+ list p_bonds, int N )
+{
+ int i, pj;
+
+ bond_data *pbond;
+ bond_data *sym_index_bond;
+ list *bonds = &p_bonds;
+ static_storage *workspace = &p_workspace;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( i >= N) return;
+
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
+
+ pbond = &(bonds->select.bond_list[pj]);
+ sym_index_bond = &( bonds->select.bond_list[ pbond->sym_index ] );
+
+ workspace->CdDelta [i] += sym_index_bond->CdDelta_ij;
+
+ rvec_Add (atoms[i].f, sym_index_bond->f );
+ }
+}
+
+
+/* this is a 3-body interaction in which the main role is
+ played by j which sits in the middle of the other two. */
+GLOBAL void k_Three_Body_Estimate ( reax_atom *atoms,
+ control_params *control,
+ list p_bonds, int N,
+ int *count)
+{
+ int i, j, pi, k, pk, t;
+ int type_i, type_j, type_k;
+ int start_j, end_j ;
+ int flag, cnt, num_thb_intrs;
+
+ real r_ij, r_jk;
+ real BOA_ij, BOA_jk;
+ list *bonds;
+
+ bond_order_data *bo_ij, *bo_jk, *bo_jt;
+ bond_data *bond_list;
+ bond_data *pbond_ij, *pbond_jk, *pbond_jt;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+ if (j >= N) return;
+
+ bonds = &p_bonds;
+ bond_list = bonds->select.bond_list;
+
+ type_j = atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+
+
+ for( pi = start_j; pi < end_j; ++pi ) {
+
+ num_thb_intrs = 0;
+ count [pi] = 0;
+
+ pbond_ij = &(bond_list[pi]);
+ bo_ij = &(pbond_ij->bo_data);
+ BOA_ij = bo_ij->BO - control->thb_cut;
+
+ if( BOA_ij/*bo_ij->BO*/ > 0.0 ) {
+ i = pbond_ij->nbr;
+ r_ij = pbond_ij->d;
+ type_i = atoms[i].type;
+
+ /*
+ for( pk = start_j; pk < pi; ++pk ) {
+ start_pk = Start_Index( pk, thb_intrs );
+ end_pk = End_Index( pk, thb_intrs );
+
+ for( t = start_pk; t < end_pk; ++t )
+ if( thb_list[t].thb == i ) {
+
+ ++num_thb_intrs;
+ break;
+ }
+ }
+ */
+
+ /* and this is the second for loop mentioned above */
+ for( pk = start_j; pk < end_j; ++pk ) {
+ if (pk == pi) continue;
+
+ pbond_jk = &(bond_list[pk]);
+ bo_jk = &(pbond_jk->bo_data);
+ BOA_jk = bo_jk->BO - control->thb_cut;
+
+ if (BOA_jk <= 0) continue;
+
+ ++num_thb_intrs;
+ }
+ }
+
+ count [pi] = num_thb_intrs;
+ }
+}
+
+
+GLOBAL void k_Hydrogen_Bonds(reax_atom *atoms,
+ single_body_parameters *sbp,
+ hbond_parameters *d_hbp,
+ control_params *control,
+ simulation_data *data,
+ static_storage p_workspace,
+ list p_bonds, list p_hbonds,
+ int N, int num_atom_types,
+ real *E_HB, rvec *aux_ext_press, rvec *atoms_f )
+{
+ extern __shared__ real t_hb[];
+ extern __shared__ real t_f[];
+ //extern __shared__ rvec t_cdbo[];
+ //extern __shared__ rvec t_hf [];
+
+ real *sh_hb = t_hb;
+ rvec *sh_atomf = (rvec *)(t_hb + blockDim.x);
+ //real *sh_cdbo = t_hb + blockDim.x;
+ //rvec *sh_hf = (rvec *) (sh_atomf + blockDim.x);
+
+ int i, j, k, pi, pk, itr, top;
+ int type_i, type_j, type_k;
+ int start_j, end_j, hb_start_j, hb_end_j;
+ int hblist[MAX_BONDS];
+ int num_hb_intrs = 0;
+ real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
+ real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
+ rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
+ rvec dvec_jk, force, ext_press;
+ ivec rel_jk;
+ // rtensor temp_rtensor, total_rtensor;
+ hbond_parameters *hbp;
+ bond_order_data *bo_ij;
+ bond_data *pbond_ij;
+ far_neighbor_data *nbr_jk;
+ list *bonds, *hbonds;
+ bond_data *bond_list;
+ hbond_data *hbond_list, *hbond_jk;
+ static_storage *workspace = &p_workspace;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+ if (j >= N) return;
+
+ //j = blockIdx.x;
+
+ bonds = &p_bonds;
+ bond_list = bonds->select.bond_list;
+
+ hbonds = &p_hbonds;
+ hbond_list = hbonds->select.hbond_list;
+
+ // loops below discover the Hydrogen bonds between i-j-k triplets.
+ // here j is H atom and there has to be some bond between i and j.
+ // Hydrogen bond is between j and k.
+ // so in this function i->X, j->H, k->Z when we map
+ // variables onto the ones in the handout.
+
+ //for( j = 0; j < system->N; ++j )
+ sh_hb [threadIdx.x] = 0;
+ rvec_MakeZero ( sh_atomf[ threadIdx.x] );
+
+ if( sbp[atoms[j].type].p_hbond==1) {// j must be H
+ //set j's variables
+ type_j = atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+ hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
+ hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
+
+ top = 0;
+ for( pi = start_j; pi < end_j; ++pi ) {
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = atoms[i].type;
+
+ if( sbp[type_i].p_hbond == 2 &&
+ bo_ij->BO >= HB_THRESHOLD )
+ hblist[top++] = pi;
+ }
+
+ // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
+ // j, top, hb_start_j, hb_end_j );
+
+ for( pk = hb_start_j; pk < hb_end_j; ++pk )
+ //pk = hb_start_j + threadIdx.x;
+ //while (pk < hb_end_j)
+ {
+ // set k's varibles
+ //TODO
+ hbond_jk = &( hbond_list[pk] );
+ //TODO
+ k = hbond_list[pk].nbr;
+ type_k = atoms[k].type;
+ nbr_jk = hbond_list[pk].ptr;
+ r_jk = nbr_jk->d;
+ rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
+
+ //TODO Double check this Hydrogen Bonds fix
+ //rvec_MakeZero ( nbr_jk->h_f );
+ rvec_MakeZero ( hbond_jk->h_f );
+ //TODO Double check this Hydrogen Bonds fix
+
+ //sh_hb [threadIdx.x] = 0;
+
+
+ //itr = threadIdx.x;
+ for( itr=0; itr < top; ++itr ) {
+ //while (itr < top) {
+ pi = hblist[itr];
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+
+ //TODO
+ //rvec_MakeZero (sh_hf [threadIdx.x]);
+ //sh_cdbo [threadIdx.x] = 0;
+
+ //rvec_MakeZero ( sh_atomf[ threadIdx.x] );
+
+
+ if( i != k ) {
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = atoms[i].type;
+ r_ij = pbond_ij->d;
+ hbp = &(d_hbp[ index_hbp(type_i, type_j, type_k, num_atom_types) ]);
+ ++num_hb_intrs;
+
+ d_Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &theta, &cos_theta );
+ // the derivative of cos(theta)
+ d_Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &dcos_theta_di, &dcos_theta_dj,
+ &dcos_theta_dk );
+
+ // hydrogen bond energy
+ sin_theta2 = SIN( theta/2.0 );
+ sin_xhz4 = SQR(sin_theta2);
+ sin_xhz4 *= sin_xhz4;
+ cos_xhz1 = ( 1.0 - cos_theta );
+ exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
+ exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
+ r_jk / hbp->r0_hb - 2.0 ) );
+
+ //PERFORMANCE IMPACT
+ e_hb = hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
+ //MYATOMICADD( &data->E_HB, e_hb );
+ //E_HB [j] += e_hb;
+ sh_hb [threadIdx.x] += e_hb;
+
+ CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
+ CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
+ CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
+ 1.0 / hbp->r0_hb);
+
+ //this is the problem here
+ //TODO
+ // hydrogen bond forces
+ bo_ij->Cdbo += CEhb1; // dbo term
+ //sh_cdbo[threadIdx.x] += CEhb1;
+ //TODO
+
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+
+ //PERFORMANCE IMPACT
+ /*
+ atomic_rvecScaledAdd( atoms[i].f,
+ +CEhb2, dcos_theta_di ); //dcos terms
+ atomic_rvecScaledAdd( atoms[j].f,
+ +CEhb2, dcos_theta_dj );
+ atomic_rvecScaledAdd( atoms[k].f,
+ +CEhb2, dcos_theta_dk );
+ //dr terms
+ atomic_rvecScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+ atomic_rvecScaledAdd( atoms[k].f, +CEhb3/r_jk, dvec_jk );
+ */
+
+ //PERFORMANCE IMPACT
+ rvec_ScaledAdd( pbond_ij->h_f, +CEhb2, dcos_theta_di ); //dcos terms
+ //rvec_ScaledAdd( sh_hf [threadIdx.x], +CEhb2, dcos_theta_di ); //dcos terms
+
+ //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
+ rvec_ScaledAdd( sh_atomf [threadIdx.x], +CEhb2, dcos_theta_dj );
+
+ //TODO you forgot here
+ //TODO Hydrogen bonds fix. -- BE VERY CAREFUL *****
+ rvec_ScaledAdd( hbond_jk->h_f,
+ +CEhb2, dcos_theta_dk );
+
+ //rvec_ScaledAdd( nbr_jk->h_f,
+ // +CEhb2, dcos_theta_dk );
+
+ //dr terms
+ //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+ rvec_ScaledAdd( sh_atomf [threadIdx.x], -CEhb3/r_jk, dvec_jk );
+
+ //atoms_f [j] ++;
+
+ //TODO you forgot
+ rvec_ScaledAdd( hbond_jk->h_f, +CEhb3/r_jk, dvec_jk );
+ //rvec_ScaledAdd( nbr_jk->h_f, +CEhb3/r_jk, dvec_jk );
+ }
+ else
+ {
+ // for pressure coupling, terms that are not related
+ // to bond order derivatives are added directly into
+ // pressure vector/tensor
+ rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
+ rvec_Add( pbond_ij->h_f, force );
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd (sh_press [threadIdx.x], 1.0, ext_press );
+
+ rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
+
+ ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
+ rvec_Scale( force, +CEhb2, dcos_theta_dk );
+
+ //rvec_Add( nbr_jk->h_f, force );
+ rvec_Add( hbond_jk->h_f, force );
+
+ rvec_iMultiply( ext_press, rel_jk, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
+
+ //dr terms
+ rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+
+ rvec_Scale( force, CEhb3/r_jk, dvec_jk );
+ rvec_Add( hbond_jk->h_f, force );
+ rvec_iMultiply( ext_press, rel_jk, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
+
+ }
+
+ //do the reduction for the bond_ij here
+ /*
+ if (threadIdx.x < 16){
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 16];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 16]);
+
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
+ }
+ if (threadIdx.x < 8){
+ //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 8];
+ //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 8]);
+
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
+ }
+ if (threadIdx.x < 4){
+ //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 4];
+ //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 4]);
+
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
+ }
+ if (threadIdx.x < 2){
+ //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 2];
+ //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 2]);
+
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
+ }
+ if (threadIdx.x < 1){
+ //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 1];
+ //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 1]);
+
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
+ }
+ if (threadIdx.x == 0){
+ //bo_ij->Cdbo += sh_cdbo [threadIdx.x];
+ //rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
+
+ E_HB [j] += sh_hb [threadIdx.x];
+ //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+ }
+ */
+
+
+ } // i != k if statement
+
+
+ //itr += blockDim.x;
+
+ } //itr for statement
+
+ /*
+ __syncthreads ();
+
+ for (int x = 1; x < blockDim.x; x++)
+ sh_hb [0] += sh_hb [x];
+
+ E_HB [j] += sh_hb[0];
+ if (threadIdx.x < 16) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
+ if (threadIdx.x < 8) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
+ if (threadIdx.x < 4) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
+ if (threadIdx.x < 2) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
+ if (threadIdx.x < 1) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
+ if (threadIdx.x == 0) E_HB [j] += sh_hb [threadIdx.x];
+ */
+
+
+ //pk += blockDim.x;
+
+ } // pk for statement
+ } // main if statment
+
+ //do the reduction for the bond_ij here
+ /*
+ if (threadIdx.x < 16){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
+ }
+ if (threadIdx.x < 8){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
+ }
+ if (threadIdx.x < 4){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
+ }
+ if (threadIdx.x < 2){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
+ }
+ if (threadIdx.x < 1){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
+ //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
+ }
+ if (threadIdx.x == 0){
+ E_HB [j] += sh_hb [threadIdx.x];
+ //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+ }
+ */
+
+ E_HB [j] += sh_hb [threadIdx.x];
+ rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+
+ //rvec_Copy (atoms_f [j], sh_atomf [threadIdx.x]);
+}
+
+
+DEVICE void warpReduce(volatile real* sdata, int tid)
+{
+ if (tid < 16) sdata[tid] += sdata[tid + 16];
+ if (tid < 8) sdata[tid] += sdata[tid + 8];
+ if (tid < 4) sdata[tid] += sdata[tid + 4];
+ if (tid < 2) sdata[tid] += sdata[tid + 2];
+ if (tid < 1) sdata[tid] += sdata[tid + 1];
+}
+
+
+
+
+GLOBAL void k_Hydrogen_Bonds_HB(reax_atom *atoms,
+ single_body_parameters *sbp,
+ hbond_parameters *d_hbp,
+ control_params *control,
+ simulation_data *data,
+ static_storage p_workspace,
+ list p_bonds, list p_hbonds,
+ int N, int num_atom_types,
+ real *E_HB, rvec *aux_ext_press, rvec *atoms_f )
+{
+ extern __shared__ real t_hb[];
+ extern __shared__ rvec t__f[];
+ extern __shared__ rvec t_cdbo[];
+ extern __shared__ rvec t_hf [];
+
+ real *sh_hb = t_hb;
+ real *sh_cdbo = t_hb + blockDim.x;
+ rvec *sh_atomf = (rvec *)(sh_cdbo + blockDim.x);
+ rvec *sh_hf = (rvec *) (sh_atomf + blockDim.x);
+
+ int __THREADS_PER_ATOM__ = HBONDS_THREADS_PER_ATOM;
+
+ int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
+ int warp_id = thread_id / __THREADS_PER_ATOM__;
+ int lane_id = thread_id & (__THREADS_PER_ATOM__ -1);
+ int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
+
+ if (warp_id >= N ) return;
+
+
+ int i, j, k, pi, pk, itr, top;
+ int type_i, type_j, type_k;
+ int start_j, end_j, hb_start_j, hb_end_j;
+ int hblist[MAX_BONDS];
+ int num_hb_intrs = 0;
+ real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
+ real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
+ rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
+ rvec dvec_jk, force, ext_press;
+ ivec rel_jk;
+ // rtensor temp_rtensor, total_rtensor;
+ hbond_parameters *hbp;
+ bond_order_data *bo_ij;
+ bond_data *pbond_ij;
+ far_neighbor_data *nbr_jk;
+ list *bonds, *hbonds;
+ bond_data *bond_list;
+ hbond_data *hbond_list, *hbond_jk;
+ static_storage *workspace = &p_workspace;
+
+ /*
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+ if (j >= N) return;
+ */
+
+ // j = blockIdx.x;
+
+ j = warp_id;
+
+ bonds = &p_bonds;
+ bond_list = bonds->select.bond_list;
+
+ hbonds = &p_hbonds;
+ hbond_list = hbonds->select.hbond_list;
+
+ // loops below discover the Hydrogen bonds between i-j-k triplets.
+ // here j is H atom and there has to be some bond between i and j.
+ // Hydrogen bond is between j and k.
+ // so in this function i->X, j->H, k->Z when we map
+ // variables onto the ones in the handout.
+
+ //for( j = 0; j < system->N; ++j )
+ sh_hb [threadIdx.x] = 0;
+ rvec_MakeZero ( sh_atomf[ threadIdx.x] );
+
+ if( sbp[atoms[j].type].p_hbond==1) {// j must be H
+ //set j's variables
+ type_j = atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+ hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
+ hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
+
+ top = 0;
+ for( pi = start_j; pi < end_j; ++pi ) {
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = atoms[i].type;
+
+ if( sbp[type_i].p_hbond == 2 &&
+ bo_ij->BO >= HB_THRESHOLD ) {
+ hblist[top++] = pi;
+ }
+ }
+
+ // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
+ // j, top, hb_start_j, hb_end_j );
+
+ for( itr=0; itr < top; ++itr ) {
+ pi = hblist[itr];
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+
+ //TODO
+ rvec_MakeZero (sh_hf [threadIdx.x]);
+ sh_cdbo [threadIdx.x] = 0;
+
+
+ //for( pk = hb_start_j; pk < hb_end_j; ++pk )
+ int loopcount = (hb_end_j - hb_start_j) / HBONDS_THREADS_PER_ATOM + (((hb_end_j - hb_start_j)%HBONDS_THREADS_PER_ATOM == 0) ? 0 : 1);
+ int count = 0;
+ //jpk = hb_start_j + threadIdx.x;
+ pk = hb_start_j + lane_id;
+ //while (pk < hb_end_j)
+ while (count < loopcount)
+ {
+
+ if (pk < hb_end_j)
+ {
+ // set k's varibles
+ //TODO
+ hbond_jk = &( hbond_list[pk] );
+ //TODO
+ k = hbond_list[pk].nbr;
+ type_k = atoms[k].type;
+ nbr_jk = hbond_list[pk].ptr;
+ r_jk = nbr_jk->d;
+ rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
+ }
+ else k = -1;
+
+ //TODO Double check this Hydrogen Bonds fix
+ //rvec_MakeZero ( nbr_jk->h_f );
+ //rvec_MakeZero ( hbond_jk->h_f );
+ //TODO Double check this Hydrogen Bonds fix
+
+ //sh_hb [threadIdx.x] = 0;
+ //rvec_MakeZero ( sh_atomf[ threadIdx.x] );
+ //__syncthreads ();
+
+
+ if(( i != k ) && (k != -1)) {
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = atoms[i].type;
+ r_ij = pbond_ij->d;
+ hbp = &(d_hbp[ index_hbp(type_i, type_j, type_k, num_atom_types) ]);
+ ++num_hb_intrs;
+
+ d_Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &theta, &cos_theta );
+ // the derivative of cos(theta)
+ d_Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &dcos_theta_di, &dcos_theta_dj,
+ &dcos_theta_dk );
+
+ // hydrogen bond energy
+ sin_theta2 = SIN( theta/2.0 );
+ sin_xhz4 = SQR(sin_theta2);
+ sin_xhz4 *= sin_xhz4;
+ cos_xhz1 = ( 1.0 - cos_theta );
+ exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
+ exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
+ r_jk / hbp->r0_hb - 2.0 ) );
+
+ //PERFORMANCE IMPACT
+ e_hb = hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
+ //MYATOMICADD( &data->E_HB, e_hb );
+ //E_HB [j] += e_hb;
+ sh_hb [threadIdx.x] += e_hb;
+
+ CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
+ CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
+ CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
+ 1.0 / hbp->r0_hb);
+
+ //this is the problem here
+ //TODO
+ // hydrogen bond forces
+ //bo_ij->Cdbo += CEhb1; // dbo term
+ sh_cdbo[threadIdx.x] += CEhb1;
+ //TODO
+ //warpReduce (sh_cdbo, threadIdx.x);
+ //if (threadIdx.x == 0)
+ // bo_ij->Cdbo += sh_cdbo [0];
+
+
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
+
+ //PERFORMANCE IMPACT
+ /*
+ atomic_rvecScaledAdd( atoms[i].f,
+ +CEhb2, dcos_theta_di ); //dcos terms
+ atomic_rvecScaledAdd( atoms[j].f,
+ +CEhb2, dcos_theta_dj );
+ atomic_rvecScaledAdd( atoms[k].f,
+ +CEhb2, dcos_theta_dk );
+ //dr terms
+ atomic_rvecScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+ atomic_rvecScaledAdd( atoms[k].f, +CEhb3/r_jk, dvec_jk );
+ */
+
+ //PERFORMANCE IMPACT
+ //rvec_ScaledAdd( pbond_ij->h_f, +CEhb2, dcos_theta_di ); //dcos terms
+ rvec_ScaledAdd( sh_hf [threadIdx.x], +CEhb2, dcos_theta_di ); //dcos terms
+
+ //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
+ rvec_ScaledAdd( sh_atomf [threadIdx.x], +CEhb2, dcos_theta_dj );
+
+
+ //TODO you forgot here
+ //TODO Hydrogen bonds fix. -- BE VERY CAREFUL *****
+ rvec_ScaledAdd( hbond_jk->h_f, +CEhb2, dcos_theta_dk );
+
+ //rvec_ScaledAdd( nbr_jk->h_f,
+ // +CEhb2, dcos_theta_dk );
+
+ //dr terms
+ //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+ rvec_ScaledAdd( sh_atomf [threadIdx.x], -CEhb3/r_jk, dvec_jk );
+
+ //TODO you forgot
+ rvec_ScaledAdd( hbond_jk->h_f, +CEhb3/r_jk, dvec_jk );
+ //rvec_ScaledAdd( nbr_jk->h_f, +CEhb3/r_jk, dvec_jk );
+ }
+ else
+ {
+ // for pressure coupling, terms that are not related
+ // to bond order derivatives are added directly into
+ // pressure vector/tensor
+ //rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
+ //rvec_Add( pbond_ij->h_f, force );
+ //rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd (sh_press [threadIdx.x], 1.0, ext_press );
+
+ //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
+
+ //ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
+ //rvec_Scale( force, +CEhb2, dcos_theta_dk );
+
+ //rvec_Add( nbr_jk->h_f, force );
+ //rvec_Add( hbond_jk->h_f, force );
+
+ //rvec_iMultiply( ext_press, rel_jk, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
+
+ //dr terms
+ //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
+
+ //rvec_Scale( force, CEhb3/r_jk, dvec_jk );
+ //rvec_Add( hbond_jk->h_f, force );
+ //rvec_iMultiply( ext_press, rel_jk, force );
+ //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
+ //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
+
+ }
+
+ } // i != k if statement
+
+ pk += __THREADS_PER_ATOM__;
+ count ++;
+
+ } // pk for statement
+
+ //__syncthreads ();
+
+ //at this point done with one bond....
+ //do the reduction now
+ //if (threadIdx.x == 0){
+ if (lane_id < 16) {
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 16];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 16]);
+ }
+ if (lane_id < 8) {
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 8];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 8]);
+ }
+ if (lane_id < 4) {
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 4];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 4]);
+ }
+ if (lane_id < 2) {
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 2];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 2]);
+ }
+ if (lane_id < 1) {
+ sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 1];
+ rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 1]);
+
+ bo_ij->Cdbo += sh_cdbo [threadIdx.x];
+ rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
+ }
+ /*
+ if (lane_id == 0){
+ for (i = 1; i < 32; i++)
+ {
+ //sh_cdbo [threadIdx.x] += sh_cdbo [i];
+ //rvec_Add (sh_hf [threadIdx.x], sh_hf [i]);
+
+ sh_cdbo [lane_id] += sh_cdbo [lane_id + i];
+ rvec_Add (sh_hf [lane_id], sh_hf [lane_id + i]);
+ }
+
+ //bo_ij->Cdbo += sh_cdbo [threadIdx.x];
+ //rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
+
+ bo_ij->Cdbo += sh_cdbo [lane_id];
+ rvec_Add (pbond_ij->h_f, sh_hf [lane_id]);
+ }
+ */
+
+ } //itr for statement
+
+ //__syncthreads ();
+ } // main if statment
+
+ //__syncthreads ();
+
+ //do the reduction for the bond_ij here
+ if (lane_id < 16){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
+ }
+ if (lane_id < 8){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
+ }
+ if (lane_id < 4){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
+ }
+ if (lane_id < 2){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
+ }
+ if (lane_id < 1){
+ sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
+ rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
+
+ E_HB [j] += sh_hb [threadIdx.x];
+ rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+ }
+ /*
+ if (lane == 0){
+ //E_HB [j] += sh_hb [threadIdx.x];
+ rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+ rvec_Copy (atoms_f [j], sh_atomf [threadIdx.x]);
+ }
+ */
+ //if (threadIdx.x == 0){
+ /*
+ if (lane_id == 0){
+ for (i = 1; i < 32; i++)
+ {
+ //sh_hb [threadIdx.x] += sh_hb [i];
+ //rvec_Add (sh_atomf [threadIdx.x], sh_atomf [i]);
+ sh_hb [lane_id] += sh_hb [lane_id + i];
+ rvec_Add (sh_atomf [lane_id], sh_atomf [lane_id + i]);
+ }
+
+ //E_HB [j] += sh_hb [threadIdx.x];
+ //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+
+ E_HB [j] += sh_hb [lane_id];
+ rvec_Add (atoms[j].f, sh_atomf [lane_id]);
+ //rvec_Copy (atoms_f[j], sh_atomf [threadIdx.x]);
+ }
+ */
+
+ //E_HB [j] += sh_hb [threadIdx.x];
+ //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
+}
+
+
+GLOBAL void k_Hydrogen_Bonds_Postprocess(reax_atom *atoms,
+ single_body_parameters *sbp,
+ static_storage p_workspace,
+ list p_bonds, list p_hbonds, list p_far_nbrs, int N,
+ real *e_hb)
+{
+
+ int i, pj, hj, nbr, k, j;
+ int start, end;
+
+ bond_data *pbond;
+ bond_data *sym_index_bond;
+ far_neighbor_data *nbr_pj, *sym_index_nbr;
+
+ list *bonds = &p_bonds;
+ list *far_nbrs = &p_far_nbrs;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( i >= N) return;
+
+ // For processing ij information
+ start = Start_Index(i, bonds);
+ end = End_Index(i, bonds);
+
+ //rvec_Scale (atoms[i].f, e_hb[i], atoms[i].f);
+
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
+
+ pbond = &(bonds->select.bond_list[pj]);
+ sym_index_bond = &( bonds->select.bond_list[ pbond->sym_index ] );
+
+ rvec_Add (atoms[i].f, sym_index_bond->h_f );
+ }
+
+ /*
+ for (pj = Start_Index (i, far_nbrs); pj < End_Index (i, far_nbrs); pj ++)
+ {
+ // check if the neighbor is of h_type
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+
+ sym_index_nbr = & (far_nbrs->select.far_nbr_list[ nbr_pj->sym_index ]);
+ rvec_Add (atoms[i].f, sym_index_nbr->h_f );
+ }
+ */
+
+ // if (workspace->hbond_index [j] != -1)
+ // {
+ // hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
+ // hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
+
+ // for ( hj = hb_start_j; hj < hb_end_j; hj ++ )
+ // {
+ // h_bond_data = &( hbonds->select.hbond_list [hj] );
+ // nbr = h_bond_data->nbr;
+
+ // if (nbr == i) {
+ // rvec_Add (atoms[i].f, h_bond_data->h_f );
+ // }
+ // }
+ // }
+}
+
+
+GLOBAL void k_Hydrogen_Bonds_Far_Nbrs(reax_atom *atoms,
+ single_body_parameters *sbp,
+ static_storage p_workspace,
+ list p_bonds, list p_hbonds, list p_far_nbrs, int N )
+{
+
+ extern __shared__ rvec __f[];
+ int i, pj,j;
+ int start, end;
+
+ far_neighbor_data *nbr_pj, *sym_index_nbr;
+ list *far_nbrs = &p_far_nbrs;
+
+ i = blockIdx.x;
+
+ start = Start_Index (i, far_nbrs);
+ end = End_Index (i, far_nbrs);
+ pj = start + threadIdx.x;
+
+ rvec_MakeZero (__f[threadIdx.x]);
+
+ while (pj < end)
+ {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+
+ //sym_index_nbr = & (far_nbrs->select.far_nbr_list[ nbr_pj->sym_index ]);
+ //
+ //rvec_Add (atoms[i].f, sym_index_nbr->h_f );
+ //
+ //rvec_Add (__f[threadIdx.x], sym_index_nbr->h_f );
+
+ pj += blockDim.x;
+ }
+
+ if (threadIdx.x < 16) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 16]);
+ if (threadIdx.x < 8) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 8]);
+ if (threadIdx.x < 4) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 4]);
+ if (threadIdx.x < 2) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 2]);
+ if (threadIdx.x < 1) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 1]);
+
+ if (threadIdx.x == 0)
+ rvec_Add (atoms[i].f, __f[0]);
+}
+
+
+GLOBAL void k_Hydrogen_Bonds_HNbrs(reax_atom *atoms,
+ single_body_parameters *sbp,
+ static_storage p_workspace,
+ list p_bonds, list p_hbonds, list p_far_nbrs, int N )
+{
+
+ extern __shared__ rvec __f[];
+ int i, pj,j;
+ int start, end;
+
+ hbond_data *nbr_pj, *sym_index_nbr;
+ list *hbonds = &p_hbonds;
+
+ i = blockIdx.x;
+
+ start = Start_Index (i, hbonds);
+ end = End_Index (i, hbonds);
+ pj = start + threadIdx.x;
+
+ rvec_MakeZero (__f[threadIdx.x]);
+
+ while (pj < end)
+ {
+ nbr_pj = &( hbonds->select.hbond_list[pj] );
+ j = nbr_pj->nbr;
+
+ sym_index_nbr = & (hbonds->select.hbond_list[ nbr_pj->sym_index ]);
+ rvec_Add (__f[threadIdx.x], sym_index_nbr->h_f );
+
+ pj += blockDim.x;
+ }
+
+ if (threadIdx.x < 16) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 16]);
+ if (threadIdx.x < 8) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 8]);
+ if (threadIdx.x < 4) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 4]);
+ if (threadIdx.x < 2) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 2]);
+ if (threadIdx.x < 1) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 1]);
+
+ if (threadIdx.x == 0)
+ rvec_Add (atoms[i].f, __f[0]);
+}
diff --git a/PuReMD-GPU/src/cuda_three_body_interactions.h b/PuReMD-GPU/src/cuda_three_body_interactions.h
new file mode 100644
index 0000000000000000000000000000000000000000..4a87fcfe42852c50ac58a1ef52e8353cf24971a2
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_three_body_interactions.h
@@ -0,0 +1,71 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_THREE_BODY_INTERACTIONS_H_
+#define __CUDA_THREE_BODY_INTERACTIONS_H_
+
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+DEVICE void d_Calculate_Theta( rvec, real, rvec, real, real*, real* );
+
+DEVICE void d_Calculate_dCos_Theta( rvec, real, rvec, real, rvec*, rvec*, rvec* );
+
+GLOBAL void k_Three_Body_Interactions( reax_atom *, single_body_parameters *, three_body_header *,
+ global_parameters , control_params *, simulation_data *, static_storage ,
+ list , list , int , int , real *, real *, real *, rvec *);
+
+GLOBAL void k_Three_Body_Interactions_results( reax_atom *,
+ control_params *, static_storage , list , int );
+
+GLOBAL void k_Three_Body_Estimate( reax_atom *atoms,
+ control_params *control, list p_bonds, int N, int *count);
+
+GLOBAL void k_Hydrogen_Bonds( reax_atom *,
+ single_body_parameters *, hbond_parameters *,
+ control_params *, simulation_data *, static_storage ,
+ list , list , int , int, real *, rvec *, rvec *);
+
+GLOBAL void k_Hydrogen_Bonds_HB( reax_atom *,
+ single_body_parameters *, hbond_parameters *,
+ control_params *, simulation_data *, static_storage ,
+ list , list , int , int, real *, rvec *, rvec *);
+
+GLOBAL void k_Hydrogen_Bonds_Postprocess( reax_atom *,
+ single_body_parameters *,
+ static_storage , list,
+ list , list , int, real * );
+
+GLOBAL void k_Hydrogen_Bonds_Far_Nbrs( reax_atom *,
+ single_body_parameters *, static_storage , list, list , list , int );
+
+GLOBAL void k_Hydrogen_Bonds_HNbrs( reax_atom *, single_body_parameters *,
+ static_storage , list, list , list , int );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PuReMD-GPU/src/cuda_two_body_interactions.cu b/PuReMD-GPU/src/cuda_two_body_interactions.cu
new file mode 100644
index 0000000000000000000000000000000000000000..d5f6abd64b16cc4582a404566560768eca49b866
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_two_body_interactions.cu
@@ -0,0 +1,1047 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "cuda_two_body_interactions.h"
+
+#include "bond_orders.h"
+#include "index_utils.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+#include "index_utils.h"
+
+#include "cuda_helpers.h"
+
+
+GLOBAL void Cuda_Bond_Energy ( reax_atom *atoms, global_parameters g_params,
+ single_body_parameters *sbp, two_body_parameters *tbp,
+ simulation_data *data,
+ static_storage p_workspace, list p_bonds,
+ int N, int num_atom_types, real *E_BE)
+{
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ real ebond, pow_BOs_be2, exp_be12, CEbo;
+ real gp3, gp4, gp7, gp10, gp37;
+ real exphu, exphua1, exphub1, exphuov, hulpov, estriph;
+ real decobdbo, decobdboua, decobdboub;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ bond_order_data *bo_ij;
+ list *bonds;
+ static_storage *workspace;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+ if ( i >= N ) return;
+
+ bonds = &p_bonds;
+ workspace = &p_workspace;
+
+ gp3 = g_params.l[3];
+ gp4 = g_params.l[4];
+ gp7 = g_params.l[7];
+ gp10 = g_params.l[10];
+ gp37 = (int) g_params.l[37];
+
+ //for( i=0; i < system->N; ++i )
+ start_i = Start_Index(i, bonds);
+ end_i = End_Index(i, bonds);
+ //fprintf( stderr, "i=%d start=%d end=%d\n", i, start_i, end_i );
+ for( pj = start_i; pj < end_i; ++pj )
+ {
+ //TODO
+ //if( i < bonds->select.bond_list[pj].nbr )
+ if( i < bonds->select.bond_list[pj].nbr )
+ {
+ //TODO
+ /* set the pointers */
+ j = bonds->select.bond_list[pj].nbr;
+ type_i = atoms[i].type;
+ type_j = atoms[j].type;
+ sbp_i = &( sbp[type_i] );
+ sbp_j = &( sbp[type_j] );
+ twbp = &( tbp[ index_tbp(type_i,type_j,num_atom_types) ] );
+ bo_ij = &( bonds->select.bond_list[pj].bo_data );
+
+ /* calculate the constants */
+ pow_BOs_be2 = POW( bo_ij->BO_s, twbp->p_be2 );
+ exp_be12 = EXP( twbp->p_be1 * ( 1.0 - pow_BOs_be2 ) );
+ CEbo = -twbp->De_s * exp_be12 *
+ ( 1.0 - twbp->p_be1 * twbp->p_be2 * pow_BOs_be2 );
+
+ /* calculate the Bond Energy */
+ ebond =
+ -twbp->De_s * bo_ij->BO_s * exp_be12
+ -twbp->De_p * bo_ij->BO_pi
+ -twbp->De_pp * bo_ij->BO_pi2;
+
+ //PERFORMANCE IMAPCT
+ //MYATOMICADD(&data->E_BE, ebond);
+ //TODO
+ //E_BE [ i ] += ebond/2.0;
+ E_BE [ i ] += ebond;
+ //data->E_BE += ebond;
+
+ /* calculate derivatives of Bond Orders */
+ bo_ij->Cdbo += CEbo;
+ bo_ij->Cdbopi -= (CEbo + twbp->De_p);
+ bo_ij->Cdbopi2 -= (CEbo + twbp->De_pp);
+
+#ifdef TEST_ENERGY
+ //TODO
+ //fprintf( out_control->ebond, "%6d%6d%24.15e%24.15e\n",
+ // workspace->orig_id[i], workspace->orig_id[j],
+ // i+1, j+1,
+ // bo_ij->BO, ebond/*, data->E_BE*/ );
+ /*
+ fprintf( out_control->ebond, "%6d%6d%12.6f%12.6f%12.6f\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ CEbo, -twbp->De_p, -twbp->De_pp );*/
+#endif
+#ifdef TEST_FORCES
+ //TODO
+ /*
+ Add_dBO( system, lists, i, pj, CEbo, workspace->f_be );
+ Add_dBOpinpi2( system, lists, i, pj,
+ -(CEbo + twbp->De_p), -(CEbo + twbp->De_pp),
+ workspace->f_be, workspace->f_be );
+ */
+ //TODO
+#endif
+
+ /* Stabilisation terminal triple bond */
+ if( bo_ij->BO >= 1.00 ) {
+ if( gp37 == 2 ||
+ (sbp_i->mass == 12.0000 && sbp_j->mass == 15.9990) ||
+ (sbp_j->mass == 12.0000 && sbp_i->mass == 15.9990) ) {
+ // ba = SQR(bo_ij->BO - 2.50);
+ exphu = EXP( -gp7 * SQR(bo_ij->BO - 2.50) );
+ //oboa=abo(j1)-boa;
+ //obob=abo(j2)-boa;
+ exphua1 = EXP(-gp3*(workspace->total_bond_order[i]-bo_ij->BO));
+ exphub1 = EXP(-gp3*(workspace->total_bond_order[j]-bo_ij->BO));
+ //ovoab=abo(j1)-aval(it1)+abo(j2)-aval(it2);
+ exphuov = EXP(gp4*(workspace->Delta[i] + workspace->Delta[j]));
+ hulpov = 1.0 / (1.0 + 25.0 * exphuov);
+
+ estriph = gp10 * exphu * hulpov * (exphua1 + exphub1);
+ //estrain(j1) = estrain(j1) + 0.50*estriph;
+ //estrain(j2) = estrain(j2) + 0.50*estriph;
+
+ //PERFORMANCE IMPACT
+ //MYATOMICADD(&data->E_BE, estriph);
+ E_BE [ i] += estriph;
+ //data->E_BE += estriph;
+
+ decobdbo = gp10 * exphu * hulpov * (exphua1 + exphub1) *
+ ( gp3 - 2.0 * gp7 * (bo_ij->BO-2.50) );
+ decobdboua = -gp10 * exphu * hulpov *
+ (gp3*exphua1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
+ decobdboub = -gp10 * exphu * hulpov *
+ (gp3*exphub1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
+
+ bo_ij->Cdbo += decobdbo;
+
+ //PERFORMANCE IMAPCT
+ workspace->CdDelta[i] += decobdboua;
+ //MYATOMICADD(&workspace->CdDelta[j], decobdboub);
+ //CdDelta [ i * N + i ] += decobdboua;
+ //CdDelta [ i * N + j ] += decobdboua;
+ //workspace->CdDelta [i] += decobdboua;
+ //workspace->CdDelta [j] += decobdboub;
+
+#ifdef TEST_ENERGY
+ /*
+ fprintf( out_control->ebond,
+ "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ //i+1, j+1,
+ estriph, decobdbo, decobdboua, decobdboub );
+ */
+#endif
+#ifdef TEST_FORCES
+ /*
+ Add_dBO( system, lists, i, pj, decobdbo, workspace->f_be );
+ Add_dDelta( system, lists, i, decobdboua, workspace->f_be );
+ Add_dDelta( system, lists, j, decobdboub, workspace->f_be );
+ */
+#endif
+ }
+ }
+ }
+ } //TODO commented out the if statement for processing i < j.
+ // we process all teh bonds and add only half the energy
+}
+
+
+/*
+
+ GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *atoms,
+ two_body_parameters *tbp,
+ global_parameters g_p,
+ control_params *control,
+ simulation_data *data,
+ list p_far_nbrs,
+ real *E_vdW, real *E_Ele, rvec *aux_ext_press,
+ int num_atom_types, int N )
+ {
+ int i, j, pj;
+ int start_i, end_i;
+ real self_coef;
+ real p_vdW1, p_vdW1i;
+ real powr_vdW1, powgi_vdW1;
+ real tmp, r_ij, fn13, exp1, exp2;
+ real Tap, dTap, dfn13, CEvd, CEclmb;
+ real dr3gamij_1, dr3gamij_3;
+ real e_ele, e_vdW, e_core, de_core;
+ rvec temp, ext_press;
+// rtensor temp_rtensor, total_rtensor;
+two_body_parameters *twbp;
+far_neighbor_data *nbr_pj;
+list *far_nbrs = &p_far_nbrs;
+
+i = blockIdx.x * blockDim.x + threadIdx.x;
+if ( i >= N ) return;
+
+p_vdW1 = g_p.l[28];
+p_vdW1i = 1.0 / p_vdW1;
+e_ele = 0;
+e_vdW = 0;
+e_core = 0;
+de_core = 0;
+
+//for( i = 0; i < system->N; ++i ) {
+start_i = Start_Index(i, far_nbrs);
+end_i = End_Index(i, far_nbrs);
+// fprintf( stderr, "i: %d, start: %d, end: %d\n",
+// i, start_i, end_i );
+
+for( pj = start_i; pj < end_i; ++pj )
+if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
+nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+j = nbr_pj->nbr;
+r_ij = nbr_pj->d;
+twbp = &(tbp[ index_tbp(atoms[i].type, atoms[j].type, num_atom_types) ]);
+self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
+
+//CHANGE ORIGINAL
+//if (i <= j) continue;
+//CHANGE ORIGINAL
+
+// Calculate Taper and its derivative
+// Tap = nbr_pj->Tap; -- precomputed during compte_H
+Tap = control->Tap7 * r_ij + control->Tap6;
+Tap = Tap * r_ij + control->Tap5;
+Tap = Tap * r_ij + control->Tap4;
+Tap = Tap * r_ij + control->Tap3;
+Tap = Tap * r_ij + control->Tap2;
+Tap = Tap * r_ij + control->Tap1;
+Tap = Tap * r_ij + control->Tap0;
+
+dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
+dTap = dTap * r_ij + 5*control->Tap5;
+dTap = dTap * r_ij + 4*control->Tap4;
+dTap = dTap * r_ij + 3*control->Tap3;
+dTap = dTap * r_ij + 2*control->Tap2;
+dTap += control->Tap1/r_ij;
+
+//vdWaals Calculations
+if(g_p.vdw_type==1 || g_p.vdw_type==3) {
+ // shielding
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+ E_vdW [i] += e_vdW / 2.0;
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
+ POW(r_ij, p_vdW1 - 2.0);
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) * dfn13 );
+}
+else{ // no shielding
+ exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+
+ e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+ E_vdW [i] += e_vdW / 2.0;
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) );
+}
+
+if(g_p.vdw_type==2 || g_p.vdw_type==3) {
+ // innner wall
+ e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
+ e_vdW = self_coef * Tap * e_core;
+
+ //TODO check this
+ E_vdW [i] += e_vdW / 2.0;
+ //TODO check this
+
+ de_core = -(twbp->acore/twbp->rcore) * e_core;
+ CEvd += self_coef * ( dTap * e_core + Tap * de_core );
+}
+
+//Coulomb Calculations
+dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+tmp = Tap / dr3gamij_3;
+//tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
+e_ele =
+self_coef * C_ele * atoms[i].q * atoms[j].q * tmp;
+E_Ele [i] += e_ele / 2.0;
+
+CEclmb = self_coef * C_ele * atoms[i].q * atoms[j].q *
+( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
+//CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
+// ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;
+
+if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ if (i >= j)
+ rvec_ScaledAdd( atoms[i].f, -(CEvd+CEclmb), nbr_pj->dvec );
+ else
+ rvec_ScaledAdd( atoms[i].f, +(CEvd+CEclmb), nbr_pj->dvec );
+}
+else { // NPT, iNPT or sNPT
+ // for pressure coupling, terms not related to bond order
+ // derivatives are added directly into pressure vector/tensor
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+
+ if ( i >= j)
+ rvec_ScaledAdd( atoms[i].f, -1., temp );
+ else
+ rvec_Add( atoms[i].f, temp );
+
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+
+ //rvec_Add( data->ext_press, ext_press );
+ rvec_Copy (aux_ext_press[i], ext_press);
+
+ //TODO CHECK THIS calculation here, it should be divided by two somehow.
+}
+}
+//}
+}
+
+*/
+
+
+GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *atoms,
+ two_body_parameters *tbp,
+ global_parameters g_p,
+ control_params *control,
+ simulation_data *data,
+ list p_far_nbrs,
+ real *E_vdW, real *E_Ele, rvec *aux_ext_press,
+ int num_atom_types, int N )
+{
+ extern __shared__ real _vdw[];
+ extern __shared__ real _ele[];
+ extern __shared__ rvec _force [];
+
+ real *sh_vdw;
+ real *sh_ele;
+ rvec *sh_force;
+
+ int i, j, pj;
+ int start_i, end_i;
+ real self_coef;
+ real p_vdW1, p_vdW1i;
+ real powr_vdW1, powgi_vdW1;
+ real tmp, r_ij, fn13, exp1, exp2;
+ real Tap, dTap, dfn13, CEvd, CEclmb;
+ real dr3gamij_1, dr3gamij_3;
+ real e_ele, e_vdW, e_core, de_core;
+ rvec temp, ext_press;
+ // rtensor temp_rtensor, total_rtensor;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ list *far_nbrs = &p_far_nbrs;
+
+ int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
+ int warpid = thread_id / VDW_THREADS_PER_ATOM;
+ int laneid = thread_id & (VDW_THREADS_PER_ATOM -1);
+
+ i = warpid;
+
+ sh_vdw = _vdw;
+ sh_ele = _vdw + blockDim.x;
+ sh_force = (rvec *)( _vdw + 2*blockDim.x);
+
+ sh_vdw[threadIdx.x] = 0.0;
+ sh_ele[threadIdx.x] = 0.0;
+ rvec_MakeZero ( sh_force [threadIdx.x] );
+
+ if (i < N)
+ {
+
+ p_vdW1 = g_p.l[28];
+ p_vdW1i = 1.0 / p_vdW1;
+ e_ele = 0;
+ e_vdW = 0;
+ e_core = 0;
+ de_core = 0;
+
+ //for( i = 0; i < system->N; ++i ) {
+ start_i = Start_Index(i, far_nbrs);
+ end_i = End_Index(i, far_nbrs);
+ // fprintf( stderr, "i: %d, start: %d, end: %d\n",
+ // i, start_i, end_i );
+
+ pj = start_i + laneid;
+ //for( pj = start_i; pj < end_i; ++pj )
+ while (pj < end_i)
+ {
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ r_ij = nbr_pj->d;
+ twbp = &(tbp[ index_tbp(atoms[i].type, atoms[j].type, num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
+
+ //CHANGE ORIGINAL
+ //if (i <= j) continue;
+ //CHANGE ORIGINAL
+
+ // Calculate Taper and its derivative
+ // Tap = nbr_pj->Tap; -- precomputed during compte_H
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
+ dTap = dTap * r_ij + 5*control->Tap5;
+ dTap = dTap * r_ij + 4*control->Tap4;
+ dTap = dTap * r_ij + 3*control->Tap3;
+ dTap = dTap * r_ij + 2*control->Tap2;
+ dTap += control->Tap1/r_ij;
+
+ //vdWaals Calculations
+ if(g_p.vdw_type==1 || g_p.vdw_type==3) {
+ // shielding
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+
+ //E_vdW [i] += e_vdW / 2.0;
+ sh_vdw [threadIdx.x] += e_vdW/2.0;
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
+ POW(r_ij, p_vdW1 - 2.0);
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) * dfn13 );
+ }
+ else{ // no shielding
+ exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+
+ e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+
+ //E_vdW [i] += e_vdW / 2.0;
+ sh_vdw [threadIdx.x] += e_vdW/2.0;
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) );
+ }
+
+ if(g_p.vdw_type==2 || g_p.vdw_type==3) {
+ // innner wall
+ e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
+ e_vdW = self_coef * Tap * e_core;
+
+ //TODO check this
+ //E_vdW [i] += e_vdW / 2.0;
+ sh_vdw [threadIdx.x] += e_vdW / 2.0;
+ //TODO check this
+
+ de_core = -(twbp->acore/twbp->rcore) * e_core;
+ CEvd += self_coef * ( dTap * e_core + Tap * de_core );
+ }
+
+ //Coulomb Calculations
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ tmp = Tap / dr3gamij_3;
+ //tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
+ e_ele =
+ self_coef * C_ele * atoms[i].q * atoms[j].q * tmp;
+
+ //E_Ele [i] += e_ele / 2.0;
+ sh_ele [threadIdx.x] += e_ele / 2.0;
+
+ CEclmb = self_coef * C_ele * atoms[i].q * atoms[j].q *
+ ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
+ //CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
+ // ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ if (i >= j){
+ //rvec_ScaledAdd( atoms[i].f, -(CEvd+CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( sh_force[threadIdx.x], -(CEvd+CEclmb), nbr_pj->dvec );
+ }
+ else
+ {
+ //rvec_ScaledAdd( atoms[i].f, +(CEvd+CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( sh_force[threadIdx.x], +(CEvd+CEclmb), nbr_pj->dvec );
+ }
+ }
+ else { // NPT, iNPT or sNPT
+ // for pressure coupling, terms not related to bond order
+ // derivatives are added directly into pressure vector/tensor
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+
+ if ( i >= j)
+ {
+ //rvec_ScaledAdd( atoms[i].f, -1., temp );
+ rvec_ScaledAdd( sh_force[threadIdx.x], -1., temp );
+ }
+ else
+ {
+ //rvec_Add( atoms[i].f, temp );
+ rvec_Add( sh_force[threadIdx.x], temp );
+ }
+
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+
+ //rvec_Add( data->ext_press, ext_press );
+ rvec_Copy (aux_ext_press[i], ext_press);
+
+ //TODO CHECK THIS calculation here, it should be divided by two somehow.
+ }
+ } // if condition for far neighbors
+
+
+ pj += VDW_THREADS_PER_ATOM;
+
+ } // end of while loop for pj < end_i condition
+ } // if (i < N ) condition
+ //}
+
+ __syncthreads ();
+
+ if (laneid < 16) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
+ }
+ __syncthreads ();
+ if (laneid < 8) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
+ }
+ __syncthreads ();
+ if (laneid < 4) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
+ }
+ __syncthreads ();
+ if (laneid < 2) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
+ }
+ __syncthreads ();
+ if (laneid < 1) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
+ }
+ __syncthreads ();
+ if (laneid == 0) {
+ E_vdW [i] += sh_vdw[threadIdx.x];
+ E_Ele [i] += sh_ele[threadIdx.x];
+ rvec_Add (atoms[i].f, sh_force [ threadIdx.x ]);
+ }
+}
+
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy(reax_atom *atoms,
+ control_params *control,
+ simulation_data *data,
+ list p_far_nbrs,
+ real *E_vdW, real *E_Ele, rvec *aux_ext_press,
+ LR_lookup_table *d_LR,
+ int num_atom_types,
+ int energy_update_freq,
+ int N )
+{
+
+ extern __shared__ real _vdw[];
+ extern __shared__ real _ele[];
+ extern __shared__ rvec _force [];
+
+ real *sh_vdw;
+ real *sh_ele;
+ rvec *sh_force;
+
+ int i, j, pj, r, steps, update_freq, update_energies;
+ int type_i, type_j, tmin, tmax;
+ int start_i, end_i;
+ real r_ij, self_coef, base, dif;
+ real e_vdW, e_ele;
+ real CEvd, CEclmb;
+ rvec temp, ext_press;
+ far_neighbor_data *nbr_pj;
+ LR_lookup_table *t;
+ list *far_nbrs = &p_far_nbrs;
+
+ int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
+ int warpid = thread_id / VDW_THREADS_PER_ATOM;
+ int laneid = thread_id & (VDW_THREADS_PER_ATOM -1);
+
+ i = warpid;
+
+ sh_vdw = _vdw;
+ sh_ele = _vdw + blockDim.x;
+ sh_force = (rvec *)( _vdw + 2*blockDim.x);
+
+ sh_vdw[threadIdx.x] = 0.0;
+ sh_ele[threadIdx.x] = 0.0;
+ rvec_MakeZero ( sh_force [threadIdx.x] );
+
+ if ( i < N )
+ {
+
+ reax_atom local_atom ;
+ local_atom.q = atoms[i].q;
+ //local_atom.q = d_far_data.q[i];
+ local_atom.type = atoms[i].type;
+ //local_atom.type = d_far_data.type[i];
+
+ /*
+ sh_vdw = _vdw;
+ sh_ele = _vdw + warpid;
+ sh_force = (rvec *)( _vdw + 2*warpid);
+
+ sh_vdw[threadIdx.x] = 0.0;
+ sh_ele[threadIdx.x] = 0.0;
+ rvec_MakeZero ( sh_force [threadIdx.x] );
+ */
+
+
+ steps = data->step - data->prev_steps;
+ update_freq = energy_update_freq;
+ update_energies = update_freq > 0 && steps % update_freq == 0;
+
+ //for( i = 0; i < system->N; ++i ) {
+ type_i = local_atom.type;
+ start_i = Start_Index(i,far_nbrs);
+ end_i = End_Index(i,far_nbrs);
+
+ pj = start_i + laneid;
+
+ //for( pj = start_i; pj < end_i; ++pj )
+ while (pj < end_i)
+ {
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
+ //if( d_far_data.d[pj] <= control->r_cut )
+ {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ //j = d_far_data.nbrs[pj];
+ type_j = atoms[j].type;
+ //type_j = d_far_data.type[j];
+ r_ij = nbr_pj->d;
+ //r_ij = d_far_data.d[pj];
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
+
+ //TODO
+ //CHANGE ORIGINAL
+ //if (i <= j) { pj += blockDim.x; continue; }
+ //CHANGE ORIGINAL
+
+ /* Cubic Spline Interpolation */
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+
+ if(( update_energies ))
+ {
+ e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
+ t->vdW[r].a;
+ e_vdW *= self_coef;
+
+ e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif + t->ele[r].a;
+ e_ele *= self_coef * local_atom.q * atoms[j].q;
+
+
+ //data->E_vdW += e_vdW;
+ //TODO
+ //E_vdW [i] += e_vdW / 2.0;
+ //E_vdW [i] = __dadd_rd (E_vdW [i], e_vdW/2.0);
+ sh_vdw [threadIdx.x] += e_vdW/2.0;
+ //E_vdW [i] += e_vdW;
+
+ //TODO
+ //data->E_Ele += e_ele;
+ //E_Ele [i] += e_ele / 2.0;
+ //E_Ele [i] = __dadd_rd ( E_Ele [i], e_ele / 2.0);
+ sh_ele [threadIdx.x] += e_ele/2.0;
+ //E_Ele [i] += e_ele;
+ }
+
+ CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
+ t->CEvd[r].a;
+ CEvd *= self_coef;
+
+ CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
+ t->CEclmb[r].a;
+ CEclmb *= self_coef * local_atom.q * atoms[j].q;
+ //CEclmb *= self_coef * local_atom.q * d_far_data.q[j];
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ if ( i >= j)
+ //rvec_ScaledAdd( atoms[i].f, -(CEvd + CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), nbr_pj->dvec );
+ //rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), d_far_data.dvec[pj] );
+ else
+ //rvec_ScaledAdd( atoms[i].f, +(CEvd + CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), nbr_pj->dvec );
+ //rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), d_far_data.dvec[pj] );
+ }
+ else { // NPT, iNPT or sNPT
+ // for pressure coupling, terms not related to bond order
+ // derivatives are added directly into pressure vector/tensor /
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+ if (i >= j)
+ rvec_ScaledAdd( atoms[i].f, -1., temp );
+ else
+ rvec_Add( atoms[i].f, temp );
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+
+ //rvec_Add( data->ext_press, ext_press );
+ rvec_Copy (aux_ext_press [i], ext_press );
+
+ //TODO CHECK THIS
+ }
+
+
+
+ }
+
+ pj += VDW_THREADS_PER_ATOM;
+ }
+
+ }// if i < n condition
+
+ __syncthreads ();
+
+ if (laneid < 16) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
+ }
+ __syncthreads ();
+ if (laneid < 8) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
+ }
+ __syncthreads ();
+ if (laneid < 4) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
+ }
+ __syncthreads ();
+ if (laneid < 2) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
+ }
+ __syncthreads ();
+ if (laneid < 1) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
+ }
+ __syncthreads ();
+ if (laneid == 0) {
+ E_vdW [i] += sh_vdw[threadIdx.x];
+ E_Ele [i] += sh_ele[threadIdx.x];
+ rvec_Add (atoms[i].f, sh_force [ threadIdx.x ]);
+ }
+}
+
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_1(reax_atom *atoms,
+ control_params *control,
+ simulation_data *data,
+ list p_far_nbrs,
+ real *E_vdW, real *E_Ele, rvec *aux_ext_press,
+ LR_lookup_table *d_LR,
+ int num_atom_types,
+ int energy_update_freq,
+ int N )
+{
+
+ extern __shared__ real _vdw[];
+ extern __shared__ real _ele[];
+
+ real *sh_vdw;
+ real *sh_ele;
+
+ int i, j, pj, r, steps, update_freq, update_energies;
+ int type_i, type_j, tmin, tmax;
+ int start_i, end_i;
+ real r_ij, self_coef, base, dif;
+ real e_vdW, e_ele;
+ real CEvd, CEclmb;
+ rvec temp, ext_press;
+ far_neighbor_data *nbr_pj;
+ LR_lookup_table *t;
+ list *far_nbrs = &p_far_nbrs;
+
+ i = blockIdx.x;
+
+ reax_atom local_atom;
+ local_atom.q = atoms[i].q;
+ local_atom.type = atoms[i].type;
+
+ sh_vdw = _vdw;
+ sh_ele = _vdw + blockDim.x;
+
+ sh_vdw[threadIdx.x] = 0.0;
+ sh_ele[threadIdx.x] = 0.0;
+
+
+ steps = data->step - data->prev_steps;
+ update_freq = energy_update_freq;
+ update_energies = update_freq > 0 && steps % update_freq == 0;
+
+ type_i = local_atom.type;
+ start_i = Start_Index(i,far_nbrs);
+ end_i = End_Index(i,far_nbrs);
+
+ pj = start_i + threadIdx.x;
+
+ while (pj < end_i)
+ {
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
+ {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ type_j = atoms[j].type;
+ r_ij = nbr_pj->d;
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
+
+ /* Cubic Spline Interpolation */
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+
+ if(( update_energies ))
+ {
+ e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
+ t->vdW[r].a;
+ e_vdW *= self_coef;
+
+ e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
+ t->ele[r].a;
+ e_ele *= self_coef * local_atom.q * atoms[j].q;
+
+ sh_vdw [threadIdx.x] += e_vdW/2.0;
+ sh_ele [threadIdx.x] += e_ele/2.0;
+ }
+ }
+
+ pj += blockDim.x;
+ }
+
+ // now do a reduce inside the warp for E_vdW, E_Ele and force.
+ if (threadIdx.x < 16) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
+ }
+ if (threadIdx.x < 8) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
+ }
+ if (threadIdx.x < 4) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
+ }
+ if (threadIdx.x < 2) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
+ }
+ if (threadIdx.x < 1) {
+ sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
+ sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
+ }
+ if (threadIdx.x == 0) {
+ E_vdW [i] += sh_vdw[0];
+ E_Ele [i] += sh_ele[0];
+ }
+}
+
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_2(reax_atom *atoms,
+ control_params *control,
+ simulation_data *data,
+ list p_far_nbrs,
+ real *E_vdW, real *E_Ele, rvec *aux_ext_press,
+ LR_lookup_table *d_LR,
+ int num_atom_types,
+ int energy_update_freq,
+ int N )
+{
+
+ extern __shared__ rvec _force [];
+
+ rvec *sh_force;
+
+ int i, j, pj, r, steps, update_freq, update_energies;
+ int type_i, type_j, tmin, tmax;
+ int start_i, end_i;
+ real r_ij, self_coef, base, dif;
+ real e_vdW, e_ele;
+ real CEvd, CEclmb;
+ rvec temp, ext_press;
+ far_neighbor_data *nbr_pj;
+ LR_lookup_table *t;
+ list *far_nbrs = &p_far_nbrs;
+
+ i = blockIdx.x;
+
+ reax_atom local_atom;
+ local_atom.q = atoms[i].q;
+ local_atom.type = atoms[i].type;
+
+ sh_force = _force;
+ rvec_MakeZero ( sh_force [threadIdx.x] );
+
+
+ steps = data->step - data->prev_steps;
+ update_freq = energy_update_freq;
+ update_energies = update_freq > 0 && steps % update_freq == 0;
+
+ //for( i = 0; i < system->N; ++i ) {
+ type_i = local_atom.type;
+ start_i = Start_Index(i,far_nbrs);
+ end_i = End_Index(i,far_nbrs);
+
+ pj = start_i + threadIdx.x;
+
+ while (pj < end_i)
+ {
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
+ {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ type_j = atoms[j].type;
+ r_ij = nbr_pj->d;
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
+
+ /* Cubic Spline Interpolation */
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+
+ CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
+ t->CEvd[r].a;
+ CEvd *= self_coef;
+
+ CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
+ t->CEclmb[r].a;
+ CEclmb *= self_coef * local_atom.q * atoms[j].q;
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
+ if ( i >= j)
+ rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), nbr_pj->dvec );
+ else
+ rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), nbr_pj->dvec );
+ }
+ else { // NPT, iNPT or sNPT
+ // for pressure coupling, terms not related to bond order
+ // derivatives are added directly into pressure vector/tensor /
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+ if (i >= j)
+ rvec_ScaledAdd( atoms[i].f, -1., temp );
+ else
+ rvec_Add( atoms[i].f, temp );
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+
+ rvec_Copy (aux_ext_press [i], ext_press );
+ }
+ }
+
+ pj += blockDim.x;
+ }
+
+ // now do a reduce inside the warp for E_vdW, E_Ele and force.
+ if (threadIdx.x < 16) {
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
+ }
+ if (threadIdx.x < 8) {
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
+ }
+ if (threadIdx.x < 4) {
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
+ }
+ if (threadIdx.x < 2) {
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
+ }
+ if (threadIdx.x < 1) {
+ rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
+ }
+ if (threadIdx.x == 0) {
+ rvec_Add (atoms[i].f, sh_force [ 0 ]);
+ }
+}
diff --git a/PuReMD-GPU/src/cuda_two_body_interactions.h b/PuReMD-GPU/src/cuda_two_body_interactions.h
new file mode 100644
index 0000000000000000000000000000000000000000..fe3e273775f67e17f88d742a52707f72bfbac56c
--- /dev/null
+++ b/PuReMD-GPU/src/cuda_two_body_interactions.h
@@ -0,0 +1,172 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#ifndef __CUDA_TWO_BODY_INTERACTIONS_H_
+#define __CUDA_TWO_BODY_INTERACTIONS_H_
+
+#include "mytypes.h"
+
+#include "index_utils.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+GLOBAL void Cuda_Bond_Energy( reax_atom *, global_parameters , single_body_parameters *, two_body_parameters *,
+ simulation_data *, static_storage , list , int , int, real * );
+
+GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *, two_body_parameters *,
+ global_parameters , control_params *, simulation_data *, list , real *, real *, rvec *,
+ int , int );
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy( reax_atom *, control_params *, simulation_data *,
+ list , real *, real *, rvec *,
+ LR_lookup_table *, int , int , int ) ;
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_1( reax_atom *, control_params *, simulation_data *,
+ list , real *, real *, rvec *,
+ LR_lookup_table *, int , int , int ) ;
+
+GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_2( reax_atom *, control_params *, simulation_data *,
+ list , real *, real *, rvec *,
+ LR_lookup_table *, int , int , int ) ;
+
+static DEVICE void d_LR_vdW_Coulomb( global_parameters g_params, two_body_parameters *tbp,
+ control_params *control, int i, int j, real r_ij, LR_data *lr, int num_atom_types )
+{
+ real p_vdW1 = g_params.l[28];
+ real p_vdW1i = 1.0 / p_vdW1;
+ real powr_vdW1, powgi_vdW1;
+ real tmp, fn13, exp1, exp2;
+ real Tap, dTap, dfn13;
+ real dr3gamij_1, dr3gamij_3;
+ real e_core, de_core;
+ two_body_parameters *twbp;
+
+ twbp = &(tbp[ index_tbp (i, j, num_atom_types) ]);
+ e_core = 0;
+ de_core = 0;
+
+ /* calculate taper and its derivative */
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dTap = 7 * control->Tap7 * r_ij + 6 * control->Tap6;
+ dTap = dTap * r_ij + 5 * control->Tap5;
+ dTap = dTap * r_ij + 4 * control->Tap4;
+ dTap = dTap * r_ij + 3 * control->Tap3;
+ dTap = dTap * r_ij + 2 * control->Tap2;
+ dTap += control->Tap1 / r_ij;
+
+
+ /* vdWaals calculations */
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+ /* fprintf(stderr,"vdW: Tap:%f, r: %f, f13:%f, D:%f, Energy:%f,\
+ Gamma_w:%f, p_vdw: %f, alpha: %f, r_vdw: %f, %lf %lf\n",
+ Tap, r_ij, fn13, twbp->D, Tap * twbp->D * (exp1 - 2.0 * exp2),
+ powgi_vdW1, p_vdW1, twbp->alpha, twbp->r_vdW, exp1, exp2); */
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) * POW(r_ij, p_vdW1 - 2.0);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
+
+ /*vdWaals Calculations*/
+ if (g_params.vdw_type == 1 || g_params.vdw_type == 3)
+ {
+ // shielding
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
+ POW(r_ij, p_vdW1 - 2.0);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
+ }
+ else // no shielding
+ {
+ exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2);
+ }
+
+ if (g_params.vdw_type == 2 || g_params.vdw_type == 3)
+ {
+ // innner wall
+ e_core = twbp->ecore * EXP(twbp->acore * (1.0 - (r_ij / twbp->rcore)));
+ lr->e_vdW += Tap * e_core;
+
+ de_core = -(twbp->acore / twbp->rcore) * e_core;
+ lr->CEvd += dTap * e_core + Tap * de_core;
+ }
+
+ /* Coulomb calculations */
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ tmp = Tap / dr3gamij_3;
+ lr->H = EV_to_KCALpMOL * tmp;
+ lr->e_ele = C_ele * tmp;
+ /* fprintf( stderr,"i:%d(%d), j:%d(%d), gamma:%f,\
+ Tap:%f, dr3gamij_3:%f, qi: %f, qj: %f\n",
+ i, system->atoms[i].type, j, system->atoms[j].type,
+ twbp->gamma, Tap, dr3gamij_3,
+ system->atoms[i].q, system->atoms[j].q ); */
+
+ lr->CEclmb = C_ele * ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
+ /* fprintf( stdout, "%d %d\t%g\t%g %g\t%g %g\t%g %g\n",
+ i+1, j+1, r_ij, e_vdW, CEvd * r_ij,
+ system->atoms[i].q, system->atoms[j].q, e_ele, CEclmb * r_ij ); */
+
+ /* fprintf( stderr,"LR_Lookup:%3d%3d%5.3f-%8.5f,%8.5f%8.5f,%8.5f%8.5f\n",
+ i, j, r_ij, lr->H, lr->e_vdW, lr->CEvd, lr->e_ele, lr->CEclmb ); */
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
+
diff --git a/PuReMD-GPU/src/cuda_utils.cu b/PuReMD-GPU/src/cuda_utils.cu
index 2c632c058e2419fbfc82310bb6f5354f863e2b39..c420db769053e46ed871973200d2de3b2cb7f871 100644
--- a/PuReMD-GPU/src/cuda_utils.cu
+++ b/PuReMD-GPU/src/cuda_utils.cu
@@ -18,120 +18,136 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
-
-
#include "cuda_utils.h"
-#include "mytypes.h"
-void cuda_malloc (void **ptr, int size, int memset, int err_code) {
+void cuda_malloc( void **ptr, int size, int memset, int err_code )
+{
cudaError_t retVal = cudaSuccess;
//fprintf (stderr, "&ptr --. %ld \n", &ptr);
//fprintf (stderr, "ptr --> %ld \n", ptr );
- retVal = cudaMalloc (ptr, size);
- if (retVal != cudaSuccess) {
- fprintf (stderr, "Failed to allocate memory on device for the res: %d... exiting with code: %d size: %d \n",
- err_code, retVal, size);
- exit (err_code);
+ retVal = cudaMalloc( ptr, size );
+ if ( retVal != cudaSuccess )
+ {
+ fprintf( stderr, "Failed to allocate memory on device for the res: %d... exiting with code: %d size: %d \n",
+ err_code, retVal, size );
+ exit( err_code );
}
//fprintf (stderr, "&ptr --. %ld \n", &ptr);
//fprintf (stderr, "ptr --> %ld \n", ptr );
- if (memset) {
- retVal = cudaMemset (*ptr, 0, size);
- if (retVal != cudaSuccess) {
- fprintf (stderr, "Failed to memset memory on device... exiting with code %d\n",
- err_code);
- exit (err_code);
+ if ( memset ) {
+ retVal = cudaMemset( *ptr, 0, size );
+ if ( retVal != cudaSuccess )
+ {
+ fprintf( stderr, "Failed to memset memory on device... exiting with code %d\n",
+ err_code );
+ exit( err_code );
}
}
}
-void cuda_free (void *ptr, int err_code) {
+void cuda_free( void *ptr, int err_code )
+{
cudaError_t retVal = cudaSuccess;
if (!ptr) return;
- retVal = cudaFree (ptr);
+ retVal = cudaFree( ptr );
- if (retVal != cudaSuccess) {
- fprintf (stderr, "Failed to release memory on device for res %d... exiting with code %d -- Address %ld\n",
- err_code, retVal, ptr);
+ if ( retVal != cudaSuccess )
+ {
+ fprintf( stderr, "Failed to release memory on device for res %d... exiting with code %d -- Address %ld\n",
+ err_code, retVal, ptr );
return;
}
}
-void cuda_memset (void *ptr, int data, size_t count, int err_code){
+
+
+void cuda_memset( void *ptr, int data, size_t count, int err_code )
+{
cudaError_t retVal = cudaSuccess;
- retVal = cudaMemset (ptr, data, count);
+ retVal = cudaMemset( ptr, data, count );
if (retVal != cudaSuccess) {
- fprintf (stderr, "ptr passed is %ld, value: %ld \n", ptr, &ptr);
- fprintf (stderr, " size to memset: %d \n", count);
- fprintf (stderr, " target data is : %d \n", data);
- fprintf (stderr, "Failed to memset memory on device... exiting with code %d, cuda code %d\n",
- err_code, retVal);
- exit (err_code);
+ fprintf( stderr, "ptr passed is %ld, value: %ld \n", ptr, &ptr );
+ fprintf( stderr, " size to memset: %d \n", count );
+ fprintf( stderr, " target data is : %d \n", data );
+ fprintf( stderr, "Failed to memset memory on device... exiting with code %d, cuda code %d\n",
+ err_code, retVal );
+ exit( err_code );
}
}
-void copy_host_device (void *host, void *dev, int size, enum cudaMemcpyKind dir, int resid)
+
+void copy_host_device( void *host, void *dev, int size, enum cudaMemcpyKind dir, int resid )
{
- cudaError_t retVal = cudaErrorNotReady;
+ cudaError_t retVal = cudaErrorNotReady;
- if (dir == cudaMemcpyHostToDevice)
- retVal = cudaMemcpy (dev, host, size, cudaMemcpyHostToDevice);
+ if ( dir == cudaMemcpyHostToDevice )
+ {
+ retVal = cudaMemcpy( dev, host, size, cudaMemcpyHostToDevice );
+ }
else
- retVal = cudaMemcpy (host, dev, size, cudaMemcpyDeviceToHost);
+ {
+ retVal = cudaMemcpy( host, dev, size, cudaMemcpyDeviceToHost );
+ }
- if (retVal != cudaSuccess) {
- fprintf (stderr, "could not copy resource %d from host to device: reason %d \n",
- resid, retVal);
- exit (resid);
+ if ( retVal != cudaSuccess ) {
+ fprintf( stderr, "could not copy resource %d from host to device: reason %d \n",
+ resid, retVal );
+ exit( resid );
}
}
-void copy_device (void *dest, void *src, int size, int resid)
+
+void copy_device( void *dest, void *src, int size, int resid )
{
- cudaError_t retVal = cudaErrorNotReady;
+ cudaError_t retVal = cudaErrorNotReady;
- retVal = cudaMemcpy (dest, src, size, cudaMemcpyDeviceToDevice);
- if (retVal != cudaSuccess) {
- fprintf (stderr, "could not copy resource %d from host to device: reason %d \n",
- resid, retVal);
- exit (resid);
+ retVal = cudaMemcpy( dest, src, size, cudaMemcpyDeviceToDevice );
+ if ( retVal != cudaSuccess )
+ {
+ fprintf( stderr, "could not copy resource %d from host to device: reason %d \n",
+ resid, retVal );
+ exit( resid );
}
}
-void compute_blocks ( int *blocks, int *block_size, int count )
+
+void compute_blocks( int *blocks, int *block_size, int count )
{
*block_size = CUDA_BLOCK_SIZE;
*blocks = (count / CUDA_BLOCK_SIZE ) + (count % CUDA_BLOCK_SIZE == 0 ? 0 : 1);
}
-void compute_nearest_pow_2 (int blocks, int *result)
+
+void compute_nearest_pow_2( int blocks, int *result )
{
int power = 1;
- while (power < blocks) power *= 2;
+ while (power < blocks)
+ {
+ power *= 2;
+ }
*result = power;
}
-void print_device_mem_usage ()
+void print_device_mem_usage( )
{
size_t total, free;
- cudaMemGetInfo (&free, &total);
- if (cudaGetLastError () != cudaSuccess )
+ cudaMemGetInfo( &free, &total );
+ if ( cudaGetLastError() != cudaSuccess )
{
- fprintf (stderr, "Error on the memory call \n");
+ fprintf( stderr, "Error on the memory call \n" );
return;
}
- fprintf (stderr, "Total %ld Mb %ld gig %ld , free %ld, Mb %ld , gig %ld \n",
+ fprintf( stderr, "Total %ld Mb %ld gig %ld , free %ld, Mb %ld , gig %ld \n",
total, total/(1024*1024), total/ (1024*1024*1024),
free, free/(1024*1024), free/ (1024*1024*1024) );
}
diff --git a/PuReMD-GPU/src/cuda_utils.h b/PuReMD-GPU/src/cuda_utils.h
index ba793e4048c7367438a88b20b758b3090200f4b4..c8976d081bbead0048faa7dd12238c40f2eda5d7 100644
--- a/PuReMD-GPU/src/cuda_utils.h
+++ b/PuReMD-GPU/src/cuda_utils.h
@@ -21,35 +21,39 @@
#ifndef __CUDA_UTILS_H_
#define __CUDA_UTILS_H_
-#include "cuda.h"
-#include "cublas_v2.h"
-#include "cusparse_v2.h"
-#include "stdlib.h"
-#include "stdio.h"
+#include "mytypes.h"
+
+#include <stdlib.h>
+#include <stdio.h>
#define IDX2C(i,j,ld) (((j)*(ld))+(i))
-static __inline__ void modify (cublasHandle_t handle, float *m, int ldm, int n, int p, int q, float alpha, float beta)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static __inline__ void modify( cublasHandle_t handle, float *m, int ldm, int n, int p, int q, float alpha, float beta )
{
- cublasSscal (handle, n - p, &alpha, &m[IDX2C(p, q, ldm)], ldm);
- cublasSscal (handle, ldm - p, &beta, &m[IDX2C(p, q, ldm)], 1);
+ cublasSscal( handle, n - p, &alpha, &m[IDX2C(p, q, ldm)], ldm );
+ cublasSscal( handle, ldm - p, &beta, &m[IDX2C(p, q, ldm)], 1 );
}
-void cuda_malloc (void **, int , int , int);
-void cuda_free (void *, int);
-void cuda_memset (void *, int , size_t , int );
-void copy_host_device (void *, void *, int , enum cudaMemcpyKind, int);
-void copy_device (void *, void *, int , int );
+void cuda_malloc( void **, int , int , int );
+void cuda_free( void *, int );
+void cuda_memset( void *, int , size_t , int );
+void copy_host_device( void *, void *, int , enum cudaMemcpyKind, int );
+void copy_device( void *, void *, int , int );
-void compute_blocks (int *, int *, int);
-void compute_nearest_pow_2 (int blocks, int *result);
+void compute_blocks( int *, int *, int );
+void compute_nearest_pow_2( int blocks, int *result );
-void print_device_mem_usage ();
+void print_device_mem_usage( );
#define cusparseCheckError(cusparseStatus) __cusparseCheckError (cusparseStatus, __FILE__, __LINE__)
-inline void __cusparseCheckError( cusparseStatus_t cusparseStatus, const char *file, const int line )
+static inline void __cusparseCheckError( cusparseStatus_t cusparseStatus, const char *file, const int line )
{
- if (cusparseStatus != CUSPARSE_STATUS_SUCCESS)
+ if ( cusparseStatus != CUSPARSE_STATUS_SUCCESS )
{
fprintf (stderr, "failed .. %s:%d -- error code %d \n", __FILE__, __LINE__, cusparseStatus);
exit (-1);
@@ -59,35 +63,42 @@ inline void __cusparseCheckError( cusparseStatus_t cusparseStatus, const char *f
#define cublasCheckError(cublasStatus) __cublasCheckError (cublasStatus, __FILE__, __LINE__)
-inline void __cublasCheckError( cublasStatus_t cublasStatus, const char *file, const int line )
+static inline void __cublasCheckError( cublasStatus_t cublasStatus, const char *file, const int line )
{
- if (cublasStatus != CUBLAS_STATUS_SUCCESS)
+ if ( cublasStatus != CUBLAS_STATUS_SUCCESS )
{
- fprintf (stderr, "failed .. %s:%d -- error code %d \n", __FILE__, __LINE__, cublasStatus);
- exit (-1);
+ fprintf( stderr, "failed .. %s:%d -- error code %d \n", __FILE__, __LINE__, cublasStatus );
+ exit( -1 );
}
return;
}
-#define cudaCheckError() __cudaCheckError( __FILE__, __LINE__ )
-inline void __cudaCheckError( const char *file, const int line )
+
+#define cudaCheckError() __cudaCheckError( __FILE__, __LINE__ )
+static inline void __cudaCheckError( const char *file, const int line )
{
- cudaError err = cudaGetLastError();
+ cudaError err = cudaGetLastError( );
if ( cudaSuccess != err )
{
- fprintf (stderr, "Failed .. %s:%d -- gpu erro code %d\n", file, line, err );
+ fprintf( stderr, "Failed .. %s:%d -- gpu erro code %d\n", file, line, err );
exit( -1 );
}
// More careful checking. However, this will affect performance.
- // Comment away if needed.
/*
- err = cudaDeviceSynchronize();
+ err = cudaDeviceSynchronize( );
if( cudaSuccess != err )
{
exit( -1 );
}
*/
+
return;
}
+
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PuReMD-GPU/src/forces.c b/PuReMD-GPU/src/forces.c
new file mode 100644
index 0000000000000000000000000000000000000000..c95d4896e32f60e954d79b0b623520afb042e9ea
--- /dev/null
+++ b/PuReMD-GPU/src/forces.c
@@ -0,0 +1,910 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "forces.h"
+
+#include "box.h"
+#include "bond_orders.h"
+#include "single_body_interactions.h"
+#include "two_body_interactions.h"
+#include "three_body_interactions.h"
+#include "four_body_interactions.h"
+#include "list.h"
+#include "print_utils.h"
+#include "system_props.h"
+#include "QEq.h"
+#include "vector.h"
+#include "index_utils.h"
+
+
+void Dummy_Interaction( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+}
+
+
+void Init_Bonded_Force_Functions( control_params *control )
+{
+ Interaction_Functions[0] = Calculate_Bond_Orders;
+ Interaction_Functions[1] = Bond_Energy; //*/Dummy_Interaction;
+ Interaction_Functions[2] = LonePair_OverUnder_Coordination_Energy;
+ //*/Dummy_Interaction;
+ Interaction_Functions[3] = Three_Body_Interactions; //*/Dummy_Interaction;
+ Interaction_Functions[4] = Four_Body_Interactions; //*/Dummy_Interaction;
+ if( control->hb_cut > 0 )
+ Interaction_Functions[5] = Hydrogen_Bonds; //*/Dummy_Interaction;
+ else Interaction_Functions[5] = Dummy_Interaction;
+ Interaction_Functions[6] = Dummy_Interaction; //empty
+ Interaction_Functions[7] = Dummy_Interaction; //empty
+ Interaction_Functions[8] = Dummy_Interaction; //empty
+ Interaction_Functions[9] = Dummy_Interaction; //empty
+}
+
+
+void Compute_Bonded_Forces( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+
+ int i;
+ real t_start, t_elapsed;
+
+#ifdef TEST_ENERGY
+ /* Mark beginning of a new timestep in each energy file */
+ fprintf( out_control->ebond, "step: %d\n%6s%6s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "bo", "ebond", "total" );
+ fprintf( out_control->elp, "step: %d\n%6s%12s%12s%12s\n",
+ data->step, "atom", "nlp", "elp", "total" );
+ fprintf( out_control->eov, "step: %d\n%6s%12s%12s\n",
+ data->step, "atom", "eov", "total" );
+ fprintf( out_control->eun, "step: %d\n%6s%12s%12s\n",
+ data->step, "atom", "eun", "total" );
+ fprintf( out_control->eval, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3",
+ "angle", "bo(12)", "bo(23)", "eval", "epen", "total" );
+ fprintf( out_control->epen, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3",
+ "angle", "bo(12)", "bo(23)", "epen", "total" );
+ fprintf( out_control->ecoa, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3",
+ "angle", "bo(12)", "bo(23)", "ecoa", "total" );
+ fprintf( out_control->ehb, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3",
+ "r(23)", "angle", "bo(12)", "ehb", "total" );
+ fprintf( out_control->etor, "step: %d\n%6s%6s%6s%6s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3", "atom4",
+ "phi", "bo(23)", "etor", "total" );
+ fprintf( out_control->econ, "step:%d\n%6s%6s%6s%6s%12s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "atom3", "atom4",
+ "phi", "bo(12)", "bo(23)", "bo(34)", "econ", "total" );
+#endif
+
+ /* Implement all the function calls as function pointers */
+ for( i = 0; i < NO_OF_INTERACTIONS; i++ ) {
+ //for( i = 0; i < 5; i++ ) {
+ t_start = Get_Time ();
+ (Interaction_Functions[i])(system, control, data, workspace,
+ lists, out_control);
+ t_elapsed = Get_Timing_Info ( t_start );
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "function %d tme %lf - \n", i, t_elapsed );
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "f%d-", i );
+#endif
+#ifdef TEST_FORCES
+ (Print_Interactions[i])(system, control, data, workspace,
+ lists, out_control);
+#endif
+ }
+}
+
+
+void Compute_NonBonded_Forces( reax_system *system, control_params *control,
+ simulation_data *data,static_storage *workspace,
+ list** lists, output_controls *out_control )
+{
+ real t_start, t_elapsed;
+#ifdef TEST_ENERGY
+ fprintf( out_control->evdw, "step: %d\n%6s%6s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "r12", "evdw", "total" );
+ fprintf( out_control->ecou, "step: %d\n%6s%6s%12s%12s%12s%12s%12s\n",
+ data->step, "atom1", "atom2", "r12", "q1", "q2", "ecou", "total" );
+#endif
+
+ t_start = Get_Time( );
+ QEq( system, control, data, workspace, lists[FAR_NBRS], out_control );
+ t_elapsed = Get_Timing_Info( t_start );
+ data->timing.QEq += t_elapsed;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "qeq - " );
+#endif
+
+ if ( control->tabulate == 0)
+ vdW_Coulomb_Energy( system, control, data, workspace, lists, out_control );
+ else
+ Tabulated_vdW_Coulomb_Energy( system, control, data, workspace,
+ lists, out_control );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "nonb forces - " );
+#endif
+
+#ifdef TEST_FORCES
+ Print_vdW_Coulomb_Forces( system, control, data, workspace,
+ lists, out_control );
+#endif
+}
+
+
+/* This version of Compute_Total_Force computes forces from coefficients
+ accumulated by all interaction functions. Saves enormous time & space! */
+void Compute_Total_Force( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists )
+{
+ int i, pj;
+ list *bonds = (*lists) + BONDS;
+
+ for( i = 0; i < system->N; ++i )
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj )
+ if( i < bonds->select.bond_list[pj].nbr ) {
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT)
+ Add_dBond_to_Forces( i, pj, system, data, workspace, lists );
+ else
+ Add_dBond_to_Forces_NPT( i, pj, system, data, workspace, lists );
+ }
+}
+
+
+void Validate_Lists( static_storage *workspace, list **lists, int step, int n,
+ int Hmax, int Htop, int num_bonds, int num_hbonds )
+{
+ int i, flag;
+ list *bonds, *hbonds;
+
+ bonds = *lists + BONDS;
+ hbonds = *lists + HBONDS;
+
+ /* far neighbors */
+ if( Htop > Hmax * DANGER_ZONE ) {
+ workspace->realloc.Htop = Htop;
+ if( Htop > Hmax ) {
+ fprintf( stderr,
+ "step%d - ran out of space on H matrix: Htop=%d, max = %d",
+ step, Htop, Hmax );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+ /* bond list */
+ flag = -1;
+ workspace->realloc.num_bonds = num_bonds;
+ for( i = 0; i < n-1; ++i )
+ if( End_Index(i, bonds) >= Start_Index(i+1, bonds)-2 ) {
+ workspace->realloc.bonds = 1;
+ if( End_Index(i, bonds) > Start_Index(i+1, bonds) )
+ flag = i;
+ }
+
+ if( flag > -1 ) {
+ fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
+ step, flag, End_Index(flag,bonds), Start_Index(flag+1,bonds) );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+ if( End_Index(i, bonds) >= bonds->num_intrs-2 ) {
+ workspace->realloc.bonds = 1;
+
+ if( End_Index(i, bonds) > bonds->num_intrs ) {
+ fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d bond_end=%d\n",
+ step, flag, End_Index(i,bonds), bonds->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+
+
+ /* hbonds list */
+ if( workspace->num_H > 0 ) {
+ flag = -1;
+ workspace->realloc.num_hbonds = num_hbonds;
+ for( i = 0; i < workspace->num_H-1; ++i )
+ if( Num_Entries(i, hbonds) >=
+ (Start_Index(i+1, hbonds) - Start_Index(i, hbonds)) * DANGER_ZONE ) {
+ workspace->realloc.hbonds = 1;
+ if( End_Index(i, hbonds) > Start_Index(i+1, hbonds) )
+ flag = i;
+ }
+
+ if( flag > -1 ) {
+ fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
+ step, flag, End_Index(flag,hbonds), Start_Index(flag+1,hbonds) );
+ exit(INSUFFICIENT_SPACE);
+ }
+
+ if( Num_Entries(i,hbonds) >=
+ (hbonds->num_intrs - Start_Index(i,hbonds)) * DANGER_ZONE ) {
+ workspace->realloc.hbonds = 1;
+
+ if( End_Index(i, hbonds) > hbonds->num_intrs ) {
+ fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d hbondend=%d\n",
+ step, flag, End_Index(i,hbonds), hbonds->num_intrs );
+ exit(INSUFFICIENT_SPACE);
+ }
+ }
+ }
+}
+
+
+void Init_Forces( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control ) {
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int Htop, btop_i, btop_j, num_bonds, num_hbonds;
+ int ihb, jhb, ihb_top, jhb_top;
+ int flag;
+ real r_ij, r2, self_coef;
+ real dr3gamij_1, dr3gamij_3, Tap;
+ //real val, dif, base;
+ real C12, C34, C56;
+ real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ sparse_matrix *H;
+ list *far_nbrs, *bonds, *hbonds;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ //LR_lookup_table *t;
+ reax_atom *atom_i, *atom_j;
+ bond_data *ibond, *jbond;
+ bond_order_data *bo_ij, *bo_ji;
+
+ far_nbrs = *lists + FAR_NBRS;
+ bonds = *lists + BONDS;
+ hbonds = *lists + HBONDS;
+
+ H = &workspace->H;
+ Htop = 0;
+ num_bonds = 0;
+ num_hbonds = 0;
+ btop_i = btop_j = 0;
+ p_boc1 = system->reaxprm.gp.l[0];
+ p_boc2 = system->reaxprm.gp.l[1];
+
+ for( i = 0; i < system->N; ++i ) {
+ atom_i = &(system->atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, far_nbrs);
+ end_i = End_Index(i, far_nbrs);
+ H->start[i] = Htop;
+ btop_i = End_Index( i, bonds );
+ sbp_i = &(system->reaxprm.sbp[type_i]);
+ ihb = ihb_top = -1;
+ if( control->hb_cut > 0 && (ihb=sbp_i->p_hbond) == 1 )
+ ihb_top = End_Index( workspace->hbond_index[i], hbonds );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(system->atoms[j]);
+
+ flag = 0;
+ if((data->step-data->prev_steps) % control->reneighbor == 0) {
+ if( nbr_pj->d <= control->r_cut)
+ flag = 1;
+ else flag = 0;
+ }
+ else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,&(system->box),
+ nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+
+ if( flag ){
+ type_j = system->atoms[j].type;
+ r_ij = nbr_pj->d;
+ sbp_j = &(system->reaxprm.sbp[type_j]);
+ twbp = &(system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0;
+
+ /* H matrix entry */
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ H->entries[Htop].j = j;
+ H->entries[Htop].val = self_coef * Tap * EV_to_KCALpMOL / dr3gamij_3;
+ ++Htop;
+
+ /* hydrogen bond lists */
+ if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ // fprintf( stderr, "%d %d\n", atom1, atom2 );
+ jhb = sbp_j->p_hbond;
+ if( ihb == 1 && jhb == 2 ) {
+ hbonds->select.hbond_list[ihb_top].nbr = j;
+ hbonds->select.hbond_list[ihb_top].scl = 1;
+ hbonds->select.hbond_list[ihb_top].ptr = nbr_pj;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ else if( ihb == 2 && jhb == 1 ) {
+ jhb_top = End_Index( workspace->hbond_index[j], hbonds );
+ hbonds->select.hbond_list[jhb_top].nbr = i;
+ hbonds->select.hbond_list[jhb_top].scl = -1;
+ hbonds->select.hbond_list[jhb_top].ptr = nbr_pj;
+ Set_End_Index( workspace->hbond_index[j], jhb_top+1, hbonds );
+ ++num_hbonds;
+ }
+ }
+
+ /* uncorrected bond orders */
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->nbr_cut ) {
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+ if( BO >= control->bo_cut ) {
+ num_bonds += 2;
+ /****** bonds i-j and j-i ******/
+ ibond = &( bonds->select.bond_list[btop_i] );
+ btop_j = End_Index( j, bonds );
+ jbond = &(bonds->select.bond_list[btop_j]);
+
+ ibond->nbr = j;
+ jbond->nbr = i;
+ ibond->d = r_ij;
+ jbond->d = r_ij;
+ rvec_Copy( ibond->dvec, nbr_pj->dvec );
+ rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
+ ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
+ ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
+ ibond->dbond_index = btop_i;
+ jbond->dbond_index = btop_i;
+ ibond->sym_index = btop_j;
+ jbond->sym_index = btop_i;
+ ++btop_i;
+ Set_End_Index( j, btop_j+1, bonds );
+
+ bo_ij = &( ibond->bo_data );
+ bo_ji = &( jbond->bo_data );
+ bo_ji->BO = bo_ij->BO = BO;
+ bo_ji->BO_s = bo_ij->BO_s = BO_s;
+ bo_ji->BO_pi = bo_ij->BO_pi = BO_pi;
+ bo_ji->BO_pi2 = bo_ij->BO_pi2 = BO_pi2;
+
+ /* Bond Order page2-3, derivative of total bond order prime */
+ Cln_BOp_s = twbp->p_bo2 * C12 / r2;
+ Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
+ Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
+
+ /* Only dln_BOp_xx wrt. dr_i is stored here, note that
+ dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
+ rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi2,
+ -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
+ rvec_Scale(bo_ji->dln_BOp_s, -1., bo_ij->dln_BOp_s);
+ rvec_Scale(bo_ji->dln_BOp_pi, -1., bo_ij->dln_BOp_pi );
+ rvec_Scale(bo_ji->dln_BOp_pi2, -1., bo_ij->dln_BOp_pi2 );
+
+ /* Only dBOp wrt. dr_i is stored here, note that
+ dBOp/dr_i = -dBOp/dr_j and all others are 0 */
+ rvec_Scale( bo_ij->dBOp,
+ -(bo_ij->BO_s * Cln_BOp_s +
+ bo_ij->BO_pi * Cln_BOp_pi +
+ bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
+ rvec_Scale( bo_ji->dBOp, -1., bo_ij->dBOp );
+
+ rvec_Add( workspace->dDeltap_self[i], bo_ij->dBOp );
+ rvec_Add( workspace->dDeltap_self[j], bo_ji->dBOp );
+
+ bo_ij->BO_s -= control->bo_cut;
+ bo_ij->BO -= control->bo_cut;
+ bo_ji->BO_s -= control->bo_cut;
+ bo_ji->BO -= control->bo_cut;
+ workspace->total_bond_order[i] += bo_ij->BO; //currently total_BOp
+ workspace->total_bond_order[j] += bo_ji->BO; //currently total_BOp
+ bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
+ bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
+
+ /*fprintf( stderr, "%d %d %g %g %g\n",
+ i+1, j+1, bo_ij->BO, bo_ij->BO_pi, bo_ij->BO_pi2 );*/
+
+ /*fprintf( stderr, "Cln_BOp_s: %f, pbo2: %f, C12:%f\n",
+ Cln_BOp_s, twbp->p_bo2, C12 );
+ fprintf( stderr, "Cln_BOp_pi: %f, pbo4: %f, C34:%f\n",
+ Cln_BOp_pi, twbp->p_bo4, C34 );
+ fprintf( stderr, "Cln_BOp_pi2: %f, pbo6: %f, C56:%f\n",
+ Cln_BOp_pi2, twbp->p_bo6, C56 );*/
+ /*fprintf(stderr, "pbo1: %f, pbo2:%f\n", twbp->p_bo1, twbp->p_bo2);
+ fprintf(stderr, "pbo3: %f, pbo4:%f\n", twbp->p_bo3, twbp->p_bo4);
+ fprintf(stderr, "pbo5: %f, pbo6:%f\n", twbp->p_bo5, twbp->p_bo6);
+ fprintf( stderr, "r_s: %f, r_p: %f, r_pp: %f\n",
+ twbp->r_s, twbp->r_p, twbp->r_pp );
+ fprintf( stderr, "C12: %g, C34:%g, C56:%g\n", C12, C34, C56 );*/
+
+ /*fprintf( stderr, "\tfactors: %g %g %g\n",
+ -(bo_ij->BO_s * Cln_BOp_s + bo_ij->BO_pi * Cln_BOp_pi +
+ bo_ij->BO_pi2 * Cln_BOp_pp),
+ -bo_ij->BO_pi * Cln_BOp_pi, -bo_ij->BO_pi2 * Cln_BOp_pi2 );*/
+ /*fprintf( stderr, "dBOpi:\t[%g, %g, %g]\n",
+ bo_ij->dBOp[0], bo_ij->dBOp[1], bo_ij->dBOp[2] );
+ fprintf( stderr, "dBOpi:\t[%g, %g, %g]\n",
+ bo_ij->dln_BOp_pi[0], bo_ij->dln_BOp_pi[1],
+ bo_ij->dln_BOp_pi[2] );
+ fprintf( stderr, "dBOpi2:\t[%g, %g, %g]\n\n",
+ bo_ij->dln_BOp_pi2[0], bo_ij->dln_BOp_pi2[1],
+ bo_ij->dln_BOp_pi2[2] );*/
+
+ Set_End_Index( j, btop_j+1, bonds );
+ }
+ }
+ }
+ }
+
+ H->entries[Htop].j = i;
+ H->entries[Htop].val = system->reaxprm.sbp[type_i].eta;
+ ++Htop;
+
+ Set_End_Index( i, btop_i, bonds );
+ if( ihb == 1 )
+ Set_End_Index( workspace->hbond_index[i], ihb_top, hbonds );
+ //fprintf( stderr, "%d bonds start: %d, end: %d\n",
+ // i, Start_Index( i, bonds ), End_Index( i, bonds ) );
+ }
+
+ // mark the end of j list
+ H->start[i] = Htop;
+ /* validate lists - decide if reallocation is required! */
+ Validate_Lists( workspace, lists,
+ data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d: Htop = %d, num_bonds = %d, num_hbonds = %d\n",
+ data->step, Htop, num_bonds, num_hbonds );
+#endif
+}
+
+
+void Init_Forces_Tab( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control ) {
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int Htop, btop_i, btop_j, num_bonds, num_hbonds;
+ int tmin, tmax, r;
+ int ihb, jhb, ihb_top, jhb_top;
+ int flag;
+ real r_ij, r2, self_coef;
+ real val, dif, base;
+ real C12, C34, C56;
+ real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ sparse_matrix *H;
+ list *far_nbrs, *bonds, *hbonds;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ LR_lookup_table *t;
+ reax_atom *atom_i, *atom_j;
+ bond_data *ibond, *jbond;
+ bond_order_data *bo_ij, *bo_ji;
+
+ far_nbrs = *lists + FAR_NBRS;
+ bonds = *lists + BONDS;
+ hbonds = *lists + HBONDS;
+
+ H = &workspace->H;
+ Htop = 0;
+ num_bonds = 0;
+ num_hbonds = 0;
+ btop_i = btop_j = 0;
+ p_boc1 = system->reaxprm.gp.l[0];
+ p_boc2 = system->reaxprm.gp.l[1];
+
+ for( i = 0; i < system->N; ++i ) {
+ atom_i = &(system->atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, far_nbrs);
+ end_i = End_Index(i, far_nbrs);
+ H->start[i] = Htop;
+ btop_i = End_Index( i, bonds );
+ sbp_i = &(system->reaxprm.sbp[type_i]);
+ ihb = ihb_top = -1;
+ if( control->hb_cut > 0 && (ihb=sbp_i->p_hbond) == 1 )
+ ihb_top = End_Index( workspace->hbond_index[i], hbonds );
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(system->atoms[j]);
+
+ flag = 0;
+ if((data->step-data->prev_steps) % control->reneighbor == 0) {
+ if(nbr_pj->d <= control->r_cut)
+ flag = 1;
+ else flag = 0;
+ }
+ else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,&(system->box),
+ nbr_pj->dvec))<=SQR(control->r_cut)){
+ nbr_pj->d = sqrt(nbr_pj->d);
+ flag = 1;
+ }
+
+ if( flag ){
+ type_j = system->atoms[j].type;
+ r_ij = nbr_pj->d;
+ sbp_j = &(system->reaxprm.sbp[type_j]);
+ twbp = &(system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( LR[ index_lr (tmin,tmax,system->reaxprm.num_atom_types) ] );
+
+ /* cubic spline interpolation */
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+ val = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
+ t->ele[r].a;
+ val *= EV_to_KCALpMOL / C_ele;
+
+ H->entries[Htop].j = j;
+ H->entries[Htop].val = self_coef * val;
+ ++Htop;
+
+ /* hydrogen bond lists */
+ if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ // fprintf( stderr, "%d %d\n", atom1, atom2 );
+ jhb = sbp_j->p_hbond;
+ if( ihb == 1 && jhb == 2 ) {
+ hbonds->select.hbond_list[ihb_top].nbr = j;
+ hbonds->select.hbond_list[ihb_top].scl = 1;
+ hbonds->select.hbond_list[ihb_top].ptr = nbr_pj;
+ ++ihb_top;
+ ++num_hbonds;
+ }
+ else if( ihb == 2 && jhb == 1 ) {
+ jhb_top = End_Index( workspace->hbond_index[j], hbonds );
+ hbonds->select.hbond_list[jhb_top].nbr = i;
+ hbonds->select.hbond_list[jhb_top].scl = -1;
+ hbonds->select.hbond_list[jhb_top].ptr = nbr_pj;
+ Set_End_Index( workspace->hbond_index[j], jhb_top+1, hbonds );
+ ++num_hbonds;
+ }
+ }
+
+ /* uncorrected bond orders */
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->nbr_cut ) {
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+ if( BO >= control->bo_cut ) {
+ num_bonds += 2;
+ /****** bonds i-j and j-i ******/
+ ibond = &( bonds->select.bond_list[btop_i] );
+ btop_j = End_Index( j, bonds );
+ jbond = &(bonds->select.bond_list[btop_j]);
+
+ ibond->nbr = j;
+ jbond->nbr = i;
+ ibond->d = r_ij;
+ jbond->d = r_ij;
+ rvec_Copy( ibond->dvec, nbr_pj->dvec );
+ rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
+ ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
+ ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
+ ibond->dbond_index = btop_i;
+ jbond->dbond_index = btop_i;
+ ibond->sym_index = btop_j;
+ jbond->sym_index = btop_i;
+ ++btop_i;
+ Set_End_Index( j, btop_j+1, bonds );
+
+ bo_ij = &( ibond->bo_data );
+ bo_ji = &( jbond->bo_data );
+ bo_ji->BO = bo_ij->BO = BO;
+ bo_ji->BO_s = bo_ij->BO_s = BO_s;
+ bo_ji->BO_pi = bo_ij->BO_pi = BO_pi;
+ bo_ji->BO_pi2 = bo_ij->BO_pi2 = BO_pi2;
+
+ /* Bond Order page2-3, derivative of total bond order prime */
+ Cln_BOp_s = twbp->p_bo2 * C12 / r2;
+ Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
+ Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
+
+ /* Only dln_BOp_xx wrt. dr_i is stored here, note that
+ dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
+ rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
+ rvec_Scale(bo_ij->dln_BOp_pi2,
+ -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
+ rvec_Scale(bo_ji->dln_BOp_s, -1., bo_ij->dln_BOp_s);
+ rvec_Scale(bo_ji->dln_BOp_pi, -1., bo_ij->dln_BOp_pi );
+ rvec_Scale(bo_ji->dln_BOp_pi2, -1., bo_ij->dln_BOp_pi2 );
+
+ /* Only dBOp wrt. dr_i is stored here, note that
+ dBOp/dr_i = -dBOp/dr_j and all others are 0 */
+ rvec_Scale( bo_ij->dBOp,
+ -(bo_ij->BO_s * Cln_BOp_s +
+ bo_ij->BO_pi * Cln_BOp_pi +
+ bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
+ rvec_Scale( bo_ji->dBOp, -1., bo_ij->dBOp );
+
+ rvec_Add( workspace->dDeltap_self[i], bo_ij->dBOp );
+ rvec_Add( workspace->dDeltap_self[j], bo_ji->dBOp );
+
+ bo_ij->BO_s -= control->bo_cut;
+ bo_ij->BO -= control->bo_cut;
+ bo_ji->BO_s -= control->bo_cut;
+ bo_ji->BO -= control->bo_cut;
+ workspace->total_bond_order[i] += bo_ij->BO; //currently total_BOp
+ workspace->total_bond_order[j] += bo_ji->BO; //currently total_BOp
+ bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
+ bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
+
+ Set_End_Index( j, btop_j+1, bonds );
+ }
+ }
+ }
+ }
+
+ H->entries[Htop].j = i;
+ H->entries[Htop].val = system->reaxprm.sbp[type_i].eta;
+ ++Htop;
+
+ Set_End_Index( i, btop_i, bonds );
+ if( ihb == 1 )
+ Set_End_Index( workspace->hbond_index[i], ihb_top, hbonds );
+ }
+
+ // mark the end of j list
+ H->start[i] = Htop;
+ /* validate lists - decide if reallocation is required! */
+ Validate_Lists( workspace, lists,
+ data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "step%d: Htop = %d, num_bonds = %d, num_hbonds = %d\n",
+ data->step, Htop, num_bonds, num_hbonds );
+ //Print_Bonds( system, bonds, "sbonds.out" );
+ //Print_Bond_List2( system, bonds, "sbonds.out" );
+ //Print_Sparse_Matrix2( H, "H.out" );
+#endif
+}
+
+
+void Estimate_Storage_Sizes( reax_system *system, control_params *control,
+ list **lists, int *Htop, int *hb_top,
+ int *bond_top, int *num_3body ) {
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ int ihb, jhb;
+ real r_ij, r2;
+ real C12, C34, C56;
+ real BO, BO_s, BO_pi, BO_pi2;
+ real p_boc1, p_boc2;
+ list *far_nbrs;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ reax_atom *atom_i, *atom_j;
+
+ far_nbrs = *lists + FAR_NBRS;
+ p_boc1 = system->reaxprm.gp.l[0];
+ p_boc2 = system->reaxprm.gp.l[1];
+
+ for( i = 0; i < system->N; ++i ) {
+ atom_i = &(system->atoms[i]);
+ type_i = atom_i->type;
+ start_i = Start_Index(i, far_nbrs);
+ end_i = End_Index(i, far_nbrs);
+ sbp_i = &(system->reaxprm.sbp[type_i]);
+ ihb = sbp_i->p_hbond;
+
+ for( pj = start_i; pj < end_i; ++pj ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ atom_j = &(system->atoms[j]);
+ type_j = atom_j->type;
+ sbp_j = &(system->reaxprm.sbp[type_j]);
+ twbp = &(system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+
+ if( nbr_pj->d <= control->r_cut ) {
+ ++(*Htop);
+
+ /* hydrogen bond lists */
+ if( control->hb_cut > 0.1 && (ihb==1 || ihb==2) &&
+ nbr_pj->d <= control->hb_cut ) {
+ jhb = sbp_j->p_hbond;
+ if( ihb == 1 && jhb == 2 )
+ ++hb_top[i];
+ else if( ihb == 2 && jhb == 1 )
+ ++hb_top[j];
+ }
+
+ /* uncorrected bond orders */
+ if( nbr_pj->d <= control->nbr_cut ) {
+ r_ij = nbr_pj->d;
+ r2 = SQR(r_ij);
+
+ if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
+ C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
+ BO_s = (1.0 + control->bo_cut) * EXP( C12 );
+ }
+ else BO_s = C12 = 0.0;
+
+ if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
+ C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
+ BO_pi = EXP( C34 );
+ }
+ else BO_pi = C34 = 0.0;
+
+ if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
+ C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
+ BO_pi2= EXP( C56 );
+ }
+ else BO_pi2 = C56 = 0.0;
+
+ /* Initially BO values are the uncorrected ones, page 1 */
+ BO = BO_s + BO_pi + BO_pi2;
+
+ if( BO >= control->bo_cut ) {
+ ++bond_top[i];
+ ++bond_top[j];
+ }
+ }
+ }
+ }
+ }
+
+ *Htop += system->N;
+ *Htop *= SAFE_ZONE;
+
+ for( i = 0; i < system->N; ++i ) {
+ hb_top[i] = MAX( hb_top[i] * SAFE_HBONDS, MIN_HBONDS );
+ *num_3body += SQR(bond_top[i]);
+ bond_top[i] = MAX( bond_top[i] * 2, MIN_BONDS );
+ }
+ *num_3body *= SAFE_ZONE;
+}
+
+
+void Compute_Forces( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list** lists, output_controls *out_control )
+{
+ real t_start, t_elapsed;
+
+ t_start = Get_Time( );
+ if( !control->tabulate )
+ Init_Forces( system, control, data, workspace, lists, out_control );
+ else Init_Forces_Tab( system, control, data, workspace, lists, out_control );
+ t_elapsed = Get_Timing_Info( t_start );
+ data->timing.init_forces += t_elapsed;
+
+#if defined(DEBUG_FOCUS)
+ print_sparse_matrix (system, workspace);
+ fprintf( stderr, "init_forces - ");
+#endif
+
+
+ //analyze_hbonds (system, workspace, lists);
+
+ t_start = Get_Time( );
+ Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
+ t_elapsed = Get_Timing_Info( t_start );
+ data->timing.bonded += t_elapsed;
+
+ //print_bond_list (system, workspace, lists);
+ //exit (0);
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "bonded_forces - ");
+#endif
+
+ t_start = Get_Time( );
+ Compute_NonBonded_Forces( system, control, data, workspace,
+ lists, out_control );
+ t_elapsed = Get_Timing_Info( t_start );
+ data->timing.nonb += t_elapsed;
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "non_bonded_forces - %lf \n", t_elapsed);
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "nonbondeds - ");
+#endif
+
+ Compute_Total_Force( system, control, data, workspace, lists );
+ //Print_Total_Force( system, control, data, workspace, lists, out_control );
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "totalforces - ");
+ //Print_Total_Force( system, control, data, workspace, lists, out_control );
+#endif
+
+#ifdef TEST_FORCES
+ Print_Total_Force( system, control, data, workspace, lists, out_control );
+ Compare_Total_Forces( system, control, data, workspace, lists, out_control );
+#endif
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "forces - ");
+#endif
+}
diff --git a/PuReMD-GPU/src/forces.cu b/PuReMD-GPU/src/forces.cu
deleted file mode 100644
index e8e1e2917b34eefa913f25cab67b1123d08f892e..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/forces.cu
+++ /dev/null
@@ -1,2880 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "forces.h"
-#include "box.h"
-#include "bond_orders.h"
-#include "single_body_interactions.h"
-#include "two_body_interactions.h"
-#include "three_body_interactions.h"
-#include "four_body_interactions.h"
-#include "list.h"
-#include "print_utils.h"
-#include "system_props.h"
-#include "QEq.h"
-#include "vector.h"
-
-#include "index_utils.h"
-#include "cuda_utils.h"
-#include "cuda_init.h"
-#include "reduction.h"
-//#include "matrix.h"
-
-#include "validation.h"
-
-#include "cudaProfiler.h"
-
-
-void Dummy_Interaction( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
-}
-
-
-void Init_Bonded_Force_Functions( control_params *control )
-{
- Interaction_Functions[0] = Calculate_Bond_Orders;
- Interaction_Functions[1] = Bond_Energy; //*/Dummy_Interaction;
- Interaction_Functions[2] = LonePair_OverUnder_Coordination_Energy;
- //*/Dummy_Interaction;
- Interaction_Functions[3] = Three_Body_Interactions; //*/Dummy_Interaction;
- Interaction_Functions[4] = Four_Body_Interactions; //*/Dummy_Interaction;
- if( control->hb_cut > 0 )
- Interaction_Functions[5] = Hydrogen_Bonds; //*/Dummy_Interaction;
- else Interaction_Functions[5] = Dummy_Interaction;
- Interaction_Functions[6] = Dummy_Interaction; //empty
- Interaction_Functions[7] = Dummy_Interaction; //empty
- Interaction_Functions[8] = Dummy_Interaction; //empty
- Interaction_Functions[9] = Dummy_Interaction; //empty
-}
-
-
-void Compute_Bonded_Forces( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
-
- int i;
- real t_start, t_elapsed;
-
-#ifdef TEST_ENERGY
- /* Mark beginning of a new timestep in each energy file */
- fprintf( out_control->ebond, "step: %d\n%6s%6s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "bo", "ebond", "total" );
- fprintf( out_control->elp, "step: %d\n%6s%12s%12s%12s\n",
- data->step, "atom", "nlp", "elp", "total" );
- fprintf( out_control->eov, "step: %d\n%6s%12s%12s\n",
- data->step, "atom", "eov", "total" );
- fprintf( out_control->eun, "step: %d\n%6s%12s%12s\n",
- data->step, "atom", "eun", "total" );
- fprintf( out_control->eval, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3",
- "angle", "bo(12)", "bo(23)", "eval", "epen", "total" );
- fprintf( out_control->epen, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3",
- "angle", "bo(12)", "bo(23)", "epen", "total" );
- fprintf( out_control->ecoa, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3",
- "angle", "bo(12)", "bo(23)", "ecoa", "total" );
- fprintf( out_control->ehb, "step: %d\n%6s%6s%6s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3",
- "r(23)", "angle", "bo(12)", "ehb", "total" );
- fprintf( out_control->etor, "step: %d\n%6s%6s%6s%6s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3", "atom4",
- "phi", "bo(23)", "etor", "total" );
- fprintf( out_control->econ, "step:%d\n%6s%6s%6s%6s%12s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "atom3", "atom4",
- "phi", "bo(12)", "bo(23)", "bo(34)", "econ", "total" );
-#endif
-
- /* Implement all the function calls as function pointers */
- for( i = 0; i < NO_OF_INTERACTIONS; i++ ) {
- //for( i = 0; i < 5; i++ ) {
- t_start = Get_Time ();
- (Interaction_Functions[i])(system, control, data, workspace,
- lists, out_control);
- t_elapsed = Get_Timing_Info ( t_start );
-
-#ifdef __DEBUG_CUDA__
- fprintf( stderr, "function %d tme %lf - \n", i, t_elapsed );
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "f%d-", i );
-#endif
-#ifdef TEST_FORCES
- (Print_Interactions[i])(system, control, data, workspace,
- lists, out_control);
-#endif
- }
- }
-
- void Cuda_Compute_Bonded_Forces( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- real t_start, t_elapsed;
- real *spad = (real *)scratch;
- rvec *rvec_spad;
-
- //Compute the bonded for interaction here.
- //Step 1.
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
- fprintf (stderr, " Begin Bonded Forces ... %d x %d\n", BLOCKS, BLOCK_SIZE);
-#endif
-
- Cuda_Calculate_Bond_Orders_Init <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
- *dev_workspace, system->reaxprm.num_atom_types, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Calculate_Bond_Orders <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
- system->reaxprm.d_tbp, *dev_workspace,
- *(dev_lists + BONDS), *(dev_lists + DDELTA), *(dev_lists + DBO),
- system->reaxprm.num_atom_types, system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Update_Uncorrected_BO <<<BLOCKS, BLOCK_SIZE>>>
- (*dev_workspace, *(dev_lists + BONDS), system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Update_Workspace_After_Bond_Orders <<<BLOCKS, BLOCK_SIZE>>>
- (system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp,
- *dev_workspace, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Bond Orders... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
- fprintf (stderr, "Cuda_Calculate_Bond_Orders Done... \n");
-#endif
-
-
-
- //Step 2.
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
- //cuda_memset (spad, 0, system->N * ( 2 * REAL_SIZE + system->N * REAL_SIZE + 16 * REAL_SIZE), RES_SCRATCH );
- cuda_memset (spad, 0, system->N * ( 2 * REAL_SIZE ) , RES_SCRATCH );
-
- Cuda_Bond_Energy <<< BLOCKS, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- ( system->d_atoms, system->reaxprm.d_gp, system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- (simulation_data *)data->d_simulation_data, *dev_workspace, *(dev_lists + BONDS),
- system->N, system->reaxprm.num_atom_types, spad );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_BE
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
- //(spad + system->N, spad + system->N + 16, 16);
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_BE, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Cuda_Bond_Energy ... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
- fprintf (stderr, "Cuda_Bond_Energy Done... \n");
-#endif
-
- //Step 3.
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
- cuda_memset (spad, 0, ( 6 * REAL_SIZE * system->N ), RES_SCRATCH );
-
- test_LonePair_OverUnder_Coordination_Energy_LP <<<BLOCKS, BLOCK_SIZE>>>( system->d_atoms, system->reaxprm.d_gp,
- system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- *dev_workspace, (simulation_data *)data->d_simulation_data,
- *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types,
- spad, spad + 2 * system->N, spad + 4*system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- test_LonePair_OverUnder_Coordination_Energy <<<BLOCKS, BLOCK_SIZE>>>( system->d_atoms, system->reaxprm.d_gp,
- system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- *dev_workspace, (simulation_data *)data->d_simulation_data,
- *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types,
- spad, spad + 2 * system->N, spad + 4*system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- test_LonePair_Postprocess <<<BLOCKS, BLOCK_SIZE, 0>>>( system->d_atoms, system->reaxprm.d_gp,
- system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- *dev_workspace, (simulation_data *)data->d_simulation_data,
- *(dev_lists + BONDS), system->N, system->reaxprm.num_atom_types);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- //Reduction for E_Lp
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Lp, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Ov
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad + 2*system->N, spad + 3*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Ov, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Un
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad + 4*system->N, spad + 5*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + 5*system->N, &((simulation_data *)data->d_simulation_data)->E_Un, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "test_LonePair_postprocess ... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
- fprintf (stderr, "test_LonePair_postprocess Done... \n");
-#endif
-
- //Step 4.
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
-
- cuda_memset(spad, 0, (dev_lists + BONDS)->num_intrs * sizeof (int), RES_SCRATCH);
- Three_Body_Estimate <<<BLOCKS, BLOCK_SIZE>>>
- (system->d_atoms,
- (control_params *)control->d_control,
- *(dev_lists + BONDS),
- system->N, (int *)spad);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Three_Body_Estimate... return value --> %d --- Timing %lf \n", cudaGetLastError (), t_elapsed );
-#endif
-
- int *thbody = (int *) malloc (sizeof (int) * (dev_lists + BONDS)->num_intrs);
- memset (thbody, 0, sizeof (int) * (dev_lists + BONDS)->num_intrs);
- copy_host_device (thbody, spad, (dev_lists + BONDS)->num_intrs * sizeof (int), cudaMemcpyDeviceToHost, RES_SCRATCH);
-
- int total_3body = thbody [0] * SAFE_ZONE;
- for (int x = 1; x < (dev_lists + BONDS)->num_intrs; x++) {
- total_3body += thbody [x]*SAFE_ZONE;
- thbody [x] += thbody [x-1];
- }
- system->num_thbodies = thbody [(dev_lists+BONDS)->num_intrs-1];
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Total Three body estimate is %d (bonds: %d) \n", total_3body, (dev_lists+BONDS)->num_intrs);
-#endif
-
- if (!system->init_thblist)
- {
- system->init_thblist = true;
- if(!Make_List((dev_lists+BONDS)->num_intrs, total_3body, TYP_THREE_BODY, dev_lists + THREE_BODIES, TYP_DEVICE)) {
- fprintf( stderr, "Problem in initializing three-body list. Terminating!\n" );
- exit( INIT_ERR );
- }
-#ifdef __CUDA_MEM__
- fprintf (stderr, "Device memory allocated: three body list = %d MB\n",
- sizeof (three_body_interaction_data) * total_3body / (1024*1024));
-#endif
- } else {
- if ((dev_workspace->realloc.bonds > 0) || (system->num_thbodies > (dev_lists+THREE_BODIES)->num_intrs )) {
- int size = MAX (dev_workspace->realloc.num_bonds, (dev_lists+BONDS)->num_intrs);
-
- /*Delete Three-body list*/
- Delete_List( dev_lists + THREE_BODIES, TYP_DEVICE );
-
-#ifdef __CUDA_MEM__
- fprintf (stderr, "Reallocating Three-body list: step: %d n - %d num_intrs - %d used: %d \n",
- data->step, dev_workspace->realloc.num_bonds, total_3body, system->num_thbodies);
-#endif
- /*Recreate Three-body list */
- if(!Make_List(size, total_3body, TYP_THREE_BODY, dev_lists + THREE_BODIES, TYP_DEVICE)) {
- fprintf( stderr, "Problem in initializing three-body list. Terminating!\n" );
- exit( INIT_ERR );
- }
- }
- }
-
- //copy the indexes into the thb list;
- copy_host_device (thbody, ((dev_lists + THREE_BODIES)->index + 1), sizeof (int) * ((dev_lists+BONDS)->num_intrs - 1),
- cudaMemcpyHostToDevice, LIST_INDEX);
- copy_host_device (thbody, ((dev_lists + THREE_BODIES)->end_index + 1), sizeof (int) * ((dev_lists+BONDS)->num_intrs - 1),
- cudaMemcpyHostToDevice, LIST_END_INDEX);
-
- free (thbody );
-
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
-
- cuda_memset (spad, 0, ( 6 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2), RES_SCRATCH );
-
- Three_Body_Interactions <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms,
- system->reaxprm.d_sbp, system->reaxprm.d_thbp, system->reaxprm.d_gp,
- (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data,
- *dev_workspace,
- *(dev_lists + BONDS), *(dev_lists + THREE_BODIES),
- system->N, system->reaxprm.num_atom_types,
- spad, spad + 2*system->N, spad + 4*system->N, (rvec *)(spad + 6*system->N));
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Not necessary to validate three-body list anymore,
- // Estimate is already done at the beginning which makes sure that
- // we have sufficient size for this list
- //Cuda_Threebody_List( system, workspace, dev_lists + THREE_BODIES, data->step );
-
- //Reduction for E_Ang
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Ang, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Pen
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad + 2*system->N, spad + 3*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Pen, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Coa
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad + 4*system->N, spad + 5*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + 5*system->N, &((simulation_data *)data->d_simulation_data)->E_Coa, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for ext_pres
- rvec_spad = (rvec *) (spad + 6*system->N);
- Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
- (rvec_spad, rvec_spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
- (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- real t_1, t_2;
- t_1 = Get_Time ();
- //Sum up the f vector for each atom and collect the CdDelta from all the bonds
- Three_Body_Interactions_results <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms,
- (control_params *)control->d_control,
- *dev_workspace,
- *(dev_lists + BONDS),
- system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- t_2 = Get_Timing_Info (t_1);
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Three_Body_Interactions post process Timing %lf \n", t_2);
- fprintf (stderr, "Three_Body_Interactions ... Timing %lf \n", t_elapsed );
- fprintf (stderr, "Three_Body_Interactions Done... \n");
-#endif
-
- //Step 5.
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
-
- cuda_memset (spad, 0, ( 4 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2), RES_SCRATCH );
- Four_Body_Interactions <<< BLOCKS, BLOCK_SIZE >>>
- //Four_Body_Interactions <<< system->N, 32, 32*( 2*REAL_SIZE + RVEC_SIZE)>>>
- ( system->d_atoms,
- system->reaxprm.d_gp,
- system->reaxprm.d_fbp,
- (control_params *)control->d_control,
- *(dev_lists + BONDS), *(dev_lists + THREE_BODIES),
- (simulation_box *)system->d_box,
- (simulation_data *)data->d_simulation_data,
- *dev_workspace,
- system->N, system->reaxprm.num_atom_types,
- spad, spad + 2*system->N, (rvec *) (spad + 4*system->N));
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Tor
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_Tor, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for E_Con
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad + 2*system->N, spad + 3*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Con, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Reduction for ext_pres
- rvec_spad = (rvec *) (spad + 4*system->N);
- Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
- (rvec_spad, rvec_spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
- (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Post process here
- Four_Body_Postprocess <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms,
- *dev_workspace,
- *(dev_lists + BONDS),
- system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Four_Body_post process return value --> %d --- Four body Timing %lf \n", cudaGetLastError (), t_elapsed );
- fprintf (stderr, " Four_Body_ Done... \n");
-#endif
-
-
- //Step 6.
- if (control->hb_cut > 0) {
-
-#ifdef __DEBUG_CUDA__
- t_start = Get_Time( );
-#endif
- cuda_memset (spad, 0, ( 2 * REAL_SIZE * system->N + RVEC_SIZE * system->N * 2 ), RES_SCRATCH );
-
- /*
- Hydrogen_Bonds <<< BLOCKS, BLOCK_SIZE, BLOCK_SIZE *( REAL_SIZE + RVEC_SIZE) >>>
- ( system->d_atoms,
- system->reaxprm.d_sbp,
- system->reaxprm.d_hbp,
- (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data,
- *dev_workspace,
- *(dev_lists + BONDS), *(dev_lists + HBONDS),
- system->N, system->reaxprm.num_atom_types,
- spad, (rvec *) (spad + 2*system->N), NULL);
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
-
-#ifdef __DEBUG_CUDA__
- real test1,test2;
- test1 = Get_Time ();
-#endif
-
- int hbs = (system->N * HBONDS_THREADS_PER_ATOM/ HBONDS_BLOCK_SIZE) +
- (((system->N * HBONDS_THREADS_PER_ATOM) % HBONDS_BLOCK_SIZE) == 0 ? 0 : 1);
- Hydrogen_Bonds_HB <<< hbs, HBONDS_BLOCK_SIZE, HBONDS_BLOCK_SIZE * ( 2 * REAL_SIZE + 2 * RVEC_SIZE ) >>>
- ( system->d_atoms,
- system->reaxprm.d_sbp,
- system->reaxprm.d_hbp,
- (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data,
- *dev_workspace,
- *(dev_lists + BONDS), *(dev_lists + HBONDS),
- system->N, system->reaxprm.num_atom_types,
- spad, (rvec *) (spad + 2*system->N), NULL);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- test2 = Get_Timing_Info (test1);
- fprintf (stderr, "Timing for the hb and forces ---> %f \n", test2);
-#endif
-
- //Reduction for E_HB
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_HB, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- //Reduction for ext_pres
- rvec_spad = (rvec *) (spad + 2*system->N);
- Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE >>>
- (rvec_spad, rvec_spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2 >>>
- (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //Post process here
-#ifdef __DEBUG_CUDA__
- real t_1, t_2;
- t_1 = Get_Time ();
-#endif
- Hydrogen_Bonds_Postprocess <<< BLOCKS, BLOCK_SIZE, BLOCK_SIZE * RVEC_SIZE >>>
- ( system->d_atoms,
- system->reaxprm.d_sbp,
- *dev_workspace,
- *(dev_lists + BONDS),
- *(dev_lists + HBONDS),
- *(dev_lists + FAR_NBRS),
- system->N,
- spad); //this is for the fix to use the shared memory
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-#ifdef __DEBUG_CUDA__
- t_2 = Get_Timing_Info ( t_1 );
- fprintf (stderr, " Hydrogen Bonds post process -----%f \n", t_2);
- t_1 = Get_Time ();
-#endif
-
- //Hydrogen_Bonds_Far_Nbrs <<< system->N, 32, 32 * RVEC_SIZE>>>
- Hydrogen_Bonds_HNbrs <<< system->N, 32, 32 * RVEC_SIZE>>>
- ( system->d_atoms,
- system->reaxprm.d_sbp,
- *dev_workspace,
- *(dev_lists + BONDS),
- *(dev_lists + HBONDS),
- *(dev_lists + FAR_NBRS),
- system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
- t_2 = Get_Timing_Info ( t_1 );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Hydrogen Bonds post process -----%f \n", t_2);
- t_elapsed = Get_Timing_Info( t_start );
- fprintf (stderr, "Hydrogen bonds post process return value --> %d --- HydrogenBonds Timing %lf \n", cudaGetLastError (), t_elapsed );
- fprintf (stderr, "Hydrogen_Bond Done... \n");
-#endif
- }
- return;
- }
-
- void Compute_NonBonded_Forces( reax_system *system, control_params *control,
- simulation_data *data,static_storage *workspace,
- list** lists, output_controls *out_control )
- {
- real t_start, t_elapsed;
-#ifdef TEST_ENERGY
- fprintf( out_control->evdw, "step: %d\n%6s%6s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "r12", "evdw", "total" );
- fprintf( out_control->ecou, "step: %d\n%6s%6s%12s%12s%12s%12s%12s\n",
- data->step, "atom1", "atom2", "r12", "q1", "q2", "ecou", "total" );
-#endif
-
- t_start = Get_Time( );
- QEq( system, control, data, workspace, lists[FAR_NBRS], out_control );
- t_elapsed = Get_Timing_Info( t_start );
- data->timing.QEq += t_elapsed;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "qeq - " );
-#endif
-
- if ( control->tabulate == 0)
- vdW_Coulomb_Energy( system, control, data, workspace, lists, out_control );
- else
- Tabulated_vdW_Coulomb_Energy( system, control, data, workspace,
- lists, out_control );
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "nonb forces - " );
-#endif
-
-#ifdef TEST_FORCES
- Print_vdW_Coulomb_Forces( system, control, data, workspace,
- lists, out_control );
-#endif
- }
-
- void Cuda_Compute_NonBonded_Forces( reax_system *system, control_params *control,
- simulation_data *data,static_storage *workspace,
- list** lists, output_controls *out_control )
- {
- real t_start, t_elapsed;
- real t1 = 0, t2 = 0;
- real *spad = (real *) scratch;
- rvec *rvec_spad;
- int cblks;
-
- t_start = Get_Time( );
- Cuda_QEq( system, control, data, workspace, lists[FAR_NBRS], out_control );
- t_elapsed = Get_Timing_Info( t_start );
- d_timing.QEq += t_elapsed;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Cuda_QEq done with timing %lf \n", t_elapsed );
-#endif
- cuda_memset (spad, 0, system->N * ( 4 * REAL_SIZE + 2 * RVEC_SIZE), RES_SCRATCH );
-
- t_start = Get_Time ();
- if ( control->tabulate == 0)
- {
- cblks = (system->N * VDW_THREADS_PER_ATOM / VDW_BLOCK_SIZE) +
- ((system->N * VDW_THREADS_PER_ATOM/VDW_BLOCK_SIZE) == 0 ? 0 : 1);
- Cuda_vdW_Coulomb_Energy <<< cblks, VDW_BLOCK_SIZE, VDW_BLOCK_SIZE * ( 2*REAL_SIZE + RVEC_SIZE) >>>
- ( system->d_atoms,
- system->reaxprm.d_tbp,
- system->reaxprm.d_gp,
- (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data,
- *(dev_lists + FAR_NBRS),
- spad , spad + 2 * system->N, (rvec *) (spad + system->N * 4),
- system->reaxprm.num_atom_types,
- system->N ) ;
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
- else
- {
- cblks = (system->N * VDW_THREADS_PER_ATOM / VDW_BLOCK_SIZE) +
- ((system->N * VDW_THREADS_PER_ATOM/VDW_BLOCK_SIZE) == 0 ? 0 : 1);
- Cuda_Tabulated_vdW_Coulomb_Energy <<< cblks, VDW_BLOCK_SIZE, VDW_BLOCK_SIZE* (2*REAL_SIZE + RVEC_SIZE)>>>
- ( (reax_atom *)system->d_atoms,
- (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data,
- *(dev_lists + FAR_NBRS),
- spad , spad + 2 * system->N, (rvec *) (spad + system->N * 4),
- d_LR,
- system->reaxprm.num_atom_types,
- out_control->energy_update_freq,
- system->N ) ;
-
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
-
- t_elapsed = Get_Timing_Info (t_start );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Cuda_Tabulated_vdW_Coulomb_Energy done... %lf \n", (t_elapsed - t2));
- fprintf (stderr, "Cuda_Tabulated_vdW_Coulomb_Energy done... %lf \n", (t_elapsed));
-#endif
-
- //Reduction on E_vdW
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (spad, spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
- (spad + system->N, &((simulation_data *)data->d_simulation_data)->E_vdW, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //reduction on E_Ele
- Cuda_reduction <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (spad + 2*system->N, spad + 3*system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2>>>
- (spad + 3*system->N, &((simulation_data *)data->d_simulation_data)->E_Ele, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
- rvec_spad = (rvec *) (spad + 4*system->N);
-
- //reduction on ext_press
- Cuda_reduction_rvec <<<BLOCKS_POW_2, BLOCK_SIZE, RVEC_SIZE * BLOCK_SIZE>>>
- (rvec_spad, rvec_spad + system->N, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction_rvec <<<1, BLOCKS_POW_2, RVEC_SIZE * BLOCKS_POW_2>>>
- (rvec_spad + system->N, &((simulation_data *)data->d_simulation_data)->ext_press, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
-
-
- /* This version of Compute_Total_Force computes forces from coefficients
- accumulated by all interaction functions. Saves enormous time & space! */
- void Compute_Total_Force( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists )
- {
- int i, pj;
- list *bonds = (*lists) + BONDS;
-
- for( i = 0; i < system->N; ++i )
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj )
- if( i < bonds->select.bond_list[pj].nbr ) {
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT)
- Add_dBond_to_Forces( i, pj, system, data, workspace, lists );
- else
- Add_dBond_to_Forces_NPT( i, pj, system, data, workspace, lists );
- }
- }
-
-
- void Validate_Lists( static_storage *workspace, list **lists, int step, int n,
- int Hmax, int Htop, int num_bonds, int num_hbonds )
- {
- int i, flag;
- list *bonds, *hbonds;
-
- bonds = *lists + BONDS;
- hbonds = *lists + HBONDS;
-
- /* far neighbors */
- if( Htop > Hmax * DANGER_ZONE ) {
- workspace->realloc.Htop = Htop;
- if( Htop > Hmax ) {
- fprintf( stderr,
- "step%d - ran out of space on H matrix: Htop=%d, max = %d",
- step, Htop, Hmax );
- exit(INSUFFICIENT_SPACE);
- }
- }
-
- /* bond list */
- flag = -1;
- workspace->realloc.num_bonds = num_bonds;
- for( i = 0; i < n-1; ++i )
- if( End_Index(i, bonds) >= Start_Index(i+1, bonds)-2 ) {
- workspace->realloc.bonds = 1;
- if( End_Index(i, bonds) > Start_Index(i+1, bonds) )
- flag = i;
- }
-
- if( flag > -1 ) {
- fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
- step, flag, End_Index(flag,bonds), Start_Index(flag+1,bonds) );
- exit(INSUFFICIENT_SPACE);
- }
-
- if( End_Index(i, bonds) >= bonds->num_intrs-2 ) {
- workspace->realloc.bonds = 1;
-
- if( End_Index(i, bonds) > bonds->num_intrs ) {
- fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d bond_end=%d\n",
- step, flag, End_Index(i,bonds), bonds->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
- }
-
-
- /* hbonds list */
- if( workspace->num_H > 0 ) {
- flag = -1;
- workspace->realloc.num_hbonds = num_hbonds;
- for( i = 0; i < workspace->num_H-1; ++i )
- if( Num_Entries(i, hbonds) >=
- (Start_Index(i+1, hbonds) - Start_Index(i, hbonds)) * DANGER_ZONE ) {
- workspace->realloc.hbonds = 1;
- if( End_Index(i, hbonds) > Start_Index(i+1, hbonds) )
- flag = i;
- }
-
- if( flag > -1 ) {
- fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
- step, flag, End_Index(flag,hbonds), Start_Index(flag+1,hbonds) );
- exit(INSUFFICIENT_SPACE);
- }
-
- if( Num_Entries(i,hbonds) >=
- (hbonds->num_intrs - Start_Index(i,hbonds)) * DANGER_ZONE ) {
- workspace->realloc.hbonds = 1;
-
- if( End_Index(i, hbonds) > hbonds->num_intrs ) {
- fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d hbondend=%d\n",
- step, flag, End_Index(i,hbonds), hbonds->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
- }
- }
- }
-
-
- void Cuda_Validate_Lists( reax_system *system, static_storage *workspace, list **lists, int step, int n,
- int num_bonds, int num_hbonds )
- {
- int i, flag;
- list *bonds, *hbonds, *thblist;
- int *bonds_start, *bonds_end;
- int *hbonds_start, *hbonds_end;
- int *mat_start, *mat_end;
- int max_sparse_entries = 0;
-
- bonds = *lists + BONDS;
- hbonds = *lists + HBONDS;
-
- bonds_start = (int *) calloc (bonds->n, INT_SIZE);
- bonds_end = (int *) calloc (bonds->n, INT_SIZE);
-
- hbonds_start = (int *) calloc (hbonds->n, INT_SIZE );
- hbonds_end = (int *) calloc (hbonds->n, INT_SIZE );
-
- mat_start = (int *) calloc (workspace->H.n, INT_SIZE );
- mat_end = (int *) calloc (workspace->H.n, INT_SIZE );
-
- copy_host_device (bonds_start, bonds->index, bonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
- copy_host_device (bonds_end, bonds->end_index, bonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
-
- copy_host_device (hbonds_start, hbonds->index, hbonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
- copy_host_device (hbonds_end, hbonds->end_index, hbonds->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
-
- copy_host_device (mat_start, workspace->H.start, workspace->H.n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
- copy_host_device (mat_end, workspace->H.end, workspace->H.n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
-
- /* Sparse Matrix entries */
-
-#ifdef __CUDA_TEST__
- /*
- workspace->realloc.Htop = 0;
- for (i = 0; i < workspace->H.n-1; i++) {
- if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i])){
- workspace->realloc.Htop = mat_end[i] - mat_start[i];
- }
- }
- */
-#endif
-
- flag = -1;
- workspace->realloc.Htop = 0;
- for ( i = 0; i < n-1; i ++){
-
- if( (mat_end[i] - mat_start[i]) >
- (system->max_sparse_matrix_entries * DANGER_ZONE )) {
- //fprintf (stderr, "step %d, Reached the water mark for sparse matrix for index: %d (%d %d) \n",
- // step, i, mat_start[i], mat_end[i]);
- if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i]))
- workspace->realloc.Htop = (mat_end[i] - mat_start[i]) ;
- }
-
- if ( (mat_end[i] > mat_start[i+1]) ){
- fprintf( stderr, "step%d-matcheck failed: i=%d end(i)=%d start(i+1)=%d\n",
- step, flag, mat_end[i], mat_start[i+1]);
- exit(INSUFFICIENT_SPACE);
- }
- }
-
- if( (mat_end[i] - mat_start[i]) > system->max_sparse_matrix_entries * DANGER_ZONE ) {
- if (workspace->realloc.Htop <= (mat_end[i] - mat_start[i]))
- workspace->realloc.Htop = (mat_end[i] - mat_start[i]) ;
- //fprintf (stderr, "step %d, Reached the water mark for sparse matrix for index %d (%d %d) -- %d \n",
- // step, i, mat_start[i], mat_end[i],
- // (int) (system->max_sparse_matrix_entries * DANGER_ZONE));
-
- if( mat_end[i] > system->N * system->max_sparse_matrix_entries ) {
- fprintf( stderr, "step%d-matchk failed: i=%d end(i)=%d mat_end=%d\n",
- step, flag, mat_end[i], system->N * system->max_sparse_matrix_entries);
- exit(INSUFFICIENT_SPACE);
- }
- }
-
-
- /* bond list */
-#ifdef __CUDA_TEST__
- //workspace->realloc.bonds = 1;
-#endif
- flag = -1;
- workspace->realloc.num_bonds = 0;
- for( i = 0; i < n-1; ++i ) {
- workspace->realloc.num_bonds += MAX((bonds_end [i] - bonds_start[i]) * 2, MIN_BONDS );
- if( bonds_end[i] >= bonds_start[i+1]-2 ) {
- workspace->realloc.bonds = 1;
- //fprintf (stderr, "step: %d, reached the water mark for bonds for atom: %d (%d %d) \n",
- // step, i, bonds_start [i], bonds_end[i]);
- if( bonds_end[i] > bonds_start[i+1] )
- flag = i;
- }
- }
-
- if( flag > -1 ) {
- fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d str(i+1)=%d\n",
- step, flag, bonds_end[flag], bonds_start[flag+1] );
- exit(INSUFFICIENT_SPACE);
- }
-
- workspace->realloc.num_bonds += MAX((bonds_end [i] - bonds_start[i]) * 2, MIN_BONDS );
- if( bonds_end[i] >= bonds->num_intrs-2 ) {
- workspace->realloc.bonds = 1;
- //fprintf (stderr, "step: %d, reached the water mark for bonds for atom: %d (%d %d) \n",
- // step, i, bonds_start [i], bonds_end[i]);
-
- if( bonds_end[i] > bonds->num_intrs ) {
- fprintf( stderr, "step%d-bondchk failed: i=%d end(i)=%d bond_end=%d\n",
- step, flag, bonds_end[i], bonds->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
- }
-
- //fprintf (stderr, "step:%d Total bonds: %d \n", step, workspace->realloc.num_bonds);
-
- /* hbonds list */
- if( workspace->num_H > 0 ) {
-#ifdef __CUDA_TEST__
- //workspace->realloc.hbonds = 1;
-#endif
- flag = -1;
- workspace->realloc.num_hbonds = 0;
- for( i = 0; i < workspace->num_H-1; ++i ) {
- workspace->realloc.num_hbonds += MAX( (hbonds_end[i] - hbonds_start[i]) * SAFE_HBONDS, MIN_HBONDS );
-
- if( (hbonds_end[i] - hbonds_start[i]) >=
- (hbonds_start[i+1] - hbonds_start[i]) * DANGER_ZONE ) {
- workspace->realloc.hbonds = 1;
- //fprintf (stderr, "step: %d, reached the water mark for hbonds for atom: %d (%d %d) \n",
- // step, i, hbonds_start [i], hbonds_end[i]);
- if( hbonds_end[i] > hbonds_start[i+1] )
- flag = i;
- }
- }
-
- if( flag > -1 ) {
- fprintf( stderr, "step%d-hbondchk failed: i=%d start(i)=%d,end(i)=%d str(i+1)=%d\n",
- step, flag, hbonds_start[(flag)],hbonds_end[(flag)], hbonds_start[(flag+1)] );
- exit(INSUFFICIENT_SPACE);
- }
-
- workspace->realloc.num_hbonds += MAX( (hbonds_end[i] - hbonds_start[i]) * SAFE_HBONDS, MIN_HBONDS );
- if( (hbonds_end[i] - hbonds_start[i]) >=
- (hbonds->num_intrs - hbonds_start[i]) * DANGER_ZONE ) {
- workspace->realloc.hbonds = 1;
- //fprintf (stderr, "step: %d, reached the water mark for hbonds for atom: %d (%d %d) \n",
- // step, i, hbonds_start [i], hbonds_end[i]);
-
- if( hbonds_end[i] > hbonds->num_intrs ) {
- fprintf( stderr, "step%d-hbondchk failed: i=%d end(i)=%d hbondend=%d\n",
- step, flag, hbonds_end[i], hbonds->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
- }
- }
-
- //fprintf (stderr, "step:%d Total Hbonds: %d \n", step, workspace->realloc.num_hbonds);
-
- free (bonds_start);
- free (bonds_end );
-
- free (hbonds_start );
- free (hbonds_end );
-
- free (mat_start );
- free (mat_end );
- }
-
- void Cuda_Threebody_List( reax_system *system, static_storage *workspace, list *thblist, int step )
- {
- int *thb_start, *thb_end;
- int i, flag;
-
- thb_start = (int *) calloc (thblist->n, INT_SIZE);
- thb_end = (int *) calloc (thblist->n, INT_SIZE );
-
- copy_host_device (thb_start, thblist->index, thblist->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
- copy_host_device (thb_end, thblist->end_index, thblist->n * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__ );
-
- /*three_body list*/
- flag = -1;
- workspace->realloc.num_3body = 0;
- for( i = 0; i < thblist->n-1; ++i ){
- if( (thb_end[i] - thb_start[i]) >= (thb_start[i+1] - thb_start[i])*DANGER_ZONE ) {
- workspace->realloc.thbody = 1;
- if( thb_end[i] > thb_end[i+1] || thb_end[i] > thblist->num_intrs ) {
- flag = i;
- break;
- }
- }
- }
-
- if( flag > -1 ) {
- //fprintf( stderr, "step%d-thbchk failed: i=%d end(i)=%d str(i+1)=%d\n",
- // step, flag, thb_end[flag], thb_start[flag+1] );
- fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
- step, flag-1, thb_start[flag-1], thb_end[flag-1], thblist->num_intrs );
- fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
- step, flag, thb_start[flag], thb_end[flag], thblist->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
-
- if( (thb_end[i]-thb_start[i]) >= (thblist->num_intrs - thb_start[i])*DANGER_ZONE ) {
- workspace->realloc.thbody = 1;
-
- if( thb_end[i] > thblist->num_intrs ) {
- fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
- step, i-1, thb_start[i-1], thb_end[i-1], thblist->num_intrs );
- fprintf( stderr, "step%d-thbchk failed: i=%d start(i)=%d end(i)=%d thb_end=%d\n",
- step, i, thb_start[i], thb_end[i], thblist->num_intrs );
- exit(INSUFFICIENT_SPACE);
- }
- }
-
- free (thb_start);
- free (thb_end);
- }
-
-
- void Init_Forces( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control ) {
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int Htop, btop_i, btop_j, num_bonds, num_hbonds;
- int ihb, jhb, ihb_top, jhb_top;
- int flag;
- real r_ij, r2, self_coef;
- real dr3gamij_1, dr3gamij_3, Tap;
- //real val, dif, base;
- real C12, C34, C56;
- real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- sparse_matrix *H;
- list *far_nbrs, *bonds, *hbonds;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- //LR_lookup_table *t;
- reax_atom *atom_i, *atom_j;
- bond_data *ibond, *jbond;
- bond_order_data *bo_ij, *bo_ji;
-
- far_nbrs = *lists + FAR_NBRS;
- bonds = *lists + BONDS;
- hbonds = *lists + HBONDS;
-
- H = &workspace->H;
- Htop = 0;
- num_bonds = 0;
- num_hbonds = 0;
- btop_i = btop_j = 0;
- p_boc1 = system->reaxprm.gp.l[0];
- p_boc2 = system->reaxprm.gp.l[1];
-
- for( i = 0; i < system->N; ++i ) {
- atom_i = &(system->atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, far_nbrs);
- end_i = End_Index(i, far_nbrs);
- H->start[i] = Htop;
- btop_i = End_Index( i, bonds );
- sbp_i = &(system->reaxprm.sbp[type_i]);
- ihb = ihb_top = -1;
- if( control->hb_cut > 0 && (ihb=sbp_i->p_hbond) == 1 )
- ihb_top = End_Index( workspace->hbond_index[i], hbonds );
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(system->atoms[j]);
-
- flag = 0;
- if((data->step-data->prev_steps) % control->reneighbor == 0) {
- if( nbr_pj->d <= control->r_cut)
- flag = 1;
- else flag = 0;
- }
- else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,&(system->box),
- nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
-
- if( flag ){
- type_j = system->atoms[j].type;
- r_ij = nbr_pj->d;
- sbp_j = &(system->reaxprm.sbp[type_j]);
- twbp = &(system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
- self_coef = (i == j) ? 0.5 : 1.0;
-
- /* H matrix entry */
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
- H->entries[Htop].j = j;
- H->entries[Htop].val = self_coef * Tap * EV_to_KCALpMOL / dr3gamij_3;
- ++Htop;
-
- /* hydrogen bond lists */
- if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
- nbr_pj->d <= control->hb_cut ) {
- // fprintf( stderr, "%d %d\n", atom1, atom2 );
- jhb = sbp_j->p_hbond;
- if( ihb == 1 && jhb == 2 ) {
- hbonds->select.hbond_list[ihb_top].nbr = j;
- hbonds->select.hbond_list[ihb_top].scl = 1;
- hbonds->select.hbond_list[ihb_top].ptr = nbr_pj;
- ++ihb_top;
- ++num_hbonds;
- }
- else if( ihb == 2 && jhb == 1 ) {
- jhb_top = End_Index( workspace->hbond_index[j], hbonds );
- hbonds->select.hbond_list[jhb_top].nbr = i;
- hbonds->select.hbond_list[jhb_top].scl = -1;
- hbonds->select.hbond_list[jhb_top].ptr = nbr_pj;
- Set_End_Index( workspace->hbond_index[j], jhb_top+1, hbonds );
- ++num_hbonds;
- }
- }
-
- /* uncorrected bond orders */
- if( far_nbrs->select.far_nbr_list[pj].d <= control->nbr_cut ) {
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
- if( BO >= control->bo_cut ) {
- num_bonds += 2;
- /****** bonds i-j and j-i ******/
- ibond = &( bonds->select.bond_list[btop_i] );
- btop_j = End_Index( j, bonds );
- jbond = &(bonds->select.bond_list[btop_j]);
-
- ibond->nbr = j;
- jbond->nbr = i;
- ibond->d = r_ij;
- jbond->d = r_ij;
- rvec_Copy( ibond->dvec, nbr_pj->dvec );
- rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
- ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
- ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
- ibond->dbond_index = btop_i;
- jbond->dbond_index = btop_i;
- ibond->sym_index = btop_j;
- jbond->sym_index = btop_i;
- ++btop_i;
- Set_End_Index( j, btop_j+1, bonds );
-
- bo_ij = &( ibond->bo_data );
- bo_ji = &( jbond->bo_data );
- bo_ji->BO = bo_ij->BO = BO;
- bo_ji->BO_s = bo_ij->BO_s = BO_s;
- bo_ji->BO_pi = bo_ij->BO_pi = BO_pi;
- bo_ji->BO_pi2 = bo_ij->BO_pi2 = BO_pi2;
-
- /* Bond Order page2-3, derivative of total bond order prime */
- Cln_BOp_s = twbp->p_bo2 * C12 / r2;
- Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
- Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
-
- /* Only dln_BOp_xx wrt. dr_i is stored here, note that
- dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
- rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi2,
- -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
- rvec_Scale(bo_ji->dln_BOp_s, -1., bo_ij->dln_BOp_s);
- rvec_Scale(bo_ji->dln_BOp_pi, -1., bo_ij->dln_BOp_pi );
- rvec_Scale(bo_ji->dln_BOp_pi2, -1., bo_ij->dln_BOp_pi2 );
-
- /* Only dBOp wrt. dr_i is stored here, note that
- dBOp/dr_i = -dBOp/dr_j and all others are 0 */
- rvec_Scale( bo_ij->dBOp,
- -(bo_ij->BO_s * Cln_BOp_s +
- bo_ij->BO_pi * Cln_BOp_pi +
- bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
- rvec_Scale( bo_ji->dBOp, -1., bo_ij->dBOp );
-
- rvec_Add( workspace->dDeltap_self[i], bo_ij->dBOp );
- rvec_Add( workspace->dDeltap_self[j], bo_ji->dBOp );
-
- bo_ij->BO_s -= control->bo_cut;
- bo_ij->BO -= control->bo_cut;
- bo_ji->BO_s -= control->bo_cut;
- bo_ji->BO -= control->bo_cut;
- workspace->total_bond_order[i] += bo_ij->BO; //currently total_BOp
- workspace->total_bond_order[j] += bo_ji->BO; //currently total_BOp
- bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
- bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
-
- /*fprintf( stderr, "%d %d %g %g %g\n",
- i+1, j+1, bo_ij->BO, bo_ij->BO_pi, bo_ij->BO_pi2 );*/
-
- /*fprintf( stderr, "Cln_BOp_s: %f, pbo2: %f, C12:%f\n",
- Cln_BOp_s, twbp->p_bo2, C12 );
- fprintf( stderr, "Cln_BOp_pi: %f, pbo4: %f, C34:%f\n",
- Cln_BOp_pi, twbp->p_bo4, C34 );
- fprintf( stderr, "Cln_BOp_pi2: %f, pbo6: %f, C56:%f\n",
- Cln_BOp_pi2, twbp->p_bo6, C56 );*/
- /*fprintf(stderr, "pbo1: %f, pbo2:%f\n", twbp->p_bo1, twbp->p_bo2);
- fprintf(stderr, "pbo3: %f, pbo4:%f\n", twbp->p_bo3, twbp->p_bo4);
- fprintf(stderr, "pbo5: %f, pbo6:%f\n", twbp->p_bo5, twbp->p_bo6);
- fprintf( stderr, "r_s: %f, r_p: %f, r_pp: %f\n",
- twbp->r_s, twbp->r_p, twbp->r_pp );
- fprintf( stderr, "C12: %g, C34:%g, C56:%g\n", C12, C34, C56 );*/
-
- /*fprintf( stderr, "\tfactors: %g %g %g\n",
- -(bo_ij->BO_s * Cln_BOp_s + bo_ij->BO_pi * Cln_BOp_pi +
- bo_ij->BO_pi2 * Cln_BOp_pp),
- -bo_ij->BO_pi * Cln_BOp_pi, -bo_ij->BO_pi2 * Cln_BOp_pi2 );*/
- /*fprintf( stderr, "dBOpi:\t[%g, %g, %g]\n",
- bo_ij->dBOp[0], bo_ij->dBOp[1], bo_ij->dBOp[2] );
- fprintf( stderr, "dBOpi:\t[%g, %g, %g]\n",
- bo_ij->dln_BOp_pi[0], bo_ij->dln_BOp_pi[1],
- bo_ij->dln_BOp_pi[2] );
- fprintf( stderr, "dBOpi2:\t[%g, %g, %g]\n\n",
- bo_ij->dln_BOp_pi2[0], bo_ij->dln_BOp_pi2[1],
- bo_ij->dln_BOp_pi2[2] );*/
-
- Set_End_Index( j, btop_j+1, bonds );
- }
- }
- }
- }
-
- H->entries[Htop].j = i;
- H->entries[Htop].val = system->reaxprm.sbp[type_i].eta;
- ++Htop;
-
- Set_End_Index( i, btop_i, bonds );
- if( ihb == 1 )
- Set_End_Index( workspace->hbond_index[i], ihb_top, hbonds );
- //fprintf( stderr, "%d bonds start: %d, end: %d\n",
- // i, Start_Index( i, bonds ), End_Index( i, bonds ) );
- }
-
- // mark the end of j list
- H->start[i] = Htop;
- /* validate lists - decide if reallocation is required! */
- Validate_Lists( workspace, lists,
- data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d: Htop = %d, num_bonds = %d, num_hbonds = %d\n",
- data->step, Htop, num_bonds, num_hbonds );
-
-#endif
- }
-
-
- GLOBAL void Estimate_Sparse_Matrix_Entries ( reax_atom *atoms, control_params *control,
- simulation_data *data, simulation_box *box, list far_nbrs, int N, int *indices ) {
-
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int Htop;
- int flag;
- far_neighbor_data *nbr_pj;
- reax_atom *atom_i, *atom_j;
-
- int temp;
-
- Htop = 0;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- atom_i = &(atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, &far_nbrs);
- end_i = End_Index(i, &far_nbrs);
- indices[i] = Htop;
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(atoms[j]);
-
- //CHANGE ORIGINAL
- //if (i < j) continue;
- //CHANGE ORIGINAL
-
- flag = 0;
- if((data->step-data->prev_steps) % control->reneighbor == 0) {
- if( nbr_pj->d <= control->r_cut)
- flag = 1;
- else flag = 0;
- }
- else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec)) <=
- SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
-
- if( flag ){
- ++Htop;
- }
- }
-
- ++Htop;
-
- // mark the end of j list
- indices[i] = Htop;
- }
-
-
-
-
- GLOBAL void Init_Forces( reax_atom *atoms, global_parameters g_params, control_params *control,
- single_body_parameters *sbp, two_body_parameters *tbp,
- simulation_data *data, simulation_box *box, static_storage workspace,
- list far_nbrs, list bonds, list hbonds,
- int N, int max_sparse_entries, int num_atom_types )
- {
-
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int Htop, btop_i, btop_j, num_bonds, num_hbonds;
- int ihb, jhb, ihb_top, jhb_top;
- int flag;
- real r_ij, r2, self_coef;
- real dr3gamij_1, dr3gamij_3, Tap;
- //real val, dif, base;
- real C12, C34, C56;
- real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- sparse_matrix *H;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- //LR_lookup_table *t;
- reax_atom *atom_i, *atom_j;
- bond_data *ibond, *jbond;
- bond_order_data *bo_ij, *bo_ji;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- H = &( workspace.H );
- //CHANGE ORIGINAL
- //Htop = 0;
- Htop = i * max_sparse_entries;
- //CHANGE ORIGINAL
- num_bonds = 0;
- num_hbonds = 0;
- btop_i = btop_j = 0;
- p_boc1 = g_params.l[0];
- p_boc2 = g_params.l[1];
-
- //for( i = 0; i < system->N; ++i )
- atom_i = &(atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, &far_nbrs);
- end_i = End_Index(i, &far_nbrs);
-
- H->start[i] = Htop;
- H->end[i] = Htop;
-
- btop_i = End_Index( i, &bonds );
- sbp_i = &(sbp[type_i]);
- ihb = ihb_top = -1;
-
- ihb = sbp_i->p_hbond;
-
- if( control->hb_cut > 0 && (ihb==1 || ihb == 2))
- ihb_top = End_Index( workspace.hbond_index[i], &hbonds );
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(atoms[j]);
-
- flag = 0;
- if((data->step-data->prev_steps) % control->reneighbor == 0) {
- if( nbr_pj->d <= control->r_cut)
- flag = 1;
- else flag = 0;
- }
- else if (i > j) {
- if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
- } else if (i < j) {
- if((nbr_pj->d=Sq_Distance_on_T3(atom_j->x,atom_i->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
- }
-
- if( flag ){
-
- type_j = atoms[j].type;
- r_ij = nbr_pj->d;
- sbp_j = &(sbp[type_j]);
- twbp = &(tbp[ index_tbp (type_i,type_j, num_atom_types) ]);
- self_coef = (i == j) ? 0.5 : 1.0;
-
- /* H matrix entry */
-
- //CHANGE ORIGINAL
- //if (i > j) {
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
- H->entries[Htop].j = j;
- H->entries[Htop].val = self_coef * Tap * EV_to_KCALpMOL / dr3gamij_3;
-
- ++Htop;
- //}
- //CHANGE ORIGINAL
-
- /* hydrogen bond lists */
- if( control->hb_cut > 0 && (ihb==1 || ihb == 2) &&
- nbr_pj->d <= control->hb_cut ) {
- // fprintf( stderr, "%d %d\n", atom1, atom2 );
- jhb = sbp_j->p_hbond;
-
- if (ihb == 1 && jhb == 2) {
- if (i > j) {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = 1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
-
- //Auxilary data structures
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- } else {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = -1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
-
- //Auxilary data structures
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- }
- } else if (ihb == 2 && jhb == 1) {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = 1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
- //TODO
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- }
- }
-
- /* uncorrected bond orders */
- if( far_nbrs.select.far_nbr_list[pj].d <= control->nbr_cut ) {
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
-
- if( BO >= control->bo_cut ) {
- //CHANGE ORIGINAL
- num_bonds += 1;
- //CHANGE ORIGINAL
-
- /****** bonds i-j and j-i ******/
-
- /* Bond Order page2-3, derivative of total bond order prime */
- Cln_BOp_s = twbp->p_bo2 * C12 / r2;
- Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
- Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
-
-
- if (i > j)
- {
- ibond = &( bonds.select.bond_list[btop_i] );
- ibond->nbr = j;
- ibond->d = r_ij;
- rvec_Copy( ibond->dvec, nbr_pj->dvec );
- ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
-
- //ibond->dbond_index = btop_i;
- //ibond->sym_index = btop_j;
- ++btop_i;
-
- bo_ij = &( ibond->bo_data );
- bo_ij->BO = BO;
- bo_ij->BO_s = BO_s;
- bo_ij->BO_pi = BO_pi;
- bo_ij->BO_pi2 = BO_pi2;
-
- //Auxilary data structures
- ibond->scratch = 0;
- ibond->CdDelta_ij = 0;
- rvec_MakeZero (ibond->f);
-
- ibond->l = -1;
- ibond->CdDelta_jk = 0;
- ibond->Cdbo_kl = 0;
- rvec_MakeZero (ibond->i_f);
- rvec_MakeZero (ibond->k_f);
-
- rvec_MakeZero (ibond->h_f);
-
- rvec_MakeZero (ibond->t_f);
-
- // Only dln_BOp_xx wrt. dr_i is stored here, note that
- // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
- rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi2,
- -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
-
- // Only dBOp wrt. dr_i is stored here, note that
- // dBOp/dr_i = -dBOp/dr_j and all others are 0
- rvec_Scale( bo_ij->dBOp,
- -(bo_ij->BO_s * Cln_BOp_s +
- bo_ij->BO_pi * Cln_BOp_pi +
- bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
-
- rvec_Add( workspace.dDeltap_self[i], bo_ij->dBOp );
-
- bo_ij->BO_s -= control->bo_cut;
- bo_ij->BO -= control->bo_cut;
- workspace.total_bond_order[i] += bo_ij->BO; //currently total_BOp
-
- bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
-
-
- } else if ( i < j )
- {
- rvec dln_BOp_s, dln_BOp_pi, dln_BOp_pi2;
- rvec dBOp;
-
- btop_j = btop_i;
-
- jbond = &(bonds.select.bond_list[btop_j]);
- jbond->nbr = j;
- jbond->d = r_ij;
- rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
- ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
-
- btop_i ++;
- //jbond->dbond_index = btop_i;
- //jbond->sym_index = btop_i;
-
- bo_ji = &( jbond->bo_data );
- bo_ji->BO = BO;
- bo_ji->BO_s = BO_s;
- bo_ji->BO_pi = BO_pi;
- bo_ji->BO_pi2 = BO_pi2;
-
- //Auxilary data structures
- jbond->scratch = 0;
- jbond->CdDelta_ij = 0;
- rvec_MakeZero (jbond->f);
-
- jbond->l = -1;
- jbond->CdDelta_jk = 0;
- jbond->Cdbo_kl = 0;
- rvec_MakeZero (jbond->i_f);
- rvec_MakeZero (jbond->k_f);
-
- rvec_MakeZero (jbond->h_f);
-
- rvec_MakeZero (jbond->t_f);
-
- // Only dln_BOp_xx wrt. dr_i is stored here, note that
- // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
- rvec_Scale(dln_BOp_s,-BO_s*Cln_BOp_s,nbr_pj->dvec);
- rvec_Scale(dln_BOp_pi,-BO_pi*Cln_BOp_pi,nbr_pj->dvec);
- rvec_Scale(dln_BOp_pi2,
- -BO_pi2*Cln_BOp_pi2,nbr_pj->dvec);
-
- rvec_Scale(bo_ji->dln_BOp_s, -1., dln_BOp_s);
- rvec_Scale(bo_ji->dln_BOp_pi, -1., dln_BOp_pi );
- rvec_Scale(bo_ji->dln_BOp_pi2, -1., dln_BOp_pi2 );
-
- // Only dBOp wrt. dr_i is stored here, note that
- // dBOp/dr_i = -dBOp/dr_j and all others are 0
- rvec_Scale( dBOp,
- -(BO_s * Cln_BOp_s +
- BO_pi * Cln_BOp_pi +
- BO_pi2 * Cln_BOp_pi2), nbr_pj->dvec );
- rvec_Scale( bo_ji->dBOp, -1., dBOp );
-
- rvec_Add( workspace.dDeltap_self[i] , bo_ji->dBOp );
-
- bo_ji->BO_s -= control->bo_cut;
- bo_ji->BO -= control->bo_cut;
- workspace.total_bond_order[i] += bo_ji->BO; //currently total_BOp
-
- bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
-
- }
- }
- }
- }
- }
-
- H->entries[Htop].j = i;
- H->entries[Htop].val = sbp[type_i].eta;
- ++Htop;
-
- H->end[i] = Htop;
-
- Set_End_Index( i, btop_i, &bonds );
- if( ihb == 1 || ihb == 2)
- Set_End_Index( workspace.hbond_index[i], ihb_top, &hbonds );
-
- //fprintf( stderr, "%d bonds start: %d, end: %d\n",
- // i, Start_Index( i, bonds ), End_Index( i, bonds ) );
- //}
-
- // mark the end of j list
- //H->start[i] = Htop;
- /* validate lists - decide if reallocation is required! */
- //Validate_Lists( workspace, lists,
- // data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
-}
-
-GLOBAL void Init_Forces_Tab ( reax_atom *atoms, global_parameters g_params, control_params *control,
- single_body_parameters *sbp, two_body_parameters *tbp,
- simulation_data *data, simulation_box *box, static_storage workspace,
- list far_nbrs, list bonds, list hbonds,
- int N, int max_sparse_entries, int num_atom_types,
- LR_lookup_table *d_LR)
-{
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int Htop, btop_i, btop_j, num_bonds, num_hbonds;
- int tmin, tmax, r;
- int ihb, jhb, ihb_top, jhb_top;
- int flag;
- real r_ij, r2, self_coef;
- real val, dif, base;
- real C12, C34, C56;
- real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- sparse_matrix *H;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- LR_lookup_table *t;
- reax_atom *atom_i, *atom_j;
- bond_data *ibond, *jbond;
- bond_order_data *bo_ij, *bo_ji;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- H = &(workspace.H);
- //CHANGE ORIGINAL
- Htop = i * max_sparse_entries;
- //CHANGE ORIGINAL
- num_bonds = 0;
- num_hbonds = 0;
- btop_i = btop_j = 0;
- p_boc1 = g_params.l[0];
- p_boc2 = g_params.l[1];
-
- //for( i = 0; i < system->N; ++i )
- atom_i = &(atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, &far_nbrs);
- end_i = End_Index(i, &far_nbrs);
- H->start[i] = Htop;
- H->end[i] = Htop;
- btop_i = End_Index( i, &bonds );
- sbp_i = &(sbp[type_i]);
- ihb = ihb_top = -1;
-
- ihb = sbp_i->p_hbond;
-
- if( control->hb_cut > 0 && (ihb==1 || ihb == 2))
- ihb_top = End_Index( workspace.hbond_index[i], &hbonds );
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(atoms[j]);
-
- flag = 0;
- if((data->step-data->prev_steps) % control->reneighbor == 0) {
- if(nbr_pj->d <= control->r_cut)
- flag = 1;
- else flag = 0;
- }
- else if (i > j) {
- if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
- }
- else if ( i < j) {
- if((nbr_pj->d=Sq_Distance_on_T3(atom_j->x,atom_i->x,box,nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
- }
-
- if( flag ){
- type_j = atoms[j].type;
- r_ij = nbr_pj->d;
- sbp_j = &(sbp[type_j]);
- twbp = &(tbp[ index_tbp (type_i,type_j,num_atom_types) ]);
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( d_LR[ index_lr (tmin, tmax, num_atom_types) ]);
-
- /* cubic spline interpolation */
- //CHANGE ORIGINAL
- //if (i > j) {
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
- val = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
- t->ele[r].a;
- val *= EV_to_KCALpMOL / C_ele;
-
- H->entries[Htop].j = j;
- H->entries[Htop].val = self_coef * val;
- //H->j [Htop] = j;
- //H->val [Htop] = self_coef * val;
- ++Htop;
- //}
- //CHANGE ORIGINAL
-
- /* hydrogen bond lists */
- if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
- nbr_pj->d <= control->hb_cut ) {
- // fprintf( stderr, "%d %d\n", atom1, atom2 );
- jhb = sbp_j->p_hbond;
-
- if ( ihb == 1 && jhb == 2 ) {
- if (i > j) {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = 1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
-
- //Auxilary data structures
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- } else {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = -1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
-
- //Auxilary data structures
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- }
- } else if (ihb == 2 && jhb == 1) {
- hbonds.select.hbond_list[ihb_top].nbr = j;
- hbonds.select.hbond_list[ihb_top].scl = 1;
- hbonds.select.hbond_list[ihb_top].ptr = nbr_pj;
-
- //Auxilary data structures
- rvec_MakeZero (hbonds.select.hbond_list[ihb_top].h_f);
- hbonds.select.hbond_list[ihb_top].sym_index= -1;
- ++ihb_top;
- ++num_hbonds;
- }
- }
-
- /* uncorrected bond orders */
- if( far_nbrs.select.far_nbr_list[pj].d <= control->nbr_cut ) {
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
- if( BO >= control->bo_cut ) {
-
- //CHANGE ORIGINAL
- num_bonds += 1;
- //CHANGE ORIGINAL
-
- /****** bonds i-j and j-i ******/
- if ( i > j )
- {
- ibond = &( bonds.select.bond_list[btop_i] );
- ibond->nbr = j;
- ibond->d = r_ij;
-
- rvec_Copy( ibond->dvec, nbr_pj->dvec );
- ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
-
- //ibond->dbond_index = btop_i;
- //ibond->sym_index = btop_j;
-
- ++btop_i;
-
- bo_ij = &( ibond->bo_data );
- bo_ij->BO = BO;
- bo_ij->BO_s = BO_s;
- bo_ij->BO_pi = BO_pi;
- bo_ij->BO_pi2 = BO_pi2;
-
- //Auxilary data strucutres to resolve dependencies
- ibond->scratch = 0;
- ibond->CdDelta_ij = 0;
- rvec_MakeZero (ibond->f);
-
- ibond->l = -1;
- ibond->CdDelta_jk = 0;
- ibond->Cdbo_kl = 0;
- rvec_MakeZero (ibond->i_f);
- rvec_MakeZero (ibond->k_f);
-
- rvec_MakeZero (ibond->h_f);
-
- rvec_MakeZero (ibond->t_f);
-
- /* Bond Order page2-3, derivative of total bond order prime */
- Cln_BOp_s = twbp->p_bo2 * C12 / r2;
- Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
- Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
-
- /* Only dln_BOp_xx wrt. dr_i is stored here, note that
- dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
- rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi2,
- -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
-
- /* Only dBOp wrt. dr_i is stored here, note that
- dBOp/dr_i = -dBOp/dr_j and all others are 0 */
- rvec_Scale( bo_ij->dBOp,
- -(bo_ij->BO_s * Cln_BOp_s +
- bo_ij->BO_pi * Cln_BOp_pi +
- bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
-
- rvec_Add( workspace.dDeltap_self[i], bo_ij->dBOp );
-
- bo_ij->BO_s -= control->bo_cut;
- bo_ij->BO -= control->bo_cut;
-
- workspace.total_bond_order[i] += bo_ij->BO; //currently total_BOp
-
- bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
- }
- else {
- rvec dln_BOp_s, dln_BOp_pi, dln_BOp_pi2;
- rvec dBOp;
-
- btop_j = btop_i;
-
- jbond = &( bonds.select.bond_list[btop_j] );
- jbond->nbr = j;
- jbond->d = r_ij;
-
- rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
- ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
-
- //jbond->dbond_index = btop_i;
- //jbond->sym_index = btop_i;
-
- ++btop_i;
-
- bo_ji = &( jbond->bo_data );
-
- bo_ji->BO = BO;
- bo_ji->BO_s = BO_s;
- bo_ji->BO_pi = BO_pi;
- bo_ji->BO_pi2 = BO_pi2;
-
- // Auxilary data structures to resolve dependencies
- jbond->scratch = 0;
- jbond->CdDelta_ij = 0;
- rvec_MakeZero (jbond->f);
-
- jbond->l = -1;
- jbond->CdDelta_jk = 0;
- jbond->Cdbo_kl = 0;
- rvec_MakeZero (jbond->i_f);
- rvec_MakeZero (jbond->k_f);
-
- rvec_MakeZero (jbond->h_f);
-
- rvec_MakeZero (jbond->t_f);
-
- // Bond Order page2-3, derivative of total bond order prime
- Cln_BOp_s = twbp->p_bo2 * C12 / r2;
- Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
- Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
-
- // Only dln_BOp_xx wrt. dr_i is stored here, note that
- // dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0
-
- rvec_Scale(dln_BOp_s,-BO_s*Cln_BOp_s,nbr_pj->dvec);
- rvec_Scale(dln_BOp_pi,-BO_pi*Cln_BOp_pi,nbr_pj->dvec);
- rvec_Scale(dln_BOp_pi2, -BO_pi2*Cln_BOp_pi2,nbr_pj->dvec);
-
- rvec_Scale(bo_ji->dln_BOp_s, -1., dln_BOp_s);
- rvec_Scale(bo_ji->dln_BOp_pi, -1., dln_BOp_pi );
- rvec_Scale(bo_ji->dln_BOp_pi2, -1., dln_BOp_pi2 );
-
- // Only dBOp wrt. dr_i is stored here, note that
- // dBOp/dr_i = -dBOp/dr_j and all others are 0
- //CHANGE ORIGINAL
- rvec_Scale( dBOp,
- -(BO_s * Cln_BOp_s +
- BO_pi * Cln_BOp_pi +
- BO_pi2 * Cln_BOp_pi2), nbr_pj->dvec);
- rvec_Scale( bo_ji->dBOp, -1., dBOp);
- //CHANGE ORIGINAL
-
- rvec_Add( workspace.dDeltap_self[i], bo_ji->dBOp );
-
- bo_ji->BO_s -= control->bo_cut;
- bo_ji->BO -= control->bo_cut;
-
- workspace.total_bond_order[i] += bo_ji->BO; //currently total_BOp
-
- bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
- }
- }
- }
- }
- }
-
- H->entries[Htop].j = i;
- H->entries[Htop].val = sbp[type_i].eta;
-
- //H->j [Htop] = i;
- //H->val [Htop] = sbp[type_i].eta;
-
- ++Htop;
-
- H->end[i] = Htop;
- Set_End_Index( i, btop_i, &bonds );
- if( ihb == 1 || ihb == 2)
- Set_End_Index( workspace.hbond_index[i], ihb_top, &hbonds );
-}
-
-GLOBAL void fix_sym_dbond_indices (list pbonds, int N)
-{
- int i, nbr;
- bond_data *ibond, *jbond;
- int atom_j;
-
- list *bonds = &pbonds;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
- {
- ibond = &( bonds->select.bond_list [j] );
- nbr = ibond->nbr;
-
- for (int k = Start_Index (nbr, bonds); k < End_Index (nbr, bonds); k ++)
- {
- jbond = &( bonds->select.bond_list[ k ] );
- atom_j = jbond->nbr;
-
- if ( (atom_j == i) )
- {
- if (i > nbr) {
- ibond->dbond_index = j;
- jbond->dbond_index = j;
-
- ibond->sym_index = k;
- jbond->sym_index = j;
- }
- }
- }
- }
-}
-
-
-GLOBAL void fix_sym_hbond_indices (static_storage p_workspace, list hbonds, int N)
-{
- static_storage *workspace = &p_workspace;
- hbond_data *ihbond, *jhbond;
- int nbr;
-
- //int i = (blockIdx.x * blockDim.x + threadIdx.x) >> 4;
- int i = (blockIdx.x);
- int start = Start_Index (workspace->hbond_index[i], &hbonds);
- int end = End_Index (workspace->hbond_index[i], &hbonds);
- //int j = start + threadIdx.x;
- //int j = start + (threadIdx.x % 16);
-
- //for (int j = Start_Index (workspace->hbond_index[i], &hbonds);
- // j < End_Index (workspace->hbond_index[i], &hbonds); j++)
- int j = start + threadIdx.x;
- while (j < end)
- //for (int j = start; j < end; j++)
- {
- ihbond = &( hbonds.select.hbond_list [j] );
- nbr = ihbond->nbr;
-
- int nbrstart = Start_Index (workspace->hbond_index[nbr], &hbonds);
- int nbrend = End_Index (workspace->hbond_index[nbr], &hbonds);
-
- for (int k = nbrstart; k < nbrend; k++)
- //k = nbrstart + threadIdx.x;
- //while (k < nbrend)
- {
- jhbond = &( hbonds.select.hbond_list [k] );
-
- if (jhbond->nbr == i){
- ihbond->sym_index = k;
- jhbond->sym_index = j;
- break;
- }
-
- //k += blockDim.x;
- }
-
- j += 32;
- }
-}
-
-GLOBAL void New_fix_sym_hbond_indices (static_storage p_workspace, list hbonds, int N )
-{
-
- static_storage *workspace = &p_workspace;
- hbond_data *ihbond, *jhbond;
-
- int __THREADS_PER_ATOM__ = HBONDS_SYM_THREADS_PER_ATOM;
- int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
- int warp_id = thread_id / __THREADS_PER_ATOM__;
- int lane_id = thread_id & (__THREADS_PER_ATOM__ - 1);
- int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
-
- if (warp_id >= N) return;
-
- int i = warp_id;
- int nbr;
- int k;
- int start = Start_Index (workspace->hbond_index[i], &hbonds);
- int end = End_Index (workspace->hbond_index[i], &hbonds);
- int j = start + lane_id;
- //for (int j = start; j < end; j++)
- while (j < end)
- {
- ihbond = &( hbonds.select.hbond_list [j] );
- nbr = ihbond->nbr;
-
- int nbrstart = Start_Index (workspace->hbond_index[nbr], &hbonds);
- int nbrend = End_Index (workspace->hbond_index[nbr], &hbonds);
-
- //k = nbrstart + lane_id;
- //if (lane_id == 0) found [my_bucket] = 0;
- //while (k < nbrend)
- for (k = nbrstart; k < nbrend; k++)
- {
- jhbond = &( hbonds.select.hbond_list [k] );
-
- if (jhbond->nbr == i){
- ihbond->sym_index = k;
- jhbond->sym_index = j;
- break;
- }
- }
-
- j += __THREADS_PER_ATOM__;
- }
-}
-
-
-void Init_Forces_Tab( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control ) {
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int Htop, btop_i, btop_j, num_bonds, num_hbonds;
- int tmin, tmax, r;
- int ihb, jhb, ihb_top, jhb_top;
- int flag;
- real r_ij, r2, self_coef;
- real val, dif, base;
- real C12, C34, C56;
- real Cln_BOp_s, Cln_BOp_pi, Cln_BOp_pi2;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- sparse_matrix *H;
- list *far_nbrs, *bonds, *hbonds;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- LR_lookup_table *t;
- reax_atom *atom_i, *atom_j;
- bond_data *ibond, *jbond;
- bond_order_data *bo_ij, *bo_ji;
-
- far_nbrs = *lists + FAR_NBRS;
- bonds = *lists + BONDS;
- hbonds = *lists + HBONDS;
-
- H = &workspace->H;
- Htop = 0;
- num_bonds = 0;
- num_hbonds = 0;
- btop_i = btop_j = 0;
- p_boc1 = system->reaxprm.gp.l[0];
- p_boc2 = system->reaxprm.gp.l[1];
-
- for( i = 0; i < system->N; ++i ) {
- atom_i = &(system->atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, far_nbrs);
- end_i = End_Index(i, far_nbrs);
- H->start[i] = Htop;
- btop_i = End_Index( i, bonds );
- sbp_i = &(system->reaxprm.sbp[type_i]);
- ihb = ihb_top = -1;
- if( control->hb_cut > 0 && (ihb=sbp_i->p_hbond) == 1 )
- ihb_top = End_Index( workspace->hbond_index[i], hbonds );
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(system->atoms[j]);
-
- flag = 0;
- if((data->step-data->prev_steps) % control->reneighbor == 0) {
- if(nbr_pj->d <= control->r_cut)
- flag = 1;
- else flag = 0;
- }
- else if((nbr_pj->d=Sq_Distance_on_T3(atom_i->x,atom_j->x,&(system->box),
- nbr_pj->dvec))<=SQR(control->r_cut)){
- nbr_pj->d = sqrt(nbr_pj->d);
- flag = 1;
- }
-
- if( flag ){
- type_j = system->atoms[j].type;
- r_ij = nbr_pj->d;
- sbp_j = &(system->reaxprm.sbp[type_j]);
- twbp = &(system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( LR[ index_lr (tmin,tmax,system->reaxprm.num_atom_types) ] );
-
- /* cubic spline interpolation */
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
- val = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
- t->ele[r].a;
- val *= EV_to_KCALpMOL / C_ele;
-
- H->entries[Htop].j = j;
- H->entries[Htop].val = self_coef * val;
- ++Htop;
-
- /* hydrogen bond lists */
- if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
- nbr_pj->d <= control->hb_cut ) {
- // fprintf( stderr, "%d %d\n", atom1, atom2 );
- jhb = sbp_j->p_hbond;
- if( ihb == 1 && jhb == 2 ) {
- hbonds->select.hbond_list[ihb_top].nbr = j;
- hbonds->select.hbond_list[ihb_top].scl = 1;
- hbonds->select.hbond_list[ihb_top].ptr = nbr_pj;
- ++ihb_top;
- ++num_hbonds;
- }
- else if( ihb == 2 && jhb == 1 ) {
- jhb_top = End_Index( workspace->hbond_index[j], hbonds );
- hbonds->select.hbond_list[jhb_top].nbr = i;
- hbonds->select.hbond_list[jhb_top].scl = -1;
- hbonds->select.hbond_list[jhb_top].ptr = nbr_pj;
- Set_End_Index( workspace->hbond_index[j], jhb_top+1, hbonds );
- ++num_hbonds;
- }
- }
-
- /* uncorrected bond orders */
- if( far_nbrs->select.far_nbr_list[pj].d <= control->nbr_cut ) {
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
- if( BO >= control->bo_cut ) {
- num_bonds += 2;
- /****** bonds i-j and j-i ******/
- ibond = &( bonds->select.bond_list[btop_i] );
- btop_j = End_Index( j, bonds );
- jbond = &(bonds->select.bond_list[btop_j]);
-
- ibond->nbr = j;
- jbond->nbr = i;
- ibond->d = r_ij;
- jbond->d = r_ij;
- rvec_Copy( ibond->dvec, nbr_pj->dvec );
- rvec_Scale( jbond->dvec, -1, nbr_pj->dvec );
- ivec_Copy( ibond->rel_box, nbr_pj->rel_box );
- ivec_Scale( jbond->rel_box, -1, nbr_pj->rel_box );
- ibond->dbond_index = btop_i;
- jbond->dbond_index = btop_i;
- ibond->sym_index = btop_j;
- jbond->sym_index = btop_i;
- ++btop_i;
- Set_End_Index( j, btop_j+1, bonds );
-
- bo_ij = &( ibond->bo_data );
- bo_ji = &( jbond->bo_data );
- bo_ji->BO = bo_ij->BO = BO;
- bo_ji->BO_s = bo_ij->BO_s = BO_s;
- bo_ji->BO_pi = bo_ij->BO_pi = BO_pi;
- bo_ji->BO_pi2 = bo_ij->BO_pi2 = BO_pi2;
-
- /* Bond Order page2-3, derivative of total bond order prime */
- Cln_BOp_s = twbp->p_bo2 * C12 / r2;
- Cln_BOp_pi = twbp->p_bo4 * C34 / r2;
- Cln_BOp_pi2 = twbp->p_bo6 * C56 / r2;
-
- /* Only dln_BOp_xx wrt. dr_i is stored here, note that
- dln_BOp_xx/dr_i = -dln_BOp_xx/dr_j and all others are 0 */
- rvec_Scale(bo_ij->dln_BOp_s,-bo_ij->BO_s*Cln_BOp_s,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi,-bo_ij->BO_pi*Cln_BOp_pi,ibond->dvec);
- rvec_Scale(bo_ij->dln_BOp_pi2,
- -bo_ij->BO_pi2*Cln_BOp_pi2,ibond->dvec);
- rvec_Scale(bo_ji->dln_BOp_s, -1., bo_ij->dln_BOp_s);
- rvec_Scale(bo_ji->dln_BOp_pi, -1., bo_ij->dln_BOp_pi );
- rvec_Scale(bo_ji->dln_BOp_pi2, -1., bo_ij->dln_BOp_pi2 );
-
- /* Only dBOp wrt. dr_i is stored here, note that
- dBOp/dr_i = -dBOp/dr_j and all others are 0 */
- rvec_Scale( bo_ij->dBOp,
- -(bo_ij->BO_s * Cln_BOp_s +
- bo_ij->BO_pi * Cln_BOp_pi +
- bo_ij->BO_pi2 * Cln_BOp_pi2), ibond->dvec );
- rvec_Scale( bo_ji->dBOp, -1., bo_ij->dBOp );
-
- rvec_Add( workspace->dDeltap_self[i], bo_ij->dBOp );
- rvec_Add( workspace->dDeltap_self[j], bo_ji->dBOp );
-
- bo_ij->BO_s -= control->bo_cut;
- bo_ij->BO -= control->bo_cut;
- bo_ji->BO_s -= control->bo_cut;
- bo_ji->BO -= control->bo_cut;
- workspace->total_bond_order[i] += bo_ij->BO; //currently total_BOp
- workspace->total_bond_order[j] += bo_ji->BO; //currently total_BOp
- bo_ij->Cdbo = bo_ij->Cdbopi = bo_ij->Cdbopi2 = 0.0;
- bo_ji->Cdbo = bo_ji->Cdbopi = bo_ji->Cdbopi2 = 0.0;
-
- Set_End_Index( j, btop_j+1, bonds );
- }
- }
- }
- }
-
- H->entries[Htop].j = i;
- H->entries[Htop].val = system->reaxprm.sbp[type_i].eta;
- ++Htop;
-
- Set_End_Index( i, btop_i, bonds );
- if( ihb == 1 )
- Set_End_Index( workspace->hbond_index[i], ihb_top, hbonds );
- }
-
- // mark the end of j list
- H->start[i] = Htop;
- /* validate lists - decide if reallocation is required! */
- Validate_Lists( workspace, lists,
- data->step, system->N, H->m, Htop, num_bonds, num_hbonds );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d: Htop = %d, num_bonds = %d, num_hbonds = %d\n",
- data->step, Htop, num_bonds, num_hbonds );
- //Print_Bonds( system, bonds, "sbonds.out" );
- //Print_Bond_List2( system, bonds, "sbonds.out" );
- //Print_Sparse_Matrix2( H, "H.out" );
-#endif
-}
-
-void Estimate_Storage_Sizes( reax_system *system, control_params *control,
- list **lists, int *Htop, int *hb_top,
- int *bond_top, int *num_3body ) {
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int ihb, jhb;
- real r_ij, r2;
- real C12, C34, C56;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- list *far_nbrs;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- reax_atom *atom_i, *atom_j;
-
- far_nbrs = *lists + FAR_NBRS;
- p_boc1 = system->reaxprm.gp.l[0];
- p_boc2 = system->reaxprm.gp.l[1];
-
- for( i = 0; i < system->N; ++i ) {
- atom_i = &(system->atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, far_nbrs);
- end_i = End_Index(i, far_nbrs);
- sbp_i = &(system->reaxprm.sbp[type_i]);
- ihb = sbp_i->p_hbond;
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &(system->atoms[j]);
- type_j = atom_j->type;
- sbp_j = &(system->reaxprm.sbp[type_j]);
- twbp = &(system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ]);
-
- if( nbr_pj->d <= control->r_cut ) {
- ++(*Htop);
-
- /* hydrogen bond lists */
- if( control->hb_cut > 0.1 && (ihb==1 || ihb==2) &&
- nbr_pj->d <= control->hb_cut ) {
- jhb = sbp_j->p_hbond;
- if( ihb == 1 && jhb == 2 )
- ++hb_top[i];
- else if( ihb == 2 && jhb == 1 )
- ++hb_top[j];
- }
-
- /* uncorrected bond orders */
- if( nbr_pj->d <= control->nbr_cut ) {
- r_ij = nbr_pj->d;
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
- if( BO >= control->bo_cut ) {
- ++bond_top[i];
- ++bond_top[j];
- }
- }
- }
- }
- }
-
- *Htop += system->N;
- *Htop *= SAFE_ZONE;
-
- for( i = 0; i < system->N; ++i ) {
- hb_top[i] = MAX( hb_top[i] * SAFE_HBONDS, MIN_HBONDS );
- *num_3body += SQR(bond_top[i]);
- bond_top[i] = MAX( bond_top[i] * 2, MIN_BONDS );
- }
- *num_3body *= SAFE_ZONE;
-}
-
-void Cuda_Estimate_Storage_Sizes (reax_system *system, control_params *control, int *output)
-{
- int *Htop, *num_3body, input_size;
- int *hb_top, *bond_top;
- int *input = (int *) scratch;
- int max_3body = 0;
-
- Htop = 0;
- num_3body = 0;
- input_size = INT_SIZE * (2 * system->N + 1 + 1);
-
- //cuda_malloc ((void **) &input, input_size, 1, __LINE__);
- cuda_memset (input, 0, input_size, RES_SCRATCH );
-
- Estimate_Storage_Sizes <<<BLOCKS_POW_2, BLOCK_SIZE>>>
- (system->d_atoms, system->N, system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- system->reaxprm.d_gp, (control_params *)control->d_control, *(dev_lists + FAR_NBRS),
- system->reaxprm.num_atom_types, input);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (output, input, input_size, cudaMemcpyDeviceToHost, __LINE__ );
-
- Htop = &output[0];
- num_3body = &output[1];
- hb_top = &output[ 2 ];
- bond_top = &output[ 2 + system->N ];
-
- *Htop += system->N;
- *Htop *= SAFE_ZONE;
-
- for( int i = 0; i < system->N; ++i ) {
- hb_top[i] = MAX( hb_top[i] * SAFE_HBONDS, MIN_HBONDS );
-
- if (max_3body <= SQR (bond_top[i]))
- max_3body = SQR (bond_top[i]);
-
- *num_3body += SQR(bond_top[i]);
- bond_top[i] = MAX( bond_top[i] * 2, MIN_BONDS );
- }
-
- *num_3body = max_3body * SAFE_ZONE;
-}
-
-
-GLOBAL void Estimate_Storage_Sizes (reax_atom *atoms,
- int N,
- single_body_parameters *sbp,
- two_body_parameters *tbp,
- global_parameters gp,
- control_params *control,
- list far_nbrs,
- int num_atom_types, int *results)
-{
- int *Htop = &results[0];
- int *num_3body = &results[1];
- int *hb_top = &results [ 2 ];
- int *bond_top = &results [ 2 + N ];
-
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- int ihb, jhb;
- real r_ij, r2;
- real C12, C34, C56;
- real BO, BO_s, BO_pi, BO_pi2;
- real p_boc1, p_boc2;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- reax_atom *atom_i, *atom_j;
-
- p_boc1 = gp.l[0];
- p_boc2 = gp.l[1];
-
- //for( i = 0; i < N; ++i ) {
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= N ) return ;
-
- atom_i = &(atoms[i]);
- type_i = atom_i->type;
- start_i = Start_Index(i, &far_nbrs);
- end_i = End_Index(i, &far_nbrs);
- sbp_i = &(sbp[type_i]);
- ihb = sbp_i->p_hbond;
-
- for( pj = start_i; pj < end_i; ++pj ) {
- nbr_pj = &( far_nbrs.select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- atom_j = &( atoms[j] );
- type_j = atom_j->type;
- sbp_j = &( sbp[type_j] );
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ] );
-
-
- if( nbr_pj->d <= control->r_cut ) {
- //++(*Htop);
- atomicAdd (Htop, 1);
-
- /* hydrogen bond lists */
- //TODO - CHANGE ORIGINAL
- if( control->hb_cut > 0 && (ihb==1 || ihb==2) &&
- nbr_pj->d <= control->hb_cut ) {
- jhb = sbp_j->p_hbond;
- if( ihb == 1 && jhb == 2 )
- //++hb_top[i];
- atomicAdd (&hb_top[i], 1);
- else if( ihb == 2 && jhb == 1 )
- //++hb_top[j];
- //atomicAdd (&hb_top[j], 1);
- atomicAdd (&hb_top[i], 1);
- }
- //TODO -- CHANGE ORIGINAL
-
- //CHANGE ORIGINAL
- if (i < j) continue;
- //CHANGE ORIGINAL
-
-
- /* uncorrected bond orders */
- if( nbr_pj->d <= control->nbr_cut ) {
- r_ij = nbr_pj->d;
- r2 = SQR(r_ij);
-
- if( sbp_i->r_s > 0.0 && sbp_j->r_s > 0.0) {
- C12 = twbp->p_bo1 * POW( r_ij / twbp->r_s, twbp->p_bo2 );
- BO_s = (1.0 + control->bo_cut) * EXP( C12 );
- }
- else BO_s = C12 = 0.0;
-
- if( sbp_i->r_pi > 0.0 && sbp_j->r_pi > 0.0) {
- C34 = twbp->p_bo3 * POW( r_ij / twbp->r_p, twbp->p_bo4 );
- BO_pi = EXP( C34 );
- }
- else BO_pi = C34 = 0.0;
-
- if( sbp_i->r_pi_pi > 0.0 && sbp_j->r_pi_pi > 0.0) {
- C56 = twbp->p_bo5 * POW( r_ij / twbp->r_pp, twbp->p_bo6 );
- BO_pi2= EXP( C56 );
- }
- else BO_pi2 = C56 = 0.0;
-
- /* Initially BO values are the uncorrected ones, page 1 */
- BO = BO_s + BO_pi + BO_pi2;
-
- if( BO >= control->bo_cut ) {
- //++bond_top[i];
- //++bond_top[j];
- atomicAdd (&bond_top[i], 1);
- atomicAdd (&bond_top[j], 1);
- }
- }
- }
- }
- //}
-}
-
-void Cuda_Compute_Forces( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list** lists, output_controls *out_control )
-{
- real t_start, t_elapsed;
- real t_1, t_2;
- int *indices;
- int *Htop;
- int max_sparse_entries = 0;
- list *far_nbrs = dev_lists + FAR_NBRS;
- int hblocks;
-
- t_start = Get_Time ();
- if ( !control->tabulate ) {
- Init_Forces <<<BLOCKS, BLOCK_SIZE>>>
- (system->d_atoms, system->reaxprm.d_gp, (control_params *)control->d_control,
- system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box, *dev_workspace,
- *(dev_lists + FAR_NBRS), *(dev_lists + BONDS), *(dev_lists + HBONDS),
- system->N, system->max_sparse_matrix_entries, system->reaxprm.num_atom_types );
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
- else
- {
- Init_Forces_Tab <<< BLOCKS, BLOCK_SIZE >>>
- ( system->d_atoms, system->reaxprm.d_gp, (control_params *)control->d_control,
- system->reaxprm.d_sbp, system->reaxprm.d_tbp,
- (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box, *dev_workspace,
- *(dev_lists + FAR_NBRS), *(dev_lists + BONDS), *(dev_lists + HBONDS),
- system->N, system->max_sparse_matrix_entries, system->reaxprm.num_atom_types,
- d_LR );
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
-
- /*This is for bonds processing to fix dbond and sym_indexes */
- t_1 = Get_Time ();
- fix_sym_dbond_indices <<<BLOCKS, BLOCK_SIZE>>> (*(dev_lists + BONDS), system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
- t_2 = Get_Timing_Info ( t_1 );
-
- //FIX -1 HYDROGEN BOND fix for cases where there are no hbonds.
- if ((control->hb_cut > 0) && (dev_workspace->num_H > 0))
- {
-
- hblocks = (system->N * HBONDS_SYM_THREADS_PER_ATOM / HBONDS_SYM_BLOCK_SIZE) +
- ((system->N * HBONDS_SYM_THREADS_PER_ATOM % HBONDS_SYM_BLOCK_SIZE) == 0 ? 0 : 1);
- t_1 = Get_Time ();
- /*
- int bs = system->N;
- int ss = 32;
- fix_sym_hbond_indices <<<bs, ss>>> (*dev_workspace, *(dev_lists + HBONDS), system->N);
- */
- New_fix_sym_hbond_indices <<<hblocks, HBONDS_SYM_BLOCK_SIZE>>> (*dev_workspace, *(dev_lists + HBONDS), system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
- }
- t_2 = Get_Timing_Info ( t_1 );
-
- t_elapsed = Get_Timing_Info (t_start);
- d_timing.init_forces+= t_elapsed;
-
- Cuda_Validate_Lists( system, dev_workspace, &dev_lists, data->step, system->N,
- system->num_bonds, system->num_hbonds );
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Done with Cuda List Validation \n");
-#endif
-
- //Bonded Force Calculations here.
- t_start = Get_Time ();
- Cuda_Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
- t_elapsed = Get_Timing_Info (t_start);
- d_timing.bonded += t_elapsed;
-
- //Compute the Non Bonded Forces here.
- t_start = Get_Time ();
- Cuda_Compute_NonBonded_Forces( system, control, data, workspace, lists, out_control );
- t_elapsed = Get_Timing_Info (t_start);
- d_timing.nonb += t_elapsed;
-
- //Compute Total Forces here
- Cuda_Compute_Total_Force<<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, (simulation_data *)data->d_simulation_data, *dev_workspace,
- *(dev_lists + BONDS), control->ensemble, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_Compute_Total_Force_PostProcess<<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, (simulation_data *)data->d_simulation_data, *dev_workspace,
- *(dev_lists + BONDS), control->ensemble, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
-
-void Compute_Forces( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list** lists, output_controls *out_control )
-{
- real t_start, t_elapsed;
-
- t_start = Get_Time( );
- if( !control->tabulate )
- Init_Forces( system, control, data, workspace, lists, out_control );
- else Init_Forces_Tab( system, control, data, workspace, lists, out_control );
- t_elapsed = Get_Timing_Info( t_start );
- data->timing.init_forces += t_elapsed;
-
-#if defined(DEBUG_FOCUS)
- print_sparse_matrix (system, workspace);
- fprintf( stderr, "init_forces - ");
-#endif
-
-
- //analyze_hbonds (system, workspace, lists);
-
- t_start = Get_Time( );
- Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
- t_elapsed = Get_Timing_Info( t_start );
- data->timing.bonded += t_elapsed;
-
- //print_bond_list (system, workspace, lists);
- //exit (0);
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "bonded_forces - ");
-#endif
-
- t_start = Get_Time( );
- Compute_NonBonded_Forces( system, control, data, workspace,
- lists, out_control );
- t_elapsed = Get_Timing_Info( t_start );
- data->timing.nonb += t_elapsed;
-
-#ifdef __DEBUG_CUDA__
- fprintf( stderr, "non_bonded_forces - %lf \n", t_elapsed);
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "nonbondeds - ");
-#endif
-
- Compute_Total_Force( system, control, data, workspace, lists );
- //Print_Total_Force( system, control, data, workspace, lists, out_control );
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "totalforces - ");
- //Print_Total_Force( system, control, data, workspace, lists, out_control );
-#endif
-
-#ifdef TEST_FORCES
- Print_Total_Force( system, control, data, workspace, lists, out_control );
- Compare_Total_Forces( system, control, data, workspace, lists, out_control );
-#endif
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "forces - ");
-#endif
-}
-
-
-bool validate_device (reax_system *system, simulation_data *data, static_storage *workspace, list **lists )
-{
- bool retval = false;
-
-#ifdef __BUILD_DEBUG__
-
- retval |= validate_neighbors (system, lists);
- retval |= validate_sym_dbond_indices (system, workspace, lists);
- retval |= validate_bonds (system, workspace, lists);
- retval |= validate_sparse_matrix (system, workspace);
- retval |= validate_three_bodies (system, workspace, lists );
- retval |= validate_hbonds (system, workspace, lists);
- retval |= validate_workspace (system, workspace, lists);
- retval |= validate_data (system, data);
- retval |= validate_atoms (system, lists);
- //analyze_hbonds (system, workspace, lists);
-
- if (!retval) {
- fprintf (stderr, "Results *DOES NOT* mattch between device and host \n");
- }
-#endif
-
- return retval;
-}
diff --git a/PuReMD-GPU/src/forces.h b/PuReMD-GPU/src/forces.h
index 10ac0ee9db54e411b9527429b9da6733d1d4072d..73323f0419baf383d6bf671158ef85584a710728 100644
--- a/PuReMD-GPU/src/forces.h
+++ b/PuReMD-GPU/src/forces.h
@@ -28,21 +28,7 @@ void Init_Bonded_Force_Functions( control_params* );
void Compute_Forces( reax_system*, control_params*, simulation_data*,
static_storage*, list**, output_controls* );
-void Cuda_Compute_Forces( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-
void Estimate_Storage_Sizes( reax_system*, control_params*, list**,
int*, int*, int*, int* );
-//Cuda
-void Cuda_Estimate_Storage_Sizes (reax_system *, control_params *, int *);
-
-GLOBAL void Estimate_Storage_Sizes (reax_atom *, int , single_body_parameters *,
- two_body_parameters *, global_parameters ,
- control_params *, list , int , int *);
-GLOBAL void Estimate_Sparse_Matrix_Entries ( reax_atom *, control_params *,
- simulation_data *, simulation_box *, list , int , int *);
-
-void Cuda_Threebody_List( reax_system *, static_storage *, list *, int );
-bool validate_device (reax_system *, simulation_data *, static_storage *, list **);
#endif
diff --git a/PuReMD-GPU/src/four_body_interactions.c b/PuReMD-GPU/src/four_body_interactions.c
new file mode 100644
index 0000000000000000000000000000000000000000..c51601fa991203a77ec4840c10e74e15cfa42c87
--- /dev/null
+++ b/PuReMD-GPU/src/four_body_interactions.c
@@ -0,0 +1,677 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "four_body_interactions.h"
+
+#include "bond_orders.h"
+#include "box.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+#include "math.h"
+#include "index_utils.h"
+
+
+real Calculate_Omega( rvec dvec_ij, real r_ij, rvec dvec_jk, real r_jk,
+ rvec dvec_kl, real r_kl, rvec dvec_li, real r_li,
+ three_body_interaction_data *p_ijk,
+ three_body_interaction_data *p_jkl,
+ rvec dcos_omega_di, rvec dcos_omega_dj,
+ rvec dcos_omega_dk, rvec dcos_omega_dl,
+ output_controls *out_control )
+{
+ real unnorm_cos_omega, unnorm_sin_omega, omega;
+ real sin_ijk, cos_ijk, sin_jkl, cos_jkl;
+ real htra, htrb, htrc, hthd, hthe, hnra, hnrc, hnhd, hnhe;
+ real arg, poem, tel;
+ rvec cross_jk_kl;
+
+ sin_ijk = SIN( p_ijk->theta );
+ cos_ijk = COS( p_ijk->theta );
+ sin_jkl = SIN( p_jkl->theta );
+ cos_jkl = COS( p_jkl->theta );
+
+ /* omega */
+ unnorm_cos_omega = -rvec_Dot( dvec_ij,dvec_jk )*rvec_Dot( dvec_jk,dvec_kl ) +
+ SQR( r_jk ) * rvec_Dot( dvec_ij,dvec_kl );
+ rvec_Cross( cross_jk_kl, dvec_jk, dvec_kl );
+ unnorm_sin_omega = -r_jk * rvec_Dot( dvec_ij, cross_jk_kl );
+ omega = atan2( unnorm_sin_omega, unnorm_cos_omega );
+
+ /* derivatives */
+ /* coef for adjusments to cos_theta's */
+ /* rla = r_ij, rlb = r_jk, rlc = r_kl, r4 = r_li;
+ coshd = cos_ijk, coshe = cos_jkl;
+ sinhd = sin_ijk, sinhe = sin_jkl; */
+ htra = r_ij + cos_ijk * ( r_kl * cos_jkl - r_jk );
+ htrb = r_jk - r_ij * cos_ijk - r_kl * cos_jkl;
+ htrc = r_kl + cos_jkl * ( r_ij * cos_ijk - r_jk );
+ hthd = r_ij * sin_ijk * ( r_jk - r_kl * cos_jkl );
+ hthe = r_kl * sin_jkl * ( r_jk - r_ij * cos_ijk );
+ hnra = r_kl * sin_ijk * sin_jkl;
+ hnrc = r_ij * sin_ijk * sin_jkl;
+ hnhd = r_ij * r_kl * cos_ijk * sin_jkl;
+ hnhe = r_ij * r_kl * sin_ijk * cos_jkl;
+
+ poem = 2.0 * r_ij * r_kl * sin_ijk * sin_jkl;
+ if( poem < 1e-20 ) poem = 1e-20;
+
+ tel = (SQR(r_ij) + SQR(r_jk) + SQR(r_kl) - SQR(r_li)) -
+ 2.0 * ( r_ij * r_jk * cos_ijk - r_ij * r_kl * cos_ijk * cos_jkl +
+ r_jk * r_kl * cos_jkl );
+
+ arg = tel / poem;
+ if( arg > 1.0 )
+ {
+ arg = 1.0;
+ }
+ if( arg < -1.0 )
+ {
+ arg = -1.0;
+ }
+
+ /*fprintf( out_control->etor,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ htra, htrb, htrc, hthd, hthe, hnra, hnrc, hnhd, hnhe );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ dvec_ij[0]/r_ij, dvec_ij[1]/r_ij, dvec_ij[2]/r_ij );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ -dvec_jk[0]/r_jk, -dvec_jk[1]/r_jk, -dvec_jk[2]/r_jk );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ -dvec_kl[0]/r_kl, -dvec_kl[1]/r_kl, -dvec_kl[2]/r_kl );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e%23.15e\n",
+ r_li, dvec_li[0], dvec_li[1], dvec_li[2] );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e%23.15e\n",
+ r_ij, r_jk, r_kl, r_li );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e%23.15e\n",
+ cos_ijk, cos_jkl, sin_ijk, sin_jkl );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ poem, tel, arg );*/
+ /* fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ -p_ijk->dcos_dk[0]/sin_ijk,
+ -p_ijk->dcos_dk[1]/sin_ijk,
+ -p_ijk->dcos_dk[2]/sin_ijk );
+ fprintf( out_control->etor, "%23.15e%23.15e%23.15e\n",
+ -p_jkl->dcos_dk[0]/sin_jkl,
+ -p_jkl->dcos_dk[1]/sin_jkl,
+ -p_jkl->dcos_dk[2]/sin_jkl );*/
+
+ if( sin_ijk >= 0 && sin_ijk <= MIN_SINE )
+ {
+ sin_ijk = MIN_SINE;
+ }
+ else if( sin_ijk <= 0 && sin_ijk >= -MIN_SINE )
+ {
+ sin_ijk = -MIN_SINE;
+ }
+ if( sin_jkl >= 0 && sin_jkl <= MIN_SINE )
+ {
+ sin_jkl = MIN_SINE;
+ }
+ else if( sin_jkl <= 0 && sin_jkl >= -MIN_SINE )
+ {
+ sin_jkl = -MIN_SINE;
+ }
+
+ // dcos_omega_di
+ rvec_ScaledSum( dcos_omega_di, (htra-arg*hnra)/r_ij, dvec_ij, -1., dvec_li );
+ rvec_ScaledAdd( dcos_omega_di,-(hthd - arg*hnhd)/sin_ijk, p_ijk->dcos_dk );
+ rvec_Scale( dcos_omega_di, 2.0 / poem, dcos_omega_di );
+
+ // dcos_omega_dj
+ rvec_ScaledSum( dcos_omega_dj,-(htra-arg*hnra)/r_ij, dvec_ij,
+ -htrb / r_jk, dvec_jk );
+ rvec_ScaledAdd( dcos_omega_dj,-(hthd-arg*hnhd) / sin_ijk, p_ijk->dcos_dj );
+ rvec_ScaledAdd( dcos_omega_dj,-(hthe-arg*hnhe) / sin_jkl, p_jkl->dcos_di );
+ rvec_Scale( dcos_omega_dj, 2.0 / poem, dcos_omega_dj );
+
+ // dcos_omega_dk
+ rvec_ScaledSum( dcos_omega_dk,-(htrc-arg*hnrc) / r_kl, dvec_kl,
+ htrb / r_jk, dvec_jk );
+ rvec_ScaledAdd( dcos_omega_dk,-(hthd-arg*hnhd) / sin_ijk, p_ijk->dcos_di );
+ rvec_ScaledAdd( dcos_omega_dk,-(hthe-arg*hnhe) / sin_jkl, p_jkl->dcos_dj );
+ rvec_Scale( dcos_omega_dk, 2.0 / poem, dcos_omega_dk );
+
+ // dcos_omega_dl
+ rvec_ScaledSum( dcos_omega_dl, (htrc-arg*hnrc) / r_kl, dvec_kl, 1., dvec_li );
+ rvec_ScaledAdd( dcos_omega_dl,-(hthe-arg*hnhe) / sin_jkl, p_jkl->dcos_dk );
+ rvec_Scale( dcos_omega_dl, 2.0 / poem, dcos_omega_dl );
+
+ return omega;
+ //return arg;
+}
+
+
+void Four_Body_Interactions( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, k, l, pi, pj, pk, pl, pij, plk;
+ int type_i, type_j, type_k, type_l;
+ int start_j, end_j, start_k, end_k;
+ int start_pj, end_pj, start_pk, end_pk;
+ int num_frb_intrs = 0;
+
+ real Delta_j, Delta_k;
+ real r_ij, r_jk, r_kl, r_li;
+ real BOA_ij, BOA_jk, BOA_kl;
+
+ real exp_tor2_ij, exp_tor2_jk, exp_tor2_kl;
+ real exp_tor1, exp_tor3_DjDk, exp_tor4_DjDk, exp_tor34_inv;
+ real exp_cot2_jk, exp_cot2_ij, exp_cot2_kl;
+ real fn10, f11_DjDk, dfn11, fn12;
+
+ real theta_ijk, theta_jkl;
+ real sin_ijk, sin_jkl;
+ real cos_ijk, cos_jkl;
+ real tan_ijk_i, tan_jkl_i;
+
+ real omega, cos_omega, cos2omega, cos3omega;
+ rvec dcos_omega_di, dcos_omega_dj, dcos_omega_dk, dcos_omega_dl;
+
+ real CV, cmn, CEtors1, CEtors2, CEtors3, CEtors4;
+ real CEtors5, CEtors6, CEtors7, CEtors8, CEtors9;
+ real Cconj, CEconj1, CEconj2, CEconj3;
+ real CEconj4, CEconj5, CEconj6;
+
+ real e_tor, e_con;
+ rvec dvec_li;
+ rvec force, ext_press;
+ ivec rel_box_jl;
+ // rtensor total_rtensor, temp_rtensor;
+
+ four_body_header *fbh;
+ four_body_parameters *fbp;
+ bond_data *pbond_ij, *pbond_jk, *pbond_kl;
+ bond_order_data *bo_ij, *bo_jk, *bo_kl;
+ three_body_interaction_data *p_ijk, *p_jkl;
+
+ real p_tor2 = system->reaxprm.gp.l[23];
+ real p_tor3 = system->reaxprm.gp.l[24];
+ real p_tor4 = system->reaxprm.gp.l[25];
+ real p_cot2 = system->reaxprm.gp.l[27];
+
+ list *bonds = (*lists) + BONDS;
+ list *thb_intrs = (*lists) + THREE_BODIES;
+
+
+ for( j = 0; j < system->N; ++j ) {
+ type_j = system->atoms[j].type;
+ Delta_j = workspace->Delta_boc[j];
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+
+
+ for( pk = start_j; pk < end_j; ++pk ) {
+ pbond_jk = &( bonds->select.bond_list[pk] );
+ k = pbond_jk->nbr;
+ bo_jk = &( pbond_jk->bo_data );
+ BOA_jk = bo_jk->BO - control->thb_cut;
+
+ /* see if there are any 3-body interactions involving j&k
+ where j is the central atom. Otherwise there is no point in
+ trying to form a 4-body interaction out of this neighborhood */
+ if( j < k && bo_jk->BO > control->thb_cut/*0*/ &&
+ Num_Entries(pk, thb_intrs) ) {
+ start_k = Start_Index(k, bonds);
+ end_k = End_Index(k, bonds);
+ pj = pbond_jk->sym_index; // pj points to j on k's list
+
+ /* do the same check as above: are there any 3-body interactions
+ involving k&j where k is the central atom */
+ if( Num_Entries(pj, thb_intrs) ) {
+ type_k = system->atoms[k].type;
+ Delta_k = workspace->Delta_boc[k];
+ r_jk = pbond_jk->d;
+
+ start_pk = Start_Index(pk, thb_intrs );
+ end_pk = End_Index(pk, thb_intrs );
+ start_pj = Start_Index(pj, thb_intrs );
+ end_pj = End_Index(pj, thb_intrs );
+
+ exp_tor2_jk = EXP( -p_tor2 * BOA_jk );
+ exp_cot2_jk = EXP( -p_cot2 * SQR(BOA_jk - 1.5) );
+ exp_tor3_DjDk = EXP( -p_tor3 * (Delta_j + Delta_k) );
+ exp_tor4_DjDk = EXP( p_tor4 * (Delta_j + Delta_k) );
+ exp_tor34_inv = 1.0 / (1.0 + exp_tor3_DjDk + exp_tor4_DjDk);
+ f11_DjDk = (2.0 + exp_tor3_DjDk) * exp_tor34_inv;
+
+
+ /* pick i up from j-k interaction where j is the centre atom */
+ for( pi = start_pk; pi < end_pk; ++pi ) {
+ p_ijk = &( thb_intrs->select.three_body_list[pi] );
+ pij = p_ijk->pthb; // pij is pointer to i on j's bond_list
+ pbond_ij = &( bonds->select.bond_list[pij] );
+ bo_ij = &( pbond_ij->bo_data );
+
+
+ if( bo_ij->BO > control->thb_cut/*0*/ ) {
+ i = p_ijk->thb;
+ type_i = system->atoms[i].type;
+ r_ij = pbond_ij->d;
+ BOA_ij = bo_ij->BO - control->thb_cut;
+
+ theta_ijk = p_ijk->theta;
+ sin_ijk = SIN( theta_ijk );
+ cos_ijk = COS( theta_ijk );
+ //tan_ijk_i = 1. / TAN( theta_ijk );
+ if( sin_ijk >= 0 && sin_ijk <= MIN_SINE )
+ tan_ijk_i = cos_ijk / MIN_SINE;
+ else if( sin_ijk <= 0 && sin_ijk >= -MIN_SINE )
+ tan_ijk_i = cos_ijk / -MIN_SINE;
+ else tan_ijk_i = cos_ijk / sin_ijk;
+
+ exp_tor2_ij = EXP( -p_tor2 * BOA_ij );
+ exp_cot2_ij = EXP( -p_cot2 * SQR(BOA_ij -1.5) );
+
+ /* pick l up from j-k intr. where k is the centre */
+ for( pl = start_pj; pl < end_pj; ++pl ) {
+ p_jkl = &( thb_intrs->select.three_body_list[pl] );
+ l = p_jkl->thb;
+ plk = p_jkl->pthb; //pointer to l on k's bond_list!
+ pbond_kl = &( bonds->select.bond_list[plk] );
+ bo_kl = &( pbond_kl->bo_data );
+ type_l = system->atoms[l].type;
+ fbh = &(system->reaxprm.fbp[ index_fbp(type_i,type_j,type_k,type_l,system->reaxprm.num_atom_types ) ]);
+ fbp = &(system->reaxprm.fbp[ index_fbp(type_i,type_j,type_k,type_l,system->reaxprm.num_atom_types )].prm[0]);
+
+ if( i != l && fbh->cnt && bo_kl->BO > control->thb_cut/*0*/ &&
+ bo_ij->BO * bo_jk->BO * bo_kl->BO > control->thb_cut/*0*/ ){
+ ++num_frb_intrs;
+ r_kl = pbond_kl->d;
+ BOA_kl = bo_kl->BO - control->thb_cut;
+
+ theta_jkl = p_jkl->theta;
+ sin_jkl = SIN( theta_jkl );
+ cos_jkl = COS( theta_jkl );
+ //tan_jkl_i = 1. / TAN( theta_jkl );
+ if( sin_jkl >= 0 && sin_jkl <= MIN_SINE )
+ tan_jkl_i = cos_jkl / MIN_SINE;
+ else if( sin_jkl <= 0 && sin_jkl >= -MIN_SINE )
+ tan_jkl_i = cos_jkl / -MIN_SINE;
+ else tan_jkl_i = cos_jkl /sin_jkl;
+
+ Sq_Distance_on_T3( system->atoms[l].x, system->atoms[i].x,
+ &(system->box), dvec_li );
+ r_li = rvec_Norm( dvec_li );
+
+
+ /* omega and its derivative */
+ //cos_omega=Calculate_Omega(pbond_ij->dvec,r_ij,pbond_jk->dvec,
+ omega = Calculate_Omega(pbond_ij->dvec, r_ij, pbond_jk->dvec,
+ r_jk, pbond_kl->dvec, r_kl,
+ dvec_li, r_li, p_ijk, p_jkl,
+ dcos_omega_di, dcos_omega_dj,
+ dcos_omega_dk, dcos_omega_dl,
+ out_control);
+ cos_omega = COS( omega );
+ cos2omega = COS( 2. * omega );
+ cos3omega = COS( 3. * omega );
+ /* end omega calculations */
+
+ /* torsion energy */
+ exp_tor1 = EXP(fbp->p_tor1 * SQR(2.-bo_jk->BO_pi-f11_DjDk));
+ exp_tor2_kl = EXP( -p_tor2 * BOA_kl );
+ exp_cot2_kl = EXP( -p_cot2 * SQR(BOA_kl-1.5) );
+ fn10 = (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_jk) *
+ (1.0 - exp_tor2_kl);
+
+ CV = 0.5 * ( fbp->V1 * (1.0 + cos_omega) +
+ fbp->V2 * exp_tor1 * (1.0 - cos2omega) +
+ fbp->V3 * (1.0 + cos3omega) );
+ //CV = 0.5 * fbp->V1 * (1.0 + cos_omega) +
+ // fbp->V2 * exp_tor1 * (1.0 - SQR(cos_omega)) +
+ // fbp->V3 * (0.5 + 2.0*CUBE(cos_omega) - 1.5 * cos_omega);
+
+ data->E_Tor += e_tor = fn10 * sin_ijk * sin_jkl * CV;
+
+ dfn11 = (-p_tor3 * exp_tor3_DjDk +
+ (p_tor3 * exp_tor3_DjDk - p_tor4 * exp_tor4_DjDk) *
+ (2.+exp_tor3_DjDk) * exp_tor34_inv) * exp_tor34_inv;
+
+ CEtors1 = sin_ijk * sin_jkl * CV;
+
+ CEtors2 = -fn10 * 2.0 * fbp->p_tor1 * fbp->V2 * exp_tor1 *
+ (2.0 - bo_jk->BO_pi - f11_DjDk) * (1.0 - SQR(cos_omega)) *
+ sin_ijk * sin_jkl;
+
+ CEtors3 = CEtors2 * dfn11;
+
+ CEtors4 = CEtors1 * p_tor2 * exp_tor2_ij *
+ (1.0 - exp_tor2_jk) * (1.0 - exp_tor2_kl);
+
+ CEtors5 = CEtors1 * p_tor2 * exp_tor2_jk *
+ (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_kl);
+
+ CEtors6 = CEtors1 * p_tor2 * exp_tor2_kl *
+ (1.0 - exp_tor2_ij) * (1.0 - exp_tor2_jk);
+
+ cmn = -fn10 * CV;
+ CEtors7 = cmn * sin_jkl * tan_ijk_i;
+ CEtors8 = cmn * sin_ijk * tan_jkl_i;
+ CEtors9 = fn10 * sin_ijk * sin_jkl *
+ (0.5 * fbp->V1 - 2.0 * fbp->V2 * exp_tor1 * cos_omega +
+ 1.5 * fbp->V3 * (cos2omega + 2. * SQR(cos_omega)));
+ //cmn = -fn10 * CV;
+ //CEtors7 = cmn * sin_jkl * cos_ijk;
+ //CEtors8 = cmn * sin_ijk * cos_jkl;
+ //CEtors9 = fn10 * sin_ijk * sin_jkl *
+ // (0.5 * fbp->V1 - 2.0 * fbp->V2 * exp_tor1 * cos_omega +
+ // fbp->V3 * (6*SQR(cos_omega) - 1.50));
+ /* end of torsion energy */
+
+
+ /* 4-body conjugation energy */
+ fn12 = exp_cot2_ij * exp_cot2_jk * exp_cot2_kl;
+ data->E_Con += e_con = fbp->p_cot1 * fn12 *
+ (1. + (SQR(cos_omega)-1.) * sin_ijk*sin_jkl);
+
+ Cconj = -2.0 * fn12 * fbp->p_cot1 * p_cot2 *
+ (1. + (SQR(cos_omega)-1.) * sin_ijk*sin_jkl);
+
+ CEconj1 = Cconj * (BOA_ij - 1.5e0);
+ CEconj2 = Cconj * (BOA_jk - 1.5e0);
+ CEconj3 = Cconj * (BOA_kl - 1.5e0);
+
+ CEconj4 = -fbp->p_cot1 * fn12 *
+ (SQR(cos_omega) - 1.0) * sin_jkl * tan_ijk_i;
+ CEconj5 = -fbp->p_cot1 * fn12 *
+ (SQR(cos_omega) - 1.0) * sin_ijk * tan_jkl_i;
+ //CEconj4 = -fbp->p_cot1 * fn12 *
+ // (SQR(cos_omega) - 1.0) * sin_jkl * cos_ijk;
+ //CEconj5 = -fbp->p_cot1 * fn12 *
+ // (SQR(cos_omega) - 1.0) * sin_ijk * cos_jkl;
+ CEconj6 = 2.0 * fbp->p_cot1 * fn12 *
+ cos_omega * sin_ijk * sin_jkl;
+ /* end 4-body conjugation energy */
+
+ //fprintf(stdout, "%6d %6d %6d %6d %7.3f %7.3f %7.3f %7.3f ",
+ // workspace->orig_id[i], workspace->orig_id[j],
+ // workspace->orig_id[k], workspace->orig_id[l],
+ // omega, cos_omega, cos2omega, cos3omega );
+ //fprintf(stdout,
+ // "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ // CEtors2, CEtors3, CEtors4, CEtors5,
+ // CEtors6, CEtors7, CEtors8, CEtors9 );
+ //fprintf(stdout, "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ // theta_ijk, theta_jkl, sin_ijk,
+ // sin_jkl, cos_jkl, tan_jkl_i );
+
+ /* forces */
+ bo_jk->Cdbopi += CEtors2;
+ workspace->CdDelta[j] += CEtors3;
+ workspace->CdDelta[k] += CEtors3;
+ bo_ij->Cdbo += (CEtors4 + CEconj1);
+ bo_jk->Cdbo += (CEtors5 + CEconj2);
+
+ bo_kl->Cdbo += (CEtors6 + CEconj3);
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ /* dcos_theta_ijk */
+ rvec_ScaledAdd( system->atoms[i].f,
+ CEtors7 + CEconj4, p_ijk->dcos_dk );
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors7 + CEconj4, p_ijk->dcos_dj );
+ rvec_ScaledAdd( system->atoms[k].f,
+ CEtors7 + CEconj4, p_ijk->dcos_di );
+
+ /* dcos_theta_jkl */
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors8 + CEconj5, p_jkl->dcos_di );
+ rvec_ScaledAdd( system->atoms[k].f,
+ CEtors8 + CEconj5, p_jkl->dcos_dj );
+ rvec_ScaledAdd( system->atoms[l].f,
+ CEtors8 + CEconj5, p_jkl->dcos_dk );
+
+ /* dcos_omega */
+ rvec_ScaledAdd( system->atoms[i].f,
+ CEtors9 + CEconj6, dcos_omega_di );
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors9 + CEconj6, dcos_omega_dj );
+ rvec_ScaledAdd( system->atoms[k].f,
+ CEtors9 + CEconj6, dcos_omega_dk );
+ rvec_ScaledAdd( system->atoms[l].f,
+ CEtors9 + CEconj6, dcos_omega_dl );
+ }
+ else {
+ ivec_Sum(rel_box_jl, pbond_jk->rel_box, pbond_kl->rel_box);
+
+ /* dcos_theta_ijk */
+ rvec_Scale( force, CEtors7 + CEconj4, p_ijk->dcos_dk );
+ rvec_Add( system->atoms[i].f, force );
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors7 + CEconj4, p_ijk->dcos_dj );
+
+ rvec_Scale( force, CEtors7 + CEconj4, p_ijk->dcos_di );
+ rvec_Add( system->atoms[k].f, force );
+ rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+
+ /* dcos_theta_jkl */
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors8 + CEconj5, p_jkl->dcos_di );
+
+ rvec_Scale( force, CEtors8 + CEconj5, p_jkl->dcos_dj );
+ rvec_Add( system->atoms[k].f, force );
+ rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+ rvec_Scale( force, CEtors8 + CEconj5, p_jkl->dcos_dk );
+ rvec_Add( system->atoms[l].f, force );
+ rvec_iMultiply( ext_press, rel_box_jl, force );
+ rvec_Add( data->ext_press, ext_press );
+
+
+ /* dcos_omega */
+ rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_di );
+ rvec_Add( system->atoms[i].f, force );
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+ rvec_ScaledAdd( system->atoms[j].f,
+ CEtors9 + CEconj6, dcos_omega_dj );
+
+ rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_dk );
+ rvec_Add( system->atoms[k].f, force );
+ rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+ rvec_Scale( force, CEtors9 + CEconj6, dcos_omega_dl );
+ rvec_Add( system->atoms[l].f, force );
+ rvec_iMultiply( ext_press, rel_box_jl, force );
+ rvec_Add( data->ext_press, ext_press );
+
+
+ /* This part is intended for a fully-flexible box */
+ /* rvec_ScaledSum( temp_rvec,
+ CEtors7 + CEconj4, p_ijk->dcos_dk, // i
+ CEtors9 + CEconj6, dcos_omega_di );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[i].x );
+ rtensor_Copy( total_rtensor, temp_rtensor );
+
+ rvec_ScaledSum( temp_rvec,
+ CEtors7 + CEconj4, p_ijk->dcos_dj, // j
+ CEtors8 + CEconj5, p_jkl->dcos_di );
+ rvec_ScaledAdd( temp_rvec,
+ CEtors9 + CEconj6, dcos_omega_dj );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[j].x );
+ rtensor_Add( total_rtensor, temp_rtensor );
+
+ rvec_ScaledSum( temp_rvec,
+ CEtors7 + CEconj4, p_ijk->dcos_di, // k
+ CEtors8 + CEconj5, p_jkl->dcos_dj );
+ rvec_ScaledAdd( temp_rvec,
+ CEtors9 + CEconj6, dcos_omega_dk );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[k].x );
+ rtensor_Add( total_rtensor, temp_rtensor );
+
+ rvec_ScaledSum( temp_rvec,
+ CEtors8 + CEconj5, p_jkl->dcos_dk, // l
+ CEtors9 + CEconj6, dcos_omega_dl );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[l].x );
+ rtensor_Copy( total_rtensor, temp_rtensor );
+
+ if( pbond_ij->imaginary || pbond_jk->imaginary ||
+ pbond_kl->imaginary )
+ rtensor_ScaledAdd( data->flex_bar.P, -1., total_rtensor );
+ else
+ rtensor_Add( data->flex_bar.P, total_rtensor ); */
+ }
+
+#ifdef TEST_ENERGY
+ /*fprintf( out_control->etor,
+ //"%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ //r_ij, r_jk, r_kl,
+ "%12.8f%12.8f%12.8f%12.8f\n",
+ cos_ijk, cos_jkl, sin_ijk, sin_jkl );*/
+ // fprintf( out_control->etor, "%12.8f\n", dfn11 );
+ fprintf( out_control->etor, "%12.8f%12.8f%12.8f\n",
+ fn10, cos_omega, CV );
+
+ fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ CEtors2, CEtors3, CEtors4, CEtors5,
+ CEtors6, CEtors7, CEtors8, CEtors9 );
+
+ /* fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ htra, htrb, htrc, hthd, hthe, hnra, hnrc, hnhd, hnhe ); */
+
+ fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f%12.8f%12.8f%12.8f\n",
+ CEconj1, CEconj2, CEconj3, CEconj4, CEconj5, CEconj6 );
+ /* fprintf(out_control->etor,"%23.15e%23.15e%23.15e%23.15e\n",
+ fbp->V1, fbp->V2, fbp->V3, fbp->p_tor1 );*/
+
+ fprintf( out_control->etor,
+ //"%6d%6d%6d%6d%23.15e%23.15e%23.15e%23.15e\n",
+ "%6d%6d%6d%6d%12.8f%12.8f\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], workspace->orig_id[l],
+ e_tor, e_con );
+ //RAD2DEG(omega), BOA_jk, e_tor, data->E_Tor );
+
+ fprintf( out_control->econ,
+ "%6d%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], workspace->orig_id[l],
+ RAD2DEG(omega), BOA_ij, BOA_jk, BOA_kl,
+ e_con,data->E_Con );
+
+ /* fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
+ (CEtors7 + CEconj4)*p_ijk->dcos_dk[0],
+ (CEtors7 + CEconj4)*p_ijk->dcos_dk[1],
+ (CEtors7 + CEconj4)*p_ijk->dcos_dk[2],
+ (CEtors7 + CEconj4)*p_ijk->dcos_dj[0],
+ (CEtors7 + CEconj4)*p_ijk->dcos_dj[1],
+ (CEtors7 + CEconj4)*p_ijk->dcos_dj[2],
+ (CEtors7 + CEconj4)*p_ijk->dcos_di[0],
+ (CEtors7 + CEconj4)*p_ijk->dcos_di[1],
+ (CEtors7 + CEconj4)*p_ijk->dcos_di[2] ); */
+
+
+ /* fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
+ (CEtors8 + CEconj5)*p_jkl->dcos_di[0],
+ (CEtors8 + CEconj5)*p_jkl->dcos_di[1],
+ (CEtors8 + CEconj5)*p_jkl->dcos_di[2],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dj[0],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dj[1],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dj[2],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dk[0],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dk[1],
+ (CEtors8 + CEconj5)*p_jkl->dcos_dk[2] ); */
+
+ fprintf( out_control->etor,
+ "%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n%12.8f%12.8f%12.8f\n",
+ dcos_omega_di[0], dcos_omega_di[1], dcos_omega_di[2],
+ dcos_omega_dj[0], dcos_omega_dj[1], dcos_omega_dj[2],
+ dcos_omega_dk[0], dcos_omega_dk[1], dcos_omega_dk[2],
+ dcos_omega_dl[0], dcos_omega_dl[1], dcos_omega_dl[2] );
+#endif
+
+#ifdef TEST_FORCES
+ // Torsion Forces
+ Add_dBOpinpi2(system, lists, j, pk, CEtors2, 0.,
+ workspace->f_tor, workspace->f_tor);
+ Add_dDelta( system, lists, j, CEtors3, workspace->f_tor );
+ Add_dDelta( system, lists, k, CEtors3, workspace->f_tor );
+ Add_dBO( system, lists, j, pij, CEtors4, workspace->f_tor );
+ Add_dBO( system, lists, j, pk, CEtors5, workspace->f_tor );
+ Add_dBO( system, lists, k, plk, CEtors6, workspace->f_tor );
+
+ rvec_ScaledAdd(workspace->f_tor[i], CEtors7, p_ijk->dcos_dk);
+ rvec_ScaledAdd(workspace->f_tor[j], CEtors7, p_ijk->dcos_dj);
+ rvec_ScaledAdd(workspace->f_tor[k], CEtors7, p_ijk->dcos_di);
+
+ rvec_ScaledAdd(workspace->f_tor[j], CEtors8, p_jkl->dcos_di);
+ rvec_ScaledAdd(workspace->f_tor[k], CEtors8, p_jkl->dcos_dj);
+ rvec_ScaledAdd(workspace->f_tor[l], CEtors8, p_jkl->dcos_dk);
+
+ rvec_ScaledAdd( workspace->f_tor[i], CEtors9, dcos_omega_di );
+ rvec_ScaledAdd( workspace->f_tor[j], CEtors9, dcos_omega_dj );
+ rvec_ScaledAdd( workspace->f_tor[k], CEtors9, dcos_omega_dk );
+ rvec_ScaledAdd( workspace->f_tor[l], CEtors9, dcos_omega_dl );
+
+ // Conjugation Forces
+ Add_dBO( system, lists, j, pij, CEconj1, workspace->f_con );
+ Add_dBO( system, lists, j, pk, CEconj2, workspace->f_con );
+ Add_dBO( system, lists, k, plk, CEconj3, workspace->f_con );
+
+ rvec_ScaledAdd(workspace->f_con[i], CEconj4, p_ijk->dcos_dk);
+ rvec_ScaledAdd(workspace->f_con[j], CEconj4, p_ijk->dcos_dj);
+ rvec_ScaledAdd(workspace->f_con[k], CEconj4, p_ijk->dcos_di);
+
+ rvec_ScaledAdd(workspace->f_con[j], CEconj5, p_jkl->dcos_di);
+ rvec_ScaledAdd(workspace->f_con[k], CEconj5, p_jkl->dcos_dj);
+ rvec_ScaledAdd(workspace->f_con[l], CEconj5, p_jkl->dcos_dk);
+
+ rvec_ScaledAdd( workspace->f_con[i], CEconj6, dcos_omega_di );
+ rvec_ScaledAdd( workspace->f_con[j], CEconj6, dcos_omega_dj );
+ rvec_ScaledAdd( workspace->f_con[k], CEconj6, dcos_omega_dk );
+ rvec_ScaledAdd( workspace->f_con[l], CEconj6, dcos_omega_dl );
+#endif
+ } // pl check ends
+ } // pl loop ends
+ } // pi check ends
+ } // pi loop ends
+ } // k-j neighbor check ends
+ } // j<k && j-k neighbor check ends
+ } // pk loop ends
+ } // j loop
+
+ /* fprintf( stderr, "4body: ext_press (%23.15e %23.15e %23.15e)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] );*/
+
+#ifdef TEST_FORCES
+ fprintf( stderr, "Number of torsion angles: %d\n", num_frb_intrs );
+ fprintf( stderr, "Torsion Energy: %g\t Conjugation Energy: %g\n",
+ data->E_Tor, data->E_Con );
+#endif
+}
diff --git a/PuReMD-GPU/src/four_body_interactions.h b/PuReMD-GPU/src/four_body_interactions.h
index 402ebe7dc807afd0d14541bf5686a0a674b43b09..8e8dd7c0991a747000e77b2d460711e433db52ef 100644
--- a/PuReMD-GPU/src/four_body_interactions.h
+++ b/PuReMD-GPU/src/four_body_interactions.h
@@ -23,20 +23,10 @@
#include "mytypes.h"
+#define MIN_SINE 1e-10
+
+
void Four_Body_Interactions( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-
-GLOBAL void Four_Body_Interactions ( reax_atom *,
- global_parameters ,
- four_body_header *,
- control_params *,
- list , list ,
- simulation_box *,
- simulation_data *,
- static_storage ,
- int , int , real *, real *, rvec *);
-
-GLOBAL void Four_Body_Postprocess (reax_atom *,
- static_storage,
- list , int );
+ static_storage*, list**, output_controls* );
+
#endif
diff --git a/PuReMD-GPU/src/grid.cu b/PuReMD-GPU/src/grid.c
similarity index 93%
rename from PuReMD-GPU/src/grid.cu
rename to PuReMD-GPU/src/grid.c
index 00e638f4b3ae3828e4e0e208f307d51283d8c2d5..fb09b409194a84b1646da3b779aad8b547ff9db3 100644
--- a/PuReMD-GPU/src/grid.cu
+++ b/PuReMD-GPU/src/grid.c
@@ -19,12 +19,11 @@
----------------------------------------------------------------------*/
#include "grid.h"
+
#include "reset_utils.h"
#include "vector.h"
#include "index_utils.h"
-#include "cuda_utils.h"
-
int Estimate_GCell_Population( reax_system* system )
{
@@ -361,23 +360,6 @@ void Bin_Atoms( reax_system* system, static_storage *workspace )
workspace->realloc.gcell_atoms = MAX(max_atoms*SAFE_ZONE,MIN_GCELL_POPL);
}
-void Cuda_Bin_Atoms (reax_system *system, static_storage *workspace )
-{
- Cuda_Reset_Grid ( &system->d_g);
-
- Bin_Atoms ( system, workspace );
-
- dev_workspace->realloc.gcell_atoms = workspace->realloc.gcell_atoms;
-}
-
-void Cuda_Bin_Atoms_Sync (reax_system *system)
-{
- copy_host_device (system->g.top, system->d_g.top,
- INT_SIZE * system->g.ncell[0]*system->g.ncell[1]*system->g.ncell[2], cudaMemcpyHostToDevice, RES_GRID_TOP);
-
- copy_host_device (system->g.atoms, system->d_g.atoms,
- INT_SIZE * system->g.max_atoms*system->g.ncell[0]*system->g.ncell[1]*system->g.ncell[2], cudaMemcpyHostToDevice, RES_GRID_ATOMS);
-}
inline void reax_atom_Copy( reax_atom *dest, reax_atom *src )
{
@@ -396,11 +378,11 @@ void Copy_Storage( reax_system *system, static_storage *workspace,
int i;
for( i = 0; i < RESTART+1; ++i )
- v[ index_wkspace_sys (i,top, system) ] = workspace->v[ index_wkspace_sys (i,old_id, system) ];
+ v[ index_wkspace_sys (i,top, system->N) ] = workspace->v[ index_wkspace_sys (i,old_id, system->N) ];
for( i = 0; i < 3; ++i ) {
- s[ index_wkspace_sys (i,top, system) ] = workspace->s[ index_wkspace_sys (i,old_id, system) ];
- t[ index_wkspace_sys (i,top, system) ] = workspace->t[ index_wkspace_sys (i,old_id, system) ];
+ s[ index_wkspace_sys (i,top, system->N) ] = workspace->s[ index_wkspace_sys (i,old_id, system->N) ];
+ t[ index_wkspace_sys (i,top, system->N) ] = workspace->t[ index_wkspace_sys (i,old_id, system->N) ];
}
orig_id[top] = workspace->orig_id[old_id];
diff --git a/PuReMD-GPU/src/grid.h b/PuReMD-GPU/src/grid.h
index b524fb76ea394c38dd5a2f7fa94e3b668b43a744..8f26f018e947bc2ebe1789736d8203dc435addfd 100644
--- a/PuReMD-GPU/src/grid.h
+++ b/PuReMD-GPU/src/grid.h
@@ -23,6 +23,11 @@
#include "mytypes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Setup_Grid( reax_system* );
void Update_Grid( reax_system* );
@@ -32,9 +37,12 @@ int Shift( int, int, int, grid* );
void Cluster_Atoms( reax_system*, static_storage* );
void Bin_Atoms( reax_system*, static_storage* );
-void Cuda_Bin_Atoms( reax_system*, static_storage* );
-void Cuda_Bin_Atoms_Sync (reax_system *);
void Reset_Marks( grid*, ivec*, int );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PuReMD-GPU/src/index_utils.h b/PuReMD-GPU/src/index_utils.h
index b856e5efe79eb7e2ca96163d9bd67744b44f9a00..cbd81cadc4113ecf8804d6bd28299336aa4705e6 100644
--- a/PuReMD-GPU/src/index_utils.h
+++ b/PuReMD-GPU/src/index_utils.h
@@ -23,101 +23,74 @@
#include "mytypes.h"
-extern inline HOST_DEVICE int index_grid_3d (int i, int j, int k, grid *g)
-{
- return (i * g->ncell[1] * g->ncell[2]) +
- (j * g->ncell[2]) +
- k;
-}
-extern inline HOST_DEVICE int index_grid_nbrs (int i, int j, int k, int l, grid *g)
+static inline HOST_DEVICE int index_grid_3d( int i, int j, int k, grid *g )
{
- return (i * g->ncell[1] * g->ncell[2] * g->max_nbrs) +
- (j * g->ncell[2] * g->max_nbrs) +
- (k * g->max_nbrs) +
- l;
+ return (i * g->ncell[1] * g->ncell[2]) + (j * g->ncell[2]) + k;
}
-extern inline HOST_DEVICE int index_grid_atoms (int i, int j, int k, int l, grid *g)
+
+static inline HOST_DEVICE int index_grid_nbrs( int i, int j, int k, int l, grid *g )
{
- return (i * g->ncell[1] * g->ncell[2] * g->max_atoms) +
- (j * g->ncell[2] * g->max_atoms) +
- (k * g->max_atoms) +
- l;
+ return (i * g->ncell[1] * g->ncell[2] * g->max_nbrs) +
+ (j * g->ncell[2] * g->max_nbrs) +
+ (k * g->max_nbrs) +
+ l;
}
-extern inline HOST_DEVICE int index_wkspace_sys (int i, int j, reax_system *system)
+
+static inline HOST_DEVICE int index_grid_atoms( int i, int j, int k, int l, grid *g )
{
- return (i * system->N) + j;
+ return (i * g->ncell[1] * g->ncell[2] * g->max_atoms) +
+ (j * g->ncell[2] * g->max_atoms) +
+ (k * g->max_atoms) +
+ l;
}
-extern inline HOST_DEVICE int index_wkspace_sys (int i, int j, int N)
+
+static inline HOST_DEVICE int index_wkspace_sys( int i, int j, int N )
{
return (i * N) + j;
}
-extern inline HOST_DEVICE int index_wkspace_res (int i, int j )
+
+static inline HOST_DEVICE int index_wkspace_res( int i, int j )
{
return (i * (RESTART + 1)) + j;
}
-extern inline HOST_DEVICE int index_tbp (int i, int j, reax_interaction *reax)
-{
- return (i * reax->num_atom_types) + j;
-}
-extern inline HOST_DEVICE int index_tbp (int i, int j, int num_atom_types)
+static inline HOST_DEVICE int index_tbp( int i, int j, int num_atom_types )
{
return (i * num_atom_types) + j;
}
-extern inline HOST_DEVICE int index_thbp (int i, int j, int k, reax_interaction *reax)
-{
- return (i * reax->num_atom_types * reax->num_atom_types ) +
- (j * reax->num_atom_types ) +
- k;
-}
-extern inline HOST_DEVICE int index_thbp (int i, int j, int k, int num_atom_types)
+static inline HOST_DEVICE int index_thbp( int i, int j, int k, int num_atom_types )
{
- return (i * num_atom_types * num_atom_types ) +
- (j * num_atom_types ) +
- k;
+ return (i * num_atom_types * num_atom_types ) + (j * num_atom_types ) + k;
}
-extern inline HOST_DEVICE int index_hbp (int i, int j, int k, reax_interaction *reax)
-{
- return (i * reax->num_atom_types * reax->num_atom_types ) +
- (j * reax->num_atom_types ) +
- k;
-}
-extern inline HOST_DEVICE int index_hbp (int i, int j, int k, int num_atom_types)
+static inline HOST_DEVICE int index_hbp( int i, int j, int k, int num_atom_types )
{
- return (i * num_atom_types * num_atom_types ) +
- (j * num_atom_types ) +
- k;
+ return (i * num_atom_types * num_atom_types ) + (j * num_atom_types ) + k;
}
-extern inline HOST_DEVICE int index_fbp (int i, int j, int k, int l, reax_interaction *reax)
-{
- return (i * reax->num_atom_types * reax->num_atom_types * reax->num_atom_types ) +
- (j * reax->num_atom_types * reax->num_atom_types ) +
- (k * reax->num_atom_types ) +
- l;
-}
-extern inline HOST_DEVICE int index_fbp (int i, int j, int k, int l, int num_atom_types)
+static inline HOST_DEVICE int index_fbp( int i, int j, int k, int l, int num_atom_types )
{
- return (i * num_atom_types * num_atom_types * num_atom_types ) +
- (j * num_atom_types * num_atom_types ) +
- (k * num_atom_types ) +
- l;
+ return (i * num_atom_types * num_atom_types * num_atom_types ) +
+ (j * num_atom_types * num_atom_types ) +
+ (k * num_atom_types ) +
+ l;
}
-extern inline HOST_DEVICE int index_lr (int i, int j, int num_atom_types )
+
+static inline HOST_DEVICE int index_lr( int i, int j, int num_atom_types )
{
return (i * num_atom_types) + j;
}
+
#endif
diff --git a/PuReMD-GPU/src/init_md.c b/PuReMD-GPU/src/init_md.c
new file mode 100644
index 0000000000000000000000000000000000000000..2a2ce1270e2c694722e489b9a3f38f8dd48177a1
--- /dev/null
+++ b/PuReMD-GPU/src/init_md.c
@@ -0,0 +1,879 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "init_md.h"
+
+#include "allocate.h"
+#include "box.h"
+#include "forces.h"
+#include "grid.h"
+#include "index_utils.h"
+#include "lin_alg.h"
+#include "integrate.h"
+#include "neighbors.h"
+#include "list.h"
+#include "lookup.h"
+#include "print_utils.h"
+#include "reset_utils.h"
+#include "system_props.h"
+#include "traj.h"
+#include "vector.h"
+
+
+void Generate_Initial_Velocities(reax_system *system, real T )
+{
+ int i;
+ real scale, norm;
+
+
+ if( T <= 0.1 )
+ {
+ for ( i = 0; i < system->N; i++ )
+ {
+ rvec_MakeZero( system->atoms[i].v );
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "no random velocities...\n" );
+#endif
+ }
+ else
+ {
+ for( i = 0; i < system->N; i++ )
+ {
+ rvec_Random( system->atoms[i].v );
+
+ norm = rvec_Norm_Sqr( system->atoms[i].v );
+ scale = SQRT( system->reaxprm.sbp[ system->atoms[i].type ].mass *
+ norm / (3.0 * K_B * T) );
+
+ rvec_Scale( system->atoms[i].v, 1.0/scale, system->atoms[i].v );
+
+ /*
+ fprintf( stderr, "v = %f %f %f\n",
+ system->atoms[i].v[0],system->atoms[i].v[1],system->atoms[i].v[2]);
+ fprintf( stderr, "scale = %f\n", scale );
+ fprintf( stderr, "v = %f %f %f\n",
+ system->atoms[i].v[0],system->atoms[i].v[1],system->atoms[i].v[2]);
+ */
+ }
+ }
+}
+
+
+void Init_System( reax_system *system, control_params *control,
+ simulation_data *data )
+{
+ int i;
+ rvec dx;
+
+ if( !control->restart )
+ {
+ Reset_Atoms( system );
+ }
+
+ Compute_Total_Mass( system, data );
+
+ Compute_Center_of_Mass( system, data, stderr );
+
+ /* reposition atoms */
+ // just fit the atoms to the periodic box
+ if( control->reposition_atoms == 0 )
+ {
+ rvec_MakeZero( dx );
+ }
+ // put the center of mass to the center of the box
+ else if( control->reposition_atoms == 1 )
+ {
+ rvec_Scale( dx, 0.5, system->box.box_norms );
+ rvec_ScaledAdd( dx, -1., data->xcm );
+ }
+ // put the center of mass to the origin
+ else if( control->reposition_atoms == 2 ) {
+ rvec_Scale( dx, -1., data->xcm );
+ }
+ else {
+ fprintf( stderr, "UNKNOWN OPTION: reposition_atoms. Terminating...\n" );
+ exit( UNKNOWN_OPTION );
+ }
+
+ for( i = 0; i < system->N; ++i ) {
+ Inc_on_T3( system->atoms[i].x, dx, &(system->box) );
+ /*fprintf( stderr, "%6d%2d%8.3f%8.3f%8.3f\n",
+ i, system->atoms[i].type,
+ system->atoms[i].x[0], system->atoms[i].x[1], system->atoms[i].x[2] );*/
+ }
+
+ /* Initialize velocities so that desired init T can be attained */
+ if( !control->restart || (control->restart && control->random_vel) ) {
+ Generate_Initial_Velocities( system, control->T_init );
+ }
+
+ Setup_Grid( system );
+}
+
+
+void Init_Simulation_Data( reax_system *system, control_params *control,
+ simulation_data *data, output_controls *out_control,
+ evolve_function *Evolve )
+{
+
+ Reset_Simulation_Data( data );
+
+ if( !control->restart )
+ data->step = data->prev_steps = 0;
+
+ switch( control->ensemble ) {
+ case NVE:
+ data->N_f = 3 * system->N;
+ *Evolve = Velocity_Verlet_NVE;
+ break;
+
+
+ case NVT:
+ data->N_f = 3 * system->N + 1;
+ //control->Tau_T = 100 * data->N_f * K_B * control->T_final;
+ if( !control->restart || (control->restart && control->random_vel) ) {
+ data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
+ data->N_f * K_B * control->T );
+ data->therm.v_xi = data->therm.G_xi * control->dt;
+ data->therm.v_xi_old = 0;
+ data->therm.xi = 0;
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "init_md: G_xi=%f Tau_T=%f E_kin=%f N_f=%f v_xi=%f\n",
+ data->therm.G_xi, control->Tau_T, data->E_Kin,
+ data->N_f, data->therm.v_xi );
+#endif
+ }
+
+ *Evolve = Velocity_Verlet_Nose_Hoover_NVT_Klein;
+ break;
+
+
+ case NPT: // Anisotropic NPT
+ fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
+ exit( UNKNOWN_OPTION );
+ data->N_f = 3 * system->N + 9;
+ if( !control->restart ) {
+ data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
+ data->N_f * K_B * control->T );
+ data->therm.v_xi = data->therm.G_xi * control->dt;
+ data->iso_bar.eps = 0.33333 * log(system->box.volume);
+ //data->inv_W = 1. / (data->N_f*K_B*control->T*SQR(control->Tau_P));
+ //Compute_Pressure( system, data, workspace );
+ }
+ *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
+ break;
+
+
+ case sNPT: // Semi-Isotropic NPT
+ data->N_f = 3 * system->N + 4;
+ *Evolve = Velocity_Verlet_Berendsen_SemiIsotropic_NPT;
+ break;
+
+
+ case iNPT: // Isotropic NPT
+ data->N_f = 3 * system->N + 2;
+ *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
+ break;
+
+ case bNVT: //berendensen NVT
+ data->N_f = 3 * system->N + 1;
+ *Evolve = Velocity_Verlet_Berendsen_NVT;
+ break;
+
+ default:
+ break;
+ }
+
+ Compute_Kinetic_Energy( system, data );
+
+ /* init timing info for the host*/
+ data->timing.start = Get_Time( );
+ data->timing.total = data->timing.start;
+ data->timing.nbrs = 0;
+ data->timing.init_forces = 0;
+ data->timing.bonded = 0;
+ data->timing.nonb = 0;
+ data->timing.QEq = 0;
+ data->timing.matvecs = 0;
+}
+
+
+void Init_Workspace( reax_system *system, control_params *control,
+ static_storage *workspace )
+{
+ int i;
+
+ /* Allocate space for hydrogen bond list */
+ workspace->hbond_index = (int *) malloc( system->N * sizeof( int ) );
+
+ /* bond order related storage */
+ workspace->total_bond_order = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Deltap = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Deltap_boc = (real *) malloc( system->N * sizeof( real ) );
+ workspace->dDeltap_self = (rvec *) malloc( system->N * sizeof( rvec ) );
+
+ workspace->Delta = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Delta_lp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Delta_lp_temp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->dDelta_lp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->dDelta_lp_temp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Delta_e = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Delta_boc = (real *) malloc( system->N * sizeof( real ) );
+ workspace->nlp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->nlp_temp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->Clp = (real *) malloc( system->N * sizeof( real ) );
+ workspace->CdDelta = (real *) malloc( system->N * sizeof( real ) );
+ workspace->vlpex = (real *) malloc( system->N * sizeof( real ) );
+
+ /* QEq storage */
+ //workspace->H = NULL;
+ //workspace->L = NULL;
+ //workspace->U = NULL;
+ //
+ workspace->H.start = NULL;
+ workspace->L.start = NULL;
+ workspace->U.start = NULL;
+
+ workspace->H.entries = NULL;
+ workspace->L.entries = NULL;
+ workspace->U.entries = NULL;
+
+ workspace->droptol = (real *) calloc( system->N, sizeof( real ) );
+ workspace->w = (real *) calloc( system->N, sizeof( real ) );
+ workspace->Hdia_inv = (real *) calloc( system->N, sizeof( real ) );
+ workspace->b = (real *) calloc( system->N * 2, sizeof( real ) );
+ workspace->b_s = (real *) calloc( system->N, sizeof( real ) );
+ workspace->b_t = (real *) calloc( system->N, sizeof( real ) );
+ workspace->b_prc = (real *) calloc( system->N * 2, sizeof( real ) );
+ workspace->b_prm = (real *) calloc( system->N * 2, sizeof( real ) );
+ workspace->s_t = (real *) calloc( system->N * 2, sizeof( real ) );
+ workspace->s = (real *) calloc( 5 * system->N, sizeof( real ) );
+ workspace->t = (real *) calloc( 5 * system->N, sizeof( real ) );
+ // workspace->s_old = (real *) calloc( system->N, sizeof( real ) );
+ // workspace->t_old = (real *) calloc( system->N, sizeof( real ) );
+ // workspace->s_oldest = (real *) calloc( system->N, sizeof( real ) );
+ // workspace->t_oldest = (real *) calloc( system->N, sizeof( real ) );
+
+ for( i = 0; i < system->N; ++i ) {
+ workspace->Hdia_inv[i] = 1./system->reaxprm.sbp[system->atoms[i].type].eta;
+ workspace->b_s[i] = -system->reaxprm.sbp[ system->atoms[i].type ].chi;
+ workspace->b_t[i] = -1.0;
+
+ workspace->b[i] = -system->reaxprm.sbp[ system->atoms[i].type ].chi;
+ workspace->b[i+system->N] = -1.0;
+ }
+
+ /* GMRES storage */
+ workspace->y = (real *) calloc( RESTART+1, sizeof( real ) );
+ workspace->z = (real *) calloc( RESTART+1, sizeof( real ) );
+ workspace->g = (real *) calloc( RESTART+1, sizeof( real ) );
+ workspace->hs = (real *) calloc( RESTART+1, sizeof( real ) );
+ workspace->hc = (real *) calloc( RESTART+1, sizeof( real ) );
+
+ workspace->rn = (real *) calloc( (RESTART+1)*system->N*2, sizeof( real) );
+ workspace->v = (real *) calloc( (RESTART+1)*system->N, sizeof( real) );
+ workspace->h = (real *) calloc( (RESTART+1)*(RESTART+1), sizeof( real) );
+
+ /* CG storage */
+ workspace->r = (real *) calloc( system->N, sizeof( real ) );
+ workspace->d = (real *) calloc( system->N, sizeof( real ) );
+ workspace->q = (real *) calloc( system->N, sizeof( real ) );
+ workspace->p = (real *) calloc( system->N, sizeof( real ) );
+
+ /* integrator storage */
+ workspace->a = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_old = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->v_const = (rvec *) malloc( system->N * sizeof( rvec ) );
+
+
+ /* storage for analysis */
+ if( control->molec_anal || control->diffusion_coef )
+ {
+ workspace->mark = (int *) calloc( system->N, sizeof(int) );
+ workspace->old_mark = (int *) calloc( system->N, sizeof(int) );
+ }
+ else
+ workspace->mark = workspace->old_mark = NULL;
+
+ if( control->diffusion_coef )
+ workspace->x_old = (rvec *) calloc( system->N, sizeof( rvec ) );
+ else workspace->x_old = NULL;
+
+
+#ifdef TEST_FORCES
+ workspace->dDelta = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_ele = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_vdw = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_bo = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_be = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_lp = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_ov = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_un = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_ang = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_coa = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_pen = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_hb = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_tor = (rvec *) malloc( system->N * sizeof( rvec ) );
+ workspace->f_con = (rvec *) malloc( system->N * sizeof( rvec ) );
+#endif
+
+ workspace->realloc.num_far = -1;
+ workspace->realloc.Htop = -1;
+ workspace->realloc.hbonds = -1;
+ workspace->realloc.bonds = -1;
+ workspace->realloc.num_3body = -1;
+ workspace->realloc.gcell_atoms = -1;
+
+ Reset_Workspace( system, workspace );
+}
+
+void compare_far_neighbors (int *test, int *start, int *end, far_neighbor_data *data, list *slist, int N)
+{
+ int index = 0;
+ int count = 0;
+ int jicount = 0;
+ int i, j, end_index, gpu_index, gpu_end, k;
+ far_neighbor_data gpu, cpu;
+
+ /*
+ for (int i = 0; i < N ; i++ )
+ {
+ if (test[i] != start[i]) {
+ fprintf (stderr, "start index does not match \n");
+ exit (0);
+ }
+
+ if (test[i+1] != (end[i]) ){
+ fprintf (stderr, "end index does not match for atom %d (cpu: %d gpu: %d) \n", i, test[i+1], end[i]);
+ exit (0);
+ }
+ }
+ */
+
+
+ for (i = 0; i < N; i++){
+ index = Start_Index (i, slist);
+ //fprintf (stderr, "GPU : Neighbors of atom --> %d (start: %d , end: %d )\n", i, start[i], end[i]);
+
+
+ for (j = start[i]; j < end[i]; j++){
+ gpu = data[j];
+
+ if (i < data[j].nbr) continue;
+ /*
+ if (i < data[j].nbr) {
+ //fprintf (stderr, " atom %d and neighbor %d @ index %d\n", i, data[j].nbr, j);
+ int src = data[j].nbr;
+ int dest = i;
+ int x;
+
+
+ for (x = start[src]; x < end[src]; x++) {
+ if (data[x].nbr != dest) continue;
+
+ gpu = data[x];
+ cpu = data[j];
+
+ if ( (gpu.d != cpu.d) ||
+ (cpu.dvec[0] != gpu.dvec[0]) || (cpu.dvec[1] != gpu.dvec[1]) || (cpu.dvec[2] != gpu.dvec[2]) ||
+ (cpu.rel_box[0] != gpu.rel_box[0]) || (cpu.rel_box[1] != gpu.rel_box[1]) || (cpu.rel_box[2] != gpu.rel_box[2])) {
+ fprintf (stderr, " atom %d neighbor %d (%f, %d, %d, %d - %f %f %f) \n", i, data[j].nbr,
+ data[j].d,
+ data[j].rel_box[0],
+ data[j].rel_box[1],
+ data[j].rel_box[2],
+ data[j].dvec[0],
+ data[j].dvec[1],
+ data[j].dvec[2]
+ );
+ fprintf (stderr, " atom %d neighbor %d (%f, %d, %d, %d - %f %f %f) \n", data[j].nbr, data[x].nbr,
+ data[x].d,
+ data[x].rel_box[0],
+ data[x].rel_box[1],
+ data[x].rel_box[2],
+ data[x].dvec[0],
+ data[x].dvec[1],
+ data[x].dvec[2]
+ );
+ jicount++;
+ }
+ break;
+ }
+
+ if (x >= end[src]) {
+ fprintf (stderr, "could not find the neighbor duplicate data for ij (%d %d)\n", i, src );
+ exit (0);
+ }
+
+ continue;
+ }
+ */
+
+ cpu = slist->select.far_nbr_list[index];
+ //if ( (gpu.nbr != cpu.nbr) || (gpu.d != cpu.d) ){
+ //if ( (gpu->d != cpu->d) ){
+ if ( (gpu.nbr != cpu.nbr) || (gpu.d != cpu.d) ||
+ (cpu.dvec[0] != gpu.dvec[0]) || (cpu.dvec[1] != gpu.dvec[1]) || (cpu.dvec[2] != gpu.dvec[2]) ||
+ (cpu.rel_box[0] != gpu.rel_box[0]) || (cpu.rel_box[1] != gpu.rel_box[1]) || (cpu.rel_box[2] != gpu.rel_box[2])) {
+ //if ( (gpu.dvec[0] != i) || (gpu.dvec[1] != i) ||(gpu.dvec[2] != i) ||
+ // (gpu.rel_box[0] != i) || (gpu.rel_box[1] != i) ||(gpu.rel_box[2] != i) ) {
+ //if (memcmp (&gpu, &cpu, FAR_NEIGHBOR_SIZE - RVEC_SIZE - INT_SIZE )){
+
+ fprintf (stderr, "GPU:atom --> %d (s: %d , e: %d, i: %d ) (%d %d %d) \n", i, start[i], end[i], j, gpu.rel_box[0], gpu.rel_box[1], gpu.rel_box[2] );
+ fprintf (stderr, "CPU:atom --> %d (s: %d , e: %d, i: %d )\n", i, Start_Index(i, slist), End_Index (i, slist), index);
+
+ /*
+ fprintf (stdout, "Far neighbors does not match atom: %d \n", i );
+ fprintf (stdout, "neighbor %d , %d \n", cpu.nbr, gpu.nbr);
+ fprintf (stdout, "d %f , %f \n", slist->select.far_nbr_list[index].d, data[j].d);
+ fprintf (stdout, "dvec (%f %f %f) (%f %f %f) \n",
+ cpu.dvec[0], cpu.dvec[1], cpu.dvec[2],
+ gpu.dvec[0], gpu.dvec[1], gpu.dvec[2] );
+
+ fprintf (stdout, "ivec (%d %d %d) (%d %d %d) \n",
+ cpu.rel_box[0], cpu.rel_box[1], cpu.rel_box[2],
+ gpu.rel_box[0], gpu.rel_box[1], gpu.rel_box[2] );
+
+ */
+ count ++;
+ }
+
+ //fprintf (stderr, "GPU (neighbor %d , d %d )\n", gpu->nbr, gpu->d);
+ index ++;
+ }
+
+ if (index != End_Index (i, slist))
+ {
+ fprintf( stderr,
+ "End index does not match for atom --> %d end index (%d) Cpu (%d, %d ) gpu (%d, %d)\n",
+ i, index, Start_Index (i, slist), End_Index(i, slist),
+ start[i], end[i]);
+ exit( 10 );
+ }
+ }
+
+ fprintf (stderr, "Far neighbors MATCH between CPU and GPU -->%d reverse %d \n", count, jicount);
+
+ /*
+ for (int i = 0; i < N; i++)
+ {
+ index = Start_Index (i, slist);
+ end_index = End_Index (i, slist);
+
+ gpu_index = start[i];
+ gpu_end = end[i];
+ for (int j = index; j < end_index; j++)
+ {
+ far_neighbor_data *cpu = &slist->select.far_nbr_list[j];
+ far_neighbor_data *gpu;
+
+ for (k = gpu_index; k < gpu_end; k++) {
+ gpu = &data[k];
+ if (gpu->nbr == cpu->nbr) break;
+ }
+
+ if (k == gpu_end) { fprintf (stderr, " could not find neighbor for atom %d \n", i); exit (1); }
+
+ if ( (gpu->nbr != cpu->nbr) || (gpu->d != cpu->d) ||
+ ((cpu->dvec[0] || gpu->dvec[0]) || (cpu->dvec[1] || gpu->dvec[1]) || (cpu->dvec[2] || gpu->dvec[2])) ||
+ ((cpu->rel_box[0] || gpu->rel_box[0]) || (cpu->rel_box[1] || gpu->rel_box[1]) || (cpu->rel_box[2] || gpu->rel_box[2])) ) {
+
+ fprintf (stderr, "Far neighbors does not match atom: %d \n", i );
+ fprintf (stderr, "neighbor %d , %d \n", cpu->nbr, gpu->nbr);
+ fprintf (stderr, "d %d , %d \n", cpu->d, gpu->d);
+ fprintf (stderr, "dvec (%f %f %f) (%f %f %f) \n",
+ cpu->dvec[0], cpu->dvec[1], cpu->dvec[2],
+ gpu->dvec[0], gpu->dvec[1], gpu->dvec[2] );
+
+ fprintf (stderr, "ivec (%d %d %d) (%d %d %d) \n",
+ cpu->rel_box[0], cpu->rel_box[1], cpu->rel_box[2],
+ gpu->rel_box[0], gpu->rel_box[1], gpu->rel_box[2] );
+ fprintf (stderr, "GPU start %d GPU End %d \n", gpu_index, gpu_end );
+
+ exit (1);
+ }
+ }
+ }
+
+ */
+}
+
+
+void Init_Lists( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, num_nbrs, num_hbonds, num_bonds, num_3body, Htop;
+ int *hb_top, *bond_top;
+
+ real t_start, t_elapsed;
+
+ num_nbrs = Estimate_NumNeighbors( system, control, workspace, lists );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, "Serial NumNeighbors ---> %d \n", num_nbrs);
+#endif
+
+ if( !Make_List(system->N, num_nbrs, TYP_FAR_NEIGHBOR, (*lists)+FAR_NBRS ) ) {
+ fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
+ exit( INIT_ERR );
+ }
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "memory allocated: far_nbrs = %ldMB\n",
+ num_nbrs * sizeof(far_neighbor_data) / (1024*1024) );
+#endif
+
+ t_start = Get_Time ();
+ Generate_Neighbor_Lists(system,control,data,workspace,lists,out_control);
+ t_elapsed = Get_Timing_Info ( t_start );
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " Timing Generate Neighbors %lf \n", t_elapsed );
+#endif
+
+ Htop = 0;
+ hb_top = (int*) calloc( system->N, sizeof(int) );
+ bond_top = (int*) calloc( system->N, sizeof(int) );
+ num_3body = 0;
+ Estimate_Storage_Sizes( system, control, lists,
+ &Htop, hb_top, bond_top, &num_3body );
+
+ Allocate_Matrix( &(workspace->H), system->N, Htop );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "estimated storage - Htop: %d\n", Htop );
+ fprintf( stderr, "memory allocated: H = %ldMB\n",
+ Htop * sizeof(sparse_matrix_entry) / (1024*1024) );
+#endif
+
+ workspace->num_H = 0;
+ if( control->hb_cut > 0 ) {
+ /* init H indexes */
+ for( i = 0; i < system->N; ++i )
+ if( system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 1 ) // H atom
+ workspace->hbond_index[i] = workspace->num_H++;
+ else workspace->hbond_index[i] = -1;
+
+ Allocate_HBond_List( system->N, workspace->num_H, workspace->hbond_index,
+ hb_top, (*lists)+HBONDS );
+ num_hbonds = hb_top[system->N-1];
+
+#ifdef __DEBUG_CUDA__
+ fprintf( stderr, "Serial num_hbonds: %d\n", num_hbonds );
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "estimated storage - num_hbonds: %d\n", num_hbonds );
+ fprintf( stderr, "memory allocated: hbonds = %ldMB\n",
+ num_hbonds * sizeof(hbond_data) / (1024*1024) );
+#endif
+ }
+
+ /* bonds list */
+ Allocate_Bond_List( system->N, bond_top, (*lists)+BONDS );
+ num_bonds = bond_top[system->N-1];
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "estimated storage - num_bonds: %d\n", num_bonds );
+ fprintf( stderr, "memory allocated: bonds = %ldMB\n",
+ num_bonds * sizeof(bond_data) / (1024*1024) );
+#endif
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " host num_3body : %d \n", num_3body);
+ fprintf (stderr, " host num_bonds : %d \n", num_bonds);
+#endif
+
+ /* 3bodies list */
+ if(!Make_List(num_bonds, num_3body, TYP_THREE_BODY, (*lists)+THREE_BODIES )) {
+ fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "estimated storage - num_3body: %d\n", num_3body );
+ fprintf( stderr, "memory allocated: 3-body = %ldMB\n",
+ num_3body * sizeof(three_body_interaction_data) / (1024*1024) );
+#endif
+
+#ifdef TEST_FORCES
+ if(!Make_List( system->N, num_bonds * 8, TYP_DDELTA, (*lists) + DDELTA )) {
+ fprintf( stderr, "Problem in initializing dDelta list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+
+ if( !Make_List( num_bonds, num_bonds*MAX_BONDS*3, TYP_DBO, (*lists)+DBO ) ) {
+ fprintf( stderr, "Problem in initializing dBO list. Terminating!\n" );
+ exit( INIT_ERR );
+ }
+#endif
+
+ free( hb_top );
+ free( bond_top );
+}
+
+
+void Init_Out_Controls(reax_system *system, control_params *control,
+ static_storage *workspace, output_controls *out_control)
+{
+ char temp[1000];
+
+ /* Init trajectory file */
+ if( out_control->write_steps > 0 ) {
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".trj" );
+ out_control->trj = fopen( temp, "w" );
+ out_control->write_header( system, control, workspace, out_control );
+ }
+
+ if( out_control->energy_update_freq > 0 ) {
+ /* Init out file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".out" );
+ out_control->out = fopen( temp, "w" );
+ fprintf( out_control->out, "%-6s%16s%16s%16s%11s%11s%13s%13s%13s\n",
+ "step", "total energy", "poten. energy", "kin. energy",
+ "temp.", "target", "volume", "press.", "target" );
+ fflush( out_control->out );
+
+ /* Init potentials file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".pot" );
+ out_control->pot = fopen( temp, "w" );
+ fprintf( out_control->pot,
+ "%-6s%13s%13s%13s%13s%13s%13s%13s%13s%13s%13s%13s\n",
+ "step", "ebond", "eatom", "elp", "eang", "ecoa", "ehb",
+ "etor", "econj", "evdw","ecoul", "epol" );
+ fflush( out_control->pot );
+
+ /* Init log file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".log" );
+ out_control->log = fopen( temp, "w" );
+ fprintf( out_control->log, "%-6s%10s%10s%10s%10s%10s%10s%10s\n",
+ "step", "total", "neighbors", "init", "bonded",
+ "nonbonded", "QEq", "matvec" );
+ }
+
+ /* Init pressure file */
+ if( control->ensemble == NPT ||
+ control->ensemble == iNPT ||
+ control->ensemble == sNPT ) {
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".prs" );
+ out_control->prs = fopen( temp, "w" );
+ fprintf( out_control->prs, "%-6s%13s%13s%13s%13s%13s%13s%13s%13s\n",
+ "step", "norm_x", "norm_y", "norm_z",
+ "press_x", "press_y", "press_z", "target_p", "volume" );
+ fflush( out_control->prs );
+ }
+
+ /* Init molecular analysis file */
+ if( control->molec_anal ) {
+ sprintf( temp, "%s.mol", control->sim_name );
+ out_control->mol = fopen( temp, "w" );
+ if( control->num_ignored ) {
+ sprintf( temp, "%s.ign", control->sim_name );
+ out_control->ign = fopen( temp, "w" );
+ }
+ }
+
+ /* Init electric dipole moment analysis file */
+ if( control->dipole_anal ) {
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".dpl" );
+ out_control->dpl = fopen( temp, "w" );
+ fprintf( out_control->dpl,
+ "Step Molecule Count Avg. Dipole Moment Norm\n" );
+ fflush( out_control->dpl );
+ }
+
+ /* Init diffusion coef analysis file */
+ if( control->diffusion_coef ) {
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".drft" );
+ out_control->drft = fopen( temp, "w" );
+ fprintf( out_control->drft, "Step Type Count Avg Squared Disp\n" );
+ fflush( out_control->drft );
+ }
+
+#ifdef TEST_ENERGY
+ /* open bond energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ebond" );
+ out_control->ebond = fopen( temp, "w" );
+
+ /* open lone-pair energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".elp" );
+ out_control->elp = fopen( temp, "w" );
+
+ /* open overcoordination energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".eov" );
+ out_control->eov = fopen( temp, "w" );
+
+ /* open undercoordination energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".eun" );
+ out_control->eun = fopen( temp, "w" );
+
+ /* open angle energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".eval" );
+ out_control->eval = fopen( temp, "w" );
+
+ /* open penalty energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".epen" );
+ out_control->epen = fopen( temp, "w" );
+
+ /* open coalition energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ecoa" );
+ out_control->ecoa = fopen( temp, "w" );
+
+ /* open hydrogen bond energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ehb" );
+ out_control->ehb = fopen( temp, "w" );
+
+ /* open torsion energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".etor" );
+ out_control->etor = fopen( temp, "w" );
+
+ /* open conjugation energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".econ" );
+ out_control->econ = fopen( temp, "w" );
+
+ /* open vdWaals energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".evdw" );
+ out_control->evdw = fopen( temp, "w" );
+
+ /* open coulomb energy file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ecou" );
+ out_control->ecou = fopen( temp, "w" );
+#endif
+
+#ifdef TEST_FORCES
+ /* open bond orders file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fbo" );
+ out_control->fbo = fopen( temp, "w" );
+
+ /* open bond orders derivatives file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fdbo" );
+ out_control->fdbo = fopen( temp, "w" );
+
+ /* open bond forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fbond" );
+ out_control->fbond = fopen( temp, "w" );
+
+ /* open lone-pair forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".flp" );
+ out_control->flp = fopen( temp, "w" );
+
+ /* open overcoordination forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fatom" );
+ out_control->fatom = fopen( temp, "w" );
+
+ /* open angle forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".f3body" );
+ out_control->f3body = fopen( temp, "w" );
+
+ /* open hydrogen bond forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fhb" );
+ out_control->fhb = fopen( temp, "w" );
+
+ /* open torsion forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".f4body" );
+ out_control->f4body = fopen( temp, "w" );
+
+ /* open nonbonded forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".fnonb" );
+ out_control->fnonb = fopen( temp, "w" );
+
+ /* open total force file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ftot" );
+ out_control->ftot = fopen( temp, "w" );
+
+ /* open coulomb forces file */
+ strcpy( temp, control->sim_name );
+ strcat( temp, ".ftot2" );
+ out_control->ftot2 = fopen( temp, "w" );
+#endif
+
+ /* Error handling */
+ /* if ( out_control->out == NULL || out_control->pot == NULL ||
+ out_control->log == NULL || out_control->mol == NULL ||
+ out_control->dpl == NULL || out_control->drft == NULL ||
+ out_control->pdb == NULL )
+ {
+ fprintf( stderr, "FILE OPEN ERROR. TERMINATING..." );
+ exit( CANNOT_OPEN_OUTFILE );
+ }*/
+}
+
+
+void Initialize(reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace, list **lists,
+ output_controls *out_control, evolve_function *Evolve)
+{
+ Randomize();
+
+ Init_System( system, control, data );
+
+ Init_Simulation_Data( system, control, data, out_control, Evolve );
+
+ Init_Workspace( system, control, workspace );
+
+ Init_Lists( system, control, data, workspace, lists, out_control );
+
+ Init_Out_Controls( system, control, workspace, out_control );
+
+ /* These are done in forces.c, only forces.c can see all those functions */
+ Init_Bonded_Force_Functions( control );
+
+#ifdef TEST_FORCES
+ Init_Force_Test_Functions( );
+#endif
+
+ if( control->tabulate )
+ Make_LR_Lookup_Table( system, control );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "data structures have been initialized...\n" );
+#endif
+}
diff --git a/PuReMD-GPU/src/init_md.cu b/PuReMD-GPU/src/init_md.cu
deleted file mode 100644
index e1912d3c2bd139e2dfcda43b23c77adfcd325782..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/init_md.cu
+++ /dev/null
@@ -1,1361 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "init_md.h"
-#include "allocate.h"
-#include "box.h"
-#include "forces.h"
-#include "grid.h"
-#include "GMRES.h"
-#include "integrate.h"
-#include "neighbors.h"
-#include "list.h"
-#include "lookup.h"
-#include "print_utils.h"
-#include "reset_utils.h"
-#include "system_props.h"
-#include "traj.h"
-#include "vector.h"
-
-
-#include "cuda_init.h"
-#include "cuda_copy.h"
-#include "cuda_utils.h"
-#include "helpers.h"
-#include "reduction.h"
-
-#include "index_utils.h"
-
-#include "validation.h"
-
-void Generate_Initial_Velocities(reax_system *system, real T )
-{
- int i;
- real scale, norm;
-
-
- if( T <= 0.1 ) {
- for (i=0; i < system->N; i++)
- rvec_MakeZero( system->atoms[i].v );
-#if defined(DEBUG)
- fprintf( stderr, "no random velocities...\n" );
-#endif
- }
- else {
- for( i = 0; i < system->N; i++ ) {
- rvec_Random( system->atoms[i].v );
-
- norm = rvec_Norm_Sqr( system->atoms[i].v );
- scale = SQRT( system->reaxprm.sbp[ system->atoms[i].type ].mass *
- norm / (3.0 * K_B * T) );
-
- rvec_Scale( system->atoms[i].v, 1.0/scale, system->atoms[i].v );
-
- /*
- fprintf( stderr, "v = %f %f %f\n",
- system->atoms[i].v[0],system->atoms[i].v[1],system->atoms[i].v[2]);
- fprintf( stderr, "scale = %f\n", scale );
- fprintf( stderr, "v = %f %f %f\n",
- system->atoms[i].v[0],system->atoms[i].v[1],system->atoms[i].v[2]);
- */
- }
- }
-}
-
-
-void Init_System( reax_system *system, control_params *control,
- simulation_data *data )
-{
- int i;
- rvec dx;
-
- if( !control->restart )
- Reset_Atoms( system );
-
- Compute_Total_Mass( system, data );
-
- Compute_Center_of_Mass( system, data, stderr );
-
- /* reposition atoms */
- // just fit the atoms to the periodic box
- if( control->reposition_atoms == 0 ) {
- rvec_MakeZero( dx );
- }
- // put the center of mass to the center of the box
- else if( control->reposition_atoms == 1 ) {
- rvec_Scale( dx, 0.5, system->box.box_norms );
- rvec_ScaledAdd( dx, -1., data->xcm );
- }
- // put the center of mass to the origin
- else if( control->reposition_atoms == 2 ) {
- rvec_Scale( dx, -1., data->xcm );
- }
- else {
- fprintf( stderr, "UNKNOWN OPTION: reposition_atoms. Terminating...\n" );
- exit( UNKNOWN_OPTION );
- }
-
- for( i = 0; i < system->N; ++i ) {
- Inc_on_T3( system->atoms[i].x, dx, &(system->box) );
- /*fprintf( stderr, "%6d%2d%8.3f%8.3f%8.3f\n",
- i, system->atoms[i].type,
- system->atoms[i].x[0], system->atoms[i].x[1], system->atoms[i].x[2] );*/
- }
-
- /* Initialize velocities so that desired init T can be attained */
- if( !control->restart || (control->restart && control->random_vel) ) {
- Generate_Initial_Velocities( system, control->T_init );
- }
-
- Setup_Grid( system );
-}
-
-
-void Cuda_Init_System( reax_system *system, control_params *control,
- simulation_data *data )
-{
- int i;
- rvec dx;
-
- if( !control->restart )
- Cuda_Reset_Atoms( system );
-
- Cuda_Compute_Total_Mass( system, data );
-
- Cuda_Compute_Center_of_Mass( system, data, stderr );
-
- /* reposition atoms */
- // just fit the atoms to the periodic box
- if( control->reposition_atoms == 0 ) {
- rvec_MakeZero( dx );
- }
- // put the center of mass to the center of the box
- else if( control->reposition_atoms == 1 ) {
- rvec_Scale( dx, 0.5, system->box.box_norms );
- rvec_ScaledAdd( dx, -1., data->xcm );
- }
- // put the center of mass to the origin
- else if( control->reposition_atoms == 2 ) {
- rvec_Scale( dx, -1., data->xcm );
- }
- else {
- fprintf( stderr, "UNKNOWN OPTION: reposition_atoms. Terminating...\n" );
- exit( UNKNOWN_OPTION );
- }
-
- compute_Inc_on_T3 <<<BLOCKS_POW_2, BLOCK_SIZE>>>
- (system->d_atoms, system->N, system->d_box, dx[0], dx[1], dx[2]);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //copy back the atoms from device to the host
- copy_host_device (system->atoms, system->d_atoms, REAX_ATOM_SIZE * system->N ,
- cudaMemcpyDeviceToHost, RES_SYSTEM_ATOMS );
-
- /* Initialize velocities so that desired init T can be attained */
- if( !control->restart || (control->restart && control->random_vel) ) {
- Generate_Initial_Velocities( system, control->T_init );
- }
-
- Setup_Grid( system );
-}
-
-
-
-void Init_Simulation_Data( reax_system *system, control_params *control,
- simulation_data *data, output_controls *out_control,
- evolve_function *Evolve )
-{
-
- Reset_Simulation_Data( data );
-
- if( !control->restart )
- data->step = data->prev_steps = 0;
-
- switch( control->ensemble ) {
- case NVE:
- data->N_f = 3 * system->N;
- *Evolve = Velocity_Verlet_NVE;
- break;
-
-
- case NVT:
- data->N_f = 3 * system->N + 1;
- //control->Tau_T = 100 * data->N_f * K_B * control->T_final;
- if( !control->restart || (control->restart && control->random_vel) ) {
- data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
- data->N_f * K_B * control->T );
- data->therm.v_xi = data->therm.G_xi * control->dt;
- data->therm.v_xi_old = 0;
- data->therm.xi = 0;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "init_md: G_xi=%f Tau_T=%f E_kin=%f N_f=%f v_xi=%f\n",
- data->therm.G_xi, control->Tau_T, data->E_Kin,
- data->N_f, data->therm.v_xi );
-#endif
- }
-
- *Evolve = Velocity_Verlet_Nose_Hoover_NVT_Klein;
- break;
-
-
- case NPT: // Anisotropic NPT
- fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
- exit( UNKNOWN_OPTION );
- data->N_f = 3 * system->N + 9;
- if( !control->restart ) {
- data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
- data->N_f * K_B * control->T );
- data->therm.v_xi = data->therm.G_xi * control->dt;
- data->iso_bar.eps = 0.33333 * log(system->box.volume);
- //data->inv_W = 1. / (data->N_f*K_B*control->T*SQR(control->Tau_P));
- //Compute_Pressure( system, data, workspace );
- }
- *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
- break;
-
-
- case sNPT: // Semi-Isotropic NPT
- data->N_f = 3 * system->N + 4;
- *Evolve = Velocity_Verlet_Berendsen_SemiIsotropic_NPT;
- break;
-
-
- case iNPT: // Isotropic NPT
- data->N_f = 3 * system->N + 2;
- *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
- break;
-
- case bNVT: //berendensen NVT
- data->N_f = 3 * system->N + 1;
- *Evolve = Velocity_Verlet_Berendsen_NVT;
- break;
-
- default:
- break;
- }
-
- Compute_Kinetic_Energy( system, data );
-
- /* init timing info for the host*/
- data->timing.start = Get_Time( );
- data->timing.total = data->timing.start;
- data->timing.nbrs = 0;
- data->timing.init_forces = 0;
- data->timing.bonded = 0;
- data->timing.nonb = 0;
- data->timing.QEq = 0;
- data->timing.matvecs = 0;
-}
-
-
-void Cuda_Init_Simulation_Data( reax_system *system, control_params *control,
- simulation_data *data, output_controls *out_control,
- evolve_function *Evolve )
-{
-
- Reset_Simulation_Data( data );
-
- if( !control->restart )
- data->step = data->prev_steps = 0;
-
- switch( control->ensemble ) {
- case NVE:
- data->N_f = 3 * system->N;
- *Evolve = Cuda_Velocity_Verlet_NVE;
- break;
-
-
- case NVT:
- data->N_f = 3 * system->N + 1;
- //control->Tau_T = 100 * data->N_f * K_B * control->T_final;
- if( !control->restart || (control->restart && control->random_vel) ) {
- data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
- data->N_f * K_B * control->T );
- data->therm.v_xi = data->therm.G_xi * control->dt;
- data->therm.v_xi_old = 0;
- data->therm.xi = 0;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "init_md: G_xi=%f Tau_T=%f E_kin=%f N_f=%f v_xi=%f\n",
- data->therm.G_xi, control->Tau_T, data->E_Kin,
- data->N_f, data->therm.v_xi );
-#endif
- }
-
- *Evolve = Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein;
- break;
-
-
- case NPT: // Anisotropic NPT
- fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
- exit( UNKNOWN_OPTION );
- data->N_f = 3 * system->N + 9;
- if( !control->restart ) {
- data->therm.G_xi = control->Tau_T * (2.0 * data->E_Kin -
- data->N_f * K_B * control->T );
- data->therm.v_xi = data->therm.G_xi * control->dt;
- data->iso_bar.eps = 0.33333 * log(system->box.volume);
- //data->inv_W = 1. / (data->N_f*K_B*control->T*SQR(control->Tau_P));
- //Compute_Pressure( system, data, workspace );
- }
- *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
- break;
-
-
- case sNPT: // Semi-Isotropic NPT
- fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
- exit( UNKNOWN_OPTION );
- data->N_f = 3 * system->N + 4;
- *Evolve = Velocity_Verlet_Berendsen_SemiIsotropic_NPT;
- break;
-
-
- case iNPT: // Isotropic NPT
- fprintf( stderr, "THIS OPTION IS NOT YET IMPLEMENTED! TERMINATING...\n" );
- exit( UNKNOWN_OPTION );
- data->N_f = 3 * system->N + 2;
- *Evolve = Velocity_Verlet_Berendsen_Isotropic_NPT;
- break;
-
- case bNVT: //berendensen NVT
- data->N_f = 3 * system->N + 1;
- *Evolve = Cuda_Velocity_Verlet_Berendsen_NVT;
- break;
-
- default:
- break;
- }
-
- Cuda_Compute_Kinetic_Energy (system, data);
-
-#ifdef __BUILD_DEBUG__
- real t_E_Kin = 0;
- t_E_Kin = data->E_Kin;
-#endif
-
- copy_host_device (&data->E_Kin, &((simulation_data *)data->d_simulation_data)->E_Kin,
- REAL_SIZE, cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
- data->therm.T = (2. * data->E_Kin) / (data->N_f * K_B);
- if ( fabs(data->therm.T) < ALMOST_ZERO ) // avoid T being an absolute zero!
- data->therm.T = ALMOST_ZERO;
-
-#ifdef __BUILD_DEBUG__
- if (check_zero (t_E_Kin, data->E_Kin)){
- fprintf (stderr, "SimulationData:E_Kin does not match between host and device (%f %f) \n", t_E_Kin, data->E_Kin );
- exit (1);
- }
- //validate_data ( system, data );
-#endif
-
- /* init timing info for the host*/
- data->timing.start = Get_Time( );
- data->timing.total = data->timing.start;
- data->timing.nbrs = 0;
- data->timing.init_forces = 0;
- data->timing.bonded = 0;
- data->timing.nonb = 0;
- data->timing.QEq = 0;
- data->timing.matvecs = 0;
-
- /* init timing info for the device */
- d_timing.start = Get_Time( );
- d_timing.total = data->timing.start;
- d_timing.nbrs = 0;
- d_timing.init_forces = 0;
- d_timing.bonded = 0;
- d_timing.nonb = 0;
- d_timing.QEq = 0;
- d_timing.matvecs = 0;
-}
-
-
-void Init_Workspace( reax_system *system, control_params *control,
- static_storage *workspace )
-{
- int i;
-
- /* Allocate space for hydrogen bond list */
- workspace->hbond_index = (int *) malloc( system->N * sizeof( int ) );
-
- /* bond order related storage */
- workspace->total_bond_order = (real *) malloc( system->N * sizeof( real ) );
- workspace->Deltap = (real *) malloc( system->N * sizeof( real ) );
- workspace->Deltap_boc = (real *) malloc( system->N * sizeof( real ) );
- workspace->dDeltap_self = (rvec *) malloc( system->N * sizeof( rvec ) );
-
- workspace->Delta = (real *) malloc( system->N * sizeof( real ) );
- workspace->Delta_lp = (real *) malloc( system->N * sizeof( real ) );
- workspace->Delta_lp_temp = (real *) malloc( system->N * sizeof( real ) );
- workspace->dDelta_lp = (real *) malloc( system->N * sizeof( real ) );
- workspace->dDelta_lp_temp = (real *) malloc( system->N * sizeof( real ) );
- workspace->Delta_e = (real *) malloc( system->N * sizeof( real ) );
- workspace->Delta_boc = (real *) malloc( system->N * sizeof( real ) );
- workspace->nlp = (real *) malloc( system->N * sizeof( real ) );
- workspace->nlp_temp = (real *) malloc( system->N * sizeof( real ) );
- workspace->Clp = (real *) malloc( system->N * sizeof( real ) );
- workspace->CdDelta = (real *) malloc( system->N * sizeof( real ) );
- workspace->vlpex = (real *) malloc( system->N * sizeof( real ) );
-
- /* QEq storage */
- //workspace->H = NULL;
- //workspace->L = NULL;
- //workspace->U = NULL;
- //
- workspace->H.start = NULL;
- workspace->L.start = NULL;
- workspace->U.start = NULL;
-
- workspace->H.entries = NULL;
- workspace->L.entries = NULL;
- workspace->U.entries = NULL;
-
- workspace->droptol = (real *) calloc( system->N, sizeof( real ) );
- workspace->w = (real *) calloc( system->N, sizeof( real ) );
- workspace->Hdia_inv = (real *) calloc( system->N, sizeof( real ) );
- workspace->b = (real *) calloc( system->N * 2, sizeof( real ) );
- workspace->b_s = (real *) calloc( system->N, sizeof( real ) );
- workspace->b_t = (real *) calloc( system->N, sizeof( real ) );
- workspace->b_prc = (real *) calloc( system->N * 2, sizeof( real ) );
- workspace->b_prm = (real *) calloc( system->N * 2, sizeof( real ) );
- workspace->s_t = (real *) calloc( system->N * 2, sizeof( real ) );
- workspace->s = (real *) calloc( 5 * system->N, sizeof( real ) );
- workspace->t = (real *) calloc( 5 * system->N, sizeof( real ) );
- // workspace->s_old = (real *) calloc( system->N, sizeof( real ) );
- // workspace->t_old = (real *) calloc( system->N, sizeof( real ) );
- // workspace->s_oldest = (real *) calloc( system->N, sizeof( real ) );
- // workspace->t_oldest = (real *) calloc( system->N, sizeof( real ) );
-
- for( i = 0; i < system->N; ++i ) {
- workspace->Hdia_inv[i] = 1./system->reaxprm.sbp[system->atoms[i].type].eta;
- workspace->b_s[i] = -system->reaxprm.sbp[ system->atoms[i].type ].chi;
- workspace->b_t[i] = -1.0;
-
- workspace->b[i] = -system->reaxprm.sbp[ system->atoms[i].type ].chi;
- workspace->b[i+system->N] = -1.0;
- }
-
- /* GMRES storage */
- workspace->y = (real *) calloc( RESTART+1, sizeof( real ) );
- workspace->z = (real *) calloc( RESTART+1, sizeof( real ) );
- workspace->g = (real *) calloc( RESTART+1, sizeof( real ) );
- workspace->hs = (real *) calloc( RESTART+1, sizeof( real ) );
- workspace->hc = (real *) calloc( RESTART+1, sizeof( real ) );
-
- workspace->rn = (real *) calloc( (RESTART+1)*system->N*2, sizeof( real) );
- workspace->v = (real *) calloc( (RESTART+1)*system->N, sizeof( real) );
- workspace->h = (real *) calloc( (RESTART+1)*(RESTART+1), sizeof( real) );
-
- /* CG storage */
- workspace->r = (real *) calloc( system->N, sizeof( real ) );
- workspace->d = (real *) calloc( system->N, sizeof( real ) );
- workspace->q = (real *) calloc( system->N, sizeof( real ) );
- workspace->p = (real *) calloc( system->N, sizeof( real ) );
-
-
- /* integrator storage */
- workspace->a = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_old = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->v_const = (rvec *) malloc( system->N * sizeof( rvec ) );
-
-
- /* storage for analysis */
- if( control->molec_anal || control->diffusion_coef )
- {
- workspace->mark = (int *) calloc( system->N, sizeof(int) );
- workspace->old_mark = (int *) calloc( system->N, sizeof(int) );
- }
- else
- workspace->mark = workspace->old_mark = NULL;
-
- if( control->diffusion_coef )
- workspace->x_old = (rvec *) calloc( system->N, sizeof( rvec ) );
- else workspace->x_old = NULL;
-
-
-#ifdef TEST_FORCES
- workspace->dDelta = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_ele = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_vdw = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_bo = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_be = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_lp = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_ov = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_un = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_ang = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_coa = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_pen = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_hb = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_tor = (rvec *) malloc( system->N * sizeof( rvec ) );
- workspace->f_con = (rvec *) malloc( system->N * sizeof( rvec ) );
-#endif
-
- workspace->realloc.num_far = -1;
- workspace->realloc.Htop = -1;
- workspace->realloc.hbonds = -1;
- workspace->realloc.bonds = -1;
- workspace->realloc.num_3body = -1;
- workspace->realloc.gcell_atoms = -1;
-
- Reset_Workspace( system, workspace );
-}
-
-void compare_far_neighbors (int *test, int *start, int *end, far_neighbor_data *data, list *slist, int N)
-{
- int index = 0;
- int count = 0;
- int jicount = 0;
- int end_index, gpu_index, gpu_end, k;
- far_neighbor_data gpu, cpu;
-
- /*
- for (int i = 0; i < N ; i++ )
- {
- if (test[i] != start[i]) {
- fprintf (stderr, "start index does not match \n");
- exit (0);
- }
-
- if (test[i+1] != (end[i]) ){
- fprintf (stderr, "end index does not match for atom %d (cpu: %d gpu: %d) \n", i, test[i+1], end[i]);
- exit (0);
- }
- }
- */
-
-
- for (int i = 0; i < N; i++){
- index = Start_Index (i, slist);
- //fprintf (stderr, "GPU : Neighbors of atom --> %d (start: %d , end: %d )\n", i, start[i], end[i]);
-
-
- for (int j = start[i]; j < end[i]; j++){
- gpu = data[j];
-
- if (i < data[j].nbr) continue;
- /*
- if (i < data[j].nbr) {
- //fprintf (stderr, " atom %d and neighbor %d @ index %d\n", i, data[j].nbr, j);
- int src = data[j].nbr;
- int dest = i;
- int x;
-
-
- for (x = start[src]; x < end[src]; x++) {
- if (data[x].nbr != dest) continue;
-
- gpu = data[x];
- cpu = data[j];
-
- if ( (gpu.d != cpu.d) ||
- (cpu.dvec[0] != gpu.dvec[0]) || (cpu.dvec[1] != gpu.dvec[1]) || (cpu.dvec[2] != gpu.dvec[2]) ||
- (cpu.rel_box[0] != gpu.rel_box[0]) || (cpu.rel_box[1] != gpu.rel_box[1]) || (cpu.rel_box[2] != gpu.rel_box[2])) {
- fprintf (stderr, " atom %d neighbor %d (%f, %d, %d, %d - %f %f %f) \n", i, data[j].nbr,
- data[j].d,
- data[j].rel_box[0],
- data[j].rel_box[1],
- data[j].rel_box[2],
- data[j].dvec[0],
- data[j].dvec[1],
- data[j].dvec[2]
- );
- fprintf (stderr, " atom %d neighbor %d (%f, %d, %d, %d - %f %f %f) \n", data[j].nbr, data[x].nbr,
- data[x].d,
- data[x].rel_box[0],
- data[x].rel_box[1],
- data[x].rel_box[2],
- data[x].dvec[0],
- data[x].dvec[1],
- data[x].dvec[2]
- );
- jicount++;
- }
- break;
- }
-
- if (x >= end[src]) {
- fprintf (stderr, "could not find the neighbor duplicate data for ij (%d %d)\n", i, src );
- exit (0);
- }
-
- continue;
- }
- */
-
- cpu = slist->select.far_nbr_list[index];
- //if ( (gpu.nbr != cpu.nbr) || (gpu.d != cpu.d) ){
- //if ( (gpu->d != cpu->d) ){
- if ( (gpu.nbr != cpu.nbr) || (gpu.d != cpu.d) ||
- (cpu.dvec[0] != gpu.dvec[0]) || (cpu.dvec[1] != gpu.dvec[1]) || (cpu.dvec[2] != gpu.dvec[2]) ||
- (cpu.rel_box[0] != gpu.rel_box[0]) || (cpu.rel_box[1] != gpu.rel_box[1]) || (cpu.rel_box[2] != gpu.rel_box[2])) {
- //if ( (gpu.dvec[0] != i) || (gpu.dvec[1] != i) ||(gpu.dvec[2] != i) ||
- // (gpu.rel_box[0] != i) || (gpu.rel_box[1] != i) ||(gpu.rel_box[2] != i) ) {
- //if (memcmp (&gpu, &cpu, FAR_NEIGHBOR_SIZE - RVEC_SIZE - INT_SIZE )){
-
- fprintf (stderr, "GPU:atom --> %d (s: %d , e: %d, i: %d ) (%d %d %d) \n", i, start[i], end[i], j, gpu.rel_box[0], gpu.rel_box[1], gpu.rel_box[2] );
- fprintf (stderr, "CPU:atom --> %d (s: %d , e: %d, i: %d )\n", i, Start_Index(i, slist), End_Index (i, slist), index);
-
- /*
- fprintf (stdout, "Far neighbors does not match atom: %d \n", i );
- fprintf (stdout, "neighbor %d , %d \n", cpu.nbr, gpu.nbr);
- fprintf (stdout, "d %f , %f \n", slist->select.far_nbr_list[index].d, data[j].d);
- fprintf (stdout, "dvec (%f %f %f) (%f %f %f) \n",
- cpu.dvec[0], cpu.dvec[1], cpu.dvec[2],
- gpu.dvec[0], gpu.dvec[1], gpu.dvec[2] );
-
- fprintf (stdout, "ivec (%d %d %d) (%d %d %d) \n",
- cpu.rel_box[0], cpu.rel_box[1], cpu.rel_box[2],
- gpu.rel_box[0], gpu.rel_box[1], gpu.rel_box[2] );
-
- */
- count ++;
- }
-
- //fprintf (stderr, "GPU (neighbor %d , d %d )\n", gpu->nbr, gpu->d);
- index ++;
- }
-
- if (index != End_Index (i, slist))
- {
- fprintf (stderr, "End index does not match for atom --> %d end index (%d) Cpu (%d, %d ) gpu (%d, %d)\n", i, index, Start_Index (i, slist), End_Index(i, slist),
- start[i], end[i]);
- exit (10);
- }
- }
-
- fprintf (stderr, "Far neighbors MATCH between CPU and GPU -->%d reverse %d \n", count, jicount);
-
- /*
- for (int i = 0; i < N; i++)
- {
- index = Start_Index (i, slist);
- end_index = End_Index (i, slist);
-
- gpu_index = start[i];
- gpu_end = end[i];
- for (int j = index; j < end_index; j++)
- {
- far_neighbor_data *cpu = &slist->select.far_nbr_list[j];
- far_neighbor_data *gpu;
-
- for (k = gpu_index; k < gpu_end; k++) {
- gpu = &data[k];
- if (gpu->nbr == cpu->nbr) break;
- }
-
- if (k == gpu_end) { fprintf (stderr, " could not find neighbor for atom %d \n", i); exit (1); }
-
- if ( (gpu->nbr != cpu->nbr) || (gpu->d != cpu->d) ||
- ((cpu->dvec[0] || gpu->dvec[0]) || (cpu->dvec[1] || gpu->dvec[1]) || (cpu->dvec[2] || gpu->dvec[2])) ||
- ((cpu->rel_box[0] || gpu->rel_box[0]) || (cpu->rel_box[1] || gpu->rel_box[1]) || (cpu->rel_box[2] || gpu->rel_box[2])) ) {
-
- fprintf (stderr, "Far neighbors does not match atom: %d \n", i );
- fprintf (stderr, "neighbor %d , %d \n", cpu->nbr, gpu->nbr);
- fprintf (stderr, "d %d , %d \n", cpu->d, gpu->d);
- fprintf (stderr, "dvec (%f %f %f) (%f %f %f) \n",
- cpu->dvec[0], cpu->dvec[1], cpu->dvec[2],
- gpu->dvec[0], gpu->dvec[1], gpu->dvec[2] );
-
- fprintf (stderr, "ivec (%d %d %d) (%d %d %d) \n",
- cpu->rel_box[0], cpu->rel_box[1], cpu->rel_box[2],
- gpu->rel_box[0], gpu->rel_box[1], gpu->rel_box[2] );
- fprintf (stderr, "GPU start %d GPU End %d \n", gpu_index, gpu_end );
-
- exit (1);
- }
- }
- }
-
- */
- }
-
- int Estimate_Device_Matrix (reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- int *indices, *Htop;
- list *far_nbrs = dev_lists + FAR_NBRS;
- int max_sparse_entries = 0;
- real t1, t2;
-
- indices = (int *) scratch;
- cuda_memset ( indices, 0, INT_SIZE * system->N, RES_SCRATCH );
-
- t1 = Get_Time ();
-
- Estimate_Sparse_Matrix_Entries <<<BLOCKS, BLOCK_SIZE>>>
- ( system->d_atoms, (control_params *)control->d_control,
- (simulation_data *)data->d_simulation_data, (simulation_box *)system->d_box,
- *far_nbrs, system->N, indices );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- t2 = Get_Timing_Info ( t1 );
-
- //fprintf (stderr, " Time to estimate sparse matrix entries --- > %f \n", t2 );
-
- Htop = (int *) malloc (INT_SIZE * (system->N + 1));
- memset (Htop, 0, INT_SIZE * (system->N + 1));
- copy_host_device (Htop, indices, system->N * INT_SIZE, cudaMemcpyDeviceToHost, __LINE__);
-
- for (int i = 0; i < system->N; i++)
- {
- if (max_sparse_entries < Htop[i]) {
- max_sparse_entries = Htop[i];
- }
- }
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Max sparse entries for this run are ---> %d \n", max_sparse_entries );
-#endif
-
- return max_sparse_entries * SAFE_ZONE;
- //return max_sparse_entries;
- }
-
- void Allocate_Device_Matrix (reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
-
- //Allocate space for the sparse Matrix entries here.
- system->max_sparse_matrix_entries =
- Estimate_Device_Matrix (system, control, data, workspace, lists, out_control );
- dev_workspace->H.n = system->N ;
- dev_workspace->H.m = system->N * system->max_sparse_matrix_entries;
- Cuda_Init_Sparse_Matrix (&dev_workspace->H, system->max_sparse_matrix_entries * system->N, system->N );
-
-#ifdef __CUDA_MEM__
- fprintf( stderr, "Device memory allocated: sparse matrix= %ld (MB)\n",
- system->max_sparse_matrix_entries * system->N * sizeof(sparse_matrix_entry) / (1024*1024) );
-#endif
- }
-
- void Cuda_Init_Lists( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- int i, num_nbrs, num_hbonds, num_bonds, num_3body, Htop;
- int *hb_top, *bond_top;
-
- real t_start, t_elapsed;
-
- grid *g = &( system->g );
- int *d_indices = (int *) scratch;
- int total = g->ncell[0] * g->ncell[1] * g->ncell[2];
-
- cuda_memset ( d_indices, 0, INT_SIZE * system->N, RES_SCRATCH );
-
-#ifdef __BUILD_DEBUG__
- for (int i = 0; i < g->max_nbrs; i ++) {
- if ((g->nbrs[i][0] >= g->ncell[0]) ||
- (g->nbrs[i][1] >= g->ncell[1]) ||
- (g->nbrs[i][2] >= g->ncell[2]) ) {
- fprintf (stderr, " Grid Incorrectly built.... \n");
- exit (1);
- }
-
- }
-#endif
-
- dim3 blockspergrid (system->g.ncell[0], system->g.ncell[1], system->g.ncell[2]);
- dim3 threadsperblock (system->g.max_atoms);
-
-#ifdef __BUILD_DEBUG__
- fprintf (stderr, "Blocks per grid (%d %d %d)\n", system->g.ncell[0], system->g.ncell[1], system->g.ncell[2]);
- fprintf (stderr, "Estimate Num Neighbors with threads per block as %d \n", system->d_g.max_atoms);
- fprintf (stderr, "Max nbrs %d \n", system->d_g.max_nbrs);
-#endif
-
-
- //First Bin atoms and they sync the host and the device for the grid.
- //This will copy the atoms from host to device.
- Cuda_Bin_Atoms (system, workspace);
- Sync_Host_Device (&system->g, &system->d_g, cudaMemcpyHostToDevice );
-
- Estimate_NumNeighbors <<<blockspergrid, threadsperblock >>>
- (system->d_atoms, system->d_g, system->d_box,
- (control_params *)control->d_control, d_indices);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- int *nbrs_indices = (int *) malloc( INT_SIZE * (system->N+1) );
- memset (nbrs_indices , 0, INT_SIZE * (system->N + 1));
-
- nbrs_indices [0] = 0;
- copy_host_device (&nbrs_indices [1], d_indices, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
-
- for (int i = 1; i <= system->N; i++)
- nbrs_indices [i] += nbrs_indices [i-1];
-
- num_nbrs = nbrs_indices [system->N] ;
- system->num_nbrs = num_nbrs;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Total neighbors %d \n", nbrs_indices[system->N]);
- fprintf (stderr, "Corrected Total neighbors %d \n", num_nbrs);
-#endif
-
-
- list *far_nbrs = (dev_lists + FAR_NBRS);
- if( !Make_List(system->N, num_nbrs, TYP_FAR_NEIGHBOR, far_nbrs, TYP_DEVICE) ) {
- fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
- exit( INIT_ERR );
- }
-
-#ifdef __CUDA_MEM__
- fprintf( stderr, "Device memory allocated: far_nbrs = %ld (MB)\n",
- num_nbrs * sizeof(far_neighbor_data) / (1024*1024) );
-#endif
-
- copy_host_device (nbrs_indices, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyHostToDevice, __LINE__ );
- copy_host_device (nbrs_indices, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyHostToDevice, __LINE__ );
- Cuda_Generate_Neighbor_Lists (system, workspace, control, false);
-
-#ifdef __BUILD_DEBUG__
-
- int *end = (int *)malloc (sizeof (int) * system->N);
- int *start = (int *) malloc (sizeof (int) * system->N );
-
- copy_host_device (start, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, 0);
- copy_host_device (end, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, 0);
-
- far_neighbor_data *far_data = (far_neighbor_data *)
- malloc (FAR_NEIGHBOR_SIZE * num_nbrs);
- copy_host_device (far_data, far_nbrs->select.far_nbr_list,
- FAR_NEIGHBOR_SIZE * num_nbrs, cudaMemcpyDeviceToHost, 0);
-
- compare_far_neighbors (nbrs_indices, start, end, far_data, *lists + FAR_NBRS, system->N);
-
- free (start);
- free (end);
-#endif
-
- int *output, size;
- size = INT_SIZE * 2 * system->N + 2;
- output = (int *) malloc (size);
- Cuda_Estimate_Storage_Sizes (system, control, output);
-
- Htop = output[0];
- num_3body = output[1];
- hb_top = &output[ 2 ];
- bond_top = &output[ 2 + system->N ];
-
-#ifdef __DEBUG_CUDA__
- int max_hbonds = 0;
- int min_hbonds = 1000;
- int max_bonds = 0;
- int min_bonds = 1000;
- for (int i = 0; i < system->N; i++) {
- if ( max_hbonds < hb_top[i])
- max_hbonds = hb_top[i];
- if (min_hbonds > hb_top[i])
- min_hbonds = hb_top[i];
-
- if (max_bonds < bond_top [i])
- max_bonds = bond_top[i];
- if (min_bonds > bond_top[i])
- min_bonds = bond_top[i];
- }
-
- fprintf (stderr, "Max Hbonds %d min Hbonds %d \n", max_hbonds, min_hbonds );
- fprintf (stderr, "Max bonds %d min bonds %d \n", max_bonds, min_bonds );
- fprintf (stderr, "Device HTop --> %d and num_3body --> %d \n", Htop, num_3body );
-#endif
-
- Allocate_Device_Matrix (system, control, data, workspace, lists, out_control );
-
- dev_workspace->num_H = 0;
-
- if( control->hb_cut > 0 ) {
-
- int *hbond_index = (int *) malloc ( INT_SIZE * system->N );
- // init H indexes
- num_hbonds = 0;
- for( i = 0; i < system->N; ++i )
- if( system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 1 ||
- system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 2 ) // H atom
- //hbond_index[i] = workspace->num_H++;
- hbond_index[i] = num_hbonds ++;
- else
- hbond_index[i] = -1;
-
- copy_host_device (hbond_index, dev_workspace->hbond_index,
- system->N * INT_SIZE, cudaMemcpyHostToDevice, RES_STORAGE_HBOND_INDEX );
- dev_workspace->num_H = num_hbonds;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Device num_H --> %d \n", dev_workspace->num_H );
-#endif
-
- Cuda_Allocate_HBond_List( system->N, dev_workspace->num_H, dev_workspace->hbond_index,
- hb_top, (dev_lists+HBONDS) );
- num_hbonds = hb_top[system->N-1];
- system->num_hbonds = num_hbonds;
-
-#ifdef __CUDA_MEM__
- fprintf (stderr, "Device memory allocated: Hydrogen Bonds list: %ld (MB) \n",
- sizeof (hbond_data) * num_hbonds / (1024*1024));
-#endif
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Device Total number of HBonds --> %d \n", num_hbonds );
-#endif
-
- free (hbond_index);
- }
-
- // bonds list
- Cuda_Allocate_Bond_List( system->N, bond_top, dev_lists+BONDS );
- num_bonds = bond_top[system->N-1];
- system->num_bonds = num_bonds;
-
-#ifdef __CUDA_MEM__
- fprintf (stderr, "Device memory allocated: Bonds list: %ld (MB) \n",
- sizeof (bond_data) * num_bonds / (1024*1024));
-#endif
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Device Total Bonds --> %d \n", num_bonds );
-#endif
-
- // system->max_thb_intrs = num_3body;
- // 3bodies list
- //if(!Make_List(num_bonds, num_bonds * MAX_THREE_BODIES, TYP_THREE_BODY, dev_lists + THREE_BODIES, TYP_DEVICE)) {
- // fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
- // exit( INIT_ERR );
- //}
-
- //fprintf( stderr, "***memory allocated: three_body = %ldMB\n",
- // num_bonds * MAX_THREE_BODIES *sizeof(three_body_interaction_data) / (1024*1024) );
- //fprintf (stderr, "size of (three_body_interaction_data) : %d \n", sizeof (three_body_interaction_data));
-
- //Free local resources
- free (output);
- free (nbrs_indices);
- }
-
-
- void Init_Lists( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- int i, num_nbrs, num_hbonds, num_bonds, num_3body, Htop;
- int *hb_top, *bond_top;
-
- real t_start, t_elapsed;
-
- num_nbrs = Estimate_NumNeighbors( system, control, workspace, lists );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Serial NumNeighbors ---> %d \n", num_nbrs);
-#endif
-
- if( !Make_List(system->N, num_nbrs, TYP_FAR_NEIGHBOR, (*lists)+FAR_NBRS) ) {
- fprintf(stderr, "Problem in initializing far nbrs list. Terminating!\n");
- exit( INIT_ERR );
- }
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "memory allocated: far_nbrs = %ldMB\n",
- num_nbrs * sizeof(far_neighbor_data) / (1024*1024) );
-#endif
-
- t_start = Get_Time ();
- Generate_Neighbor_Lists(system,control,data,workspace,lists,out_control);
- t_elapsed = Get_Timing_Info ( t_start );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Timing Generate Neighbors %lf \n", t_elapsed );
-#endif
-
- Htop = 0;
- hb_top = (int*) calloc( system->N, sizeof(int) );
- bond_top = (int*) calloc( system->N, sizeof(int) );
- num_3body = 0;
- Estimate_Storage_Sizes( system, control, lists,
- &Htop, hb_top, bond_top, &num_3body );
-
- Allocate_Matrix( &(workspace->H), system->N, Htop );
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "estimated storage - Htop: %d\n", Htop );
- fprintf( stderr, "memory allocated: H = %ldMB\n",
- Htop * sizeof(sparse_matrix_entry) / (1024*1024) );
-#endif
-
- workspace->num_H = 0;
- if( control->hb_cut > 0 ) {
- /* init H indexes */
- for( i = 0; i < system->N; ++i )
- if( system->reaxprm.sbp[ system->atoms[i].type ].p_hbond == 1 ) // H atom
- workspace->hbond_index[i] = workspace->num_H++;
- else workspace->hbond_index[i] = -1;
-
- Allocate_HBond_List( system->N, workspace->num_H, workspace->hbond_index,
- hb_top, (*lists)+HBONDS );
- num_hbonds = hb_top[system->N-1];
-
-#ifdef __DEBUG_CUDA__
- fprintf( stderr, "Serial num_hbonds: %d\n", num_hbonds );
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "estimated storage - num_hbonds: %d\n", num_hbonds );
- fprintf( stderr, "memory allocated: hbonds = %ldMB\n",
- num_hbonds * sizeof(hbond_data) / (1024*1024) );
-#endif
- }
-
- /* bonds list */
- Allocate_Bond_List( system->N, bond_top, (*lists)+BONDS );
- num_bonds = bond_top[system->N-1];
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "estimated storage - num_bonds: %d\n", num_bonds );
- fprintf( stderr, "memory allocated: bonds = %ldMB\n",
- num_bonds * sizeof(bond_data) / (1024*1024) );
-#endif
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " host num_3body : %d \n", num_3body);
- fprintf (stderr, " host num_bonds : %d \n", num_bonds);
-#endif
-
- /* 3bodies list */
- if(!Make_List(num_bonds, num_3body, TYP_THREE_BODY, (*lists)+THREE_BODIES)) {
- fprintf( stderr, "Problem in initializing angles list. Terminating!\n" );
- exit( INIT_ERR );
- }
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "estimated storage - num_3body: %d\n", num_3body );
- fprintf( stderr, "memory allocated: 3-body = %ldMB\n",
- num_3body * sizeof(three_body_interaction_data) / (1024*1024) );
-#endif
-#ifdef TEST_FORCES
- if(!Make_List( system->N, num_bonds * 8, TYP_DDELTA, (*lists) + DDELTA )) {
- fprintf( stderr, "Problem in initializing dDelta list. Terminating!\n" );
- exit( INIT_ERR );
- }
-
- if( !Make_List( num_bonds, num_bonds*MAX_BONDS*3, TYP_DBO, (*lists)+DBO ) ) {
- fprintf( stderr, "Problem in initializing dBO list. Terminating!\n" );
- exit( INIT_ERR );
- }
-#endif
-
- free( hb_top );
- free( bond_top );
- }
-
-
- void Init_Out_Controls(reax_system *system, control_params *control,
- static_storage *workspace, output_controls *out_control)
- {
- char temp[1000];
-
- /* Init trajectory file */
- if( out_control->write_steps > 0 ) {
- strcpy( temp, control->sim_name );
- strcat( temp, ".trj" );
- out_control->trj = fopen( temp, "w" );
- out_control->write_header( system, control, workspace, out_control );
- }
-
- if( out_control->energy_update_freq > 0 ) {
- /* Init out file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".out" );
- out_control->out = fopen( temp, "w" );
- fprintf( out_control->out, "%-6s%16s%16s%16s%11s%11s%13s%13s%13s\n",
- "step", "total energy", "poten. energy", "kin. energy",
- "temp.", "target", "volume", "press.", "target" );
- fflush( out_control->out );
-
- /* Init potentials file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".pot" );
- out_control->pot = fopen( temp, "w" );
- fprintf( out_control->pot,
- "%-6s%13s%13s%13s%13s%13s%13s%13s%13s%13s%13s%13s\n",
- "step", "ebond", "eatom", "elp", "eang", "ecoa", "ehb",
- "etor", "econj", "evdw","ecoul", "epol" );
- fflush( out_control->pot );
-
- /* Init log file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".log" );
- out_control->log = fopen( temp, "w" );
- fprintf( out_control->log, "%-6s%10s%10s%10s%10s%10s%10s%10s\n",
- "step", "total", "neighbors", "init", "bonded",
- "nonbonded", "QEq", "matvec" );
- }
-
- /* Init pressure file */
- if( control->ensemble == NPT ||
- control->ensemble == iNPT ||
- control->ensemble == sNPT ) {
- strcpy( temp, control->sim_name );
- strcat( temp, ".prs" );
- out_control->prs = fopen( temp, "w" );
- fprintf( out_control->prs, "%-6s%13s%13s%13s%13s%13s%13s%13s%13s\n",
- "step", "norm_x", "norm_y", "norm_z",
- "press_x", "press_y", "press_z", "target_p", "volume" );
- fflush( out_control->prs );
- }
-
- /* Init molecular analysis file */
- if( control->molec_anal ) {
- sprintf( temp, "%s.mol", control->sim_name );
- out_control->mol = fopen( temp, "w" );
- if( control->num_ignored ) {
- sprintf( temp, "%s.ign", control->sim_name );
- out_control->ign = fopen( temp, "w" );
- }
- }
-
- /* Init electric dipole moment analysis file */
- if( control->dipole_anal ) {
- strcpy( temp, control->sim_name );
- strcat( temp, ".dpl" );
- out_control->dpl = fopen( temp, "w" );
- fprintf( out_control->dpl,
- "Step Molecule Count Avg. Dipole Moment Norm\n" );
- fflush( out_control->dpl );
- }
-
- /* Init diffusion coef analysis file */
- if( control->diffusion_coef ) {
- strcpy( temp, control->sim_name );
- strcat( temp, ".drft" );
- out_control->drft = fopen( temp, "w" );
- fprintf( out_control->drft, "Step Type Count Avg Squared Disp\n" );
- fflush( out_control->drft );
- }
-
-
-#ifdef TEST_ENERGY
- /* open bond energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ebond" );
- out_control->ebond = fopen( temp, "w" );
-
- /* open lone-pair energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".elp" );
- out_control->elp = fopen( temp, "w" );
-
- /* open overcoordination energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".eov" );
- out_control->eov = fopen( temp, "w" );
-
- /* open undercoordination energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".eun" );
- out_control->eun = fopen( temp, "w" );
-
- /* open angle energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".eval" );
- out_control->eval = fopen( temp, "w" );
-
- /* open penalty energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".epen" );
- out_control->epen = fopen( temp, "w" );
-
- /* open coalition energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ecoa" );
- out_control->ecoa = fopen( temp, "w" );
-
- /* open hydrogen bond energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ehb" );
- out_control->ehb = fopen( temp, "w" );
-
- /* open torsion energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".etor" );
- out_control->etor = fopen( temp, "w" );
-
- /* open conjugation energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".econ" );
- out_control->econ = fopen( temp, "w" );
-
- /* open vdWaals energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".evdw" );
- out_control->evdw = fopen( temp, "w" );
-
- /* open coulomb energy file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ecou" );
- out_control->ecou = fopen( temp, "w" );
-#endif
-
-
-#ifdef TEST_FORCES
- /* open bond orders file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fbo" );
- out_control->fbo = fopen( temp, "w" );
-
- /* open bond orders derivatives file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fdbo" );
- out_control->fdbo = fopen( temp, "w" );
-
- /* open bond forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fbond" );
- out_control->fbond = fopen( temp, "w" );
-
- /* open lone-pair forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".flp" );
- out_control->flp = fopen( temp, "w" );
-
- /* open overcoordination forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fatom" );
- out_control->fatom = fopen( temp, "w" );
-
- /* open angle forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".f3body" );
- out_control->f3body = fopen( temp, "w" );
-
- /* open hydrogen bond forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fhb" );
- out_control->fhb = fopen( temp, "w" );
-
- /* open torsion forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".f4body" );
- out_control->f4body = fopen( temp, "w" );
-
- /* open nonbonded forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".fnonb" );
- out_control->fnonb = fopen( temp, "w" );
-
- /* open total force file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ftot" );
- out_control->ftot = fopen( temp, "w" );
-
- /* open coulomb forces file */
- strcpy( temp, control->sim_name );
- strcat( temp, ".ftot2" );
- out_control->ftot2 = fopen( temp, "w" );
-#endif
-
-
- /* Error handling */
- /* if ( out_control->out == NULL || out_control->pot == NULL ||
- out_control->log == NULL || out_control->mol == NULL ||
- out_control->dpl == NULL || out_control->drft == NULL ||
- out_control->pdb == NULL )
- {
- fprintf( stderr, "FILE OPEN ERROR. TERMINATING..." );
- exit( CANNOT_OPEN_OUTFILE );
- }*/
- }
-
-
- void Initialize(reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace, list **lists,
- output_controls *out_control, evolve_function *Evolve)
- {
- Randomize();
-
- Init_System( system, control, data );
-
- Init_Simulation_Data( system, control, data, out_control, Evolve );
-
- Init_Workspace( system, control, workspace );
-
- Init_Lists( system, control, data, workspace, lists, out_control );
-
- Init_Out_Controls( system, control, workspace, out_control );
-
- /* These are done in forces.c, only forces.c can see all those functions */
- Init_Bonded_Force_Functions( control );
-#ifdef TEST_FORCES
- Init_Force_Test_Functions( );
-#endif
-
- if( control->tabulate )
- Make_LR_Lookup_Table( system, control );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "data structures have been initialized...\n" );
-#endif
- }
-
- void Cuda_Initialize(reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace, list **lists,
- output_controls *out_control, evolve_function *Evolve)
- {
- Randomize ();
-
- Cuda_Init_Scratch ();
-
- //System
- Cuda_Init_System (system);
- Sync_Host_Device ( system, cudaMemcpyHostToDevice );
- Cuda_Init_System (system, control, data );
-
- //Simulation Data
- copy_host_device (system->atoms, system->d_atoms, REAX_ATOM_SIZE * system->N ,
- cudaMemcpyHostToDevice, RES_SYSTEM_ATOMS );
- Cuda_Init_Simulation_Data (data);
- //Sync_Host_Device (data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice);
- Cuda_Init_Simulation_Data( system, control, data, out_control, Evolve );
- Sync_Host_Device (data, (simulation_data *)data->d_simulation_data, cudaMemcpyHostToDevice);
-
- //static storage
- Cuda_Init_Workspace_System ( system, dev_workspace );
- Cuda_Init_Workspace ( system, control, dev_workspace );
- Cuda_Init_Workspace_Device (workspace);
-
- //control
- Cuda_Init_Control (control);
-
- //Grid
- Cuda_Init_Grid (&system->g, &system->d_g );
-
- //lists
- Cuda_Init_Lists (system, control, data, workspace, lists, out_control );
-
- Init_Out_Controls( system, control, workspace, out_control );
-
- if( control->tabulate ) {
- real start, end;
- start = Get_Time ();
- Make_LR_Lookup_Table( system, control );
- copy_LR_table_to_device (system, control );
- end = Get_Timing_Info ( start );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Done copying the LR table to the device ---> %f \n", end );
-#endif
- }
- }
diff --git a/PuReMD-GPU/src/init_md.h b/PuReMD-GPU/src/init_md.h
index 65e12348eef4830a6eede11c881b9f46ffa283e8..8c23806594a8f2b107ddb884efbf68e7b5fe27ff 100644
--- a/PuReMD-GPU/src/init_md.h
+++ b/PuReMD-GPU/src/init_md.h
@@ -23,10 +23,22 @@
#include "mytypes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Initialize( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls*, evolve_function* );
+ static_storage*, list**, output_controls*, evolve_function* );
+
+void Generate_Initial_Velocities(reax_system *, real );
+
+void Init_Out_Controls(reax_system *, control_params *, static_storage *,
+ output_controls *);
+
+#ifdef __cplusplus
+}
+#endif
-void Cuda_Initialize( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls*, evolve_function* );
#endif
diff --git a/PuReMD-GPU/src/integrate.cu b/PuReMD-GPU/src/integrate.c
similarity index 65%
rename from PuReMD-GPU/src/integrate.cu
rename to PuReMD-GPU/src/integrate.c
index d079028653d79dfeb5117fa6d95f33b700dee5b1..482a9c89a302c052e9ac44ae2de446c61b1c6a3e 100644
--- a/PuReMD-GPU/src/integrate.cu
+++ b/PuReMD-GPU/src/integrate.c
@@ -19,6 +19,7 @@
----------------------------------------------------------------------*/
#include "integrate.h"
+
#include "allocate.h"
#include "box.h"
#include "forces.h"
@@ -32,10 +33,6 @@
#include "vector.h"
#include "list.h"
-#include "cuda_utils.h"
-#include "reduction.h"
-#include "validation.h"
-
void Velocity_Verlet_NVE(reax_system* system, control_params* control,
simulation_data *data, static_storage *workspace,
@@ -49,6 +46,7 @@ void Velocity_Verlet_NVE(reax_system* system, control_params* control,
dt_sqr = SQR(dt);
steps = data->step - data->prev_steps;
renbr = (steps % control->reneighbor == 0);
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "step%d: ", data->step );
#endif
@@ -63,6 +61,7 @@ void Velocity_Verlet_NVE(reax_system* system, control_params* control,
rvec_ScaledAdd( system->atoms[i].v,
0.5 * dt * -F_CONV * inv_m, system->atoms[i].f );
}
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet1 - ");
#endif
@@ -70,106 +69,25 @@ void Velocity_Verlet_NVE(reax_system* system, control_params* control,
Reallocate( system, workspace, lists, renbr );
Reset( system, control, data, workspace, lists );
if( renbr )
+ {
Generate_Neighbor_Lists( system, control, data, workspace,
lists, out_control );
+ }
Compute_Forces( system, control, data, workspace, lists, out_control );
- for( i = 0; i < system->N; i++ ) {
+ for( i = 0; i < system->N; i++ )
+ {
inv_m = 1.0 / system->reaxprm.sbp[system->atoms[i].type].mass;
rvec_ScaledAdd( system->atoms[i].v,
0.5 * dt * -F_CONV * inv_m, system->atoms[i].f );
}
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "verlet2\n");
-#endif
-}
-
-///////////////////////////////////////////////////////////////////
-//Cuda Function -- Velocity Verlet NVE
-///////////////////////////////////////////////////////////////////
-
-GLOBAL void Cuda_Velocity_Verlet_NVE_atoms1 (reax_atom *atoms,
- single_body_parameters *sbp,
- simulation_box *box,
- int N, real dt)
-{
- real inv_m, dt_sqr;
- rvec dx;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- dt_sqr = SQR(dt);
- //for( i = 0; i < system->N; i++ ) {
- inv_m = 1.0 / sbp[atoms[i].type].mass;
-
- rvec_ScaledSum( dx, dt, atoms[i].v,
- 0.5 * dt_sqr * -F_CONV * inv_m, atoms[i].f );
- Inc_on_T3( atoms[i].x, dx, box );
-
- rvec_ScaledAdd( atoms[i].v,
- 0.5 * dt * -F_CONV * inv_m, atoms[i].f );
- //}
-}
-
-GLOBAL void Cuda_Velocity_Verlet_NVE_atoms2 (reax_atom *atoms, single_body_parameters *sbp, int N, real dt)
-{
- real inv_m;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- //for( i = 0; i < system->N; i++ ) {
- inv_m = 1.0 / sbp[atoms[i].type].mass;
- rvec_ScaledAdd( atoms[i].v,
- 0.5 * dt * -F_CONV * inv_m, atoms[i].f );
- //}
-}
-
-void Cuda_Velocity_Verlet_NVE(reax_system* system, control_params* control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, steps, renbr;
- real inv_m, dt, dt_sqr;
- rvec dx;
- int blocks, block_size;
-
- dt = control->dt;
- dt_sqr = SQR(dt);
- steps = data->step - data->prev_steps;
- renbr = (steps % control->reneighbor == 0);
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d: ", data->step );
-#endif
-
- compute_blocks (&blocks, &block_size, system->N);
- Cuda_Velocity_Verlet_NVE_atoms1 <<<blocks, block_size>>>
- (system->d_atoms, system->reaxprm.d_sbp,
- (simulation_box *)system->d_box, system->N, dt);
- cudaThreadSynchronize ();
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "verlet1 - ");
-#endif
-
- Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
- Cuda_Reset( system, control, data, workspace, lists );
-
- if( renbr ) {
- Cuda_Generate_Neighbor_Lists (system, dev_workspace, control, true);
- }
-
- Cuda_Compute_Forces( system, control, data, workspace, lists, out_control );
-
- Cuda_Velocity_Verlet_NVE_atoms2<<<blocks, block_size>>>
- (system->d_atoms, system->reaxprm.d_sbp, system->N, dt);
- cudaThreadSynchronize ();
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet2\n");
#endif
}
+
void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
control_params* control,
simulation_data *data,
@@ -188,6 +106,7 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
therm = &( data->therm );
steps = data->step - data->prev_steps;
renbr = (steps % control->reneighbor == 0);
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "step%d: ", data->step );
#endif
@@ -197,7 +116,8 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
#endif
/* Compute x(t + dt) and copy old forces */
- for (i=0; i < system->N; i++) {
+ for (i=0; i < system->N; i++)
+ {
inv_m = 1.0 / system->reaxprm.sbp[system->atoms[i].type].mass;
rvec_ScaledSum( dx, dt - 0.5 * dt_sqr * therm->v_xi, system->atoms[i].v,
@@ -209,6 +129,7 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
}
/* Compute xi(t + dt) */
therm->xi += ( therm->v_xi * dt + 0.5 * dt_sqr * therm->G_xi );
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet1 - " );
#endif
@@ -217,14 +138,17 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
Reset( system, control, data, workspace, lists );
if( renbr )
+ {
Generate_Neighbor_Lists( system, control, data, workspace,
lists, out_control );
+ }
/* Calculate Forces at time (t + dt) */
Compute_Forces( system,control,data, workspace, lists, out_control );
/* Compute iteration constants for each atom's velocity */
- for( i = 0; i < system->N; ++i ) {
+ for( i = 0; i < system->N; ++i )
+ {
inv_m = 1.0 / system->reaxprm.sbp[system->atoms[i].type].mass;
rvec_Scale( workspace->v_const[i],
@@ -241,7 +165,6 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
#endif
}
-
v_xi_new = therm->v_xi_old + 2.0 * dt * therm->G_xi;
E_kin_new = G_xi_new = v_xi_old = 0;
itr = 0;
@@ -258,7 +181,8 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
//print_sys_atoms (system);
#endif
- for( i = 0; i < system->N; ++i ) {
+ for( i = 0; i < system->N; ++i )
+ {
rvec_Scale( system->atoms[i].v, coef_v, workspace->v_const[i] );
E_kin_new += ( 0.5*system->reaxprm.sbp[system->atoms[i].type].mass *
@@ -272,6 +196,7 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
G_xi_new = control->Tau_T * ( 2.0 * E_kin_new -
data->N_f * K_B * control->T );
v_xi_new = therm->v_xi + 0.5 * dt * ( therm->G_xi + G_xi_new );
+
#if defined(DEBUG)
fprintf( stderr, "itr%d: G_xi_new = %f, v_xi_new = %f, v_xi_old = %f\n",
itr, G_xi_new, v_xi_new, v_xi_old );
@@ -283,7 +208,6 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
fprintf (stderr, " Iteration Count in NVE --> %d \n", itr );
#endif
-
#ifndef __BUILD_DEBUG__
therm->v_xi_old = therm->v_xi;
therm->v_xi = v_xi_new;
@@ -296,215 +220,6 @@ void Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
}
-
-
-///////////////////////////////////////////////////////////////////
-//Cuda Function -- Velocity_Verlet_Nose_Hoover_NVT_Klein
-///////////////////////////////////////////////////////////////////
-
-GLOBAL void Compute_X_t_dt (real dt, real dt_sqr, thermostat p_therm,
- reax_atom *atoms, single_body_parameters *sbp,
- simulation_box *box,
- static_storage p_workspace, int N)
-{
-
- real inv_m;
- rvec dx;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= N) return;
-
- static_storage *workspace = &p_workspace;
- thermostat *therm = &p_therm;
-
- /* Compute x(t + dt) and copy old forces */
- //for (i=0; i < system->N; i++) {
- inv_m = 1.0 / sbp[atoms[i].type].mass;
-
- rvec_ScaledSum( dx, dt - 0.5 * dt_sqr * therm->v_xi, atoms[i].v,
- 0.5 * dt_sqr * inv_m * -F_CONV, atoms[i].f );
-
- Inc_on_T3( atoms[i].x, dx, box );
-
- rvec_Copy( workspace->f_old[i], atoms[i].f );
- //}
-
-}
-
-GLOBAL void Update_Velocity (reax_atom *atoms, single_body_parameters *sbp,
- static_storage p_workspace, real dt, thermostat p_therm,
- int N)
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- real inv_m;
- static_storage *workspace = &p_workspace;
- thermostat *therm = &p_therm;
-
- //for( i = 0; i < system->N; ++i ) {
- inv_m = 1.0 / sbp[atoms[i].type].mass;
-
- rvec_Scale( workspace->v_const[i],
- 1.0 - 0.5 * dt * therm->v_xi, atoms[i].v );
- rvec_ScaledAdd( workspace->v_const[i],
- 0.5 * dt * inv_m * -F_CONV, workspace->f_old[i] );
- rvec_ScaledAdd( workspace->v_const[i],
- 0.5 * dt * inv_m * -F_CONV, atoms[i].f );
- //}
-}
-
-GLOBAL void E_Kin_Reduction (reax_atom *atoms, static_storage p_workspace,
- single_body_parameters *sbp,
- real *per_block_results, real coef_v, const size_t n)
-{
- extern __shared__ real sdata[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real x = 0;
- static_storage *workspace = &p_workspace;
-
- if(i < n)
- {
- rvec_Scale( atoms[i].v, coef_v, workspace->v_const[i] );
- x = ( 0.5 * sbp[atoms[i].type].mass *
- rvec_Dot( atoms[i].v, atoms[i].v ) );
- }
- sdata[threadIdx.x] = x;
- __syncthreads();
-
- for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
- {
- if(threadIdx.x < offset)
- {
- sdata[threadIdx.x] += sdata[threadIdx.x + offset];
- }
-
- __syncthreads();
- }
-
- if(threadIdx.x == 0)
- {
- per_block_results[blockIdx.x] = sdata[0];
- }
-}
-
-
-void Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein(reax_system* system,
- control_params* control,
- simulation_data *data,
- static_storage *workspace,
- list **lists,
- output_controls *out_control )
-{
- int i, itr, steps, renbr;
- real inv_m, coef_v, dt, dt_sqr;
- real E_kin_new, G_xi_new, v_xi_new, v_xi_old;
- rvec dx;
- thermostat *therm;
-
- real *results = (real *)scratch;
-
- dt = control->dt;
- dt_sqr = SQR( dt );
- therm = &( data->therm );
- steps = data->step - data->prev_steps;
- renbr = (steps % control->reneighbor == 0);
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Device: Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein --> coef to update velocity --> %6.10f\n", therm->v_xi_old);
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d: ", data->step );
-#endif
-
- Compute_X_t_dt <<< BLOCKS, BLOCK_SIZE >>>
- (dt, dt_sqr, data->therm, system->d_atoms,
- system->reaxprm.d_sbp, system->d_box, *dev_workspace, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- /* Compute xi(t + dt) */
- therm->xi += ( therm->v_xi * dt + 0.5 * dt_sqr * therm->G_xi );
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "verlet1 - " );
-#endif
-
- Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
- Cuda_Reset( system, control, data, workspace, lists );
-
- if( renbr ) {
- //generate_neighbor_lists here
- Cuda_Generate_Neighbor_Lists (system, dev_workspace, control, true);
- }
-
- /* Calculate Forces at time (t + dt) */
- Cuda_Compute_Forces( system,control,data, workspace, lists, out_control );
-
- /* Compute iteration constants for each atom's velocity */
- Update_Velocity <<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, system->reaxprm.d_sbp, *dev_workspace,
- dt, *therm, system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- v_xi_new = therm->v_xi_old + 2.0 * dt * therm->G_xi;
- E_kin_new = G_xi_new = v_xi_old = 0;
- itr = 0;
- do {
- itr++;
-
- /* new values become old in this iteration */
- v_xi_old = v_xi_new;
- coef_v = 1.0 / (1.0 + 0.5 * dt * v_xi_old);
- E_kin_new = 0;
-
- /*reduction for the E_Kin_new here*/
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Device: coef to update velocity --> %6.10f, %6.10f, %6.10f\n", coef_v, dt, therm->v_xi_old);
-#endif
- cuda_memset (results, 0, 2 * BLOCK_SIZE * REAL_SIZE, RES_SCRATCH );
- E_Kin_Reduction <<< BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE >>>
- (system->d_atoms, *dev_workspace, system->reaxprm.d_sbp,
- results, coef_v, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- Cuda_reduction <<<1, BLOCKS_POW_2, REAL_SIZE * BLOCKS_POW_2 >>>
- (results, results + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (&E_kin_new, results + BLOCKS_POW_2, REAL_SIZE, cudaMemcpyDeviceToHost, RES_SCRATCH );
-
- G_xi_new = control->Tau_T * ( 2.0 * E_kin_new -
- data->N_f * K_B * control->T );
- v_xi_new = therm->v_xi + 0.5 * dt * ( therm->G_xi + G_xi_new );
-#if defined(DEBUG)
- fprintf( stderr, "itr%d: G_xi_new = %f, v_xi_new = %f, v_xi_old = %f\n",
- itr, G_xi_new, v_xi_new, v_xi_old );
-#endif
- }
- while( fabs(v_xi_new - v_xi_old ) > 1e-5 );
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Iteration Count in NVE --> %d \n", itr );
-#endif
-
- therm->v_xi_old = therm->v_xi;
- therm->v_xi = v_xi_new;
- therm->G_xi = G_xi_new;
-#if defined(DEBUG_FOCUS)
- fprintf( stderr,"vel scale\n" );
-#endif
-}
-
-///////////////////////////////////////////////////////////////////
-//Cuda Function -- Velocity_Verlet_Nose_Hoover_NVT_Klein
-///////////////////////////////////////////////////////////////////
-
-
/* uses Berendsen-type coupling for both T and P.
All box dimensions are scaled by the same amount,
there is no change in the angles between axes. */
@@ -522,6 +237,7 @@ void Velocity_Verlet_Berendsen_Isotropic_NPT( reax_system* system,
dt = control->dt;
steps = data->step - data->prev_steps;
renbr = (steps % control->reneighbor == 0);
+
#if defined(DEBUG_FOCUS)
//fprintf( out_control->prs,
// "tau_t: %g tau_p: %g dt/tau_t: %g dt/tau_p: %g\n",
@@ -530,7 +246,8 @@ void Velocity_Verlet_Berendsen_Isotropic_NPT( reax_system* system,
#endif
/* velocity verlet, 1st part */
- for( i = 0; i < system->N; i++ ) {
+ for( i = 0; i < system->N; i++ )
+ {
inv_m = 1.0 / system->reaxprm.sbp[system->atoms[i].type].mass;
/* Compute x(t + dt) */
rvec_ScaledSum( dx, dt, system->atoms[i].v,
@@ -546,6 +263,7 @@ void Velocity_Verlet_Berendsen_Isotropic_NPT( reax_system* system,
0.5 * SQR(dt) * -F_CONV * inv_m * system->atoms[i].f[1],
0.5 * SQR(dt) * -F_CONV * inv_m * system->atoms[i].f[2] ); */
}
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet1 - " );
#endif
@@ -574,6 +292,7 @@ void Velocity_Verlet_Berendsen_Isotropic_NPT( reax_system* system,
}
//Compute_Kinetic_Energy( system, data );
Compute_Pressure_Isotropic( system, control, data, out_control );
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet2 - " );
#endif
@@ -633,6 +352,7 @@ void Velocity_Verlet_Berendsen_SemiIsotropic_NPT( reax_system* system,
dt = control->dt;
steps = data->step - data->prev_steps;
renbr = (steps % control->reneighbor == 0);
+
#if defined(DEBUG_FOCUS)
//fprintf( out_control->prs,
// "tau_t: %g tau_p: %g dt/tau_t: %g dt/tau_p: %g\n",
@@ -657,6 +377,7 @@ void Velocity_Verlet_Berendsen_SemiIsotropic_NPT( reax_system* system,
0.5 * SQR(dt) * -F_CONV * inv_m * system->atoms[i].f[1],
0.5 * SQR(dt) * -F_CONV * inv_m * system->atoms[i].f[2] ); */
}
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet1 - " );
#endif
@@ -685,6 +406,7 @@ void Velocity_Verlet_Berendsen_SemiIsotropic_NPT( reax_system* system,
}
//Compute_Kinetic_Energy( system, data );
Compute_Pressure_Isotropic( system, control, data, out_control );
+
#if defined(DEBUG_FOCUS)
fprintf( stderr, "verlet2 - " );
#endif
@@ -730,14 +452,12 @@ void Velocity_Verlet_Berendsen_SemiIsotropic_NPT( reax_system* system,
}
-
/************************************************/
/* BELOW FUNCTIONS ARE NOT BEING USED ANYMORE! */
/* */
/*!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!*/
/*!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!*/
/************************************************/
-
#ifdef ANISOTROPIC
void Velocity_Verlet_Nose_Hoover_NVT(reax_system* system,
@@ -810,7 +530,6 @@ void Velocity_Verlet_Nose_Hoover_NVT(reax_system* system,
}
-
void Velocity_Verlet_Isotropic_NPT( reax_system* system,
control_params* control,
simulation_data *data,
@@ -983,8 +702,6 @@ void Velocity_Verlet_Isotropic_NPT( reax_system* system,
#endif
-////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////////////////////////////////
/* uses Berendsen-type coupling for both T and P.
All box dimensions are scaled by the same amount,
there is no change in the angles between axes. */
@@ -1006,6 +723,7 @@ void Velocity_Verlet_Berendsen_NVT( reax_system* system,
#if defined(DEBUG_FOCUS)
fprintf( stderr, "step%d\n", data->step );
#endif
+
dt = control->dt;
steps = data->step - data->prev_steps;
renbr = (steps % control->reneighbor == 0);
@@ -1020,6 +738,7 @@ void Velocity_Verlet_Berendsen_NVT( reax_system* system,
/* Compute v(t + dt/2) */
rvec_ScaledAdd( atom->v, 0.5 * -F_CONV * inv_m * dt, atom->f );
}
+
#if defined(DEBUG_FOCUS)
fprintf(stderr, "step%d: verlet1 done\n", data->step);
#endif
@@ -1040,6 +759,7 @@ void Velocity_Verlet_Berendsen_NVT( reax_system* system,
/* Compute v(t + dt) */
rvec_ScaledAdd( atom->v, 0.5 * dt * -F_CONV * inv_m, atom->f );
}
+
#if defined(DEBUG_FOCUS)
fprintf(stderr, "step%d: verlet2 done\n", data->step);
#endif
@@ -1065,182 +785,3 @@ void Velocity_Verlet_Berendsen_NVT( reax_system* system,
data->step );
#endif
}
-
-GLOBAL void ker_update_velocity_1 (reax_atom *atoms,
- single_body_parameters *sbp,
- real dt,
- simulation_box *box,
- int N)
-{
- real inv_m;
- rvec dx;
- reax_atom *atom;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if ( i >= N ) return;
-
- /* velocity verlet, 1st part */
- //for( i = 0; i < system->n; i++ ) {
- atom = &(atoms[i]);
- inv_m = 1.0 / sbp[atom->type].mass;
- /* Compute x(t + dt) */
- rvec_ScaledSum( dx, dt, atom->v, 0.5 * -F_CONV * inv_m * SQR(dt), atom->f );
- rvec_Add( atom->x, dx );
-
- /* Metin's suggestion to rebox the atoms */
- /* bNVT fix */
- Inc_on_T3( atoms[i].x, dx, box );
- /* bNVT fix */
-
- /* Compute v(t + dt/2) */
- rvec_ScaledAdd( atom->v, 0.5 * -F_CONV * inv_m * dt, atom->f );
- //}
-}
-
-void bNVT_update_velocity_part1 (reax_system *system, simulation_box *box, real dt)
-{
- ker_update_velocity_1 <<< BLOCKS, BLOCK_SIZE>>>
- (system->d_atoms, system->reaxprm.d_sbp, dt, box, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
-
-GLOBAL void ker_update_velocity_2 (reax_atom *atoms,
- single_body_parameters *sbp,
- real dt,
- int N)
-{
- reax_atom *atom;
- real inv_m;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if ( i >= N ) return;
-
- /* velocity verlet, 2nd part */
- //for( i = 0; i < system->n; i++ ) {
- atom = &(atoms[i]);
- inv_m = 1.0 / sbp[atom->type].mass;
- /* Compute v(t + dt) */
- rvec_ScaledAdd( atom->v, 0.5 * dt * -F_CONV * inv_m, atom->f );
- //}
-}
-
-void bNVT_update_velocity_part2 (reax_system *system, real dt)
-{
- ker_update_velocity_2 <<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, system->reaxprm.d_sbp, dt, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
-
-GLOBAL void ker_scale_velocities (reax_atom *atoms, real lambda, int N)
-{
- reax_atom *atom;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if ( i >= N ) return;
-
- /* Scale velocities and positions at t+dt */
- //for( i = 0; i < system->n; ++i ) {
- atom = &(atoms[i]);
- rvec_Scale( atom->v, lambda, atom->v );
- //}
-}
-
-void bNVT_scale_velocities (reax_system *system, real lambda)
-{
- ker_scale_velocities <<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, lambda, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-}
-
-void Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* system,
- control_params* control,
- simulation_data *data,
- static_storage *workspace,
- list **lists,
- output_controls *out_control
- )
-{
- int i, steps, renbr;
- real inv_m, dt, lambda;
- rvec dx;
- reax_atom *atom;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d\n", data->step );
-#endif
- dt = control->dt;
- steps = data->step - data->prev_steps;
- renbr = (steps % control->reneighbor == 0);
-
- /* velocity verlet, 1st part
- for( i = 0; i < system->N; i++ ) {
- atom = &(system->atoms[i]);
- inv_m = 1.0 / system->reaxprm.sbp[atom->type].mass;
- // Compute x(t + dt)
- rvec_ScaledSum( dx, dt, atom->v, 0.5 * -F_CONV * inv_m * SQR(dt), atom->f );
- rvec_Add( atom->x, dx );
- // Compute v(t + dt/2)
- rvec_ScaledAdd( atom->v, 0.5 * -F_CONV * inv_m * dt, atom->f );
- }
- */
- bNVT_update_velocity_part1 (system, (simulation_box *) system->d_box, dt);
-
-#if defined(DEBUG_FOCUS)
- fprintf(stderr, "step%d: verlet1 done\n", data->step);
-#endif
-
- Cuda_Reallocate( system, dev_workspace, dev_lists, renbr, data->step );
- Cuda_Reset( system, control, data, workspace, lists );
-
- if( renbr ) {
- Cuda_Generate_Neighbor_Lists( system, workspace, control, true);
- }
-
- Cuda_Compute_Forces( system, control, data, workspace,
- lists, out_control );
-
- /* velocity verlet, 2nd part
- for( i = 0; i < system->N; i++ ) {
- atom = &(system->atoms[i]);
- inv_m = 1.0 / system->reaxprm.sbp[atom->type].mass;
- // Compute v(t + dt)
- rvec_ScaledAdd( atom->v, 0.5 * dt * -F_CONV * inv_m, atom->f );
- }
- */
- bNVT_update_velocity_part2 (system, dt);
-#if defined(DEBUG_FOCUS)
- fprintf(stderr, "step%d: verlet2 done\n", data->step);
-#endif
-
- /* temperature scaler */
- Cuda_Compute_Kinetic_Energy( system, data );
- //get the latest temperature from the device to the host.
- copy_host_device (&data->therm, &((simulation_data *)data->d_simulation_data)->therm,
- sizeof (thermostat), cudaMemcpyDeviceToHost, RES_SIMULATION_DATA );
-
- lambda = 1.0 + (dt / control->Tau_T) * (control->T / data->therm.T - 1.0);
- if( lambda < MIN_dT )
- lambda = MIN_dT;
- else if (lambda > MAX_dT )
- lambda = MAX_dT;
- lambda = SQRT( lambda );
-
- //fprintf (stderr, "step:%d lambda -> %f \n", data->step, lambda);
-
- /* Scale velocities and positions at t+dt
- for( i = 0; i < system->N; ++i ) {
- atom = &(system->atoms[i]);
- rvec_Scale( atom->v, lambda, atom->v );
- }
- */
- bNVT_scale_velocities (system, lambda);
- Cuda_Compute_Kinetic_Energy( system, data );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "step%d: scaled velocities\n",
- data->step );
-#endif
-
-}
-
-
diff --git a/PuReMD-GPU/src/integrate.h b/PuReMD-GPU/src/integrate.h
index 945b29fa09fb59e8e801b55a7527ca88814e7bc2..6f5848f0de84e8a50ef2c5090194618b61f185fc 100644
--- a/PuReMD-GPU/src/integrate.h
+++ b/PuReMD-GPU/src/integrate.h
@@ -24,24 +24,19 @@
#include "mytypes.h"
void Velocity_Verlet_NVE( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-void Cuda_Velocity_Verlet_NVE( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
+ static_storage*, list**, output_controls* );
void Velocity_Verlet_Nose_Hoover_NVT( reax_system*, control_params*,
- simulation_data*, static_storage*,
- list**, output_controls* );
-void Velocity_Verlet_Nose_Hoover_NVT_Klein( reax_system*, control_params*,
simulation_data*, static_storage*,
list**, output_controls* );
-void Cuda_Velocity_Verlet_Nose_Hoover_NVT_Klein( reax_system*, control_params*,
+void Velocity_Verlet_Nose_Hoover_NVT_Klein( reax_system*, control_params*,
simulation_data*, static_storage*,
list**, output_controls* );
void Velocity_Verlet_Flexible_NPT( reax_system*, control_params*,
- simulation_data*, static_storage*,
- list**, output_controls* );
+ simulation_data*, static_storage*,
+ list**, output_controls* );
void Velocity_Verlet_Isotropic_NPT( reax_system*, control_params*,
- simulation_data*, static_storage*,
- list**, output_controls* );
+ simulation_data*, static_storage*,
+ list**, output_controls* );
void Velocity_Verlet_Berendsen_Isotropic_NPT( reax_system*, control_params*,
simulation_data*, static_storage*,
list**, output_controls* );
@@ -50,9 +45,7 @@ void Velocity_Verlet_Berendsen_SemiIsotropic_NPT( reax_system*, control_params*,
static_storage*, list**,
output_controls* );
void Velocity_Verlet_Berendsen_NVT( reax_system* , control_params* ,
- simulation_data *, static_storage *,
- list **, output_controls * );
-void Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* , control_params* ,
simulation_data *, static_storage *,
list **, output_controls * );
+
#endif
diff --git a/PuReMD-GPU/src/lin_alg.c b/PuReMD-GPU/src/lin_alg.c
new file mode 100644
index 0000000000000000000000000000000000000000..cb141d475b0e2cf702901ed551287e0e238cdcd6
--- /dev/null
+++ b/PuReMD-GPU/src/lin_alg.c
@@ -0,0 +1,676 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "lin_alg.h"
+
+#include "list.h"
+#include "vector.h"
+#include "index_utils.h"
+
+
+void Sparse_MatVec( sparse_matrix *A, real *x, real *b )
+{
+ int i, j, k, n, si, ei;
+ real H;
+
+ n = A->n;
+ for( i = 0; i < n; ++i )
+ b[i] = 0;
+
+ for( i = 0; i < n; ++i ) {
+ si = A->start[i];
+ ei = A->start[i+1]-1;
+
+ for( k = si; k < ei; ++k ) {
+ j = A->entries[k].j;
+ H = A->entries[k].val;
+ b[j] += H * x[i];
+ b[i] += H * x[j];
+ }
+
+ // the diagonal entry is the last one in
+ b[i] += A->entries[k].val * x[i];
+ }
+}
+
+
+void Forward_Subs( sparse_matrix *L, real *b, real *y )
+{
+ int i, pj, j, si, ei;
+ real val;
+
+ for( i = 0; i < L->n; ++i ) {
+ y[i] = b[i];
+ si = L->start[i];
+ ei = L->start[i+1];
+ for( pj = si; pj < ei-1; ++pj ){
+ j = L->entries[pj].j;
+ val = L->entries[pj].val;
+ y[i] -= val * y[j];
+ }
+ y[i] /= L->entries[pj].val;
+ }
+}
+
+
+void Backward_Subs( sparse_matrix *U, real *y, real *x )
+{
+ int i, pj, j, si, ei;
+ real val;
+
+ for( i = U->n-1; i >= 0; --i ) {
+ x[i] = y[i];
+ si = U->start[i];
+ ei = U->start[i+1];
+ for( pj = si+1; pj < ei; ++pj ){
+ j = U->entries[pj].j;
+ val = U->entries[pj].val;
+ x[i] -= val * x[j];
+ }
+ x[i] /= U->entries[si].val;
+ }
+}
+
+
+int GMRES( static_storage *workspace, sparse_matrix *H,
+ real *b, real tol, real *x, FILE *fout, reax_system* system)
+{
+ int i, j, k, itr, N;
+ real cc, tmp1, tmp2, temp, bnorm;
+
+ N = H->n;
+ bnorm = Norm( b, N );
+
+ /* apply the diagonal pre-conditioner to rhs */
+ for( i = 0; i < N; ++i )
+ workspace->b_prc[i] = b[i] * workspace->Hdia_inv[i];
+
+ /* GMRES outer-loop */
+ for( itr = 0; itr < MAX_ITR; ++itr ) {
+ /* calculate r0 */
+ Sparse_MatVec( H, x, workspace->b_prm );
+
+ for( i = 0; i < N; ++i )
+ workspace->b_prm[i] *= workspace->Hdia_inv[i]; /* pre-conditioner */
+
+
+ Vector_Sum(&workspace->v[ index_wkspace_sys (0,0,system->N) ], 1.,workspace->b_prc, -1., workspace->b_prm, N);
+ workspace->g[0] = Norm( &workspace->v[index_wkspace_sys (0,0,system->N)], N );
+ Vector_Scale( &workspace->v[ index_wkspace_sys (0,0,system->N) ], 1.0/workspace->g[0], &workspace->v[index_wkspace_sys(0,0,system->N)], N );
+
+ /* GMRES inner-loop */
+ for( j = 0; j < RESTART && fabs(workspace->g[j]) / bnorm > tol; j++ ) {
+ /* matvec */
+ Sparse_MatVec( H, &workspace->v[index_wkspace_sys(j,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)] );
+
+ for( k = 0; k < N; ++k )
+ workspace->v[ index_wkspace_sys (j+1,k,system->N)] *= workspace->Hdia_inv[k]; /*pre-conditioner*/
+
+ /* apply modified Gram-Schmidt to orthogonalize the new residual */
+ for( i = 0; i <= j; i++ ) {
+ workspace->h[ index_wkspace_res (i,j) ] = Dot( &workspace->v[index_wkspace_sys(i,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ Vector_Add( &workspace->v[index_wkspace_sys(j+1,0,system->N)],
+ -workspace->h[index_wkspace_res (i,j) ], &workspace->v[index_wkspace_sys(i,0,system->N)], N );
+ }
+
+
+ workspace->h[ index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ Vector_Scale( &workspace->v[index_wkspace_sys(j+1,0,system->N)],
+ 1. / workspace->h[ index_wkspace_res (j+1,j) ], &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ // fprintf( stderr, "%d-%d: orthogonalization completed.\n", itr, j );
+
+
+ /* Givens rotations on the upper-Hessenberg matrix to make it U */
+ for( i = 0; i <= j; i++ ) {
+ if( i == j ) {
+ cc = SQRT( SQR(workspace->h[ index_wkspace_res (j,j) ])+SQR(workspace->h[ index_wkspace_res (j+1,j) ]) );
+ workspace->hc[j] = workspace->h[ index_wkspace_res (j,j) ] / cc;
+ workspace->hs[j] = workspace->h[ index_wkspace_res (j+1,j) ] / cc;
+ }
+
+ tmp1 = workspace->hc[i] * workspace->h[ index_wkspace_res (i,j) ] +
+ workspace->hs[i] * workspace->h[ index_wkspace_res (i+1,j) ];
+ tmp2 = -workspace->hs[i] * workspace->h[ index_wkspace_res (i,j) ] +
+ workspace->hc[i] * workspace->h[ index_wkspace_res (i+1,j) ];
+
+ workspace->h[ index_wkspace_res (i,j) ] = tmp1;
+ workspace->h[ index_wkspace_res (i+1,j) ] = tmp2;
+ }
+
+ /* apply Givens rotations to the rhs as well */
+ tmp1 = workspace->hc[j] * workspace->g[j];
+ tmp2 = -workspace->hs[j] * workspace->g[j];
+ workspace->g[j] = tmp1;
+ workspace->g[j+1] = tmp2;
+
+ // fprintf( stderr, "h: " );
+ // for( i = 0; i <= j+1; ++i )
+ // fprintf( stderr, "%.6f ", workspace->h[i][j] );
+ // fprintf( stderr, "\n" );
+ //fprintf( stderr, "res: %.15e\n", workspace->g[j+1] );
+ }
+
+
+ /* solve Hy = g.
+ H is now upper-triangular, do back-substitution */
+ for( i = j-1; i >= 0; i-- ) {
+ temp = workspace->g[i];
+ for( k = j-1; k > i; k-- )
+ temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
+
+ workspace->y[i] = temp / workspace->h[ index_wkspace_res (i,i) ];
+ }
+
+
+ /* update x = x_0 + Vy */
+ for( i = 0; i < j; i++ )
+ Vector_Add( x, workspace->y[i], &workspace->v[index_wkspace_sys(i,0,system->N)], N );
+
+ /* stopping condition */
+ if( fabs(workspace->g[j]) / bnorm <= tol )
+ break;
+ }
+
+ // Sparse_MatVec( H, x, workspace->b_prm );
+ // for( i = 0; i < N; ++i )
+ // workspace->b_prm[i] *= workspace->Hdia_inv[i];
+ // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
+ // for( i = 0; i < N; ++i )
+ // fprintf( fout, "%10.5f%15.12f%15.12f\n",
+ // workspace->b_prc[i], workspace->b_prm[i], x[i] );*/
+
+ // fprintf(fout,"GMRES outer:%d, inner:%d iters - residual norm: %25.20f\n",
+ // itr, j, fabs( workspace->g[j] ) / bnorm );
+ // data->timing.matvec += itr * RESTART + j;
+
+ if( itr >= MAX_ITR ) {
+ fprintf( stderr, "GMRES convergence failed\n" );
+ // return -1;
+ return itr * (RESTART+1) + j + 1;
+ }
+
+ return itr * (RESTART+1) + j + 1;
+}
+
+
+int GMRES_HouseHolder( static_storage *workspace, sparse_matrix *H,
+ real *b, real tol, real *x, FILE *fout, reax_system *system)
+{
+ int i, j, k, itr, N;
+ real cc, tmp1, tmp2, temp, bnorm;
+ real v[10000], z[RESTART+2][10000], w[RESTART+2];
+ real u[RESTART+2][10000];
+
+ N = H->n;
+ bnorm = Norm( b, N );
+
+ /* apply the diagonal pre-conditioner to rhs */
+ for( i = 0; i < N; ++i )
+ workspace->b_prc[i] = b[i] * workspace->Hdia_inv[i];
+
+ // memset( x, 0, sizeof(real) * N );
+
+ /* GMRES outer-loop */
+ for( itr = 0; itr < MAX_ITR; ++itr ) {
+ /* compute z = r0 */
+ Sparse_MatVec( H, x, workspace->b_prm );
+ for( i = 0; i < N; ++i )
+ workspace->b_prm[i] *= workspace->Hdia_inv[i]; /* pre-conditioner */
+ Vector_Sum( z[0], 1., workspace->b_prc, -1., workspace->b_prm, N );
+
+ Vector_MakeZero( w, RESTART+1 );
+ w[0] = Norm( z[0], N );
+
+ Vector_Copy( u[0], z[0], N );
+ u[0][0] += ( u[0][0] < 0.0 ? -1 : 1 ) * w[0];
+ Vector_Scale( u[0], 1 / Norm( u[0], N ), u[0], N );
+
+ w[0] *= ( u[0][0] < 0.0 ? 1 :-1 );
+ // fprintf( stderr, "\n\n%12.6f\n", w[0] );
+
+ /* GMRES inner-loop */
+ for( j = 0; j < RESTART && fabs( w[j] ) / bnorm > tol; j++ ) {
+ /* compute v_j */
+ Vector_Scale( z[j], -2 * u[j][j], u[j], N );
+ z[j][j] += 1.; /* due to e_j */
+
+ for( i = j-1; i >= 0; --i )
+ Vector_Add( z[j]+i, -2 * Dot( u[i]+i, z[j]+i, N-i ), u[i]+i, N-i );
+
+
+ /* matvec */
+ Sparse_MatVec( H, z[j], v );
+
+ for( k = 0; k < N; ++k )
+ v[k] *= workspace->Hdia_inv[k]; /* pre-conditioner */
+
+ for( i = 0; i <= j; ++i )
+ Vector_Add( v+i, -2 * Dot( u[i]+i, v+i, N-i ), u[i]+i, N-i );
+
+
+ if( !Vector_isZero( v + (j+1), N - (j+1) ) ) {
+ /* compute the HouseHolder unit vector u_j+1 */
+ for( i = 0; i <= j; ++i )
+ u[j+1][i] = 0;
+
+ Vector_Copy( u[j+1] + (j+1), v + (j+1), N - (j+1) );
+
+ u[j+1][j+1] += ( v[j+1]<0.0 ? -1:1 ) * Norm( v+(j+1), N-(j+1) );
+
+ Vector_Scale( u[j+1], 1 / Norm( u[j+1], N ), u[j+1], N );
+
+ /* overwrite v with P_m+1 * v */
+ v[j+1] -= 2 * Dot( u[j+1]+(j+1), v+(j+1), N-(j+1) ) * u[j+1][j+1];
+ Vector_MakeZero( v + (j+2), N - (j+2) );
+ // Vector_Add( v, -2 * Dot( u[j+1], v, N ), u[j+1], N );
+ }
+
+
+ /* prev Givens rots on the upper-Hessenberg matrix to make it U */
+ for( i = 0; i < j; i++ ) {
+ tmp1 = workspace->hc[i] * v[i] + workspace->hs[i] * v[i+1];
+ tmp2 = -workspace->hs[i] * v[i] + workspace->hc[i] * v[i+1];
+
+ v[i] = tmp1;
+ v[i+1] = tmp2;
+ }
+
+ /* apply the new Givens rotation to H and right-hand side */
+ if( fabs(v[j+1]) >= ALMOST_ZERO ) {
+ cc = SQRT( SQR( v[j] ) + SQR( v[j+1] ) );
+ workspace->hc[j] = v[j] / cc;
+ workspace->hs[j] = v[j+1] / cc;
+
+ tmp1 = workspace->hc[j] * v[j] + workspace->hs[j] * v[j+1];
+ tmp2 = -workspace->hs[j] * v[j] + workspace->hc[j] * v[j+1];
+
+ v[j] = tmp1;
+ v[j+1] = tmp2;
+
+ /* Givens rotations to rhs */
+ tmp1 = workspace->hc[j] * w[j];
+ tmp2 = -workspace->hs[j] * w[j];
+ w[j] = tmp1;
+ w[j+1] = tmp2;
+ }
+
+ /* extend R */
+ for( i = 0; i <= j; ++i )
+ workspace->h[ index_wkspace_res (i,j) ] = v[i];
+
+
+ // fprintf( stderr, "h:" );
+ // for( i = 0; i <= j+1 ; ++i )
+ // fprintf( stderr, "%.6f ", h[i][j] );
+ // fprintf( stderr, "\n" );
+ // fprintf( stderr, "%12.6f\n", w[j+1] );
+ }
+
+
+ /* solve Hy = w.
+ H is now upper-triangular, do back-substitution */
+ for( i = j-1; i >= 0; i-- ) {
+ temp = w[i];
+ for( k = j-1; k > i; k-- )
+ temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
+
+ workspace->y[i] = temp / workspace->h[ index_wkspace_res (i,i) ];
+ }
+
+ // fprintf( stderr, "y: " );
+ // for( i = 0; i < RESTART+1; ++i )
+ // fprintf( stderr, "%8.3f ", workspace->y[i] );
+
+
+ /* update x = x_0 + Vy */
+ // memset( z, 0, sizeof(real) * N );
+ // for( i = j-1; i >= 0; i-- )
+ // {
+ // Vector_Copy( v, z, N );
+ // v[i] += workspace->y[i];
+ //
+ // Vector_Sum( z, 1., v, -2 * Dot( u[i], v, N ), u[i], N );
+ // }
+ //
+ // fprintf( stderr, "\nz: " );
+ // for( k = 0; k < N; ++k )
+ // fprintf( stderr, "%6.2f ", z[k] );
+
+ // fprintf( stderr, "\nx_bef: " );
+ // for( i = 0; i < N; ++i )
+ // fprintf( stderr, "%6.2f ", x[i] );
+
+ // Vector_Add( x, 1, z, N );
+ for( i = j-1; i >= 0; i-- )
+ Vector_Add( x, workspace->y[i], z[i], N );
+
+ // fprintf( stderr, "\nx_aft: " );
+ // for( i = 0; i < N; ++i )
+ // fprintf( stderr, "%6.2f ", x[i] );
+
+ /* stopping condition */
+ if( fabs( w[j] ) / bnorm <= tol )
+ break;
+ }
+
+ // Sparse_MatVec( H, x, workspace->b_prm );
+ // for( i = 0; i < N; ++i )
+ // workspace->b_prm[i] *= workspace->Hdia_inv[i];
+
+ // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
+ // for( i = 0; i < N; ++i )
+ // fprintf( fout, "%10.5f%15.12f%15.12f\n",
+ // workspace->b_prc[i], workspace->b_prm[i], x[i] );
+
+ //fprintf( fout,"GMRES outer:%d, inner:%d iters - residual norm: %15.10f\n",
+ // itr, j, fabs( workspace->g[j] ) / bnorm );
+
+ if( itr >= MAX_ITR ) {
+ fprintf( stderr, "GMRES convergence failed\n" );
+ // return -1;
+ return itr * (RESTART+1) + j + 1;
+ }
+
+ return itr * (RESTART+1) + j + 1;
+}
+
+
+int PGMRES( static_storage *workspace, sparse_matrix *H, real *b, real tol,
+ sparse_matrix *L, sparse_matrix *U, real *x, FILE *fout, reax_system *system )
+{
+ int i, j, k, itr, N;
+ real cc, tmp1, tmp2, temp, bnorm;
+
+ N = H->n;
+ bnorm = Norm( b, N );
+
+ /* GMRES outer-loop */
+ for( itr = 0; itr < MAX_ITR; ++itr )
+ {
+ /* calculate r0 */
+ Sparse_MatVec( H, x, workspace->b_prm );
+ Vector_Sum( &workspace->v[index_wkspace_sys(0,0,system->N)], 1., b, -1., workspace->b_prm, N );
+ Forward_Subs( L, &workspace->v[index_wkspace_sys(0,0,system->N)], &workspace->v[index_wkspace_sys(0,0,system->N)] );
+ Backward_Subs( U, &workspace->v[index_wkspace_sys(0,0,system->N)], &workspace->v[index_wkspace_sys(0,0,system->N)] );
+ workspace->g[0] = Norm( &workspace->v[index_wkspace_sys(0,0,system->N)], N );
+ Vector_Scale( &workspace->v[index_wkspace_sys(0,0,system->N)], 1. / workspace->g[0], &workspace->v[index_wkspace_sys (0,0,system->N)], N );
+ //fprintf( stderr, "res: %.15e\n", workspace->g[0] );
+
+ /* GMRES inner-loop */
+ for( j = 0; j < RESTART && fabs(workspace->g[j]) / bnorm > tol; j++ )
+ {
+ /* matvec */
+ Sparse_MatVec( H, &workspace->v[index_wkspace_sys (j,0,system->N)], &workspace->v[index_wkspace_sys (j+1,0,system->N)] );
+ Forward_Subs( L, &workspace->v[index_wkspace_sys(j+1,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)] );
+ Backward_Subs( U, &workspace->v[index_wkspace_sys(j+1,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)] );
+
+ /* apply modified Gram-Schmidt to orthogonalize the new residual */
+ for( i = 0; i < j-1; i++ )
+ {
+ workspace->h[ index_wkspace_res (i,j)] = 0;
+ }
+
+ //for( i = 0; i <= j; i++ ) {
+ for( i = MAX(j-1,0); i <= j; i++ ) {
+ workspace->h[index_wkspace_res (i,j)] = Dot( &workspace->v[index_wkspace_sys (i,0,system->N)], &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ Vector_Add( &workspace->v[index_wkspace_sys(j+1,0,system->N)],-workspace->h[ index_wkspace_res (i,j) ], &workspace->v[index_wkspace_sys(i,0,system->N)], N );
+ }
+
+ workspace->h[index_wkspace_res (j+1,j) ] = Norm( &workspace->v[index_wkspace_sys (j+1,0,system->N)], N );
+ Vector_Scale( &workspace->v[index_wkspace_sys(j+1,0,system->N)],
+ 1. / workspace->h[ index_wkspace_res (j+1,j)], &workspace->v[index_wkspace_sys(j+1,0,system->N)], N );
+ // fprintf( stderr, "%d-%d: orthogonalization completed.\n", itr, j );
+
+ /* Givens rotations on the upper-Hessenberg matrix to make it U */
+ for( i = MAX(j-1,0); i <= j; i++ )
+ {
+ if( i == j )
+ {
+ cc = SQRT( SQR(workspace->h[ index_wkspace_res (j,j) ])+SQR(workspace->h[ index_wkspace_res (j+1,j) ]) );
+ workspace->hc[j] = workspace->h[ index_wkspace_res (j,j) ] / cc;
+ workspace->hs[j] = workspace->h[ index_wkspace_res (j+1,j) ] / cc;
+ }
+
+ tmp1 = workspace->hc[i] * workspace->h[ index_wkspace_res (i,j) ] +
+ workspace->hs[i] * workspace->h[index_wkspace_res (i+1,j) ];
+ tmp2 = -workspace->hs[i] * workspace->h[index_wkspace_res (i,j)] +
+ workspace->hc[i] * workspace->h[index_wkspace_res (i+1,j) ];
+
+ workspace->h[ index_wkspace_res (i,j) ] = tmp1;
+ workspace->h[ index_wkspace_res (i+1,j) ] = tmp2;
+ }
+
+ /* apply Givens rotations to the rhs as well */
+ tmp1 = workspace->hc[j] * workspace->g[j];
+ tmp2 = -workspace->hs[j] * workspace->g[j];
+ workspace->g[j] = tmp1;
+ workspace->g[j+1] = tmp2;
+
+ //fprintf( stderr, "h: " );
+ //for( i = 0; i <= j+1; ++i )
+ //fprintf( stderr, "%.6f ", workspace->h[i][j] );
+ //fprintf( stderr, "\n" );
+ //fprintf( stderr, "res: %.15e\n", workspace->g[j+1] );
+ }
+
+
+ /* solve Hy = g: H is now upper-triangular, do back-substitution */
+ for( i = j-1; i >= 0; i-- )
+ {
+ temp = workspace->g[i];
+ for( k = j-1; k > i; k-- )
+ {
+ temp -= workspace->h[ index_wkspace_res (i,k) ] * workspace->y[k];
+ }
+
+ workspace->y[i] = temp / workspace->h[index_wkspace_res (i,i)];
+ }
+
+ /* update x = x_0 + Vy */
+ Vector_MakeZero( workspace->p, N );
+ for( i = 0; i < j; i++ )
+ Vector_Add( workspace->p, workspace->y[i], &workspace->v[index_wkspace_sys(i,0,system->N)], N );
+ //Backward_Subs( U, workspace->p, workspace->p );
+ //Forward_Subs( L, workspace->p, workspace->p );
+ Vector_Add( x, 1., workspace->p, N );
+
+ /* stopping condition */
+ if( fabs(workspace->g[j]) / bnorm <= tol )
+ {
+ break;
+ }
+ }
+
+ // Sparse_MatVec( H, x, workspace->b_prm );
+ // for( i = 0; i < N; ++i )
+ // workspace->b_prm[i] *= workspace->Hdia_inv[i];
+ // fprintf( fout, "\n%10s%15s%15s\n", "b_prc", "b_prm", "x" );
+ // for( i = 0; i < N; ++i )
+ // fprintf( fout, "%10.5f%15.12f%15.12f\n",
+ // workspace->b_prc[i], workspace->b_prm[i], x[i] );*/
+
+ // fprintf(fout,"GMRES outer:%d, inner:%d iters - residual norm: %25.20f\n",
+ // itr, j, fabs( workspace->g[j] ) / bnorm );
+ // data->timing.matvec += itr * RESTART + j;
+
+ if( itr >= MAX_ITR ) {
+ fprintf( stderr, "GMRES convergence failed\n" );
+ // return -1;
+ return itr * (RESTART+1) + j + 1;
+ }
+
+ return itr * (RESTART+1) + j + 1;
+}
+
+
+int PCG( static_storage *workspace, sparse_matrix *A, real *b, real tol,
+ sparse_matrix *L, sparse_matrix *U, real *x, FILE *fout, reax_system* system )
+{
+ int i, N;
+ real tmp, alpha, beta, b_norm, r_norm;
+ real sig0, sig_old, sig_new;
+
+ N = A->n;
+ b_norm = Norm( b, N );
+ //fprintf( stderr, "b_norm: %.15e\n", b_norm );
+
+ Sparse_MatVec( A, x, workspace->q );
+ Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
+ r_norm = Norm(workspace->r, N);
+ //Print_Soln( workspace, x, q, b, N );
+ //fprintf( stderr, "res: %.15e\n", r_norm );
+
+ Forward_Subs( L, workspace->r, workspace->d );
+ Backward_Subs( U, workspace->d, workspace->p );
+ sig_new = Dot( workspace->r, workspace->p, N );
+ sig0 = sig_new;
+
+ for( i = 0; i < 200 && r_norm/b_norm > tol; ++i )
+ {
+ //for( i = 0; i < 200 && sig_new > SQR(tol) * sig0; ++i ) {
+ Sparse_MatVec( A, workspace->p, workspace->q );
+ tmp = Dot( workspace->q, workspace->p, N );
+ alpha = sig_new / tmp;
+ Vector_Add( x, alpha, workspace->p, N );
+ //fprintf( stderr, "iter%d: |p|=%.15e |q|=%.15e tmp=%.15e\n",
+ // i+1, Norm(workspace->p,N), Norm(workspace->q,N), tmp );
+
+ Vector_Add( workspace->r, -alpha, workspace->q, N );
+ r_norm = Norm(workspace->r, N);
+ //fprintf( stderr, "res: %.15e\n", r_norm );
+
+ Forward_Subs( L, workspace->r, workspace->d );
+ Backward_Subs( U, workspace->d, workspace->d );
+ sig_old = sig_new;
+ sig_new = Dot( workspace->r, workspace->d, N );
+ beta = sig_new / sig_old;
+ Vector_Sum( workspace->p, 1., workspace->d, beta, workspace->p, N );
+ }
+
+ //fprintf( fout, "CG took %d iterations\n", i );
+ if( i >= 200 ) {
+ fprintf( stderr, "CG convergence failed!\n" );
+ return i;
+ }
+
+ return i;
+}
+
+
+int CG( static_storage *workspace, sparse_matrix *H,
+ real *b, real tol, real *x, FILE *fout, reax_system *system)
+{
+ int i, j, N;
+ real tmp, alpha, beta, b_norm;
+ real sig_old, sig_new, sig0;
+
+ N = H->n;
+ b_norm = Norm( b, N );
+ //fprintf( stderr, "b_norm: %10.6f\n", b_norm );
+
+ Sparse_MatVec( H, x, workspace->q );
+ Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
+ for( j = 0; j < N; ++j )
+ workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
+
+ sig_new = Dot( workspace->r, workspace->d, N );
+ sig0 = sig_new;
+ //Print_Soln( workspace, x, q, b, N );
+ //fprintf( stderr, "sig_new: %24.15e, d_norm:%24.15e, q_norm:%24.15e\n",
+ // sqrt(sig_new), Norm(workspace->d,N), Norm(workspace->q,N) );
+ //fprintf( stderr, "sig_new: %f\n", sig_new );
+
+ for( i = 0; i < 300 && SQRT(sig_new) / b_norm > tol; ++i ) {
+ //for( i = 0; i < 300 && sig_new > SQR(tol)*sig0; ++i ) {
+ Sparse_MatVec( H, workspace->d, workspace->q );
+ tmp = Dot( workspace->d, workspace->q, N );
+ //fprintf( stderr, "tmp: %f\n", tmp );
+ alpha = sig_new / tmp;
+ Vector_Add( x, alpha, workspace->d, N );
+ //fprintf( stderr, "d_norm:%24.15e, q_norm:%24.15e, tmp:%24.15e\n",
+ // Norm(workspace->d,N), Norm(workspace->q,N), tmp );
+
+ Vector_Add( workspace->r, -alpha, workspace->q, N );
+ for( j = 0; j < N; ++j )
+ workspace->p[j] = workspace->r[j] * workspace->Hdia_inv[j];
+
+ sig_old = sig_new;
+ sig_new = Dot( workspace->r, workspace->p, N );
+ beta = sig_new / sig_old;
+ Vector_Sum( workspace->d, 1., workspace->p, beta, workspace->d, N );
+ //fprintf( stderr, "sig_new: %f\n", sig_new );
+ }
+
+ fprintf( stderr, "CG took %d iterations\n", i );
+
+ if( i >= 300 ) {
+ fprintf( stderr, "CG convergence failed!\n" );
+ return i;
+ }
+
+ return i;
+}
+
+
+/* Steepest Descent */
+int SDM( static_storage *workspace, sparse_matrix *H,
+ real *b, real tol, real *x, FILE *fout )
+{
+ int i, j, N;
+ real tmp, alpha, beta, b_norm;
+ real sig0, sig;
+
+ N = H->n;
+ b_norm = Norm( b, N );
+ //fprintf( stderr, "b_norm: %10.6f\n", b_norm );
+
+ Sparse_MatVec( H, x, workspace->q );
+ Vector_Sum( workspace->r , 1., b, -1., workspace->q, N );
+ for( j = 0; j < N; ++j )
+ workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
+
+ sig = Dot( workspace->r, workspace->d, N );
+ sig0 = sig;
+
+ for( i = 0; i < 300 && SQRT(sig) / b_norm > tol; ++i ) {
+ Sparse_MatVec( H, workspace->d, workspace->q );
+
+ sig = Dot( workspace->r, workspace->d, N );
+ tmp = Dot( workspace->d, workspace->q, N );
+ alpha = sig / tmp;
+
+ Vector_Add( x, alpha, workspace->d, N );
+ Vector_Add( workspace->r, -alpha, workspace->q, N );
+ for( j = 0; j < N; ++j )
+ workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j];
+
+ //fprintf( stderr, "d_norm:%24.15e, q_norm:%24.15e, tmp:%24.15e\n",
+ // Norm(workspace->d,N), Norm(workspace->q,N), tmp );
+ }
+
+ fprintf( stderr, "SDM took %d iterations\n", i );
+
+ if( i >= 300 ) {
+ fprintf( stderr, "SDM convergence failed!\n" );
+ return i;
+ }
+
+ return i;
+}
diff --git a/PuReMD-GPU/src/GMRES.h b/PuReMD-GPU/src/lin_alg.h
similarity index 89%
rename from PuReMD-GPU/src/GMRES.h
rename to PuReMD-GPU/src/lin_alg.h
index 5f9dc46bcd853ec1b2036c995b9bfc0675fd635c..a515a959494a6eca40fe9f338d2a08118ff3e39a 100644
--- a/PuReMD-GPU/src/GMRES.h
+++ b/PuReMD-GPU/src/lin_alg.h
@@ -18,19 +18,17 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __GMRES_H_
-#define __GMRES_H_
+#ifndef __LIN_ALG_H_
+#define __LIN_ALG_H_
#define SIGN(x) (x < 0.0 ? -1 : 1);
#include "mytypes.h"
+
int GMRES( static_storage*, sparse_matrix*,
real*, real, real*, FILE* , reax_system* );
-int Cuda_GMRES( static_storage *, real *b, real tol, real *x );
-int Cublas_GMRES( reax_system *, static_storage *, real *b, real tol, real *x );
-
int GMRES_HouseHolder( static_storage*, sparse_matrix*,
real*, real, real*, FILE* , reax_system* );
@@ -46,4 +44,5 @@ int CG( static_storage*, sparse_matrix*,
int uyduruk_GMRES( static_storage*, sparse_matrix*,
real*, real, real*, int, FILE*, reax_system* );
+
#endif
diff --git a/PuReMD-GPU/src/list.c b/PuReMD-GPU/src/list.c
new file mode 100644
index 0000000000000000000000000000000000000000..c6f0e55ebad4fc59c07f253a1d216d3242115aff
--- /dev/null
+++ b/PuReMD-GPU/src/list.c
@@ -0,0 +1,146 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "list.h"
+
+
+char Make_List(int n, int num_intrs, int type, list* l)
+{
+ char success=1;
+
+ l->n = n;
+ l->num_intrs = num_intrs;
+
+ l->index = (int*) malloc( n * sizeof(int) );
+ l->end_index = (int*) malloc( n * sizeof(int) );
+
+ if (l->index == NULL) success = 0;
+ if (l->end_index == NULL) success = 0;
+
+ l->type = type;
+
+ switch(type)
+ {
+ case TYP_VOID:
+ l->select.v = (void *) malloc(l->num_intrs*sizeof(void));
+ if (l->select.v == NULL) success = 0;
+ break;
+
+ case TYP_THREE_BODY:
+ l->select.three_body_list = (three_body_interaction_data*)
+ malloc(l->num_intrs*sizeof(three_body_interaction_data));
+ if (l->select.three_body_list == NULL) success = 0;
+ break;
+
+ case TYP_BOND:
+ l->select.bond_list = (bond_data*)
+ malloc(l->num_intrs * sizeof(bond_data));
+ if (l->select.bond_list == NULL) success = 0;
+ break;
+
+ case TYP_DBO:
+ l->select.dbo_list = (dbond_data*)
+ malloc(l->num_intrs * sizeof(dbond_data));
+ if (l->select.dbo_list == NULL) success = 0;
+ break;
+
+ case TYP_DDELTA:
+ l->select.dDelta_list = (dDelta_data*)
+ malloc(l->num_intrs*sizeof(dDelta_data));
+ if (l->select.dDelta_list == NULL) success = 0;
+ break;
+
+ case TYP_FAR_NEIGHBOR:
+ l->select.far_nbr_list = (far_neighbor_data*)
+ malloc(l->num_intrs*sizeof(far_neighbor_data));
+ if (l->select.far_nbr_list == NULL) success = 0;
+ break;
+
+ case TYP_NEAR_NEIGHBOR:
+ l->select.near_nbr_list = (near_neighbor_data*)
+ malloc(l->num_intrs*sizeof(near_neighbor_data));
+ if (l->select.near_nbr_list == NULL) success = 0;
+ break;
+
+ case TYP_HBOND:
+ l->select.hbond_list = (hbond_data*)
+ malloc( l->num_intrs * sizeof(hbond_data) );
+ if (l->select.hbond_list == NULL) success = 0;
+ break;
+
+ default:
+ l->select.v = (void *) malloc(l->num_intrs*sizeof(void));
+ if (l->select.v == NULL) success = 0;
+ l->type = TYP_VOID;
+ break;
+ }
+
+ return success;
+}
+
+
+void Delete_List(list* l)
+{
+ if( l->index != NULL )
+ free(l->index);
+ if( l->end_index != NULL )
+ free(l->end_index);
+
+ switch(l->type)
+ {
+ case TYP_VOID:
+ if( l->select.v != NULL )
+ free(l->select.v);
+ break;
+ case TYP_THREE_BODY:
+ if( l->select.three_body_list != NULL )
+ free(l->select.three_body_list);
+ break;
+ case TYP_BOND:
+ if( l->select.bond_list != NULL )
+ free(l->select.bond_list);
+ break;
+ case TYP_DBO:
+ if( l->select.dbo_list != NULL )
+ free(l->select.dbo_list);
+ break;
+ case TYP_DDELTA:
+ if( l->select.dDelta_list != NULL )
+ free(l->select.dDelta_list);
+ break;
+ case TYP_FAR_NEIGHBOR:
+ if( l->select.far_nbr_list != NULL )
+ free(l->select.far_nbr_list);
+ break;
+ case TYP_NEAR_NEIGHBOR:
+ if( l->select.near_nbr_list != NULL )
+ free(l->select.near_nbr_list);
+ break;
+ case TYP_HBOND:
+ if( l->select.hbond_list != NULL )
+ free(l->select.hbond_list);
+ break;
+
+ default:
+ // Report fatal error
+ break;
+ }
+}
+
diff --git a/PuReMD-GPU/src/list.cu b/PuReMD-GPU/src/list.cu
deleted file mode 100644
index 095409aa4fdc96102a5e321d1511737e37371eef..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/list.cu
+++ /dev/null
@@ -1,235 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "list.h"
-#include "cuda_utils.h"
-
-HOST char Make_List(int n, int num_intrs, int type, list* l, int proc)
-{
- char success=1;
-
- if (proc == TYP_HOST) {
-
- l->n = n;
- l->num_intrs = num_intrs;
-
- l->index = (int*) malloc( n * sizeof(int) );
- l->end_index = (int*) malloc( n * sizeof(int) );
-
- if (l->index == NULL) success = 0;
- if (l->end_index == NULL) success = 0;
-
- l->type = type;
-
- switch(type)
- {
- case TYP_VOID:
- l->select.v = (void *) malloc(l->num_intrs*sizeof(void));
- if (l->select.v == NULL) success = 0;
- break;
-
- case TYP_THREE_BODY:
- l->select.three_body_list = (three_body_interaction_data*)
- malloc(l->num_intrs*sizeof(three_body_interaction_data));
- if (l->select.three_body_list == NULL) success = 0;
- break;
-
- case TYP_BOND:
- l->select.bond_list = (bond_data*)
- malloc(l->num_intrs * sizeof(bond_data));
- if (l->select.bond_list == NULL) success = 0;
- break;
-
- case TYP_DBO:
- l->select.dbo_list = (dbond_data*)
- malloc(l->num_intrs * sizeof(dbond_data));
- if (l->select.dbo_list == NULL) success = 0;
- break;
-
- case TYP_DDELTA:
- l->select.dDelta_list = (dDelta_data*)
- malloc(l->num_intrs*sizeof(dDelta_data));
- if (l->select.dDelta_list == NULL) success = 0;
- break;
-
- case TYP_FAR_NEIGHBOR:
- l->select.far_nbr_list = (far_neighbor_data*)
- malloc(l->num_intrs*sizeof(far_neighbor_data));
- if (l->select.far_nbr_list == NULL) success = 0;
- break;
-
- case TYP_NEAR_NEIGHBOR:
- l->select.near_nbr_list = (near_neighbor_data*)
- malloc(l->num_intrs*sizeof(near_neighbor_data));
- if (l->select.near_nbr_list == NULL) success = 0;
- break;
-
- case TYP_HBOND:
- l->select.hbond_list = (hbond_data*)
- malloc( l->num_intrs * sizeof(hbond_data) );
- if (l->select.hbond_list == NULL) success = 0;
- break;
-
- default:
- l->select.v = (void *) malloc(l->num_intrs*sizeof(void));
- if (l->select.v == NULL) success = 0;
- l->type = TYP_VOID;
- break;
- }
-
- }
- else
- {
- l->n = n;
- l->num_intrs = num_intrs;
-
- cuda_malloc ((void **)&l->index, n * sizeof(int), 1, LIST_INDEX );
- cuda_malloc ((void **)&l->end_index, n * sizeof(int), 1, LIST_END_INDEX );
-
- switch(type)
- {
- case TYP_FAR_NEIGHBOR:
- cuda_malloc ((void **) &l->select.far_nbr_list,
- l->num_intrs*sizeof(far_neighbor_data),
- 1, LIST_FAR_NEIGHBOR_DATA);
- /*
- cudaHostAlloc ((void **) &l->select.far_nbr_list,
- l->num_intrs*sizeof(far_neighbor_data),
- cudaHostAllocMapped);
-
- cudaHostGetDevicePointer ( (void **) &l->select.far_nbr_list,
- (void *)l->select.far_nbr_list, 0);
- */
- break;
-
- case TYP_HBOND:
- cuda_malloc ((void **) &l->select.hbond_list,
- l->num_intrs * sizeof(hbond_data),
- 1, LIST_HBOND_DATA );
- break;
-
- case TYP_BOND:
- cuda_malloc ((void **) &l->select.bond_list,
- l->num_intrs * sizeof(bond_data),
- 1, LIST_BOND_DATA );
- break;
-
- case TYP_THREE_BODY:
- cuda_malloc ( (void **) &l->select.three_body_list,
- l->num_intrs * sizeof(three_body_interaction_data),
- 1, LIST_THREE_BODY_DATA );
- break;
-
- default:
- fprintf (stderr, "Unknown list creation \n" );
- exit (1);
- }
- }
-
- return success;
-}
-
-
-HOST void Delete_List(list* l, int type)
-{
-
- if (type == TYP_HOST )
- {
- if( l->index != NULL )
- free(l->index);
- if( l->end_index != NULL )
- free(l->end_index);
-
- switch(l->type)
- {
- case TYP_VOID:
- if( l->select.v != NULL )
- free(l->select.v);
- break;
- case TYP_THREE_BODY:
- if( l->select.three_body_list != NULL )
- free(l->select.three_body_list);
- break;
- case TYP_BOND:
- if( l->select.bond_list != NULL )
- free(l->select.bond_list);
- break;
- case TYP_DBO:
- if( l->select.dbo_list != NULL )
- free(l->select.dbo_list);
- break;
- case TYP_DDELTA:
- if( l->select.dDelta_list != NULL )
- free(l->select.dDelta_list);
- break;
- case TYP_FAR_NEIGHBOR:
- if( l->select.far_nbr_list != NULL )
- free(l->select.far_nbr_list);
- break;
- case TYP_NEAR_NEIGHBOR:
- if( l->select.near_nbr_list != NULL )
- free(l->select.near_nbr_list);
- break;
- case TYP_HBOND:
- if( l->select.hbond_list != NULL )
- free(l->select.hbond_list);
- break;
-
- default:
- // Report fatal error
- break;
- }
- }
- else
- {
- if (l->index != NULL)
- cuda_free (l->index, LIST_INDEX );
- if (l->end_index != NULL)
- cuda_free (l->end_index, LIST_END_INDEX );
-
- switch(type)
- {
- case TYP_FAR_NEIGHBOR:
- if (l->select.far_nbr_list != NULL)
- cuda_free (l->select.far_nbr_list, LIST_FAR_NEIGHBOR_DATA);
- break;
-
- case TYP_HBOND:
- if (l->select.hbond_list != NULL)
- cuda_free (l->select.hbond_list, LIST_HBOND_DATA );
- break;
-
- case TYP_BOND:
- if (l->select.bond_list != NULL)
- cuda_free (l->select.bond_list, LIST_BOND_DATA );
- break;
-
- case TYP_THREE_BODY:
- if (l->select.three_body_list != NULL)
- cuda_free ( l->select.three_body_list, LIST_THREE_BODY_DATA );
- break;
-
- default:
- fprintf (stderr, "Unknown list deletion \n" );
- exit (1);
- }
- }
-}
-
diff --git a/PuReMD-GPU/src/list.h b/PuReMD-GPU/src/list.h
index f341c2e270912e57beee4867bca9927a5c905633..b90c41419271ca6b859be08ea4005fbe9107c029 100644
--- a/PuReMD-GPU/src/list.h
+++ b/PuReMD-GPU/src/list.h
@@ -23,31 +23,36 @@
#include "mytypes.h"
-HOST char Make_List( int, int, int, list* , int proc = TYP_HOST);
-HOST void Delete_List( list* , int proc = TYP_HOST);
+char Make_List( int, int, int, list* );
+void Delete_List( list* );
-inline HOST_DEVICE int Num_Entries(int i, list* l)
+
+static inline HOST_DEVICE int Num_Entries(int i, list* l)
{
return l->end_index[i] - l->index[i];
}
-inline HOST_DEVICE int Start_Index(int i, list *l )
+
+static inline HOST_DEVICE int Start_Index(int i, list *l )
{
return l->index[i];
}
-inline HOST_DEVICE int End_Index( int i, list *l )
+
+static inline HOST_DEVICE int End_Index( int i, list *l )
{
return l->end_index[i];
}
-inline HOST_DEVICE void Set_Start_Index(int i, int val, list *l)
+
+static inline HOST_DEVICE void Set_Start_Index(int i, int val, list *l)
{
l->index[i] = val;
}
-inline HOST_DEVICE void Set_End_Index(int i, int val, list *l)
+
+static inline HOST_DEVICE void Set_End_Index(int i, int val, list *l)
{
l->end_index[i] = val;
}
diff --git a/PuReMD-GPU/src/lookup.c b/PuReMD-GPU/src/lookup.c
new file mode 100644
index 0000000000000000000000000000000000000000..c439709dc09c77775ed716a39db797fa8c831585
--- /dev/null
+++ b/PuReMD-GPU/src/lookup.c
@@ -0,0 +1,406 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "lookup.h"
+
+#include "two_body_interactions.h"
+
+#include "index_utils.h"
+
+
+void Make_Lookup_Table(real xmin, real xmax, int n,
+ lookup_function f, lookup_table* t)
+{
+ int i;
+
+ t->xmin = xmin;
+ t->xmax = xmax;
+ t->n = n;
+ t->dx = (xmax - xmin)/(n-1);
+ t->inv_dx = 1.0 / t->dx;
+ t->a = (n-1)/(xmax-xmin);
+ t->y = (real*) malloc(n*sizeof(real));
+
+ for(i=0; i < n; i++)
+ t->y[i] = f(i*t->dx + t->xmin);
+
+ // //fprintf(stdout,"dx = %lf\n",t->dx);
+ // for(i=0; i < n; i++)
+ // //fprintf( stdout,"%d %lf %lf %lf\n",
+ // i, i/t->a+t->xmin, t->y[i], exp(i/t->a+t->xmin) );
+}
+
+
+/* Fills solution into x. Warning: will modify c and d! */
+void Tridiagonal_Solve( const real *a, const real *b,
+ real *c, real *d, real *x, unsigned int n){
+ int i;
+ real id;
+
+ /* Modify the coefficients. */
+ c[0] /= b[0]; /* Division by zero risk. */
+ d[0] /= b[0]; /* Division by zero would imply a singular matrix. */
+ for(i = 1; i < n; i++){
+ id = (b[i] - c[i-1] * a[i]); /* Division by zero risk. */
+ c[i] /= id; /* Last value calculated is redundant. */
+ d[i] = (d[i] - d[i-1] * a[i])/id;
+ }
+
+ /* Now back substitute. */
+ x[n - 1] = d[n - 1];
+ for(i = n - 2; i >= 0; i--)
+ x[i] = d[i] - c[i] * x[i + 1];
+}
+
+
+void Natural_Cubic_Spline( const real *h, const real *f,
+ cubic_spline_coef *coef, unsigned int n )
+{
+ int i;
+ real *a, *b, *c, *d, *v;
+
+ /* allocate space for the linear system */
+ a = (real*) malloc( n * sizeof(real) );
+ b = (real*) malloc( n * sizeof(real) );
+ c = (real*) malloc( n * sizeof(real) );
+ d = (real*) malloc( n * sizeof(real) );
+ v = (real*) malloc( n * sizeof(real) );
+
+ /* build the linear system */
+ a[0] = a[1] = a[n-1] = 0;
+ for( i = 2; i < n-1; ++i )
+ a[i] = h[i-1];
+
+ b[0] = b[n-1] = 0;
+ for( i = 1; i < n-1; ++i )
+ b[i] = 2 * (h[i-1] + h[i]);
+
+ c[0] = c[n-2] = c[n-1] = 0;
+ for( i = 1; i < n-2; ++i )
+ c[i] = h[i];
+
+ d[0] = d[n-1] = 0;
+ for( i = 1; i < n-1; ++i )
+ d[i] = 6 * ((f[i+1]-f[i])/h[i] - (f[i]-f[i-1])/h[i-1]);
+
+ /*//fprintf( stderr, "i a b c d\n" );
+ for( i = 0; i < n; ++i )
+ //fprintf( stderr, "%d %f %f %f %f\n", i, a[i], b[i], c[i], d[i] );*/
+ v[0] = 0;
+ v[n-1] = 0;
+ Tridiagonal_Solve( &(a[1]), &(b[1]), &(c[1]), &(d[1]), &(v[1]), n-2 );
+
+ for( i = 1; i < n; ++i ){
+ coef[i-1].d = (v[i] - v[i-1]) / (6*h[i-1]);
+ coef[i-1].c = v[i]/2;
+ coef[i-1].b = (f[i]-f[i-1])/h[i-1] + h[i-1]*(2*v[i] + v[i-1])/6;
+ coef[i-1].a = f[i];
+ }
+
+ /*//fprintf( stderr, "i v coef\n" );
+ for( i = 0; i < n; ++i )
+ //fprintf( stderr, "%d %f %f %f %f %f\n",
+ i, v[i], coef[i].a, coef[i].b, coef[i].c, coef[i].d ); */
+}
+
+
+void Complete_Cubic_Spline( const real *h, const real *f, real v0, real vlast,
+ cubic_spline_coef *coef, unsigned int n )
+{
+ int i;
+ real *a, *b, *c, *d, *v;
+
+ /* allocate space for the linear system */
+ a = (real*) malloc( n * sizeof(real) );
+ b = (real*) malloc( n * sizeof(real) );
+ c = (real*) malloc( n * sizeof(real) );
+ d = (real*) malloc( n * sizeof(real) );
+ v = (real*) malloc( n * sizeof(real) );
+
+ /* build the linear system */
+ a[0] = 0;
+ for( i = 1; i < n; ++i )
+ a[i] = h[i-1];
+
+ b[0] = 2*h[0];
+ for( i = 1; i < n; ++i )
+ b[i] = 2 * (h[i-1] + h[i]);
+
+ c[n-1] = 0;
+ for( i = 0; i < n-1; ++i )
+ c[i] = h[i];
+
+ d[0] = 6 * (f[1]-f[0])/h[0] - 6 * v0;
+ d[n-1] = 6 * vlast - 6 * (f[n-1]-f[n-2]/h[n-2]);
+ for( i = 1; i < n-1; ++i )
+ d[i] = 6 * ((f[i+1]-f[i])/h[i] - (f[i]-f[i-1])/h[i-1]);
+
+ /*//fprintf( stderr, "i a b c d\n" );
+ for( i = 0; i < n; ++i )
+ //fprintf( stderr, "%d %f %f %f %f\n", i, a[i], b[i], c[i], d[i] );*/
+ Tridiagonal_Solve( &(a[0]), &(b[0]), &(c[0]), &(d[0]), &(v[0]), n );
+ // Tridiagonal_Solve( &(a[1]), &(b[1]), &(c[1]), &(d[1]), &(v[1]), n-2 );
+
+ for( i = 1; i < n; ++i ){
+ coef[i-1].d = (v[i] - v[i-1]) / (6*h[i-1]);
+ coef[i-1].c = v[i]/2;
+ coef[i-1].b = (f[i]-f[i-1])/h[i-1] + h[i-1]*(2*v[i] + v[i-1])/6;
+ coef[i-1].a = f[i];
+ }
+
+ /*//fprintf( stderr, "i v coef\n" );
+ for( i = 0; i < n; ++i )
+ //fprintf( stderr, "%d %f %f %f %f %f\n",
+ i, v[i], coef[i].a, coef[i].b, coef[i].c, coef[i].d ); */
+}
+
+
+void LR_Lookup( LR_lookup_table *t, real r, LR_data *y )
+{
+ int i;
+ real base, dif;
+
+ i = (int)(r * t->inv_dx);
+ if( i == 0 ) ++i;
+ base = (real)(i+1) * t->dx;
+ dif = r - base;
+ ////fprintf( stderr, "r: %f, i: %d, base: %f, dif: %f\n", r, i, base, dif );
+
+ y->e_vdW = ((t->vdW[i].d*dif + t->vdW[i].c)*dif + t->vdW[i].b)*dif +
+ t->vdW[i].a;
+ y->CEvd = ((t->CEvd[i].d*dif + t->CEvd[i].c)*dif +
+ t->CEvd[i].b)*dif + t->CEvd[i].a;
+ //y->CEvd = (3*t->vdW[i].d*dif + 2*t->vdW[i].c)*dif + t->vdW[i].b;
+
+ y->e_ele = ((t->ele[i].d*dif + t->ele[i].c)*dif + t->ele[i].b)*dif +
+ t->ele[i].a;
+ y->CEclmb = ((t->CEclmb[i].d*dif + t->CEclmb[i].c)*dif + t->CEclmb[i].b)*dif +
+ t->CEclmb[i].a;
+
+ y->H = y->e_ele * EV_to_KCALpMOL / C_ele;
+ //y->H = ((t->H[i].d*dif + t->H[i].c)*dif + t->H[i].b)*dif + t->H[i].a;
+}
+
+
+void Make_LR_Lookup_Table( reax_system *system, control_params *control )
+{
+ int i, j, r;
+ int num_atom_types;
+ int existing_types[MAX_ATOM_TYPES];
+ real dr;
+ real *h, *fh, *fvdw, *fele, *fCEvd, *fCEclmb;
+ real v0_vdw, v0_ele, vlast_vdw, vlast_ele;
+ /* real rand_dist;
+ real evdw_abserr, evdw_relerr, fvdw_abserr, fvdw_relerr;
+ real eele_abserr, eele_relerr, fele_abserr, fele_relerr;
+ real evdw_maxerr, eele_maxerr;
+ LR_data y, y_spline; */
+
+ /* initializations */
+ vlast_ele = 0;
+ vlast_vdw = 0;
+ v0_ele = 0;
+ v0_vdw = 0;
+
+ num_atom_types = system->reaxprm.num_atom_types;
+ dr = control->r_cut / control->tabulate;
+ h = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+ fh = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+ fvdw = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+ fCEvd = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+ fele = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+ fCEclmb = (real*) malloc( (control->tabulate+1) * sizeof(real) );
+
+ /* allocate Long-Range LookUp Table space based on
+ number of atom types in the ffield file */
+ //LR = (LR_lookup_table**) malloc( num_atom_types * sizeof(LR_lookup_table*) );
+ //for( i = 0; i < num_atom_types; ++i )
+ // LR[i] = (LR_lookup_table*) malloc(num_atom_types * sizeof(LR_lookup_table));
+
+ LR = (LR_lookup_table*) malloc(num_atom_types * num_atom_types * sizeof(LR_lookup_table));
+
+ /* most atom types in ffield file will not exist in the current
+ simulation. to avoid unnecessary lookup table space, determine
+ the atom types that exist in the current simulation */
+ for( i = 0; i < MAX_ATOM_TYPES; ++i )
+ existing_types[i] = 0;
+ for( i = 0; i < system->N; ++i )
+ existing_types[ system->atoms[i].type ] = 1;
+
+ /* fill in the lookup table entries for existing atom types.
+ only lower half should be enough. */
+ for( i = 0; i < num_atom_types; ++i )
+ if( existing_types[i] )
+ for( j = i; j < num_atom_types; ++j )
+ if( existing_types[j] ) {
+ LR[ index_lr (i,j,num_atom_types) ].xmin = 0;
+ LR[ index_lr (i,j,num_atom_types) ].xmax = control->r_cut;
+ LR[ index_lr (i,j,num_atom_types) ].n = control->tabulate + 1;
+ LR[ index_lr (i,j,num_atom_types) ].dx = dr;
+ LR[ index_lr (i,j,num_atom_types) ].inv_dx = control->tabulate / control->r_cut;
+ LR[ index_lr (i,j,num_atom_types) ].y = (LR_data*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(LR_data));
+ LR[ index_lr (i,j,num_atom_types) ].H = (cubic_spline_coef*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
+ LR[ index_lr (i,j,num_atom_types) ].vdW = (cubic_spline_coef*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
+ LR[ index_lr (i,j,num_atom_types) ].CEvd = (cubic_spline_coef*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
+ LR[ index_lr (i,j,num_atom_types) ].ele = (cubic_spline_coef*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
+ LR[ index_lr (i,j,num_atom_types) ].CEclmb = (cubic_spline_coef*)
+ malloc(LR[ index_lr (i,j,num_atom_types) ].n * sizeof(cubic_spline_coef));
+
+ for( r = 1; r <= control->tabulate; ++r ) {
+ LR_vdW_Coulomb( system, control, i, j, r * dr, &(LR[ index_lr (i,j,num_atom_types) ].y[r]) );
+ h[r] = LR[ index_lr (i,j,num_atom_types) ].dx;
+ fh[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].H;
+ fvdw[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].e_vdW;
+ fCEvd[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
+ fele[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].e_ele;
+ fCEclmb[r] = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
+
+ if( r == 1 ){
+ v0_vdw = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
+ v0_ele = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
+ }
+ else if( r == control->tabulate ){
+ vlast_vdw = LR[ index_lr (i,j,num_atom_types) ].y[r].CEvd;
+ vlast_ele = LR[ index_lr (i,j,num_atom_types) ].y[r].CEclmb;
+ }
+ }
+
+ /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fh" );
+ for( r = 1; r <= control->tabulate; ++r )
+ //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fh[r] ); */
+ Natural_Cubic_Spline( &h[1], &fh[1],
+ &(LR[ index_lr (i,j,num_atom_types) ].H[1]), control->tabulate+1 );
+
+ /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fvdw" );
+ for( r = 1; r <= control->tabulate; ++r )
+ //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fvdw[r] );
+ //fprintf( stderr, "v0_vdw: %f, vlast_vdw: %f\n", v0_vdw, vlast_vdw );
+ */
+ Complete_Cubic_Spline( &h[1], &fvdw[1], v0_vdw, vlast_vdw,
+ &(LR[ index_lr (i,j,num_atom_types) ].vdW[1]), control->tabulate+1 );
+ Natural_Cubic_Spline( &h[1], &fCEvd[1],
+ &(LR[ index_lr (i,j,num_atom_types) ].CEvd[1]), control->tabulate+1 );
+
+ /*//fprintf( stderr, "%-6s %-6s %-6s\n", "r", "h", "fele" );
+ for( r = 1; r <= control->tabulate; ++r )
+ //fprintf( stderr, "%f %f %f\n", r * dr, h[r], fele[r] );
+ //fprintf( stderr, "v0_ele: %f, vlast_ele: %f\n", v0_ele, vlast_ele );
+ */
+ Complete_Cubic_Spline( &h[1], &fele[1], v0_ele, vlast_ele,
+ &(LR[ index_lr (i,j,num_atom_types) ].ele[1]), control->tabulate+1 );
+ Natural_Cubic_Spline( &h[1], &fCEclmb[1],
+ &(LR[ index_lr (i,j,num_atom_types) ].CEclmb[1]), control->tabulate+1 );
+ }
+
+ /***** //test LR-Lookup table
+ evdw_maxerr = 0;
+ eele_maxerr = 0;
+ for( i = 0; i < num_atom_types; ++i )
+ if( existing_types[i] )
+ for( j = i; j < num_atom_types; ++j )
+ if( existing_types[j] ) {
+ for( r = 1; r <= 100; ++r ) {
+ rand_dist = (real)rand()/RAND_MAX * control->r_cut;
+ LR_vdW_Coulomb( system, control, i, j, rand_dist, &y );
+ LR_Lookup( &(LR[i][j]), rand_dist, &y_spline );
+
+ evdw_abserr = fabs(y.e_vdW - y_spline.e_vdW);
+ evdw_relerr = fabs(evdw_abserr / y.e_vdW);
+ fvdw_abserr = fabs(y.CEvd - y_spline.CEvd);
+ fvdw_relerr = fabs(fvdw_abserr / y.CEvd);
+ eele_abserr = fabs(y.e_ele - y_spline.e_ele);
+ eele_relerr = fabs(eele_abserr / y.e_ele);
+ fele_abserr = fabs(y.CEclmb - y_spline.CEclmb);
+ fele_relerr = fabs(fele_abserr / y.CEclmb);
+
+ if( evdw_relerr > 1e-10 || eele_relerr > 1e-10 ){
+ //fprintf( stderr, "rand_dist = %24.15e\n", rand_dist );
+ //fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
+ y.H, y_spline.H,
+ fabs(y.H-y_spline.H), fabs((y.H-y_spline.H)/y.H) );
+
+ //fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
+ y.e_vdW, y_spline.e_vdW, evdw_abserr, evdw_relerr );
+ //fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
+ y.CEvd, y_spline.CEvd, fvdw_abserr, fvdw_relerr );
+
+ //fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
+ y.e_ele, y_spline.e_ele, eele_abserr, eele_relerr );
+ //fprintf( stderr, "%24.15e %24.15e %24.15e %24.15e\n",
+ y.CEclmb, y_spline.CEclmb, fele_abserr, fele_relerr );
+ }
+
+ if( evdw_relerr > evdw_maxerr )
+ evdw_maxerr = evdw_relerr;
+ if( eele_relerr > eele_maxerr )
+ eele_maxerr = eele_relerr;
+ }
+ }
+ //fprintf( stderr, "evdw_maxerr: %24.15e\n", evdw_maxerr );
+ //fprintf( stderr, "eele_maxerr: %24.15e\n", eele_maxerr );
+ *******/
+
+ free(h);
+ free(fh);
+ free(fvdw);
+ free(fCEvd);
+ free(fele);
+ free(fCEclmb);
+}
+
+
+int Lookup_Index_Of( real x, lookup_table* t )
+{
+ return (int)( t->a * ( x - t->xmin ) );
+}
+
+
+real Lookup( real x, lookup_table* t )
+{
+ real x1, x2;
+ real b;
+ int i;
+
+ /* if ( x < t->xmin)
+ {
+ //fprintf(stderr,"Domain check %lf > %lf\n",t->xmin,x);
+ exit(0);
+ }
+ if ( x > t->xmax)
+ {
+ //fprintf(stderr,"Domain check %lf < %lf\n",t->xmax,x);
+ exit(0);
+ } */
+
+ i = Lookup_Index_Of( x, t );
+ x1 = i * t->dx + t->xmin;
+ x2 = (i+1) * t->dx + t->xmin;
+
+ b = ( x2 * t->y[i] - x1 * t->y[i+1] ) * t->inv_dx;
+ // //fprintf( stdout,"SLookup_Entry: %d, %lf, %lf, %lf, %lf: %lf, %lf\n",
+ // i,x1,x2,x,b,t->one_over_dx*(t->y[i+1]-t->y[i])*x+b,exp(x));
+
+ return t->inv_dx * ( t->y[i+1] - t->y[i] ) * x + b;
+}
diff --git a/PuReMD-GPU/src/lookup.h b/PuReMD-GPU/src/lookup.h
index 9ea972f8eeaa07e307516149658c07bfc68f37f4..7dac3e4f41764caa53cf0e2313f8c566f4e0e634 100644
--- a/PuReMD-GPU/src/lookup.h
+++ b/PuReMD-GPU/src/lookup.h
@@ -23,14 +23,20 @@
#include "mytypes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Make_Lookup_Table( real, real, int, lookup_function, lookup_table* );
int Lookup_Index_Of( real, lookup_table* );
real Lookup( real, lookup_table* );
void Make_LR_Lookup_Table( reax_system*, control_params* );
-//CUDA Functions
-void Cuda_Make_LR_Lookup_Table( reax_system*, control_params* );
-void copy_LR_table_to_device ( reax_system*, control_params* );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PuReMD-GPU/src/matvec.cu b/PuReMD-GPU/src/matvec.cu
deleted file mode 100644
index bf08cdf83dc8e14e31b48a6eaac41b5bfa8cf97e..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/matvec.cu
+++ /dev/null
@@ -1,89 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-
-#include "matvec.h"
-
-//one thread per row
-GLOBAL void Cuda_Matvec (sparse_matrix H, real *vec, real *results, int rows)
-{
- real results_row = 0;
- int col;
- real val;
-
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if ( i >= rows) return;
-
- for (int c = H.start[i]; c < H.end[i]; c++)
- {
- col = H.entries [c].j;
- val = H.entries[c].val;
-
- results_row += val * vec [col];
- }
-
- results [i] = results_row;
-}
-
-//32 thread warp per matrix row.
-//invoked as follows
-// <<< system->N, 32 >>>
-GLOBAL void Cuda_Matvec_csr (sparse_matrix H, real *vec, real *results, int num_rows)
-{
- extern __shared__ real vals [];
- int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
- int warp_id = thread_id / 32;
- int lane = thread_id & (32 - 1);
-
- int row_start;
- int row_end;
-
- // one warp per row
- //int row = warp_id;
- int row = warp_id;
- //if (row < num_rows)
- {
- vals[threadIdx.x] = 0;
-
- if (row < num_rows) {
- row_start = H.start[row];
- row_end = H.end[row];
-
- // compute running sum per thread
- for(int jj = row_start + lane; jj < row_end; jj += 32)
- vals[threadIdx.x] += H.entries[jj].val * vec [ H.entries[jj].j ];
- //vals[threadIdx.x] += H.val[jj] * vec [ H.j[jj] ];
- }
-
- __syncthreads ();
-
- // parallel reduction in shared memory
- //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
- if (lane < 16) vals[threadIdx.x] += vals[threadIdx.x + 16]; __syncthreads();
- if (lane < 8) vals[threadIdx.x] += vals[threadIdx.x + 8]; __syncthreads ();
- if (lane < 4) vals[threadIdx.x] += vals[threadIdx.x + 4]; __syncthreads ();
- if (lane < 2) vals[threadIdx.x] += vals[threadIdx.x + 2]; __syncthreads ();
- if (lane < 1) vals[threadIdx.x] += vals[threadIdx.x + 1]; __syncthreads ();
-
- // first thread writes the result
- if (lane == 0 && row < num_rows)
- results[row] = vals[threadIdx.x];
- }
-}
diff --git a/PuReMD-GPU/src/mytypes.h b/PuReMD-GPU/src/mytypes.h
index c7d42ee97e386aae84f4f9d9bc51d2767d866f29..273f95976159c55461a84668c4a5f2494f1d9522 100644
--- a/PuReMD-GPU/src/mytypes.h
+++ b/PuReMD-GPU/src/mytypes.h
@@ -18,8 +18,42 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __MYTYPES_H_
-#define __MYTYPES_H_
+#if !(defined(__MYTYPES_H_) || defined(__CUDA_MYTYPES_H_))
+
+#ifdef __CUDACC__
+ #ifndef __CUDA_MYTYPES_H_
+ #define __CUDA_MYTYPES_H_
+ #define HOST __host__
+ #define DEVICE __device__
+ #define GLOBAL __global__
+ #define HOST_DEVICE __host__ __device__
+
+ #include <cuda_runtime.h>
+ #include <cuda.h>
+ #include <cuda_runtime_api.h>
+
+ #include <cublas_v2.h>
+ #include <cusparse_v2.h>
+ #if __CUDA_ARCH__ < 600
+ #define MYATOMICADD myAtomicAdd
+ #else
+ #define MYATOMICADD atomicAdd
+ #endif
+ #endif
+#else
+ #ifndef __MYTYPES_H_
+ #define __MYTYPES_H_
+ #define HOST
+ #define DEVICE
+ #define GLOBAL
+ #define HOST_DEVICE
+ #endif
+#endif
+
+#if (defined(HAVE_CONFIG_H) && !defined(__CONFIG_H_))
+ #define __CONFIG_H_
+ #include "config.h"
+#endif
#include "math.h"
//#include "random.h"
@@ -30,28 +64,16 @@
#include "time.h"
#include "zlib.h"
-
//#define DEBUG_FOCUS
//#define TEST_FORCES
//#define TEST_ENERGY
//#define REORDER_ATOMS // turns on nbrgen opt by re-ordering atoms
//#define LGJ
-#ifdef __USE_GPU__
-
-#include "cublas_v2.h"
-#include "cusparse_v2.h"
-
-#define HOST __host__
-#define DEVICE __device__
-#define GLOBAL __global__
-#define HOST_DEVICE __host__ __device__
-#else
-#define HOST
-#define DEVICE
-#define GLOBAL
-#define HOST_DEVICE
-#endif
+#define SUCCESS 1
+#define FAILURE 0
+#define TRUE 1
+#define FALSE 0
#define EXP exp
#define SQRT sqrt
@@ -140,10 +162,10 @@
#define RES_GRID_MARK 0x03
#define RES_GRID_START 0x04
#define RES_GRID_END 0x05
-#define RES_GRID_NBRS 0x06
-#define RES_GRID_NBRS_CP 0x07
+#define RES_GRID_NBRS 0x06
+#define RES_GRID_NBRS_CP 0x07
-#define RES_SYSTEM_ATOMS 0x10
+#define RES_SYSTEM_ATOMS 0x10
#define RES_SYSTEM_SIMULATION_BOX 0x11
#define RES_REAX_INT_SBP 0x20
@@ -154,58 +176,58 @@
#define RES_SIMULATION_DATA 0x30
-#define RES_STORAGE 0x401
-#define RES_STORAGE_HBOND_INDEX 0x402
-#define RES_STORAGE_TOTAL_BOND_ORDER 0x403
-#define RES_STORAGE_DELTAP 0x404
-#define RES_STORAGE_DELTAP_BOC 0x404
-#define RES_STORAGE_DDELTAP_SELF 0x405
-#define RES_STORAGE_DELTA 0x406
-#define RES_STORAGE_DELTA_LP 0x407
-#define RES_STORAGE_DELTA_LP_TEMP 0x408
-#define RES_STORAGE_DDELTA_LP 0x409
-#define RES_STORAGE_DDELTA_LP_TEMP 0x40A
-#define RES_STORAGE_DELTA_E 0x40B
-#define RES_STORAGE_DELTA_BOC 0x40C
-#define RES_STORAGE_NL 0x40D
-#define RES_STORAGE_NLP_TEMP 0x40E
-#define RES_STORAGE_CLP 0x40F
-#define RES_STORAGE_CDDELTA 0x410
-#define RES_STORAGE_VLPEX 0x411
-#define RES_STORAGE_DROPTOL 0x412
-#define RES_STORAGE_W 0x413
-#define RES_STORAGE_HDIA_INV 0x414
-#define RES_STORAGE_B 0x415
-#define RES_STORAGE_B_S 0x416
-#define RES_STORAGE_B_T 0x417
-#define RES_STORAGE_B_PRC 0x418
-#define RES_STORAGE_B_PRM 0x419
-#define RES_STORAGE_S_T 0x41A
-#define RES_STORAGE_S 0x41B
-#define RES_STORAGE_T 0x41C
-#define RES_STORAGE_Y 0x41D
-#define RES_STORAGE_Z 0x41E
-#define RES_STORAGE_G 0x41F
-#define RES_STORAGE_HS 0x420
-#define RES_STORAGE_HC 0x421
-#define RES_STORAGE_RN 0x422
-#define RES_STORAGE_V 0x423
-#define RES_STORAGE_H 0x424
-#define RES_STORAGE_R 0x425
-#define RES_STORAGE_D 0x426
-#define RES_STORAGE_Q 0x427
-#define RES_STORAGE_P 0x428
-#define RES_STORAGE_A 0x429
-#define RES_STORAGE_F_OLD 0x42A
-#define RES_STORAGE_V_CONST 0x42B
-#define RES_STORAGE_MARK 0x42C
-#define RES_STORAGE_OLD_MARK 0x42D
-#define RES_STORAGE_X_OLD 0x42E
-#define RES_STORAGE_NLP 0x42F
-#define RES_STORAGE_MAP_SERIALS 0x430
-#define RES_STORAGE_RESTRICTED 0x431
-#define RES_STORAGE_RESTRICTED_LIST 0x432
-#define RES_STORAGE_ORIG_ID 0x433
+#define RES_STORAGE 0x401
+#define RES_STORAGE_HBOND_INDEX 0x402
+#define RES_STORAGE_TOTAL_BOND_ORDER 0x403
+#define RES_STORAGE_DELTAP 0x404
+#define RES_STORAGE_DELTAP_BOC 0x404
+#define RES_STORAGE_DDELTAP_SELF 0x405
+#define RES_STORAGE_DELTA 0x406
+#define RES_STORAGE_DELTA_LP 0x407
+#define RES_STORAGE_DELTA_LP_TEMP 0x408
+#define RES_STORAGE_DDELTA_LP 0x409
+#define RES_STORAGE_DDELTA_LP_TEMP 0x40A
+#define RES_STORAGE_DELTA_E 0x40B
+#define RES_STORAGE_DELTA_BOC 0x40C
+#define RES_STORAGE_NL 0x40D
+#define RES_STORAGE_NLP_TEMP 0x40E
+#define RES_STORAGE_CLP 0x40F
+#define RES_STORAGE_CDDELTA 0x410
+#define RES_STORAGE_VLPEX 0x411
+#define RES_STORAGE_DROPTOL 0x412
+#define RES_STORAGE_W 0x413
+#define RES_STORAGE_HDIA_INV 0x414
+#define RES_STORAGE_B 0x415
+#define RES_STORAGE_B_S 0x416
+#define RES_STORAGE_B_T 0x417
+#define RES_STORAGE_B_PRC 0x418
+#define RES_STORAGE_B_PRM 0x419
+#define RES_STORAGE_S_T 0x41A
+#define RES_STORAGE_S 0x41B
+#define RES_STORAGE_T 0x41C
+#define RES_STORAGE_Y 0x41D
+#define RES_STORAGE_Z 0x41E
+#define RES_STORAGE_G 0x41F
+#define RES_STORAGE_HS 0x420
+#define RES_STORAGE_HC 0x421
+#define RES_STORAGE_RN 0x422
+#define RES_STORAGE_V 0x423
+#define RES_STORAGE_H 0x424
+#define RES_STORAGE_R 0x425
+#define RES_STORAGE_D 0x426
+#define RES_STORAGE_Q 0x427
+#define RES_STORAGE_P 0x428
+#define RES_STORAGE_A 0x429
+#define RES_STORAGE_F_OLD 0x42A
+#define RES_STORAGE_V_CONST 0x42B
+#define RES_STORAGE_MARK 0x42C
+#define RES_STORAGE_OLD_MARK 0x42D
+#define RES_STORAGE_X_OLD 0x42E
+#define RES_STORAGE_NLP 0x42F
+#define RES_STORAGE_MAP_SERIALS 0x430
+#define RES_STORAGE_RESTRICTED 0x431
+#define RES_STORAGE_RESTRICTED_LIST 0x432
+#define RES_STORAGE_ORIG_ID 0x433
#define RES_CONTROL_PARAMS 0x50
@@ -224,7 +246,6 @@
#define RES_SCRATCH 0x90
-
#define LIST_INDEX 0x00
#define LIST_END_INDEX 0x01
#define LIST_FAR_NEIGHBOR_DATA 0x10
@@ -288,9 +309,6 @@
#define MATVEC_THREADS_PER_ROW 32
-
-enum {TYP_HOST, TYP_DEVICE};
-
typedef double real;
typedef real rvec[3];
typedef int ivec[3];
@@ -309,7 +327,6 @@ enum {WRITE_ASCII, WRITE_BINARY, RF_N};
enum {XYZ, PDB, BGF, ASCII_RESTART, BINARY_RESTART, GF_N};
-
/* Global params mapping */
/*
l[0] = p_boc1
@@ -352,7 +369,6 @@ l[36] = N/A
l[37] = version number
l[38] = p_coa3
*/
-
typedef struct
{
int n_global;
@@ -361,7 +377,6 @@ typedef struct
} global_parameters;
-
typedef struct
{
/* Line one in field file */
@@ -405,7 +420,6 @@ typedef struct
} single_body_parameters;
-
/* Two Body Parameters */
typedef struct
{
@@ -435,7 +449,6 @@ typedef struct
} two_body_parameters;
-
/* 3-body parameters */
typedef struct
{
@@ -458,7 +471,6 @@ typedef struct
} three_body_header;
-
/* hydrogen-bond parameters */
typedef struct
{
@@ -466,7 +478,6 @@ typedef struct
} hbond_parameters;
-
/* 4-body parameters */
typedef struct
{
@@ -560,7 +571,6 @@ typedef struct
int *end;
ivec *nbrs;
rvec *nbrs_cp;
-
} grid;
@@ -768,8 +778,6 @@ typedef struct
reax_timing timing;
//CUDA
reax_timing d_timing;
-
-
void *d_simulation_data;
} simulation_data;
@@ -837,6 +845,7 @@ typedef struct
rvec dBO, dBOpi, dBOpi2;
} dbond_data;
+
typedef struct
{
real BO, BO_s, BO_pi, BO_pi2;
@@ -847,6 +856,7 @@ typedef struct
rvec dBOp, dln_BOp_s, dln_BOp_pi, dln_BOp_pi2;
} bond_order_data;
+
typedef struct
{
int nbr;
@@ -886,6 +896,7 @@ typedef struct
real val;
} sparse_matrix_entry;
+
typedef struct
{
int n, m;
@@ -914,6 +925,7 @@ typedef struct
int gcell_atoms;
} reallocate_data;
+
typedef struct
{
/* bond order related storage */
@@ -999,7 +1011,6 @@ typedef struct
} list;
-
typedef struct
{
FILE *trj;
@@ -1070,12 +1081,12 @@ typedef struct
} LR_data;
-
typedef struct
{
real a, b, c, d;
} cubic_spline_coef;
+
typedef struct
{
real xmin, xmax;
@@ -1126,8 +1137,7 @@ typedef void (*get_far_neighbors_function)(rvec, rvec, simulation_box*,
int*);
-// CUDA structures
-//
+/* CUDA structures */
extern list *dev_lists;
extern static_storage *dev_workspace;
extern LR_lookup_table *d_LR;
@@ -1138,15 +1148,5 @@ extern void *scratch;
extern int BLOCKS, BLOCKS_POW_2, BLOCK_SIZE;
extern int MATVEC_BLOCKS;
-#ifdef __USE_GPU__
-extern cublasStatus_t cublasStatus;
-extern cublasHandle_t cublasHandle;
-
-extern cusparseHandle_t cusparseHandle;
-extern cusparseStatus_t cusparseStatus;
-extern cusparseMatDescr_t matdescriptor;
-#endif
-
-
#endif
diff --git a/PuReMD-GPU/src/neighbors.c b/PuReMD-GPU/src/neighbors.c
new file mode 100644
index 0000000000000000000000000000000000000000..5f425e672080d2d4a272f7aca1859c45d8dde17d
--- /dev/null
+++ b/PuReMD-GPU/src/neighbors.c
@@ -0,0 +1,698 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "neighbors.h"
+
+#include "box.h"
+#include "grid.h"
+#include "index_utils.h"
+#include "list.h"
+#include "reset_utils.h"
+#include "system_props.h"
+#include "vector.h"
+
+
+int Are_Far_Neighbors( rvec x1, rvec x2, simulation_box *box,
+ real cutoff, far_neighbor_data *data )
+{
+ real norm_sqr, d, tmp;
+ int i;
+
+ norm_sqr = 0;
+
+ for( i = 0; i < 3; i++ ) {
+ d = x2[i] - x1[i];
+ tmp = SQR(d);
+
+ if( tmp >= SQR( box->box_norms[i] / 2.0 ) ) {
+ if( x2[i] > x1[i] ) {
+ d -= box->box_norms[i];
+ data->rel_box[i] = -1;
+ }
+ else {
+ d += box->box_norms[i];
+ data->rel_box[i] = +1;
+ }
+
+ data->dvec[i] = d;
+ norm_sqr += SQR(d);
+ }
+ else {
+ data->dvec[i] = d;
+ norm_sqr += tmp;
+ data->rel_box[i] = 0;
+ }
+ }
+
+ if( norm_sqr <= SQR(cutoff) ){
+ data->d = sqrt(norm_sqr);
+ return 1;
+ }
+
+ return 0;
+}
+
+
+void Generate_Neighbor_Lists( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, k, l, m, itr;
+ int x, y, z;
+ int atom1, atom2, max;
+ int num_far;
+ int *nbr_atoms;
+ ivec *nbrs;
+ rvec *nbrs_cp;
+ grid *g;
+ list *far_nbrs;
+ far_neighbor_data *nbr_data;
+ real t_start, t_elapsed;
+
+ // fprintf( stderr, "\n\tentered nbrs - " );
+ g = &( system->g );
+ far_nbrs = (*lists) + FAR_NBRS;
+ Bin_Atoms( system, workspace );
+
+ t_start = Get_Time( );
+
+ // fprintf( stderr, "atoms sorted - " );
+ num_far = 0;
+
+ /* first pick up a cell in the grid */
+ for( i = 0; i < g->ncell[0]; i++ )
+ for( j = 0; j < g->ncell[1]; j++ )
+ for( k = 0; k < g->ncell[2]; k++ ) {
+ nbrs = &g->nbrs[ index_grid_nbrs (i,j,k,0,g) ];
+ nbrs_cp = &g->nbrs_cp[ index_grid_nbrs (i,j,k,0,g) ];
+ //fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
+
+ /* pick up an atom from the current cell */
+ for(l = 0; l < g->top[ index_grid_3d (i,j,k,g) ]; ++l ){
+ atom1 = g->atoms[ index_grid_atoms (i,j,k,l,g) ];
+ Set_Start_Index( atom1, num_far, far_nbrs );
+ //fprintf( stderr, "\tatom %d\n", atom1 );
+
+ itr = 0;
+ while( nbrs[itr][0] >= 0 ){
+ x = nbrs[itr][0];
+ y = nbrs[itr][1];
+ z = nbrs[itr][2];
+ //fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
+
+ if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
+ SQR(control->vlist_cut) ) {
+ nbr_atoms = &g->atoms[ index_grid_atoms (x,y,z,0,g) ];
+ max = g->top[ index_grid_3d (x,y,z,g) ];
+ //fprintf( stderr, "\t\tmax: %d\n", max );
+
+ /* pick up another atom from the neighbor cell */
+ for( m = 0; m < max; ++m ) {
+ atom2 = nbr_atoms[m];
+ if( atom1 > atom2 ) {
+ nbr_data = &(far_nbrs->select.far_nbr_list[num_far]);
+ if(Are_Far_Neighbors(system->atoms[atom1].x,
+ system->atoms[atom2].x,
+ &(system->box), control->vlist_cut,
+ nbr_data)) {
+ nbr_data->nbr = atom2;
+
+ ++num_far;
+ }
+ }
+ }
+ }
+
+ ++itr;
+ }
+
+ Set_End_Index( atom1, num_far, far_nbrs );
+ //fprintf(stderr, "i:%d, start: %d, end: %d - itr: %d\n",
+ // atom1,Start_Index(atom1,far_nbrs),End_Index(atom1,far_nbrs),
+ // itr);
+ }
+ }
+
+ fprintf (stderr, " TOTAL HOST NEIGHBORS : %d \n", num_far);
+
+ if( num_far > far_nbrs->num_intrs * DANGER_ZONE ) {
+ workspace->realloc.num_far = num_far;
+ if( num_far > far_nbrs->num_intrs ){
+ fprintf( stderr, "step%d-ran out of space on far_nbrs: top=%d, max=%d",
+ data->step, num_far, far_nbrs->num_intrs );
+ exit( INSUFFICIENT_SPACE );
+ }
+ }
+
+ t_elapsed = Get_Timing_Info( t_start );
+ data->timing.nbrs += t_elapsed;
+
+#if defined(DEBUG)
+ for( i = 0; i < system->N; ++i ) {
+ qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
+ Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
+ compare_far_nbrs );
+ }
+#endif
+
+#if defined(DEBUG_FOCUS)
+ //fprintf( stderr, "nbrs - ");
+ //fprintf( stderr, "nbrs done, num_far: %d\n", num_far );
+#endif
+
+#if defined(TEST_ENERGY)
+ //Print_Far_Neighbors( system, control, workspace, lists );
+#endif
+}
+
+
+int Estimate_NumNeighbors( reax_system *system, control_params *control,
+ static_storage *workspace, list **lists )
+{
+ int i, j, k, l, m, itr;
+ int x, y, z;
+ int atom1, atom2, max;
+ int num_far;
+ int *nbr_atoms;
+ ivec *nbrs;
+ rvec *nbrs_cp;
+ grid *g;
+ far_neighbor_data nbr_data;
+
+ int start = 0, finish = 0;
+
+ // fprintf( stderr, "\n\tentered nbrs - " );
+ g = &( system->g );
+ Bin_Atoms( system, workspace );
+ // fprintf( stderr, "atoms sorted - " );
+ num_far = 0;
+ g->max_cuda_nbrs = 0;
+
+ /* first pick up a cell in the grid */
+ for( i = 0; i < g->ncell[0]; i++ )
+ for( j = 0; j < g->ncell[1]; j++ )
+ for( k = 0; k < g->ncell[2]; k++ ) {
+ nbrs = &g->nbrs[index_grid_nbrs (i,j,k,0,g) ];
+ nbrs_cp = &g->nbrs_cp[index_grid_nbrs (i,j,k,0,g) ];
+ //fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
+
+ /* pick up an atom from the current cell */
+ for(l = 0; l < g->top[index_grid_3d (i,j,k,g) ]; ++l ){
+ atom1 = g->atoms[index_grid_atoms (i,j,k,l,g) ];
+ start = num_far;
+
+ itr = 0;
+ while( nbrs[itr][0] >= 0 ){
+ x = nbrs[itr][0];
+ y = nbrs[itr][1];
+ z = nbrs[itr][2];
+ //fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
+
+ if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
+ SQR(control->vlist_cut) ) {
+ nbr_atoms = &g->atoms[index_grid_atoms (x,y,z,0,g) ];
+ max = g->top[index_grid_3d (x,y,z,g) ];
+ //fprintf( stderr, "\t\tmax: %d\n", max );
+
+ /* pick up another atom from the neighbor cell -
+ we have to compare atom1 with its own periodic images as well,
+ that's why there is also equality in the if stmt below */
+ for( m = 0; m < max; ++m ) {
+ atom2 = nbr_atoms[m];
+ //if( nbrs[itr+1][0] >= 0 || atom1 > atom2 ) {
+ if( atom1 > atom2 ) {
+ if(Are_Far_Neighbors(system->atoms[atom1].x,
+ system->atoms[atom2].x,
+ &(system->box), control->vlist_cut,
+ &nbr_data))
+ ++num_far;
+ }
+ }
+ }
+
+ ++itr;
+ }
+
+ // finish note
+ finish = num_far;
+ if (g->max_cuda_nbrs <= (finish - start)){
+ g->max_cuda_nbrs = finish - start;
+ }
+ }
+ }
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "estimate nbrs done, num_far: %d\n", num_far );
+#endif
+
+ return num_far * SAFE_ZONE;
+}
+
+
+//Code not used anymore
+#if defined DONE
+
+void Choose_Neighbor_Finder( reax_system *system, control_params *control,
+ get_far_neighbors_function *Get_Far_Neighbors )
+{
+ if( control->periodic_boundaries )
+ {
+ if( system->box.box_norms[0] > 2.0 * control->vlist_cut &&
+ system->box.box_norms[1] > 2.0 * control->vlist_cut &&
+ system->box.box_norms[2] > 2.0 * control->vlist_cut )
+ (*Get_Far_Neighbors) = Get_Periodic_Far_Neighbors_Big_Box;
+ else (*Get_Far_Neighbors) = Get_Periodic_Far_Neighbors_Small_Box;
+ }
+ else
+ (*Get_Far_Neighbors) = Get_NonPeriodic_Far_Neighbors;
+}
+
+
+int compare_near_nbrs(const void *v1, const void *v2)
+{
+ return ((*(near_neighbor_data *)v1).nbr - (*(near_neighbor_data *)v2).nbr);
+}
+
+
+int compare_far_nbrs(const void *v1, const void *v2)
+{
+ return ((*(far_neighbor_data *)v1).nbr - (*(far_neighbor_data *)v2).nbr);
+}
+
+
+inline void Set_Far_Neighbor( far_neighbor_data *dest, int nbr, real d, real C,
+ rvec dvec, ivec rel_box/*, rvec ext_factor*/ )
+{
+ dest->nbr = nbr;
+ dest->d = d;
+ rvec_Scale( dest->dvec, C, dvec );
+ ivec_Copy( dest->rel_box, rel_box );
+ // rvec_Scale( dest->ext_factor, C, ext_factor );
+}
+
+
+inline void Set_Near_Neighbor(near_neighbor_data *dest, int nbr, real d, real C,
+ rvec dvec, ivec rel_box/*, rvec ext_factor*/)
+{
+ dest->nbr = nbr;
+ dest->d = d;
+ rvec_Scale( dest->dvec, C, dvec );
+ ivec_Scale( dest->rel_box, C, rel_box );
+ // rvec_Scale( dest->ext_factor, C, ext_factor );
+}
+
+
+/* In case bond restrictions are applied, this method checks if
+ atom1 and atom2 are allowed to bond with each other */
+inline int can_Bond( static_storage *workspace, int atom1, int atom2 )
+{
+ int i;
+
+ // fprintf( stderr, "can bond %6d %6d?\n", atom1, atom2 );
+
+ if( !workspace->restricted[ atom1 ] && !workspace->restricted[ atom2 ] )
+ return 1;
+
+ for( i = 0; i < workspace->restricted[ atom1 ]; ++i )
+ if( workspace->restricted_list[ atom1 ][i] == atom2 )
+ return 1;
+
+ for( i = 0; i < workspace->restricted[ atom2 ]; ++i )
+ if( workspace->restricted_list[ atom2 ][i] == atom1 )
+ return 1;
+
+ return 0;
+}
+
+
+/* check if atom2 is on atom1's near neighbor list */
+inline int is_Near_Neighbor( list *near_nbrs, int atom1, int atom2 )
+{
+ int i;
+
+ for( i=Start_Index(atom1,near_nbrs); i<End_Index(atom1,near_nbrs); ++i )
+ if( near_nbrs->select.near_nbr_list[i].nbr == atom2 )
+ {
+ // fprintf( stderr, "near neighbors %6d %6d\n", atom1, atom2 );
+ return 1;
+ }
+
+ return 0;
+}
+
+void Generate_Neighbor_Lists( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, k;
+ int x, y, z;
+ int *nbr_atoms;
+ int atom1, atom2, max;
+ int num_far;
+ int c, count;
+ int grid_top;
+ grid *g = &( system->g );
+ list *far_nbrs = (*lists) + FAR_NBRS;
+ //int hb_type1, hb_type2;
+ //list *hbonds = (*lists) + HBOND;
+ //int top_hbond1, top_hbond2;
+ get_far_neighbors_function Get_Far_Neighbors;
+ far_neighbor_data new_nbrs[125];
+#ifndef REORDER_ATOMS
+ int l, m;
+#endif
+
+ // fprintf( stderr, "\n\tentered nbrs - " );
+ if( control->ensemble == iNPT || control->ensemble == sNPT ||
+ control->ensemble == NPT )
+ Update_Grid( system );
+ // fprintf( stderr, "grid updated - " );
+
+ Bin_Atoms( system, out_control );
+ // fprintf( stderr, "atoms sorted - " );
+
+#ifdef REORDER_ATOMS
+ Cluster_Atoms( system, workspace );
+ // fprintf( stderr, "atoms clustered - " );
+#endif
+
+ Choose_Neighbor_Finder( system, control, &Get_Far_Neighbors );
+ // fprintf( stderr, "function chosen - " );
+
+ Reset_Neighbor_Lists( system, workspace, lists );
+ // fprintf( stderr, "lists cleared - " );
+
+ num_far = 0;
+ num_near = 0;
+ c = 0;
+
+ /* first pick up a cell in the grid */
+ for( i = 0; i < g->ncell[0]; i++ )
+ for( j = 0; j < g->ncell[1]; j++ )
+ for( k = 0; k < g->ncell[2]; k++ ) {
+ nbrs = g->nbrs[i][j][k];
+ nbrs_cp = g->nbrs_cp[i][j][k];
+
+ /* pick up an atom from the current cell */
+ //#ifdef REORDER_ATOMS
+ // for(atom1 = g->start[i][j][k]; atom1 < g->end[i][j][k]; atom1++)
+ //#else
+ for(l = 0; l < g->top[i][j][k]; ++l ){
+ atom1 = g->atoms[i][j][k][l];
+ Set_End_Index( atom1, num_far, far_nbrs );
+ // fprintf( stderr, "atom %d:\n", atom1 );
+
+ itr = 0;
+ while( nbrs[itr][0] > 0 ){
+ x = nbrs[itr][0];
+ y = nbrs[itr][1];
+ z = nbrs[itr][2];
+
+ // if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
+ // SQR(control->r_cut))
+ nbr_atoms = g->atoms[x][y][z];
+ max_atoms = g->top[x][y][z];
+
+ /* pick up another atom from the neighbor cell -
+ we have to compare atom1 with its own periodic images as well,
+ that's why there is also equality in the if stmt below */
+ //#ifdef REORDER_ATOMS
+ //for(atom2=g->start[x][y][z]; atom2<g->end[x][y][z]; atom2++)
+ //#else
+ for( m = 0, atom2=nbr_atoms[m]; m < max; ++m, atom2=nbr_atoms[m] )
+ if( atom1 >= atom2 ) {
+ //fprintf( stderr, "\tatom2 %d", atom2 );
+ //top_near1 = End_Index( atom1, near_nbrs );
+ //Set_Start_Index( atom1, num_far, far_nbrs );
+ //hb_type1=system->reaxprm.sbp[system->atoms[atom1].type].p_hbond;
+ Get_Far_Neighbors( system->atoms[atom1].x,
+ system->atoms[atom2].x,
+ &(system->box), control, new_nbrs, &count );
+ fprintf( stderr, "\t%d count:%d\n", atom2, count );
+
+ for( c = 0; c < count; ++c )
+ if(atom1 != atom2 || (atom1 == atom2 && new_nbrs[c].d>=0.1)){
+ Set_Far_Neighbor(&(far_nbrs->select.far_nbr_list[num_far]),
+ atom2, new_nbrs[c].d, 1.0,
+ new_nbrs[c].dvec, new_nbrs[c].rel_box );
+ ++num_far;
+
+ /*fprintf(stderr,"FARNBR:%6d%6d%8.3f[%8.3f%8.3f%8.3f]\n",
+ atom1, atom2, new_nbrs[c].d,
+ new_nbrs[c].dvec[0], new_nbrs[c].dvec[1],
+ new_nbrs[c].dvec[2] ); */
+
+
+ /* hydrogen bond lists */
+ /*if( control->hb_cut > 0.1 &&
+ new_nbrs[c].d <= control->hb_cut ) {
+ // fprintf( stderr, "%d %d\n", atom1, atom2 );
+ hb_type2=system->reaxprm.sbp[system->atoms[atom2].type].p_hbond;
+ if( hb_type1 == 1 && hb_type2 == 2 ) {
+ top_hbond1=End_Index(workspace->hbond_index[atom1],hbonds);
+ Set_Near_Neighbor(&(hbonds->select.hbond_list[top_hbond1]),
+ atom2, new_nbrs[c].d, 1.0, new_nbrs[c].dvec,
+ new_nbrs[c].rel_box );
+ Set_End_Index( workspace->hbond_index[atom1],
+ top_hbond1 + 1, hbonds );
+ }
+ else if( hb_type1 == 2 && hb_type2 == 1 ) {
+ top_hbond2 = End_Index( workspace->hbond_index[atom2], hbonds );
+ Set_Near_Neighbor(&(hbonds->select.hbond_list[top_hbond2]),
+ atom1, new_nbrs[c].d, -1.0, new_nbrs[c].dvec,
+ new_nbrs[c].rel_box );
+ Set_End_Index( workspace->hbond_index[atom2],
+ top_hbond2 + 1, hbonds );
+ }*/
+ }
+ }
+ }
+
+ Set_End_Index( atom1, top_far1, far_nbrs );
+ }
+ }
+
+ fprintf( stderr, "nbrs done-" );
+
+ /* apply restrictions on near neighbors only */
+ if( (data->step - data->prev_steps) < control->restrict_bonds ) {
+ for( atom1 = 0; atom1 < system->N; ++atom1 )
+ if( workspace->restricted[ atom1 ] ) {
+ // fprintf( stderr, "atom1: %d\n", atom1 );
+
+ top_near1 = End_Index( atom1, near_nbrs );
+
+ for( j = 0; j < workspace->restricted[ atom1 ]; ++j )
+ if(!is_Near_Neighbor(near_nbrs, atom1,
+ atom2 = workspace->restricted_list[atom1][j])) {
+ fprintf( stderr, "%3d-%3d: added bond by applying restrictions!\n",
+ atom1, atom2 );
+
+ top_near2 = End_Index( atom2, near_nbrs );
+
+ /* we just would like to get the nearest image, so a call to
+ Get_Periodic_Far_Neighbors_Big_Box is good enough. */
+ Get_Periodic_Far_Neighbors_Big_Box( system->atoms[ atom1 ].x,
+ system->atoms[ atom2 ].x,
+ &(system->box), control,
+ new_nbrs, &count );
+
+ Set_Near_Neighbor( &(near_nbrs->select.near_nbr_list[ top_near1 ]),
+ atom2, new_nbrs[c].d, 1.0,
+ new_nbrs[c].dvec, new_nbrs[c].rel_box );
+ ++top_near1;
+
+ Set_Near_Neighbor( &(near_nbrs->select.near_nbr_list[ top_near2 ]),
+ atom1, new_nbrs[c].d, -1.0,
+ new_nbrs[c].dvec, new_nbrs[c].rel_box );
+ Set_End_Index( atom2, top_near2+1, near_nbrs );
+ }
+
+ Set_End_Index( atom1, top_near1, near_nbrs );
+ }
+ }
+ // fprintf( stderr, "restrictions applied-" );
+
+
+ /* verify nbrlists, count num_intrs, sort nearnbrs */
+ near_nbrs->num_intrs = 0;
+ far_nbrs->num_intrs = 0;
+ for( i = 0; i < system->N-1; ++i ) {
+ if( End_Index(i, near_nbrs) > Start_Index(i+1, near_nbrs) ) {
+ fprintf( stderr,
+ "step%3d: nearnbr list of atom%d is overwritten by atom%d\n",
+ data->step, i+1, i );
+ exit( 1 );
+ }
+
+ near_nbrs->num_intrs += Num_Entries(i, near_nbrs);
+
+ if( End_Index(i, far_nbrs) > Start_Index(i+1, far_nbrs) ) {
+ fprintf( stderr,
+ "step%3d: farnbr list of atom%d is overwritten by atom%d\n",
+ data->step, i+1, i );
+ exit( 1 );
+ }
+
+ far_nbrs->num_intrs += Num_Entries(i, far_nbrs);
+ }
+
+ for( i = 0; i < system->N; ++i ) {
+ qsort( &(near_nbrs->select.near_nbr_list[ Start_Index(i, near_nbrs) ]),
+ Num_Entries(i, near_nbrs), sizeof(near_neighbor_data),
+ compare_near_nbrs );
+ }
+ // fprintf( stderr, "near nbrs sorted\n" );
+
+#ifdef TEST_ENERGY
+ /* for( i = 0; i < system->N; ++i ) {
+ qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
+ Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
+ compare_far_nbrs );
+ } */
+
+ fprintf( stderr, "Near neighbors/atom: %d (compare to 150)\n",
+ num_near / system->N );
+ fprintf( stderr, "Far neighbors per atom: %d (compare to %d)\n",
+ num_far / system->N, control->max_far_nbrs );
+#endif
+
+ //fprintf( stderr, "step%d: num of nearnbrs = %6d num of farnbrs: %6d\n",
+ // data->step, num_near, num_far );
+
+ //fprintf( stderr, "\talloc nearnbrs = %6d alloc farnbrs: %6d\n",
+ // system->N * near_nbrs->intrs_per_unit,
+ // system->N * far_nbrs->intrs_per_unit );
+}
+
+
+void Generate_Neighbor_Lists( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, k, l, m, itr;
+ int x, y, z;
+ int atom1, atom2, max;
+ int num_far, c, count;
+ int *nbr_atoms;
+ ivec *nbrs;
+ rvec *nbrs_cp;
+ grid *g;
+ list *far_nbrs;
+ get_far_neighbors_function Get_Far_Neighbors;
+ far_neighbor_data new_nbrs[125];
+
+ g = &( system->g );
+ far_nbrs = (*lists) + FAR_NBRS;
+
+ // fprintf( stderr, "\n\tentered nbrs - " );
+ if( control->ensemble == iNPT ||
+ control->ensemble == sNPT ||
+ control->ensemble == NPT )
+ Update_Grid( system );
+ // fprintf( stderr, "grid updated - " );
+
+ Bin_Atoms( system, out_control );
+ // fprintf( stderr, "atoms sorted - " );
+ Choose_Neighbor_Finder( system, control, &Get_Far_Neighbors );
+ // fprintf( stderr, "function chosen - " );
+ Reset_Neighbor_Lists( system, workspace, lists );
+ // fprintf( stderr, "lists cleared - " );
+
+ num_far = 0;
+ c = 0;
+
+ /* first pick up a cell in the grid */
+ for( i = 0; i < g->ncell[0]; i++ )
+ for( j = 0; j < g->ncell[1]; j++ )
+ for( k = 0; k < g->ncell[2]; k++ ) {
+ nbrs = g->nbrs[i][j][k];
+ nbrs_cp = g->nbrs_cp[i][j][k];
+ fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
+
+ /* pick up an atom from the current cell */
+ for(l = 0; l < g->top[i][j][k]; ++l ){
+ atom1 = g->atoms[i][j][k][l];
+ Set_Start_Index( atom1, num_far, far_nbrs );
+ fprintf( stderr, "\tatom %d\n", atom1 );
+
+ itr = 0;
+ while( nbrs[itr][0] > 0 ){
+ x = nbrs[itr][0];
+ y = nbrs[itr][1];
+ z = nbrs[itr][2];
+ fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
+
+ // if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
+ // SQR(control->r_cut))
+ nbr_atoms = g->atoms[x][y][z];
+ max = g->top[x][y][z];
+ fprintf( stderr, "\t\tmax: %d\n", max );
+
+
+ /* pick up another atom from the neighbor cell -
+ we have to compare atom1 with its own periodic images as well,
+ that's why there is also equality in the if stmt below */
+ for( m = 0, atom2=nbr_atoms[m]; m < max; ++m, atom2=nbr_atoms[m] )
+ if( atom1 >= atom2 ) {
+ Get_Far_Neighbors( system->atoms[atom1].x,
+ system->atoms[atom2].x,
+ &(system->box), control, new_nbrs, &count );
+ fprintf( stderr, "\t\t\t%d count:%d\n", atom2, count );
+
+ for( c = 0; c < count; ++c )
+ if(atom1 != atom2 || (atom1 == atom2 && new_nbrs[c].d>=0.1)){
+ Set_Far_Neighbor(&(far_nbrs->select.far_nbr_list[num_far]),
+ atom2, new_nbrs[c].d, 1.0,
+ new_nbrs[c].dvec, new_nbrs[c].rel_box );
+ ++num_far;
+
+ /*fprintf(stderr,"FARNBR:%6d%6d%8.3f[%8.3f%8.3f%8.3f]\n",
+ atom1, atom2, new_nbrs[c].d,
+ new_nbrs[c].dvec[0], new_nbrs[c].dvec[1],
+ new_nbrs[c].dvec[2] ); */
+ }
+ }
+
+ ++itr;
+ }
+
+ Set_End_Index( atom1, num_far, far_nbrs );
+ }
+ }
+
+ far_nbrs->num_intrs = num_far;
+ fprintf( stderr, "nbrs done, num_far: %d\n", num_far );
+
+#if defined(DEBUG)
+ for( i = 0; i < system->N; ++i ) {
+ qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
+ Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
+ compare_far_nbrs );
+ }
+
+ fprintf( stderr, "step%d: num of farnbrs=%6d\n", data->step, num_far );
+ fprintf( stderr, "\tallocated farnbrs: %6d\n",
+ system->N * far_nbrs->intrs_per_unit );
+#endif
+}
+
+
+#endif
diff --git a/PuReMD-GPU/src/neighbors.cu b/PuReMD-GPU/src/neighbors.cu
deleted file mode 100644
index 90779538353a05eca7bb04215ebd33b3ffd81e35..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/neighbors.cu
+++ /dev/null
@@ -1,1413 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "neighbors.h"
-#include "box.h"
-#include "grid.h"
-#include "list.h"
-#include "reset_utils.h"
-#include "system_props.h"
-#include "vector.h"
-#include "index_utils.h"
-#include "cuda_utils.h"
-
-extern inline DEVICE int index_grid (int blocksize)
-{
- return blockIdx.x * gridDim.y * gridDim.z * blocksize +
- blockIdx.y * gridDim.z * blocksize +
- blockIdx.z * blocksize ;
-}
-
-extern inline HOST_DEVICE int index_grid_debug (int x, int y, int z, int blocksize)
-{
- return x * 8 * 8 * blocksize +
- y * 8 * blocksize +
- z * blocksize ;
-}
-
-inline HOST_DEVICE real DistSqr_to_CP( rvec cp, rvec x )
-{
- int i;
- real d_sqr = 0;
-
- for( i = 0; i < 3; ++i )
- if( cp[i] > NEG_INF )
- d_sqr += SQR( cp[i] - x[i] );
-
- return d_sqr;
-}
-
-HOST_DEVICE int Are_Far_Neighbors( rvec x1, rvec x2, simulation_box *box,
- real cutoff, far_neighbor_data *data )
-{
- real norm_sqr, d, tmp;
- int i;
-
- norm_sqr = 0;
-
- for( i = 0; i < 3; i++ ) {
- d = x2[i] - x1[i];
- tmp = SQR(d);
-
- if( tmp >= SQR( box->box_norms[i] / 2.0 ) ) {
- if( x2[i] > x1[i] ) {
- d -= box->box_norms[i];
- data->rel_box[i] = -1;
- }
- else {
- d += box->box_norms[i];
- data->rel_box[i] = +1;
- }
-
- data->dvec[i] = d;
- norm_sqr += SQR(d);
- }
- else {
- data->dvec[i] = d;
- norm_sqr += tmp;
- data->rel_box[i] = 0;
- }
- }
-
- if( norm_sqr <= SQR(cutoff) ){
- data->d = sqrt(norm_sqr);
- return 1;
- }
-
- return 0;
-}
-
-void Generate_Neighbor_Lists( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, k, l, m, itr;
- int x, y, z;
- int atom1, atom2, max;
- int num_far;
- int *nbr_atoms;
- ivec *nbrs;
- rvec *nbrs_cp;
- grid *g;
- list *far_nbrs;
- far_neighbor_data *nbr_data;
- real t_start, t_elapsed;
-
- // fprintf( stderr, "\n\tentered nbrs - " );
- g = &( system->g );
- far_nbrs = (*lists) + FAR_NBRS;
- Bin_Atoms( system, workspace );
-
- t_start = Get_Time( );
-
- // fprintf( stderr, "atoms sorted - " );
- num_far = 0;
-
- /* first pick up a cell in the grid */
- for( i = 0; i < g->ncell[0]; i++ )
- for( j = 0; j < g->ncell[1]; j++ )
- for( k = 0; k < g->ncell[2]; k++ ) {
- nbrs = &g->nbrs[ index_grid_nbrs (i,j,k,0,g) ];
- nbrs_cp = &g->nbrs_cp[ index_grid_nbrs (i,j,k,0,g) ];
- //fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
-
- /* pick up an atom from the current cell */
- for(l = 0; l < g->top[ index_grid_3d (i,j,k,g) ]; ++l ){
- atom1 = g->atoms[ index_grid_atoms (i,j,k,l,g) ];
- Set_Start_Index( atom1, num_far, far_nbrs );
- //fprintf( stderr, "\tatom %d\n", atom1 );
-
- itr = 0;
- while( nbrs[itr][0] >= 0 ){
- x = nbrs[itr][0];
- y = nbrs[itr][1];
- z = nbrs[itr][2];
- //fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
-
- if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
- SQR(control->vlist_cut) ) {
- nbr_atoms = &g->atoms[ index_grid_atoms (x,y,z,0,g) ];
- max = g->top[ index_grid_3d (x,y,z,g) ];
- //fprintf( stderr, "\t\tmax: %d\n", max );
-
- /* pick up another atom from the neighbor cell */
- for( m = 0; m < max; ++m ) {
- atom2 = nbr_atoms[m];
- if( atom1 > atom2 ) {
- nbr_data = &(far_nbrs->select.far_nbr_list[num_far]);
- if(Are_Far_Neighbors(system->atoms[atom1].x,
- system->atoms[atom2].x,
- &(system->box), control->vlist_cut,
- nbr_data)) {
- nbr_data->nbr = atom2;
-
- ++num_far;
- }
- }
- }
- }
-
- ++itr;
- }
-
- Set_End_Index( atom1, num_far, far_nbrs );
- //fprintf(stderr, "i:%d, start: %d, end: %d - itr: %d\n",
- // atom1,Start_Index(atom1,far_nbrs),End_Index(atom1,far_nbrs),
- // itr);
- }
- }
-
- fprintf (stderr, " TOTAL HOST NEIGHBORS : %d \n", num_far);
-
- if( num_far > far_nbrs->num_intrs * DANGER_ZONE ) {
- workspace->realloc.num_far = num_far;
- if( num_far > far_nbrs->num_intrs ){
- fprintf( stderr, "step%d-ran out of space on far_nbrs: top=%d, max=%d",
- data->step, num_far, far_nbrs->num_intrs );
- exit( INSUFFICIENT_SPACE );
- }
- }
-
- t_elapsed = Get_Timing_Info( t_start );
- data->timing.nbrs += t_elapsed;
-
-#if defined(DEBUG)
- for( i = 0; i < system->N; ++i ) {
- qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
- Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
- compare_far_nbrs );
- }
-#endif
-#if defined(DEBUG_FOCUS)
- //fprintf( stderr, "nbrs - ");
- //fprintf( stderr, "nbrs done, num_far: %d\n", num_far );
-#endif
-#if defined(TEST_ENERGY)
- //Print_Far_Neighbors( system, control, workspace, lists );
-#endif
-}
-
-
-int Estimate_NumNeighbors( reax_system *system, control_params *control,
- static_storage *workspace, list **lists )
-{
- int i, j, k, l, m, itr;
- int x, y, z;
- int atom1, atom2, max;
- int num_far;
- int *nbr_atoms;
- ivec *nbrs;
- rvec *nbrs_cp;
- grid *g;
- far_neighbor_data nbr_data;
-
- int start = 0, finish = 0;
-
- // fprintf( stderr, "\n\tentered nbrs - " );
- g = &( system->g );
- Bin_Atoms( system, workspace );
- // fprintf( stderr, "atoms sorted - " );
- num_far = 0;
- g->max_cuda_nbrs = 0;
-
- /* first pick up a cell in the grid */
- for( i = 0; i < g->ncell[0]; i++ )
- for( j = 0; j < g->ncell[1]; j++ )
- for( k = 0; k < g->ncell[2]; k++ ) {
- nbrs = &g->nbrs[index_grid_nbrs (i,j,k,0,g) ];
- nbrs_cp = &g->nbrs_cp[index_grid_nbrs (i,j,k,0,g) ];
- //fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
-
- /* pick up an atom from the current cell */
- for(l = 0; l < g->top[index_grid_3d (i,j,k,g) ]; ++l ){
- atom1 = g->atoms[index_grid_atoms (i,j,k,l,g) ];
- start = num_far;
-
- itr = 0;
- while( nbrs[itr][0] >= 0 ){
- x = nbrs[itr][0];
- y = nbrs[itr][1];
- z = nbrs[itr][2];
- //fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
-
- if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
- SQR(control->vlist_cut) ) {
- nbr_atoms = &g->atoms[index_grid_atoms (x,y,z,0,g) ];
- max = g->top[index_grid_3d (x,y,z,g) ];
- //fprintf( stderr, "\t\tmax: %d\n", max );
-
- /* pick up another atom from the neighbor cell -
- we have to compare atom1 with its own periodic images as well,
- that's why there is also equality in the if stmt below */
- for( m = 0; m < max; ++m ) {
- atom2 = nbr_atoms[m];
- //if( nbrs[itr+1][0] >= 0 || atom1 > atom2 ) {
- if( atom1 > atom2 ) {
- if(Are_Far_Neighbors(system->atoms[atom1].x,
- system->atoms[atom2].x,
- &(system->box), control->vlist_cut,
- &nbr_data))
- ++num_far;
- }
- }
- }
-
- ++itr;
- }
-
- // finish note
- finish = num_far;
- if (g->max_cuda_nbrs <= (finish - start)){
- g->max_cuda_nbrs = finish - start;
- }
- }
- }
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "estimate nbrs done, num_far: %d\n", num_far );
-#endif
- return num_far * SAFE_ZONE;
- }
-
- GLOBAL void Estimate_NumNeighbors ( reax_atom *sys_atoms,
- grid g,
- simulation_box *box,
- control_params *control,
- int *indices)
- {
- int *atoms = g.atoms;
- int *top = g.top;
- ivec *nbrs = g.nbrs;
- rvec *nbrs_cp = g.nbrs_cp;
-
- int *nbr_atoms;
- int atom1, atom2, l, iter, max, m, num_far;
- far_neighbor_data nbr_data;
- int x, y, z, i;
-
- if (threadIdx.x >= *(top + index_grid(1))){
- return;
- }
-
- nbrs = nbrs + index_grid (g.max_nbrs);
- nbrs_cp = nbrs_cp + index_grid (g.max_nbrs);
- atom1 = atoms [ index_grid (g.max_atoms) + threadIdx.x];
-
- num_far = 0;
- iter = 0;
-
- while (nbrs[iter][0] >= 0) {
- x = nbrs[iter][0];
- y = nbrs[iter][1];
- z = nbrs[iter][2];
-
- //condition check for cutoff here
- if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
- SQR (control->vlist_cut))
- {
- nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
- max = top [index_grid_3d(x, y, z, &g)];
- for (m = 0; m < max; m++) {
- atom2 = nbr_atoms[m];
-
- //CHANGE ORIGINAL
- /*
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
- control->vlist_cut, &nbr_data)){
- ++num_far;
- }
- }
- */
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
- control->vlist_cut, &nbr_data)){
- ++num_far;
- }
- }
- else if (atom1 < atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
- control->vlist_cut, &nbr_data)){
- ++num_far;
- }
- }
- //CHANGE ORIGINAL
- }
- }
- ++iter;
- }
-
- //indices[ atom1 ] = num_far;// * SAFE_ZONE;
- indices[ atom1 ] = num_far * SAFE_ZONE;
- }
-
- /*One thread per atom Implementation */
- GLOBAL void New_Estimate_NumNeighbors ( reax_atom *sys_atoms,
- grid g,
- simulation_box *box,
- control_params* control,
- int N, int *indices)
- {
- int *atoms = g.atoms;
- int *top = g.top;
- ivec *nbrs = g.nbrs;
- rvec *nbrs_cp = g.nbrs_cp;
-
- int *nbr_atoms;
- int atom1, atom2, iter, max, m, num_far;
- int x, y, z, i;
- int atom_x, atom_y, atom_z;
- far_neighbor_data temp;
- rvec atom1_x;
-
- int index = blockIdx.x * blockDim.x + threadIdx.x;
- if (index > N) return;
-
- atom_x = (int)(sys_atoms[index].x[0] * g.inv_len[0]);
- atom_y = (int)(sys_atoms[index].x[1] * g.inv_len[1]);
- atom_z = (int)(sys_atoms[index].x[2] * g.inv_len[2]);
-
-#ifdef __BNVT_FIX__
- if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
- if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
- if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
-#endif
-
- nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
- nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
- atom1 = index;
-
- rvec_Copy (atom1_x, sys_atoms [atom1].x );
-
- num_far = 0;
- iter = 0;
-
- while (nbrs[iter][0] >= 0) {
- x = nbrs[iter][0];
- y = nbrs[iter][1];
- z = nbrs[iter][2];
-
- if (DistSqr_to_CP (nbrs_cp[iter], atom1_x) <=
- SQR (control->vlist_cut))
- {
- nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
- max = top [index_grid_3d(x, y, z, &g)];
-
- for (m = 0; m < max; m++)
- {
- atom2 = nbr_atoms[m];
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (atom1_x, sys_atoms[atom2].x, box,
- control->vlist_cut, &temp)){
- num_far++;
- }
- }
- else if (atom1 < atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom2].x, atom1_x, box,
- control->vlist_cut, &temp)){
- num_far ++;
- }
- }
- }
- }
- ++iter;
- }
- indices [atom1] = num_far * SAFE_ZONE;
- }
-
-
-
- /*One thread per entry in the gcell implementation */
- GLOBAL void Generate_Neighbor_Lists ( reax_atom *sys_atoms,
- grid g,
- simulation_box *box,
- control_params* control,
- list far_nbrs)
- {
- int *atoms = g.atoms;
- int *top = g.top;
- ivec *nbrs = g.nbrs;
- rvec *nbrs_cp = g.nbrs_cp;
-
- int *nbr_atoms;
- int atom1, atom2, l, iter, max, m, num_far;
- int x, y, z, i;
- far_neighbor_data *nbr_data;
- far_neighbor_data temp;
-
- if (threadIdx.x >= *(top + index_grid(1))){
- return;
- }
-
- nbrs = nbrs + index_grid (g.max_nbrs);
- nbrs_cp = nbrs_cp + index_grid (g.max_nbrs);
- atom1 = atoms [ index_grid (g.max_atoms) + threadIdx.x];
-
- num_far = Start_Index (atom1, &far_nbrs);
- //Set_Start_Index (atom1, 0, &far_nbrs);
- //num_far = 0;
- iter = 0;
-
- while (nbrs[iter][0] >= 0) {
- x = nbrs[iter][0];
- y = nbrs[iter][1];
- z = nbrs[iter][2];
-
- //condition check for cutoff here
- if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
- SQR (control->vlist_cut))
- {
- nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
- max = top [index_grid_3d(x, y, z, &g)];
-
- for (m = 0; m < max; m++) {
- atom2 = nbr_atoms[m];
-
- //nbr_data = & ( far_nbrs.select.far_nbr_list[atom1 * g.max_cuda_nbrs + num_far] );
-
- //CHANGE ORIGINAL
- /*
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
- control->vlist_cut, &temp)){
-
- nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
- ++num_far;
- }
- }
- */
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
- control->vlist_cut, &temp)){
- nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
- ++num_far;
- }
- }
- else if (atom1 < atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
- control->vlist_cut, &temp)){
- nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
- ++num_far;
- }
- }
- //CHANGE ORIGINAL
- }
- }
- ++iter;
- }
-
- //end the far_neighbor list here
- Set_End_Index (atom1, num_far, &far_nbrs);
- }
-
-
- /*One thread per atom Implementation */
- GLOBAL void New_Generate_Neighbor_Lists ( reax_atom *sys_atoms,
- grid g,
- simulation_box *box,
- control_params* control,
- list far_nbrs, int N)
- {
- int *atoms = g.atoms;
- int *top = g.top;
- ivec *nbrs = g.nbrs;
- rvec *nbrs_cp = g.nbrs_cp;
-
- int *nbr_atoms;
- int atom1, atom2, l, iter, max, m, num_far;
- int x, y, z, i;
- far_neighbor_data *nbr_data, *my_start;
- far_neighbor_data temp;
- int atom_x, atom_y, atom_z;
- rvec atom1_x;
-
- int index = blockIdx.x * blockDim.x + threadIdx.x;
- if (index > N) return;
-
- atom_x = (int)(sys_atoms[index].x[0] * g.inv_len[0]);
- atom_y = (int)(sys_atoms[index].x[1] * g.inv_len[1]);
- atom_z = (int)(sys_atoms[index].x[2] * g.inv_len[2]);
-
-#ifdef __BNVT_FIX__
- if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
- if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
- if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
-#endif
-
- nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
- nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
- atom1 = index;
-
- rvec_Copy (atom1_x, sys_atoms [atom1].x );
-
- num_far = Start_Index (atom1, &far_nbrs);
- my_start = & (far_nbrs.select.far_nbr_list [num_far] );
-
- //Set_Start_Index (atom1, 0, &far_nbrs);
- //num_far = 0;
- iter = 0;
-
- while (nbrs[iter][0] >= 0) {
- x = nbrs[iter][0];
- y = nbrs[iter][1];
- z = nbrs[iter][2];
-
- //condition check for cutoff here
- //if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms[atom1].x) <=
- if (DistSqr_to_CP (nbrs_cp[iter], atom1_x) <=
- SQR (control->vlist_cut))
- {
- nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
- max = top [index_grid_3d(x, y, z, &g)];
-
- for (m = 0; m < max; m++)
- {
- atom2 = nbr_atoms[m];
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (atom1_x, sys_atoms[atom2].x, box,
- control->vlist_cut, &temp)){
- //nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
- nbr_data = my_start;
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
- num_far++;
- my_start ++;
- }
- }
- else if (atom1 < atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom2].x, atom1_x, box,
- control->vlist_cut, &temp)){
- //nbr_data = & ( far_nbrs.select.far_nbr_list[num_far] );
- nbr_data = my_start;
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
- num_far ++;
- my_start ++;
- }
- }
- //CHANGE ORIGINAL
- }
- }
- ++iter;
- }
-
- //end the far_neighbor list here
- Set_End_Index (atom1, num_far, &far_nbrs);
- }
-
- /*Multiple threads per atom Implementation */
- GLOBAL void Test_Generate_Neighbor_Lists ( reax_atom *sys_atoms,
- grid g,
- simulation_box *box,
- control_params* control,
- list far_nbrs, int N )
- {
-
- extern __shared__ int __nbr[];
- extern __shared__ int __sofar [];
- bool nbrgen;
-
- int __THREADS_PER_ATOM__ = NBRS_THREADS_PER_ATOM;
-
- int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
- int warp_id = thread_id / __THREADS_PER_ATOM__;
- int lane_id = thread_id & (__THREADS_PER_ATOM__ -1);
- int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
-
- if (warp_id >= N ) return;
-
- int *tnbr = __nbr;
- //int *nbrssofar = __nbr + __THREADS_PER_ATOM__;
- int *nbrssofar = __nbr + blockDim.x;
-
- int *atoms = g.atoms;
- int *top = g.top;
- ivec *nbrs = g.nbrs;
- rvec *nbrs_cp = g.nbrs_cp;
-
- int *nbr_atoms;
- int atom1, atom2, l, iter, max, m, num_far;
- int leader = -10;
- int x, y, z, i;
- far_neighbor_data *nbr_data, *my_start;
- far_neighbor_data temp;
- int atom_x, atom_y, atom_z;
-
-
- atom1 = warp_id;
- atom_x = (int)(sys_atoms[atom1].x[0] * g.inv_len[0]);
- atom_y = (int)(sys_atoms[atom1].x[1] * g.inv_len[1]);
- atom_z = (int)(sys_atoms[atom1].x[2] * g.inv_len[2]);
-
-#ifdef __BNVT_FIX__
- if (atom_x >= g.ncell[0]) atom_x = g.ncell[0]-1;
- if (atom_y >= g.ncell[1]) atom_y = g.ncell[1]-1;
- if (atom_z >= g.ncell[2]) atom_z = g.ncell[2]-1;
-#endif
-
- nbrs = nbrs + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
- nbrs_cp = nbrs_cp + index_grid_nbrs (atom_x, atom_y, atom_z, 0, &g);
-
- num_far = Start_Index (atom1, &far_nbrs);
- my_start = & (far_nbrs.select.far_nbr_list [num_far] );
-
- iter = 0;
- tnbr[threadIdx.x] = 0;
-
- if (lane_id == 0) {
- //nbrssofar [threadIdx.x /__THREADS_PER_ATOM__] = 0;
- nbrssofar [my_bucket] = 0;
- }
-
- __syncthreads ();
-
- while ((nbrs[iter][0] >= 0)) {
- x = nbrs[iter][0];
- y = nbrs[iter][1];
- z = nbrs[iter][2];
-
- tnbr[threadIdx.x] = 0;
- nbrgen = false;
-
- if (DistSqr_to_CP (nbrs_cp[iter], sys_atoms [atom1].x) <=
- SQR (control->vlist_cut))
- {
- nbr_atoms = &(atoms [index_grid_atoms (x, y, z, 0, &g) ]);
- max = top [index_grid_3d(x, y, z, &g)];
-
- tnbr[threadIdx.x] = 0;
- nbrgen = false;
- m = lane_id ; //0-31
- int loopcount = max / __THREADS_PER_ATOM__ + ((max % __THREADS_PER_ATOM__) == 0 ? 0 : 1);
- int iterations = 0;
- //while (m < max)
- while (iterations < loopcount)
- {
- tnbr [threadIdx.x] = 0;
- nbrgen = false;
-
- if (m < max) {
- atom2 = nbr_atoms[m];
- if (atom1 > atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom1].x, sys_atoms[atom2].x, box,
- control->vlist_cut, &temp))
- {
- tnbr [threadIdx.x] = 1;
- nbrgen = true;
- }
- }
- else if (atom1 < atom2) {
- if (Are_Far_Neighbors (sys_atoms[atom2].x, sys_atoms[atom1].x, box,
- control->vlist_cut, &temp)){
- tnbr [threadIdx.x] = 1;
- nbrgen = true;
- }
- }
- }
-
- if (nbrgen)
- {
- //do leader selection here
- leader = -1;
- //for (l = threadIdx.x / __THREADS_PER_ATOM__; l < threadIdx.x / __THREADS_PER_ATOM__ + __THREADS_PER_ATOM__; l++)
- for (l = my_bucket *__THREADS_PER_ATOM__; l < (my_bucket)*__THREADS_PER_ATOM__ + __THREADS_PER_ATOM__; l++)
- if (tnbr[l]){
- leader = l;
- break;
- }
-
- //do the reduction;
- if (threadIdx.x == leader)
- for (l = 1; l < __THREADS_PER_ATOM__; l++)
- //tnbr [(threadIdx.x / __THREADS_PER_ATOM__) * __THREADS_PER_ATOM__ + l] += tnbr [(threadIdx.x / __THREADS_PER_ATOM__) * __THREADS_PER_ATOM__ + (l-1)];
- tnbr [my_bucket * __THREADS_PER_ATOM__ + l] += tnbr [my_bucket * __THREADS_PER_ATOM__ + (l-1)];
- }
-
- //__syncthreads ();
- //atomicAdd ( &warp_sync [threadIdx.x / __THREADS_PER_ATOM__ ], 1);
- //while ( warp_sync [threadIdx.x / __THREADS_PER_ATOM__ ] < __THREADS_PER_ATOM__ ) ;
-
- if (nbrgen)
- {
- //got the indices
- //nbr_data = my_start + nbrssofar[threadIdx.x / __THREADS_PER_ATOM__] + tnbr [threadIdx.x] - 1;
- nbr_data = my_start + nbrssofar[my_bucket] + tnbr [threadIdx.x] - 1;
- nbr_data->nbr = atom2;
- nbr_data->rel_box[0] = temp.rel_box[0];
- nbr_data->rel_box[1] = temp.rel_box[1];
- nbr_data->rel_box[2] = temp.rel_box[2];
-
- nbr_data->d = temp.d;
- nbr_data->dvec[0] = temp.dvec[0];
- nbr_data->dvec[1] = temp.dvec[1];
- nbr_data->dvec[2] = temp.dvec[2];
-
- if (threadIdx.x == leader)
- //nbrssofar[threadIdx.x / __THREADS_PER_ATOM__] += tnbr[(threadIdx.x / __THREADS_PER_ATOM__)*__THREADS_PER_ATOM__ + (__THREADS_PER_ATOM__ - 1)];
- nbrssofar[my_bucket] += tnbr[my_bucket *__THREADS_PER_ATOM__ + (__THREADS_PER_ATOM__ - 1)];
- }
-
- m += __THREADS_PER_ATOM__;
- iterations ++;
-
- //cleanup
- nbrgen = false;
- tnbr [threadIdx.x] = 0;
- }
- }
- ++iter;
- }
-
- __syncthreads ();
-
- //end the far_neighbor list here
- if (lane_id == 0)
- Set_End_Index (atom1, num_far + nbrssofar[my_bucket], &far_nbrs);
- //Set_End_Index (atom1, num_far + tnbr[63], &far_nbrs);
- }
-
- void Cuda_Generate_Neighbor_Lists (reax_system *system, static_storage *workspace, control_params *control, bool estimate)
- {
- real t_start, t_elapsed;
- real t_1, t_2;
-
- list *far_nbrs = dev_lists + FAR_NBRS;
-
- int *d_indices = (int *) scratch;
- int *nbrs_start, *nbrs_end;
- int i, max_nbrs = 0;
- int nbs;
-
- t_start = Get_Time ();
-
- Cuda_Bin_Atoms (system, workspace);
- Cuda_Bin_Atoms_Sync ( system );
-
- if (dev_workspace->realloc.estimate_nbrs > -1) {
-
- /*reset the re-neighbor condition */
- dev_workspace->realloc.estimate_nbrs = -1;
-
- //#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Recomputing the neighbors estimate.... \n");
- //#endif
- cuda_memset (d_indices, 0, INT_SIZE * system->N, RES_SCRATCH );
- /*
- dim3 blockspergrid (system->g.ncell[0], system->g.ncell[1], system->g.ncell[2]);
- dim3 threadsperblock (system->g.max_atoms);
-
- Estimate_NumNeighbors <<<blockspergrid, threadsperblock >>>
- (system->d_atoms, system->d_g, system->d_box,
- (control_params *)control->d_control, d_indices);
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
- nbs = (system->N / NBRS_BLOCK_SIZE) + (((system->N) % NBRS_BLOCK_SIZE) == 0 ? 0 : 1);
- New_Estimate_NumNeighbors <<<nbs, NBRS_BLOCK_SIZE>>>
- ( system->d_atoms, system->d_g,
- system->d_box, (control_params *)control->d_control,
- system->N, d_indices);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
-
- int *nbrs_indices = NULL;
- nbrs_indices = (int *) malloc( INT_SIZE * (system->N+1) );
- if (nbrs_indices == NULL)
- {
- fprintf (stderr, "Malloc failed for nbrs indices .... \n");
- exit (1);
- }
- memset (nbrs_indices , 0, INT_SIZE * (system->N+1) );
-
- copy_host_device (nbrs_indices+1, d_indices, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 1; i <= system->N; i++)
- nbrs_indices [i] += nbrs_indices [i-1];
-
- copy_host_device (nbrs_indices, (far_nbrs->index), INT_SIZE * (system->N), cudaMemcpyHostToDevice, __LINE__ );
- copy_host_device (nbrs_indices, (far_nbrs->end_index), INT_SIZE * (system->N), cudaMemcpyHostToDevice, __LINE__ );
-
- free (nbrs_indices);
- }
-
- /*
- One thread per atom Implementation
- Generate_Neighbor_Lists <<<blockspergrid, threadsperblock >>>
- (system->d_atoms, system->d_g, system->d_box,
- (control_params *)control->d_control, *far_nbrs);
- */
- nbs = (system->N * NBRS_THREADS_PER_ATOM/ NBRS_BLOCK_SIZE) +
- (((system->N *NBRS_THREADS_PER_ATOM) % NBRS_BLOCK_SIZE) == 0 ? 0 : 1);
-
- /* Multiple threads per atom Implementation */
- Test_Generate_Neighbor_Lists <<<nbs, NBRS_BLOCK_SIZE,
- INT_SIZE * (NBRS_BLOCK_SIZE+ NBRS_BLOCK_SIZE/NBRS_THREADS_PER_ATOM) >>>
- (system->d_atoms, system->d_g, system->d_box,
- (control_params *)control->d_control, *far_nbrs, system->N );
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- t_elapsed = Get_Timing_Info (t_start);
- d_timing.nbrs += t_elapsed;
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Done with neighbor generation ---> %f \n", t_elapsed);
-#endif
-
- /*validate neighbors list*/
- nbrs_start = (int *) calloc (system->N, INT_SIZE);
- nbrs_end = (int *) calloc (system->N, INT_SIZE);
-
- copy_host_device (nbrs_start, far_nbrs->index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__ );
- copy_host_device (nbrs_end, far_nbrs->end_index, INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__ );
-
- int device_nbrs = 0;
- for(i = 0; i < system->N; i++)
- {
- if ((nbrs_end[i] - nbrs_start[i]) > max_nbrs)
- max_nbrs = nbrs_end[i] - nbrs_start[i];
-
- device_nbrs += nbrs_end[i] - nbrs_start[i];
- }
-#ifdef __CUDA_TEST__
- //fprintf (stderr, " New Device count is : %d \n", device_nbrs);
- //dev_workspace->realloc.num_far = device_nbrs;
-#endif
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, "Max neighbors is ---> %d \n", max_nbrs );
- fprintf (stderr, "DEVICE NEIGHBORS ---> %d \n", device_nbrs);
-#endif
-
- //validate check here
- //get the num_far from the list here
- for (i = 0; i < system->N-1; i++)
- {
- if ((nbrs_end[i] - nbrs_start[i]) > (nbrs_start[i+1] - nbrs_start[i]) * DANGER_ZONE )
- {
- dev_workspace->realloc.num_far = device_nbrs;
- //#ifdef __CUDA_MEM__
- //fprintf (stderr, "Need to reallocate the neighbors ----> %d \n", dev_workspace->realloc.num_far);
- //fprintf (stderr, "Reaching the limits of neighbors for index ----> %d (%d %d %d) \n",
- // i, nbrs_start[i], nbrs_end[i], nbrs_start[i+1]);
- //#endif
- }
-
- if (nbrs_end[i] > nbrs_start[i+1]) {
- fprintf( stderr, "**ran out of space on far_nbrs: start[i] = %d, end[i]=%d, start[i+1]=%d, end[i+1] = %d",
- nbrs_start[i], nbrs_end[i], nbrs_start[i+1], nbrs_end[i+1]);
- exit( INSUFFICIENT_SPACE );
- }
- }
-
- if ((nbrs_end[i] - nbrs_start[i]) > (far_nbrs->num_intrs - nbrs_start[i]) * DANGER_ZONE ) {
- dev_workspace->realloc.num_far = device_nbrs;
- //#ifdef __CUDA_MEM__
- //fprintf (stderr, "Need to reallocate the neighbors ----> %d \n", dev_workspace->realloc.num_far);
- //fprintf (stderr, "Reaching the limits of neighbors for index ----> %d start: %d, end: %d, count: %d\n"
- // , i, nbrs_start[i], nbrs_end[i], far_nbrs->num_intrs);
- //#endif
- }
- if (nbrs_end[i] > far_nbrs->num_intrs) {
- fprintf( stderr, "**ran out of space on far_nbrs: top=%d, max=%d",
- nbrs_end[i], far_nbrs->num_intrs );
- exit( INSUFFICIENT_SPACE );
- }
-
- free (nbrs_start);
- free (nbrs_end);
- }
-
- //Code not used anymore
-#if defined DONE
-
- void Choose_Neighbor_Finder( reax_system *system, control_params *control,
- get_far_neighbors_function *Get_Far_Neighbors )
- {
- if( control->periodic_boundaries )
- {
- if( system->box.box_norms[0] > 2.0 * control->vlist_cut &&
- system->box.box_norms[1] > 2.0 * control->vlist_cut &&
- system->box.box_norms[2] > 2.0 * control->vlist_cut )
- (*Get_Far_Neighbors) = Get_Periodic_Far_Neighbors_Big_Box;
- else (*Get_Far_Neighbors) = Get_Periodic_Far_Neighbors_Small_Box;
- }
- else
- (*Get_Far_Neighbors) = Get_NonPeriodic_Far_Neighbors;
- }
-
-
- int compare_near_nbrs(const void *v1, const void *v2)
- {
- return ((*(near_neighbor_data *)v1).nbr - (*(near_neighbor_data *)v2).nbr);
- }
-
-
- int compare_far_nbrs(const void *v1, const void *v2)
- {
- return ((*(far_neighbor_data *)v1).nbr - (*(far_neighbor_data *)v2).nbr);
- }
-
-
- inline void Set_Far_Neighbor( far_neighbor_data *dest, int nbr, real d, real C,
- rvec dvec, ivec rel_box/*, rvec ext_factor*/ )
- {
- dest->nbr = nbr;
- dest->d = d;
- rvec_Scale( dest->dvec, C, dvec );
- ivec_Copy( dest->rel_box, rel_box );
- // rvec_Scale( dest->ext_factor, C, ext_factor );
- }
-
-
- inline void Set_Near_Neighbor(near_neighbor_data *dest, int nbr, real d, real C,
- rvec dvec, ivec rel_box/*, rvec ext_factor*/)
- {
- dest->nbr = nbr;
- dest->d = d;
- rvec_Scale( dest->dvec, C, dvec );
- ivec_Scale( dest->rel_box, C, rel_box );
- // rvec_Scale( dest->ext_factor, C, ext_factor );
- }
-
-
- /* In case bond restrictions are applied, this method checks if
- atom1 and atom2 are allowed to bond with each other */
- inline int can_Bond( static_storage *workspace, int atom1, int atom2 )
- {
- int i;
-
- // fprintf( stderr, "can bond %6d %6d?\n", atom1, atom2 );
-
- if( !workspace->restricted[ atom1 ] && !workspace->restricted[ atom2 ] )
- return 1;
-
- for( i = 0; i < workspace->restricted[ atom1 ]; ++i )
- if( workspace->restricted_list[ atom1 ][i] == atom2 )
- return 1;
-
- for( i = 0; i < workspace->restricted[ atom2 ]; ++i )
- if( workspace->restricted_list[ atom2 ][i] == atom1 )
- return 1;
-
- return 0;
- }
-
-
- /* check if atom2 is on atom1's near neighbor list */
- inline int is_Near_Neighbor( list *near_nbrs, int atom1, int atom2 )
- {
- int i;
-
- for( i=Start_Index(atom1,near_nbrs); i<End_Index(atom1,near_nbrs); ++i )
- if( near_nbrs->select.near_nbr_list[i].nbr == atom2 )
- {
- // fprintf( stderr, "near neighbors %6d %6d\n", atom1, atom2 );
- return 1;
- }
-
- return 0;
- }
-
- void Generate_Neighbor_Lists( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- int i, j, k;
- int x, y, z;
- int *nbr_atoms;
- int atom1, atom2, max;
- int num_far;
- int c, count;
- int grid_top;
- grid *g = &( system->g );
- list *far_nbrs = (*lists) + FAR_NBRS;
- //int hb_type1, hb_type2;
- //list *hbonds = (*lists) + HBOND;
- //int top_hbond1, top_hbond2;
- get_far_neighbors_function Get_Far_Neighbors;
- far_neighbor_data new_nbrs[125];
-#ifndef REORDER_ATOMS
- int l, m;
-#endif
-
- // fprintf( stderr, "\n\tentered nbrs - " );
- if( control->ensemble == iNPT || control->ensemble == sNPT ||
- control->ensemble == NPT )
- Update_Grid( system );
- // fprintf( stderr, "grid updated - " );
-
- Bin_Atoms( system, out_control );
- // fprintf( stderr, "atoms sorted - " );
-
-#ifdef REORDER_ATOMS
- Cluster_Atoms( system, workspace );
- // fprintf( stderr, "atoms clustered - " );
-#endif
-
- Choose_Neighbor_Finder( system, control, &Get_Far_Neighbors );
- // fprintf( stderr, "function chosen - " );
-
- Reset_Neighbor_Lists( system, workspace, lists );
- // fprintf( stderr, "lists cleared - " );
-
- num_far = 0;
- num_near = 0;
- c = 0;
-
- /* first pick up a cell in the grid */
- for( i = 0; i < g->ncell[0]; i++ )
- for( j = 0; j < g->ncell[1]; j++ )
- for( k = 0; k < g->ncell[2]; k++ ) {
- nbrs = g->nbrs[i][j][k];
- nbrs_cp = g->nbrs_cp[i][j][k];
-
- /* pick up an atom from the current cell */
- //#ifdef REORDER_ATOMS
- // for(atom1 = g->start[i][j][k]; atom1 < g->end[i][j][k]; atom1++)
- //#else
- for(l = 0; l < g->top[i][j][k]; ++l ){
- atom1 = g->atoms[i][j][k][l];
- Set_End_Index( atom1, num_far, far_nbrs );
- // fprintf( stderr, "atom %d:\n", atom1 );
-
- itr = 0;
- while( nbrs[itr][0] > 0 ){
- x = nbrs[itr][0];
- y = nbrs[itr][1];
- z = nbrs[itr][2];
-
- // if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
- // SQR(control->r_cut))
- nbr_atoms = g->atoms[x][y][z];
- max_atoms = g->top[x][y][z];
-
- /* pick up another atom from the neighbor cell -
- we have to compare atom1 with its own periodic images as well,
- that's why there is also equality in the if stmt below */
- //#ifdef REORDER_ATOMS
- //for(atom2=g->start[x][y][z]; atom2<g->end[x][y][z]; atom2++)
- //#else
- for( m = 0, atom2=nbr_atoms[m]; m < max; ++m, atom2=nbr_atoms[m] )
- if( atom1 >= atom2 ) {
- //fprintf( stderr, "\tatom2 %d", atom2 );
- //top_near1 = End_Index( atom1, near_nbrs );
- //Set_Start_Index( atom1, num_far, far_nbrs );
- //hb_type1=system->reaxprm.sbp[system->atoms[atom1].type].p_hbond;
- Get_Far_Neighbors( system->atoms[atom1].x,
- system->atoms[atom2].x,
- &(system->box), control, new_nbrs, &count );
- fprintf( stderr, "\t%d count:%d\n", atom2, count );
-
- for( c = 0; c < count; ++c )
- if(atom1 != atom2 || (atom1 == atom2 && new_nbrs[c].d>=0.1)){
- Set_Far_Neighbor(&(far_nbrs->select.far_nbr_list[num_far]),
- atom2, new_nbrs[c].d, 1.0,
- new_nbrs[c].dvec, new_nbrs[c].rel_box );
- ++num_far;
-
- /*fprintf(stderr,"FARNBR:%6d%6d%8.3f[%8.3f%8.3f%8.3f]\n",
- atom1, atom2, new_nbrs[c].d,
- new_nbrs[c].dvec[0], new_nbrs[c].dvec[1],
- new_nbrs[c].dvec[2] ); */
-
-
- /* hydrogen bond lists */
- /*if( control->hb_cut > 0.1 &&
- new_nbrs[c].d <= control->hb_cut ) {
- // fprintf( stderr, "%d %d\n", atom1, atom2 );
- hb_type2=system->reaxprm.sbp[system->atoms[atom2].type].p_hbond;
- if( hb_type1 == 1 && hb_type2 == 2 ) {
- top_hbond1=End_Index(workspace->hbond_index[atom1],hbonds);
- Set_Near_Neighbor(&(hbonds->select.hbond_list[top_hbond1]),
- atom2, new_nbrs[c].d, 1.0, new_nbrs[c].dvec,
- new_nbrs[c].rel_box );
- Set_End_Index( workspace->hbond_index[atom1],
- top_hbond1 + 1, hbonds );
- }
- else if( hb_type1 == 2 && hb_type2 == 1 ) {
- top_hbond2 = End_Index( workspace->hbond_index[atom2], hbonds );
- Set_Near_Neighbor(&(hbonds->select.hbond_list[top_hbond2]),
- atom1, new_nbrs[c].d, -1.0, new_nbrs[c].dvec,
- new_nbrs[c].rel_box );
- Set_End_Index( workspace->hbond_index[atom2],
- top_hbond2 + 1, hbonds );
- }*/
- }
- }
- }
-
- Set_End_Index( atom1, top_far1, far_nbrs );
- }
- }
-
-
- fprintf( stderr, "nbrs done-" );
-
-
- /* apply restrictions on near neighbors only */
- if( (data->step - data->prev_steps) < control->restrict_bonds ) {
- for( atom1 = 0; atom1 < system->N; ++atom1 )
- if( workspace->restricted[ atom1 ] ) {
- // fprintf( stderr, "atom1: %d\n", atom1 );
-
- top_near1 = End_Index( atom1, near_nbrs );
-
- for( j = 0; j < workspace->restricted[ atom1 ]; ++j )
- if(!is_Near_Neighbor(near_nbrs, atom1,
- atom2 = workspace->restricted_list[atom1][j])) {
- fprintf( stderr, "%3d-%3d: added bond by applying restrictions!\n",
- atom1, atom2 );
-
- top_near2 = End_Index( atom2, near_nbrs );
-
- /* we just would like to get the nearest image, so a call to
- Get_Periodic_Far_Neighbors_Big_Box is good enough. */
- Get_Periodic_Far_Neighbors_Big_Box( system->atoms[ atom1 ].x,
- system->atoms[ atom2 ].x,
- &(system->box), control,
- new_nbrs, &count );
-
- Set_Near_Neighbor( &(near_nbrs->select.near_nbr_list[ top_near1 ]),
- atom2, new_nbrs[c].d, 1.0,
- new_nbrs[c].dvec, new_nbrs[c].rel_box );
- ++top_near1;
-
- Set_Near_Neighbor( &(near_nbrs->select.near_nbr_list[ top_near2 ]),
- atom1, new_nbrs[c].d, -1.0,
- new_nbrs[c].dvec, new_nbrs[c].rel_box );
- Set_End_Index( atom2, top_near2+1, near_nbrs );
- }
-
- Set_End_Index( atom1, top_near1, near_nbrs );
- }
- }
- // fprintf( stderr, "restrictions applied-" );
-
-
- /* verify nbrlists, count num_intrs, sort nearnbrs */
- near_nbrs->num_intrs = 0;
- far_nbrs->num_intrs = 0;
- for( i = 0; i < system->N-1; ++i ) {
- if( End_Index(i, near_nbrs) > Start_Index(i+1, near_nbrs) ) {
- fprintf( stderr,
- "step%3d: nearnbr list of atom%d is overwritten by atom%d\n",
- data->step, i+1, i );
- exit( 1 );
- }
-
- near_nbrs->num_intrs += Num_Entries(i, near_nbrs);
-
- if( End_Index(i, far_nbrs) > Start_Index(i+1, far_nbrs) ) {
- fprintf( stderr,
- "step%3d: farnbr list of atom%d is overwritten by atom%d\n",
- data->step, i+1, i );
- exit( 1 );
- }
-
- far_nbrs->num_intrs += Num_Entries(i, far_nbrs);
- }
-
- for( i = 0; i < system->N; ++i ) {
- qsort( &(near_nbrs->select.near_nbr_list[ Start_Index(i, near_nbrs) ]),
- Num_Entries(i, near_nbrs), sizeof(near_neighbor_data),
- compare_near_nbrs );
- }
- // fprintf( stderr, "near nbrs sorted\n" );
-
-
-#ifdef TEST_ENERGY
- /* for( i = 0; i < system->N; ++i ) {
- qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
- Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
- compare_far_nbrs );
- } */
-
- fprintf( stderr, "Near neighbors/atom: %d (compare to 150)\n",
- num_near / system->N );
- fprintf( stderr, "Far neighbors per atom: %d (compare to %d)\n",
- num_far / system->N, control->max_far_nbrs );
-#endif
-
- //fprintf( stderr, "step%d: num of nearnbrs = %6d num of farnbrs: %6d\n",
- // data->step, num_near, num_far );
-
- //fprintf( stderr, "\talloc nearnbrs = %6d alloc farnbrs: %6d\n",
- // system->N * near_nbrs->intrs_per_unit,
- // system->N * far_nbrs->intrs_per_unit );
- }
-
-
-
- void Generate_Neighbor_Lists( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
- {
- int i, j, k, l, m, itr;
- int x, y, z;
- int atom1, atom2, max;
- int num_far, c, count;
- int *nbr_atoms;
- ivec *nbrs;
- rvec *nbrs_cp;
- grid *g;
- list *far_nbrs;
- get_far_neighbors_function Get_Far_Neighbors;
- far_neighbor_data new_nbrs[125];
-
- g = &( system->g );
- far_nbrs = (*lists) + FAR_NBRS;
-
- // fprintf( stderr, "\n\tentered nbrs - " );
- if( control->ensemble == iNPT ||
- control->ensemble == sNPT ||
- control->ensemble == NPT )
- Update_Grid( system );
- // fprintf( stderr, "grid updated - " );
-
- Bin_Atoms( system, out_control );
- // fprintf( stderr, "atoms sorted - " );
- Choose_Neighbor_Finder( system, control, &Get_Far_Neighbors );
- // fprintf( stderr, "function chosen - " );
- Reset_Neighbor_Lists( system, workspace, lists );
- // fprintf( stderr, "lists cleared - " );
-
- num_far = 0;
- c = 0;
-
- /* first pick up a cell in the grid */
- for( i = 0; i < g->ncell[0]; i++ )
- for( j = 0; j < g->ncell[1]; j++ )
- for( k = 0; k < g->ncell[2]; k++ ) {
- nbrs = g->nbrs[i][j][k];
- nbrs_cp = g->nbrs_cp[i][j][k];
- fprintf( stderr, "gridcell %d %d %d\n", i, j, k );
-
- /* pick up an atom from the current cell */
- for(l = 0; l < g->top[i][j][k]; ++l ){
- atom1 = g->atoms[i][j][k][l];
- Set_Start_Index( atom1, num_far, far_nbrs );
- fprintf( stderr, "\tatom %d\n", atom1 );
-
- itr = 0;
- while( nbrs[itr][0] > 0 ){
- x = nbrs[itr][0];
- y = nbrs[itr][1];
- z = nbrs[itr][2];
- fprintf( stderr, "\t\tgridcell %d %d %d\n", x, y, z );
-
- // if( DistSqr_to_CP(nbrs_cp[itr], system->atoms[atom1].x ) <=
- // SQR(control->r_cut))
- nbr_atoms = g->atoms[x][y][z];
- max = g->top[x][y][z];
- fprintf( stderr, "\t\tmax: %d\n", max );
-
-
- /* pick up another atom from the neighbor cell -
- we have to compare atom1 with its own periodic images as well,
- that's why there is also equality in the if stmt below */
- for( m = 0, atom2=nbr_atoms[m]; m < max; ++m, atom2=nbr_atoms[m] )
- if( atom1 >= atom2 ) {
- Get_Far_Neighbors( system->atoms[atom1].x,
- system->atoms[atom2].x,
- &(system->box), control, new_nbrs, &count );
- fprintf( stderr, "\t\t\t%d count:%d\n", atom2, count );
-
- for( c = 0; c < count; ++c )
- if(atom1 != atom2 || (atom1 == atom2 && new_nbrs[c].d>=0.1)){
- Set_Far_Neighbor(&(far_nbrs->select.far_nbr_list[num_far]),
- atom2, new_nbrs[c].d, 1.0,
- new_nbrs[c].dvec, new_nbrs[c].rel_box );
- ++num_far;
-
- /*fprintf(stderr,"FARNBR:%6d%6d%8.3f[%8.3f%8.3f%8.3f]\n",
- atom1, atom2, new_nbrs[c].d,
- new_nbrs[c].dvec[0], new_nbrs[c].dvec[1],
- new_nbrs[c].dvec[2] ); */
- }
- }
-
- ++itr;
- }
-
- Set_End_Index( atom1, num_far, far_nbrs );
- }
- }
-
- far_nbrs->num_intrs = num_far;
- fprintf( stderr, "nbrs done, num_far: %d\n", num_far );
-
-#if defined(DEBUG)
- for( i = 0; i < system->N; ++i ) {
- qsort( &(far_nbrs->select.far_nbr_list[ Start_Index(i, far_nbrs) ]),
- Num_Entries(i, far_nbrs), sizeof(far_neighbor_data),
- compare_far_nbrs );
- }
-
- fprintf( stderr, "step%d: num of farnbrs=%6d\n", data->step, num_far );
- fprintf( stderr, "\tallocated farnbrs: %6d\n",
- system->N * far_nbrs->intrs_per_unit );
-#endif
- }
-
-
-
-#endif
diff --git a/PuReMD-GPU/src/neighbors.h b/PuReMD-GPU/src/neighbors.h
index 465d61de9775bdc20130d5f69f537031aaa98ff8..64c14ad29d5194006aacb057a7d80ef54aeee8e4 100644
--- a/PuReMD-GPU/src/neighbors.h
+++ b/PuReMD-GPU/src/neighbors.h
@@ -23,25 +23,35 @@
#include "mytypes.h"
+
void Generate_Neighbor_Lists( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-void Cuda_Generate_Neighbor_Lists (reax_system *system,
- static_storage *workspace, control_params *control, bool);
+ static_storage*, list**, output_controls* );
int Estimate_NumNeighbors( reax_system*, control_params*,
- static_storage*, list** );
+ static_storage*, list** );
+
+int Are_Far_Neighbors( rvec, rvec, simulation_box*, real, far_neighbor_data* );
+
+
+static inline HOST_DEVICE int index_grid_debug (int x, int y, int z, int blocksize)
+{
+ return x * 8 * 8 * blocksize +
+ y * 8 * blocksize +
+ z * blocksize ;
+}
+
-HOST_DEVICE int Are_Far_Neighbors( rvec, rvec, simulation_box*, real, far_neighbor_data* );
+static inline HOST_DEVICE real DistSqr_to_CP( rvec cp, rvec x )
+{
+ int i;
+ real d_sqr = 0;
-GLOBAL void Estimate_NumNeighbors ( reax_atom *, grid , simulation_box *, control_params *, int *);
-GLOBAL void Generate_Neighbor_Lists( reax_atom *, grid , simulation_box *, control_params *, list );
+ for( i = 0; i < 3; ++i )
+ if( cp[i] > NEG_INF )
+ d_sqr += SQR( cp[i] - x[i] );
-GLOBAL void Estimate_NumNeighbors ( reax_atom *,
- grid ,
- simulation_box *,
- control_params *,
- int *, int *, int, int , int, int);
-GLOBAL void fix_sym_indices_far_nbrs (list , int );
+ return d_sqr;
+}
#endif
diff --git a/PuReMD-GPU/src/param.h b/PuReMD-GPU/src/param.h
index f8101896932a4f06a53d3d99c6b18a5be1710078..2b24b056983233840966a8de29ce902ca6beb981 100644
--- a/PuReMD-GPU/src/param.h
+++ b/PuReMD-GPU/src/param.h
@@ -27,13 +27,15 @@
#define MAX_TOKENS 20
#define MAX_TOKEN_LEN 1024
-int Get_Atom_Type( reax_interaction*, char* );
-int Tokenize( char*, char*** );
+int Get_Atom_Type( reax_interaction*, char* );
+
+int Tokenize( char*, char*** );
char Read_Force_Field( FILE*, reax_interaction* );
char Read_Control_File( FILE*, reax_system*, control_params*,
- output_controls* );
+ output_controls* );
+
#endif
diff --git a/PuReMD-GPU/src/random.h b/PuReMD-GPU/src/random.h
index f7edb397293676c6e5a7a7d34ba5d12b8f3dab4a..b19bc58e3dcef04a324b108be718bfbff3e5c06c 100644
--- a/PuReMD-GPU/src/random.h
+++ b/PuReMD-GPU/src/random.h
@@ -23,31 +23,29 @@
#include "mytypes.h"
-HOST_DEVICE inline double Random(double);
-HOST_DEVICE inline void Randomize();
-HOST_DEVICE inline double GRandom(double , double );
-
/* System random number generator used linear congruance method with
large periodicity for generation of pseudo random number. function
Random returns this random number appropriately scaled so that
0 <= Random(range) < range */
-HOST_DEVICE inline double Random(double range)
+static inline HOST_DEVICE double Random(double range)
{
return (random() * range) / 2147483647L;
}
+
/* This function seeds the system pseudo random number generator with
current time. Use this function once in the begining to initialize
the system */
-HOST_DEVICE inline void Randomize()
+static inline HOST_DEVICE void Randomize( )
{
- srandom(time(NULL));
+ srandom( time(NULL) );
}
+
/* GRandom return random number with gaussian distribution with mean
and standard deviation "sigma" */
-HOST_DEVICE inline double GRandom(double mean, double sigma)
+static inline HOST_DEVICE double GRandom(double mean, double sigma)
{
double v1 = Random(2.0) - 1.0;
double v2 = Random(2.0) - 1.0;
@@ -63,4 +61,5 @@ HOST_DEVICE inline double GRandom(double mean, double sigma)
return mean + v1 * sigma * sqrt(-2.0 * log(rsq) / rsq);
}
+
#endif
diff --git a/PuReMD-GPU/src/reset_utils.c b/PuReMD-GPU/src/reset_utils.c
new file mode 100644
index 0000000000000000000000000000000000000000..f79596aa9d29a65f673448d18a28c73c00444e43
--- /dev/null
+++ b/PuReMD-GPU/src/reset_utils.c
@@ -0,0 +1,162 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "reset_utils.h"
+
+#include "list.h"
+#include "vector.h"
+
+
+void Reset_Atoms( reax_system* system )
+{
+ int i;
+
+ for( i = 0; i < system->N; ++i )
+ memset( system->atoms[i].f, 0.0, RVEC_SIZE );
+}
+
+
+void Reset_Pressures( simulation_data *data )
+{
+ rtensor_MakeZero( data->flex_bar.P );
+ data->iso_bar.P = 0;
+ rvec_MakeZero( data->int_press );
+ rvec_MakeZero( data->ext_press );
+ /* fprintf( stderr, "reset: ext_press (%12.6f %12.6f %12.6f)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] ); */
+}
+
+
+void Reset_Simulation_Data( simulation_data* data )
+{
+ data->E_BE = 0;
+ data->E_Ov = 0;
+ data->E_Un = 0;
+ data->E_Lp = 0;
+ data->E_Ang = 0;
+ data->E_Pen = 0;
+ data->E_Coa = 0;
+ data->E_HB = 0;
+ data->E_Tor = 0;
+ data->E_Con = 0;
+ data->E_vdW = 0;
+ data->E_Ele = 0;
+ data->E_Kin = 0;
+}
+
+
+#ifdef TEST_FORCES
+void Reset_Test_Forces( reax_system *system, static_storage *workspace )
+{
+ memset( workspace->f_ele, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_vdw, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_bo, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_be, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_lp, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_ov, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_un, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_ang, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_coa, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_pen, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_hb, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_tor, 0, system->N * sizeof(rvec) );
+ memset( workspace->f_con, 0, system->N * sizeof(rvec) );
+}
+#endif
+
+
+void Reset_Workspace( reax_system *system, static_storage *workspace )
+{
+ memset( workspace->total_bond_order, 0, system->N * sizeof( real ) );
+ memset( workspace->dDeltap_self, 0, system->N * sizeof( rvec ) );
+
+ memset( workspace->CdDelta, 0, system->N * sizeof( real ) );
+ //memset( workspace->virial_forces, 0, system->N * sizeof( rvec ) );
+
+#ifdef TEST_FORCES
+ memset( workspace->dDelta, 0, sizeof(rvec) * system->N );
+ Reset_Test_Forces( system, workspace );
+#endif
+}
+
+
+void Reset_Neighbor_Lists( reax_system *system, control_params *control,
+ static_storage *workspace, list **lists )
+{
+ int i, tmp;
+ list *bonds = (*lists) + BONDS;
+ list *hbonds = (*lists) + HBONDS;
+
+ for( i = 0; i < system->N; ++i ) {
+ tmp = Start_Index( i, bonds );
+ Set_End_Index( i, tmp, bonds );
+ }
+
+ //TODO check if this is needed
+ memset (bonds->select.bond_list, 0, BOND_DATA_SIZE * bonds->num_intrs );
+
+ if( control->hb_cut > 0 )
+ for( i = 0; i < system->N; ++i )
+ if( system->reaxprm.sbp[system->atoms[i].type].p_hbond == 1) {
+ tmp = Start_Index( workspace->hbond_index[i], hbonds );
+ Set_End_Index( workspace->hbond_index[i], tmp, hbonds );
+ /* fprintf( stderr, "i:%d, hbond: %d-%d\n",
+ i, Start_Index( workspace->hbond_index[i], hbonds ),
+ End_Index( workspace->hbond_index[i], hbonds ) );*/
+ }
+}
+
+
+void Reset( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace, list **lists )
+{
+ Reset_Atoms( system );
+
+ Reset_Simulation_Data( data );
+
+ if( control->ensemble == NPT || control->ensemble == sNPT ||
+ control->ensemble == iNPT )
+ Reset_Pressures( data );
+
+ Reset_Workspace( system, workspace );
+
+ Reset_Neighbor_Lists( system, control, workspace, lists );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "reset - ");
+#endif
+}
+
+
+void Reset_Grid( grid *g )
+{
+ memset (g->top, 0, INT_SIZE * g->ncell[0]*g->ncell[1]*g->ncell[2]);
+}
+
+
+void Reset_Marks( grid *g, ivec *grid_stack, int grid_top )
+{
+ int i;
+
+ for( i = 0; i < grid_top; ++i )
+ g->mark[grid_stack[i][0] * g->ncell[1]*g->ncell[2] +
+ grid_stack[i][1] * g->ncell[2] +
+ grid_stack[i][2]] = 0;
+}
diff --git a/PuReMD-GPU/src/reset_utils.h b/PuReMD-GPU/src/reset_utils.h
index 7fb318e8e6b9fec7941ea01f1102886aa7425e3e..190bd7f5632f3b0a2b3291edf82e03aa4922793e 100644
--- a/PuReMD-GPU/src/reset_utils.h
+++ b/PuReMD-GPU/src/reset_utils.h
@@ -23,6 +23,11 @@
#include "mytypes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Reset_Atoms( reax_system* );
void Reset_Pressures( simulation_data* );
@@ -36,10 +41,10 @@ void Reset_Test_Forces( reax_system*, static_storage* );
void Reset_Workspace( reax_system*, static_storage* );
void Reset_Neighbor_Lists( reax_system*, control_params*,
- static_storage*, list** );
+ static_storage*, list** );
void Reset( reax_system*, control_params*, simulation_data*,
- static_storage*, list** );
+ static_storage*, list** );
//void Reset_Neighbor_Lists( reax_system*, static_storage*, list** );
@@ -47,12 +52,9 @@ void Reset_Grid( grid* );
void Reset_Marks( grid*, ivec*, int );
-void Cuda_Reset_Grid( grid* );
+#ifdef __cplusplus
+}
+#endif
-//CUDA functions
-void Cuda_Reset_Workspace (reax_system *, static_storage *);
-void Cuda_Reset( reax_system*, control_params*, simulation_data*,
- static_storage*, list** );
-void Cuda_Reset_Atoms (reax_system *);
#endif
diff --git a/PuReMD-GPU/src/single_body_interactions.c b/PuReMD-GPU/src/single_body_interactions.c
new file mode 100644
index 0000000000000000000000000000000000000000..b26f493e703819f066389991a4845acab113b326
--- /dev/null
+++ b/PuReMD-GPU/src/single_body_interactions.c
@@ -0,0 +1,314 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "single_body_interactions.h"
+
+#include "bond_orders.h"
+#include "index_utils.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+
+
+void LonePair_OverUnder_Coordination_Energy( reax_system *system,
+ control_params *control,
+ simulation_data *data,
+ static_storage *workspace,
+ list **lists,
+ output_controls *out_control )
+{
+ int i, j, pj, type_i, type_j;
+ real Delta_lpcorr, dfvl;
+ real e_lp, expvd2, inv_expvd2, dElp, CElp, DlpVi;
+ real e_lph, Di, vov3, deahu2dbo, deahu2dsbo;
+ real e_ov, CEover1, CEover2, CEover3, CEover4;
+ real exp_ovun1, exp_ovun2, sum_ovun1, sum_ovun2;
+ real exp_ovun2n, exp_ovun6, exp_ovun8;
+ real inv_exp_ovun1, inv_exp_ovun2, inv_exp_ovun2n, inv_exp_ovun8;
+ real e_un, CEunder1, CEunder2, CEunder3, CEunder4;
+ real p_lp1, p_lp2, p_lp3;
+ real p_ovun2, p_ovun3, p_ovun4, p_ovun5, p_ovun6, p_ovun7, p_ovun8;
+
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ bond_data *pbond;
+ bond_order_data *bo_ij;
+ list *bonds = (*lists) + BONDS;
+
+ /* Initialize parameters */
+ p_lp1 = system->reaxprm.gp.l[15];
+ p_lp3 = system->reaxprm.gp.l[5];
+ p_ovun3 = system->reaxprm.gp.l[32];
+ p_ovun4 = system->reaxprm.gp.l[31];
+ p_ovun6 = system->reaxprm.gp.l[6];
+ p_ovun7 = system->reaxprm.gp.l[8];
+ p_ovun8 = system->reaxprm.gp.l[9];
+
+ for( i = 0; i < system->N; ++i ) {
+ /* set the parameter pointer */
+ type_i = system->atoms[i].type;
+ sbp_i = &(system->reaxprm.sbp[ type_i ]);
+
+ /* lone-pair Energy */
+ p_lp2 = sbp_i->p_lp2;
+ expvd2 = EXP( -75 * workspace->Delta_lp[i] );
+ inv_expvd2 = 1. / (1. + expvd2 );
+
+ /* calculate the energy */
+ data->E_Lp += e_lp =
+ p_lp2 * workspace->Delta_lp[i] * inv_expvd2;
+
+ dElp = p_lp2 * inv_expvd2 +
+ 75 * p_lp2 * workspace->Delta_lp[i] * expvd2 * SQR(inv_expvd2);
+ CElp = dElp * workspace->dDelta_lp[i];
+
+ workspace->CdDelta[i] += CElp; // lp - 1st term
+
+#ifdef TEST_ENERGY
+ fprintf( out_control->elp, "%23.15e%23.15e%23.15e%23.15e\n",
+ p_lp2, workspace->Delta_lp_temp[i], expvd2, dElp );
+ fprintf( out_control->elp, "%6d%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i]+1, workspace->nlp[i], e_lp, data->E_Lp );
+#endif
+#ifdef TEST_FORCES
+ Add_dDelta( system, lists, i, CElp, workspace->f_lp ); // lp - 1st term
+#endif
+
+ /* correction for C2 */
+ if( system->reaxprm.gp.l[5] > 0.001 &&
+ !strcmp( system->reaxprm.sbp[type_i].name, "C" ) )
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj )
+ if( i < bonds->select.bond_list[pj].nbr ) {
+ j = bonds->select.bond_list[pj].nbr;
+ type_j = system->atoms[j].type;
+
+ if( !strcmp( system->reaxprm.sbp[type_j].name, "C" ) ) {
+ twbp = &( system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+ bo_ij = &( bonds->select.bond_list[pj].bo_data );
+ Di = workspace->Delta[i];
+ vov3 = bo_ij->BO - Di - 0.040*POW(Di, 4.);
+
+ if( vov3 > 3. ) {
+ data->E_Lp += e_lph = p_lp3 * SQR(vov3-3.0);
+ //estrain(i) += e_lph;
+
+ deahu2dbo = 2.*p_lp3*(vov3 - 3.);
+ deahu2dsbo = 2.*p_lp3*(vov3 - 3.)*(-1. - 0.16*POW(Di, 3.));
+
+ bo_ij->Cdbo += deahu2dbo;
+ workspace->CdDelta[i] += deahu2dsbo;
+#ifdef TEST_ENERGY
+ fprintf(out_control->elp,"C2cor%6d%6d%23.15e%23.15e%23.15e\n",
+ // workspace->orig_id[i], workspace->orig_id[j],
+ i+1, j+1, e_lph, deahu2dbo, deahu2dsbo );
+#endif
+#ifdef TEST_FORCES
+ Add_dBO(system, lists, i, pj, deahu2dbo, workspace->f_lp);
+ Add_dDelta(system, lists, i, deahu2dsbo, workspace->f_lp);
+#endif
+ }
+ }
+
+ }
+ }
+
+
+ for( i = 0; i < system->N; ++i ) {
+ type_i = system->atoms[i].type;
+ sbp_i = &(system->reaxprm.sbp[ type_i ]);
+
+ /* over-coordination energy */
+ if( sbp_i->mass > 21.0 )
+ dfvl = 0.0;
+ else dfvl = 1.0; // only for 1st-row elements
+
+ p_ovun2 = sbp_i->p_ovun2;
+ sum_ovun1 = 0;
+ sum_ovun2 = 0;
+
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ) {
+ j = bonds->select.bond_list[pj].nbr;
+ type_j = system->atoms[j].type;
+ bo_ij = &(bonds->select.bond_list[pj].bo_data);
+ sbp_j = &(system->reaxprm.sbp[ type_j ]);
+ twbp = &(system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+
+ sum_ovun1 += twbp->p_ovun1 * twbp->De_s * bo_ij->BO;
+ sum_ovun2 += (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j])*
+ ( bo_ij->BO_pi + bo_ij->BO_pi2 );
+
+ /*fprintf( stdout, "%4d%4d%23.15e%23.15e%23.15e\n",
+ i+1, j+1,
+ dfvl * workspace->Delta_lp_temp[j],
+ sbp_j->nlp_opt,
+ workspace->nlp_temp[j] );*/
+ }
+
+ exp_ovun1 = p_ovun3 * EXP( p_ovun4 * sum_ovun2 );
+ inv_exp_ovun1 = 1.0 / (1 + exp_ovun1);
+ Delta_lpcorr = workspace->Delta[i] -
+ (dfvl*workspace->Delta_lp_temp[i]) * inv_exp_ovun1;
+
+ exp_ovun2 = EXP( p_ovun2 * Delta_lpcorr );
+ inv_exp_ovun2 = 1.0 / (1.0 + exp_ovun2);
+
+ DlpVi = 1.0 / (Delta_lpcorr + sbp_i->valency + 1e-8 );
+ CEover1 = Delta_lpcorr * DlpVi * inv_exp_ovun2;
+
+ data->E_Ov += e_ov = sum_ovun1 * CEover1;
+
+ CEover2 = sum_ovun1 * DlpVi * inv_exp_ovun2 *
+ ( 1.0 - Delta_lpcorr*( DlpVi + p_ovun2 * exp_ovun2 * inv_exp_ovun2 ) );
+
+ CEover3 = CEover2 * (1.0 - dfvl*workspace->dDelta_lp[i]*inv_exp_ovun1 );
+
+ CEover4 = CEover2 * (dfvl*workspace->Delta_lp_temp[i]) *
+ p_ovun4 * exp_ovun1 * SQR(inv_exp_ovun1);
+
+
+ /* under-coordination potential */
+ p_ovun2 = sbp_i->p_ovun2;
+ p_ovun5 = sbp_i->p_ovun5;
+
+ exp_ovun2n = 1.0 / exp_ovun2;
+ exp_ovun6 = EXP( p_ovun6 * Delta_lpcorr );
+ exp_ovun8 = p_ovun7 * EXP(p_ovun8 * sum_ovun2);
+ inv_exp_ovun2n = 1.0 / (1.0 + exp_ovun2n);
+ inv_exp_ovun8 = 1.0 / (1.0 + exp_ovun8);
+
+ data->E_Un += e_un =
+ -p_ovun5 * (1.0 - exp_ovun6) * inv_exp_ovun2n * inv_exp_ovun8;
+
+ CEunder1 = inv_exp_ovun2n * ( p_ovun5*p_ovun6*exp_ovun6*inv_exp_ovun8 +
+ p_ovun2 * e_un * exp_ovun2n);
+ CEunder2 = -e_un * p_ovun8 * exp_ovun8 * inv_exp_ovun8;
+ CEunder3 = CEunder1 * (1.0 - dfvl*workspace->dDelta_lp[i]*inv_exp_ovun1);
+ CEunder4 = CEunder1 * (dfvl*workspace->Delta_lp_temp[i]) *
+ p_ovun4 * exp_ovun1 * SQR(inv_exp_ovun1) + CEunder2;
+
+ //fprintf( stdout, "%6d%23.15e%23.15e%23.15e\n",
+ // i+1, sum_ovun2, e_ov, e_un );
+
+ /* forces */
+ workspace->CdDelta[i] += CEover3; // OvCoor - 2nd term
+ workspace->CdDelta[i] += CEunder3; // UnCoor - 1st term
+
+#ifdef TEST_FORCES
+ Add_dDelta( system, lists, i, CEover3, workspace->f_ov ); // OvCoor - 2nd
+ Add_dDelta( system, lists, i, CEunder3, workspace->f_un ); // UnCoor - 1st
+#endif
+
+
+ for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
+ pbond = &(bonds->select.bond_list[pj]);
+ j = pbond->nbr;
+ type_j = system->atoms[j].type;
+ bo_ij = &(pbond->bo_data);
+ twbp = &(system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ]);
+
+
+ bo_ij->Cdbo += CEover1 * twbp->p_ovun1 * twbp->De_s; // OvCoor - 1st
+ workspace->CdDelta[j] += CEover4*(1.0 - dfvl*workspace->dDelta_lp[j])*
+ (bo_ij->BO_pi + bo_ij->BO_pi2); // OvCoor - 3a
+ bo_ij->Cdbopi += CEover4 *
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//OvCoor-3b
+ bo_ij->Cdbopi2 += CEover4 *
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//OvCoor-3b
+
+
+ workspace->CdDelta[j] += CEunder4*(1.0-dfvl*workspace->dDelta_lp[j]) *
+ (bo_ij->BO_pi + bo_ij->BO_pi2); // UnCoor - 2a
+ bo_ij->Cdbopi += CEunder4 *
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//UnCoor-2b
+ bo_ij->Cdbopi2 += CEunder4 *
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]);//UnCoor-2b
+
+
+#ifdef TEST_ENERGY
+ /* fprintf( out_control->eov, "%6d%23.15e%23.15e"
+ workspace->orig_id[j]+1,
+ //twbp->p_ovun1,twbp->De_s,Delta_lpcorr*DlpVi*inv_exp_ovun2,
+ CEover1*twbp->p_ovun1*twbp->De_s, CEover3 ); */
+
+ /*fprintf( out_control->eov, "%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[j]+1,
+ CEover4,
+ CEover4*
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
+ CEover4 * (bo_ij->BO_pi + bo_ij->BO_pi2),
+ (1.0 - dfvl*workspace->dDelta_lp[j]),
+ CEover4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
+ (bo_ij->BO_pi + bo_ij->BO_pi2) );*/
+
+ /* fprintf( out_control->eun, "%6d%23.15e\n",
+ workspace->orig_id[j]+1, CEunder3 ); */
+
+ /*fprintf( out_control->eun, "%6d%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[j]+1,
+ CEunder4,
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
+ CEunder4*
+ (workspace->Delta[j] - dfvl*workspace->Delta_lp_temp[j]),
+ CEunder4*(1.0 - dfvl*workspace->dDelta_lp[j])*
+ (bo_ij->BO_pi + bo_ij->BO_pi2) );*/
+#endif
+
+#ifdef TEST_FORCES
+ Add_dBO( system, lists, i, pj, CEover1 * twbp->p_ovun1 * twbp->De_s,
+ workspace->f_ov ); // OvCoor - 1st term
+
+ Add_dDelta( system, lists, j,
+ CEover4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
+ (bo_ij->BO_pi+bo_ij->BO_pi2), workspace->f_ov );//OvCoor3a
+
+ Add_dBOpinpi2( system, lists, i, pj,
+ CEover4 * (workspace->Delta[j] -
+ dfvl * workspace->Delta_lp_temp[j]),
+ CEover4 * (workspace->Delta[j] -
+ dfvl * workspace->Delta_lp_temp[j]),
+ workspace->f_ov, workspace->f_ov ); // OvCoor - 3b
+
+ Add_dDelta( system, lists, j,
+ CEunder4 * (1.0 - dfvl*workspace->dDelta_lp[j]) *
+ (bo_ij->BO_pi + bo_ij->BO_pi2),
+ workspace->f_un ); // UnCoor - 2a
+
+ Add_dBOpinpi2( system, lists, i, pj,
+ CEunder4 * (workspace->Delta[j] -
+ dfvl * workspace->Delta_lp_temp[j]),
+ CEunder4 * (workspace->Delta[j] -
+ dfvl * workspace->Delta_lp_temp[j]),
+ workspace->f_un, workspace->f_un ); // UnCoor - 2b
+#endif
+ }
+
+#ifdef TEST_ENERGY
+
+ fprintf( out_control->eov, "%6d%15.8f%15.8f%15.8f\n",
+ i+1, DlpVi, Delta_lpcorr, sbp_i->valency );
+
+ fprintf( out_control->eov, "%6d%15.8f%15.8f\n",
+ i+1/*workspace->orig_id[i]+1*/, e_ov, data->E_Ov + data->E_Un );
+
+ fprintf( out_control->eov, "%6d%15.8f%15.8f\n",
+ i+1/*workspace->orig_id[i]+1*/, e_un, data->E_Ov + data->E_Un );
+#endif
+ }
+}
diff --git a/PuReMD-GPU/src/single_body_interactions.h b/PuReMD-GPU/src/single_body_interactions.h
index dd26679755915e5faeecb1c3da90e81ff686132f..5ebe03b85785e956074423a17f37494f1c6ae36c 100644
--- a/PuReMD-GPU/src/single_body_interactions.h
+++ b/PuReMD-GPU/src/single_body_interactions.h
@@ -21,33 +21,13 @@
#ifndef __SINGLE_BODY_INTERACTIONS_H_
#define __SINGLE_BODY_INTERACTIONS_H_
-#include <mytypes.h>
+#include "mytypes.h"
+
void LonePair_OverUnder_Coordination_Energy( reax_system*, control_params*,
simulation_data*, static_storage*,
list**, output_controls* );
-//CUDA Functions...
-GLOBAL void Cuda_LonePair_OverUnder_Coordination_Energy ( reax_atom *, global_parameters ,
- single_body_parameters *, two_body_parameters *,
- static_storage , simulation_data *,
- list , int , int );
-
-GLOBAL void test_LonePair_OverUnder_Coordination_Energy_LP ( reax_atom *, global_parameters ,
- single_body_parameters *, two_body_parameters *,
- static_storage , simulation_data *,
- list , int , int,
- real *, real *, real *);
-
-GLOBAL void test_LonePair_OverUnder_Coordination_Energy ( reax_atom *, global_parameters ,
- single_body_parameters *, two_body_parameters *,
- static_storage , simulation_data *,
- list , int , int,
- real *, real *, real *);
-
-GLOBAL void test_LonePair_Postprocess ( reax_atom *, global_parameters ,
- single_body_parameters *, two_body_parameters *,
- static_storage , simulation_data *,
- list , int , int );
+
#endif
diff --git a/PuReMD-GPU/src/sort.h b/PuReMD-GPU/src/sort.h
index 1ccd7116ea529a817496d1265e83b6d81d186430..11bb61288384191a057e58788b03883a7703e5a1 100644
--- a/PuReMD-GPU/src/sort.h
+++ b/PuReMD-GPU/src/sort.h
@@ -23,14 +23,16 @@
#include "mytypes.h"
-HOST_DEVICE inline void h_swap(sparse_matrix_entry *array, int index1, int index2)
+
+static inline HOST_DEVICE void h_swap(sparse_matrix_entry *array, int index1, int index2)
{
sparse_matrix_entry temp = array[index1];
array[index1] = array[index2];
array[index2] = temp;
}
-HOST_DEVICE inline void h_quick_sort(sparse_matrix_entry *array, int start, int end)
+
+static inline HOST_DEVICE void h_quick_sort(sparse_matrix_entry *array, int start, int end)
{
int i = start;
int k = end;
@@ -51,14 +53,16 @@ HOST_DEVICE inline void h_quick_sort(sparse_matrix_entry *array, int start, int
}
}
-inline void d_swap(sparse_matrix_entry *array, int index1, int index2)
+
+static inline void d_swap(sparse_matrix_entry *array, int index1, int index2)
{
sparse_matrix_entry temp = array[index1];
array[index1] = array[index2];
array[index2] = temp;
}
-inline void d_quick_sort(sparse_matrix_entry *array, int start, int end)
+
+static inline void d_quick_sort(sparse_matrix_entry *array, int start, int end)
{
int i = start;
int k = end;
@@ -82,5 +86,4 @@ inline void d_quick_sort(sparse_matrix_entry *array, int start, int end)
}
-
#endif
diff --git a/PuReMD-GPU/src/system_props.c b/PuReMD-GPU/src/system_props.c
new file mode 100644
index 0000000000000000000000000000000000000000..0126b86b776dce8fd30aea0c228731b95104b216
--- /dev/null
+++ b/PuReMD-GPU/src/system_props.c
@@ -0,0 +1,348 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "system_props.h"
+
+#include "box.h"
+#include "vector.h"
+
+
+HOST real Get_Time( )
+{
+ struct timeval tim;
+
+ gettimeofday(&tim, NULL );
+ return( tim.tv_sec + (tim.tv_usec / 1000000.0) );
+}
+
+
+HOST real Get_Timing_Info( real t_start )
+{
+ struct timeval tim;
+ real t_end;
+
+ gettimeofday(&tim, NULL );
+ t_end = tim.tv_sec + (tim.tv_usec / 1000000.0);
+ return (t_end - t_start);
+}
+
+
+void Temperature_Control( control_params *control, simulation_data *data,
+ output_controls *out_control )
+{
+ real tmp;
+
+ if( control->T_mode == 1 ) { // step-wise temperature control
+ if( (data->step - data->prev_steps) %
+ ((int)(control->T_freq / control->dt)) == 0 ) {
+ if( fabs( control->T - control->T_final ) >= fabs( control->T_rate ) )
+ control->T += control->T_rate;
+ else control->T = control->T_final;
+ }
+ }
+ else if( control->T_mode == 2 ) { // constant slope control
+ tmp = control->T_rate * control->dt / control->T_freq;
+
+ if( fabs( control->T - control->T_final ) >= fabs( tmp ) )
+ control->T += tmp;
+ }
+}
+
+
+void Compute_Total_Mass( reax_system *system, simulation_data *data )
+{
+ int i;
+ int blocks;
+ int block_size;
+ real *partial_sums = 0;
+
+ data->M = 0;
+
+ for( i = 0; i < system->N; i++ )
+ data->M += system->reaxprm.sbp[ system->atoms[i].type ].mass;
+
+ data->inv_M = 1. / data->M;
+}
+
+
+void Compute_Center_of_Mass( reax_system *system, simulation_data *data,
+ FILE *fout )
+{
+ int i;
+ real m, xx, xy, xz, yy, yz, zz, det;
+ rvec tvec, diff;
+ rtensor mat, inv;
+
+ int blocks;
+ int block_size;
+ rvec *l_xcm, *l_vcm, *l_amcm;
+ real t_start, t_end;
+
+ rvec_MakeZero( data->xcm ); // position of CoM
+ rvec_MakeZero( data->vcm ); // velocity of CoM
+ rvec_MakeZero( data->amcm ); // angular momentum of CoM
+ rvec_MakeZero( data->avcm ); // angular velocity of CoM
+
+ /* Compute the position, velocity and angular momentum about the CoM */
+ for( i = 0; i < system->N; ++i ) {
+ m = system->reaxprm.sbp[ system->atoms[i].type ].mass;
+
+ rvec_ScaledAdd( data->xcm, m, system->atoms[i].x );
+ rvec_ScaledAdd( data->vcm, m, system->atoms[i].v );
+
+ rvec_Cross( tvec, system->atoms[i].x, system->atoms[i].v );
+ rvec_ScaledAdd( data->amcm, m, tvec );
+
+ /*fprintf( fout,"%3d %g %g %g\n",
+ i+1,
+ system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2] );
+ fprintf( fout, "vcm: %g %g %g\n",
+ data->vcm[0], data->vcm[1], data->vcm[2] );
+ */
+ }
+
+ rvec_Scale( data->xcm, data->inv_M, data->xcm );
+ rvec_Scale( data->vcm, data->inv_M, data->vcm );
+
+ rvec_Cross( tvec, data->xcm, data->vcm );
+ rvec_ScaledAdd( data->amcm, -data->M, tvec );
+
+ data->etran_cm = 0.5 * data->M * rvec_Norm_Sqr( data->vcm );
+
+ /* Calculate and then invert the inertial tensor */
+ xx = xy = xz = yy = yz = zz = 0;
+
+ for( i = 0; i < system->N; ++i ) {
+ m = system->reaxprm.sbp[ system->atoms[i].type ].mass;
+
+ rvec_ScaledSum( diff, 1., system->atoms[i].x, -1., data->xcm );
+ xx += diff[0] * diff[0] * m;
+ xy += diff[0] * diff[1] * m;
+ xz += diff[0] * diff[2] * m;
+ yy += diff[1] * diff[1] * m;
+ yz += diff[1] * diff[2] * m;
+ zz += diff[2] * diff[2] * m;
+ }
+
+#ifdef __DEBUG_CUDA__
+ fprintf (stderr, " xx: %f \n", xx);
+ fprintf (stderr, " xy: %f \n", xy);
+ fprintf (stderr, " xz: %f \n", xz);
+ fprintf (stderr, " yy: %f \n", yy);
+ fprintf (stderr, " yz: %f \n", yz);
+ fprintf (stderr, " zz: %f \n", zz);
+#endif
+
+ mat[0][0] = yy + zz;
+ mat[0][1] = mat[1][0] = -xy;
+ mat[0][2] = mat[2][0] = -xz;
+ mat[1][1] = xx + zz;
+ mat[2][1] = mat[1][2] = -yz;
+ mat[2][2] = xx + yy;
+
+ /* invert the inertial tensor */
+ det = ( mat[0][0] * mat[1][1] * mat[2][2] +
+ mat[0][1] * mat[1][2] * mat[2][0] +
+ mat[0][2] * mat[1][0] * mat[2][1] ) -
+ ( mat[0][0] * mat[1][2] * mat[2][1] +
+ mat[0][1] * mat[1][0] * mat[2][2] +
+ mat[0][2] * mat[1][1] * mat[2][0] );
+
+ inv[0][0] = mat[1][1] * mat[2][2] - mat[1][2] * mat[2][1];
+ inv[0][1] = mat[0][2] * mat[2][1] - mat[0][1] * mat[2][2];
+ inv[0][2] = mat[0][1] * mat[1][2] - mat[0][2] * mat[1][1];
+ inv[1][0] = mat[1][2] * mat[2][0] - mat[1][0] * mat[2][2];
+ inv[1][1] = mat[0][0] * mat[2][2] - mat[0][2] * mat[2][0];
+ inv[1][2] = mat[0][2] * mat[1][0] - mat[0][0] * mat[1][2];
+ inv[2][0] = mat[1][0] * mat[2][1] - mat[2][0] * mat[1][1];
+ inv[2][1] = mat[2][0] * mat[0][1] - mat[0][0] * mat[2][1];
+ inv[2][2] = mat[0][0] * mat[1][1] - mat[1][0] * mat[0][1];
+
+ if( fabs(det) > ALMOST_ZERO )
+ rtensor_Scale( inv, 1./det, inv );
+ else
+ rtensor_MakeZero( inv );
+
+ /* Compute the angular velocity about the centre of mass */
+ rtensor_MatVec( data->avcm, inv, data->amcm );
+ data->erot_cm = 0.5 * E_CONV * rvec_Dot( data->avcm, data->amcm );
+
+#if defined(DEBUG)
+ fprintf( stderr, "xcm: %24.15e %24.15e %24.15e\n",
+ data->xcm[0], data->xcm[1], data->xcm[2] );
+ fprintf( stderr, "vcm: %24.15e %24.15e %24.15e\n",
+ data->vcm[0], data->vcm[1], data->vcm[2] );
+ fprintf( stderr, "amcm: %24.15e %24.15e %24.15e\n",
+ data->amcm[0], data->amcm[1], data->amcm[2] );
+ /* fprintf( fout, "mat: %f %f %f\n %f %f %f\n %f %f %f\n",
+ mat[0][0], mat[0][1], mat[0][2],
+ mat[1][0], mat[1][1], mat[1][2],
+ mat[2][0], mat[2][1], mat[2][2] );
+ fprintf( fout, "inv: %g %g %g\n %g %g %g\n %g %g %g\n",
+ inv[0][0], inv[0][1], inv[0][2],
+ inv[1][0], inv[1][1], inv[1][2],
+ inv[2][0], inv[2][1], inv[2][2] );
+ fflush( fout ); */
+ fprintf( stderr, "avcm: %24.15e %24.15e %24.15e\n",
+ data->avcm[0], data->avcm[1], data->avcm[2] );
+#endif
+}
+
+
+void Compute_Kinetic_Energy( reax_system* system, simulation_data* data )
+{
+ int i;
+ rvec p;
+ real m;
+
+ data->E_Kin = 0.0;
+
+ for (i=0; i < system->N; i++) {
+ m = system->reaxprm.sbp[system->atoms[i].type].mass;
+
+ rvec_Scale( p, m, system->atoms[i].v );
+ data->E_Kin += 0.5 * rvec_Dot( p, system->atoms[i].v );
+
+ /* fprintf(stderr,"%d, %lf, %lf, %lf %lf\n",
+ i,system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2],
+ system->reaxprm.sbp[system->atoms[i].type].mass); */
+ }
+
+ data->therm.T = (2. * data->E_Kin) / (data->N_f * K_B);
+
+ if ( fabs(data->therm.T) < ALMOST_ZERO ) /* avoid T being an absolute zero! */
+ data->therm.T = ALMOST_ZERO;
+}
+
+
+/* IMPORTANT: This function assumes that current kinetic energy and
+ * the center of mass of the system is already computed before.
+ *
+ * IMPORTANT: In Klein's paper, it is stated that a dU/dV term needs
+ * to be added when there are long-range interactions or long-range
+ * corrections to short-range interactions present.
+ * We may want to add that for more accuracy.
+ */
+void Compute_Pressure_Isotropic( reax_system* system, control_params *control,
+ simulation_data* data,
+ output_controls *out_control )
+{
+ int i;
+ reax_atom *p_atom;
+ rvec tx;
+ rvec tmp;
+ simulation_box *box = &(system->box);
+
+ /* Calculate internal pressure */
+ rvec_MakeZero( data->int_press );
+
+ // 0: both int and ext, 1: ext only, 2: int only
+ if( control->press_mode == 0 || control->press_mode == 2 ) {
+ for( i = 0; i < system->N; ++i ) {
+ p_atom = &( system->atoms[i] );
+
+ /* transform x into unitbox coordinates */
+ Transform_to_UnitBox( p_atom->x, box, 1, tx );
+
+ /* this atom's contribution to internal pressure */
+ rvec_Multiply( tmp, p_atom->f, tx );
+ rvec_Add( data->int_press, tmp );
+
+ if( out_control->debug_level > 0 ) {
+ fprintf( out_control->prs, "%-8d%8.2f%8.2f%8.2f",
+ i+1, p_atom->x[0], p_atom->x[1], p_atom->x[2] );
+ fprintf( out_control->prs, "%8.2f%8.2f%8.2f",
+ p_atom->f[0], p_atom->f[1], p_atom->f[2] );
+ fprintf( out_control->prs, "%8.2f%8.2f%8.2f\n",
+ data->int_press[0],data->int_press[1],data->int_press[2]);
+ }
+ }
+ }
+
+ /* kinetic contribution */
+ data->kin_press = 2. * (E_CONV * data->E_Kin) / ( 3. * box->volume * P_CONV );
+
+ /* Calculate total pressure in each direction */
+ data->tot_press[0] = data->kin_press -
+ ((data->int_press[0] + data->ext_press[0]) /
+ (box->box_norms[1] * box->box_norms[2] * P_CONV));
+
+ data->tot_press[1] = data->kin_press -
+ ((data->int_press[1] + data->ext_press[1])/
+ (box->box_norms[0] * box->box_norms[2] * P_CONV));
+
+ data->tot_press[2] = data->kin_press -
+ ((data->int_press[2] + data->ext_press[2])/
+ (box->box_norms[0] * box->box_norms[1] * P_CONV));
+
+ /* Average pressure for the whole box */
+ data->iso_bar.P=(data->tot_press[0]+data->tot_press[1]+data->tot_press[2])/3;
+}
+
+
+void Compute_Pressure_Isotropic_Klein( reax_system* system,
+ simulation_data* data )
+{
+ int i;
+ reax_atom *p_atom;
+ rvec dx;
+
+ // IMPORTANT: This function assumes that current kinetic energy and
+ // the center of mass of the system is already computed before.
+ data->iso_bar.P = 2.0 * data->E_Kin;
+
+ for( i = 0; i < system->N; ++i )
+ {
+ p_atom = &( system->atoms[i] );
+ rvec_ScaledSum(dx,1.0,p_atom->x,-1.0,data->xcm);
+ data->iso_bar.P += ( -F_CONV * rvec_Dot(p_atom->f, dx) );
+ }
+
+ data->iso_bar.P /= (3.0 * system->box.volume);
+
+ // IMPORTANT: In Klein's paper, it is stated that a dU/dV term needs
+ // to be added when there are long-range interactions or long-range
+ // corrections to short-range interactions present.
+ // We may want to add that for more accuracy.
+}
+
+
+void Compute_Pressure( reax_system* system, simulation_data* data,
+ static_storage *workspace )
+{
+ int i;
+ reax_atom *p_atom;
+ rtensor temp;
+
+ rtensor_MakeZero( data->flex_bar.P );
+
+ for( i = 0; i < system->N; ++i ) {
+ p_atom = &( system->atoms[i] );
+ // Distance_on_T3_Gen( data->rcm, p_atom->x, &(system->box), &dx );
+ rvec_OuterProduct( temp, p_atom->v, p_atom->v );
+ rtensor_ScaledAdd( data->flex_bar.P,
+ system->reaxprm.sbp[ p_atom->type ].mass, temp );
+ // rvec_OuterProduct(temp, workspace->virial_forces[i], p_atom->x );
+ rtensor_ScaledAdd( data->flex_bar.P, -F_CONV, temp );
+ }
+
+ rtensor_Scale( data->flex_bar.P, 1.0 / system->box.volume, data->flex_bar.P );
+ data->iso_bar.P = rtensor_Trace( data->flex_bar.P ) / 3.0;
+}
diff --git a/PuReMD-GPU/src/system_props.h b/PuReMD-GPU/src/system_props.h
index d287992f7b178f589808acfbaae1ed474a7a65e2..874132451d02b2d62d87c82065874f04a35b2d37 100644
--- a/PuReMD-GPU/src/system_props.h
+++ b/PuReMD-GPU/src/system_props.h
@@ -21,7 +21,12 @@
#ifndef __SYSTEM_PROP_H_
#define __SYSTEM_PROP_H_
-#include <mytypes.h>
+#include "mytypes.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
real Get_Time( );
@@ -30,21 +35,18 @@ real Get_Timing_Info( real );
void Temperature_Control( control_params*, simulation_data*, output_controls* );
void Compute_Total_Mass( reax_system*, simulation_data* );
-void Cuda_Compute_Total_Mass( reax_system*, simulation_data* );
void Compute_Center_of_Mass( reax_system*, simulation_data*, FILE* );
-void Cuda_Compute_Center_of_Mass( reax_system*, simulation_data*, FILE* );
void Compute_Kinetic_Energy( reax_system*, simulation_data* );
-void Cuda_Compute_Kinetic_Energy( reax_system*, simulation_data* );
void Compute_Pressure( reax_system*, simulation_data*, static_storage* );
void Compute_Pressure_Isotropic( reax_system*, control_params*, simulation_data*, output_controls* );
-void prep_dev_system (reax_system *system);
-GLOBAL void Compute_Total_Mass (single_body_parameters *, reax_atom *, real *, size_t );
-//GLOBAL void Compute_Kinetic_Energy (single_body_parameters *, reax_atom *, unsigned int , simulation_data *, real *);
+#ifdef __cplusplus
+}
+#endif
#endif
diff --git a/PuReMD-GPU/src/testmd.cu b/PuReMD-GPU/src/testmd.c
similarity index 51%
rename from PuReMD-GPU/src/testmd.cu
rename to PuReMD-GPU/src/testmd.c
index 93f286cc2c0f99cc4aa9788195881dd7ffa15f9f..afe0cd4a5cf7e7282d7e2409f7f0edc66c34296a 100644
--- a/PuReMD-GPU/src/testmd.cu
+++ b/PuReMD-GPU/src/testmd.c
@@ -19,9 +19,11 @@
----------------------------------------------------------------------*/
#include "mytypes.h"
+
#include "analyze.h"
#include "box.h"
#include "forces.h"
+#include "grid.h"
#include "init_md.h"
#include "integrate.h"
#include "neighbors.h"
@@ -34,11 +36,17 @@
#include "traj.h"
#include "vector.h"
-#include "grid.h"
-#include "cuda_utils.h"
-#include "cuda_copy.h"
-#include "validation.h"
+#include "cuda_environment.h"
+#include "cuda_forces.h"
+#include "cuda_init_md.h"
+#include "cuda_neighbors.h"
+#include "cuda_post_evolve.h"
+#include "cuda_reset_utils.h"
+#include "cuda_system_props.h"
+#ifdef __BUILD_DEBUG__
+ #include "validation.h"
+#endif
interaction_function Interaction_Functions[NO_OF_INTERACTIONS];
@@ -48,11 +56,10 @@ print_interaction Print_Interactions[NO_OF_INTERACTIONS];
LR_lookup_table *LR;
LR_lookup_table *d_LR;
-list *dev_lists;
+list *dev_lists;
static_storage *dev_workspace;
reax_timing d_timing;
-
//TODO
real *testdata;
@@ -61,13 +68,6 @@ void *scratch;
int BLOCKS, BLOCKS_POW_2, BLOCK_SIZE;
int MATVEC_BLOCKS;
-cublasStatus_t cublasStatus;
-cublasHandle_t cublasHandle;
-
-cusparseHandle_t cusparseHandle;
-cusparseStatus_t cusparseStatus;
-cusparseMatDescr_t matdescriptor;
-
void Post_Evolve( reax_system* system, control_params* control,
simulation_data* data, static_storage* workspace,
@@ -90,12 +90,14 @@ void Post_Evolve( reax_system* system, control_params* control,
/* remove rotational and translational velocity of the center of mass */
if( control->ensemble != NVE &&
control->remove_CoM_vel &&
- data->step && data->step % control->remove_CoM_vel == 0 ) {
+ data->step && data->step % control->remove_CoM_vel == 0 )
+ {
/* compute velocity of the center of mass */
Compute_Center_of_Mass( system, data, out_control->prs );
- for( i = 0; i < system->N; i++ ) {
+ for( i = 0; i < system->N; i++ )
+ {
// remove translational
rvec_ScaledAdd( system->atoms[i].v, -1., data->vcm );
@@ -107,98 +109,6 @@ void Post_Evolve( reax_system* system, control_params* control,
}
}
-GLOBAL void Update_Atoms_Post_Evolve (reax_atom *atoms, simulation_data *data, int N)
-{
- rvec diff, cross;
- int i = blockIdx.x * blockDim.x + threadIdx.x;
- if (i >= N) return;
-
- //for( i = 0; i < system->N; i++ ) {
- // remove translational
- rvec_ScaledAdd( atoms[i].v, -1., data->vcm );
-
- // remove rotational
- rvec_ScaledSum( diff, 1., atoms[i].x, -1., data->xcm );
- rvec_Cross( cross, data->avcm, diff );
- rvec_ScaledAdd( atoms[i].v, -1., cross );
- //}
-}
-
-void Cuda_Post_Evolve( reax_system* system, control_params* control,
- simulation_data* data, static_storage* workspace,
- list** lists, output_controls *out_control )
-{
- int i;
- rvec diff, cross;
-
- /* compute kinetic energy of the system */
- /*
- real *results = (real *) scratch;
- cuda_memset (results, 0, REAL_SIZE * BLOCKS_POW_2, RES_SCRATCH);
- Compute_Kinetic_Energy <<<BLOCKS_POW_2, BLOCK_SIZE, REAL_SIZE * BLOCK_SIZE>>>
- (system->reaxprm.d_sbp, system->d_atoms, system->N,
- (simulation_data *)data->d_simulation_data, (real *) results);
- cudaThreadSynchronize ();
- cudaCheckError ();
- */
-
- //fprintf (stderr, "Cuda_Post_Evolve: Begin\n");
- Cuda_Compute_Kinetic_Energy (system, data);
- //fprintf (stderr, " Cuda_Compute_Kinetic_Energy done.... \n");
-
- /* remove rotational and translational velocity of the center of mass */
- if( control->ensemble != NVE &&
- control->remove_CoM_vel &&
- data->step && data->step % control->remove_CoM_vel == 0 ) {
-
- /*
- rvec t_xcm, t_vcm, t_avcm;
- rvec_MakeZero (t_xcm);
- rvec_MakeZero (t_vcm);
- rvec_MakeZero (t_avcm);
-
- rvec_Copy (t_xcm, data->xcm);
- rvec_Copy (t_vcm, data->vcm);
- rvec_Copy (t_avcm, data->avcm);
- */
-
- /* compute velocity of the center of mass */
- Cuda_Compute_Center_of_Mass( system, data, out_control->prs );
- //fprintf (stderr, "Cuda_Compute_Center_of_Mass done... \n");
- /*
- fprintf (stderr, "center of mass done on the device \n");
-
- fprintf (stderr, "xcm --> %4.10f %4.10f \n", t_xcm, data->xcm );
- fprintf (stderr, "vcm --> %4.10f %4.10f \n", t_vcm, data->vcm );
- fprintf (stderr, "avcm --> %4.10f %4.10f \n", t_avcm, data->avcm );
-
- if (check_zero (t_xcm, data->xcm) ||
- check_zero (t_vcm, data->vcm) ||
- check_zero (t_avcm, data->avcm)){
- fprintf (stderr, "SimulationData (xcm, vcm, avcm) does not match between device and host \n");
- exit (0);
- }
- */
-
- //xcm, avcm,
- copy_host_device (data->vcm, ((simulation_data *)data->d_simulation_data)->vcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
- copy_host_device (data->xcm, ((simulation_data *)data->d_simulation_data)->xcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
- copy_host_device (data->avcm, ((simulation_data *)data->d_simulation_data)->avcm, RVEC_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
-
- //fprintf (stderr, "data copied.... \n");
-
- Update_Atoms_Post_Evolve <<< BLOCKS, BLOCK_SIZE >>>
- (system->d_atoms, (simulation_data *)data->d_simulation_data, system->N);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- //fprintf (stderr, " Cuda_Post_Evolve:End \n");
-
- }
-}
-
-
-
void Read_System( char *geof, char *ff, char *ctrlf,
reax_system *system, control_params *control,
@@ -217,15 +127,21 @@ void Read_System( char *geof, char *ff, char *ctrlf,
Read_Control_File( ctrl, system, control, out_control );
/* geo file */
- if( control->geo_format == XYZ ) {
+ if( control->geo_format == XYZ )
+ {
fprintf( stderr, "xyz input is not implemented yet\n" );
- exit(1);
+ exit( 1 );
}
else if( control->geo_format == PDB )
+ {
Read_PDB( geof, system, control, data, workspace );
+ }
else if( control->geo_format == BGF )
+ {
Read_BGF( geof, system, control, data, workspace );
- else if( control->geo_format == ASCII_RESTART ) {
+ }
+ else if( control->geo_format == ASCII_RESTART )
+ {
Read_ASCII_Restart( geof, system, control, data, workspace );
control->restart = 1;
}
@@ -233,9 +149,10 @@ void Read_System( char *geof, char *ff, char *ctrlf,
Read_Binary_Restart( geof, system, control, data, workspace );
control->restart = 1;
}
- else {
+ else
+ {
fprintf( stderr, "unknown geo file format. terminating!\n" );
- exit(1);
+ exit( 1 );
}
#if defined(DEBUG_FOCUS)
@@ -244,17 +161,18 @@ void Read_System( char *geof, char *ff, char *ctrlf,
#endif
}
-void Init_Data_Structures (simulation_data *data)
+
+void Init_Data_Structures( simulation_data *data )
{
//data->step = 0;
//data->prev_steps = 0;
//data->time = 0;
- memset (data, 0, SIMULATION_DATA_SIZE );
+ memset( data, 0, SIMULATION_DATA_SIZE );
}
-int main(int argc, char* argv[])
+int main( int argc, char* argv[] )
{
reax_system system;
control_params control;
@@ -271,16 +189,7 @@ int main(int argc, char* argv[])
lists = (list*) malloc( sizeof(list) * LIST_N );
- cudaDeviceSetLimit (cudaLimitStackSize, 8192);
- cudaDeviceSetCacheConfig (cudaFuncCachePreferL1);
- cudaCheckError ();
-
- cublasCheckError (cublasStatus = cublasCreate (&cublasHandle));
-
- cusparseCheckError (cusparseStatus = cusparseCreate (&cusparseHandle));
- cusparseCheckError (cusparseCreateMatDescr (&matdescriptor));
- cusparseSetMatType (matdescriptor, CUSPARSE_MATRIX_TYPE_GENERAL);
- cusparseSetMatIndexBase (matdescriptor, CUSPARSE_INDEX_BASE_ZERO);
+ Setup_Cuda_Environment( 0, 1, 1 );
dev_lists = (list *) malloc (sizeof (list) * LIST_N );
dev_workspace = (static_storage *) malloc (STORAGE_SIZE);
@@ -289,32 +198,14 @@ int main(int argc, char* argv[])
dev_workspace->realloc.estimate_nbrs = -1;
//Cleanup before usage.
- Init_Data_Structures (&data);
- system.init_thblist = false;
+ Init_Data_Structures( &data );
+ system.init_thblist = FALSE;
Read_System( argv[1], argv[2], argv[3], &system, &control,
&data, &workspace, &out_control );
- compute_blocks (&BLOCKS, &BLOCK_SIZE, system.N);
- compute_nearest_pow_2 (BLOCKS, &BLOCKS_POW_2);
-
- //MATVEC_BLOCKS = system.N;
- //MATVEC_BLOCK_SIZE = 32;
-
- MATVEC_BLOCKS = (system.N * MATVEC_THREADS_PER_ROW / MATVEC_BLOCK_SIZE) +
- ((system.N * MATVEC_THREADS_PER_ROW / MATVEC_BLOCK_SIZE) == 0 ? 0 : 1);
-
-#ifdef __DEBUG_CUDA__
- fprintf (stderr, " MATVEC Blocks : %d, Block_Size : %d \n", MATVEC_BLOCKS, MATVEC_BLOCK_SIZE );
- fprintf (stderr, " Blocks : %d, Blocks_Pow_2 : %d, Block_Size : %d \n", BLOCKS, BLOCKS_POW_2, BLOCK_SIZE );
- fprintf (stderr, " Size of far neighbor data %d \n", sizeof (far_neighbor_data));
- fprintf (stderr, " Size of reax_atom %d \n", sizeof (reax_atom));
- fprintf (stderr, " size of sparse matrix entry %d \n", sizeof (sparse_matrix_entry));
- fprintf (stderr, " TOTAL NUMBER OF ATOMS IN THE SYSTEM --> %d \n", system.N);
-#endif
-
#ifdef __CUDA_MEM__
- print_device_mem_usage ();
+ print_device_mem_usage( );
#endif
#ifdef __BUILD_DEBUG__
@@ -322,71 +213,51 @@ int main(int argc, char* argv[])
&out_control, &Evolve );
#endif
- t_start = Get_Time ();
+ t_start = Get_Time( );
Cuda_Initialize( &system, &control, &data, &workspace, &lists,
- &out_control, &Cuda_Evolve);
- t_elapsed = Get_Timing_Info (t_start);
+ &out_control, &Cuda_Evolve );
+ t_elapsed = Get_Timing_Info( t_start );
#ifdef __DEBUG_CUDA__
- fprintf (stderr, " Cuda Initialize timing ---> %f \n", t_elapsed );
+ fprintf( stderr, " Cuda Initialize timing ---> %f \n", t_elapsed );
#endif
-
#ifdef __CUDA_MEM__
- print_device_mem_usage ();
+ print_device_mem_usage( );
#endif
#ifdef __BUILD_DEBUG__
Reset( &system, &control, &data, &workspace, &lists );
#endif
- Cuda_Reset( &system, &control, &data, &workspace, &lists );
-
+ Cuda_Reset( &system, &control, &data, &workspace, &lists );
#ifdef __BUILD_DEBUG__
- Generate_Neighbor_Lists ( &system, &control, &data, &workspace,
+ Generate_Neighbor_Lists( &system, &control, &data, &workspace,
&lists, &out_control );
#endif
- /*
- dim3 blockspergrid (system.g.ncell[0], system.g.ncell[1], system.g.ncell[2]);
- dim3 threadsperblock (system.g.max_atoms);
-
- t_start = Get_Time ();
- Cuda_Bin_Atoms (&system, &workspace);
- Cuda_Bin_Atoms_Sync ( &system );
-
- Generate_Neighbor_Lists <<<blockspergrid, threadsperblock >>>
- (system.d_atoms, system.d_g, system.d_box,
- (control_params *)control.d_control, *(dev_lists + FAR_NBRS));
- cudaThreadSynchronize ();
- cudaCheckError ();
- t_elapsed = Get_Timing_Info (t_start);
- d_timing.nbrs += t_elapsed;
- */
-
- Cuda_Generate_Neighbor_Lists (&system, &workspace, &control, false);
+ Cuda_Generate_Neighbor_Lists( &system, &workspace, &control, FALSE );
#ifdef __BUILD_DEBUG__
Compute_Forces(&system, &control, &data, &workspace, &lists, &out_control);
#endif
- Cuda_Compute_Forces(&system, &control, &data, &workspace, &lists, &out_control);
+ Cuda_Compute_Forces(&system, &control, &data, &workspace, &lists, &out_control);
#ifdef __BUILD_DEBUG__
Compute_Kinetic_Energy( &system, &data );
#endif
- Cuda_Compute_Kinetic_Energy (&system, &data);
+ Cuda_Compute_Kinetic_Energy (&system, &data);
#ifndef __BUILD_DEBUG__
- // Here sync the simulation data, because it has been changed.
- Prep_Device_For_Output ( &system, &data );
+ Cuda_Setup_Output( &system, &data );
Output_Results(&system, &control, &data, &workspace, &lists, &out_control);
#endif
#ifdef __BUILD_DEBUG__
- if (!validate_device (&system, &data, &workspace, &lists) )
+ if( !validate_device (&system, &data, &workspace, &lists) )
{
fprintf (stderr, " Results does not match between Device and host @ step --> %d \n", data.step);
exit (1);
@@ -397,50 +268,45 @@ int main(int argc, char* argv[])
fprintf (stderr, "step -> %d <- done. \n", data.step);
#endif
-
++data.step;
-
- for( ; data.step <= control.nsteps; data.step++ ) {
-
- //fprintf (stderr, "Begin ... \n");
- //to Sync step to the device.
- //Sync_Host_Device (&data, (simulation_data *)data.d_simulation_data, cudaMemcpyHostToDevice );
- copy_host_device (&data.step, &((simulation_data *)data.d_simulation_data)->step,
- INT_SIZE, cudaMemcpyHostToDevice, RES_SIMULATION_DATA );
+ for( ; data.step <= control.nsteps; data.step++ )
+ {
+ Cuda_Setup_Evolve( &system, &control, &data, &workspace, &lists, &out_control );
//fprintf (stderr, "Synched data .... \n");
- if( control.T_mode ) {
+ if( control.T_mode )
+ {
Temperature_Control( &control, &data, &out_control );
- Sync_Host_Device (&control, (control_params *)control.d_control, cudaMemcpyHostToDevice );
+ Cuda_Sync_Temp( &control );
}
//fprintf (stderr, "Temp. Control done ... \n");
#ifdef __BUILD_DEBUG__
Evolve( &system, &control, &data, &workspace, &lists, &out_control );
#endif
- Cuda_Evolve( &system, &control, &data, &workspace, &lists, &out_control );
+ Cuda_Evolve( &system, &control, &data, &workspace, &lists, &out_control );
//fprintf (stderr, "Evolve done \n");
-
#ifdef __BUILD_DEBUG__
Post_Evolve( &system, &control, &data, &workspace, &lists, &out_control );
#endif
+
Cuda_Post_Evolve( &system, &control, &data, &workspace, &lists, &out_control );
//fprintf (stderr, "Post Evolve done \n");
#ifndef __BUILD_DEBUG__
- Prep_Device_For_Output ( &system, &data );
- Output_Results(&system, &control, &data, &workspace, &lists, &out_control);
+ Cuda_Setup_Output( &system, &data );
+ Output_Results( &system, &control, &data, &workspace, &lists, &out_control );
- /*
- Analysis( &system, &control, &data, &workspace, &lists, &out_control );
- */
+ //Analysis( &system, &control, &data, &workspace, &lists, &out_control );
steps = data.step - data.prev_steps;
if( steps && out_control.restart_freq &&
steps % out_control.restart_freq == 0 )
+ {
Write_Restart( &system, &control, &data, &workspace, &out_control );
+ }
#endif
#ifdef __BUILD_DEBUG__
@@ -463,5 +329,7 @@ int main(int argc, char* argv[])
data.timing.elapsed = Get_Timing_Info( data.timing.start );
fprintf( out_control.log, "total: %.2f secs\n", data.timing.elapsed );
+ Cleanup_Cuda_Environment( );
+
return 0;
}
diff --git a/PuReMD-GPU/src/three_body_interactions.c b/PuReMD-GPU/src/three_body_interactions.c
new file mode 100644
index 0000000000000000000000000000000000000000..7ac96e057c6c799ba88204f3f6339fe54b3c61da
--- /dev/null
+++ b/PuReMD-GPU/src/three_body_interactions.c
@@ -0,0 +1,801 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "three_body_interactions.h"
+
+#include "bond_orders.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+#include "index_utils.h"
+
+
+/* calculates the theta angle between i-j-k */
+void Calculate_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
+ real *theta, real *cos_theta )
+{
+ (*cos_theta) = Dot( dvec_ji, dvec_jk, 3 ) / ( d_ji * d_jk );
+ if( *cos_theta > 1. ) *cos_theta = 1.0;
+ if( *cos_theta < -1. ) *cos_theta = -1.0;
+
+ (*theta) = ACOS( *cos_theta );
+}
+
+
+/* calculates the derivative of the cosine of the angle between i-j-k */
+void Calculate_dCos_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
+ rvec* dcos_theta_di, rvec* dcos_theta_dj,
+ rvec* dcos_theta_dk )
+{
+ int t;
+ real sqr_d_ji = SQR(d_ji);
+ real sqr_d_jk = SQR(d_jk);
+ real inv_dists = 1.0 / (d_ji * d_jk);
+ real inv_dists3 = POW( inv_dists, 3 );
+ real dot_dvecs = Dot( dvec_ji, dvec_jk, 3 );
+ real Cdot_inv3 = dot_dvecs * inv_dists3;
+
+ for( t = 0; t < 3; ++t ) {
+ (*dcos_theta_di)[t] = dvec_jk[t] * inv_dists -
+ Cdot_inv3 * sqr_d_jk * dvec_ji[t];
+
+ (*dcos_theta_dj)[t] = -(dvec_jk[t] + dvec_ji[t]) * inv_dists +
+ Cdot_inv3 * ( sqr_d_jk * dvec_ji[t] + sqr_d_ji * dvec_jk[t] );
+
+ (*dcos_theta_dk)[t] = dvec_ji[t] * inv_dists -
+ Cdot_inv3 * sqr_d_ji * dvec_jk[t];
+ }
+
+ /*fprintf( stderr,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ dvec_jk[t] * inv_dists*/
+}
+
+
+/* this is a 3-body interaction in which the main role is
+ played by j which sits in the middle of the other two. */
+void Three_Body_Interactions( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, pi, k, pk, t;
+ int type_i, type_j, type_k;
+ int start_j, end_j, start_pk, end_pk;
+ int flag, cnt, num_thb_intrs;
+
+ real temp, temp_bo_jt, pBOjt7;
+ real p_val1, p_val2, p_val3, p_val4, p_val5;
+ real p_val6, p_val7, p_val8, p_val9, p_val10;
+ real p_pen1, p_pen2, p_pen3, p_pen4;
+ real p_coa1, p_coa2, p_coa3, p_coa4;
+ real trm8, expval6, expval7, expval2theta, expval12theta, exp3ij, exp3jk;
+ real exp_pen2ij, exp_pen2jk, exp_pen3, exp_pen4, trm_pen34, exp_coa2;
+ real dSBO1, dSBO2, SBO, SBO2, CSBO2, SBOp, prod_SBO;
+ real CEval1, CEval2, CEval3, CEval4, CEval5, CEval6, CEval7, CEval8;
+ real CEpen1, CEpen2, CEpen3;
+ real e_ang, e_coa, e_pen;
+ real CEcoa1, CEcoa2, CEcoa3, CEcoa4, CEcoa5;
+ real Cf7ij, Cf7jk, Cf8j, Cf9j;
+ real f7_ij, f7_jk, f8_Dj, f9_Dj;
+ real Ctheta_0, theta_0, theta_00, theta, cos_theta, sin_theta;
+ real r_ij, r_jk;
+ real BOA_ij, BOA_jk;
+ real vlpadj;
+ rvec force, ext_press;
+ // rtensor temp_rtensor, total_rtensor;
+ real *total_bo;
+ three_body_header *thbh;
+ three_body_parameters *thbp;
+ three_body_interaction_data *p_ijk, *p_kji;
+ bond_data *pbond_ij, *pbond_jk, *pbond_jt;
+ bond_order_data *bo_ij, *bo_jk, *bo_jt;
+ list *bonds, *thb_intrs;
+ bond_data *bond_list;
+ three_body_interaction_data *thb_list;
+
+ total_bo = workspace->total_bond_order;
+ bonds = (*lists) + BONDS;
+ bond_list = bonds->select.bond_list;
+ thb_intrs = (*lists) + THREE_BODIES;
+ thb_list = thb_intrs->select.three_body_list;
+
+ /* global parameters used in these calculations */
+ p_val6 = system->reaxprm.gp.l[14];
+ p_val8 = system->reaxprm.gp.l[33];
+ p_val9 = system->reaxprm.gp.l[16];
+ p_val10 = system->reaxprm.gp.l[17];
+ num_thb_intrs = 0;
+
+ for( j = 0; j < system->N; ++j ) {
+ // fprintf( out_control->eval, "j: %d\n", j );
+ type_j = system->atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+
+ p_val3 = system->reaxprm.sbp[ type_j ].p_val3;
+ p_val5 = system->reaxprm.sbp[ type_j ].p_val5;
+
+ SBOp = 0, prod_SBO = 1;
+ for( t = start_j; t < end_j; ++t ) {
+ bo_jt = &(bond_list[t].bo_data);
+ SBOp += (bo_jt->BO_pi + bo_jt->BO_pi2);
+ temp = SQR( bo_jt->BO );
+ temp *= temp;
+ temp *= temp;
+ prod_SBO *= EXP( -temp );
+ }
+
+ /* modifications to match Adri's code - 09/01/09 */
+ if( workspace->vlpex[j] >= 0 ){
+ vlpadj = 0;
+ dSBO2 = prod_SBO - 1;
+ }
+ else{
+ vlpadj = workspace->nlp[j];
+ dSBO2 = (prod_SBO - 1) * (1 - p_val8 * workspace->dDelta_lp[j]);
+ }
+
+ SBO = SBOp + (1 - prod_SBO) * (-workspace->Delta_boc[j] - p_val8 * vlpadj);
+ dSBO1 = -8 * prod_SBO * ( workspace->Delta_boc[j] + p_val8 * vlpadj );
+
+ if( SBO <= 0 )
+ SBO2 = 0, CSBO2 = 0;
+ else if( SBO > 0 && SBO <= 1 ) {
+ SBO2 = POW( SBO, p_val9 );
+ CSBO2 = p_val9 * POW( SBO, p_val9 - 1 );
+ }
+ else if( SBO > 1 && SBO < 2 ) {
+ SBO2 = 2 - POW( 2-SBO, p_val9 );
+ CSBO2 = p_val9 * POW( 2 - SBO, p_val9 - 1 );
+ }
+ else
+ SBO2 = 2, CSBO2 = 0;
+
+ expval6 = EXP( p_val6 * workspace->Delta_boc[j] );
+
+ /* unlike 2-body intrs where we enforce i<j, we cannot put any such
+ restrictions here. such a restriction would prevent us from producing
+ all 4-body intrs correctly */
+ for( pi = start_j; pi < end_j; ++pi ) {
+ Set_Start_Index( pi, num_thb_intrs, thb_intrs );
+
+ pbond_ij = &(bond_list[pi]);
+ bo_ij = &(pbond_ij->bo_data);
+ BOA_ij = bo_ij->BO - control->thb_cut;
+
+
+ if( BOA_ij/*bo_ij->BO*/ > (real) 0.0 ) {
+ i = pbond_ij->nbr;
+ r_ij = pbond_ij->d;
+ type_i = system->atoms[i].type;
+ // fprintf( out_control->eval, "i: %d\n", i );
+
+
+ /* first copy 3-body intrs from previously computed ones where i>k.
+ IMPORTANT: if it is less costly to compute theta and its
+ derivative, we should definitely re-compute them,
+ instead of copying!
+ in the second for-loop below, we compute only new 3-body intrs
+ where i < k */
+ for( pk = start_j; pk < pi; ++pk ) {
+ // fprintf( out_control->eval, "pk: %d\n", pk );
+ start_pk = Start_Index( pk, thb_intrs );
+ end_pk = End_Index( pk, thb_intrs );
+
+ for( t = start_pk; t < end_pk; ++t )
+ if( thb_list[t].thb == i ) {
+ p_ijk = &(thb_list[num_thb_intrs]);
+ p_kji = &(thb_list[t]);
+
+ p_ijk->thb = bond_list[pk].nbr;
+ p_ijk->pthb = pk;
+ p_ijk->theta = p_kji->theta;
+ rvec_Copy( p_ijk->dcos_di, p_kji->dcos_dk );
+ rvec_Copy( p_ijk->dcos_dj, p_kji->dcos_dj );
+ rvec_Copy( p_ijk->dcos_dk, p_kji->dcos_di );
+
+ //if (j == 12)
+ //fprintf (stderr, "Adding one for matched atom %d \n", i);
+
+ ++num_thb_intrs;
+ break;
+ }
+ }
+
+
+ /* and this is the second for loop mentioned above */
+ for( pk = pi+1; pk < end_j; ++pk ) {
+ pbond_jk = &(bond_list[pk]);
+ bo_jk = &(pbond_jk->bo_data);
+ BOA_jk = bo_jk->BO - control->thb_cut;
+ k = pbond_jk->nbr;
+ type_k = system->atoms[k].type;
+ p_ijk = &( thb_list[num_thb_intrs] );
+
+ //TODO - CHANGE ORIGINAL
+ if (BOA_jk <= 0) continue;
+
+ Calculate_Theta( pbond_ij->dvec, pbond_ij->d,
+ pbond_jk->dvec, pbond_jk->d,
+ &theta, &cos_theta );
+
+ Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d,
+ pbond_jk->dvec, pbond_jk->d,
+ &(p_ijk->dcos_di), &(p_ijk->dcos_dj),
+ &(p_ijk->dcos_dk) );
+
+ p_ijk->thb = k;
+ p_ijk->pthb = pk;
+ p_ijk->theta = theta;
+
+ //if (j == 12)
+ //fprintf (stderr, "Adding one for the rest %d \n", k);
+
+ sin_theta = SIN( theta );
+ if( sin_theta < 1.0e-5 )
+ sin_theta = 1.0e-5;
+
+ ++num_thb_intrs;
+
+
+ if( BOA_jk > 0.0 &&
+ (bo_ij->BO * bo_jk->BO) > SQR(control->thb_cut)/*0*/) {
+ r_jk = pbond_jk->d;
+ thbh = &( system->reaxprm.thbp[ index_thbp(type_i,type_j,type_k,system->reaxprm.num_atom_types) ] );
+ flag = 0;
+
+ /* if( workspace->orig_id[i] < workspace->orig_id[k] )
+ fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], bo_ij->BO, bo_jk->BO, p_ijk->theta );
+ else
+ fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
+ workspace->orig_id[k], workspace->orig_id[j],
+ workspace->orig_id[i], bo_jk->BO, bo_ij->BO, p_ijk->theta ); */
+
+
+ for( cnt = 0; cnt < thbh->cnt; ++cnt ) {
+ // fprintf( out_control->eval,
+ // "%6d%6d%6d -- exists in thbp\n", i+1, j+1, k+1 );
+
+ if( fabs(thbh->prm[cnt].p_val1) > 0.001 ) {
+ thbp = &( thbh->prm[cnt] );
+
+ /* ANGLE ENERGY */
+ p_val1 = thbp->p_val1;
+ p_val2 = thbp->p_val2;
+ p_val4 = thbp->p_val4;
+ p_val7 = thbp->p_val7;
+ theta_00 = thbp->theta_00;
+
+ exp3ij = EXP( -p_val3 * POW( BOA_ij, p_val4 ) );
+ f7_ij = 1.0 - exp3ij;
+ Cf7ij = p_val3 * p_val4 *
+ POW( BOA_ij, p_val4 - 1.0 ) * exp3ij;
+
+ exp3jk = EXP( -p_val3 * POW( BOA_jk, p_val4 ) );
+ f7_jk = 1.0 - exp3jk;
+ Cf7jk = p_val3 * p_val4 *
+ POW( BOA_jk, p_val4 - 1.0 ) * exp3jk;
+
+ expval7 = EXP( -p_val7 * workspace->Delta_boc[j] );
+ trm8 = 1.0 + expval6 + expval7;
+ f8_Dj = p_val5 - ( (p_val5 - 1.0) * (2.0 + expval6) / trm8 );
+ Cf8j = ( (1.0 - p_val5) / SQR(trm8) ) *
+ (p_val6 * expval6 * trm8 -
+ (2.0 + expval6) * ( p_val6 * expval6 - p_val7 * expval7 ));
+
+ theta_0 = 180.0 -
+ theta_00 * (1.0 - EXP(-p_val10 * (2.0 - SBO2)));
+ theta_0 = DEG2RAD( theta_0 );
+
+ expval2theta = EXP(-p_val2 * SQR(theta_0-theta));
+ if( p_val1 >= 0 )
+ expval12theta = p_val1 * (1.0 - expval2theta);
+ else // To avoid linear Me-H-Me angles (6/6/06)
+ expval12theta = p_val1 * -expval2theta;
+
+ CEval1 = Cf7ij * f7_jk * f8_Dj * expval12theta;
+ CEval2 = Cf7jk * f7_ij * f8_Dj * expval12theta;
+ CEval3 = Cf8j * f7_ij * f7_jk * expval12theta;
+ CEval4 = -2.0 * p_val1 * p_val2 * f7_ij * f7_jk * f8_Dj *
+ expval2theta * (theta_0 - theta);
+
+ Ctheta_0 = p_val10 * DEG2RAD(theta_00) *
+ exp( -p_val10 * (2.0 - SBO2) );
+
+ CEval5 = -CEval4 * Ctheta_0 * CSBO2;
+ CEval6 = CEval5 * dSBO1;
+ CEval7 = CEval5 * dSBO2;
+ CEval8 = -CEval4 / sin_theta;
+
+ data->E_Ang += e_ang = f7_ij * f7_jk * f8_Dj * expval12theta;
+ /* END ANGLE ENERGY*/
+
+
+ /* PENALTY ENERGY */
+ p_pen1 = thbp->p_pen1;
+ p_pen2 = system->reaxprm.gp.l[19];
+ p_pen3 = system->reaxprm.gp.l[20];
+ p_pen4 = system->reaxprm.gp.l[21];
+
+ exp_pen2ij = EXP( -p_pen2 * SQR( BOA_ij - 2.0 ) );
+ exp_pen2jk = EXP( -p_pen2 * SQR( BOA_jk - 2.0 ) );
+ exp_pen3 = EXP( -p_pen3 * workspace->Delta[j] );
+ exp_pen4 = EXP( p_pen4 * workspace->Delta[j] );
+ trm_pen34 = 1.0 + exp_pen3 + exp_pen4;
+ f9_Dj = ( 2.0 + exp_pen3 ) / trm_pen34;
+ Cf9j = (-p_pen3 * exp_pen3 * trm_pen34 -
+ (2.0 + exp_pen3) * ( -p_pen3 * exp_pen3 +
+ p_pen4 * exp_pen4 )) /
+ SQR( trm_pen34 );
+
+ data->E_Pen += e_pen =
+ p_pen1 * f9_Dj * exp_pen2ij * exp_pen2jk;
+
+ CEpen1 = e_pen * Cf9j / f9_Dj;
+ temp = -2.0 * p_pen2 * e_pen;
+ CEpen2 = temp * (BOA_ij - 2.0);
+ CEpen3 = temp * (BOA_jk - 2.0);
+ /* END PENALTY ENERGY */
+
+
+ /* COALITION ENERGY */
+ p_coa1 = thbp->p_coa1;
+ p_coa2 = system->reaxprm.gp.l[2];
+ p_coa3 = system->reaxprm.gp.l[38];
+ p_coa4 = system->reaxprm.gp.l[30];
+
+ exp_coa2 = EXP( p_coa2 * workspace->Delta_boc[j] );
+ data->E_Coa += e_coa =
+ p_coa1 / (1. + exp_coa2) *
+ EXP( -p_coa3 * SQR(total_bo[i] - BOA_ij) ) *
+ EXP( -p_coa3 * SQR(total_bo[k] - BOA_jk) ) *
+ EXP( -p_coa4 * SQR(BOA_ij - 1.5) ) *
+ EXP( -p_coa4 * SQR(BOA_jk - 1.5) );
+
+ CEcoa1 = -2 * p_coa4 * (BOA_ij - 1.5) * e_coa;
+ CEcoa2 = -2 * p_coa4 * (BOA_jk - 1.5) * e_coa;
+ CEcoa3 = -p_coa2 * exp_coa2 * e_coa / (1+exp_coa2);
+ CEcoa4 = -2*p_coa3 * (total_bo[i]-BOA_ij) * e_coa;
+ CEcoa5 = -2*p_coa3 * (total_bo[k]-BOA_jk) * e_coa;
+ /* END COALITION ENERGY */
+
+ /* FORCES */
+ bo_ij->Cdbo += (CEval1 + CEpen2 + (CEcoa1-CEcoa4));
+ bo_jk->Cdbo += (CEval2 + CEpen3 + (CEcoa2-CEcoa5));
+ workspace->CdDelta[j] += ((CEval3 + CEval7) +
+ CEpen1 + CEcoa3);
+ workspace->CdDelta[i] += CEcoa4;
+ workspace->CdDelta[k] += CEcoa5;
+
+ for( t = start_j; t < end_j; ++t ) {
+ pbond_jt = &( bond_list[t] );
+ bo_jt = &(pbond_jt->bo_data);
+ temp_bo_jt = bo_jt->BO;
+ temp = CUBE( temp_bo_jt );
+ pBOjt7 = temp * temp * temp_bo_jt;
+
+ // fprintf( out_control->eval, "%6d%12.8f\n",
+ // workspace->orig_id[ bond_list[t].nbr ],
+ // (CEval6 * pBOjt7) );
+
+ bo_jt->Cdbo += (CEval6 * pBOjt7);
+ bo_jt->Cdbopi += CEval5;
+ bo_jt->Cdbopi2 += CEval5;
+ }
+
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+
+ rvec_ScaledAdd( system->atoms[i].f, CEval8, p_ijk->dcos_di );
+ rvec_ScaledAdd( system->atoms[j].f, CEval8, p_ijk->dcos_dj );
+ rvec_ScaledAdd( system->atoms[k].f, CEval8, p_ijk->dcos_dk );
+
+ /*
+ if (i == 0) fprintf (stderr, " atom %d adding to i (j) = 0\n", j);
+ if (k == 0) fprintf (stderr, " atom %d adding to i (k) = 0\n", j);
+ */
+ }
+ else {
+ /* terms not related to bond order derivatives
+ are added directly into
+ forces and pressure vector/tensor */
+ rvec_Scale( force, CEval8, p_ijk->dcos_di );
+ rvec_Add( system->atoms[i].f, force );
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+ rvec_ScaledAdd( system->atoms[j].f, CEval8, p_ijk->dcos_dj );
+
+ rvec_Scale( force, CEval8, p_ijk->dcos_dk );
+ rvec_Add( system->atoms[k].f, force );
+ rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
+ rvec_Add( data->ext_press, ext_press );
+
+
+ /* This part is for a fully-flexible box */
+ /* rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_di, system->atoms[i].x );
+ rtensor_Scale( total_rtensor, +CEval8, temp_rtensor );
+
+ rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_dj, system->atoms[j].x );
+ rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
+
+ rvec_OuterProduct( temp_rtensor,
+ p_ijk->dcos_dk, system->atoms[k].x );
+ rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
+
+ if( pbond_ij->imaginary || pbond_jk->imaginary )
+ rtensor_ScaledAdd( data->flex_bar.P,
+ -1.0, total_rtensor );
+ else
+ rtensor_Add( data->flex_bar.P, total_rtensor ); */
+ }
+
+#ifdef TEST_ENERGY
+ fprintf( out_control->eval,
+ //"%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e",
+ "%6d%6d%6d%23.15e%23.15e%23.15e\n",
+ i+1, j+1, k+1,
+ //workspace->orig_id[i]+1,
+ //workspace->orig_id[j]+1,
+ //workspace->orig_id[k]+1,
+ //workspace->Delta_boc[j],
+ RAD2DEG(theta), /*BOA_ij, BOA_jk, */
+ e_ang, data->E_Ang );
+
+ /*fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e",
+ p_val3, p_val4, BOA_ij, BOA_jk );
+ fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e",
+ f7_ij, f7_jk, f8_Dj, expval12theta );
+ fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ CEval1, CEval2, CEval3, CEval4, CEval5
+ //CEval6, CEval7, CEval8 );*/
+
+ /*fprintf( out_control->eval,
+ "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ -p_ijk->dcos_di[0]/sin_theta,
+ -p_ijk->dcos_di[1]/sin_theta,
+ -p_ijk->dcos_di[2]/sin_theta,
+ -p_ijk->dcos_dj[0]/sin_theta,
+ -p_ijk->dcos_dj[1]/sin_theta,
+ -p_ijk->dcos_dj[2]/sin_theta,
+ -p_ijk->dcos_dk[0]/sin_theta,
+ -p_ijk->dcos_dk[1]/sin_theta,
+ -p_ijk->dcos_dk[2]/sin_theta );*/
+
+ /* fprintf( out_control->epen,
+ "%23.15e%23.15e%23.15e\n",
+ CEpen1, CEpen2, CEpen3 );
+ fprintf( out_control->epen,
+ "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ workspace->orig_id[k], RAD2DEG(theta),
+ BOA_ij, BOA_jk, e_pen, data->E_Pen ); */
+
+ fprintf( out_control->ecoa,
+ "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i],
+ workspace->orig_id[j],
+ workspace->orig_id[k],
+ RAD2DEG(theta), BOA_ij, BOA_jk,
+ e_coa, data->E_Coa );
+#endif
+
+#ifdef TEST_FORCES /* angle forces */
+ Add_dBO( system, lists, j, pi, CEval1, workspace->f_ang );
+ Add_dBO( system, lists, j, pk, CEval2, workspace->f_ang );
+ Add_dDelta( system, lists,
+ j, CEval3 + CEval7, workspace->f_ang );
+
+ for( t = start_j; t < end_j; ++t ) {
+ pbond_jt = &( bond_list[t] );
+ bo_jt = &(pbond_jt->bo_data);
+ temp_bo_jt = bo_jt->BO;
+ temp = CUBE( temp_bo_jt );
+ pBOjt7 = temp * temp * temp_bo_jt;
+
+ Add_dBO( system, lists, j, t, pBOjt7 * CEval6,
+ workspace->f_ang );
+ Add_dBOpinpi2( system, lists, j, t,
+ CEval5, CEval5,
+ workspace->f_ang, workspace->f_ang );
+ }
+
+ rvec_ScaledAdd( workspace->f_ang[i], CEval8, p_ijk->dcos_di );
+ rvec_ScaledAdd( workspace->f_ang[j], CEval8, p_ijk->dcos_dj );
+ rvec_ScaledAdd( workspace->f_ang[k], CEval8, p_ijk->dcos_dk );
+ /* end angle forces */
+
+ /* penalty forces */
+ Add_dDelta( system, lists, j, CEpen1, workspace->f_pen );
+ Add_dBO( system, lists, j, pi, CEpen2, workspace->f_pen );
+ Add_dBO( system, lists, j, pk, CEpen3, workspace->f_pen );
+ /* end penalty forces */
+
+ /* coalition forces */
+ Add_dBO( system, lists,
+ j, pi, CEcoa1-CEcoa4, workspace->f_coa );
+ Add_dBO( system, lists,
+ j, pk, CEcoa2-CEcoa5, workspace->f_coa );
+ Add_dDelta( system, lists, j, CEcoa3, workspace->f_coa );
+ Add_dDelta( system, lists, i, CEcoa4, workspace->f_coa );
+ Add_dDelta( system, lists, k, CEcoa5, workspace->f_coa );
+ /* end coalition forces */
+#endif
+ }
+ }
+ }
+ }
+ }
+
+ Set_End_Index(pi, num_thb_intrs, thb_intrs );
+ }
+ }
+
+ if( num_thb_intrs >= thb_intrs->num_intrs * DANGER_ZONE ) {
+ workspace->realloc.num_3body = num_thb_intrs;
+ if( num_thb_intrs > thb_intrs->num_intrs ) {
+ fprintf( stderr, "step%d-ran out of space on angle_list: top=%d, max=%d",
+ data->step, num_thb_intrs, thb_intrs->num_intrs );
+ exit( INSUFFICIENT_SPACE );
+ }
+ }
+
+ //fprintf( stderr,"%d: Number of angle interactions: %d\n",
+ // data->step, num_thb_intrs );
+#ifdef TEST_ENERGY
+ fprintf( stderr,"Number of angle interactions: %d\n", num_thb_intrs );
+
+ fprintf( stderr,"Angle Energy:%g\t Penalty Energy:%g\t Coalition Energy:%g\n",
+ data->E_Ang, data->E_Pen, data->E_Coa );
+
+ fprintf( stderr,"3body: ext_press (%23.15e %23.15e %23.15e)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] );
+#endif
+}
+
+
+void Hydrogen_Bonds( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, k, pi, pk, itr, top;
+ int type_i, type_j, type_k;
+ int start_j, end_j, hb_start_j, hb_end_j;
+ int hblist[MAX_BONDS];
+ int num_hb_intrs = 0;
+ real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
+ real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
+ rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
+ rvec dvec_jk, force, ext_press;
+ ivec rel_jk;
+ // rtensor temp_rtensor, total_rtensor;
+ hbond_parameters *hbp;
+ bond_order_data *bo_ij;
+ bond_data *pbond_ij;
+ far_neighbor_data *nbr_jk;
+ list *bonds, *hbonds;
+ bond_data *bond_list;
+ hbond_data *hbond_list;
+
+ bonds = (*lists) + BONDS;
+ bond_list = bonds->select.bond_list;
+
+ hbonds = (*lists) + HBONDS;
+ hbond_list = hbonds->select.hbond_list;
+
+ /* loops below discover the Hydrogen bonds between i-j-k triplets.
+ here j is H atom and there has to be some bond between i and j.
+ Hydrogen bond is between j and k.
+ so in this function i->X, j->H, k->Z when we map
+ variables onto the ones in the handout.*/
+ for( j = 0; j < system->N; ++j )
+ if( system->reaxprm.sbp[system->atoms[j].type].p_hbond==1 ) {// j must be H
+ /*set j's variables */
+ type_j = system->atoms[j].type;
+ start_j = Start_Index(j, bonds);
+ end_j = End_Index(j, bonds);
+ hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
+ hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
+
+ top = 0;
+ for( pi = start_j; pi < end_j; ++pi ) {
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = system->atoms[i].type;
+
+ if( system->reaxprm.sbp[type_i].p_hbond == 2 &&
+ bo_ij->BO >= HB_THRESHOLD )
+ hblist[top++] = pi;
+ }
+
+ // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
+ // j, top, hb_start_j, hb_end_j );
+
+ for( pk = hb_start_j; pk < hb_end_j; ++pk ) {
+ /* set k's varibles */
+ k = hbond_list[pk].nbr;
+ type_k = system->atoms[k].type;
+ nbr_jk = hbond_list[pk].ptr;
+ r_jk = nbr_jk->d;
+ rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
+
+ for( itr=0; itr < top; ++itr ) {
+ pi = hblist[itr];
+ pbond_ij = &( bond_list[pi] );
+ i = pbond_ij->nbr;
+
+ if( i != k ) {
+ bo_ij = &(pbond_ij->bo_data);
+ type_i = system->atoms[i].type;
+ r_ij = pbond_ij->d;
+ hbp = &(system->reaxprm.hbp[ index_hbp(type_i, type_j, type_k, system->reaxprm.num_atom_types) ]);
+ ++num_hb_intrs;
+
+ Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &theta, &cos_theta );
+ /* the derivative of cos(theta) */
+ Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
+ &dcos_theta_di, &dcos_theta_dj,
+ &dcos_theta_dk );
+
+ /* hydrogen bond energy*/
+ sin_theta2 = SIN( theta/2.0 );
+ sin_xhz4 = SQR(sin_theta2);
+ sin_xhz4 *= sin_xhz4;
+ cos_xhz1 = ( 1.0 - cos_theta );
+ exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
+ exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
+ r_jk / hbp->r0_hb - 2.0 ) );
+
+ data->E_HB += e_hb =
+ hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
+
+ CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
+ CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
+ CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
+ 1.0 / hbp->r0_hb);
+
+ /* hydrogen bond forces */
+ bo_ij->Cdbo += CEhb1; // dbo term
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
+ rvec_ScaledAdd( system->atoms[i].f,
+ +CEhb2, dcos_theta_di ); //dcos terms
+ rvec_ScaledAdd( system->atoms[j].f,
+ +CEhb2, dcos_theta_dj );
+
+
+
+
+ //TODO
+ rvec_ScaledAdd( system->atoms[k].f,
+ +CEhb2, dcos_theta_dk );
+
+ //dr terms
+ rvec_ScaledAdd( system->atoms[j].f, -CEhb3/r_jk, dvec_jk );
+
+
+ //TODO
+ rvec_ScaledAdd( system->atoms[k].f, +CEhb3/r_jk, dvec_jk );
+ }
+ else
+ {
+ /* for pressure coupling, terms that are not related
+ to bond order derivatives are added directly into
+ pressure vector/tensor */
+ rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
+ rvec_Add( system->atoms[i].f, force );
+ rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
+ rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
+
+ rvec_ScaledAdd( system->atoms[j].f, +CEhb2, dcos_theta_dj );
+
+ ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
+ rvec_Scale( force, +CEhb2, dcos_theta_dk );
+
+
+
+ //TODO
+ rvec_Add( system->atoms[k].f, force );
+
+
+
+ rvec_iMultiply( ext_press, rel_jk, force );
+ rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
+
+ //dr terms
+ rvec_ScaledAdd( system->atoms[j].f, -CEhb3/r_jk, dvec_jk );
+
+ rvec_Scale( force, CEhb3/r_jk, dvec_jk );
+ rvec_Add( system->atoms[k].f, force );
+ rvec_iMultiply( ext_press, rel_jk, force );
+ rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
+
+ /* This part is intended for a fully-flexible box */
+ /* rvec_OuterProduct( temp_rtensor,
+ dcos_theta_di, system->atoms[i].x );
+ rtensor_Scale( total_rtensor, -CEhb2, temp_rtensor );
+
+ rvec_ScaledSum( temp_rvec, -CEhb2, dcos_theta_dj,
+ -CEhb3/r_jk, pbond_jk->dvec );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[j].x );
+ rtensor_Add( total_rtensor, temp_rtensor );
+
+ rvec_ScaledSum( temp_rvec, -CEhb2, dcos_theta_dk,
+ +CEhb3/r_jk, pbond_jk->dvec );
+ rvec_OuterProduct( temp_rtensor,
+ temp_rvec, system->atoms[k].x );
+ rtensor_Add( total_rtensor, temp_rtensor );
+
+ if( pbond_ij->imaginary || pbond_jk->imaginary )
+ rtensor_ScaledAdd( data->flex_bar.P, -1.0, total_rtensor );
+ else
+ rtensor_Add( data->flex_bar.P, total_rtensor ); */
+ }
+
+#ifdef TEST_ENERGY
+ /*fprintf( out_control->ehb,
+ "%23.15e%23.15e%23.15e\n%23.15e%23.15e%23.15e\n%23.15e%23.15e%23.15e\n",
+ dcos_theta_di[0], dcos_theta_di[1], dcos_theta_di[2],
+ dcos_theta_dj[0], dcos_theta_dj[1], dcos_theta_dj[2],
+ dcos_theta_dk[0], dcos_theta_dk[1], dcos_theta_dk[2]);
+ fprintf( out_control->ehb, "%23.15e%23.15e%23.15e\n",
+ CEhb1, CEhb2, CEhb3 ); */
+ fprintf( stderr, //out_control->ehb,
+ "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
+ workspace->orig_id[i],
+ workspace->orig_id[j],
+ workspace->orig_id[k],
+ r_jk, theta, bo_ij->BO, e_hb, data->E_HB );
+
+#endif
+#ifdef TEST_FORCES
+ // dbo term
+ Add_dBO( system, lists, j, pi, +CEhb1, workspace->f_hb );
+ // dcos terms
+ rvec_ScaledAdd( workspace->f_hb[i], +CEhb2, dcos_theta_di );
+ rvec_ScaledAdd( workspace->f_hb[j], +CEhb2, dcos_theta_dj );
+ rvec_ScaledAdd( workspace->f_hb[k], +CEhb2, dcos_theta_dk );
+ // dr terms
+ rvec_ScaledAdd( workspace->f_hb[j], -CEhb3/r_jk, dvec_jk );
+ rvec_ScaledAdd( workspace->f_hb[k], +CEhb3/r_jk, dvec_jk );
+#endif
+ }
+ }
+ }
+ }
+
+ /* fprintf( stderr, "hydbonds: ext_press (%23.15e %23.15e %23.15e)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] ); */
+
+#ifdef TEST_FORCES
+ fprintf( stderr, "Number of hydrogen bonds: %d\n", num_hb_intrs );
+ fprintf( stderr, "Hydrogen Bond Energy: %g\n", data->E_HB );
+#endif
+}
diff --git a/PuReMD-GPU/src/three_body_interactions.cu b/PuReMD-GPU/src/three_body_interactions.cu
deleted file mode 100644
index c2eed63bc52c8e5f3db040d74d9d8d083478ec82..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/three_body_interactions.cu
+++ /dev/null
@@ -1,2462 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "three_body_interactions.h"
-#include "bond_orders.h"
-#include "list.h"
-#include "lookup.h"
-#include "vector.h"
-#include "index_utils.h"
-
-#include "cuda_helpers.h"
-
-
-/* calculates the theta angle between i-j-k */
-HOST_DEVICE void Calculate_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
- real *theta, real *cos_theta )
-{
- (*cos_theta) = Dot( dvec_ji, dvec_jk, 3 ) / ( d_ji * d_jk );
- if( *cos_theta > 1. ) *cos_theta = 1.0;
- if( *cos_theta < -1. ) *cos_theta = -1.0;
-
- (*theta) = ACOS( *cos_theta );
-}
-
-
-/* calculates the derivative of the cosine of the angle between i-j-k */
-HOST_DEVICE void Calculate_dCos_Theta( rvec dvec_ji, real d_ji, rvec dvec_jk, real d_jk,
- rvec* dcos_theta_di, rvec* dcos_theta_dj,
- rvec* dcos_theta_dk )
-{
- int t;
- real sqr_d_ji = SQR(d_ji);
- real sqr_d_jk = SQR(d_jk);
- real inv_dists = 1.0 / (d_ji * d_jk);
- real inv_dists3 = POW( inv_dists, 3 );
- real dot_dvecs = Dot( dvec_ji, dvec_jk, 3 );
- real Cdot_inv3 = dot_dvecs * inv_dists3;
-
- for( t = 0; t < 3; ++t ) {
- (*dcos_theta_di)[t] = dvec_jk[t] * inv_dists -
- Cdot_inv3 * sqr_d_jk * dvec_ji[t];
-
- (*dcos_theta_dj)[t] = -(dvec_jk[t] + dvec_ji[t]) * inv_dists +
- Cdot_inv3 * ( sqr_d_jk * dvec_ji[t] + sqr_d_ji * dvec_jk[t] );
-
- (*dcos_theta_dk)[t] = dvec_ji[t] * inv_dists -
- Cdot_inv3 * sqr_d_ji * dvec_jk[t];
- }
-
- /*fprintf( stderr,
- "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- dvec_jk[t] * inv_dists*/
-}
-
-
-
-
-
-
-
-
-
-
-
-
-/* this is a 3-body interaction in which the main role is
- played by j which sits in the middle of the other two. */
-void Three_Body_Interactions( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, pi, k, pk, t;
- int type_i, type_j, type_k;
- int start_j, end_j, start_pk, end_pk;
- int flag, cnt, num_thb_intrs;
-
- real temp, temp_bo_jt, pBOjt7;
- real p_val1, p_val2, p_val3, p_val4, p_val5;
- real p_val6, p_val7, p_val8, p_val9, p_val10;
- real p_pen1, p_pen2, p_pen3, p_pen4;
- real p_coa1, p_coa2, p_coa3, p_coa4;
- real trm8, expval6, expval7, expval2theta, expval12theta, exp3ij, exp3jk;
- real exp_pen2ij, exp_pen2jk, exp_pen3, exp_pen4, trm_pen34, exp_coa2;
- real dSBO1, dSBO2, SBO, SBO2, CSBO2, SBOp, prod_SBO;
- real CEval1, CEval2, CEval3, CEval4, CEval5, CEval6, CEval7, CEval8;
- real CEpen1, CEpen2, CEpen3;
- real e_ang, e_coa, e_pen;
- real CEcoa1, CEcoa2, CEcoa3, CEcoa4, CEcoa5;
- real Cf7ij, Cf7jk, Cf8j, Cf9j;
- real f7_ij, f7_jk, f8_Dj, f9_Dj;
- real Ctheta_0, theta_0, theta_00, theta, cos_theta, sin_theta;
- real r_ij, r_jk;
- real BOA_ij, BOA_jk;
- real vlpadj;
- rvec force, ext_press;
- // rtensor temp_rtensor, total_rtensor;
- real *total_bo;
- three_body_header *thbh;
- three_body_parameters *thbp;
- three_body_interaction_data *p_ijk, *p_kji;
- bond_data *pbond_ij, *pbond_jk, *pbond_jt;
- bond_order_data *bo_ij, *bo_jk, *bo_jt;
- list *bonds, *thb_intrs;
- bond_data *bond_list;
- three_body_interaction_data *thb_list;
-
- total_bo = workspace->total_bond_order;
- bonds = (*lists) + BONDS;
- bond_list = bonds->select.bond_list;
- thb_intrs = (*lists) + THREE_BODIES;
- thb_list = thb_intrs->select.three_body_list;
-
- /* global parameters used in these calculations */
- p_val6 = system->reaxprm.gp.l[14];
- p_val8 = system->reaxprm.gp.l[33];
- p_val9 = system->reaxprm.gp.l[16];
- p_val10 = system->reaxprm.gp.l[17];
- num_thb_intrs = 0;
-
- for( j = 0; j < system->N; ++j ) {
- // fprintf( out_control->eval, "j: %d\n", j );
- type_j = system->atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
-
- p_val3 = system->reaxprm.sbp[ type_j ].p_val3;
- p_val5 = system->reaxprm.sbp[ type_j ].p_val5;
-
- SBOp = 0, prod_SBO = 1;
- for( t = start_j; t < end_j; ++t ) {
- bo_jt = &(bond_list[t].bo_data);
- SBOp += (bo_jt->BO_pi + bo_jt->BO_pi2);
- temp = SQR( bo_jt->BO );
- temp *= temp;
- temp *= temp;
- prod_SBO *= EXP( -temp );
- }
-
- /* modifications to match Adri's code - 09/01/09 */
- if( workspace->vlpex[j] >= 0 ){
- vlpadj = 0;
- dSBO2 = prod_SBO - 1;
- }
- else{
- vlpadj = workspace->nlp[j];
- dSBO2 = (prod_SBO - 1) * (1 - p_val8 * workspace->dDelta_lp[j]);
- }
-
- SBO = SBOp + (1 - prod_SBO) * (-workspace->Delta_boc[j] - p_val8 * vlpadj);
- dSBO1 = -8 * prod_SBO * ( workspace->Delta_boc[j] + p_val8 * vlpadj );
-
- if( SBO <= 0 )
- SBO2 = 0, CSBO2 = 0;
- else if( SBO > 0 && SBO <= 1 ) {
- SBO2 = POW( SBO, p_val9 );
- CSBO2 = p_val9 * POW( SBO, p_val9 - 1 );
- }
- else if( SBO > 1 && SBO < 2 ) {
- SBO2 = 2 - POW( 2-SBO, p_val9 );
- CSBO2 = p_val9 * POW( 2 - SBO, p_val9 - 1 );
- }
- else
- SBO2 = 2, CSBO2 = 0;
-
- expval6 = EXP( p_val6 * workspace->Delta_boc[j] );
-
- /* unlike 2-body intrs where we enforce i<j, we cannot put any such
- restrictions here. such a restriction would prevent us from producing
- all 4-body intrs correctly */
- for( pi = start_j; pi < end_j; ++pi ) {
- Set_Start_Index( pi, num_thb_intrs, thb_intrs );
-
- pbond_ij = &(bond_list[pi]);
- bo_ij = &(pbond_ij->bo_data);
- BOA_ij = bo_ij->BO - control->thb_cut;
-
-
- if( BOA_ij/*bo_ij->BO*/ > (real) 0.0 ) {
- i = pbond_ij->nbr;
- r_ij = pbond_ij->d;
- type_i = system->atoms[i].type;
- // fprintf( out_control->eval, "i: %d\n", i );
-
-
- /* first copy 3-body intrs from previously computed ones where i>k.
-IMPORTANT: if it is less costly to compute theta and its
-derivative, we should definitely re-compute them,
-instead of copying!
-in the second for-loop below, we compute only new 3-body intrs
-where i < k */
- for( pk = start_j; pk < pi; ++pk ) {
- // fprintf( out_control->eval, "pk: %d\n", pk );
- start_pk = Start_Index( pk, thb_intrs );
- end_pk = End_Index( pk, thb_intrs );
-
- for( t = start_pk; t < end_pk; ++t )
- if( thb_list[t].thb == i ) {
- p_ijk = &(thb_list[num_thb_intrs]);
- p_kji = &(thb_list[t]);
-
- p_ijk->thb = bond_list[pk].nbr;
- p_ijk->pthb = pk;
- p_ijk->theta = p_kji->theta;
- rvec_Copy( p_ijk->dcos_di, p_kji->dcos_dk );
- rvec_Copy( p_ijk->dcos_dj, p_kji->dcos_dj );
- rvec_Copy( p_ijk->dcos_dk, p_kji->dcos_di );
-
- //if (j == 12)
- //fprintf (stderr, "Adding one for matched atom %d \n", i);
-
- ++num_thb_intrs;
- break;
- }
- }
-
-
- /* and this is the second for loop mentioned above */
- for( pk = pi+1; pk < end_j; ++pk ) {
- pbond_jk = &(bond_list[pk]);
- bo_jk = &(pbond_jk->bo_data);
- BOA_jk = bo_jk->BO - control->thb_cut;
- k = pbond_jk->nbr;
- type_k = system->atoms[k].type;
- p_ijk = &( thb_list[num_thb_intrs] );
-
- //TODO - CHANGE ORIGINAL
- if (BOA_jk <= 0) continue;
-
- Calculate_Theta( pbond_ij->dvec, pbond_ij->d,
- pbond_jk->dvec, pbond_jk->d,
- &theta, &cos_theta );
-
- Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d,
- pbond_jk->dvec, pbond_jk->d,
- &(p_ijk->dcos_di), &(p_ijk->dcos_dj),
- &(p_ijk->dcos_dk) );
-
- p_ijk->thb = k;
- p_ijk->pthb = pk;
- p_ijk->theta = theta;
-
- //if (j == 12)
- //fprintf (stderr, "Adding one for the rest %d \n", k);
-
- sin_theta = SIN( theta );
- if( sin_theta < 1.0e-5 )
- sin_theta = 1.0e-5;
-
- ++num_thb_intrs;
-
-
- if( BOA_jk > 0.0 &&
- (bo_ij->BO * bo_jk->BO) > SQR(control->thb_cut)/*0*/) {
- r_jk = pbond_jk->d;
- thbh = &( system->reaxprm.thbp[ index_thbp (type_i,type_j,type_k,&system->reaxprm) ] );
- flag = 0;
-
- /* if( workspace->orig_id[i] < workspace->orig_id[k] )
- fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], bo_ij->BO, bo_jk->BO, p_ijk->theta );
- else
- fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
- workspace->orig_id[k], workspace->orig_id[j],
- workspace->orig_id[i], bo_jk->BO, bo_ij->BO, p_ijk->theta ); */
-
-
- for( cnt = 0; cnt < thbh->cnt; ++cnt ) {
- // fprintf( out_control->eval,
- // "%6d%6d%6d -- exists in thbp\n", i+1, j+1, k+1 );
-
- if( fabs(thbh->prm[cnt].p_val1) > 0.001 ) {
- thbp = &( thbh->prm[cnt] );
-
- /* ANGLE ENERGY */
- p_val1 = thbp->p_val1;
- p_val2 = thbp->p_val2;
- p_val4 = thbp->p_val4;
- p_val7 = thbp->p_val7;
- theta_00 = thbp->theta_00;
-
- exp3ij = EXP( -p_val3 * POW( BOA_ij, p_val4 ) );
- f7_ij = 1.0 - exp3ij;
- Cf7ij = p_val3 * p_val4 *
- POW( BOA_ij, p_val4 - 1.0 ) * exp3ij;
-
- exp3jk = EXP( -p_val3 * POW( BOA_jk, p_val4 ) );
- f7_jk = 1.0 - exp3jk;
- Cf7jk = p_val3 * p_val4 *
- POW( BOA_jk, p_val4 - 1.0 ) * exp3jk;
-
- expval7 = EXP( -p_val7 * workspace->Delta_boc[j] );
- trm8 = 1.0 + expval6 + expval7;
- f8_Dj = p_val5 - ( (p_val5 - 1.0) * (2.0 + expval6) / trm8 );
- Cf8j = ( (1.0 - p_val5) / SQR(trm8) ) *
- (p_val6 * expval6 * trm8 -
- (2.0 + expval6) * ( p_val6 * expval6 - p_val7 * expval7 ));
-
- theta_0 = 180.0 -
- theta_00 * (1.0 - EXP(-p_val10 * (2.0 - SBO2)));
- theta_0 = DEG2RAD( theta_0 );
-
- expval2theta = EXP(-p_val2 * SQR(theta_0-theta));
- if( p_val1 >= 0 )
- expval12theta = p_val1 * (1.0 - expval2theta);
- else // To avoid linear Me-H-Me angles (6/6/06)
- expval12theta = p_val1 * -expval2theta;
-
- CEval1 = Cf7ij * f7_jk * f8_Dj * expval12theta;
- CEval2 = Cf7jk * f7_ij * f8_Dj * expval12theta;
- CEval3 = Cf8j * f7_ij * f7_jk * expval12theta;
- CEval4 = -2.0 * p_val1 * p_val2 * f7_ij * f7_jk * f8_Dj *
- expval2theta * (theta_0 - theta);
-
- Ctheta_0 = p_val10 * DEG2RAD(theta_00) *
- exp( -p_val10 * (2.0 - SBO2) );
-
- CEval5 = -CEval4 * Ctheta_0 * CSBO2;
- CEval6 = CEval5 * dSBO1;
- CEval7 = CEval5 * dSBO2;
- CEval8 = -CEval4 / sin_theta;
-
- data->E_Ang += e_ang = f7_ij * f7_jk * f8_Dj * expval12theta;
- /* END ANGLE ENERGY*/
-
-
- /* PENALTY ENERGY */
- p_pen1 = thbp->p_pen1;
- p_pen2 = system->reaxprm.gp.l[19];
- p_pen3 = system->reaxprm.gp.l[20];
- p_pen4 = system->reaxprm.gp.l[21];
-
- exp_pen2ij = EXP( -p_pen2 * SQR( BOA_ij - 2.0 ) );
- exp_pen2jk = EXP( -p_pen2 * SQR( BOA_jk - 2.0 ) );
- exp_pen3 = EXP( -p_pen3 * workspace->Delta[j] );
- exp_pen4 = EXP( p_pen4 * workspace->Delta[j] );
- trm_pen34 = 1.0 + exp_pen3 + exp_pen4;
- f9_Dj = ( 2.0 + exp_pen3 ) / trm_pen34;
- Cf9j = (-p_pen3 * exp_pen3 * trm_pen34 -
- (2.0 + exp_pen3) * ( -p_pen3 * exp_pen3 +
- p_pen4 * exp_pen4 )) /
- SQR( trm_pen34 );
-
- data->E_Pen += e_pen =
- p_pen1 * f9_Dj * exp_pen2ij * exp_pen2jk;
-
- CEpen1 = e_pen * Cf9j / f9_Dj;
- temp = -2.0 * p_pen2 * e_pen;
- CEpen2 = temp * (BOA_ij - 2.0);
- CEpen3 = temp * (BOA_jk - 2.0);
- /* END PENALTY ENERGY */
-
-
- /* COALITION ENERGY */
- p_coa1 = thbp->p_coa1;
- p_coa2 = system->reaxprm.gp.l[2];
- p_coa3 = system->reaxprm.gp.l[38];
- p_coa4 = system->reaxprm.gp.l[30];
-
- exp_coa2 = EXP( p_coa2 * workspace->Delta_boc[j] );
- data->E_Coa += e_coa =
- p_coa1 / (1. + exp_coa2) *
- EXP( -p_coa3 * SQR(total_bo[i] - BOA_ij) ) *
- EXP( -p_coa3 * SQR(total_bo[k] - BOA_jk) ) *
- EXP( -p_coa4 * SQR(BOA_ij - 1.5) ) *
- EXP( -p_coa4 * SQR(BOA_jk - 1.5) );
-
- CEcoa1 = -2 * p_coa4 * (BOA_ij - 1.5) * e_coa;
- CEcoa2 = -2 * p_coa4 * (BOA_jk - 1.5) * e_coa;
- CEcoa3 = -p_coa2 * exp_coa2 * e_coa / (1+exp_coa2);
- CEcoa4 = -2*p_coa3 * (total_bo[i]-BOA_ij) * e_coa;
- CEcoa5 = -2*p_coa3 * (total_bo[k]-BOA_jk) * e_coa;
- /* END COALITION ENERGY */
-
- /* FORCES */
- bo_ij->Cdbo += (CEval1 + CEpen2 + (CEcoa1-CEcoa4));
- bo_jk->Cdbo += (CEval2 + CEpen3 + (CEcoa2-CEcoa5));
- workspace->CdDelta[j] += ((CEval3 + CEval7) +
- CEpen1 + CEcoa3);
- workspace->CdDelta[i] += CEcoa4;
- workspace->CdDelta[k] += CEcoa5;
-
- for( t = start_j; t < end_j; ++t ) {
- pbond_jt = &( bond_list[t] );
- bo_jt = &(pbond_jt->bo_data);
- temp_bo_jt = bo_jt->BO;
- temp = CUBE( temp_bo_jt );
- pBOjt7 = temp * temp * temp_bo_jt;
-
- // fprintf( out_control->eval, "%6d%12.8f\n",
- // workspace->orig_id[ bond_list[t].nbr ],
- // (CEval6 * pBOjt7) );
-
- bo_jt->Cdbo += (CEval6 * pBOjt7);
- bo_jt->Cdbopi += CEval5;
- bo_jt->Cdbopi2 += CEval5;
- }
-
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
-
- rvec_ScaledAdd( system->atoms[i].f, CEval8, p_ijk->dcos_di );
- rvec_ScaledAdd( system->atoms[j].f, CEval8, p_ijk->dcos_dj );
- rvec_ScaledAdd( system->atoms[k].f, CEval8, p_ijk->dcos_dk );
-
- /*
- if (i == 0) fprintf (stderr, " atom %d adding to i (j) = 0\n", j);
- if (k == 0) fprintf (stderr, " atom %d adding to i (k) = 0\n", j);
- */
- }
- else {
- /* terms not related to bond order derivatives
- are added directly into
- forces and pressure vector/tensor */
- rvec_Scale( force, CEval8, p_ijk->dcos_di );
- rvec_Add( system->atoms[i].f, force );
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
- rvec_ScaledAdd( system->atoms[j].f, CEval8, p_ijk->dcos_dj );
-
- rvec_Scale( force, CEval8, p_ijk->dcos_dk );
- rvec_Add( system->atoms[k].f, force );
- rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
- rvec_Add( data->ext_press, ext_press );
-
-
- /* This part is for a fully-flexible box */
- /* rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_di, system->atoms[i].x );
- rtensor_Scale( total_rtensor, +CEval8, temp_rtensor );
-
- rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_dj, system->atoms[j].x );
- rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
-
- rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_dk, system->atoms[k].x );
- rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
-
- if( pbond_ij->imaginary || pbond_jk->imaginary )
- rtensor_ScaledAdd( data->flex_bar.P,
- -1.0, total_rtensor );
- else
- rtensor_Add( data->flex_bar.P, total_rtensor ); */
- }
-
-#ifdef TEST_ENERGY
- fprintf( out_control->eval,
- //"%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e",
- "%6d%6d%6d%23.15e%23.15e%23.15e\n",
- i+1, j+1, k+1,
- //workspace->orig_id[i]+1,
- //workspace->orig_id[j]+1,
- //workspace->orig_id[k]+1,
- //workspace->Delta_boc[j],
- RAD2DEG(theta), /*BOA_ij, BOA_jk, */
- e_ang, data->E_Ang );
-
- /*fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e",
- p_val3, p_val4, BOA_ij, BOA_jk );
- fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e",
- f7_ij, f7_jk, f8_Dj, expval12theta );
- fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- CEval1, CEval2, CEval3, CEval4, CEval5
- //CEval6, CEval7, CEval8 );*/
-
- /*fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- -p_ijk->dcos_di[0]/sin_theta,
- -p_ijk->dcos_di[1]/sin_theta,
- -p_ijk->dcos_di[2]/sin_theta,
- -p_ijk->dcos_dj[0]/sin_theta,
- -p_ijk->dcos_dj[1]/sin_theta,
- -p_ijk->dcos_dj[2]/sin_theta,
- -p_ijk->dcos_dk[0]/sin_theta,
- -p_ijk->dcos_dk[1]/sin_theta,
- -p_ijk->dcos_dk[2]/sin_theta );*/
-
- /* fprintf( out_control->epen,
- "%23.15e%23.15e%23.15e\n",
- CEpen1, CEpen2, CEpen3 );
- fprintf( out_control->epen,
- "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], RAD2DEG(theta),
- BOA_ij, BOA_jk, e_pen, data->E_Pen ); */
-
- fprintf( out_control->ecoa,
- "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i],
- workspace->orig_id[j],
- workspace->orig_id[k],
- RAD2DEG(theta), BOA_ij, BOA_jk,
- e_coa, data->E_Coa );
-#endif
-
-#ifdef TEST_FORCES /* angle forces */
- Add_dBO( system, lists, j, pi, CEval1, workspace->f_ang );
- Add_dBO( system, lists, j, pk, CEval2, workspace->f_ang );
- Add_dDelta( system, lists,
- j, CEval3 + CEval7, workspace->f_ang );
-
- for( t = start_j; t < end_j; ++t ) {
- pbond_jt = &( bond_list[t] );
- bo_jt = &(pbond_jt->bo_data);
- temp_bo_jt = bo_jt->BO;
- temp = CUBE( temp_bo_jt );
- pBOjt7 = temp * temp * temp_bo_jt;
-
- Add_dBO( system, lists, j, t, pBOjt7 * CEval6,
- workspace->f_ang );
- Add_dBOpinpi2( system, lists, j, t,
- CEval5, CEval5,
- workspace->f_ang, workspace->f_ang );
- }
-
- rvec_ScaledAdd( workspace->f_ang[i], CEval8, p_ijk->dcos_di );
- rvec_ScaledAdd( workspace->f_ang[j], CEval8, p_ijk->dcos_dj );
- rvec_ScaledAdd( workspace->f_ang[k], CEval8, p_ijk->dcos_dk );
- /* end angle forces */
-
- /* penalty forces */
- Add_dDelta( system, lists, j, CEpen1, workspace->f_pen );
- Add_dBO( system, lists, j, pi, CEpen2, workspace->f_pen );
- Add_dBO( system, lists, j, pk, CEpen3, workspace->f_pen );
- /* end penalty forces */
-
- /* coalition forces */
- Add_dBO( system, lists,
- j, pi, CEcoa1-CEcoa4, workspace->f_coa );
- Add_dBO( system, lists,
- j, pk, CEcoa2-CEcoa5, workspace->f_coa );
- Add_dDelta( system, lists, j, CEcoa3, workspace->f_coa );
- Add_dDelta( system, lists, i, CEcoa4, workspace->f_coa );
- Add_dDelta( system, lists, k, CEcoa5, workspace->f_coa );
- /* end coalition forces */
-#endif
- }
- }
- }
- }
- }
-
- Set_End_Index(pi, num_thb_intrs, thb_intrs );
- }
- }
-
- if( num_thb_intrs >= thb_intrs->num_intrs * DANGER_ZONE ) {
- workspace->realloc.num_3body = num_thb_intrs;
- if( num_thb_intrs > thb_intrs->num_intrs ) {
- fprintf( stderr, "step%d-ran out of space on angle_list: top=%d, max=%d",
- data->step, num_thb_intrs, thb_intrs->num_intrs );
- exit( INSUFFICIENT_SPACE );
- }
- }
-
- //fprintf( stderr,"%d: Number of angle interactions: %d\n",
- // data->step, num_thb_intrs );
-#ifdef TEST_ENERGY
- fprintf( stderr,"Number of angle interactions: %d\n", num_thb_intrs );
-
- fprintf( stderr,"Angle Energy:%g\t Penalty Energy:%g\t Coalition Energy:%g\n",
- data->E_Ang, data->E_Pen, data->E_Coa );
-
- fprintf( stderr,"3body: ext_press (%23.15e %23.15e %23.15e)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] );
-#endif
-}
-
-//////////////////////////////////////////////////////////////////////
-//Cuda Function for the Three body interactions.
-//////////////////////////////////////////////////////////////////////
-
-
-
-/* this is a 3-body interaction in which the main role is
- played by j which sits in the middle of the other two. */
-GLOBAL void Three_Body_Interactions( reax_atom *atoms,
- single_body_parameters *sbp,
- three_body_header *d_thbp,
- global_parameters g_params,
- control_params *control,
- simulation_data *data,
- static_storage p_workspace,
- list p_bonds, list p_thb_intrs,
- int N, int num_atom_types,
- real *E_Ang, real *E_Pen, real *E_Coa, rvec *aux_ext_press )
-{
- int i, j, pi, k, pk, t;
- int type_i, type_j, type_k;
- int start_j, end_j, start_pk, end_pk;
- int flag, cnt, num_thb_intrs;
-
- real temp, temp_bo_jt, pBOjt7;
- real p_val1, p_val2, p_val3, p_val4, p_val5;
- real p_val6, p_val7, p_val8, p_val9, p_val10;
- real p_pen1, p_pen2, p_pen3, p_pen4;
- real p_coa1, p_coa2, p_coa3, p_coa4;
- real trm8, expval6, expval7, expval2theta, expval12theta, exp3ij, exp3jk;
- real exp_pen2ij, exp_pen2jk, exp_pen3, exp_pen4, trm_pen34, exp_coa2;
- real dSBO1, dSBO2, SBO, SBO2, CSBO2, SBOp, prod_SBO;
- real CEval1, CEval2, CEval3, CEval4, CEval5, CEval6, CEval7, CEval8;
- real CEpen1, CEpen2, CEpen3;
- real e_ang, e_coa, e_pen;
- real CEcoa1, CEcoa2, CEcoa3, CEcoa4, CEcoa5;
- real Cf7ij, Cf7jk, Cf8j, Cf9j;
- real f7_ij, f7_jk, f8_Dj, f9_Dj;
- real Ctheta_0, theta_0, theta_00, theta, cos_theta, sin_theta;
- real r_ij, r_jk;
- real BOA_ij, BOA_jk;
- real vlpadj;
- rvec force, ext_press;
- // rtensor temp_rtensor, total_rtensor;
- real *total_bo;
- three_body_header *thbh;
- three_body_parameters *thbp;
- three_body_interaction_data *p_ijk, *p_kji;
- bond_data *pbond_ij, *pbond_jk, *pbond_jt;
- bond_order_data *bo_ij, *bo_jk, *bo_jt;
- list *bonds, *thb_intrs;
- bond_data *bond_list;
- three_body_interaction_data *thb_list;
- static_storage *workspace = &p_workspace;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
- if (j >= N) return;
-
-
- total_bo = workspace->total_bond_order;
- bonds = &p_bonds;
- bond_list = bonds->select.bond_list;
- thb_intrs = &p_thb_intrs;
- thb_list = thb_intrs->select.three_body_list;
-
- /* global parameters used in these calculations */
- p_val6 = g_params.l[14];
- p_val8 = g_params.l[33];
- p_val9 = g_params.l[16];
- p_val10 = g_params.l[17];
-
- //TODO check this, initially this was zero,
- // I am changing it to the starting index for this atom.
- //num_thb_intrs = j * MAX_TH_BODY;
-
- //for( j = 0; j < system->N; ++j ) {
- // fprintf( out_control->eval, "j: %d\n", j );
- type_j = atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
-
- p_val3 = sbp[ type_j ].p_val3;
- p_val5 = sbp[ type_j ].p_val5;
-
- SBOp = 0, prod_SBO = 1;
- for( t = start_j; t < end_j; ++t ) {
- bo_jt = &(bond_list[t].bo_data);
- SBOp += (bo_jt->BO_pi + bo_jt->BO_pi2);
- temp = SQR( bo_jt->BO );
- temp *= temp;
- temp *= temp;
- prod_SBO *= EXP( -temp );
- }
-
- /* modifications to match Adri's code - 09/01/09 */
- if( workspace->vlpex[j] >= 0 ){
- vlpadj = 0;
- dSBO2 = prod_SBO - 1;
- }
- else{
- vlpadj = workspace->nlp[j];
- dSBO2 = (prod_SBO - 1) * (1 - p_val8 * workspace->dDelta_lp[j]);
- }
-
- SBO = SBOp + (1 - prod_SBO) * (-workspace->Delta_boc[j] - p_val8 * vlpadj);
- dSBO1 = -8 * prod_SBO * ( workspace->Delta_boc[j] + p_val8 * vlpadj );
-
- if( SBO <= 0 )
- SBO2 = 0, CSBO2 = 0;
- else if( SBO > 0 && SBO <= 1 ) {
- SBO2 = POW( SBO, p_val9 );
- CSBO2 = p_val9 * POW( SBO, p_val9 - 1 );
- }
- else if( SBO > 1 && SBO < 2 ) {
- SBO2 = 2 - POW( 2-SBO, p_val9 );
- CSBO2 = p_val9 * POW( 2 - SBO, p_val9 - 1 );
- }
- else
- SBO2 = 2, CSBO2 = 0;
-
- expval6 = EXP( p_val6 * workspace->Delta_boc[j] );
-
- /* unlike 2-body intrs where we enforce i<j, we cannot put any such
- restrictions here. such a restriction would prevent us from producing
- all 4-body intrs correctly */
- for( pi = start_j; pi < end_j; ++pi ) {
-
- //TODO
- //num_thb_intrs = pi * MAX_THREE_BODIES;
- //TODO
-
- //Set_Start_Index( pi, num_thb_intrs, thb_intrs );
- num_thb_intrs = Start_Index (pi, thb_intrs);
-
- pbond_ij = &(bond_list[pi]);
- bo_ij = &(pbond_ij->bo_data);
- BOA_ij = bo_ij->BO - control->thb_cut;
-
-
- if( BOA_ij/*bo_ij->BO*/ > 0.0 ) {
- i = pbond_ij->nbr;
- r_ij = pbond_ij->d;
- type_i = atoms[i].type;
- // fprintf( out_control->eval, "i: %d\n", i );
-
-
- /* first copy 3-body intrs from previously computed ones where i>k.
-IMPORTANT: if it is less costly to compute theta and its
-derivative, we should definitely re-compute them,
-instead of copying!
-in the second for-loop below, we compute only new 3-body intrs
-where i < k */
- for( pk = start_j; pk < pi; ++pk ) {
- // fprintf( out_control->eval, "pk: %d\n", pk );
- start_pk = Start_Index( pk, thb_intrs );
- end_pk = End_Index( pk, thb_intrs );
-
- for( t = start_pk; t < end_pk; ++t )
- if( thb_list[t].thb == i ) {
- p_ijk = &(thb_list[num_thb_intrs]);
- p_kji = &(thb_list[t]);
-
- p_ijk->thb = bond_list[pk].nbr;
- p_ijk->pthb = pk;
- p_ijk->theta = p_kji->theta;
- rvec_Copy( p_ijk->dcos_di, p_kji->dcos_dk );
- rvec_Copy( p_ijk->dcos_dj, p_kji->dcos_dj );
- rvec_Copy( p_ijk->dcos_dk, p_kji->dcos_di );
-
- ++num_thb_intrs;
- break;
- }
- }
-
-
- /* and this is the second for loop mentioned above */
- for( pk = pi+1; pk < end_j; ++pk ) {
- pbond_jk = &(bond_list[pk]);
- bo_jk = &(pbond_jk->bo_data);
- BOA_jk = bo_jk->BO - control->thb_cut;
- k = pbond_jk->nbr;
- type_k = atoms[k].type;
- p_ijk = &( thb_list[num_thb_intrs] );
-
- //CHANGE ORIGINAL
- if (BOA_jk <= 0) continue;
- //CHANGE ORIGINAL
-
- Calculate_Theta( pbond_ij->dvec, pbond_ij->d,
- pbond_jk->dvec, pbond_jk->d,
- &theta, &cos_theta );
-
- Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d,
- pbond_jk->dvec, pbond_jk->d,
- &(p_ijk->dcos_di), &(p_ijk->dcos_dj),
- &(p_ijk->dcos_dk) );
-
- p_ijk->thb = k;
- p_ijk->pthb = pk;
- p_ijk->theta = theta;
-
- sin_theta = SIN( theta );
- if( sin_theta < 1.0e-5 )
- sin_theta = 1.0e-5;
-
- ++num_thb_intrs;
-
-
- if( BOA_jk > 0.0 &&
- (bo_ij->BO * bo_jk->BO) > SQR(control->thb_cut)/*0*/) {
- r_jk = pbond_jk->d;
- thbh = &( d_thbp[ index_thbp (type_i,type_j,type_k,num_atom_types) ] );
- flag = 0;
-
- /* if( workspace->orig_id[i] < workspace->orig_id[k] )
- fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], bo_ij->BO, bo_jk->BO, p_ijk->theta );
- else
- fprintf( stdout, "%6d %6d %6d %7.3f %7.3f %7.3f\n",
- workspace->orig_id[k], workspace->orig_id[j],
- workspace->orig_id[i], bo_jk->BO, bo_ij->BO, p_ijk->theta ); */
-
- //TODO:
- //pbond_jk->scratch = thbh->cnt;
-
- for( cnt = 0; cnt < thbh->cnt; ++cnt ) {
- // fprintf( out_control->eval,
- // "%6d%6d%6d -- exists in thbp\n", i+1, j+1, k+1 );
-
- if( fabs(thbh->prm[cnt].p_val1) > 0.001 ) {
- thbp = &( thbh->prm[cnt] );
-
- /* ANGLE ENERGY */
- p_val1 = thbp->p_val1;
- p_val2 = thbp->p_val2;
- p_val4 = thbp->p_val4;
- p_val7 = thbp->p_val7;
- theta_00 = thbp->theta_00;
-
- exp3ij = EXP( -p_val3 * POW( BOA_ij, p_val4 ) );
- f7_ij = 1.0 - exp3ij;
- Cf7ij = p_val3 * p_val4 *
- POW( BOA_ij, p_val4 - 1.0 ) * exp3ij;
-
- exp3jk = EXP( -p_val3 * POW( BOA_jk, p_val4 ) );
- f7_jk = 1.0 - exp3jk;
- Cf7jk = p_val3 * p_val4 *
- POW( BOA_jk, p_val4 - 1.0 ) * exp3jk;
-
- expval7 = EXP( -p_val7 * workspace->Delta_boc[j] );
- trm8 = 1.0 + expval6 + expval7;
- f8_Dj = p_val5 - ( (p_val5 - 1.0) * (2.0 + expval6) / trm8 );
- Cf8j = ( (1.0 - p_val5) / SQR(trm8) ) *
- (p_val6 * expval6 * trm8 -
- (2.0 + expval6) * ( p_val6 * expval6 - p_val7 * expval7 ));
-
- theta_0 = 180.0 -
- theta_00 * (1.0 - EXP(-p_val10 * (2.0 - SBO2)));
- theta_0 = DEG2RAD( theta_0 );
-
- expval2theta = EXP(-p_val2 * SQR(theta_0-theta));
- if( p_val1 >= 0 )
- expval12theta = p_val1 * (1.0 - expval2theta);
- else // To avoid linear Me-H-Me angles (6/6/06)
- expval12theta = p_val1 * -expval2theta;
-
- CEval1 = Cf7ij * f7_jk * f8_Dj * expval12theta;
- CEval2 = Cf7jk * f7_ij * f8_Dj * expval12theta;
- CEval3 = Cf8j * f7_ij * f7_jk * expval12theta;
- CEval4 = -2.0 * p_val1 * p_val2 * f7_ij * f7_jk * f8_Dj *
- expval2theta * (theta_0 - theta);
-
- Ctheta_0 = p_val10 * DEG2RAD(theta_00) *
- exp( -p_val10 * (2.0 - SBO2) );
-
- CEval5 = -CEval4 * Ctheta_0 * CSBO2;
- CEval6 = CEval5 * dSBO1;
- CEval7 = CEval5 * dSBO2;
- CEval8 = -CEval4 / sin_theta;
-
- e_ang = f7_ij * f7_jk * f8_Dj * expval12theta;
- //PERFORMANCE IMPACT
- //atomicAdd (&data->E_Ang, e_ang);
- E_Ang [j] += e_ang;
- /* END ANGLE ENERGY*/
-
-
- /* PENALTY ENERGY */
- p_pen1 = thbp->p_pen1;
- p_pen2 = g_params.l[19];
- p_pen3 = g_params.l[20];
- p_pen4 = g_params.l[21];
-
- exp_pen2ij = EXP( -p_pen2 * SQR( BOA_ij - 2.0 ) );
- exp_pen2jk = EXP( -p_pen2 * SQR( BOA_jk - 2.0 ) );
- exp_pen3 = EXP( -p_pen3 * workspace->Delta[j] );
- exp_pen4 = EXP( p_pen4 * workspace->Delta[j] );
- trm_pen34 = 1.0 + exp_pen3 + exp_pen4;
- f9_Dj = ( 2.0 + exp_pen3 ) / trm_pen34;
- Cf9j = (-p_pen3 * exp_pen3 * trm_pen34 -
- (2.0 + exp_pen3) * ( -p_pen3 * exp_pen3 +
- p_pen4 * exp_pen4 )) /
- SQR( trm_pen34 );
-
- e_pen = p_pen1 * f9_Dj * exp_pen2ij * exp_pen2jk;
- //PERFORMANCE IMPACT
- //atomicAdd (&data->E_Pen, e_pen);
- E_Pen [j] += e_pen;
-
-
- CEpen1 = e_pen * Cf9j / f9_Dj;
- temp = -2.0 * p_pen2 * e_pen;
- CEpen2 = temp * (BOA_ij - 2.0);
- CEpen3 = temp * (BOA_jk - 2.0);
- /* END PENALTY ENERGY */
-
-
- /* COALITION ENERGY */
- p_coa1 = thbp->p_coa1;
- p_coa2 = g_params.l[2];
- p_coa3 = g_params.l[38];
- p_coa4 = g_params.l[30];
-
- exp_coa2 = EXP( p_coa2 * workspace->Delta_boc[j] );
- e_coa =
- p_coa1 / (1. + exp_coa2) *
- EXP( -p_coa3 * SQR(total_bo[i] - BOA_ij) ) *
- EXP( -p_coa3 * SQR(total_bo[k] - BOA_jk) ) *
- EXP( -p_coa4 * SQR(BOA_ij - 1.5) ) *
- EXP( -p_coa4 * SQR(BOA_jk - 1.5) );
-
- //PERFORMANCE IMPACT
- //atomicAdd (&data->E_Coa, e_coa);
- E_Coa [j] += e_coa;
-
- CEcoa1 = -2 * p_coa4 * (BOA_ij - 1.5) * e_coa;
- CEcoa2 = -2 * p_coa4 * (BOA_jk - 1.5) * e_coa;
- CEcoa3 = -p_coa2 * exp_coa2 * e_coa / (1+exp_coa2);
- CEcoa4 = -2*p_coa3 * (total_bo[i]-BOA_ij) * e_coa;
- CEcoa5 = -2*p_coa3 * (total_bo[k]-BOA_jk) * e_coa;
- /* END COALITION ENERGY */
-
- /* FORCES */
- /*
- atomicAdd (&bo_ij->Cdbo, (CEval1 + CEpen2 + (CEcoa1-CEcoa4)) );
- atomicAdd (&bo_jk->Cdbo, (CEval2 + CEpen3 + (CEcoa2-CEcoa5)) );
- atomicAdd (&workspace->CdDelta[j], ((CEval3 + CEval7) + CEpen1 + CEcoa3) );
- atomicAdd (&workspace->CdDelta[i], CEcoa4 );
- atomicAdd (&workspace->CdDelta[k], CEcoa5 );
- */
-
- bo_ij->Cdbo += (CEval1 + CEpen2 + (CEcoa1-CEcoa4)) ;
- bo_jk->Cdbo += (CEval2 + CEpen3 + (CEcoa2-CEcoa5)) ;
- workspace->CdDelta[j] += ((CEval3 + CEval7) + CEpen1 + CEcoa3) ;
- //atomicAdd (&workspace->CdDelta[i], CEcoa4 );
- pbond_ij->CdDelta_ij += CEcoa4 ;
- //atomicAdd (&workspace->CdDelta[k], CEcoa5 );
- pbond_jk->CdDelta_ij += CEcoa5;
-
- for( t = start_j; t < end_j; ++t ) {
- pbond_jt = &( bond_list[t] );
- bo_jt = &(pbond_jt->bo_data);
- temp_bo_jt = bo_jt->BO;
- temp = CUBE( temp_bo_jt );
- pBOjt7 = temp * temp * temp_bo_jt;
-
- // fprintf( out_control->eval, "%6d%12.8f\n",
- // workspace->orig_id[ bond_list[t].nbr ],
- // (CEval6 * pBOjt7) );
-
- /*
- atomicAdd (&bo_jt->Cdbo, (CEval6 * pBOjt7) );
- atomicAdd (&bo_jt->Cdbopi, CEval5 );
- atomicAdd (&bo_jt->Cdbopi2, CEval5 );
- */
- bo_jt->Cdbo += (CEval6 * pBOjt7) ;
- bo_jt->Cdbopi += CEval5 ;
- bo_jt->Cdbopi2 += CEval5 ;
- }
-
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- /*
- atomic_rvecScaledAdd( atoms[i].f, CEval8, p_ijk->dcos_di );
- atomic_rvecScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
- atomic_rvecScaledAdd( atoms[k].f, CEval8, p_ijk->dcos_dk );
- */
- rvec_ScaledAdd( pbond_ij->f, CEval8, p_ijk->dcos_di );
- rvec_ScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
- rvec_ScaledAdd( pbond_jk->f, CEval8, p_ijk->dcos_dk );
-
-
- }
- else {
- /* terms not related to bond order derivatives
- are added directly into
- forces and pressure vector/tensor */
- rvec_Scale( force, CEval8, p_ijk->dcos_di );
- //atomic_rvecAdd( atoms[i].f, force );
- rvec_Add( pbond_ij->f, force );
-
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- //atomic_rvecAdd( data->ext_press, ext_press );
- rvec_Add( aux_ext_press [j], ext_press );
-
- //atomic_rvecScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
- rvec_ScaledAdd( atoms[j].f, CEval8, p_ijk->dcos_dj );
-
- rvec_Scale( force, CEval8, p_ijk->dcos_dk );
- //atomic_rvecAdd( atoms[k].f, force );
- rvec_Add( pbond_jk->f, force );
- rvec_iMultiply( ext_press, pbond_jk->rel_box, force );
- //atomic_rvecAdd( data->ext_press, ext_press );
- rvec_Add( aux_ext_press [j], ext_press );
-
-
- /* This part is for a fully-flexible box */
- /* rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_di, system->atoms[i].x );
- rtensor_Scale( total_rtensor, +CEval8, temp_rtensor );
-
- rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_dj, system->atoms[j].x );
- rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
-
- rvec_OuterProduct( temp_rtensor,
- p_ijk->dcos_dk, system->atoms[k].x );
- rtensor_ScaledAdd(total_rtensor, CEval8, temp_rtensor);
-
- if( pbond_ij->imaginary || pbond_jk->imaginary )
- rtensor_ScaledAdd( data->flex_bar.P,
- -1.0, total_rtensor );
- else
- rtensor_Add( data->flex_bar.P, total_rtensor ); */
- }
-
-#ifdef TEST_ENERGY
- //TODO -- check this
- // fprintf( out_control->eval,
- //"%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e",
- // "%6d%6d%6d%23.15e%23.15e%23.15e\n",
- // i+1, j+1, k+1,
- //workspace->orig_id[i]+1,
- //workspace->orig_id[j]+1,
- //workspace->orig_id[k]+1,
- //workspace->Delta_boc[j],
- // RAD2DEG(theta), /*BOA_ij, BOA_jk, */
- // e_ang, data->E_Ang );
-
- /*fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e",
- p_val3, p_val4, BOA_ij, BOA_jk );
- fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e",
- f7_ij, f7_jk, f8_Dj, expval12theta );
- fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- CEval1, CEval2, CEval3, CEval4, CEval5
- //CEval6, CEval7, CEval8 );*/
-
- /*fprintf( out_control->eval,
- "%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- -p_ijk->dcos_di[0]/sin_theta,
- -p_ijk->dcos_di[1]/sin_theta,
- -p_ijk->dcos_di[2]/sin_theta,
- -p_ijk->dcos_dj[0]/sin_theta,
- -p_ijk->dcos_dj[1]/sin_theta,
- -p_ijk->dcos_dj[2]/sin_theta,
- -p_ijk->dcos_dk[0]/sin_theta,
- -p_ijk->dcos_dk[1]/sin_theta,
- -p_ijk->dcos_dk[2]/sin_theta );*/
-
- /* fprintf( out_control->epen,
- "%23.15e%23.15e%23.15e\n",
- CEpen1, CEpen2, CEpen3 );
- fprintf( out_control->epen,
- "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- workspace->orig_id[k], RAD2DEG(theta),
- BOA_ij, BOA_jk, e_pen, data->E_Pen ); */
-
- // fprintf( out_control->ecoa,
- // "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- // workspace->orig_id[i],
- // workspace->orig_id[j],
- // workspace->orig_id[k],
- // RAD2DEG(theta), BOA_ij, BOA_jk,
- // e_coa, data->E_Coa );
-#endif
-
-#ifdef TEST_FORCES /* angle forces */
- //TODO -- check this
- /*
- Add_dBO( system, lists, j, pi, CEval1, workspace->f_ang );
- Add_dBO( system, lists, j, pk, CEval2, workspace->f_ang );
- Add_dDelta( system, lists,
- j, CEval3 + CEval7, workspace->f_ang );
-
- for( t = start_j; t < end_j; ++t ) {
- pbond_jt = &( bond_list[t] );
- bo_jt = &(pbond_jt->bo_data);
- temp_bo_jt = bo_jt->BO;
- temp = CUBE( temp_bo_jt );
- pBOjt7 = temp * temp * temp_bo_jt;
-
- Add_dBO( system, lists, j, t, pBOjt7 * CEval6,
- workspace->f_ang );
- Add_dBOpinpi2( system, lists, j, t,
- CEval5, CEval5,
- workspace->f_ang, workspace->f_ang );
- }
-
- rvec_ScaledAdd( workspace->f_ang[i], CEval8, p_ijk->dcos_di );
- rvec_ScaledAdd( workspace->f_ang[j], CEval8, p_ijk->dcos_dj );
- rvec_ScaledAdd( workspace->f_ang[k], CEval8, p_ijk->dcos_dk );
- // end angle forces
-
- // penalty forces
- Add_dDelta( system, lists, j, CEpen1, workspace->f_pen );
- Add_dBO( system, lists, j, pi, CEpen2, workspace->f_pen );
- Add_dBO( system, lists, j, pk, CEpen3, workspace->f_pen );
- // end penalty forces
-
- // coalition forces
- Add_dBO( system, lists,
- j, pi, CEcoa1-CEcoa4, workspace->f_coa );
- Add_dBO( system, lists,
- j, pk, CEcoa2-CEcoa5, workspace->f_coa );
- Add_dDelta( system, lists, j, CEcoa3, workspace->f_coa );
- Add_dDelta( system, lists, i, CEcoa4, workspace->f_coa );
- Add_dDelta( system, lists, k, CEcoa5, workspace->f_coa );
- // end coalition forces
-
- */
-#endif
- }
- }
- }
- }
- }
-
- Set_End_Index(pi, num_thb_intrs, thb_intrs );
- }
- // } // end of the main for loop here
-
-
- //TODO - to be done on the CPU
- /*
-
- if( num_thb_intrs >= thb_intrs->num_intrs * DANGER_ZONE ) {
- workspace->realloc.num_3body = num_thb_intrs;
- if( num_thb_intrs > thb_intrs->num_intrs ) {
- fprintf( stderr, "step%d-ran out of space on angle_list: top=%d, max=%d",
- data->step, num_thb_intrs, thb_intrs->num_intrs );
- exit( INSUFFICIENT_SPACE );
- }
- }
- */
-
- //fprintf( stderr,"%d: Number of angle interactions: %d\n",
- // data->step, num_thb_intrs );
-
-#ifdef TEST_ENERGY
- /*
- fprintf( stderr,"Number of angle interactions: %d\n", num_thb_intrs );
-
- fprintf( stderr,"Angle Energy:%g\t Penalty Energy:%g\t Coalition Energy:%g\n",
- data->E_Ang, data->E_Pen, data->E_Coa );
-
- fprintf( stderr,"3body: ext_press (%23.15e %23.15e %23.15e)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] );
- */
-#endif
-}
-
-
-GLOBAL void Three_Body_Interactions_results ( reax_atom *atoms, control_params *control,
- static_storage p_workspace,
- list p_bonds, int N )
-{
- int i, pj;
-
- bond_data *pbond;
- bond_data *sym_index_bond;
- list *bonds = &p_bonds;
- static_storage *workspace = &p_workspace;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( i >= N) return;
-
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
-
- pbond = &(bonds->select.bond_list[pj]);
- sym_index_bond = &( bonds->select.bond_list[ pbond->sym_index ] );
-
- workspace->CdDelta [i] += sym_index_bond->CdDelta_ij;
-
- rvec_Add (atoms[i].f, sym_index_bond->f );
- }
-}
-
-
-
-//////////////////////////////////////////////////////////////////////////
-// Three Body Estimation
-//////////////////////////////////////////////////////////////////////////
-
-
-/* this is a 3-body interaction in which the main role is
- played by j which sits in the middle of the other two. */
-GLOBAL void Three_Body_Estimate ( reax_atom *atoms,
- control_params *control,
- list p_bonds, int N,
- int *count)
-{
- int i, j, pi, k, pk, t;
- int type_i, type_j, type_k;
- int start_j, end_j ;
- int flag, cnt, num_thb_intrs;
-
- real r_ij, r_jk;
- real BOA_ij, BOA_jk;
- list *bonds;
-
- bond_order_data *bo_ij, *bo_jk, *bo_jt;
- bond_data *bond_list;
- bond_data *pbond_ij, *pbond_jk, *pbond_jt;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
- if (j >= N) return;
-
- bonds = &p_bonds;
- bond_list = bonds->select.bond_list;
-
- type_j = atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
-
-
- for( pi = start_j; pi < end_j; ++pi ) {
-
- num_thb_intrs = 0;
- count [pi] = 0;
-
- pbond_ij = &(bond_list[pi]);
- bo_ij = &(pbond_ij->bo_data);
- BOA_ij = bo_ij->BO - control->thb_cut;
-
- if( BOA_ij/*bo_ij->BO*/ > 0.0 ) {
- i = pbond_ij->nbr;
- r_ij = pbond_ij->d;
- type_i = atoms[i].type;
-
- /*
- for( pk = start_j; pk < pi; ++pk ) {
- start_pk = Start_Index( pk, thb_intrs );
- end_pk = End_Index( pk, thb_intrs );
-
- for( t = start_pk; t < end_pk; ++t )
- if( thb_list[t].thb == i ) {
-
- ++num_thb_intrs;
- break;
- }
- }
- */
-
- /* and this is the second for loop mentioned above */
- for( pk = start_j; pk < end_j; ++pk ) {
- if (pk == pi) continue;
-
- pbond_jk = &(bond_list[pk]);
- bo_jk = &(pbond_jk->bo_data);
- BOA_jk = bo_jk->BO - control->thb_cut;
-
- if (BOA_jk <= 0) continue;
-
- ++num_thb_intrs;
- }
- }
-
- count [pi] = num_thb_intrs;
- }
-}
-
-
-
-//////////////////////////////////////////////////////////////////////
-//End here
-//////////////////////////////////////////////////////////////////////
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-void Hydrogen_Bonds( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, k, pi, pk, itr, top;
- int type_i, type_j, type_k;
- int start_j, end_j, hb_start_j, hb_end_j;
- int hblist[MAX_BONDS];
- int num_hb_intrs = 0;
- real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
- real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
- rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
- rvec dvec_jk, force, ext_press;
- ivec rel_jk;
- // rtensor temp_rtensor, total_rtensor;
- hbond_parameters *hbp;
- bond_order_data *bo_ij;
- bond_data *pbond_ij;
- far_neighbor_data *nbr_jk;
- list *bonds, *hbonds;
- bond_data *bond_list;
- hbond_data *hbond_list;
-
- bonds = (*lists) + BONDS;
- bond_list = bonds->select.bond_list;
-
- hbonds = (*lists) + HBONDS;
- hbond_list = hbonds->select.hbond_list;
-
- /* loops below discover the Hydrogen bonds between i-j-k triplets.
- here j is H atom and there has to be some bond between i and j.
- Hydrogen bond is between j and k.
- so in this function i->X, j->H, k->Z when we map
- variables onto the ones in the handout.*/
- for( j = 0; j < system->N; ++j )
- if( system->reaxprm.sbp[system->atoms[j].type].p_hbond==1 ) {// j must be H
- /*set j's variables */
- type_j = system->atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
- hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
- hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
-
- top = 0;
- for( pi = start_j; pi < end_j; ++pi ) {
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
- bo_ij = &(pbond_ij->bo_data);
- type_i = system->atoms[i].type;
-
- if( system->reaxprm.sbp[type_i].p_hbond == 2 &&
- bo_ij->BO >= HB_THRESHOLD )
- hblist[top++] = pi;
- }
-
- // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
- // j, top, hb_start_j, hb_end_j );
-
- for( pk = hb_start_j; pk < hb_end_j; ++pk ) {
- /* set k's varibles */
- k = hbond_list[pk].nbr;
- type_k = system->atoms[k].type;
- nbr_jk = hbond_list[pk].ptr;
- r_jk = nbr_jk->d;
- rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
-
- for( itr=0; itr < top; ++itr ) {
- pi = hblist[itr];
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
-
- if( i != k ) {
- bo_ij = &(pbond_ij->bo_data);
- type_i = system->atoms[i].type;
- r_ij = pbond_ij->d;
- hbp = &(system->reaxprm.hbp[ index_hbp(type_i, type_j, type_k, &system->reaxprm) ]);
- ++num_hb_intrs;
-
- Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &theta, &cos_theta );
- /* the derivative of cos(theta) */
- Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &dcos_theta_di, &dcos_theta_dj,
- &dcos_theta_dk );
-
- /* hydrogen bond energy*/
- sin_theta2 = SIN( theta/2.0 );
- sin_xhz4 = SQR(sin_theta2);
- sin_xhz4 *= sin_xhz4;
- cos_xhz1 = ( 1.0 - cos_theta );
- exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
- exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
- r_jk / hbp->r0_hb - 2.0 ) );
-
- data->E_HB += e_hb =
- hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
-
- CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
- CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
- CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
- 1.0 / hbp->r0_hb);
-
- /* hydrogen bond forces */
- bo_ij->Cdbo += CEhb1; // dbo term
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
- rvec_ScaledAdd( system->atoms[i].f,
- +CEhb2, dcos_theta_di ); //dcos terms
- rvec_ScaledAdd( system->atoms[j].f,
- +CEhb2, dcos_theta_dj );
-
-
-
-
- //TODO
- rvec_ScaledAdd( system->atoms[k].f,
- +CEhb2, dcos_theta_dk );
-
- //dr terms
- rvec_ScaledAdd( system->atoms[j].f, -CEhb3/r_jk, dvec_jk );
-
-
- //TODO
- rvec_ScaledAdd( system->atoms[k].f, +CEhb3/r_jk, dvec_jk );
- }
- else
- {
- /* for pressure coupling, terms that are not related
- to bond order derivatives are added directly into
- pressure vector/tensor */
- rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
- rvec_Add( system->atoms[i].f, force );
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
-
- rvec_ScaledAdd( system->atoms[j].f, +CEhb2, dcos_theta_dj );
-
- ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
- rvec_Scale( force, +CEhb2, dcos_theta_dk );
-
-
-
- //TODO
- rvec_Add( system->atoms[k].f, force );
-
-
-
- rvec_iMultiply( ext_press, rel_jk, force );
- rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
-
- //dr terms
- rvec_ScaledAdd( system->atoms[j].f, -CEhb3/r_jk, dvec_jk );
-
- rvec_Scale( force, CEhb3/r_jk, dvec_jk );
- rvec_Add( system->atoms[k].f, force );
- rvec_iMultiply( ext_press, rel_jk, force );
- rvec_ScaledAdd( data->ext_press, 1.0, ext_press );
-
- /* This part is intended for a fully-flexible box */
- /* rvec_OuterProduct( temp_rtensor,
- dcos_theta_di, system->atoms[i].x );
- rtensor_Scale( total_rtensor, -CEhb2, temp_rtensor );
-
- rvec_ScaledSum( temp_rvec, -CEhb2, dcos_theta_dj,
- -CEhb3/r_jk, pbond_jk->dvec );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[j].x );
- rtensor_Add( total_rtensor, temp_rtensor );
-
- rvec_ScaledSum( temp_rvec, -CEhb2, dcos_theta_dk,
- +CEhb3/r_jk, pbond_jk->dvec );
- rvec_OuterProduct( temp_rtensor,
- temp_rvec, system->atoms[k].x );
- rtensor_Add( total_rtensor, temp_rtensor );
-
- if( pbond_ij->imaginary || pbond_jk->imaginary )
- rtensor_ScaledAdd( data->flex_bar.P, -1.0, total_rtensor );
- else
- rtensor_Add( data->flex_bar.P, total_rtensor ); */
- }
-
-#ifdef TEST_ENERGY
- /*fprintf( out_control->ehb,
- "%23.15e%23.15e%23.15e\n%23.15e%23.15e%23.15e\n%23.15e%23.15e%23.15e\n",
- dcos_theta_di[0], dcos_theta_di[1], dcos_theta_di[2],
- dcos_theta_dj[0], dcos_theta_dj[1], dcos_theta_dj[2],
- dcos_theta_dk[0], dcos_theta_dk[1], dcos_theta_dk[2]);
- fprintf( out_control->ehb, "%23.15e%23.15e%23.15e\n",
- CEhb1, CEhb2, CEhb3 ); */
- fprintf( stderr, //out_control->ehb,
- "%6d%6d%6d%23.15e%23.15e%23.15e%23.15e%23.15e\n",
- workspace->orig_id[i],
- workspace->orig_id[j],
- workspace->orig_id[k],
- r_jk, theta, bo_ij->BO, e_hb, data->E_HB );
-
-#endif
-#ifdef TEST_FORCES
- // dbo term
- Add_dBO( system, lists, j, pi, +CEhb1, workspace->f_hb );
- // dcos terms
- rvec_ScaledAdd( workspace->f_hb[i], +CEhb2, dcos_theta_di );
- rvec_ScaledAdd( workspace->f_hb[j], +CEhb2, dcos_theta_dj );
- rvec_ScaledAdd( workspace->f_hb[k], +CEhb2, dcos_theta_dk );
- // dr terms
- rvec_ScaledAdd( workspace->f_hb[j], -CEhb3/r_jk, dvec_jk );
- rvec_ScaledAdd( workspace->f_hb[k], +CEhb3/r_jk, dvec_jk );
-#endif
- }
- }
- }
- }
-
- /* fprintf( stderr, "hydbonds: ext_press (%23.15e %23.15e %23.15e)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] ); */
-
-#ifdef TEST_FORCES
- fprintf( stderr, "Number of hydrogen bonds: %d\n", num_hb_intrs );
- fprintf( stderr, "Hydrogen Bond Energy: %g\n", data->E_HB );
-#endif
-}
-
-
-
-
-
-
-////////////////////////////////////////////////////////////////////
-// Cuda Function
-////////////////////////////////////////////////////////////////////
-
-GLOBAL void Hydrogen_Bonds ( reax_atom *atoms,
- single_body_parameters *sbp,
- hbond_parameters *d_hbp,
- control_params *control,
- simulation_data *data,
- static_storage p_workspace,
- list p_bonds, list p_hbonds,
- int N, int num_atom_types,
- real *E_HB, rvec *aux_ext_press, rvec *atoms_f )
-{
- extern __shared__ real t_hb[];
- extern __shared__ real t_f[];
- //extern __shared__ rvec t_cdbo[];
- //extern __shared__ rvec t_hf [];
-
- real *sh_hb = t_hb;
- rvec *sh_atomf = (rvec *)(t_hb + blockDim.x);
- //real *sh_cdbo = t_hb + blockDim.x;
- //rvec *sh_hf = (rvec *) (sh_atomf + blockDim.x);
-
- int i, j, k, pi, pk, itr, top;
- int type_i, type_j, type_k;
- int start_j, end_j, hb_start_j, hb_end_j;
- int hblist[MAX_BONDS];
- int num_hb_intrs = 0;
- real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
- real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
- rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
- rvec dvec_jk, force, ext_press;
- ivec rel_jk;
- // rtensor temp_rtensor, total_rtensor;
- hbond_parameters *hbp;
- bond_order_data *bo_ij;
- bond_data *pbond_ij;
- far_neighbor_data *nbr_jk;
- list *bonds, *hbonds;
- bond_data *bond_list;
- hbond_data *hbond_list, *hbond_jk;
- static_storage *workspace = &p_workspace;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
- if (j >= N) return;
-
- //j = blockIdx.x;
-
- bonds = &p_bonds;
- bond_list = bonds->select.bond_list;
-
- hbonds = &p_hbonds;
- hbond_list = hbonds->select.hbond_list;
-
- // loops below discover the Hydrogen bonds between i-j-k triplets.
- // here j is H atom and there has to be some bond between i and j.
- // Hydrogen bond is between j and k.
- // so in this function i->X, j->H, k->Z when we map
- // variables onto the ones in the handout.
-
- //for( j = 0; j < system->N; ++j )
- sh_hb [threadIdx.x] = 0;
- rvec_MakeZero ( sh_atomf[ threadIdx.x] );
-
- if( sbp[atoms[j].type].p_hbond==1) {// j must be H
- //set j's variables
- type_j = atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
- hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
- hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
-
- top = 0;
- for( pi = start_j; pi < end_j; ++pi ) {
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
- bo_ij = &(pbond_ij->bo_data);
- type_i = atoms[i].type;
-
- if( sbp[type_i].p_hbond == 2 &&
- bo_ij->BO >= HB_THRESHOLD )
- hblist[top++] = pi;
- }
-
- // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
- // j, top, hb_start_j, hb_end_j );
-
- for( pk = hb_start_j; pk < hb_end_j; ++pk )
- //pk = hb_start_j + threadIdx.x;
- //while (pk < hb_end_j)
- {
- // set k's varibles
- //TODO
- hbond_jk = &( hbond_list[pk] );
- //TODO
- k = hbond_list[pk].nbr;
- type_k = atoms[k].type;
- nbr_jk = hbond_list[pk].ptr;
- r_jk = nbr_jk->d;
- rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
-
- //TODO Double check this Hydrogen Bonds fix
- //rvec_MakeZero ( nbr_jk->h_f );
- rvec_MakeZero ( hbond_jk->h_f );
- //TODO Double check this Hydrogen Bonds fix
-
- //sh_hb [threadIdx.x] = 0;
-
-
- //itr = threadIdx.x;
- for( itr=0; itr < top; ++itr ) {
- //while (itr < top) {
- pi = hblist[itr];
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
-
- //TODO
- //rvec_MakeZero (sh_hf [threadIdx.x]);
- //sh_cdbo [threadIdx.x] = 0;
-
- //rvec_MakeZero ( sh_atomf[ threadIdx.x] );
-
-
- if( i != k ) {
- bo_ij = &(pbond_ij->bo_data);
- type_i = atoms[i].type;
- r_ij = pbond_ij->d;
- hbp = &(d_hbp[ index_hbp(type_i, type_j, type_k, num_atom_types) ]);
- ++num_hb_intrs;
-
- Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &theta, &cos_theta );
- // the derivative of cos(theta)
- Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &dcos_theta_di, &dcos_theta_dj,
- &dcos_theta_dk );
-
- // hydrogen bond energy
- sin_theta2 = SIN( theta/2.0 );
- sin_xhz4 = SQR(sin_theta2);
- sin_xhz4 *= sin_xhz4;
- cos_xhz1 = ( 1.0 - cos_theta );
- exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
- exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
- r_jk / hbp->r0_hb - 2.0 ) );
-
- //PERFORMANCE IMPACT
- e_hb = hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
- //atomicAdd ( &data->E_HB, e_hb );
- //E_HB [j] += e_hb;
- sh_hb [threadIdx.x] += e_hb;
-
- CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
- CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
- CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
- 1.0 / hbp->r0_hb);
-
- //this is the problem here
- //TODO
- // hydrogen bond forces
- bo_ij->Cdbo += CEhb1; // dbo term
- //sh_cdbo[threadIdx.x] += CEhb1;
- //TODO
-
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
-
- //PERFORMANCE IMPACT
- /*
- atomic_rvecScaledAdd( atoms[i].f,
- +CEhb2, dcos_theta_di ); //dcos terms
- atomic_rvecScaledAdd( atoms[j].f,
- +CEhb2, dcos_theta_dj );
- atomic_rvecScaledAdd( atoms[k].f,
- +CEhb2, dcos_theta_dk );
- //dr terms
- atomic_rvecScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
- atomic_rvecScaledAdd( atoms[k].f, +CEhb3/r_jk, dvec_jk );
- */
-
- //PERFORMANCE IMPACT
- rvec_ScaledAdd( pbond_ij->h_f, +CEhb2, dcos_theta_di ); //dcos terms
- //rvec_ScaledAdd( sh_hf [threadIdx.x], +CEhb2, dcos_theta_di ); //dcos terms
-
- //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
- rvec_ScaledAdd( sh_atomf [threadIdx.x], +CEhb2, dcos_theta_dj );
-
- //TODO you forgot here
- //TODO Hydrogen bonds fix. -- BE VERY CAREFUL *****
- rvec_ScaledAdd( hbond_jk->h_f,
- +CEhb2, dcos_theta_dk );
-
- //rvec_ScaledAdd( nbr_jk->h_f,
- // +CEhb2, dcos_theta_dk );
-
- //dr terms
- //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
- rvec_ScaledAdd( sh_atomf [threadIdx.x], -CEhb3/r_jk, dvec_jk );
-
- //atoms_f [j] ++;
-
- //TODO you forgot
- rvec_ScaledAdd( hbond_jk->h_f, +CEhb3/r_jk, dvec_jk );
- //rvec_ScaledAdd( nbr_jk->h_f, +CEhb3/r_jk, dvec_jk );
- }
- else
- {
- // for pressure coupling, terms that are not related
- // to bond order derivatives are added directly into
- // pressure vector/tensor
- rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
- rvec_Add( pbond_ij->h_f, force );
- rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd (sh_press [threadIdx.x], 1.0, ext_press );
-
- rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
-
- ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
- rvec_Scale( force, +CEhb2, dcos_theta_dk );
-
- //rvec_Add( nbr_jk->h_f, force );
- rvec_Add( hbond_jk->h_f, force );
-
- rvec_iMultiply( ext_press, rel_jk, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
-
- //dr terms
- rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
-
- rvec_Scale( force, CEhb3/r_jk, dvec_jk );
- rvec_Add( hbond_jk->h_f, force );
- rvec_iMultiply( ext_press, rel_jk, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
-
- }
-
- //do the reduction for the bond_ij here
- /*
- if (threadIdx.x < 16){
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 16];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 16]);
-
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
- }
- if (threadIdx.x < 8){
- //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 8];
- //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 8]);
-
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
- }
- if (threadIdx.x < 4){
- //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 4];
- //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 4]);
-
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
- }
- if (threadIdx.x < 2){
- //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 2];
- //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 2]);
-
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
- }
- if (threadIdx.x < 1){
- //sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 1];
- //rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 1]);
-
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
- }
- if (threadIdx.x == 0){
- //bo_ij->Cdbo += sh_cdbo [threadIdx.x];
- //rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
-
- E_HB [j] += sh_hb [threadIdx.x];
- //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
- }
- */
-
-
- } // i != k if statement
-
-
- //itr += blockDim.x;
-
- } //itr for statement
-
- /*
- __syncthreads ();
-
- for (int x = 1; x < blockDim.x; x++)
- sh_hb [0] += sh_hb [x];
-
- E_HB [j] += sh_hb[0];
- if (threadIdx.x < 16) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
- if (threadIdx.x < 8) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
- if (threadIdx.x < 4) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
- if (threadIdx.x < 2) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
- if (threadIdx.x < 1) sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
- if (threadIdx.x == 0) E_HB [j] += sh_hb [threadIdx.x];
- */
-
-
- //pk += blockDim.x;
-
- } // pk for statement
- } // main if statment
-
- //do the reduction for the bond_ij here
- /*
- if (threadIdx.x < 16){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
- }
- if (threadIdx.x < 8){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
- }
- if (threadIdx.x < 4){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
- }
- if (threadIdx.x < 2){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
- }
- if (threadIdx.x < 1){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
- //rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
- }
- if (threadIdx.x == 0){
- E_HB [j] += sh_hb [threadIdx.x];
- //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
- }
- */
-
- E_HB [j] += sh_hb [threadIdx.x];
- rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
-
- //rvec_Copy (atoms_f [j], sh_atomf [threadIdx.x]);
- }
-
-
- DEVICE void warpReduce(volatile real* sdata, int tid)
- {
- if (tid < 16) sdata[tid] += sdata[tid + 16];
- if (tid < 8) sdata[tid] += sdata[tid + 8];
- if (tid < 4) sdata[tid] += sdata[tid + 4];
- if (tid < 2) sdata[tid] += sdata[tid + 2];
- if (tid < 1) sdata[tid] += sdata[tid + 1];
- }
-
-
-
-
- GLOBAL void Hydrogen_Bonds_HB ( reax_atom *atoms,
- single_body_parameters *sbp,
- hbond_parameters *d_hbp,
- control_params *control,
- simulation_data *data,
- static_storage p_workspace,
- list p_bonds, list p_hbonds,
- int N, int num_atom_types,
- real *E_HB, rvec *aux_ext_press, rvec *atoms_f )
- {
- extern __shared__ real t_hb[];
- extern __shared__ rvec t__f[];
- extern __shared__ rvec t_cdbo[];
- extern __shared__ rvec t_hf [];
-
- real *sh_hb = t_hb;
- real *sh_cdbo = t_hb + blockDim.x;
- rvec *sh_atomf = (rvec *)(sh_cdbo + blockDim.x);
- rvec *sh_hf = (rvec *) (sh_atomf + blockDim.x);
-
- int __THREADS_PER_ATOM__ = HBONDS_THREADS_PER_ATOM;
-
- int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
- int warp_id = thread_id / __THREADS_PER_ATOM__;
- int lane_id = thread_id & (__THREADS_PER_ATOM__ -1);
- int my_bucket = threadIdx.x / __THREADS_PER_ATOM__;
-
- if (warp_id >= N ) return;
-
-
- int i, j, k, pi, pk, itr, top;
- int type_i, type_j, type_k;
- int start_j, end_j, hb_start_j, hb_end_j;
- int hblist[MAX_BONDS];
- int num_hb_intrs = 0;
- real r_ij, r_jk, theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
- real e_hb, exp_hb2, exp_hb3, CEhb1, CEhb2, CEhb3;
- rvec dcos_theta_di, dcos_theta_dj, dcos_theta_dk;
- rvec dvec_jk, force, ext_press;
- ivec rel_jk;
- // rtensor temp_rtensor, total_rtensor;
- hbond_parameters *hbp;
- bond_order_data *bo_ij;
- bond_data *pbond_ij;
- far_neighbor_data *nbr_jk;
- list *bonds, *hbonds;
- bond_data *bond_list;
- hbond_data *hbond_list, *hbond_jk;
- static_storage *workspace = &p_workspace;
-
- /*
- j = blockIdx.x * blockDim.x + threadIdx.x;
- if (j >= N) return;
- */
-
- // j = blockIdx.x;
-
- j = warp_id;
-
- bonds = &p_bonds;
- bond_list = bonds->select.bond_list;
-
- hbonds = &p_hbonds;
- hbond_list = hbonds->select.hbond_list;
-
- // loops below discover the Hydrogen bonds between i-j-k triplets.
- // here j is H atom and there has to be some bond between i and j.
- // Hydrogen bond is between j and k.
- // so in this function i->X, j->H, k->Z when we map
- // variables onto the ones in the handout.
-
- //for( j = 0; j < system->N; ++j )
- sh_hb [threadIdx.x] = 0;
- rvec_MakeZero ( sh_atomf[ threadIdx.x] );
-
- if( sbp[atoms[j].type].p_hbond==1) {// j must be H
- //set j's variables
- type_j = atoms[j].type;
- start_j = Start_Index(j, bonds);
- end_j = End_Index(j, bonds);
- hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
- hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
-
- top = 0;
- for( pi = start_j; pi < end_j; ++pi ) {
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
- bo_ij = &(pbond_ij->bo_data);
- type_i = atoms[i].type;
-
- if( sbp[type_i].p_hbond == 2 &&
- bo_ij->BO >= HB_THRESHOLD ) {
- hblist[top++] = pi;
- }
- }
-
- // fprintf( stderr, "j: %d, top: %d, hb_start_j: %d, hb_end_j:%d\n",
- // j, top, hb_start_j, hb_end_j );
-
- for( itr=0; itr < top; ++itr ) {
- pi = hblist[itr];
- pbond_ij = &( bond_list[pi] );
- i = pbond_ij->nbr;
-
- //TODO
- rvec_MakeZero (sh_hf [threadIdx.x]);
- sh_cdbo [threadIdx.x] = 0;
-
-
- //for( pk = hb_start_j; pk < hb_end_j; ++pk )
- int loopcount = (hb_end_j - hb_start_j) / HBONDS_THREADS_PER_ATOM + (((hb_end_j - hb_start_j)%HBONDS_THREADS_PER_ATOM == 0) ? 0 : 1);
- int count = 0;
- //jpk = hb_start_j + threadIdx.x;
- pk = hb_start_j + lane_id;
- //while (pk < hb_end_j)
- while (count < loopcount)
- {
-
- if (pk < hb_end_j)
- {
- // set k's varibles
- //TODO
- hbond_jk = &( hbond_list[pk] );
- //TODO
- k = hbond_list[pk].nbr;
- type_k = atoms[k].type;
- nbr_jk = hbond_list[pk].ptr;
- r_jk = nbr_jk->d;
- rvec_Scale( dvec_jk, hbond_list[pk].scl, nbr_jk->dvec );
- }
- else k = -1;
-
- //TODO Double check this Hydrogen Bonds fix
- //rvec_MakeZero ( nbr_jk->h_f );
- //rvec_MakeZero ( hbond_jk->h_f );
- //TODO Double check this Hydrogen Bonds fix
-
- //sh_hb [threadIdx.x] = 0;
- //rvec_MakeZero ( sh_atomf[ threadIdx.x] );
- //__syncthreads ();
-
-
- if(( i != k ) && (k != -1)) {
- bo_ij = &(pbond_ij->bo_data);
- type_i = atoms[i].type;
- r_ij = pbond_ij->d;
- hbp = &(d_hbp[ index_hbp(type_i, type_j, type_k, num_atom_types) ]);
- ++num_hb_intrs;
-
- Calculate_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &theta, &cos_theta );
- // the derivative of cos(theta)
- Calculate_dCos_Theta( pbond_ij->dvec, pbond_ij->d, dvec_jk, r_jk,
- &dcos_theta_di, &dcos_theta_dj,
- &dcos_theta_dk );
-
- // hydrogen bond energy
- sin_theta2 = SIN( theta/2.0 );
- sin_xhz4 = SQR(sin_theta2);
- sin_xhz4 *= sin_xhz4;
- cos_xhz1 = ( 1.0 - cos_theta );
- exp_hb2 = EXP( -hbp->p_hb2 * bo_ij->BO );
- exp_hb3 = EXP( -hbp->p_hb3 * ( hbp->r0_hb / r_jk +
- r_jk / hbp->r0_hb - 2.0 ) );
-
- //PERFORMANCE IMPACT
- e_hb = hbp->p_hb1 * (1.0 - exp_hb2) * exp_hb3 * sin_xhz4;
- //atomicAdd ( &data->E_HB, e_hb );
- //E_HB [j] += e_hb;
- sh_hb [threadIdx.x] += e_hb;
-
- CEhb1 = hbp->p_hb1*hbp->p_hb2 * exp_hb2*exp_hb3 * sin_xhz4;
- CEhb2 = -hbp->p_hb1/2.0*(1.0 - exp_hb2) * exp_hb3 * cos_xhz1;
- CEhb3 = -hbp->p_hb3 * e_hb * (-hbp->r0_hb / SQR(r_jk) +
- 1.0 / hbp->r0_hb);
-
- //this is the problem here
- //TODO
- // hydrogen bond forces
- //bo_ij->Cdbo += CEhb1; // dbo term
- sh_cdbo[threadIdx.x] += CEhb1;
- //TODO
- //warpReduce (sh_cdbo, threadIdx.x);
- //if (threadIdx.x == 0)
- // bo_ij->Cdbo += sh_cdbo [0];
-
-
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
-
- //PERFORMANCE IMPACT
- /*
- atomic_rvecScaledAdd( atoms[i].f,
- +CEhb2, dcos_theta_di ); //dcos terms
- atomic_rvecScaledAdd( atoms[j].f,
- +CEhb2, dcos_theta_dj );
- atomic_rvecScaledAdd( atoms[k].f,
- +CEhb2, dcos_theta_dk );
- //dr terms
- atomic_rvecScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
- atomic_rvecScaledAdd( atoms[k].f, +CEhb3/r_jk, dvec_jk );
- */
-
- //PERFORMANCE IMPACT
- //rvec_ScaledAdd( pbond_ij->h_f, +CEhb2, dcos_theta_di ); //dcos terms
- rvec_ScaledAdd( sh_hf [threadIdx.x], +CEhb2, dcos_theta_di ); //dcos terms
-
- //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
- rvec_ScaledAdd( sh_atomf [threadIdx.x], +CEhb2, dcos_theta_dj );
-
-
- //TODO you forgot here
- //TODO Hydrogen bonds fix. -- BE VERY CAREFUL *****
- rvec_ScaledAdd( hbond_jk->h_f, +CEhb2, dcos_theta_dk );
-
- //rvec_ScaledAdd( nbr_jk->h_f,
- // +CEhb2, dcos_theta_dk );
-
- //dr terms
- //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
- rvec_ScaledAdd( sh_atomf [threadIdx.x], -CEhb3/r_jk, dvec_jk );
-
- //TODO you forgot
- rvec_ScaledAdd( hbond_jk->h_f, +CEhb3/r_jk, dvec_jk );
- //rvec_ScaledAdd( nbr_jk->h_f, +CEhb3/r_jk, dvec_jk );
- }
- else
- {
- // for pressure coupling, terms that are not related
- // to bond order derivatives are added directly into
- // pressure vector/tensor
- //rvec_Scale( force, +CEhb2, dcos_theta_di ); // dcos terms
- //rvec_Add( pbond_ij->h_f, force );
- //rvec_iMultiply( ext_press, pbond_ij->rel_box, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd (sh_press [threadIdx.x], 1.0, ext_press );
-
- //rvec_ScaledAdd( atoms[j].f, +CEhb2, dcos_theta_dj );
-
- //ivec_Scale( rel_jk, hbond_list[pk].scl, nbr_jk->rel_box );
- //rvec_Scale( force, +CEhb2, dcos_theta_dk );
-
- //rvec_Add( nbr_jk->h_f, force );
- //rvec_Add( hbond_jk->h_f, force );
-
- //rvec_iMultiply( ext_press, rel_jk, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
-
- //dr terms
- //rvec_ScaledAdd( atoms[j].f, -CEhb3/r_jk, dvec_jk );
-
- //rvec_Scale( force, CEhb3/r_jk, dvec_jk );
- //rvec_Add( hbond_jk->h_f, force );
- //rvec_iMultiply( ext_press, rel_jk, force );
- //rvec_ScaledAdd( aux_ext_press [j], 1.0, ext_press );
- //rvec_ScaledAdd( sh_press [threadIdx.x], 1.0, ext_press );
-
- }
-
- } // i != k if statement
-
- pk += __THREADS_PER_ATOM__;
- count ++;
-
- } // pk for statement
-
- //__syncthreads ();
-
- //at this point done with one bond....
- //do the reduction now
- //if (threadIdx.x == 0){
- if (lane_id < 16) {
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 16];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 16]);
- }
- if (lane_id < 8) {
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 8];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 8]);
- }
- if (lane_id < 4) {
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 4];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 4]);
- }
- if (lane_id < 2) {
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 2];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 2]);
- }
- if (lane_id < 1) {
- sh_cdbo [threadIdx.x] += sh_cdbo [threadIdx.x + 1];
- rvec_Add (sh_hf [threadIdx.x], sh_hf [threadIdx.x + 1]);
-
- bo_ij->Cdbo += sh_cdbo [threadIdx.x];
- rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
- }
- /*
- if (lane_id == 0){
- for (i = 1; i < 32; i++)
- {
- //sh_cdbo [threadIdx.x] += sh_cdbo [i];
- //rvec_Add (sh_hf [threadIdx.x], sh_hf [i]);
-
- sh_cdbo [lane_id] += sh_cdbo [lane_id + i];
- rvec_Add (sh_hf [lane_id], sh_hf [lane_id + i]);
- }
-
- //bo_ij->Cdbo += sh_cdbo [threadIdx.x];
- //rvec_Add (pbond_ij->h_f, sh_hf [threadIdx.x]);
-
- bo_ij->Cdbo += sh_cdbo [lane_id];
- rvec_Add (pbond_ij->h_f, sh_hf [lane_id]);
- }
- */
-
- } //itr for statement
-
- //__syncthreads ();
- } // main if statment
-
- //__syncthreads ();
-
-
- //do the reduction for the bond_ij here
- if (lane_id < 16){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 16];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 16] );
- }
- if (lane_id < 8){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 8];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 8] );
- }
- if (lane_id < 4){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 4];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 4] );
- }
- if (lane_id < 2){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 2];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 2] );
- }
- if (lane_id < 1){
- sh_hb [threadIdx.x] += sh_hb [threadIdx.x + 1];
- rvec_Add ( sh_atomf [threadIdx.x], sh_atomf [threadIdx.x + 1] );
-
- E_HB [j] += sh_hb [threadIdx.x];
- rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
- }
- /*
- if (lane == 0){
- //E_HB [j] += sh_hb [threadIdx.x];
- rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
- rvec_Copy (atoms_f [j], sh_atomf [threadIdx.x]);
- }
- */
- //if (threadIdx.x == 0){
- /*
- if (lane_id == 0){
- for (i = 1; i < 32; i++)
- {
- //sh_hb [threadIdx.x] += sh_hb [i];
- //rvec_Add (sh_atomf [threadIdx.x], sh_atomf [i]);
- sh_hb [lane_id] += sh_hb [lane_id + i];
- rvec_Add (sh_atomf [lane_id], sh_atomf [lane_id + i]);
- }
-
- //E_HB [j] += sh_hb [threadIdx.x];
- //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
-
- E_HB [j] += sh_hb [lane_id];
- rvec_Add (atoms[j].f, sh_atomf [lane_id]);
- //rvec_Copy (atoms_f[j], sh_atomf [threadIdx.x]);
- }
- */
-
- //E_HB [j] += sh_hb [threadIdx.x];
- //rvec_Add (atoms[j].f, sh_atomf [threadIdx.x]);
- }
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
- GLOBAL void Hydrogen_Bonds_Postprocess ( reax_atom *atoms,
- single_body_parameters *sbp,
- static_storage p_workspace,
- list p_bonds, list p_hbonds, list p_far_nbrs, int N,
- real *e_hb)
- {
-
- int i, pj, hj, nbr, k, j;
- int start, end;
-
- bond_data *pbond;
- bond_data *sym_index_bond;
- far_neighbor_data *nbr_pj, *sym_index_nbr;
-
- list *bonds = &p_bonds;
- list *far_nbrs = &p_far_nbrs;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( i >= N) return;
-
- // For processing ij information
- start = Start_Index(i, bonds);
- end = End_Index(i, bonds);
-
- //rvec_Scale (atoms[i].f, e_hb[i], atoms[i].f);
-
- for( pj = Start_Index(i, bonds); pj < End_Index(i, bonds); ++pj ){
-
- pbond = &(bonds->select.bond_list[pj]);
- sym_index_bond = &( bonds->select.bond_list[ pbond->sym_index ] );
-
- rvec_Add (atoms[i].f, sym_index_bond->h_f );
- }
-
- /*
- for (pj = Start_Index (i, far_nbrs); pj < End_Index (i, far_nbrs); pj ++)
- {
- // check if the neighbor is of h_type
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
-
- sym_index_nbr = & (far_nbrs->select.far_nbr_list[ nbr_pj->sym_index ]);
- rvec_Add (atoms[i].f, sym_index_nbr->h_f );
- }
- */
-
- // if (workspace->hbond_index [j] != -1)
- // {
- // hb_start_j = Start_Index( workspace->hbond_index[j], hbonds );
- // hb_end_j = End_Index ( workspace->hbond_index[j], hbonds );
-
- // for ( hj = hb_start_j; hj < hb_end_j; hj ++ )
- // {
- // h_bond_data = &( hbonds->select.hbond_list [hj] );
- // nbr = h_bond_data->nbr;
-
- // if (nbr == i) {
- // rvec_Add (atoms[i].f, h_bond_data->h_f );
- // }
- // }
- // }
- }
-
- GLOBAL void Hydrogen_Bonds_Far_Nbrs ( reax_atom *atoms,
- single_body_parameters *sbp,
- static_storage p_workspace,
- list p_bonds, list p_hbonds, list p_far_nbrs, int N )
- {
-
- extern __shared__ rvec __f[];
- int i, pj,j;
- int start, end;
-
- far_neighbor_data *nbr_pj, *sym_index_nbr;
- list *far_nbrs = &p_far_nbrs;
-
- i = blockIdx.x;
-
- start = Start_Index (i, far_nbrs);
- end = End_Index (i, far_nbrs);
- pj = start + threadIdx.x;
-
- rvec_MakeZero (__f[threadIdx.x]);
-
- while (pj < end)
- {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
-
- //sym_index_nbr = & (far_nbrs->select.far_nbr_list[ nbr_pj->sym_index ]);
- //
- //rvec_Add (atoms[i].f, sym_index_nbr->h_f );
- //
- //rvec_Add (__f[threadIdx.x], sym_index_nbr->h_f );
-
- pj += blockDim.x;
- }
-
- if (threadIdx.x < 16) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 16]);
- if (threadIdx.x < 8) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 8]);
- if (threadIdx.x < 4) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 4]);
- if (threadIdx.x < 2) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 2]);
- if (threadIdx.x < 1) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 1]);
-
- if (threadIdx.x == 0)
- rvec_Add (atoms[i].f, __f[0]);
- }
-
- GLOBAL void Hydrogen_Bonds_HNbrs ( reax_atom *atoms,
- single_body_parameters *sbp,
- static_storage p_workspace,
- list p_bonds, list p_hbonds, list p_far_nbrs, int N )
- {
-
- extern __shared__ rvec __f[];
- int i, pj,j;
- int start, end;
-
- hbond_data *nbr_pj, *sym_index_nbr;
- list *hbonds = &p_hbonds;
-
- i = blockIdx.x;
-
- start = Start_Index (i, hbonds);
- end = End_Index (i, hbonds);
- pj = start + threadIdx.x;
-
- rvec_MakeZero (__f[threadIdx.x]);
-
- while (pj < end)
- {
- nbr_pj = &( hbonds->select.hbond_list[pj] );
- j = nbr_pj->nbr;
-
- sym_index_nbr = & (hbonds->select.hbond_list[ nbr_pj->sym_index ]);
- rvec_Add (__f[threadIdx.x], sym_index_nbr->h_f );
-
- pj += blockDim.x;
- }
-
- if (threadIdx.x < 16) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 16]);
- if (threadIdx.x < 8) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 8]);
- if (threadIdx.x < 4) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 4]);
- if (threadIdx.x < 2) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 2]);
- if (threadIdx.x < 1) rvec_Add (__f[threadIdx.x], __f[threadIdx.x + 1]);
-
- if (threadIdx.x == 0)
- rvec_Add (atoms[i].f, __f[0]);
- }
-
diff --git a/PuReMD-GPU/src/three_body_interactions.h b/PuReMD-GPU/src/three_body_interactions.h
index 2aa0d4434a7001793a5065a82e238b6d349f49d7..dcbadb0697f951b11b98d53488bd466a633a3fb6 100644
--- a/PuReMD-GPU/src/three_body_interactions.h
+++ b/PuReMD-GPU/src/three_body_interactions.h
@@ -23,52 +23,15 @@
#include "mytypes.h"
+
void Three_Body_Interactions( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
+ static_storage*, list**, output_controls* );
void Hydrogen_Bonds( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-
-
-//CUDA Functions.
-HOST_DEVICE void Calculate_Theta( rvec, real, rvec, real, real*, real* );
-
-HOST_DEVICE void Calculate_dCos_Theta( rvec, real, rvec, real, rvec*, rvec*, rvec* );
-
-GLOBAL void Three_Body_Interactions( reax_atom *, single_body_parameters *, three_body_header *,
- global_parameters , control_params *, simulation_data *,
- static_storage ,
- list , list , int , int , real *, real *, real *, rvec *);
-
-GLOBAL void Three_Body_Interactions_results ( reax_atom *,
- control_params *,
- static_storage ,
- list , int );
+ static_storage*, list**, output_controls* );
-GLOBAL void Three_Body_Estimate ( reax_atom *atoms,
- control_params *control,
- list p_bonds, int N,
- int *count);
+void Calculate_Theta( rvec, real, rvec, real, real*, real* );
-GLOBAL void Hydrogen_Bonds ( reax_atom *,
- single_body_parameters *, hbond_parameters *,
- control_params *, simulation_data *, static_storage ,
- list , list , int , int, real *, rvec *, rvec *);
-GLOBAL void Hydrogen_Bonds_HB ( reax_atom *,
- single_body_parameters *, hbond_parameters *,
- control_params *, simulation_data *, static_storage ,
- list , list , int , int, real *, rvec *, rvec *);
+void Calculate_dCos_Theta( rvec, real, rvec, real, rvec*, rvec*, rvec* );
-GLOBAL void Hydrogen_Bonds_Postprocess ( reax_atom *,
- single_body_parameters *,
- static_storage , list,
- list , list , int, real * );
-GLOBAL void Hydrogen_Bonds_Far_Nbrs ( reax_atom *,
- single_body_parameters *,
- static_storage , list,
- list , list , int );
-GLOBAL void Hydrogen_Bonds_HNbrs ( reax_atom *,
- single_body_parameters *,
- static_storage , list,
- list , list , int );
#endif
diff --git a/PuReMD-GPU/src/traj.cu b/PuReMD-GPU/src/traj.c
similarity index 98%
rename from PuReMD-GPU/src/traj.cu
rename to PuReMD-GPU/src/traj.c
index 97496e7f7b8dc4e9add824ee198d0c2907180a98..2844c370ee79702ed0c75d090afe545149aae185 100644
--- a/PuReMD-GPU/src/traj.cu
+++ b/PuReMD-GPU/src/traj.c
@@ -19,13 +19,17 @@
----------------------------------------------------------------------*/
#include "traj.h"
+
#include "list.h"
-#include "cuda_copy.h"
+
+#ifdef __PRINT_CPU_RESULTS__
+ #include "cuda_copy.h"
+#endif
+
/************************************************/
/* CUSTOM FORMAT ROUTINES */
/************************************************/
-
int Write_Custom_Header(reax_system *system, control_params *control,
static_storage *workspace, output_controls *out_control)
{
@@ -207,9 +211,9 @@ int Append_Custom_Frame( reax_system *system, control_params *control,
if( write_bonds )
{
-#ifndef __PRINT_CPU_RESULTS__
+#ifdef __PRINT_CPU_RESULTS__
//fprintf (stderr, "Synching bonds from device for printing ....\n");
- Sync_Host_Device (bonds, (dev_lists + BONDS), TYP_BOND );
+ Sync_Host_Device_List( bonds, (dev_lists + BONDS), TYP_BOND );
#endif
for( i = 0; i < system->N; ++i )
@@ -239,12 +243,12 @@ int Append_Custom_Frame( reax_system *system, control_params *control,
num_thb_intrs = 0;
if( write_angles ) {
-#ifndef __PRINT_CPU_RESULTS__
+#ifdef __PRINT_CPU_RESULTS__
//fprintf (stderr, "Synching three bodies from deivce for printing ... \n");
- Sync_Host_Device (thb_intrs, dev_lists + THREE_BODIES, TYP_THREE_BODY );
+ Sync_Host_Device_List( thb_intrs, dev_lists + THREE_BODIES, TYP_THREE_BODY );
if ( !write_bonds) {
//fprintf (stderr, "Synching bonds for three bodies from device for printing ... \n");
- Sync_Host_Device (bonds, (dev_lists + BONDS), TYP_BOND );
+ Sync_Host_Device_List( bonds, (dev_lists + BONDS), TYP_BOND );
}
#endif
diff --git a/PuReMD-GPU/src/traj.h b/PuReMD-GPU/src/traj.h
index d8c1792d0f6941d6881939d559bd9003b286bfa1..35d92602eee7c2d0b5ee83889623df2cb2106c71 100644
--- a/PuReMD-GPU/src/traj.h
+++ b/PuReMD-GPU/src/traj.h
@@ -22,8 +22,8 @@
#define __TRAJ_H__
#include "mytypes.h"
-#include "zlib.h"
+#include <zlib.h>
#define BLOCK_MARK "REAX_BLOCK_MARK "
#define BLOCK_MARK_LEN 16
@@ -73,12 +73,14 @@
#define SIZE_INFO_LINE3 "%-10d %-10d %-10d\n"
#define SIZE_INFO_LEN3 33
+
enum ATOM_LINE_OPTS {OPT_NOATOM = 0, OPT_ATOM_BASIC = 4, OPT_ATOM_wF = 5,
OPT_ATOM_wV = 6, OPT_ATOM_FULL = 7
};
enum BOND_LINE_OPTS {OPT_NOBOND, OPT_BOND_BASIC, OPT_BOND_FULL};
enum ANGLE_LINE_OPTS {OPT_NOANGLE, OPT_ANGLE_BASIC};
+
struct
{
int no_of_sub_blocks;
@@ -89,11 +91,11 @@ struct
typedef struct __block block;
+
int Write_Block( gzFile, block* );
int Read_Next_Block( gzFile, block*, int* );
int Skip_Next_Block( gzFile, int*);
-
/*
Format for trajectory file
@@ -141,8 +143,6 @@ int Skip_Next_Block( gzFile, int*);
No. of torsion entries (int)
Torsion info lines as per torsion format.
*/
-
-
int Write_Custom_Header( reax_system*, control_params*,
static_storage*, output_controls* );
int Write_xyz_Header ( reax_system*, control_params*,
diff --git a/PuReMD-GPU/src/two_body_interactions.c b/PuReMD-GPU/src/two_body_interactions.c
new file mode 100644
index 0000000000000000000000000000000000000000..2e7a6daf9039ea26c22b2fcfda5913e46255ad75
--- /dev/null
+++ b/PuReMD-GPU/src/two_body_interactions.c
@@ -0,0 +1,571 @@
+/*----------------------------------------------------------------------
+ PuReMD-GPU - Reax Force Field Simulator
+
+ Copyright (2014) Purdue University
+ Sudhir Kylasa, skylasa@purdue.edu
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Ananth Y Grama, ayg@cs.purdue.edu
+
+ This program is free software; you can redistribute it and/or
+ modify it under the terms of the GNU General Public License as
+ published by the Free Software Foundation; either version 2 of
+ the License, or (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ See the GNU General Public License for more details:
+ <http://www.gnu.org/licenses/>.
+ ----------------------------------------------------------------------*/
+
+#include "two_body_interactions.h"
+
+#include "bond_orders.h"
+#include "list.h"
+#include "lookup.h"
+#include "vector.h"
+#include "index_utils.h"
+
+
+void Bond_Energy( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, pj;
+ int start_i, end_i;
+ int type_i, type_j;
+ real ebond, pow_BOs_be2, exp_be12, CEbo;
+ real gp3, gp4, gp7, gp10, gp37;
+ real exphu, exphua1, exphub1, exphuov, hulpov, estriph;
+ real decobdbo, decobdboua, decobdboub;
+ single_body_parameters *sbp_i, *sbp_j;
+ two_body_parameters *twbp;
+ bond_order_data *bo_ij;
+ list *bonds;
+
+ bonds = (*lists) + BONDS;
+ gp3 = system->reaxprm.gp.l[3];
+ gp4 = system->reaxprm.gp.l[4];
+ gp7 = system->reaxprm.gp.l[7];
+ gp10 = system->reaxprm.gp.l[10];
+ gp37 = (int) system->reaxprm.gp.l[37];
+
+ for( i=0; i < system->N; ++i ) {
+ start_i = Start_Index(i, bonds);
+ end_i = End_Index(i, bonds);
+ //fprintf( stderr, "i=%d start=%d end=%d\n", i, start_i, end_i );
+ for( pj = start_i; pj < end_i; ++pj )
+ if( i < bonds->select.bond_list[pj].nbr ) {
+ /* set the pointers */
+ j = bonds->select.bond_list[pj].nbr;
+ type_i = system->atoms[i].type;
+ type_j = system->atoms[j].type;
+ sbp_i = &( system->reaxprm.sbp[type_i] );
+ sbp_j = &( system->reaxprm.sbp[type_j] );
+ twbp = &( system->reaxprm.tbp[ index_tbp(type_i,type_j,system->reaxprm.num_atom_types) ] );
+ bo_ij = &( bonds->select.bond_list[pj].bo_data );
+
+ /* calculate the constants */
+ pow_BOs_be2 = POW( bo_ij->BO_s, twbp->p_be2 );
+ exp_be12 = EXP( twbp->p_be1 * ( 1.0 - pow_BOs_be2 ) );
+ CEbo = -twbp->De_s * exp_be12 *
+ ( 1.0 - twbp->p_be1 * twbp->p_be2 * pow_BOs_be2 );
+
+ /* calculate the Bond Energy */
+ ebond =
+ -twbp->De_s * bo_ij->BO_s * exp_be12
+ -twbp->De_p * bo_ij->BO_pi
+ -twbp->De_pp * bo_ij->BO_pi2;
+
+ data->E_BE += ebond;
+
+ /* calculate derivatives of Bond Orders */
+ bo_ij->Cdbo += CEbo;
+ bo_ij->Cdbopi -= (CEbo + twbp->De_p);
+ bo_ij->Cdbopi2 -= (CEbo + twbp->De_pp);
+
+#ifdef TEST_ENERGY
+ fprintf( out_control->ebond, "%6d%6d%24.15e%24.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ // i+1, j+1,
+ bo_ij->BO, ebond/*, data->E_BE*/ );
+ /* fprintf( out_control->ebond, "%6d%6d%12.6f%12.6f%12.6f\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ CEbo, -twbp->De_p, -twbp->De_pp );*/
+#endif
+#ifdef TEST_FORCES
+ Add_dBO( system, lists, i, pj, CEbo, workspace->f_be );
+ Add_dBOpinpi2( system, lists, i, pj,
+ -(CEbo + twbp->De_p), -(CEbo + twbp->De_pp),
+ workspace->f_be, workspace->f_be );
+#endif
+
+ /* Stabilisation terminal triple bond */
+ if( bo_ij->BO >= 1.00 ) {
+ if( gp37 == 2 ||
+ (sbp_i->mass == 12.0000 && sbp_j->mass == 15.9990) ||
+ (sbp_j->mass == 12.0000 && sbp_i->mass == 15.9990) ) {
+ // ba = SQR(bo_ij->BO - 2.50);
+ exphu = EXP( -gp7 * SQR(bo_ij->BO - 2.50) );
+ //oboa=abo(j1)-boa;
+ //obob=abo(j2)-boa;
+ exphua1 = EXP(-gp3*(workspace->total_bond_order[i]-bo_ij->BO));
+ exphub1 = EXP(-gp3*(workspace->total_bond_order[j]-bo_ij->BO));
+ //ovoab=abo(j1)-aval(it1)+abo(j2)-aval(it2);
+ exphuov = EXP(gp4*(workspace->Delta[i] + workspace->Delta[j]));
+ hulpov = 1.0 / (1.0 + 25.0 * exphuov);
+
+ estriph = gp10 * exphu * hulpov * (exphua1 + exphub1);
+ //estrain(j1) = estrain(j1) + 0.50*estriph;
+ //estrain(j2) = estrain(j2) + 0.50*estriph;
+ data->E_BE += estriph;
+
+ decobdbo = gp10 * exphu * hulpov * (exphua1 + exphub1) *
+ ( gp3 - 2.0 * gp7 * (bo_ij->BO-2.50) );
+ decobdboua = -gp10 * exphu * hulpov *
+ (gp3*exphua1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
+ decobdboub = -gp10 * exphu * hulpov *
+ (gp3*exphub1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
+
+ bo_ij->Cdbo += decobdbo;
+ workspace->CdDelta[i] += decobdboua;
+ workspace->CdDelta[j] += decobdboub;
+ //loop_j ++;
+ //fprintf (stderr, "incrementing loopj %d \n", loop_j);
+#ifdef TEST_ENERGY
+ fprintf( out_control->ebond,
+ "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ //i+1, j+1,
+ estriph, decobdbo, decobdboua, decobdboub );
+#endif
+#ifdef TEST_FORCES
+ Add_dBO( system, lists, i, pj, decobdbo, workspace->f_be );
+ Add_dDelta( system, lists, i, decobdboua, workspace->f_be );
+ Add_dDelta( system, lists, j, decobdboub, workspace->f_be );
+#endif
+ }
+ }
+ }
+ }
+}
+
+
+void vdW_Coulomb_Energy( reax_system *system, control_params *control,
+ simulation_data *data, static_storage *workspace,
+ list **lists, output_controls *out_control )
+{
+ int i, j, pj;
+ int start_i, end_i;
+ real self_coef;
+ real p_vdW1, p_vdW1i;
+ real powr_vdW1, powgi_vdW1;
+ real tmp, r_ij, fn13, exp1, exp2;
+ real Tap, dTap, dfn13, CEvd, CEclmb;
+ real dr3gamij_1, dr3gamij_3;
+ real e_ele, e_vdW, e_core, de_core;
+ rvec temp, ext_press;
+ // rtensor temp_rtensor, total_rtensor;
+ two_body_parameters *twbp;
+ far_neighbor_data *nbr_pj;
+ list *far_nbrs;
+
+ p_vdW1 = system->reaxprm.gp.l[28];
+ p_vdW1i = 1.0 / p_vdW1;
+ far_nbrs = (*lists) + FAR_NBRS;
+ e_ele = 0;
+ e_vdW = 0;
+ e_core = 0;
+ de_core = 0;
+
+ for( i = 0; i < system->N; ++i ) {
+ start_i = Start_Index(i, far_nbrs);
+ end_i = End_Index(i, far_nbrs);
+ // fprintf( stderr, "i: %d, start: %d, end: %d\n",
+ // i, start_i, end_i );
+
+ for( pj = start_i; pj < end_i; ++pj )
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ r_ij = nbr_pj->d;
+ twbp = &(system->reaxprm.tbp[ index_tbp(system->atoms[i].type, system->atoms[j].type, system->reaxprm.num_atom_types) ]);
+ self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
+
+ /* Calculate Taper and its derivative */
+ // Tap = nbr_pj->Tap; -- precomputed during compte_H
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
+ dTap = dTap * r_ij + 5*control->Tap5;
+ dTap = dTap * r_ij + 4*control->Tap4;
+ dTap = dTap * r_ij + 3*control->Tap3;
+ dTap = dTap * r_ij + 2*control->Tap2;
+ dTap += control->Tap1/r_ij;
+
+ /*vdWaals Calculations*/
+ if(system->reaxprm.gp.vdw_type==1 || system->reaxprm.gp.vdw_type==3) {
+ // shielding
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ data->E_vdW += e_vdW =
+ self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
+ POW(r_ij, p_vdW1 - 2.0);
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) * dfn13 );
+ }
+ else{ // no shielding
+ exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+
+ data->E_vdW += e_vdW =
+ self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
+ (exp1 - exp2) );
+ }
+
+ if(system->reaxprm.gp.vdw_type==2 || system->reaxprm.gp.vdw_type==3) {
+ // innner wall
+ e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
+ e_vdW += self_coef * Tap * e_core;
+ data->E_vdW += self_coef * Tap * e_core;
+
+ de_core = -(twbp->acore/twbp->rcore) * e_core;
+ CEvd += self_coef * ( dTap * e_core + Tap * de_core );
+ }
+
+ /*Coulomb Calculations*/
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ tmp = Tap / dr3gamij_3;
+ //tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
+ data->E_Ele += e_ele =
+ self_coef * C_ele * system->atoms[i].q * system->atoms[j].q * tmp;
+
+
+ CEclmb = self_coef * C_ele * system->atoms[i].q * system->atoms[j].q *
+ ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
+ /*CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
+ ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;*/
+
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ rvec_ScaledAdd( system->atoms[i].f,
+ -(CEvd+CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( system->atoms[j].f,
+ +(CEvd+CEclmb), nbr_pj->dvec );
+ }
+ else { // NPT, iNPT or sNPT
+ /* for pressure coupling, terms not related to bond order
+ derivatives are added directly into pressure vector/tensor */
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+
+ rvec_ScaledAdd( system->atoms[i].f, -1., temp );
+ rvec_Add( system->atoms[j].f, temp );
+
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+ rvec_Add( data->ext_press, ext_press );
+
+ /*fprintf( stderr, "nonbonded(%d,%d): rel_box (%f %f %f)",
+ i,j,nbr_pj->rel_box[0],nbr_pj->rel_box[1],nbr_pj->rel_box[2] );
+
+ fprintf( stderr, "force(%f %f %f)", temp[0], temp[1], temp[2] );
+
+ fprintf( stderr, "ext_press (%12.6f %12.6f %12.6f)\n",
+ data->ext_press[0], data->ext_press[1], data->ext_press[2] );*/
+
+ /* This part is intended for a fully-flexible box */
+ /* rvec_OuterProduct( temp_rtensor, nbr_pj->dvec,
+ system->atoms[i].x );
+ rtensor_Scale( total_rtensor,
+ F_C * -(CEvd + CEclmb), temp_rtensor );
+ rvec_OuterProduct( temp_rtensor,
+ nbr_pj->dvec, system->atoms[j].x );
+ rtensor_ScaledAdd( total_rtensor,
+ F_C * +(CEvd + CEclmb), temp_rtensor );
+
+ if( nbr_pj->imaginary )
+ // This is an external force due to an imaginary nbr
+ rtensor_ScaledAdd( data->flex_bar.P, -1.0, total_rtensor );
+ else
+ // This interaction is completely internal
+ rtensor_Add( data->flex_bar.P, total_rtensor ); */
+ }
+
+#ifdef TEST_ENERGY
+ rvec_MakeZero( temp );
+ rvec_ScaledAdd( temp, +CEvd, nbr_pj->dvec );
+ fprintf( out_control->evdw,
+ "%6d%6d%24.15e%24.15e%24.15e%24.15e%24.15e\n",
+ //i+1, j+1,
+ MIN( workspace->orig_id[i], workspace->orig_id[j] ),
+ MAX( workspace->orig_id[i], workspace->orig_id[j] ),
+ r_ij, e_vdW, temp[0], temp[1], temp[2]/*, data->E_vdW*/ );
+
+ fprintf( out_control->ecou, "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
+ MIN( workspace->orig_id[i], workspace->orig_id[j] ),
+ MAX( workspace->orig_id[i], workspace->orig_id[j] ),
+ r_ij, system->atoms[i].q, system->atoms[j].q,
+ e_ele/*, data->E_Ele*/ );
+#endif
+#ifdef TEST_FORCES
+ rvec_ScaledAdd( workspace->f_vdw[i], -CEvd, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_vdw[j], +CEvd, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_ele[i], -CEclmb, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_ele[j], +CEclmb, nbr_pj->dvec );
+#endif
+ }
+ }
+
+ // fclose( fout );
+
+ // fprintf( stderr, "nonbonded: ext_press (%24.15e %24.15e %24.15e)\n",
+ // data->ext_press[0], data->ext_press[1], data->ext_press[2] );
+}
+
+
+void LR_vdW_Coulomb( reax_system *system, control_params *control,
+ int i, int j, real r_ij, LR_data *lr )
+{
+ real p_vdW1 = system->reaxprm.gp.l[28];
+ real p_vdW1i = 1.0 / p_vdW1;
+ real powr_vdW1, powgi_vdW1;
+ real tmp, fn13, exp1, exp2;
+ real Tap, dTap, dfn13;
+ real dr3gamij_1, dr3gamij_3;
+ real e_core, de_core;
+ two_body_parameters *twbp;
+
+ twbp = &(system->reaxprm.tbp[ index_tbp(i,j,system->reaxprm.num_atom_types) ]);
+ e_core = 0;
+ de_core = 0;
+
+ /* calculate taper and its derivative */
+ Tap = control->Tap7 * r_ij + control->Tap6;
+ Tap = Tap * r_ij + control->Tap5;
+ Tap = Tap * r_ij + control->Tap4;
+ Tap = Tap * r_ij + control->Tap3;
+ Tap = Tap * r_ij + control->Tap2;
+ Tap = Tap * r_ij + control->Tap1;
+ Tap = Tap * r_ij + control->Tap0;
+
+ dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
+ dTap = dTap * r_ij + 5*control->Tap5;
+ dTap = dTap * r_ij + 4*control->Tap4;
+ dTap = dTap * r_ij + 3*control->Tap3;
+ dTap = dTap * r_ij + 2*control->Tap2;
+ dTap += control->Tap1/r_ij;
+
+
+ /* vdWaals calculations */
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+ /* fprintf(stderr,"vdW: Tap:%f, r: %f, f13:%f, D:%f, Energy:%f,\
+Gamma_w:%f, p_vdw: %f, alpha: %f, r_vdw: %f, %lf %lf\n",
+Tap, r_ij, fn13, twbp->D, Tap * twbp->D * (exp1 - 2.0 * exp2),
+powgi_vdW1, p_vdW1, twbp->alpha, twbp->r_vdW, exp1, exp2); */
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) * POW(r_ij, p_vdW1 - 2.0);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
+
+ /*vdWaals Calculations*/
+ if(system->reaxprm.gp.vdw_type==1 || system->reaxprm.gp.vdw_type==3)
+ { // shielding
+ powr_vdW1 = POW(r_ij, p_vdW1);
+ powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
+
+ fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
+ exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
+ POW(r_ij, p_vdW1 - 2.0);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
+ }
+ else{ // no shielding
+ exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+ exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
+
+ lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
+
+ lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
+ Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2);
+ }
+
+ if(system->reaxprm.gp.vdw_type==2 || system->reaxprm.gp.vdw_type==3)
+ { // innner wall
+ e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
+ lr->e_vdW += Tap * e_core;
+
+ de_core = -(twbp->acore/twbp->rcore) * e_core;
+ lr->CEvd += dTap * e_core + Tap * de_core;
+ }
+
+ /* Coulomb calculations */
+ dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
+ dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+
+ tmp = Tap / dr3gamij_3;
+ lr->H = EV_to_KCALpMOL * tmp;
+ lr->e_ele = C_ele * tmp;
+ /* fprintf( stderr,"i:%d(%d), j:%d(%d), gamma:%f,\
+Tap:%f, dr3gamij_3:%f, qi: %f, qj: %f\n",
+i, system->atoms[i].type, j, system->atoms[j].type,
+twbp->gamma, Tap, dr3gamij_3,
+system->atoms[i].q, system->atoms[j].q ); */
+
+ lr->CEclmb = C_ele * ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
+ /* fprintf( stdout, "%d %d\t%g\t%g %g\t%g %g\t%g %g\n",
+ i+1, j+1, r_ij, e_vdW, CEvd * r_ij,
+ system->atoms[i].q, system->atoms[j].q, e_ele, CEclmb * r_ij ); */
+
+ /* fprintf( stderr,"LR_Lookup:%3d%3d%5.3f-%8.5f,%8.5f%8.5f,%8.5f%8.5f\n",
+ i, j, r_ij, lr->H, lr->e_vdW, lr->CEvd, lr->e_ele, lr->CEclmb ); */
+}
+
+
+void Tabulated_vdW_Coulomb_Energy( reax_system *system, control_params *control,
+ simulation_data *data,
+ static_storage *workspace, list **lists,
+ output_controls *out_control )
+{
+ int i, j, pj, r, steps, update_freq, update_energies;
+ int type_i, type_j, tmin, tmax;
+ int start_i, end_i;
+ real r_ij, self_coef, base, dif;
+ real e_vdW, e_ele;
+ real CEvd, CEclmb;
+ rvec temp, ext_press;
+ far_neighbor_data *nbr_pj;
+ list *far_nbrs = (*lists) + FAR_NBRS;
+ LR_lookup_table *t;
+
+ steps = data->step - data->prev_steps;
+ update_freq = out_control->energy_update_freq;
+ update_energies = update_freq > 0 && steps % update_freq == 0;
+
+ for( i = 0; i < system->N; ++i ) {
+ type_i = system->atoms[i].type;
+ start_i = Start_Index(i,far_nbrs);
+ end_i = End_Index(i,far_nbrs);
+
+ for( pj = start_i; pj < end_i; ++pj )
+ if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
+ nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
+ j = nbr_pj->nbr;
+ type_j = system->atoms[j].type;
+ r_ij = nbr_pj->d;
+ self_coef = (i == j) ? 0.5 : 1.0;
+ tmin = MIN( type_i, type_j );
+ tmax = MAX( type_i, type_j );
+ t = &( LR[ index_lr (tmin,tmax,system->reaxprm.num_atom_types) ] );
+
+ /* Cubic Spline Interpolation */
+ r = (int)(r_ij * t->inv_dx);
+ if( r == 0 ) ++r;
+ base = (real)(r+1) * t->dx;
+ dif = r_ij - base;
+ //fprintf(stderr, "r: %f, i: %d, base: %f, dif: %f\n", r, i, base, dif);
+
+ if( update_energies ) {
+ e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
+ t->vdW[r].a;
+ e_vdW *= self_coef;
+
+ e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
+ t->ele[r].a;
+ e_ele *= self_coef * system->atoms[i].q * system->atoms[j].q;
+
+ data->E_vdW += e_vdW;
+ data->E_Ele += e_ele;
+ }
+
+ CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
+ t->CEvd[r].a;
+ CEvd *= self_coef;
+ //CEvd = (3*t->vdW[r].d*dif + 2*t->vdW[r].c)*dif + t->vdW[r].b;
+
+ CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
+ t->CEclmb[r].a;
+ CEclmb *= self_coef * system->atoms[i].q * system->atoms[j].q;
+
+ if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
+ rvec_ScaledAdd( system->atoms[i].f, -(CEvd + CEclmb), nbr_pj->dvec );
+ rvec_ScaledAdd( system->atoms[j].f, +(CEvd + CEclmb), nbr_pj->dvec );
+ }
+ else { // NPT, iNPT or sNPT
+ /* for pressure coupling, terms not related to bond order
+ derivatives are added directly into pressure vector/tensor */
+ rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
+ rvec_ScaledAdd( system->atoms[i].f, -1., temp );
+ rvec_Add( system->atoms[j].f, temp );
+ rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
+ rvec_Add( data->ext_press, ext_press );
+ }
+
+#ifdef TEST_ENERGY
+ fprintf(out_control->evdw, "%6d%6d%24.15e%24.15e%24.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ r_ij, e_vdW, data->E_vdW );
+ fprintf(out_control->ecou,"%6d%6d%24.15e%24.15e%24.15e%24.15e%24.15e\n",
+ workspace->orig_id[i], workspace->orig_id[j],
+ r_ij, system->atoms[i].q, system->atoms[j].q,
+ e_ele, data->E_Ele );
+#endif
+#ifdef TEST_FORCES
+ rvec_ScaledAdd( workspace->f_vdw[i], -CEvd, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_vdw[j], +CEvd, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_ele[i], -CEclmb, nbr_pj->dvec );
+ rvec_ScaledAdd( workspace->f_ele[j], +CEclmb, nbr_pj->dvec );
+#endif
+ }
+ }
+}
+
+
+#if defined(OLD)
+ /* Linear extrapolation */
+ /*p = (r_ij * t->inv_dx;
+ r = (int) p;
+ prev = &( t->y[r] );
+ next = &( t->y[r+1] );
+
+ tmp = p - r;
+ e_vdW = self_coef * (prev->e_vdW + tmp*(next->e_vdW - prev->e_vdW ));
+ CEvd = self_coef * (prev->CEvd + tmp*(next->CEvd - prev->CEvd ));
+
+ e_ele = self_coef * (prev->e_ele + tmp*(next->e_ele - prev->e_ele ));
+ e_ele = e_ele * system->atoms[i].q * system->atoms[j].q;
+ CEclmb = self_coef * (prev->CEclmb+tmp*(next->CEclmb - prev->CEclmb));
+ CEclmb = CEclmb * system->atoms[i].q * system->atoms[j].q;*/
+#endif
diff --git a/PuReMD-GPU/src/two_body_interactions.cu b/PuReMD-GPU/src/two_body_interactions.cu
deleted file mode 100644
index f53b0cfb0fb1c23626032e7a257fce5b1447953e..0000000000000000000000000000000000000000
--- a/PuReMD-GPU/src/two_body_interactions.cu
+++ /dev/null
@@ -1,1630 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD-GPU - Reax Force Field Simulator
-
- Copyright (2014) Purdue University
- Sudhir Kylasa, skylasa@purdue.edu
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Ananth Y Grama, ayg@cs.purdue.edu
-
- This program is free software; you can redistribute it and/or
- modify it under the terms of the GNU General Public License as
- published by the Free Software Foundation; either version 2 of
- the License, or (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- See the GNU General Public License for more details:
- <http://www.gnu.org/licenses/>.
- ----------------------------------------------------------------------*/
-
-#include "two_body_interactions.h"
-#include "bond_orders.h"
-#include "list.h"
-#include "lookup.h"
-#include "vector.h"
-#include "index_utils.h"
-
-#include "cuda_helpers.h"
-
-
-void Bond_Energy( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- real ebond, pow_BOs_be2, exp_be12, CEbo;
- real gp3, gp4, gp7, gp10, gp37;
- real exphu, exphua1, exphub1, exphuov, hulpov, estriph;
- real decobdbo, decobdboua, decobdboub;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- bond_order_data *bo_ij;
- list *bonds;
-
- bonds = (*lists) + BONDS;
- gp3 = system->reaxprm.gp.l[3];
- gp4 = system->reaxprm.gp.l[4];
- gp7 = system->reaxprm.gp.l[7];
- gp10 = system->reaxprm.gp.l[10];
- gp37 = (int) system->reaxprm.gp.l[37];
-
- for( i=0; i < system->N; ++i ) {
- start_i = Start_Index(i, bonds);
- end_i = End_Index(i, bonds);
- //fprintf( stderr, "i=%d start=%d end=%d\n", i, start_i, end_i );
- for( pj = start_i; pj < end_i; ++pj )
- if( i < bonds->select.bond_list[pj].nbr ) {
- /* set the pointers */
- j = bonds->select.bond_list[pj].nbr;
- type_i = system->atoms[i].type;
- type_j = system->atoms[j].type;
- sbp_i = &( system->reaxprm.sbp[type_i] );
- sbp_j = &( system->reaxprm.sbp[type_j] );
- twbp = &( system->reaxprm.tbp[ index_tbp (type_i,type_j,&system->reaxprm) ] );
- bo_ij = &( bonds->select.bond_list[pj].bo_data );
-
- /* calculate the constants */
- pow_BOs_be2 = POW( bo_ij->BO_s, twbp->p_be2 );
- exp_be12 = EXP( twbp->p_be1 * ( 1.0 - pow_BOs_be2 ) );
- CEbo = -twbp->De_s * exp_be12 *
- ( 1.0 - twbp->p_be1 * twbp->p_be2 * pow_BOs_be2 );
-
- /* calculate the Bond Energy */
- ebond =
- -twbp->De_s * bo_ij->BO_s * exp_be12
- -twbp->De_p * bo_ij->BO_pi
- -twbp->De_pp * bo_ij->BO_pi2;
-
- data->E_BE += ebond;
-
- /* calculate derivatives of Bond Orders */
- bo_ij->Cdbo += CEbo;
- bo_ij->Cdbopi -= (CEbo + twbp->De_p);
- bo_ij->Cdbopi2 -= (CEbo + twbp->De_pp);
-
-#ifdef TEST_ENERGY
- fprintf( out_control->ebond, "%6d%6d%24.15e%24.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- // i+1, j+1,
- bo_ij->BO, ebond/*, data->E_BE*/ );
- /* fprintf( out_control->ebond, "%6d%6d%12.6f%12.6f%12.6f\n",
- workspace->orig_id[i], workspace->orig_id[j],
- CEbo, -twbp->De_p, -twbp->De_pp );*/
-#endif
-#ifdef TEST_FORCES
- Add_dBO( system, lists, i, pj, CEbo, workspace->f_be );
- Add_dBOpinpi2( system, lists, i, pj,
- -(CEbo + twbp->De_p), -(CEbo + twbp->De_pp),
- workspace->f_be, workspace->f_be );
-#endif
-
- /* Stabilisation terminal triple bond */
- if( bo_ij->BO >= 1.00 ) {
- if( gp37 == 2 ||
- (sbp_i->mass == 12.0000 && sbp_j->mass == 15.9990) ||
- (sbp_j->mass == 12.0000 && sbp_i->mass == 15.9990) ) {
- // ba = SQR(bo_ij->BO - 2.50);
- exphu = EXP( -gp7 * SQR(bo_ij->BO - 2.50) );
- //oboa=abo(j1)-boa;
- //obob=abo(j2)-boa;
- exphua1 = EXP(-gp3*(workspace->total_bond_order[i]-bo_ij->BO));
- exphub1 = EXP(-gp3*(workspace->total_bond_order[j]-bo_ij->BO));
- //ovoab=abo(j1)-aval(it1)+abo(j2)-aval(it2);
- exphuov = EXP(gp4*(workspace->Delta[i] + workspace->Delta[j]));
- hulpov = 1.0 / (1.0 + 25.0 * exphuov);
-
- estriph = gp10 * exphu * hulpov * (exphua1 + exphub1);
- //estrain(j1) = estrain(j1) + 0.50*estriph;
- //estrain(j2) = estrain(j2) + 0.50*estriph;
- data->E_BE += estriph;
-
- decobdbo = gp10 * exphu * hulpov * (exphua1 + exphub1) *
- ( gp3 - 2.0 * gp7 * (bo_ij->BO-2.50) );
- decobdboua = -gp10 * exphu * hulpov *
- (gp3*exphua1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
- decobdboub = -gp10 * exphu * hulpov *
- (gp3*exphub1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
-
- bo_ij->Cdbo += decobdbo;
- workspace->CdDelta[i] += decobdboua;
- workspace->CdDelta[j] += decobdboub;
- //loop_j ++;
- //fprintf (stderr, "incrementing loopj %d \n", loop_j);
-#ifdef TEST_ENERGY
- fprintf( out_control->ebond,
- "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- //i+1, j+1,
- estriph, decobdbo, decobdboua, decobdboub );
-#endif
-#ifdef TEST_FORCES
- Add_dBO( system, lists, i, pj, decobdbo, workspace->f_be );
- Add_dDelta( system, lists, i, decobdboua, workspace->f_be );
- Add_dDelta( system, lists, j, decobdboub, workspace->f_be );
-#endif
- }
- }
- }
- }
-}
-
-
-
-
-
-
-
-GLOBAL void Cuda_Bond_Energy ( reax_atom *atoms, global_parameters g_params,
- single_body_parameters *sbp, two_body_parameters *tbp,
- simulation_data *data,
- static_storage p_workspace, list p_bonds,
- int N, int num_atom_types, real *E_BE)
-{
- int i, j, pj;
- int start_i, end_i;
- int type_i, type_j;
- real ebond, pow_BOs_be2, exp_be12, CEbo;
- real gp3, gp4, gp7, gp10, gp37;
- real exphu, exphua1, exphub1, exphuov, hulpov, estriph;
- real decobdbo, decobdboua, decobdboub;
- single_body_parameters *sbp_i, *sbp_j;
- two_body_parameters *twbp;
- bond_order_data *bo_ij;
- list *bonds;
- static_storage *workspace;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
- if ( i >= N ) return;
-
- bonds = &p_bonds;
- workspace = &p_workspace;
-
- gp3 = g_params.l[3];
- gp4 = g_params.l[4];
- gp7 = g_params.l[7];
- gp10 = g_params.l[10];
- gp37 = (int) g_params.l[37];
-
- //for( i=0; i < system->N; ++i )
- start_i = Start_Index(i, bonds);
- end_i = End_Index(i, bonds);
- //fprintf( stderr, "i=%d start=%d end=%d\n", i, start_i, end_i );
- for( pj = start_i; pj < end_i; ++pj )
- {
- //TODO
- //if( i < bonds->select.bond_list[pj].nbr )
- if( i < bonds->select.bond_list[pj].nbr )
- {
- //TODO
- /* set the pointers */
- j = bonds->select.bond_list[pj].nbr;
- type_i = atoms[i].type;
- type_j = atoms[j].type;
- sbp_i = &( sbp[type_i] );
- sbp_j = &( sbp[type_j] );
- twbp = &( tbp[ index_tbp (type_i,type_j,num_atom_types) ] );
- bo_ij = &( bonds->select.bond_list[pj].bo_data );
-
- /* calculate the constants */
- pow_BOs_be2 = POW( bo_ij->BO_s, twbp->p_be2 );
- exp_be12 = EXP( twbp->p_be1 * ( 1.0 - pow_BOs_be2 ) );
- CEbo = -twbp->De_s * exp_be12 *
- ( 1.0 - twbp->p_be1 * twbp->p_be2 * pow_BOs_be2 );
-
- /* calculate the Bond Energy */
- ebond =
- -twbp->De_s * bo_ij->BO_s * exp_be12
- -twbp->De_p * bo_ij->BO_pi
- -twbp->De_pp * bo_ij->BO_pi2;
-
- //PERFORMANCE IMAPCT
- //atomicAdd (&data->E_BE, ebond);
- //TODO
- //E_BE [ i ] += ebond/2.0;
- E_BE [ i ] += ebond;
- //data->E_BE += ebond;
-
- /* calculate derivatives of Bond Orders */
- bo_ij->Cdbo += CEbo;
- bo_ij->Cdbopi -= (CEbo + twbp->De_p);
- bo_ij->Cdbopi2 -= (CEbo + twbp->De_pp);
-
-#ifdef TEST_ENERGY
- //TODO
- //fprintf( out_control->ebond, "%6d%6d%24.15e%24.15e\n",
- // workspace->orig_id[i], workspace->orig_id[j],
- // i+1, j+1,
- // bo_ij->BO, ebond/*, data->E_BE*/ );
- /*
- fprintf( out_control->ebond, "%6d%6d%12.6f%12.6f%12.6f\n",
- workspace->orig_id[i], workspace->orig_id[j],
- CEbo, -twbp->De_p, -twbp->De_pp );*/
-#endif
-#ifdef TEST_FORCES
- //TODO
- /*
- Add_dBO( system, lists, i, pj, CEbo, workspace->f_be );
- Add_dBOpinpi2( system, lists, i, pj,
- -(CEbo + twbp->De_p), -(CEbo + twbp->De_pp),
- workspace->f_be, workspace->f_be );
- */
- //TODO
-#endif
-
- /* Stabilisation terminal triple bond */
- if( bo_ij->BO >= 1.00 ) {
- if( gp37 == 2 ||
- (sbp_i->mass == 12.0000 && sbp_j->mass == 15.9990) ||
- (sbp_j->mass == 12.0000 && sbp_i->mass == 15.9990) ) {
- // ba = SQR(bo_ij->BO - 2.50);
- exphu = EXP( -gp7 * SQR(bo_ij->BO - 2.50) );
- //oboa=abo(j1)-boa;
- //obob=abo(j2)-boa;
- exphua1 = EXP(-gp3*(workspace->total_bond_order[i]-bo_ij->BO));
- exphub1 = EXP(-gp3*(workspace->total_bond_order[j]-bo_ij->BO));
- //ovoab=abo(j1)-aval(it1)+abo(j2)-aval(it2);
- exphuov = EXP(gp4*(workspace->Delta[i] + workspace->Delta[j]));
- hulpov = 1.0 / (1.0 + 25.0 * exphuov);
-
- estriph = gp10 * exphu * hulpov * (exphua1 + exphub1);
- //estrain(j1) = estrain(j1) + 0.50*estriph;
- //estrain(j2) = estrain(j2) + 0.50*estriph;
-
- //PERFORMANCE IMPACT
- //atomicAdd (&data->E_BE, estriph);
- E_BE [ i] += estriph;
- //data->E_BE += estriph;
-
- decobdbo = gp10 * exphu * hulpov * (exphua1 + exphub1) *
- ( gp3 - 2.0 * gp7 * (bo_ij->BO-2.50) );
- decobdboua = -gp10 * exphu * hulpov *
- (gp3*exphua1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
- decobdboub = -gp10 * exphu * hulpov *
- (gp3*exphub1 + 25.0*gp4*exphuov*hulpov*(exphua1+exphub1));
-
- bo_ij->Cdbo += decobdbo;
-
- //PERFORMANCE IMAPCT
- workspace->CdDelta[i] += decobdboua;
- //atomicAdd (&workspace->CdDelta[j], decobdboub);
- //CdDelta [ i * N + i ] += decobdboua;
- //CdDelta [ i * N + j ] += decobdboua;
- //workspace->CdDelta [i] += decobdboua;
- //workspace->CdDelta [j] += decobdboub;
-
-#ifdef TEST_ENERGY
- /*
- fprintf( out_control->ebond,
- "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- //i+1, j+1,
- estriph, decobdbo, decobdboua, decobdboub );
- */
-#endif
-#ifdef TEST_FORCES
- /*
- Add_dBO( system, lists, i, pj, decobdbo, workspace->f_be );
- Add_dDelta( system, lists, i, decobdboua, workspace->f_be );
- Add_dDelta( system, lists, j, decobdboub, workspace->f_be );
- */
-#endif
- }
- }
- }
- } //TODO commented out the if statement for processing i < j.
- // we process all teh bonds and add only half the energy
-}
-
-
-void vdW_Coulomb_Energy( reax_system *system, control_params *control,
- simulation_data *data, static_storage *workspace,
- list **lists, output_controls *out_control )
-{
- int i, j, pj;
- int start_i, end_i;
- real self_coef;
- real p_vdW1, p_vdW1i;
- real powr_vdW1, powgi_vdW1;
- real tmp, r_ij, fn13, exp1, exp2;
- real Tap, dTap, dfn13, CEvd, CEclmb;
- real dr3gamij_1, dr3gamij_3;
- real e_ele, e_vdW, e_core, de_core;
- rvec temp, ext_press;
- // rtensor temp_rtensor, total_rtensor;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- list *far_nbrs;
-
- p_vdW1 = system->reaxprm.gp.l[28];
- p_vdW1i = 1.0 / p_vdW1;
- far_nbrs = (*lists) + FAR_NBRS;
- e_ele = 0;
- e_vdW = 0;
- e_core = 0;
- de_core = 0;
-
- for( i = 0; i < system->N; ++i ) {
- start_i = Start_Index(i, far_nbrs);
- end_i = End_Index(i, far_nbrs);
- // fprintf( stderr, "i: %d, start: %d, end: %d\n",
- // i, start_i, end_i );
-
- for( pj = start_i; pj < end_i; ++pj )
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- r_ij = nbr_pj->d;
- twbp = &(system->reaxprm.tbp[ index_tbp (system->atoms[i].type, system->atoms[j].type, &system->reaxprm) ]);
- self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
-
- /* Calculate Taper and its derivative */
- // Tap = nbr_pj->Tap; -- precomputed during compte_H
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
- dTap = dTap * r_ij + 5*control->Tap5;
- dTap = dTap * r_ij + 4*control->Tap4;
- dTap = dTap * r_ij + 3*control->Tap3;
- dTap = dTap * r_ij + 2*control->Tap2;
- dTap += control->Tap1/r_ij;
-
- /*vdWaals Calculations*/
- if(system->reaxprm.gp.vdw_type==1 || system->reaxprm.gp.vdw_type==3) {
- // shielding
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- data->E_vdW += e_vdW =
- self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
- POW(r_ij, p_vdW1 - 2.0);
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) * dfn13 );
- }
- else{ // no shielding
- exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
-
- data->E_vdW += e_vdW =
- self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) );
- }
-
- if(system->reaxprm.gp.vdw_type==2 || system->reaxprm.gp.vdw_type==3) {
- // innner wall
- e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
- e_vdW += self_coef * Tap * e_core;
- data->E_vdW += self_coef * Tap * e_core;
-
- de_core = -(twbp->acore/twbp->rcore) * e_core;
- CEvd += self_coef * ( dTap * e_core + Tap * de_core );
- }
-
- /*Coulomb Calculations*/
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
- tmp = Tap / dr3gamij_3;
- //tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
- data->E_Ele += e_ele =
- self_coef * C_ele * system->atoms[i].q * system->atoms[j].q * tmp;
-
-
- CEclmb = self_coef * C_ele * system->atoms[i].q * system->atoms[j].q *
- ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
- /*CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
- ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;*/
-
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- rvec_ScaledAdd( system->atoms[i].f,
- -(CEvd+CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( system->atoms[j].f,
- +(CEvd+CEclmb), nbr_pj->dvec );
- }
- else { // NPT, iNPT or sNPT
- /* for pressure coupling, terms not related to bond order
- derivatives are added directly into pressure vector/tensor */
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
-
- rvec_ScaledAdd( system->atoms[i].f, -1., temp );
- rvec_Add( system->atoms[j].f, temp );
-
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
- rvec_Add( data->ext_press, ext_press );
-
- /*fprintf( stderr, "nonbonded(%d,%d): rel_box (%f %f %f)",
- i,j,nbr_pj->rel_box[0],nbr_pj->rel_box[1],nbr_pj->rel_box[2] );
-
- fprintf( stderr, "force(%f %f %f)", temp[0], temp[1], temp[2] );
-
- fprintf( stderr, "ext_press (%12.6f %12.6f %12.6f)\n",
- data->ext_press[0], data->ext_press[1], data->ext_press[2] );*/
-
- /* This part is intended for a fully-flexible box */
- /* rvec_OuterProduct( temp_rtensor, nbr_pj->dvec,
- system->atoms[i].x );
- rtensor_Scale( total_rtensor,
- F_C * -(CEvd + CEclmb), temp_rtensor );
- rvec_OuterProduct( temp_rtensor,
- nbr_pj->dvec, system->atoms[j].x );
- rtensor_ScaledAdd( total_rtensor,
- F_C * +(CEvd + CEclmb), temp_rtensor );
-
- if( nbr_pj->imaginary )
- // This is an external force due to an imaginary nbr
- rtensor_ScaledAdd( data->flex_bar.P, -1.0, total_rtensor );
- else
- // This interaction is completely internal
- rtensor_Add( data->flex_bar.P, total_rtensor ); */
- }
-
-#ifdef TEST_ENERGY
- rvec_MakeZero( temp );
- rvec_ScaledAdd( temp, +CEvd, nbr_pj->dvec );
- fprintf( out_control->evdw,
- "%6d%6d%24.15e%24.15e%24.15e%24.15e%24.15e\n",
- //i+1, j+1,
- MIN( workspace->orig_id[i], workspace->orig_id[j] ),
- MAX( workspace->orig_id[i], workspace->orig_id[j] ),
- r_ij, e_vdW, temp[0], temp[1], temp[2]/*, data->E_vdW*/ );
-
- fprintf( out_control->ecou, "%6d%6d%24.15e%24.15e%24.15e%24.15e\n",
- MIN( workspace->orig_id[i], workspace->orig_id[j] ),
- MAX( workspace->orig_id[i], workspace->orig_id[j] ),
- r_ij, system->atoms[i].q, system->atoms[j].q,
- e_ele/*, data->E_Ele*/ );
-#endif
-#ifdef TEST_FORCES
- rvec_ScaledAdd( workspace->f_vdw[i], -CEvd, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_vdw[j], +CEvd, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_ele[i], -CEclmb, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_ele[j], +CEclmb, nbr_pj->dvec );
-#endif
- }
- }
-
- // fclose( fout );
-
- // fprintf( stderr, "nonbonded: ext_press (%24.15e %24.15e %24.15e)\n",
- // data->ext_press[0], data->ext_press[1], data->ext_press[2] );
-}
-
-
-/*
-
- GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *atoms,
- two_body_parameters *tbp,
- global_parameters g_p,
- control_params *control,
- simulation_data *data,
- list p_far_nbrs,
- real *E_vdW, real *E_Ele, rvec *aux_ext_press,
- int num_atom_types, int N )
- {
- int i, j, pj;
- int start_i, end_i;
- real self_coef;
- real p_vdW1, p_vdW1i;
- real powr_vdW1, powgi_vdW1;
- real tmp, r_ij, fn13, exp1, exp2;
- real Tap, dTap, dfn13, CEvd, CEclmb;
- real dr3gamij_1, dr3gamij_3;
- real e_ele, e_vdW, e_core, de_core;
- rvec temp, ext_press;
-// rtensor temp_rtensor, total_rtensor;
-two_body_parameters *twbp;
-far_neighbor_data *nbr_pj;
-list *far_nbrs = &p_far_nbrs;
-
-i = blockIdx.x * blockDim.x + threadIdx.x;
-if ( i >= N ) return;
-
-p_vdW1 = g_p.l[28];
-p_vdW1i = 1.0 / p_vdW1;
-e_ele = 0;
-e_vdW = 0;
-e_core = 0;
-de_core = 0;
-
-//for( i = 0; i < system->N; ++i ) {
-start_i = Start_Index(i, far_nbrs);
-end_i = End_Index(i, far_nbrs);
-// fprintf( stderr, "i: %d, start: %d, end: %d\n",
-// i, start_i, end_i );
-
-for( pj = start_i; pj < end_i; ++pj )
-if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
-nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
-j = nbr_pj->nbr;
-r_ij = nbr_pj->d;
-twbp = &(tbp[ index_tbp (atoms[i].type, atoms[j].type, num_atom_types) ]);
-self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
-
-//CHANGE ORIGINAL
-//if (i <= j) continue;
-//CHANGE ORIGINAL
-
-// Calculate Taper and its derivative
-// Tap = nbr_pj->Tap; -- precomputed during compte_H
-Tap = control->Tap7 * r_ij + control->Tap6;
-Tap = Tap * r_ij + control->Tap5;
-Tap = Tap * r_ij + control->Tap4;
-Tap = Tap * r_ij + control->Tap3;
-Tap = Tap * r_ij + control->Tap2;
-Tap = Tap * r_ij + control->Tap1;
-Tap = Tap * r_ij + control->Tap0;
-
-dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
-dTap = dTap * r_ij + 5*control->Tap5;
-dTap = dTap * r_ij + 4*control->Tap4;
-dTap = dTap * r_ij + 3*control->Tap3;
-dTap = dTap * r_ij + 2*control->Tap2;
-dTap += control->Tap1/r_ij;
-
-//vdWaals Calculations
-if(g_p.vdw_type==1 || g_p.vdw_type==3) {
- // shielding
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
- E_vdW [i] += e_vdW / 2.0;
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
- POW(r_ij, p_vdW1 - 2.0);
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) * dfn13 );
-}
-else{ // no shielding
- exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
-
- e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
- E_vdW [i] += e_vdW / 2.0;
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) );
-}
-
-if(g_p.vdw_type==2 || g_p.vdw_type==3) {
- // innner wall
- e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
- e_vdW = self_coef * Tap * e_core;
-
- //TODO check this
- E_vdW [i] += e_vdW / 2.0;
- //TODO check this
-
- de_core = -(twbp->acore/twbp->rcore) * e_core;
- CEvd += self_coef * ( dTap * e_core + Tap * de_core );
-}
-
-//Coulomb Calculations
-dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
-dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
-tmp = Tap / dr3gamij_3;
-//tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
-e_ele =
-self_coef * C_ele * atoms[i].q * atoms[j].q * tmp;
-E_Ele [i] += e_ele / 2.0;
-
-CEclmb = self_coef * C_ele * atoms[i].q * atoms[j].q *
-( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
-//CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
-// ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;
-
-if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- if (i >= j)
- rvec_ScaledAdd( atoms[i].f, -(CEvd+CEclmb), nbr_pj->dvec );
- else
- rvec_ScaledAdd( atoms[i].f, +(CEvd+CEclmb), nbr_pj->dvec );
-}
-else { // NPT, iNPT or sNPT
- // for pressure coupling, terms not related to bond order
- // derivatives are added directly into pressure vector/tensor
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
-
- if ( i >= j)
- rvec_ScaledAdd( atoms[i].f, -1., temp );
- else
- rvec_Add( atoms[i].f, temp );
-
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
-
- //rvec_Add( data->ext_press, ext_press );
- rvec_Copy (aux_ext_press[i], ext_press);
-
- //TODO CHECK THIS calculation here, it should be divided by two somehow.
-}
-}
-//}
-}
-
-*/
-
-
-
-
-GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *atoms,
- two_body_parameters *tbp,
- global_parameters g_p,
- control_params *control,
- simulation_data *data,
- list p_far_nbrs,
- real *E_vdW, real *E_Ele, rvec *aux_ext_press,
- int num_atom_types, int N )
-{
- extern __shared__ real _vdw[];
- extern __shared__ real _ele[];
- extern __shared__ rvec _force [];
-
- real *sh_vdw;
- real *sh_ele;
- rvec *sh_force;
-
- int i, j, pj;
- int start_i, end_i;
- real self_coef;
- real p_vdW1, p_vdW1i;
- real powr_vdW1, powgi_vdW1;
- real tmp, r_ij, fn13, exp1, exp2;
- real Tap, dTap, dfn13, CEvd, CEclmb;
- real dr3gamij_1, dr3gamij_3;
- real e_ele, e_vdW, e_core, de_core;
- rvec temp, ext_press;
- // rtensor temp_rtensor, total_rtensor;
- two_body_parameters *twbp;
- far_neighbor_data *nbr_pj;
- list *far_nbrs = &p_far_nbrs;
-
- int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
- int warpid = thread_id / VDW_THREADS_PER_ATOM;
- int laneid = thread_id & (VDW_THREADS_PER_ATOM -1);
-
- i = warpid;
-
- sh_vdw = _vdw;
- sh_ele = _vdw + blockDim.x;
- sh_force = (rvec *)( _vdw + 2*blockDim.x);
-
- sh_vdw[threadIdx.x] = 0.0;
- sh_ele[threadIdx.x] = 0.0;
- rvec_MakeZero ( sh_force [threadIdx.x] );
-
- if (i < N)
- {
-
- p_vdW1 = g_p.l[28];
- p_vdW1i = 1.0 / p_vdW1;
- e_ele = 0;
- e_vdW = 0;
- e_core = 0;
- de_core = 0;
-
- //for( i = 0; i < system->N; ++i ) {
- start_i = Start_Index(i, far_nbrs);
- end_i = End_Index(i, far_nbrs);
- // fprintf( stderr, "i: %d, start: %d, end: %d\n",
- // i, start_i, end_i );
-
- pj = start_i + laneid;
- //for( pj = start_i; pj < end_i; ++pj )
- while (pj < end_i)
- {
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- r_ij = nbr_pj->d;
- twbp = &(tbp[ index_tbp (atoms[i].type, atoms[j].type, num_atom_types) ]);
- self_coef = (i == j) ? 0.5 : 1.0; // for supporting small boxes!
-
- //CHANGE ORIGINAL
- //if (i <= j) continue;
- //CHANGE ORIGINAL
-
- // Calculate Taper and its derivative
- // Tap = nbr_pj->Tap; -- precomputed during compte_H
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
- dTap = dTap * r_ij + 5*control->Tap5;
- dTap = dTap * r_ij + 4*control->Tap4;
- dTap = dTap * r_ij + 3*control->Tap3;
- dTap = dTap * r_ij + 2*control->Tap2;
- dTap += control->Tap1/r_ij;
-
- //vdWaals Calculations
- if(g_p.vdw_type==1 || g_p.vdw_type==3) {
- // shielding
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
-
-
- //E_vdW [i] += e_vdW / 2.0;
- sh_vdw [threadIdx.x] += e_vdW/2.0;
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
- POW(r_ij, p_vdW1 - 2.0);
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) * dfn13 );
- }
- else{ // no shielding
- exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
-
- e_vdW = self_coef * Tap * twbp->D * (exp1 - 2.0 * exp2);
-
-
- //E_vdW [i] += e_vdW / 2.0;
- sh_vdw [threadIdx.x] += e_vdW/2.0;
-
- CEvd = self_coef * ( dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) *
- (exp1 - exp2) );
- }
-
- if(g_p.vdw_type==2 || g_p.vdw_type==3) {
- // innner wall
- e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
- e_vdW = self_coef * Tap * e_core;
-
- //TODO check this
- //E_vdW [i] += e_vdW / 2.0;
- sh_vdw [threadIdx.x] += e_vdW / 2.0;
- //TODO check this
-
- de_core = -(twbp->acore/twbp->rcore) * e_core;
- CEvd += self_coef * ( dTap * e_core + Tap * de_core );
- }
-
- //Coulomb Calculations
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
- tmp = Tap / dr3gamij_3;
- //tmp = Tap * nbr_pj->inv_dr3gamij_3; -- precomputed during compte_H
- e_ele =
- self_coef * C_ele * atoms[i].q * atoms[j].q * tmp;
-
- //E_Ele [i] += e_ele / 2.0;
- sh_ele [threadIdx.x] += e_ele / 2.0;
-
- CEclmb = self_coef * C_ele * atoms[i].q * atoms[j].q *
- ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
- //CEclmb = self_coef*C_ele*system->atoms[i].q*system->atoms[j].q*
- // ( dTap- Tap*r_ij*nbr_pj->inv_dr3gamij_1 )*nbr_pj->inv_dr3gamij_3;
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- if (i >= j){
- //rvec_ScaledAdd( atoms[i].f, -(CEvd+CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( sh_force[threadIdx.x], -(CEvd+CEclmb), nbr_pj->dvec );
- }
- else
- {
- //rvec_ScaledAdd( atoms[i].f, +(CEvd+CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( sh_force[threadIdx.x], +(CEvd+CEclmb), nbr_pj->dvec );
- }
- }
- else { // NPT, iNPT or sNPT
- // for pressure coupling, terms not related to bond order
- // derivatives are added directly into pressure vector/tensor
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
-
- if ( i >= j)
- {
- //rvec_ScaledAdd( atoms[i].f, -1., temp );
- rvec_ScaledAdd( sh_force[threadIdx.x], -1., temp );
- }
- else
- {
- //rvec_Add( atoms[i].f, temp );
- rvec_Add( sh_force[threadIdx.x], temp );
- }
-
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
-
- //rvec_Add( data->ext_press, ext_press );
- rvec_Copy (aux_ext_press[i], ext_press);
-
- //TODO CHECK THIS calculation here, it should be divided by two somehow.
- }
- } // if condition for far neighbors
-
-
- pj += VDW_THREADS_PER_ATOM;
-
- } // end of while loop for pj < end_i condition
- } // if (i < N ) condition
- //}
-
- __syncthreads ();
-
- if (laneid < 16) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
- }
- __syncthreads ();
- if (laneid < 8) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
- }
- __syncthreads ();
- if (laneid < 4) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
- }
- __syncthreads ();
- if (laneid < 2) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
- }
- __syncthreads ();
- if (laneid < 1) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
- }
- __syncthreads ();
- if (laneid == 0) {
- E_vdW [i] += sh_vdw[threadIdx.x];
- E_Ele [i] += sh_ele[threadIdx.x];
- rvec_Add (atoms[i].f, sh_force [ threadIdx.x ]);
- }
-
-
-}
-
-void LR_vdW_Coulomb( reax_system *system, control_params *control,
- int i, int j, real r_ij, LR_data *lr )
-{
- real p_vdW1 = system->reaxprm.gp.l[28];
- real p_vdW1i = 1.0 / p_vdW1;
- real powr_vdW1, powgi_vdW1;
- real tmp, fn13, exp1, exp2;
- real Tap, dTap, dfn13;
- real dr3gamij_1, dr3gamij_3;
- real e_core, de_core;
- two_body_parameters *twbp;
-
- twbp = &(system->reaxprm.tbp[ index_tbp (i,j,&system->reaxprm) ]);
- e_core = 0;
- de_core = 0;
-
- /* calculate taper and its derivative */
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dTap = 7*control->Tap7 * r_ij + 6*control->Tap6;
- dTap = dTap * r_ij + 5*control->Tap5;
- dTap = dTap * r_ij + 4*control->Tap4;
- dTap = dTap * r_ij + 3*control->Tap3;
- dTap = dTap * r_ij + 2*control->Tap2;
- dTap += control->Tap1/r_ij;
-
-
- /* vdWaals calculations */
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
- /* fprintf(stderr,"vdW: Tap:%f, r: %f, f13:%f, D:%f, Energy:%f,\
-Gamma_w:%f, p_vdw: %f, alpha: %f, r_vdw: %f, %lf %lf\n",
-Tap, r_ij, fn13, twbp->D, Tap * twbp->D * (exp1 - 2.0 * exp2),
-powgi_vdW1, p_vdW1, twbp->alpha, twbp->r_vdW, exp1, exp2); */
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) * POW(r_ij, p_vdW1 - 2.0);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
-
- /*vdWaals Calculations*/
- if(system->reaxprm.gp.vdw_type==1 || system->reaxprm.gp.vdw_type==3)
- { // shielding
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
- POW(r_ij, p_vdW1 - 2.0);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
- }
- else{ // no shielding
- exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
-
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2);
- }
-
- if(system->reaxprm.gp.vdw_type==2 || system->reaxprm.gp.vdw_type==3)
- { // innner wall
- e_core = twbp->ecore * EXP(twbp->acore * (1.0-(r_ij/twbp->rcore)));
- lr->e_vdW += Tap * e_core;
-
- de_core = -(twbp->acore/twbp->rcore) * e_core;
- lr->CEvd += dTap * e_core + Tap * de_core;
- }
-
- /* Coulomb calculations */
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
-
- tmp = Tap / dr3gamij_3;
- lr->H = EV_to_KCALpMOL * tmp;
- lr->e_ele = C_ele * tmp;
- /* fprintf( stderr,"i:%d(%d), j:%d(%d), gamma:%f,\
-Tap:%f, dr3gamij_3:%f, qi: %f, qj: %f\n",
-i, system->atoms[i].type, j, system->atoms[j].type,
-twbp->gamma, Tap, dr3gamij_3,
-system->atoms[i].q, system->atoms[j].q ); */
-
- lr->CEclmb = C_ele * ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
- /* fprintf( stdout, "%d %d\t%g\t%g %g\t%g %g\t%g %g\n",
- i+1, j+1, r_ij, e_vdW, CEvd * r_ij,
- system->atoms[i].q, system->atoms[j].q, e_ele, CEclmb * r_ij ); */
-
- /* fprintf( stderr,"LR_Lookup:%3d%3d%5.3f-%8.5f,%8.5f%8.5f,%8.5f%8.5f\n",
- i, j, r_ij, lr->H, lr->e_vdW, lr->CEvd, lr->e_ele, lr->CEclmb ); */
-}
-
-
-void Tabulated_vdW_Coulomb_Energy( reax_system *system, control_params *control,
- simulation_data *data,
- static_storage *workspace, list **lists,
- output_controls *out_control )
-{
- int i, j, pj, r, steps, update_freq, update_energies;
- int type_i, type_j, tmin, tmax;
- int start_i, end_i;
- real r_ij, self_coef, base, dif;
- real e_vdW, e_ele;
- real CEvd, CEclmb;
- rvec temp, ext_press;
- far_neighbor_data *nbr_pj;
- list *far_nbrs = (*lists) + FAR_NBRS;
- LR_lookup_table *t;
-
- steps = data->step - data->prev_steps;
- update_freq = out_control->energy_update_freq;
- update_energies = update_freq > 0 && steps % update_freq == 0;
-
- for( i = 0; i < system->N; ++i ) {
- type_i = system->atoms[i].type;
- start_i = Start_Index(i,far_nbrs);
- end_i = End_Index(i,far_nbrs);
-
- for( pj = start_i; pj < end_i; ++pj )
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut ) {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- type_j = system->atoms[j].type;
- r_ij = nbr_pj->d;
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( LR[ index_lr (tmin,tmax,system->reaxprm.num_atom_types) ] );
-
- /* Cubic Spline Interpolation */
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
- //fprintf(stderr, "r: %f, i: %d, base: %f, dif: %f\n", r, i, base, dif);
-
- if( update_energies ) {
- e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
- t->vdW[r].a;
- e_vdW *= self_coef;
-
- e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
- t->ele[r].a;
- e_ele *= self_coef * system->atoms[i].q * system->atoms[j].q;
-
- data->E_vdW += e_vdW;
- data->E_Ele += e_ele;
- }
-
- CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
- t->CEvd[r].a;
- CEvd *= self_coef;
- //CEvd = (3*t->vdW[r].d*dif + 2*t->vdW[r].c)*dif + t->vdW[r].b;
-
- CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
- t->CEclmb[r].a;
- CEclmb *= self_coef * system->atoms[i].q * system->atoms[j].q;
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- rvec_ScaledAdd( system->atoms[i].f, -(CEvd + CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( system->atoms[j].f, +(CEvd + CEclmb), nbr_pj->dvec );
- }
- else { // NPT, iNPT or sNPT
- /* for pressure coupling, terms not related to bond order
- derivatives are added directly into pressure vector/tensor */
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
- rvec_ScaledAdd( system->atoms[i].f, -1., temp );
- rvec_Add( system->atoms[j].f, temp );
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
- rvec_Add( data->ext_press, ext_press );
- }
-
-#ifdef TEST_ENERGY
- fprintf(out_control->evdw, "%6d%6d%24.15e%24.15e%24.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- r_ij, e_vdW, data->E_vdW );
- fprintf(out_control->ecou,"%6d%6d%24.15e%24.15e%24.15e%24.15e%24.15e\n",
- workspace->orig_id[i], workspace->orig_id[j],
- r_ij, system->atoms[i].q, system->atoms[j].q,
- e_ele, data->E_Ele );
-#endif
-#ifdef TEST_FORCES
- rvec_ScaledAdd( workspace->f_vdw[i], -CEvd, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_vdw[j], +CEvd, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_ele[i], -CEclmb, nbr_pj->dvec );
- rvec_ScaledAdd( workspace->f_ele[j], +CEclmb, nbr_pj->dvec );
-#endif
- }
- }
-}
-
-GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy( reax_atom *atoms,
- control_params *control,
- simulation_data *data,
- list p_far_nbrs,
- real *E_vdW, real *E_Ele, rvec *aux_ext_press,
- LR_lookup_table *d_LR,
- int num_atom_types,
- int energy_update_freq,
- int N )
-{
-
- extern __shared__ real _vdw[];
- extern __shared__ real _ele[];
- extern __shared__ rvec _force [];
-
- real *sh_vdw;
- real *sh_ele;
- rvec *sh_force;
-
- int i, j, pj, r, steps, update_freq, update_energies;
- int type_i, type_j, tmin, tmax;
- int start_i, end_i;
- real r_ij, self_coef, base, dif;
- real e_vdW, e_ele;
- real CEvd, CEclmb;
- rvec temp, ext_press;
- far_neighbor_data *nbr_pj;
- LR_lookup_table *t;
- list *far_nbrs = &p_far_nbrs;
-
- int thread_id = blockIdx.x * blockDim.x + threadIdx.x;
- int warpid = thread_id / VDW_THREADS_PER_ATOM;
- int laneid = thread_id & (VDW_THREADS_PER_ATOM -1);
-
- i = warpid;
-
- sh_vdw = _vdw;
- sh_ele = _vdw + blockDim.x;
- sh_force = (rvec *)( _vdw + 2*blockDim.x);
-
- sh_vdw[threadIdx.x] = 0.0;
- sh_ele[threadIdx.x] = 0.0;
- rvec_MakeZero ( sh_force [threadIdx.x] );
-
- if ( i < N )
- {
-
- reax_atom local_atom ;
- local_atom.q = atoms[i].q;
- //local_atom.q = d_far_data.q[i];
- local_atom.type = atoms[i].type;
- //local_atom.type = d_far_data.type[i];
-
- /*
- sh_vdw = _vdw;
- sh_ele = _vdw + warpid;
- sh_force = (rvec *)( _vdw + 2*warpid);
-
- sh_vdw[threadIdx.x] = 0.0;
- sh_ele[threadIdx.x] = 0.0;
- rvec_MakeZero ( sh_force [threadIdx.x] );
- */
-
-
- steps = data->step - data->prev_steps;
- update_freq = energy_update_freq;
- update_energies = update_freq > 0 && steps % update_freq == 0;
-
- //for( i = 0; i < system->N; ++i ) {
- type_i = local_atom.type;
- start_i = Start_Index(i,far_nbrs);
- end_i = End_Index(i,far_nbrs);
-
- pj = start_i + laneid;
-
- //for( pj = start_i; pj < end_i; ++pj )
- while (pj < end_i)
- {
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
- //if( d_far_data.d[pj] <= control->r_cut )
- {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- //j = d_far_data.nbrs[pj];
- type_j = atoms[j].type;
- //type_j = d_far_data.type[j];
- r_ij = nbr_pj->d;
- //r_ij = d_far_data.d[pj];
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
-
- //TODO
- //CHANGE ORIGINAL
- //if (i <= j) { pj += blockDim.x; continue; }
- //CHANGE ORIGINAL
-
- /* Cubic Spline Interpolation */
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
-
- if(( update_energies ))
- {
- e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
- t->vdW[r].a;
- e_vdW *= self_coef;
-
- e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif + t->ele[r].a;
- e_ele *= self_coef * local_atom.q * atoms[j].q;
-
-
- //data->E_vdW += e_vdW;
- //TODO
- //E_vdW [i] += e_vdW / 2.0;
- //E_vdW [i] = __dadd_rd (E_vdW [i], e_vdW/2.0);
- sh_vdw [threadIdx.x] += e_vdW/2.0;
- //E_vdW [i] += e_vdW;
-
- //TODO
- //data->E_Ele += e_ele;
- //E_Ele [i] += e_ele / 2.0;
- //E_Ele [i] = __dadd_rd ( E_Ele [i], e_ele / 2.0);
- sh_ele [threadIdx.x] += e_ele/2.0;
- //E_Ele [i] += e_ele;
- }
-
- CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
- t->CEvd[r].a;
- CEvd *= self_coef;
-
- CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
- t->CEclmb[r].a;
- CEclmb *= self_coef * local_atom.q * atoms[j].q;
- //CEclmb *= self_coef * local_atom.q * d_far_data.q[j];
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT) {
- if ( i >= j)
- //rvec_ScaledAdd( atoms[i].f, -(CEvd + CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), nbr_pj->dvec );
- //rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), d_far_data.dvec[pj] );
- else
- //rvec_ScaledAdd( atoms[i].f, +(CEvd + CEclmb), nbr_pj->dvec );
- rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), nbr_pj->dvec );
- //rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), d_far_data.dvec[pj] );
- }
- else { // NPT, iNPT or sNPT
- // for pressure coupling, terms not related to bond order
- // derivatives are added directly into pressure vector/tensor /
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
- if (i >= j)
- rvec_ScaledAdd( atoms[i].f, -1., temp );
- else
- rvec_Add( atoms[i].f, temp );
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
-
- //rvec_Add( data->ext_press, ext_press );
- rvec_Copy (aux_ext_press [i], ext_press );
-
- //TODO CHECK THIS
- }
-
-
-
- }
-
- pj += VDW_THREADS_PER_ATOM;
- }
-
- }// if i < n condition
-
- __syncthreads ();
-
- if (laneid < 16) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
- }
- __syncthreads ();
- if (laneid < 8) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
- }
- __syncthreads ();
- if (laneid < 4) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
- }
- __syncthreads ();
- if (laneid < 2) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
- }
- __syncthreads ();
- if (laneid < 1) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
- }
- __syncthreads ();
- if (laneid == 0) {
- E_vdW [i] += sh_vdw[threadIdx.x];
- E_Ele [i] += sh_ele[threadIdx.x];
- rvec_Add (atoms[i].f, sh_force [ threadIdx.x ]);
- }
-
-
- }
-
-
-
-
- GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_1( reax_atom *atoms,
- control_params *control,
- simulation_data *data,
- list p_far_nbrs,
- real *E_vdW, real *E_Ele, rvec *aux_ext_press,
- LR_lookup_table *d_LR,
- int num_atom_types,
- int energy_update_freq,
- int N )
- {
-
- extern __shared__ real _vdw[];
- extern __shared__ real _ele[];
-
- real *sh_vdw;
- real *sh_ele;
-
- int i, j, pj, r, steps, update_freq, update_energies;
- int type_i, type_j, tmin, tmax;
- int start_i, end_i;
- real r_ij, self_coef, base, dif;
- real e_vdW, e_ele;
- real CEvd, CEclmb;
- rvec temp, ext_press;
- far_neighbor_data *nbr_pj;
- LR_lookup_table *t;
- list *far_nbrs = &p_far_nbrs;
-
- i = blockIdx.x;
-
- reax_atom local_atom;
- local_atom.q = atoms[i].q;
- local_atom.type = atoms[i].type;
-
- sh_vdw = _vdw;
- sh_ele = _vdw + blockDim.x;
-
- sh_vdw[threadIdx.x] = 0.0;
- sh_ele[threadIdx.x] = 0.0;
-
-
- steps = data->step - data->prev_steps;
- update_freq = energy_update_freq;
- update_energies = update_freq > 0 && steps % update_freq == 0;
-
- type_i = local_atom.type;
- start_i = Start_Index(i,far_nbrs);
- end_i = End_Index(i,far_nbrs);
-
- pj = start_i + threadIdx.x;
-
- while (pj < end_i)
- {
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
- {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- type_j = atoms[j].type;
- r_ij = nbr_pj->d;
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
-
- /* Cubic Spline Interpolation */
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
-
- if(( update_energies ))
- {
- e_vdW = ((t->vdW[r].d*dif + t->vdW[r].c)*dif + t->vdW[r].b)*dif +
- t->vdW[r].a;
- e_vdW *= self_coef;
-
- e_ele = ((t->ele[r].d*dif + t->ele[r].c)*dif + t->ele[r].b)*dif +
- t->ele[r].a;
- e_ele *= self_coef * local_atom.q * atoms[j].q;
-
- sh_vdw [threadIdx.x] += e_vdW/2.0;
- sh_ele [threadIdx.x] += e_ele/2.0;
- }
- }
-
- pj += blockDim.x;
- }
-
- // now do a reduce inside the warp for E_vdW, E_Ele and force.
- if (threadIdx.x < 16) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 16];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 16];
- }
- if (threadIdx.x < 8) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 8];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 8];
- }
- if (threadIdx.x < 4) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 4];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 4];
- }
- if (threadIdx.x < 2) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 2];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 2];
- }
- if (threadIdx.x < 1) {
- sh_vdw[threadIdx.x] += sh_vdw[threadIdx.x + 1];
- sh_ele[threadIdx.x] += sh_ele[threadIdx.x + 1];
- }
- if (threadIdx.x == 0) {
- E_vdW [i] += sh_vdw[0];
- E_Ele [i] += sh_ele[0];
- }
-
- }
-
-
-
-
-
-
- GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_2( reax_atom *atoms,
- control_params *control,
- simulation_data *data,
- list p_far_nbrs,
- real *E_vdW, real *E_Ele, rvec *aux_ext_press,
- LR_lookup_table *d_LR,
- int num_atom_types,
- int energy_update_freq,
- int N )
- {
-
- extern __shared__ rvec _force [];
-
- rvec *sh_force;
-
- int i, j, pj, r, steps, update_freq, update_energies;
- int type_i, type_j, tmin, tmax;
- int start_i, end_i;
- real r_ij, self_coef, base, dif;
- real e_vdW, e_ele;
- real CEvd, CEclmb;
- rvec temp, ext_press;
- far_neighbor_data *nbr_pj;
- LR_lookup_table *t;
- list *far_nbrs = &p_far_nbrs;
-
- i = blockIdx.x;
-
- reax_atom local_atom;
- local_atom.q = atoms[i].q;
- local_atom.type = atoms[i].type;
-
- sh_force = _force;
- rvec_MakeZero ( sh_force [threadIdx.x] );
-
-
- steps = data->step - data->prev_steps;
- update_freq = energy_update_freq;
- update_energies = update_freq > 0 && steps % update_freq == 0;
-
- //for( i = 0; i < system->N; ++i ) {
- type_i = local_atom.type;
- start_i = Start_Index(i,far_nbrs);
- end_i = End_Index(i,far_nbrs);
-
- pj = start_i + threadIdx.x;
-
- while (pj < end_i)
- {
- if( far_nbrs->select.far_nbr_list[pj].d <= control->r_cut )
- {
- nbr_pj = &( far_nbrs->select.far_nbr_list[pj] );
- j = nbr_pj->nbr;
- type_j = atoms[j].type;
- r_ij = nbr_pj->d;
- self_coef = (i == j) ? 0.5 : 1.0;
- tmin = MIN( type_i, type_j );
- tmax = MAX( type_i, type_j );
- t = &( d_LR[ index_lr (tmin,tmax,num_atom_types) ] );
-
- /* Cubic Spline Interpolation */
- r = (int)(r_ij * t->inv_dx);
- if( r == 0 ) ++r;
- base = (real)(r+1) * t->dx;
- dif = r_ij - base;
-
- CEvd = ((t->CEvd[r].d*dif + t->CEvd[r].c)*dif + t->CEvd[r].b)*dif +
- t->CEvd[r].a;
- CEvd *= self_coef;
-
- CEclmb = ((t->CEclmb[r].d*dif+t->CEclmb[r].c)*dif+t->CEclmb[r].b)*dif +
- t->CEclmb[r].a;
- CEclmb *= self_coef * local_atom.q * atoms[j].q;
-
- if( control->ensemble == NVE || control->ensemble == NVT || control->ensemble == bNVT ) {
- if ( i >= j)
- rvec_ScaledAdd( sh_force [threadIdx.x], -(CEvd + CEclmb), nbr_pj->dvec );
- else
- rvec_ScaledAdd( sh_force [threadIdx.x], +(CEvd + CEclmb), nbr_pj->dvec );
- }
- else { // NPT, iNPT or sNPT
- // for pressure coupling, terms not related to bond order
- // derivatives are added directly into pressure vector/tensor /
- rvec_Scale( temp, CEvd + CEclmb, nbr_pj->dvec );
- if (i >= j)
- rvec_ScaledAdd( atoms[i].f, -1., temp );
- else
- rvec_Add( atoms[i].f, temp );
- rvec_iMultiply( ext_press, nbr_pj->rel_box, temp );
-
- rvec_Copy (aux_ext_press [i], ext_press );
- }
- }
-
- pj += blockDim.x;
- }
-
-
- // now do a reduce inside the warp for E_vdW, E_Ele and force.
- if (threadIdx.x < 16) {
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 16] );
- }
- if (threadIdx.x < 8) {
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 8] );
- }
- if (threadIdx.x < 4) {
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 4] );
- }
- if (threadIdx.x < 2) {
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 2] );
- }
- if (threadIdx.x < 1) {
- rvec_Add (sh_force [threadIdx.x], sh_force [threadIdx.x + 1] );
- }
- if (threadIdx.x == 0) {
- rvec_Add (atoms[i].f, sh_force [ 0 ]);
- }
-
-
- }
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-#if defined(OLD)
- /* Linear extrapolation */
- /*p = (r_ij * t->inv_dx;
- r = (int) p;
- prev = &( t->y[r] );
- next = &( t->y[r+1] );
-
- tmp = p - r;
- e_vdW = self_coef * (prev->e_vdW + tmp*(next->e_vdW - prev->e_vdW ));
- CEvd = self_coef * (prev->CEvd + tmp*(next->CEvd - prev->CEvd ));
-
- e_ele = self_coef * (prev->e_ele + tmp*(next->e_ele - prev->e_ele ));
- e_ele = e_ele * system->atoms[i].q * system->atoms[j].q;
- CEclmb = self_coef * (prev->CEclmb+tmp*(next->CEclmb - prev->CEclmb));
- CEclmb = CEclmb * system->atoms[i].q * system->atoms[j].q;*/
-#endif
diff --git a/PuReMD-GPU/src/two_body_interactions.h b/PuReMD-GPU/src/two_body_interactions.h
index 41483222a1d1e66eb21805f7e1659b5eecd3f40b..f689290e9e680ca62352b6f627737b8a8b006472 100644
--- a/PuReMD-GPU/src/two_body_interactions.h
+++ b/PuReMD-GPU/src/two_body_interactions.h
@@ -21,156 +21,19 @@
#ifndef __TWO_BODY_INTERACTIONS_H_
#define __TWO_BODY_INTERACTIONS_H_
-#include <mytypes.h>
-#include "index_utils.h"
+#include "mytypes.h"
-void Bond_Energy( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-void vdW_Coulomb_Energy( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-void LR_vdW_Coulomb( reax_system*, control_params*, int, int, real, LR_data* );
-void Tabulated_vdW_Coulomb_Energy( reax_system*, control_params*, simulation_data*,
- static_storage*, list**, output_controls* );
-
-//CUDA functions
-GLOBAL void Cuda_Bond_Energy ( reax_atom *, global_parameters , single_body_parameters *, two_body_parameters *,
- simulation_data *, static_storage , list , int , int, real *);
-
-GLOBAL void Cuda_vdW_Coulomb_Energy( reax_atom *,
- two_body_parameters *,
- global_parameters ,
- control_params *,
- simulation_data *,
- list ,
- real *, real *, rvec *,
- int , int );
-
-GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy ( reax_atom *, control_params *, simulation_data *,
- list , real *, real *, rvec *,
- LR_lookup_table *, int , int , int ) ;
-GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_1 ( reax_atom *, control_params *, simulation_data *,
- list , real *, real *, rvec *,
- LR_lookup_table *, int , int , int ) ;
-GLOBAL void Cuda_Tabulated_vdW_Coulomb_Energy_2 ( reax_atom *, control_params *, simulation_data *,
- list , real *, real *, rvec *,
- LR_lookup_table *, int , int , int ) ;
-
-HOST_DEVICE void LR_vdW_Coulomb( global_parameters , two_body_parameters *,
- control_params *, int , int , real , LR_data * , int);
-
-HOST_DEVICE inline void LR_vdW_Coulomb( global_parameters g_params, two_body_parameters *tbp,
- control_params *control,
- int i, int j, real r_ij, LR_data *lr, int num_atom_types )
-{
- real p_vdW1 = g_params.l[28];
- real p_vdW1i = 1.0 / p_vdW1;
- real powr_vdW1, powgi_vdW1;
- real tmp, fn13, exp1, exp2;
- real Tap, dTap, dfn13;
- real dr3gamij_1, dr3gamij_3;
- real e_core, de_core;
- two_body_parameters *twbp;
-
- twbp = &(tbp[ index_tbp (i, j, num_atom_types) ]);
- e_core = 0;
- de_core = 0;
-
- /* calculate taper and its derivative */
- Tap = control->Tap7 * r_ij + control->Tap6;
- Tap = Tap * r_ij + control->Tap5;
- Tap = Tap * r_ij + control->Tap4;
- Tap = Tap * r_ij + control->Tap3;
- Tap = Tap * r_ij + control->Tap2;
- Tap = Tap * r_ij + control->Tap1;
- Tap = Tap * r_ij + control->Tap0;
-
- dTap = 7 * control->Tap7 * r_ij + 6 * control->Tap6;
- dTap = dTap * r_ij + 5 * control->Tap5;
- dTap = dTap * r_ij + 4 * control->Tap4;
- dTap = dTap * r_ij + 3 * control->Tap3;
- dTap = dTap * r_ij + 2 * control->Tap2;
- dTap += control->Tap1 / r_ij;
-
-
- /* vdWaals calculations */
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
- /* fprintf(stderr,"vdW: Tap:%f, r: %f, f13:%f, D:%f, Energy:%f,\
- Gamma_w:%f, p_vdw: %f, alpha: %f, r_vdw: %f, %lf %lf\n",
- Tap, r_ij, fn13, twbp->D, Tap * twbp->D * (exp1 - 2.0 * exp2),
- powgi_vdW1, p_vdW1, twbp->alpha, twbp->r_vdW, exp1, exp2); */
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) * POW(r_ij, p_vdW1 - 2.0);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
-
- /*vdWaals Calculations*/
- if (g_params.vdw_type == 1 || g_params.vdw_type == 3)
- {
- // shielding
- powr_vdW1 = POW(r_ij, p_vdW1);
- powgi_vdW1 = POW( 1.0 / twbp->gamma_w, p_vdW1);
-
- fn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i );
- exp1 = EXP( twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - fn13 / twbp->r_vdW) );
-
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- dfn13 = POW( powr_vdW1 + powgi_vdW1, p_vdW1i - 1.0) *
- POW(r_ij, p_vdW1 - 2.0);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2) * dfn13;
- }
- else // no shielding
- {
- exp1 = EXP( twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
- exp2 = EXP( 0.5 * twbp->alpha * (1.0 - r_ij / twbp->r_vdW) );
-
- lr->e_vdW = Tap * twbp->D * (exp1 - 2.0 * exp2);
-
- lr->CEvd = dTap * twbp->D * (exp1 - 2.0 * exp2) -
- Tap * twbp->D * (twbp->alpha / twbp->r_vdW) * (exp1 - exp2);
- }
-
- if (g_params.vdw_type == 2 || g_params.vdw_type == 3)
- {
- // innner wall
- e_core = twbp->ecore * EXP(twbp->acore * (1.0 - (r_ij / twbp->rcore)));
- lr->e_vdW += Tap * e_core;
-
- de_core = -(twbp->acore / twbp->rcore) * e_core;
- lr->CEvd += dTap * e_core + Tap * de_core;
- }
+void Bond_Energy( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls* );
- /* Coulomb calculations */
- dr3gamij_1 = ( r_ij * r_ij * r_ij + twbp->gamma );
- dr3gamij_3 = POW( dr3gamij_1 , 0.33333333333333 );
+void vdW_Coulomb_Energy( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls* );
- tmp = Tap / dr3gamij_3;
- lr->H = EV_to_KCALpMOL * tmp;
- lr->e_ele = C_ele * tmp;
- /* fprintf( stderr,"i:%d(%d), j:%d(%d), gamma:%f,\
- Tap:%f, dr3gamij_3:%f, qi: %f, qj: %f\n",
- i, system->atoms[i].type, j, system->atoms[j].type,
- twbp->gamma, Tap, dr3gamij_3,
- system->atoms[i].q, system->atoms[j].q ); */
+void LR_vdW_Coulomb( reax_system*, control_params*, int, int, real, LR_data* );
- lr->CEclmb = C_ele * ( dTap - Tap * r_ij / dr3gamij_1 ) / dr3gamij_3;
- /* fprintf( stdout, "%d %d\t%g\t%g %g\t%g %g\t%g %g\n",
- i+1, j+1, r_ij, e_vdW, CEvd * r_ij,
- system->atoms[i].q, system->atoms[j].q, e_ele, CEclmb * r_ij ); */
+void Tabulated_vdW_Coulomb_Energy( reax_system*, control_params*, simulation_data*,
+ static_storage*, list**, output_controls* );
- /* fprintf( stderr,"LR_Lookup:%3d%3d%5.3f-%8.5f,%8.5f%8.5f,%8.5f%8.5f\n",
- i, j, r_ij, lr->H, lr->e_vdW, lr->CEvd, lr->e_ele, lr->CEclmb ); */
-}
#endif
diff --git a/PuReMD-GPU/src/validation.cu b/PuReMD-GPU/src/validation.cu
index f8261555a595a06dcbdfbffe5d62c7a0375c4c97..21cd2145e689621ee0b3827889b106ed7c05af7f 100644
--- a/PuReMD-GPU/src/validation.cu
+++ b/PuReMD-GPU/src/validation.cu
@@ -18,7 +18,6 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-
#include "validation.h"
#include "cuda_utils.h"
@@ -27,33 +26,37 @@
#include "sort.h"
#include "index_utils.h"
-bool check_zero (real p1, real p2)
+
+int check_zero (real p1, real p2)
{
if (abs (p1 - p2) >= GPU_TOLERANCE)
- return true;
+ return TRUE;
else
- return false;
+ return FALSE;
}
-bool check_zero (rvec p1, rvec p2)
+
+int check_zero (rvec p1, rvec p2)
{
if (((abs (p1[0] - p2[0])) >= GPU_TOLERANCE) ||
((abs (p1[1] - p2[1])) >= GPU_TOLERANCE) ||
((abs (p1[2] - p2[2])) >= GPU_TOLERANCE ))
- return true;
- else return false;
+ return TRUE;
+ else return FALSE;
}
-bool check_same (ivec p1, ivec p2)
+
+int check_same (ivec p1, ivec p2)
{
if ( (p1[0] == p2[0]) || (p1[1] == p2[1]) || (p1[2] == p2[2]) )
- return true;
+ return TRUE;
else
- return false;
+ return FALSE;
}
-bool validate_box (simulation_box *host, simulation_box *dev)
+
+int validate_box (simulation_box *host, simulation_box *dev)
{
simulation_box test;
@@ -62,14 +65,15 @@ bool validate_box (simulation_box *host, simulation_box *dev)
if (memcmp (&test, host, SIMULATION_BOX_SIZE)) {
fprintf (stderr, " Simulation box is not in synch between host and device \n");
- return false;
+ return FALSE;
}
fprintf (stderr, " Simulation box is in **synch** between host and device \n");
- return true;
+ return TRUE;
}
-bool validate_atoms (reax_system *system, list **lists)
+
+int validate_atoms (reax_system *system, list **lists)
{
int start, end, index, count, miscount;
@@ -154,9 +158,10 @@ bool validate_atoms (reax_system *system, list **lists)
//fprintf (stderr, "Reax Atoms DOES **match** between host and device --> %d miscount --> %d \n", count, miscount);
free (test);
- return true;
+ return TRUE;
}
+
void Print_Matrix( sparse_matrix *A )
{
int i, j;
@@ -170,6 +175,7 @@ void Print_Matrix( sparse_matrix *A )
}
}
+
void Print_Matrix_L( sparse_matrix *A )
{
int i, j;
@@ -184,7 +190,7 @@ void Print_Matrix_L( sparse_matrix *A )
}
-bool validate_sort_matrix (reax_system *system, static_storage *workspace)
+int validate_sort_matrix (reax_system *system, static_storage *workspace)
{
sparse_matrix test;
int index, count;
@@ -221,7 +227,7 @@ bool validate_sort_matrix (reax_system *system, static_storage *workspace)
}
-bool validate_sparse_matrix( reax_system *system, static_storage *workspace )
+int validate_sparse_matrix( reax_system *system, static_storage *workspace )
{
sparse_matrix test;
int index, count;
@@ -287,10 +293,10 @@ bool validate_sparse_matrix( reax_system *system, static_storage *workspace )
free (test.start);
free (test.end);
free (test.entries);
- return true;
+ return TRUE;
}
-bool validate_lu (static_storage *workspace)
+int validate_lu (static_storage *workspace)
{
sparse_matrix test;
int index, count;
@@ -354,7 +360,7 @@ bool validate_lu (static_storage *workspace)
}
//fprintf (stderr, "L and U match on device and host \n");
- return true;
+ return TRUE;
}
void print_sparse_matrix (reax_system *system, static_storage *workspace)
@@ -405,7 +411,7 @@ void print_sparse_matrix (reax_system *system, static_storage *workspace)
}
-bool validate_bonds (reax_system *system, static_storage *workspace, list **lists)
+int validate_bonds (reax_system *system, static_storage *workspace, list **lists)
{
int start, end, index, count, miscount;
int *d_start, *d_end;
@@ -601,10 +607,10 @@ bool validate_bonds (reax_system *system, static_storage *workspace, list **list
free (d_start);
free (d_end);
free (d_bond_data);
- return true;
+ return TRUE;
}
-bool validate_sym_dbond_indices (reax_system *system, static_storage *workspace, list **lists)
+int validate_sym_dbond_indices (reax_system *system, static_storage *workspace, list **lists)
{
int start, end, index, count, miscount;
int *d_start, *d_end;
@@ -660,10 +666,11 @@ bool validate_sym_dbond_indices (reax_system *system, static_storage *workspace,
free (d_start);
free (d_end);
free (d_bond_data);
- return true;
+ return TRUE;
}
-bool analyze_hbonds (reax_system *system, static_storage *workspace, list **lists)
+
+int analyze_hbonds (reax_system *system, static_storage *workspace, list **lists)
{
int hindex, nbr_hindex;
int pj, hj, hb_start_j, hb_end_j, j, nbr;
@@ -748,7 +755,7 @@ bool analyze_hbonds (reax_system *system, static_storage *workspace, list **list
}
-bool validate_hbonds (reax_system *system, static_storage *workspace, list **lists)
+int validate_hbonds (reax_system *system, static_storage *workspace, list **lists)
{
int *hbond_index, count;
int *d_start, *d_end, index, d_index;
@@ -858,10 +865,10 @@ bool validate_hbonds (reax_system *system, static_storage *workspace, list **lis
free (d_start);
free (d_end);
free (data);
- return true;
+ return TRUE;
}
-bool validate_neighbors (reax_system *system, list **lists)
+int validate_neighbors (reax_system *system, list **lists)
{
list *far_nbrs = *lists + FAR_NBRS;
list *d_nbrs = dev_lists + FAR_NBRS;
@@ -989,971 +996,975 @@ bool validate_neighbors (reax_system *system, list **lists)
start[i], end[i]);
exit (10);
}
+ }
+
+ //fprintf (stderr, "FAR Neighbors match between device and host \n");
+ free (start);
+ free (end);
+ free (data);
+ return TRUE;
+}
+
+
+int validate_workspace (reax_system *system, static_storage *workspace, list **lists)
+{
+ real *total_bond_order;
+ int count, tcount;
+
+ total_bond_order = (real *) malloc ( system->N * REAL_SIZE );
+ copy_host_device (total_bond_order, dev_workspace->total_bond_order, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+
+ count = 0;
+ for (int i = 0; i < system->N; i++) {
+
+ //if (abs (workspace->total_bond_order[i] - total_bond_order[i]) >= GPU_TOLERANCE){
+ if ( check_zero (workspace->total_bond_order[i], total_bond_order[i])){
+ fprintf (stderr, "Total bond order does not match for atom %d (%4.15e %4.15e)\n",
+ i, workspace->total_bond_order[i], total_bond_order[i]);
+ exit (-1);
+ count ++;
}
+ }
+ free (total_bond_order);
+ //fprintf (stderr, "TOTAL Bond Order mismatch count %d\n", count);
- //fprintf (stderr, "FAR Neighbors match between device and host \n");
- free (start);
- free (end);
- free (data);
- return true;
+
+ rvec *dDeltap_self;
+ dDeltap_self = (rvec *) calloc (system->N, RVEC_SIZE);
+ copy_host_device (dDeltap_self, dev_workspace->dDeltap_self, system->N * RVEC_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+
+ count = 0;
+ for (int i = 0; i < system->N; i++ )
+ {
+ if (check_zero (workspace->dDeltap_self[i], dDeltap_self[i]))
+ {
+ fprintf (stderr, "index: %d c (%f %f %f) g (%f %f %f )\n", i,
+ workspace->dDeltap_self[i][0],
+ workspace->dDeltap_self[i][1],
+ workspace->dDeltap_self[i][2],
+ dDeltap_self[3*i+0],
+ dDeltap_self[3*i+1],
+ dDeltap_self[3*i+2] );
+ exit (-1);
+ count ++;
}
+ }
+ free (dDeltap_self);
+ //fprintf (stderr, "dDeltap_self mismatch count %d\n", count);
- bool validate_workspace (reax_system *system, static_storage *workspace, list **lists)
+ //exit for init_forces
+
+ real *test;
+ test = (real *) malloc (system->N * REAL_SIZE);
+
+ copy_host_device (test, dev_workspace->Deltap, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ )
+ {
+ if (check_zero (workspace->Deltap[i], test[i]))
{
- real *total_bond_order;
- int count, tcount;
+ fprintf (stderr, "Deltap: Mismatch index --> %d (%f %f) \n", i, workspace->Deltap[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Deltap mismatch count %d\n", count);
- total_bond_order = (real *) malloc ( system->N * REAL_SIZE );
- copy_host_device (total_bond_order, dev_workspace->total_bond_order, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (test, dev_workspace->Deltap_boc, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ )
+ {
+ if (check_zero (workspace->Deltap_boc[i], test[i]))
+ {
+ fprintf (stderr, "Deltap_boc: Mismatch index --> %d (%f %f) \n", i, workspace->Deltap_boc[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "dDeltap_boc mismatch count %d\n", count);
- count = 0;
- for (int i = 0; i < system->N; i++) {
+ copy_host_device (test, dev_workspace->Delta, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->Delta[i], test[i])) {
+ fprintf (stderr, "Delta: Mismatch index --> %d (%f %f) \n", i, workspace->Delta[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Delta mismatch count %d\n", count);
- //if (abs (workspace->total_bond_order[i] - total_bond_order[i]) >= GPU_TOLERANCE){
- if ( check_zero (workspace->total_bond_order[i], total_bond_order[i])){
- fprintf (stderr, "Total bond order does not match for atom %d (%4.15e %4.15e)\n",
- i, workspace->total_bond_order[i], total_bond_order[i]);
- exit (-1);
- count ++;
- }
- }
- free (total_bond_order);
- //fprintf (stderr, "TOTAL Bond Order mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->Delta_e, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->Delta_e[i], test[i])) {
+ fprintf (stderr, "Delta_e: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_e[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Delta_e mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->vlpex, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->vlpex[i], test[i])) {
+ fprintf (stderr, "vlpex: Mismatch index --> %d (%f %f) \n", i, workspace->vlpex[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "vlpex mismatch count %d\n", count);
- rvec *dDeltap_self;
- dDeltap_self = (rvec *) calloc (system->N, RVEC_SIZE);
- copy_host_device (dDeltap_self, dev_workspace->dDeltap_self, system->N * RVEC_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (test, dev_workspace->nlp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->nlp[i], test[i])) {
+ fprintf (stderr, "nlp: Mismatch index --> %d (%f %f) \n", i, workspace->nlp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "nlp mismatch count %d\n", count);
- count = 0;
- for (int i = 0; i < system->N; i++ )
- {
- if (check_zero (workspace->dDeltap_self[i], dDeltap_self[i]))
- {
- fprintf (stderr, "index: %d c (%f %f %f) g (%f %f %f )\n", i,
- workspace->dDeltap_self[i][0],
- workspace->dDeltap_self[i][1],
- workspace->dDeltap_self[i][2],
- dDeltap_self[3*i+0],
- dDeltap_self[3*i+1],
- dDeltap_self[3*i+2] );
- exit (-1);
- count ++;
- }
- }
- free (dDeltap_self);
- //fprintf (stderr, "dDeltap_self mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->Delta_lp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->Delta_lp[i], test[i])) {
+ fprintf (stderr, "Delta_lp: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_lp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Delta_lp mismatch count %d\n", count);
- //exit for init_forces
+ copy_host_device (test, dev_workspace->Clp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->Clp[i], test[i])) {
+ fprintf (stderr, "Clp: Mismatch index --> %d (%f %f) \n", i, workspace->Clp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Clp mismatch count %d\n", count);
- real *test;
- test = (real *) malloc (system->N * REAL_SIZE);
+ copy_host_device (test, dev_workspace->dDelta_lp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->dDelta_lp[i], test[i])) {
+ fprintf (stderr, "dDelta_lp: Mismatch index --> %d (%f %f) \n", i, workspace->dDelta_lp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "dDelta_lp mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->Deltap, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ )
- {
- if (check_zero (workspace->Deltap[i], test[i]))
- {
- fprintf (stderr, "Deltap: Mismatch index --> %d (%f %f) \n", i, workspace->Deltap[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Deltap mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->nlp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->nlp_temp[i], test[i])) {
+ fprintf (stderr, "nlp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->nlp_temp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "nlp_temp mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->Deltap_boc, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ )
- {
- if (check_zero (workspace->Deltap_boc[i], test[i]))
- {
- fprintf (stderr, "Deltap_boc: Mismatch index --> %d (%f %f) \n", i, workspace->Deltap_boc[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "dDeltap_boc mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->Delta_lp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->Delta_lp_temp[i], test[i])) {
+ fprintf (stderr, "Delta_lp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_lp_temp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "Delta_lp_temp mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->Delta, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->Delta[i], test[i])) {
- fprintf (stderr, "Delta: Mismatch index --> %d (%f %f) \n", i, workspace->Delta[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Delta mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->dDelta_lp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->dDelta_lp_temp[i], test[i])) {
+ fprintf (stderr, "dDelta_lp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->dDelta_lp_temp[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "dDelta_lp_temp mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->Delta_e, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->Delta_e[i], test[i])) {
- fprintf (stderr, "Delta_e: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_e[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Delta_e mismatch count %d\n", count);
+ //exit for Bond order calculations
- copy_host_device (test, dev_workspace->vlpex, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->vlpex[i], test[i])) {
- fprintf (stderr, "vlpex: Mismatch index --> %d (%f %f) \n", i, workspace->vlpex[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "vlpex mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->nlp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->nlp[i], test[i])) {
- fprintf (stderr, "nlp: Mismatch index --> %d (%f %f) \n", i, workspace->nlp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "nlp mismatch count %d\n", count);
+ copy_host_device (test, dev_workspace->CdDelta, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->CdDelta[i], test[i])) {
+ fprintf (stderr, " CdDelta does NOT match (%f %f) for atom %d \n", workspace->CdDelta[i], test[i], i);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "CdDelta mismatch count %d\n", count);
+ //exit for Bond Energy calculations
- copy_host_device (test, dev_workspace->Delta_lp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->Delta_lp[i], test[i])) {
- fprintf (stderr, "Delta_lp: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_lp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Delta_lp mismatch count %d\n", count);
+ /*
+ copy_host_device (test, dev_workspace->droptol, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->droptol[i], test[i])) {
+ fprintf (stderr, " Droptol Does not match (%f %f) \n", workspace->droptol[i], test[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "droptol mismatch count %d\n", count);
+ */
- copy_host_device (test, dev_workspace->Clp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->Clp[i], test[i])) {
- fprintf (stderr, "Clp: Mismatch index --> %d (%f %f) \n", i, workspace->Clp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Clp mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->dDelta_lp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->dDelta_lp[i], test[i])) {
- fprintf (stderr, "dDelta_lp: Mismatch index --> %d (%f %f) \n", i, workspace->dDelta_lp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "dDelta_lp mismatch count %d\n", count);
+ //exit for QEa calculations
+ /*
+ real *t_s;
- copy_host_device (test, dev_workspace->nlp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->nlp_temp[i], test[i])) {
- fprintf (stderr, "nlp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->nlp_temp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "nlp_temp mismatch count %d\n", count);
+ t_s = (real *) malloc (REAL_SIZE * (system->N * 2) );
+ copy_host_device (t_s, dev_workspace->b_prm, REAL_SIZE * (system->N * 2), cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device (test, dev_workspace->Delta_lp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->Delta_lp_temp[i], test[i])) {
- fprintf (stderr, "Delta_lp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->Delta_lp_temp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "Delta_lp_temp mismatch count %d\n", count);
+ count = 0;
+ for (int i = 0; i < (system->N * 2); i++ ) {
+ if (check_zero (workspace->b_prm[i], t_s[i])) {
+ fprintf (stderr, " (%f %f) \n", workspace->b_prm[i], t_s[i]);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "b_prm mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->dDelta_lp_temp, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->dDelta_lp_temp[i], test[i])) {
- fprintf (stderr, "dDelta_lp_temp: Mismatch index --> %d (%f %f) \n", i, workspace->dDelta_lp_temp[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "dDelta_lp_temp mismatch count %d\n", count);
+ t_s = (real *) malloc (REAL_SIZE * 5 * system->N);
+ copy_host_device (t_s, dev_workspace->s, system->N * REAL_SIZE * 5, cudaMemcpyDeviceToHost, __LINE__);
- //exit for Bond order calculations
+ count = 0;
+ for (int i = 0; i < 5*system->N; i++ ) {
+ if (check_zero (workspace->s[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) @ index %d \n", workspace->s[i], t_s[i], i);
+ count ++;
+ }
+ }
+ fprintf (stderr, "s mismatch count %d\n", count);
- copy_host_device (test, dev_workspace->CdDelta, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->CdDelta[i], test[i])) {
- fprintf (stderr, " CdDelta does NOT match (%f %f) for atom %d \n", workspace->CdDelta[i], test[i], i);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "CdDelta mismatch count %d\n", count);
- //exit for Bond Energy calculations
+ t_s = (real *) malloc (REAL_SIZE * 5 * system->N);
+ copy_host_device (t_s, dev_workspace->t, system->N * REAL_SIZE * 5, cudaMemcpyDeviceToHost, __LINE__);
- /*
- copy_host_device (test, dev_workspace->droptol, system->N * REAL_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->droptol[i], test[i])) {
- fprintf (stderr, " Droptol Does not match (%f %f) \n", workspace->droptol[i], test[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "droptol mismatch count %d\n", count);
- */
+ count = 0;
+ for (int i = 0; i < 5*system->N; i++ ) {
+ if (check_zero (workspace->t[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) @ index : %d\n", workspace->t[i], t_s[i], i);
+ count ++;
+ }
+ }
+ fprintf (stderr, "t mismatch count %d\n", count);
- //exit for QEa calculations
- /*
- real *t_s;
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) * system->N);
+ copy_host_device (t_s, dev_workspace->v, system->N * REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
- t_s = (real *) malloc (REAL_SIZE * (system->N * 2) );
- copy_host_device (t_s, dev_workspace->b_prm, REAL_SIZE * (system->N * 2), cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < (RESTART + 1)*system->N; i++ ) {
+ if (check_zero (workspace->v[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) @ index %d \n", workspace->v[i], t_s[i], i);
+ count ++;
+ }
+ }
+ fprintf (stderr, "v mismatch count %d\n", count);
- count = 0;
- for (int i = 0; i < (system->N * 2); i++ ) {
- if (check_zero (workspace->b_prm[i], t_s[i])) {
- fprintf (stderr, " (%f %f) \n", workspace->b_prm[i], t_s[i]);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "b_prm mismatch count %d\n", count);
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
+ copy_host_device (t_s, dev_workspace->y, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
- t_s = (real *) malloc (REAL_SIZE * 5 * system->N);
- copy_host_device (t_s, dev_workspace->s, system->N * REAL_SIZE * 5, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < (RESTART + 1); i++ ) {
+ if (check_zero (workspace->y[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) \n", workspace->y[i], t_s[i]);
+ count ++;
+ }
+ }
+ fprintf (stderr, "y mismatch count %d\n", count);
- count = 0;
- for (int i = 0; i < 5*system->N; i++ ) {
- if (check_zero (workspace->s[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) @ index %d \n", workspace->s[i], t_s[i], i);
- count ++;
- }
- }
- fprintf (stderr, "s mismatch count %d\n", count);
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
+ copy_host_device (t_s, dev_workspace->hc, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < (RESTART + 1); i++ ) {
+ if (check_zero (workspace->hc[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) \n", workspace->hc[i], t_s[i]);
+ count ++;
+ }
+ }
+ fprintf (stderr, "hc mismatch count %d\n", count);
- t_s = (real *) malloc (REAL_SIZE * 5 * system->N);
- copy_host_device (t_s, dev_workspace->t, system->N * REAL_SIZE * 5, cudaMemcpyDeviceToHost, __LINE__);
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
+ copy_host_device (t_s, dev_workspace->hs, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < 5*system->N; i++ ) {
- if (check_zero (workspace->t[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) @ index : %d\n", workspace->t[i], t_s[i], i);
+ count = 0;
+ for (int i = 0; i < (RESTART + 1); i++ ) {
+ if (check_zero (workspace->hs[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) \n", workspace->hs[i], t_s[i]);
count ++;
- }
- }
- fprintf (stderr, "t mismatch count %d\n", count);
-
+ }
+ }
+ fprintf (stderr, "hs mismatch count %d\n", count);
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) * system->N);
- copy_host_device (t_s, dev_workspace->v, system->N * REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) * (RESTART+1) );
+ copy_host_device (t_s, dev_workspace->h, REAL_SIZE * (RESTART+1)*(RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < (RESTART + 1)*system->N; i++ ) {
- if (check_zero (workspace->v[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) @ index %d \n", workspace->v[i], t_s[i], i);
+ count = 0;
+ for (int i = 0; i < (RESTART+1)*(RESTART+1); i++ ) {
+ if (check_zero (workspace->h[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) \n", workspace->h[i], t_s[i]);
count ++;
- }
- }
- fprintf (stderr, "v mismatch count %d\n", count);
+ }
+ }
+ fprintf (stderr, "h mismatch count %d\n", count);
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
- copy_host_device (t_s, dev_workspace->y, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
+ copy_host_device (t_s, dev_workspace->g, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < (RESTART + 1); i++ ) {
- if (check_zero (workspace->y[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) \n", workspace->y[i], t_s[i]);
+ count = 0;
+ for (int i = 0; i < (RESTART + 1); i++ ) {
+ if (check_zero (workspace->g[i], t_s[i])) {
+ //fprintf (stderr, " (%f %f) @ index %d\n", workspace->g[i], t_s[i], i);
count ++;
- }
- }
- fprintf (stderr, "y mismatch count %d\n", count);
+ }
+ }
+ fprintf (stderr, "g mismatch count %d\n", count);
+ */
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
- copy_host_device (t_s, dev_workspace->hc, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ rvec *r_s = (rvec *) malloc (RVEC_SIZE * system->N );
+ copy_host_device (r_s, dev_workspace->v_const, RVEC_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- count = 0;
- for (int i = 0; i < (RESTART + 1); i++ ) {
- if (check_zero (workspace->hc[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) \n", workspace->hc[i], t_s[i]);
- count ++;
- }
- }
- fprintf (stderr, "hc mismatch count %d\n", count);
+ count = 0;
+ for (int i = 0; i < system->N; i++ ) {
+ if (check_zero (workspace->v_const[i], r_s[i])) {
+ fprintf (stderr, " v_const (%f %f %f) (%f %f %f) @ index %d\n",
+ workspace->v_const[i][0],
+ workspace->v_const[i][1],
+ workspace->v_const[i][2],
+ r_s[i][0],
+ r_s[i][1],
+ r_s[i][2],
+ i);
+ exit (-1);
+ count ++;
+ }
+ }
+ //fprintf (stderr, "v_const mismatch count %d\n", count);
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
- copy_host_device (t_s, dev_workspace->hs, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ free (test);
+ free (r_s);
+ return TRUE;
+}
- count = 0;
- for (int i = 0; i < (RESTART + 1); i++ ) {
- if (check_zero (workspace->hs[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) \n", workspace->hs[i], t_s[i]);
- count ++;
- }
- }
- fprintf (stderr, "hs mismatch count %d\n", count);
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) * (RESTART+1) );
- copy_host_device (t_s, dev_workspace->h, REAL_SIZE * (RESTART+1)*(RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+int validate_data (reax_system *system, simulation_data *host)
+{
+ simulation_data device;
- count = 0;
- for (int i = 0; i < (RESTART+1)*(RESTART+1); i++ ) {
- if (check_zero (workspace->h[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) \n", workspace->h[i], t_s[i]);
- count ++;
- }
- }
- fprintf (stderr, "h mismatch count %d\n", count);
+ copy_host_device (&device, host->d_simulation_data, SIMULATION_DATA_SIZE, cudaMemcpyDeviceToHost, __LINE__);
- t_s = (real *) malloc (REAL_SIZE * (RESTART+1) );
- copy_host_device (t_s, dev_workspace->g, REAL_SIZE * (RESTART+1), cudaMemcpyDeviceToHost, __LINE__);
+ if (check_zero (host->E_BE, device.E_BE)){
+ fprintf (stderr, "E_BE does not match (%4.15e %4.15e) \n", host->E_BE, device.E_BE);
+ exit (-1);
+ }
- count = 0;
- for (int i = 0; i < (RESTART + 1); i++ ) {
- if (check_zero (workspace->g[i], t_s[i])) {
- //fprintf (stderr, " (%f %f) @ index %d\n", workspace->g[i], t_s[i], i);
- count ++;
- }
- }
- fprintf (stderr, "g mismatch count %d\n", count);
- */
-
- rvec *r_s = (rvec *) malloc (RVEC_SIZE * system->N );
- copy_host_device (r_s, dev_workspace->v_const, RVEC_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
-
- count = 0;
- for (int i = 0; i < system->N; i++ ) {
- if (check_zero (workspace->v_const[i], r_s[i])) {
- fprintf (stderr, " v_const (%f %f %f) (%f %f %f) @ index %d\n",
- workspace->v_const[i][0],
- workspace->v_const[i][1],
- workspace->v_const[i][2],
- r_s[i][0],
- r_s[i][1],
- r_s[i][2],
- i);
- exit (-1);
- count ++;
- }
- }
- //fprintf (stderr, "v_const mismatch count %d\n", count);
+ if (check_zero (host->E_Lp, device.E_Lp)){
+ fprintf (stderr, "E_Lp does not match (%4.10e %4.10e) \n", host->E_Lp, device.E_Lp);
+ exit (-1);
+ }
- free (test);
- free (r_s);
- return true;
- }
+ if (check_zero (host->E_Ov, device.E_Ov)){
+ fprintf (stderr, "E_Ov does not match (%4.10e %4.10e) \n", host->E_Ov, device.E_Ov);
+ exit (-1);
+ }
- bool validate_data (reax_system *system, simulation_data *host)
- {
- simulation_data device;
+ if (check_zero (host->E_Un, device.E_Un)){
+ fprintf (stderr, "E_Un does not match (%4.10e %4.10e) \n", host->E_Un, device.E_Un);
+ exit (-1);
+ }
- copy_host_device (&device, host->d_simulation_data, SIMULATION_DATA_SIZE, cudaMemcpyDeviceToHost, __LINE__);
+ if (check_zero (host->E_Tor, device.E_Tor)) {
+ fprintf (stderr, "E_Tor does not match (%4.10e %4.10e) \n", host->E_Tor, device.E_Tor);
+ exit (-1);
+ }
- if (check_zero (host->E_BE, device.E_BE)){
- fprintf (stderr, "E_BE does not match (%4.15e %4.15e) \n", host->E_BE, device.E_BE);
- exit (-1);
- }
+ if (check_zero (host->E_Con, device.E_Con)) {
+ fprintf (stderr, "E_Con does not match (%4.10e %4.10e) \n", host->E_Con, device.E_Con);
+ exit (-1);
+ }
- if (check_zero (host->E_Lp, device.E_Lp)){
- fprintf (stderr, "E_Lp does not match (%4.10e %4.10e) \n", host->E_Lp, device.E_Lp);
- exit (-1);
- }
+ if (check_zero (host->ext_press, device.ext_press)) {
+ fprintf (stderr, "ext_press does not match (%4.10e %4.10e) \n", host->ext_press, device.ext_press);
+ exit (-1);
+ }
- if (check_zero (host->E_Ov, device.E_Ov)){
- fprintf (stderr, "E_Ov does not match (%4.10e %4.10e) \n", host->E_Ov, device.E_Ov);
- exit (-1);
- }
+ if (check_zero (host->E_HB, device.E_HB)) {
+ fprintf (stderr, "E_Hb does not match (%4.10e %4.10e) \n", host->E_HB, device.E_HB);
+ exit (-1);
+ }
- if (check_zero (host->E_Un, device.E_Un)){
- fprintf (stderr, "E_Un does not match (%4.10e %4.10e) \n", host->E_Un, device.E_Un);
- exit (-1);
- }
+ if (check_zero (host->E_Ang, device.E_Ang)) {
+ fprintf (stderr, "E_Ang does not match (%4.10e %4.10e) \n", host->E_Ang, device.E_Ang);
+ exit (-1);
+ }
- if (check_zero (host->E_Tor, device.E_Tor)) {
- fprintf (stderr, "E_Tor does not match (%4.10e %4.10e) \n", host->E_Tor, device.E_Tor);
- exit (-1);
- }
+ if (check_zero (host->E_Pen, device.E_Pen)) {
+ fprintf (stderr, "E_Pen does not match (%4.10e %4.10e) \n", host->E_Pen, device.E_Pen);
+ exit (-1);
+ }
- if (check_zero (host->E_Con, device.E_Con)) {
- fprintf (stderr, "E_Con does not match (%4.10e %4.10e) \n", host->E_Con, device.E_Con);
- exit (-1);
- }
+ if (check_zero (host->E_Coa, device.E_Coa)) {
+ fprintf (stderr, "E_Coa does not match (%4.10e %4.10e) \n", host->E_Coa, device.E_Coa);
+ exit (-1);
+ }
- if (check_zero (host->ext_press, device.ext_press)) {
- fprintf (stderr, "ext_press does not match (%4.10e %4.10e) \n", host->ext_press, device.ext_press);
- exit (-1);
- }
+ if (check_zero (host->E_vdW, device.E_vdW)) {
+ fprintf (stderr, "E_vdW does not match (%4.20e %4.20e) \n", host->E_vdW, device.E_vdW);
+ exit (-1);
+ }
- if (check_zero (host->E_HB, device.E_HB)) {
- fprintf (stderr, "E_Hb does not match (%4.10e %4.10e) \n", host->E_HB, device.E_HB);
- exit (-1);
- }
+ if (check_zero (host->E_Ele, device.E_Ele)) {
+ fprintf (stderr, "E_Ele does not match (%4.20e %4.20e) \n", host->E_Ele, device.E_Ele);
+ exit (-1);
+ }
- if (check_zero (host->E_Ang, device.E_Ang)) {
- fprintf (stderr, "E_Ang does not match (%4.10e %4.10e) \n", host->E_Ang, device.E_Ang);
- exit (-1);
- }
+ if (check_zero (host->E_Pol, device.E_Pol)) {
+ fprintf (stderr, "E_Pol does not match (%4.10e %4.10e) \n", host->E_Pol, device.E_Pol);
+ exit (-1);
+ }
- if (check_zero (host->E_Pen, device.E_Pen)) {
- fprintf (stderr, "E_Pen does not match (%4.10e %4.10e) \n", host->E_Pen, device.E_Pen);
- exit (-1);
- }
- if (check_zero (host->E_Coa, device.E_Coa)) {
- fprintf (stderr, "E_Coa does not match (%4.10e %4.10e) \n", host->E_Coa, device.E_Coa);
- exit (-1);
- }
+ //fprintf (stderr, "Simulation Data match between host and device \n");
+ return TRUE;
+}
- if (check_zero (host->E_vdW, device.E_vdW)) {
- fprintf (stderr, "E_vdW does not match (%4.20e %4.20e) \n", host->E_vdW, device.E_vdW);
- exit (-1);
- }
- if (check_zero (host->E_Ele, device.E_Ele)) {
- fprintf (stderr, "E_Ele does not match (%4.20e %4.20e) \n", host->E_Ele, device.E_Ele);
- exit (-1);
- }
+void print_bond_data (bond_order_data *s)
+{
+ /*
+ fprintf (stderr, "Bond_Order_Data BO (%f ) BO_s (%f ) BO_pi (%f ) BO_pi2 (%f ) ",
+ s->BO,
+ s->BO_s,
+ s->BO_pi,
+ s->BO_pi2 );
+ */
+ fprintf (stderr, " Cdbo (%e) ", s->Cdbo );
+ fprintf (stderr, " Cdbopi (%e) ", s->Cdbopi );
+ fprintf (stderr, " Cdbopi2 (%e) ", s->Cdbopi2 );
+}
- if (check_zero (host->E_Pol, device.E_Pol)) {
- fprintf (stderr, "E_Pol does not match (%4.10e %4.10e) \n", host->E_Pol, device.E_Pol);
- exit (-1);
- }
+void print_bond_list (reax_system *system, static_storage *workspace, list **lists)
+{
+ list *bonds = *lists + BONDS;
- //fprintf (stderr, "Simulation Data match between host and device \n");
- return true;
- }
+ for (int i = 1; i < 2; i++)
+ {
+ fprintf (stderr, "Atom %d Bond_data ( nbrs \n", i);
+ for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
+ {
+ bond_data *data = &bonds->select.bond_list [j];
+ fprintf (stderr, " %d, ", data->nbr );
+ print_bond_data (&data->bo_data);
+ fprintf (stderr, ")\n");
+ }
+ }
- void print_bond_data (bond_order_data *s)
- {
- /*
- fprintf (stderr, "Bond_Order_Data BO (%f ) BO_s (%f ) BO_pi (%f ) BO_pi2 (%f ) ",
- s->BO,
- s->BO_s,
- s->BO_pi,
- s->BO_pi2 );
- */
- fprintf (stderr, " Cdbo (%e) ", s->Cdbo );
- fprintf (stderr, " Cdbopi (%e) ", s->Cdbopi );
- fprintf (stderr, " Cdbopi2 (%e) ", s->Cdbopi2 );
- }
+ int *b_start = (int *) malloc (INT_SIZE * system->N);
+ int *b_end = (int *) malloc (INT_SIZE * system->N);
+ list *d_bonds = dev_lists + BONDS;
+ bond_data *d_bond_data;
- void print_bond_list (reax_system *system, static_storage *workspace, list **lists)
- {
- list *bonds = *lists + BONDS;
+ d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
- for (int i = 1; i < 2; i++)
- {
- fprintf (stderr, "Atom %d Bond_data ( nbrs \n", i);
- for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
- {
- bond_data *data = &bonds->select.bond_list [j];
- fprintf (stderr, " %d, ", data->nbr );
- print_bond_data (&data->bo_data);
- fprintf (stderr, ")\n");
- }
- }
+ copy_host_device ( b_start, d_bonds->index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( b_end, d_bonds->end_index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ for (int i = 0; i < 2; i++)
+ {
+ fprintf (stderr, "Atom %d Bond_data ( nbrs \n", i);
+ for (int j = b_start[i]; j < b_end[i]; j ++) {
+ bond_data *src = &d_bond_data[j];
+ fprintf (stderr, " %d, ", src->nbr );
+ print_bond_data (&src->bo_data);
+ fprintf (stderr, ")\n");
+ }
+ }
+}
- int *b_start = (int *) malloc (INT_SIZE * system->N);
- int *b_end = (int *) malloc (INT_SIZE * system->N);
- list *d_bonds = dev_lists + BONDS;
- bond_data *d_bond_data;
- d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
+void count_three_bodies (reax_system *system, static_storage *workspace, list **lists)
+{
+ list *three = *lists + THREE_BODIES;
+ list *bonds = *lists + BONDS;
- copy_host_device ( b_start, d_bonds->index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( b_end, d_bonds->end_index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 0; i < 2; i++)
- {
- fprintf (stderr, "Atom %d Bond_data ( nbrs \n", i);
- for (int j = b_start[i]; j < b_end[i]; j ++) {
- bond_data *src = &d_bond_data[j];
- fprintf (stderr, " %d, ", src->nbr );
- print_bond_data (&src->bo_data);
- fprintf (stderr, ")\n");
- }
- }
- }
+ list *d_three = dev_lists + THREE_BODIES;
+ list *d_bonds = dev_lists + BONDS;
+ bond_data *d_bond_data;
+ real *test;
+
+ three_body_interaction_data *data = (three_body_interaction_data *)
+ malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
+ int *start = (int *) malloc (INT_SIZE * system->num_bonds);
+ int *end = (int *) malloc (INT_SIZE * system->num_bonds);
+
+ int *b_start = (int *) malloc (INT_SIZE * system->N);
+ int *b_end = (int *) malloc (INT_SIZE * system->N);
+ int count;
+ int hcount, dcount;
+
+ copy_host_device ( start, d_three->index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( end, d_three->end_index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( data, d_three->select.three_body_list,
+ sizeof (three_body_interaction_data) * system->num_thbodies,
+ cudaMemcpyDeviceToHost, __LINE__);
+ d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
+ copy_host_device ( b_start, d_bonds->index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( b_end, d_bonds->end_index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- void count_three_bodies (reax_system *system, static_storage *workspace, list **lists)
- {
- list *three = *lists + THREE_BODIES;
- list *bonds = *lists + BONDS;
-
- list *d_three = dev_lists + THREE_BODIES;
- list *d_bonds = dev_lists + BONDS;
- bond_data *d_bond_data;
- real *test;
-
- three_body_interaction_data *data = (three_body_interaction_data *)
- malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
- int *start = (int *) malloc (INT_SIZE * system->num_bonds);
- int *end = (int *) malloc (INT_SIZE * system->num_bonds);
-
- int *b_start = (int *) malloc (INT_SIZE * system->N);
- int *b_end = (int *) malloc (INT_SIZE * system->N);
- int count;
- int hcount, dcount;
-
- copy_host_device ( start, d_three->index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( end, d_three->end_index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( data, d_three->select.three_body_list,
- sizeof (three_body_interaction_data) * system->num_thbodies,
- cudaMemcpyDeviceToHost, __LINE__);
-
- d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
-
- copy_host_device ( b_start, d_bonds->index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( b_end, d_bonds->end_index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
-
- count = 0;
- hcount = dcount = 0;
- for (int i = 0; i < system->N; i++)
- {
- for (int j = b_start[i]; j < b_end[i]; j ++) {
- dcount += end[j] - start[j];
- }
- }
+ count = 0;
+ hcount = dcount = 0;
+ for (int i = 0; i < system->N; i++)
+ {
+ for (int j = b_start[i]; j < b_end[i]; j ++) {
+ dcount += end[j] - start[j];
+ }
+ }
- fprintf (stderr, "Total Actual Three Body Count ---> %d \n", dcount);
+ fprintf (stderr, "Total Actual Three Body Count ---> %d \n", dcount);
- free (data);
- free (start);
- free (end);
- free (b_start);
- free (b_end);
- free (d_bond_data);
- }
+ free (data);
+ free (start);
+ free (end);
+ free (b_start);
+ free (b_end);
+ free (d_bond_data);
+}
+int validate_three_bodies (reax_system *system, static_storage *workspace, list **lists)
+{
+ list *three = *lists + THREE_BODIES;
+ list *bonds = *lists + BONDS;
- bool validate_three_bodies (reax_system *system, static_storage *workspace, list **lists)
- {
- list *three = *lists + THREE_BODIES;
- list *bonds = *lists + BONDS;
+ list *d_three = dev_lists + THREE_BODIES;
+ list *d_bonds = dev_lists + BONDS;
+ bond_data *d_bond_data;
+ real *test;
- list *d_three = dev_lists + THREE_BODIES;
- list *d_bonds = dev_lists + BONDS;
- bond_data *d_bond_data;
- real *test;
+ three_body_interaction_data *data = (three_body_interaction_data *)
+ malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
+ int *start = (int *) malloc (INT_SIZE * system->num_bonds);
+ int *end = (int *) malloc (INT_SIZE * system->num_bonds);
- three_body_interaction_data *data = (three_body_interaction_data *)
- malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
- int *start = (int *) malloc (INT_SIZE * system->num_bonds);
- int *end = (int *) malloc (INT_SIZE * system->num_bonds);
+ int *b_start = (int *) malloc (INT_SIZE * system->N);
+ int *b_end = (int *) malloc (INT_SIZE * system->N);
+ int count;
+ int hcount, dcount;
- int *b_start = (int *) malloc (INT_SIZE * system->N);
- int *b_end = (int *) malloc (INT_SIZE * system->N);
- int count;
- int hcount, dcount;
+ copy_host_device ( start, d_three->index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( end, d_three->end_index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( data, d_three->select.three_body_list,
+ sizeof (three_body_interaction_data) * system->num_thbodies,
+ cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( start, d_three->index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( end, d_three->end_index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( data, d_three->select.three_body_list,
- sizeof (three_body_interaction_data) * system->num_thbodies,
- cudaMemcpyDeviceToHost, __LINE__);
+ d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
- d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
+ copy_host_device ( b_start, d_bonds->index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( b_end, d_bonds->end_index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( b_start, d_bonds->index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( b_end, d_bonds->end_index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ //test = (real *) malloc (REAL_SIZE * system->num_bonds);
+ //memset (test, 0, REAL_SIZE * system->num_bonds);
+ //copy_host_device (test, testdata, REAL_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- //test = (real *) malloc (REAL_SIZE * system->num_bonds);
- //memset (test, 0, REAL_SIZE * system->num_bonds);
- //copy_host_device (test, testdata, REAL_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ count = 0;
+ for (int i = 0; i < system->N; i++)
+ {
+ //for (int j = bonds->index[i]; j < bonds->end_index[i]; j ++)
- count = 0;
- for (int i = 0; i < system->N; i++)
- {
- //for (int j = bonds->index[i]; j < bonds->end_index[i]; j ++)
+ hcount = dcount = 0;
+ for (int j = b_start[i]; j < b_end[i]; j ++) {
+ dcount += end[j] - start[j];
+ hcount += Num_Entries (j, three);
- hcount = dcount = 0;
- for (int j = b_start[i]; j < b_end[i]; j ++) {
- dcount += end[j] - start[j];
- hcount += Num_Entries (j, three);
+ /*
+ if ((end[j] - start[j]) != (End_Index (j, three) - Start_Index (j, three)))
+ {
+ fprintf (stderr, " Three body count does not match between host and device\n");
+ fprintf (stderr, " Host count : (%d, %d)\n", Start_Index (j, three), End_Index (j, three));
+ fprintf (stderr, " Device count: (%d, %d)\n", start[j], end[j]);
+ }
+ */
+ }
- /*
- if ((end[j] - start[j]) != (End_Index (j, three) - Start_Index (j, three)))
- {
- fprintf (stderr, " Three body count does not match between host and device\n");
- fprintf (stderr, " Host count : (%d, %d)\n", Start_Index (j, three), End_Index (j, three));
- fprintf (stderr, " Device count: (%d, %d)\n", start[j], end[j]);
- }
- */
- }
+ if ((dcount != hcount)) {
- if ((dcount != hcount)) {
+ fprintf (stderr, " Three body count does not match for the bond %d - %d \n", hcount, dcount);
- fprintf (stderr, " Three body count does not match for the bond %d - %d \n", hcount, dcount);
+ for (int j = b_start[i]; j < b_end[i]; j ++) {
+ bond_order_data *src = &d_bond_data[j].bo_data;
+ dcount = end[j] - start[j];
+ hcount = Num_Entries (j, three);
+ fprintf (stderr, "device \n");
+ print_bond_data (src);
- for (int j = b_start[i]; j < b_end[i]; j ++) {
- bond_order_data *src = &d_bond_data[j].bo_data;
- dcount = end[j] - start[j];
- hcount = Num_Entries (j, three);
- fprintf (stderr, "device \n");
- print_bond_data (src);
+ fprintf (stderr, "\n");
+ src = &bonds->select.bond_list[j].bo_data;
+ fprintf (stderr, "host \n");
+ print_bond_data (src);
+ fprintf (stderr, "\n");
- fprintf (stderr, "\n");
- src = &bonds->select.bond_list[j].bo_data;
- fprintf (stderr, "host \n");
- print_bond_data (src);
- fprintf (stderr, "\n");
+ //fprintf (stderr, "--- Device bo is %f \n", test[j]);
+ fprintf (stderr, "Device %d %d bonds (%d %d) - Host %d %d bonds (%d %d) \n", start[j], end[j],b_start[i], b_end[i],
+ Start_Index (j, three), End_Index (j, three), Start_Index (i, bonds), End_Index (i, bonds));
+ fprintf (stderr, "Host %d Device %d -- atom %d index %d \n", hcount, dcount, i, j);
+ fprintf (stderr, "------\n");
+ }
+ fprintf (stderr, " Three Bodies count does not match between host and device \n");
+ exit (-1);
+ }
+ }
- //fprintf (stderr, "--- Device bo is %f \n", test[j]);
- fprintf (stderr, "Device %d %d bonds (%d %d) - Host %d %d bonds (%d %d) \n", start[j], end[j],b_start[i], b_end[i],
- Start_Index (j, three), End_Index (j, three), Start_Index (i, bonds), End_Index (i, bonds));
- fprintf (stderr, "Host %d Device %d -- atom %d index %d \n", hcount, dcount, i, j);
- fprintf (stderr, "------\n");
- }
- fprintf (stderr, " Three Bodies count does not match between host and device \n");
- exit (-1);
- }
- }
+ //fprintf (stderr, "Three body count on DEVICE %d HOST %d \n", dcount, hcount);
- //fprintf (stderr, "Three body count on DEVICE %d HOST %d \n", dcount, hcount);
+ count = 0;
+ for (int i = 0; i < system->N; i++)
+ {
+ int x, y, z;
+ for (x = b_start[i]; x < b_end[i]; x++)
+ {
+ int t_start = start[x];
+ int t_end = end[x];
- count = 0;
- for (int i = 0; i < system->N; i++)
+ bond_data *dev_bond = &d_bond_data [x];
+ bond_data *host_bond;
+ for (z = Start_Index (i, bonds); z < End_Index (i, bonds); z++)
+ {
+ host_bond = &bonds->select.bond_list [z];
+ if ((dev_bond->nbr == host_bond->nbr) &&
+ check_same (dev_bond->rel_box, host_bond->rel_box) &&
+ !check_zero (dev_bond->dvec, host_bond->dvec) &&
+ !check_zero (dev_bond->d, host_bond->d) )
{
- int x, y, z;
- for (x = b_start[i]; x < b_end[i]; x++)
- {
- int t_start = start[x];
- int t_end = end[x];
-
- bond_data *dev_bond = &d_bond_data [x];
- bond_data *host_bond;
- for (z = Start_Index (i, bonds); z < End_Index (i, bonds); z++)
- {
- host_bond = &bonds->select.bond_list [z];
- if ((dev_bond->nbr == host_bond->nbr) &&
- check_same (dev_bond->rel_box, host_bond->rel_box) &&
- !check_zero (dev_bond->dvec, host_bond->dvec) &&
- !check_zero (dev_bond->d, host_bond->d) )
- {
- break;
- }
- }
- if (z >= End_Index (i, bonds)){
- fprintf (stderr, "Could not find the matching bond on host and device \n");
- exit (-1);
- }
-
- //find this bond in the bonds on the host side.
-
- for (y = t_start; y < t_end; y++)
- {
-
- three_body_interaction_data *device = data + y;
- three_body_interaction_data *host;
-
- //fprintf (stderr, "Device thb %d pthb %d \n", device->thb, device->pthb);
-
- int xx;
- for (xx = Start_Index (z, three); xx < End_Index (z, three); xx++)
- {
- host = &three->select.three_body_list [xx];
- //fprintf (stderr, "Host thb %d pthb %d \n", host->thb, host->pthb);
- //if ((host->thb == device->thb) && (host->pthb == device->pthb))
- if ((host->thb == device->thb) && !check_zero (host->theta, device->theta))
- {
- count ++;
- break;
- }
- }
-
- if ( xx >= End_Index (z, three) ) {
- fprintf (stderr, " Could not match for atom %d bonds %d (%d) Three body(%d %d) (%d %d) \n", i, x, z,
- Start_Index (z, three), End_Index (z, three), start[x], end[x] );
- exit (-1);
- }// else fprintf (stderr, "----------------- \n");
- }
- }
+ break;
}
- free (data);
- free (start);
- free (end);
- free (b_start);
- free (b_end);
- free (d_bond_data);
-
- //fprintf (stderr, "Three Body Interaction Data MATCH on device and HOST --> %d \n", count);
- return true;
}
+ if (z >= End_Index (i, bonds)){
+ fprintf (stderr, "Could not find the matching bond on host and device \n");
+ exit (-1);
+ }
+
+ //find this bond in the bonds on the host side.
- bool bin_three_bodies (reax_system *system, static_storage *workspace, list **lists)
+ for (y = t_start; y < t_end; y++)
{
- list *d_three = dev_lists + THREE_BODIES;
- list *d_bonds = dev_lists + BONDS;
- list *three = *lists + THREE_BODIES;
- list *bonds = *lists + BONDS;
- bond_data *d_bond_data;
-
- three_body_interaction_data *data = (three_body_interaction_data *)
- malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
- int *start = (int *) malloc (INT_SIZE * system->num_bonds);
- int *end = (int *) malloc (INT_SIZE * system->num_bonds);
-
- int *b_start = (int *) malloc (INT_SIZE * system->N);
- int *b_end = (int *) malloc (INT_SIZE * system->N);
-
- int *a = (int *) malloc (2 * INT_SIZE * system->N );
- int *b = (int *) malloc (2 * INT_SIZE * system->N );
- int *c = (int *) malloc (2 * INT_SIZE * system->N );
- int *d = (int *) malloc (2 * INT_SIZE * system->N );
-
- for (int i = 0; i < 2 * system->N; i++)
- a[i] = b[i] = c[i] = d[i] = -1;
-
- int count;
- int hcount, dcount;
- int index_a, index_b, index_c, index_d;
- index_a = index_b = index_c = index_d = 0;
-
- copy_host_device ( start, d_three->index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( end, d_three->end_index,
- INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( data, d_three->select.three_body_list,
- sizeof (three_body_interaction_data) * system->num_thbodies,
- cudaMemcpyDeviceToHost, __LINE__);
-
- d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
-
- copy_host_device ( b_start, d_bonds->index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device ( b_end, d_bonds->end_index,
- INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
- copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
-
- count = 0;
- hcount = dcount = 0;
-
- /*
- for (int i = 0; i < 20; i++)
- {
- for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
- {
- for ( int k = Start_Index (j, three); k < End_Index (j, three); k ++)
- {
- three_body_interaction_data *host = &three->select.three_body_list [k];
- fprintf (stderr, " atom %d bond (%d %d) -- %d, (%d %d)\n",
- i, Start_Index (i, bonds), End_Index (i, bonds), j, host->thb, host->pthb );
-
- }
- }
- }
- exit (-1);
- */
-
- count = 0;
- for (int i = 0; i < system->N; i++)
- {
- for (int j = b_start[i]; j < b_end[i]; j ++) {
- /*
- bond_data *src;
- src = &d_bond_data[j];
- fprintf (stderr, " atom %d Neighbor %d \n", i, src->nbr );
- */
+ three_body_interaction_data *device = data + y;
+ three_body_interaction_data *host;
- for (int x = start[j]; x < end[j]; x ++)
- {
- three_body_interaction_data *device = data + x;
-
- int center = device->j;
- int d_i = device->i;
- int d_k = device->k;
-
-
- //fprintf (stderr, " atom %d bond (%d %d) -- %d, (%d %d %d) -- (%d %d)\n",
- //i, b_start[i], b_end[i], j, center, d_i, d_k, device->thb, device->pthb);
-
- if ((a[system->N + center] != -1)) {
- a[d_i] = a[d_k] = 1;
- continue;
- } else if ((b[system->N + center] != -1)) {
- b[d_i] = b[d_k] = 1;
- continue;
- } else if ((c[system->N + center] != -1)) {
- c[d_i] = c[d_k] = 1;
- continue;
- } else if ((d[system->N + center] != -1)) {
- d[d_i] = d[d_k] = 1;
- continue;
- }
-
- if ((a[center] == -1) && (a[d_i] == -1) && (a[d_k] == -1)) {
- a[center] = a[d_i] = a[d_k] = 1;
- a[system->N + center] = 1;
- } else if ((b[center] == -1) && (b[d_i] == -1) && (b[d_k] == -1)) {
- b[center] = b[d_i] = b[d_k] = 1;
- b[system->N + center] = 1;
- } else if ((c[center] == -1) && (c[d_i] == -1) && (c[d_k] == -1)) {
- c[center] = c[d_i] = c[d_k] = 1;
- c[system->N + center] = 1;
- } else if ((d[center] == -1) && (d[d_i] == -1) && (d[d_k] == -1)) {
- d[center] = d[d_i] = d[d_k] = 1;
- d[system->N + center]= 1;
- }
- else {
- count ++;
- break;
- fprintf (stderr, "We have a problem with the four bins atom %d bond (%d %d) -- %d, (%d %d %d)\n",
- i, b_start[i], b_end[i], j, center, d_i, d_k);
- fprintf (stderr, "A's contents %d %d %d (%d %d %d)\n",
- a[system->N + center], a[system->N + d_i], a[system->N + d_k], a[center], a[d_i], a[d_k]);
- fprintf (stderr, "B's contents %d %d %d (%d %d %d)\n",
- b[system->N + center], b[system->N + d_i], b[system->N + d_k], b[center], b[d_i], b[d_k]);
- fprintf (stderr, "C's contents %d %d %d (%d %d %d)\n",
- c[system->N + center], c[system->N + d_i], c[system->N + d_k], c[center], c[d_i], c[d_k]);
- fprintf (stderr, "D's contents %d %d %d (%d %d %d)\n",
- d[system->N + center], d[system->N + d_i], d[system->N + d_k], d[center], d[d_i], d[d_k]);
-
- }
- }
- }
- }
- fprintf (stderr, "Miscount is %d \n", count);
- exit (-1);
+ //fprintf (stderr, "Device thb %d pthb %d \n", device->thb, device->pthb);
- count = 0;
- for (int i = 0; i < system->N; i++)
+ int xx;
+ for (xx = Start_Index (z, three); xx < End_Index (z, three); xx++)
{
- if (a[system->N + i] != -1) count ++;
- if (b[system->N + i] != -1) count ++;
- if (c[system->N + i] != -1) count ++;
- if (d[system->N + i] != -1) count ++;
+ host = &three->select.three_body_list [xx];
+ //fprintf (stderr, "Host thb %d pthb %d \n", host->thb, host->pthb);
+ //if ((host->thb == device->thb) && (host->pthb == device->pthb))
+ if ((host->thb == device->thb) && !check_zero (host->theta, device->theta))
+ {
+ count ++;
+ break;
+ }
}
- fprintf (stderr, "binned so many atoms --> %d \n", count );
+ if ( xx >= End_Index (z, three) ) {
+ fprintf (stderr, " Could not match for atom %d bonds %d (%d) Three body(%d %d) (%d %d) \n", i, x, z,
+ Start_Index (z, three), End_Index (z, three), start[x], end[x] );
+ exit (-1);
+ }// else fprintf (stderr, "----------------- \n");
}
+ }
+ }
+ free (data);
+ free (start);
+ free (end);
+ free (b_start);
+ free (b_end);
+ free (d_bond_data);
- bool validate_grid (reax_system *system)
- {
- int total = system->g.ncell[0] * system->g.ncell[1] * system->g.ncell[2];
- int count = 0;
+ //fprintf (stderr, "Three Body Interaction Data MATCH on device and HOST --> %d \n", count);
+ return TRUE;
+}
- int *dtop = (int *) malloc (INT_SIZE * total );
- copy_host_device (dtop, system->d_g.top, INT_SIZE * total, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 0; i < total; i++){
- if (system->g.top[i] != dtop[i]){
- fprintf (stderr, " top count does not match (%d %d) @ index %d \n", system->g.top[i], dtop[i], i );
- exit (-1);
- }
- }
- free (dtop);
+int bin_three_bodies (reax_system *system, static_storage *workspace, list **lists)
+{
+ list *d_three = dev_lists + THREE_BODIES;
+ list *d_bonds = dev_lists + BONDS;
+ list *three = *lists + THREE_BODIES;
+ list *bonds = *lists + BONDS;
+ bond_data *d_bond_data;
- int *datoms = (int *) malloc (INT_SIZE * total * system->d_g.max_atoms);
- copy_host_device (datoms, system->d_g.atoms, INT_SIZE * total * system->d_g.max_atoms, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 0; i < total*system->d_g.max_atoms; i++){
- if (system->g.atoms[i] != datoms[i]){
- fprintf (stderr, " atoms count does not match (%d %d) @ index %d \n", system->g.atoms[i], datoms[i], i );
- exit (-1);
- }
- }
- free (datoms);
+ three_body_interaction_data *data = (three_body_interaction_data *)
+ malloc ( sizeof (three_body_interaction_data) * system->num_thbodies);
+ int *start = (int *) malloc (INT_SIZE * system->num_bonds);
+ int *end = (int *) malloc (INT_SIZE * system->num_bonds);
- ivec *dnbrs = (ivec *) malloc (IVEC_SIZE * total * system->d_g.max_nbrs);
- copy_host_device (dnbrs, system->d_g.nbrs, IVEC_SIZE * total * system->d_g.max_nbrs, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 0; i < total*system->d_g.max_nbrs; i++){
- if (!check_same (system->g.nbrs[i], dnbrs[i])){
- fprintf (stderr, " nbrs count does not match @ index %d \n", i );
- exit (-1);
- }
- }
- free (dnbrs);
+ int *b_start = (int *) malloc (INT_SIZE * system->N);
+ int *b_end = (int *) malloc (INT_SIZE * system->N);
- rvec *dnbrs_cp = (rvec *) malloc (RVEC_SIZE * total * system->d_g.max_nbrs);
- copy_host_device (dnbrs_cp, system->d_g.nbrs_cp, RVEC_SIZE * total * system->d_g.max_nbrs, cudaMemcpyDeviceToHost, __LINE__);
- for (int i = 0; i < total*system->d_g.max_nbrs; i++){
- if (check_zero (system->g.nbrs_cp[i], dnbrs_cp[i])){
- fprintf (stderr, " nbrs_cp count does not match @ index %d \n", i );
- exit (-1);
- }
- }
- free (dnbrs_cp);
+ int *a = (int *) malloc (2 * INT_SIZE * system->N );
+ int *b = (int *) malloc (2 * INT_SIZE * system->N );
+ int *c = (int *) malloc (2 * INT_SIZE * system->N );
+ int *d = (int *) malloc (2 * INT_SIZE * system->N );
- //fprintf (stderr, " Grid match between device and host \n");
- return true;
- }
+ for (int i = 0; i < 2 * system->N; i++)
+ a[i] = b[i] = c[i] = d[i] = -1;
- void print_atoms (reax_system *system)
- {
- int start, end, index;
+ int count;
+ int hcount, dcount;
+ int index_a, index_b, index_c, index_d;
+ index_a = index_b = index_c = index_d = 0;
- reax_atom *test = (reax_atom *) malloc (REAX_ATOM_SIZE * system->N);
- copy_host_device (test, system->d_atoms, REAX_ATOM_SIZE * system->N, cudaMemcpyDeviceToHost, RES_SYSTEM_ATOMS );
+ copy_host_device ( start, d_three->index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( end, d_three->end_index,
+ INT_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( data, d_three->select.three_body_list,
+ sizeof (three_body_interaction_data) * system->num_thbodies,
+ cudaMemcpyDeviceToHost, __LINE__);
- //for (int i = 0; i < system->N; i++)
- for (int i = 0; i < 10; i++)
- {
- fprintf (stderr, "Atom:%d: Type:%d", i, test[i].type);
- fprintf (stderr, " x(%6.10f %6.10f %6.10f)", test[i].x[0], test[i].x[1], test[i].x[2] );
- fprintf (stderr, " v(%6.10f %6.10f %6.10f)", test[i].v[0], test[i].v[1], test[i].v[2] );
- fprintf (stderr, " f(%6.10f %6.10f %6.10f)", test[i].f[0], test[i].f[1], test[i].f[2] );
- fprintf (stderr, " q(%6.10f) \n", test[i].q );
- }
- }
+ d_bond_data = (bond_data *) malloc (BOND_DATA_SIZE * system->num_bonds );
+
+ copy_host_device ( b_start, d_bonds->index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device ( b_end, d_bonds->end_index,
+ INT_SIZE * system->N, cudaMemcpyDeviceToHost, __LINE__);
+ copy_host_device (d_bond_data, d_bonds->select.bond_list, BOND_DATA_SIZE * system->num_bonds, cudaMemcpyDeviceToHost, __LINE__);
+
+ count = 0;
+ hcount = dcount = 0;
+
+ /*
+ for (int i = 0; i < 20; i++)
+ {
+ for (int j = Start_Index (i, bonds); j < End_Index (i, bonds); j++)
+ {
+ for ( int k = Start_Index (j, three); k < End_Index (j, three); k ++)
+ {
+ three_body_interaction_data *host = &three->select.three_body_list [k];
+ fprintf (stderr, " atom %d bond (%d %d) -- %d, (%d %d)\n",
+ i, Start_Index (i, bonds), End_Index (i, bonds), j, host->thb, host->pthb );
+
+ }
+ }
+ }
+ exit (-1);
+ */
+
+ count = 0;
+ for (int i = 0; i < system->N; i++)
+ {
+ for (int j = b_start[i]; j < b_end[i]; j ++) {
- void print_sys_atoms (reax_system *system)
+ /*
+ bond_data *src;
+ src = &d_bond_data[j];
+ fprintf (stderr, " atom %d Neighbor %d \n", i, src->nbr );
+ */
+
+ for (int x = start[j]; x < end[j]; x ++)
{
- for (int i = 0; i < 10; i++)
- {
- fprintf (stderr, "Atom:%d: Type:%d", i, system->atoms[i].type);
- fprintf (stderr, " x(%6.10f %6.10f %6.10f)",system->atoms[i].x[0], system->atoms[i].x[1], system->atoms[i].x[2] );
- fprintf (stderr, " v(%6.10f %6.10f %6.10f)",system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2] );
- fprintf (stderr, " f(%6.10f %6.10f %6.10f)", system->atoms[i].f[0], system->atoms[i].f[1], system->atoms[i].f[2] );
- fprintf (stderr, " q(%6.10f) \n", system->atoms[i].q );
+ three_body_interaction_data *device = data + x;
+
+ int center = device->j;
+ int d_i = device->i;
+ int d_k = device->k;
+
+
+ //fprintf (stderr, " atom %d bond (%d %d) -- %d, (%d %d %d) -- (%d %d)\n",
+ //i, b_start[i], b_end[i], j, center, d_i, d_k, device->thb, device->pthb);
+
+ if ((a[system->N + center] != -1)) {
+ a[d_i] = a[d_k] = 1;
+ continue;
+ } else if ((b[system->N + center] != -1)) {
+ b[d_i] = b[d_k] = 1;
+ continue;
+ } else if ((c[system->N + center] != -1)) {
+ c[d_i] = c[d_k] = 1;
+ continue;
+ } else if ((d[system->N + center] != -1)) {
+ d[d_i] = d[d_k] = 1;
+ continue;
}
- }
+ if ((a[center] == -1) && (a[d_i] == -1) && (a[d_k] == -1)) {
+ a[center] = a[d_i] = a[d_k] = 1;
+ a[system->N + center] = 1;
+ } else if ((b[center] == -1) && (b[d_i] == -1) && (b[d_k] == -1)) {
+ b[center] = b[d_i] = b[d_k] = 1;
+ b[system->N + center] = 1;
+ } else if ((c[center] == -1) && (c[d_i] == -1) && (c[d_k] == -1)) {
+ c[center] = c[d_i] = c[d_k] = 1;
+ c[system->N + center] = 1;
+ } else if ((d[center] == -1) && (d[d_i] == -1) && (d[d_k] == -1)) {
+ d[center] = d[d_i] = d[d_k] = 1;
+ d[system->N + center]= 1;
+ }
+ else {
+ count ++;
+ break;
+ fprintf (stderr, "We have a problem with the four bins atom %d bond (%d %d) -- %d, (%d %d %d)\n",
+ i, b_start[i], b_end[i], j, center, d_i, d_k);
+ fprintf (stderr, "A's contents %d %d %d (%d %d %d)\n",
+ a[system->N + center], a[system->N + d_i], a[system->N + d_k], a[center], a[d_i], a[d_k]);
+ fprintf (stderr, "B's contents %d %d %d (%d %d %d)\n",
+ b[system->N + center], b[system->N + d_i], b[system->N + d_k], b[center], b[d_i], b[d_k]);
+ fprintf (stderr, "C's contents %d %d %d (%d %d %d)\n",
+ c[system->N + center], c[system->N + d_i], c[system->N + d_k], c[center], c[d_i], c[d_k]);
+ fprintf (stderr, "D's contents %d %d %d (%d %d %d)\n",
+ d[system->N + center], d[system->N + d_i], d[system->N + d_k], d[center], d[d_i], d[d_k]);
- void print_grid (reax_system *system)
- {
- int i, j, k, x;
- grid *g = &system->g;
-
- for( i = 0; i < g->ncell[0]; i++ )
- for( j = 0; j < g->ncell[1]; j++ )
- for( k = 0; k < g->ncell[2]; k++ ){
- fprintf (stderr, "Cell [%d,%d,%d]--(", i, j, k);
- for (x = 0; x < g->top[index_grid_3d (i,j,k,g) ]; x++){
- fprintf (stderr, "%d,", g->atoms[ index_grid_atoms (i,j,k,x,g) ]);
- }
- fprintf (stderr, ")\n");
- }
+ }
}
+ }
+ }
+ fprintf (stderr, "Miscount is %d \n", count);
+ exit (-1);
+
+ count = 0;
+ for (int i = 0; i < system->N; i++)
+ {
+ if (a[system->N + i] != -1) count ++;
+ if (b[system->N + i] != -1) count ++;
+ if (c[system->N + i] != -1) count ++;
+ if (d[system->N + i] != -1) count ++;
+ }
+
+ fprintf (stderr, "binned so many atoms --> %d \n", count );
+}
+
+
+int validate_grid (reax_system *system)
+{
+ int total = system->g.ncell[0] * system->g.ncell[1] * system->g.ncell[2];
+ int count = 0;
+ int *dtop = (int *) malloc (INT_SIZE * total );
+ copy_host_device (dtop, system->d_g.top, INT_SIZE * total, cudaMemcpyDeviceToHost, __LINE__);
+
+ for (int i = 0; i < total; i++){
+ if (system->g.top[i] != dtop[i]){
+ fprintf (stderr, " top count does not match (%d %d) @ index %d \n", system->g.top[i], dtop[i], i );
+ exit (-1);
+ }
+ }
+ free (dtop);
+
+ int *datoms = (int *) malloc (INT_SIZE * total * system->d_g.max_atoms);
+ copy_host_device (datoms, system->d_g.atoms, INT_SIZE * total * system->d_g.max_atoms, cudaMemcpyDeviceToHost, __LINE__);
+ for (int i = 0; i < total*system->d_g.max_atoms; i++){
+ if (system->g.atoms[i] != datoms[i]){
+ fprintf (stderr, " atoms count does not match (%d %d) @ index %d \n", system->g.atoms[i], datoms[i], i );
+ exit (-1);
+ }
+ }
+ free (datoms);
+ ivec *dnbrs = (ivec *) malloc (IVEC_SIZE * total * system->d_g.max_nbrs);
+ copy_host_device (dnbrs, system->d_g.nbrs, IVEC_SIZE * total * system->d_g.max_nbrs, cudaMemcpyDeviceToHost, __LINE__);
+ for (int i = 0; i < total*system->d_g.max_nbrs; i++){
+ if (!check_same (system->g.nbrs[i], dnbrs[i])){
+ fprintf (stderr, " nbrs count does not match @ index %d \n", i );
+ exit (-1);
+ }
+ }
+ free (dnbrs);
+
+ rvec *dnbrs_cp = (rvec *) malloc (RVEC_SIZE * total * system->d_g.max_nbrs);
+ copy_host_device (dnbrs_cp, system->d_g.nbrs_cp, RVEC_SIZE * total * system->d_g.max_nbrs, cudaMemcpyDeviceToHost, __LINE__);
+ for (int i = 0; i < total*system->d_g.max_nbrs; i++){
+ if (check_zero (system->g.nbrs_cp[i], dnbrs_cp[i])){
+ fprintf (stderr, " nbrs_cp count does not match @ index %d \n", i );
+ exit (-1);
+ }
+ }
+ free (dnbrs_cp);
+
+ //fprintf (stderr, " Grid match between device and host \n");
+ return TRUE;
+}
+
+
+void print_atoms (reax_system *system)
+{
+ int start, end, index;
+
+ reax_atom *test = (reax_atom *) malloc (REAX_ATOM_SIZE * system->N);
+ copy_host_device (test, system->d_atoms, REAX_ATOM_SIZE * system->N, cudaMemcpyDeviceToHost, RES_SYSTEM_ATOMS );
+
+ //for (int i = 0; i < system->N; i++)
+ for (int i = 0; i < 10; i++)
+ {
+ fprintf (stderr, "Atom:%d: Type:%d", i, test[i].type);
+ fprintf (stderr, " x(%6.10f %6.10f %6.10f)", test[i].x[0], test[i].x[1], test[i].x[2] );
+ fprintf (stderr, " v(%6.10f %6.10f %6.10f)", test[i].v[0], test[i].v[1], test[i].v[2] );
+ fprintf (stderr, " f(%6.10f %6.10f %6.10f)", test[i].f[0], test[i].f[1], test[i].f[2] );
+ fprintf (stderr, " q(%6.10f) \n", test[i].q );
+ }
+}
+
+
+void print_sys_atoms (reax_system *system)
+{
+ for (int i = 0; i < 10; i++)
+ {
+ fprintf (stderr, "Atom:%d: Type:%d", i, system->atoms[i].type);
+ fprintf (stderr, " x(%6.10f %6.10f %6.10f)",system->atoms[i].x[0], system->atoms[i].x[1], system->atoms[i].x[2] );
+ fprintf (stderr, " v(%6.10f %6.10f %6.10f)",system->atoms[i].v[0], system->atoms[i].v[1], system->atoms[i].v[2] );
+ fprintf (stderr, " f(%6.10f %6.10f %6.10f)", system->atoms[i].f[0], system->atoms[i].f[1], system->atoms[i].f[2] );
+ fprintf (stderr, " q(%6.10f) \n", system->atoms[i].q );
+ }
+}
+
+
+void print_grid (reax_system *system)
+{
+ int i, j, k, x;
+ grid *g = &system->g;
+
+ for( i = 0; i < g->ncell[0]; i++ )
+ for( j = 0; j < g->ncell[1]; j++ )
+ for( k = 0; k < g->ncell[2]; k++ ){
+ fprintf (stderr, "Cell [%d,%d,%d]--(", i, j, k);
+ for (x = 0; x < g->top[index_grid_3d (i,j,k,g) ]; x++){
+ fprintf (stderr, "%d,", g->atoms[ index_grid_atoms (i,j,k,x,g) ]);
+ }
+ fprintf (stderr, ")\n");
+ }
+}
diff --git a/PuReMD-GPU/src/validation.h b/PuReMD-GPU/src/validation.h
index 5ef9d2f37393c37eb4430238bdf0c68734842bcc..5eccf7d94f8c7716d45b45305155b0492b70dda2 100644
--- a/PuReMD-GPU/src/validation.h
+++ b/PuReMD-GPU/src/validation.h
@@ -23,35 +23,35 @@
#include "mytypes.h"
-bool check_zero (real , real );
-bool check_zero (rvec , rvec );
-bool check_same (ivec , ivec );
+int check_zero (real , real );
+int check_zero (rvec , rvec );
+int check_same (ivec , ivec );
-bool validate_box (simulation_box *host, simulation_box *dev);
-bool validate_atoms (reax_system *, list **);
-bool validate_grid (reax_system *);
+int validate_box (simulation_box *host, simulation_box *dev);
+int validate_atoms (reax_system *, list **);
+int validate_grid (reax_system *);
-bool validate_bonds (reax_system *, static_storage *, list **);
-bool validate_hbonds (reax_system *, static_storage *, list **);
-bool validate_sym_dbond_indices (reax_system *, static_storage *, list **);
-bool validate_three_bodies (reax_system *, static_storage *, list **);
+int validate_bonds (reax_system *, static_storage *, list **);
+int validate_hbonds (reax_system *, static_storage *, list **);
+int validate_sym_dbond_indices (reax_system *, static_storage *, list **);
+int validate_three_bodies (reax_system *, static_storage *, list **);
void count_three_bodies (reax_system *system, static_storage *workspace, list **lists);
-bool bin_three_bodies (reax_system *, static_storage *, list **);
+int bin_three_bodies (reax_system *, static_storage *, list **);
-bool validate_sort_matrix (reax_system *, static_storage *);
-bool validate_sparse_matrix (reax_system *, static_storage *);
-bool validate_lu (static_storage *);
+int validate_sort_matrix (reax_system *, static_storage *);
+int validate_sparse_matrix (reax_system *, static_storage *);
+int validate_lu (static_storage *);
void print_sparse_matrix (reax_system *, static_storage *);
void print_bond_list (reax_system *, static_storage *, list **);
-bool validate_workspace (reax_system *, static_storage *, list **);
-bool validate_neighbors (reax_system *, list **lists);
+int validate_workspace (reax_system *, static_storage *, list **);
+int validate_neighbors (reax_system *, list **lists);
-bool validate_data (reax_system *, simulation_data *);
+int validate_data (reax_system *, simulation_data *);
-bool analyze_hbonds (reax_system *, static_storage *, list **);
+int analyze_hbonds (reax_system *, static_storage *, list **);
void Print_Matrix (sparse_matrix *);
void Print_Matrix_L (sparse_matrix *);
diff --git a/PuReMD-GPU/src/vector.cu b/PuReMD-GPU/src/vector.c
similarity index 100%
rename from PuReMD-GPU/src/vector.cu
rename to PuReMD-GPU/src/vector.c
diff --git a/PuReMD-GPU/src/vector.h b/PuReMD-GPU/src/vector.h
index 336534784e50fd5552aaf27ff0c9531c3b97a02a..e1111e514928e79fc79197a0f2486d5eefb1cfa3 100644
--- a/PuReMD-GPU/src/vector.h
+++ b/PuReMD-GPU/src/vector.h
@@ -22,8 +22,14 @@
#define __VECTOR_H_
#include "mytypes.h"
+
#include "random.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
int Vector_isZero( real*, int );
void Vector_MakeZero( real*, int );
void Vector_Copy( real*, real*, int );
@@ -33,8 +39,6 @@ void Vector_Copy( real*, real*, int );
void Vector_Print( FILE*, char*, real*, int );
real Norm( real*, int );
-HOST_DEVICE inline real Dot( real*, real*, int );
-
void rvec_Sum( rvec, rvec, rvec );
real rvec_ScaledDot( real, rvec, real, rvec );
void rvec_Multiply( rvec, rvec, rvec );
@@ -44,19 +48,6 @@ void rvec_Invert( rvec, rvec );
void rvec_OuterProduct( rtensor, rvec, rvec );
int rvec_isZero( rvec );
-HOST_DEVICE inline real rvec_Dot( rvec, rvec );
-HOST_DEVICE inline void rvec_Scale( rvec, real, rvec );
-HOST_DEVICE inline real rvec_Norm_Sqr( rvec );
-HOST_DEVICE inline void rvec_Random( rvec );
-HOST_DEVICE inline void rvec_MakeZero( rvec );
-HOST_DEVICE inline void rvec_Add( rvec, rvec );
-HOST_DEVICE inline void rvec_Copy( rvec, rvec );
-HOST_DEVICE inline void rvec_Cross( rvec, rvec, rvec );
-HOST_DEVICE inline void rvec_ScaledAdd( rvec, real, rvec );
-HOST_DEVICE inline void rvec_ScaledSum( rvec, real, rvec, real, rvec );
-HOST_DEVICE inline void rvec_iMultiply( rvec, ivec, rvec );
-HOST_DEVICE inline real rvec_Norm( rvec );
-
void rtensor_MakeZero( rtensor );
void rtensor_Multiply( rtensor, rtensor, rtensor );
void rtensor_MatVec( rvec, rtensor, rvec );
@@ -80,16 +71,7 @@ void ivec_MakeZero( ivec );
void ivec_rScale( ivec, real, rvec );
-HOST_DEVICE inline void ivec_Copy( ivec, ivec );
-HOST_DEVICE inline void ivec_Scale( ivec, real, ivec );
-HOST_DEVICE inline void ivec_Sum( ivec, ivec, ivec );
-
-/*
- * Code which is common to multiple HOST and DEVICE
- *
- */
-
-HOST_DEVICE inline real Dot( real* v1, real* v2, int k )
+static inline HOST_DEVICE real Dot( real* v1, real* v2, int k )
{
real ret = 0;
@@ -100,102 +82,109 @@ HOST_DEVICE inline real Dot( real* v1, real* v2, int k )
}
-
-
/////////////////////////////
//rvec functions
/////////////////////////////
-
-HOST_DEVICE inline void rvec_MakeZero( rvec v )
+static inline HOST_DEVICE void rvec_MakeZero( rvec v )
{
v[0] = v[1] = v[2] = ZERO;
}
-HOST_DEVICE inline void rvec_Add( rvec ret, rvec v )
+
+static inline HOST_DEVICE void rvec_Add( rvec ret, rvec v )
{
ret[0] += v[0];
ret[1] += v[1];
ret[2] += v[2];
}
-HOST_DEVICE inline void rvec_Copy( rvec dest, rvec src )
+
+static inline HOST_DEVICE void rvec_Copy( rvec dest, rvec src )
{
dest[0] = src[0], dest[1] = src[1], dest[2] = src[2];
}
-HOST_DEVICE inline void rvec_Cross( rvec ret, rvec v1, rvec v2 )
+
+static inline HOST_DEVICE void rvec_Cross( rvec ret, rvec v1, rvec v2 )
{
ret[0] = v1[1] * v2[2] - v1[2] * v2[1];
ret[1] = v1[2] * v2[0] - v1[0] * v2[2];
ret[2] = v1[0] * v2[1] - v1[1] * v2[0];
}
-HOST_DEVICE inline void rvec_ScaledAdd( rvec ret, real c, rvec v )
+
+static inline HOST_DEVICE void rvec_ScaledAdd( rvec ret, real c, rvec v )
{
ret[0] += c * v[0], ret[1] += c * v[1], ret[2] += c * v[2];
}
-HOST_DEVICE inline void rvec_ScaledSum( rvec ret, real c1, rvec v1 , real c2, rvec v2 )
+
+static inline HOST_DEVICE void rvec_ScaledSum( rvec ret, real c1, rvec v1 , real c2, rvec v2 )
{
ret[0] = c1 * v1[0] + c2 * v2[0];
ret[1] = c1 * v1[1] + c2 * v2[1];
ret[2] = c1 * v1[2] + c2 * v2[2];
}
-HOST_DEVICE inline void rvec_Random( rvec v )
+
+static inline HOST_DEVICE void rvec_Random( rvec v )
{
v[0] = Random(2.0) - 1.0;
v[1] = Random(2.0) - 1.0;
v[2] = Random(2.0) - 1.0;
}
-HOST_DEVICE inline real rvec_Norm_Sqr( rvec v )
+
+static inline HOST_DEVICE real rvec_Norm_Sqr( rvec v )
{
return SQR(v[0]) + SQR(v[1]) + SQR(v[2]);
}
-HOST_DEVICE inline void rvec_Scale( rvec ret, real c, rvec v )
+
+static inline HOST_DEVICE void rvec_Scale( rvec ret, real c, rvec v )
{
ret[0] = c * v[0], ret[1] = c * v[1], ret[2] = c * v[2];
}
-HOST_DEVICE inline real rvec_Dot( rvec v1, rvec v2 )
+
+static inline HOST_DEVICE real rvec_Dot( rvec v1, rvec v2 )
{
return v1[0] * v2[0] + v1[1] * v2[1] + v1[2] * v2[2];
}
-HOST_DEVICE inline void rvec_iMultiply( rvec r, ivec v1, rvec v2 )
+
+static inline HOST_DEVICE void rvec_iMultiply( rvec r, ivec v1, rvec v2 )
{
r[0] = v1[0] * v2[0];
r[1] = v1[1] * v2[1];
r[2] = v1[2] * v2[2];
}
-HOST_DEVICE inline real rvec_Norm( rvec v )
+
+static inline HOST_DEVICE real rvec_Norm( rvec v )
{
return SQRT( SQR(v[0]) + SQR(v[1]) + SQR(v[2]) );
}
-
/////////////////
//ivec functions
/////////////////
-
-
-HOST_DEVICE inline void ivec_Copy( ivec dest , ivec src )
+static inline HOST_DEVICE void ivec_Copy( ivec dest , ivec src )
{
dest[0] = src[0], dest[1] = src[1], dest[2] = src[2];
}
-HOST_DEVICE inline void ivec_Scale( ivec dest, real C, ivec src )
+
+static inline HOST_DEVICE void ivec_Scale( ivec dest, real C, ivec src )
{
dest[0] = C * src[0];
dest[1] = C * src[1];
dest[2] = C * src[2];
}
-HOST_DEVICE inline void ivec_Sum( ivec dest, ivec v1, ivec v2 )
+
+static inline HOST_DEVICE void ivec_Sum( ivec dest, ivec v1, ivec v2 )
{
dest[0] = v1[0] + v2[0];
dest[1] = v1[1] + v2[1];
@@ -203,26 +192,32 @@ HOST_DEVICE inline void ivec_Sum( ivec dest, ivec v1, ivec v2 )
}
-
/////////////////
//vector functions
/////////////////
-HOST_DEVICE inline void Vector_Sum( real* dest, real c, real* v, real d, real* y, int k )
+static inline HOST_DEVICE void Vector_Sum( real* dest, real c, real* v, real d, real* y, int k )
{
for (k--; k >= 0; k--)
dest[k] = c * v[k] + d * y[k];
}
-HOST_DEVICE inline void Vector_Scale( real* dest, real c, real* v, int k )
+
+static inline HOST_DEVICE void Vector_Scale( real* dest, real c, real* v, int k )
{
for (k--; k >= 0; k--)
dest[k] = c * v[k];
}
-HOST_DEVICE inline void Vector_Add( real* dest, real c, real* v, int k )
+
+static inline HOST_DEVICE void Vector_Add( real* dest, real c, real* v, int k )
{
for (k--; k >= 0; k--)
dest[k] += c * v[k];
}
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PuReMD/src/linear_solvers.c b/PuReMD/src/linear_solvers.c
index 835ffa358e4e6ff995e2f69dc694bb1c3f7798c1..d08dfe1ed75780bc90b5aa65bb13462393d9af47 100644
--- a/PuReMD/src/linear_solvers.c
+++ b/PuReMD/src/linear_solvers.c
@@ -36,7 +36,9 @@ void dual_Sparse_MatVec( sparse_matrix *A, rvec2 *x, rvec2 *b, int N )
real H;
for ( i = 0; i < N; ++i )
+ {
b[i][0] = b[i][1] = 0;
+ }
/* perform multiplication */
for ( i = 0; i < A->n; ++i )
@@ -64,7 +66,7 @@ void dual_Sparse_MatVec( sparse_matrix *A, rvec2 *x, rvec2 *b, int N )
int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
- rvec2 *b, real tol, rvec2 *x, mpi_datatypes* mpi_data, FILE *fout )
+ rvec2 *b, real tol, rvec2 *x, mpi_datatypes* mpi_data, FILE *fout )
{
int i, j, n, N, matvecs, scale;
rvec2 tmp, alpha, beta;
@@ -86,13 +88,17 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
t_start = Get_Time( );
}
#endif
+
Dist( system, mpi_data, x, mpi_data->mpi_rvec2, scale, rvec2_packer );
dual_Sparse_MatVec( H, x, workspace->q2, N );
// tryQEq
Coll(system, mpi_data, workspace->q2, mpi_data->mpi_rvec2, scale, rvec2_unpacker);
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &matvec_time );
+ }
#endif
for ( j = 0; j < system->n; ++j )
@@ -126,6 +132,7 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
}
MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
//fprintf( stderr, "sig_new: %f %f\n", sig_new[0], sig_new[1] );
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
Update_Timing_Info( &t_start, &dot_time );
@@ -137,9 +144,12 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
dual_Sparse_MatVec( H, workspace->d2, workspace->q2, N );
// tryQEq
Coll(system, mpi_data, workspace->q2, mpi_data->mpi_rvec2, scale, rvec2_unpacker);
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &matvec_time );
+ }
#endif
/* dot product: d.q */
@@ -174,12 +184,18 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
sig_old[1] = sig_new[1];
MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
//fprintf( stderr, "sig_new: %f %f\n", sig_new[0], sig_new[1] );
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &dot_time );
+ }
#endif
+
if ( sqrt(sig_new[0]) / b_norm[0] <= tol || sqrt(sig_new[1]) / b_norm[1] <= tol )
+ {
break;
+ }
beta[0] = sig_new[0] / sig_old[0];
beta[1] = sig_new[1] / sig_old[1];
@@ -194,30 +210,41 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
if ( sqrt(sig_new[0]) / b_norm[0] <= tol )
{
for ( j = 0; j < n; ++j )
+ {
workspace->t[j] = workspace->x[j][1];
- matvecs = CG( system, workspace, H, workspace->b_t, tol, workspace->t,
- mpi_data, fout );
+ }
+ matvecs = CG( system, workspace, H, workspace->b_t, tol,
+ workspace->t,mpi_data, fout );
for ( j = 0; j < n; ++j )
+ {
workspace->x[j][1] = workspace->t[j];
+ }
}
else if ( sqrt(sig_new[1]) / b_norm[1] <= tol )
{
for ( j = 0; j < n; ++j )
+ {
workspace->s[j] = workspace->x[j][0];
+ }
matvecs = CG( system, workspace, H, workspace->b_s, tol, workspace->s,
- mpi_data, fout );
+ mpi_data, fout );
for ( j = 0; j < system->n; ++j )
+ {
workspace->x[j][0] = workspace->s[j];
+ }
}
-
if ( i >= 300 )
+ {
fprintf( stderr, "CG convergence failed!\n" );
+ }
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
- fprintf( fout, "QEq %d + %d iters. matvecs: %f dot: %f\n",
- i + 1, matvecs, matvec_time, dot_time );
+ {
+ fprintf( fout, "QEq %d + %d iters. matvecs: %f dot: %f\n", i + 1,
+ matvecs, matvec_time, dot_time );
+ }
#endif
return (i + 1) + matvecs;
@@ -230,7 +257,9 @@ void Sparse_MatVec( sparse_matrix *A, real *x, real *b, int N )
real H;
for ( i = 0; i < N; ++i )
+ {
b[i] = 0;
+ }
/* perform multiplication */
for ( i = 0; i < A->n; ++i )
@@ -249,8 +278,8 @@ void Sparse_MatVec( sparse_matrix *A, real *x, real *b, int N )
}
-int CG( reax_system *system, storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, mpi_datatypes* mpi_data, FILE *fout )
+int CG( reax_system *system, storage *workspace, sparse_matrix *H, real *b,
+ real tol, real *x, mpi_datatypes* mpi_data, FILE *fout )
{
int i, j, scale;
real tmp, alpha, beta, b_norm;
@@ -269,21 +298,29 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H,
Sparse_MatVec( H, x, workspace->q, system->N );
// tryQEq
Coll( system, mpi_data, workspace->q, MPI_DOUBLE, scale, real_unpacker );
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &matvec_time );
+ }
#endif
Vector_Sum( workspace->r , 1., b, -1., workspace->q, system->n );
for ( j = 0; j < system->n; ++j )
+ {
workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j]; //pre-condition
+ }
b_norm = Parallel_Norm( b, system->n, mpi_data->world );
sig_new = Parallel_Dot(workspace->r, workspace->d, system->n, mpi_data->world);
sig0 = sig_new;
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &dot_time );
+ }
#endif
for ( i = 1; i < 300 && sqrt(sig_new) / b_norm > tol; ++i )
@@ -292,9 +329,12 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H,
Sparse_MatVec( H, workspace->d, workspace->q, system->N );
//tryQEq
Coll(system, mpi_data, workspace->q, MPI_DOUBLE, scale, real_unpacker);
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &matvec_time );
+ }
#endif
tmp = Parallel_Dot(workspace->d, workspace->q, system->n, mpi_data->world);
@@ -303,15 +343,20 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H,
Vector_Add( workspace->r, -alpha, workspace->q, system->n );
/* pre-conditioning */
for ( j = 0; j < system->n; ++j )
+ {
workspace->p[j] = workspace->r[j] * workspace->Hdia_inv[j];
+ }
sig_old = sig_new;
sig_new = Parallel_Dot(workspace->r, workspace->p, system->n, mpi_data->world);
beta = sig_new / sig_old;
Vector_Sum( workspace->d, 1., workspace->p, beta, workspace->d, system->n );
+
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
+ {
Update_Timing_Info( &t_start, &dot_time );
+ }
#endif
}
@@ -323,8 +368,10 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H,
#if defined(CG_PERFORMANCE)
if ( system->my_rank == MASTER_NODE )
- fprintf( fout, "QEq %d iters. matvecs: %f dot: %f\n",
- i, matvec_time, dot_time );
+ {
+ fprintf( fout, "QEq %d iters. matvecs: %f dot: %f\n", i, matvec_time,
+ dot_time );
+ }
#endif
return i;
@@ -332,7 +379,7 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H,
int CG_test( reax_system *system, storage *workspace, sparse_matrix *H,
- real *b, real tol, real *x, mpi_datatypes* mpi_data, FILE *fout )
+ real *b, real tol, real *x, mpi_datatypes* mpi_data, FILE *fout )
{
int i, j, scale;
real tmp, alpha, beta, b_norm;