From 5299a3154c6112e9f002a919442ed32ed8fc5b8f Mon Sep 17 00:00:00 2001
From: "Kurt A. O'Hearn" <ohearnku@msu.edu>
Date: Sun, 16 Jul 2017 23:19:55 -0400
Subject: [PATCH] PG-PuReMD: minor project refactoring. Separate CUDA code.
Change header includes.
---
PG-PuReMD/Makefile.am | 38 +-
PG-PuReMD/src/allocate.c | 3 +-
PG-PuReMD/src/allocate.h | 4 +-
PG-PuReMD/src/analyze.c | 3 +
PG-PuReMD/src/analyze.h | 12 +-
PG-PuReMD/src/basic_comm.h | 20 +-
PG-PuReMD/src/bond_orders.c | 1 +
PG-PuReMD/src/bond_orders.h | 34 +-
PG-PuReMD/src/bonds.c | 28 +-
PG-PuReMD/src/bonds.h | 13 +-
PG-PuReMD/src/box.c | 3 +
PG-PuReMD/src/box.h | 35 +-
PG-PuReMD/src/center_mass.cu | 551 --------
PG-PuReMD/src/center_mass.h | 49 -
PG-PuReMD/src/charges.c | 102 +-
PG-PuReMD/src/charges.h | 9 +-
PG-PuReMD/src/comm_tools.c | 2 +
PG-PuReMD/src/comm_tools.h | 23 +-
PG-PuReMD/src/control.h | 10 +
PG-PuReMD/src/{ => cuda}/cuda_allocate.cu | 8 +-
PG-PuReMD/src/{ => cuda}/cuda_allocate.h | 2 +-
PG-PuReMD/src/{ => cuda}/cuda_bond_orders.cu | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_bond_orders.h | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_bonds.cu | 7 +-
PG-PuReMD/src/{ => cuda}/cuda_bonds.h | 18 +-
PG-PuReMD/src/{ => cuda}/cuda_charges.cu | 95 +-
PG-PuReMD/src/{ => cuda}/cuda_charges.h | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_copy.cu | 3 +-
PG-PuReMD/src/{ => cuda}/cuda_copy.h | 10 +-
PG-PuReMD/src/{ => cuda}/cuda_environment.cu | 0
PG-PuReMD/src/{ => cuda}/cuda_environment.h | 8 +-
PG-PuReMD/src/{ => cuda}/cuda_forces.cu | 227 +++-
PG-PuReMD/src/{ => cuda}/cuda_forces.h | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_helpers.h | 2 +-
.../src/{ => cuda}/cuda_hydrogen_bonds.cu | 8 +-
PG-PuReMD/src/cuda/cuda_hydrogen_bonds.h | 48 +
PG-PuReMD/src/cuda/cuda_init_md.cu | 412 ++++++
PG-PuReMD/src/cuda/cuda_init_md.h | 22 +
PG-PuReMD/src/cuda/cuda_integrate.cu | 249 ++++
PG-PuReMD/src/{ => cuda}/cuda_integrate.h | 10 +-
PG-PuReMD/src/cuda/cuda_lin_alg.cu | 1113 +++++++++++++++++
PG-PuReMD/src/{ => cuda}/cuda_lin_alg.h | 39 +-
PG-PuReMD/src/{ => cuda}/cuda_list.cu | 9 +-
PG-PuReMD/src/{ => cuda}/cuda_list.h | 4 +-
PG-PuReMD/src/{ => cuda}/cuda_lookup.cu | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_lookup.h | 6 +-
PG-PuReMD/src/{ => cuda}/cuda_multi_body.cu | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_multi_body.h | 25 +-
PG-PuReMD/src/{ => cuda}/cuda_neighbors.cu | 8 +-
PG-PuReMD/src/{ => cuda}/cuda_neighbors.h | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_nonbonded.cu | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_nonbonded.h | 14 +-
PG-PuReMD/src/{ => cuda}/cuda_post_evolve.cu | 5 +-
PG-PuReMD/src/{ => cuda}/cuda_post_evolve.h | 6 +-
PG-PuReMD/src/{ => cuda}/cuda_reduction.cu | 6 +-
PG-PuReMD/src/{ => cuda}/cuda_reduction.h | 15 +-
PG-PuReMD/src/{ => cuda}/cuda_reset_tools.cu | 2 +-
PG-PuReMD/src/{ => cuda}/cuda_reset_tools.h | 6 +-
PG-PuReMD/src/{ => cuda}/cuda_shuffle.h | 3 +-
PG-PuReMD/src/cuda/cuda_system_props.cu | 1026 +++++++++++++++
PG-PuReMD/src/{ => cuda}/cuda_system_props.h | 15 +-
.../src/{ => cuda}/cuda_torsion_angles.cu | 7 +-
.../src/{ => cuda}/cuda_torsion_angles.h | 32 +-
PG-PuReMD/src/{ => cuda}/cuda_utils.cu | 17 +
PG-PuReMD/src/{ => cuda}/cuda_utils.h | 32 +-
.../src/{ => cuda}/cuda_valence_angles.cu | 5 +-
.../src/{ => cuda}/cuda_valence_angles.h | 6 +-
PG-PuReMD/src/{ => cuda}/cuda_validation.cu | 9 +-
PG-PuReMD/src/{ => cuda}/cuda_validation.h | 18 +-
PG-PuReMD/src/cuda_hydrogen_bonds.h | 66 -
PG-PuReMD/src/cuda_init_md.cu | 14 -
PG-PuReMD/src/cuda_init_md.h | 15 -
PG-PuReMD/src/cuda_integrate.cu | 105 --
PG-PuReMD/src/cuda_lin_alg.cu | 624 ---------
PG-PuReMD/src/cuda_system_props.cu | 406 ------
PG-PuReMD/src/ffield.c | 3 +-
PG-PuReMD/src/ffield.h | 10 +
PG-PuReMD/src/forces.c | 217 +---
PG-PuReMD/src/forces.h | 11 +-
PG-PuReMD/src/geo_tools.c | 3 +
PG-PuReMD/src/geo_tools.h | 21 +-
PG-PuReMD/src/grid.c | 6 +-
PG-PuReMD/src/grid.h | 14 +
PG-PuReMD/src/hydrogen_bonds.c | 4 +-
PG-PuReMD/src/hydrogen_bonds.h | 12 +-
PG-PuReMD/src/init_md.c | 377 ------
PG-PuReMD/src/init_md.h | 15 +-
PG-PuReMD/src/integrate.c | 147 +--
PG-PuReMD/src/integrate.h | 13 +-
PG-PuReMD/src/io_tools.c | 4 +-
PG-PuReMD/src/io_tools.h | 107 +-
PG-PuReMD/src/lin_alg.c | 503 +-------
PG-PuReMD/src/lin_alg.h | 29 +-
PG-PuReMD/src/list.c | 8 +-
PG-PuReMD/src/list.h | 11 +-
PG-PuReMD/src/lookup.c | 12 +-
PG-PuReMD/src/lookup.h | 12 +-
PG-PuReMD/src/multi_body.c | 1 +
PG-PuReMD/src/multi_body.h | 12 +-
PG-PuReMD/src/neighbors.c | 6 +-
PG-PuReMD/src/neighbors.h | 14 +-
PG-PuReMD/src/nonbonded.c | 8 +-
PG-PuReMD/src/nonbonded.h | 16 +-
PG-PuReMD/src/parallelreax.c | 33 +-
PG-PuReMD/src/random.c | 2 +
PG-PuReMD/src/random.h | 15 +-
PG-PuReMD/src/reax_types.h | 15 +-
PG-PuReMD/src/reset_tools.c | 4 +-
PG-PuReMD/src/reset_tools.h | 4 +-
PG-PuReMD/src/restart.c | 3 +
PG-PuReMD/src/restart.h | 19 +-
PG-PuReMD/src/system_props.c | 157 +--
PG-PuReMD/src/system_props.h | 19 +-
PG-PuReMD/src/tool_box.h | 3 +-
PG-PuReMD/src/torsion_angles.c | 3 +-
PG-PuReMD/src/torsion_angles.h | 12 +-
PG-PuReMD/src/traj.c | 2 +-
PG-PuReMD/src/traj.h | 10 +-
PG-PuReMD/src/valence_angles.h | 8 +
PG-PuReMD/src/vector.h | 5 +-
120 files changed, 3967 insertions(+), 3710 deletions(-)
delete mode 100644 PG-PuReMD/src/center_mass.cu
delete mode 100644 PG-PuReMD/src/center_mass.h
rename PG-PuReMD/src/{ => cuda}/cuda_allocate.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_allocate.h (96%)
rename PG-PuReMD/src/{ => cuda}/cuda_bond_orders.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_bond_orders.h (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_bonds.cu (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_bonds.h (69%)
rename PG-PuReMD/src/{ => cuda}/cuda_charges.cu (68%)
rename PG-PuReMD/src/{ => cuda}/cuda_charges.h (89%)
rename PG-PuReMD/src/{ => cuda}/cuda_copy.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_copy.h (93%)
rename PG-PuReMD/src/{ => cuda}/cuda_environment.cu (100%)
rename PG-PuReMD/src/{ => cuda}/cuda_environment.h (56%)
rename PG-PuReMD/src/{ => cuda}/cuda_forces.cu (89%)
rename PG-PuReMD/src/{ => cuda}/cuda_forces.h (86%)
rename PG-PuReMD/src/{ => cuda}/cuda_helpers.h (97%)
rename PG-PuReMD/src/{ => cuda}/cuda_hydrogen_bonds.cu (99%)
create mode 100644 PG-PuReMD/src/cuda/cuda_hydrogen_bonds.h
create mode 100644 PG-PuReMD/src/cuda/cuda_init_md.cu
create mode 100644 PG-PuReMD/src/cuda/cuda_init_md.h
create mode 100644 PG-PuReMD/src/cuda/cuda_integrate.cu
rename PG-PuReMD/src/{ => cuda}/cuda_integrate.h (86%)
create mode 100644 PG-PuReMD/src/cuda/cuda_lin_alg.cu
rename PG-PuReMD/src/{ => cuda}/cuda_lin_alg.h (52%)
rename PG-PuReMD/src/{ => cuda}/cuda_list.cu (96%)
rename PG-PuReMD/src/{ => cuda}/cuda_list.h (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_lookup.cu (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_lookup.h (56%)
rename PG-PuReMD/src/{ => cuda}/cuda_multi_body.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_multi_body.h (58%)
rename PG-PuReMD/src/{ => cuda}/cuda_neighbors.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_neighbors.h (95%)
rename PG-PuReMD/src/{ => cuda}/cuda_nonbonded.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_nonbonded.h (79%)
rename PG-PuReMD/src/{ => cuda}/cuda_post_evolve.cu (95%)
rename PG-PuReMD/src/{ => cuda}/cuda_post_evolve.h (60%)
rename PG-PuReMD/src/{ => cuda}/cuda_reduction.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_reduction.h (96%)
rename PG-PuReMD/src/{ => cuda}/cuda_reset_tools.cu (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_reset_tools.h (94%)
rename PG-PuReMD/src/{ => cuda}/cuda_shuffle.h (97%)
create mode 100644 PG-PuReMD/src/cuda/cuda_system_props.cu
rename PG-PuReMD/src/{ => cuda}/cuda_system_props.h (65%)
rename PG-PuReMD/src/{ => cuda}/cuda_torsion_angles.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_torsion_angles.h (57%)
rename PG-PuReMD/src/{ => cuda}/cuda_utils.cu (88%)
rename PG-PuReMD/src/{ => cuda}/cuda_utils.h (80%)
rename PG-PuReMD/src/{ => cuda}/cuda_valence_angles.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_valence_angles.h (98%)
rename PG-PuReMD/src/{ => cuda}/cuda_validation.cu (99%)
rename PG-PuReMD/src/{ => cuda}/cuda_validation.h (97%)
delete mode 100644 PG-PuReMD/src/cuda_hydrogen_bonds.h
delete mode 100644 PG-PuReMD/src/cuda_init_md.cu
delete mode 100644 PG-PuReMD/src/cuda_init_md.h
delete mode 100644 PG-PuReMD/src/cuda_integrate.cu
delete mode 100644 PG-PuReMD/src/cuda_lin_alg.cu
delete mode 100644 PG-PuReMD/src/cuda_system_props.cu
diff --git a/PG-PuReMD/Makefile.am b/PG-PuReMD/Makefile.am
index b0c1c871..3b051035 100644
--- a/PG-PuReMD/Makefile.am
+++ b/PG-PuReMD/Makefile.am
@@ -34,25 +34,25 @@ include_HEADERS = src/reax_types.h src/index_utils.h \
src/integrate.h src/init_md.h
if USE_CUDA
-bin_pg_puremd_SOURCES += src/cuda_utils.cu src/cuda_allocate.cu src/cuda_environment.cu \
- src/cuda_system_props.cu src/cuda_reduction.cu src/center_mass.cu \
- src/cuda_copy.cu src/cuda_reset_tools.cu src/cuda_list.cu \
- src/cuda_neighbors.cu src/cuda_bond_orders.cu src/cuda_bonds.cu \
- src/cuda_multi_body.cu src/cuda_valence_angles.cu \
- src/cuda_torsion_angles.cu src/cuda_hydrogen_bonds.cu src/cuda_forces.cu \
- src/cuda_charges.cu src/cuda_lin_alg.cu \
- src/cuda_nonbonded.cu src/cuda_integrate.cu src/cuda_post_evolve.cu \
- src/cuda_init_md.cu src/cuda_validation.cu src/cuda_lookup.cu
-include_HEADERS += src/cuda_helpers.h src/cuda_shuffle.h \
- src/cuda_utils.h src/cuda_allocate.h src/cuda_environment.h \
- src/cuda_system_props.h src/cuda_reduction.h src/center_mass.h \
- src/cuda_copy.h src/cuda_reset_tools.h src/cuda_list.h \
- src/cuda_neighbors.h src/cuda_bond_orders.h src/cuda_bonds.h \
- src/cuda_multi_body.h src/cuda_valence_angles.h \
- src/cuda_torsion_angles.h src/cuda_hydrogen_bonds.h src/cuda_forces.h \
- src/cuda_charges.h src/cuda_lin_alg.h \
- src/cuda_nonbonded.h src/cuda_integrate.h src/cuda_post_evolve.h \
- src/cuda_init_md.h src/cuda_validation.h src/cuda_lookup.h
+bin_pg_puremd_SOURCES += src/cuda/cuda_utils.cu src/cuda/cuda_allocate.cu src/cuda/cuda_environment.cu \
+ src/cuda/cuda_system_props.cu src/cuda/cuda_reduction.cu \
+ src/cuda/cuda_copy.cu src/cuda/cuda_reset_tools.cu src/cuda/cuda_list.cu \
+ src/cuda/cuda_neighbors.cu src/cuda/cuda_bond_orders.cu src/cuda/cuda_bonds.cu \
+ src/cuda/cuda_multi_body.cu src/cuda/cuda_valence_angles.cu \
+ src/cuda/cuda_torsion_angles.cu src/cuda/cuda_hydrogen_bonds.cu src/cuda/cuda_forces.cu \
+ src/cuda/cuda_charges.cu src/cuda/cuda_lin_alg.cu \
+ src/cuda/cuda_nonbonded.cu src/cuda/cuda_integrate.cu src/cuda/cuda_post_evolve.cu \
+ src/cuda/cuda_init_md.cu src/cuda/cuda_validation.cu src/cuda/cuda_lookup.cu
+include_HEADERS += src/cuda/cuda_helpers.h src/cuda/cuda_shuffle.h \
+ src/cuda/cuda_utils.h src/cuda/cuda_allocate.h src/cuda/cuda_environment.h \
+ src/cuda/cuda_system_props.h src/cuda/cuda_reduction.h \
+ src/cuda/cuda_copy.h src/cuda/cuda_reset_tools.h src/cuda/cuda_list.h \
+ src/cuda/cuda_neighbors.h src/cuda/cuda_bond_orders.h src/cuda/cuda_bonds.h \
+ src/cuda/cuda_multi_body.h src/cuda/cuda_valence_angles.h \
+ src/cuda/cuda_torsion_angles.h src/cuda/cuda_hydrogen_bonds.h src/cuda/cuda_forces.h \
+ src/cuda/cuda_charges.h src/cuda/cuda_lin_alg.h \
+ src/cuda/cuda_nonbonded.h src/cuda/cuda_integrate.h src/cuda/cuda_post_evolve.h \
+ src/cuda/cuda_init_md.h src/cuda/cuda_validation.h src/cuda/cuda_lookup.h
# dummy source to cause C linking
nodist_EXTRA_bin_pg_puremd_SOURCES = src/dummy.c
diff --git a/PG-PuReMD/src/allocate.c b/PG-PuReMD/src/allocate.c
index 1d85b8f9..54614694 100644
--- a/PG-PuReMD/src/allocate.c
+++ b/PG-PuReMD/src/allocate.c
@@ -20,7 +20,6 @@
----------------------------------------------------------------------*/
#include "reax_types.h"
-#include "index_utils.h"
#if defined(PURE_REAX)
#include "allocate.h"
@@ -36,6 +35,8 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
/* allocate space for my_atoms
important: we cannot know the exact number of atoms that will fall into a
diff --git a/PG-PuReMD/src/allocate.h b/PG-PuReMD/src/allocate.h
index 5fd27315..a2876453 100644
--- a/PG-PuReMD/src/allocate.h
+++ b/PG-PuReMD/src/allocate.h
@@ -24,11 +24,11 @@
#include "reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-
int PreAllocate_Space( reax_system*, control_params*, storage* );
void Allocate_System( reax_system*, int, int, char* );
@@ -53,9 +53,9 @@ void Deallocate_MPI_Buffers( mpi_datatypes * );
void ReAllocate( reax_system*, control_params*, simulation_data*, storage*,
reax_list**, mpi_datatypes* );
-
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/analyze.c b/PG-PuReMD/src/analyze.c
index 283d7e47..0f47ba48 100644
--- a/PG-PuReMD/src/analyze.c
+++ b/PG-PuReMD/src/analyze.c
@@ -19,7 +19,10 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "analyze.h"
+
#include "box.h"
#include "list.h"
#include "vector.h"
diff --git a/PG-PuReMD/src/analyze.h b/PG-PuReMD/src/analyze.h
index e4703341..a772dcb2 100644
--- a/PG-PuReMD/src/analyze.h
+++ b/PG-PuReMD/src/analyze.h
@@ -24,7 +24,17 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Analysis( reax_system*, control_params*, simulation_data*, storage*,
- reax_list**, output_controls*, mpi_datatypes* );
+ reax_list**, output_controls*, mpi_datatypes* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/basic_comm.h b/PG-PuReMD/src/basic_comm.h
index e1effc50..4d8f1c34 100644
--- a/PG-PuReMD/src/basic_comm.h
+++ b/PG-PuReMD/src/basic_comm.h
@@ -24,33 +24,43 @@
#include "reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
void real_packer( void*, mpi_out_data* );
+
void rvec_packer( void*, mpi_out_data* );
+
void rvec2_packer( void*, mpi_out_data* );
+
void Dist(reax_system*, mpi_datatypes*, void*, MPI_Datatype, int, dist_packer);
void real_unpacker( void*, void*, mpi_out_data* );
+
void rvec_unpacker( void*, void*, mpi_out_data* );
+
void rvec2_unpacker( void*, void*, mpi_out_data* );
+
void Coll( reax_system*, mpi_datatypes*, void*, MPI_Datatype,
- int, coll_unpacker );
+ int, coll_unpacker );
real Parallel_Norm( real*, int, MPI_Comm );
+
real Parallel_Dot( real*, real*, int, MPI_Comm );
+
real Parallel_Vector_Acc( real*, int, MPI_Comm );
+#if defined(TEST_FORCES)
+void Coll_ids_at_Master( reax_system*, storage*, mpi_datatypes* );
+
+void Coll_rvecs_at_Master( reax_system*, storage*, mpi_datatypes*, rvec* );
+#endif
#ifdef __cplusplus
}
#endif
-#if defined(TEST_FORCES)
-void Coll_ids_at_Master( reax_system*, storage*, mpi_datatypes* );
-void Coll_rvecs_at_Master( reax_system*, storage*, mpi_datatypes*, rvec* );
-#endif
#endif
diff --git a/PG-PuReMD/src/bond_orders.c b/PG-PuReMD/src/bond_orders.c
index 4e023e97..da23e002 100644
--- a/PG-PuReMD/src/bond_orders.c
+++ b/PG-PuReMD/src/bond_orders.c
@@ -31,6 +31,7 @@
#include "reax_list.h"
#include "reax_vector.h"
#endif
+
#include "index_utils.h"
diff --git a/PG-PuReMD/src/bond_orders.h b/PG-PuReMD/src/bond_orders.h
index 1975e20b..8cfa2e18 100644
--- a/PG-PuReMD/src/bond_orders.h
+++ b/PG-PuReMD/src/bond_orders.h
@@ -24,6 +24,7 @@
#include "reax_types.h"
+
typedef struct
{
real C1dbo, C2dbo, C3dbo;
@@ -32,28 +33,45 @@ typedef struct
real C1dDelta, C2dDelta, C3dDelta;
} dbond_coefficients;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
#ifdef TEST_FORCES
void Get_dBO( reax_system*, reax_list**, int, int, real, rvec* );
+
void Get_dBOpinpi2( reax_system*, reax_list**,
- int, int, real, real, rvec*, rvec* );
+ int, int, real, real, rvec*, rvec* );
void Add_dBO( reax_system*, reax_list**, int, int, real, rvec* );
+
void Add_dBOpinpi2( reax_system*, reax_list**,
- int, int, real, real, rvec*, rvec* );
+ int, int, real, real, rvec*, rvec* );
void Add_dBO_to_Forces( reax_system*, reax_list**, int, int, real );
+
void Add_dBOpinpi2_to_Forces( reax_system*, reax_list**,
- int, int, real, real );
+ int, int, real, real );
void Add_dDelta( reax_system*, reax_list**, int, real, rvec* );
+
void Add_dDelta_to_Forces( reax_system *, reax_list**, int, real );
#endif
void Add_dBond_to_Forces( int, int, storage*, reax_list** );
-void Add_dBond_to_Forces_NPT( int, int, simulation_data*,
- storage*, reax_list** );
-int BOp(storage*, reax_list*, real, int, int, far_neighbor_data*,
- single_body_parameters*, single_body_parameters*, two_body_parameters*);
+
+void Add_dBond_to_Forces_NPT( int, int, simulation_data*, storage*, reax_list** );
+
+int BOp( storage*, reax_list*, real, int, int, far_neighbor_data*,
+ single_body_parameters*, single_body_parameters*, two_body_parameters* );
+
void BO( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/bonds.c b/PG-PuReMD/src/bonds.c
index 9c2839eb..8fb160ec 100644
--- a/PG-PuReMD/src/bonds.c
+++ b/PG-PuReMD/src/bonds.c
@@ -20,25 +20,27 @@
----------------------------------------------------------------------*/
#include "reax_types.h"
-#include "index_utils.h"
+
#if defined(PURE_REAX)
-#include "bonds.h"
-#include "bond_orders.h"
-#include "list.h"
-#include "tool_box.h"
-#include "vector.h"
+ #include "bonds.h"
+ #include "bond_orders.h"
+ #include "list.h"
+ #include "tool_box.h"
+ #include "vector.h"
#elif defined(LAMMPS_REAX)
-#include "reax_bonds.h"
-#include "reax_bond_orders.h"
-#include "reax_list.h"
-#include "reax_tool_box.h"
-#include "reax_vector.h"
+ #include "reax_bonds.h"
+ #include "reax_bond_orders.h"
+ #include "reax_list.h"
+ #include "reax_tool_box.h"
+ #include "reax_vector.h"
#endif
+#include "index_utils.h"
+
void Bonds( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace, reax_list **lists,
- output_controls *out_control )
+ simulation_data *data, storage *workspace, reax_list **lists,
+ output_controls *out_control )
{
int i, j, pj, natoms;
int start_i, end_i;
diff --git a/PG-PuReMD/src/bonds.h b/PG-PuReMD/src/bonds.h
index 2aa3c1f9..89090386 100644
--- a/PG-PuReMD/src/bonds.h
+++ b/PG-PuReMD/src/bonds.h
@@ -24,6 +24,17 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Bonds( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/box.c b/PG-PuReMD/src/box.c
index 86ebd6eb..525f24e5 100644
--- a/PG-PuReMD/src/box.c
+++ b/PG-PuReMD/src/box.c
@@ -19,7 +19,10 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "box.h"
+
#include "comm_tools.h"
#include "io_tools.h"
#include "system_props.h"
diff --git a/PG-PuReMD/src/box.h b/PG-PuReMD/src/box.h
index 841e3679..00e51d06 100644
--- a/PG-PuReMD/src/box.h
+++ b/PG-PuReMD/src/box.h
@@ -24,30 +24,51 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
/* initializes simulation boxes */
void Setup_Big_Box( real, real, real, real, real, real, simulation_box* );
+
void Init_Box( rtensor, simulation_box* );
+
//void Setup_My_Box( reax_system*, control_params* );
+
//void Setup_My_Ext_Box( reax_system*, control_params* );
+
void Setup_Environment( reax_system*, control_params*, mpi_datatypes* );
/* scales simulation box for NPT ensembles */
void Scale_Box( reax_system*, control_params*,
- simulation_data*, mpi_datatypes* );
+ simulation_data*, mpi_datatypes* );
/* applies transformation to/from Cartesian/ Triclinic coordinates */
/* use -1 flag for Cartesian -> Triclinic and +1 for otherway */
-// void Transform( rvec, simulation_box*, char, rvec );
-// void Distance_on_T3_Gen( rvec, rvec, simulation_box*, rvec );
-// void Inc_on_T3_Gen( rvec, rvec, simulation_box* );
-// int Get_Nbr_Box( simulation_box*, int, int, int );
-// rvec Get_Nbr_Box_Press( simulation_box*, int, int, int );
-// void Inc_Nbr_Box_Press( simulation_box*, int, int, int, rvec );
+//void Transform( rvec, simulation_box*, char, rvec );
+
+//void Distance_on_T3_Gen( rvec, rvec, simulation_box*, rvec );
+
+//void Inc_on_T3_Gen( rvec, rvec, simulation_box* );
+
+//int Get_Nbr_Box( simulation_box*, int, int, int );
+
+//rvec Get_Nbr_Box_Press( simulation_box*, int, int, int );
+
+//void Inc_Nbr_Box_Press( simulation_box*, int, int, int, rvec );
/* these functions assume that the coordinates are in triclinic system
this function returns cartesian norm but triclinic distance vector */
//real Sq_Distance_on_T3( rvec, rvec, simulation_box*, rvec );
+
//void Inc_on_T3( rvec, rvec, simulation_box* );
+
//real Metric_Product( rvec, rvec, simulation_box* );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/center_mass.cu b/PG-PuReMD/src/center_mass.cu
deleted file mode 100644
index 725cafbb..00000000
--- a/PG-PuReMD/src/center_mass.cu
+++ /dev/null
@@ -1,551 +0,0 @@
-#include "center_mass.h"
-#include "vector.h"
-#include "cuda_shuffle.h"
-
-CUDA_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)
-{
- extern __shared__ rvec xcm[];
- extern __shared__ rvec vcm[];
- extern __shared__ rvec amcm[];
-
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- //unsigned int xcm_id = threadIdx.x;
- unsigned int vcm_id = blockDim.x;
- unsigned int amcm_id = 2 *(blockDim.x);
-
- unsigned int index = 0;
- rvec tmp;
- real m;
-
- rvec_MakeZero (xcm [threadIdx.x]);
- rvec_MakeZero (vcm [vcm_id + threadIdx.x]);
- rvec_MakeZero (amcm[amcm_id + threadIdx.x]);
- rvec_MakeZero (tmp);
-
- if (i < n){
- m = sbp [ atoms[i].type ].mass;
- rvec_ScaledAdd (xcm [threadIdx.x], m, atoms [i].x);
- rvec_ScaledAdd (vcm [vcm_id + threadIdx.x], m, atoms [i].v);
- rvec_Cross (tmp, atoms[i].x, atoms [i].v);
- rvec_ScaledAdd (amcm[amcm_id + threadIdx.x], m, tmp);
- }
- __syncthreads ();
-
- for( int offset = blockDim.x / 2; offset > 0; offset >>= 1 ) {
-
- if ((threadIdx.x < offset)) {
- index = threadIdx.x + offset;
- rvec_Add (xcm [threadIdx.x], xcm[index]);
- rvec_Add (vcm [vcm_id + threadIdx.x], vcm[vcm_id + index]);
- rvec_Add (amcm[amcm_id + threadIdx.x], amcm[amcm_id + index]);
- }
- __syncthreads ();
- }
-
- if ((threadIdx.x == 0)){
- rvec_Copy (res_xcm[blockIdx.x], xcm[0]);
- rvec_Copy (res_vcm[blockIdx.x], vcm[vcm_id]);
- rvec_Copy (res_amcm[blockIdx.x], amcm[amcm_id]);
- }
-}
-
-#if defined( __SM_35__)
-CUDA_GLOBAL void center_of_mass_blocks_xcm (single_body_parameters *sbp, reax_atom *atoms,
- rvec *res_xcm,
- size_t n)
-{
- extern __shared__ rvec my_xcm[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int xcm_id = threadIdx.x;
- unsigned int index = 0;
- rvec xcm;
- real m;
-
- rvec_MakeZero (xcm);
-
- if (i < n){
- m = sbp [ atoms[i].type ].mass;
- rvec_ScaledAdd (xcm , m, atoms [i].x);
- }
- __syncthreads ();
-
- for (int z = 16; z >= 1; z /= 2){
- xcm[0] += shfl( xcm[0], z);
- xcm[1] += shfl( xcm[1], z);
- xcm[2] += shfl( xcm[2], z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0)
- rvec_Copy( my_xcm[ threadIdx.x >> 5], xcm );
- __syncthreads ();
-
- for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
-
- if ((threadIdx.x < offset)) {
- index = threadIdx.x + offset;
- rvec_Add (my_xcm [threadIdx.x], my_xcm[index]);
- }
- __syncthreads ();
- }
-
- if ((threadIdx.x == 0))
- rvec_Copy (res_xcm[blockIdx.x], my_xcm[0]);
-}
-
-CUDA_GLOBAL void center_of_mass_blocks_vcm (single_body_parameters *sbp, reax_atom *atoms,
- rvec *res_vcm,
- size_t n)
-{
- extern __shared__ rvec my_vcm[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
- rvec vcm;
- real m;
-
- rvec_MakeZero (vcm);
-
- if (i < n){
- m = sbp [ atoms[i].type ].mass;
- rvec_ScaledAdd (vcm , m, atoms [i].v);
- }
- __syncthreads ();
-
- for (int z = 16; z >= 1; z /= 2){
- vcm[0] += shfl( vcm[0], z);
- vcm[1] += shfl( vcm[1], z);
- vcm[2] += shfl( vcm[2], z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0)
- rvec_Copy( my_vcm[ threadIdx.x >> 5], vcm );
- __syncthreads ();
-
- for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
-
- if ((threadIdx.x < offset)) {
- index = threadIdx.x + offset;
- rvec_Add (my_vcm [threadIdx.x], my_vcm[index]);
- }
- __syncthreads ();
- }
-
- if ((threadIdx.x == 0))
- rvec_Copy (res_vcm[blockIdx.x], my_vcm[0]);
-}
-
-CUDA_GLOBAL void center_of_mass_blocks_amcm (single_body_parameters *sbp, reax_atom *atoms,
- rvec *res_amcm,
- size_t n)
-{
- extern __shared__ rvec my_amcm[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
- rvec amcm;
- real m;
- rvec tmp;
-
- rvec_MakeZero (amcm);
- rvec_MakeZero( tmp );
-
- if (i < n){
- m = sbp [ atoms[i].type ].mass;
- rvec_Cross (tmp, atoms[i].x, atoms [i].v);
- rvec_ScaledAdd (amcm, m, tmp);
- }
- __syncthreads ();
-
- for (int z = 16; z >= 1; z /= 2){
- amcm[0] += shfl( amcm[0], z);
- amcm[1] += shfl( amcm[1], z);
- amcm[2] += shfl( amcm[2], z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0)
- rvec_Copy( my_amcm[ threadIdx.x >> 5], amcm );
- __syncthreads ();
-
-
- for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
-
- if ((threadIdx.x < offset)) {
- index = threadIdx.x + offset;
- rvec_Add (my_amcm[threadIdx.x], my_amcm[index]);
- }
- __syncthreads ();
- }
-
- if ((threadIdx.x == 0)){
- rvec_Copy (res_amcm[blockIdx.x], my_amcm[0]);
- }
-}
-
-#endif
-
-
-CUDA_GLOBAL void 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[];
- extern __shared__ rvec sh_amcm[];
-
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- unsigned int xcm_id = threadIdx.x;
- unsigned int vcm_id = blockDim.x;
- unsigned int amcm_id = 2 * (blockDim.x);
-
- unsigned int index = 0;
- rvec t_xcm, t_vcm, t_amcm;
-
- rvec_MakeZero (t_xcm);
- rvec_MakeZero (t_vcm);
- rvec_MakeZero (t_amcm);
-
- if (i < n){
- rvec_Copy ( t_xcm, xcm[threadIdx.x]);
- rvec_Copy ( t_vcm, vcm[threadIdx.x]);
- rvec_Copy ( t_amcm, amcm[threadIdx.x]);
- }
-
- rvec_Copy (sh_xcm[xcm_id], t_xcm);
- rvec_Copy (sh_vcm[vcm_id + threadIdx.x], t_vcm);
- rvec_Copy (sh_amcm[amcm_id + threadIdx.x], t_amcm);
-
- __syncthreads ();
-
- for( int offset = blockDim.x / 2; offset > 0; offset >>= 1 ) {
-
- if (threadIdx.x < offset) {
- index = threadIdx.x + offset;
- rvec_Add (sh_xcm [threadIdx.x], sh_xcm[index]);
- rvec_Add (sh_vcm [vcm_id + threadIdx.x], sh_vcm[vcm_id + index]);
- rvec_Add (sh_amcm [amcm_id + threadIdx.x], sh_amcm[amcm_id + index]);
- }
- __syncthreads ();
- }
-
- if (threadIdx.x == 0){
- rvec_Copy (res_xcm[blockIdx.x], sh_xcm[0]);
- rvec_Copy (res_vcm[blockIdx.x], sh_vcm[vcm_id]);
- rvec_Copy (res_amcm[blockIdx.x], sh_amcm[amcm_id]);
- }
-}
-
-CUDA_GLOBAL void compute_center_mass (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[];
- extern __shared__ real xz[];
- extern __shared__ real yy[];
- extern __shared__ real yz[];
- extern __shared__ real zz[];
-
- unsigned int xx_i = threadIdx.x;
- unsigned int xy_i = blockDim.x;
- unsigned int xz_i = 2 * blockDim.x;
- unsigned int yy_i = 3 * blockDim.x;
- unsigned int yz_i = 4 * blockDim.x;
- unsigned int zz_i = 5 * blockDim.x;
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
-
- rvec diff, xcm;
- real m = 0;
- rvec_MakeZero (diff);
- xcm[0] = xcm0;
- 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){
- m = sbp[ atoms[i].type ].mass;
- rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
- xx[ xx_i ] = diff[0] * diff[0] * m;
- xy[ xy_i + threadIdx.x ] = diff[0] * diff[1] * m;
- xz[ xz_i + threadIdx.x ] = diff[0] * diff[2] * m;
- yy[ yy_i + threadIdx.x ] = diff[1] * diff[1] * m;
- yz[ yz_i + threadIdx.x ] = diff[1] * diff[2] * m;
- zz[ zz_i + threadIdx.x ] = diff[2] * diff[2] * m;
- }
- __syncthreads ();
-
- 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 ];
- xz[ xz_i + threadIdx.x ] += xz [ xz_i + index ];
- yy[ yy_i + threadIdx.x ] += yy [ yy_i + index ];
- yz[ yz_i + threadIdx.x ] += yz [ yz_i + index ];
- zz[ zz_i + threadIdx.x ] += zz [ zz_i + index ];
- }
- __syncthreads ();
- }
-
- 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 ];
- results [ blockIdx.x*6 + 3 ] = yy [ yy_i + 0 ];
- results [ blockIdx.x*6 + 4 ] = yz [ yz_i + 0 ];
- results [ blockIdx.x*6 + 5 ] = zz [ zz_i + 0 ];
- }
-}
-
-CUDA_GLOBAL void compute_center_mass (real *input, real *output, size_t n)
-{
- extern __shared__ real xx[];
- extern __shared__ real xy[];
- extern __shared__ real xz[];
- extern __shared__ real yy[];
- extern __shared__ real yz[];
- extern __shared__ real zz[];
-
- unsigned int xx_i = threadIdx.x;
- unsigned int xy_i = blockDim.x;
- unsigned int xz_i = 2 * blockDim.x;
- unsigned int yy_i = 3 * blockDim.x;
- unsigned int yz_i = 4 * blockDim.x;
- unsigned int zz_i = 5 * blockDim.x;
-
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
-
- 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)
- {
- xx [ xx_i ] = input [ threadIdx.x*6 + 0 ];
- xy [ xy_i + threadIdx.x ] = input [ threadIdx.x*6 + 1 ];
- xz [ xz_i + threadIdx.x ] = input [ threadIdx.x*6 + 2 ];
- yy [ yy_i + threadIdx.x ] = input [ threadIdx.x*6 + 3 ];
- yz [ yz_i + threadIdx.x ] = input [ threadIdx.x*6 + 4 ];
- zz [ zz_i + threadIdx.x ] = input [ threadIdx.x*6 + 5 ];
- }
- __syncthreads ();
-
- 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 ];
- xz [ xz_i + threadIdx.x ] += xz [ xz_i + index ];
- yy [ yy_i + threadIdx.x ] += yy [ yy_i + index ];
- yz [ yz_i + threadIdx.x ] += yz [ yz_i + index ];
- zz [ zz_i + threadIdx.x ] += zz [ zz_i + index ];
- }
- __syncthreads ();
- }
-
- if (threadIdx.x == 0)
- {
- output[0] = xx[0];
- output[1] = xy[xy_i];
- output[2] = xz[xz_i];
- output[3] = xz[yy_i];
- output[4] = xz[yz_i];
- output[5] = xz[zz_i];
- }
-}
-
-#if defined( __SM_35__)
-
-CUDA_GLOBAL void compute_center_mass_xx_xy (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n)
-{
- extern __shared__ real my_results_xx[];
- extern __shared__ real my_results_xy[];
-
- unsigned int xx_i = threadIdx.x;
- unsigned int xy_i = blockDim.x;
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
- real xx = 0;
- real xy = 0;
-
- rvec diff, xcm;
- real m = 0;
- rvec_MakeZero (diff);
- xcm[0] = xcm0;
- xcm[1] = xcm1;
- xcm[2] = xcm2;
-
-
- if (i < n){
- m = sbp[ atoms[i].type ].mass;
- rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
- xx = diff[0] * diff[0] * m;
- xy = diff[0] * diff[1] * m;
- }
- __syncthreads ();
-
- for (int z = 16; z <= 1; z++){
- xx += shfl( xx, z);
- xy += shfl( xy, z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0){
- my_results_xx[threadIdx.x >> 5] = xx;
- my_results_xy[threadIdx.x >> 5] = xy;
- }
- __syncthreads ();
-
- for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
- if (threadIdx.x < offset){
- index = threadIdx.x + offset;
- my_results_xx[ threadIdx.x ] += my_results_xx[ index ];
- my_results_xy[ xy_i + threadIdx.x ] += my_results_xy [ xy_i + index ];
- }
- __syncthreads ();
- }
-
- if (threadIdx.x == 0) {
- results [ blockIdx.x*6 ] = my_results_xx [ 0 ];
- results [ blockIdx.x*6 + 1 ] = my_results_xy [ xy_i + 0 ];
- }
-}
-
-CUDA_GLOBAL void compute_center_mass_xz_yy (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n)
-{
- extern __shared__ real my_results_xz[];
- extern __shared__ real my_results_yy[];
-
- unsigned int yy_i = blockDim.x;
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
- real xz = 0;
- real yy = 0;
-
- rvec diff, xcm;
- real m = 0;
- rvec_MakeZero (diff);
- xcm[0] = xcm0;
- xcm[1] = xcm1;
- xcm[2] = xcm2;
-
- if (i < n){
- m = sbp[ atoms[i].type ].mass;
- rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
- xz = diff[0] * diff[2] * m;
- yy = diff[1] * diff[1] * m;
- }
- __syncthreads ();
-
- for (int z = 16; z <= 1; z++){
- xz += shfl( xz, z);
- yy += shfl( yy, z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0){
- my_results_xz[threadIdx.x >> 5] = xz;
- my_results_yy[threadIdx.x >> 5] = yy;
- }
- __syncthreads ();
-
- for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
- if (threadIdx.x < offset){
- index = threadIdx.x + offset;
- my_results_xz[ threadIdx.x ] += my_results_xz [ index ];
- my_results_yy[ yy_i + threadIdx.x ] += my_results_yy [ yy_i + index ];
- }
- __syncthreads ();
- }
-
- if (threadIdx.x == 0) {
- results [ blockIdx.x*6 + 2 ] = my_results_xz [ 0 ];
- results [ blockIdx.x*6 + 3 ] = my_results_yy [ yy_i + 0 ];
- }
-}
-
-CUDA_GLOBAL void compute_center_mass_yz_zz (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n)
-{
- extern __shared__ real my_results_yz[];
- extern __shared__ real my_results_zz[];
-
- unsigned int zz_i = blockDim.x;
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- unsigned int index = 0;
- real yz = 0;
- real zz = 0;
-
- rvec diff, xcm;
- real m = 0;
- rvec_MakeZero (diff);
- xcm[0] = xcm0;
- xcm[1] = xcm1;
- xcm[2] = xcm2;
-
-
- if (i < n){
- m = sbp[ atoms[i].type ].mass;
- rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
- yz = diff[1] * diff[2] * m;
- zz = diff[2] * diff[2] * m;
- }
- __syncthreads ();
-
- for (int z = 16; z <= 1; z++){
- yz += shfl( yz, z);
- zz += shfl( zz, z);
- }
- __syncthreads ();
-
- if (threadIdx.x % 32 == 0){
- my_results_yz[threadIdx.x >> 5] = yz;
- my_results_zz[threadIdx.x >> 5] = zz;
- }
- __syncthreads ();
-
- for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
- if (threadIdx.x < offset){
- index = threadIdx.x + offset;
- my_results_yz[ threadIdx.x ] += my_results_yz [ index ];
- my_results_zz[ zz_i + threadIdx.x ] += my_results_zz [ zz_i + index ];
- }
- __syncthreads ();
- }
-
- if (threadIdx.x == 0) {
- results [ blockIdx.x*6 + 4 ] = my_results_yz [ 0 ];
- results [ blockIdx.x*6 + 5 ] = my_results_zz [ zz_i + 0 ];
- }
-}
-
-#endif
diff --git a/PG-PuReMD/src/center_mass.h b/PG-PuReMD/src/center_mass.h
deleted file mode 100644
index 113971ad..00000000
--- a/PG-PuReMD/src/center_mass.h
+++ /dev/null
@@ -1,49 +0,0 @@
-
-#ifndef __CENTER_MASS_H__
-#define __CENTER_MASS_H__
-
-#include "reax_types.h"
-#include "reax_types.h"
-
-CUDA_GLOBAL void center_of_mass_blocks (single_body_parameters *, reax_atom *,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n);
-
-#if defined(__SM_35__)
-CUDA_GLOBAL void center_of_mass_blocks_xcm (single_body_parameters *, reax_atom *,
- rvec *res_xcm,
- size_t n);
-CUDA_GLOBAL void center_of_mass_blocks_vcm (single_body_parameters *, reax_atom *,
- rvec *res_vcm,
- size_t n);
-CUDA_GLOBAL void center_of_mass_blocks_amcm (single_body_parameters *, reax_atom *,
- rvec *res_amcm,
- size_t n);
-#endif
-
-
-CUDA_GLOBAL void center_of_mass (rvec *xcm,
- rvec *vcm,
- rvec *amcm,
- rvec *res_xcm,
- rvec *res_vcm,
- rvec *res_amcm,
- size_t n);
-
-CUDA_GLOBAL void compute_center_mass (single_body_parameters *sbp,
- reax_atom *atoms,
- real *results,
- real xcm0, real xcm1, real xcm2,
- size_t n);
-
-CUDA_GLOBAL void compute_center_mass (real *input, real *output, size_t n);
-
-#if defined(__SM_35__)
-CUDA_GLOBAL void compute_center_mass_xx_xy (single_body_parameters *, reax_atom *, real *, real , real , real , size_t );
-CUDA_GLOBAL void compute_center_mass_xz_yy (single_body_parameters *, reax_atom *, real *, real , real , real , size_t );
-CUDA_GLOBAL void compute_center_mass_yz_zz (single_body_parameters *, reax_atom *, real *, real , real , real , size_t );
-#endif
-
-#endif
diff --git a/PG-PuReMD/src/charges.c b/PG-PuReMD/src/charges.c
index 6d695f56..8f53b65d 100644
--- a/PG-PuReMD/src/charges.c
+++ b/PG-PuReMD/src/charges.c
@@ -19,6 +19,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "charges.h"
#include "allocate.h"
@@ -27,12 +29,6 @@
#include "lin_alg.h"
#include "tool_box.h"
-#ifdef HAVE_CUDA
- #include "cuda_charges.h"
- #include "cuda_lin_alg.h"
- #include "cuda_validation.h"
-#endif
-
int compare_matrix_entry(const void *v1, const void *v2)
{
@@ -406,46 +402,6 @@ void Calculate_Charges( reax_system *system, storage *workspace,
}
-#ifdef HAVE_CUDA
-void Cuda_Calculate_Charges( reax_system *system, storage *workspace,
- mpi_datatypes *mpi_data )
-{
- int i, scale;
- real u;//, s_sum, t_sum;
- rvec2 my_sum, all_sum;
- reax_atom *atom;
- real *q;
-
- my_sum[0] = 0.0;
- my_sum[1] = 0.0;
- scale = sizeof(real) / sizeof(void);
- q = (real *) host_scratch;
- memset( q, 0, system->N * sizeof (real));
-
- cuda_charges_x( system, my_sum );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "Device: my_sum[0]: %f, my_sum[1]: %f\n",
- my_sum[0], my_sum[1] );
-#endif
-
- MPI_Allreduce( &my_sum, &all_sum, 2, MPI_DOUBLE, MPI_SUM, mpi_data->world );
-
- u = all_sum[0] / all_sum[1];
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "Device: u: %f \n", u );
-#endif
-
- cuda_charges_st( system, workspace, q, u );
-
- Dist( system, mpi_data, q, MPI_DOUBLE, scale, real_packer );
-
- cuda_charges_updateq( system, q );
-}
-#endif
-
-
void QEq( reax_system *system, control_params *control, simulation_data *data,
storage *workspace, output_controls *out_control,
mpi_datatypes *mpi_data )
@@ -504,57 +460,3 @@ void QEq( reax_system *system, control_params *control, simulation_data *data,
}
#endif
}
-
-
-#ifdef HAVE_CUDA
-void Cuda_QEq( reax_system *system, control_params *control, simulation_data
- *data, storage *workspace, output_controls *out_control, mpi_datatypes
- *mpi_data )
-{
- int s_matvecs, t_matvecs;
-
- Cuda_Init_MatVec( system, workspace );
-
- //if (data->step > 0) {
- // compare_rvec2 (workspace->b, dev_workspace->b, system->n, "b");
- // compare_rvec2 (workspace->x, dev_workspace->x, system->n, "x");
- // compare_array (workspace->b_s, dev_workspace->b_s, system->n, "b_s");
- // compare_array (workspace->b_t, dev_workspace->b_t, system->n, "b_t");
- //}
-
-//#ifdef __CUDA_DEBUG__
-// Init_MatVec( system, data, control, workspace, mpi_data );
-//#endif
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: initialized qEq\n", system->my_rank );
- //Print_Linear_System( system, control, workspace, data->step );
-#endif
-
- //MATRIX CHANGES
- s_matvecs = Cuda_dual_CG(system, workspace, &dev_workspace->H,
- dev_workspace->b, control->q_err, dev_workspace->x, mpi_data,
- out_control->log, data);
- t_matvecs = 0;
- //fprintf (stderr, "Device: First CG complated with iterations: %d \n", s_matvecs);
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: first CG completed\n", system->my_rank );
-#endif
-
- Cuda_Calculate_Charges( system, workspace, mpi_data );
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: computed charges\n", system->my_rank );
- //Print_Charges( system );
-#endif
-
-#if defined(LOG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- data->timing.s_matvecs += s_matvecs;
- data->timing.t_matvecs += t_matvecs;
- }
-#endif
-}
-#endif
diff --git a/PG-PuReMD/src/charges.h b/PG-PuReMD/src/charges.h
index faad0d09..08af5641 100644
--- a/PG-PuReMD/src/charges.h
+++ b/PG-PuReMD/src/charges.h
@@ -25,11 +25,16 @@
#include "reax_types.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void QEq( reax_system*, control_params*, simulation_data*,
storage*, output_controls*, mpi_datatypes* );
-void Cuda_QEq( reax_system*, control_params*, simulation_data*,
- storage*, output_controls*, mpi_datatypes* );
+#ifdef __cplusplus
+}
+#endif
#endif
diff --git a/PG-PuReMD/src/comm_tools.c b/PG-PuReMD/src/comm_tools.c
index 5a832aff..a8d46fcb 100644
--- a/PG-PuReMD/src/comm_tools.c
+++ b/PG-PuReMD/src/comm_tools.c
@@ -19,6 +19,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "comm_tools.h"
#include "grid.h"
diff --git a/PG-PuReMD/src/comm_tools.h b/PG-PuReMD/src/comm_tools.h
index 3b0b645f..a0e8d7e5 100644
--- a/PG-PuReMD/src/comm_tools.h
+++ b/PG-PuReMD/src/comm_tools.h
@@ -24,21 +24,36 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Check_MPI_Error( int, const char * );
+
void Setup_Comm( reax_system*, control_params*, mpi_datatypes* );
+
void Update_Comm( reax_system* );
void Sort_Boundary_Atoms( reax_system*, int, int, int, mpi_out_data* );
+
void Estimate_Boundary_Atoms( reax_system*, int, int, int, mpi_out_data* );
+
void Unpack_Exchange_Message( reax_system*, int, void*, int,
- neighbor_proc*, int );
+ neighbor_proc*, int );
+
void Unpack_Estimate_Message( reax_system*, int, void*, int,
- neighbor_proc*, int );
+ neighbor_proc*, int );
int SendRecv( reax_system*, mpi_datatypes*_data, MPI_Datatype, int*,
- message_sorter, unpacker, int );
+ message_sorter, unpacker, int );
void Comm_Atoms( reax_system*, control_params*, simulation_data*, storage*,
- reax_list**, mpi_datatypes*, int );
+ reax_list**, mpi_datatypes*, int );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/control.h b/PG-PuReMD/src/control.h
index c6c6ce6c..24cf0451 100644
--- a/PG-PuReMD/src/control.h
+++ b/PG-PuReMD/src/control.h
@@ -24,6 +24,16 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
char Read_Control_File( char*, control_params*, output_controls* );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/cuda_allocate.cu b/PG-PuReMD/src/cuda/cuda_allocate.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_allocate.cu
rename to PG-PuReMD/src/cuda/cuda_allocate.cu
index 7970e9f6..5c722e56 100644
--- a/PG-PuReMD/src/cuda_allocate.cu
+++ b/PG-PuReMD/src/cuda/cuda_allocate.cu
@@ -6,10 +6,10 @@
#include "cuda_neighbors.h"
#include "cuda_utils.h"
-#include "allocate.h"
-#include "index_utils.h"
-#include "tool_box.h"
-#include "vector.h"
+#include "../allocate.h"
+#include "../index_utils.h"
+#include "../tool_box.h"
+#include "../vector.h"
extern "C"
{
diff --git a/PG-PuReMD/src/cuda_allocate.h b/PG-PuReMD/src/cuda/cuda_allocate.h
similarity index 96%
rename from PG-PuReMD/src/cuda_allocate.h
rename to PG-PuReMD/src/cuda/cuda_allocate.h
index 57124046..0d78d932 100644
--- a/PG-PuReMD/src/cuda_allocate.h
+++ b/PG-PuReMD/src/cuda/cuda_allocate.h
@@ -1,7 +1,7 @@
#ifndef __CUDA_ALLOCATE_H_
#define __CUDA_ALLOCATE_H_
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
extern "C" {
diff --git a/PG-PuReMD/src/cuda_bond_orders.cu b/PG-PuReMD/src/cuda/cuda_bond_orders.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_bond_orders.cu
rename to PG-PuReMD/src/cuda/cuda_bond_orders.cu
index 6e4344aa..bb478a3a 100644
--- a/PG-PuReMD/src/cuda_bond_orders.cu
+++ b/PG-PuReMD/src/cuda/cuda_bond_orders.cu
@@ -2,11 +2,12 @@
#include "cuda_bond_orders.h"
#include "cuda_list.h"
-#include "index_utils.h"
-#include "bond_orders.h"
#include "cuda_utils.h"
#include "cuda_reduction.h"
+#include "../index_utils.h"
+#include "../bond_orders.h"
+
CUDA_GLOBAL void Cuda_Calculate_BO_init( reax_atom *my_atoms,
single_body_parameters *sbp, storage p_workspace, int N )
diff --git a/PG-PuReMD/src/cuda_bond_orders.h b/PG-PuReMD/src/cuda/cuda_bond_orders.h
similarity index 98%
rename from PG-PuReMD/src/cuda_bond_orders.h
rename to PG-PuReMD/src/cuda/cuda_bond_orders.h
index 8be3a592..a957b11b 100644
--- a/PG-PuReMD/src/cuda_bond_orders.h
+++ b/PG-PuReMD/src/cuda/cuda_bond_orders.h
@@ -2,10 +2,9 @@
#ifndef __CUDA_BOND_ORDERS_H__
#define __CUDA_BOND_ORDERS_H__
-#include "reax_types.h"
-#include "reax_types.h"
+#include "../reax_types.h"
-#include "vector.h"
+#include "../vector.h"
extern "C" {
diff --git a/PG-PuReMD/src/cuda_bonds.cu b/PG-PuReMD/src/cuda/cuda_bonds.cu
similarity index 98%
rename from PG-PuReMD/src/cuda_bonds.cu
rename to PG-PuReMD/src/cuda/cuda_bonds.cu
index 81f3444b..e3592630 100644
--- a/PG-PuReMD/src/cuda_bonds.cu
+++ b/PG-PuReMD/src/cuda/cuda_bonds.cu
@@ -19,13 +19,12 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reax_types.h"
-
-#include "index_utils.h"
-#include "reax_types.h"
+#include "cuda_bonds.h"
#include "cuda_list.h"
+#include "../index_utils.h"
+
CUDA_GLOBAL void Cuda_Bonds( reax_atom *my_atoms, global_parameters gp,
single_body_parameters *sbp, two_body_parameters *tbp,
diff --git a/PG-PuReMD/src/cuda_bonds.h b/PG-PuReMD/src/cuda/cuda_bonds.h
similarity index 69%
rename from PG-PuReMD/src/cuda_bonds.h
rename to PG-PuReMD/src/cuda/cuda_bonds.h
index d8a7d273..fd9126be 100644
--- a/PG-PuReMD/src/cuda_bonds.h
+++ b/PG-PuReMD/src/cuda/cuda_bonds.h
@@ -22,16 +22,12 @@
#ifndef __CUDA_BONDS_H_
#define __CUDA_BONDS_H_
-#include "reax_types.h"
-
-CUDA_GLOBAL void Cuda_Bonds( reax_atom *,
- global_parameters ,
- single_body_parameters *,
- two_body_parameters *,
- storage ,
- reax_list ,
- int , int ,
- real *
- );
+#include "../reax_types.h"
+
+
+CUDA_GLOBAL void Cuda_Bonds( reax_atom *, global_parameters,
+ single_body_parameters *, two_body_parameters *, storage,
+ reax_list, int, int, real * );
+
#endif
diff --git a/PG-PuReMD/src/cuda_charges.cu b/PG-PuReMD/src/cuda/cuda_charges.cu
similarity index 68%
rename from PG-PuReMD/src/cuda_charges.cu
rename to PG-PuReMD/src/cuda/cuda_charges.cu
index c8410952..ada6bf2f 100644
--- a/PG-PuReMD/src/cuda_charges.cu
+++ b/PG-PuReMD/src/cuda/cuda_charges.cu
@@ -21,12 +21,13 @@
#include "cuda_charges.h"
-#include "reax_types.h"
+#include "cuda_lin_alg.h"
#include "cuda_reduction.h"
#include "cuda_utils.h"
-
#include "cuda_validation.h"
+#include "../basic_comm.h"
+
CUDA_GLOBAL void k_init_matvec( reax_atom *my_atoms, single_body_parameters
*sbp, storage p_workspace, int n )
@@ -204,3 +205,93 @@ void cuda_charges_updateq( reax_system *system, real *q )
cudaThreadSynchronize( );
cudaCheckError( );
}
+
+
+void Cuda_Calculate_Charges( reax_system *system, storage *workspace,
+ mpi_datatypes *mpi_data )
+{
+ int i, scale;
+ real u;//, s_sum, t_sum;
+ rvec2 my_sum, all_sum;
+ reax_atom *atom;
+ real *q;
+
+ my_sum[0] = 0.0;
+ my_sum[1] = 0.0;
+ scale = sizeof(real) / sizeof(void);
+ q = (real *) host_scratch;
+ memset( q, 0, system->N * sizeof (real));
+
+ cuda_charges_x( system, my_sum );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "Device: my_sum[0]: %f, my_sum[1]: %f\n",
+ my_sum[0], my_sum[1] );
+#endif
+
+ MPI_Allreduce( &my_sum, &all_sum, 2, MPI_DOUBLE, MPI_SUM, mpi_data->world );
+
+ u = all_sum[0] / all_sum[1];
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "Device: u: %f \n", u );
+#endif
+
+ cuda_charges_st( system, workspace, q, u );
+
+ Dist( system, mpi_data, q, MPI_DOUBLE, scale, real_packer );
+
+ cuda_charges_updateq( system, q );
+}
+
+
+void Cuda_QEq( reax_system *system, control_params *control, simulation_data
+ *data, storage *workspace, output_controls *out_control, mpi_datatypes
+ *mpi_data )
+{
+ int s_matvecs, t_matvecs;
+
+ Cuda_Init_MatVec( system, workspace );
+
+ //if (data->step > 0) {
+ // compare_rvec2 (workspace->b, dev_workspace->b, system->n, "b");
+ // compare_rvec2 (workspace->x, dev_workspace->x, system->n, "x");
+ // compare_array (workspace->b_s, dev_workspace->b_s, system->n, "b_s");
+ // compare_array (workspace->b_t, dev_workspace->b_t, system->n, "b_t");
+ //}
+
+//#ifdef __CUDA_DEBUG__
+// Init_MatVec( system, data, control, workspace, mpi_data );
+//#endif
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: initialized qEq\n", system->my_rank );
+ //Print_Linear_System( system, control, workspace, data->step );
+#endif
+
+ //MATRIX CHANGES
+ s_matvecs = Cuda_dual_CG(system, workspace, &dev_workspace->H,
+ dev_workspace->b, control->q_err, dev_workspace->x, mpi_data,
+ out_control->log, data);
+ t_matvecs = 0;
+ //fprintf (stderr, "Device: First CG complated with iterations: %d \n", s_matvecs);
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: first CG completed\n", system->my_rank );
+#endif
+
+ Cuda_Calculate_Charges( system, workspace, mpi_data );
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: computed charges\n", system->my_rank );
+ //Print_Charges( system );
+#endif
+
+#if defined(LOG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ data->timing.s_matvecs += s_matvecs;
+ data->timing.t_matvecs += t_matvecs;
+ }
+#endif
+}
diff --git a/PG-PuReMD/src/cuda_charges.h b/PG-PuReMD/src/cuda/cuda_charges.h
similarity index 89%
rename from PG-PuReMD/src/cuda_charges.h
rename to PG-PuReMD/src/cuda/cuda_charges.h
index 2d421389..d1922a48 100644
--- a/PG-PuReMD/src/cuda_charges.h
+++ b/PG-PuReMD/src/cuda/cuda_charges.h
@@ -22,7 +22,7 @@
#ifndef __CUDA_CHARGES_H_
#define __CUDA_CHARGES_H_
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
extern "C" {
@@ -37,6 +37,9 @@ void cuda_charges_st( reax_system *, storage *, real *, real );
void cuda_charges_updateq( reax_system *, real * );
+void Cuda_QEq( reax_system*, control_params*, simulation_data*,
+ storage*, output_controls*, mpi_datatypes* );
+
#ifdef __cplusplus
}
diff --git a/PG-PuReMD/src/cuda_copy.cu b/PG-PuReMD/src/cuda/cuda_copy.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_copy.cu
rename to PG-PuReMD/src/cuda/cuda_copy.cu
index a3bfca30..42055875 100644
--- a/PG-PuReMD/src/cuda_copy.cu
+++ b/PG-PuReMD/src/cuda/cuda_copy.cu
@@ -2,7 +2,8 @@
#include "cuda_copy.h"
#include "cuda_utils.h"
-#include "vector.h"
+
+#include "../vector.h"
/* Copy grid info from host to device */
diff --git a/PG-PuReMD/src/cuda_copy.h b/PG-PuReMD/src/cuda/cuda_copy.h
similarity index 93%
rename from PG-PuReMD/src/cuda_copy.h
rename to PG-PuReMD/src/cuda/cuda_copy.h
index 51c4314c..72bf992c 100644
--- a/PG-PuReMD/src/cuda_copy.h
+++ b/PG-PuReMD/src/cuda/cuda_copy.h
@@ -1,24 +1,30 @@
#ifndef __CUDA_COPY_H_
#define __CUDA_COPY_H_
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
void Sync_Atoms( reax_system * );
+
void Sync_Grid( grid *, grid * );
+
void Sync_System( reax_system * );
void Prep_Device_For_Output( reax_system *, simulation_data * );
+
void Output_Sync_Lists( reax_list *host, reax_list *device, int type );
+
void Output_Sync_Atoms( reax_system * );
-void Output_Sync_Simulation_Data( simulation_data *, simulation_data * );
+void Output_Sync_Simulation_Data( simulation_data *, simulation_data * );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_environment.cu b/PG-PuReMD/src/cuda/cuda_environment.cu
similarity index 100%
rename from PG-PuReMD/src/cuda_environment.cu
rename to PG-PuReMD/src/cuda/cuda_environment.cu
diff --git a/PG-PuReMD/src/cuda_environment.h b/PG-PuReMD/src/cuda/cuda_environment.h
similarity index 56%
rename from PG-PuReMD/src/cuda_environment.h
rename to PG-PuReMD/src/cuda/cuda_environment.h
index f8ae3cd0..1cbcc92c 100644
--- a/PG-PuReMD/src/cuda_environment.h
+++ b/PG-PuReMD/src/cuda/cuda_environment.h
@@ -2,15 +2,19 @@
#ifndef __CUDA_ENVIRONMENT_H__
#define __CUDA_ENVIRONMENT_H__
+#include "../reax_types.h"
+
+
#ifdef __cplusplus
extern "C" {
#endif
-void Setup_Cuda_Environment (int, int, int);
-void Cleanup_Cuda_Environment ();
+void Setup_Cuda_Environment( int, int, int );
+void Cleanup_Cuda_Environment( );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_forces.cu b/PG-PuReMD/src/cuda/cuda_forces.cu
similarity index 89%
rename from PG-PuReMD/src/cuda_forces.cu
rename to PG-PuReMD/src/cuda/cuda_forces.cu
index 831a5694..a790b1a8 100644
--- a/PG-PuReMD/src/cuda_forces.cu
+++ b/PG-PuReMD/src/cuda/cuda_forces.cu
@@ -1,24 +1,27 @@
#include "cuda_forces.h"
-#include "reax_types.h"
-#include "cuda_list.h"
-#include "cuda_utils.h"
+#include "cuda_bonds.h"
+#include "cuda_bond_orders.h"
+#include "cuda_charges.h"
#include "cuda_helpers.h"
+#include "cuda_hydrogen_bonds.h"
+#include "cuda_lin_alg.h"
+#include "cuda_list.h"
+#include "cuda_multi_body.h"
#include "cuda_neighbors.h"
-#include "cuda_bond_orders.h"
+#include "cuda_nonbonded.h"
#include "cuda_reduction.h"
-#include "cuda_bonds.h"
-#include "cuda_multi_body.h"
-#include "cuda_valence_angles.h"
#include "cuda_torsion_angles.h"
-#include "cuda_hydrogen_bonds.h"
-#include "tool_box.h"
-#include "cuda_nonbonded.h"
+#include "cuda_utils.h"
+#include "cuda_valence_angles.h"
+#include "cuda_validation.h"
-#include "index_utils.h"
-#include "vector.h"
-#include "forces.h"
+#include "../basic_comm.h"
+#include "../forces.h"
+#include "../index_utils.h"
+#include "../tool_box.h"
+#include "../vector.h"
CUDA_GLOBAL void k_disable_hydrogen_bonding( control_params *control )
@@ -1719,3 +1722,201 @@ void Cuda_Compute_NonBonded_Forces( reax_system *system, control_params *control
Cuda_NonBonded_Energy( system, control, workspace, data,
lists, out_control, (control->tabulate == 0) ? false: true );
}
+
+
+void Cuda_Compute_Total_Force( reax_system *system, control_params *control,
+ simulation_data *data, storage *workspace,
+ reax_list **lists, mpi_datatypes *mpi_data )
+{
+ rvec *f;
+
+ f = (rvec *) host_scratch;
+ memset( f, 0, sizeof(rvec) * system->N );
+
+ Cuda_Total_Forces( system, control, data, workspace );
+
+#if defined(PURE_REAX)
+ /* now all forces are computed to their partially-final values
+ * based on the neighbors information each processor has had.
+ * final values of force on each atom needs to be computed by adding up
+ * all partially-final pieces */
+
+ //MVAPICH2
+ copy_host_device( f, dev_workspace->f, sizeof(rvec) * system->N ,
+ cudaMemcpyDeviceToHost, "total_force:f:get" );
+
+ Coll( system, mpi_data, f, mpi_data->mpi_rvec,
+ sizeof(rvec) / sizeof(void), rvec_unpacker );
+
+ copy_host_device( f, dev_workspace->f, sizeof(rvec) * system->N,
+ cudaMemcpyHostToDevice, "total_force:f:put" );
+
+ Cuda_Total_Forces_PURE( system, dev_workspace );
+#endif
+
+}
+
+
+int Cuda_Compute_Forces( reax_system *system, control_params *control,
+ simulation_data *data, storage *workspace, reax_list **lists,
+ output_controls *out_control, mpi_datatypes *mpi_data )
+{
+ int charge_flag, retVal;
+
+#if defined(LOG_PERFORMANCE)
+ real t_start = 0;
+
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ t_start = Get_Time( );
+ }
+#endif
+
+ retVal = SUCCESS;
+
+ /********* init forces ************/
+ if ( control->charge_freq && (data->step - data->prev_steps) % control->charge_freq == 0 )
+ {
+ charge_flag = TRUE;
+ }
+ else
+ {
+ charge_flag = FALSE;
+ }
+
+ if ( charge_flag == TRUE )
+ {
+ retVal = Cuda_Init_Forces( system, control, data, workspace, lists, out_control );
+
+// int i;
+// static reax_list **temp_lists;
+//
+// if ( data->step == 0 )
+// {
+// temp_lists = (reax_list **) smalloc( LIST_N * sizeof (reax_list *), "temp_lists" );
+// for ( i = 0; i < LIST_N; ++i )
+// {
+// temp_lists[i] = (reax_list *) smalloc( sizeof(reax_list), "lists[i]" );
+// temp_lists[i]->allocated = FALSE;
+// }
+// Make_List( (*dev_lists + BONDS)->n, (*dev_lists + BONDS)->num_intrs,
+// TYP_BOND, *temp_lists + BONDS );
+// Make_List( (*dev_lists + HBONDS)->n, (*dev_lists + HBONDS)->num_intrs,
+// TYP_HBOND, *temp_lists + HBONDS );
+// }
+// else
+// {
+// Delete_List( *temp_lists + BONDS );
+// Make_List( (*dev_lists + BONDS)->n, (*dev_lists + BONDS)->num_intrs,
+// TYP_BOND, *temp_lists + BONDS );
+// Delete_List( *temp_lists + HBONDS );
+// Make_List( (*dev_lists + HBONDS)->n, (*dev_lists + HBONDS)->num_intrs,
+// TYP_HBOND, *temp_lists + HBONDS );
+//
+// }
+// Output_Sync_Lists( *temp_lists + BONDS, *dev_lists + BONDS, TYP_BOND );
+// Print_Bonds( system, temp_lists, control );
+// Output_Sync_Lists( *temp_lists + HBONDS, *dev_lists + HBONDS, TYP_HBOND );
+// Print_HBonds( system, temp_lists, control, data->step );
+// Print_HBond_Indices( system, temp_lists, control, data->step );
+// exit( 0 );
+ }
+ else
+ {
+ retVal = Cuda_Init_Forces_No_Charges( system, control, data, workspace, lists, out_control );
+ }
+
+ if ( retVal == SUCCESS )
+ {
+ //validate_sparse_matrix( system, workspace );
+
+#if defined(LOG_PERFORMANCE)
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &(data->timing.init_forces) );
+ }
+#endif
+
+ /********* bonded interactions ************/
+ retVal = Cuda_Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
+
+#if defined(LOG_PERFORMANCE)
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &(data->timing.bonded) );
+ }
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d: completed bonded\n",
+ system->my_rank, data->step );
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+ }
+
+ if ( retVal == SUCCESS )
+ {
+ /**************** charges ************************/
+#if defined(PURE_REAX)
+ if ( charge_flag == TRUE )
+ {
+ Cuda_QEq( system, control, data, workspace, out_control, mpi_data );
+ }
+
+#if defined(LOG_PERFORMANCE)
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &(data->timing.qEq) );
+ }
+#endif
+
+#if defined(DEBUG_FOCUS)
+ fprintf(stderr, "p%d @ step%d: qeq completed\n", system->my_rank, data->step);
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+#endif //PURE_REAX
+
+ /********* nonbonded interactions ************/
+ Cuda_Compute_NonBonded_Forces( system, control, data, workspace,
+ lists, out_control, mpi_data );
+
+#if defined(LOG_PERFORMANCE)
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &(data->timing.nonb) );
+ }
+#endif
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d: nonbonded forces completed\n",
+ system->my_rank, data->step );
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+
+ /*********** total force ***************/
+ Cuda_Compute_Total_Force( system, control, data, workspace, lists, mpi_data );
+
+#if defined(LOG_PERFORMANCE)
+ //MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &(data->timing.bonded) );
+ }
+#endif
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d: total forces computed\n",
+ system->my_rank, data->step );
+ //Print_Total_Force( system, data, workspace );
+ MPI_Barrier( MPI_COMM_WORLD );
+
+#endif
+
+// Print_Forces( system );
+ }
+
+ return retVal;
+}
diff --git a/PG-PuReMD/src/cuda_forces.h b/PG-PuReMD/src/cuda/cuda_forces.h
similarity index 86%
rename from PG-PuReMD/src/cuda_forces.h
rename to PG-PuReMD/src/cuda/cuda_forces.h
index 9dc3da66..4abdb52f 100644
--- a/PG-PuReMD/src/cuda_forces.h
+++ b/PG-PuReMD/src/cuda/cuda_forces.h
@@ -2,7 +2,7 @@
#ifndef __CUDA_FORCES_H__
#define __CUDA_FORCES_H__
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
@@ -32,6 +32,9 @@ void Cuda_Compute_NonBonded_Forces( reax_system *, control_params *,
simulation_data *, storage *, reax_list **, output_controls *,
mpi_datatypes * );
+int Cuda_Compute_Forces( reax_system*, control_params*, simulation_data*,
+ storage*, reax_list**, output_controls*, mpi_datatypes* );
+
void Print_Forces( reax_system * );
diff --git a/PG-PuReMD/src/cuda_helpers.h b/PG-PuReMD/src/cuda/cuda_helpers.h
similarity index 97%
rename from PG-PuReMD/src/cuda_helpers.h
rename to PG-PuReMD/src/cuda/cuda_helpers.h
index a4943a5f..b14f45b3 100644
--- a/PG-PuReMD/src/cuda_helpers.h
+++ b/PG-PuReMD/src/cuda/cuda_helpers.h
@@ -1,7 +1,7 @@
#ifndef __CUDA_HELPERS__
#define __CUDA_HELPERS__
-#include "reax_types.h"
+#include "../reax_types.h"
CUDA_DEVICE static inline int cuda_strcmp( char * a,
diff --git a/PG-PuReMD/src/cuda_hydrogen_bonds.cu b/PG-PuReMD/src/cuda/cuda_hydrogen_bonds.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_hydrogen_bonds.cu
rename to PG-PuReMD/src/cuda/cuda_hydrogen_bonds.cu
index 95eda081..18cdbb57 100644
--- a/PG-PuReMD/src/cuda_hydrogen_bonds.cu
+++ b/PG-PuReMD/src/cuda/cuda_hydrogen_bonds.cu
@@ -21,16 +21,14 @@
#include "cuda_hydrogen_bonds.h"
-#include "reax_types.h"
-#include "index_utils.h"
-
#include "cuda_valence_angles.h"
#include "cuda_helpers.h"
#include "cuda_list.h"
-#include "vector.h"
-
#include "cuda_shuffle.h"
+#include "../index_utils.h"
+#include "../vector.h"
+
CUDA_GLOBAL void Cuda_Hydrogen_Bonds( reax_atom *my_atoms, single_body_parameters *sbp,
hbond_parameters *d_hbp, global_parameters gp, control_params *control,
diff --git a/PG-PuReMD/src/cuda/cuda_hydrogen_bonds.h b/PG-PuReMD/src/cuda/cuda_hydrogen_bonds.h
new file mode 100644
index 00000000..606196b4
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_hydrogen_bonds.h
@@ -0,0 +1,48 @@
+/*----------------------------------------------------------------------
+ PuReMD - Purdue ReaxFF Molecular Dynamics Program
+
+ Copyright (2010) Purdue University
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Joseph Fogarty, jcfogart@mail.usf.edu
+ Sagar Pandit, pandit@usf.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_HBONDS_H_
+#define __CUDA_HBONDS_H_
+
+#include "../reax_types.h"
+
+
+CUDA_GLOBAL void Cuda_Hydrogen_Bonds_HNbrs( reax_atom *,
+ storage, reax_list );
+
+CUDA_GLOBAL void Cuda_Hydrogen_Bonds_HNbrs_BL( reax_atom *,
+ storage, reax_list, int );
+
+CUDA_GLOBAL void Cuda_Hydrogen_Bonds_PostProcess( reax_atom *,
+ storage, reax_list, int );
+
+CUDA_GLOBAL void Cuda_Hydrogen_Bonds( reax_atom *,
+ single_body_parameters *, hbond_parameters *,
+ global_parameters, control_params *, storage ,
+ reax_list, reax_list, int, int, real *, rvec * );
+
+CUDA_GLOBAL void Cuda_Hydrogen_Bonds_MT( reax_atom *,
+ single_body_parameters *, hbond_parameters *,
+ global_parameters , control_params *, storage,
+ reax_list, reax_list, int, int, real *, rvec * );
+
+
+#endif
diff --git a/PG-PuReMD/src/cuda/cuda_init_md.cu b/PG-PuReMD/src/cuda/cuda_init_md.cu
new file mode 100644
index 00000000..fb1ac0df
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_init_md.cu
@@ -0,0 +1,412 @@
+
+#include "cuda_init_md.h"
+
+#include "cuda_allocate.h"
+#include "cuda_list.h"
+#include "cuda_copy.h"
+#include "cuda_forces.h"
+#include "cuda_integrate.h"
+#include "cuda_neighbors.h"
+#include "cuda_reset_tools.h"
+#include "cuda_system_props.h"
+#include "cuda_utils.h"
+#include "cuda_validation.h"
+
+#if defined(PURE_REAX)
+ #include "../box.h"
+ #include "../comm_tools.h"
+ #include "../grid.h"
+ #include "../init_md.h"
+ #include "../integrate.h"
+ #include "../io_tools.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+ #include "../lookup.h"
+#ifdef __cplusplus
+}
+#endif
+ #include "../random.h"
+ #include "../reset_tools.h"
+ #include "../tool_box.h"
+ #include "../vector.h"
+#elif defined(LAMMPS_REAX)
+ #include "../reax_box.h"
+ #include "../reax_comm_tools.h"
+ #include "../reax_grid.h"
+ #include "../reax_init_md.h"
+ #include "../reax_integrate.h"
+ #include "../reax_io_tools.h"
+ #include "../reax_list.h"
+ #include "../reax_lookup.h"
+ #include "../reax_random.h"
+ #include "../reax_reset_tools.h"
+ #include "../reax_tool_box.h"
+ #include "../reax_vector.h"
+#endif
+
+
+void Cuda_Init_ScratchArea( )
+{
+ cuda_malloc( (void **)&scratch, DEVICE_SCRATCH_SIZE, TRUE, "device:scratch" );
+
+ host_scratch = (void *) smalloc( HOST_SCRATCH_SIZE, "host:scratch" );
+}
+
+
+int Cuda_Init_System( reax_system *system, control_params *control,
+ simulation_data *data, storage *workspace,
+ mpi_datatypes *mpi_data, char *msg )
+{
+ int i, ret;
+ reax_atom *atom;
+ int nrecv[MAX_NBRS];
+
+ Setup_New_Grid( system, control, MPI_COMM_WORLD );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d GRID:\n", system->my_rank );
+ Print_Grid( &(system->my_grid), stderr );
+#endif
+
+ Bin_My_Atoms( system, &(workspace->realloc) );
+ Reorder_My_Atoms( system, workspace );
+
+ /* estimate N and total capacity */
+ for ( i = 0; i < MAX_NBRS; ++i )
+ {
+ nrecv[i] = 0;
+ }
+
+ MPI_Barrier( MPI_COMM_WORLD );
+ system->max_recved = 0;
+ system->N = SendRecv( system, mpi_data, mpi_data->boundary_atom_type, nrecv,
+ Estimate_Boundary_Atoms, Unpack_Estimate_Message, TRUE );
+ system->total_cap = MAX( (int)(system->N * SAFE_ZONE), MIN_CAP );
+ Bin_Boundary_Atoms( system );
+
+ /* Sync atoms here to continue the computation */
+ dev_alloc_system( system );
+ Sync_System( system );
+
+ /* estimate numH and Hcap */
+ Cuda_Reset_Atoms( system, control );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d: n=%d local_cap=%d\n",
+ system->my_rank, system->n, system->local_cap );
+ fprintf( stderr, "p%d: N=%d total_cap=%d\n",
+ system->my_rank, system->N, system->total_cap );
+ fprintf( stderr, "p%d: numH=%d H_cap=%d\n",
+ system->my_rank, system->numH, system->Hcap );
+#endif
+
+ Cuda_Compute_Total_Mass( system, data, mpi_data->comm_mesh3D );
+
+ Cuda_Compute_Center_of_Mass( system, data, mpi_data, mpi_data->comm_mesh3D );
+
+// if( Reposition_Atoms( system, control, data, mpi_data, msg ) == FAILURE )
+// {
+// return FAILURE;
+// }
+
+ /* 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 );
+ }
+
+ Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
+
+ return SUCCESS;
+}
+
+
+void Cuda_Init_Simulation_Data( reax_system *system, control_params *control,
+ simulation_data *data, char *msg )
+{
+ dev_alloc_simulation_data( data );
+
+ Reset_Simulation_Data( data );
+
+ if ( !control->restart )
+ {
+ data->step = data->prev_steps = 0;
+ }
+
+ switch ( control->ensemble )
+ {
+ case NVE:
+ data->N_f = 3 * system->bigN;
+ Cuda_Evolve = Velocity_Verlet_NVE;
+ control->virial = 0;
+ break;
+
+ case bNVT:
+ data->N_f = 3 * system->bigN + 1;
+ Cuda_Evolve = Cuda_Velocity_Verlet_Berendsen_NVT;
+ control->virial = 0;
+ break;
+
+ case nhNVT:
+ fprintf( stderr, "[WARNING] Nose-Hoover NVT is still under testing.\n" );
+ data->N_f = 3 * system->bigN + 1;
+ Cuda_Evolve = Velocity_Verlet_Nose_Hoover_NVT_Klein;
+ control->virial = 0;
+ if ( !control->restart || (control->restart && control->random_vel) )
+ {
+ data->therm.G_xi = control->Tau_T *
+ (2.0 * data->sys_en.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;
+ }
+ break;
+
+ case sNPT: /* Semi-Isotropic NPT */
+ data->N_f = 3 * system->bigN + 4;
+ Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
+ control->virial = 1;
+ if ( !control->restart )
+ {
+ Reset_Pressures( data );
+ }
+ break;
+
+ case iNPT: /* Isotropic NPT */
+ data->N_f = 3 * system->bigN + 2;
+ Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
+ control->virial = 1;
+ if ( !control->restart )
+ {
+ Reset_Pressures( data );
+ }
+ break;
+
+ case NPT: /* Anisotropic NPT */
+ data->N_f = 3 * system->bigN + 9;
+ Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
+ control->virial = 1;
+
+ fprintf( stderr, "p%d: init_simulation_data: option not yet implemented\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, INVALID_INPUT );
+ break;
+
+ default:
+ fprintf( stderr, "p%d: init_simulation_data: ensemble not recognized\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, INVALID_INPUT );
+ }
+
+ /* initialize the timer(s) */
+ MPI_Barrier( MPI_COMM_WORLD );
+ if ( system->my_rank == MASTER_NODE )
+ {
+ data->timing.start = Get_Time( );
+
+#if defined(LOG_PERFORMANCE)
+ Reset_Timing( &data->timing );
+#endif
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "data->N_f: %8.3f\n", data->N_f );
+#endif
+}
+
+
+void Cuda_Init_Workspace( reax_system *system, control_params *control,
+ storage *workspace, char *msg )
+{
+ dev_alloc_workspace( system, control, dev_workspace,
+ system->local_cap, system->total_cap, msg );
+
+ memset( &(workspace->realloc), 0, sizeof(reallocate_data) );
+ Cuda_Reset_Workspace( system, workspace );
+
+ /* Initialize the Taper function */
+ Init_Taper( control, dev_workspace );
+}
+
+
+int Cuda_Init_Lists( reax_system *system, control_params *control,
+ simulation_data *data, storage *workspace, reax_list **lists,
+ mpi_datatypes *mpi_data, char *msg )
+{
+ int ret;
+ int Htop;
+
+ /* ignore returned error, as system->d_max_far_nbrs was not valid */
+ ret = Cuda_Estimate_Neighbors( system, data->step );
+
+ Dev_Make_List( system->total_cap, system->total_far_nbrs,
+ TYP_FAR_NEIGHBOR, *dev_lists + FAR_NBRS );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d: allocated far_nbrs: num_far=%d, space=%dMB\n",
+ system->my_rank, system->total_far_nbrs,
+ (int)(system->total_far_nbrs * sizeof(far_neighbor_data) / (1024 * 1024)) );
+ fprintf( stderr, "N: %d and total_cap: %d \n", system->N, system->total_cap );
+#endif
+
+ Cuda_Init_Neighbor_Indices( system );
+
+ Cuda_Generate_Neighbor_Lists( system, data, workspace, dev_lists );
+
+ /* estimate storage for bonds and hbonds */
+ Cuda_Estimate_Storages( system, control, dev_lists, &(dev_workspace->H), data->step );
+
+ /* estimate storage for charge sparse matrix */
+// Cuda_Estimate_Storage_Sparse_Matrix( system, control, data, dev_lists );
+
+ dev_alloc_matrix( &(dev_workspace->H), system->total_cap, system->total_cm_entries );
+
+ Cuda_Init_Sparse_Matrix_Indices( system, &(dev_workspace->H) );
+
+ //MATRIX CHANGES
+ //workspace->L = NULL;
+ //workspace->U = NULL;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p:%d - allocated H matrix: max_entries: %d, cap: %d \n",
+ system->my_rank, system->total_cm_entries, dev_workspace->H.m );
+ fprintf( stderr, "p%d: allocated H matrix: Htop=%d, space=%dMB\n",
+ system->my_rank, Htop,
+ (int)(Htop * sizeof(sparse_matrix_entry) / (1024 * 1024)) );
+#endif
+
+ if ( control->hbond_cut > 0.0 && system->numH > 0 )
+ {
+ Dev_Make_List( system->total_cap, system->total_hbonds, TYP_HBOND, *dev_lists + HBONDS );
+// Make_List( system->total_cap, system->total_hbonds, TYP_HBOND, *lists + HBONDS );
+
+ Cuda_Init_HBond_Indices( system );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d: allocated hbonds: total_hbonds=%d, space=%dMB\n",
+ system->my_rank, system->total_hbonds,
+ (int)(system->total_hbonds * sizeof(hbond_data) / (1024 * 1024)) );
+#endif
+ }
+
+ /* bonds list */
+ Dev_Make_List( system->total_cap, system->total_bonds, TYP_BOND, *dev_lists + BONDS );
+// Make_List( system->total_cap, system->total_bonds, TYP_BOND, *lists + BONDS );
+
+ Cuda_Init_Bond_Indices( system );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d: allocated bonds: total_bonds=%d, space=%dMB\n",
+ system->my_rank, total_bonds,
+ (int)(total_bonds * sizeof(bond_data) / (1024 * 1024)) );
+#endif
+
+ /* 3bodies list: since a more accurate estimate of the num.
+ * of three body interactions requires that bond orders have
+ * been computed, delay estimation until for computation */
+
+ return SUCCESS;
+}
+
+
+void Cuda_Initialize( reax_system *system, control_params *control,
+ simulation_data *data, storage *workspace,
+ reax_list **lists, output_controls *out_control,
+ mpi_datatypes *mpi_data )
+{
+ char msg[MAX_STR];
+ real t_start, t_end;
+
+ /* HOST/DEVICE SCRATCH */
+ Cuda_Init_ScratchArea( );
+
+ /* MPI_DATATYPES */
+ if ( Init_MPI_Datatypes( system, workspace, mpi_data, msg ) == FAILURE )
+ {
+ fprintf( stderr, "p%d: init_mpi_datatypes: could not create datatypes\n",
+ system->my_rank );
+ fprintf( stderr, "p%d: mpi_data couldn't be initialized! terminating.\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
+ }
+
+ /* SYSTEM */
+ if ( Cuda_Init_System( system, control, data, workspace, mpi_data, msg ) == FAILURE )
+ {
+ fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
+ fprintf( stderr, "p%d: system could not be initialized! terminating.\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
+ }
+
+ /* GRID */
+ dev_alloc_grid( system );
+ Sync_Grid( &system->my_grid, &system->d_my_grid );
+
+ //validate_grid( system );
+
+ /* SIMULATION_DATA */
+ Cuda_Init_Simulation_Data( system, control, data, msg );
+
+ /* WORKSPACE */
+ Cuda_Init_Workspace( system, control, workspace, msg );
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: initialized workspace\n", system->my_rank );
+#endif
+
+ //Sync the taper here from host to device.
+
+ /* CONTROL */
+ dev_alloc_control( control );
+
+ /* LISTS */
+ if ( Cuda_Init_Lists( system, control, data, workspace, lists, mpi_data, msg ) ==
+ FAILURE )
+ {
+ fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
+ fprintf( stderr, "p%d: system could not be initialized! terminating.\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: initialized lists\n", system->my_rank );
+#endif
+
+ /* OUTPUT Files */
+ if ( Init_Output_Files( system, control, out_control, mpi_data, msg ) == FAILURE )
+ {
+ fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
+ fprintf( stderr, "p%d: could not open output files! terminating...\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: output files opened\n", system->my_rank );
+#endif
+
+ /* Lookup Tables */
+ if ( control->tabulate )
+ {
+ if ( Init_Lookup_Tables( system, control, dev_workspace->Tap, mpi_data, msg ) == FAILURE )
+ {
+ fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
+ fprintf( stderr, "p%d: couldn't create lookup table! terminating.\n",
+ system->my_rank );
+ MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: initialized lookup tables\n", system->my_rank );
+#endif
+ }
+
+#if defined(DEBUG)
+ fprintf( stderr, "p%d: Device Initialization Done \n", system->my_rank );
+#endif
+}
+
+
diff --git a/PG-PuReMD/src/cuda/cuda_init_md.h b/PG-PuReMD/src/cuda/cuda_init_md.h
new file mode 100644
index 00000000..328674a5
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_init_md.h
@@ -0,0 +1,22 @@
+
+#ifndef __CUDA_INIT_MD_H__
+#define __CUDA_INIT_MD_H__
+
+#include "../reax_types.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Cuda_Init_ScratchArea( );
+
+void Cuda_Initialize( reax_system*, control_params*, simulation_data*,
+ storage*, reax_list**, output_controls*, mpi_datatypes* );
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/PG-PuReMD/src/cuda/cuda_integrate.cu b/PG-PuReMD/src/cuda/cuda_integrate.cu
new file mode 100644
index 00000000..dcb97292
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_integrate.cu
@@ -0,0 +1,249 @@
+
+#include "cuda_integrate.h"
+
+#include "cuda_allocate.h"
+#include "cuda_forces.h"
+#include "cuda_integrate.h"
+#include "cuda_copy.h"
+#include "cuda_neighbors.h"
+#include "cuda_reset_tools.h"
+#include "cuda_system_props.h"
+#include "cuda_utils.h"
+
+#include "../comm_tools.h"
+#include "../grid.h"
+#include "../vector.h"
+
+
+CUDA_GLOBAL void k_update_velocity_1( reax_atom *my_atoms,
+ single_body_parameters *sbp, real dt, 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 */
+ atom = &(my_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 );
+ /* 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, real dt )
+{
+ int blocks;
+
+ blocks = system->n / DEF_BLOCK_SIZE +
+ ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+ k_update_velocity_1 <<< blocks, DEF_BLOCK_SIZE >>>
+ (system->d_my_atoms, system->reax_param.d_sbp, dt, system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_update_velocity_2( reax_atom *my_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 */
+ atom = &(my_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 )
+{
+ int blocks;
+
+ blocks = system->n / DEF_BLOCK_SIZE +
+ ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+ k_update_velocity_2 <<< blocks, DEF_BLOCK_SIZE >>>
+ (system->d_my_atoms, system->reax_param.d_sbp, dt, system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_scale_velocities( reax_atom *my_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 */
+ atom = &(my_atoms[i]);
+ rvec_Scale( atom->v, lambda, atom->v );
+}
+
+
+void bNVT_scale_velocities( reax_system *system, real lambda )
+{
+ int blocks;
+
+ blocks = system->n / DEF_BLOCK_SIZE +
+ ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+ k_scale_velocities <<< blocks, DEF_BLOCK_SIZE >>>
+ (system->d_my_atoms, lambda, system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+int Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* system, control_params* control,
+ simulation_data *data, storage *workspace, reax_list **lists,
+ output_controls *out_control, mpi_datatypes *mpi_data )
+{
+ int i, steps, renbr, ret;
+ static int verlet_part1_done = FALSE, estimate_nbrs_done = 0;
+ real inv_m, dt, lambda;
+ rvec dx;
+ reax_atom *atom;
+ int *bond_top, *hb_top;
+ int Htop, num_3body;
+ int total_hbonds, count, total_bonds;
+ int bond_cap, cap_3body;
+ real t_over_start, t_over_elapsed;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d\n", system->my_rank, data->step );
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+
+ dt = control->dt;
+ steps = data->step - data->prev_steps;
+ renbr = steps % control->reneighbor == 0 ? TRUE : FALSE;
+ ret = SUCCESS;
+
+ Cuda_ReAllocate( system, control, data, workspace, lists, mpi_data );
+
+ if ( verlet_part1_done == FALSE )
+ {
+ /* velocity verlet, 1st part */
+ bNVT_update_velocity_part1( system, dt );
+ verlet_part1_done = TRUE;
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d: verlet1 done\n", system->my_rank, data->step );
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+
+ if ( renbr )
+ {
+ Update_Grid( system, control, mpi_data->world );
+ }
+
+ Output_Sync_Atoms( system );
+ Comm_Atoms( system, control, data, workspace, lists, mpi_data, renbr );
+ Sync_Atoms( system );
+
+ /* synch the Grid to the Device here */
+ Sync_Grid( &system->my_grid, &system->d_my_grid );
+
+ init_blocks( system );
+
+#if defined(__CUDA_DEBUG_LOG__)
+ fprintf( stderr, "p:%d - Matvec BLocks: %d, blocksize: %d \n",
+ system->my_rank, MATVEC_BLOCKS, MATVEC_BLOCK_SIZE );
+#endif
+ }
+
+ Cuda_Reset( system, control, data, workspace, lists );
+
+ if ( renbr )
+ {
+#if defined(DEBUG)
+ t_over_start = Get_Time ();
+#endif
+
+ if ( estimate_nbrs_done == 0 )
+ {
+ //TODO: move far_nbrs reallocation checks outside of renbr frequency check
+ ret = Cuda_Estimate_Neighbors( system, data->step );
+ estimate_nbrs_done = 1;
+ }
+
+ if ( ret == SUCCESS && estimate_nbrs_done == 1 )
+ {
+ Cuda_Generate_Neighbor_Lists( system, data, workspace, lists );
+ estimate_nbrs_done = 2;
+
+#if defined(DEBUG)
+ t_over_elapsed = Get_Timing_Info( t_over_start );
+ fprintf( stderr, "p%d --> Overhead (Step-%d) %f \n",
+ system->my_rank, data->step, t_over_elapsed );
+#endif
+ }
+ }
+
+ if ( ret == SUCCESS )
+ {
+ ret = Cuda_Compute_Forces( system, control, data, workspace,
+ lists, out_control, mpi_data );
+ }
+
+ if ( ret == SUCCESS )
+ {
+ /* velocity verlet, 2nd part */
+ bNVT_update_velocity_part2( system, dt );
+
+#if defined(DEBUG_FOCUS)
+ fprintf(stderr, "p%d @ step%d: verlet2 done\n", system->my_rank, data->step);
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+
+ /* temperature scaler */
+ Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
+
+ 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 );
+
+ /* Scale velocities and positions at t+dt */
+ bNVT_scale_velocities( system, lambda );
+
+ Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
+
+#if defined(DEBUG_FOCUS)
+ fprintf( stderr, "p%d @ step%d: scaled velocities\n",
+ system->my_rank, data->step );
+ MPI_Barrier( MPI_COMM_WORLD );
+#endif
+
+ verlet_part1_done = FALSE;
+ estimate_nbrs_done = 0;
+ }
+
+ return ret;
+}
diff --git a/PG-PuReMD/src/cuda_integrate.h b/PG-PuReMD/src/cuda/cuda_integrate.h
similarity index 86%
rename from PG-PuReMD/src/cuda_integrate.h
rename to PG-PuReMD/src/cuda/cuda_integrate.h
index b71e14e3..2797b3e3 100644
--- a/PG-PuReMD/src/cuda_integrate.h
+++ b/PG-PuReMD/src/cuda/cuda_integrate.h
@@ -22,18 +22,26 @@
#ifndef __CUDA_INTEGRATE_H_
#define __CUDA_INTEGRATE_H_
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
void bNVT_update_velocity_part1( reax_system *, real );
+
void bNVT_update_velocity_part2( reax_system *, real );
+
void bNVT_scale_velocities( reax_system *, real );
+int Cuda_Velocity_Verlet_Berendsen_NVT( reax_system*, control_params*,
+ simulation_data*, storage*, reax_list**, output_controls*,
+ mpi_datatypes* );
+
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda/cuda_lin_alg.cu b/PG-PuReMD/src/cuda/cuda_lin_alg.cu
new file mode 100644
index 00000000..dc7a2fc3
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_lin_alg.cu
@@ -0,0 +1,1113 @@
+/*----------------------------------------------------------------------
+ PuReMD - Purdue ReaxFF Molecular Dynamics Program
+
+ Copyright (2010) Purdue University
+ Hasan Metin Aktulga, haktulga@cs.purdue.edu
+ Joseph Fogarty, jcfogart@mail.usf.edu
+ Sagar Pandit, pandit@usf.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 "cuda_shuffle.h"
+#include "cuda_utils.h"
+#include "cuda_reduction.h"
+
+#include "../basic_comm.h"
+
+
+//one thread per row
+CUDA_GLOBAL void k_matvec( sparse_matrix H, real *vec, real *results,
+ int rows )
+{
+ int i, col;
+ real results_row;
+ real val;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( i >= rows )
+ {
+ return;
+ }
+
+ results_row = 0;
+
+ 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 >>>
+//CUDA_GLOBAL void __launch_bounds__(384, 16) k_matvec_csr(sparse_matrix H, real *vec, real *results, int num_rows)
+CUDA_GLOBAL void k_matvec_csr( sparse_matrix H, real *vec, real *results,
+ int num_rows )
+{
+#if defined(__SM_35__)
+ real vals;
+ int x;
+#else
+ extern __shared__ real vals[];
+#endif
+ int jj;
+ int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
+ int warp_id = thread_id / MATVEC_KER_THREADS_PER_ROW;
+ int lane = thread_id & ( MATVEC_KER_THREADS_PER_ROW - 1);
+ int row_start;
+ int row_end;
+ // one warp per row
+ int row = warp_id;
+
+#if defined(__SM_35__)
+ vals = 0;
+#else
+ vals[threadIdx.x] = 0;
+#endif
+
+ if (row < num_rows)
+ {
+ row_start = H.start[row];
+ row_end = H.end[row];
+
+ // compute running sum per thread
+ for ( jj = row_start + lane; jj < row_end;
+ jj += MATVEC_KER_THREADS_PER_ROW )
+#if defined(__SM_35__)
+ {
+ vals += H.entries[jj].val * vec[ H.entries[jj].j ];
+ }
+ }
+#else
+ {
+ vals[threadIdx.x] += H.entries[jj].val * vec[ H.entries[jj].j ];
+ }
+ }
+
+ __syncthreads( );
+#endif
+
+ // parallel reduction in shared memory
+ //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
+#if defined(__SM_35__)
+ for (x = MATVEC_KER_THREADS_PER_ROW >> 1; x >= 1; x/=2)
+ {
+ vals += shfl( vals, x );
+ }
+
+ if (lane == 0 && row < num_rows)
+ {
+ results[row] = vals;
+ }
+#else
+ 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];
+ }
+#endif
+}
+
+
+//one thread per row
+CUDA_GLOBAL void k_dual_matvec( sparse_matrix H, rvec2 *vec, rvec2 *results,
+ int rows )
+{
+ int i, c, col;
+ rvec2 results_row;
+ real val;
+
+ i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( i >= rows)
+ {
+ return;
+ }
+
+ results_row[0] = 0.0;
+ results_row[1] = 0.0;
+
+ for (c = H.start[i]; c < H.end[i]; c++)
+ {
+ col = H.entries [c].j;
+ val = H.entries[c].val;
+
+ results_row[0] += val * vec [col][0];
+ results_row[1] += val * vec [col][1];
+ }
+
+ results[i][0] = results_row[0];
+ results[i][1] = results_row[1];
+}
+
+
+//32 thread warp per matrix row.
+//invoked as follows
+// <<< system->N, 32 >>>
+//CUDA_GLOBAL void __launch_bounds__(384, 8) k_dual_matvec_csr(sparse_matrix H, rvec2 *vec, rvec2 *results, int num_rows)
+CUDA_GLOBAL void k_dual_matvec_csr( sparse_matrix H, rvec2 *vec,
+ rvec2 *results, int num_rows )
+{
+#if defined(__SM_35__)
+ rvec2 rvals;
+ int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
+ int warp_id = thread_id / MATVEC_KER_THREADS_PER_ROW;
+ int lane = thread_id & (MATVEC_KER_THREADS_PER_ROW - 1);
+ int row_start;
+ int row_end;
+ // one warp per row
+ int row = warp_id;
+
+ rvals[0] = 0;
+ rvals[1] = 0;
+
+ if (row < num_rows)
+ {
+ row_start = H.start[row];
+ row_end = H.end[row];
+
+ for(int jj = row_start + lane; jj < row_end; jj += MATVEC_KER_THREADS_PER_ROW)
+ {
+ rvals[0] += H.entries[jj].val * vec [ H.entries[jj].j ][0];
+ rvals[1] += H.entries[jj].val * vec [ H.entries[jj].j ][1];
+ }
+ }
+
+ for (int s = MATVEC_KER_THREADS_PER_ROW >> 1; s >= 1; s /= 2)
+ {
+ rvals[0] += shfl( rvals[0], s);
+ rvals[1] += shfl( rvals[1], s);
+ }
+
+ if (lane == 0 && row < num_rows)
+ {
+ results[row][0] = rvals[0];
+ results[row][1] = rvals[1];
+ }
+
+#else
+ extern __shared__ rvec2 rvals[];
+ 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;
+
+ rvals[threadIdx.x][0] = 0;
+ rvals[threadIdx.x][1] = 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)
+ {
+ rvals[threadIdx.x][0] += H.entries[jj].val * vec [ H.entries[jj].j ][0];
+ rvals[threadIdx.x][1] += H.entries[jj].val * vec [ H.entries[jj].j ][1];
+ }
+ }
+
+ __syncthreads( );
+
+ // parallel reduction in shared memory
+ //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
+ if (lane < 16)
+ {
+ rvals[threadIdx.x][0] += rvals[threadIdx.x + 16][0];
+ rvals[threadIdx.x][1] += rvals[threadIdx.x + 16][1];
+ }
+ __syncthreads( );
+ if (lane < 8)
+ {
+ rvals[threadIdx.x][0] += rvals[threadIdx.x + 8][0];
+ rvals[threadIdx.x][1] += rvals[threadIdx.x + 8][1];
+ }
+ __syncthreads( );
+ if (lane < 4)
+ {
+ rvals[threadIdx.x][0] += rvals[threadIdx.x + 4][0];
+ rvals[threadIdx.x][1] += rvals[threadIdx.x + 4][1];
+ }
+ __syncthreads( );
+ if (lane < 2)
+ {
+ rvals[threadIdx.x][0] += rvals[threadIdx.x + 2][0];
+ rvals[threadIdx.x][1] += rvals[threadIdx.x + 2][1];
+ }
+ __syncthreads( );
+ if (lane < 1)
+ {
+ rvals[threadIdx.x][0] += rvals[threadIdx.x + 1][0];
+ rvals[threadIdx.x][1] += rvals[threadIdx.x + 1][1];
+ }
+ __syncthreads( );
+
+ // first thread writes the result
+ if (lane == 0 && row < num_rows)
+ {
+ results[row][0] = rvals[threadIdx.x][0];
+ results[row][1] = rvals[threadIdx.x][1];
+ }
+
+#endif
+}
+
+
+void Cuda_Vector_Sum( real *res, real a, real *x, real b, real *y, int count )
+{
+ //res = ax + by
+ //use the cublas here
+ int blocks;
+
+ blocks = (count / DEF_BLOCK_SIZE) +
+ ((count % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_vector_sum <<< blocks, DEF_BLOCK_SIZE >>>
+ ( res, a, x, b, y, count );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+void Cuda_CG_Preconditioner( real *res, real *a, real *b, int count )
+{
+ //res = a*b - vector multiplication
+ //use the cublas here.
+ int blocks;
+
+ blocks = (count / DEF_BLOCK_SIZE) +
+ ((count % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_vector_mul <<< blocks, DEF_BLOCK_SIZE >>>
+ ( res, a, b, count );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_diagonal_preconditioner(storage p_workspace, rvec2 *b, int n)
+{
+ storage *workspace;
+ int j;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( j >= n )
+ {
+ return;
+ }
+
+ workspace = &( p_workspace );
+
+ //for( j = 0; j < system->n; ++j ) {
+ // residual
+ workspace->r2[j][0] = b[j][0] - workspace->q2[j][0];
+ workspace->r2[j][1] = b[j][1] - workspace->q2[j][1];
+
+ // apply diagonal pre-conditioner
+ workspace->d2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
+ workspace->d2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
+ //}
+}
+
+
+void Cuda_CG_Diagonal_Preconditioner( storage *workspace, rvec2 *b, int n )
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_diagonal_preconditioner <<< blocks, DEF_BLOCK_SIZE >>>
+ (*workspace, b, n);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_dual_cg_preconditioner( storage p_workspace, rvec2 *x,
+ real alpha_0, real alpha_1, int n, rvec2 *my_dot )
+{
+ storage *workspace;
+ rvec2 alpha;
+ int j;
+
+ j = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if ( j >= n )
+ {
+ return;
+ }
+
+ workspace = &( p_workspace );
+ alpha[0] = alpha_0;
+ alpha[1] = alpha_1;
+ my_dot[j][0] = my_dot[j][1] = 0.0;
+
+ //for( j = 0; j < system->n; ++j ) {
+ // update x
+ x[j][0] += alpha[0] * workspace->d2[j][0];
+ x[j][1] += alpha[1] * workspace->d2[j][1];
+
+ // update residual
+ workspace->r2[j][0] -= alpha[0] * workspace->q2[j][0];
+ workspace->r2[j][1] -= alpha[1] * workspace->q2[j][1];
+
+ // apply diagonal pre-conditioner
+ workspace->p2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
+ workspace->p2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
+
+ // dot product: r.p
+ my_dot[j][0] = workspace->r2[j][0] * workspace->p2[j][0];
+ my_dot[j][1] = workspace->r2[j][1] * workspace->p2[j][1];
+ //}
+}
+
+
+void Cuda_DualCG_Preconditioner( storage *workspace, rvec2 *x, rvec2 alpha,
+ int n, rvec2 result )
+{
+ int blocks;
+ rvec2 *tmp = (rvec2 *) scratch;
+
+ cuda_memset( tmp, 0, sizeof(rvec2) * ( 2 * n + 1),
+ "cuda_dualcg_preconditioner" );
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_dual_cg_preconditioner <<< blocks, DEF_BLOCK_SIZE >>>
+ (*workspace, x, alpha[0], alpha[1], n, tmp);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ //Reduction to calculate my_dot
+ k_reduction_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
+ ( tmp, tmp + n, n);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
+ ( tmp + n, tmp + 2*n, blocks);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( result, (tmp + 2*n), sizeof(rvec2),
+ cudaMemcpyDeviceToHost, "my_dot" );
+}
+
+
+void Cuda_Norm( rvec2 *arr, int n, rvec2 result )
+{
+ int blocks;
+ rvec2 *tmp = (rvec2 *) scratch;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_norm_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
+ (arr, tmp, n, INITIAL);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_norm_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
+ (tmp, tmp + BLOCKS_POW_2, blocks, FINAL );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( result, tmp + BLOCKS_POW_2, sizeof(rvec2),
+ cudaMemcpyDeviceToHost, "cuda_norm_rvec2" );
+}
+
+
+void Cuda_Dot( rvec2 *a, rvec2 *b, rvec2 result, int n )
+{
+ int blocks;
+ rvec2 *tmp = (rvec2 *) scratch;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_dot_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
+ ( a, b, tmp, n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_norm_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
+ //k_norm_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * BLOCKS_POW_2 >>>
+ ( tmp, tmp + BLOCKS_POW_2, blocks, FINAL );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( result, tmp + BLOCKS_POW_2, sizeof(rvec2),
+ cudaMemcpyDeviceToHost, "cuda_dot" );
+}
+
+
+void Cuda_Vector_Sum_Rvec2(rvec2 *x, rvec2 *a, rvec2 b, rvec2 *c, int n)
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_rvec2_pbetad <<< blocks, DEF_BLOCK_SIZE >>>
+ ( x, a, b[0], b[1], c, n);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_rvec2_to_real_copy( real *dst, rvec2 *src, int index, int n )
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= n)
+ {
+ return;
+ }
+
+ dst[i] = src[i][index];
+}
+
+
+void Cuda_RvecCopy_From( real *dst, rvec2 *src, int index, int n )
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_rvec2_to_real_copy <<< blocks, DEF_BLOCK_SIZE >>>
+ ( dst, src, index, n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+CUDA_GLOBAL void k_real_to_rvec2_copy( rvec2 *dst, real *src, int index, int n)
+{
+ int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ if (i >= n)
+ {
+ return;
+ }
+
+ dst[i][index] = src[i];
+}
+
+
+void Cuda_RvecCopy_To(rvec2 *dst, real *src, int index, int n)
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
+
+ k_real_to_rvec2_copy <<< blocks, DEF_BLOCK_SIZE >>>
+ ( dst, src, index, n);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+void Cuda_Dual_Matvec( sparse_matrix *H, rvec2 *a, rvec2 *b, int n, int size )
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE) == 0 ? 0 : 1);
+
+ cuda_memset( b, 0, sizeof(rvec2) * size, "dual_matvec:result" );
+
+ //One thread per row implementation
+ //k_dual_matvec <<< blocks, DEF_BLOCK_SIZE >>>
+ // (*H, a, b, n);
+ //cudaThreadSynchronize ();
+ //cudaCheckError ();
+
+ //One warp per row implementation
+#if defined(__SM_35__)
+ k_dual_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE >>>
+#else
+ k_dual_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE,
+ sizeof(rvec2) * MATVEC_BLOCK_SIZE >>>
+#endif
+ ( *H, a, b, n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+void Cuda_Matvec( sparse_matrix *H, real *a, real *b, int n, int size )
+{
+ int blocks;
+
+ blocks = (n / DEF_BLOCK_SIZE) +
+ (( n % DEF_BLOCK_SIZE) == 0 ? 0 : 1);
+
+ cuda_memset( b, 0, sizeof(real) * size, "dual_matvec:result" );
+
+ //one thread per row implementation
+ //k_matvec <<< blocks, DEF_BLOCK_SIZE >>>
+ // (*H, a, b, n);
+ //cudaThreadSynchronize ();
+ //cudaCheckError ();
+
+#if defined(__SM_35__)
+ k_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE >>>
+#else
+ k_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE,
+ sizeof(real) * MATVEC_BLOCK_SIZE>>>
+#endif
+ (*H, a, b, n);
+
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+}
+
+
+int Cuda_dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
+ rvec2 *b, real tol, rvec2 *x, mpi_datatypes* mpi_data, FILE *fout,
+ simulation_data *data )
+{
+ int i, j, n, N, matvecs, scale;
+ rvec2 tmp, alpha, beta;
+ rvec2 my_sum, norm_sqr, b_norm, my_dot;
+ rvec2 sig_old, sig_new;
+ MPI_Comm comm;
+ rvec2 *spad = (rvec2 *) host_scratch;
+ int a;
+
+ n = system->n;
+ N = system->N;
+ comm = mpi_data->world;
+ matvecs = 0;
+ scale = sizeof(rvec2) / sizeof(void);
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ matvecs = 0;
+ t_start = matvec_time = dot_time = 0;
+ t_start = Get_Time( );
+ }
+#endif
+
+ //MVAPICH2
+//#ifdef __CUDA_DEBUG__
+// Dist( system, mpi_data, workspace->x, mpi_data->mpi_rvec2, scale, rvec2_packer );
+//#endif
+
+// check_zeros_device( x, system->N, "x" );
+
+ copy_host_device( spad, x, sizeof(rvec2) * system->total_cap, cudaMemcpyDeviceToHost, "CG:x:get" );
+ Dist( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_packer );
+ copy_host_device( spad, x, sizeof(rvec2) * system->total_cap, cudaMemcpyHostToDevice, "CG:x:put" );
+
+// check_zeros_device( x, system->N, "x" );
+
+// compare_rvec2 (workspace->x, x, N, "x");
+// if (data->step > 0) {
+// compare_rvec2 (workspace->b, dev_workspace->b, system->N, "b");
+// compare_rvec2 (workspace->x, dev_workspace->x, system->N, "x");
+//
+// exit (0);
+// }
+
+
+//#ifdef __CUDA_DEBUG__
+// dual_Sparse_MatVec( &workspace->H, workspace->x, workspace->q2, N );
+//#endif
+ //originally we were using only H->n which was system->n (init_md.c)
+ //Cuda_Dual_Matvec ( H, x, dev_workspace->q2, H->n, system->total_cap);
+
+ Cuda_Dual_Matvec ( H, x, dev_workspace->q2, system->N, system->total_cap);
+
+// compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
+
+// if (data->step > 0) exit (0);
+
+ // tryQEq
+ //MVAPICH2
+//#ifdef __CUDA_DEBUG__
+// Coll(system,mpi_data,workspace->q2,mpi_data->mpi_rvec2,scale,rvec2_unpacker);
+//#endif
+
+ copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyDeviceToHost, "CG:q2:get" );
+ Coll(system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_unpacker);
+ copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyHostToDevice,"CG:q2:put" );
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &matvec_time );
+ }
+#endif
+
+//#ifdef __CUDA_DEBUG__
+// for( j = 0; j < system->n; ++j ) {
+// // residual
+// workspace->r2[j][0] = workspace->b[j][0] - workspace->q2[j][0];
+// workspace->r2[j][1] = workspace->b[j][1] - workspace->q2[j][1];
+// // apply diagonal pre-conditioner
+// workspace->d2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
+// workspace->d2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
+// }
+//#endif
+
+ Cuda_CG_Diagonal_Preconditioner( dev_workspace, b, system->n );
+
+// compare_rvec2 (workspace->r2, dev_workspace->r2, n, "r2");
+// compare_rvec2 (workspace->d2, dev_workspace->d2, n, "d2");
+
+ /* norm of b */
+//#ifdef __CUDA_DEBUG__
+// my_sum[0] = my_sum[1] = 0;
+// for( j = 0; j < n; ++j ) {
+// my_sum[0] += SQR( workspace->b[j][0] );
+// my_sum[1] += SQR( workspace->b[j][1] );
+// }
+// fprintf (stderr, "cg: my_sum[ %f, %f] \n", my_sum[0], my_sum[1]);
+//#endif
+
+ my_sum[0] = my_sum[1] = 0;
+ Cuda_Norm (b, n, my_sum);
+
+// fprintf (stderr, "cg: my_sum[ %f, %f] \n", my_sum[0], my_sum[1]);
+
+ MPI_Allreduce( &my_sum, &norm_sqr, 2, MPI_DOUBLE, MPI_SUM, comm );
+ b_norm[0] = SQRT( norm_sqr[0] );
+ b_norm[1] = SQRT( norm_sqr[1] );
+ //fprintf( stderr, "bnorm = %f %f\n", b_norm[0], b_norm[1] );
+
+ /* dot product: r.d */
+//#ifdef __CUDA_DEBUG__
+// my_dot[0] = my_dot[1] = 0;
+// for( j = 0; j < n; ++j ) {
+// my_dot[0] += workspace->r2[j][0] * workspace->d2[j][0];
+// my_dot[1] += workspace->r2[j][1] * workspace->d2[j][1];
+// }
+// fprintf( stderr, "my_dot: %f %f\n", my_dot[0], my_dot[1] );
+//#endif
+
+ my_dot[0] = my_dot[1] = 0;
+ Cuda_Dot (dev_workspace->r2, dev_workspace->d2, my_dot, n);
+
+// fprintf( stderr, "my_dot: %f %f\n", my_dot[0], my_dot[1] );
+
+ MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
+
+ //fprintf( stderr, "DEVICE: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
+
+ for ( i = 1; i < 300; ++i )
+ {
+ //MVAPICH2
+//#ifdef __CUDA_DEBUG__
+// Dist(system,mpi_data,workspace->d2,mpi_data->mpi_rvec2,scale,rvec2_packer);
+//#endif
+
+ copy_host_device( spad, dev_workspace->d2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cg:d2:get" );
+ Dist( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_packer );
+ copy_host_device( spad, dev_workspace->d2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyHostToDevice, "cg:d2:put" );
+
+ //print_device_rvec2 (dev_workspace->d2, N);
+
+//#ifdef __CUDA_DEBUG__
+// dual_Sparse_MatVec( &workspace->H, workspace->d2, workspace->q2, N );
+//#endif
+
+ Cuda_Dual_Matvec( H, dev_workspace->d2, dev_workspace->q2, system->N,
+ system->total_cap );
+
+ /*
+ fprintf (stderr, "******************* Device sparse Matrix--------> %d \n", H->n );
+ fprintf (stderr, " ******* HOST SPARSE MATRIX ******** \n");
+ print_sparse_matrix_host (&workspace->H);
+ fprintf (stderr, " ******* HOST Vector ***************\n");
+ print_host_rvec2 (workspace->d2, system->N);
+ fprintf (stderr, " ******* Device SPARSE MATRIX ******** \n");
+ print_sparse_matrix (&dev_workspace->H);
+ fprintf (stderr, " ******* Device Vector ***************\n");
+ print_device_rvec2 (dev_workspace->d2, system->N);
+ */
+ //compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
+
+ // tryQEq
+ // MVAPICH2
+//#ifdef __CUDA_DEBUG__
+// Coll(system,mpi_data,workspace->q2,mpi_data->mpi_rvec2,scale,rvec2_unpacker);
+//#endif
+
+ copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cg:q2:get" );
+ Coll( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_unpacker );
+ copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
+ cudaMemcpyHostToDevice, "cg:q2:put" );
+
+// compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &matvec_time );
+ }
+#endif
+
+ /* dot product: d.q */
+//#ifdef __CUDA_DEBUG__
+// my_dot[0] = my_dot[1] = 0;
+// for( j = 0; j < n; ++j ) {
+// my_dot[0] += workspace->d2[j][0] * workspace->q2[j][0];
+// my_dot[1] += workspace->d2[j][1] * workspace->q2[j][1];
+// }
+// fprintf( stderr, "H:my_dot: %f %f\n", my_dot[0], my_dot[1] );
+//#endif
+
+ my_dot[0] = my_dot[1] = 0;
+ Cuda_Dot (dev_workspace->d2, dev_workspace->q2, my_dot, n);
+ //fprintf( stderr, "D:my_dot: %f %f\n", my_dot[0], my_dot[1] );
+
+ MPI_Allreduce( &my_dot, &tmp, 2, MPI_DOUBLE, MPI_SUM, comm );
+ //fprintf( stderr, "tmp: %f %f\n", tmp[0], tmp[1] );
+
+ alpha[0] = sig_new[0] / tmp[0];
+ alpha[1] = sig_new[1] / tmp[1];
+ my_dot[0] = my_dot[1] = 0;
+
+//#ifdef __CUDA_DEBUG__
+// for( j = 0; j < system->n; ++j ) {
+// // update x
+// workspace->x[j][0] += alpha[0] * workspace->d2[j][0];
+// workspace->x[j][1] += alpha[1] * workspace->d2[j][1];
+// // update residual
+// workspace->r2[j][0] -= alpha[0] * workspace->q2[j][0];
+// workspace->r2[j][1] -= alpha[1] * workspace->q2[j][1];
+// // apply diagonal pre-conditioner
+// workspace->p2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
+// workspace->p2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
+// // dot product: r.p
+// my_dot[0] += workspace->r2[j][0] * workspace->p2[j][0];
+// my_dot[1] += workspace->r2[j][1] * workspace->p2[j][1];
+// }
+// fprintf( stderr, "H:my_dot: %f %f\n", my_dot[0], my_dot[1] );
+//#endif
+
+ my_dot[0] = my_dot[1] = 0;
+ Cuda_DualCG_Preconditioner( dev_workspace, x, alpha, system->n, my_dot );
+
+ //fprintf( stderr, "D:my_dot: %f %f\n", my_dot[0], my_dot[1] );
+
+// compare_rvec2 (workspace->x, dev_workspace->x, N, "x");
+// compare_rvec2 (workspace->r2, dev_workspace->r2, N, "r2");
+// compare_rvec2 (workspace->p2, dev_workspace->p2, N, "p2");
+
+ sig_old[0] = sig_new[0];
+ sig_old[1] = sig_new[1];
+ MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
+
+ //fprintf( stderr, "DEVICE: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];
+
+//#ifdef __CUDA_DEBUG__
+// for( j = 0; j < system->n; ++j ) {
+// // d = p + beta * d
+// workspace->d2[j][0] = workspace->p2[j][0] + beta[0] * workspace->d2[j][0];
+// workspace->d2[j][1] = workspace->p2[j][1] + beta[1] * workspace->d2[j][1];
+// }
+//#endif
+
+ Cuda_Vector_Sum_Rvec2( dev_workspace->d2, dev_workspace->p2, beta,
+ dev_workspace->d2, system->n );
+
+// compare_rvec2 (workspace->d2, dev_workspace->d2, N, "q2");
+ }
+
+
+ if ( SQRT(sig_new[0]) / b_norm[0] <= tol )
+ {
+ //for( j = 0; j < n; ++j )
+ // workspace->t[j] = workspace->x[j][1];
+ //fprintf (stderr, "Getting started with Cuda_CG1 \n");
+
+ Cuda_RvecCopy_From( dev_workspace->t, dev_workspace->x, 1, system->n );
+
+ //compare_array (workspace->b_t, dev_workspace->b_t, system->n, "b_t");
+ //compare_array (workspace->t, dev_workspace->t, system->n, "t");
+
+ matvecs = Cuda_CG( system, workspace, H, dev_workspace->b_t, tol, dev_workspace->t,
+ mpi_data, fout );
+
+ //fprintf (stderr, " Cuda_CG1: iterations --> %d \n", matvecs );
+ //for( j = 0; j < n; ++j )
+ // workspace->x[j][1] = workspace->t[j];
+
+ Cuda_RvecCopy_To( dev_workspace->x, dev_workspace->t, 1, system->n );
+ }
+ else if ( SQRT(sig_new[1]) / b_norm[1] <= tol )
+ {
+ //for( j = 0; j < n; ++j )
+ // workspace->s[j] = workspace->x[j][0];
+
+ Cuda_RvecCopy_From( dev_workspace->s, dev_workspace->x, 0, system->n );
+
+ //compare_array (workspace->s, dev_workspace->s, system->n, "s");
+ //compare_array (workspace->b_s, dev_workspace->b_s, system->n, "b_s");
+
+ //fprintf (stderr, "Getting started with Cuda_CG2 \n");
+
+ matvecs = Cuda_CG( system, workspace, H, dev_workspace->b_s, tol, dev_workspace->s,
+ mpi_data, fout );
+
+ //fprintf (stderr, " Cuda_CG2: iterations --> %d \n", matvecs );
+ //for( j = 0; j < system->n; ++j )
+ // workspace->x[j][0] = workspace->s[j];
+
+ Cuda_RvecCopy_To( dev_workspace->x, dev_workspace->s, 0, system->n );
+ }
+
+ if ( i >= 300 )
+ {
+ fprintf( stderr, "[WARNING] p%d: dual CG convergence failed! (%d steps)\n",
+ system->my_rank, i );
+ fprintf( stderr, " [INFO] s lin solve error: %f\n", SQRT(sig_new[0]) / b_norm[0] );
+ fprintf( stderr, " [INFO] t lin solve error: %f\n", SQRT(sig_new[1]) / b_norm[1] );
+ }
+
+#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 );
+ }
+#endif
+
+ return (i + 1) + matvecs;
+}
+
+
+int Cuda_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;
+ real sig_old, sig_new, sig0;
+ real *spad = (real *) host_scratch;
+
+ scale = sizeof(real) / sizeof(void);
+
+ /* x is on the device */
+ //MVAPICH2
+ memset( spad, 0, sizeof(real) * system->total_cap );
+ copy_host_device( spad, x, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:x:get" );
+ Dist( system, mpi_data, spad, MPI_DOUBLE, scale, real_packer );
+
+ //MVAPICH2
+ copy_host_device( spad, x, sizeof(real) * system->total_cap,
+ cudaMemcpyHostToDevice, "cuda_cg:x:put" );
+ Cuda_Matvec( H, x, dev_workspace->q, system->N, system->total_cap );
+
+ // tryQEq
+ // MVAPICH2
+ copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
+ Coll( system, mpi_data, spad, MPI_DOUBLE, scale, real_unpacker );
+
+ //MVAPICH2
+ copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
+ cudaMemcpyHostToDevice, "cuda_cg:q:put" );
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &matvec_time );
+ }
+#endif
+
+ Cuda_Vector_Sum( dev_workspace->r , 1., b, -1., dev_workspace->q,
+ system->n );
+ //for( j = 0; j < system->n; ++j )
+ // workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j]; //pre-condition
+ Cuda_CG_Preconditioner( dev_workspace->d, dev_workspace->r,
+ dev_workspace->Hdia_inv, system->n );
+
+ //TODO do the parallel_norm on the device for the local sum
+ copy_host_device( spad, b, sizeof(real) * system->n,
+ cudaMemcpyDeviceToHost, "cuda_cg:b:get" );
+ b_norm = Parallel_Norm( spad, system->n, mpi_data->world );
+
+ //TODO do the parallel dot on the device for the local sum
+ copy_host_device( spad, dev_workspace->r, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:r:get" );
+ copy_host_device( spad + system->total_cap, dev_workspace->d, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
+ sig_new = Parallel_Dot( spad, spad + system->total_cap, 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 )
+ {
+ //MVAPICH2
+ copy_host_device( spad, dev_workspace->d, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
+ Dist( system, mpi_data, spad, MPI_DOUBLE, scale, real_packer );
+ copy_host_device( spad, dev_workspace->d, sizeof(real) * system->total_cap,
+ cudaMemcpyHostToDevice, "cuda_cg:d:put" );
+
+ Cuda_Matvec( H, dev_workspace->d, dev_workspace->q, system->N, system->total_cap );
+
+ //tryQEq
+ copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
+ cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
+ Coll( system, mpi_data, spad, MPI_DOUBLE, scale, real_unpacker );
+ copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
+ cudaMemcpyHostToDevice, "cuda_cg:q:get" );
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &matvec_time );
+ }
+#endif
+
+ //TODO do the parallel dot on the device for the local sum
+ copy_host_device( spad, dev_workspace->d, sizeof(real) * system->n,
+ cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
+ copy_host_device( spad + system->n, dev_workspace->q, sizeof(real) * system->n,
+ cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
+ tmp = Parallel_Dot( spad, spad + system->n, system->n, mpi_data->world );
+
+ alpha = sig_new / tmp;
+ //Cuda_Vector_Add( x, alpha, dev_workspace->d, system->n );
+ Cuda_Vector_Sum( x, alpha, dev_workspace->d, 1.0, x, system->n );
+
+ //Cuda_Vector_Add( workspace->r, -alpha, workspace->q, system->n );
+ Cuda_Vector_Sum( dev_workspace->r, -alpha, dev_workspace->q, 1.0,
+ dev_workspace->r, system->n );
+ /* pre-conditioning */
+ //for( j = 0; j < system->n; ++j )
+ // workspace->p[j] = workspace->r[j] * workspace->Hdia_inv[j];
+ Cuda_CG_Preconditioner( dev_workspace->p, dev_workspace->r,
+ dev_workspace->Hdia_inv, system->n );
+
+ sig_old = sig_new;
+
+ //TODO do the parallel dot on the device for the local sum
+ copy_host_device( spad, dev_workspace->r, sizeof(real) * system->n,
+ cudaMemcpyDeviceToHost, "cuda_cg:r:get" );
+ copy_host_device( spad + system->n, dev_workspace->p, sizeof(real) * system->n,
+ cudaMemcpyDeviceToHost, "cuda_cg:p:get" );
+ sig_new = Parallel_Dot( spad , spad + system->n, system->n, mpi_data->world );
+ //fprintf (stderr, "Device: sig_new: %f \n", sig_new );
+
+ beta = sig_new / sig_old;
+ Cuda_Vector_Sum( dev_workspace->d, 1., dev_workspace->p, beta,
+ dev_workspace->d, system->n );
+
+#if defined(CG_PERFORMANCE)
+ if ( system->my_rank == MASTER_NODE )
+ {
+ Update_Timing_Info( &t_start, &dot_time );
+ }
+#endif
+ }
+
+ if ( i >= 300 )
+ {
+ fprintf( stderr, "CG convergence failed!\n" );
+ return i;
+ }
+
+ return i;
+}
diff --git a/PG-PuReMD/src/cuda_lin_alg.h b/PG-PuReMD/src/cuda/cuda_lin_alg.h
similarity index 52%
rename from PG-PuReMD/src/cuda_lin_alg.h
rename to PG-PuReMD/src/cuda/cuda_lin_alg.h
index a7e3cc5f..aa31c126 100644
--- a/PG-PuReMD/src/cuda_lin_alg.h
+++ b/PG-PuReMD/src/cuda/cuda_lin_alg.h
@@ -22,29 +22,44 @@
#ifndef __CUDA_LIN_ALG_H_
#define __CUDA_LIN_ALG_H_
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
extern "C" {
#endif
+void Cuda_Vector_Sum( real *, real, real *, real, real *, int );
-void Cuda_Vector_Sum(real *res, real a, real *x, real b, real *y, int count);
-void Cuda_CG_Preconditioner(real *res, real *a, real *b, int count);
-void Cuda_CG_Diagonal_Preconditioner(storage *workspace, rvec2 *b, int n);
-void Cuda_DualCG_Preconditioner(storage *workspace, rvec2 *, rvec2 alpha, int n, rvec2 result);
-void Cuda_Norm(rvec2 *arr, int n, rvec2 result);
-void Cuda_Dot(rvec2 *a, rvec2 *b, rvec2 result, int n);
-void Cuda_Vector_Sum_Rvec2(rvec2 *x, rvec2 *, rvec2 , rvec2 *c, int n);
-void Cuda_RvecCopy_From(real *dst, rvec2 *src, int index, int n);
-void Cuda_RvecCopy_To(rvec2 *dst, real *src, int index, int n);
-void Cuda_Dual_Matvec(sparse_matrix *, rvec2 *, rvec2 *, int , int);
-void Cuda_Matvec(sparse_matrix *, real *, real *, int , int);
+void Cuda_CG_Preconditioner( real *, real *, real *, int );
+void Cuda_CG_Diagonal_Preconditioner( storage *, rvec2 *, int );
+
+void Cuda_DualCG_Preconditioner( storage *, rvec2 *, rvec2, int, rvec2 );
+
+void Cuda_Norm( rvec2 *, int, rvec2 );
+
+void Cuda_Dot( rvec2 *, rvec2 *, rvec2, int );
+
+void Cuda_Vector_Sum_Rvec2( rvec2 *, rvec2 *, rvec2, rvec2 *, int );
+
+void Cuda_RvecCopy_From( real *, rvec2 *, int, int );
+
+void Cuda_RvecCopy_To( rvec2 *, real *, int, int );
+
+void Cuda_Dual_Matvec( sparse_matrix *, rvec2 *, rvec2 *, int , int );
+
+void Cuda_Matvec( sparse_matrix *, real *, real *, int , int );
+
+int Cuda_dual_CG( reax_system*, storage*, sparse_matrix*,
+ rvec2*, real, rvec2*, mpi_datatypes*, FILE* , simulation_data * );
+
+int Cuda_CG( reax_system*, storage*, sparse_matrix*,
+ real*, real, real*, mpi_datatypes*, FILE* );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_list.cu b/PG-PuReMD/src/cuda/cuda_list.cu
similarity index 96%
rename from PG-PuReMD/src/cuda_list.cu
rename to PG-PuReMD/src/cuda/cuda_list.cu
index 21d8d091..9d0626f1 100644
--- a/PG-PuReMD/src/cuda_list.cu
+++ b/PG-PuReMD/src/cuda/cuda_list.cu
@@ -19,15 +19,14 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reax_types.h"
#include "cuda_utils.h"
#if defined(PURE_REAX)
- #include "list.h"
- #include "tool_box.h"
+ #include "../list.h"
+ #include "../tool_box.h"
#elif defined(LAMMPS_REAX)
- #include "reax_list.h"
- #include "reax_tool_box.h"
+ #include "../reax_list.h"
+ #include "../reax_tool_box.h"
#endif
diff --git a/PG-PuReMD/src/cuda_list.h b/PG-PuReMD/src/cuda/cuda_list.h
similarity index 98%
rename from PG-PuReMD/src/cuda_list.h
rename to PG-PuReMD/src/cuda/cuda_list.h
index 0b4e7aa0..fe06f4ce 100644
--- a/PG-PuReMD/src/cuda_list.h
+++ b/PG-PuReMD/src/cuda/cuda_list.h
@@ -22,13 +22,15 @@
#ifndef __CUDA_LIST_H_
#define __CUDA_LIST_H_
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
void Dev_Make_List( int, int, int, reax_list* );
+
void Dev_Delete_List( reax_list* );
#ifdef __cplusplus
diff --git a/PG-PuReMD/src/cuda_lookup.cu b/PG-PuReMD/src/cuda/cuda_lookup.cu
similarity index 98%
rename from PG-PuReMD/src/cuda_lookup.cu
rename to PG-PuReMD/src/cuda/cuda_lookup.cu
index 837a3c71..01bc8a79 100644
--- a/PG-PuReMD/src/cuda_lookup.cu
+++ b/PG-PuReMD/src/cuda/cuda_lookup.cu
@@ -1,8 +1,9 @@
#include "cuda_lookup.h"
-#include "index_utils.h"
+
#include "cuda_utils.h"
-#include "reax_types.h"
+
+#include "../index_utils.h"
void copy_LR_table_to_device( reax_system *system, control_params *control,
diff --git a/PG-PuReMD/src/cuda_lookup.h b/PG-PuReMD/src/cuda/cuda_lookup.h
similarity index 56%
rename from PG-PuReMD/src/cuda_lookup.h
rename to PG-PuReMD/src/cuda/cuda_lookup.h
index 88f5cfce..87026f7d 100644
--- a/PG-PuReMD/src/cuda_lookup.h
+++ b/PG-PuReMD/src/cuda/cuda_lookup.h
@@ -2,16 +2,18 @@
#ifndef __CUDA_LOOKUP_H__
#define __CUDA_LOOKUP_H__
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-void copy_LR_table_to_device (reax_system *, control_params *, int *);
+void copy_LR_table_to_device( reax_system *, control_params *, int * );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_multi_body.cu b/PG-PuReMD/src/cuda/cuda_multi_body.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_multi_body.cu
rename to PG-PuReMD/src/cuda/cuda_multi_body.cu
index 09a12963..cb741571 100644
--- a/PG-PuReMD/src/cuda_multi_body.cu
+++ b/PG-PuReMD/src/cuda/cuda_multi_body.cu
@@ -19,12 +19,13 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reax_types.h"
#include "cuda_multi_body.h"
-#include "index_utils.h"
+
#include "cuda_helpers.h"
#include "cuda_list.h"
+#include "../index_utils.h"
+
CUDA_GLOBAL void Cuda_Atom_Energy( reax_atom *my_atoms, global_parameters gp,
single_body_parameters *sbp, two_body_parameters *tbp,
diff --git a/PG-PuReMD/src/cuda_multi_body.h b/PG-PuReMD/src/cuda/cuda_multi_body.h
similarity index 58%
rename from PG-PuReMD/src/cuda_multi_body.h
rename to PG-PuReMD/src/cuda/cuda_multi_body.h
index 332e6f06..06014b3a 100644
--- a/PG-PuReMD/src/cuda_multi_body.h
+++ b/PG-PuReMD/src/cuda/cuda_multi_body.h
@@ -22,21 +22,14 @@
#ifndef __CUDA_MULTI_BODY_H_
#define __CUDA_MULTI_BODY_H_
-#include "reax_types.h"
-
-CUDA_GLOBAL void Cuda_Atom_Energy( reax_atom *,
- global_parameters ,
- single_body_parameters *,
- two_body_parameters *,
- storage ,
- reax_list ,
- int ,
- int ,
- real *,
- real *,
- real *
- );
-
-CUDA_GLOBAL void Cuda_Atom_Energy_PostProcess (reax_list, storage, int );
+#include "../reax_types.h"
+
+
+CUDA_GLOBAL void Cuda_Atom_Energy( reax_atom *, global_parameters,
+ single_body_parameters *, two_body_parameters *, storage,
+ reax_list, int, int, real *, real *, real *);
+
+CUDA_GLOBAL void Cuda_Atom_Energy_PostProcess( reax_list, storage, int );
+
#endif
diff --git a/PG-PuReMD/src/cuda_neighbors.cu b/PG-PuReMD/src/cuda/cuda_neighbors.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_neighbors.cu
rename to PG-PuReMD/src/cuda/cuda_neighbors.cu
index f9a20ebd..b1f2b85d 100644
--- a/PG-PuReMD/src/cuda_neighbors.cu
+++ b/PG-PuReMD/src/cuda/cuda_neighbors.cu
@@ -21,15 +21,13 @@
#include "cuda_neighbors.h"
-#include "reax_types.h"
-
#include "cuda_list.h"
#include "cuda_utils.h"
#include "cuda_reduction.h"
-#include "vector.h"
-#include "index_utils.h"
-#include "tool_box.h"
+#include "../index_utils.h"
+#include "../tool_box.h"
+#include "../vector.h"
CUDA_DEVICE real Dev_DistSqr_to_Special_Point( rvec cp, rvec x )
diff --git a/PG-PuReMD/src/cuda_neighbors.h b/PG-PuReMD/src/cuda/cuda_neighbors.h
similarity index 95%
rename from PG-PuReMD/src/cuda_neighbors.h
rename to PG-PuReMD/src/cuda/cuda_neighbors.h
index f7d7cb15..4d4a9c4e 100644
--- a/PG-PuReMD/src/cuda_neighbors.h
+++ b/PG-PuReMD/src/cuda/cuda_neighbors.h
@@ -2,14 +2,13 @@
#ifndef __CUDA_NEIGHBORS_H__
#define __CUDA_NEIGHBORS_H__
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
extern "C" {
#endif
-
void Cuda_Generate_Neighbor_Lists( reax_system *, simulation_data *, storage *, reax_list ** );
int Cuda_Estimate_Neighbors( reax_system *, int );
@@ -24,9 +23,9 @@ void Cuda_Init_Sparse_Matrix_Indices( reax_system *, sparse_matrix * );
void Cuda_Init_Three_Body_Indices( int *, int );
-
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_nonbonded.cu b/PG-PuReMD/src/cuda/cuda_nonbonded.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_nonbonded.cu
rename to PG-PuReMD/src/cuda/cuda_nonbonded.cu
index 93bca2da..25c0b17d 100644
--- a/PG-PuReMD/src/cuda_nonbonded.cu
+++ b/PG-PuReMD/src/cuda/cuda_nonbonded.cu
@@ -25,10 +25,9 @@
#include "cuda_utils.h"
#include "cuda_reduction.h"
#include "cuda_shuffle.h"
-#include "vector.h"
-#include "reax_types.h"
-#include "index_utils.h"
+#include "../index_utils.h"
+#include "../vector.h"
//CUDA_GLOBAL void __launch_bounds__ (960) ker_vdW_coulomb_energy(
diff --git a/PG-PuReMD/src/cuda_nonbonded.h b/PG-PuReMD/src/cuda/cuda_nonbonded.h
similarity index 79%
rename from PG-PuReMD/src/cuda_nonbonded.h
rename to PG-PuReMD/src/cuda/cuda_nonbonded.h
index 1c9916bf..238d49d7 100644
--- a/PG-PuReMD/src/cuda_nonbonded.h
+++ b/PG-PuReMD/src/cuda/cuda_nonbonded.h
@@ -19,15 +19,17 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __NONBONDED_H_
-#define __NONBONDED_H_
+#ifndef __CUDA_NONBONDED_H_
+#define __CUDA_NONBONDED_H_
-#include "reax_types.h"
+#include "../reax_types.h"
void Cuda_Compute_Polarization_Energy( reax_system *, simulation_data *);
-void Cuda_NonBonded_Energy ( reax_system *, control_params *,
- storage *, simulation_data *, reax_list **,
- output_controls *, bool );
+
+void Cuda_NonBonded_Energy( reax_system *, control_params *,
+ storage *, simulation_data *, reax_list **,
+ output_controls *, bool );
+
#endif
diff --git a/PG-PuReMD/src/cuda_post_evolve.cu b/PG-PuReMD/src/cuda/cuda_post_evolve.cu
similarity index 95%
rename from PG-PuReMD/src/cuda_post_evolve.cu
rename to PG-PuReMD/src/cuda/cuda_post_evolve.cu
index 9a478192..828a0e4b 100644
--- a/PG-PuReMD/src/cuda_post_evolve.cu
+++ b/PG-PuReMD/src/cuda/cuda_post_evolve.cu
@@ -1,9 +1,10 @@
#include "cuda_post_evolve.h"
-#include "reax_types.h"
-#include "vector.h"
+
#include "cuda_utils.h"
+#include "../vector.h"
+
CUDA_GLOBAL void ker_post_evolve( reax_atom *my_atoms,
simulation_data *data, int n )
diff --git a/PG-PuReMD/src/cuda_post_evolve.h b/PG-PuReMD/src/cuda/cuda_post_evolve.h
similarity index 60%
rename from PG-PuReMD/src/cuda_post_evolve.h
rename to PG-PuReMD/src/cuda/cuda_post_evolve.h
index dcdcd50c..a1a0571a 100644
--- a/PG-PuReMD/src/cuda_post_evolve.h
+++ b/PG-PuReMD/src/cuda/cuda_post_evolve.h
@@ -2,16 +2,18 @@
#ifndef __CUDA_POST_EVOLVE_H__
#define __CUDA_POST_EVOLVE_H__
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-void post_evolve_velocities (reax_system *, simulation_data *);
+void post_evolve_velocities( reax_system *, simulation_data * );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_reduction.cu b/PG-PuReMD/src/cuda/cuda_reduction.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_reduction.cu
rename to PG-PuReMD/src/cuda/cuda_reduction.cu
index 02d800ee..01bd3c81 100644
--- a/PG-PuReMD/src/cuda_reduction.cu
+++ b/PG-PuReMD/src/cuda/cuda_reduction.cu
@@ -4,10 +4,10 @@
#include "cuda_shuffle.h"
#include "cuda_utils.h"
-#include "vector.h"
+#include "../vector.h"
-#include "cub/cub/device/device_reduce.cuh"
-#include "cub/cub/device/device_scan.cuh"
+#include "../cub/cub/device/device_reduce.cuh"
+#include "../cub/cub/device/device_scan.cuh"
//struct RvecSum
diff --git a/PG-PuReMD/src/cuda_reduction.h b/PG-PuReMD/src/cuda/cuda_reduction.h
similarity index 96%
rename from PG-PuReMD/src/cuda_reduction.h
rename to PG-PuReMD/src/cuda/cuda_reduction.h
index 15ca538f..cf9efc5d 100644
--- a/PG-PuReMD/src/cuda_reduction.h
+++ b/PG-PuReMD/src/cuda/cuda_reduction.h
@@ -2,32 +2,45 @@
#ifndef __CUDA_REDUCTION_H__
#define __CUDA_REDUCTION_H__
-#include "reax_types.h"
+#include "../reax_types.h"
#define INITIAL 0
#define FINAL 1
void Cuda_Reduction_Sum( int *, int *, size_t );
+
void Cuda_Reduction_Sum( real *, real *, size_t );
+
//void Cuda_Reduction_Sum( rvec *, rvec *, size_t );
+
void Cuda_Reduction_Max( int *, int *, size_t );
+
void Cuda_Scan_Excl_Sum( int *, int *, size_t );
CUDA_GLOBAL void k_reduction( const real *, real *, const size_t );
+
CUDA_GLOBAL void k_reduction_rvec( rvec *, rvec *, size_t );
+
CUDA_GLOBAL void k_reduction_rvec2( rvec2 *, rvec2 *, size_t );
+
CUDA_GLOBAL void k_norm( const real *, real *, const size_t, int );
+
CUDA_GLOBAL void k_dot( const real *, const real *, real *,
const size_t );
CUDA_GLOBAL void k_vector_sum( real*, real, real*, real,
real*, int );
+
CUDA_GLOBAL void k_rvec2_pbetad( rvec2 *, rvec2 *, real, real,
rvec2 *, int );
+
CUDA_GLOBAL void k_rvec2_mul( rvec2*, rvec2*, rvec2*, int );
+
CUDA_GLOBAL void k_vector_mul( real*, real*, real*, int );
+
CUDA_GLOBAL void k_norm_rvec2( const rvec2 *, rvec2 *, const size_t, int );
+
CUDA_GLOBAL void k_dot_rvec2( const rvec2 *, rvec2 *, rvec2 *, const size_t );
diff --git a/PG-PuReMD/src/cuda_reset_tools.cu b/PG-PuReMD/src/cuda/cuda_reset_tools.cu
similarity index 98%
rename from PG-PuReMD/src/cuda_reset_tools.cu
rename to PG-PuReMD/src/cuda/cuda_reset_tools.cu
index 27cb4580..ca435269 100644
--- a/PG-PuReMD/src/cuda_reset_tools.cu
+++ b/PG-PuReMD/src/cuda/cuda_reset_tools.cu
@@ -5,7 +5,7 @@
#include "cuda_utils.h"
#include "cuda_reduction.h"
-#include "reset_tools.h"
+#include "../reset_tools.h"
extern "C"
diff --git a/PG-PuReMD/src/cuda_reset_tools.h b/PG-PuReMD/src/cuda/cuda_reset_tools.h
similarity index 94%
rename from PG-PuReMD/src/cuda_reset_tools.h
rename to PG-PuReMD/src/cuda/cuda_reset_tools.h
index f158afec..2e90b8eb 100644
--- a/PG-PuReMD/src/cuda_reset_tools.h
+++ b/PG-PuReMD/src/cuda/cuda_reset_tools.h
@@ -2,13 +2,13 @@
#ifndef __CUDA_RESET_TOOLS_H__
#define __CUDA_RESET_TOOLS_H__
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-
void Cuda_Reset_Workspace( reax_system *, storage * );
void Cuda_Reset_Atoms( reax_system *, control_params * );
@@ -19,9 +19,9 @@ int Cuda_Reset_Neighbor_Lists( reax_system *, control_params *,
void Cuda_Reset( reax_system*, control_params*, simulation_data*,
storage*, reax_list** );
-
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_shuffle.h b/PG-PuReMD/src/cuda/cuda_shuffle.h
similarity index 97%
rename from PG-PuReMD/src/cuda_shuffle.h
rename to PG-PuReMD/src/cuda/cuda_shuffle.h
index f8dfddfa..0d687271 100644
--- a/PG-PuReMD/src/cuda_shuffle.h
+++ b/PG-PuReMD/src/cuda/cuda_shuffle.h
@@ -22,8 +22,7 @@
#ifndef __CUDA_SHUFFLE_H_
#define __CUDA_SHUFFLE_H_
-#include "reax_types.h"
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
diff --git a/PG-PuReMD/src/cuda/cuda_system_props.cu b/PG-PuReMD/src/cuda/cuda_system_props.cu
new file mode 100644
index 00000000..54957d00
--- /dev/null
+++ b/PG-PuReMD/src/cuda/cuda_system_props.cu
@@ -0,0 +1,1026 @@
+
+#include "cuda_system_props.h"
+
+#include "cuda_utils.h"
+#include "cuda_reduction.h"
+#include "cuda_copy.h"
+#include "cuda_shuffle.h"
+
+#include "../vector.h"
+
+
+CUDA_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 )
+{
+ extern __shared__ rvec xcm[];
+ extern __shared__ rvec vcm[];
+ extern __shared__ rvec amcm[];
+
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ //unsigned int xcm_id = threadIdx.x;
+ unsigned int vcm_id = blockDim.x;
+ unsigned int amcm_id = 2 *(blockDim.x);
+
+ unsigned int index = 0;
+ rvec tmp;
+ real m;
+
+ rvec_MakeZero (xcm [threadIdx.x]);
+ rvec_MakeZero (vcm [vcm_id + threadIdx.x]);
+ rvec_MakeZero (amcm[amcm_id + threadIdx.x]);
+ rvec_MakeZero (tmp);
+
+ if (i < n){
+ m = sbp [ atoms[i].type ].mass;
+ rvec_ScaledAdd (xcm [threadIdx.x], m, atoms [i].x);
+ rvec_ScaledAdd (vcm [vcm_id + threadIdx.x], m, atoms [i].v);
+ rvec_Cross (tmp, atoms[i].x, atoms [i].v);
+ rvec_ScaledAdd (amcm[amcm_id + threadIdx.x], m, tmp);
+ }
+ __syncthreads ();
+
+ for( int offset = blockDim.x / 2; offset > 0; offset >>= 1 ) {
+
+ if ((threadIdx.x < offset)) {
+ index = threadIdx.x + offset;
+ rvec_Add (xcm [threadIdx.x], xcm[index]);
+ rvec_Add (vcm [vcm_id + threadIdx.x], vcm[vcm_id + index]);
+ rvec_Add (amcm[amcm_id + threadIdx.x], amcm[amcm_id + index]);
+ }
+ __syncthreads ();
+ }
+
+ if ((threadIdx.x == 0)){
+ rvec_Copy (res_xcm[blockIdx.x], xcm[0]);
+ rvec_Copy (res_vcm[blockIdx.x], vcm[vcm_id]);
+ rvec_Copy (res_amcm[blockIdx.x], amcm[amcm_id]);
+ }
+}
+
+
+#if defined( __SM_35__)
+CUDA_GLOBAL void center_of_mass_blocks_xcm( single_body_parameters *sbp, reax_atom *atoms,
+ rvec *res_xcm, size_t n )
+{
+ extern __shared__ rvec my_xcm[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int xcm_id = threadIdx.x;
+ unsigned int index = 0;
+ rvec xcm;
+ real m;
+
+ rvec_MakeZero (xcm);
+
+ if (i < n){
+ m = sbp [ atoms[i].type ].mass;
+ rvec_ScaledAdd (xcm , m, atoms [i].x);
+ }
+ __syncthreads ();
+
+ for (int z = 16; z >= 1; z /= 2){
+ xcm[0] += shfl( xcm[0], z);
+ xcm[1] += shfl( xcm[1], z);
+ xcm[2] += shfl( xcm[2], z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0)
+ rvec_Copy( my_xcm[ threadIdx.x >> 5], xcm );
+ __syncthreads ();
+
+ for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
+
+ if ((threadIdx.x < offset)) {
+ index = threadIdx.x + offset;
+ rvec_Add (my_xcm [threadIdx.x], my_xcm[index]);
+ }
+ __syncthreads ();
+ }
+
+ if ((threadIdx.x == 0))
+ rvec_Copy (res_xcm[blockIdx.x], my_xcm[0]);
+}
+
+
+CUDA_GLOBAL void center_of_mass_blocks_vcm( single_body_parameters *sbp, reax_atom *atoms,
+ rvec *res_vcm, size_t n )
+{
+ extern __shared__ rvec my_vcm[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+ rvec vcm;
+ real m;
+
+ rvec_MakeZero (vcm);
+
+ if (i < n){
+ m = sbp [ atoms[i].type ].mass;
+ rvec_ScaledAdd (vcm , m, atoms [i].v);
+ }
+ __syncthreads ();
+
+ for (int z = 16; z >= 1; z /= 2){
+ vcm[0] += shfl( vcm[0], z);
+ vcm[1] += shfl( vcm[1], z);
+ vcm[2] += shfl( vcm[2], z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0)
+ rvec_Copy( my_vcm[ threadIdx.x >> 5], vcm );
+ __syncthreads ();
+
+ for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
+
+ if ((threadIdx.x < offset)) {
+ index = threadIdx.x + offset;
+ rvec_Add (my_vcm [threadIdx.x], my_vcm[index]);
+ }
+ __syncthreads ();
+ }
+
+ if ((threadIdx.x == 0))
+ rvec_Copy (res_vcm[blockIdx.x], my_vcm[0]);
+}
+
+
+CUDA_GLOBAL void center_of_mass_blocks_amcm( single_body_parameters *sbp, reax_atom *atoms,
+ rvec *res_amcm, size_t n )
+{
+ extern __shared__ rvec my_amcm[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+ rvec amcm;
+ real m;
+ rvec tmp;
+
+ rvec_MakeZero (amcm);
+ rvec_MakeZero( tmp );
+
+ if (i < n){
+ m = sbp [ atoms[i].type ].mass;
+ rvec_Cross (tmp, atoms[i].x, atoms [i].v);
+ rvec_ScaledAdd (amcm, m, tmp);
+ }
+ __syncthreads ();
+
+ for (int z = 16; z >= 1; z /= 2){
+ amcm[0] += shfl( amcm[0], z);
+ amcm[1] += shfl( amcm[1], z);
+ amcm[2] += shfl( amcm[2], z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0)
+ rvec_Copy( my_amcm[ threadIdx.x >> 5], amcm );
+ __syncthreads ();
+
+
+ for( int offset = blockDim.x >> 6; offset > 0; offset >>= 1 ) {
+
+ if ((threadIdx.x < offset)) {
+ index = threadIdx.x + offset;
+ rvec_Add (my_amcm[threadIdx.x], my_amcm[index]);
+ }
+ __syncthreads ();
+ }
+
+ if ((threadIdx.x == 0)){
+ rvec_Copy (res_amcm[blockIdx.x], my_amcm[0]);
+ }
+}
+#endif
+
+
+CUDA_GLOBAL void 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[];
+ extern __shared__ rvec sh_amcm[];
+
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+
+ unsigned int xcm_id = threadIdx.x;
+ unsigned int vcm_id = blockDim.x;
+ unsigned int amcm_id = 2 * (blockDim.x);
+
+ unsigned int index = 0;
+ rvec t_xcm, t_vcm, t_amcm;
+
+ rvec_MakeZero (t_xcm);
+ rvec_MakeZero (t_vcm);
+ rvec_MakeZero (t_amcm);
+
+ if (i < n){
+ rvec_Copy ( t_xcm, xcm[threadIdx.x]);
+ rvec_Copy ( t_vcm, vcm[threadIdx.x]);
+ rvec_Copy ( t_amcm, amcm[threadIdx.x]);
+ }
+
+ rvec_Copy (sh_xcm[xcm_id], t_xcm);
+ rvec_Copy (sh_vcm[vcm_id + threadIdx.x], t_vcm);
+ rvec_Copy (sh_amcm[amcm_id + threadIdx.x], t_amcm);
+
+ __syncthreads ();
+
+ for( int offset = blockDim.x / 2; offset > 0; offset >>= 1 ) {
+
+ if (threadIdx.x < offset) {
+ index = threadIdx.x + offset;
+ rvec_Add (sh_xcm [threadIdx.x], sh_xcm[index]);
+ rvec_Add (sh_vcm [vcm_id + threadIdx.x], sh_vcm[vcm_id + index]);
+ rvec_Add (sh_amcm [amcm_id + threadIdx.x], sh_amcm[amcm_id + index]);
+ }
+ __syncthreads ();
+ }
+
+ if (threadIdx.x == 0){
+ rvec_Copy (res_xcm[blockIdx.x], sh_xcm[0]);
+ rvec_Copy (res_vcm[blockIdx.x], sh_vcm[vcm_id]);
+ rvec_Copy (res_amcm[blockIdx.x], sh_amcm[amcm_id]);
+ }
+}
+
+
+CUDA_GLOBAL void compute_center_mass( 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[];
+ extern __shared__ real xz[];
+ extern __shared__ real yy[];
+ extern __shared__ real yz[];
+ extern __shared__ real zz[];
+
+ unsigned int xx_i = threadIdx.x;
+ unsigned int xy_i = blockDim.x;
+ unsigned int xz_i = 2 * blockDim.x;
+ unsigned int yy_i = 3 * blockDim.x;
+ unsigned int yz_i = 4 * blockDim.x;
+ unsigned int zz_i = 5 * blockDim.x;
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+
+ rvec diff, xcm;
+ real m = 0;
+ rvec_MakeZero (diff);
+ xcm[0] = xcm0;
+ 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){
+ m = sbp[ atoms[i].type ].mass;
+ rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
+ xx[ xx_i ] = diff[0] * diff[0] * m;
+ xy[ xy_i + threadIdx.x ] = diff[0] * diff[1] * m;
+ xz[ xz_i + threadIdx.x ] = diff[0] * diff[2] * m;
+ yy[ yy_i + threadIdx.x ] = diff[1] * diff[1] * m;
+ yz[ yz_i + threadIdx.x ] = diff[1] * diff[2] * m;
+ zz[ zz_i + threadIdx.x ] = diff[2] * diff[2] * m;
+ }
+ __syncthreads ();
+
+ 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 ];
+ xz[ xz_i + threadIdx.x ] += xz [ xz_i + index ];
+ yy[ yy_i + threadIdx.x ] += yy [ yy_i + index ];
+ yz[ yz_i + threadIdx.x ] += yz [ yz_i + index ];
+ zz[ zz_i + threadIdx.x ] += zz [ zz_i + index ];
+ }
+ __syncthreads ();
+ }
+
+ 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 ];
+ results [ blockIdx.x*6 + 3 ] = yy [ yy_i + 0 ];
+ results [ blockIdx.x*6 + 4 ] = yz [ yz_i + 0 ];
+ results [ blockIdx.x*6 + 5 ] = zz [ zz_i + 0 ];
+ }
+}
+
+
+CUDA_GLOBAL void compute_center_mass( real *input, real *output, size_t n )
+{
+ extern __shared__ real xx[];
+ extern __shared__ real xy[];
+ extern __shared__ real xz[];
+ extern __shared__ real yy[];
+ extern __shared__ real yz[];
+ extern __shared__ real zz[];
+
+ unsigned int xx_i = threadIdx.x;
+ unsigned int xy_i = blockDim.x;
+ unsigned int xz_i = 2 * blockDim.x;
+ unsigned int yy_i = 3 * blockDim.x;
+ unsigned int yz_i = 4 * blockDim.x;
+ unsigned int zz_i = 5 * blockDim.x;
+
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+
+ 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)
+ {
+ xx [ xx_i ] = input [ threadIdx.x*6 + 0 ];
+ xy [ xy_i + threadIdx.x ] = input [ threadIdx.x*6 + 1 ];
+ xz [ xz_i + threadIdx.x ] = input [ threadIdx.x*6 + 2 ];
+ yy [ yy_i + threadIdx.x ] = input [ threadIdx.x*6 + 3 ];
+ yz [ yz_i + threadIdx.x ] = input [ threadIdx.x*6 + 4 ];
+ zz [ zz_i + threadIdx.x ] = input [ threadIdx.x*6 + 5 ];
+ }
+ __syncthreads ();
+
+ 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 ];
+ xz [ xz_i + threadIdx.x ] += xz [ xz_i + index ];
+ yy [ yy_i + threadIdx.x ] += yy [ yy_i + index ];
+ yz [ yz_i + threadIdx.x ] += yz [ yz_i + index ];
+ zz [ zz_i + threadIdx.x ] += zz [ zz_i + index ];
+ }
+ __syncthreads ();
+ }
+
+ if (threadIdx.x == 0)
+ {
+ output[0] = xx[0];
+ output[1] = xy[xy_i];
+ output[2] = xz[xz_i];
+ output[3] = xz[yy_i];
+ output[4] = xz[yz_i];
+ output[5] = xz[zz_i];
+ }
+}
+
+
+#if defined( __SM_35__)
+CUDA_GLOBAL void compute_center_mass_xx_xy( single_body_parameters *sbp,
+ reax_atom *atoms, real *results, real xcm0, real xcm1, real xcm2,
+ size_t n )
+{
+ extern __shared__ real my_results_xx[];
+ extern __shared__ real my_results_xy[];
+
+ unsigned int xx_i = threadIdx.x;
+ unsigned int xy_i = blockDim.x;
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+ real xx = 0;
+ real xy = 0;
+
+ rvec diff, xcm;
+ real m = 0;
+ rvec_MakeZero (diff);
+ xcm[0] = xcm0;
+ xcm[1] = xcm1;
+ xcm[2] = xcm2;
+
+
+ if (i < n){
+ m = sbp[ atoms[i].type ].mass;
+ rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
+ xx = diff[0] * diff[0] * m;
+ xy = diff[0] * diff[1] * m;
+ }
+ __syncthreads ();
+
+ for (int z = 16; z <= 1; z++){
+ xx += shfl( xx, z);
+ xy += shfl( xy, z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0){
+ my_results_xx[threadIdx.x >> 5] = xx;
+ my_results_xy[threadIdx.x >> 5] = xy;
+ }
+ __syncthreads ();
+
+ for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
+ if (threadIdx.x < offset){
+ index = threadIdx.x + offset;
+ my_results_xx[ threadIdx.x ] += my_results_xx[ index ];
+ my_results_xy[ xy_i + threadIdx.x ] += my_results_xy [ xy_i + index ];
+ }
+ __syncthreads ();
+ }
+
+ if (threadIdx.x == 0) {
+ results [ blockIdx.x*6 ] = my_results_xx [ 0 ];
+ results [ blockIdx.x*6 + 1 ] = my_results_xy [ xy_i + 0 ];
+ }
+}
+
+
+CUDA_GLOBAL void compute_center_mass_xz_yy( single_body_parameters *sbp,
+ reax_atom *atoms, real *results, real xcm0, real xcm1, real xcm2,
+ size_t n )
+{
+ extern __shared__ real my_results_xz[];
+ extern __shared__ real my_results_yy[];
+
+ unsigned int yy_i = blockDim.x;
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+ real xz = 0;
+ real yy = 0;
+
+ rvec diff, xcm;
+ real m = 0;
+ rvec_MakeZero (diff);
+ xcm[0] = xcm0;
+ xcm[1] = xcm1;
+ xcm[2] = xcm2;
+
+ if (i < n){
+ m = sbp[ atoms[i].type ].mass;
+ rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
+ xz = diff[0] * diff[2] * m;
+ yy = diff[1] * diff[1] * m;
+ }
+ __syncthreads ();
+
+ for (int z = 16; z <= 1; z++){
+ xz += shfl( xz, z);
+ yy += shfl( yy, z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0){
+ my_results_xz[threadIdx.x >> 5] = xz;
+ my_results_yy[threadIdx.x >> 5] = yy;
+ }
+ __syncthreads ();
+
+ for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
+ if (threadIdx.x < offset){
+ index = threadIdx.x + offset;
+ my_results_xz[ threadIdx.x ] += my_results_xz [ index ];
+ my_results_yy[ yy_i + threadIdx.x ] += my_results_yy [ yy_i + index ];
+ }
+ __syncthreads ();
+ }
+
+ if (threadIdx.x == 0) {
+ results [ blockIdx.x*6 + 2 ] = my_results_xz [ 0 ];
+ results [ blockIdx.x*6 + 3 ] = my_results_yy [ yy_i + 0 ];
+ }
+}
+
+
+CUDA_GLOBAL void compute_center_mass_yz_zz( single_body_parameters *sbp,
+ reax_atom *atoms, real *results, real xcm0, real xcm1, real xcm2,
+ size_t n )
+{
+ extern __shared__ real my_results_yz[];
+ extern __shared__ real my_results_zz[];
+
+ unsigned int zz_i = blockDim.x;
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ unsigned int index = 0;
+ real yz = 0;
+ real zz = 0;
+
+ rvec diff, xcm;
+ real m = 0;
+ rvec_MakeZero (diff);
+ xcm[0] = xcm0;
+ xcm[1] = xcm1;
+ xcm[2] = xcm2;
+
+ if (i < n)
+ {
+ m = sbp[ atoms[i].type ].mass;
+ rvec_ScaledSum( diff, 1., atoms[i].x, -1., xcm );
+ yz = diff[1] * diff[2] * m;
+ zz = diff[2] * diff[2] * m;
+ }
+ __syncthreads ();
+
+ for (int z = 16; z <= 1; z++){
+ yz += shfl( yz, z);
+ zz += shfl( zz, z);
+ }
+ __syncthreads ();
+
+ if (threadIdx.x % 32 == 0){
+ my_results_yz[threadIdx.x >> 5] = yz;
+ my_results_zz[threadIdx.x >> 5] = zz;
+ }
+ __syncthreads ();
+
+ for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1){
+ if (threadIdx.x < offset){
+ index = threadIdx.x + offset;
+ my_results_yz[ threadIdx.x ] += my_results_yz [ index ];
+ my_results_zz[ zz_i + threadIdx.x ] += my_results_zz [ zz_i + index ];
+ }
+ __syncthreads ();
+ }
+
+ if (threadIdx.x == 0) {
+ results [ blockIdx.x*6 + 4 ] = my_results_yz [ 0 ];
+ results [ blockIdx.x*6 + 5 ] = my_results_zz [ zz_i + 0 ];
+ }
+}
+#endif
+
+
+CUDA_GLOBAL void k_compute_total_mass( single_body_parameters *sbp, reax_atom *my_atoms,
+ real *block_results, int n )
+{
+#if defined(__SM_35__)
+ extern __shared__ real my_sbp[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real sdata = 0;
+
+ if (i < n)
+ {
+ sdata = sbp[ my_atoms[i].type ].mass;
+ }
+ __syncthreads( );
+
+ for(int z = 16; z >=1; z/=2)
+ {
+ sdata += shfl( sdata, z);
+ }
+
+ if (threadIdx.x % 32 == 0)
+ {
+ my_sbp[threadIdx.x >> 5] = sdata;
+ }
+
+ __syncthreads( );
+
+ for(int offset = blockDim.x >> 6; offset > 0; offset >>= 1)
+ {
+ if(threadIdx.x < offset)
+ {
+ my_sbp[threadIdx.x] += my_sbp[threadIdx.x + offset];
+ }
+
+ __syncthreads( );
+ }
+
+ if(threadIdx.x == 0)
+ {
+ block_results[blockIdx.x] = my_sbp[0];
+ }
+
+#else
+ extern __shared__ real sdata[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real x = 0;
+
+ if (i < n)
+ {
+ x = sbp[ my_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)
+ {
+ block_results[ blockIdx.x] = sdata [0];
+ }
+
+#endif
+}
+
+
+extern "C" void dev_compute_total_mass( reax_system *system, real *local_val )
+{
+ real *block_mass = (real *) scratch;
+ cuda_memset( block_mass, 0, sizeof(real) * (1 + BLOCKS_POW_2), "total_mass:tmp" );
+
+ k_compute_total_mass <<<BLOCKS, BLOCK_SIZE, sizeof(real) * BLOCK_SIZE >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, block_mass, system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction <<<1, BLOCKS_POW_2, sizeof(real) * BLOCKS_POW_2 >>>
+ (block_mass, block_mass + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device (local_val, block_mass + BLOCKS_POW_2, sizeof(real),
+ cudaMemcpyDeviceToHost, "total_mass:tmp");
+}
+
+
+CUDA_GLOBAL void k_compute_kinetic_energy( single_body_parameters *sbp, reax_atom *my_atoms,
+ real *block_results, int n )
+{
+#if defined(__SM_35__)
+ extern __shared__ real my_sbpdot[];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real sdata = 0;
+ rvec p;
+
+ if (i < n)
+ {
+ sdata = sbp[ my_atoms[i].type ].mass;
+ rvec_Scale( p, sdata, my_atoms[ i ].v );
+ sdata = 0.5 * rvec_Dot( p, my_atoms[ i ].v );
+ }
+
+ __syncthreads( );
+
+ for(int z = 16; z >=1; z/=2)
+ {
+ sdata += shfl( sdata, z);
+ }
+
+ if (threadIdx.x % 32 == 0)
+ {
+ my_sbpdot[threadIdx.x >> 5] = sdata;
+ }
+
+ __syncthreads( );
+
+ for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1)
+ {
+ if (threadIdx.x < offset)
+ {
+ my_sbpdot[threadIdx.x] += my_sbpdot[threadIdx.x + offset];
+ }
+
+ __syncthreads( );
+ }
+
+ if (threadIdx.x == 0)
+ {
+ block_results[blockIdx.x] = my_sbpdot[0];
+ }
+
+#else
+ extern __shared__ real sdata [];
+ unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
+ real m = 0;
+ rvec p;
+
+ if (i < n)
+ {
+ m = sbp[ my_atoms[i].type ].mass;
+ rvec_Scale( p, m, my_atoms[ i ].v );
+ m = 0.5 * rvec_Dot( p, my_atoms[ i ].v );
+ }
+
+ sdata[ threadIdx.x ] = m;
+ __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)
+ {
+ block_results[blockIdx.x] = sdata[0];
+ }
+#endif
+}
+
+extern "C" void dev_compute_kinetic_energy( reax_system *system,
+ simulation_data *data, real *local_val )
+{
+ real *block_energy = (real *) scratch;
+ cuda_memset( block_energy, 0, sizeof(real) * (BLOCKS_POW_2 + 1), "kinetic_energy:tmp" );
+
+ k_compute_kinetic_energy <<<BLOCKS, BLOCK_SIZE, sizeof(real) * BLOCK_SIZE >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, block_energy, system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction <<<1, BLOCKS_POW_2, sizeof(real) * BLOCKS_POW_2 >>>
+ (block_energy, block_energy + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( local_val, block_energy + BLOCKS_POW_2,
+ //copy_host_device (local_val, &((simulation_data *)data->d_simulation_data)->my_en.e_kin,
+ sizeof(real), cudaMemcpyDeviceToHost, "kinetic_energy:tmp" );
+ //copy_device (block_energy + BLOCKS_POW_2, &((simulation_data *)data->d_simulation_data)->my_en.e_kin,
+ // sizeof (real), "kinetic_energy");
+}
+
+
+extern "C" void dev_compute_momentum( reax_system *system, rvec xcm,
+ rvec vcm, rvec amcm )
+{
+ rvec *l_xcm, *l_vcm, *l_amcm;
+ rvec *r_scratch = (rvec *)scratch;
+
+#if defined( __SM_35__)
+ // xcm
+ cuda_memset( scratch, 0, sizeof(rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
+ l_xcm = r_scratch;
+
+ center_of_mass_blocks_xcm <<< BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
+ ( system->reax_param.d_sbp, system->d_my_atoms, l_xcm, system->n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
+ (l_xcm, l_xcm + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+ copy_host_device( xcm, l_xcm + BLOCKS_POW_2,
+ sizeof(rvec), cudaMemcpyDeviceToHost, "momentum:xcm" );
+
+ // vcm
+ cuda_memset( scratch, 0, sizeof(rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
+ l_vcm = r_scratch;
+
+ center_of_mass_blocks_vcm <<< BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
+ ( system->reax_param.d_sbp, system->d_my_atoms, l_vcm, system->n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
+ (l_vcm, l_vcm + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+ copy_host_device( vcm, l_vcm + BLOCKS_POW_2, sizeof(rvec),
+ cudaMemcpyDeviceToHost, "momentum:vcm" );
+
+ // amcm
+ cuda_memset( scratch, 0, sizeof (rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp");
+ l_amcm = r_scratch;
+
+ center_of_mass_blocks_amcm <<< BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
+ ( system->reax_param.d_sbp, system->d_my_atoms, l_amcm, system->n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
+ (l_amcm, l_amcm + BLOCKS_POW_2, BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+ copy_host_device( amcm, l_amcm + BLOCKS_POW_2, sizeof(rvec),
+ cudaMemcpyDeviceToHost, "momemtum:amcm" );
+
+#else
+ cuda_memset( scratch, 0, 3 * sizeof (rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
+
+ l_xcm = r_scratch;
+ 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 * (sizeof (rvec) * BLOCK_SIZE) >>>
+ ( system->reax_param.d_sbp, system->d_my_atoms, l_xcm, l_vcm, l_amcm, system->n );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ center_of_mass <<< 1, BLOCKS_POW_2, 3 * (sizeof (rvec) * BLOCKS_POW_2) >>>
+ ( l_xcm, l_vcm, l_amcm, l_xcm + BLOCKS_POW_2, l_vcm + BLOCKS_POW_2,
+ l_amcm + BLOCKS_POW_2, BLOCKS_POW_2 );
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( xcm, l_xcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost, "momemtum:xcm" );
+ copy_host_device( vcm, l_vcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost, "momentum:vcm" );
+ copy_host_device( amcm, l_amcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost,"momentum:amcm" );
+#endif
+}
+
+
+extern "C" void dev_compute_inertial_tensor( reax_system *system, real *local_results, rvec my_xcm )
+{
+#if defined(__SM_35__)
+ real *partial_results = (real *) scratch;
+ cuda_memset( partial_results, 0, sizeof (real) * 6 * (BLOCKS_POW_2 + 1), "tensor:tmp" );
+
+ compute_center_mass_xx_xy <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
+ my_xcm[0], my_xcm[1], my_xcm[2], system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ compute_center_mass_xz_yy <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
+ my_xcm[0], my_xcm[1], my_xcm[2], system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ compute_center_mass_yz_zz <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
+ my_xcm[0], my_xcm[1], my_xcm[2], system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ compute_center_mass <<<1, BLOCKS_POW_2, 6 * (sizeof (real) * BLOCKS_POW_2) >>>
+ (partial_results, partial_results + (BLOCKS_POW_2 * 6), BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device( local_results, partial_results + 6 * BLOCKS_POW_2,
+ sizeof(real) * 6, cudaMemcpyDeviceToHost, "tensor:local_results" );
+
+#else
+ real *partial_results = (real *) scratch;
+ //real *local_results;
+
+ cuda_memset (partial_results, 0, sizeof (real) * 6 * (BLOCKS_POW_2 + 1), "tensor:tmp");
+ //local_results = (real *) malloc (sizeof (real) * 6 *(BLOCKS_POW_2+ 1));
+
+ compute_center_mass <<<BLOCKS_POW_2, BLOCK_SIZE, 6 * (sizeof (real) * BLOCK_SIZE) >>>
+ (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
+ my_xcm[0], my_xcm[1], my_xcm[2], system->n);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ compute_center_mass <<<1, BLOCKS_POW_2, 6 * (sizeof (real) * BLOCKS_POW_2) >>>
+ (partial_results, partial_results + (BLOCKS_POW_2 * 6), BLOCKS_POW_2);
+ cudaThreadSynchronize( );
+ cudaCheckError( );
+
+ copy_host_device (local_results, partial_results + 6 * BLOCKS_POW_2,
+ sizeof(real) * 6, cudaMemcpyDeviceToHost, "tensor:local_results");
+#endif
+}
+
+
+extern "C" void dev_sync_simulation_data( simulation_data *data )
+{
+ Output_Sync_Simulation_Data( data, (simulation_data *)data->d_simulation_data );
+}
+
+
+void Cuda_Compute_Kinetic_Energy( reax_system* system, simulation_data* data,
+ MPI_Comm comm )
+{
+ int i;
+ rvec p;
+ real m;
+
+ data->my_en.e_kin = 0.0;
+
+ dev_compute_kinetic_energy( system, data, &data->my_en.e_kin );
+
+ MPI_Allreduce( &data->my_en.e_kin, &data->sys_en.e_kin,
+ 1, MPI_DOUBLE, MPI_SUM, comm );
+
+ data->therm.T = (2. * data->sys_en.e_kin) / (data->N_f * K_B);
+
+ // avoid T being an absolute zero, might cause F.P.E!
+ if ( FABS(data->therm.T) < ALMOST_ZERO )
+ {
+ data->therm.T = ALMOST_ZERO;
+ }
+}
+
+
+void Cuda_Compute_Total_Mass( reax_system *system, simulation_data *data,
+ MPI_Comm comm )
+{
+ int i;
+ real tmp;
+
+ //compute local total mass of the system
+ dev_compute_total_mass( system, &tmp );
+
+ MPI_Allreduce( &tmp, &data->M, 1, MPI_DOUBLE, MPI_SUM, comm );
+
+ data->inv_M = 1. / data->M;
+}
+
+
+void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
+ mpi_datatypes *mpi_data, MPI_Comm comm )
+{
+ int i;
+ real m, det; //xx, xy, xz, yy, yz, zz;
+ real tmp_mat[6], tot_mat[6];
+ rvec my_xcm, my_vcm, my_amcm, my_avcm;
+ rvec tvec, diff;
+ rtensor mat, inv;
+
+ rvec_MakeZero( my_xcm ); // position of CoM
+ rvec_MakeZero( my_vcm ); // velocity of CoM
+ rvec_MakeZero( my_amcm ); // angular momentum of CoM
+ rvec_MakeZero( my_avcm ); // angular velocity of CoM
+
+ /* Compute the position, vel. and ang. momentum about the centre of mass */
+ dev_compute_momentum ( system, my_xcm, my_vcm, my_amcm );
+
+ MPI_Allreduce( my_xcm, data->xcm, 3, MPI_DOUBLE, MPI_SUM, comm );
+ MPI_Allreduce( my_vcm, data->vcm, 3, MPI_DOUBLE, MPI_SUM, comm );
+ MPI_Allreduce( my_amcm, data->amcm, 3, MPI_DOUBLE, MPI_SUM, comm );
+
+ 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 */
+ for ( i = 0; i < 6; ++i )
+ {
+ tmp_mat[i] = 0;
+ }
+
+ dev_compute_inertial_tensor( system, tmp_mat, my_xcm );
+
+ MPI_Reduce( tmp_mat, tot_mat, 6, MPI_DOUBLE, MPI_SUM, MASTER_NODE, comm );
+
+ if ( system->my_rank == MASTER_NODE )
+ {
+ mat[0][0] = tot_mat[3] + tot_mat[5]; // yy + zz;
+ mat[0][1] = mat[1][0] = -tot_mat[1]; // -xy;
+ mat[0][2] = mat[2][0] = -tot_mat[2]; // -xz;
+ mat[1][1] = tot_mat[0] + tot_mat[5]; // xx + zz;
+ mat[2][1] = mat[1][2] = -tot_mat[4]; // -yz;
+ mat[2][2] = tot_mat[0] + tot_mat[3]; // 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 ( 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 );
+ }
+
+ MPI_Bcast( data->avcm, 3, MPI_DOUBLE, MASTER_NODE, comm );
+
+ /* Compute the rotational energy */
+ 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( stderr, "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( stderr, "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] ); */
+ fprintf( stderr, "avcm: %24.15e %24.15e %24.15e\n",
+ data->avcm[0], data->avcm[1], data->avcm[2] );
+#endif
+}
+
+
diff --git a/PG-PuReMD/src/cuda_system_props.h b/PG-PuReMD/src/cuda/cuda_system_props.h
similarity index 65%
rename from PG-PuReMD/src/cuda_system_props.h
rename to PG-PuReMD/src/cuda/cuda_system_props.h
index ce6fccc1..66f620b3 100644
--- a/PG-PuReMD/src/cuda_system_props.h
+++ b/PG-PuReMD/src/cuda/cuda_system_props.h
@@ -2,24 +2,35 @@
#ifndef __CUDA_SYSTEM_PROPS_H__
#define __CUDA_SYSTEM_PROPS_H__
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-
void dev_compute_total_mass( reax_system *, real * );
+
void dev_compute_kinetic_energy( reax_system *, simulation_data *, real * );
+
void dev_compute_momentum( reax_system *, rvec, rvec, rvec );
+
void dev_compute_inertial_tensor( reax_system *, real *, rvec my_xcm );
void dev_sync_simulation_data( simulation_data * );
+
//void dev_compute_kinetic_energy( reax_system *, simulation_data *, real * );
+void Cuda_Compute_Total_Mass( reax_system*, simulation_data*, MPI_Comm );
+
+void Cuda_Compute_Kinetic_Energy( reax_system*, simulation_data*, MPI_Comm );
+
+void Cuda_Compute_Center_of_Mass( reax_system*, simulation_data*,
+ mpi_datatypes*, MPI_Comm );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_torsion_angles.cu b/PG-PuReMD/src/cuda/cuda_torsion_angles.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_torsion_angles.cu
rename to PG-PuReMD/src/cuda/cuda_torsion_angles.cu
index e70c378b..47c087d2 100644
--- a/PG-PuReMD/src/cuda_torsion_angles.cu
+++ b/PG-PuReMD/src/cuda/cuda_torsion_angles.cu
@@ -19,13 +19,14 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#include "reax_types.h"
-#include "index_utils.h"
#include "cuda_torsion_angles.h"
-#include "vector.h"
+
#include "cuda_list.h"
#include "cuda_helpers.h"
+#include "../index_utils.h"
+#include "../vector.h"
+
#define MIN_SINE 1e-10
diff --git a/PG-PuReMD/src/cuda_torsion_angles.h b/PG-PuReMD/src/cuda/cuda_torsion_angles.h
similarity index 57%
rename from PG-PuReMD/src/cuda_torsion_angles.h
rename to PG-PuReMD/src/cuda/cuda_torsion_angles.h
index 235e91b0..a7d9c3cb 100644
--- a/PG-PuReMD/src/cuda_torsion_angles.h
+++ b/PG-PuReMD/src/cuda/cuda_torsion_angles.h
@@ -19,24 +19,18 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
-#ifndef __TORSION_ANGLES_H_
-#define __TORSION_ANGLES_H_
-
-#include "reax_types.h"
-#include "reax_types.h"
-
-CUDA_GLOBAL void Cuda_Torsion_Angles( reax_atom *,
- global_parameters ,
- four_body_header *,
- control_params *,
- reax_list , reax_list ,
- storage ,
- int , int ,
- real *, real *,
- rvec *);
-
-CUDA_GLOBAL void Cuda_Torsion_Angles_PostProcess ( reax_atom *,
- storage ,
- reax_list , int );
+#ifndef __CUDA_TORSION_ANGLES_H_
+#define __CUDA_TORSION_ANGLES_H_
+
+#include "../reax_types.h"
+
+
+CUDA_GLOBAL void Cuda_Torsion_Angles( reax_atom *, global_parameters,
+ four_body_header *, control_params *, reax_list, reax_list,
+ storage, int, int, real *, real *, rvec * );
+
+CUDA_GLOBAL void Cuda_Torsion_Angles_PostProcess( reax_atom *,
+ storage, reax_list, int );
+
#endif
diff --git a/PG-PuReMD/src/cuda_utils.cu b/PG-PuReMD/src/cuda/cuda_utils.cu
similarity index 88%
rename from PG-PuReMD/src/cuda_utils.cu
rename to PG-PuReMD/src/cuda/cuda_utils.cu
index 5899a1ec..7e1757bc 100644
--- a/PG-PuReMD/src/cuda_utils.cu
+++ b/PG-PuReMD/src/cuda/cuda_utils.cu
@@ -149,3 +149,20 @@ extern "C" void print_device_mem_usage( )
total, (long long int)total/(1024.0*1024.0),
free, (long long int)free/(1024.0*1024.0) );
}
+
+
+extern "C" void init_blocks( reax_system *system )
+{
+ compute_blocks( &BLOCKS, &BLOCK_SIZE, system->n );
+ compute_nearest_pow_2( BLOCKS, &BLOCKS_POW_2 );
+
+ compute_blocks( &BLOCKS_N, &BLOCK_SIZE, system->N );
+ compute_nearest_pow_2( BLOCKS_N, &BLOCKS_POW_2_N );
+
+ compute_matvec_blocks( &MATVEC_BLOCKS, system->N );
+
+#if defined(__CUDA_DEBUG_LOG__)
+ fprintf( stderr, " MATVEC_BLOCKS: %d BLOCKSIZE: %d - N:%d \n",
+ MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, system->N );
+#endif
+}
diff --git a/PG-PuReMD/src/cuda_utils.h b/PG-PuReMD/src/cuda/cuda_utils.h
similarity index 80%
rename from PG-PuReMD/src/cuda_utils.h
rename to PG-PuReMD/src/cuda/cuda_utils.h
index 3d63d5e3..bfc4256d 100644
--- a/PG-PuReMD/src/cuda_utils.h
+++ b/PG-PuReMD/src/cuda/cuda_utils.h
@@ -1,7 +1,7 @@
#ifndef __CUDA_UTILS_H_
#define __CUDA_UTILS_H_
-#include "reax_types.h"
+#include "../reax_types.h"
#ifdef __cplusplus
@@ -9,22 +9,33 @@ extern "C" {
#endif
void cuda_malloc( void **, size_t, int, const char * );
+
void cuda_free( void *, const char * );
+
void cuda_memset( void *, int , size_t , const char * );
+
void copy_host_device( void *, void *, size_t, enum cudaMemcpyKind, const char * );
+
void copy_device( void *, void *, size_t, const char * );
void compute_blocks( int *, int *, int );
+
void compute_matvec_blocks( int *, int );
+
void compute_nearest_pow_2( int, int * );
+void init_blocks( reax_system * );
+
void print_device_mem_usage( );
+
#ifdef __cplusplus
#define cudaCheckError() __cudaCheckError( __FILE__, __LINE__ )
static inline void __cudaCheckError( const char *file, const int line )
{
- cudaError err = cudaGetLastError();
+ cudaError err;
+
+ err = cudaGetLastError();
if ( cudaSuccess != err )
{
fprintf( stderr, "[ERROR] runtime error encountered: %s:%d\n", file, line );
@@ -32,19 +43,22 @@ static inline void __cudaCheckError( const char *file, const int line )
exit( RUNTIME_ERROR );
}
+#if defined(DEBUG)
/* More careful checking. However, this will affect performance. */
-// err = cudaDeviceSynchronize();
-// if( cudaSuccess != err )
-// {
-// exit( -1 );
-// }
+ err = cudaDeviceSynchronize( );
+ if( cudaSuccess != err )
+ {
+ exit( RUNTIME_ERROR );
+ }
+#endif
return;
}
#endif
-#endif
-
#ifdef __cplusplus
}
#endif
+
+
+#endif
diff --git a/PG-PuReMD/src/cuda_valence_angles.cu b/PG-PuReMD/src/cuda/cuda_valence_angles.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_valence_angles.cu
rename to PG-PuReMD/src/cuda/cuda_valence_angles.cu
index d778c3b2..21b8d2c8 100644
--- a/PG-PuReMD/src/cuda_valence_angles.cu
+++ b/PG-PuReMD/src/cuda/cuda_valence_angles.cu
@@ -21,9 +21,10 @@
#include "cuda_valence_angles.h"
-#include "index_utils.h"
#include "cuda_list.h"
-#include "vector.h"
+
+#include "../index_utils.h"
+#include "../vector.h"
/* Compute 3-body interactions, in which the main role is played by
diff --git a/PG-PuReMD/src/cuda_valence_angles.h b/PG-PuReMD/src/cuda/cuda_valence_angles.h
similarity index 98%
rename from PG-PuReMD/src/cuda_valence_angles.h
rename to PG-PuReMD/src/cuda/cuda_valence_angles.h
index 65109597..d8abac25 100644
--- a/PG-PuReMD/src/cuda_valence_angles.h
+++ b/PG-PuReMD/src/cuda/cuda_valence_angles.h
@@ -22,8 +22,10 @@
#ifndef __CUDA_VALENCE_ANGLES_H_
#define __CUDA_VALENCE_ANGLES_H_
-#include "reax_types.h"
-#include "vector.h"
+#include "../reax_types.h"
+
+#include "../vector.h"
+
CUDA_GLOBAL void Cuda_Valence_Angles( reax_atom *, global_parameters,
single_body_parameters *, three_body_header *, control_params *,
diff --git a/PG-PuReMD/src/cuda_validation.cu b/PG-PuReMD/src/cuda/cuda_validation.cu
similarity index 99%
rename from PG-PuReMD/src/cuda_validation.cu
rename to PG-PuReMD/src/cuda/cuda_validation.cu
index 34a42430..34ebf6e5 100644
--- a/PG-PuReMD/src/cuda_validation.cu
+++ b/PG-PuReMD/src/cuda/cuda_validation.cu
@@ -1,13 +1,12 @@
-#include "reax_types.h"
#include "cuda_validation.h"
#include "cuda_utils.h"
-#include "index_utils.h"
-#include "list.h"
-#include "tool_box.h"
-#include "vector.h"
+#include "../index_utils.h"
+#include "../list.h"
+#include "../tool_box.h"
+#include "../vector.h"
bool check_zero( real p1, real p2 )
diff --git a/PG-PuReMD/src/cuda_validation.h b/PG-PuReMD/src/cuda/cuda_validation.h
similarity index 97%
rename from PG-PuReMD/src/cuda_validation.h
rename to PG-PuReMD/src/cuda/cuda_validation.h
index 42eb37a4..7faa773b 100644
--- a/PG-PuReMD/src/cuda_validation.h
+++ b/PG-PuReMD/src/cuda/cuda_validation.h
@@ -3,50 +3,60 @@
#ifndef __CUDA_VALIDATION_H__
#define __CUDA_VALIDATION_H__
-#include "reax_types.h"
+#include "../reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-
int validate_neighbors( reax_system *, reax_list **lists );
+
int validate_sym_dbond_indices( reax_system *system,
storage *workspace, reax_list **lists );
int validate_bonds( reax_system *, storage *, reax_list ** );
+
int validate_hbonds( reax_system *, storage *, reax_list ** );
+
int validate_sparse_matrix( reax_system *, storage * );
int validate_grid( reax_system * );
+
int validate_workspace( reax_system *, storage * );
int validate_data( reax_system *, simulation_data * );
+
int validate_three_bodies( reax_system *, storage *,
reax_list ** );
+
int validate_atoms( reax_system *, reax_list ** );
int print_sparse_matrix( sparse_matrix *H );
+
int print_sparse_matrix_host( sparse_matrix *H );
int print_host_rvec2( rvec2 *, int );
+
int print_device_rvec2( rvec2 *, int );
int print_host_array( real *, int );
+
int print_device_array( real *, int );
void compare_rvec2( rvec2 *host, rvec2 *device, int N,
const char *msg );
+
void compare_array( real *host, real *device, int N,
const char *msg );
int check_zeros_host( rvec2 *host, int n, const char * );
-int check_zeros_device( rvec2 *device, int n, const char * );
-
+int check_zeros_device( rvec2 *device, int n, const char * );
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/cuda_hydrogen_bonds.h b/PG-PuReMD/src/cuda_hydrogen_bonds.h
deleted file mode 100644
index 7e1644f1..00000000
--- a/PG-PuReMD/src/cuda_hydrogen_bonds.h
+++ /dev/null
@@ -1,66 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD - Purdue ReaxFF Molecular Dynamics Program
-
- Copyright (2010) Purdue University
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Joseph Fogarty, jcfogart@mail.usf.edu
- Sagar Pandit, pandit@usf.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 __HBONDS_H_
-#define __HBONDS_H_
-
-#include "reax_types.h"
-#include "reax_types.h"
-
-CUDA_GLOBAL void Cuda_Hydrogen_Bonds_HNbrs ( reax_atom *,
- storage ,
- reax_list );
-
-CUDA_GLOBAL void Cuda_Hydrogen_Bonds_HNbrs_BL ( reax_atom *,
- storage ,
- reax_list, int );
-
-CUDA_GLOBAL void Cuda_Hydrogen_Bonds_PostProcess ( reax_atom *,
- storage ,
- reax_list , int );
-
-CUDA_GLOBAL void Cuda_Hydrogen_Bonds( reax_atom *,
- single_body_parameters *,
- hbond_parameters *,
- global_parameters ,
- control_params *,
- storage ,
- reax_list ,
- reax_list ,
- int ,
- int ,
- real *,
- rvec *);
-
-CUDA_GLOBAL void Cuda_Hydrogen_Bonds_MT( reax_atom *,
- single_body_parameters *,
- hbond_parameters *,
- global_parameters ,
- control_params *,
- storage ,
- reax_list ,
- reax_list ,
- int ,
- int ,
- real *,
- rvec *);
-
-#endif
diff --git a/PG-PuReMD/src/cuda_init_md.cu b/PG-PuReMD/src/cuda_init_md.cu
deleted file mode 100644
index 044e8e73..00000000
--- a/PG-PuReMD/src/cuda_init_md.cu
+++ /dev/null
@@ -1,14 +0,0 @@
-
-#include "cuda_init_md.h"
-
-#include "reax_types.h"
-#include "cuda_utils.h"
-
-#include "tool_box.h"
-
-void Cuda_Init_ScratchArea( )
-{
- cuda_malloc( (void **)&scratch, DEVICE_SCRATCH_SIZE, TRUE, "device:scratch" );
-
- host_scratch = (void *) smalloc( HOST_SCRATCH_SIZE, "host:scratch" );
-}
diff --git a/PG-PuReMD/src/cuda_init_md.h b/PG-PuReMD/src/cuda_init_md.h
deleted file mode 100644
index cf7b5249..00000000
--- a/PG-PuReMD/src/cuda_init_md.h
+++ /dev/null
@@ -1,15 +0,0 @@
-
-#ifndef __CUDA_INIT_MD_H__
-#define __CUDA_INIT_MD_H__
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-void Cuda_Init_ScratchArea( );
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
diff --git a/PG-PuReMD/src/cuda_integrate.cu b/PG-PuReMD/src/cuda_integrate.cu
deleted file mode 100644
index 936c6816..00000000
--- a/PG-PuReMD/src/cuda_integrate.cu
+++ /dev/null
@@ -1,105 +0,0 @@
-
-#include "cuda_integrate.h"
-#include "reax_types.h"
-
-#include "vector.h"
-#include "cuda_utils.h"
-
-
-CUDA_GLOBAL void k_update_velocity_1( reax_atom *my_atoms,
- single_body_parameters *sbp, real dt, 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 */
- atom = &(my_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 );
- /* 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, real dt )
-{
- int blocks;
-
- blocks = system->n / DEF_BLOCK_SIZE +
- ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
- k_update_velocity_1 <<< blocks, DEF_BLOCK_SIZE >>>
- (system->d_my_atoms, system->reax_param.d_sbp, dt, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_update_velocity_2( reax_atom *my_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 */
- atom = &(my_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 )
-{
- int blocks;
-
- blocks = system->n / DEF_BLOCK_SIZE +
- ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
- k_update_velocity_2 <<< blocks, DEF_BLOCK_SIZE >>>
- (system->d_my_atoms, system->reax_param.d_sbp, dt, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_scale_velocities( reax_atom *my_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 */
- atom = &(my_atoms[i]);
- rvec_Scale( atom->v, lambda, atom->v );
-}
-
-
-void bNVT_scale_velocities( reax_system *system, real lambda )
-{
- int blocks;
-
- blocks = system->n / DEF_BLOCK_SIZE +
- ((system->n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
- k_scale_velocities <<< blocks, DEF_BLOCK_SIZE >>>
- (system->d_my_atoms, lambda, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
diff --git a/PG-PuReMD/src/cuda_lin_alg.cu b/PG-PuReMD/src/cuda_lin_alg.cu
deleted file mode 100644
index 4f37d577..00000000
--- a/PG-PuReMD/src/cuda_lin_alg.cu
+++ /dev/null
@@ -1,624 +0,0 @@
-/*----------------------------------------------------------------------
- PuReMD - Purdue ReaxFF Molecular Dynamics Program
-
- Copyright (2010) Purdue University
- Hasan Metin Aktulga, haktulga@cs.purdue.edu
- Joseph Fogarty, jcfogart@mail.usf.edu
- Sagar Pandit, pandit@usf.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 "reax_types.h"
-
-#include "cuda_shuffle.h"
-#include "cuda_utils.h"
-#include "cuda_reduction.h"
-
-
-//one thread per row
-CUDA_GLOBAL void k_matvec( sparse_matrix H, real *vec, real *results,
- int rows )
-{
- int i, col;
- real results_row;
- real val;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( i >= rows )
- {
- return;
- }
-
- results_row = 0;
-
- 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 >>>
-//CUDA_GLOBAL void __launch_bounds__(384, 16) k_matvec_csr(sparse_matrix H, real *vec, real *results, int num_rows)
-CUDA_GLOBAL void k_matvec_csr( sparse_matrix H, real *vec, real *results,
- int num_rows )
-{
-#if defined(__SM_35__)
- real vals;
- int x;
-#else
- extern __shared__ real vals[];
-#endif
- int jj;
- int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
- int warp_id = thread_id / MATVEC_KER_THREADS_PER_ROW;
- int lane = thread_id & ( MATVEC_KER_THREADS_PER_ROW - 1);
- int row_start;
- int row_end;
- // one warp per row
- int row = warp_id;
-
-#if defined(__SM_35__)
- vals = 0;
-#else
- vals[threadIdx.x] = 0;
-#endif
-
- if (row < num_rows)
- {
- row_start = H.start[row];
- row_end = H.end[row];
-
- // compute running sum per thread
- for ( jj = row_start + lane; jj < row_end;
- jj += MATVEC_KER_THREADS_PER_ROW )
-#if defined(__SM_35__)
- {
- vals += H.entries[jj].val * vec[ H.entries[jj].j ];
- }
- }
-#else
- {
- vals[threadIdx.x] += H.entries[jj].val * vec[ H.entries[jj].j ];
- }
- }
-
- __syncthreads( );
-#endif
-
- // parallel reduction in shared memory
- //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
-#if defined(__SM_35__)
- for (x = MATVEC_KER_THREADS_PER_ROW >> 1; x >= 1; x/=2)
- {
- vals += shfl( vals, x );
- }
-
- if (lane == 0 && row < num_rows)
- {
- results[row] = vals;
- }
-#else
- 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];
- }
-#endif
-}
-
-
-//one thread per row
-CUDA_GLOBAL void k_dual_matvec( sparse_matrix H, rvec2 *vec, rvec2 *results,
- int rows )
-{
- int i, c, col;
- rvec2 results_row;
- real val;
-
- i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( i >= rows)
- {
- return;
- }
-
- results_row[0] = 0.0;
- results_row[1] = 0.0;
-
- for (c = H.start[i]; c < H.end[i]; c++)
- {
- col = H.entries [c].j;
- val = H.entries[c].val;
-
- results_row[0] += val * vec [col][0];
- results_row[1] += val * vec [col][1];
- }
-
- results[i][0] = results_row[0];
- results[i][1] = results_row[1];
-}
-
-
-//32 thread warp per matrix row.
-//invoked as follows
-// <<< system->N, 32 >>>
-//CUDA_GLOBAL void __launch_bounds__(384, 8) k_dual_matvec_csr(sparse_matrix H, rvec2 *vec, rvec2 *results, int num_rows)
-CUDA_GLOBAL void k_dual_matvec_csr( sparse_matrix H, rvec2 *vec,
- rvec2 *results, int num_rows )
-{
-#if defined(__SM_35__)
- rvec2 rvals;
- int thread_id = blockDim.x * blockIdx.x + threadIdx.x;
- int warp_id = thread_id / MATVEC_KER_THREADS_PER_ROW;
- int lane = thread_id & (MATVEC_KER_THREADS_PER_ROW - 1);
- int row_start;
- int row_end;
- // one warp per row
- int row = warp_id;
-
- rvals[0] = 0;
- rvals[1] = 0;
-
- if (row < num_rows)
- {
- row_start = H.start[row];
- row_end = H.end[row];
-
- for(int jj = row_start + lane; jj < row_end; jj += MATVEC_KER_THREADS_PER_ROW)
- {
- rvals[0] += H.entries[jj].val * vec [ H.entries[jj].j ][0];
- rvals[1] += H.entries[jj].val * vec [ H.entries[jj].j ][1];
- }
- }
-
- for (int s = MATVEC_KER_THREADS_PER_ROW >> 1; s >= 1; s /= 2)
- {
- rvals[0] += shfl( rvals[0], s);
- rvals[1] += shfl( rvals[1], s);
- }
-
- if (lane == 0 && row < num_rows)
- {
- results[row][0] = rvals[0];
- results[row][1] = rvals[1];
- }
-
-#else
- extern __shared__ rvec2 rvals[];
- 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;
-
- rvals[threadIdx.x][0] = 0;
- rvals[threadIdx.x][1] = 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)
- {
- rvals[threadIdx.x][0] += H.entries[jj].val * vec [ H.entries[jj].j ][0];
- rvals[threadIdx.x][1] += H.entries[jj].val * vec [ H.entries[jj].j ][1];
- }
- }
-
- __syncthreads( );
-
- // parallel reduction in shared memory
- //SIMD instructions with a WARP are synchronous -- so we do not need to synch here
- if (lane < 16)
- {
- rvals[threadIdx.x][0] += rvals[threadIdx.x + 16][0];
- rvals[threadIdx.x][1] += rvals[threadIdx.x + 16][1];
- }
- __syncthreads( );
- if (lane < 8)
- {
- rvals[threadIdx.x][0] += rvals[threadIdx.x + 8][0];
- rvals[threadIdx.x][1] += rvals[threadIdx.x + 8][1];
- }
- __syncthreads( );
- if (lane < 4)
- {
- rvals[threadIdx.x][0] += rvals[threadIdx.x + 4][0];
- rvals[threadIdx.x][1] += rvals[threadIdx.x + 4][1];
- }
- __syncthreads( );
- if (lane < 2)
- {
- rvals[threadIdx.x][0] += rvals[threadIdx.x + 2][0];
- rvals[threadIdx.x][1] += rvals[threadIdx.x + 2][1];
- }
- __syncthreads( );
- if (lane < 1)
- {
- rvals[threadIdx.x][0] += rvals[threadIdx.x + 1][0];
- rvals[threadIdx.x][1] += rvals[threadIdx.x + 1][1];
- }
- __syncthreads( );
-
- // first thread writes the result
- if (lane == 0 && row < num_rows)
- {
- results[row][0] = rvals[threadIdx.x][0];
- results[row][1] = rvals[threadIdx.x][1];
- }
-
-#endif
-}
-
-
-void Cuda_Vector_Sum( real *res, real a, real *x, real b, real *y, int count )
-{
- //res = ax + by
- //use the cublas here
- int blocks;
-
- blocks = (count / DEF_BLOCK_SIZE) +
- ((count % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_vector_sum <<< blocks, DEF_BLOCK_SIZE >>>
- ( res, a, x, b, y, count );
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-void Cuda_CG_Preconditioner( real *res, real *a, real *b, int count )
-{
- //res = a*b - vector multiplication
- //use the cublas here.
- int blocks;
-
- blocks = (count / DEF_BLOCK_SIZE) +
- ((count % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_vector_mul <<< blocks, DEF_BLOCK_SIZE >>>
- ( res, a, b, count );
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_diagonal_preconditioner(storage p_workspace, rvec2 *b, int n)
-{
- storage *workspace;
- int j;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( j >= n )
- {
- return;
- }
-
- workspace = &( p_workspace );
-
- //for( j = 0; j < system->n; ++j ) {
- // residual
- workspace->r2[j][0] = b[j][0] - workspace->q2[j][0];
- workspace->r2[j][1] = b[j][1] - workspace->q2[j][1];
-
- // apply diagonal pre-conditioner
- workspace->d2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
- workspace->d2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
- //}
-}
-
-
-void Cuda_CG_Diagonal_Preconditioner( storage *workspace, rvec2 *b, int n )
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_diagonal_preconditioner <<< blocks, DEF_BLOCK_SIZE >>>
- (*workspace, b, n);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_dual_cg_preconditioner( storage p_workspace, rvec2 *x,
- real alpha_0, real alpha_1, int n, rvec2 *my_dot )
-{
- storage *workspace;
- rvec2 alpha;
- int j;
-
- j = blockIdx.x * blockDim.x + threadIdx.x;
-
- if ( j >= n )
- {
- return;
- }
-
- workspace = &( p_workspace );
- alpha[0] = alpha_0;
- alpha[1] = alpha_1;
- my_dot[j][0] = my_dot[j][1] = 0.0;
-
- //for( j = 0; j < system->n; ++j ) {
- // update x
- x[j][0] += alpha[0] * workspace->d2[j][0];
- x[j][1] += alpha[1] * workspace->d2[j][1];
-
- // update residual
- workspace->r2[j][0] -= alpha[0] * workspace->q2[j][0];
- workspace->r2[j][1] -= alpha[1] * workspace->q2[j][1];
-
- // apply diagonal pre-conditioner
- workspace->p2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
- workspace->p2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
-
- // dot product: r.p
- my_dot[j][0] = workspace->r2[j][0] * workspace->p2[j][0];
- my_dot[j][1] = workspace->r2[j][1] * workspace->p2[j][1];
- //}
-}
-
-
-void Cuda_DualCG_Preconditioner( storage *workspace, rvec2 *x, rvec2 alpha,
- int n, rvec2 result )
-{
- int blocks;
- rvec2 *tmp = (rvec2 *) scratch;
-
- cuda_memset( tmp, 0, sizeof(rvec2) * ( 2 * n + 1),
- "cuda_dualcg_preconditioner" );
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_dual_cg_preconditioner <<< blocks, DEF_BLOCK_SIZE >>>
- (*workspace, x, alpha[0], alpha[1], n, tmp);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- //Reduction to calculate my_dot
- k_reduction_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
- ( tmp, tmp + n, n);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
- ( tmp + n, tmp + 2*n, blocks);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( result, (tmp + 2*n), sizeof(rvec2),
- cudaMemcpyDeviceToHost, "my_dot" );
-}
-
-
-void Cuda_Norm( rvec2 *arr, int n, rvec2 result )
-{
- int blocks;
- rvec2 *tmp = (rvec2 *) scratch;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_norm_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
- (arr, tmp, n, INITIAL);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_norm_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
- (tmp, tmp + BLOCKS_POW_2, blocks, FINAL );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( result, tmp + BLOCKS_POW_2, sizeof(rvec2),
- cudaMemcpyDeviceToHost, "cuda_norm_rvec2" );
-}
-
-
-void Cuda_Dot( rvec2 *a, rvec2 *b, rvec2 result, int n )
-{
- int blocks;
- rvec2 *tmp = (rvec2 *) scratch;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_dot_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * DEF_BLOCK_SIZE >>>
- ( a, b, tmp, n );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_norm_rvec2 <<< 1, BLOCKS_POW_2, sizeof(rvec2) * BLOCKS_POW_2 >>>
- //k_norm_rvec2 <<< blocks, DEF_BLOCK_SIZE, sizeof(rvec2) * BLOCKS_POW_2 >>>
- ( tmp, tmp + BLOCKS_POW_2, blocks, FINAL );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( result, tmp + BLOCKS_POW_2, sizeof(rvec2),
- cudaMemcpyDeviceToHost, "cuda_dot" );
-}
-
-
-void Cuda_Vector_Sum_Rvec2(rvec2 *x, rvec2 *a, rvec2 b, rvec2 *c, int n)
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_rvec2_pbetad <<< blocks, DEF_BLOCK_SIZE >>>
- ( x, a, b[0], b[1], c, n);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_rvec2_to_real_copy( real *dst, rvec2 *src, int index, int n )
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= n)
- {
- return;
- }
-
- dst[i] = src[i][index];
-}
-
-
-void Cuda_RvecCopy_From( real *dst, rvec2 *src, int index, int n )
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_rvec2_to_real_copy <<< blocks, DEF_BLOCK_SIZE >>>
- ( dst, src, index, n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-CUDA_GLOBAL void k_real_to_rvec2_copy( rvec2 *dst, real *src, int index, int n)
-{
- int i = blockIdx.x * blockDim.x + threadIdx.x;
-
- if (i >= n)
- {
- return;
- }
-
- dst[i][index] = src[i];
-}
-
-
-void Cuda_RvecCopy_To(rvec2 *dst, real *src, int index, int n)
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE == 0) ? 0 : 1);
-
- k_real_to_rvec2_copy <<< blocks, DEF_BLOCK_SIZE >>>
- ( dst, src, index, n);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-void Cuda_Dual_Matvec( sparse_matrix *H, rvec2 *a, rvec2 *b, int n, int size )
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE) == 0 ? 0 : 1);
-
- cuda_memset( b, 0, sizeof(rvec2) * size, "dual_matvec:result" );
-
- //One thread per row implementation
- //k_dual_matvec <<< blocks, DEF_BLOCK_SIZE >>>
- // (*H, a, b, n);
- //cudaThreadSynchronize ();
- //cudaCheckError ();
-
- //One warp per row implementation
-#if defined(__SM_35__)
- k_dual_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE >>>
-#else
- k_dual_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE,
- sizeof(rvec2) * MATVEC_BLOCK_SIZE >>>
-#endif
- ( *H, a, b, n );
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
-
-
-void Cuda_Matvec( sparse_matrix *H, real *a, real *b, int n, int size )
-{
- int blocks;
-
- blocks = (n / DEF_BLOCK_SIZE) +
- (( n % DEF_BLOCK_SIZE) == 0 ? 0 : 1);
-
- cuda_memset( b, 0, sizeof(real) * size, "dual_matvec:result" );
-
- //one thread per row implementation
- //k_matvec <<< blocks, DEF_BLOCK_SIZE >>>
- // (*H, a, b, n);
- //cudaThreadSynchronize ();
- //cudaCheckError ();
-
-#if defined(__SM_35__)
- k_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE >>>
-#else
- k_matvec_csr <<< MATVEC_BLOCKS, MATVEC_BLOCK_SIZE,
- sizeof(real) * MATVEC_BLOCK_SIZE>>>
-#endif
- (*H, a, b, n);
-
- cudaThreadSynchronize( );
- cudaCheckError( );
-}
diff --git a/PG-PuReMD/src/cuda_system_props.cu b/PG-PuReMD/src/cuda_system_props.cu
deleted file mode 100644
index 3202f64a..00000000
--- a/PG-PuReMD/src/cuda_system_props.cu
+++ /dev/null
@@ -1,406 +0,0 @@
-
-#include "cuda_system_props.h"
-
-#include "cuda_utils.h"
-#include "cuda_reduction.h"
-#include "center_mass.h"
-#include "cuda_copy.h"
-#include "cuda_shuffle.h"
-
-#include "vector.h"
-
-
-CUDA_GLOBAL void k_compute_total_mass( single_body_parameters *sbp, reax_atom *my_atoms,
- real *block_results, int n )
-{
-#if defined(__SM_35__)
- extern __shared__ real my_sbp[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real sdata = 0;
-
- if (i < n)
- {
- sdata = sbp[ my_atoms[i].type ].mass;
- }
- __syncthreads( );
-
- for(int z = 16; z >=1; z/=2)
- {
- sdata += shfl( sdata, z);
- }
-
- if (threadIdx.x % 32 == 0)
- {
- my_sbp[threadIdx.x >> 5] = sdata;
- }
-
- __syncthreads( );
-
- for(int offset = blockDim.x >> 6; offset > 0; offset >>= 1)
- {
- if(threadIdx.x < offset)
- {
- my_sbp[threadIdx.x] += my_sbp[threadIdx.x + offset];
- }
-
- __syncthreads( );
- }
-
- if(threadIdx.x == 0)
- {
- block_results[blockIdx.x] = my_sbp[0];
- }
-
-#else
- extern __shared__ real sdata[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real x = 0;
-
- if (i < n)
- {
- x = sbp[ my_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)
- {
- block_results[ blockIdx.x] = sdata [0];
- }
-
-#endif
-}
-
-
-extern "C" void dev_compute_total_mass( reax_system *system, real *local_val )
-{
- real *block_mass = (real *) scratch;
- cuda_memset( block_mass, 0, sizeof(real) * (1 + BLOCKS_POW_2), "total_mass:tmp" );
-
- k_compute_total_mass <<<BLOCKS, BLOCK_SIZE, sizeof(real) * BLOCK_SIZE >>>
- (system->reax_param.d_sbp, system->d_my_atoms, block_mass, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction <<<1, BLOCKS_POW_2, sizeof(real) * BLOCKS_POW_2 >>>
- (block_mass, block_mass + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device (local_val, block_mass + BLOCKS_POW_2, sizeof(real),
- cudaMemcpyDeviceToHost, "total_mass:tmp");
-}
-
-
-CUDA_GLOBAL void k_compute_kinetic_energy( single_body_parameters *sbp, reax_atom *my_atoms,
- real *block_results, int n )
-{
-#if defined(__SM_35__)
- extern __shared__ real my_sbpdot[];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real sdata = 0;
- rvec p;
-
- if (i < n)
- {
- sdata = sbp[ my_atoms[i].type ].mass;
- rvec_Scale( p, sdata, my_atoms[ i ].v );
- sdata = 0.5 * rvec_Dot( p, my_atoms[ i ].v );
- }
-
- __syncthreads( );
-
- for(int z = 16; z >=1; z/=2)
- {
- sdata += shfl( sdata, z);
- }
-
- if (threadIdx.x % 32 == 0)
- {
- my_sbpdot[threadIdx.x >> 5] = sdata;
- }
-
- __syncthreads( );
-
- for (int offset = blockDim.x >> 6; offset > 0; offset >>= 1)
- {
- if (threadIdx.x < offset)
- {
- my_sbpdot[threadIdx.x] += my_sbpdot[threadIdx.x + offset];
- }
-
- __syncthreads( );
- }
-
- if (threadIdx.x == 0)
- {
- block_results[blockIdx.x] = my_sbpdot[0];
- }
-
-#else
- extern __shared__ real sdata [];
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real m = 0;
- rvec p;
-
- if (i < n)
- {
- m = sbp[ my_atoms[i].type ].mass;
- rvec_Scale( p, m, my_atoms[ i ].v );
- m = 0.5 * rvec_Dot( p, my_atoms[ i ].v );
- }
-
- sdata[ threadIdx.x ] = m;
- __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)
- {
- block_results[blockIdx.x] = sdata[0];
- }
-#endif
-}
-
-extern "C" void dev_compute_kinetic_energy( reax_system *system,
- simulation_data *data, real *local_val )
-{
- real *block_energy = (real *) scratch;
- cuda_memset( block_energy, 0, sizeof(real) * (BLOCKS_POW_2 + 1), "kinetic_energy:tmp" );
-
- k_compute_kinetic_energy <<<BLOCKS, BLOCK_SIZE, sizeof(real) * BLOCK_SIZE >>>
- (system->reax_param.d_sbp, system->d_my_atoms, block_energy, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction <<<1, BLOCKS_POW_2, sizeof(real) * BLOCKS_POW_2 >>>
- (block_energy, block_energy + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( local_val, block_energy + BLOCKS_POW_2,
- //copy_host_device (local_val, &((simulation_data *)data->d_simulation_data)->my_en.e_kin,
- sizeof(real), cudaMemcpyDeviceToHost, "kinetic_energy:tmp" );
- //copy_device (block_energy + BLOCKS_POW_2, &((simulation_data *)data->d_simulation_data)->my_en.e_kin,
- // sizeof (real), "kinetic_energy");
-}
-
-
-extern "C" void dev_compute_momentum( reax_system *system, rvec xcm,
- rvec vcm, rvec amcm )
-{
- rvec *l_xcm, *l_vcm, *l_amcm;
- rvec *r_scratch = (rvec *)scratch;
-
-#if defined( __SM_35__)
- // xcm
- cuda_memset( scratch, 0, sizeof(rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
- l_xcm = r_scratch;
-
- center_of_mass_blocks_xcm <<<BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, l_xcm, system->n );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
- (l_xcm, l_xcm + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
- copy_host_device( xcm, l_xcm + BLOCKS_POW_2,
- sizeof(rvec), cudaMemcpyDeviceToHost, "momentum:xcm" );
-
- // vcm
- cuda_memset( scratch, 0, sizeof(rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
- l_vcm = r_scratch;
-
- center_of_mass_blocks_vcm <<<BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, l_vcm, system->n );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
- (l_vcm, l_vcm + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
- copy_host_device( vcm, l_vcm + BLOCKS_POW_2, sizeof(rvec),
- cudaMemcpyDeviceToHost, "momentum:vcm" );
-
- // amcm
- cuda_memset( scratch, 0, sizeof (rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp");
- l_amcm = r_scratch;
-
- center_of_mass_blocks_amcm <<<BLOCKS_POW_2,BLOCK_SIZE,(sizeof(rvec) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, l_amcm, system->n );
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- k_reduction_rvec <<<1, BLOCKS_POW_2, (sizeof(rvec) * BLOCKS_POW_2) >>>
- (l_amcm, l_amcm + BLOCKS_POW_2, BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
- copy_host_device( amcm, l_amcm + BLOCKS_POW_2, sizeof(rvec),
- cudaMemcpyDeviceToHost, "momemtum:amcm" );
-
-#else
- cuda_memset( scratch, 0, 3 * sizeof (rvec) * (BLOCKS_POW_2 + 1), "momentum:tmp" );
-
- l_xcm = r_scratch;
- 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 * (sizeof (rvec) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, l_xcm, l_vcm, l_amcm, system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- center_of_mass <<<1, BLOCKS_POW_2, 3 * (sizeof (rvec) * BLOCKS_POW_2) >>>
- (l_xcm, l_vcm, l_amcm,
- l_xcm + BLOCKS_POW_2,
- l_vcm + BLOCKS_POW_2,
- l_amcm + BLOCKS_POW_2,
- BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( xcm, l_xcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost, "momemtum:xcm" );
- copy_host_device( vcm, l_vcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost, "momentum:vcm" );
- copy_host_device( amcm, l_amcm + BLOCKS_POW_2, sizeof (rvec), cudaMemcpyDeviceToHost,"momentum:amcm" );
-#endif
-}
-
-
-extern "C" void dev_compute_inertial_tensor( reax_system *system, real *local_results, rvec my_xcm )
-{
-#if defined(__SM_35__)
- real *partial_results = (real *) scratch;
- cuda_memset( partial_results, 0, sizeof (real) * 6 * (BLOCKS_POW_2 + 1), "tensor:tmp" );
-
- compute_center_mass_xx_xy <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
- my_xcm[0], my_xcm[1], my_xcm[2], system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- compute_center_mass_xz_yy <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
- my_xcm[0], my_xcm[1], my_xcm[2], system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- compute_center_mass_yz_zz <<<BLOCKS_POW_2, BLOCK_SIZE, 2 * (sizeof (real) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
- my_xcm[0], my_xcm[1], my_xcm[2], system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- compute_center_mass <<<1, BLOCKS_POW_2, 6 * (sizeof (real) * BLOCKS_POW_2) >>>
- (partial_results, partial_results + (BLOCKS_POW_2 * 6), BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device( local_results, partial_results + 6 * BLOCKS_POW_2,
- sizeof(real) * 6, cudaMemcpyDeviceToHost, "tensor:local_results" );
-
-#else
- real *partial_results = (real *) scratch;
- //real *local_results;
-
- cuda_memset (partial_results, 0, sizeof (real) * 6 * (BLOCKS_POW_2 + 1), "tensor:tmp");
- //local_results = (real *) malloc (sizeof (real) * 6 *(BLOCKS_POW_2+ 1));
-
- compute_center_mass <<<BLOCKS_POW_2, BLOCK_SIZE, 6 * (sizeof (real) * BLOCK_SIZE) >>>
- (system->reax_param.d_sbp, system->d_my_atoms, partial_results,
- my_xcm[0], my_xcm[1], my_xcm[2], system->n);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- compute_center_mass <<<1, BLOCKS_POW_2, 6 * (sizeof (real) * BLOCKS_POW_2) >>>
- (partial_results, partial_results + (BLOCKS_POW_2 * 6), BLOCKS_POW_2);
- cudaThreadSynchronize( );
- cudaCheckError( );
-
- copy_host_device (local_results, partial_results + 6 * BLOCKS_POW_2,
- sizeof(real) * 6, cudaMemcpyDeviceToHost, "tensor:local_results");
-#endif
-}
-
-
-extern "C" void dev_sync_simulation_data( simulation_data *data )
-{
- Output_Sync_Simulation_Data( data, (simulation_data *)data->d_simulation_data );
-}
-
-
-/*
-CUDA_GLOBAL void ker_kinetic_energy (reax_atom *my_atoms,
- single_body_parameters *sbp, int n, real *block_results)
-{
- extern __shared__ real sken[];
- rvec p;
- unsigned int i = blockIdx.x * blockDim.x + threadIdx.x;
- real x = 0;
-
- if(i < n)
- {
- m = sbp[my_atoms[i].type].mass;
- rvec_Scale( p, m, my_atoms[i].v );
- x = 0.5 * rvec_Dot( p, my_atoms[i].v );
- }
- sken[threadIdx.x] = x;
- __syncthreads();
-
- for(int offset = blockDim.x / 2; offset > 0; offset >>= 1)
- {
- if(threadIdx.x < offset)
- {
- sken[threadIdx.x] += sken[threadIdx.x + offset];
- }
-
- __syncthreads();
- }
-
- if(threadIdx.x == 0)
- {
- per_block_results[blockIdx.x] = sken[0];
- }
-}
-
-
-void dev_compute_kinetic_energy (reax_system *system, simulation_data *data, real *p_ekin)
-{
- real *spad = (real *) scratch;
- cuda_memset (spad, 0, sizeof (real) * 2 * system->n, "kinetic_energy");
-
- ker_kinetic_energy <<<BLOCKS, BLOCK_SIZE, sizeof (real) * BLOCK_SIZE >>>
- (spad, spad + system->n, system->n);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- k_reduction <<<1, BLOCKS_POW_2, sizeof (real) * BLOCKS_POW_2 >>>
- (spad + system->n, &((simulation_data *)data->d_simulation_data)->my_en.e_kin, BLOCKS);
- cudaThreadSynchronize ();
- cudaCheckError ();
-
- copy_host_device (p_ekin, &((simulation_data *)data->d_simulation_data)->my_en.e_kin,
- sizeof (real), cudaMemcpyDeviceToHost, "kinetic_energy");
-}
-*/
diff --git a/PG-PuReMD/src/ffield.c b/PG-PuReMD/src/ffield.c
index 443d9051..d985339b 100644
--- a/PG-PuReMD/src/ffield.c
+++ b/PG-PuReMD/src/ffield.c
@@ -20,7 +20,8 @@
----------------------------------------------------------------------*/
#include "reax_types.h"
- #if defined(PURE_REAX)
+
+#if defined(PURE_REAX)
#include "ffield.h"
#include "tool_box.h"
#elif defined(LAMMPS_REAX)
diff --git a/PG-PuReMD/src/ffield.h b/PG-PuReMD/src/ffield.h
index 9aa2a27f..313c3e67 100644
--- a/PG-PuReMD/src/ffield.h
+++ b/PG-PuReMD/src/ffield.h
@@ -24,6 +24,16 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
char Read_Force_Field( char*, reax_interaction*, control_params* );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/forces.c b/PG-PuReMD/src/forces.c
index c57527fe..19133fce 100644
--- a/PG-PuReMD/src/forces.c
+++ b/PG-PuReMD/src/forces.c
@@ -21,15 +21,6 @@
#include "reax_types.h"
-#include "index_utils.h"
-#ifdef HAVE_CUDA
- #include "cuda_forces.h"
- #include "cuda_lin_alg.h"
- #include "cuda_neighbors.h"
- #include "cuda_utils.h"
- #include "cuda_validation.h"
-#endif
-
#if defined(PURE_REAX)
#include "forces.h"
#include "bond_orders.h"
@@ -63,11 +54,7 @@
#include "reax_vector.h"
#endif
-
-#ifdef HAVE_CUDA
-void Cuda_Total_Forces( reax_system *, control_params *, simulation_data *, storage * );
-void Cuda_Total_Forces_PURE( reax_system *, storage * );
-#endif
+#include "index_utils.h"
interaction_function Interaction_Functions[NUM_INTRS];
@@ -221,41 +208,6 @@ void Compute_Total_Force( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-void Cuda_Compute_Total_Force( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace,
- reax_list **lists, mpi_datatypes *mpi_data )
-{
- rvec *f;
-
- f = (rvec *) host_scratch;
- memset( f, 0, sizeof(rvec) * system->N );
-
- Cuda_Total_Forces( system, control, data, workspace );
-
-#if defined(PURE_REAX)
- /* now all forces are computed to their partially-final values
- * based on the neighbors information each processor has had.
- * final values of force on each atom needs to be computed by adding up
- * all partially-final pieces */
-
- //MVAPICH2
- copy_host_device( f, dev_workspace->f, sizeof(rvec) * system->N ,
- cudaMemcpyDeviceToHost, "total_force:f:get" );
-
- Coll( system, mpi_data, f, mpi_data->mpi_rvec,
- sizeof(rvec) / sizeof(void), rvec_unpacker );
-
- copy_host_device( f, dev_workspace->f, sizeof(rvec) * system->N,
- cudaMemcpyHostToDevice, "total_force:f:put" );
-
- Cuda_Total_Forces_PURE( system, dev_workspace );
-#endif
-
-}
-#endif
-
-
// Essentially no-cuda copies of cuda kernels, to be used only in the mpi-not-gpu version
////////////////////////
// HBOND ISSUE
@@ -1851,173 +1803,6 @@ int Compute_Forces( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-int Cuda_Compute_Forces( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace, reax_list **lists,
- output_controls *out_control, mpi_datatypes *mpi_data )
-{
- int charge_flag, retVal;
-
-#if defined(LOG_PERFORMANCE)
- real t_start = 0;
-
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- t_start = Get_Time( );
- }
-#endif
-
- retVal = SUCCESS;
-
- /********* init forces ************/
- if ( control->charge_freq && (data->step - data->prev_steps) % control->charge_freq == 0 )
- {
- charge_flag = TRUE;
- }
- else
- {
- charge_flag = FALSE;
- }
-
- if ( charge_flag == TRUE )
- {
- retVal = Cuda_Init_Forces( system, control, data, workspace, lists, out_control );
-
-// int i;
-// static reax_list **temp_lists;
-//
-// if ( data->step == 0 )
-// {
-// temp_lists = (reax_list **) smalloc( LIST_N * sizeof (reax_list *), "temp_lists" );
-// for ( i = 0; i < LIST_N; ++i )
-// {
-// temp_lists[i] = (reax_list *) smalloc( sizeof(reax_list), "lists[i]" );
-// temp_lists[i]->allocated = FALSE;
-// }
-// Make_List( (*dev_lists + BONDS)->n, (*dev_lists + BONDS)->num_intrs,
-// TYP_BOND, *temp_lists + BONDS );
-// Make_List( (*dev_lists + HBONDS)->n, (*dev_lists + HBONDS)->num_intrs,
-// TYP_HBOND, *temp_lists + HBONDS );
-// }
-// else
-// {
-// Delete_List( *temp_lists + BONDS );
-// Make_List( (*dev_lists + BONDS)->n, (*dev_lists + BONDS)->num_intrs,
-// TYP_BOND, *temp_lists + BONDS );
-// Delete_List( *temp_lists + HBONDS );
-// Make_List( (*dev_lists + HBONDS)->n, (*dev_lists + HBONDS)->num_intrs,
-// TYP_HBOND, *temp_lists + HBONDS );
-//
-// }
-// Output_Sync_Lists( *temp_lists + BONDS, *dev_lists + BONDS, TYP_BOND );
-// Print_Bonds( system, temp_lists, control );
-// Output_Sync_Lists( *temp_lists + HBONDS, *dev_lists + HBONDS, TYP_HBOND );
-// Print_HBonds( system, temp_lists, control, data->step );
-// Print_HBond_Indices( system, temp_lists, control, data->step );
-// exit( 0 );
- }
- else
- {
- retVal = Cuda_Init_Forces_No_Charges( system, control, data, workspace, lists, out_control );
- }
-
- if ( retVal == SUCCESS )
- {
- //validate_sparse_matrix( system, workspace );
-
-#if defined(LOG_PERFORMANCE)
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &(data->timing.init_forces) );
- }
-#endif
-
- /********* bonded interactions ************/
- retVal = Cuda_Compute_Bonded_Forces( system, control, data, workspace, lists, out_control );
-
-#if defined(LOG_PERFORMANCE)
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &(data->timing.bonded) );
- }
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d: completed bonded\n",
- system->my_rank, data->step );
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
- }
-
- if ( retVal == SUCCESS )
- {
- /**************** charges ************************/
-#if defined(PURE_REAX)
- if ( charge_flag == TRUE )
- {
- Cuda_QEq( system, control, data, workspace, out_control, mpi_data );
- }
-
-#if defined(LOG_PERFORMANCE)
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &(data->timing.qEq) );
- }
-#endif
-
-#if defined(DEBUG_FOCUS)
- fprintf(stderr, "p%d @ step%d: qeq completed\n", system->my_rank, data->step);
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-#endif //PURE_REAX
-
- /********* nonbonded interactions ************/
- Cuda_Compute_NonBonded_Forces( system, control, data, workspace,
- lists, out_control, mpi_data );
-
-#if defined(LOG_PERFORMANCE)
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &(data->timing.nonb) );
- }
-#endif
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d: nonbonded forces completed\n",
- system->my_rank, data->step );
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-
- /*********** total force ***************/
- Cuda_Compute_Total_Force( system, control, data, workspace, lists, mpi_data );
-
-#if defined(LOG_PERFORMANCE)
- //MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &(data->timing.bonded) );
- }
-#endif
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d: total forces computed\n",
- system->my_rank, data->step );
- //Print_Total_Force( system, data, workspace );
- MPI_Barrier( MPI_COMM_WORLD );
-
-#endif
-
-// Print_Forces( system );
- }
-
- return retVal;
-}
-#endif
-
-
int validate_device( reax_system *system, simulation_data *data,
storage *workspace, reax_list **lists )
{
diff --git a/PG-PuReMD/src/forces.h b/PG-PuReMD/src/forces.h
index 6b4218e8..0579f092 100644
--- a/PG-PuReMD/src/forces.h
+++ b/PG-PuReMD/src/forces.h
@@ -28,6 +28,10 @@
extern interaction_function Interaction_Functions[NUM_INTRS];
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Init_Force_Functions( control_params* );
int Compute_Forces( reax_system*, control_params*, simulation_data*,
@@ -36,10 +40,11 @@ int Compute_Forces( reax_system*, control_params*, simulation_data*,
void Estimate_Storages( reax_system*, control_params*, reax_list**,
int*, int*, int*, int* );
-int Cuda_Compute_Forces( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls*, mpi_datatypes* );
-
int validate_device( reax_system *, simulation_data *, storage *, reax_list ** );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/geo_tools.c b/PG-PuReMD/src/geo_tools.c
index b97123a9..dff292e7 100644
--- a/PG-PuReMD/src/geo_tools.c
+++ b/PG-PuReMD/src/geo_tools.c
@@ -19,7 +19,10 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "geo_tools.h"
+
#include "allocate.h"
#include "box.h"
#include "tool_box.h"
diff --git a/PG-PuReMD/src/geo_tools.h b/PG-PuReMD/src/geo_tools.h
index 80786856..628e8f74 100644
--- a/PG-PuReMD/src/geo_tools.h
+++ b/PG-PuReMD/src/geo_tools.h
@@ -29,10 +29,6 @@
// CUSTOM ATOM: serial element name x y z
#define CUSTOM_ATOM_FORMAT " %d %s %s %lf %lf %lf"
-char Read_Geo( char*, reax_system*, control_params*,
- simulation_data*, storage*, mpi_datatypes* );
-
-
/*PDB format :
http://www.rcsb.org/pdb/file_formats/pdb/pdbguide2.2/guide2.2_frame.html
@@ -114,10 +110,23 @@ COLUMNS DATA TYPE FIELD DEFINITION
#define PDB_ATOM_FORMAT_O_LENGTH 81
#define PDB_CRYST1_FORMAT_O "%6s%9.3f%9.3f%9.3f%7.2f%7.2f%7.2f%11s%4d\n"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+char Read_Geo( char*, reax_system*, control_params*,
+ simulation_data*, storage*, mpi_datatypes* );
+
char Read_PDB( char*, reax_system*, control_params*,
- simulation_data*, storage*, mpi_datatypes* );
+ simulation_data*, storage*, mpi_datatypes* );
char Write_PDB( reax_system*, reax_list*, simulation_data*,
- control_params*, mpi_datatypes*, output_controls* );
+ control_params*, mpi_datatypes*, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/grid.c b/PG-PuReMD/src/grid.c
index 3714766c..d893f6c6 100644
--- a/PG-PuReMD/src/grid.c
+++ b/PG-PuReMD/src/grid.c
@@ -19,15 +19,17 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "grid.h"
+
#include "allocate.h"
+#include "index_utils.h"
#include "io_tools.h"
#include "reset_tools.h"
#include "tool_box.h"
#include "vector.h"
-#include "index_utils.h"
-
/* determines the exchange boundaries with nbrs in terms of gcells */
void Mark_GCells( reax_system* system, grid *g, ivec procs, MPI_Comm comm )
diff --git a/PG-PuReMD/src/grid.h b/PG-PuReMD/src/grid.h
index ad51e699..cb124da7 100644
--- a/PG-PuReMD/src/grid.h
+++ b/PG-PuReMD/src/grid.h
@@ -24,10 +24,24 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Setup_New_Grid( reax_system*, control_params*, MPI_Comm );
+
void Update_Grid( reax_system*, control_params*, MPI_Comm );
+
void Bin_My_Atoms( reax_system*, reallocate_data* );
+
void Reorder_My_Atoms( reax_system*, storage* );
+
void Bin_Boundary_Atoms( reax_system* );
+#ifdef __cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/hydrogen_bonds.c b/PG-PuReMD/src/hydrogen_bonds.c
index 5743feb5..dfd7abac 100644
--- a/PG-PuReMD/src/hydrogen_bonds.c
+++ b/PG-PuReMD/src/hydrogen_bonds.c
@@ -21,8 +21,6 @@
#include "reax_types.h"
-#include "index_utils.h"
-
#if defined(PURE_REAX)
#include "hydrogen_bonds.h"
#include "bond_orders.h"
@@ -37,6 +35,8 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
// DANIEL
// This function is taken straight from PuReMD, with minimal changes to accomodate the new datastructures
diff --git a/PG-PuReMD/src/hydrogen_bonds.h b/PG-PuReMD/src/hydrogen_bonds.h
index 346f0045..e4f58e10 100644
--- a/PG-PuReMD/src/hydrogen_bonds.h
+++ b/PG-PuReMD/src/hydrogen_bonds.h
@@ -24,7 +24,17 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Hydrogen_Bonds( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/init_md.c b/PG-PuReMD/src/init_md.c
index 595724bc..2e406d1a 100644
--- a/PG-PuReMD/src/init_md.c
+++ b/PG-PuReMD/src/init_md.c
@@ -23,17 +23,6 @@
#include <stddef.h>
-#ifdef HAVE_CUDA
- #include "cuda_allocate.h"
- #include "cuda_list.h"
- #include "cuda_copy.h"
- #include "cuda_forces.h"
- #include "cuda_init_md.h"
- #include "cuda_neighbors.h"
- #include "cuda_reset_tools.h"
- #include "cuda_validation.h"
-#endif
-
#if defined(PURE_REAX)
#include "init_md.h"
#include "allocate.h"
@@ -239,76 +228,6 @@ int Init_System( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-int Cuda_Init_System( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace,
- mpi_datatypes *mpi_data, char *msg )
-{
- int i, ret;
- reax_atom *atom;
- int nrecv[MAX_NBRS];
-
- Setup_New_Grid( system, control, MPI_COMM_WORLD );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d GRID:\n", system->my_rank );
- Print_Grid( &(system->my_grid), stderr );
-#endif
-
- Bin_My_Atoms( system, &(workspace->realloc) );
- Reorder_My_Atoms( system, workspace );
-
- /* estimate N and total capacity */
- for ( i = 0; i < MAX_NBRS; ++i )
- {
- nrecv[i] = 0;
- }
-
- MPI_Barrier( MPI_COMM_WORLD );
- system->max_recved = 0;
- system->N = SendRecv( system, mpi_data, mpi_data->boundary_atom_type, nrecv,
- Estimate_Boundary_Atoms, Unpack_Estimate_Message, TRUE );
- system->total_cap = MAX( (int)(system->N * SAFE_ZONE), MIN_CAP );
- Bin_Boundary_Atoms( system );
-
- /* Sync atoms here to continue the computation */
- dev_alloc_system( system );
- Sync_System( system );
-
- /* estimate numH and Hcap */
- Cuda_Reset_Atoms( system, control );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d: n=%d local_cap=%d\n",
- system->my_rank, system->n, system->local_cap );
- fprintf( stderr, "p%d: N=%d total_cap=%d\n",
- system->my_rank, system->N, system->total_cap );
- fprintf( stderr, "p%d: numH=%d H_cap=%d\n",
- system->my_rank, system->numH, system->Hcap );
-#endif
-
- Cuda_Compute_Total_Mass( system, data, mpi_data->comm_mesh3D );
-
- Cuda_Compute_Center_of_Mass( system, data, mpi_data, mpi_data->comm_mesh3D );
-
-// if( Reposition_Atoms( system, control, data, mpi_data, msg ) == FAILURE )
-// {
-// return FAILURE;
-// }
-
- /* 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 );
- }
-
- Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
-
- return SUCCESS;
-}
-#endif
-
-
/************************ initialize simulation data ************************/
void Init_Simulation_Data( reax_system *system, control_params *control,
simulation_data *data, char *msg )
@@ -411,102 +330,6 @@ void Init_Simulation_Data( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-void Cuda_Init_Simulation_Data( reax_system *system, control_params *control,
- simulation_data *data, char *msg )
-{
- dev_alloc_simulation_data( data );
-
- Reset_Simulation_Data( data );
-
- if ( !control->restart )
- {
- data->step = data->prev_steps = 0;
- }
-
- switch ( control->ensemble )
- {
- case NVE:
- data->N_f = 3 * system->bigN;
- Cuda_Evolve = Velocity_Verlet_NVE;
- control->virial = 0;
- break;
-
- case bNVT:
- data->N_f = 3 * system->bigN + 1;
- Cuda_Evolve = Cuda_Velocity_Verlet_Berendsen_NVT;
- control->virial = 0;
- break;
-
- case nhNVT:
- fprintf( stderr, "[WARNING] Nose-Hoover NVT is still under testing.\n" );
- data->N_f = 3 * system->bigN + 1;
- Cuda_Evolve = Velocity_Verlet_Nose_Hoover_NVT_Klein;
- control->virial = 0;
- if ( !control->restart || (control->restart && control->random_vel) )
- {
- data->therm.G_xi = control->Tau_T *
- (2.0 * data->sys_en.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;
- }
- break;
-
- case sNPT: /* Semi-Isotropic NPT */
- data->N_f = 3 * system->bigN + 4;
- Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
- control->virial = 1;
- if ( !control->restart )
- {
- Reset_Pressures( data );
- }
- break;
-
- case iNPT: /* Isotropic NPT */
- data->N_f = 3 * system->bigN + 2;
- Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
- control->virial = 1;
- if ( !control->restart )
- {
- Reset_Pressures( data );
- }
- break;
-
- case NPT: /* Anisotropic NPT */
- data->N_f = 3 * system->bigN + 9;
- Cuda_Evolve = Velocity_Verlet_Berendsen_NPT;
- control->virial = 1;
-
- fprintf( stderr, "p%d: init_simulation_data: option not yet implemented\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, INVALID_INPUT );
- break;
-
- default:
- fprintf( stderr, "p%d: init_simulation_data: ensemble not recognized\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, INVALID_INPUT );
- }
-
- /* initialize the timer(s) */
- MPI_Barrier( MPI_COMM_WORLD );
- if ( system->my_rank == MASTER_NODE )
- {
- data->timing.start = Get_Time( );
-
-#if defined(LOG_PERFORMANCE)
- Reset_Timing( &data->timing );
-#endif
- }
-
-#if defined(DEBUG)
- fprintf( stderr, "data->N_f: %8.3f\n", data->N_f );
-#endif
-}
-#endif
-
-
#elif defined(LAMMPS_REAX)
int Init_System( reax_system *system, char *msg )
{
@@ -603,22 +426,6 @@ void Init_Workspace( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-void Cuda_Init_Workspace( reax_system *system, control_params *control,
- storage *workspace, char *msg )
-{
- dev_alloc_workspace( system, control, dev_workspace,
- system->local_cap, system->total_cap, msg );
-
- memset( &(workspace->realloc), 0, sizeof(reallocate_data) );
- Cuda_Reset_Workspace( system, workspace );
-
- /* Initialize the Taper function */
- Init_Taper( control, dev_workspace );
-}
-#endif
-
-
/************** setup communication data structures **************/
int Init_MPI_Datatypes( reax_system *system, storage *workspace,
mpi_datatypes *mpi_data, char *msg )
@@ -885,88 +692,6 @@ int Init_Lists( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-int Cuda_Init_Lists( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace, reax_list **lists,
- mpi_datatypes *mpi_data, char *msg )
-{
- int ret;
- int Htop;
-
- /* ignore returned error, as system->d_max_far_nbrs was not valid */
- ret = Cuda_Estimate_Neighbors( system, data->step );
-
- Dev_Make_List( system->total_cap, system->total_far_nbrs,
- TYP_FAR_NEIGHBOR, *dev_lists + FAR_NBRS );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d: allocated far_nbrs: num_far=%d, space=%dMB\n",
- system->my_rank, system->total_far_nbrs,
- (int)(system->total_far_nbrs * sizeof(far_neighbor_data) / (1024 * 1024)) );
- fprintf( stderr, "N: %d and total_cap: %d \n", system->N, system->total_cap );
-#endif
-
- Cuda_Init_Neighbor_Indices( system );
-
- Cuda_Generate_Neighbor_Lists( system, data, workspace, dev_lists );
-
- /* estimate storage for bonds and hbonds */
- Cuda_Estimate_Storages( system, control, dev_lists, &(dev_workspace->H), data->step );
-
- /* estimate storage for charge sparse matrix */
-// Cuda_Estimate_Storage_Sparse_Matrix( system, control, data, dev_lists );
-
- dev_alloc_matrix( &(dev_workspace->H), system->total_cap, system->total_cm_entries );
-
- Cuda_Init_Sparse_Matrix_Indices( system, &(dev_workspace->H) );
-
- //MATRIX CHANGES
- //workspace->L = NULL;
- //workspace->U = NULL;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p:%d - allocated H matrix: max_entries: %d, cap: %d \n",
- system->my_rank, system->total_cm_entries, dev_workspace->H.m );
- fprintf( stderr, "p%d: allocated H matrix: Htop=%d, space=%dMB\n",
- system->my_rank, Htop,
- (int)(Htop * sizeof(sparse_matrix_entry) / (1024 * 1024)) );
-#endif
-
- if ( control->hbond_cut > 0.0 && system->numH > 0 )
- {
- Dev_Make_List( system->total_cap, system->total_hbonds, TYP_HBOND, *dev_lists + HBONDS );
-// Make_List( system->total_cap, system->total_hbonds, TYP_HBOND, *lists + HBONDS );
-
- Cuda_Init_HBond_Indices( system );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d: allocated hbonds: total_hbonds=%d, space=%dMB\n",
- system->my_rank, system->total_hbonds,
- (int)(system->total_hbonds * sizeof(hbond_data) / (1024 * 1024)) );
-#endif
- }
-
- /* bonds list */
- Dev_Make_List( system->total_cap, system->total_bonds, TYP_BOND, *dev_lists + BONDS );
-// Make_List( system->total_cap, system->total_bonds, TYP_BOND, *lists + BONDS );
-
- Cuda_Init_Bond_Indices( system );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d: allocated bonds: total_bonds=%d, space=%dMB\n",
- system->my_rank, total_bonds,
- (int)(total_bonds * sizeof(bond_data) / (1024 * 1024)) );
-#endif
-
- /* 3bodies list: since a more accurate estimate of the num.
- * of three body interactions requires that bond orders have
- * been computed, delay estimation until for computation */
-
- return SUCCESS;
-}
-#endif
-
-
#if defined(PURE_REAX)
void Initialize( reax_system *system, control_params *control,
simulation_data *data, storage *workspace,
@@ -1106,108 +831,6 @@ void Pure_Initialize( reax_system *system, control_params *control,
}
-#ifdef HAVE_CUDA
-void Cuda_Initialize( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace,
- reax_list **lists, output_controls *out_control,
- mpi_datatypes *mpi_data )
-{
- char msg[MAX_STR];
- real t_start, t_end;
-
- /* HOST/DEVICE SCRATCH */
- Cuda_Init_ScratchArea( );
-
- /* MPI_DATATYPES */
- if ( Init_MPI_Datatypes( system, workspace, mpi_data, msg ) == FAILURE )
- {
- fprintf( stderr, "p%d: init_mpi_datatypes: could not create datatypes\n",
- system->my_rank );
- fprintf( stderr, "p%d: mpi_data couldn't be initialized! terminating.\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
- }
-
- /* SYSTEM */
- if ( Cuda_Init_System( system, control, data, workspace, mpi_data, msg ) == FAILURE )
- {
- fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
- fprintf( stderr, "p%d: system could not be initialized! terminating.\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
- }
-
- /* GRID */
- dev_alloc_grid( system );
- Sync_Grid( &system->my_grid, &system->d_my_grid );
-
- //validate_grid( system );
-
- /* SIMULATION_DATA */
- Cuda_Init_Simulation_Data( system, control, data, msg );
-
- /* WORKSPACE */
- Cuda_Init_Workspace( system, control, workspace, msg );
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: initialized workspace\n", system->my_rank );
-#endif
-
- //Sync the taper here from host to device.
-
- /* CONTROL */
- dev_alloc_control( control );
-
- /* LISTS */
- if ( Cuda_Init_Lists( system, control, data, workspace, lists, mpi_data, msg ) ==
- FAILURE )
- {
- fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
- fprintf( stderr, "p%d: system could not be initialized! terminating.\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
- }
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: initialized lists\n", system->my_rank );
-#endif
-
- /* OUTPUT Files */
- if ( Init_Output_Files( system, control, out_control, mpi_data, msg ) == FAILURE )
- {
- fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
- fprintf( stderr, "p%d: could not open output files! terminating...\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
- }
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: output files opened\n", system->my_rank );
-#endif
-
- /* Lookup Tables */
- if ( control->tabulate )
- {
- if ( Init_Lookup_Tables( system, control, dev_workspace->Tap, mpi_data, msg ) == FAILURE )
- {
- fprintf( stderr, "p%d: %s\n", system->my_rank, msg );
- fprintf( stderr, "p%d: couldn't create lookup table! terminating.\n",
- system->my_rank );
- MPI_Abort( MPI_COMM_WORLD, CANNOT_INITIALIZE );
- }
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: initialized lookup tables\n", system->my_rank );
-#endif
- }
-
-#if defined(DEBUG)
- fprintf( stderr, "p%d: Device Initialization Done \n", system->my_rank );
-#endif
-}
-#endif
-
-
#elif defined(LAMMPS_REAX)
void Initialize( reax_system *system, control_params *control,
simulation_data *data, storage *workspace,
diff --git a/PG-PuReMD/src/init_md.h b/PG-PuReMD/src/init_md.h
index 5a66e4fb..c5222cbd 100644
--- a/PG-PuReMD/src/init_md.h
+++ b/PG-PuReMD/src/init_md.h
@@ -25,14 +25,25 @@
#include "reax_types.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void Generate_Initial_Velocities( reax_system *, real );
+
+int Init_MPI_Datatypes( reax_system *, storage *, mpi_datatypes *, char * );
+
void Initialize( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls*, mpi_datatypes* );
void Pure_Initialize( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls*, mpi_datatypes* );
-void Cuda_Initialize( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls*, mpi_datatypes* );
+void Init_Taper( control_params *, storage * );
+
+#ifdef __cplusplus
+}
+#endif
#endif
diff --git a/PG-PuReMD/src/integrate.c b/PG-PuReMD/src/integrate.c
index 88b406b5..b0200897 100644
--- a/PG-PuReMD/src/integrate.c
+++ b/PG-PuReMD/src/integrate.c
@@ -19,6 +19,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "integrate.h"
#include "allocate.h"
@@ -33,14 +35,6 @@
#include "tool_box.h"
#include "vector.h"
-#ifdef HAVE_CUDA
- #include "cuda_allocate.h"
- #include "cuda_integrate.h"
- #include "cuda_copy.h"
- #include "cuda_neighbors.h"
- #include "cuda_reset_tools.h"
-#endif
-
int Velocity_Verlet_NVE( reax_system* system, control_params* control,
simulation_data *data, storage *workspace, reax_list **lists,
@@ -339,143 +333,6 @@ int Velocity_Verlet_Berendsen_NVT( reax_system* system, control_params* control,
}
-#ifdef HAVE_CUDA
-int Cuda_Velocity_Verlet_Berendsen_NVT( reax_system* system, control_params* control,
- simulation_data *data, storage *workspace, reax_list **lists,
- output_controls *out_control, mpi_datatypes *mpi_data )
-{
- int i, steps, renbr, ret;
- static int verlet_part1_done = FALSE, estimate_nbrs_done = 0;
- real inv_m, dt, lambda;
- rvec dx;
- reax_atom *atom;
- int *bond_top, *hb_top;
- int Htop, num_3body;
- int total_hbonds, count, total_bonds;
- int bond_cap, cap_3body;
- real t_over_start, t_over_elapsed;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d\n", system->my_rank, data->step );
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-
- dt = control->dt;
- steps = data->step - data->prev_steps;
- renbr = steps % control->reneighbor == 0 ? TRUE : FALSE;
- ret = SUCCESS;
-
- Cuda_ReAllocate( system, control, data, workspace, lists, mpi_data );
-
- if ( verlet_part1_done == FALSE )
- {
- /* velocity verlet, 1st part */
- bNVT_update_velocity_part1( system, dt );
- verlet_part1_done = TRUE;
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d: verlet1 done\n", system->my_rank, data->step );
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-
- if ( renbr )
- {
- Update_Grid( system, control, mpi_data->world );
- }
-
- Output_Sync_Atoms( system );
- Comm_Atoms( system, control, data, workspace, lists, mpi_data, renbr );
- Sync_Atoms( system );
-
- /* synch the Grid to the Device here */
- Sync_Grid( &system->my_grid, &system->d_my_grid );
-
- init_blocks( system );
-
-#if defined(__CUDA_DEBUG_LOG__)
- fprintf( stderr, "p:%d - Matvec BLocks: %d, blocksize: %d \n",
- system->my_rank, MATVEC_BLOCKS, MATVEC_BLOCK_SIZE );
-#endif
- }
-
- Cuda_Reset( system, control, data, workspace, lists );
-
- if ( renbr )
- {
-#if defined(DEBUG)
- t_over_start = Get_Time ();
-#endif
-
- if ( estimate_nbrs_done == 0 )
- {
- //TODO: move far_nbrs reallocation checks outside of renbr frequency check
- ret = Cuda_Estimate_Neighbors( system, data->step );
- estimate_nbrs_done = 1;
- }
-
- if ( ret == SUCCESS && estimate_nbrs_done == 1 )
- {
- Cuda_Generate_Neighbor_Lists( system, data, workspace, lists );
- estimate_nbrs_done = 2;
-
-#if defined(DEBUG)
- t_over_elapsed = Get_Timing_Info( t_over_start );
- fprintf( stderr, "p%d --> Overhead (Step-%d) %f \n",
- system->my_rank, data->step, t_over_elapsed );
-#endif
- }
- }
-
- if ( ret == SUCCESS )
- {
- ret = Cuda_Compute_Forces( system, control, data, workspace,
- lists, out_control, mpi_data );
- }
-
- if ( ret == SUCCESS )
- {
- /* velocity verlet, 2nd part */
- bNVT_update_velocity_part2( system, dt );
-
-#if defined(DEBUG_FOCUS)
- fprintf(stderr, "p%d @ step%d: verlet2 done\n", system->my_rank, data->step);
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-
- /* temperature scaler */
- Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
-
- 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 );
-
- /* Scale velocities and positions at t+dt */
- bNVT_scale_velocities( system, lambda );
-
- Cuda_Compute_Kinetic_Energy( system, data, mpi_data->comm_mesh3D );
-
-#if defined(DEBUG_FOCUS)
- fprintf( stderr, "p%d @ step%d: scaled velocities\n",
- system->my_rank, data->step );
- MPI_Barrier( MPI_COMM_WORLD );
-#endif
-
- verlet_part1_done = FALSE;
- estimate_nbrs_done = 0;
- }
-
- return ret;
-}
-#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. */
diff --git a/PG-PuReMD/src/integrate.h b/PG-PuReMD/src/integrate.h
index 63fa9cbf..9a25c761 100644
--- a/PG-PuReMD/src/integrate.h
+++ b/PG-PuReMD/src/integrate.h
@@ -24,6 +24,11 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
int Velocity_Verlet_NVE( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls*, mpi_datatypes* );
@@ -49,9 +54,9 @@ int Velocity_Verlet_Flexible_NPT( reax_system*, control_params*,
output_controls*, mpi_datatypes* );
*/
-//CUDA SPECIFIC FUNCTIONS
-int Cuda_Velocity_Verlet_Berendsen_NVT( reax_system*, control_params*,
- simulation_data*, storage*, reax_list**, output_controls*,
- mpi_datatypes* );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/io_tools.c b/PG-PuReMD/src/io_tools.c
index c7c0f2fe..131f8a2e 100644
--- a/PG-PuReMD/src/io_tools.c
+++ b/PG-PuReMD/src/io_tools.c
@@ -20,7 +20,7 @@
----------------------------------------------------------------------*/
#include "reax_types.h"
-#include "index_utils.h"
+
#if defined(PURE_REAX)
#include "io_tools.h"
#include "basic_comm.h"
@@ -41,6 +41,8 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
print_interaction Print_Interactions[NUM_INTRS];
diff --git a/PG-PuReMD/src/io_tools.h b/PG-PuReMD/src/io_tools.h
index 6ae2d6d8..f83c9686 100644
--- a/PG-PuReMD/src/io_tools.h
+++ b/PG-PuReMD/src/io_tools.h
@@ -25,45 +25,71 @@
#include "reax_types.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+
int Init_Output_Files( reax_system*, control_params*,
- output_controls*, mpi_datatypes*, char* );
+ output_controls*, mpi_datatypes*, char* );
+
int Close_Output_Files( reax_system*, control_params*,
- output_controls*, mpi_datatypes* );
-
-void Print_Box( simulation_box*, char*, FILE* );
-
-void Print_Grid( grid*, FILE* );
-void Print_GCell_Exchange_Bounds( int, neighbor_proc* );
-void Print_Native_GCells( reax_system* );
-void Print_All_GCells( reax_system*);
-
-void Print_Init_Atoms( reax_system*, storage* );
-void Print_My_Atoms( reax_system* );
-void Print_My_Ext_Atoms( reax_system* );
-
-void Print_Far_Neighbors( reax_system*, reax_list**, control_params *);
-void Print_Sparse_Matrix( reax_system*, sparse_matrix* );
-void Print_Sparse_Matrix2( reax_system*, sparse_matrix*, char* );
-void Print_Linear_System( reax_system*, control_params*, storage*, int );
-void Print_LinSys_Soln( reax_system*, real*, real*, real* );
-void Print_Charges( reax_system* );
-void Print_HBonds( reax_system*, reax_list**, control_params *, int );
-void Print_HBond_Indices( reax_system*, reax_list**, control_params *, int );
-void Print_Bonds( reax_system*, reax_list**, control_params *);
-void Print_Bond_List2( reax_system*, reax_list*, char* );
-void Print_Total_Force( reax_system*, simulation_data*, storage* );
-void Output_Results( reax_system*, control_params*, simulation_data*,
- reax_list**, output_controls*, mpi_datatypes* );
+ output_controls*, mpi_datatypes* );
+
+void Print_Box( simulation_box*, char*, FILE* );
+
+void Print_Grid( grid*, FILE* );
+
+void Print_GCell_Exchange_Bounds( int, neighbor_proc* );
+
+void Print_Native_GCells( reax_system* );
+
+void Print_All_GCells( reax_system*);
+
+void Print_Init_Atoms( reax_system*, storage* );
+
+void Print_My_Atoms( reax_system* );
+
+void Print_My_Ext_Atoms( reax_system* );
+
+void Print_Far_Neighbors( reax_system*, reax_list**, control_params *);
+
+void Print_Sparse_Matrix( reax_system*, sparse_matrix* );
+
+void Print_Sparse_Matrix2( reax_system*, sparse_matrix*, char* );
+
+void Print_Linear_System( reax_system*, control_params*, storage*, int );
+
+void Print_LinSys_Soln( reax_system*, real*, real*, real* );
+
+void Print_Charges( reax_system* );
+
+void Print_HBonds( reax_system*, reax_list**, control_params *, int );
+
+void Print_HBond_Indices( reax_system*, reax_list**, control_params *, int );
+
+void Print_Bonds( reax_system*, reax_list**, control_params *);
+
+void Print_Bond_List2( reax_system*, reax_list*, char* );
+
+void Print_Total_Force( reax_system*, simulation_data*, storage* );
+
+void Output_Results( reax_system*, control_params*, simulation_data*,
+ reax_list**, output_controls*, mpi_datatypes* );
#if defined(DEBUG_FOCUS) || defined(TEST_FORCES) || defined(TEST_ENERGY)
void Debug_Marker_Bonded( output_controls*, int );
+
void Debug_Marker_Nonbonded( output_controls*, int );
-void Print_Near_Neighbors_List( reax_system*, reax_list**, control_params*,
- simulation_data*, output_controls* );
-void Print_Far_Neighbors_List( reax_system*, reax_list**, control_params*,
- simulation_data*, output_controls* );
-void Print_Bond_List( reax_system*, control_params*, simulation_data*,
- reax_list**, output_controls* );
+
+void Print_Near_Neighbors_List( reax_system*, reax_list**, control_params*,
+ simulation_data*, output_controls* );
+
+void Print_Far_Neighbors_List( reax_system*, reax_list**, control_params*,
+ simulation_data*, output_controls* );
+
+void Print_Bond_List( reax_system*, control_params*, simulation_data*,
+ reax_list**, output_controls* );
+
/*void Dummy_Printer( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls* );
void Print_Bond_Orders( reax_system*, control_params*, simulation_data*,
@@ -89,23 +115,28 @@ void Print_Total_Force( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls* );
void Compare_Total_Forces( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls* );*/
+
//void Print_Total_Force( reax_system*, control_params* );
+
void Print_Force_Files( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls*,
- mpi_datatypes * );
+ storage*, reax_list**, output_controls*, mpi_datatypes * );
+
//void Init_Force_Test_Functions( );
int fn_qsort_intcmp( const void *, const void * );
void Print_Far_Neighbors_List( reax_system*, reax_list**, control_params*,
- simulation_data*, output_controls* );
+ simulation_data*, output_controls* );
void Print_Near_Neighbors_List( reax_system*, reax_list**, control_params*,
- simulation_data*, output_controls* );
+ simulation_data*, output_controls* );
void Print_Bond_List( reax_system*, control_params*, simulation_data*,
- reax_list**, output_controls*);
+ reax_list**, output_controls*);
+#endif
+#ifdef __cplusplus
+}
#endif
diff --git a/PG-PuReMD/src/lin_alg.c b/PG-PuReMD/src/lin_alg.c
index bac272a0..e9ce62e7 100644
--- a/PG-PuReMD/src/lin_alg.c
+++ b/PG-PuReMD/src/lin_alg.c
@@ -19,6 +19,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "lin_alg.h"
#include "basic_comm.h"
@@ -27,9 +29,7 @@
#include "vector.h"
#ifdef HAVE_CUDA
- #include "cuda_lin_alg.h"
- #include "cuda_utils.h"
- #include "cuda_validation.h"
+ #include "cuda/cuda_validation.h"
#endif
#if defined(CG_PERFORMANCE)
@@ -100,13 +100,13 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H, rvec2
#endif
#ifdef HAVE_CUDA
- check_zeros_host (x, system->N, "x");
+ check_zeros_host( x, system->N, "x" );
#endif
Dist( system, mpi_data, x, mpi_data->mpi_rvec2, scale, rvec2_packer );
#ifdef HAVE_CUDA
- check_zeros_host (x, system->N, "x");
+ check_zeros_host( x, system->N, "x" );
#endif
dual_Sparse_MatVec( H, x, workspace->q2, N );
@@ -285,352 +285,6 @@ int dual_CG( reax_system *system, storage *workspace, sparse_matrix *H, rvec2
}
-#ifdef HAVE_CUDA
-int Cuda_dual_CG( reax_system *system, storage *workspace, sparse_matrix *H,
- rvec2 *b, real tol, rvec2 *x, mpi_datatypes* mpi_data, FILE *fout,
- simulation_data *data )
-{
- int i, j, n, N, matvecs, scale;
- rvec2 tmp, alpha, beta;
- rvec2 my_sum, norm_sqr, b_norm, my_dot;
- rvec2 sig_old, sig_new;
- MPI_Comm comm;
- rvec2 *spad = (rvec2 *) host_scratch;
- int a;
-
- n = system->n;
- N = system->N;
- comm = mpi_data->world;
- matvecs = 0;
- scale = sizeof(rvec2) / sizeof(void);
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- matvecs = 0;
- t_start = matvec_time = dot_time = 0;
- t_start = Get_Time( );
- }
-#endif
-
- //MVAPICH2
-//#ifdef __CUDA_DEBUG__
-// Dist( system, mpi_data, workspace->x, mpi_data->mpi_rvec2, scale, rvec2_packer );
-//#endif
-
-// check_zeros_device( x, system->N, "x" );
-
- copy_host_device( spad, x, sizeof(rvec2) * system->total_cap, cudaMemcpyDeviceToHost, "CG:x:get" );
- Dist( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_packer );
- copy_host_device( spad, x, sizeof(rvec2) * system->total_cap, cudaMemcpyHostToDevice, "CG:x:put" );
-
-// check_zeros_device( x, system->N, "x" );
-
-// compare_rvec2 (workspace->x, x, N, "x");
-// if (data->step > 0) {
-// compare_rvec2 (workspace->b, dev_workspace->b, system->N, "b");
-// compare_rvec2 (workspace->x, dev_workspace->x, system->N, "x");
-//
-// exit (0);
-// }
-
-
-//#ifdef __CUDA_DEBUG__
-// dual_Sparse_MatVec( &workspace->H, workspace->x, workspace->q2, N );
-//#endif
- //originally we were using only H->n which was system->n (init_md.c)
- //Cuda_Dual_Matvec ( H, x, dev_workspace->q2, H->n, system->total_cap);
-
- Cuda_Dual_Matvec ( H, x, dev_workspace->q2, system->N, system->total_cap);
-
-// compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
-
-// if (data->step > 0) exit (0);
-
- // tryQEq
- //MVAPICH2
-//#ifdef __CUDA_DEBUG__
-// Coll(system,mpi_data,workspace->q2,mpi_data->mpi_rvec2,scale,rvec2_unpacker);
-//#endif
-
- copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyDeviceToHost, "CG:q2:get" );
- Coll(system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_unpacker);
- copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyHostToDevice,"CG:q2:put" );
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &matvec_time );
- }
-#endif
-
-//#ifdef __CUDA_DEBUG__
-// for( j = 0; j < system->n; ++j ) {
-// // residual
-// workspace->r2[j][0] = workspace->b[j][0] - workspace->q2[j][0];
-// workspace->r2[j][1] = workspace->b[j][1] - workspace->q2[j][1];
-// // apply diagonal pre-conditioner
-// workspace->d2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
-// workspace->d2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
-// }
-//#endif
-
- Cuda_CG_Diagonal_Preconditioner( dev_workspace, b, system->n );
-
-// compare_rvec2 (workspace->r2, dev_workspace->r2, n, "r2");
-// compare_rvec2 (workspace->d2, dev_workspace->d2, n, "d2");
-
- /* norm of b */
-//#ifdef __CUDA_DEBUG__
-// my_sum[0] = my_sum[1] = 0;
-// for( j = 0; j < n; ++j ) {
-// my_sum[0] += SQR( workspace->b[j][0] );
-// my_sum[1] += SQR( workspace->b[j][1] );
-// }
-// fprintf (stderr, "cg: my_sum[ %f, %f] \n", my_sum[0], my_sum[1]);
-//#endif
-
- my_sum[0] = my_sum[1] = 0;
- Cuda_Norm (b, n, my_sum);
-
-// fprintf (stderr, "cg: my_sum[ %f, %f] \n", my_sum[0], my_sum[1]);
-
- MPI_Allreduce( &my_sum, &norm_sqr, 2, MPI_DOUBLE, MPI_SUM, comm );
- b_norm[0] = SQRT( norm_sqr[0] );
- b_norm[1] = SQRT( norm_sqr[1] );
- //fprintf( stderr, "bnorm = %f %f\n", b_norm[0], b_norm[1] );
-
- /* dot product: r.d */
-//#ifdef __CUDA_DEBUG__
-// my_dot[0] = my_dot[1] = 0;
-// for( j = 0; j < n; ++j ) {
-// my_dot[0] += workspace->r2[j][0] * workspace->d2[j][0];
-// my_dot[1] += workspace->r2[j][1] * workspace->d2[j][1];
-// }
-// fprintf( stderr, "my_dot: %f %f\n", my_dot[0], my_dot[1] );
-//#endif
-
- my_dot[0] = my_dot[1] = 0;
- Cuda_Dot (dev_workspace->r2, dev_workspace->d2, my_dot, n);
-
-// fprintf( stderr, "my_dot: %f %f\n", my_dot[0], my_dot[1] );
-
- MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
-
- //fprintf( stderr, "DEVICE: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
-
- for ( i = 1; i < 300; ++i )
- {
- //MVAPICH2
-//#ifdef __CUDA_DEBUG__
-// Dist(system,mpi_data,workspace->d2,mpi_data->mpi_rvec2,scale,rvec2_packer);
-//#endif
-
- copy_host_device( spad, dev_workspace->d2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyDeviceToHost, "cg:d2:get" );
- Dist( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_packer );
- copy_host_device( spad, dev_workspace->d2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyHostToDevice, "cg:d2:put" );
-
- //print_device_rvec2 (dev_workspace->d2, N);
-
-//#ifdef __CUDA_DEBUG__
-// dual_Sparse_MatVec( &workspace->H, workspace->d2, workspace->q2, N );
-//#endif
-
- Cuda_Dual_Matvec( H, dev_workspace->d2, dev_workspace->q2, system->N,
- system->total_cap );
-
- /*
- fprintf (stderr, "******************* Device sparse Matrix--------> %d \n", H->n );
- fprintf (stderr, " ******* HOST SPARSE MATRIX ******** \n");
- print_sparse_matrix_host (&workspace->H);
- fprintf (stderr, " ******* HOST Vector ***************\n");
- print_host_rvec2 (workspace->d2, system->N);
- fprintf (stderr, " ******* Device SPARSE MATRIX ******** \n");
- print_sparse_matrix (&dev_workspace->H);
- fprintf (stderr, " ******* Device Vector ***************\n");
- print_device_rvec2 (dev_workspace->d2, system->N);
- */
- //compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
-
- // tryQEq
- // MVAPICH2
-//#ifdef __CUDA_DEBUG__
-// Coll(system,mpi_data,workspace->q2,mpi_data->mpi_rvec2,scale,rvec2_unpacker);
-//#endif
-
- copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyDeviceToHost, "cg:q2:get" );
- Coll( system, mpi_data, spad, mpi_data->mpi_rvec2, scale, rvec2_unpacker );
- copy_host_device( spad, dev_workspace->q2, sizeof(rvec2) * system->total_cap,
- cudaMemcpyHostToDevice, "cg:q2:put" );
-
-// compare_rvec2 (workspace->q2, dev_workspace->q2, N, "q2");
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &matvec_time );
- }
-#endif
-
- /* dot product: d.q */
-//#ifdef __CUDA_DEBUG__
-// my_dot[0] = my_dot[1] = 0;
-// for( j = 0; j < n; ++j ) {
-// my_dot[0] += workspace->d2[j][0] * workspace->q2[j][0];
-// my_dot[1] += workspace->d2[j][1] * workspace->q2[j][1];
-// }
-// fprintf( stderr, "H:my_dot: %f %f\n", my_dot[0], my_dot[1] );
-//#endif
-
- my_dot[0] = my_dot[1] = 0;
- Cuda_Dot (dev_workspace->d2, dev_workspace->q2, my_dot, n);
- //fprintf( stderr, "D:my_dot: %f %f\n", my_dot[0], my_dot[1] );
-
- MPI_Allreduce( &my_dot, &tmp, 2, MPI_DOUBLE, MPI_SUM, comm );
- //fprintf( stderr, "tmp: %f %f\n", tmp[0], tmp[1] );
-
- alpha[0] = sig_new[0] / tmp[0];
- alpha[1] = sig_new[1] / tmp[1];
- my_dot[0] = my_dot[1] = 0;
-
-//#ifdef __CUDA_DEBUG__
-// for( j = 0; j < system->n; ++j ) {
-// // update x
-// workspace->x[j][0] += alpha[0] * workspace->d2[j][0];
-// workspace->x[j][1] += alpha[1] * workspace->d2[j][1];
-// // update residual
-// workspace->r2[j][0] -= alpha[0] * workspace->q2[j][0];
-// workspace->r2[j][1] -= alpha[1] * workspace->q2[j][1];
-// // apply diagonal pre-conditioner
-// workspace->p2[j][0] = workspace->r2[j][0] * workspace->Hdia_inv[j];
-// workspace->p2[j][1] = workspace->r2[j][1] * workspace->Hdia_inv[j];
-// // dot product: r.p
-// my_dot[0] += workspace->r2[j][0] * workspace->p2[j][0];
-// my_dot[1] += workspace->r2[j][1] * workspace->p2[j][1];
-// }
-// fprintf( stderr, "H:my_dot: %f %f\n", my_dot[0], my_dot[1] );
-//#endif
-
- my_dot[0] = my_dot[1] = 0;
- Cuda_DualCG_Preconditioner( dev_workspace, x, alpha, system->n, my_dot );
-
- //fprintf( stderr, "D:my_dot: %f %f\n", my_dot[0], my_dot[1] );
-
-// compare_rvec2 (workspace->x, dev_workspace->x, N, "x");
-// compare_rvec2 (workspace->r2, dev_workspace->r2, N, "r2");
-// compare_rvec2 (workspace->p2, dev_workspace->p2, N, "p2");
-
- sig_old[0] = sig_new[0];
- sig_old[1] = sig_new[1];
- MPI_Allreduce( &my_dot, &sig_new, 2, MPI_DOUBLE, MPI_SUM, comm );
-
- //fprintf( stderr, "DEVICE: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];
-
-//#ifdef __CUDA_DEBUG__
-// for( j = 0; j < system->n; ++j ) {
-// // d = p + beta * d
-// workspace->d2[j][0] = workspace->p2[j][0] + beta[0] * workspace->d2[j][0];
-// workspace->d2[j][1] = workspace->p2[j][1] + beta[1] * workspace->d2[j][1];
-// }
-//#endif
-
- Cuda_Vector_Sum_Rvec2( dev_workspace->d2, dev_workspace->p2, beta,
- dev_workspace->d2, system->n );
-
-// compare_rvec2 (workspace->d2, dev_workspace->d2, N, "q2");
- }
-
-
- if ( SQRT(sig_new[0]) / b_norm[0] <= tol )
- {
- //for( j = 0; j < n; ++j )
- // workspace->t[j] = workspace->x[j][1];
- //fprintf (stderr, "Getting started with Cuda_CG1 \n");
-
- Cuda_RvecCopy_From( dev_workspace->t, dev_workspace->x, 1, system->n );
-
- //compare_array (workspace->b_t, dev_workspace->b_t, system->n, "b_t");
- //compare_array (workspace->t, dev_workspace->t, system->n, "t");
-
- matvecs = Cuda_CG( system, workspace, H, dev_workspace->b_t, tol, dev_workspace->t,
- mpi_data, fout );
-
- //fprintf (stderr, " Cuda_CG1: iterations --> %d \n", matvecs );
- //for( j = 0; j < n; ++j )
- // workspace->x[j][1] = workspace->t[j];
-
- Cuda_RvecCopy_To( dev_workspace->x, dev_workspace->t, 1, system->n );
- }
- else if ( SQRT(sig_new[1]) / b_norm[1] <= tol )
- {
- //for( j = 0; j < n; ++j )
- // workspace->s[j] = workspace->x[j][0];
-
- Cuda_RvecCopy_From( dev_workspace->s, dev_workspace->x, 0, system->n );
-
- //compare_array (workspace->s, dev_workspace->s, system->n, "s");
- //compare_array (workspace->b_s, dev_workspace->b_s, system->n, "b_s");
-
- //fprintf (stderr, "Getting started with Cuda_CG2 \n");
-
- matvecs = Cuda_CG( system, workspace, H, dev_workspace->b_s, tol, dev_workspace->s,
- mpi_data, fout );
-
- //fprintf (stderr, " Cuda_CG2: iterations --> %d \n", matvecs );
- //for( j = 0; j < system->n; ++j )
- // workspace->x[j][0] = workspace->s[j];
-
- Cuda_RvecCopy_To( dev_workspace->x, dev_workspace->s, 0, system->n );
- }
-
- if ( i >= 300 )
- {
- fprintf( stderr, "[WARNING] p%d: dual CG convergence failed! (%d steps)\n",
- system->my_rank, i );
- fprintf( stderr, " [INFO] s lin solve error: %f\n", SQRT(sig_new[0]) / b_norm[0] );
- fprintf( stderr, " [INFO] t lin solve error: %f\n", SQRT(sig_new[1]) / b_norm[1] );
- }
-
-#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 );
- }
-#endif
-
- return (i + 1) + matvecs;
-}
-#endif
-
-
void Sparse_MatVec( sparse_matrix *A, real *x, real *b, int N )
{
int i, j, k, si;
@@ -745,153 +399,6 @@ int CG( reax_system *system, storage *workspace, sparse_matrix *H, real *b,
}
-#ifdef HAVE_CUDA
-int Cuda_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;
- real sig_old, sig_new, sig0;
- real *spad = (real *) host_scratch;
-
- scale = sizeof(real) / sizeof(void);
-
- /* x is on the device */
- //MVAPICH2
- memset( spad, 0, sizeof(real) * system->total_cap );
- copy_host_device( spad, x, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:x:get" );
- Dist( system, mpi_data, spad, MPI_DOUBLE, scale, real_packer );
-
- //MVAPICH2
- copy_host_device( spad, x, sizeof(real) * system->total_cap,
- cudaMemcpyHostToDevice, "cuda_cg:x:put" );
- Cuda_Matvec( H, x, dev_workspace->q, system->N, system->total_cap );
-
- // tryQEq
- // MVAPICH2
- copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
- Coll( system, mpi_data, spad, MPI_DOUBLE, scale, real_unpacker );
-
- //MVAPICH2
- copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
- cudaMemcpyHostToDevice, "cuda_cg:q:put" );
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &matvec_time );
- }
-#endif
-
- Cuda_Vector_Sum( dev_workspace->r , 1., b, -1., dev_workspace->q,
- system->n );
- //for( j = 0; j < system->n; ++j )
- // workspace->d[j] = workspace->r[j] * workspace->Hdia_inv[j]; //pre-condition
- Cuda_CG_Preconditioner( dev_workspace->d, dev_workspace->r,
- dev_workspace->Hdia_inv, system->n );
-
- //TODO do the parallel_norm on the device for the local sum
- copy_host_device( spad, b, sizeof(real) * system->n,
- cudaMemcpyDeviceToHost, "cuda_cg:b:get" );
- b_norm = Parallel_Norm( spad, system->n, mpi_data->world );
-
- //TODO do the parallel dot on the device for the local sum
- copy_host_device( spad, dev_workspace->r, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:r:get" );
- copy_host_device( spad + system->total_cap, dev_workspace->d, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
- sig_new = Parallel_Dot( spad, spad + system->total_cap, 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 )
- {
- //MVAPICH2
- copy_host_device( spad, dev_workspace->d, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
- Dist( system, mpi_data, spad, MPI_DOUBLE, scale, real_packer );
- copy_host_device( spad, dev_workspace->d, sizeof(real) * system->total_cap,
- cudaMemcpyHostToDevice, "cuda_cg:d:put" );
-
- Cuda_Matvec( H, dev_workspace->d, dev_workspace->q, system->N, system->total_cap );
-
- //tryQEq
- copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
- cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
- Coll( system, mpi_data, spad, MPI_DOUBLE, scale, real_unpacker );
- copy_host_device( spad, dev_workspace->q, sizeof(real) * system->total_cap,
- cudaMemcpyHostToDevice, "cuda_cg:q:get" );
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &matvec_time );
- }
-#endif
-
- //TODO do the parallel dot on the device for the local sum
- copy_host_device( spad, dev_workspace->d, sizeof(real) * system->n,
- cudaMemcpyDeviceToHost, "cuda_cg:d:get" );
- copy_host_device( spad + system->n, dev_workspace->q, sizeof(real) * system->n,
- cudaMemcpyDeviceToHost, "cuda_cg:q:get" );
- tmp = Parallel_Dot( spad, spad + system->n, system->n, mpi_data->world );
-
- alpha = sig_new / tmp;
- //Cuda_Vector_Add( x, alpha, dev_workspace->d, system->n );
- Cuda_Vector_Sum( x, alpha, dev_workspace->d, 1.0, x, system->n );
-
- //Cuda_Vector_Add( workspace->r, -alpha, workspace->q, system->n );
- Cuda_Vector_Sum( dev_workspace->r, -alpha, dev_workspace->q, 1.0,
- dev_workspace->r, system->n );
- /* pre-conditioning */
- //for( j = 0; j < system->n; ++j )
- // workspace->p[j] = workspace->r[j] * workspace->Hdia_inv[j];
- Cuda_CG_Preconditioner( dev_workspace->p, dev_workspace->r,
- dev_workspace->Hdia_inv, system->n );
-
- sig_old = sig_new;
-
- //TODO do the parallel dot on the device for the local sum
- copy_host_device( spad, dev_workspace->r, sizeof(real) * system->n,
- cudaMemcpyDeviceToHost, "cuda_cg:r:get" );
- copy_host_device( spad + system->n, dev_workspace->p, sizeof(real) * system->n,
- cudaMemcpyDeviceToHost, "cuda_cg:p:get" );
- sig_new = Parallel_Dot( spad , spad + system->n, system->n, mpi_data->world );
- //fprintf (stderr, "Device: sig_new: %f \n", sig_new );
-
- beta = sig_new / sig_old;
- Cuda_Vector_Sum( dev_workspace->d, 1., dev_workspace->p, beta,
- dev_workspace->d, system->n );
-
-#if defined(CG_PERFORMANCE)
- if ( system->my_rank == MASTER_NODE )
- {
- Update_Timing_Info( &t_start, &dot_time );
- }
-#endif
- }
-
- if ( i >= 300 )
- {
- fprintf( stderr, "CG convergence failed!\n" );
- return i;
- }
-
- return i;
-}
-#endif
-
-
int CG_test( reax_system *system, storage *workspace, sparse_matrix *H, real
*b, real tol, real *x, mpi_datatypes* mpi_data, FILE *fout )
{
diff --git a/PG-PuReMD/src/lin_alg.h b/PG-PuReMD/src/lin_alg.h
index f401fb2d..3663978e 100644
--- a/PG-PuReMD/src/lin_alg.h
+++ b/PG-PuReMD/src/lin_alg.h
@@ -24,23 +24,32 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
int GMRES( reax_system*, storage*, sparse_matrix*,
- real*, real, real*, mpi_datatypes*, FILE* );
+ real*, real, real*, mpi_datatypes*, FILE* );
+
int GMRES_HouseHolder( reax_system*, storage*, sparse_matrix*,
- real*, real, real*, mpi_datatypes*, FILE* );
+ real*, real, real*, mpi_datatypes*, FILE* );
+
int dual_CG( reax_system*, storage*, sparse_matrix*,
- rvec2*, real, rvec2*, mpi_datatypes*, FILE* , simulation_data *);
+ rvec2*, real, rvec2*, mpi_datatypes*, FILE* , simulation_data *);
+
int CG( reax_system*, storage*, sparse_matrix*,
real*, real, real*, mpi_datatypes*, FILE* );
+
int PCG( reax_system*, storage*, sparse_matrix*, real*, real,
- sparse_matrix*, sparse_matrix*, real*, mpi_datatypes*, FILE* );
+ sparse_matrix*, sparse_matrix*, real*, mpi_datatypes*, FILE* );
+
int sCG( reax_system*, storage*, sparse_matrix*,
- real*, real, real*, mpi_datatypes*, FILE* );
+ real*, real, real*, mpi_datatypes*, FILE* );
+
+#ifdef __cplusplus
+}
+#endif
-//CUDA Functions
-int Cuda_dual_CG( reax_system*, storage*, sparse_matrix*,
- rvec2*, real, rvec2*, mpi_datatypes*, FILE* , simulation_data *);
-int Cuda_CG( reax_system*, storage*, sparse_matrix*,
- real*, real, real*, mpi_datatypes*, FILE* );
#endif
diff --git a/PG-PuReMD/src/list.c b/PG-PuReMD/src/list.c
index 05213cb3..69736afb 100644
--- a/PG-PuReMD/src/list.c
+++ b/PG-PuReMD/src/list.c
@@ -22,11 +22,11 @@
#include "reax_types.h"
#if defined(PURE_REAX)
-#include "list.h"
-#include "tool_box.h"
+ #include "list.h"
+ #include "tool_box.h"
#elif defined(LAMMPS_REAX)
-#include "reax_list.h"
-#include "reax_tool_box.h"
+ #include "reax_list.h"
+ #include "reax_tool_box.h"
#endif
diff --git a/PG-PuReMD/src/list.h b/PG-PuReMD/src/list.h
index 1f29f5f8..df6ec82f 100644
--- a/PG-PuReMD/src/list.h
+++ b/PG-PuReMD/src/list.h
@@ -24,17 +24,21 @@
#include "reax_types.h"
+
#ifdef _cplusplus
extern "C" {
#endif
-
void Print_List( reax_list* );
void Make_List( int, int, int, reax_list* );
void Delete_List( reax_list* );
+#ifdef _cplusplus
+}
+#endif
+
#if defined(LAMMPS_REAX) || defined(PURE_REAX)
static inline int Num_Entries( int i, reax_list *l )
{
@@ -60,12 +64,7 @@ static inline void Set_End_Index( int i, int val, reax_list *l )
{
l->end_index[i] = val;
}
-
#endif
-#ifdef _cplusplus
-}
-#endif
-
#endif
diff --git a/PG-PuReMD/src/lookup.c b/PG-PuReMD/src/lookup.c
index 2c6652f9..b071ea89 100644
--- a/PG-PuReMD/src/lookup.c
+++ b/PG-PuReMD/src/lookup.c
@@ -21,12 +21,6 @@
#include "reax_types.h"
-#include "index_utils.h"
-
-#ifdef HAVE_CUDA
- #include "cuda_lookup.h"
-#endif
-
#if defined(PURE_REAX)
#include "lookup.h"
#include "nonbonded.h"
@@ -37,6 +31,12 @@
#include "reax_tool_box.h"
#endif
+#include "index_utils.h"
+
+#ifdef HAVE_CUDA
+ #include "cuda/cuda_lookup.h"
+#endif
+
/* Fills solution into x. Warning: will modify c and d! */
void Tridiagonal_Solve( const real *a, const real *b,
diff --git a/PG-PuReMD/src/lookup.h b/PG-PuReMD/src/lookup.h
index f6e45bd1..4db34ce0 100644
--- a/PG-PuReMD/src/lookup.h
+++ b/PG-PuReMD/src/lookup.h
@@ -26,7 +26,17 @@
//extern LR_lookup_table **LR;
+
+#ifdef _cplusplus
+extern "C" {
+#endif
+
int Init_Lookup_Tables( reax_system*, control_params*, real *,
- mpi_datatypes*, char* );
+ mpi_datatypes*, char* );
+
+#ifdef _cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/multi_body.c b/PG-PuReMD/src/multi_body.c
index b480d3bb..aab4957d 100644
--- a/PG-PuReMD/src/multi_body.c
+++ b/PG-PuReMD/src/multi_body.c
@@ -32,6 +32,7 @@
#include "reax_list.h"
#include "reax_vector.h"
#endif
+
#include "index_utils.h"
diff --git a/PG-PuReMD/src/multi_body.h b/PG-PuReMD/src/multi_body.h
index aaed59e5..9cc865b4 100644
--- a/PG-PuReMD/src/multi_body.h
+++ b/PG-PuReMD/src/multi_body.h
@@ -24,7 +24,17 @@
#include "reax_types.h"
+
+#ifdef _cplusplus
+extern "C" {
+#endif
+
void Atom_Energy( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
+
+#ifdef _cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/neighbors.c b/PG-PuReMD/src/neighbors.c
index 753ecc36..e938329a 100644
--- a/PG-PuReMD/src/neighbors.c
+++ b/PG-PuReMD/src/neighbors.c
@@ -19,14 +19,16 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "neighbors.h"
+
+#include "index_utils.h"
#include "io_tools.h"
#include "list.h"
#include "tool_box.h"
#include "vector.h"
-#include "index_utils.h"
-
int compare_far_nbrs( const void *p1, const void *p2 )
{
diff --git a/PG-PuReMD/src/neighbors.h b/PG-PuReMD/src/neighbors.h
index 0a1e3daf..37c3642b 100644
--- a/PG-PuReMD/src/neighbors.h
+++ b/PG-PuReMD/src/neighbors.h
@@ -31,8 +31,18 @@
int, int*, int*, int*, int,
int, int, real, rvec, ivec );*/
-void Generate_Neighbor_Lists( reax_system*, simulation_data*, storage*,
- reax_list** );
+
+#ifdef _cplusplus
+extern "C" {
+#endif
+
+void Generate_Neighbor_Lists( reax_system*, simulation_data*, storage*, reax_list** );
+
int Estimate_NumNeighbors( reax_system*, reax_list** );
+#ifdef _cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/nonbonded.c b/PG-PuReMD/src/nonbonded.c
index 8edd2b11..e073ec62 100644
--- a/PG-PuReMD/src/nonbonded.c
+++ b/PG-PuReMD/src/nonbonded.c
@@ -20,7 +20,7 @@
----------------------------------------------------------------------*/
#include "reax_types.h"
-#include "index_utils.h"
+
#if defined(PURE_REAX)
#include "nonbonded.h"
#include "bond_orders.h"
@@ -34,10 +34,12 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
void vdW_Coulomb_Energy( reax_system *system, control_params *control,
- simulation_data *data, storage *workspace,
- reax_list **lists, output_controls *out_control )
+ simulation_data *data, storage *workspace,
+ reax_list **lists, output_controls *out_control )
{
int i, j, pj, natoms;
int start_i, end_i, orig_i, orig_j;
diff --git a/PG-PuReMD/src/nonbonded.h b/PG-PuReMD/src/nonbonded.h
index 81613be5..45137bf8 100644
--- a/PG-PuReMD/src/nonbonded.h
+++ b/PG-PuReMD/src/nonbonded.h
@@ -24,14 +24,24 @@
#include "reax_types.h"
+
+#ifdef _cplusplus
+extern "C" {
+#endif
+
void vdW_Coulomb_Energy( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
void Tabulated_vdW_Coulomb_Energy( reax_system*, control_params*,
- simulation_data*, storage*,
- reax_list**, output_controls* );
+ simulation_data*, storage*, reax_list**, output_controls* );
void Compute_Polarization_Energy( reax_system*, simulation_data* );
void LR_vdW_Coulomb( reax_system*, real *, int, int, real, LR_data* );
+
+#ifdef _cplusplus
+}
+#endif
+
+
#endif
diff --git a/PG-PuReMD/src/parallelreax.c b/PG-PuReMD/src/parallelreax.c
index 4d677687..30c23722 100644
--- a/PG-PuReMD/src/parallelreax.c
+++ b/PG-PuReMD/src/parallelreax.c
@@ -40,13 +40,13 @@
#include "vector.h"
#ifdef HAVE_CUDA
- #include "cuda_copy.h"
- #include "cuda_environment.h"
- #include "cuda_neighbors.h"
- #include "cuda_post_evolve.h"
- #include "cuda_reset_tools.h"
- #include "cuda_utils.h"
- #include "cuda_validation.h"
+ #include "cuda/cuda_copy.h"
+ #include "cuda/cuda_environment.h"
+ #include "cuda/cuda_neighbors.h"
+ #include "cuda/cuda_post_evolve.h"
+ #include "cuda/cuda_reset_tools.h"
+ #include "cuda/cuda_utils.h"
+ #include "cuda/cuda_validation.h"
#endif
evolve_function Evolve;
@@ -156,25 +156,6 @@ int Cuda_Post_Evolve( reax_system* system, control_params* control,
#endif
-#ifdef HAVE_CUDA
-void init_blocks( reax_system *system )
-{
- compute_blocks( &BLOCKS, &BLOCK_SIZE, system->n );
- compute_nearest_pow_2( BLOCKS, &BLOCKS_POW_2 );
-
- compute_blocks( &BLOCKS_N, &BLOCK_SIZE, system->N );
- compute_nearest_pow_2( BLOCKS_N, &BLOCKS_POW_2_N );
-
- compute_matvec_blocks( &MATVEC_BLOCKS, system->N );
-
-#if defined(__CUDA_DEBUG_LOG__)
- fprintf( stderr, " MATVEC_BLOCKS: %d BLOCKSIZE: %d - N:%d \n",
- MATVEC_BLOCKS, MATVEC_BLOCK_SIZE, system->N );
-#endif
-}
-#endif
-
-
static void usage( char* argv[] )
{
fprintf( stderr, "usage: ./%s geometry ffield control\n", argv[0] );
diff --git a/PG-PuReMD/src/random.c b/PG-PuReMD/src/random.c
index 2811a6b5..ffe55458 100644
--- a/PG-PuReMD/src/random.c
+++ b/PG-PuReMD/src/random.c
@@ -19,6 +19,8 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "random.h"
/* System random number generator used linear congruance method with
diff --git a/PG-PuReMD/src/random.h b/PG-PuReMD/src/random.h
index a3ce3526..66a5d59d 100644
--- a/PG-PuReMD/src/random.h
+++ b/PG-PuReMD/src/random.h
@@ -24,19 +24,28 @@
#include "reax_types.h"
+
+#ifdef _cplusplus
+extern "C" {
+#endif
+
/* 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 */
-double Random(double);
+double Random( double );
/* This function seeds the system pseudo random number generator with
current time. Use this function once in the begining to initialize
the system */
-void Randomize();
+void Randomize( );
/* GRandom return random number with gaussian distribution with mean
and standard deviation "sigma" */
-double GRandom(double, double);
+double GRandom( double, double );
+
+#ifdef _cplusplus
+}
+#endif
#endif
diff --git a/PG-PuReMD/src/reax_types.h b/PG-PuReMD/src/reax_types.h
index c39277b8..38810bd6 100644
--- a/PG-PuReMD/src/reax_types.h
+++ b/PG-PuReMD/src/reax_types.h
@@ -96,6 +96,14 @@
#define FABS fabs
#define FMOD fmod
+/* transcendental constant pi */
+#if defined(M_PI)
+ /* GNU C library (libc), defined in math.h */
+ #define PI (M_PI)
+#else
+ #define PI 3.14159265
+#endif
+
#define SQR(x) ((x)*(x))
#define CUBE(x) ((x)*(x)*(x))
#define DEG2RAD(a) ((a)*PI/180.0)
@@ -104,13 +112,6 @@
#define MIN(x,y) (((x) < (y)) ? (x) : (y))
#define MAX3(x,y,z) MAX( MAX(x,y), z)
-/* transcendental constant pi */
-#if defined(M_PI)
- /* GNU C library (libc), defined in math.h */
- #define PI (M_PI)
-#else
- #define PI 3.14159265
-#endif
/* ??? */
#define C_ele 332.06371
/* ??? */
diff --git a/PG-PuReMD/src/reset_tools.c b/PG-PuReMD/src/reset_tools.c
index a605cc79..c3778145 100644
--- a/PG-PuReMD/src/reset_tools.c
+++ b/PG-PuReMD/src/reset_tools.c
@@ -21,8 +21,6 @@
#include "reax_types.h"
-#include "index_utils.h"
-
#if defined(PURE_REAX)
#include "reset_tools.h"
#include "list.h"
@@ -35,6 +33,8 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
void Reset_Atoms( reax_system* system, control_params *control )
{
diff --git a/PG-PuReMD/src/reset_tools.h b/PG-PuReMD/src/reset_tools.h
index 34f38760..001b7f57 100644
--- a/PG-PuReMD/src/reset_tools.h
+++ b/PG-PuReMD/src/reset_tools.h
@@ -24,11 +24,11 @@
#include "reax_types.h"
+
#ifdef __cplusplus
extern "C" {
#endif
-
void Reset_Pressures( simulation_data* );
void Reset_Simulation_Data( simulation_data* );
@@ -49,9 +49,9 @@ void Reset( reax_system*, control_params*, simulation_data*, storage*, reax_list
void Reset_Test_Forces( reax_system*, storage* );
#endif
-
#ifdef __cplusplus
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/restart.c b/PG-PuReMD/src/restart.c
index 967e025d..6b8ddcdf 100644
--- a/PG-PuReMD/src/restart.c
+++ b/PG-PuReMD/src/restart.c
@@ -19,7 +19,10 @@
<http://www.gnu.org/licenses/>.
----------------------------------------------------------------------*/
+#include "reax_types.h"
+
#include "restart.h"
+
#include "allocate.h"
#include "box.h"
#include "tool_box.h"
diff --git a/PG-PuReMD/src/restart.h b/PG-PuReMD/src/restart.h
index 39a5dcd5..3d13a5a1 100644
--- a/PG-PuReMD/src/restart.h
+++ b/PG-PuReMD/src/restart.h
@@ -24,6 +24,7 @@
#include "reax_types.h"
+
#define RESTART_HEADER "%8d%12d%8.3f%8.3f%8.3f%8.3f%8.3f\n%15.5f%15.5f%15.5f\n%15.5f%15.5f%15.5f\n%15.5f%15.5f%15.5f\n"
#define RESTART_HEADER_LINE_LEN 200
/* step, system->bigN, data->therm.T, data->therm.xi,
@@ -39,16 +40,26 @@
#define READ_RESTART_HEADER " %d %d %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf %lf"
#define READ_RESTART_LINE " %d %d %s %lf %lf %lf %lf %lf %lf"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Write_Binary_Restart( reax_system*, control_params*,
- simulation_data*, output_controls*, mpi_datatypes* );
+ simulation_data*, output_controls*, mpi_datatypes* );
void Write_Restart( reax_system*, control_params*,
- simulation_data*, output_controls*, mpi_datatypes* );
+ simulation_data*, output_controls*, mpi_datatypes* );
void Read_Binary_Restart( char*, reax_system*, control_params*,
- simulation_data*, storage*, mpi_datatypes* );
+ simulation_data*, storage*, mpi_datatypes* );
void Read_Restart( char*, reax_system*, control_params*,
- simulation_data*, storage*, mpi_datatypes* );
+ simulation_data*, storage*, mpi_datatypes* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/system_props.c b/PG-PuReMD/src/system_props.c
index e2852a4b..ea4465c5 100644
--- a/PG-PuReMD/src/system_props.c
+++ b/PG-PuReMD/src/system_props.c
@@ -21,10 +21,6 @@
#include "reax_types.h"
-#ifdef HAVE_CUDA
- #include "cuda_system_props.h"
-#endif
-
#if defined(PURE_REAX)
#include "system_props.h"
#include "tool_box.h"
@@ -35,6 +31,10 @@
#include "reax_vector.h"
#endif
+#ifdef HAVE_CUDA
+ #include "cuda/cuda_system_props.h"
+#endif
+
void Temperature_Control( control_params *control, simulation_data *data )
{
@@ -83,29 +83,6 @@ void Compute_Kinetic_Energy( reax_system* system, simulation_data* data,
data->my_en.e_kin += 0.5 * rvec_Dot( p, system->my_atoms[i].v );
}
- MPI_Allreduce( &data->my_en.e_kin, &data->sys_en.e_kin,
- 1, MPI_DOUBLE, MPI_SUM, comm );
-
- data->therm.T = (2. * data->sys_en.e_kin) / (data->N_f * K_B);
-
- // avoid T being an absolute zero, might cause F.P.E!
- if ( FABS(data->therm.T) < ALMOST_ZERO )
- data->therm.T = ALMOST_ZERO;
-}
-
-
-#ifdef HAVE_CUDA
-void Cuda_Compute_Kinetic_Energy( reax_system* system, simulation_data* data,
- MPI_Comm comm )
-{
- int i;
- rvec p;
- real m;
-
- data->my_en.e_kin = 0.0;
-
- dev_compute_kinetic_energy( system, data, &data->my_en.e_kin );
-
MPI_Allreduce( &data->my_en.e_kin, &data->sys_en.e_kin,
1, MPI_DOUBLE, MPI_SUM, comm );
@@ -117,7 +94,6 @@ void Cuda_Compute_Kinetic_Energy( reax_system* system, simulation_data* data,
data->therm.T = ALMOST_ZERO;
}
}
-#endif
void Compute_System_Energy( reax_system *system, simulation_data *data,
@@ -130,7 +106,7 @@ void Compute_System_Energy( reax_system *system, simulation_data *data,
#ifdef HAVE_CUDA
//Cuda Wrapper here
- dev_sync_simulation_data ( data );
+ dev_sync_simulation_data( data );
#endif
my_en[0] = data->my_en.e_bond;
@@ -205,23 +181,6 @@ void Compute_Total_Mass( reax_system *system, simulation_data *data,
}
-#ifdef HAVE_CUDA
-void Cuda_Compute_Total_Mass( reax_system *system, simulation_data *data,
- MPI_Comm comm )
-{
- int i;
- real tmp;
-
- //compute local total mass of the system
- dev_compute_total_mass( system, &tmp );
-
- MPI_Allreduce( &tmp, &data->M, 1, MPI_DOUBLE, MPI_SUM, comm );
-
- data->inv_M = 1. / data->M;
-}
-#endif
-
-
void Compute_Center_of_Mass( reax_system *system, simulation_data *data,
mpi_datatypes *mpi_data, MPI_Comm comm )
{
@@ -342,112 +301,6 @@ void Compute_Center_of_Mass( reax_system *system, simulation_data *data,
}
-#ifdef HAVE_CUDA
-void Cuda_Compute_Center_of_Mass( reax_system *system, simulation_data *data,
- mpi_datatypes *mpi_data, MPI_Comm comm )
-{
- int i;
- real m, det; //xx, xy, xz, yy, yz, zz;
- real tmp_mat[6], tot_mat[6];
- rvec my_xcm, my_vcm, my_amcm, my_avcm;
- rvec tvec, diff;
- rtensor mat, inv;
-
- rvec_MakeZero( my_xcm ); // position of CoM
- rvec_MakeZero( my_vcm ); // velocity of CoM
- rvec_MakeZero( my_amcm ); // angular momentum of CoM
- rvec_MakeZero( my_avcm ); // angular velocity of CoM
-
- /* Compute the position, vel. and ang. momentum about the centre of mass */
- dev_compute_momentum ( system, my_xcm, my_vcm, my_amcm );
-
- MPI_Allreduce( my_xcm, data->xcm, 3, MPI_DOUBLE, MPI_SUM, comm );
- MPI_Allreduce( my_vcm, data->vcm, 3, MPI_DOUBLE, MPI_SUM, comm );
- MPI_Allreduce( my_amcm, data->amcm, 3, MPI_DOUBLE, MPI_SUM, comm );
-
- 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 */
- for ( i = 0; i < 6; ++i )
- {
- tmp_mat[i] = 0;
- }
-
- dev_compute_inertial_tensor( system, tmp_mat, my_xcm );
-
- MPI_Reduce( tmp_mat, tot_mat, 6, MPI_DOUBLE, MPI_SUM, MASTER_NODE, comm );
-
- if ( system->my_rank == MASTER_NODE )
- {
- mat[0][0] = tot_mat[3] + tot_mat[5]; // yy + zz;
- mat[0][1] = mat[1][0] = -tot_mat[1]; // -xy;
- mat[0][2] = mat[2][0] = -tot_mat[2]; // -xz;
- mat[1][1] = tot_mat[0] + tot_mat[5]; // xx + zz;
- mat[2][1] = mat[1][2] = -tot_mat[4]; // -yz;
- mat[2][2] = tot_mat[0] + tot_mat[3]; // 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 ( 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 );
- }
-
- MPI_Bcast( data->avcm, 3, MPI_DOUBLE, MASTER_NODE, comm );
-
- /* Compute the rotational energy */
- 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( stderr, "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( stderr, "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] ); */
- fprintf( stderr, "avcm: %24.15e %24.15e %24.15e\n",
- data->avcm[0], data->avcm[1], data->avcm[2] );
-#endif
-}
-#endif
-
-
/* IMPORTANT: This function assumes that current kinetic energy
* the system is already computed
*
diff --git a/PG-PuReMD/src/system_props.h b/PG-PuReMD/src/system_props.h
index 5efff3c5..f04a9590 100644
--- a/PG-PuReMD/src/system_props.h
+++ b/PG-PuReMD/src/system_props.h
@@ -24,6 +24,11 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Temperature_Control( control_params*, simulation_data* );
void Compute_Kinetic_Energy( reax_system*, simulation_data*, MPI_Comm );
@@ -33,16 +38,16 @@ void Compute_System_Energy( reax_system*, simulation_data*, MPI_Comm );
void Compute_Total_Mass( reax_system*, simulation_data*, MPI_Comm );
void Compute_Center_of_Mass( reax_system*, simulation_data*,
- mpi_datatypes*, MPI_Comm );
+ mpi_datatypes*, MPI_Comm );
void Compute_Pressure( reax_system*, control_params*,
- simulation_data*, mpi_datatypes* );
+ simulation_data*, mpi_datatypes* );
+
//void Compute_Pressure( reax_system*, simulation_data* );
-//CUDA Functions
-void Cuda_Compute_Total_Mass( reax_system*, simulation_data*, MPI_Comm );
-void Cuda_Compute_Kinetic_Energy( reax_system*, simulation_data*, MPI_Comm );
-void Cuda_Compute_Center_of_Mass( reax_system*, simulation_data*,
- mpi_datatypes*, MPI_Comm );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/tool_box.h b/PG-PuReMD/src/tool_box.h
index ecaee197..a1f55910 100644
--- a/PG-PuReMD/src/tool_box.h
+++ b/PG-PuReMD/src/tool_box.h
@@ -29,7 +29,6 @@
extern "C" {
#endif
-
/* from comm_tools.h */
int SumScan( int, int, int, MPI_Comm );
@@ -76,7 +75,6 @@ void *scalloc( size_t, size_t, const char* );
void sfree( void*, const char* );
-
#ifdef __cplusplus
}
#endif
@@ -227,4 +225,5 @@ static inline real DistSqr_to_Special_Point( rvec cp, rvec x )
}
#endif
+
#endif
diff --git a/PG-PuReMD/src/torsion_angles.c b/PG-PuReMD/src/torsion_angles.c
index 58e71f4e..29cfb444 100644
--- a/PG-PuReMD/src/torsion_angles.c
+++ b/PG-PuReMD/src/torsion_angles.c
@@ -21,7 +21,6 @@
#include "reax_types.h"
-#include "index_utils.h"
#if defined(PURE_REAX)
#include "torsion_angles.h"
#include "bond_orders.h"
@@ -36,6 +35,8 @@
#include "reax_vector.h"
#endif
+#include "index_utils.h"
+
#define MIN_SINE 1e-10
diff --git a/PG-PuReMD/src/torsion_angles.h b/PG-PuReMD/src/torsion_angles.h
index d0762a4e..454f0679 100644
--- a/PG-PuReMD/src/torsion_angles.h
+++ b/PG-PuReMD/src/torsion_angles.h
@@ -24,7 +24,17 @@
#include "reax_types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Torsion_Angles( reax_system*, control_params*, simulation_data*,
- storage*, reax_list**, output_controls* );
+ storage*, reax_list**, output_controls* );
+
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/traj.c b/PG-PuReMD/src/traj.c
index d561a45f..b7ba1112 100644
--- a/PG-PuReMD/src/traj.c
+++ b/PG-PuReMD/src/traj.c
@@ -32,7 +32,7 @@
#endif
#ifdef HAVE_CUDA
- #include "cuda_copy.h"
+ #include "cuda/cuda_copy.h"
#endif
diff --git a/PG-PuReMD/src/traj.h b/PG-PuReMD/src/traj.h
index 8f09c4a7..13435ecb 100644
--- a/PG-PuReMD/src/traj.h
+++ b/PG-PuReMD/src/traj.h
@@ -22,10 +22,8 @@
#ifndef __TRAJ_H__
#define __TRAJ_H__
-
#include "reax_types.h"
-
#define MAX_TRJ_LINE_LEN 120
#define MAX_TRJ_BUFFER_SIZE (MAX_TRJ_LINE_LEN * 100)
@@ -80,6 +78,10 @@ enum ANGLE_LINE_OPTS
};
+#ifdef __cplusplus
+extern "C" {
+#endif
+
int Init_Traj( reax_system*, control_params*, output_controls*, mpi_datatypes*, char* );
int End_Traj( int, output_controls* );
@@ -87,5 +89,9 @@ int End_Traj( int, output_controls* );
int Append_Frame( reax_system*, control_params*, simulation_data*, reax_list**,
output_controls*, mpi_datatypes* );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/valence_angles.h b/PG-PuReMD/src/valence_angles.h
index c7a56eaa..1958b0cb 100644
--- a/PG-PuReMD/src/valence_angles.h
+++ b/PG-PuReMD/src/valence_angles.h
@@ -25,6 +25,10 @@
#include "reax_types.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+
void Valence_Angles( reax_system*, control_params*, simulation_data*,
storage*, reax_list**, output_controls* );
@@ -32,5 +36,9 @@ void Calculate_Theta( rvec, real, rvec, real, real*, real* );
void Calculate_dCos_Theta( rvec, real, rvec, real, rvec*, rvec*, rvec* );
+#ifdef __cplusplus
+}
+#endif
+
#endif
diff --git a/PG-PuReMD/src/vector.h b/PG-PuReMD/src/vector.h
index adfe7da2..14250909 100644
--- a/PG-PuReMD/src/vector.h
+++ b/PG-PuReMD/src/vector.h
@@ -136,6 +136,7 @@ CUDA_HOST_DEVICE static inline void rvec_Copy( rvec dest, rvec src )
dest[2] = src[2];
}
+
CUDA_HOST_DEVICE static inline void rvec_Scale( rvec ret, real c, rvec v )
{
ret[0] = c * v[0];
@@ -497,8 +498,8 @@ CUDA_HOST_DEVICE static inline void rtensor_Transpose( rtensor ret, rtensor t )
CUDA_HOST_DEVICE static inline real rtensor_Det( rtensor t )
{
return ( t[0][0] * (t[1][1] * t[2][2] - t[1][2] * t[2][1] ) +
- t[0][1] * (t[1][2] * t[2][0] - t[1][0] * t[2][2] ) +
- t[0][2] * (t[1][0] * t[2][1] - t[1][1] * t[2][0] ) );
+ t[0][1] * (t[1][2] * t[2][0] - t[1][0] * t[2][2] ) +
+ t[0][2] * (t[1][0] * t[2][1] - t[1][1] * t[2][0] ) );
}
--
GitLab