From 854b27254cfd24e5deaba109c90cd02ac09c5f61 Mon Sep 17 00:00:00 2001
From: "Kurt A. O'Hearn" <ohearnku@msu.edu>
Date: Fri, 8 Jul 2022 18:23:03 -0400
Subject: [PATCH] PG-PuReMD: add OpenMP support for SAI preconditioner
computation (--enable-openmp=yes flag in configure script). Other code
clean-up.
---
Makefile.am | 6 +-
PG-PuReMD/src/control.c | 169 +++++++-------
PG-PuReMD/src/cuda/cuda_charges.cu | 9 -
PG-PuReMD/src/cuda/cuda_init_md.cu | 19 ++
PG-PuReMD/src/init_md.c | 34 +--
PG-PuReMD/src/init_md.h | 3 -
PG-PuReMD/src/lin_alg.c | 347 +++++++++++++++--------------
PG-PuReMD/src/reax_types.h | 8 +
configure.ac | 33 ++-
9 files changed, 353 insertions(+), 275 deletions(-)
diff --git a/Makefile.am b/Makefile.am
index fe042f2b..f20ca082 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -331,7 +331,7 @@ include cuda.am
endif
if USE_CUDA
-AM_NVCCFLAGS = --compiler-options="$(DEFS) $(AM_CFLAGS) -std=c++14 @CFLAGS_EXTRA@ $(CFLAGS) @M_CFLAGS@" -std=c++14 @NFLAGS@ @NFLAGS_EXTRA@
+AM_NVCCFLAGS = --compiler-options="$(DEFS) $(AM_CFLAGS) -std=c++14 @CFLAGS_EXTRA@ $(CFLAGS) @OMP_CFLAGS@ @M_CFLAGS@" -std=c++14 @NFLAGS@ @NFLAGS_EXTRA@
endif
bin_PROGRAMS = PG-PuReMD/bin/pg-puremd
@@ -464,9 +464,9 @@ PG_PuReMD_bin_pg_puremd_SOURCES += PG-PuReMD/src/cuda/cuda_allocate.cu \
nodist_EXTRA_PG_PuReMD_bin_pg_puremd_SOURCES = src/dummy.c
endif
-PG_PuReMD_bin_pg_puremd_CFLAGS = -std=c11 @CFLAGS_EXTRA@ @M_CFLAGS@
+PG_PuReMD_bin_pg_puremd_CFLAGS = -std=c11 @CFLAGS_EXTRA@ @OMP_CFLAGS@ @M_CFLAGS@
PG_PuReMD_bin_pg_puremd_CPPFLAGS = -I PG-PuReMD/src @CPPFLAGS_EXTRA@
-PG_PuReMD_bin_pg_puremd_LDADD = @M_LIBS@ @L_LIBS@ -lstdc++
+PG_PuReMD_bin_pg_puremd_LDADD = @OMP_LIBS@ @M_LIBS@ @L_LIBS@ -lstdc++
if USE_CUDA
PG_PuReMD_bin_pg_puremd_CFLAGS += @CUDA_CFLAGS@
PG_PuReMD_bin_pg_puremd_LDADD += @CUDA_LIBS@
diff --git a/PG-PuReMD/src/control.c b/PG-PuReMD/src/control.c
index dcac33b0..dd4bf787 100644
--- a/PG-PuReMD/src/control.c
+++ b/PG-PuReMD/src/control.c
@@ -52,6 +52,7 @@ void Read_Control_File( const char * const control_file, control_params * const
control->ensemble = NVE;
control->nsteps = 0;
control->dt = 0.25;
+ control->num_threads_set = FALSE;
control->nprocs = 1;
control->procs_by_dim[0] = 1;
control->procs_by_dim[1] = 1;
@@ -155,27 +156,32 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "ensemble_type", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->ensemble = ival;
}
else if ( strncmp(tmp[0], "nsteps", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->nsteps = ival;
}
- else if ( strncmp(tmp[0], "dt", MAX_LINE) == 0)
+ else if ( strncmp(tmp[0], "dt", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->dt = val * 1.e-3; // convert dt from fs to ps!
}
+ else if ( strncmp(tmp[0], "num_threads", MAX_LINE) == 0 )
+ {
+ control->num_threads = sstrtol( tmp[1], __FILE__, __LINE__ );
+ control->num_threads_set = TRUE;
+ }
else if ( strncmp(tmp[0], "gpus_per_node", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->gpus_per_node = ival;
}
else if ( strncmp(tmp[0], "gpu_streams", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->gpu_streams = ival;
if ( control->gpu_streams <= 0 || control->gpu_streams > MAX_CUDA_STREAMS )
@@ -193,11 +199,11 @@ void Read_Control_File( const char * const control_file, control_params * const
MPI_Abort( MPI_COMM_WORLD, INVALID_INPUT );
}
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->procs_by_dim[0] = ival;
- ival = atoi(tmp[2]);
+ ival = sstrtol( tmp[2], __FILE__, __LINE__ );
control->procs_by_dim[1] = ival;
- ival = atoi(tmp[3]);
+ ival = sstrtol( tmp[3], __FILE__, __LINE__ );
control->procs_by_dim[2] = ival;
control->nprocs = control->procs_by_dim[0] * control->procs_by_dim[1] *
@@ -210,7 +216,7 @@ void Read_Control_File( const char * const control_file, control_params * const
}
// else if( strncmp(tmp[0], "restart", MAX_LINE) == 0 )
// {
-// ival = atoi(tmp[1]);
+// ival = sstrtol( tmp[1], __FILE__, __LINE__ );
// control->restart = ival;
// }
// else if( strncmp(tmp[0], "restart_from", MAX_LINE) == 0 )
@@ -220,122 +226,122 @@ void Read_Control_File( const char * const control_file, control_params * const
// }
else if ( strncmp(tmp[0], "random_vel", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->random_vel = ival;
}
else if ( strncmp(tmp[0], "restart_format", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->restart_format = ival;
}
else if ( strncmp(tmp[0], "restart_freq", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->restart_freq = ival;
}
else if ( strncmp(tmp[0], "reposition_atoms", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->reposition_atoms = ival;
}
else if ( strncmp(tmp[0], "restrict_bonds", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->restrict_bonds = ival;
}
else if ( strncmp(tmp[0], "remove_CoM_vel", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->remove_CoM_vel = ival;
}
else if ( strncmp(tmp[0], "debug_level", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->debug_level = ival;
}
else if ( strncmp(tmp[0], "energy_update_freq", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->energy_update_freq = ival;
}
else if ( strncmp(tmp[0], "reneighbor", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->reneighbor = ival;
}
else if ( strncmp(tmp[0], "vlist_buffer", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->vlist_cut = val + control->nonb_cut;
}
else if ( strncmp(tmp[0], "nbrhood_cutoff", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->bond_cut = val;
}
else if ( strncmp(tmp[0], "bond_graph_cutoff", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->bg_cut = val;
}
else if ( strncmp(tmp[0], "thb_cutoff", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->thb_cut = val;
}
else if ( strncmp(tmp[0], "hbond_cutoff", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->hbond_cut = val;
}
else if ( strncmp(tmp[0], "ghost_cutoff", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->user_ghost_cut = val;
}
else if ( strncmp(tmp[0], "tabulate_long_range", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->tabulate = ival;
}
else if ( strncmp(tmp[0], "charge_method", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->charge_method = ival;
}
else if ( strncmp(tmp[0], "charge_freq", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->charge_freq = ival;
}
else if ( strncmp(tmp[0], "cm_q_net", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->cm_q_net = val;
}
else if ( strncmp(tmp[0], "cm_solver_type", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_type = ival;
}
else if ( strncmp(tmp[0], "cm_solver_max_iters", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_max_iters = ival;
}
else if ( strncmp(tmp[0], "cm_solver_restart", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_restart = ival;
}
else if ( strncmp(tmp[0], "cm_solver_q_err", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->cm_solver_q_err = val;
}
else if ( strncmp(tmp[0], "cm_domain_sparsity", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->cm_domain_sparsity = val;
if ( val < 1.0 )
{
@@ -344,57 +350,57 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "cm_init_guess_extrap1", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_init_guess_extrap1 = ival;
}
else if ( strncmp(tmp[0], "cm_init_guess_extrap2", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_init_guess_extrap2 = ival;
}
else if ( strncmp(tmp[0], "cm_solver_pre_comp_type", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_comp_type = ival;
}
else if ( strncmp(tmp[0], "cm_solver_pre_comp_refactor", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_comp_refactor = ival;
}
else if ( strncmp(tmp[0], "cm_solver_pre_comp_droptol", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_comp_droptol = val;
}
else if ( strncmp(tmp[0], "cm_solver_pre_comp_sweeps", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_comp_sweeps = ival;
}
else if ( strncmp(tmp[0], "cm_solver_pre_comp_sai_thres", MAX_LINE) == 0 )
{
- val = atof( tmp[1] );
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_comp_sai_thres = val;
}
else if ( strncmp(tmp[0], "cm_solver_pre_app_type", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_app_type = ival;
}
else if ( strncmp(tmp[0], "cm_solver_pre_app_jacobi_iters", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->cm_solver_pre_app_jacobi_iters = ival;
}
else if ( strncmp(tmp[0], "include_polarization_energy", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->polarization_energy_enabled = ival;
}
else if ( strncmp(tmp[0], "temp_init", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->T_init = val;
if ( control->T_init < 0.001 )
@@ -404,7 +410,7 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "temp_final", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->T_final = val;
if ( control->T_final < 0.1 )
@@ -414,36 +420,39 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "t_mass", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
/* convert from fs to s */
control->Tau_T = val * 1.0e-15;
}
else if ( strncmp(tmp[0], "t_mode", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->T_mode = ival;
}
else if ( strncmp(tmp[0], "t_rate", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->T_rate = val;
}
else if ( strncmp(tmp[0], "t_freq", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->T_freq = val;
}
else if ( strncmp(tmp[0], "pressure", MAX_LINE) == 0 )
{
if ( control->ensemble == iNPT )
{
- control->P[0] = control->P[1] = control->P[2] = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
+ control->P[0] = val;
+ control->P[1] = val;
+ control->P[2] = val;
}
else if ( control->ensemble == sNPT )
{
- control->P[0] = atof(tmp[1]);
- control->P[1] = atof(tmp[2]);
- control->P[2] = atof(tmp[3]);
+ control->P[0] = sstrtod( tmp[1], __FILE__, __LINE__ );
+ control->P[1] = sstrtod( tmp[2], __FILE__, __LINE__ );
+ control->P[2] = sstrtod( tmp[3], __FILE__, __LINE__ );
}
}
else if ( strncmp(tmp[0], "p_mass", MAX_LINE) == 0 )
@@ -451,45 +460,47 @@ void Read_Control_File( const char * const control_file, control_params * const
// convert p_mass from fs to ps
if ( control->ensemble == iNPT )
{
- control->Tau_P[0] = control->Tau_P[1] = control->Tau_P[2] =
- atof(tmp[1]) * 1.e-3;
+ val = sstrtod( tmp[1], __FILE__, __LINE__ ) * 1.e-3;
+ control->Tau_P[0] = val;
+ control->Tau_P[1] = val;
+ control->Tau_P[2] = val;
}
else if ( control->ensemble == sNPT )
{
- control->Tau_P[0] = atof(tmp[1]) * 1.e-3;
- control->Tau_P[1] = atof(tmp[2]) * 1.e-3;
- control->Tau_P[2] = atof(tmp[3]) * 1.e-3;
+ control->Tau_P[0] = sstrtod( tmp[1], __FILE__, __LINE__ ) * 1.e-3;
+ control->Tau_P[1] = sstrtod( tmp[2], __FILE__, __LINE__ ) * 1.e-3;
+ control->Tau_P[2] = sstrtod( tmp[3], __FILE__, __LINE__ ) * 1.e-3;
}
}
else if ( strncmp(tmp[0], "pt_mass", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->Tau_PT[0] = control->Tau_PT[1] = control->Tau_PT[2] =
val * 1.e-3; // convert pt_mass from fs to ps
}
else if ( strncmp(tmp[0], "compress", MAX_LINE) == 0 )
{
- val = atof(tmp[1]);
+ val = sstrtod( tmp[1], __FILE__, __LINE__ );
control->compressibility = val;
}
else if ( strncmp(tmp[0], "press_mode", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->press_mode = ival;
}
else if ( strncmp(tmp[0], "geo_format", MAX_LINE) == 0 )
{
- ival = atoi( tmp[1] );
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->geo_format = ival;
}
else if ( strncmp(tmp[0], "write_freq", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->write_steps = ival;
}
else if ( strncmp(tmp[0], "traj_compress", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->traj_compress = ival;
}
else if ( strncmp(tmp[0], "traj_format", MAX_LINE) == 0 )
@@ -499,7 +510,7 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "traj_method", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->traj_method = ival;
}
else if ( strncmp(tmp[0], "traj_title", MAX_LINE) == 0 )
@@ -509,27 +520,27 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "atom_info", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->atom_info += ival * 4;
}
else if ( strncmp(tmp[0], "atom_velocities", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->atom_info += ival * 2;
}
else if ( strncmp(tmp[0], "atom_forces", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->atom_info += ival * 1;
}
else if ( strncmp(tmp[0], "bond_info", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->bond_info = ival;
}
else if ( strncmp(tmp[0], "angle_info", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
out_control->angle_info = ival;
}
else if ( strncmp(tmp[0], "test_forces", MAX_LINE) == 0 )
@@ -540,15 +551,15 @@ void Read_Control_File( const char * const control_file, control_params * const
else if ( strncmp(tmp[0], "molecular_analysis", MAX_LINE) == 0
|| strncmp(tmp[0], "molec_anal", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->molecular_analysis = ival;
}
else if ( strncmp(tmp[0], "ignore", MAX_LINE) == 0 )
{
- control->num_ignored = atoi(tmp[1]);
+ control->num_ignored = sstrtol( tmp[1], __FILE__, __LINE__ );
for ( i = 0; i < control->num_ignored; ++i )
{
- control->ignore[atoi(tmp[i + 2])] = 1;
+ control->ignore[sstrtol( tmp[i + 2], __FILE__, __LINE__ )] = 1;
}
}
else if ( strncmp(tmp[0], "freq_molec_anal", MAX_LINE) == 0 )
@@ -558,27 +569,27 @@ void Read_Control_File( const char * const control_file, control_params * const
}
else if ( strncmp(tmp[0], "dipole_anal", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->dipole_anal = ival;
}
else if ( strncmp(tmp[0], "freq_dipole_anal", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->freq_dipole_anal = ival;
}
else if ( strncmp(tmp[0], "diffusion_coef", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->diffusion_coef = ival;
}
else if ( strncmp(tmp[0], "freq_diffusion_coef", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->freq_diffusion_coef = ival;
}
else if ( strncmp(tmp[0], "restrict_type", MAX_LINE) == 0 )
{
- ival = atoi(tmp[1]);
+ ival = sstrtol( tmp[1], __FILE__, __LINE__ );
control->restrict_type = ival;
}
else
diff --git a/PG-PuReMD/src/cuda/cuda_charges.cu b/PG-PuReMD/src/cuda/cuda_charges.cu
index 0f4d4280..6eeda69b 100644
--- a/PG-PuReMD/src/cuda/cuda_charges.cu
+++ b/PG-PuReMD/src/cuda/cuda_charges.cu
@@ -280,13 +280,6 @@ static void Setup_Preconditioner_QEq( reax_system const * const system,
time = Get_Time( );
#endif
- /* sort H needed for SpMV's in linear solver, H or H_sp needed for preconditioning */
-// Sort_Matrix_Rows( &workspace->d_workspace->H, workspace, s );
-
-#if defined(LOG_PERFORMANCE)
- Update_Timing_Info( &time, &data->timing.cm_sort );
-#endif
-
switch ( control->cm_solver_pre_comp_type )
{
case NONE_PC:
@@ -323,8 +316,6 @@ static void Setup_Preconditioner_QEq( reax_system const * const system,
Cuda_Copy_Matrix_Device_to_Host( &workspace->H,
&workspace->d_workspace->H, s );
- Sort_Matrix_Rows( &workspace->H );
-
setup_sparse_approx_inverse( system, data,
&workspace->H, &workspace->H_spar_patt,
control->nprocs, control->cm_solver_pre_comp_sai_thres );
diff --git a/PG-PuReMD/src/cuda/cuda_init_md.cu b/PG-PuReMD/src/cuda/cuda_init_md.cu
index c46a1ffc..b37430ef 100644
--- a/PG-PuReMD/src/cuda/cuda_init_md.cu
+++ b/PG-PuReMD/src/cuda/cuda_init_md.cu
@@ -250,6 +250,25 @@ extern "C" void Cuda_Initialize( reax_system *system, control_params *control,
{
int i;
+#if defined(_OPENMP)
+ #pragma omp parallel default(none) shared(control)
+ {
+ #pragma omp single
+ {
+ if ( control->num_threads_set == FALSE )
+ {
+ /* set using OMP_NUM_THREADS environment variable */
+ control->num_threads = omp_get_num_threads( );
+ control->num_threads_set = TRUE;
+ }
+ }
+ }
+
+ omp_set_num_threads( control->num_threads );
+#else
+ control->num_threads = 1;
+#endif
+
Init_MPI_Datatypes( system, workspace, mpi_data );
#if defined(CUDA_DEVICE_PACK)
diff --git a/PG-PuReMD/src/init_md.c b/PG-PuReMD/src/init_md.c
index 5b3e0727..1934388d 100644
--- a/PG-PuReMD/src/init_md.c
+++ b/PG-PuReMD/src/init_md.c
@@ -710,6 +710,25 @@ void Initialize( reax_system * const system, control_params * const control,
reax_list ** const lists, output_controls * const out_control,
mpi_datatypes * const mpi_data )
{
+#if defined(_OPENMP)
+ #pragma omp parallel default(none) shared(control)
+ {
+ #pragma omp single
+ {
+ if ( control->num_threads_set == FALSE )
+ {
+ /* set using OMP_NUM_THREADS environment variable */
+ control->num_threads = omp_get_num_threads( );
+ control->num_threads_set = TRUE;
+ }
+ }
+ }
+
+ omp_set_num_threads( control->num_threads );
+#else
+ control->num_threads = 1;
+#endif
+
Init_MPI_Datatypes( system, workspace, mpi_data );
Init_Simulation_Data( system, control, data, mpi_data );
@@ -733,21 +752,6 @@ void Initialize( reax_system * const system, control_params * const control,
}
-void Pure_Initialize( reax_system * const system, control_params * const control,
- simulation_data * const data, storage * const workspace,
- reax_list ** const lists, output_controls * const out_control,
- mpi_datatypes * const mpi_data )
-{
- Init_Simulation_Data( system, control, data, mpi_data );
-
- Init_Workspace( system, control, workspace, mpi_data );
-
- Init_Lists( system, control, data, workspace, lists, mpi_data );
-
- Init_Force_Functions( control );
-}
-
-
#elif defined(LAMMPS_REAX)
void Initialize( reax_system * const system, control_params * const control,
simulation_data * const data, storage * const workspace,
diff --git a/PG-PuReMD/src/init_md.h b/PG-PuReMD/src/init_md.h
index f4fa2a32..44ccb6b5 100644
--- a/PG-PuReMD/src/init_md.h
+++ b/PG-PuReMD/src/init_md.h
@@ -36,9 +36,6 @@ void Init_MPI_Datatypes( reax_system * const, storage * const, mpi_datatypes * c
void Initialize( reax_system * const, control_params * const, simulation_data * const,
storage * const, reax_list** const, output_controls * const, mpi_datatypes * const );
-void Pure_Initialize( reax_system * const, control_params * const, simulation_data * const,
- storage * const, reax_list** const, output_controls * const, mpi_datatypes * const );
-
void Init_Taper( control_params * const, storage * const,
mpi_datatypes * const );
diff --git a/PG-PuReMD/src/lin_alg.c b/PG-PuReMD/src/lin_alg.c
index b341a045..30cc2dd6 100644
--- a/PG-PuReMD/src/lin_alg.c
+++ b/PG-PuReMD/src/lin_alg.c
@@ -1494,7 +1494,7 @@ void sparse_approx_inverse( reax_system const * const system,
int local_pos, identity_pos;
int *X, *q;
int q_size, q_top;
- int e_size, D_size;
+ int e_i_size, D_size;
int cnt;
int *row_nnz, *row_start, total_entries;
int *col_inds;
@@ -1544,10 +1544,6 @@ void sparse_approx_inverse( reax_system const * const system,
nz_vals_send_buf1_size = 0;
nz_vals_send_buf2 = NULL;
nz_vals_send_buf2_size = 0;
- e_i = NULL;
- e_size = 0;
- D = NULL;
- D_size = 0;
row_nnz = smalloc( sizeof(int) * system->total_cap, __FILE__, __LINE__ );
row_start = smalloc( sizeof(int) * (system->total_cap + 1 - A->n), __FILE__, __LINE__ );
@@ -1778,222 +1774,249 @@ void sparse_approx_inverse( reax_system const * const system,
sfree( nz_vals_send_buf1, __FILE__, __LINE__ );
sfree( nz_vals_send_buf2, __FILE__, __LINE__ );
- X = smalloc( sizeof(int) * system->bigN, __FILE__, __LINE__ );
- q_size = MIN_QUEUE_SIZE;
- q = smalloc( sizeof(int) * q_size, __FILE__, __LINE__ );
-
- for ( i = 0; i < A_spar_patt->n; ++i )
+#if defined(_OPENMP)
+ #pragma omp parallel default(none) \
+ private(i, pj, pk, ret, N, M, d_i, d_j, mark, local_pos, identity_pos, \
+ X, q, q_size, q_top, e_i, e_i_size, D, D_size, \
+ m, n, lda, info) \
+ firstprivate(system, A, A_spar_patt, A_app_inv, row_nnz, row_start, col_inds, nz_vals) \
+ shared(stderr)
+#endif
{
- M = 0;
- N = 0;
- q_top = 0;
- mark = -999;
-
- /* column indices for the non-zeros in the i-th row in the sparsity pattern
- * for H_app_inv delineate which columns of H must be searched for non-zeros
- * (when constructing the dense matrix D) */
- for ( pj = A_spar_patt->start[i]; pj < A_spar_patt->end[i]; ++pj )
+ e_i = NULL;
+ e_i_size = 0;
+ D = NULL;
+ D_size = 0;
+ X = smalloc( sizeof(int) * system->bigN, __FILE__, __LINE__ );
+ q_size = MIN_QUEUE_SIZE;
+ q = smalloc( sizeof(int) * q_size, __FILE__, __LINE__ );
+
+#if defined(_OPENMP)
+ #pragma omp for schedule(dynamic,32)
+#endif
+ for ( i = 0; i < A_spar_patt->n; ++i )
{
- ++N;
+ M = 0;
+ N = 0;
+ q_top = 0;
+ mark = -999;
+
+ /* column indices for the non-zeros in the i-th row in the sparsity pattern
+ * for H_app_inv delineate which columns of H must be searched for non-zeros
+ * (when constructing the dense matrix D) */
+ for ( pj = A_spar_patt->start[i]; pj < A_spar_patt->end[i]; ++pj )
+ {
+ ++N;
- assert( 0 <= A_spar_patt->j[pj] );
- assert( A_spar_patt->j[pj] < system->N );
+ assert( 0 <= A_spar_patt->j[pj] );
+ assert( A_spar_patt->j[pj] < system->N );
- /* due to symmetry, search rows of H instead of columns for non-zeros
- *
- * case: index's row in local matrix */
- if ( A_spar_patt->j[pj] < A->n )
- {
- for ( pk = A->start[A_spar_patt->j[pj]]; pk < A->end[A_spar_patt->j[pj]]; ++pk )
+ /* due to symmetry, search rows of H instead of columns for non-zeros
+ *
+ * case: index's row in local matrix */
+ if ( A_spar_patt->j[pj] < A->n )
{
- assert( 1 <= system->my_atoms[A->j[pk]].orig_id
- && system->my_atoms[A->j[pk]].orig_id <= system->bigN );
+ for ( pk = A->start[A_spar_patt->j[pj]]; pk < A->end[A_spar_patt->j[pj]]; ++pk )
+ {
+ assert( 1 <= system->my_atoms[A->j[pk]].orig_id
+ && system->my_atoms[A->j[pk]].orig_id <= system->bigN );
- /* accumulate the nonzero column indices to serve as the row indices of the dense matrix */
- X[system->my_atoms[A->j[pk]].orig_id - 1] = mark;
+ /* accumulate the nonzero column indices to serve as the row indices of the dense matrix */
+ X[system->my_atoms[A->j[pk]].orig_id - 1] = mark;
- if ( q_top >= q_size )
- {
- q_size = (int) CEIL( q_size * SAFE_ZONE );
- q = srealloc( q, sizeof(int) * q_size, __FILE__, __LINE__ );
- }
+ if ( q_top >= q_size )
+ {
+ q_size = (int) CEIL( q_size * SAFE_ZONE );
+ q = srealloc( q, sizeof(int) * q_size, __FILE__, __LINE__ );
+ }
- q[q_top] = system->my_atoms[A->j[pk]].orig_id - 1;
- ++q_top;
+ q[q_top] = system->my_atoms[A->j[pk]].orig_id - 1;
+ ++q_top;
+ }
}
- }
- /* case: index's row communicated */
- else
- {
- assert( 0 <= row_start[A_spar_patt->j[pj] - A->n] );
- assert( row_start[A_spar_patt->j[pj] - A->n] < total_entries );
- assert( 0 <= row_start[A_spar_patt->j[pj] - A->n + 1] );
- assert( row_start[A_spar_patt->j[pj] - A->n + 1] < total_entries );
-
- for ( pk = row_start[A_spar_patt->j[pj] - A->n];
- pk < row_start[A_spar_patt->j[pj] - A->n + 1]; ++pk )
+ /* case: index's row communicated */
+ else
{
- assert( 0 <= col_inds[pk] );
- assert( col_inds[pk] < system->bigN );
-
- /* accumulate the nonzero column indices to serve as the row indices of the dense matrix */
- X[col_inds[pk]] = mark;
+ assert( 0 <= row_start[A_spar_patt->j[pj] - A->n] );
+ assert( row_start[A_spar_patt->j[pj] - A->n] < total_entries );
+ assert( 0 <= row_start[A_spar_patt->j[pj] - A->n + 1] );
+ assert( row_start[A_spar_patt->j[pj] - A->n + 1] < total_entries );
- if ( q_top >= q_size )
+ for ( pk = row_start[A_spar_patt->j[pj] - A->n];
+ pk < row_start[A_spar_patt->j[pj] - A->n + 1]; ++pk )
{
- q_size = (int) CEIL( q_size * SAFE_ZONE );
- q = srealloc( q, sizeof(int) * q_size, __FILE__, __LINE__ );
- }
+ assert( 0 <= col_inds[pk] );
+ assert( col_inds[pk] < system->bigN );
+
+ /* accumulate the nonzero column indices to serve as the row indices of the dense matrix */
+ X[col_inds[pk]] = mark;
+
+ if ( q_top >= q_size )
+ {
+ q_size = (int) CEIL( q_size * SAFE_ZONE );
+ q = srealloc( q, sizeof(int) * q_size, __FILE__, __LINE__ );
+ }
- q[q_top] = col_inds[pk];
- ++q_top;
+ q[q_top] = col_inds[pk];
+ ++q_top;
+ }
}
}
- }
- /* enumerate the row indices from 0 to (# of nonzero rows - 1) for the dense matrix */
- for ( pj = 0; pj < q_top; ++pj )
- {
- if ( X[q[pj]] == mark )
+ /* enumerate the row indices from 0 to (# of nonzero rows - 1) for the dense matrix */
+ for ( pj = 0; pj < q_top; ++pj )
{
- X[q[pj]] = M;
- ++M;
+ if ( X[q[pj]] == mark )
+ {
+ X[q[pj]] = M;
+ ++M;
+ }
}
- }
- identity_pos = X[system->my_atoms[i].orig_id - 1];
+ identity_pos = X[system->my_atoms[i].orig_id - 1];
- assert( 0 <= identity_pos );
- assert( identity_pos < M );
+ assert( 0 <= identity_pos );
+ assert( identity_pos < M );
- if ( M * N > D_size )
- {
- if ( D_size > 0 )
+ if ( M * N > D_size )
{
- sfree( D, __FILE__, __LINE__ );
+ if ( D_size > 0 )
+ {
+ sfree( D, __FILE__, __LINE__ );
+ }
+
+ D_size = (int) CEIL( M * N * SAFE_ZONE );
+
+ D = smalloc( sizeof(real) * D_size, __FILE__, __LINE__ );
}
-
- D_size = (int) CEIL( M * N * SAFE_ZONE );
- D = smalloc( sizeof(real) * D_size, __FILE__, __LINE__ );
- }
+ for ( d_i = 0; d_i < M; ++d_i )
+ {
+ for ( d_j = 0; d_j < N; ++d_j )
+ {
+ D[d_i * N + d_j] = 0.0;
+ }
+ }
- for ( d_i = 0; d_i < M; ++d_i )
- {
+ /* fill in entries of D column-wise */
for ( d_j = 0; d_j < N; ++d_j )
{
- D[d_i * N + d_j] = 0.0;
- }
- }
+ local_pos = A_spar_patt->j[A_spar_patt->start[i] + d_j];
- /* fill in entries of D column-wise */
- for ( d_j = 0; d_j < N; ++d_j )
- {
- local_pos = A_spar_patt->j[A_spar_patt->start[i] + d_j];
+ assert( 0 <= local_pos );
+ assert( local_pos < system->N );
- assert( 0 <= local_pos );
- assert( local_pos < system->N );
+ if ( local_pos < A->n )
+ {
+ for ( d_i = A->start[local_pos]; d_i < A->end[local_pos]; ++d_i )
+ {
+ assert( 0 <= X[system->my_atoms[A->j[d_i]].orig_id - 1] );
+ assert( X[system->my_atoms[A->j[d_i]].orig_id - 1] < M );
- if ( local_pos < A->n )
- {
- for ( d_i = A->start[local_pos]; d_i < A->end[local_pos]; ++d_i )
+ D[X[system->my_atoms[A->j[d_i]].orig_id - 1] * N + d_j] = A->val[d_i];
+ }
+ }
+ else
{
- assert( 0 <= X[system->my_atoms[A->j[d_i]].orig_id - 1] );
- assert( X[system->my_atoms[A->j[d_i]].orig_id - 1] < M );
+ assert( 0 <= row_start[local_pos - A->n] );
+ assert( row_start[local_pos - A->n] < total_entries );
+ assert( 0 <= row_start[local_pos - A->n + 1] );
+ assert( row_start[local_pos - A->n + 1] < total_entries );
+
+ for ( d_i = row_start[local_pos - A->n];
+ d_i < row_start[local_pos - A->n + 1]; ++d_i )
+ {
+ assert( 0 <= col_inds[d_i] );
+ assert( col_inds[d_i] < system->bigN );
+ assert( 0 <= X[col_inds[d_i]] );
+ assert( X[col_inds[d_i]] < M );
- D[X[system->my_atoms[A->j[d_i]].orig_id - 1] * N + d_j] = A->val[d_i];
+ D[X[col_inds[d_i]] * N + d_j] = nz_vals[d_i];
+ }
}
}
- else
- {
- assert( 0 <= row_start[local_pos - A->n] );
- assert( row_start[local_pos - A->n] < total_entries );
- assert( 0 <= row_start[local_pos - A->n + 1] );
- assert( row_start[local_pos - A->n + 1] < total_entries );
- for ( d_i = row_start[local_pos - A->n];
- d_i < row_start[local_pos - A->n + 1]; ++d_i )
+ /* create the right hand side of the linear equation
+ * that is the full column of the identity matrix */
+ if ( e_i_size < M )
+ {
+ if ( e_i_size > 0 )
{
- assert( 0 <= col_inds[d_i] );
- assert( col_inds[d_i] < system->bigN );
- assert( 0 <= X[col_inds[d_i]] );
- assert( X[col_inds[d_i]] < M );
-
- D[X[col_inds[d_i]] * N + d_j] = nz_vals[d_i];
+ sfree( e_i, __FILE__, __LINE__ );
}
+
+ e_i_size = (int) CEIL( M * SAFE_ZONE );
+
+ e_i = smalloc( sizeof(real) * e_i_size, __FILE__, __LINE__ );
}
- }
- /* create the right hand side of the linear equation
- * that is the full column of the identity matrix */
- if ( e_size < M )
- {
- if ( e_size > 0 )
+ for ( pk = 0; pk < M; ++pk )
{
- sfree( e_i, __FILE__, __LINE__ );
+ e_i[pk] = 0.0;
}
+ e_i[identity_pos] = 1.0;
- e_size = (int) CEIL( M * SAFE_ZONE );
-
- e_i = smalloc( sizeof(real) * e_size, __FILE__, __LINE__ );
- }
+ /* solve the overdetermined system from the i-th columns of min_{M} ||I - HM||_F
+ * through the least-squares problem min_{m_i} ||e_i - Dm_i||_{2}^{2}
+ * where D represents the dense matrix composed of relevant non-zeros from H,
+ * e_i is the i-th column of the identify matrix, and m_i is the i-th column of M
+ *
+ * after solve, D contains the QR factorization and e_i contains
+ * the solution (if info == 0) */
+ m = (lapack_int) M;
+ n = (lapack_int) N;
+ lda = n;
- for ( pk = 0; pk < M; ++pk )
- {
- e_i[pk] = 0.0;
- }
- e_i[identity_pos] = 1.0;
+ assert( M >= N );
- /* solve the overdetermined system from the i-th columns of min_{M} ||I - HM||_F
- * through the least-squares problem min_{m_i} ||e_i - Dm_i||_{2}^{2}
- * where D represents the dense matrix composed of relevant non-zeros from H,
- * e_i is the i-th column of the identify matrix, and m_i is the i-th column of M
- *
- * after solve, D contains the QR factorization and e_i contains
- * the solution (if info == 0) */
- m = (lapack_int) M;
- n = (lapack_int) N;
- lda = n;
+ info = LAPACKE_dgels( LAPACK_ROW_MAJOR, 'N', m, n, 1, D, lda, e_i, 1 );
- assert( M >= N );
+ /* check for full rank */
+ if ( info > 0 )
+ {
+ fprintf( stderr, "[ERROR] SAI preconditioner computation failure\n" );
+ fprintf( stderr, " [INFO] The diagonal element %i of the triangular factor\n", info );
+ fprintf( stderr, " of A is zero, so A does not have full rank.\n" );
+// MPI_Abort( MPI_COMM_WORLD, RUNTIME_ERROR );
+ exit( RUNTIME_ERROR );
+ }
- info = LAPACKE_dgels( LAPACK_ROW_MAJOR, 'N', m, n, 1, D, lda, e_i, 1 );
+ if ( A_spar_patt->end[i] - A_spar_patt->start[i] < N )
+ {
+ fprintf( stderr, "[ERROR] SAI preconditioner computation failure\n" );
+ fprintf( stderr, " [INFO] out of memory for row i = %d\n", i );
+// MPI_Abort( MPI_COMM_WORLD, RUNTIME_ERROR );
+ exit( RUNTIME_ERROR );
+ }
- /* check for full rank */
- if ( info > 0 )
- {
- fprintf( stderr, "[ERROR] SAI preconditioner computation failure\n" );
- fprintf( stderr, " [INFO] The diagonal element %i of the triangular factor\n", info );
- fprintf( stderr, " of A is zero, so A does not have full rank.\n" );
- MPI_Abort( MPI_COMM_WORLD, RUNTIME_ERROR );
+ /* least-squares solution in e_i comprises non-zero values for i-th row in A_app_inv
+ * (computed as right preconditioner, store transpose as left preconditioner) */
+ A_app_inv->start[i] = A_spar_patt->start[i];
+ A_app_inv->end[i] = A_spar_patt->end[i];
+ for ( pj = A_app_inv->start[i]; pj < A_app_inv->end[i]; ++pj )
+ {
+ A_app_inv->j[pj] = A_spar_patt->j[pj];
+ A_app_inv->val[pj] = e_i[pj - A_spar_patt->start[i]];
+ }
}
- if ( A_spar_patt->end[i] - A_spar_patt->start[i] < N )
+ if ( D != NULL )
{
- fprintf( stderr, "[ERROR] SAI preconditioner computation failure\n" );
- fprintf( stderr, " [INFO] out of memory for row i = %d\n", i );
- MPI_Abort( MPI_COMM_WORLD, RUNTIME_ERROR );
+ sfree( D, __FILE__, __LINE__ );
}
-
- /* least-squares solution in e_i comprises non-zero values for i-th row in A_app_inv
- * (computed as right preconditioner, store transpose as left preconditioner) */
- A_app_inv->start[i] = A_spar_patt->start[i];
- A_app_inv->end[i] = A_spar_patt->end[i];
- for ( pj = A_app_inv->start[i]; pj < A_app_inv->end[i]; ++pj )
+ if ( e_i != NULL )
{
- A_app_inv->j[pj] = A_spar_patt->j[pj];
- A_app_inv->val[pj] = e_i[pj - A_spar_patt->start[i]];
+ sfree( e_i, __FILE__, __LINE__ );
}
+ sfree( X, __FILE__, __LINE__ );
+ sfree( q, __FILE__, __LINE__ );
}
+
for ( i = A_app_inv->n; i < A_app_inv->n_max; ++i )
{
A_app_inv->start[i] = 0;
A_app_inv->end[i] = 0;
}
- sfree( q, __FILE__, __LINE__ );
- sfree( D, __FILE__, __LINE__ );
- sfree( e_i, __FILE__, __LINE__ );
- sfree( X, __FILE__, __LINE__ );
sfree( col_inds, __FILE__, __LINE__ );
sfree( nz_vals, __FILE__, __LINE__ );
sfree( row_nnz, __FILE__, __LINE__ );
diff --git a/PG-PuReMD/src/reax_types.h b/PG-PuReMD/src/reax_types.h
index 8865fd7e..ae75d7ee 100644
--- a/PG-PuReMD/src/reax_types.h
+++ b/PG-PuReMD/src/reax_types.h
@@ -63,6 +63,10 @@
#endif
#endif
+#if defined(_OPENMP)
+ #include <omp.h>
+#endif
+
/* IBMC */
#if defined(__IBMC__)
#define inline __inline__
@@ -1702,6 +1706,10 @@ struct control_params
int freq_diffusion_coef;
/**/
int restrict_type;
+ /* number of OpenMP threads to use during the simulation */
+ int num_threads;
+ /* TRUE if the num. OpenMP has bet set, FALSE otherwise */
+ int num_threads_set;
/* function pointer for ensemble used to evolve atomic system */
evolve_function Evolve;
/* function pointers for bonded interactions */
diff --git a/configure.ac b/configure.ac
index dd8a0295..35356b34 100644
--- a/configure.ac
+++ b/configure.ac
@@ -143,7 +143,7 @@ AC_PROG_CXXCPP
AC_CONFIG_HEADERS([common/include/config.h])
# sPuReMD
-if test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "xyes"; then
+if test "x${pack_mpi_enabled}" != "xyes" && test "x${pack_mpi_gpu_enabled}" != "xyes" && { test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "xyes"; }; then
if test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" != "xyes"; then
BUILD_OPENMP="no"
else
@@ -217,8 +217,6 @@ if test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "
OPENMP_LIBS="-lgomp"
fi
- # AC_MSG_NOTICE([OPENMP_CFLAGS=${OPENMP_CFLAGS}])
- # AC_MSG_NOTICE([OPENMP_LIBS=${OPENMP_LIBS}])
AC_SUBST([OMP_CFLAGS], ["${OPENMP_CFLAGS}"])
AC_SUBST([OMP_LIBS], ["${OPENMP_LIBS}"])
fi
@@ -351,7 +349,7 @@ if test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "
AC_LANG_POP([C])
fi
-AM_CONDITIONAL([BUILD_S_OMP], [test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "xyes"])
+AM_CONDITIONAL([BUILD_S_OMP], [test "x${pack_mpi_enabled}" != "xyes" && test "x${pack_mpi_gpu_enabled}" != "xyes" && { test "x${pack_serial_enabled}" = "xyes" || test "x${pack_openmp_enabled}" = "xyes"; }])
# PuReMD-GPU
@@ -633,6 +631,11 @@ AM_CONDITIONAL([BUILD_MPI], [test "x${pack_mpi_old_enabled}" = "xyes"])
# PG-PuReMD
if test "x${pack_mpi_enabled}" = "xyes" || test "x${pack_mpi_gpu_enabled}" = "xyes"; then
+ if test "x${pack_openmp_enabled}" = "xyes"; then
+ BUILD_OPENMP_MPI="yes"
+ else
+ BUILD_OPENMP_MPI="no"
+ fi
if test "x${pack_mpi_enabled}" = "xyes" || test "x${pack_mpi_gpu_enabled}" != "xyes"; then
export BUILD_GPU="no"
else
@@ -691,6 +694,28 @@ if test "x${pack_mpi_enabled}" = "xyes" || test "x${pack_mpi_gpu_enabled}" = "xy
AC_SUBST(CPPFLAGS_EXTRA)
AC_SUBST(CFLAGS_EXTRA)
+ # Check for OpenMP support.
+ if test "x${BUILD_OPENMP_MPI}" = "xyes"; then
+ AX_OPENMP
+
+ if test "x${OPENMP_CFLAGS}" = "x"; then
+ AC_MSG_WARN([
+-----------------------------------------------
+ Unable to find OpenMP support on this system.
+ Building a single-threaded version.
+-----------------------------------------------])
+ else
+ # bug due to recent Intel compiler change (?)
+ if test "x${ax_cv_c_compiler_vendor}" = "xintel"; then
+ OPENMP_CFLAGS="-qopenmp"
+ fi
+ OPENMP_LIBS="-lgomp"
+ fi
+
+ AC_SUBST([OMP_CFLAGS], ["${OPENMP_CFLAGS}"])
+ AC_SUBST([OMP_LIBS], ["${OPENMP_LIBS}"])
+ fi
+
# Check for MPI support.
CONFIGURE_HEADLINE([ MPI compiler ])
ACX_MPI([], [AC_MSG_ERROR([could not find MPI library])])
--
GitLab