Merge pull request #536 from akohlmey/fix-nvcc-openmp-conflicts (326a8a12) · Commits · 郑智淋 / lammps

src/USER-OMP/pair_comb_omp.cpp

+3 −3

Original line number	Diff line number	Diff line
		@@ -484,7 +484,7 @@ double PairCombOMP::yasu_char(double *qf_fix, int &igroup)

		qfo_field(&params[iparam_ij],rsq1,iq,jq,fqji,fqjj);
		fqi += jq * fqij + fqji;
		#if defined(_OPENMP)
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp atomic
		#endif
		qf[j] += (iq * fqij + fqjj);
		@@ -511,13 +511,13 @@ double PairCombOMP::yasu_char(double *qf_fix, int &igroup)

		qfo_short(&params[iparam_ij],i,nj,rsq1,iq,jq,fqij,fqjj);
		fqi += fqij;
		#if defined(_OPENMP)
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp atomic
		#endif
		qf[j] += fqjj;
		}

		#if defined(_OPENMP)
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp atomic
		#endif
		qf[i] += fqi;

src/USER-OMP/reaxc_torsion_angles_omp.cpp

+7 −1

Original line number	Diff line number	Diff line
		@@ -69,7 +69,9 @@ void Torsion_AnglesOMP( reax_system system, control_params control,
		double total_Econ = 0;
		int nthreads = control->nthreads;

		#if defined(_OPENMP)
		#pragma omp parallel default(shared) reduction(+: total_Etor, total_Econ)
		#endif
		{
		int i, j, k, l, pi, pj, pk, pl, pij, plk;
		int type_i, type_j, type_k, type_l;
		@@ -125,7 +127,9 @@ void Torsion_AnglesOMP( reax_system system, control_params control,
		system->N, system->pair_ptr->eatom,
		system->pair_ptr->vatom, thr);

		#if defined(_OPENMP)
		#pragma omp for schedule(static)
		#endif
		for (j = 0; j < system->N; ++j) {
		start_j = Start_Index(j, bonds);
		end_j = End_Index(j, bonds);
		@@ -137,7 +141,9 @@ void Torsion_AnglesOMP( reax_system system, control_params control,
		}
		}

		#if defined(_OPENMP)
		#pragma omp for schedule(dynamic,50)
		#endif
		for (j = 0; j < natoms; ++j) {
		type_j = system->my_atoms[j].type;
		Delta_j = workspace->Delta_boc[j];

src/USER-OMP/reaxc_valence_angles_omp.cpp

+12 −3

Original line number	Diff line number	Diff line
		@@ -124,8 +124,9 @@ void Valence_AnglesOMP( reax_system system, control_params control,
		int nthreads = control->nthreads;
		int num_thb_intrs = 0;
		int TWICE = 2;

		#if defined(_OPENMP)
		#pragma omp parallel default(shared) reduction(+:total_Eang, total_Epen, total_Ecoa, num_thb_intrs)
		#endif
		{
		int i, j, pi, k, pk, t;
		int type_i, type_j, type_k;
		@@ -180,7 +181,9 @@ void Valence_AnglesOMP( reax_system system, control_params control,

		const int per_thread = thb_intrs->num_intrs / nthreads;

		#if defined(_OPENMP)
		#pragma omp for schedule(dynamic,50)
		#endif
		for (j = 0; j < system->N; ++j) {
		type_j = system->my_atoms[j].type;
		_my_offset[j] = 0;
		@@ -251,11 +254,14 @@ void Valence_AnglesOMP( reax_system system, control_params control,
		} // for(j)

		// Wait for all threads to finish counting angles
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp barrier

		#endif
		// Master thread uses angle counts to compute offsets
		// This can be threaded
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp master
		#endif
		{
		int current_count = 0;
		int m = _my_offset[0];
		@@ -269,12 +275,15 @@ void Valence_AnglesOMP( reax_system system, control_params control,
		}

		// All threads wait till master thread finished computing offsets
		#if defined(_OPENMP) && !defined(__NVCC__)
		#pragma omp barrier

		#endif
		// Original loop, but now using precomputed offsets
		// Safe to use all threads available, regardless of threads tasked above
		// We also now skip over atoms that have no angles assigned
		#if defined(_OPENMP)
		#pragma omp for schedule(dynamic,50)//(dynamic,chunksize)//(guided)
		#endif
		for (j = 0; j < system->N; ++j) { // Ray: the first one with system->N
		type_j = system->my_atoms[j].type;
		if(type_j < 0) continue;

Admin message