[ViewVC] Diff of: OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1721 by gezelter, Thu May 24 14:17:42 2012 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1931 by gezelter, Mon Aug 19 19:20:32 2013 UTC

#	Line 35 \| Line 35
35		*
36		* [1] Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2] Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38	<	* [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
38	>	* [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).
39		* [4] Kuang & Gezelter, J. Chem. Phys. 133, 164101 (2010).
40		* [5] Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
#	Line 95 \| Line 95 \| namespace OpenMD {
95		storageLayout_ = sman_->getStorageLayout();
96		ff_ = info_->getForceField();
97		nLocal_ = snap_->getNumberOfAtoms();
98	<
98	>
99		nGroups_ = info_->getNLocalCutoffGroups();
100		// gather the information for atomtype IDs (atids):
101		idents = info_->getIdentArray();
102	+	regions = info_->getRegions();
103		AtomLocalToGlobal = info_->getGlobalAtomIndices();
104		cgLocalToGlobal = info_->getGlobalGroupIndices();
105		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
#	Line 109 \| Line 110 \| namespace OpenMD {
110		PairList* oneTwo = info_->getOneTwoInteractions();
111		PairList* oneThree = info_->getOneThreeInteractions();
112		PairList* oneFour = info_->getOneFourInteractions();
113	<
113	>
114	>	if (needVelocities_)
115	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition \|
116	>	DataStorage::dslVelocity);
117	>	else
118	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
119	>
120		#ifdef IS_MPI
121
122		MPI::Intracomm row = rowComm.getComm();
#	Line 145 \| Line 152 \| namespace OpenMD {
152		cgRowData.resize(nGroupsInRow_);
153		cgRowData.setStorageLayout(DataStorage::dslPosition);
154		cgColData.resize(nGroupsInCol_);
155	<	cgColData.setStorageLayout(DataStorage::dslPosition);
156	<
155	>	if (needVelocities_)
156	>	// we only need column velocities if we need them.
157	>	cgColData.setStorageLayout(DataStorage::dslPosition \|
158	>	DataStorage::dslVelocity);
159	>	else
160	>	cgColData.setStorageLayout(DataStorage::dslPosition);
161	>
162		identsRow.resize(nAtomsInRow_);
163		identsCol.resize(nAtomsInCol_);
164
165		AtomPlanIntRow->gather(idents, identsRow);
166		AtomPlanIntColumn->gather(idents, identsCol);
167	+
168	+	regionsRow.resize(nAtomsInRow_);
169	+	regionsCol.resize(nAtomsInCol_);
170
171	+	AtomPlanIntRow->gather(regions, regionsRow);
172	+	AtomPlanIntColumn->gather(regions, regionsCol);
173	+
174		// allocate memory for the parallel objects
175		atypesRow.resize(nAtomsInRow_);
176		atypesCol.resize(nAtomsInCol_);
#	Line 165 \| Line 183 \| namespace OpenMD {
183		pot_row.resize(nAtomsInRow_);
184		pot_col.resize(nAtomsInCol_);
185
186	+	expot_row.resize(nAtomsInRow_);
187	+	expot_col.resize(nAtomsInCol_);
188	+
189		AtomRowToGlobal.resize(nAtomsInRow_);
190		AtomColToGlobal.resize(nAtomsInCol_);
191		AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
#	Line 294 \| Line 315 \| namespace OpenMD {
315
316		void ForceMatrixDecomposition::createGtypeCutoffMap() {
317
318	+	GrCut.clear();
319	+	GrCutSq.clear();
320	+	GrlistSq.clear();
321	+
322		RealType tol = 1e-6;
323		largestRcut_ = 0.0;
299	–	RealType rc;
324		int atid;
325		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
326
#	Line 381 \| Line 405 \| namespace OpenMD {
405		}
406
407		bool gTypeFound = false;
408	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
408	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
409		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
410		groupToGtype[cg1] = gt;
411		gTypeFound = true;
#	Line 406 \| Line 430 \| namespace OpenMD {
430
431		RealType tradRcut = groupMax;
432
433	<	for (int i = 0; i < gTypeCutoffs.size(); i++) {
434	<	for (int j = 0; j < gTypeCutoffs.size(); j++) {
433	>	GrCut.resize( gTypeCutoffs.size() );
434	>	GrCutSq.resize( gTypeCutoffs.size() );
435	>	GrlistSq.resize( gTypeCutoffs.size() );
436	>
437	>
438	>	for (unsigned int i = 0; i < gTypeCutoffs.size(); i++) {
439	>	GrCut[i].resize( gTypeCutoffs.size() , 0.0);
440	>	GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 );
441	>	GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 );
442	>
443	>	for (unsigned int j = 0; j < gTypeCutoffs.size(); j++) {
444		RealType thisRcut;
445		switch(cutoffPolicy_) {
446		case TRADITIONAL:
#	Line 429 \| Line 462 \| namespace OpenMD {
462		break;
463		}
464
465	<	pair<int,int> key = make_pair(i,j);
433	<	gTypeCutoffMap[key].first = thisRcut;
465	>	GrCut[i][j] = thisRcut;
466		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
467	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
468	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
467	>	GrCutSq[i][j] = thisRcut * thisRcut;
468	>	GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
469	>
470	>	// pair<int,int> key = make_pair(i,j);
471	>	// gTypeCutoffMap[key].first = thisRcut;
472	>	// gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
473		// sanity check
474
475		if (userChoseCutoff_) {
476	<	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
476	>	if (abs(GrCut[i][j] - userCutoff_) > 0.0001) {
477		sprintf(painCave.errMsg,
478		"ForceMatrixDecomposition::createGtypeCutoffMap "
479		"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
#	Line 450 \| Line 486 \| namespace OpenMD {
486		}
487		}
488
489	<
454	<	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
489	>	void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) {
490		int i, j;
491		#ifdef IS_MPI
492		i = groupRowToGtype[cg1];
#	Line 460 \| Line 495 \| namespace OpenMD {
495		i = groupToGtype[cg1];
496		j = groupToGtype[cg2];
497		#endif
498	<	return gTypeCutoffMap[make_pair(i,j)];
498	>	rcut = GrCut[i][j];
499	>	rcutsq = GrCutSq[i][j];
500	>	rlistsq = GrlistSq[i][j];
501	>	return;
502	>	//return gTypeCutoffMap[make_pair(i,j)];
503		}
504
505		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
506	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
506	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
507		if (toposForAtom[atom1][j] == atom2)
508		return topoDist[atom1][j];
509	<	}
509	>	}
510		return 0;
511		}
512
513		void ForceMatrixDecomposition::zeroWorkArrays() {
514		pairwisePot = 0.0;
515		embeddingPot = 0.0;
516	+	excludedPot = 0.0;
517	+	excludedSelfPot = 0.0;
518
519		#ifdef IS_MPI
520		if (storageLayout_ & DataStorage::dslForce) {
#	Line 492 \| Line 533 \| namespace OpenMD {
533		fill(pot_col.begin(), pot_col.end(),
534		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
535
536	+	fill(expot_row.begin(), expot_row.end(),
537	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
538	+
539	+	fill(expot_col.begin(), expot_col.end(),
540	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
541	+
542		if (storageLayout_ & DataStorage::dslParticlePot) {
543		fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
544		0.0);
#	Line 539 \| Line 586 \| namespace OpenMD {
586		atomColData.electricField.end(), V3Zero);
587		}
588
542	–	if (storageLayout_ & DataStorage::dslFlucQForce) {
543	–	fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
544	–	0.0);
545	–	fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
546	–	0.0);
547	–	}
548	–
589		#endif
590		// even in parallel, we need to zero out the local arrays:
591
#	Line 600 \| Line 640 \| namespace OpenMD {
640		cgPlanVectorColumn->gather(snap_->cgData.position,
641		cgColData.position);
642
643	+
644	+
645	+	if (needVelocities_) {
646	+	// gather up the atomic velocities
647	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
648	+	atomColData.velocity);
649	+
650	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
651	+	cgColData.velocity);
652	+	}
653	+
654
655		// if needed, gather the atomic rotation matrices
656		if (storageLayout_ & DataStorage::dslAmat) {
#	Line 608 \| Line 659 \| namespace OpenMD {
659		AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
660		atomColData.aMat);
661		}
662	<
663	<	// if needed, gather the atomic eletrostatic frames
664	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
665	<	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
666	<	atomRowData.electroFrame);
667	<	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
668	<	atomColData.electroFrame);
662	>
663	>	// if needed, gather the atomic eletrostatic information
664	>	if (storageLayout_ & DataStorage::dslDipole) {
665	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
666	>	atomRowData.dipole);
667	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
668	>	atomColData.dipole);
669		}
670
671	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
672	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
673	+	atomRowData.quadrupole);
674	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
675	+	atomColData.quadrupole);
676	+	}
677	+
678		// if needed, gather the atomic fluctuating charge values
679		if (storageLayout_ & DataStorage::dslFlucQPosition) {
680		AtomPlanRealRow->gather(snap_->atomData.flucQPos,
#	Line 648 \| Line 706 \| namespace OpenMD {
706		snap_->atomData.density[i] += rho_tmp[i];
707		}
708
709	+	// this isn't necessary if we don't have polarizable atoms, but
710	+	// we'll leave it here for now.
711		if (storageLayout_ & DataStorage::dslElectricField) {
712
713		AtomPlanVectorRow->scatter(atomRowData.electricField,
#	Line 655 \| Line 715 \| namespace OpenMD {
715
716		int n = snap_->atomData.electricField.size();
717		vector<Vector3d> field_tmp(n, V3Zero);
718	<	AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
718	>	AtomPlanVectorColumn->scatter(atomColData.electricField,
719	>	field_tmp);
720		for (int i = 0; i < n; i++)
721		snap_->atomData.electricField[i] += field_tmp[i];
722		}
#	Line 755 \| Line 816 \| namespace OpenMD {
816
817		}
818
819	+	if (storageLayout_ & DataStorage::dslElectricField) {
820	+
821	+	int nef = snap_->atomData.electricField.size();
822	+	vector<Vector3d> efield_tmp(nef, V3Zero);
823	+
824	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
825	+	for (int i = 0; i < nef; i++) {
826	+	snap_->atomData.electricField[i] += efield_tmp[i];
827	+	efield_tmp[i] = 0.0;
828	+	}
829	+
830	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
831	+	for (int i = 0; i < nef; i++)
832	+	snap_->atomData.electricField[i] += efield_tmp[i];
833	+	}
834	+
835	+
836		nLocal_ = snap_->getNumberOfAtoms();
837
838		vector<potVec> pot_temp(nLocal_,
839		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
840	+	vector<potVec> expot_temp(nLocal_,
841	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
842
843		// scatter/gather pot_row into the members of my column
844
845		AtomPlanPotRow->scatter(pot_row, pot_temp);
846	+	AtomPlanPotRow->scatter(expot_row, expot_temp);
847
848	<	for (int ii = 0; ii < pot_temp.size(); ii++ )
848	>	for (int ii = 0; ii < pot_temp.size(); ii++ )
849		pairwisePot += pot_temp[ii];
850	<
850	>
851	>	for (int ii = 0; ii < expot_temp.size(); ii++ )
852	>	excludedPot += expot_temp[ii];
853	>
854	>	if (storageLayout_ & DataStorage::dslParticlePot) {
855	>	// This is the pairwise contribution to the particle pot. The
856	>	// embedding contribution is added in each of the low level
857	>	// non-bonded routines. In single processor, this is done in
858	>	// unpackInteractionData, not in collectData.
859	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
860	>	for (int i = 0; i < nLocal_; i++) {
861	>	// factor of two is because the total potential terms are divided
862	>	// by 2 in parallel due to row/ column scatter
863	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
864	>	}
865	>	}
866	>	}
867	>
868		fill(pot_temp.begin(), pot_temp.end(),
869		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
870	+	fill(expot_temp.begin(), expot_temp.end(),
871	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
872
873		AtomPlanPotColumn->scatter(pot_col, pot_temp);
874	+	AtomPlanPotColumn->scatter(expot_col, expot_temp);
875
876		for (int ii = 0; ii < pot_temp.size(); ii++ )
877		pairwisePot += pot_temp[ii];
878	+
879	+	for (int ii = 0; ii < expot_temp.size(); ii++ )
880	+	excludedPot += expot_temp[ii];
881	+
882	+	if (storageLayout_ & DataStorage::dslParticlePot) {
883	+	// This is the pairwise contribution to the particle pot. The
884	+	// embedding contribution is added in each of the low level
885	+	// non-bonded routines. In single processor, this is done in
886	+	// unpackInteractionData, not in collectData.
887	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
888	+	for (int i = 0; i < nLocal_; i++) {
889	+	// factor of two is because the total potential terms are divided
890	+	// by 2 in parallel due to row/ column scatter
891	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
892	+	}
893	+	}
894	+	}
895
896	+	if (storageLayout_ & DataStorage::dslParticlePot) {
897	+	int npp = snap_->atomData.particlePot.size();
898	+	vector<RealType> ppot_temp(npp, 0.0);
899	+
900	+	// This is the direct or embedding contribution to the particle
901	+	// pot.
902	+
903	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
904	+	for (int i = 0; i < npp; i++) {
905	+	snap_->atomData.particlePot[i] += ppot_temp[i];
906	+	}
907	+
908	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
909	+
910	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
911	+	for (int i = 0; i < npp; i++) {
912	+	snap_->atomData.particlePot[i] += ppot_temp[i];
913	+	}
914	+	}
915	+
916		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
917		RealType ploc1 = pairwisePot[ii];
918		RealType ploc2 = 0.0;
#	Line 783 \| Line 921 \| namespace OpenMD {
921		}
922
923		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
924	<	RealType ploc1 = embeddingPot[ii];
924	>	RealType ploc1 = excludedPot[ii];
925		RealType ploc2 = 0.0;
926		MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
927	<	embeddingPot[ii] = ploc2;
927	>	excludedPot[ii] = ploc2;
928		}
929
930	+	// Here be dragons.
931	+	MPI::Intracomm col = colComm.getComm();
932	+
933	+	col.Allreduce(MPI::IN_PLACE,
934	+	&snap_->frameData.conductiveHeatFlux[0], 3,
935	+	MPI::REALTYPE, MPI::SUM);
936	+
937	+
938		#endif
939
940		}
941
942	<	int ForceMatrixDecomposition::getNAtomsInRow() {
942	>	/**
943	>	* Collects information obtained during the post-pair (and embedding
944	>	* functional) loops onto local data structures.
945	>	*/
946	>	void ForceMatrixDecomposition::collectSelfData() {
947	>	snap_ = sman_->getCurrentSnapshot();
948	>	storageLayout_ = sman_->getStorageLayout();
949	>
950		#ifdef IS_MPI
951	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
952	+	RealType ploc1 = embeddingPot[ii];
953	+	RealType ploc2 = 0.0;
954	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
955	+	embeddingPot[ii] = ploc2;
956	+	}
957	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
958	+	RealType ploc1 = excludedSelfPot[ii];
959	+	RealType ploc2 = 0.0;
960	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
961	+	excludedSelfPot[ii] = ploc2;
962	+	}
963	+	#endif
964	+
965	+	}
966	+
967	+
968	+
969	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
970	+	#ifdef IS_MPI
971		return nAtomsInRow_;
972		#else
973		return nLocal_;
#	Line 804 \| Line 977 \| namespace OpenMD {
977		/**
978		* returns the list of atoms belonging to this group.
979		*/
980	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
980	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
981		#ifdef IS_MPI
982		return groupListRow_[cg1];
983		#else
#	Line 812 \| Line 985 \| namespace OpenMD {
985		#endif
986		}
987
988	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
988	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
989		#ifdef IS_MPI
990		return groupListCol_[cg2];
991		#else
#	Line 829 \| Line 1002 \| namespace OpenMD {
1002		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
1003		#endif
1004
1005	<	snap_->wrapVector(d);
1005	>	if (usePeriodicBoundaryConditions_) {
1006	>	snap_->wrapVector(d);
1007	>	}
1008		return d;
1009		}
1010
1011	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1012	+	#ifdef IS_MPI
1013	+	return cgColData.velocity[cg2];
1014	+	#else
1015	+	return snap_->cgData.velocity[cg2];
1016	+	#endif
1017	+	}
1018
1019	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1020	+	#ifdef IS_MPI
1021	+	return atomColData.velocity[atom2];
1022	+	#else
1023	+	return snap_->atomData.velocity[atom2];
1024	+	#endif
1025	+	}
1026	+
1027	+
1028		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1029
1030		Vector3d d;
#	Line 843 \| Line 1034 \| namespace OpenMD {
1034		#else
1035		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1036		#endif
1037	<
1038	<	snap_->wrapVector(d);
1037	>	if (usePeriodicBoundaryConditions_) {
1038	>	snap_->wrapVector(d);
1039	>	}
1040		return d;
1041		}
1042
#	Line 856 \| Line 1048 \| namespace OpenMD {
1048		#else
1049		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1050		#endif
1051	<
1052	<	snap_->wrapVector(d);
1051	>	if (usePeriodicBoundaryConditions_) {
1052	>	snap_->wrapVector(d);
1053	>	}
1054		return d;
1055		}
1056
1057	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1057	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1058		#ifdef IS_MPI
1059		return massFactorsRow[atom1];
1060		#else
#	Line 869 \| Line 1062 \| namespace OpenMD {
1062		#endif
1063		}
1064
1065	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1065	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1066		#ifdef IS_MPI
1067		return massFactorsCol[atom2];
1068		#else
#	Line 886 \| Line 1079 \| namespace OpenMD {
1079		#else
1080		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1081		#endif
1082	<
1083	<	snap_->wrapVector(d);
1082	>	if (usePeriodicBoundaryConditions_) {
1083	>	snap_->wrapVector(d);
1084	>	}
1085		return d;
1086		}
1087
1088	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1088	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1089		return excludesForAtom[atom1];
1090		}
1091
#	Line 899 \| Line 1093 \| namespace OpenMD {
1093		* We need to exclude some overcounted interactions that result from
1094		* the parallel decomposition.
1095		*/
1096	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1096	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1097		int unique_id_1, unique_id_2;
1098
1099		#ifdef IS_MPI
1100		// in MPI, we have to look up the unique IDs for each atom
1101		unique_id_1 = AtomRowToGlobal[atom1];
1102		unique_id_2 = AtomColToGlobal[atom2];
1103	+	// group1 = cgRowToGlobal[cg1];
1104	+	// group2 = cgColToGlobal[cg2];
1105		#else
1106		unique_id_1 = AtomLocalToGlobal[atom1];
1107		unique_id_2 = AtomLocalToGlobal[atom2];
1108	+	int group1 = cgLocalToGlobal[cg1];
1109	+	int group2 = cgLocalToGlobal[cg2];
1110		#endif
1111
1112		if (unique_id_1 == unique_id_2) return true;
#	Line 920 \| Line 1118 \| namespace OpenMD {
1118		} else {
1119		if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1120		}
1121	+	#endif
1122	+
1123	+	#ifndef IS_MPI
1124	+	if (group1 == group2) {
1125	+	if (unique_id_1 < unique_id_2) return true;
1126	+	}
1127		#endif
1128
1129		return false;
#	Line 971 \| Line 1175 \| namespace OpenMD {
1175		idat.excluded = excludeAtomPair(atom1, atom2);
1176
1177		#ifdef IS_MPI
1178	<	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1179	<	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1180	<	// ff_->getAtomType(identsCol[atom2]) );
1181	<
1178	>	//idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1179	>	idat.atid1 = identsRow[atom1];
1180	>	idat.atid2 = identsCol[atom2];
1181	>
1182	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1183	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1184	>	} else {
1185	>	idat.sameRegion = false;
1186	>	}
1187	>
1188		if (storageLayout_ & DataStorage::dslAmat) {
1189		idat.A1 = &(atomRowData.aMat[atom1]);
1190		idat.A2 = &(atomColData.aMat[atom2]);
1191		}
1192
983	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
984	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
985	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
986	–	}
987	–
1193		if (storageLayout_ & DataStorage::dslTorque) {
1194		idat.t1 = &(atomRowData.torque[atom1]);
1195		idat.t2 = &(atomColData.torque[atom2]);
1196		}
1197
1198	+	if (storageLayout_ & DataStorage::dslDipole) {
1199	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1200	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1201	+	}
1202	+
1203	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1204	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1205	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1206	+	}
1207	+
1208		if (storageLayout_ & DataStorage::dslDensity) {
1209		idat.rho1 = &(atomRowData.density[atom1]);
1210		idat.rho2 = &(atomColData.density[atom2]);
#	Line 1022 \| Line 1237 \| namespace OpenMD {
1237
1238		#else
1239
1240	+	//idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1241	+	idat.atid1 = idents[atom1];
1242	+	idat.atid2 = idents[atom2];
1243
1244	<	// cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1245	<	// cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
1246	<	// cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1244	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1245	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1246	>	} else {
1247	>	idat.sameRegion = false;
1248	>	}
1249
1030	–	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1031	–	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1032	–	// ff_->getAtomType(idents[atom2]) );
1033	–
1250		if (storageLayout_ & DataStorage::dslAmat) {
1251		idat.A1 = &(snap_->atomData.aMat[atom1]);
1252		idat.A2 = &(snap_->atomData.aMat[atom2]);
1253		}
1254
1039	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
1040	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1041	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1042	–	}
1043	–
1255		if (storageLayout_ & DataStorage::dslTorque) {
1256		idat.t1 = &(snap_->atomData.torque[atom1]);
1257		idat.t2 = &(snap_->atomData.torque[atom2]);
1258		}
1259
1260	+	if (storageLayout_ & DataStorage::dslDipole) {
1261	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1262	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1263	+	}
1264	+
1265	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1266	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1267	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1268	+	}
1269	+
1270		if (storageLayout_ & DataStorage::dslDensity) {
1271		idat.rho1 = &(snap_->atomData.density[atom1]);
1272		idat.rho2 = &(snap_->atomData.density[atom2]);
#	Line 1084 \| Line 1305 \| namespace OpenMD {
1305		#ifdef IS_MPI
1306		pot_row[atom1] += RealType(0.5) * *(idat.pot);
1307		pot_col[atom2] += RealType(0.5) * *(idat.pot);
1308	+	expot_row[atom1] += RealType(0.5) * *(idat.excludedPot);
1309	+	expot_col[atom2] += RealType(0.5) * *(idat.excludedPot);
1310
1311		atomRowData.force[atom1] += *(idat.f1);
1312		atomColData.force[atom2] -= *(idat.f1);
1313
1314		if (storageLayout_ & DataStorage::dslFlucQForce) {
1315	<	atomRowData.flucQFrc[atom1] += *(idat.dVdFQ1);
1316	<	atomColData.flucQFrc[atom2] += *(idat.dVdFQ2);
1315	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1316	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1317		}
1318
1319		if (storageLayout_ & DataStorage::dslElectricField) {
#	Line 1098 \| Line 1321 \| namespace OpenMD {
1321		atomColData.electricField[atom2] += *(idat.eField2);
1322		}
1323
1101	–	// should particle pot be done here also?
1324		#else
1325		pairwisePot += *(idat.pot);
1326	+	excludedPot += *(idat.excludedPot);
1327
1328		snap_->atomData.force[atom1] += *(idat.f1);
1329		snap_->atomData.force[atom2] -= *(idat.f1);
1330
1331		if (idat.doParticlePot) {
1332	+	// This is the pairwise contribution to the particle pot. The
1333	+	// embedding contribution is added in each of the low level
1334	+	// non-bonded routines. In parallel, this calculation is done
1335	+	// in collectData, not in unpackInteractionData.
1336		snap_->atomData.particlePot[atom1] += (idat.vpair) *(idat.sw);
1337	<	snap_->atomData.particlePot[atom2] -= (idat.vpair) *(idat.sw);
1337	>	snap_->atomData.particlePot[atom2] += (idat.vpair) *(idat.sw);
1338		}
1339
1340		if (storageLayout_ & DataStorage::dslFlucQForce) {
1341	<	snap_->atomData.flucQFrc[atom1] += *(idat.dVdFQ1);
1341	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1342		snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1343		}
1344
#	Line 1130 \| Line 1357 \| namespace OpenMD {
1357		* first element of pair is row-indexed CutoffGroup
1358		* second element of pair is column-indexed CutoffGroup
1359		*/
1360	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1361	<
1362	<	vector<pair<int, int> > neighborList;
1360	>	void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1361	>
1362	>	neighborList.clear();
1363		groupCutoffs cuts;
1364		bool doAllPairs = false;
1365
1366	+	RealType rList_ = (largestRcut_ + skinThickness_);
1367	+	RealType rcut, rcutsq, rlistsq;
1368	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1369	+	Mat3x3d box;
1370	+	Mat3x3d invBox;
1371	+
1372	+	Vector3d rs, scaled, dr;
1373	+	Vector3i whichCell;
1374	+	int cellIndex;
1375	+
1376		#ifdef IS_MPI
1377		cellListRow_.clear();
1378		cellListCol_.clear();
1379		#else
1380		cellList_.clear();
1381		#endif
1382	<
1383	<	RealType rList_ = (largestRcut_ + skinThickness_);
1384	<	RealType rl2 = rList_ * rList_;
1385	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1386	<	Mat3x3d Hmat = snap_->getHmat();
1387	<	Vector3d Hx = Hmat.getColumn(0);
1388	<	Vector3d Hy = Hmat.getColumn(1);
1389	<	Vector3d Hz = Hmat.getColumn(2);
1390	<
1391	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1392	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1393	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1394	<
1382	>
1383	>	if (!usePeriodicBoundaryConditions_) {
1384	>	box = snap_->getBoundingBox();
1385	>	invBox = snap_->getInvBoundingBox();
1386	>	} else {
1387	>	box = snap_->getHmat();
1388	>	invBox = snap_->getInvHmat();
1389	>	}
1390	>
1391	>	Vector3d boxX = box.getColumn(0);
1392	>	Vector3d boxY = box.getColumn(1);
1393	>	Vector3d boxZ = box.getColumn(2);
1394	>
1395	>	nCells_.x() = (int) ( boxX.length() )/ rList_;
1396	>	nCells_.y() = (int) ( boxY.length() )/ rList_;
1397	>	nCells_.z() = (int) ( boxZ.length() )/ rList_;
1398	>
1399		// handle small boxes where the cell offsets can end up repeating cells
1400
1401		if (nCells_.x() < 3) doAllPairs = true;
1402		if (nCells_.y() < 3) doAllPairs = true;
1403		if (nCells_.z() < 3) doAllPairs = true;
1404	<
1164	<	Mat3x3d invHmat = snap_->getInvHmat();
1165	<	Vector3d rs, scaled, dr;
1166	<	Vector3i whichCell;
1167	<	int cellIndex;
1404	>
1405		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1406	<
1406	>
1407		#ifdef IS_MPI
1408		cellListRow_.resize(nCtot);
1409		cellListCol_.resize(nCtot);
1410		#else
1411		cellList_.resize(nCtot);
1412		#endif
1413	<
1413	>
1414		if (!doAllPairs) {
1415		#ifdef IS_MPI
1416	<
1416	>
1417		for (int i = 0; i < nGroupsInRow_; i++) {
1418		rs = cgRowData.position[i];
1419
1420		// scaled positions relative to the box vectors
1421	<	scaled = invHmat * rs;
1421	>	scaled = invBox * rs;
1422
1423		// wrap the vector back into the unit box by subtracting integer box
1424		// numbers
1425		for (int j = 0; j < 3; j++) {
1426		scaled[j] -= roundMe(scaled[j]);
1427		scaled[j] += 0.5;
1428	+	// Handle the special case when an object is exactly on the
1429	+	// boundary (a scaled coordinate of 1.0 is the same as
1430	+	// scaled coordinate of 0.0)
1431	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1432		}
1433
1434		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1205 \| Line 1446 \| namespace OpenMD {
1446		rs = cgColData.position[i];
1447
1448		// scaled positions relative to the box vectors
1449	<	scaled = invHmat * rs;
1449	>	scaled = invBox * rs;
1450
1451		// wrap the vector back into the unit box by subtracting integer box
1452		// numbers
1453		for (int j = 0; j < 3; j++) {
1454		scaled[j] -= roundMe(scaled[j]);
1455		scaled[j] += 0.5;
1456	+	// Handle the special case when an object is exactly on the
1457	+	// boundary (a scaled coordinate of 1.0 is the same as
1458	+	// scaled coordinate of 0.0)
1459	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1460		}
1461
1462		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1225 \| Line 1470 \| namespace OpenMD {
1470		// add this cutoff group to the list of groups in this cell;
1471		cellListCol_[cellIndex].push_back(i);
1472		}
1473	<
1473	>
1474		#else
1475		for (int i = 0; i < nGroups_; i++) {
1476		rs = snap_->cgData.position[i];
1477
1478		// scaled positions relative to the box vectors
1479	<	scaled = invHmat * rs;
1479	>	scaled = invBox * rs;
1480
1481		// wrap the vector back into the unit box by subtracting integer box
1482		// numbers
1483		for (int j = 0; j < 3; j++) {
1484		scaled[j] -= roundMe(scaled[j]);
1485		scaled[j] += 0.5;
1486	+	// Handle the special case when an object is exactly on the
1487	+	// boundary (a scaled coordinate of 1.0 is the same as
1488	+	// scaled coordinate of 0.0)
1489	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1490		}
1491
1492		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1296 \| Line 1545 \| namespace OpenMD {
1545		// & column indicies and will divide labor in the
1546		// force evaluation later.
1547		dr = cgColData.position[(j2)] - cgRowData.position[(j1)];
1548	<	snap_->wrapVector(dr);
1549	<	cuts = getGroupCutoffs( (j1), (j2) );
1550	<	if (dr.lengthSquare() < cuts.third) {
1548	>	if (usePeriodicBoundaryConditions_) {
1549	>	snap_->wrapVector(dr);
1550	>	}
1551	>	getGroupCutoffs( (j1), (j2), rcut, rcutsq, rlistsq );
1552	>	if (dr.lengthSquare() < rlistsq) {
1553		neighborList.push_back(make_pair((j1), (j2)));
1554		}
1555		}
#	Line 1318 \| Line 1569 \| namespace OpenMD {
1569		// allows atoms within a single cutoff group to
1570		// interact with each other.
1571
1321	–
1322	–
1572		if (m2 != m1 \|\| (j2) >= (j1) ) {
1573
1574		dr = snap_->cgData.position[(j2)] - snap_->cgData.position[(j1)];
1575	<	snap_->wrapVector(dr);
1576	<	cuts = getGroupCutoffs( (j1), (j2) );
1577	<	if (dr.lengthSquare() < cuts.third) {
1575	>	if (usePeriodicBoundaryConditions_) {
1576	>	snap_->wrapVector(dr);
1577	>	}
1578	>	getGroupCutoffs( (j1), (j2), rcut, rcutsq, rlistsq );
1579	>	if (dr.lengthSquare() < rlistsq) {
1580		neighborList.push_back(make_pair((j1), (j2)));
1581		}
1582		}
#	Line 1342 \| Line 1593 \| namespace OpenMD {
1593		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1594		for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1595		dr = cgColData.position[j2] - cgRowData.position[j1];
1596	<	snap_->wrapVector(dr);
1597	<	cuts = getGroupCutoffs( j1, j2 );
1598	<	if (dr.lengthSquare() < cuts.third) {
1596	>	if (usePeriodicBoundaryConditions_) {
1597	>	snap_->wrapVector(dr);
1598	>	}
1599	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1600	>	if (dr.lengthSquare() < rlistsq) {
1601		neighborList.push_back(make_pair(j1, j2));
1602		}
1603		}
#	Line 1355 \| Line 1608 \| namespace OpenMD {
1608		// include self group interactions j2 == j1
1609		for (int j2 = j1; j2 < nGroups_; j2++) {
1610		dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1611	<	snap_->wrapVector(dr);
1612	<	cuts = getGroupCutoffs( j1, j2 );
1613	<	if (dr.lengthSquare() < cuts.third) {
1611	>	if (usePeriodicBoundaryConditions_) {
1612	>	snap_->wrapVector(dr);
1613	>	}
1614	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1615	>	if (dr.lengthSquare() < rlistsq) {
1616		neighborList.push_back(make_pair(j1, j2));
1617		}
1618		}
#	Line 1370 \| Line 1625 \| namespace OpenMD {
1625		saved_CG_positions_.clear();
1626		for (int i = 0; i < nGroups_; i++)
1627		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1373	–
1374	–	return neighborList;
1628		}
1629		} //end namespace OpenMD

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing: branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1721 by gezelter, Thu May 24 14:17:42 2012 UTC vs. trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1931 by gezelter, Mon Aug 19 19:20:32 2013 UTC

Diff Legend

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1721 by gezelter, Thu May 24 14:17:42 2012 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1931 by gezelter, Mon Aug 19 19:20:32 2013 UTC