1 |
< |
/** |
2 |
< |
* @file ForceDecomposition.cpp |
3 |
< |
* @author Charles Vardeman <cvardema.at.nd.edu> |
4 |
< |
* @date 08/18/2010 |
5 |
< |
* @time 11:56am |
6 |
< |
* @version 1.0 |
1 |
> |
/* |
2 |
> |
* Copyright (c) 2005 The University of Notre Dame. All Rights Reserved. |
3 |
|
* |
8 |
– |
* @section LICENSE |
9 |
– |
* Copyright (c) 2010 The University of Notre Dame. All Rights Reserved. |
10 |
– |
* |
4 |
|
* The University of Notre Dame grants you ("Licensee") a |
5 |
|
* non-exclusive, royalty free, license to use, modify and |
6 |
|
* redistribute this software in source and binary code form, provided |
38 |
|
* [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008). |
39 |
|
* [4] Vardeman & Gezelter, in progress (2009). |
40 |
|
*/ |
41 |
+ |
#include "parallel/ForceDecomposition.hpp" |
42 |
+ |
#include "math/SquareMatrix3.hpp" |
43 |
+ |
#include "nonbonded/NonBondedInteraction.hpp" |
44 |
+ |
#include "brains/SnapshotManager.hpp" |
45 |
|
|
46 |
+ |
using namespace std; |
47 |
+ |
namespace OpenMD { |
48 |
|
|
49 |
+ |
/** |
50 |
+ |
* distributeInitialData is essentially a copy of the older fortran |
51 |
+ |
* SimulationSetup |
52 |
+ |
*/ |
53 |
+ |
|
54 |
+ |
void ForceDecomposition::distributeInitialData() { |
55 |
+ |
#ifdef IS_MPI |
56 |
+ |
Snapshot* snap = sman_->getCurrentSnapshot(); |
57 |
+ |
int nLocal = snap->getNumberOfAtoms(); |
58 |
+ |
int nGroups = snap->getNumberOfCutoffGroups(); |
59 |
|
|
60 |
< |
/* -*- c++ -*- */ |
61 |
< |
#include "config.h" |
62 |
< |
#include <stdlib.h> |
63 |
< |
#ifdef IS_MPI |
55 |
< |
#include <mpi.h> |
56 |
< |
#endif |
60 |
> |
AtomCommIntI = new Communicator<Row,int>(nLocal); |
61 |
> |
AtomCommRealI = new Communicator<Row,RealType>(nLocal); |
62 |
> |
AtomCommVectorI = new Communicator<Row,Vector3d>(nLocal); |
63 |
> |
AtomCommMatrixI = new Communicator<Row,Mat3x3d>(nLocal); |
64 |
|
|
65 |
< |
#include <iostream> |
66 |
< |
#include <vector> |
67 |
< |
#include <algorithm> |
68 |
< |
#include <cmath> |
62 |
< |
#include "parallel/ForceDecomposition.hpp" |
65 |
> |
AtomCommIntJ = new Communicator<Column,int>(nLocal); |
66 |
> |
AtomCommRealJ = new Communicator<Column,RealType>(nLocal); |
67 |
> |
AtomCommVectorJ = new Communicator<Column,Vector3d>(nLocal); |
68 |
> |
AtomCommMatrixJ = new Communicator<Column,Mat3x3d>(nLocal); |
69 |
|
|
70 |
+ |
cgCommIntI = new Communicator<Row,int>(nGroups); |
71 |
+ |
cgCommVectorI = new Communicator<Row,Vector3d>(nGroups); |
72 |
+ |
cgCommIntJ = new Communicator<Column,int>(nGroups); |
73 |
+ |
cgCommVectorJ = new Communicator<Column,Vector3d>(nGroups); |
74 |
|
|
75 |
< |
using namespace std; |
76 |
< |
using namespace OpenMD; |
75 |
> |
int nAtomsInRow = AtomCommIntI->getSize(); |
76 |
> |
int nAtomsInCol = AtomCommIntJ->getSize(); |
77 |
> |
int nGroupsInRow = cgCommIntI->getSize(); |
78 |
> |
int nGroupsInCol = cgCommIntJ->getSize(); |
79 |
|
|
80 |
< |
//__static |
81 |
< |
#ifdef IS_MPI |
82 |
< |
static vector<MPI:Comm> communictors; |
83 |
< |
#endif |
80 |
> |
vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES, |
81 |
> |
vector<RealType> (nAtomsInRow, 0.0)); |
82 |
> |
vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES, |
83 |
> |
vector<RealType> (nAtomsInCol, 0.0)); |
84 |
> |
|
85 |
> |
vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0); |
86 |
|
|
87 |
< |
//____ MPITypeTraits |
88 |
< |
template<typename T> |
89 |
< |
struct MPITypeTraits; |
87 |
> |
// gather the information for atomtype IDs (atids): |
88 |
> |
AtomCommIntI->gather(info_->getIdentArray(), identsRow); |
89 |
> |
AtomCommIntJ->gather(info_->getIdentArray(), identsCol); |
90 |
|
|
91 |
< |
#ifdef IS_MPI |
92 |
< |
template<> |
93 |
< |
struct MPITypeTraits<RealType> { |
80 |
< |
static const MPI::Datatype datatype; |
81 |
< |
}; |
82 |
< |
const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL; |
91 |
> |
AtomLocalToGlobal = info_->getLocalToGlobalAtomIndex(); |
92 |
> |
AtomCommIntI->gather(AtomLocalToGlobal, AtomRowToGlobal); |
93 |
> |
AtomCommIntJ->gather(AtomLocalToGlobal, AtomColToGlobal); |
94 |
|
|
95 |
< |
template<> |
96 |
< |
struct MPITypeTraits<int> { |
97 |
< |
static const MPI::Datatype datatype; |
87 |
< |
}; |
88 |
< |
const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT; |
89 |
< |
#endif |
95 |
> |
cgLocalToGlobal = info_->getLocalToGlobalCutoffGroupIndex(); |
96 |
> |
cgCommIntI->gather(cgLocalToGlobal, cgRowToGlobal); |
97 |
> |
cgCommIntJ->gather(cgLocalToGlobal, cgColToGlobal); |
98 |
|
|
99 |
< |
/** |
100 |
< |
* Constructor for ForceDecomposition Parallel Decomposition Method |
93 |
< |
* Will try to construct a symmetric grid of processors. Ideally, the |
94 |
< |
* number of processors will be a square ex: 4, 9, 16, 25. |
95 |
< |
* |
96 |
< |
*/ |
99 |
> |
|
100 |
> |
|
101 |
|
|
98 |
– |
ForceDecomposition::ForceDecomposition() { |
102 |
|
|
103 |
< |
#ifdef IS_MPI |
104 |
< |
int nProcs = MPI::COMM_WORLD.Get_size(); |
105 |
< |
int worldRank = MPI::COMM_WORLD.Get_rank(); |
103 |
> |
|
104 |
> |
// still need: |
105 |
> |
// topoDist |
106 |
> |
// exclude |
107 |
|
#endif |
108 |
+ |
} |
109 |
+ |
|
110 |
|
|
111 |
< |
// First time through, construct column stride. |
112 |
< |
if (communicators.size() == 0) |
113 |
< |
{ |
114 |
< |
int nColumnsMax = (int) round(sqrt((float) nProcs)); |
115 |
< |
for (int i = 0; i < nProcs; ++i) |
116 |
< |
{ |
117 |
< |
if (nProcs%i==0) nColumns=i; |
111 |
> |
|
112 |
> |
void ForceDecomposition::distributeData() { |
113 |
> |
#ifdef IS_MPI |
114 |
> |
Snapshot* snap = sman_->getCurrentSnapshot(); |
115 |
> |
|
116 |
> |
// gather up the atomic positions |
117 |
> |
AtomCommVectorI->gather(snap->atomData.position, |
118 |
> |
snap->atomIData.position); |
119 |
> |
AtomCommVectorJ->gather(snap->atomData.position, |
120 |
> |
snap->atomJData.position); |
121 |
> |
|
122 |
> |
// gather up the cutoff group positions |
123 |
> |
cgCommVectorI->gather(snap->cgData.position, |
124 |
> |
snap->cgIData.position); |
125 |
> |
cgCommVectorJ->gather(snap->cgData.position, |
126 |
> |
snap->cgJData.position); |
127 |
> |
|
128 |
> |
// if needed, gather the atomic rotation matrices |
129 |
> |
if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) { |
130 |
> |
AtomCommMatrixI->gather(snap->atomData.aMat, |
131 |
> |
snap->atomIData.aMat); |
132 |
> |
AtomCommMatrixJ->gather(snap->atomData.aMat, |
133 |
> |
snap->atomJData.aMat); |
134 |
|
} |
135 |
+ |
|
136 |
+ |
// if needed, gather the atomic eletrostatic frames |
137 |
+ |
if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) { |
138 |
+ |
AtomCommMatrixI->gather(snap->atomData.electroFrame, |
139 |
+ |
snap->atomIData.electroFrame); |
140 |
+ |
AtomCommMatrixJ->gather(snap->atomData.electroFrame, |
141 |
+ |
snap->atomJData.electroFrame); |
142 |
+ |
} |
143 |
+ |
#endif |
144 |
+ |
} |
145 |
+ |
|
146 |
+ |
void ForceDecomposition::collectIntermediateData() { |
147 |
+ |
#ifdef IS_MPI |
148 |
+ |
Snapshot* snap = sman_->getCurrentSnapshot(); |
149 |
+ |
|
150 |
+ |
if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) { |
151 |
|
|
152 |
< |
int nRows = nProcs/nColumns; |
153 |
< |
myRank_ = (int) worldRank%nColumns; |
152 |
> |
AtomCommRealI->scatter(snap->atomIData.density, |
153 |
> |
snap->atomData.density); |
154 |
> |
|
155 |
> |
int n = snap->atomData.density.size(); |
156 |
> |
std::vector<RealType> rho_tmp(n, 0.0); |
157 |
> |
AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp); |
158 |
> |
for (int i = 0; i < n; i++) |
159 |
> |
snap->atomData.density[i] += rho_tmp[i]; |
160 |
> |
} |
161 |
> |
#endif |
162 |
|
} |
163 |
< |
else |
164 |
< |
{ |
165 |
< |
myRank_ = myRank/nColumns; |
163 |
> |
|
164 |
> |
void ForceDecomposition::distributeIntermediateData() { |
165 |
> |
#ifdef IS_MPI |
166 |
> |
Snapshot* snap = sman_->getCurrentSnapshot(); |
167 |
> |
if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) { |
168 |
> |
AtomCommRealI->gather(snap->atomData.functional, |
169 |
> |
snap->atomIData.functional); |
170 |
> |
AtomCommRealJ->gather(snap->atomData.functional, |
171 |
> |
snap->atomJData.functional); |
172 |
> |
} |
173 |
> |
|
174 |
> |
if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) { |
175 |
> |
AtomCommRealI->gather(snap->atomData.functionalDerivative, |
176 |
> |
snap->atomIData.functionalDerivative); |
177 |
> |
AtomCommRealJ->gather(snap->atomData.functionalDerivative, |
178 |
> |
snap->atomJData.functionalDerivative); |
179 |
> |
} |
180 |
> |
#endif |
181 |
|
} |
121 |
– |
MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0); |
182 |
|
|
123 |
– |
isColumn_ = false; |
183 |
|
|
184 |
< |
} |
184 |
> |
void ForceDecomposition::collectData() { |
185 |
> |
#ifdef IS_MPI |
186 |
> |
Snapshot* snap = sman_->getCurrentSnapshot(); |
187 |
> |
|
188 |
> |
int n = snap->atomData.force.size(); |
189 |
> |
vector<Vector3d> frc_tmp(n, V3Zero); |
190 |
> |
|
191 |
> |
AtomCommVectorI->scatter(snap->atomIData.force, frc_tmp); |
192 |
> |
for (int i = 0; i < n; i++) { |
193 |
> |
snap->atomData.force[i] += frc_tmp[i]; |
194 |
> |
frc_tmp[i] = 0.0; |
195 |
> |
} |
196 |
> |
|
197 |
> |
AtomCommVectorJ->scatter(snap->atomJData.force, frc_tmp); |
198 |
> |
for (int i = 0; i < n; i++) |
199 |
> |
snap->atomData.force[i] += frc_tmp[i]; |
200 |
> |
|
201 |
> |
|
202 |
> |
if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) { |
203 |
|
|
204 |
< |
ForceDecomposition::gather(sendbuf, receivebuf){ |
205 |
< |
communicators(myIndex_).Allgatherv(); |
129 |
< |
} |
204 |
> |
int nt = snap->atomData.force.size(); |
205 |
> |
vector<Vector3d> trq_tmp(nt, V3Zero); |
206 |
|
|
207 |
+ |
AtomCommVectorI->scatter(snap->atomIData.torque, trq_tmp); |
208 |
+ |
for (int i = 0; i < n; i++) { |
209 |
+ |
snap->atomData.torque[i] += trq_tmp[i]; |
210 |
+ |
trq_tmp[i] = 0.0; |
211 |
+ |
} |
212 |
+ |
|
213 |
+ |
AtomCommVectorJ->scatter(snap->atomJData.torque, trq_tmp); |
214 |
+ |
for (int i = 0; i < n; i++) |
215 |
+ |
snap->atomData.torque[i] += trq_tmp[i]; |
216 |
+ |
} |
217 |
+ |
|
218 |
+ |
int nLocal = snap->getNumberOfAtoms(); |
219 |
|
|
220 |
< |
|
221 |
< |
ForceDecomposition::scatter(sbuffer, rbuffer){ |
222 |
< |
communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM); |
223 |
< |
} |
224 |
< |
|
225 |
< |
|
220 |
> |
vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES, |
221 |
> |
vector<RealType> (nLocal, 0.0)); |
222 |
> |
|
223 |
> |
for (int i = 0; i < N_INTERACTION_FAMILIES; i++) { |
224 |
> |
AtomCommRealI->scatter(pot_row[i], pot_temp[i]); |
225 |
> |
for (int ii = 0; ii < pot_temp[i].size(); ii++ ) { |
226 |
> |
pot_local[i] += pot_temp[i][ii]; |
227 |
> |
} |
228 |
> |
} |
229 |
> |
#endif |
230 |
> |
} |
231 |
> |
|
232 |
> |
} //end namespace OpenMD |