| 1 |
|
#ifdef IS_MPI |
| 2 |
< |
|
| 3 |
< |
#include <cstdlib> |
| 4 |
< |
#include <cstring> |
| 5 |
< |
#include <cmath> |
| 2 |
> |
#include <iostream> |
| 3 |
> |
#include <stdlib.h> |
| 4 |
> |
#include <string.h> |
| 5 |
> |
#include <math.h> |
| 6 |
|
#include <mpi.h> |
| 7 |
– |
#include <mpi++.h> |
| 7 |
|
|
| 8 |
|
#include "mpiSimulation.hpp" |
| 9 |
|
#include "simError.h" |
| 10 |
|
#include "fortranWrappers.hpp" |
| 11 |
|
#include "randomSPRNG.hpp" |
| 12 |
|
|
| 14 |
– |
#define BASE_SEED 123456789 |
| 15 |
– |
|
| 13 |
|
mpiSimulation* mpiSim; |
| 14 |
|
|
| 15 |
|
mpiSimulation::mpiSimulation(SimInfo* the_entryPlug) |
| 17 |
|
entryPlug = the_entryPlug; |
| 18 |
|
mpiPlug = new mpiSimData; |
| 19 |
|
|
| 20 |
< |
mpiPlug->numberProcessors = MPI::COMM_WORLD.Get_size(); |
| 20 |
> |
MPI_Comm_size(MPI_COMM_WORLD, &(mpiPlug->numberProcessors) ); |
| 21 |
|
mpiPlug->myNode = worldRank; |
| 22 |
|
|
| 23 |
|
MolToProcMap = new int[entryPlug->n_mol]; |
| 40 |
|
|
| 41 |
|
} |
| 42 |
|
|
| 43 |
< |
int* mpiSimulation::divideLabor( void ){ |
| 43 |
> |
void mpiSimulation::divideLabor( ){ |
| 44 |
|
|
| 48 |
– |
int* globalIndex; |
| 49 |
– |
|
| 45 |
|
int nComponents; |
| 46 |
|
MoleculeStamp** compStamps; |
| 47 |
|
randomSPRNG *myRandom; |
| 55 |
|
int old_atoms, add_atoms, new_atoms; |
| 56 |
|
|
| 57 |
|
int nTarget; |
| 58 |
< |
int molIndex, atomIndex, compIndex, compStart; |
| 58 |
> |
int molIndex, atomIndex; |
| 59 |
|
int done; |
| 65 |
– |
int nLocal, molLocal; |
| 60 |
|
int i, j, loops, which_proc, nmol_local, natoms_local; |
| 61 |
|
int nmol_global, natoms_global; |
| 62 |
< |
int local_index, index; |
| 63 |
< |
int smallDiff, bigDiff; |
| 70 |
< |
int baseSeed = BASE_SEED; |
| 62 |
> |
int local_index; |
| 63 |
> |
int baseSeed = entryPlug->getSeed(); |
| 64 |
|
|
| 72 |
– |
int testSum; |
| 73 |
– |
|
| 65 |
|
nComponents = entryPlug->nComponents; |
| 66 |
|
compStamps = entryPlug->compStamps; |
| 67 |
|
componentsNmol = entryPlug->componentsNmol; |
| 76 |
|
|
| 77 |
|
myRandom = new randomSPRNG( baseSeed ); |
| 78 |
|
|
| 79 |
< |
a = (double)mpiPlug->nMolGlobal / (double)mpiPlug->nAtomsGlobal; |
| 79 |
> |
a = 3.0 * (double)mpiPlug->nMolGlobal / (double)mpiPlug->nAtomsGlobal; |
| 80 |
|
|
| 81 |
|
// Initialize things that we'll send out later: |
| 82 |
|
for (i = 0; i < mpiPlug->numberProcessors; i++ ) { |
| 124 |
|
// How many atoms does this processor have? |
| 125 |
|
|
| 126 |
|
old_atoms = AtomsPerProc[which_proc]; |
| 136 |
– |
|
| 137 |
– |
// If the processor already had too many atoms, just skip this |
| 138 |
– |
// processor and try again. |
| 139 |
– |
|
| 140 |
– |
if (old_atoms >= nTarget) continue; |
| 141 |
– |
|
| 127 |
|
add_atoms = compStamps[MolComponentType[i]]->getNAtoms(); |
| 128 |
|
new_atoms = old_atoms + add_atoms; |
| 144 |
– |
|
| 145 |
– |
// If we can add this molecule to this processor without sending |
| 146 |
– |
// it above nTarget, then go ahead and do it: |
| 147 |
– |
|
| 148 |
– |
if (new_atoms <= nTarget) { |
| 149 |
– |
MolToProcMap[i] = which_proc; |
| 150 |
– |
AtomsPerProc[which_proc] += add_atoms; |
| 151 |
– |
for (j = 0 ; j < add_atoms; j++ ) { |
| 152 |
– |
atomIndex++; |
| 153 |
– |
AtomToProcMap[atomIndex] = which_proc; |
| 154 |
– |
} |
| 155 |
– |
done = 1; |
| 156 |
– |
continue; |
| 157 |
– |
} |
| 129 |
|
|
| 130 |
|
// If we've been through this loop too many times, we need |
| 131 |
|
// to just give up and assign the molecule to this processor |
| 143 |
|
MolToProcMap[i] = which_proc; |
| 144 |
|
AtomsPerProc[which_proc] += add_atoms; |
| 145 |
|
for (j = 0 ; j < add_atoms; j++ ) { |
| 146 |
< |
atomIndex++; |
| 147 |
< |
AtomToProcMap[atomIndex] = which_proc; |
| 146 |
> |
AtomToProcMap[atomIndex] = which_proc; |
| 147 |
> |
atomIndex++; |
| 148 |
|
} |
| 149 |
|
done = 1; |
| 150 |
|
continue; |
| 151 |
|
} |
| 152 |
+ |
|
| 153 |
+ |
// If we can add this molecule to this processor without sending |
| 154 |
+ |
// it above nTarget, then go ahead and do it: |
| 155 |
+ |
|
| 156 |
+ |
if (new_atoms <= nTarget) { |
| 157 |
+ |
MolToProcMap[i] = which_proc; |
| 158 |
+ |
AtomsPerProc[which_proc] += add_atoms; |
| 159 |
+ |
for (j = 0 ; j < add_atoms; j++ ) { |
| 160 |
+ |
AtomToProcMap[atomIndex] = which_proc; |
| 161 |
+ |
atomIndex++; |
| 162 |
+ |
} |
| 163 |
+ |
done = 1; |
| 164 |
+ |
continue; |
| 165 |
+ |
} |
| 166 |
|
|
| 182 |
– |
// The only situation left is where old_atoms < nTarget, but |
| 183 |
– |
// new_atoms > nTarget. We want to accept this with some |
| 184 |
– |
// probability that dies off the farther we are from nTarget |
| 167 |
|
|
| 168 |
+ |
// The only situation left is when new_atoms > nTarget. We |
| 169 |
+ |
// want to accept this with some probability that dies off the |
| 170 |
+ |
// farther we are from nTarget |
| 171 |
+ |
|
| 172 |
|
// roughly: x = new_atoms - nTarget |
| 173 |
|
// Pacc(x) = exp(- a * x) |
| 174 |
< |
// where a = 1 / (average atoms per molecule) |
| 174 |
> |
// where a = penalty / (average atoms per molecule) |
| 175 |
|
|
| 176 |
|
x = (double) (new_atoms - nTarget); |
| 177 |
|
y = myRandom->getRandom(); |
| 178 |
< |
|
| 179 |
< |
if (exp(- a * x) > y) { |
| 178 |
> |
|
| 179 |
> |
if (y < exp(- a * x)) { |
| 180 |
|
MolToProcMap[i] = which_proc; |
| 181 |
|
AtomsPerProc[which_proc] += add_atoms; |
| 182 |
|
for (j = 0 ; j < add_atoms; j++ ) { |
| 183 |
< |
atomIndex++; |
| 184 |
< |
AtomToProcMap[atomIndex] = which_proc; |
| 185 |
< |
} |
| 183 |
> |
AtomToProcMap[atomIndex] = which_proc; |
| 184 |
> |
atomIndex++; |
| 185 |
> |
} |
| 186 |
|
done = 1; |
| 187 |
|
continue; |
| 188 |
|
} else { |
| 194 |
|
|
| 195 |
|
// Spray out this nonsense to all other processors: |
| 196 |
|
|
| 197 |
< |
MPI::COMM_WORLD.Bcast(MolToProcMap, mpiPlug->nMolGlobal, |
| 198 |
< |
MPI_INT, 0); |
| 197 |
> |
MPI_Bcast(MolToProcMap, mpiPlug->nMolGlobal, |
| 198 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 199 |
|
|
| 200 |
< |
MPI::COMM_WORLD.Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal, |
| 201 |
< |
MPI_INT, 0); |
| 200 |
> |
MPI_Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal, |
| 201 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 202 |
|
|
| 203 |
< |
MPI::COMM_WORLD.Bcast(MolComponentType, mpiPlug->nMolGlobal, |
| 204 |
< |
MPI_INT, 0); |
| 203 |
> |
MPI_Bcast(MolComponentType, mpiPlug->nMolGlobal, |
| 204 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 205 |
|
|
| 206 |
< |
MPI::COMM_WORLD.Bcast(AtomsPerProc, mpiPlug->numberProcessors, |
| 207 |
< |
MPI_INT, 0); |
| 206 |
> |
MPI_Bcast(AtomsPerProc, mpiPlug->numberProcessors, |
| 207 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 208 |
|
} else { |
| 209 |
|
|
| 210 |
|
// Listen to your marching orders from processor 0: |
| 211 |
|
|
| 212 |
< |
MPI::COMM_WORLD.Bcast(MolToProcMap, mpiPlug->nMolGlobal, |
| 213 |
< |
MPI_INT, 0); |
| 212 |
> |
MPI_Bcast(MolToProcMap, mpiPlug->nMolGlobal, |
| 213 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 214 |
|
|
| 215 |
< |
MPI::COMM_WORLD.Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal, |
| 216 |
< |
MPI_INT, 0); |
| 215 |
> |
MPI_Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal, |
| 216 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 217 |
|
|
| 218 |
< |
MPI::COMM_WORLD.Bcast(MolComponentType, mpiPlug->nMolGlobal, |
| 219 |
< |
MPI_INT, 0); |
| 218 |
> |
MPI_Bcast(MolComponentType, mpiPlug->nMolGlobal, |
| 219 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 220 |
|
|
| 221 |
< |
MPI::COMM_WORLD.Bcast(AtomsPerProc, mpiPlug->numberProcessors, |
| 222 |
< |
MPI_INT, 0); |
| 221 |
> |
MPI_Bcast(AtomsPerProc, mpiPlug->numberProcessors, |
| 222 |
> |
MPI_INT, 0, MPI_COMM_WORLD); |
| 223 |
> |
|
| 224 |
> |
|
| 225 |
|
} |
| 226 |
|
|
| 227 |
|
|
| 241 |
|
} |
| 242 |
|
} |
| 243 |
|
|
| 244 |
< |
MPI::COMM_WORLD.Allreduce(&nmol_local,&nmol_global,1,MPI_INT,MPI_SUM); |
| 245 |
< |
MPI::COMM_WORLD.Allreduce(&natoms_local,&natoms_global,1,MPI_INT,MPI_SUM); |
| 244 |
> |
MPI_Allreduce(&nmol_local,&nmol_global,1,MPI_INT,MPI_SUM, |
| 245 |
> |
MPI_COMM_WORLD); |
| 246 |
> |
MPI_Allreduce(&natoms_local,&natoms_global,1,MPI_INT, |
| 247 |
> |
MPI_SUM, MPI_COMM_WORLD); |
| 248 |
|
|
| 249 |
|
if( nmol_global != entryPlug->n_mol ){ |
| 250 |
|
sprintf( painCave.errMsg, |
| 271 |
|
mpiPlug->myNMol = nmol_local; |
| 272 |
|
mpiPlug->myNlocal = natoms_local; |
| 273 |
|
|
| 274 |
< |
globalIndex = new int[mpiPlug->myNlocal]; |
| 274 |
> |
globalAtomIndex.resize(mpiPlug->myNlocal); |
| 275 |
|
local_index = 0; |
| 276 |
|
for (i = 0; i < mpiPlug->nAtomsGlobal; i++) { |
| 277 |
|
if (AtomToProcMap[i] == mpiPlug->myNode) { |
| 278 |
+ |
globalAtomIndex[local_index] = i; |
| 279 |
+ |
|
| 280 |
+ |
globalToLocalAtom[i] = local_index; |
| 281 |
|
local_index++; |
| 282 |
< |
globalIndex[local_index] = i; |
| 282 |
> |
|
| 283 |
|
} |
| 284 |
+ |
else |
| 285 |
+ |
globalToLocalAtom[i] = -1; |
| 286 |
|
} |
| 287 |
< |
|
| 288 |
< |
return globalIndex; |
| 287 |
> |
|
| 288 |
> |
globalMolIndex.resize(mpiPlug->myNMol); |
| 289 |
> |
local_index = 0; |
| 290 |
> |
for (i = 0; i < mpiPlug->nMolGlobal; i++) { |
| 291 |
> |
if (MolToProcMap[i] == mpiPlug->myNode) { |
| 292 |
> |
globalMolIndex[local_index] = i; |
| 293 |
> |
globalToLocalMol[i] = local_index; |
| 294 |
> |
local_index++; |
| 295 |
> |
} |
| 296 |
> |
else |
| 297 |
> |
globalToLocalMol[i] = -1; |
| 298 |
> |
} |
| 299 |
> |
|
| 300 |
|
} |
| 301 |
|
|
| 302 |
|
|
| 305 |
|
int isError, i; |
| 306 |
|
int *globalIndex = new int[mpiPlug->myNlocal]; |
| 307 |
|
|
| 308 |
< |
for(i=0; i<mpiPlug->myNlocal; i++) globalIndex[i] = entryPlug->atoms[i]->getGlobalIndex(); |
| 308 |
> |
// Fortran indexing needs to be increased by 1 in order to get the 2 languages to |
| 309 |
> |
// not barf |
| 310 |
|
|
| 311 |
+ |
for(i=0; i<mpiPlug->myNlocal; i++) globalIndex[i] = entryPlug->atoms[i]->getGlobalIndex()+1; |
| 312 |
+ |
|
| 313 |
|
|
| 314 |
|
isError = 0; |
| 315 |
|
setFsimParallel( mpiPlug, &(entryPlug->n_atoms), globalIndex, &isError ); |