28 |
|
if(worldRank == 0 ){ |
29 |
|
#endif // is_mpi |
30 |
|
|
31 |
– |
|
31 |
|
dumpFile.open(entry_plug->sampleName, ios::out | ios::trunc ); |
32 |
|
|
33 |
|
if( !dumpFile ){ |
39 |
|
simError(); |
40 |
|
} |
41 |
|
|
43 |
– |
//outFile.setf( ios::scientific ); |
44 |
– |
|
42 |
|
#ifdef IS_MPI |
43 |
|
} |
44 |
|
|
91 |
|
|
92 |
|
indexArray.clear(); |
93 |
|
|
94 |
< |
for(int i = 0; i < mpiSim->getMyNlocal();i++) |
94 |
> |
for(int i = 0; i < mpiSim->getMyNlocal();i++) |
95 |
|
indexArray.push_back(make_pair(i, atoms[i]->getGlobalIndex())); |
96 |
|
|
97 |
|
sort(indexArray.begin(), indexArray.end(), indexSortingCriterion); |
98 |
+ |
|
99 |
+ |
//for (int i = 0; i < mpiSim->getMyNlocal(); i++) { |
100 |
+ |
// printf("node %d has global %d at local %d\n", worldRank, indexArray[i].second, indexArray[i].first); |
101 |
+ |
//} |
102 |
+ |
|
103 |
|
} |
104 |
+ |
|
105 |
|
#endif |
106 |
|
|
107 |
|
void DumpWriter::writeDump(double currentTime){ |
105 |
– |
|
106 |
– |
// write to eor file |
107 |
– |
writeFinal(currentTime); |
108 |
|
|
109 |
< |
//write to dump file |
110 |
< |
writeFrame(dumpFile, currentTime); |
109 |
> |
ofstream finalOut; |
110 |
> |
vector<ofstream*> fileStreams; |
111 |
> |
|
112 |
> |
#ifdef IS_MPI |
113 |
> |
printf("Hello from node %d\n", worldRank); |
114 |
> |
sortByGlobalIndex(); |
115 |
> |
if(worldRank == 0 ){ |
116 |
> |
|
117 |
> |
finalOut.open( entry_plug->finalName, ios::out | ios::trunc ); |
118 |
> |
if( !finalOut ){ |
119 |
> |
sprintf( painCave.errMsg, |
120 |
> |
"Could not open \"%s\" for final dump output.\n", |
121 |
> |
entry_plug->finalName ); |
122 |
> |
painCave.isFatal = 1; |
123 |
> |
simError(); |
124 |
> |
} |
125 |
> |
} |
126 |
> |
#endif // is_mpi |
127 |
> |
|
128 |
> |
fileStreams.push_back(&finalOut); |
129 |
> |
fileStreams.push_back(&dumpFile); |
130 |
> |
|
131 |
> |
writeFrame(fileStreams, currentTime); |
132 |
> |
|
133 |
> |
#ifdef IS_MPI |
134 |
> |
finalOut.close(); |
135 |
> |
#endif |
136 |
|
|
137 |
|
} |
138 |
|
|
139 |
|
void DumpWriter::writeFinal(double currentTime){ |
140 |
|
|
141 |
< |
ofstream finalOut; |
142 |
< |
|
143 |
< |
//Open eor file |
141 |
> |
ofstream finalOut; |
142 |
> |
vector<ofstream*> fileStreams; |
143 |
> |
|
144 |
|
#ifdef IS_MPI |
145 |
|
if(worldRank == 0 ){ |
146 |
|
#endif // is_mpi |
147 |
|
|
148 |
|
finalOut.open( entry_plug->finalName, ios::out | ios::trunc ); |
149 |
+ |
|
150 |
|
if( !finalOut ){ |
151 |
|
sprintf( painCave.errMsg, |
152 |
|
"Could not open \"%s\" for final dump output.\n", |
154 |
|
painCave.isFatal = 1; |
155 |
|
simError(); |
156 |
|
} |
157 |
< |
|
157 |
> |
|
158 |
|
#ifdef IS_MPI |
159 |
|
} |
134 |
– |
#endif |
135 |
– |
|
136 |
– |
//write to eor file |
137 |
– |
writeFrame(finalOut, currentTime); |
138 |
– |
|
139 |
– |
//close eor file |
140 |
– |
#ifdef IS_MPI |
141 |
– |
if(worldRank == 0 ){ |
142 |
– |
finalOut.close(); |
143 |
– |
} |
160 |
|
#endif // is_mpi |
161 |
+ |
|
162 |
+ |
fileStreams.push_back(&finalOut); |
163 |
+ |
writeFrame(fileStreams, currentTime); |
164 |
|
|
165 |
+ |
#ifdef IS_MPI |
166 |
+ |
finalOut.close(); |
167 |
+ |
#endif |
168 |
+ |
|
169 |
|
} |
170 |
|
|
171 |
< |
void DumpWriter::writeFrame( ofstream& outFile, double currentTime ){ |
171 |
> |
void DumpWriter::writeFrame( vector<ofstream*>& outFile, double currentTime ){ |
172 |
|
|
173 |
|
const int BUFFERSIZE = 2000; |
174 |
|
const int MINIBUFFERSIZE = 100; |
175 |
|
|
176 |
< |
char tempBuffer[BUFFERSIZE]; |
176 |
> |
char tempBuffer[BUFFERSIZE]; |
177 |
|
char writeLine[BUFFERSIZE]; |
178 |
|
|
179 |
< |
int i; |
179 |
> |
int i, k; |
180 |
|
|
181 |
|
#ifdef IS_MPI |
182 |
|
|
183 |
+ |
/********************************************************************* |
184 |
+ |
* Documentation? You want DOCUMENTATION? |
185 |
+ |
* |
186 |
+ |
* Why all the potatoes below? |
187 |
+ |
* |
188 |
+ |
* To make a long story short, the original version of DumpWriter |
189 |
+ |
* worked in the most inefficient way possible. Node 0 would |
190 |
+ |
* poke each of the node for an individual atom's formatted data |
191 |
+ |
* as node 0 worked its way down the global index. This was particularly |
192 |
+ |
* inefficient since the method blocked all processors at every atom |
193 |
+ |
* (and did it twice!). |
194 |
+ |
* |
195 |
+ |
* An intermediate version of DumpWriter could be described from Node |
196 |
+ |
* zero's perspective as follows: |
197 |
+ |
* |
198 |
+ |
* 1) Have 100 of your friends stand in a circle. |
199 |
+ |
* 2) When you say go, have all of them start tossing potatoes at |
200 |
+ |
* you (one at a time). |
201 |
+ |
* 3) Catch the potatoes. |
202 |
+ |
* |
203 |
+ |
* It was an improvement, but MPI has buffers and caches that could |
204 |
+ |
* best be described in this analogy as "potato nets", so there's no |
205 |
+ |
* need to block the processors atom-by-atom. |
206 |
+ |
* |
207 |
+ |
* This new and improved DumpWriter works in an even more efficient |
208 |
+ |
* way: |
209 |
+ |
* |
210 |
+ |
* 1) Have 100 of your friend stand in a circle. |
211 |
+ |
* 2) When you say go, have them start tossing 5-pound bags of |
212 |
+ |
* potatoes at you. |
213 |
+ |
* 3) Once you've caught a friend's bag of potatoes, |
214 |
+ |
* toss them a spud to let them know they can toss another bag. |
215 |
+ |
* |
216 |
+ |
* How's THAT for documentation? |
217 |
+ |
* |
218 |
+ |
*********************************************************************/ |
219 |
+ |
|
220 |
|
int *potatoes; |
221 |
|
int myPotato; |
222 |
|
|
238 |
|
double pos[3], vel[3]; |
239 |
|
|
240 |
|
#ifndef IS_MPI |
241 |
+ |
|
242 |
+ |
for(k = 0; k < outFile.size(); k++){ |
243 |
+ |
*outFile[k] << nAtoms << "\n"; |
244 |
|
|
245 |
< |
outFile << nAtoms << "\n"; |
245 |
> |
*outFile[k] << currentTime << ";\t" |
246 |
> |
<< entry_plug->Hmat[0][0] << "\t" |
247 |
> |
<< entry_plug->Hmat[1][0] << "\t" |
248 |
> |
<< entry_plug->Hmat[2][0] << ";\t" |
249 |
> |
|
250 |
> |
<< entry_plug->Hmat[0][1] << "\t" |
251 |
> |
<< entry_plug->Hmat[1][1] << "\t" |
252 |
> |
<< entry_plug->Hmat[2][1] << ";\t" |
253 |
|
|
254 |
< |
outFile << currentTime << ";\t" |
255 |
< |
<< entry_plug->Hmat[0][0] << "\t" |
256 |
< |
<< entry_plug->Hmat[1][0] << "\t" |
187 |
< |
<< entry_plug->Hmat[2][0] << ";\t" |
254 |
> |
<< entry_plug->Hmat[0][2] << "\t" |
255 |
> |
<< entry_plug->Hmat[1][2] << "\t" |
256 |
> |
<< entry_plug->Hmat[2][2] << ";"; |
257 |
|
|
258 |
< |
<< entry_plug->Hmat[0][1] << "\t" |
259 |
< |
<< entry_plug->Hmat[1][1] << "\t" |
260 |
< |
<< entry_plug->Hmat[2][1] << ";\t" |
261 |
< |
|
193 |
< |
<< entry_plug->Hmat[0][2] << "\t" |
194 |
< |
<< entry_plug->Hmat[1][2] << "\t" |
195 |
< |
<< entry_plug->Hmat[2][2] << ";"; |
196 |
< |
//write out additional parameters, such as chi and eta |
197 |
< |
outFile << entry_plug->the_integrator->getAdditionalParameters(); |
198 |
< |
outFile << endl; |
199 |
< |
|
258 |
> |
//write out additional parameters, such as chi and eta |
259 |
> |
*outFile[k] << entry_plug->the_integrator->getAdditionalParameters() << endl; |
260 |
> |
} |
261 |
> |
|
262 |
|
for( i=0; i<nAtoms; i++ ){ |
263 |
|
|
264 |
|
atoms[i]->getPos(pos); |
294 |
|
else |
295 |
|
strcat( writeLine, "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0\n" ); |
296 |
|
|
297 |
< |
outFile << writeLine; |
297 |
> |
for(k = 0; k < outFile.size(); k++) |
298 |
> |
*outFile[k] << writeLine; |
299 |
|
} |
300 |
|
|
301 |
|
#else // is_mpi |
324 |
|
nProc = mpiSim->getNumberProcessors(); |
325 |
|
potatoes = new int[nProc]; |
326 |
|
|
327 |
+ |
//write out the comment lines |
328 |
|
for (i = 0; i < nProc; i++) |
329 |
|
potatoes[i] = 0; |
330 |
|
|
331 |
< |
outFile << mpiSim->getTotAtoms() << "\n"; |
331 |
> |
for(k = 0; k < outFile.size(); k++){ |
332 |
> |
*outFile[k] << mpiSim->getTotAtoms() << "\n"; |
333 |
|
|
334 |
< |
outFile << currentTime << ";\t" |
335 |
< |
<< entry_plug->Hmat[0][0] << "\t" |
336 |
< |
<< entry_plug->Hmat[1][0] << "\t" |
337 |
< |
<< entry_plug->Hmat[2][0] << ";\t" |
334 |
> |
*outFile[k] << currentTime << ";\t" |
335 |
> |
<< entry_plug->Hmat[0][0] << "\t" |
336 |
> |
<< entry_plug->Hmat[1][0] << "\t" |
337 |
> |
<< entry_plug->Hmat[2][0] << ";\t" |
338 |
|
|
339 |
< |
<< entry_plug->Hmat[0][1] << "\t" |
340 |
< |
<< entry_plug->Hmat[1][1] << "\t" |
341 |
< |
<< entry_plug->Hmat[2][1] << ";\t" |
339 |
> |
<< entry_plug->Hmat[0][1] << "\t" |
340 |
> |
<< entry_plug->Hmat[1][1] << "\t" |
341 |
> |
<< entry_plug->Hmat[2][1] << ";\t" |
342 |
|
|
343 |
< |
<< entry_plug->Hmat[0][2] << "\t" |
344 |
< |
<< entry_plug->Hmat[1][2] << "\t" |
345 |
< |
<< entry_plug->Hmat[2][2] << ";"; |
343 |
> |
<< entry_plug->Hmat[0][2] << "\t" |
344 |
> |
<< entry_plug->Hmat[1][2] << "\t" |
345 |
> |
<< entry_plug->Hmat[2][2] << ";"; |
346 |
> |
|
347 |
> |
*outFile[k] << entry_plug->the_integrator->getAdditionalParameters() << endl; |
348 |
> |
} |
349 |
|
|
282 |
– |
outFile << entry_plug->the_integrator->getAdditionalParameters(); |
283 |
– |
outFile << endl; |
284 |
– |
outFile.flush(); |
285 |
– |
|
350 |
|
currentIndex = 0; |
351 |
+ |
|
352 |
|
for (i = 0 ; i < mpiSim->getTotAtoms(); i++ ) { |
353 |
|
|
354 |
|
// Get the Node number which has this atom; |
393 |
|
|
394 |
|
} else { |
395 |
|
|
396 |
< |
haveError = 0; |
396 |
> |
haveError = 0; |
397 |
|
which_atom = i; |
398 |
|
|
399 |
< |
local_index = indexArray[currentIndex].first; |
335 |
< |
|
336 |
< |
if (which_atom == indexArray[currentIndex].second) { |
337 |
< |
|
338 |
< |
atomTypeString = atoms[local_index]->getType(); |
399 |
> |
//local_index = -1; |
400 |
|
|
401 |
+ |
//for (j=0; (j<mpiSim->getMyNlocal()) && (local_index < 0); j++) { |
402 |
+ |
// if (atoms[j]->getGlobalIndex() == which_atom) local_index = j; |
403 |
+ |
//} |
404 |
+ |
|
405 |
+ |
//if (local_index != -1) { |
406 |
+ |
|
407 |
+ |
local_index = indexArray[currentIndex].first; |
408 |
+ |
|
409 |
+ |
if (which_atom == indexArray[currentIndex].second) { |
410 |
+ |
|
411 |
+ |
atomTypeString = atoms[local_index]->getType(); |
412 |
+ |
|
413 |
|
atoms[local_index]->getPos(pos); |
414 |
|
atoms[local_index]->getVel(vel); |
415 |
< |
|
415 |
> |
|
416 |
|
atomData6[0] = pos[0]; |
417 |
|
atomData6[1] = pos[1]; |
418 |
|
atomData6[2] = pos[2]; |
445 |
|
|
446 |
|
} else { |
447 |
|
sprintf(painCave.errMsg, |
448 |
< |
"Atom %d not found on processor %d\n", |
449 |
< |
i, worldRank ); |
448 |
> |
"Atom %d not found on processor %d, currentIndex = %d, local_index = %d\n", |
449 |
> |
which_atom, worldRank, currentIndex, local_index ); |
450 |
|
haveError= 1; |
451 |
|
simError(); |
452 |
|
} |
453 |
|
|
454 |
< |
if(haveError) DieDieDie(); |
454 |
> |
if(haveError) DieDieDie(); |
455 |
|
|
456 |
< |
currentIndex ++; |
456 |
> |
currentIndex++; |
457 |
|
} |
458 |
|
// If we've survived to here, format the line: |
459 |
|
|
460 |
|
if (!isDirectional) { |
461 |
|
|
462 |
< |
sprintf( writeLine, |
462 |
> |
sprintf( writeLine, |
463 |
|
"%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t", |
464 |
|
atomTypeString, |
465 |
|
atomData6[0], |
468 |
|
atomData6[3], |
469 |
|
atomData6[4], |
470 |
|
atomData6[5]); |
471 |
< |
|
471 |
> |
|
472 |
|
strcat( writeLine, "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0\n" ); |
473 |
|
|
474 |
|
} else { |
492 |
|
|
493 |
|
} |
494 |
|
|
495 |
< |
outFile << writeLine; |
495 |
> |
for(k = 0; k < outFile.size(); k++) |
496 |
> |
*outFile[k] << writeLine; |
497 |
|
} |
498 |
|
|
499 |
< |
|
500 |
< |
outFile.flush(); |
499 |
> |
for(k = 0; k < outFile.size(); k++) |
500 |
> |
outFile[k]->flush(); |
501 |
> |
|
502 |
|
sprintf( checkPointMsg, |
503 |
|
"Sucessfully took a dump.\n"); |
504 |
+ |
|
505 |
|
MPIcheckPoint(); |
506 |
+ |
|
507 |
|
delete[] potatoes; |
508 |
+ |
|
509 |
|
} else { |
510 |
|
|
511 |
|
// worldRank != 0, so I'm a remote node. |
522 |
|
if (AtomToProcMap[i] == worldRank) { |
523 |
|
|
524 |
|
if (myPotato + 3 >= MAXTAG) { |
525 |
< |
|
525 |
> |
|
526 |
|
// The potato was going to exceed the maximum value, |
527 |
|
// so wrap this processor potato back to 0 (and block until |
528 |
|
// node 0 says we can go: |
529 |
< |
|
529 |
> |
|
530 |
|
MPI_Recv(&myPotato, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, &istatus); |
531 |
|
|
532 |
|
} |
533 |
|
which_atom = i; |
534 |
< |
local_index = indexArray[currentIndex].first; |
534 |
> |
|
535 |
> |
//local_index = -1; |
536 |
> |
|
537 |
> |
//for (j=0; (j<mpiSim->getMyNlocal()) && (local_index < 0); j++) { |
538 |
> |
// if (atoms[j]->getGlobalIndex() == which_atom) local_index = j; |
539 |
> |
//} |
540 |
> |
|
541 |
> |
//if (local_index != -1) { |
542 |
> |
|
543 |
> |
local_index = indexArray[currentIndex].first; |
544 |
|
|
545 |
< |
if (which_atom == indexArray[currentIndex].second) { |
545 |
> |
if (which_atom == indexArray[currentIndex].second) { |
546 |
|
|
547 |
|
atomTypeString = atoms[local_index]->getType(); |
548 |
< |
|
548 |
> |
|
549 |
|
atoms[local_index]->getPos(pos); |
550 |
|
atoms[local_index]->getVel(vel); |
551 |
< |
|
551 |
> |
|
552 |
|
atomData6[0] = pos[0]; |
553 |
|
atomData6[1] = pos[1]; |
554 |
|
atomData6[2] = pos[2]; |
573 |
|
atomData13[7] = q[1]; |
574 |
|
atomData13[8] = q[2]; |
575 |
|
atomData13[9] = q[3]; |
576 |
< |
|
576 |
> |
|
577 |
|
atomData13[10] = dAtom->getJx(); |
578 |
|
atomData13[11] = dAtom->getJy(); |
579 |
|
atomData13[12] = dAtom->getJz(); |
581 |
|
|
582 |
|
} else { |
583 |
|
sprintf(painCave.errMsg, |
584 |
< |
"Atom %d not found on processor %d\n", |
585 |
< |
i, worldRank ); |
584 |
> |
"Atom %d not found on processor %d, currentIndex = %d, local_index = %d\n", |
585 |
> |
which_atom, worldRank, currentIndex, local_index ); |
586 |
|
haveError= 1; |
587 |
|
simError(); |
588 |
|
} |
589 |
< |
|
589 |
> |
|
590 |
|
strncpy(MPIatomTypeString, atomTypeString, MINIBUFFERSIZE); |
591 |
|
|
592 |
|
// null terminate the string before sending (just in case): |
593 |
|
MPIatomTypeString[MINIBUFFERSIZE-1] = '\0'; |
594 |
|
|
595 |
|
MPI_Send(MPIatomTypeString, MINIBUFFERSIZE, MPI_CHAR, 0, |
596 |
< |
myPotato, MPI_COMM_WORLD); |
596 |
> |
myPotato, MPI_COMM_WORLD); |
597 |
|
|
598 |
|
myPotato++; |
599 |
|
|
600 |
|
MPI_Send(&isDirectional, 1, MPI_INT, 0, |
601 |
< |
myPotato, MPI_COMM_WORLD); |
601 |
> |
myPotato, MPI_COMM_WORLD); |
602 |
|
|
603 |
|
myPotato++; |
604 |
|
|