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 |
+ |
|
100 |
|
#endif |
101 |
|
|
102 |
|
void DumpWriter::writeDump(double currentTime){ |
105 |
– |
|
106 |
– |
// write to eor file |
107 |
– |
writeFinal(currentTime); |
103 |
|
|
104 |
< |
//write to dump file |
105 |
< |
writeFrame(dumpFile, currentTime); |
104 |
> |
ofstream finalOut; |
105 |
> |
vector<ofstream*> fileStreams; |
106 |
> |
|
107 |
> |
#ifdef IS_MPI |
108 |
> |
if(worldRank == 0 ){ |
109 |
> |
#endif |
110 |
> |
finalOut.open( entry_plug->finalName, ios::out | ios::trunc ); |
111 |
> |
if( !finalOut ){ |
112 |
> |
sprintf( painCave.errMsg, |
113 |
> |
"Could not open \"%s\" for final dump output.\n", |
114 |
> |
entry_plug->finalName ); |
115 |
> |
painCave.isFatal = 1; |
116 |
> |
simError(); |
117 |
> |
} |
118 |
> |
#ifdef IS_MPI |
119 |
> |
} |
120 |
> |
#endif // is_mpi |
121 |
> |
|
122 |
> |
fileStreams.push_back(&finalOut); |
123 |
> |
fileStreams.push_back(&dumpFile); |
124 |
> |
|
125 |
> |
writeFrame(fileStreams, currentTime); |
126 |
> |
|
127 |
> |
#ifdef IS_MPI |
128 |
> |
finalOut.close(); |
129 |
> |
#endif |
130 |
|
|
131 |
|
} |
132 |
|
|
133 |
|
void DumpWriter::writeFinal(double currentTime){ |
134 |
|
|
135 |
< |
ofstream finalOut; |
136 |
< |
|
137 |
< |
//Open eor file |
135 |
> |
ofstream finalOut; |
136 |
> |
vector<ofstream*> fileStreams; |
137 |
> |
|
138 |
|
#ifdef IS_MPI |
139 |
|
if(worldRank == 0 ){ |
140 |
|
#endif // is_mpi |
141 |
|
|
142 |
|
finalOut.open( entry_plug->finalName, ios::out | ios::trunc ); |
143 |
+ |
|
144 |
|
if( !finalOut ){ |
145 |
|
sprintf( painCave.errMsg, |
146 |
|
"Could not open \"%s\" for final dump output.\n", |
148 |
|
painCave.isFatal = 1; |
149 |
|
simError(); |
150 |
|
} |
151 |
< |
|
151 |
> |
|
152 |
|
#ifdef IS_MPI |
153 |
|
} |
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 |
– |
} |
154 |
|
#endif // is_mpi |
155 |
+ |
|
156 |
+ |
fileStreams.push_back(&finalOut); |
157 |
+ |
writeFrame(fileStreams, currentTime); |
158 |
|
|
159 |
+ |
#ifdef IS_MPI |
160 |
+ |
finalOut.close(); |
161 |
+ |
#endif |
162 |
+ |
|
163 |
|
} |
164 |
|
|
165 |
< |
void DumpWriter::writeFrame( ofstream& outFile, double currentTime ){ |
165 |
> |
void DumpWriter::writeFrame( vector<ofstream*>& outFile, double currentTime ){ |
166 |
|
|
167 |
|
const int BUFFERSIZE = 2000; |
168 |
|
const int MINIBUFFERSIZE = 100; |
169 |
|
|
170 |
< |
char tempBuffer[BUFFERSIZE]; |
170 |
> |
char tempBuffer[BUFFERSIZE]; |
171 |
|
char writeLine[BUFFERSIZE]; |
172 |
|
|
173 |
< |
int i; |
173 |
> |
int i, k; |
174 |
|
|
175 |
|
#ifdef IS_MPI |
176 |
|
|
177 |
+ |
/********************************************************************* |
178 |
+ |
* Documentation? You want DOCUMENTATION? |
179 |
+ |
* |
180 |
+ |
* Why all the potatoes below? |
181 |
+ |
* |
182 |
+ |
* To make a long story short, the original version of DumpWriter |
183 |
+ |
* worked in the most inefficient way possible. Node 0 would |
184 |
+ |
* poke each of the node for an individual atom's formatted data |
185 |
+ |
* as node 0 worked its way down the global index. This was particularly |
186 |
+ |
* inefficient since the method blocked all processors at every atom |
187 |
+ |
* (and did it twice!). |
188 |
+ |
* |
189 |
+ |
* An intermediate version of DumpWriter could be described from Node |
190 |
+ |
* zero's perspective as follows: |
191 |
+ |
* |
192 |
+ |
* 1) Have 100 of your friends stand in a circle. |
193 |
+ |
* 2) When you say go, have all of them start tossing potatoes at |
194 |
+ |
* you (one at a time). |
195 |
+ |
* 3) Catch the potatoes. |
196 |
+ |
* |
197 |
+ |
* It was an improvement, but MPI has buffers and caches that could |
198 |
+ |
* best be described in this analogy as "potato nets", so there's no |
199 |
+ |
* need to block the processors atom-by-atom. |
200 |
+ |
* |
201 |
+ |
* This new and improved DumpWriter works in an even more efficient |
202 |
+ |
* way: |
203 |
+ |
* |
204 |
+ |
* 1) Have 100 of your friend stand in a circle. |
205 |
+ |
* 2) When you say go, have them start tossing 5-pound bags of |
206 |
+ |
* potatoes at you. |
207 |
+ |
* 3) Once you've caught a friend's bag of potatoes, |
208 |
+ |
* toss them a spud to let them know they can toss another bag. |
209 |
+ |
* |
210 |
+ |
* How's THAT for documentation? |
211 |
+ |
* |
212 |
+ |
*********************************************************************/ |
213 |
+ |
|
214 |
|
int *potatoes; |
215 |
|
int myPotato; |
216 |
|
|
232 |
|
double pos[3], vel[3]; |
233 |
|
|
234 |
|
#ifndef IS_MPI |
235 |
+ |
|
236 |
+ |
for(k = 0; k < outFile.size(); k++){ |
237 |
+ |
*outFile[k] << nAtoms << "\n"; |
238 |
|
|
239 |
< |
outFile << nAtoms << "\n"; |
239 |
> |
*outFile[k] << currentTime << ";\t" |
240 |
> |
<< entry_plug->Hmat[0][0] << "\t" |
241 |
> |
<< entry_plug->Hmat[1][0] << "\t" |
242 |
> |
<< entry_plug->Hmat[2][0] << ";\t" |
243 |
> |
|
244 |
> |
<< entry_plug->Hmat[0][1] << "\t" |
245 |
> |
<< entry_plug->Hmat[1][1] << "\t" |
246 |
> |
<< entry_plug->Hmat[2][1] << ";\t" |
247 |
|
|
248 |
< |
outFile << currentTime << ";\t" |
249 |
< |
<< entry_plug->Hmat[0][0] << "\t" |
250 |
< |
<< entry_plug->Hmat[1][0] << "\t" |
187 |
< |
<< entry_plug->Hmat[2][0] << ";\t" |
188 |
< |
|
189 |
< |
<< entry_plug->Hmat[0][1] << "\t" |
190 |
< |
<< entry_plug->Hmat[1][1] << "\t" |
191 |
< |
<< entry_plug->Hmat[2][1] << ";\t" |
248 |
> |
<< entry_plug->Hmat[0][2] << "\t" |
249 |
> |
<< entry_plug->Hmat[1][2] << "\t" |
250 |
> |
<< entry_plug->Hmat[2][2] << ";"; |
251 |
|
|
252 |
< |
<< entry_plug->Hmat[0][2] << "\t" |
253 |
< |
<< entry_plug->Hmat[1][2] << "\t" |
254 |
< |
<< entry_plug->Hmat[2][2] << ";"; |
255 |
< |
//write out additional parameters, such as chi and eta |
197 |
< |
outFile << entry_plug->the_integrator->getAdditionalParameters(); |
198 |
< |
outFile << endl; |
199 |
< |
|
252 |
> |
//write out additional parameters, such as chi and eta |
253 |
> |
*outFile[k] << entry_plug->the_integrator->getAdditionalParameters() << endl; |
254 |
> |
} |
255 |
> |
|
256 |
|
for( i=0; i<nAtoms; i++ ){ |
257 |
|
|
258 |
|
atoms[i]->getPos(pos); |
288 |
|
else |
289 |
|
strcat( writeLine, "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0\n" ); |
290 |
|
|
291 |
< |
outFile << writeLine; |
291 |
> |
for(k = 0; k < outFile.size(); k++) |
292 |
> |
*outFile[k] << writeLine; |
293 |
|
} |
294 |
|
|
295 |
|
#else // is_mpi |
318 |
|
nProc = mpiSim->getNumberProcessors(); |
319 |
|
potatoes = new int[nProc]; |
320 |
|
|
321 |
+ |
//write out the comment lines |
322 |
|
for (i = 0; i < nProc; i++) |
323 |
|
potatoes[i] = 0; |
324 |
|
|
325 |
< |
outFile << mpiSim->getTotAtoms() << "\n"; |
325 |
> |
for(k = 0; k < outFile.size(); k++){ |
326 |
> |
*outFile[k] << mpiSim->getTotAtoms() << "\n"; |
327 |
|
|
328 |
< |
outFile << currentTime << ";\t" |
329 |
< |
<< entry_plug->Hmat[0][0] << "\t" |
330 |
< |
<< entry_plug->Hmat[1][0] << "\t" |
331 |
< |
<< entry_plug->Hmat[2][0] << ";\t" |
328 |
> |
*outFile[k] << currentTime << ";\t" |
329 |
> |
<< entry_plug->Hmat[0][0] << "\t" |
330 |
> |
<< entry_plug->Hmat[1][0] << "\t" |
331 |
> |
<< entry_plug->Hmat[2][0] << ";\t" |
332 |
|
|
333 |
< |
<< entry_plug->Hmat[0][1] << "\t" |
334 |
< |
<< entry_plug->Hmat[1][1] << "\t" |
335 |
< |
<< entry_plug->Hmat[2][1] << ";\t" |
333 |
> |
<< entry_plug->Hmat[0][1] << "\t" |
334 |
> |
<< entry_plug->Hmat[1][1] << "\t" |
335 |
> |
<< entry_plug->Hmat[2][1] << ";\t" |
336 |
|
|
337 |
< |
<< entry_plug->Hmat[0][2] << "\t" |
338 |
< |
<< entry_plug->Hmat[1][2] << "\t" |
339 |
< |
<< entry_plug->Hmat[2][2] << ";"; |
337 |
> |
<< entry_plug->Hmat[0][2] << "\t" |
338 |
> |
<< entry_plug->Hmat[1][2] << "\t" |
339 |
> |
<< entry_plug->Hmat[2][2] << ";"; |
340 |
> |
|
341 |
> |
*outFile[k] << entry_plug->the_integrator->getAdditionalParameters() << endl; |
342 |
> |
} |
343 |
|
|
282 |
– |
outFile << entry_plug->the_integrator->getAdditionalParameters(); |
283 |
– |
outFile << endl; |
284 |
– |
outFile.flush(); |
285 |
– |
|
344 |
|
currentIndex = 0; |
345 |
+ |
|
346 |
|
for (i = 0 ; i < mpiSim->getTotAtoms(); i++ ) { |
347 |
|
|
348 |
|
// Get the Node number which has this atom; |
387 |
|
|
388 |
|
} else { |
389 |
|
|
390 |
< |
haveError = 0; |
390 |
> |
haveError = 0; |
391 |
|
which_atom = i; |
392 |
|
|
393 |
< |
local_index = indexArray[currentIndex].first; |
335 |
< |
|
336 |
< |
if (which_atom == indexArray[currentIndex].second) { |
393 |
> |
local_index = indexArray[currentIndex].first; |
394 |
|
|
395 |
< |
atomTypeString = atoms[local_index]->getType(); |
396 |
< |
|
395 |
> |
if (which_atom == indexArray[currentIndex].second) { |
396 |
> |
|
397 |
> |
atomTypeString = atoms[local_index]->getType(); |
398 |
> |
|
399 |
|
atoms[local_index]->getPos(pos); |
400 |
|
atoms[local_index]->getVel(vel); |
401 |
< |
|
401 |
> |
|
402 |
|
atomData6[0] = pos[0]; |
403 |
|
atomData6[1] = pos[1]; |
404 |
|
atomData6[2] = pos[2]; |
431 |
|
|
432 |
|
} else { |
433 |
|
sprintf(painCave.errMsg, |
434 |
< |
"Atom %d not found on processor %d\n", |
435 |
< |
i, worldRank ); |
434 |
> |
"Atom %d not found on processor %d, currentIndex = %d, local_index = %d\n", |
435 |
> |
which_atom, worldRank, currentIndex, local_index ); |
436 |
|
haveError= 1; |
437 |
|
simError(); |
438 |
|
} |
439 |
|
|
440 |
< |
if(haveError) DieDieDie(); |
440 |
> |
if(haveError) DieDieDie(); |
441 |
|
|
442 |
< |
currentIndex ++; |
442 |
> |
currentIndex++; |
443 |
|
} |
444 |
|
// If we've survived to here, format the line: |
445 |
|
|
446 |
|
if (!isDirectional) { |
447 |
|
|
448 |
< |
sprintf( writeLine, |
448 |
> |
sprintf( writeLine, |
449 |
|
"%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t", |
450 |
|
atomTypeString, |
451 |
|
atomData6[0], |
454 |
|
atomData6[3], |
455 |
|
atomData6[4], |
456 |
|
atomData6[5]); |
457 |
< |
|
457 |
> |
|
458 |
|
strcat( writeLine, "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0\n" ); |
459 |
|
|
460 |
|
} else { |
478 |
|
|
479 |
|
} |
480 |
|
|
481 |
< |
outFile << writeLine; |
481 |
> |
for(k = 0; k < outFile.size(); k++) |
482 |
> |
*outFile[k] << writeLine; |
483 |
|
} |
484 |
|
|
485 |
< |
|
486 |
< |
outFile.flush(); |
485 |
> |
for(k = 0; k < outFile.size(); k++) |
486 |
> |
outFile[k]->flush(); |
487 |
> |
|
488 |
|
sprintf( checkPointMsg, |
489 |
|
"Sucessfully took a dump.\n"); |
490 |
+ |
|
491 |
|
MPIcheckPoint(); |
492 |
+ |
|
493 |
|
delete[] potatoes; |
494 |
+ |
|
495 |
|
} else { |
496 |
|
|
497 |
|
// worldRank != 0, so I'm a remote node. |
508 |
|
if (AtomToProcMap[i] == worldRank) { |
509 |
|
|
510 |
|
if (myPotato + 3 >= MAXTAG) { |
511 |
< |
|
511 |
> |
|
512 |
|
// The potato was going to exceed the maximum value, |
513 |
|
// so wrap this processor potato back to 0 (and block until |
514 |
|
// node 0 says we can go: |
515 |
< |
|
515 |
> |
|
516 |
|
MPI_Recv(&myPotato, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, &istatus); |
517 |
|
|
518 |
|
} |
519 |
|
which_atom = i; |
520 |
< |
local_index = indexArray[currentIndex].first; |
520 |
> |
|
521 |
> |
local_index = indexArray[currentIndex].first; |
522 |
|
|
523 |
< |
if (which_atom == indexArray[currentIndex].second) { |
523 |
> |
if (which_atom == indexArray[currentIndex].second) { |
524 |
|
|
525 |
|
atomTypeString = atoms[local_index]->getType(); |
526 |
< |
|
526 |
> |
|
527 |
|
atoms[local_index]->getPos(pos); |
528 |
|
atoms[local_index]->getVel(vel); |
529 |
< |
|
529 |
> |
|
530 |
|
atomData6[0] = pos[0]; |
531 |
|
atomData6[1] = pos[1]; |
532 |
|
atomData6[2] = pos[2]; |
551 |
|
atomData13[7] = q[1]; |
552 |
|
atomData13[8] = q[2]; |
553 |
|
atomData13[9] = q[3]; |
554 |
< |
|
554 |
> |
|
555 |
|
atomData13[10] = dAtom->getJx(); |
556 |
|
atomData13[11] = dAtom->getJy(); |
557 |
|
atomData13[12] = dAtom->getJz(); |
559 |
|
|
560 |
|
} else { |
561 |
|
sprintf(painCave.errMsg, |
562 |
< |
"Atom %d not found on processor %d\n", |
563 |
< |
i, worldRank ); |
562 |
> |
"Atom %d not found on processor %d, currentIndex = %d, local_index = %d\n", |
563 |
> |
which_atom, worldRank, currentIndex, local_index ); |
564 |
|
haveError= 1; |
565 |
|
simError(); |
566 |
|
} |
567 |
< |
|
567 |
> |
|
568 |
|
strncpy(MPIatomTypeString, atomTypeString, MINIBUFFERSIZE); |
569 |
|
|
570 |
|
// null terminate the string before sending (just in case): |
571 |
|
MPIatomTypeString[MINIBUFFERSIZE-1] = '\0'; |
572 |
|
|
573 |
|
MPI_Send(MPIatomTypeString, MINIBUFFERSIZE, MPI_CHAR, 0, |
574 |
< |
myPotato, MPI_COMM_WORLD); |
574 |
> |
myPotato, MPI_COMM_WORLD); |
575 |
|
|
576 |
|
myPotato++; |
577 |
|
|
578 |
|
MPI_Send(&isDirectional, 1, MPI_INT, 0, |
579 |
< |
myPotato, MPI_COMM_WORLD); |
579 |
> |
myPotato, MPI_COMM_WORLD); |
580 |
|
|
581 |
|
myPotato++; |
582 |
|
|