Newer
Older
#endif
#include <stdlib.h>
#include <stdio.h>
#include "mpi.h"
//#include "pardef.h"
#include "util.h"
// NOTA: vedere come e cosa cambia aumentando i valori nAstroPSolved, nAttP e nInstrConst facendoli non usuali ma molto grossi
void aprod(int mode, long int m, long int n, double *vVect, double *knownTerms,
double *systemMatrix, long int *matrixIndex, int *instrCol, int *instrConstrIlung, struct comData comlsqr, time_t *ompSec)
// Parallel definitions
int myid, nproc;
long int *mapNoss, *mapNcoeff;
int nthreads, tid, ntasks;
long **mapForThread;
double zero = 0.0;
double sum, yi;
long int l1, l2;
long int i, i1;
long int l, j, k;
int i2 = 0, j2 = 0, j3 = 0, na = 0;
int setBound[4];
short nAstroPSolved = comlsqr.nAstroPSolved;
long localAstro = comlsqr.VrIdAstroPDim * nAstroPSolved;
long localAstroMax = comlsqr.VrIdAstroPDimMax * nAstroPSolved;
// Initialize.
myid = comlsqr.myid;
nproc = comlsqr.nproc;
mapNcoeff = comlsqr.mapNcoeff;
mapNoss = comlsqr.mapNoss;
nthreads = comlsqr.nthreads;
ntasks= comlsqr.ntasks;
mapForThread = comlsqr.mapForThread;
int multMI = comlsqr.multMI;
long nparam = comlsqr.parOss;
short nAttAxes = comlsqr.nAttAxes;
int numOfExtStar = comlsqr.numOfExtStar;
int numOfBarStar = comlsqr.numOfBarStar;
int numOfExtAttCol = comlsqr.numOfExtAttCol;
long VrIdAstroPDimMax = comlsqr.VrIdAstroPDimMax;
int startingAttColExtConstr = comlsqr.startingAttColExtConstr;
int nOfElextObs = comlsqr.nOfElextObs;
int nEqExtConstr = comlsqr.nEqExtConstr;
int nOfElBarObs = comlsqr.nOfElBarObs;
int nEqBarConstr = comlsqr.nEqBarConstr;
int debugMode = comlsqr.debugMode;
short nInstrPSolved = comlsqr.nInstrPSolved;
int nOfInstrConstr = comlsqr.nOfInstrConstr;
int nElemIC = comlsqr.nElemIC;
short nAttP = comlsqr.nAttP;
short nGlobP = comlsqr.nGlobP;
setBound[0] = comlsqr.setBound[0];
setBound[1] = comlsqr.setBound[1];
setBound[2] = comlsqr.setBound[2];
setBound[3] = comlsqr.setBound[3];
long nDegFreedomAtt = comlsqr.nDegFreedomAtt;
short nAttParAxis = comlsqr.nAttParAxis;
long offsetAttParam = comlsqr.offsetAttParam;
long offsetInstrParam = comlsqr.offsetInstrParam;
long offsetGlobParam = comlsqr.offsetGlobParam;
printf("ERROR: Invalid mode=%d in aprod function\n", mode);
exit(1);
l1 = 0;
l2 = 0;
myid = comlsqr.myid;
if (mode == 1)
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
myid = comlsqr.myid;
if (comlsqr.itn == 1 && debugMode)
printf("PE=%d Aprod1 OpenMP num of threads =%d from thread =%d icycle=%ld comlsqr.itn=%d\n", myid, nthreads, tid, i, comlsqr.itn);
long miValAstro = 0;
long miValAtt = 0;
long jstartAtt = 0;
long jstartAstro = 0;
long lset = 0;
long offLocalAstro = 0;
long offLocalAtt = 0;
long offLocalInstr = 0; //Offset on Instruments
long ixInstr = 0;
int nInstrVal = 0;
offLocalInstr = offsetInstrParam + (localAstroMax - offsetAttParam); //Offset on Instruments
nInstrVal = nAstroPSolved + nAttP;
offLocalAstro = comlsqr.mapStar[myid][0] * nAstroPSolved; //Offset on my mstars
offLocalAtt = localAstroMax - offsetAttParam; //Offset on attitude
long offLocalGlob = offsetGlobParam + (localAstroMax - offsetAttParam); //Offset on GlobP
int nGlobVal = nAstroPSolved + nAttP + nInstrPSolved;
jstartAstro = miValAstro - offLocalAstro;
for(int nt=0; nt < ntasks; nt++ )
{
#pragma omp task label(mode1)
{
for (long ix = mapForThread[nt][0]; ix < mapForThread[nt][2]; ix++) //FUNZIONE DA FARE
{
/*================= INIZIO
sum = 0.; // FARE UNA FUNZIONE CHIAMATA NEL CICLO FOR
/////////////////////////////////////////////////////
/// Mode 1 Astrometric Sect
if (nAstroPSolved)
{
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
lset = ix * nparam;
if (matrixIndex[multMI * ix] != miValAstro)
{
miValAstro = matrixIndex[multMI * ix];
jstartAstro = miValAstro - offLocalAstro;
}
for (long jx = jstartAstro; jx < jstartAstro + nAstroPSolved; jx++)
{
sum = sum + systemMatrix[lset] * vVect[jx];
lset++;
}
}
//////////////////////////////////////////////////////
/// Mode 1 Attitude Sect
if (nAttP)
{
lset = ix * nparam + nAstroPSolved;
miValAtt = matrixIndex[multMI * ix + (multMI - 1)];
for (int nax = 0; nax < nAttAxes; nax++)
{
jstartAtt = miValAtt + offLocalAtt + nax * nDegFreedomAtt;
for (long inpax = jstartAtt; inpax < jstartAtt + nAttParAxis; inpax++)
{
sum = sum + systemMatrix[lset] * vVect[inpax];
lset++;
}
}
}
//////////////////////////////////////////////////////
/// Mode 1 Instrument Sect
if (nInstrPSolved)
{
lset = ix * nparam + nInstrVal;
long iiVal = ix * nInstrPSolved;
for (int inInstr = 0; inInstr < nInstrPSolved; inInstr++)
{
ixInstr = offLocalInstr + instrCol[iiVal + inInstr];
sum = sum + systemMatrix[lset] * vVect[ixInstr];
lset++;
}
}
//////////////////////////////////////////////////////
/// Mode 1 Global sect
if (nGlobP)
{
lset = ix * nparam + nGlobVal;
for (long inGlob = offLocalGlob; inGlob < offLocalGlob + nGlobP; inGlob++)
{
sum = sum + systemMatrix[lset] * vVect[inGlob];
lset++;
}
}
//////////////////////////////////////////////////////
// knownTerms[ix] += sum;
================ FINE */
knownTerms[ix] += aprodM1Obs(ix,comlsqr, vVect, systemMatrix, matrixIndex, instrCol);
/// Mode 1 ExtConstr
if (nEqExtConstr)
{
long offExtAtt;
long offExtAttConstr = VrIdAstroPDimMax * nAstroPSolved + startingAttColExtConstr;
long vVIx;
long ktIx = mapNoss[myid];
long offExtConstr;
for (int iexc = 0; iexc < nEqExtConstr; iexc++)
{
sum = 0.0;
offExtConstr = mapNcoeff[myid] + iexc * nOfElextObs;
for (int j3 = 0; j3 < numOfExtStar * nAstroPSolved; j3++)
sum += systemMatrix[offExtConstr + j3] * vVect[j3];
for (int nax = 0; nax < nAttAxes; nax++)
{
offExtAtt = offExtConstr + numOfExtStar * nAstroPSolved + nax * numOfExtAttCol;
vVIx = offExtAttConstr + nax * nDegFreedomAtt;
for (int j3 = 0; j3 < numOfExtAttCol; j3++)
sum += systemMatrix[offExtAtt + j3] * vVect[vVIx + j3];
}
knownTerms[ktIx + iexc] += sum;
} //for iexc
}
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
//////////////////////////////////////////////////////
/// Mode 1 BarConstr
if (nEqBarConstr)
{
long offBarConstr = mapNcoeff[myid] + nOfElextObs * nEqExtConstr;
long offBarConstrIx;
long ktIx = mapNoss[myid] + nEqExtConstr;
for (int iexc = 0; iexc < nEqBarConstr; iexc++)
{
sum = 0.0;
offBarConstrIx = offBarConstr + iexc * nOfElBarObs;
for (int j3 = 0; j3 < numOfBarStar * nAstroPSolved; j3++)
sum += systemMatrix[offBarConstrIx + j3] * vVect[j3];
knownTerms[ktIx + iexc] += sum;
} //for iexc
}
//////////////////////////////////////////////////////
/// Mode 1 InstrConstr
if (nOfInstrConstr)
{
long offSetInstr = 0;
long offSetInstrInc = 0;
long offSetInstrInc1 = 0;
long vVix = 0;
long offvV = 0;
long offSetInstrConstr = mapNcoeff[myid] + nOfElextObs * nEqExtConstr + nOfElBarObs * nEqBarConstr;
long offSetInstrConstr1 = VrIdAstroPDimMax * nAstroPSolved + nDegFreedomAtt * nAttAxes;
long ktIx = mapNoss[myid] + nEqExtConstr + nEqBarConstr;
for (int i1 = myid; i1 < nOfInstrConstr; i1 = i1 + nproc)
{
sum = 0.0;
offSetInstrInc = offSetInstrConstr;
offSetInstr = 0;
for (int m = 0; m < i1; m++)
{
offSetInstrInc += instrConstrIlung[m];
offSetInstr += instrConstrIlung[m];
}
offvV = mapNoss[myid] * nInstrPSolved + offSetInstr;
for (int j3 = 0; j3 < instrConstrIlung[i1]; j3++)
{
vVix = instrCol[offvV + j3];
sum += systemMatrix[offSetInstrInc + j3] * vVect[offSetInstrConstr1 + vVix];
}
knownTerms[ktIx + i1] += sum;
}
}
//////////////////////////////////////////////////////
} //pragma
*ompSec += time(NULL) - startTime;
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("PE=%d AprodTiming: mode=1 OmpSec=%ld\n", myid, time(NULL) - startTime);
#pragma omp taskwait
time_t startTime = time(NULL);
int count = 0;
myid = comlsqr.myid;
printf("vVect[1]: %f:\n", vVect[1] );
/////////////////////////////////////////////////////
/// Mode 2 Astrometric Sect
if (nAstroPSolved)
{
long offLocalAstro = comlsqr.mapStar[myid][0] * nAstroPSolved;
long lset;
long jstartAstro;
for(int nt=0; nt < ntasks; nt++ )
{
#pragma omp target device (fpga) copy_deps
#pragma omp task label(vVect_nAstroPSolved) //out(vVect[jstartAstro : jstartAstro + nAstroPSolved-1])
{
lset = mapForThread[nt][0] * nparam;
double taskLocalSum;
jstartAstro = 0;
for (long ix = mapForThread[nt][0]; ix < mapForThread[nt][2]; ix++) //QUI e altrove in for(long ix..) richiama una funzione
{
//======== INIZIO
lset=ix*nparam;
long miValAstro = 0;
if (matrixIndex[multMI * ix] != miValAstro)
{
miValAstro = matrixIndex[multMI * ix];
jstartAstro = miValAstro - offLocalAstro;
}
for (long jx = jstartAstro; jx < jstartAstro + nAstroPSolved; jx++)
{
taskLocalSum = systemMatrix[lset] * knownTerms[ix];
//vVect[jx] = 0.000001;
if(jx == 1 && false)
{
printf("nAstroPSolved nt:%d ix:%ld vVect[jx]:%f\n\n",nt,ix,vVect[jx] );
}
vVect[jx] += taskLocalSum;
lset++;
} //for jx
//lset += nparam - nAstroPSolved; //**QUI attenzio0ne che nel caso di "funzioni avviate in parallelo succede casino! è modificata!! in lste messa all'inizio!
//========== FINE
aprodM2AstroP(ix, nt, comlsqr, vVect, systemMatrix, matrixIndex, knownTerms);
} //for ix
}
}
}
/////////////////////////////////////////////////////
/////////////////////////////////////////////////////
/// Mode 2 Attitude Sect
if (nAttP)
{
long mix;
long offj = (localAstroMax - offsetAttParam);
long vVix;
long lset;
for(int nt=0; nt < ntasks; nt++ )
{
#pragma omp task label(nAttP) commutative(vVect)//out(vVect[mix : mix + (nAttAxes-1) * nDegFreedomAtt])
{
for (long ix = mapForThread[nt][0]; ix < mapForThread[nt][2]; ix++)
{
//============= INIZIO
/*
lset = nparam * ix + nAstroPSolved;
mix = matrixIndex[multMI * ix + (multMI - 1)] + offj;
for (int ly = 0; ly < nAttAxes; ly++)
{
vVix = mix + ly * nDegFreedomAtt;
if(vVix == 1)
{
printf("nAttP nt %d %ld\n\n",nt,ix);
}
for (int lx = 0; lx < nAttParAxis; lx++)
{
localSum = systemMatrix[lset] * knownTerms[ix];
// #pragma omp atomic
// vVect[vVix] = 0.000001;
vVect[vVix] += localSum;
lset++;
vVix++;
} //for lx
} //for ly
//========== FINE
*/
aprodM2AttP(ix, nt, comlsqr, vVect, systemMatrix, matrixIndex, knownTerms);
} //for ix
}
}
}
/////////////////////////////////////////////////////
/// Mode 2 Instrument Sect
printf("\t\t\t\tnOfInstrConstr: %d", nOfInstrConstr);
if (nOfInstrConstr )
{
int offLset = nAstroPSolved + nAttP;
long iix;
long offj = offsetInstrParam + (localAstroMax - offsetAttParam);
long vVix;
for(int nt=0; nt < ntasks; nt++ )
{
#pragma omp task label(nOfInstrConstr) commutative(vVect) //out(vVect[mix : mix + (nAttAxes-1) * nDegFreedomAtt])
{
long lset;
for (long ix = mapForThread[nt][0]; ix < mapForThread[nt][2]; ix++)
{
//========== INIZIO
/*
lset = nparam * ix + offLset;
iix = ix * nInstrPSolved;
for (int ly = 0; ly < nInstrPSolved; ly++)
{
vVix = offj + instrCol[iix + ly];
if(vVix == 1)
{
printf("nOfInstrConstr nt %d %ld\n\n",nt,ix);
}
localSum = systemMatrix[lset] * knownTerms[ix];
//#pragma omp atomic
//vVect[vVix] = 0.000001;
vVect[vVix] += localSum;
lset++;
} //for ly
//========== FINE
*/
aprodM2InstrP(ix, nt, comlsqr, vVect, systemMatrix, instrCol, knownTerms);
} //for ix
}
}
}
*ompSec += time(NULL) - startTime;
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("PE=%d AprodTiming: mode=2.1 OmpSec=%ld\n", myid, time(NULL) - startTime);
/////////////////////////////////////////////////////
/// Mode 2 Global Sect
if (nGlobP )
{
time_t startTime_nGlobP = time(NULL);
long lset;
int offLset = nAstroPSolved + nAttP + nInstrPSolved;
long offj = offsetGlobParam + (localAstroMax - offsetAttParam);
long vVix;
for (long ix = 0; ix < mapNoss[myid]; ix++)
{
// ==================== INIZIO
/*
lset = nparam * ix + offLset;
for (long ly = 0; ly < nGlobP; ly++)
{
vVix = offj + ly;
if(vVix == 1)
{
printf("nGlobP nt %ld\n\n",ix);
}
localSum = systemMatrix[lset] * knownTerms[ix];
vVect[vVix] += localSum;
lset++;
} //for ly
// ============ FINE
*/
aprodM2GlobP(ix, ntasks, comlsqr, vVect, systemMatrix, knownTerms);
} //for ix
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("\tPE=%d AprodTiming: nGlobP sec=%ld\n", myid, time(NULL) - startTime_nGlobP);
}
////////////////////////////////////////////////////
myid = comlsqr.myid;
localSum = 0.0;
printf("\t\t\t\nEqExtConstr: %d nEqBarConstr: %d nOfInstrConstr: %d", nEqExtConstr,nEqBarConstr,nOfInstrConstr);
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
{
time_t startTime_nEqExtConstr = time(NULL);
long offExtStarConstrEq;
long off2;
long off3;
for (int ix = 0; ix < nEqExtConstr; ix++)
{ //stars
yi = knownTerms[mapNoss[myid] + ix];
offExtStarConstrEq = mapNcoeff[myid] + ix * nOfElextObs; //Star offset on the row ix (all the row)
//FV_ EDIT ompSs
//#pragma omp for
for (long yx = 0; yx < numOfExtStar; yx++)
{
off3 = yx * nAstroPSolved;
off2 = offExtStarConstrEq + off3;
for (int j2 = 0; j2 < nAstroPSolved; j2++)
{
localSum = systemMatrix[off2 + j2] * yi;
vVect[j2 + off3] += localSum;
}
}
//FV_ EDIT ompSs
//#pragma omp barrier
} //for ix
long offExtAttConstrEq;
long offExtUnk;
long off1 = VrIdAstroPDimMax * nAstroPSolved + startingAttColExtConstr;
for (int ix = 0; ix < nEqExtConstr; ix++)
{ //att
yi = knownTerms[mapNoss[myid] + ix];
offExtAttConstrEq = mapNcoeff[myid] + ix * nOfElextObs; //Att offset on the row ix (all the row)
offExtAttConstrEq += numOfExtStar * nAstroPSolved; //Att offset inside ix row
for (int nax = 0; nax < nAttAxes; nax++)
{
offExtUnk = off1 + nax * nDegFreedomAtt; // Att offset for Unk array on extConstr
off2 = offExtAttConstrEq + nax * numOfExtAttCol;
//FV_ EDIT ompSs
//#pragma omp for
for (int jx = 0; jx < numOfExtAttCol; jx++)
{
localSum = systemMatrix[off2 + jx] * yi;
vVect[offExtUnk + jx] += localSum;
}
//FV_ EDIT ompSs
//#pragma omp barrier
}
}
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("\tPE=%d AprodTiming: nEqExtConstr sec=%ld\n", myid, time(NULL) - startTime_nEqExtConstr);
}
//////////////////////////////////////////////////////
/// Mode 2 BarConstr
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
{
time_t startTime_nEqBarConstr = time(NULL);
localSum = 0.0;
long off3;
long off2;
for (int ix = 0; ix < nEqBarConstr; ix++)
{ //stars
yi = knownTerms[mapNoss[myid] + nEqExtConstr + ix];
long offBarStarConstrEq = mapNcoeff[myid] + nEqExtConstr * nOfElextObs + ix * nOfElBarObs; //Star offset on the row i2 (all the row)
//FV_ EDIT ompSs
//#pragma omp for
for (long yx = 0; yx < numOfBarStar; yx++)
{
off3 = yx * nAstroPSolved;
off2 = offBarStarConstrEq + off3;
for (int j2 = 0; j2 < nAstroPSolved; j2++)
{
localSum = systemMatrix[off2 + j2] * yi;
vVect[j2 + off3] += localSum;
}
}
//FV_ EDIT ompSs
//#pragma omp barrier
} //for i2
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("\tPE=%d AprodTiming: nEqBarConstr sec=%ld\n", myid, time(NULL) - startTime_nEqBarConstr);
}
//////////////////////////////////////////////////////
/// Mode 2 InstrConstr
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
{
time_t startTime_nOfInstrConstr = time(NULL);
localSum = 0.0;
int offSetInstr;
long off1;
long offInstrUnk = VrIdAstroPDimMax * nAstroPSolved + nAttAxes * nDegFreedomAtt;
long off2 = mapNoss[myid] + nEqExtConstr + nEqBarConstr;
long off3 = mapNcoeff[myid] + nOfElextObs * nEqExtConstr + nOfElBarObs * nEqBarConstr;
long off4 = mapNoss[myid] * nInstrPSolved;
long off5;
long off6;
for (int k1 = myid; k1 < nOfInstrConstr; k1 = k1 + nproc)
{
yi = knownTerms[off2 + k1];
offSetInstr = 0;
for (int m = 0; m < k1; m++)
offSetInstr += instrConstrIlung[m];
off1 = off3 + offSetInstr;
off5 = off4 + offSetInstr;
//FV_ EDIT ompSs
//#pragma omp for
for (int j = 0; j < instrConstrIlung[k1]; j++)
{
localSum = systemMatrix[off1 + j] * yi;
off6 = offInstrUnk + instrCol[off5 + j];
vVect[off6] += localSum;
}
}
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("\tPE=%d AprodTiming: nOfInstrConstr sec=%ld\n", myid, time(NULL) - startTime_nOfInstrConstr);
}
////////////
*ompSec += time(NULL) - startTime;
if (comlsqr.itn <= 2 && (myid == 0 || myid == nproc - 1 || debugMode == 1))
printf("PE=%d AprodTiming: mode=2.2 OmpSec=%ld\n", myid, time(NULL) - startTime);
} // else if(mode==2)
}