1 2 /* 3 Defines projective product routines where A is a MPIAIJ matrix 4 C = P^T * A * P 5 */ 6 7 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 8 #include <../src/mat/utils/freespace.h> 9 #include <../src/mat/impls/aij/mpi/mpiaij.h> 10 #include <petscbt.h> 11 #include <petsctime.h> 12 13 #define PTAP_PROFILE 14 15 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 16 #undef __FUNCT__ 17 #define __FUNCT__ "MatDestroy_MPIAIJ_PtAP" 18 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 19 { 20 PetscErrorCode ierr; 21 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data; 22 Mat_PtAPMPI *ptap=a->ptap; 23 24 PetscFunctionBegin; 25 if (ptap) { 26 Mat_Merge_SeqsToMPI *merge=ptap->merge; 27 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 28 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 29 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 30 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 31 ierr = MatDestroy(&ptap->A_loc);CHKERRQ(ierr); /* used by MatTransposeMatMult() */ 32 ierr = MatDestroy(&ptap->Rd);CHKERRQ(ierr); 33 ierr = MatDestroy(&ptap->Ro);CHKERRQ(ierr); 34 ierr = MatDestroy(&ptap->AP);CHKERRQ(ierr); 35 if (ptap->api) {ierr = PetscFree(ptap->api);CHKERRQ(ierr);} 36 if (ptap->apj) {ierr = PetscFree(ptap->apj);CHKERRQ(ierr);} 37 if (ptap->apa) {ierr = PetscFree(ptap->apa);CHKERRQ(ierr);} 38 if (merge) { 39 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 40 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 41 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 42 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 43 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 44 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 45 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 46 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 47 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 48 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 49 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 50 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 51 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 52 ierr = merge->destroy(A);CHKERRQ(ierr); 53 ierr = PetscFree(ptap->merge);CHKERRQ(ierr); 54 } 55 ierr = PetscFree(ptap);CHKERRQ(ierr); 56 } 57 PetscFunctionReturn(0); 58 } 59 60 #undef __FUNCT__ 61 #define __FUNCT__ "MatDuplicate_MPIAIJ_MatPtAP" 62 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 63 { 64 PetscErrorCode ierr; 65 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 66 Mat_PtAPMPI *ptap = a->ptap; 67 Mat_Merge_SeqsToMPI *merge = ptap->merge; 68 69 PetscFunctionBegin; 70 ierr = (*merge->duplicate)(A,op,M);CHKERRQ(ierr); 71 72 (*M)->ops->destroy = merge->destroy; 73 (*M)->ops->duplicate = merge->duplicate; 74 PetscFunctionReturn(0); 75 } 76 77 #undef __FUNCT__ 78 #define __FUNCT__ "MatPtAP_MPIAIJ_MPIAIJ" 79 PetscErrorCode MatPtAP_MPIAIJ_MPIAIJ(Mat A,Mat P,MatReuse scall,PetscReal fill,Mat *C) 80 { 81 PetscErrorCode ierr; 82 83 PetscFunctionBegin; 84 if (scall == MAT_INITIAL_MATRIX) { 85 ierr = PetscLogEventBegin(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 86 ierr = MatPtAPSymbolic_MPIAIJ_MPIAIJ(A,P,fill,C);CHKERRQ(ierr); 87 ierr = PetscLogEventEnd(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 88 } 89 ierr = PetscLogEventBegin(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 90 ierr = MatPtAPNumeric_MPIAIJ_MPIAIJ(A,P,*C);CHKERRQ(ierr); 91 ierr = PetscLogEventEnd(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 92 PetscFunctionReturn(0); 93 } 94 95 #undef __FUNCT__ 96 #define __FUNCT__ "MatPtAPSymbolic_MPIAIJ_MPIAIJ" 97 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 98 { 99 PetscErrorCode ierr; 100 Mat Cmpi; 101 Mat_PtAPMPI *ptap; 102 PetscFreeSpaceList free_space=NULL,current_space=NULL; 103 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 104 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 105 Mat_SeqAIJ *p_loc,*p_oth; 106 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pdti,*pdtj,*poti,*potj,*ptJ; 107 PetscInt *adi=ad->i,*aj,*aoi=ao->i,nnz; 108 PetscInt *lnk,*owners_co,*coi,*coj,i,k,pnz,row; 109 PetscInt am=A->rmap->n,pN=P->cmap->N,pm=P->rmap->n,pn=P->cmap->n; 110 PetscBT lnkbt; 111 MPI_Comm comm; 112 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0; 113 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 114 PetscInt len,proc,*dnz,*onz,*owners; 115 PetscInt nzi,*pti,*ptj; 116 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 117 MPI_Request *swaits,*rwaits; 118 MPI_Status *sstatus,rstatus; 119 Mat_Merge_SeqsToMPI *merge; 120 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,pon,nspacedouble=0,j,ap_rmax=0; 121 PetscReal afill=1.0,afill_tmp; 122 PetscInt rmax; 123 124 PetscFunctionBegin; 125 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 126 127 /* check if matrix local sizes are compatible */ 128 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend) { 129 SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, Arow (%D, %D) != Prow (%D,%D)",A->rmap->rstart,A->rmap->rend,P->rmap->rstart,P->rmap->rend); 130 } 131 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend) { 132 SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, Acol (%D, %D) != Prow (%D,%D)",A->cmap->rstart,A->cmap->rend,P->rmap->rstart,P->rmap->rend); 133 } 134 135 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 136 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 137 138 /* create struct Mat_PtAPMPI and attached it to C later */ 139 ierr = PetscNew(&ptap);CHKERRQ(ierr); 140 ierr = PetscNew(&merge);CHKERRQ(ierr); 141 ptap->merge = merge; 142 ptap->reuse = MAT_INITIAL_MATRIX; 143 144 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 145 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 146 147 /* get P_loc by taking all local rows of P */ 148 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 149 150 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 151 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 152 pi_loc = p_loc->i; pj_loc = p_loc->j; 153 pi_oth = p_oth->i; pj_oth = p_oth->j; 154 155 /* (1) compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth (api,apj) */ 156 /*--------------------------------------------------------------------------*/ 157 ierr = PetscMalloc1(am+1,&api);CHKERRQ(ierr); 158 api[0] = 0; 159 160 /* create and initialize a linked list */ 161 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 162 163 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) -OOM for ex56, np=8k on Intrepid! */ 164 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 165 166 current_space = free_space; 167 168 for (i=0; i<am; i++) { 169 /* diagonal portion of A */ 170 nzi = adi[i+1] - adi[i]; 171 aj = ad->j + adi[i]; 172 for (j=0; j<nzi; j++) { 173 row = aj[j]; 174 pnz = pi_loc[row+1] - pi_loc[row]; 175 Jptr = pj_loc + pi_loc[row]; 176 /* add non-zero cols of P into the sorted linked list lnk */ 177 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 178 } 179 /* off-diagonal portion of A */ 180 nzi = aoi[i+1] - aoi[i]; 181 aj = ao->j + aoi[i]; 182 for (j=0; j<nzi; j++) { 183 row = aj[j]; 184 pnz = pi_oth[row+1] - pi_oth[row]; 185 Jptr = pj_oth + pi_oth[row]; 186 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 187 } 188 apnz = lnk[0]; 189 api[i+1] = api[i] + apnz; 190 if (ap_rmax < apnz) ap_rmax = apnz; 191 192 /* if free space is not available, double the total space in the list */ 193 if (current_space->local_remaining<apnz) { 194 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 195 nspacedouble++; 196 } 197 198 /* Copy data into free space, then initialize lnk */ 199 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 200 201 current_space->array += apnz; 202 current_space->local_used += apnz; 203 current_space->local_remaining -= apnz; 204 } 205 206 /* Allocate space for apj, initialize apj, and */ 207 /* destroy list of free space and other temporary array(s) */ 208 ierr = PetscMalloc1(api[am]+1,&apj);CHKERRQ(ierr); 209 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 210 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1); 211 if (afill_tmp > afill) afill = afill_tmp; 212 213 /* (2) determine symbolic Co=(p->B)^T*AP - send to others (coi,coj)*/ 214 /*-----------------------------------------------------------------*/ 215 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 216 217 /* then, compute symbolic Co = (p->B)^T*AP */ 218 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 219 >= (num of nonzero rows of C_seq) - pn */ 220 ierr = PetscMalloc1(pon+1,&coi);CHKERRQ(ierr); 221 coi[0] = 0; 222 223 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 224 nnz = fill*(poti[pon] + api[am]); 225 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 226 current_space = free_space; 227 228 for (i=0; i<pon; i++) { 229 pnz = poti[i+1] - poti[i]; 230 ptJ = potj + poti[i]; 231 for (j=0; j<pnz; j++) { 232 row = ptJ[j]; /* row of AP == col of Pot */ 233 apnz = api[row+1] - api[row]; 234 Jptr = apj + api[row]; 235 /* add non-zero cols of AP into the sorted linked list lnk */ 236 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 237 } 238 nnz = lnk[0]; 239 240 /* If free space is not available, double the total space in the list */ 241 if (current_space->local_remaining<nnz) { 242 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 243 nspacedouble++; 244 } 245 246 /* Copy data into free space, and zero out denserows */ 247 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 248 249 current_space->array += nnz; 250 current_space->local_used += nnz; 251 current_space->local_remaining -= nnz; 252 253 coi[i+1] = coi[i] + nnz; 254 } 255 256 ierr = PetscMalloc1(coi[pon],&coj);CHKERRQ(ierr); 257 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 258 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]+1); 259 if (afill_tmp > afill) afill = afill_tmp; 260 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 261 262 /* (3) send j-array (coj) of Co to other processors */ 263 /*--------------------------------------------------*/ 264 ierr = PetscCalloc1(size,&merge->len_s);CHKERRQ(ierr); 265 len_s = merge->len_s; 266 merge->nsend = 0; 267 268 269 /* determine row ownership */ 270 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 271 merge->rowmap->n = pn; 272 merge->rowmap->bs = 1; 273 274 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 275 owners = merge->rowmap->range; 276 277 /* determine the number of messages to send, their lengths */ 278 ierr = PetscMalloc2(size,&len_si,size,&sstatus);CHKERRQ(ierr); 279 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 280 ierr = PetscMalloc1(size+2,&owners_co);CHKERRQ(ierr); 281 282 proc = 0; 283 for (i=0; i<pon; i++) { 284 while (prmap[i] >= owners[proc+1]) proc++; 285 len_si[proc]++; /* num of rows in Co(=Pt*AP) to be sent to [proc] */ 286 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 287 } 288 289 len = 0; /* max length of buf_si[], see (4) */ 290 owners_co[0] = 0; 291 for (proc=0; proc<size; proc++) { 292 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 293 if (len_s[proc]) { 294 merge->nsend++; 295 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 296 len += len_si[proc]; 297 } 298 } 299 300 /* determine the number and length of messages to receive for coi and coj */ 301 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 302 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 303 304 /* post the Irecv and Isend of coj */ 305 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 306 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 307 ierr = PetscMalloc1(merge->nsend+1,&swaits);CHKERRQ(ierr); 308 for (proc=0, k=0; proc<size; proc++) { 309 if (!len_s[proc]) continue; 310 i = owners_co[proc]; 311 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 312 k++; 313 } 314 315 /* receives and sends of coj are complete */ 316 for (i=0; i<merge->nrecv; i++) { 317 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 318 } 319 ierr = PetscFree(rwaits);CHKERRQ(ierr); 320 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 321 322 /* (4) send and recv coi */ 323 /*-----------------------*/ 324 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 325 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 326 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 327 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 328 for (proc=0,k=0; proc<size; proc++) { 329 if (!len_s[proc]) continue; 330 /* form outgoing message for i-structure: 331 buf_si[0]: nrows to be sent 332 [1:nrows]: row index (global) 333 [nrows+1:2*nrows+1]: i-structure index 334 */ 335 /*-------------------------------------------*/ 336 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 337 buf_si_i = buf_si + nrows+1; 338 buf_si[0] = nrows; 339 buf_si_i[0] = 0; 340 nrows = 0; 341 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 342 nzi = coi[i+1] - coi[i]; 343 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 344 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 345 nrows++; 346 } 347 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 348 k++; 349 buf_si += len_si[proc]; 350 } 351 i = merge->nrecv; 352 while (i--) { 353 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 354 } 355 ierr = PetscFree(rwaits);CHKERRQ(ierr); 356 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 357 358 ierr = PetscFree2(len_si,sstatus);CHKERRQ(ierr); 359 ierr = PetscFree(len_ri);CHKERRQ(ierr); 360 ierr = PetscFree(swaits);CHKERRQ(ierr); 361 ierr = PetscFree(buf_s);CHKERRQ(ierr); 362 363 /* (5) compute the local portion of C (mpi mat) */ 364 /*----------------------------------------------*/ 365 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 366 367 /* allocate pti array and free space for accumulating nonzero column info */ 368 ierr = PetscMalloc1(pn+1,&pti);CHKERRQ(ierr); 369 pti[0] = 0; 370 371 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 372 nnz = fill*(pi_loc[pm] + api[am]); 373 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 374 current_space = free_space; 375 376 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 377 for (k=0; k<merge->nrecv; k++) { 378 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 379 nrows = *buf_ri_k[k]; 380 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 381 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 382 } 383 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 384 rmax = 0; 385 for (i=0; i<pn; i++) { 386 /* add pdt[i,:]*AP into lnk */ 387 pnz = pdti[i+1] - pdti[i]; 388 ptJ = pdtj + pdti[i]; 389 for (j=0; j<pnz; j++) { 390 row = ptJ[j]; /* row of AP == col of Pt */ 391 apnz = api[row+1] - api[row]; 392 Jptr = apj + api[row]; 393 /* add non-zero cols of AP into the sorted linked list lnk */ 394 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 395 } 396 397 /* add received col data into lnk */ 398 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 399 if (i == *nextrow[k]) { /* i-th row */ 400 nzi = *(nextci[k]+1) - *nextci[k]; 401 Jptr = buf_rj[k] + *nextci[k]; 402 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 403 nextrow[k]++; nextci[k]++; 404 } 405 } 406 nnz = lnk[0]; 407 408 /* if free space is not available, make more free space */ 409 if (current_space->local_remaining<nnz) { 410 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 411 nspacedouble++; 412 } 413 /* copy data into free space, then initialize lnk */ 414 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 415 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 416 417 current_space->array += nnz; 418 current_space->local_used += nnz; 419 current_space->local_remaining -= nnz; 420 421 pti[i+1] = pti[i] + nnz; 422 if (nnz > rmax) rmax = nnz; 423 } 424 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 425 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 426 427 ierr = PetscMalloc1(pti[pn]+1,&ptj);CHKERRQ(ierr); 428 ierr = PetscFreeSpaceContiguous(&free_space,ptj);CHKERRQ(ierr); 429 afill_tmp = (PetscReal)pti[pn]/(pi_loc[pm] + api[am]+1); 430 if (afill_tmp > afill) afill = afill_tmp; 431 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 432 433 /* (6) create symbolic parallel matrix Cmpi */ 434 /*------------------------------------------*/ 435 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 436 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 437 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 438 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 439 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 440 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 441 442 merge->bi = pti; /* Cseq->i */ 443 merge->bj = ptj; /* Cseq->j */ 444 merge->coi = coi; /* Co->i */ 445 merge->coj = coj; /* Co->j */ 446 merge->buf_ri = buf_ri; 447 merge->buf_rj = buf_rj; 448 merge->owners_co = owners_co; 449 merge->destroy = Cmpi->ops->destroy; 450 merge->duplicate = Cmpi->ops->duplicate; 451 452 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 453 Cmpi->assembled = PETSC_FALSE; 454 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 455 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 456 457 /* attach the supporting struct to Cmpi for reuse */ 458 c = (Mat_MPIAIJ*)Cmpi->data; 459 c->ptap = ptap; 460 ptap->api = api; 461 ptap->apj = apj; 462 ptap->rmax = ap_rmax; 463 *C = Cmpi; 464 465 /* flag 'scalable' determines which implementations to be used: 466 0: do dense axpy in MatPtAPNumeric() - fast, but requires storage of a nonscalable dense array apa; 467 1: do sparse axpy in MatPtAPNumeric() - might slow, uses a sparse array apa */ 468 /* set default scalable */ 469 ptap->scalable = PETSC_FALSE; //PETSC_TRUE; 470 471 ierr = PetscOptionsGetBool(((PetscObject)Cmpi)->prefix,"-matptap_scalable",&ptap->scalable,NULL);CHKERRQ(ierr); 472 if (!ptap->scalable) { /* Do dense axpy */ 473 ierr = PetscCalloc1(pN,&ptap->apa);CHKERRQ(ierr); 474 ierr = MatTranspose_SeqAIJ(p->A,MAT_INITIAL_MATRIX,&ptap->Rd);CHKERRQ(ierr); 475 ierr = MatTranspose_SeqAIJ(p->B,MAT_INITIAL_MATRIX,&ptap->Ro);CHKERRQ(ierr); 476 ierr = MatMatMult(A,P,MAT_INITIAL_MATRIX,2.0,&ptap->AP);CHKERRQ(ierr); 477 } else { 478 ierr = PetscCalloc1(ap_rmax+1,&ptap->apa);CHKERRQ(ierr); 479 } 480 481 #if defined(PETSC_USE_INFO) 482 if (pti[pn] != 0) { 483 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 484 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%g,&C) for best performance.\n",(double)afill);CHKERRQ(ierr); 485 } else { 486 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 487 } 488 #endif 489 PetscFunctionReturn(0); 490 } 491 492 #undef __FUNCT__ 493 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 494 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 495 { 496 PetscErrorCode ierr; 497 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 498 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 499 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 500 Mat_SeqAIJ *p_loc,*p_oth; 501 Mat_PtAPMPI *ptap; 502 Mat_Merge_SeqsToMPI *merge; 503 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 504 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 505 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 506 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 507 PetscInt am =A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 508 MPI_Comm comm; 509 PetscMPIInt size,rank,taga,*len_s; 510 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 511 PetscInt **buf_ri,**buf_rj; 512 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 513 MPI_Request *s_waits,*r_waits; 514 MPI_Status *status; 515 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 516 PetscInt *api,*apj,*coi,*coj; 517 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 518 PetscBool scalable; 519 #if defined(PTAP_PROFILE) 520 PetscLogDouble t0,t1,t2,eP,t3,t4,et2_AP=0.0,ePtAP=0.0,t2_0,t2_1,t2_2,t_tran0,t_tran1,t_tran2,t_tran3; 521 PetscLogDouble eR,eAP,eCseq,eCmpi; 522 #endif 523 524 PetscFunctionBegin; 525 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 526 #if defined(PTAP_PROFILE) 527 ierr = PetscTime(&t0);CHKERRQ(ierr); 528 #endif 529 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 530 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 531 532 ptap = c->ptap; 533 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_INCOMP,"MatPtAP() has not been called to create matrix C yet, cannot use MAT_REUSE_MATRIX"); 534 merge = ptap->merge; 535 apa = ptap->apa; 536 scalable = ptap->scalable; 537 538 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 539 /*-----------------------------------------------------*/ 540 if (ptap->reuse == MAT_INITIAL_MATRIX) { 541 /* P_oth and P_loc are obtained in MatPtASymbolic(), skip calling MatGetBrowsOfAoCols() and MatMPIAIJGetLocalMat() */ 542 ptap->reuse = MAT_REUSE_MATRIX; 543 } else { /* update numerical values of P_oth and P_loc */ 544 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 545 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 546 } 547 #if defined(PTAP_PROFILE) 548 ierr = PetscTime(&t1);CHKERRQ(ierr); 549 eP = t1-t0; 550 #endif 551 /* 552 printf("[%d] Ad: %d, %d; Ao: %d, %d; P_loc: %d, %d; P_oth %d, %d;\n",rank, 553 a->A->rmap->N,a->A->cmap->N,a->B->rmap->N,a->B->cmap->N, 554 ptap->P_loc->rmap->N,ptap->P_loc->cmap->N, 555 ptap->P_oth->rmap->N,ptap->P_oth->cmap->N); 556 */ 557 558 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 559 /*--------------------------------------------------------------*/ 560 /* get data from symbolic products */ 561 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 562 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 563 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 564 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 565 566 coi = merge->coi; coj = merge->coj; 567 ierr = PetscCalloc1(coi[pon]+1,&coa);CHKERRQ(ierr); 568 569 bi = merge->bi; bj = merge->bj; 570 owners = merge->rowmap->range; 571 ierr = PetscCalloc1(bi[cm]+1,&ba);CHKERRQ(ierr); /* ba: Cseq->a */ 572 573 api = ptap->api; apj = ptap->apj; 574 575 if (!scalable) { /* Do dense axpy on apa (length of pN, stores A[i,:]*P) - nonscalable, but faster (could take 1/3 scalable time) */ 576 ierr = PetscInfo(C,"Using non-scalable dense axpy\n");CHKERRQ(ierr); 577 /*-----------------------------------------------------------------------------------------------------*/ 578 Mat AP_loc,C_loc,Co; 579 580 // R = Pd^T,Ro = Po^T 581 ierr = MatZeroEntries(C);CHKERRQ(ierr); 582 ierr = PetscTime(&t0);CHKERRQ(ierr); 583 ierr = MatTranspose_SeqAIJ(p->A,MAT_REUSE_MATRIX,&ptap->Rd);CHKERRQ(ierr); 584 ierr = MatTranspose_SeqAIJ(p->B,MAT_REUSE_MATRIX,&ptap->Ro);CHKERRQ(ierr); 585 ierr = PetscTime(&t1);CHKERRQ(ierr); 586 eR = t1 - t0; 587 588 // AP = A*P 589 ierr = MatMatMult(A,P,MAT_REUSE_MATRIX,2.0,&ptap->AP);CHKERRQ(ierr); 590 591 // Get AP_loc 592 ierr = MatMPIAIJGetLocalMat(ptap->AP,MAT_INITIAL_MATRIX,&AP_loc);CHKERRQ(ierr); 593 ierr = PetscTime(&t2);CHKERRQ(ierr); 594 eAP = t2 - t1; 595 596 // C_loc = R*AP_loc, Co = Ro*AP_loc 597 ierr = MatMatMult(ptap->Rd,AP_loc,MAT_INITIAL_MATRIX,2.0,&C_loc);CHKERRQ(ierr); 598 ierr = MatMatMult(ptap->Ro,AP_loc,MAT_INITIAL_MATRIX,2.0,&Co);CHKERRQ(ierr); 599 //printf("[%d] Co %d, %d\n", rank,Co->rmap->N,Co->cmap->N); 600 ierr = PetscTime(&t3);CHKERRQ(ierr); 601 eCseq = t3 - t2; 602 603 // SetValues in C 604 PetscInt rstart,rend,cm,ncols,row; 605 const PetscInt *cols; 606 const PetscScalar *vals; 607 608 ierr = MatGetOwnershipRange(C,&rstart,&rend);CHKERRQ(ierr); 609 //if (cm != rend - rstart) SETERRQ(MPI_COMM_SELF,0,"cm != rend - rstart"); 610 611 // C_loc -> C 612 cm = C_loc->rmap->N; 613 for (i=0; i<cm; i++) { 614 ierr = MatGetRow(C_loc,i,&ncols,&cols,&vals);CHKERRQ(ierr); 615 row = rstart + i; 616 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 617 ierr = MatRestoreRow(C_loc,i,&ncols,&cols,&vals);CHKERRQ(ierr); 618 } 619 620 // Co -> C, off-diagonal part, send to others 621 //printf("[%d] p->B %d, %d\n",rank,p->B->rmap->N,p->B->cmap->N); 622 for (i=0; i<Co->rmap->N; i++) { 623 ierr = MatGetRow(Co,i,&ncols,&cols,&vals);CHKERRQ(ierr); 624 row = p->garray[i]; 625 //printf("[%d] row[%d] = %d\n",rank,i,row); 626 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 627 ierr = MatRestoreRow(Co,i,&ncols,&cols,&vals);CHKERRQ(ierr); 628 } 629 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 630 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 631 ierr = PetscTime(&t4);CHKERRQ(ierr); 632 eCmpi = t4 - t3; 633 634 ierr = MatDestroy(&AP_loc);CHKERRQ(ierr); 635 ierr = MatDestroy(&C_loc);CHKERRQ(ierr); 636 ierr = MatDestroy(&Co);CHKERRQ(ierr); 637 638 ierr = PetscFree(ba);CHKERRQ(ierr); 639 ierr = PetscFree(coa);CHKERRQ(ierr); 640 641 642 if (rank==1) { 643 ierr = PetscPrintf(MPI_COMM_SELF," R %g, AP %g, Cseq %g, Cmpi %g = %g\n", eR,eAP,eCseq,eCmpi,eR+eAP+eCseq+eCmpi);CHKERRQ(ierr); 644 } 645 646 PetscFunctionReturn(0); 647 #if 0 648 Mat R = ptap->R; 649 Mat_SeqAIJ *r = (Mat_SeqAIJ*)R->data; 650 PetscInt *ri=r->i,*rj=r->j,rnz,arow,l,prow,pcol,pN=P->cmap->N; 651 PetscScalar *ra=r->a,tmp,cdense[pN]; 652 653 ierr = PetscMemzero(cdense,pN*sizeof(PetscScalar));CHKERRQ(ierr); 654 for (i=0; i<cm; i++) { /* each row of C or R */ 655 rnz = ri[i+1] - ri[i]; 656 657 for (j=0; j<rnz; j++) { /* each nz of R */ 658 arow = rj[ri[i] + j]; 659 660 /* diagonal portion of A */ 661 anz = ad->i[arow+1] - ad->i[arow]; 662 for (k=0; k<anz; k++) { /* each nz of Ad */ 663 tmp = ra[ri[i] + j]*ad->a[ad->i[arow] + k]; 664 prow = ad->j[ad->i[arow] + k]; 665 pnz = pi_loc[prow+1] - pi_loc[prow]; 666 667 for (l=0; l<pnz; l++) { /* each nz of P_loc */ 668 pcol = pj_loc[pi_loc[prow] + l]; 669 cdense[pcol] += tmp*pa_loc[pi_loc[prow] + l]; 670 } 671 } 672 673 /* off-diagonal portion of A */ 674 anz = ao->i[arow+1] - ao->i[arow]; 675 for (k=0; k<anz; k++) { /* each nz of Ao */ 676 tmp = ra[ri[i] + j]*ao->a[ao->i[arow] + k]; 677 prow = ao->j[ao->i[arow] + k]; 678 pnz = pi_oth[prow+1] - pi_oth[prow]; 679 680 for (l=0; l<pnz; l++) { /* each nz of P_oth */ 681 pcol = pj_oth[pi_oth[prow] + l]; 682 cdense[pcol] += tmp*pa_oth[pi_oth[prow] + l]; 683 } 684 } 685 686 } //for (j=0; j<rnz; j++) 687 688 /* copy cdense[] into ca; zero cdense[] */ 689 cnz = bi[i+1] - bi[i]; 690 cj = bj + bi[i]; 691 ca = ba + bi[i]; 692 for (j=0; j<cnz; j++) { 693 ca[j] += cdense[cj[j]]; 694 cdense[cj[j]] = 0.0; 695 } 696 #if 0 697 if (rank == 0) { 698 printf("[%d] row %d: ",rank,i); 699 for (j=0; j<pN; j++) printf(" %g,",cdense[j]); 700 printf("\n"); 701 } 702 for (j=0; j<pN; j++) cdense[j]=0.0; // zero cdnese[] 703 #endif 704 } //for (i=0; i<cm; i++) { 705 #endif 706 ierr = PetscTime(&t_tran3);CHKERRQ(ierr); 707 708 //========================================== 709 710 ierr = PetscTime(&t1);CHKERRQ(ierr); 711 for (i=0; i<am; i++) { 712 #if defined(PTAP_PROFILE) 713 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 714 #endif 715 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 716 /*------------------------------------------------------------*/ 717 apJ = apj + api[i]; 718 719 /* diagonal portion of A */ 720 anz = adi[i+1] - adi[i]; 721 adj = ad->j + adi[i]; 722 ada = ad->a + adi[i]; 723 for (j=0; j<anz; j++) { 724 row = adj[j]; 725 pnz = pi_loc[row+1] - pi_loc[row]; 726 pj = pj_loc + pi_loc[row]; 727 pa = pa_loc + pi_loc[row]; 728 729 /* perform dense axpy */ 730 valtmp = ada[j]; 731 for (k=0; k<pnz; k++) { 732 apa[pj[k]] += valtmp*pa[k]; 733 } 734 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 735 } 736 737 /* off-diagonal portion of A */ 738 anz = aoi[i+1] - aoi[i]; 739 aoj = ao->j + aoi[i]; 740 aoa = ao->a + aoi[i]; 741 for (j=0; j<anz; j++) { 742 row = aoj[j]; 743 pnz = pi_oth[row+1] - pi_oth[row]; 744 pj = pj_oth + pi_oth[row]; 745 pa = pa_oth + pi_oth[row]; 746 747 /* perform dense axpy */ 748 valtmp = aoa[j]; 749 for (k=0; k<pnz; k++) { 750 apa[pj[k]] += valtmp*pa[k]; 751 } 752 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 753 } 754 #if defined(PTAP_PROFILE) 755 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 756 et2_AP += t2_1 - t2_0; 757 #endif 758 759 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 760 /*--------------------------------------------------------------*/ 761 apnz = api[i+1] - api[i]; 762 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 763 pnz = po->i[i+1] - po->i[i]; 764 poJ = po->j + po->i[i]; 765 pA = po->a + po->i[i]; 766 for (j=0; j<pnz; j++) { 767 row = poJ[j]; 768 cnz = coi[row+1] - coi[row]; 769 cj = coj + coi[row]; 770 ca = coa + coi[row]; 771 /* perform dense axpy */ 772 valtmp = pA[j]; 773 for (k=0; k<cnz; k++) { 774 ca[k] += valtmp*apa[cj[k]]; 775 } 776 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 777 } 778 #if 1 779 /* put the value into Cd (diagonal part) */ 780 pnz = pd->i[i+1] - pd->i[i]; 781 pdJ = pd->j + pd->i[i]; 782 pA = pd->a + pd->i[i]; 783 for (j=0; j<pnz; j++) { 784 row = pdJ[j]; 785 cnz = bi[row+1] - bi[row]; 786 cj = bj + bi[row]; 787 ca = ba + bi[row]; 788 /* perform dense axpy */ 789 valtmp = pA[j]; 790 for (k=0; k<cnz; k++) { 791 ca[k] += valtmp*apa[cj[k]]; 792 } 793 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 794 } 795 #endif 796 /* zero the current row of A*P */ 797 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 798 #if defined(PTAP_PROFILE) 799 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 800 ePtAP += t2_2 - t2_1; 801 #endif 802 } 803 804 if (rank == 100) { 805 for (row=0; row<cm; row++) { 806 printf("[%d] row %d: ",rank,row); 807 cnz = bi[row+1] - bi[row]; 808 for (j=0; j<cnz; j++) printf(" %g,",ba[bi[row]+j]); 809 printf("\n"); 810 } 811 } 812 813 } else { /* Do sparse axpy on apa (length of ap_rmax, stores A[i,:]*P) - scalable, but slower */ 814 ierr = PetscInfo(C,"Using scalable sparse axpy\n");CHKERRQ(ierr); 815 /*-----------------------------------------------------------------------------------------*/ 816 pA=pa_loc; 817 for (i=0; i<am; i++) { 818 #if defined(PTAP_PROFILE) 819 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 820 #endif 821 /* form i-th sparse row of A*P */ 822 apnz = api[i+1] - api[i]; 823 apJ = apj + api[i]; 824 /* diagonal portion of A */ 825 anz = adi[i+1] - adi[i]; 826 adj = ad->j + adi[i]; 827 ada = ad->a + adi[i]; 828 for (j=0; j<anz; j++) { 829 row = adj[j]; 830 pnz = pi_loc[row+1] - pi_loc[row]; 831 pj = pj_loc + pi_loc[row]; 832 pa = pa_loc + pi_loc[row]; 833 valtmp = ada[j]; 834 nextp = 0; 835 for (k=0; nextp<pnz; k++) { 836 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 837 apa[k] += valtmp*pa[nextp++]; 838 } 839 } 840 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 841 } 842 /* off-diagonal portion of A */ 843 anz = aoi[i+1] - aoi[i]; 844 aoj = ao->j + aoi[i]; 845 aoa = ao->a + aoi[i]; 846 for (j=0; j<anz; j++) { 847 row = aoj[j]; 848 pnz = pi_oth[row+1] - pi_oth[row]; 849 pj = pj_oth + pi_oth[row]; 850 pa = pa_oth + pi_oth[row]; 851 valtmp = aoa[j]; 852 nextp = 0; 853 for (k=0; nextp<pnz; k++) { 854 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 855 apa[k] += valtmp*pa[nextp++]; 856 } 857 } 858 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 859 } 860 #if defined(PTAP_PROFILE) 861 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 862 et2_AP += t2_1 - t2_0; 863 #endif 864 865 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 866 /*--------------------------------------------------------------*/ 867 pnz = pi_loc[i+1] - pi_loc[i]; 868 pJ = pj_loc + pi_loc[i]; 869 for (j=0; j<pnz; j++) { 870 nextap = 0; 871 row = pJ[j]; /* global index */ 872 if (row < pcstart || row >=pcend) { /* put the value into Co */ 873 row = *poJ; 874 cj = coj + coi[row]; 875 ca = coa + coi[row]; poJ++; 876 } else { /* put the value into Cd */ 877 row = *pdJ; 878 cj = bj + bi[row]; 879 ca = ba + bi[row]; pdJ++; 880 } 881 valtmp = pA[j]; 882 for (k=0; nextap<apnz; k++) { 883 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 884 } 885 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 886 } 887 pA += pnz; 888 /* zero the current row info for A*P */ 889 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 890 #if defined(PTAP_PROFILE) 891 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 892 ePtAP += t2_2 - t2_1; 893 #endif 894 } 895 } 896 #if defined(PTAP_PROFILE) 897 ierr = PetscTime(&t2);CHKERRQ(ierr); 898 #endif 899 900 /* 3) send and recv matrix values coa */ 901 /*------------------------------------*/ 902 buf_ri = merge->buf_ri; 903 buf_rj = merge->buf_rj; 904 len_s = merge->len_s; 905 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 906 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 907 908 ierr = PetscMalloc2(merge->nsend+1,&s_waits,size,&status);CHKERRQ(ierr); 909 for (proc=0,k=0; proc<size; proc++) { 910 if (!len_s[proc]) continue; 911 i = merge->owners_co[proc]; 912 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 913 k++; 914 } 915 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 916 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 917 918 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 919 ierr = PetscFree(r_waits);CHKERRQ(ierr); 920 ierr = PetscFree(coa);CHKERRQ(ierr); 921 #if defined(PTAP_PROFILE) 922 ierr = PetscTime(&t3);CHKERRQ(ierr); 923 #endif 924 925 /* 4) insert local Cseq and received values into Cmpi */ 926 /*------------------------------------------------------*/ 927 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 928 for (k=0; k<merge->nrecv; k++) { 929 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 930 nrows = *(buf_ri_k[k]); 931 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 932 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 933 } 934 935 for (i=0; i<cm; i++) { 936 row = owners[rank] + i; /* global row index of C_seq */ 937 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 938 ba_i = ba + bi[i]; 939 bnz = bi[i+1] - bi[i]; 940 /* add received vals into ba */ 941 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 942 /* i-th row */ 943 if (i == *nextrow[k]) { 944 cnz = *(nextci[k]+1) - *nextci[k]; 945 cj = buf_rj[k] + *(nextci[k]); 946 ca = abuf_r[k] + *(nextci[k]); 947 nextcj = 0; 948 for (j=0; nextcj<cnz; j++) { 949 if (bj_i[j] == cj[nextcj]) { /* bcol == ccol */ 950 ba_i[j] += ca[nextcj++]; 951 } 952 } 953 nextrow[k]++; nextci[k]++; 954 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 955 } 956 } 957 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 958 } 959 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 960 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 961 962 ierr = PetscFree(ba);CHKERRQ(ierr); 963 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 964 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 965 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 966 #if defined(PTAP_PROFILE) 967 ierr = PetscTime(&t4);CHKERRQ(ierr); 968 if (rank==1) { 969 ierr = PetscPrintf(MPI_COMM_SELF," R=Pd^T %g; AP %g\n", t_tran1-t_tran0,t_tran2-t_tran1);CHKERRQ(ierr); 970 ierr = PetscPrintf(MPI_COMM_SELF," [%d] PtAPNum %g/P + %g/PtAP( %g/A*P + %g/Pt*AP ) + %g/comm + %g/Cloc = %g\n\n",rank,eP,t2-t1,et2_AP,ePtAP,t3-t2,t4-t3,t4-t0);CHKERRQ(ierr); 971 } 972 #endif 973 PetscFunctionReturn(0); 974 } 975