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 12 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 13 #undef __FUNCT__ 14 #define __FUNCT__ "MatDestroy_MPIAIJ_PtAP" 15 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 16 { 17 PetscErrorCode ierr; 18 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 19 Mat_PtAPMPI *ptap=a->ptap; 20 21 PetscFunctionBegin; 22 if (ptap){ 23 ierr = PetscFree2(ptap->startsj,ptap->startsj_r);CHKERRQ(ierr); 24 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 25 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 26 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 27 ierr = PetscFree(ptap->api);CHKERRQ(ierr); 28 ierr = PetscFree(ptap->apj);CHKERRQ(ierr); 29 } 30 if (ptap->merge) { 31 Mat_Merge_SeqsToMPI *merge=ptap->merge; 32 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 33 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 34 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 35 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 36 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 37 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 38 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 39 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 40 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 41 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 42 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 43 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 44 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 45 ierr = merge->destroy(A);CHKERRQ(ierr); 46 ierr = PetscFree(ptap->merge);CHKERRQ(ierr); 47 } 48 ierr = PetscFree(ptap);CHKERRQ(ierr); 49 PetscFunctionReturn(0); 50 } 51 52 #undef __FUNCT__ 53 #define __FUNCT__ "MatDuplicate_MPIAIJ_MatPtAP" 54 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 55 { 56 PetscErrorCode ierr; 57 Mat_Merge_SeqsToMPI *merge; 58 PetscContainer container; 59 60 PetscFunctionBegin; 61 ierr = PetscObjectQuery((PetscObject)A,"MatMergeSeqsToMPI",(PetscObject *)&container);CHKERRQ(ierr); 62 if (container) { 63 ierr = PetscContainerGetPointer(container,(void **)&merge);CHKERRQ(ierr); 64 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 65 ierr = (*merge->duplicate)(A,op,M);CHKERRQ(ierr); 66 (*M)->ops->destroy = merge->destroy; /* =MatDestroy_MPIAIJ, *M doesn't duplicate A's container! */ 67 (*M)->ops->duplicate = merge->duplicate; /* =MatDuplicate_ MPIAIJ */ 68 PetscFunctionReturn(0); 69 } 70 71 #undef __FUNCT__ 72 #define __FUNCT__ "MatPtAPSymbolic_MPIAIJ" 73 PetscErrorCode MatPtAPSymbolic_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 74 { 75 PetscErrorCode ierr; 76 77 PetscFunctionBegin; 78 if (!P->ops->ptapsymbolic_mpiaij) SETERRQ2(((PetscObject)A)->comm,PETSC_ERR_SUP,"Not implemented for A=%s and P=%s",((PetscObject)A)->type_name,((PetscObject)P)->type_name); 79 ierr = (*P->ops->ptapsymbolic_mpiaij)(A,P,fill,C);CHKERRQ(ierr); 80 PetscFunctionReturn(0); 81 } 82 83 #undef __FUNCT__ 84 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ" 85 PetscErrorCode MatPtAPNumeric_MPIAIJ(Mat A,Mat P,Mat C) 86 { 87 PetscErrorCode ierr; 88 89 PetscFunctionBegin; 90 if (!P->ops->ptapnumeric_mpiaij) SETERRQ2(((PetscObject)A)->comm,PETSC_ERR_SUP,"Not implemented for A=%s and P=%s",((PetscObject)A)->type_name,((PetscObject)P)->type_name); 91 ierr = (*P->ops->ptapnumeric_mpiaij)(A,P,C);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 B_mpi; 101 Mat_PtAPMPI *ptap; 102 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_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,*adj=ad->j,*aoi=ao->i,*aoj=ao->j,nnz; 108 PetscInt nlnk,*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=((PetscObject)A)->comm; 112 PetscMPIInt size,rank,tag,*len_si,*len_s,*len_ri; 113 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 114 PetscInt len,proc,*dnz,*onz,*owners; 115 PetscInt nzi,*bi,*bj; 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; 121 PetscReal afill=1.0,afill_tmp; 122 123 PetscFunctionBegin; 124 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 125 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 126 127 /* create struct Mat_PtAPMPI and attached it to C later */ 128 ierr = PetscNew(Mat_PtAPMPI,&ptap);CHKERRQ(ierr); 129 ptap->api=PETSC_NULL; ptap->apj=PETSC_NULL; 130 ptap->abnz_max = 0; 131 ptap->reuse = MAT_INITIAL_MATRIX; 132 133 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 134 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 135 /* get P_loc by taking all local rows of P */ 136 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 137 138 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 139 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 140 pi_loc = p_loc->i; pj_loc = p_loc->j; 141 pi_oth = p_oth->i; pj_oth = p_oth->j; 142 143 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 144 /*-------------------------------------------------------------------*/ 145 ierr = PetscMalloc((am+2)*sizeof(PetscInt),&api);CHKERRQ(ierr); 146 ptap->api = api; 147 api[0] = 0; 148 149 /* create and initialize a linked list */ 150 nlnk = pN+1; 151 ierr = PetscLLCreate(pN,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 152 153 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 154 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 155 current_space = free_space; 156 157 for (i=0;i<am;i++) { 158 apnz = 0; 159 /* diagonal portion of A */ 160 nzi = adi[i+1] - adi[i]; 161 for (j=0; j<nzi; j++){ 162 row = *adj++; 163 pnz = pi_loc[row+1] - pi_loc[row]; 164 Jptr = pj_loc + pi_loc[row]; 165 /* add non-zero cols of P into the sorted linked list lnk */ 166 ierr = PetscLLAdd(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 167 apnz += nlnk; 168 } 169 /* off-diagonal portion of A */ 170 nzi = aoi[i+1] - aoi[i]; 171 for (j=0; j<nzi; j++){ 172 row = *aoj++; 173 pnz = pi_oth[row+1] - pi_oth[row]; 174 Jptr = pj_oth + pi_oth[row]; 175 ierr = PetscLLAdd(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 176 apnz += nlnk; 177 } 178 179 api[i+1] = api[i] + apnz; 180 if (ptap->abnz_max < apnz) ptap->abnz_max = apnz; 181 182 /* if free space is not available, double the total space in the list */ 183 if (current_space->local_remaining<apnz) { 184 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 185 nspacedouble++; 186 } 187 188 /* Copy data into free space, then initialize lnk */ 189 ierr = PetscLLClean(pN,pN,apnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 190 current_space->array += apnz; 191 current_space->local_used += apnz; 192 current_space->local_remaining -= apnz; 193 } 194 /* Allocate space for apj, initialize apj, and */ 195 /* destroy list of free space and other temporary array(s) */ 196 ierr = PetscMalloc((api[am]+1)*sizeof(PetscInt),&ptap->apj);CHKERRQ(ierr); 197 apj = ptap->apj; 198 ierr = PetscFreeSpaceContiguous(&free_space,ptap->apj);CHKERRQ(ierr); 199 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]); 200 if (afill_tmp > afill) afill = afill_tmp; 201 202 /* determine symbolic Co=(p->B)^T*AP - send to others */ 203 /*----------------------------------------------------*/ 204 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 205 206 /* then, compute symbolic Co = (p->B)^T*AP */ 207 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 208 >= (num of nonzero rows of C_seq) - pn */ 209 ierr = PetscMalloc((pon+1)*sizeof(PetscInt),&coi);CHKERRQ(ierr); 210 coi[0] = 0; 211 212 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 213 nnz = fill*(poti[pon] + api[am]); 214 ierr = PetscFreeSpaceGet(nnz,&free_space); 215 current_space = free_space; 216 217 for (i=0; i<pon; i++) { 218 nnz = 0; 219 pnz = poti[i+1] - poti[i]; 220 j = pnz; 221 ptJ = potj + poti[i+1]; 222 while (j){/* assume cols are almost in increasing order, starting from its end saves computation */ 223 j--; ptJ--; 224 row = *ptJ; /* row of AP == col of Pot */ 225 apnz = api[row+1] - api[row]; 226 Jptr = apj + api[row]; 227 /* add non-zero cols of AP into the sorted linked list lnk */ 228 ierr = PetscLLAdd(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 229 nnz += nlnk; 230 } 231 232 /* If free space is not available, double the total space in the list */ 233 if (current_space->local_remaining<nnz) { 234 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 235 nspacedouble++; 236 } 237 238 /* Copy data into free space, and zero out denserows */ 239 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 240 current_space->array += nnz; 241 current_space->local_used += nnz; 242 current_space->local_remaining -= nnz; 243 coi[i+1] = coi[i] + nnz; 244 } 245 ierr = PetscMalloc((coi[pon]+1)*sizeof(PetscInt),&coj);CHKERRQ(ierr); 246 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 247 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]); 248 if (afill_tmp > afill) afill = afill_tmp; 249 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 250 251 /* send j-array (coj) of Co to other processors */ 252 /*----------------------------------------------*/ 253 /* determine row ownership */ 254 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 255 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 256 merge->rowmap->n = pn; 257 merge->rowmap->bs = 1; 258 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 259 owners = merge->rowmap->range; 260 261 /* determine the number of messages to send, their lengths */ 262 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 263 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 264 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 265 len_s = merge->len_s; 266 merge->nsend = 0; 267 268 ierr = PetscMalloc((size+2)*sizeof(PetscInt),&owners_co);CHKERRQ(ierr); 269 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 270 271 proc = 0; 272 for (i=0; i<pon; i++){ 273 while (prmap[i] >= owners[proc+1]) proc++; 274 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 275 len_s[proc] += coi[i+1] - coi[i]; 276 } 277 278 len = 0; /* max length of buf_si[] */ 279 owners_co[0] = 0; 280 for (proc=0; proc<size; proc++){ 281 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 282 if (len_si[proc]){ 283 merge->nsend++; 284 len_si[proc] = 2*(len_si[proc] + 1); 285 len += len_si[proc]; 286 } 287 } 288 289 /* determine the number and length of messages to receive for coi and coj */ 290 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 291 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 292 293 /* post the Irecv and Isend of coj */ 294 ierr = PetscCommGetNewTag(comm,&tag);CHKERRQ(ierr); 295 ierr = PetscPostIrecvInt(comm,tag,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 296 297 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&swaits);CHKERRQ(ierr); 298 299 for (proc=0, k=0; proc<size; proc++){ 300 if (!len_s[proc]) continue; 301 i = owners_co[proc]; 302 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tag,comm,swaits+k);CHKERRQ(ierr); 303 k++; 304 } 305 306 /* receives and sends of coj are complete */ 307 ierr = PetscMalloc(size*sizeof(MPI_Status),&sstatus);CHKERRQ(ierr); 308 i = merge->nrecv; 309 while (i--) { 310 ierr = MPI_Waitany(merge->nrecv,rwaits,&j,&rstatus);CHKERRQ(ierr); 311 } 312 ierr = PetscFree(rwaits);CHKERRQ(ierr); 313 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 314 315 /* send and recv coi */ 316 /*-------------------*/ 317 ierr = PetscPostIrecvInt(comm,tag,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 318 319 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 320 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 321 for (proc=0,k=0; proc<size; proc++){ 322 if (!len_s[proc]) continue; 323 /* form outgoing message for i-structure: 324 buf_si[0]: nrows to be sent 325 [1:nrows]: row index (global) 326 [nrows+1:2*nrows+1]: i-structure index 327 */ 328 /*-------------------------------------------*/ 329 nrows = len_si[proc]/2 - 1; 330 buf_si_i = buf_si + nrows+1; 331 buf_si[0] = nrows; 332 buf_si_i[0] = 0; 333 nrows = 0; 334 for (i=owners_co[proc]; i<owners_co[proc+1]; i++){ 335 nzi = coi[i+1] - coi[i]; 336 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 337 buf_si[nrows+1] =prmap[i] -owners[proc]; /* local row index */ 338 nrows++; 339 } 340 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tag,comm,swaits+k);CHKERRQ(ierr); 341 k++; 342 buf_si += len_si[proc]; 343 } 344 i = merge->nrecv; 345 while (i--) { 346 ierr = MPI_Waitany(merge->nrecv,rwaits,&j,&rstatus);CHKERRQ(ierr); 347 } 348 ierr = PetscFree(rwaits);CHKERRQ(ierr); 349 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 350 /* 351 ierr = PetscInfo2(A,"nsend: %d, nrecv: %d\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 352 for (i=0; i<merge->nrecv; i++){ 353 ierr = PetscInfo3(A,"recv len_ri=%d, len_rj=%d from [%d]\n",len_ri[i],merge->len_r[i],merge->id_r[i]);CHKERRQ(ierr); 354 } 355 */ 356 ierr = PetscFree(len_si);CHKERRQ(ierr); 357 ierr = PetscFree(len_ri);CHKERRQ(ierr); 358 ierr = PetscFree(swaits);CHKERRQ(ierr); 359 ierr = PetscFree(sstatus);CHKERRQ(ierr); 360 ierr = PetscFree(buf_s);CHKERRQ(ierr); 361 362 /* compute the local portion of C (mpi mat) */ 363 /*------------------------------------------*/ 364 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 365 366 /* allocate bi array and free space for accumulating nonzero column info */ 367 ierr = PetscMalloc((pn+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 368 bi[0] = 0; 369 370 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 371 nnz = fill*(pi_loc[pm] + api[am]); 372 ierr = PetscFreeSpaceGet(nnz,&free_space); 373 current_space = free_space; 374 375 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 376 for (k=0; k<merge->nrecv; k++){ 377 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 378 nrows = *buf_ri_k[k]; 379 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 380 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 381 } 382 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 383 for (i=0; i<pn; i++) { 384 /* add pdt[i,:]*AP into lnk */ 385 nnz = 0; 386 pnz = pdti[i+1] - pdti[i]; 387 j = pnz; 388 ptJ = pdtj + pdti[i+1]; 389 while (j){/* assume cols are almost in increasing order, starting from its end saves computation */ 390 j--; ptJ--; 391 row = *ptJ; /* row of AP == col of Pt */ 392 apnz = api[row+1] - api[row]; 393 Jptr = apj + api[row]; 394 /* add non-zero cols of AP into the sorted linked list lnk */ 395 ierr = PetscLLAdd(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 396 nnz += nlnk; 397 } 398 /* add received col data into lnk */ 399 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 400 if (i == *nextrow[k]) { /* i-th row */ 401 nzi = *(nextci[k]+1) - *nextci[k]; 402 Jptr = buf_rj[k] + *nextci[k]; 403 ierr = PetscLLAdd(nzi,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 404 nnz += nlnk; 405 nextrow[k]++; nextci[k]++; 406 } 407 } 408 409 /* if free space is not available, make more free space */ 410 if (current_space->local_remaining<nnz) { 411 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 412 nspacedouble++; 413 } 414 /* copy data into free space, then initialize lnk */ 415 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 416 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 417 current_space->array += nnz; 418 current_space->local_used += nnz; 419 current_space->local_remaining -= nnz; 420 bi[i+1] = bi[i] + nnz; 421 } 422 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 423 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 424 425 ierr = PetscMalloc((bi[pn]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 426 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 427 afill_tmp = (PetscReal)bi[pn]/(pi_loc[pm] + api[am]); 428 if (afill_tmp > afill) afill = afill_tmp; 429 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 430 431 /* create symbolic parallel matrix B_mpi */ 432 /*---------------------------------------*/ 433 ierr = MatCreate(comm,&B_mpi);CHKERRQ(ierr); 434 ierr = MatSetSizes(B_mpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 435 ierr = MatSetType(B_mpi,MATMPIAIJ);CHKERRQ(ierr); 436 ierr = MatMPIAIJSetPreallocation(B_mpi,0,dnz,0,onz);CHKERRQ(ierr); 437 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 438 439 merge->bi = bi; 440 merge->bj = bj; 441 merge->coi = coi; 442 merge->coj = coj; 443 merge->buf_ri = buf_ri; 444 merge->buf_rj = buf_rj; 445 merge->owners_co = owners_co; 446 merge->destroy = B_mpi->ops->destroy; 447 merge->duplicate = B_mpi->ops->duplicate; 448 449 /* B_mpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 450 B_mpi->assembled = PETSC_FALSE; 451 B_mpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 452 B_mpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 453 ierr = MatSetBlockSize(B_mpi,1);CHKERRQ(ierr); 454 455 /* attach the supporting struct to B_mpi for reuse */ 456 c = (Mat_MPIAIJ*)B_mpi->data; 457 c->ptap = ptap; 458 ptap->merge = merge; 459 460 *C = B_mpi; 461 #if defined(PETSC_USE_INFO) 462 if (bi[pn] != 0) { 463 ierr = PetscInfo3(B_mpi,"Reallocs %D; Fill ratio: given %G needed %G.\n",nspacedouble,fill,afill);CHKERRQ(ierr); 464 ierr = PetscInfo1(B_mpi,"Use MatPtAP(A,P,MatReuse,%G,&C) for best performance.\n",afill);CHKERRQ(ierr); 465 } else { 466 ierr = PetscInfo(B_mpi,"Empty matrix product\n");CHKERRQ(ierr); 467 } 468 #endif 469 PetscFunctionReturn(0); 470 } 471 472 #undef __FUNCT__ 473 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 474 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 475 { 476 PetscErrorCode ierr; 477 Mat_Merge_SeqsToMPI *merge; 478 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 479 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 480 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 481 Mat_SeqAIJ *p_loc,*p_oth; 482 Mat_PtAPMPI *ptap; 483 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 484 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 485 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 486 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 487 PetscInt am=A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 488 MPI_Comm comm=((PetscObject)C)->comm; 489 PetscMPIInt size,rank,taga,*len_s; 490 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 491 PetscInt **buf_ri,**buf_rj; 492 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 493 MPI_Request *s_waits,*r_waits; 494 MPI_Status *status; 495 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 496 PetscInt *api,*apj,*coi,*coj; 497 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 498 PetscInt sparse_axpy; 499 PetscLogDouble t0,tf,etime=0.0,t00,tff,time_matupdate=0.0,time_malloc=0.0,time_Cseq0=0.0,time_Cseq1=0.0,time_setvals=0.0; 500 501 PetscFunctionBegin; 502 /* ierr = MPI_Barrier(comm);CHKERRQ(ierr); */ 503 ierr = PetscGetTime(&t00);CHKERRQ(ierr); 504 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 505 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 506 507 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 508 ptap = c->ptap; 509 merge = ptap->merge; 510 511 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 512 /*--------------------------------------------------*/ 513 if (ptap->reuse == MAT_INITIAL_MATRIX){ 514 ptap->reuse = MAT_REUSE_MATRIX; 515 } else { /* update numerical values of P_oth and P_loc */ 516 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 517 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 518 } 519 520 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 521 time_matupdate += tf-t0; 522 523 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 524 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 525 /*--------------------------------------------------------------*/ 526 /* get data from symbolic products */ 527 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 528 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 529 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 530 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 531 532 coi = merge->coi; coj = merge->coj; 533 ierr = PetscMalloc((coi[pon]+1)*sizeof(MatScalar),&coa);CHKERRQ(ierr); 534 ierr = PetscMemzero(coa,coi[pon]*sizeof(MatScalar));CHKERRQ(ierr); 535 536 bi = merge->bi; bj = merge->bj; 537 owners = merge->rowmap->range; 538 ierr = PetscMalloc((bi[cm]+1)*sizeof(MatScalar),&ba);CHKERRQ(ierr); 539 ierr = PetscMemzero(ba,bi[cm]*sizeof(MatScalar));CHKERRQ(ierr); 540 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 541 time_malloc += tf-t0; 542 543 api = ptap->api; apj = ptap->apj; 544 /* flag 'sparse_axpy' determines which implementations to be used: 545 0: do dense axpy in MatPtAPNumeric() - fastest, but requires storage of a dense array apa; (default) 546 1: do one sparse axpy - uses same memory as sparse_axpy=0 and might execute less flops 547 (apnz vs. cnz in the outerproduct), slower than case '0' when cnz is not too large than apnz; 548 2: do two sparse axpy in MatPtAPNumeric() - slowest, uses a sparse array apa */ 549 /* set default sparse_axpy */ 550 sparse_axpy = 0; 551 ierr = PetscOptionsGetInt(PETSC_NULL,"-matptap_sparseaxpy",&sparse_axpy,PETSC_NULL);CHKERRQ(ierr); 552 if (sparse_axpy == 0){ /* Do not perform sparse axpy */ 553 /*--------------------------------------------------*/ 554 /* malloc apa to store dense row A[i,:]*P */ 555 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 556 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 557 558 for (i=0; i<am; i++) { 559 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 560 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 561 /*------------------------------------------------------------*/ 562 apJ = apj + api[i]; 563 564 /* diagonal portion of A */ 565 anz = adi[i+1] - adi[i]; 566 adj = ad->j + adi[i]; 567 ada = ad->a + adi[i]; 568 for (j=0; j<anz; j++) { 569 row = adj[j]; 570 pnz = pi_loc[row+1] - pi_loc[row]; 571 pj = pj_loc + pi_loc[row]; 572 pa = pa_loc + pi_loc[row]; 573 574 /* perform dense axpy */ 575 valtmp = ada[j]; 576 for (k=0; k<pnz; k++){ 577 apa[pj[k]] += valtmp*pa[k]; 578 } 579 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 580 } 581 582 /* off-diagonal portion of A */ 583 anz = aoi[i+1] - aoi[i]; 584 aoj = ao->j + aoi[i]; 585 aoa = ao->a + aoi[i]; 586 for (j=0; j<anz; j++) { 587 row = aoj[j]; 588 pnz = pi_oth[row+1] - pi_oth[row]; 589 pj = pj_oth + pi_oth[row]; 590 pa = pa_oth + pi_oth[row]; 591 592 /* perform dense axpy */ 593 valtmp = aoa[j]; 594 for (k=0; k<pnz; k++){ 595 apa[pj[k]] += valtmp*pa[k]; 596 } 597 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 598 } 599 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 600 time_Cseq0 += tf - t0; 601 602 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 603 /*--------------------------------------------------------------*/ 604 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 605 apnz = api[i+1] - api[i]; 606 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 607 pnz = po->i[i+1] - po->i[i]; 608 poJ = po->j + po->i[i]; 609 pA = po->a + po->i[i]; 610 for (j=0; j<pnz; j++){ 611 row = poJ[j]; 612 cnz = coi[row+1] - coi[row]; 613 cj = coj + coi[row]; 614 ca = coa + coi[row]; 615 /* perform dense axpy */ 616 valtmp = pA[j]; 617 for (k=0; k<cnz; k++) { 618 ca[k] += valtmp*apa[cj[k]]; 619 } 620 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 621 } 622 623 /* put the value into Cd (diagonal part) */ 624 pnz = pd->i[i+1] - pd->i[i]; 625 pdJ = pd->j + pd->i[i]; 626 pA = pd->a + pd->i[i]; 627 for (j=0; j<pnz; j++){ 628 row = pdJ[j]; 629 cnz = bi[row+1] - bi[row]; 630 cj = bj + bi[row]; 631 ca = ba + bi[row]; 632 /* perform dense axpy */ 633 valtmp = pA[j]; 634 for (k=0; k<cnz; k++) { 635 ca[k] += valtmp*apa[cj[k]]; 636 } 637 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 638 } 639 640 /* zero the current row of A*P */ 641 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 642 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 643 time_Cseq1 += tf - t0; 644 } 645 } else if (sparse_axpy == 1){ /* Perform one sparse axpy */ 646 /*------------------------------------------------------*/ 647 /* malloc apa to store dense row A[i,:]*P */ 648 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 649 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 650 651 for (i=0; i<am; i++) { 652 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 653 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 654 /*------------------------------------------------------------*/ 655 apJ = apj + api[i]; 656 657 /* diagonal portion of A */ 658 anz = adi[i+1] - adi[i]; 659 adj = ad->j + adi[i]; 660 ada = ad->a + adi[i]; 661 for (j=0; j<anz; j++) { 662 row = adj[j]; 663 pnz = pi_loc[row+1] - pi_loc[row]; 664 pj = pj_loc + pi_loc[row]; 665 pa = pa_loc + pi_loc[row]; 666 667 /* perform dense axpy */ 668 valtmp = ada[j]; 669 for (k=0; k<pnz; k++){ 670 apa[pj[k]] += valtmp*pa[k]; 671 } 672 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 673 } 674 675 /* off-diagonal portion of A */ 676 anz = aoi[i+1] - aoi[i]; 677 aoj = ao->j + aoi[i]; 678 aoa = ao->a + aoi[i]; 679 for (j=0; j<anz; j++) { 680 row = aoj[j]; 681 pnz = pi_oth[row+1] - pi_oth[row]; 682 pj = pj_oth + pi_oth[row]; 683 pa = pa_oth + pi_oth[row]; 684 685 /* perform dense axpy */ 686 valtmp = aoa[j]; 687 for (k=0; k<pnz; k++){ 688 apa[pj[k]] += valtmp*pa[k]; 689 } 690 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 691 } 692 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 693 time_Cseq0 += tf - t0; 694 695 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 696 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 697 /*--------------------------------------------------------------*/ 698 apnz = api[i+1] - api[i]; 699 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 700 pnz = po->i[i+1] - po->i[i]; 701 poJ = po->j + po->i[i]; 702 pA = po->a + po->i[i]; 703 for (j=0; j<pnz; j++){ 704 row = poJ[j]; 705 cj = coj + coi[row]; 706 ca = coa + coi[row]; 707 valtmp = pA[j]; 708 /* perform sparse axpy */ 709 nextap = 0; 710 for (k=0; nextap<apnz; k++) { 711 if (cj[k]==apJ[nextap]) { /* global column index */ 712 ca[k] += valtmp*apa[cj[k]]; nextap++; 713 } 714 } 715 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 716 } 717 718 /* put the value into Cd (diagonal part) */ 719 pnz = pd->i[i+1] - pd->i[i]; 720 pdJ = pd->j + pd->i[i]; 721 pA = pd->a + pd->i[i]; 722 for (j=0; j<pnz; j++){ 723 row = pdJ[j]; 724 cj = bj + bi[row]; 725 ca = ba + bi[row]; 726 valtmp = pA[j]; 727 /* perform sparse axpy */ 728 nextap = 0; 729 for (k=0; nextap<apnz; k++) { 730 if (cj[k]==apJ[nextap]) { /* global column index */ 731 ca[k] += valtmp*apa[cj[k]]; 732 nextap++; 733 } 734 } 735 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 736 } 737 738 /* zero the current row of A*P */ 739 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 740 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 741 time_Cseq1 += tf - t0; 742 } 743 } else if (sparse_axpy == 2){/* Perform two sparse axpy */ 744 /*----------------------------------------------------*/ 745 /* malloc apa to store sparse row A[i,:]*P */ 746 ierr = PetscMalloc((ptap->abnz_max+1)*sizeof(MatScalar),&apa);CHKERRQ(ierr); 747 ierr = PetscMemzero(apa,ptap->abnz_max*sizeof(MatScalar));CHKERRQ(ierr); 748 749 pA=pa_loc; 750 for (i=0; i<am; i++) { 751 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 752 /* form i-th sparse row of A*P */ 753 apnz = api[i+1] - api[i]; 754 apJ = apj + api[i]; 755 /* diagonal portion of A */ 756 anz = adi[i+1] - adi[i]; 757 adj = ad->j + adi[i]; 758 ada = ad->a + adi[i]; 759 for (j=0; j<anz; j++) { 760 row = adj[j]; 761 pnz = pi_loc[row+1] - pi_loc[row]; 762 pj = pj_loc + pi_loc[row]; 763 pa = pa_loc + pi_loc[row]; 764 valtmp = ada[j]; 765 nextp = 0; 766 for (k=0; nextp<pnz; k++) { 767 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 768 apa[k] += valtmp*pa[nextp++]; 769 } 770 } 771 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 772 } 773 /* off-diagonal portion of A */ 774 anz = aoi[i+1] - aoi[i]; 775 aoj = ao->j + aoi[i]; 776 aoa = ao->a + aoi[i]; 777 for (j=0; j<anz; j++) { 778 row = aoj[j]; 779 pnz = pi_oth[row+1] - pi_oth[row]; 780 pj = pj_oth + pi_oth[row]; 781 pa = pa_oth + pi_oth[row]; 782 valtmp = aoa[j]; 783 nextp = 0; 784 for (k=0; nextp<pnz; k++) { 785 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 786 apa[k] += valtmp*pa[nextp++]; 787 } 788 } 789 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 790 } 791 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 792 time_Cseq0 += tf - t0; 793 794 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 795 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 796 /*--------------------------------------------------------------*/ 797 pnz = pi_loc[i+1] - pi_loc[i]; 798 pJ = pj_loc + pi_loc[i]; 799 for (j=0; j<pnz; j++) { 800 nextap = 0; 801 row = pJ[j]; /* global index */ 802 if (row < pcstart || row >=pcend) { /* put the value into Co */ 803 row = *poJ; 804 cj = coj + coi[row]; 805 ca = coa + coi[row]; poJ++; 806 } else { /* put the value into Cd */ 807 row = *pdJ; 808 cj = bj + bi[row]; 809 ca = ba + bi[row]; pdJ++; 810 } 811 valtmp = pA[j]; 812 for (k=0; nextap<apnz; k++) { 813 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 814 } 815 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 816 } 817 pA += pnz; 818 /* zero the current row info for A*P */ 819 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 820 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 821 time_Cseq1 += tf - t0; 822 } 823 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"sparse_axpy only takes values 0, 1 and 2"); 824 ierr = PetscFree(apa);CHKERRQ(ierr); 825 826 /* 3) send and recv matrix values coa */ 827 /*------------------------------------*/ 828 buf_ri = merge->buf_ri; 829 buf_rj = merge->buf_rj; 830 len_s = merge->len_s; 831 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 832 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 833 834 ierr = PetscMalloc2(merge->nsend+1,MPI_Request,&s_waits,size,MPI_Status,&status);CHKERRQ(ierr); 835 for (proc=0,k=0; proc<size; proc++){ 836 if (!len_s[proc]) continue; 837 i = merge->owners_co[proc]; 838 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 839 k++; 840 } 841 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 842 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 843 844 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 845 ierr = PetscFree(r_waits);CHKERRQ(ierr); 846 ierr = PetscFree(coa);CHKERRQ(ierr); 847 848 /* 4) insert local Cseq and received values into Cmpi */ 849 /*------------------------------------------------------*/ 850 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 851 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 852 853 for (k=0; k<merge->nrecv; k++){ 854 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 855 nrows = *(buf_ri_k[k]); 856 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 857 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 858 } 859 860 for (i=0; i<cm; i++) { 861 row = owners[rank] + i; /* global row index of C_seq */ 862 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 863 ba_i = ba + bi[i]; 864 bnz = bi[i+1] - bi[i]; 865 /* add received vals into ba */ 866 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 867 /* i-th row */ 868 if (i == *nextrow[k]) { 869 cnz = *(nextci[k]+1) - *nextci[k]; 870 cj = buf_rj[k] + *(nextci[k]); 871 ca = abuf_r[k] + *(nextci[k]); 872 nextcj = 0; 873 for (j=0; nextcj<cnz; j++){ 874 if (bj_i[j] == cj[nextcj]){ /* bcol == ccol */ 875 ba_i[j] += ca[nextcj++]; 876 } 877 } 878 nextrow[k]++; nextci[k]++; 879 } 880 } 881 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 882 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 883 } 884 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 885 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 886 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 887 time_setvals += tf-t0; 888 889 ierr = PetscFree(ba);CHKERRQ(ierr); 890 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 891 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 892 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 893 894 ierr = PetscGetTime(&tff);CHKERRQ(ierr); 895 etime += tff - t00; 896 /* 897 PetscInt prid=0; 898 if (rank == prid){ 899 ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] PtAPNum time %g = matupdate %g + malloc %g + Cseq %g + %g + setvals %g\n",rank,etime,time_matupdate,time_malloc,time_Cseq0,time_Cseq1,time_setvals); 900 } 901 */ 902 PetscFunctionReturn(0); 903 } 904