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 Cmpi; 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,*aj,*aoi=ao->i,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,ap_rmax=0; 121 PetscReal afill=1.0,afill_tmp; 122 PetscLogDouble t0,tf,etime=0.0,t00,tff,time_matupdate=0.0,time_Cd=0.0,time_AP=0.0,time_Co=0.0; 123 PetscInt rstart = P->cmap->rstart,rmax; 124 PetscScalar *vals; 125 126 PetscFunctionBegin; 127 ierr = PetscGetTime(&t00);CHKERRQ(ierr); 128 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 129 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 130 131 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 132 /* create struct Mat_PtAPMPI and attached it to C later */ 133 ierr = PetscNew(Mat_PtAPMPI,&ptap);CHKERRQ(ierr); 134 ptap->reuse = MAT_INITIAL_MATRIX; 135 136 137 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 138 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 139 /* get P_loc by taking all local rows of P */ 140 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 141 142 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 143 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 144 pi_loc = p_loc->i; pj_loc = p_loc->j; 145 pi_oth = p_oth->i; pj_oth = p_oth->j; 146 147 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 148 time_matupdate += tf-t0; 149 150 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 151 /*-------------------------------------------------------------------*/ 152 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 153 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&api);CHKERRQ(ierr); 154 api[0] = 0; 155 156 /* create and initialize a linked list */ 157 nlnk = pN+1; 158 ierr = PetscLLCreate(pN,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 159 160 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 161 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 162 current_space = free_space; 163 164 for (i=0; i<am; i++) { 165 apnz = 0; 166 /* diagonal portion of A */ 167 nzi = adi[i+1] - adi[i]; 168 aj = ad->j + adi[i]; 169 for (j=0; j<nzi; j++){ 170 row = aj[j]; 171 pnz = pi_loc[row+1] - pi_loc[row]; 172 Jptr = pj_loc + pi_loc[row]; 173 /* add non-zero cols of P into the sorted linked list lnk */ 174 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 175 apnz += nlnk; 176 } 177 /* off-diagonal portion of A */ 178 nzi = aoi[i+1] - aoi[i]; 179 aj = ao->j + aoi[i]; 180 for (j=0; j<nzi; j++){ 181 row = aj[j]; 182 pnz = pi_oth[row+1] - pi_oth[row]; 183 Jptr = pj_oth + pi_oth[row]; 184 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 185 apnz += nlnk; 186 } 187 188 api[i+1] = api[i] + apnz; 189 if (ap_rmax < apnz) ap_rmax = apnz; 190 191 /* if free space is not available, double the total space in the list */ 192 if (current_space->local_remaining<apnz) { 193 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 194 nspacedouble++; 195 } 196 197 /* Copy data into free space, then initialize lnk */ 198 ierr = PetscLLClean(pN,pN,apnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 199 current_space->array += apnz; 200 current_space->local_used += apnz; 201 current_space->local_remaining -= apnz; 202 } 203 /* Allocate space for apj, initialize apj, and */ 204 /* destroy list of free space and other temporary array(s) */ 205 ierr = PetscMalloc((api[am]+1)*sizeof(PetscInt),&apj);CHKERRQ(ierr); 206 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 207 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]); 208 if (afill_tmp > afill) afill = afill_tmp; 209 210 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 211 time_AP += tf-t0; 212 213 /* determine symbolic Co=(p->B)^T*AP - send to others */ 214 /*----------------------------------------------------*/ 215 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 216 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 217 218 /* then, compute symbolic Co = (p->B)^T*AP */ 219 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 220 >= (num of nonzero rows of C_seq) - pn */ 221 ierr = PetscMalloc((pon+1)*sizeof(PetscInt),&coi);CHKERRQ(ierr); 222 coi[0] = 0; 223 224 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 225 nnz = fill*(poti[pon] + api[am]); 226 ierr = PetscFreeSpaceGet(nnz,&free_space); 227 current_space = free_space; 228 229 for (i=0; i<pon; i++) { 230 nnz = 0; 231 pnz = poti[i+1] - poti[i]; 232 ptJ = potj + poti[i]; 233 for (j=0; j<pnz; j++){ 234 row = ptJ[j]; /* row of AP == col of Pot */ 235 apnz = api[row+1] - api[row]; 236 Jptr = apj + api[row]; 237 /* add non-zero cols of AP into the sorted linked list lnk */ 238 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 239 nnz += nlnk; 240 } 241 242 /* If free space is not available, double the total space in the list */ 243 if (current_space->local_remaining<nnz) { 244 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 245 nspacedouble++; 246 } 247 248 /* Copy data into free space, and zero out denserows */ 249 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 250 current_space->array += nnz; 251 current_space->local_used += nnz; 252 current_space->local_remaining -= nnz; 253 coi[i+1] = coi[i] + nnz; 254 } 255 ierr = PetscMalloc((coi[pon]+1)*sizeof(PetscInt),&coj);CHKERRQ(ierr); 256 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 257 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]); 258 if (afill_tmp > afill) afill = afill_tmp; 259 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 260 261 /* send j-array (coj) of Co to other processors */ 262 /*----------------------------------------------*/ 263 /* determine row ownership */ 264 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 265 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 266 merge->rowmap->n = pn; 267 merge->rowmap->bs = 1; 268 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 269 owners = merge->rowmap->range; 270 271 /* determine the number of messages to send, their lengths */ 272 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 273 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 274 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 275 len_s = merge->len_s; 276 merge->nsend = 0; 277 278 ierr = PetscMalloc((size+2)*sizeof(PetscInt),&owners_co);CHKERRQ(ierr); 279 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 280 281 proc = 0; 282 for (i=0; i<pon; i++){ 283 while (prmap[i] >= owners[proc+1]) proc++; 284 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 285 len_s[proc] += coi[i+1] - coi[i]; 286 } 287 288 len = 0; /* max length of buf_si[] */ 289 owners_co[0] = 0; 290 for (proc=0; proc<size; proc++){ 291 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 292 if (len_si[proc]){ 293 merge->nsend++; 294 len_si[proc] = 2*(len_si[proc] + 1); 295 len += len_si[proc]; 296 } 297 } 298 299 /* determine the number and length of messages to receive for coi and coj */ 300 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 301 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 302 303 /* post the Irecv and Isend of coj */ 304 ierr = PetscCommGetNewTag(comm,&tag);CHKERRQ(ierr); 305 ierr = PetscPostIrecvInt(comm,tag,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 306 307 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&swaits);CHKERRQ(ierr); 308 309 for (proc=0, k=0; proc<size; proc++){ 310 if (!len_s[proc]) continue; 311 i = owners_co[proc]; 312 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tag,comm,swaits+k);CHKERRQ(ierr); 313 k++; 314 } 315 316 /* receives and sends of coj are complete */ 317 ierr = PetscMalloc(size*sizeof(MPI_Status),&sstatus);CHKERRQ(ierr); 318 for (i=0; i<merge->nrecv; i++){ 319 ierr = MPI_Waitany(merge->nrecv,rwaits,&j,&rstatus);CHKERRQ(ierr); 320 } 321 ierr = PetscFree(rwaits);CHKERRQ(ierr); 322 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 323 324 /* send and recv coi */ 325 /*-------------------*/ 326 ierr = PetscPostIrecvInt(comm,tag,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 327 328 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 329 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 330 for (proc=0,k=0; proc<size; proc++){ 331 if (!len_s[proc]) continue; 332 /* form outgoing message for i-structure: 333 buf_si[0]: nrows to be sent 334 [1:nrows]: row index (global) 335 [nrows+1:2*nrows+1]: i-structure index 336 */ 337 /*-------------------------------------------*/ 338 nrows = len_si[proc]/2 - 1; 339 buf_si_i = buf_si + nrows+1; 340 buf_si[0] = nrows; 341 buf_si_i[0] = 0; 342 nrows = 0; 343 for (i=owners_co[proc]; i<owners_co[proc+1]; i++){ 344 nzi = coi[i+1] - coi[i]; 345 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 346 buf_si[nrows+1] =prmap[i] -owners[proc]; /* local row index */ 347 nrows++; 348 } 349 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tag,comm,swaits+k);CHKERRQ(ierr); 350 k++; 351 buf_si += len_si[proc]; 352 } 353 i = merge->nrecv; 354 while (i--) { 355 ierr = MPI_Waitany(merge->nrecv,rwaits,&j,&rstatus);CHKERRQ(ierr); 356 } 357 ierr = PetscFree(rwaits);CHKERRQ(ierr); 358 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 359 /* 360 ierr = PetscInfo2(A,"nsend: %d, nrecv: %d\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 361 for (i=0; i<merge->nrecv; i++){ 362 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); 363 } 364 */ 365 ierr = PetscFree(len_si);CHKERRQ(ierr); 366 ierr = PetscFree(len_ri);CHKERRQ(ierr); 367 ierr = PetscFree(swaits);CHKERRQ(ierr); 368 ierr = PetscFree(sstatus);CHKERRQ(ierr); 369 ierr = PetscFree(buf_s);CHKERRQ(ierr); 370 371 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 372 time_Co += tf-t0; 373 374 /* compute the local portion of C (mpi mat) */ 375 /*------------------------------------------*/ 376 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 377 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 378 379 /* allocate bi array and free space for accumulating nonzero column info */ 380 ierr = PetscMalloc((pn+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 381 bi[0] = 0; 382 383 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 384 nnz = fill*(pi_loc[pm] + api[am]); 385 ierr = PetscFreeSpaceGet(nnz,&free_space); 386 current_space = free_space; 387 388 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 389 for (k=0; k<merge->nrecv; k++){ 390 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 391 nrows = *buf_ri_k[k]; 392 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 393 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 394 } 395 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 396 rmax = 0; 397 for (i=0; i<pn; i++) { 398 /* add pdt[i,:]*AP into lnk */ 399 nnz = 0; 400 pnz = pdti[i+1] - pdti[i]; 401 ptJ = pdtj + pdti[i]; 402 for (j=0; j<pnz; j++){ 403 row = ptJ[j]; /* row of AP == col of Pt */ 404 apnz = api[row+1] - api[row]; 405 Jptr = apj + api[row]; 406 /* add non-zero cols of AP into the sorted linked list lnk */ 407 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 408 nnz += nlnk; 409 } 410 /* add received col data into lnk */ 411 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 412 if (i == *nextrow[k]) { /* i-th row */ 413 nzi = *(nextci[k]+1) - *nextci[k]; 414 Jptr = buf_rj[k] + *nextci[k]; 415 ierr = PetscLLAddSorted(nzi,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 416 nnz += nlnk; 417 nextrow[k]++; nextci[k]++; 418 } 419 } 420 421 /* if free space is not available, make more free space */ 422 if (current_space->local_remaining<nnz) { 423 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 424 nspacedouble++; 425 } 426 /* copy data into free space, then initialize lnk */ 427 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 428 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 429 current_space->array += nnz; 430 current_space->local_used += nnz; 431 current_space->local_remaining -= nnz; 432 bi[i+1] = bi[i] + nnz; 433 if (nnz > rmax) rmax = nnz; 434 } 435 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 436 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 437 438 ierr = PetscMalloc((bi[pn]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 439 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 440 afill_tmp = (PetscReal)bi[pn]/(pi_loc[pm] + api[am]); 441 if (afill_tmp > afill) afill = afill_tmp; 442 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 443 444 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 445 time_Cd += tf-t0; 446 447 /* create symbolic parallel matrix Cmpi - why cannot be assembled in Numeric part - check runex55_SA ? */ 448 /*------------------------------------------------------------------------------------------------------*/ 449 ierr = PetscMalloc((rmax+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 450 ierr = PetscMemzero(vals,rmax*sizeof(PetscScalar));CHKERRQ(ierr); 451 /* 452 ierr = PetscSynchronizedPrintf(comm, "[%d] pn %D, pN %D, cstart %D, rmax %D\n", rank,pn,pN,rstart,rmax);CHKERRQ(ierr); 453 ierr = PetscSynchronizedFlush(comm);CHKERRQ(ierr); 454 */ 455 456 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 457 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 458 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 459 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 460 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 461 ierr = MatSetBlockSize(Cmpi,1);CHKERRQ(ierr); 462 for (i=0; i<pn; i++){ 463 row = i + rstart; 464 nnz = bi[i+1] - bi[i]; 465 Jptr = bj + bi[i]; 466 ierr = MatSetValues(Cmpi,1,&row,nnz,Jptr,vals,INSERT_VALUES);CHKERRQ(ierr); 467 } 468 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 469 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 470 ierr = PetscFree(vals);CHKERRQ(ierr); 471 472 merge->bi = bi; 473 merge->bj = bj; 474 merge->coi = coi; 475 merge->coj = coj; 476 merge->buf_ri = buf_ri; 477 merge->buf_rj = buf_rj; 478 merge->owners_co = owners_co; 479 merge->destroy = Cmpi->ops->destroy; 480 merge->duplicate = Cmpi->ops->duplicate; 481 482 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 483 /* Cmpi->assembled = PETSC_FALSE; */ 484 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 485 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 486 487 /* attach the supporting struct to Cmpi for reuse */ 488 c = (Mat_MPIAIJ*)Cmpi->data; 489 c->ptap = ptap; 490 ptap->api = api; 491 ptap->apj = apj; 492 ptap->merge = merge; 493 ptap->abnz_max = ap_rmax; 494 495 *C = Cmpi; 496 #if defined(PETSC_USE_INFO) 497 if (bi[pn] != 0) { 498 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %G needed %G.\n",nspacedouble,fill,afill);CHKERRQ(ierr); 499 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%G,&C) for best performance.\n",afill);CHKERRQ(ierr); 500 } else { 501 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 502 } 503 #endif 504 ierr = PetscGetTime(&tff);CHKERRQ(ierr); 505 etime += tff - t00; 506 /* 507 PetscInt prid=0; 508 if (rank == prid){ 509 ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] PtAPSym time %g = matsetup %g + AP %g + Co %g + Cd %g\n",rank,etime,time_matupdate,time_AP,time_Co,time_Cd); 510 } 511 */ 512 PetscFunctionReturn(0); 513 } 514 515 #undef __FUNCT__ 516 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 517 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 518 { 519 PetscErrorCode ierr; 520 Mat_Merge_SeqsToMPI *merge; 521 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 522 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 523 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 524 Mat_SeqAIJ *p_loc,*p_oth; 525 Mat_PtAPMPI *ptap; 526 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 527 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 528 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 529 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 530 PetscInt am=A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 531 MPI_Comm comm=((PetscObject)C)->comm; 532 PetscMPIInt size,rank,taga,*len_s; 533 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 534 PetscInt **buf_ri,**buf_rj; 535 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 536 MPI_Request *s_waits,*r_waits; 537 MPI_Status *status; 538 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 539 PetscInt *api,*apj,*coi,*coj; 540 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 541 PetscInt sparse_axpy; 542 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; 543 544 PetscFunctionBegin; 545 /* ierr = MPI_Barrier(comm);CHKERRQ(ierr); */ 546 ierr = PetscGetTime(&t00);CHKERRQ(ierr); 547 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 548 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 549 550 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 551 ptap = c->ptap; 552 merge = ptap->merge; 553 554 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 555 /*--------------------------------------------------*/ 556 if (ptap->reuse == MAT_INITIAL_MATRIX){ 557 ptap->reuse = MAT_REUSE_MATRIX; 558 } else { /* update numerical values of P_oth and P_loc */ 559 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 560 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 561 } 562 563 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 564 time_matupdate += tf-t0; 565 566 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 567 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 568 /*--------------------------------------------------------------*/ 569 /* get data from symbolic products */ 570 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 571 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 572 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 573 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 574 575 coi = merge->coi; coj = merge->coj; 576 ierr = PetscMalloc((coi[pon]+1)*sizeof(MatScalar),&coa);CHKERRQ(ierr); 577 ierr = PetscMemzero(coa,coi[pon]*sizeof(MatScalar));CHKERRQ(ierr); 578 579 bi = merge->bi; bj = merge->bj; 580 owners = merge->rowmap->range; 581 ierr = PetscMalloc((bi[cm]+1)*sizeof(MatScalar),&ba);CHKERRQ(ierr); 582 ierr = PetscMemzero(ba,bi[cm]*sizeof(MatScalar));CHKERRQ(ierr); 583 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 584 time_malloc += tf-t0; 585 586 api = ptap->api; apj = ptap->apj; 587 /* flag 'sparse_axpy' determines which implementations to be used: 588 0: do dense axpy in MatPtAPNumeric() - fastest, but requires storage of a dense array apa; (default) 589 1: do one sparse axpy - uses same memory as sparse_axpy=0 and might execute less flops 590 (apnz vs. cnz in the outerproduct), slower than case '0' when cnz is not too large than apnz; 591 2: do two sparse axpy in MatPtAPNumeric() - slowest, uses a sparse array apa */ 592 /* set default sparse_axpy */ 593 sparse_axpy = 0; 594 ierr = PetscOptionsGetInt(PETSC_NULL,"-matptap_sparseaxpy",&sparse_axpy,PETSC_NULL);CHKERRQ(ierr); 595 if (sparse_axpy == 0){ /* Do not perform sparse axpy */ 596 /*--------------------------------------------------*/ 597 /* malloc apa to store dense row A[i,:]*P */ 598 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 599 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 600 601 for (i=0; i<am; i++) { 602 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 603 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 604 /*------------------------------------------------------------*/ 605 apJ = apj + api[i]; 606 607 /* diagonal portion of A */ 608 anz = adi[i+1] - adi[i]; 609 adj = ad->j + adi[i]; 610 ada = ad->a + adi[i]; 611 for (j=0; j<anz; j++) { 612 row = adj[j]; 613 pnz = pi_loc[row+1] - pi_loc[row]; 614 pj = pj_loc + pi_loc[row]; 615 pa = pa_loc + pi_loc[row]; 616 617 /* perform dense axpy */ 618 valtmp = ada[j]; 619 for (k=0; k<pnz; k++){ 620 apa[pj[k]] += valtmp*pa[k]; 621 } 622 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 623 } 624 625 /* off-diagonal portion of A */ 626 anz = aoi[i+1] - aoi[i]; 627 aoj = ao->j + aoi[i]; 628 aoa = ao->a + aoi[i]; 629 for (j=0; j<anz; j++) { 630 row = aoj[j]; 631 pnz = pi_oth[row+1] - pi_oth[row]; 632 pj = pj_oth + pi_oth[row]; 633 pa = pa_oth + pi_oth[row]; 634 635 /* perform dense axpy */ 636 valtmp = aoa[j]; 637 for (k=0; k<pnz; k++){ 638 apa[pj[k]] += valtmp*pa[k]; 639 } 640 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 641 } 642 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 643 time_Cseq0 += tf - t0; 644 645 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 646 /*--------------------------------------------------------------*/ 647 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 648 apnz = api[i+1] - api[i]; 649 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 650 pnz = po->i[i+1] - po->i[i]; 651 poJ = po->j + po->i[i]; 652 pA = po->a + po->i[i]; 653 for (j=0; j<pnz; j++){ 654 row = poJ[j]; 655 cnz = coi[row+1] - coi[row]; 656 cj = coj + coi[row]; 657 ca = coa + coi[row]; 658 /* perform dense axpy */ 659 valtmp = pA[j]; 660 for (k=0; k<cnz; k++) { 661 ca[k] += valtmp*apa[cj[k]]; 662 } 663 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 664 } 665 666 /* put the value into Cd (diagonal part) */ 667 pnz = pd->i[i+1] - pd->i[i]; 668 pdJ = pd->j + pd->i[i]; 669 pA = pd->a + pd->i[i]; 670 for (j=0; j<pnz; j++){ 671 row = pdJ[j]; 672 cnz = bi[row+1] - bi[row]; 673 cj = bj + bi[row]; 674 ca = ba + bi[row]; 675 /* perform dense axpy */ 676 valtmp = pA[j]; 677 for (k=0; k<cnz; k++) { 678 ca[k] += valtmp*apa[cj[k]]; 679 } 680 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 681 } 682 683 /* zero the current row of A*P */ 684 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 685 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 686 time_Cseq1 += tf - t0; 687 } 688 } else if (sparse_axpy == 1){ /* Perform one sparse axpy */ 689 /*------------------------------------------------------*/ 690 /* malloc apa to store dense row A[i,:]*P */ 691 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 692 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 693 694 for (i=0; i<am; i++) { 695 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 696 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 697 /*------------------------------------------------------------*/ 698 apJ = apj + api[i]; 699 700 /* diagonal portion of A */ 701 anz = adi[i+1] - adi[i]; 702 adj = ad->j + adi[i]; 703 ada = ad->a + adi[i]; 704 for (j=0; j<anz; j++) { 705 row = adj[j]; 706 pnz = pi_loc[row+1] - pi_loc[row]; 707 pj = pj_loc + pi_loc[row]; 708 pa = pa_loc + pi_loc[row]; 709 710 /* perform dense axpy */ 711 valtmp = ada[j]; 712 for (k=0; k<pnz; k++){ 713 apa[pj[k]] += valtmp*pa[k]; 714 } 715 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 716 } 717 718 /* off-diagonal portion of A */ 719 anz = aoi[i+1] - aoi[i]; 720 aoj = ao->j + aoi[i]; 721 aoa = ao->a + aoi[i]; 722 for (j=0; j<anz; j++) { 723 row = aoj[j]; 724 pnz = pi_oth[row+1] - pi_oth[row]; 725 pj = pj_oth + pi_oth[row]; 726 pa = pa_oth + pi_oth[row]; 727 728 /* perform dense axpy */ 729 valtmp = aoa[j]; 730 for (k=0; k<pnz; k++){ 731 apa[pj[k]] += valtmp*pa[k]; 732 } 733 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 734 } 735 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 736 time_Cseq0 += tf - t0; 737 738 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 739 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 740 /*--------------------------------------------------------------*/ 741 apnz = api[i+1] - api[i]; 742 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 743 pnz = po->i[i+1] - po->i[i]; 744 poJ = po->j + po->i[i]; 745 pA = po->a + po->i[i]; 746 for (j=0; j<pnz; j++){ 747 row = poJ[j]; 748 cj = coj + coi[row]; 749 ca = coa + coi[row]; 750 valtmp = pA[j]; 751 /* perform sparse axpy */ 752 nextap = 0; 753 for (k=0; nextap<apnz; k++) { 754 if (cj[k]==apJ[nextap]) { /* global column index */ 755 ca[k] += valtmp*apa[cj[k]]; nextap++; 756 } 757 } 758 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 759 } 760 761 /* put the value into Cd (diagonal part) */ 762 pnz = pd->i[i+1] - pd->i[i]; 763 pdJ = pd->j + pd->i[i]; 764 pA = pd->a + pd->i[i]; 765 for (j=0; j<pnz; j++){ 766 row = pdJ[j]; 767 cj = bj + bi[row]; 768 ca = ba + bi[row]; 769 valtmp = pA[j]; 770 /* perform sparse axpy */ 771 nextap = 0; 772 for (k=0; nextap<apnz; k++) { 773 if (cj[k]==apJ[nextap]) { /* global column index */ 774 ca[k] += valtmp*apa[cj[k]]; 775 nextap++; 776 } 777 } 778 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 779 } 780 781 /* zero the current row of A*P */ 782 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 783 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 784 time_Cseq1 += tf - t0; 785 } 786 } else if (sparse_axpy == 2){/* Perform two sparse axpy */ 787 /*----------------------------------------------------*/ 788 /* malloc apa to store sparse row A[i,:]*P */ 789 ierr = PetscMalloc((ptap->abnz_max+1)*sizeof(MatScalar),&apa);CHKERRQ(ierr); 790 ierr = PetscMemzero(apa,ptap->abnz_max*sizeof(MatScalar));CHKERRQ(ierr); 791 792 pA=pa_loc; 793 for (i=0; i<am; i++) { 794 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 795 /* form i-th sparse row of A*P */ 796 apnz = api[i+1] - api[i]; 797 apJ = apj + api[i]; 798 /* diagonal portion of A */ 799 anz = adi[i+1] - adi[i]; 800 adj = ad->j + adi[i]; 801 ada = ad->a + adi[i]; 802 for (j=0; j<anz; j++) { 803 row = adj[j]; 804 pnz = pi_loc[row+1] - pi_loc[row]; 805 pj = pj_loc + pi_loc[row]; 806 pa = pa_loc + pi_loc[row]; 807 valtmp = ada[j]; 808 nextp = 0; 809 for (k=0; nextp<pnz; k++) { 810 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 811 apa[k] += valtmp*pa[nextp++]; 812 } 813 } 814 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 815 } 816 /* off-diagonal portion of A */ 817 anz = aoi[i+1] - aoi[i]; 818 aoj = ao->j + aoi[i]; 819 aoa = ao->a + aoi[i]; 820 for (j=0; j<anz; j++) { 821 row = aoj[j]; 822 pnz = pi_oth[row+1] - pi_oth[row]; 823 pj = pj_oth + pi_oth[row]; 824 pa = pa_oth + pi_oth[row]; 825 valtmp = aoa[j]; 826 nextp = 0; 827 for (k=0; nextp<pnz; k++) { 828 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 829 apa[k] += valtmp*pa[nextp++]; 830 } 831 } 832 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 833 } 834 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 835 time_Cseq0 += tf - t0; 836 837 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 838 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 839 /*--------------------------------------------------------------*/ 840 pnz = pi_loc[i+1] - pi_loc[i]; 841 pJ = pj_loc + pi_loc[i]; 842 for (j=0; j<pnz; j++) { 843 nextap = 0; 844 row = pJ[j]; /* global index */ 845 if (row < pcstart || row >=pcend) { /* put the value into Co */ 846 row = *poJ; 847 cj = coj + coi[row]; 848 ca = coa + coi[row]; poJ++; 849 } else { /* put the value into Cd */ 850 row = *pdJ; 851 cj = bj + bi[row]; 852 ca = ba + bi[row]; pdJ++; 853 } 854 valtmp = pA[j]; 855 for (k=0; nextap<apnz; k++) { 856 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 857 } 858 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 859 } 860 pA += pnz; 861 /* zero the current row info for A*P */ 862 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 863 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 864 time_Cseq1 += tf - t0; 865 } 866 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"sparse_axpy only takes values 0, 1 and 2"); 867 ierr = PetscFree(apa);CHKERRQ(ierr); 868 869 /* 3) send and recv matrix values coa */ 870 /*------------------------------------*/ 871 buf_ri = merge->buf_ri; 872 buf_rj = merge->buf_rj; 873 len_s = merge->len_s; 874 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 875 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 876 877 ierr = PetscMalloc2(merge->nsend+1,MPI_Request,&s_waits,size,MPI_Status,&status);CHKERRQ(ierr); 878 for (proc=0,k=0; proc<size; proc++){ 879 if (!len_s[proc]) continue; 880 i = merge->owners_co[proc]; 881 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 882 k++; 883 } 884 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 885 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 886 887 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 888 ierr = PetscFree(r_waits);CHKERRQ(ierr); 889 ierr = PetscFree(coa);CHKERRQ(ierr); 890 891 /* 4) insert local Cseq and received values into Cmpi */ 892 /*------------------------------------------------------*/ 893 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 894 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 895 896 for (k=0; k<merge->nrecv; k++){ 897 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 898 nrows = *(buf_ri_k[k]); 899 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 900 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 901 } 902 903 for (i=0; i<cm; i++) { 904 row = owners[rank] + i; /* global row index of C_seq */ 905 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 906 ba_i = ba + bi[i]; 907 bnz = bi[i+1] - bi[i]; 908 /* add received vals into ba */ 909 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 910 /* i-th row */ 911 if (i == *nextrow[k]) { 912 cnz = *(nextci[k]+1) - *nextci[k]; 913 cj = buf_rj[k] + *(nextci[k]); 914 ca = abuf_r[k] + *(nextci[k]); 915 nextcj = 0; 916 for (j=0; nextcj<cnz; j++){ 917 if (bj_i[j] == cj[nextcj]){ /* bcol == ccol */ 918 ba_i[j] += ca[nextcj++]; 919 } 920 } 921 nextrow[k]++; nextci[k]++; 922 } 923 } 924 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 925 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 926 } 927 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 928 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 929 ierr = PetscGetTime(&tf);CHKERRQ(ierr); 930 time_setvals += tf-t0; 931 932 ierr = PetscFree(ba);CHKERRQ(ierr); 933 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 934 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 935 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 936 937 ierr = PetscGetTime(&tff);CHKERRQ(ierr); 938 etime += tff - t00; 939 /* 940 PetscInt prid=0; 941 if (rank == prid){ 942 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); 943 } 944 */ 945 PetscFunctionReturn(0); 946 } 947