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 /* 13 #define DEBUG_MATPTAP 14 */ 15 16 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 17 #undef __FUNCT__ 18 #define __FUNCT__ "MatDestroy_MPIAIJ_PtAP" 19 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 20 { 21 PetscErrorCode ierr; 22 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 23 Mat_PtAPMPI *ptap=a->ptap; 24 25 PetscFunctionBegin; 26 if (ptap){ 27 Mat_Merge_SeqsToMPI *merge=ptap->merge; 28 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 29 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 30 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 31 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 32 ierr = MatDestroy(&ptap->A_loc);CHKERRQ(ierr); /* used by MatTransposeMatMult() */ 33 if (ptap->api){ierr = PetscFree(ptap->api);CHKERRQ(ierr);} 34 if (ptap->apj){ierr = PetscFree(ptap->apj);CHKERRQ(ierr);} 35 if (merge) { 36 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 37 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 38 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 39 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 40 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 41 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 42 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 43 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 44 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 45 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 46 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 47 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 48 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 49 ierr = merge->destroy(A);CHKERRQ(ierr); 50 ierr = PetscFree(ptap->merge);CHKERRQ(ierr); 51 } 52 ierr = PetscFree(ptap);CHKERRQ(ierr); 53 } 54 55 PetscFunctionReturn(0); 56 } 57 58 #undef __FUNCT__ 59 #define __FUNCT__ "MatDuplicate_MPIAIJ_MatPtAP" 60 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 61 { 62 PetscErrorCode ierr; 63 Mat_Merge_SeqsToMPI *merge; 64 PetscContainer container; 65 66 PetscFunctionBegin; 67 ierr = PetscObjectQuery((PetscObject)A,"MatMergeSeqsToMPI",(PetscObject *)&container);CHKERRQ(ierr); 68 if (container) { 69 ierr = PetscContainerGetPointer(container,(void **)&merge);CHKERRQ(ierr); 70 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 71 ierr = (*merge->duplicate)(A,op,M);CHKERRQ(ierr); 72 (*M)->ops->destroy = merge->destroy; /* =MatDestroy_MPIAIJ, *M doesn't duplicate A's container! */ 73 (*M)->ops->duplicate = merge->duplicate; /* =MatDuplicate_ MPIAIJ */ 74 PetscFunctionReturn(0); 75 } 76 77 #undef __FUNCT__ 78 #define __FUNCT__ "MatPtAPSymbolic_MPIAIJ" 79 PetscErrorCode MatPtAPSymbolic_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 80 { 81 PetscErrorCode ierr; 82 83 PetscFunctionBegin; 84 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); 85 ierr = (*P->ops->ptapsymbolic_mpiaij)(A,P,fill,C);CHKERRQ(ierr); 86 PetscFunctionReturn(0); 87 } 88 89 #undef __FUNCT__ 90 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ" 91 PetscErrorCode MatPtAPNumeric_MPIAIJ(Mat A,Mat P,Mat C) 92 { 93 PetscErrorCode ierr; 94 95 PetscFunctionBegin; 96 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); 97 ierr = (*P->ops->ptapnumeric_mpiaij)(A,P,C);CHKERRQ(ierr); 98 PetscFunctionReturn(0); 99 } 100 101 #undef __FUNCT__ 102 #define __FUNCT__ "MatPtAPSymbolic_MPIAIJ_MPIAIJ" 103 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 104 { 105 PetscErrorCode ierr; 106 Mat Cmpi; 107 Mat_PtAPMPI *ptap; 108 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 109 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 110 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 111 Mat_SeqAIJ *p_loc,*p_oth; 112 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pdti,*pdtj,*poti,*potj,*ptJ; 113 PetscInt *adi=ad->i,*aj,*aoi=ao->i,nnz; 114 PetscInt nlnk,*lnk,*owners_co,*coi,*coj,i,k,pnz,row; 115 PetscInt am=A->rmap->n,pN=P->cmap->N,pm=P->rmap->n,pn=P->cmap->n; 116 PetscBT lnkbt; 117 MPI_Comm comm=((PetscObject)A)->comm; 118 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri; 119 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 120 PetscInt len,proc,*dnz,*onz,*owners; 121 PetscInt nzi,*bi,*bj; 122 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 123 MPI_Request *swaits,*rwaits; 124 MPI_Status *sstatus,rstatus; 125 Mat_Merge_SeqsToMPI *merge; 126 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,pon,nspacedouble=0,j,ap_rmax=0; 127 PetscReal afill=1.0,afill_tmp; 128 PetscInt rstart = P->cmap->rstart,rmax; 129 PetscScalar *vals; 130 131 PetscFunctionBegin; 132 /* check if matrix local sizes are compatible */ 133 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend){ 134 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); 135 } 136 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend){ 137 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); 138 } 139 140 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 141 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 142 143 /* create struct Mat_PtAPMPI and attached it to C later */ 144 ierr = PetscNew(Mat_PtAPMPI,&ptap);CHKERRQ(ierr); 145 ptap->reuse = MAT_INITIAL_MATRIX; 146 147 148 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 149 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 150 #if defined(DEBUG_MATPTAP) 151 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] P_oth is done...\n",rank); 152 #endif 153 154 /* get P_loc by taking all local rows of P */ 155 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 156 #if defined(DEBUG_MATPTAP) 157 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 158 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] P_loc is done...\n",rank); 159 #endif 160 161 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 162 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 163 pi_loc = p_loc->i; pj_loc = p_loc->j; 164 pi_oth = p_oth->i; pj_oth = p_oth->j; 165 166 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 167 /*-------------------------------------------------------------------*/ 168 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&api);CHKERRQ(ierr); 169 api[0] = 0; 170 171 /* create and initialize a linked list */ 172 nlnk = pN+1; 173 #if defined(DEBUG_MATPTAP) 174 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 175 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] call PetscLLCreate(), pN %D, P_rmax %D %D; Annz %D\n",rank,pN,p_loc->rmax,p_oth->rmax,adi[am]+aoi[am]); 176 #endif 177 ierr = PetscLLCreate(pN,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 178 #if defined(DEBUG_MATPTAP) 179 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 180 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] call PetscLLCreate() is done\n",rank); 181 #endif 182 183 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 184 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 185 current_space = free_space; 186 187 for (i=0; i<am; i++) { 188 apnz = 0; 189 /* diagonal portion of A */ 190 nzi = adi[i+1] - adi[i]; 191 aj = ad->j + adi[i]; 192 for (j=0; j<nzi; j++){ 193 row = aj[j]; 194 pnz = pi_loc[row+1] - pi_loc[row]; 195 Jptr = pj_loc + pi_loc[row]; 196 /* add non-zero cols of P into the sorted linked list lnk */ 197 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 198 apnz += nlnk; 199 } 200 /* off-diagonal portion of A */ 201 nzi = aoi[i+1] - aoi[i]; 202 aj = ao->j + aoi[i]; 203 for (j=0; j<nzi; j++){ 204 row = aj[j]; 205 pnz = pi_oth[row+1] - pi_oth[row]; 206 Jptr = pj_oth + pi_oth[row]; 207 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 208 apnz += nlnk; 209 } 210 211 api[i+1] = api[i] + apnz; 212 if (ap_rmax < apnz) ap_rmax = apnz; 213 214 /* if free space is not available, double the total space in the list */ 215 if (current_space->local_remaining<apnz) { 216 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 217 nspacedouble++; 218 } 219 220 /* Copy data into free space, then initialize lnk */ 221 ierr = PetscLLClean(pN,pN,apnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 222 current_space->array += apnz; 223 current_space->local_used += apnz; 224 current_space->local_remaining -= apnz; 225 } 226 /* Allocate space for apj, initialize apj, and */ 227 /* destroy list of free space and other temporary array(s) */ 228 ierr = PetscMalloc((api[am]+1)*sizeof(PetscInt),&apj);CHKERRQ(ierr); 229 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 230 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]); 231 if (afill_tmp > afill) afill = afill_tmp; 232 233 /* determine symbolic Co=(p->B)^T*AP - send to others */ 234 /*----------------------------------------------------*/ 235 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 236 237 /* then, compute symbolic Co = (p->B)^T*AP */ 238 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 239 >= (num of nonzero rows of C_seq) - pn */ 240 ierr = PetscMalloc((pon+1)*sizeof(PetscInt),&coi);CHKERRQ(ierr); 241 coi[0] = 0; 242 243 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 244 nnz = fill*(poti[pon] + api[am]); 245 ierr = PetscFreeSpaceGet(nnz,&free_space); 246 current_space = free_space; 247 248 for (i=0; i<pon; i++) { 249 nnz = 0; 250 pnz = poti[i+1] - poti[i]; 251 ptJ = potj + poti[i]; 252 for (j=0; j<pnz; j++){ 253 row = ptJ[j]; /* row of AP == col of Pot */ 254 apnz = api[row+1] - api[row]; 255 Jptr = apj + api[row]; 256 /* add non-zero cols of AP into the sorted linked list lnk */ 257 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 258 nnz += nlnk; 259 } 260 261 /* If free space is not available, double the total space in the list */ 262 if (current_space->local_remaining<nnz) { 263 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 264 nspacedouble++; 265 } 266 267 /* Copy data into free space, and zero out denserows */ 268 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 269 current_space->array += nnz; 270 current_space->local_used += nnz; 271 current_space->local_remaining -= nnz; 272 coi[i+1] = coi[i] + nnz; 273 } 274 ierr = PetscMalloc((coi[pon]+1)*sizeof(PetscInt),&coj);CHKERRQ(ierr); 275 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 276 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]); 277 if (afill_tmp > afill) afill = afill_tmp; 278 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 279 280 #if defined(DEBUG_MATPTAP) 281 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 282 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] local Co is done\n",rank); 283 #endif 284 285 /* send j-array (coj) of Co to other processors */ 286 /*----------------------------------------------*/ 287 /* determine row ownership */ 288 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 289 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 290 merge->rowmap->n = pn; 291 merge->rowmap->bs = 1; 292 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 293 owners = merge->rowmap->range; 294 295 /* determine the number of messages to send, their lengths */ 296 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 297 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 298 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 299 len_s = merge->len_s; 300 merge->nsend = 0; 301 302 ierr = PetscMalloc((size+2)*sizeof(PetscInt),&owners_co);CHKERRQ(ierr); 303 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 304 305 proc = 0; 306 for (i=0; i<pon; i++){ 307 while (prmap[i] >= owners[proc+1]) proc++; 308 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 309 len_s[proc] += coi[i+1] - coi[i]; 310 } 311 312 len = 0; /* max length of buf_si[] */ 313 owners_co[0] = 0; 314 for (proc=0; proc<size; proc++){ 315 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 316 if (len_si[proc]){ 317 merge->nsend++; 318 len_si[proc] = 2*(len_si[proc] + 1); 319 len += len_si[proc]; 320 } 321 } 322 323 /* determine the number and length of messages to receive for coi and coj */ 324 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 325 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 326 327 /* post the Irecv and Isend of coj */ 328 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 329 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 330 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&swaits);CHKERRQ(ierr); 331 for (proc=0, k=0; proc<size; proc++){ 332 if (!len_s[proc]) continue; 333 i = owners_co[proc]; 334 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 335 k++; 336 } 337 338 /* receives and sends of coj are complete */ 339 ierr = PetscMalloc(size*sizeof(MPI_Status),&sstatus);CHKERRQ(ierr); 340 for (i=0; i<merge->nrecv; i++){ 341 PetscMPIInt icompleted; 342 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 343 } 344 ierr = PetscFree(rwaits);CHKERRQ(ierr); 345 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 346 347 /* send and recv coi */ 348 /*-------------------*/ 349 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 350 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 351 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 352 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 353 for (proc=0,k=0; proc<size; proc++){ 354 if (!len_s[proc]) continue; 355 /* form outgoing message for i-structure: 356 buf_si[0]: nrows to be sent 357 [1:nrows]: row index (global) 358 [nrows+1:2*nrows+1]: i-structure index 359 */ 360 /*-------------------------------------------*/ 361 nrows = len_si[proc]/2 - 1; 362 buf_si_i = buf_si + nrows+1; 363 buf_si[0] = nrows; 364 buf_si_i[0] = 0; 365 nrows = 0; 366 for (i=owners_co[proc]; i<owners_co[proc+1]; i++){ 367 nzi = coi[i+1] - coi[i]; 368 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 369 buf_si[nrows+1] =prmap[i] -owners[proc]; /* local row index */ 370 nrows++; 371 } 372 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 373 k++; 374 buf_si += len_si[proc]; 375 } 376 i = merge->nrecv; 377 while (i--) { 378 PetscMPIInt icompleted; 379 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 380 } 381 ierr = PetscFree(rwaits);CHKERRQ(ierr); 382 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 383 /* 384 ierr = PetscInfo2(A,"nsend: %d, nrecv: %d\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 385 for (i=0; i<merge->nrecv; i++){ 386 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); 387 } 388 */ 389 ierr = PetscFree(len_si);CHKERRQ(ierr); 390 ierr = PetscFree(len_ri);CHKERRQ(ierr); 391 ierr = PetscFree(swaits);CHKERRQ(ierr); 392 ierr = PetscFree(sstatus);CHKERRQ(ierr); 393 ierr = PetscFree(buf_s);CHKERRQ(ierr); 394 395 #if defined(DEBUG_MATPTAP) 396 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 397 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] Co is done\n",rank); 398 #endif 399 400 /* compute the local portion of C (mpi mat) */ 401 /*------------------------------------------*/ 402 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 403 404 /* allocate bi array and free space for accumulating nonzero column info */ 405 ierr = PetscMalloc((pn+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 406 bi[0] = 0; 407 408 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 409 nnz = fill*(pi_loc[pm] + api[am]); 410 ierr = PetscFreeSpaceGet(nnz,&free_space); 411 current_space = free_space; 412 413 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 414 for (k=0; k<merge->nrecv; k++){ 415 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 416 nrows = *buf_ri_k[k]; 417 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 418 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 419 } 420 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 421 rmax = 0; 422 for (i=0; i<pn; i++) { 423 /* add pdt[i,:]*AP into lnk */ 424 nnz = 0; 425 pnz = pdti[i+1] - pdti[i]; 426 ptJ = pdtj + pdti[i]; 427 for (j=0; j<pnz; j++){ 428 row = ptJ[j]; /* row of AP == col of Pt */ 429 apnz = api[row+1] - api[row]; 430 Jptr = apj + api[row]; 431 /* add non-zero cols of AP into the sorted linked list lnk */ 432 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 433 nnz += nlnk; 434 } 435 /* add received col data into lnk */ 436 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 437 if (i == *nextrow[k]) { /* i-th row */ 438 nzi = *(nextci[k]+1) - *nextci[k]; 439 Jptr = buf_rj[k] + *nextci[k]; 440 ierr = PetscLLAddSorted(nzi,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 441 nnz += nlnk; 442 nextrow[k]++; nextci[k]++; 443 } 444 } 445 446 /* if free space is not available, make more free space */ 447 if (current_space->local_remaining<nnz) { 448 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 449 nspacedouble++; 450 } 451 /* copy data into free space, then initialize lnk */ 452 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 453 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 454 current_space->array += nnz; 455 current_space->local_used += nnz; 456 current_space->local_remaining -= nnz; 457 bi[i+1] = bi[i] + nnz; 458 if (nnz > rmax) rmax = nnz; 459 } 460 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 461 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 462 463 ierr = PetscMalloc((bi[pn]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 464 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 465 afill_tmp = (PetscReal)bi[pn]/(pi_loc[pm] + api[am]); 466 if (afill_tmp > afill) afill = afill_tmp; 467 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 468 469 /* create symbolic parallel matrix Cmpi - why cannot be assembled in Numeric part - check runex55_SA ? */ 470 /*------------------------------------------------------------------------------------------------------*/ 471 ierr = PetscMalloc((rmax+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 472 ierr = PetscMemzero(vals,rmax*sizeof(PetscScalar));CHKERRQ(ierr); 473 474 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 475 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 476 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 477 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 478 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 479 ierr = MatSetBlockSize(Cmpi,1);CHKERRQ(ierr); 480 for (i=0; i<pn; i++){ 481 row = i + rstart; 482 nnz = bi[i+1] - bi[i]; 483 Jptr = bj + bi[i]; 484 ierr = MatSetValues(Cmpi,1,&row,nnz,Jptr,vals,INSERT_VALUES);CHKERRQ(ierr); 485 } 486 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 487 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 488 ierr = PetscFree(vals);CHKERRQ(ierr); 489 490 merge->bi = bi; 491 merge->bj = bj; 492 merge->coi = coi; 493 merge->coj = coj; 494 merge->buf_ri = buf_ri; 495 merge->buf_rj = buf_rj; 496 merge->owners_co = owners_co; 497 merge->destroy = Cmpi->ops->destroy; 498 merge->duplicate = Cmpi->ops->duplicate; 499 500 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 501 /* Cmpi->assembled = PETSC_FALSE; */ 502 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 503 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 504 505 /* attach the supporting struct to Cmpi for reuse */ 506 c = (Mat_MPIAIJ*)Cmpi->data; 507 c->ptap = ptap; 508 ptap->api = api; 509 ptap->apj = apj; 510 ptap->merge = merge; 511 ptap->rmax = ap_rmax; 512 513 *C = Cmpi; 514 #if defined(PETSC_USE_INFO) 515 if (bi[pn] != 0) { 516 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %G needed %G.\n",nspacedouble,fill,afill);CHKERRQ(ierr); 517 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%G,&C) for best performance.\n",afill);CHKERRQ(ierr); 518 } else { 519 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 520 } 521 #endif 522 PetscFunctionReturn(0); 523 } 524 525 #undef __FUNCT__ 526 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 527 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 528 { 529 PetscErrorCode ierr; 530 Mat_Merge_SeqsToMPI *merge; 531 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 532 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 533 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 534 Mat_SeqAIJ *p_loc,*p_oth; 535 Mat_PtAPMPI *ptap; 536 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 537 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 538 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 539 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 540 PetscInt am=A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 541 MPI_Comm comm=((PetscObject)C)->comm; 542 PetscMPIInt size,rank,taga,*len_s; 543 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 544 PetscInt **buf_ri,**buf_rj; 545 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 546 MPI_Request *s_waits,*r_waits; 547 MPI_Status *status; 548 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 549 PetscInt *api,*apj,*coi,*coj; 550 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 551 PetscInt sparse_axpy; 552 553 PetscFunctionBegin; 554 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 555 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 556 557 ptap = c->ptap; 558 merge = ptap->merge; 559 560 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 561 /*--------------------------------------------------*/ 562 if (ptap->reuse == MAT_INITIAL_MATRIX){ 563 ptap->reuse = MAT_REUSE_MATRIX; 564 } else { /* update numerical values of P_oth and P_loc */ 565 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 566 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 567 } 568 569 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 570 /*--------------------------------------------------------------*/ 571 /* get data from symbolic products */ 572 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 573 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 574 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 575 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 576 577 coi = merge->coi; coj = merge->coj; 578 ierr = PetscMalloc((coi[pon]+1)*sizeof(MatScalar),&coa);CHKERRQ(ierr); 579 ierr = PetscMemzero(coa,coi[pon]*sizeof(MatScalar));CHKERRQ(ierr); 580 581 bi = merge->bi; bj = merge->bj; 582 owners = merge->rowmap->range; 583 ierr = PetscMalloc((bi[cm]+1)*sizeof(MatScalar),&ba);CHKERRQ(ierr); 584 ierr = PetscMemzero(ba,bi[cm]*sizeof(MatScalar));CHKERRQ(ierr); 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 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 603 /*------------------------------------------------------------*/ 604 apJ = apj + api[i]; 605 606 /* diagonal portion of A */ 607 anz = adi[i+1] - adi[i]; 608 adj = ad->j + adi[i]; 609 ada = ad->a + adi[i]; 610 for (j=0; j<anz; j++) { 611 row = adj[j]; 612 pnz = pi_loc[row+1] - pi_loc[row]; 613 pj = pj_loc + pi_loc[row]; 614 pa = pa_loc + pi_loc[row]; 615 616 /* perform dense axpy */ 617 valtmp = ada[j]; 618 for (k=0; k<pnz; k++){ 619 apa[pj[k]] += valtmp*pa[k]; 620 } 621 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 622 } 623 624 /* off-diagonal portion of A */ 625 anz = aoi[i+1] - aoi[i]; 626 aoj = ao->j + aoi[i]; 627 aoa = ao->a + aoi[i]; 628 for (j=0; j<anz; j++) { 629 row = aoj[j]; 630 pnz = pi_oth[row+1] - pi_oth[row]; 631 pj = pj_oth + pi_oth[row]; 632 pa = pa_oth + pi_oth[row]; 633 634 /* perform dense axpy */ 635 valtmp = aoa[j]; 636 for (k=0; k<pnz; k++){ 637 apa[pj[k]] += valtmp*pa[k]; 638 } 639 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 640 } 641 642 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 643 /*--------------------------------------------------------------*/ 644 apnz = api[i+1] - api[i]; 645 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 646 pnz = po->i[i+1] - po->i[i]; 647 poJ = po->j + po->i[i]; 648 pA = po->a + po->i[i]; 649 for (j=0; j<pnz; j++){ 650 row = poJ[j]; 651 cnz = coi[row+1] - coi[row]; 652 cj = coj + coi[row]; 653 ca = coa + coi[row]; 654 /* perform dense axpy */ 655 valtmp = pA[j]; 656 for (k=0; k<cnz; k++) { 657 ca[k] += valtmp*apa[cj[k]]; 658 } 659 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 660 } 661 662 /* put the value into Cd (diagonal part) */ 663 pnz = pd->i[i+1] - pd->i[i]; 664 pdJ = pd->j + pd->i[i]; 665 pA = pd->a + pd->i[i]; 666 for (j=0; j<pnz; j++){ 667 row = pdJ[j]; 668 cnz = bi[row+1] - bi[row]; 669 cj = bj + bi[row]; 670 ca = ba + bi[row]; 671 /* perform dense axpy */ 672 valtmp = pA[j]; 673 for (k=0; k<cnz; k++) { 674 ca[k] += valtmp*apa[cj[k]]; 675 } 676 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 677 } 678 679 /* zero the current row of A*P */ 680 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 681 } 682 } else if (sparse_axpy == 1){ /* Perform one sparse axpy */ 683 /*------------------------------------------------------*/ 684 /* malloc apa to store dense row A[i,:]*P */ 685 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 686 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 687 688 for (i=0; i<am; i++) { 689 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 690 /*------------------------------------------------------------*/ 691 apJ = apj + api[i]; 692 693 /* diagonal portion of A */ 694 anz = adi[i+1] - adi[i]; 695 adj = ad->j + adi[i]; 696 ada = ad->a + adi[i]; 697 for (j=0; j<anz; j++) { 698 row = adj[j]; 699 pnz = pi_loc[row+1] - pi_loc[row]; 700 pj = pj_loc + pi_loc[row]; 701 pa = pa_loc + pi_loc[row]; 702 703 /* perform dense axpy */ 704 valtmp = ada[j]; 705 for (k=0; k<pnz; k++){ 706 apa[pj[k]] += valtmp*pa[k]; 707 } 708 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 709 } 710 711 /* off-diagonal portion of A */ 712 anz = aoi[i+1] - aoi[i]; 713 aoj = ao->j + aoi[i]; 714 aoa = ao->a + aoi[i]; 715 for (j=0; j<anz; j++) { 716 row = aoj[j]; 717 pnz = pi_oth[row+1] - pi_oth[row]; 718 pj = pj_oth + pi_oth[row]; 719 pa = pa_oth + pi_oth[row]; 720 721 /* perform dense axpy */ 722 valtmp = aoa[j]; 723 for (k=0; k<pnz; k++){ 724 apa[pj[k]] += valtmp*pa[k]; 725 } 726 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 727 } 728 729 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 730 /*--------------------------------------------------------------*/ 731 apnz = api[i+1] - api[i]; 732 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 733 pnz = po->i[i+1] - po->i[i]; 734 poJ = po->j + po->i[i]; 735 pA = po->a + po->i[i]; 736 for (j=0; j<pnz; j++){ 737 row = poJ[j]; 738 cj = coj + coi[row]; 739 ca = coa + coi[row]; 740 valtmp = pA[j]; 741 /* perform sparse axpy */ 742 nextap = 0; 743 for (k=0; nextap<apnz; k++) { 744 if (cj[k]==apJ[nextap]) { /* global column index */ 745 ca[k] += valtmp*apa[cj[k]]; nextap++; 746 } 747 } 748 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 749 } 750 751 /* put the value into Cd (diagonal part) */ 752 pnz = pd->i[i+1] - pd->i[i]; 753 pdJ = pd->j + pd->i[i]; 754 pA = pd->a + pd->i[i]; 755 for (j=0; j<pnz; j++){ 756 row = pdJ[j]; 757 cj = bj + bi[row]; 758 ca = ba + bi[row]; 759 valtmp = pA[j]; 760 /* perform sparse axpy */ 761 nextap = 0; 762 for (k=0; nextap<apnz; k++) { 763 if (cj[k]==apJ[nextap]) { /* global column index */ 764 ca[k] += valtmp*apa[cj[k]]; 765 nextap++; 766 } 767 } 768 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 769 } 770 771 /* zero the current row of A*P */ 772 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 773 } 774 } else if (sparse_axpy == 2){/* Perform two sparse axpy */ 775 /*----------------------------------------------------*/ 776 /* malloc apa to store sparse row A[i,:]*P */ 777 ierr = PetscMalloc((ptap->rmax+1)*sizeof(MatScalar),&apa);CHKERRQ(ierr); 778 ierr = PetscMemzero(apa,ptap->rmax*sizeof(MatScalar));CHKERRQ(ierr); 779 780 pA=pa_loc; 781 for (i=0; i<am; i++) { 782 /* form i-th sparse row of A*P */ 783 apnz = api[i+1] - api[i]; 784 apJ = apj + api[i]; 785 /* diagonal portion of A */ 786 anz = adi[i+1] - adi[i]; 787 adj = ad->j + adi[i]; 788 ada = ad->a + adi[i]; 789 for (j=0; j<anz; j++) { 790 row = adj[j]; 791 pnz = pi_loc[row+1] - pi_loc[row]; 792 pj = pj_loc + pi_loc[row]; 793 pa = pa_loc + pi_loc[row]; 794 valtmp = ada[j]; 795 nextp = 0; 796 for (k=0; nextp<pnz; k++) { 797 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 798 apa[k] += valtmp*pa[nextp++]; 799 } 800 } 801 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 802 } 803 /* off-diagonal portion of A */ 804 anz = aoi[i+1] - aoi[i]; 805 aoj = ao->j + aoi[i]; 806 aoa = ao->a + aoi[i]; 807 for (j=0; j<anz; j++) { 808 row = aoj[j]; 809 pnz = pi_oth[row+1] - pi_oth[row]; 810 pj = pj_oth + pi_oth[row]; 811 pa = pa_oth + pi_oth[row]; 812 valtmp = aoa[j]; 813 nextp = 0; 814 for (k=0; nextp<pnz; k++) { 815 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 816 apa[k] += valtmp*pa[nextp++]; 817 } 818 } 819 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 820 } 821 822 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 823 /*--------------------------------------------------------------*/ 824 pnz = pi_loc[i+1] - pi_loc[i]; 825 pJ = pj_loc + pi_loc[i]; 826 for (j=0; j<pnz; j++) { 827 nextap = 0; 828 row = pJ[j]; /* global index */ 829 if (row < pcstart || row >=pcend) { /* put the value into Co */ 830 row = *poJ; 831 cj = coj + coi[row]; 832 ca = coa + coi[row]; poJ++; 833 } else { /* put the value into Cd */ 834 row = *pdJ; 835 cj = bj + bi[row]; 836 ca = ba + bi[row]; pdJ++; 837 } 838 valtmp = pA[j]; 839 for (k=0; nextap<apnz; k++) { 840 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 841 } 842 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 843 } 844 pA += pnz; 845 /* zero the current row info for A*P */ 846 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 847 } 848 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"sparse_axpy only takes values 0, 1 and 2"); 849 ierr = PetscFree(apa);CHKERRQ(ierr); 850 851 /* 3) send and recv matrix values coa */ 852 /*------------------------------------*/ 853 buf_ri = merge->buf_ri; 854 buf_rj = merge->buf_rj; 855 len_s = merge->len_s; 856 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 857 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 858 859 ierr = PetscMalloc2(merge->nsend+1,MPI_Request,&s_waits,size,MPI_Status,&status);CHKERRQ(ierr); 860 for (proc=0,k=0; proc<size; proc++){ 861 if (!len_s[proc]) continue; 862 i = merge->owners_co[proc]; 863 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 864 k++; 865 } 866 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 867 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 868 869 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 870 ierr = PetscFree(r_waits);CHKERRQ(ierr); 871 ierr = PetscFree(coa);CHKERRQ(ierr); 872 873 /* 4) insert local Cseq and received values into Cmpi */ 874 /*------------------------------------------------------*/ 875 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 876 for (k=0; k<merge->nrecv; k++){ 877 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 878 nrows = *(buf_ri_k[k]); 879 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 880 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 881 } 882 883 for (i=0; i<cm; i++) { 884 row = owners[rank] + i; /* global row index of C_seq */ 885 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 886 ba_i = ba + bi[i]; 887 bnz = bi[i+1] - bi[i]; 888 /* add received vals into ba */ 889 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 890 /* i-th row */ 891 if (i == *nextrow[k]) { 892 cnz = *(nextci[k]+1) - *nextci[k]; 893 cj = buf_rj[k] + *(nextci[k]); 894 ca = abuf_r[k] + *(nextci[k]); 895 nextcj = 0; 896 for (j=0; nextcj<cnz; j++){ 897 if (bj_i[j] == cj[nextcj]){ /* bcol == ccol */ 898 ba_i[j] += ca[nextcj++]; 899 } 900 } 901 nextrow[k]++; nextci[k]++; 902 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 903 } 904 } 905 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 906 } 907 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 908 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 909 910 ierr = PetscFree(ba);CHKERRQ(ierr); 911 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 912 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 913 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 914 PetscFunctionReturn(0); 915 } 916