1 2 /* 3 Defines matrix-matrix product routines for pairs of MPIAIJ matrices 4 C = A * B 5 */ 6 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 7 #include <../src/mat/utils/freespace.h> 8 #include <../src/mat/impls/aij/mpi/mpiaij.h> 9 #include <petscbt.h> 10 #include <../src/mat/impls/dense/mpi/mpidense.h> 11 #include <petsc/private/vecimpl.h> 12 13 #if defined(PETSC_HAVE_HYPRE) 14 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat*); 15 #endif 16 17 PETSC_INTERN PetscErrorCode MatMatMult_MPIAIJ_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill, Mat *C) 18 { 19 PetscErrorCode ierr; 20 #if defined(PETSC_HAVE_HYPRE) 21 const char *algTypes[4] = {"scalable","nonscalable","seqmpi","hypre"}; 22 PetscInt nalg = 4; 23 #else 24 const char *algTypes[3] = {"scalable","nonscalable","seqmpi"}; 25 PetscInt nalg = 3; 26 #endif 27 PetscInt alg = 1; /* set nonscalable algorithm as default */ 28 MPI_Comm comm; 29 PetscBool flg; 30 31 PetscFunctionBegin; 32 if (scall == MAT_INITIAL_MATRIX) { 33 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 34 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, (%D, %D) != (%D,%D)",A->cmap->rstart,A->cmap->rend,B->rmap->rstart,B->rmap->rend); 35 36 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"MatMatMult","Mat");CHKERRQ(ierr); 37 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[1],&alg,&flg);CHKERRQ(ierr); 38 ierr = PetscOptionsEnd();CHKERRQ(ierr); 39 40 if (!flg && B->cmap->N > 100000) { /* may switch to scalable algorithm as default */ 41 MatInfo Ainfo,Binfo; 42 PetscInt nz_local; 43 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 44 45 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 46 ierr = MatGetInfo(B,MAT_LOCAL,&Binfo);CHKERRQ(ierr); 47 nz_local = (PetscInt)(Ainfo.nz_allocated + Binfo.nz_allocated); 48 49 if (B->cmap->N > fill*nz_local) alg_scalable_loc = PETSC_TRUE; 50 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 51 52 if (alg_scalable) { 53 alg = 0; /* scalable algorithm would 50% slower than nonscalable algorithm */ 54 ierr = PetscInfo2(B,"Use scalable algorithm, BN %D, fill*nz_allocated %g\n",B->cmap->N,fill*nz_local);CHKERRQ(ierr); 55 } 56 } 57 58 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 59 switch (alg) { 60 case 1: 61 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(A,B,fill,C);CHKERRQ(ierr); 62 break; 63 case 2: 64 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ_seqMPI(A,B,fill,C);CHKERRQ(ierr); 65 break; 66 #if defined(PETSC_HAVE_HYPRE) 67 case 3: 68 ierr = MatMatMultSymbolic_AIJ_AIJ_wHYPRE(A,B,fill,C);CHKERRQ(ierr); 69 break; 70 #endif 71 default: 72 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 73 break; 74 } 75 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 76 } 77 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 78 ierr = (*(*C)->ops->matmultnumeric)(A,B,*C);CHKERRQ(ierr); 79 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 80 PetscFunctionReturn(0); 81 } 82 83 PetscErrorCode MatDestroy_MPIAIJ_MatMatMult(Mat A) 84 { 85 PetscErrorCode ierr; 86 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 87 Mat_APMPI *ptap = a->ap; 88 89 PetscFunctionBegin; 90 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 91 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 92 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 93 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 94 ierr = MatDestroy(&ptap->Pt);CHKERRQ(ierr); 95 ierr = PetscFree(ptap->api);CHKERRQ(ierr); 96 ierr = PetscFree(ptap->apj);CHKERRQ(ierr); 97 ierr = PetscFree(ptap->apa);CHKERRQ(ierr); 98 ierr = ptap->destroy(A);CHKERRQ(ierr); 99 ierr = PetscFree(ptap);CHKERRQ(ierr); 100 PetscFunctionReturn(0); 101 } 102 103 PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable(Mat A,Mat P,Mat C) 104 { 105 PetscErrorCode ierr; 106 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*c=(Mat_MPIAIJ*)C->data; 107 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 108 Mat_SeqAIJ *cd =(Mat_SeqAIJ*)(c->A)->data,*co=(Mat_SeqAIJ*)(c->B)->data; 109 PetscScalar *cda=cd->a,*coa=co->a; 110 Mat_SeqAIJ *p_loc,*p_oth; 111 PetscScalar *apa,*ca; 112 PetscInt cm =C->rmap->n; 113 Mat_APMPI *ptap=c->ap; 114 PetscInt *api,*apj,*apJ,i,k; 115 PetscInt cstart=C->cmap->rstart; 116 PetscInt cdnz,conz,k0,k1; 117 MPI_Comm comm; 118 PetscMPIInt size; 119 PetscBool freestruct; 120 121 PetscFunctionBegin; 122 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 123 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 124 125 if (!ptap) { 126 SETERRQ(comm,PETSC_ERR_ARG_WRONGSTATE,"AP cannot be reused. Do not call MatFreeIntermediateDataStructures() or use '-mat_freeintermediatedatastructures'"); 127 } 128 129 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 130 /*-----------------------------------------------------*/ 131 /* update numerical values of P_oth and P_loc */ 132 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 133 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 134 135 /* 2) compute numeric C_loc = A_loc*P = Ad*P_loc + Ao*P_oth */ 136 /*----------------------------------------------------------*/ 137 /* get data from symbolic products */ 138 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 139 p_oth = NULL; 140 if (size >1) { 141 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 142 } 143 144 /* get apa for storing dense row A[i,:]*P */ 145 apa = ptap->apa; 146 147 api = ptap->api; 148 apj = ptap->apj; 149 for (i=0; i<cm; i++) { 150 /* compute apa = A[i,:]*P */ 151 AProw_nonscalable(i,ad,ao,p_loc,p_oth,apa); 152 153 /* set values in C */ 154 apJ = apj + api[i]; 155 cdnz = cd->i[i+1] - cd->i[i]; 156 conz = co->i[i+1] - co->i[i]; 157 158 /* 1st off-diagoanl part of C */ 159 ca = coa + co->i[i]; 160 k = 0; 161 for (k0=0; k0<conz; k0++) { 162 if (apJ[k] >= cstart) break; 163 ca[k0] = apa[apJ[k]]; 164 apa[apJ[k++]] = 0.0; 165 } 166 167 /* diagonal part of C */ 168 ca = cda + cd->i[i]; 169 for (k1=0; k1<cdnz; k1++) { 170 ca[k1] = apa[apJ[k]]; 171 apa[apJ[k++]] = 0.0; 172 } 173 174 /* 2nd off-diagoanl part of C */ 175 ca = coa + co->i[i]; 176 for (; k0<conz; k0++) { 177 ca[k0] = apa[apJ[k]]; 178 apa[apJ[k++]] = 0.0; 179 } 180 } 181 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 182 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 183 184 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatFreeIntermediateDataStructures","Mat");CHKERRQ(ierr); 185 /* supporting struct ptap consumes almost same amount of memory as C=AP, release it if C will not be updated by A and P */ 186 freestruct = PETSC_FALSE; 187 ierr = PetscOptionsBool("-mat_freeintermediatedatastructures","Free intermediate data structures", "MatFreeIntermediateDataStructures",freestruct, &freestruct, NULL);CHKERRQ(ierr); 188 if (freestruct) { 189 ierr = MatFreeIntermediateDataStructures(C);CHKERRQ(ierr); 190 } 191 ierr = PetscOptionsEnd();CHKERRQ(ierr); 192 PetscFunctionReturn(0); 193 } 194 195 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(Mat A,Mat P,PetscReal fill,Mat *C) 196 { 197 PetscErrorCode ierr; 198 MPI_Comm comm; 199 PetscMPIInt size; 200 Mat Cmpi; 201 Mat_APMPI *ptap; 202 PetscFreeSpaceList free_space=NULL,current_space=NULL; 203 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*c; 204 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc,*p_oth; 205 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*dnz,*onz; 206 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,*aoj=ao->j,rstart=A->rmap->rstart; 207 PetscInt *lnk,i,pnz,row,*api,*apj,*Jptr,apnz,nspacedouble=0,j,nzi; 208 PetscInt am=A->rmap->n,pN=P->cmap->N,pn=P->cmap->n,pm=P->rmap->n; 209 PetscBT lnkbt; 210 PetscScalar *apa; 211 PetscReal afill; 212 MatType mtype; 213 214 PetscFunctionBegin; 215 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 216 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 217 218 /* create struct Mat_APMPI and attached it to C later */ 219 ierr = PetscNew(&ptap);CHKERRQ(ierr); 220 221 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 222 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 223 224 /* get P_loc by taking all local rows of P */ 225 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 226 227 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 228 pi_loc = p_loc->i; pj_loc = p_loc->j; 229 if (size > 1) { 230 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 231 pi_oth = p_oth->i; pj_oth = p_oth->j; 232 } else { 233 p_oth = NULL; 234 pi_oth = NULL; pj_oth = NULL; 235 } 236 237 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 238 /*-------------------------------------------------------------------*/ 239 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 240 ptap->api = api; 241 api[0] = 0; 242 243 /* create and initialize a linked list */ 244 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 245 246 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 247 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space);CHKERRQ(ierr); 248 current_space = free_space; 249 250 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 251 for (i=0; i<am; i++) { 252 /* diagonal portion of A */ 253 nzi = adi[i+1] - adi[i]; 254 for (j=0; j<nzi; j++) { 255 row = *adj++; 256 pnz = pi_loc[row+1] - pi_loc[row]; 257 Jptr = pj_loc + pi_loc[row]; 258 /* add non-zero cols of P into the sorted linked list lnk */ 259 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 260 } 261 /* off-diagonal portion of A */ 262 nzi = aoi[i+1] - aoi[i]; 263 for (j=0; j<nzi; j++) { 264 row = *aoj++; 265 pnz = pi_oth[row+1] - pi_oth[row]; 266 Jptr = pj_oth + pi_oth[row]; 267 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 268 } 269 270 apnz = lnk[0]; 271 api[i+1] = api[i] + apnz; 272 273 /* if free space is not available, double the total space in the list */ 274 if (current_space->local_remaining<apnz) { 275 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 276 nspacedouble++; 277 } 278 279 /* Copy data into free space, then initialize lnk */ 280 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 281 ierr = MatPreallocateSet(i+rstart,apnz,current_space->array,dnz,onz);CHKERRQ(ierr); 282 283 current_space->array += apnz; 284 current_space->local_used += apnz; 285 current_space->local_remaining -= apnz; 286 } 287 288 /* Allocate space for apj, initialize apj, and */ 289 /* destroy list of free space and other temporary array(s) */ 290 ierr = PetscMalloc1(api[am]+1,&ptap->apj);CHKERRQ(ierr); 291 apj = ptap->apj; 292 ierr = PetscFreeSpaceContiguous(&free_space,ptap->apj);CHKERRQ(ierr); 293 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 294 295 /* malloc apa to store dense row A[i,:]*P */ 296 ierr = PetscCalloc1(pN,&apa);CHKERRQ(ierr); 297 298 ptap->apa = apa; 299 300 /* create and assemble symbolic parallel matrix Cmpi */ 301 /*----------------------------------------------------*/ 302 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 303 ierr = MatSetSizes(Cmpi,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 304 ierr = MatSetBlockSizesFromMats(Cmpi,A,P);CHKERRQ(ierr); 305 306 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 307 ierr = MatSetType(Cmpi,mtype);CHKERRQ(ierr); 308 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 309 310 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(Cmpi, apj, api);CHKERRQ(ierr); 311 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 312 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 313 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 314 315 ptap->destroy = Cmpi->ops->destroy; 316 ptap->duplicate = Cmpi->ops->duplicate; 317 Cmpi->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 318 Cmpi->ops->destroy = MatDestroy_MPIAIJ_MatMatMult; 319 Cmpi->ops->freeintermediatedatastructures = MatFreeIntermediateDataStructures_MPIAIJ_AP; 320 321 /* attach the supporting struct to Cmpi for reuse */ 322 c = (Mat_MPIAIJ*)Cmpi->data; 323 c->ap = ptap; 324 325 *C = Cmpi; 326 327 /* set MatInfo */ 328 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 329 if (afill < 1.0) afill = 1.0; 330 Cmpi->info.mallocs = nspacedouble; 331 Cmpi->info.fill_ratio_given = fill; 332 Cmpi->info.fill_ratio_needed = afill; 333 334 #if defined(PETSC_USE_INFO) 335 if (api[am]) { 336 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 337 ierr = PetscInfo1(Cmpi,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 338 } else { 339 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 340 } 341 #endif 342 PetscFunctionReturn(0); 343 } 344 345 PETSC_INTERN PetscErrorCode MatMatMult_MPIAIJ_MPIDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 346 { 347 PetscErrorCode ierr; 348 349 PetscFunctionBegin; 350 if (scall == MAT_INITIAL_MATRIX) { 351 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 352 ierr = MatMatMultSymbolic_MPIAIJ_MPIDense(A,B,fill,C);CHKERRQ(ierr); 353 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 354 } 355 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 356 ierr = MatMatMultNumeric_MPIAIJ_MPIDense(A,B,*C);CHKERRQ(ierr); 357 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 358 PetscFunctionReturn(0); 359 } 360 361 typedef struct { 362 Mat workB; 363 PetscScalar *rvalues,*svalues; 364 MPI_Request *rwaits,*swaits; 365 } MPIAIJ_MPIDense; 366 367 PetscErrorCode MatMPIAIJ_MPIDenseDestroy(void *ctx) 368 { 369 MPIAIJ_MPIDense *contents = (MPIAIJ_MPIDense*) ctx; 370 PetscErrorCode ierr; 371 372 PetscFunctionBegin; 373 ierr = MatDestroy(&contents->workB);CHKERRQ(ierr); 374 ierr = PetscFree4(contents->rvalues,contents->svalues,contents->rwaits,contents->swaits);CHKERRQ(ierr); 375 ierr = PetscFree(contents);CHKERRQ(ierr); 376 PetscFunctionReturn(0); 377 } 378 379 #include <petsc/private/vecscatterimpl.h> 380 /* 381 This is a "dummy function" that handles the case where matrix C was created as a dense matrix 382 directly by the user and passed to MatMatMult() with the MAT_REUSE_MATRIX option 383 384 It is the same as MatMatMultSymbolic_MPIAIJ_MPIDense() except does not create C 385 386 Developer Notes: This directly accesses information inside the VecScatter associated with the matrix-vector product 387 for this matrix. This is not desirable.. 388 389 */ 390 PetscErrorCode MatMatMultNumeric_MPIDense(Mat A,Mat B,Mat C) 391 { 392 PetscErrorCode ierr; 393 PetscBool flg; 394 Mat_MPIAIJ *aij = (Mat_MPIAIJ*) A->data; 395 PetscInt nz = aij->B->cmap->n; 396 PetscContainer container; 397 MPIAIJ_MPIDense *contents; 398 VecScatter ctx = aij->Mvctx; 399 VecScatter_MPI_General *from = (VecScatter_MPI_General*) ctx->fromdata; 400 VecScatter_MPI_General *to = (VecScatter_MPI_General*) ctx->todata; 401 402 PetscFunctionBegin; 403 ierr = PetscObjectTypeCompare((PetscObject)B,MATMPIDENSE,&flg);CHKERRQ(ierr); 404 if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Second matrix must be mpidense"); 405 406 /* Handle case where where user provided the final C matrix rather than calling MatMatMult() with MAT_INITIAL_MATRIX*/ 407 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&flg);CHKERRQ(ierr); 408 if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"First matrix must be MPIAIJ"); 409 410 C->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIDense; 411 412 ierr = PetscNew(&contents);CHKERRQ(ierr); 413 /* Create work matrix used to store off processor rows of B needed for local product */ 414 ierr = MatCreateSeqDense(PETSC_COMM_SELF,nz,B->cmap->N,NULL,&contents->workB);CHKERRQ(ierr); 415 /* Create work arrays needed */ 416 ierr = PetscMalloc4(B->cmap->N*from->starts[from->n],&contents->rvalues, 417 B->cmap->N*to->starts[to->n],&contents->svalues, 418 from->n,&contents->rwaits, 419 to->n,&contents->swaits);CHKERRQ(ierr); 420 421 ierr = PetscContainerCreate(PetscObjectComm((PetscObject)A),&container);CHKERRQ(ierr); 422 ierr = PetscContainerSetPointer(container,contents);CHKERRQ(ierr); 423 ierr = PetscContainerSetUserDestroy(container,MatMPIAIJ_MPIDenseDestroy);CHKERRQ(ierr); 424 ierr = PetscObjectCompose((PetscObject)C,"workB",(PetscObject)container);CHKERRQ(ierr); 425 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 426 427 ierr = (*C->ops->matmultnumeric)(A,B,C);CHKERRQ(ierr); 428 PetscFunctionReturn(0); 429 } 430 431 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIDense(Mat A,Mat B,PetscReal fill,Mat *C) 432 { 433 PetscErrorCode ierr; 434 Mat_MPIAIJ *aij = (Mat_MPIAIJ*) A->data; 435 PetscInt nz = aij->B->cmap->n; 436 PetscContainer container; 437 MPIAIJ_MPIDense *contents; 438 VecScatter ctx = aij->Mvctx; 439 VecScatter_MPI_General *from = (VecScatter_MPI_General*) ctx->fromdata; 440 VecScatter_MPI_General *to = (VecScatter_MPI_General*) ctx->todata; 441 PetscInt m = A->rmap->n,n=B->cmap->n; 442 443 PetscFunctionBegin; 444 ierr = MatCreate(PetscObjectComm((PetscObject)B),C);CHKERRQ(ierr); 445 ierr = MatSetSizes(*C,m,n,A->rmap->N,B->cmap->N);CHKERRQ(ierr); 446 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 447 ierr = MatSetType(*C,MATMPIDENSE);CHKERRQ(ierr); 448 ierr = MatMPIDenseSetPreallocation(*C,NULL);CHKERRQ(ierr); 449 ierr = MatAssemblyBegin(*C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 450 ierr = MatAssemblyEnd(*C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 451 452 (*C)->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIDense; 453 454 ierr = PetscNew(&contents);CHKERRQ(ierr); 455 /* Create work matrix used to store off processor rows of B needed for local product */ 456 ierr = MatCreateSeqDense(PETSC_COMM_SELF,nz,B->cmap->N,NULL,&contents->workB);CHKERRQ(ierr); 457 /* Create work arrays needed */ 458 ierr = PetscMalloc4(B->cmap->N*from->starts[from->n],&contents->rvalues, 459 B->cmap->N*to->starts[to->n],&contents->svalues, 460 from->n,&contents->rwaits, 461 to->n,&contents->swaits);CHKERRQ(ierr); 462 463 ierr = PetscContainerCreate(PetscObjectComm((PetscObject)A),&container);CHKERRQ(ierr); 464 ierr = PetscContainerSetPointer(container,contents);CHKERRQ(ierr); 465 ierr = PetscContainerSetUserDestroy(container,MatMPIAIJ_MPIDenseDestroy);CHKERRQ(ierr); 466 ierr = PetscObjectCompose((PetscObject)(*C),"workB",(PetscObject)container);CHKERRQ(ierr); 467 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 468 PetscFunctionReturn(0); 469 } 470 471 /* 472 Performs an efficient scatter on the rows of B needed by this process; this is 473 a modification of the VecScatterBegin_() routines. 474 */ 475 PetscErrorCode MatMPIDenseScatter(Mat A,Mat B,Mat C,Mat *outworkB) 476 { 477 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 478 PetscErrorCode ierr; 479 PetscScalar *b,*w,*svalues,*rvalues; 480 VecScatter ctx = aij->Mvctx; 481 VecScatter_MPI_General *from = (VecScatter_MPI_General*) ctx->fromdata; 482 VecScatter_MPI_General *to = (VecScatter_MPI_General*) ctx->todata; 483 PetscInt i,j,k; 484 PetscInt *sindices,*sstarts,*rindices,*rstarts; 485 PetscMPIInt *sprocs,*rprocs,nrecvs; 486 MPI_Request *swaits,*rwaits; 487 MPI_Comm comm; 488 PetscMPIInt tag = ((PetscObject)ctx)->tag,ncols = B->cmap->N, nrows = aij->B->cmap->n,imdex,nrowsB = B->rmap->n; 489 MPI_Status status; 490 MPIAIJ_MPIDense *contents; 491 PetscContainer container; 492 Mat workB; 493 494 PetscFunctionBegin; 495 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 496 ierr = PetscObjectQuery((PetscObject)C,"workB",(PetscObject*)&container);CHKERRQ(ierr); 497 if (!container) SETERRQ(comm,PETSC_ERR_PLIB,"Container does not exist"); 498 ierr = PetscContainerGetPointer(container,(void**)&contents);CHKERRQ(ierr); 499 500 workB = *outworkB = contents->workB; 501 if (nrows != workB->rmap->n) SETERRQ2(comm,PETSC_ERR_PLIB,"Number of rows of workB %D not equal to columns of aij->B %D",nrows,workB->cmap->n); 502 sindices = to->indices; 503 sstarts = to->starts; 504 sprocs = to->procs; 505 swaits = contents->swaits; 506 svalues = contents->svalues; 507 508 rindices = from->indices; 509 rstarts = from->starts; 510 rprocs = from->procs; 511 rwaits = contents->rwaits; 512 rvalues = contents->rvalues; 513 514 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 515 ierr = MatDenseGetArray(workB,&w);CHKERRQ(ierr); 516 517 for (i=0; i<from->n; i++) { 518 ierr = MPI_Irecv(rvalues+ncols*rstarts[i],ncols*(rstarts[i+1]-rstarts[i]),MPIU_SCALAR,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 519 } 520 521 for (i=0; i<to->n; i++) { 522 /* pack a message at a time */ 523 for (j=0; j<sstarts[i+1]-sstarts[i]; j++) { 524 for (k=0; k<ncols; k++) { 525 svalues[ncols*(sstarts[i] + j) + k] = b[sindices[sstarts[i]+j] + nrowsB*k]; 526 } 527 } 528 ierr = MPI_Isend(svalues+ncols*sstarts[i],ncols*(sstarts[i+1]-sstarts[i]),MPIU_SCALAR,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 529 } 530 531 nrecvs = from->n; 532 while (nrecvs) { 533 ierr = MPI_Waitany(from->n,rwaits,&imdex,&status);CHKERRQ(ierr); 534 nrecvs--; 535 /* unpack a message at a time */ 536 for (j=0; j<rstarts[imdex+1]-rstarts[imdex]; j++) { 537 for (k=0; k<ncols; k++) { 538 w[rindices[rstarts[imdex]+j] + nrows*k] = rvalues[ncols*(rstarts[imdex] + j) + k]; 539 } 540 } 541 } 542 if (to->n) {ierr = MPI_Waitall(to->n,swaits,to->sstatus);CHKERRQ(ierr);} 543 544 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 545 ierr = MatDenseRestoreArray(workB,&w);CHKERRQ(ierr); 546 ierr = MatAssemblyBegin(workB,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 547 ierr = MatAssemblyEnd(workB,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 548 PetscFunctionReturn(0); 549 } 550 extern PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat,Mat,Mat); 551 552 PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIDense(Mat A,Mat B,Mat C) 553 { 554 PetscErrorCode ierr; 555 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 556 Mat_MPIDense *bdense = (Mat_MPIDense*)B->data; 557 Mat_MPIDense *cdense = (Mat_MPIDense*)C->data; 558 Mat workB; 559 560 PetscFunctionBegin; 561 /* diagonal block of A times all local rows of B*/ 562 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(aij->A,bdense->A,cdense->A);CHKERRQ(ierr); 563 564 /* get off processor parts of B needed to complete the product */ 565 ierr = MatMPIDenseScatter(A,B,C,&workB);CHKERRQ(ierr); 566 567 /* off-diagonal block of A times nonlocal rows of B */ 568 ierr = MatMatMultNumericAdd_SeqAIJ_SeqDense(aij->B,workB,cdense->A);CHKERRQ(ierr); 569 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 570 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 571 PetscFunctionReturn(0); 572 } 573 574 PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 575 { 576 PetscErrorCode ierr; 577 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data,*c=(Mat_MPIAIJ*)C->data; 578 Mat_SeqAIJ *ad = (Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 579 Mat_SeqAIJ *cd = (Mat_SeqAIJ*)(c->A)->data,*co=(Mat_SeqAIJ*)(c->B)->data; 580 PetscInt *adi = ad->i,*adj,*aoi=ao->i,*aoj; 581 PetscScalar *ada,*aoa,*cda=cd->a,*coa=co->a; 582 Mat_SeqAIJ *p_loc,*p_oth; 583 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pj; 584 PetscScalar *pa_loc,*pa_oth,*pa,valtmp,*ca; 585 PetscInt cm = C->rmap->n,anz,pnz; 586 Mat_APMPI *ptap = c->ap; 587 PetscScalar *apa_sparse; 588 PetscInt *api,*apj,*apJ,i,j,k,row; 589 PetscInt cstart = C->cmap->rstart; 590 PetscInt cdnz,conz,k0,k1,nextp; 591 MPI_Comm comm; 592 PetscMPIInt size; 593 PetscBool freestruct; 594 595 PetscFunctionBegin; 596 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 597 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 598 599 if (!ptap) { 600 SETERRQ(comm,PETSC_ERR_ARG_WRONGSTATE,"AP cannot be reused. Do not call MatFreeIntermediateDataStructures() or use '-mat_freeintermediatedatastructures'"); 601 } 602 apa_sparse = ptap->apa; 603 604 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 605 /*-----------------------------------------------------*/ 606 /* update numerical values of P_oth and P_loc */ 607 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 608 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 609 610 /* 2) compute numeric C_loc = A_loc*P = Ad*P_loc + Ao*P_oth */ 611 /*----------------------------------------------------------*/ 612 /* get data from symbolic products */ 613 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 614 pi_loc = p_loc->i; pj_loc = p_loc->j; pa_loc = p_loc->a; 615 if (size >1) { 616 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 617 pi_oth = p_oth->i; pj_oth = p_oth->j; pa_oth = p_oth->a; 618 } else { 619 p_oth = NULL; pi_oth = NULL; pj_oth = NULL; pa_oth = NULL; 620 } 621 622 api = ptap->api; 623 apj = ptap->apj; 624 for (i=0; i<cm; i++) { 625 apJ = apj + api[i]; 626 627 /* diagonal portion of A */ 628 anz = adi[i+1] - adi[i]; 629 adj = ad->j + adi[i]; 630 ada = ad->a + adi[i]; 631 for (j=0; j<anz; j++) { 632 row = adj[j]; 633 pnz = pi_loc[row+1] - pi_loc[row]; 634 pj = pj_loc + pi_loc[row]; 635 pa = pa_loc + pi_loc[row]; 636 /* perform sparse axpy */ 637 valtmp = ada[j]; 638 nextp = 0; 639 for (k=0; nextp<pnz; k++) { 640 if (apJ[k] == pj[nextp]) { /* column of AP == column of P */ 641 apa_sparse[k] += valtmp*pa[nextp++]; 642 } 643 } 644 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 645 } 646 647 /* off-diagonal portion of A */ 648 anz = aoi[i+1] - aoi[i]; 649 aoj = ao->j + aoi[i]; 650 aoa = ao->a + aoi[i]; 651 for (j=0; j<anz; j++) { 652 row = aoj[j]; 653 pnz = pi_oth[row+1] - pi_oth[row]; 654 pj = pj_oth + pi_oth[row]; 655 pa = pa_oth + pi_oth[row]; 656 /* perform sparse axpy */ 657 valtmp = aoa[j]; 658 nextp = 0; 659 for (k=0; nextp<pnz; k++) { 660 if (apJ[k] == pj[nextp]) { /* column of AP == column of P */ 661 apa_sparse[k] += valtmp*pa[nextp++]; 662 } 663 } 664 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 665 } 666 667 /* set values in C */ 668 cdnz = cd->i[i+1] - cd->i[i]; 669 conz = co->i[i+1] - co->i[i]; 670 671 /* 1st off-diagoanl part of C */ 672 ca = coa + co->i[i]; 673 k = 0; 674 for (k0=0; k0<conz; k0++) { 675 if (apJ[k] >= cstart) break; 676 ca[k0] = apa_sparse[k]; 677 apa_sparse[k] = 0.0; 678 k++; 679 } 680 681 /* diagonal part of C */ 682 ca = cda + cd->i[i]; 683 for (k1=0; k1<cdnz; k1++) { 684 ca[k1] = apa_sparse[k]; 685 apa_sparse[k] = 0.0; 686 k++; 687 } 688 689 /* 2nd off-diagoanl part of C */ 690 ca = coa + co->i[i]; 691 for (; k0<conz; k0++) { 692 ca[k0] = apa_sparse[k]; 693 apa_sparse[k] = 0.0; 694 k++; 695 } 696 } 697 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 698 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 699 700 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatFreeIntermediateDataStructures","Mat");CHKERRQ(ierr); 701 freestruct = PETSC_FALSE; 702 ierr = PetscOptionsBool("-mat_freeintermediatedatastructures","Free intermediate data structures", "MatFreeIntermediateDataStructures",freestruct, &freestruct, NULL);CHKERRQ(ierr); 703 if (freestruct) { 704 ierr = MatFreeIntermediateDataStructures(C);CHKERRQ(ierr); 705 } 706 ierr = PetscOptionsEnd();CHKERRQ(ierr); 707 PetscFunctionReturn(0); 708 } 709 710 /* same as MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(), except using LLCondensed to avoid O(BN) memory requirement */ 711 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 712 { 713 PetscErrorCode ierr; 714 MPI_Comm comm; 715 PetscMPIInt size; 716 Mat Cmpi; 717 Mat_APMPI *ptap; 718 PetscFreeSpaceList free_space = NULL,current_space=NULL; 719 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data,*c; 720 Mat_SeqAIJ *ad = (Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc,*p_oth; 721 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*dnz,*onz; 722 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,*aoj=ao->j,rstart=A->rmap->rstart; 723 PetscInt i,pnz,row,*api,*apj,*Jptr,apnz,nspacedouble=0,j,nzi,*lnk,apnz_max=0; 724 PetscInt am=A->rmap->n,pn=P->cmap->n,pm=P->rmap->n,lsize=pn+20; 725 PetscReal afill; 726 PetscScalar *apa; 727 MatType mtype; 728 729 PetscFunctionBegin; 730 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 731 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 732 733 /* create struct Mat_APMPI and attached it to C later */ 734 ierr = PetscNew(&ptap);CHKERRQ(ierr); 735 736 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 737 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 738 739 /* get P_loc by taking all local rows of P */ 740 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 741 742 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 743 pi_loc = p_loc->i; pj_loc = p_loc->j; 744 if (size > 1) { 745 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 746 pi_oth = p_oth->i; pj_oth = p_oth->j; 747 } else { 748 p_oth = NULL; 749 pi_oth = NULL; pj_oth = NULL; 750 } 751 752 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 753 /*-------------------------------------------------------------------*/ 754 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 755 ptap->api = api; 756 api[0] = 0; 757 758 ierr = PetscLLCondensedCreate_Scalable(lsize,&lnk);CHKERRQ(ierr); 759 760 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 761 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space);CHKERRQ(ierr); 762 current_space = free_space; 763 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 764 for (i=0; i<am; i++) { 765 /* diagonal portion of A */ 766 nzi = adi[i+1] - adi[i]; 767 for (j=0; j<nzi; j++) { 768 row = *adj++; 769 pnz = pi_loc[row+1] - pi_loc[row]; 770 Jptr = pj_loc + pi_loc[row]; 771 /* Expand list if it is not long enough */ 772 if (pnz+apnz_max > lsize) { 773 lsize = pnz+apnz_max; 774 ierr = PetscLLCondensedExpand_Scalable(lsize, &lnk);CHKERRQ(ierr); 775 } 776 /* add non-zero cols of P into the sorted linked list lnk */ 777 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 778 apnz = *lnk; /* The first element in the list is the number of items in the list */ 779 api[i+1] = api[i] + apnz; 780 if (apnz > apnz_max) apnz_max = apnz; 781 } 782 /* off-diagonal portion of A */ 783 nzi = aoi[i+1] - aoi[i]; 784 for (j=0; j<nzi; j++) { 785 row = *aoj++; 786 pnz = pi_oth[row+1] - pi_oth[row]; 787 Jptr = pj_oth + pi_oth[row]; 788 /* Expand list if it is not long enough */ 789 if (pnz+apnz_max > lsize) { 790 lsize = pnz + apnz_max; 791 ierr = PetscLLCondensedExpand_Scalable(lsize, &lnk);CHKERRQ(ierr); 792 } 793 /* add non-zero cols of P into the sorted linked list lnk */ 794 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 795 apnz = *lnk; /* The first element in the list is the number of items in the list */ 796 api[i+1] = api[i] + apnz; 797 if (apnz > apnz_max) apnz_max = apnz; 798 } 799 apnz = *lnk; 800 api[i+1] = api[i] + apnz; 801 if (apnz > apnz_max) apnz_max = apnz; 802 803 /* if free space is not available, double the total space in the list */ 804 if (current_space->local_remaining<apnz) { 805 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 806 nspacedouble++; 807 } 808 809 /* Copy data into free space, then initialize lnk */ 810 ierr = PetscLLCondensedClean_Scalable(apnz,current_space->array,lnk);CHKERRQ(ierr); 811 ierr = MatPreallocateSet(i+rstart,apnz,current_space->array,dnz,onz);CHKERRQ(ierr); 812 813 current_space->array += apnz; 814 current_space->local_used += apnz; 815 current_space->local_remaining -= apnz; 816 } 817 818 /* Allocate space for apj, initialize apj, and */ 819 /* destroy list of free space and other temporary array(s) */ 820 ierr = PetscMalloc1(api[am]+1,&ptap->apj);CHKERRQ(ierr); 821 apj = ptap->apj; 822 ierr = PetscFreeSpaceContiguous(&free_space,ptap->apj);CHKERRQ(ierr); 823 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 824 825 /* create and assemble symbolic parallel matrix Cmpi */ 826 /*----------------------------------------------------*/ 827 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 828 ierr = MatSetSizes(Cmpi,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 829 ierr = MatSetBlockSizesFromMats(Cmpi,A,P);CHKERRQ(ierr); 830 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 831 ierr = MatSetType(Cmpi,mtype);CHKERRQ(ierr); 832 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 833 834 /* malloc apa for assembly Cmpi */ 835 ierr = PetscCalloc1(apnz_max,&apa);CHKERRQ(ierr); 836 ptap->apa = apa; 837 838 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(Cmpi, apj, api);CHKERRQ(ierr); 839 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 840 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 841 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 842 843 ptap->destroy = Cmpi->ops->destroy; 844 ptap->duplicate = Cmpi->ops->duplicate; 845 Cmpi->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ; 846 Cmpi->ops->destroy = MatDestroy_MPIAIJ_MatMatMult; 847 Cmpi->ops->freeintermediatedatastructures = MatFreeIntermediateDataStructures_MPIAIJ_AP; 848 849 /* attach the supporting struct to Cmpi for reuse */ 850 c = (Mat_MPIAIJ*)Cmpi->data; 851 c->ap = ptap; 852 *C = Cmpi; 853 854 /* set MatInfo */ 855 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 856 if (afill < 1.0) afill = 1.0; 857 Cmpi->info.mallocs = nspacedouble; 858 Cmpi->info.fill_ratio_given = fill; 859 Cmpi->info.fill_ratio_needed = afill; 860 861 #if defined(PETSC_USE_INFO) 862 if (api[am]) { 863 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 864 ierr = PetscInfo1(Cmpi,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 865 } else { 866 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 867 } 868 #endif 869 PetscFunctionReturn(0); 870 } 871 872 /* This function is needed for the seqMPI matrix-matrix multiplication. */ 873 /* Three input arrays are merged to one output array. The size of the */ 874 /* output array is also output. Duplicate entries only show up once. */ 875 static void Merge3SortedArrays(PetscInt size1, PetscInt *in1, 876 PetscInt size2, PetscInt *in2, 877 PetscInt size3, PetscInt *in3, 878 PetscInt *size4, PetscInt *out) 879 { 880 int i = 0, j = 0, k = 0, l = 0; 881 882 /* Traverse all three arrays */ 883 while (i<size1 && j<size2 && k<size3) { 884 if (in1[i] < in2[j] && in1[i] < in3[k]) { 885 out[l++] = in1[i++]; 886 } 887 else if(in2[j] < in1[i] && in2[j] < in3[k]) { 888 out[l++] = in2[j++]; 889 } 890 else if(in3[k] < in1[i] && in3[k] < in2[j]) { 891 out[l++] = in3[k++]; 892 } 893 else if(in1[i] == in2[j] && in1[i] < in3[k]) { 894 out[l++] = in1[i]; 895 i++, j++; 896 } 897 else if(in1[i] == in3[k] && in1[i] < in2[j]) { 898 out[l++] = in1[i]; 899 i++, k++; 900 } 901 else if(in3[k] == in2[j] && in2[j] < in1[i]) { 902 out[l++] = in2[j]; 903 k++, j++; 904 } 905 else if(in1[i] == in2[j] && in1[i] == in3[k]) { 906 out[l++] = in1[i]; 907 i++, j++, k++; 908 } 909 } 910 911 /* Traverse two remaining arrays */ 912 while (i<size1 && j<size2) { 913 if (in1[i] < in2[j]) { 914 out[l++] = in1[i++]; 915 } 916 else if(in1[i] > in2[j]) { 917 out[l++] = in2[j++]; 918 } 919 else { 920 out[l++] = in1[i]; 921 i++, j++; 922 } 923 } 924 925 while (i<size1 && k<size3) { 926 if (in1[i] < in3[k]) { 927 out[l++] = in1[i++]; 928 } 929 else if(in1[i] > in3[k]) { 930 out[l++] = in3[k++]; 931 } 932 else { 933 out[l++] = in1[i]; 934 i++, k++; 935 } 936 } 937 938 while (k<size3 && j<size2) { 939 if (in3[k] < in2[j]) { 940 out[l++] = in3[k++]; 941 } 942 else if(in3[k] > in2[j]) { 943 out[l++] = in2[j++]; 944 } 945 else { 946 out[l++] = in3[k]; 947 k++, j++; 948 } 949 } 950 951 /* Traverse one remaining array */ 952 while (i<size1) out[l++] = in1[i++]; 953 while (j<size2) out[l++] = in2[j++]; 954 while (k<size3) out[l++] = in3[k++]; 955 956 *size4 = l; 957 } 958 959 /* This matrix-matrix multiplication algorithm divides the multiplication into three multiplications and */ 960 /* adds up the products. Two of these three multiplications are performed with existing (sequential) */ 961 /* matrix-matrix multiplications. */ 962 #undef __FUNCT__ 963 #define __FUNCT__ "MatMatMultSymbolic_MPIAIJ_MPIAIJ_seqMPI" 964 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ_seqMPI(Mat A, Mat P, PetscReal fill, Mat *C) 965 { 966 PetscErrorCode ierr; 967 MPI_Comm comm; 968 PetscMPIInt size; 969 Mat Cmpi; 970 Mat_APMPI *ptap; 971 PetscFreeSpaceList free_space_diag=NULL, current_space=NULL; 972 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data; 973 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc; 974 Mat_MPIAIJ *p =(Mat_MPIAIJ*)P->data; 975 Mat_MPIAIJ *c; 976 Mat_SeqAIJ *adpd_seq, *p_off, *aopoth_seq; 977 PetscInt adponz, adpdnz; 978 PetscInt *pi_loc,*dnz,*onz; 979 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,rstart=A->rmap->rstart; 980 PetscInt *lnk,i, i1=0,pnz,row,*adpoi,*adpoj, *api, *adpoJ, *aopJ, *apJ,*Jptr, aopnz, nspacedouble=0,j,nzi, 981 *apj,apnz, *adpdi, *adpdj, *adpdJ, *poff_i, *poff_j, *j_temp, *aopothi, *aopothj; 982 PetscInt am=A->rmap->n,pN=P->cmap->N,pn=P->cmap->n,pm=P->rmap->n, p_colstart, p_colend; 983 PetscBT lnkbt; 984 PetscScalar *apa; 985 PetscReal afill; 986 PetscMPIInt rank; 987 Mat adpd, aopoth; 988 MatType mtype; 989 990 PetscFunctionBegin; 991 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 992 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 993 ierr = MPI_Comm_rank(comm, &rank);CHKERRQ(ierr); 994 ierr = MatGetOwnershipRangeColumn(P, &p_colstart, &p_colend); CHKERRQ(ierr); 995 996 /* create struct Mat_APMPI and attached it to C later */ 997 ierr = PetscNew(&ptap);CHKERRQ(ierr); 998 999 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 1000 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 1001 1002 /* get P_loc by taking all local rows of P */ 1003 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 1004 1005 1006 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 1007 pi_loc = p_loc->i; 1008 1009 /* Allocate memory for the i arrays of the matrices A*P, A_diag*P_off and A_offd * P */ 1010 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 1011 ierr = PetscMalloc1(am+2,&adpoi);CHKERRQ(ierr); 1012 1013 adpoi[0] = 0; 1014 ptap->api = api; 1015 api[0] = 0; 1016 1017 /* create and initialize a linked list, will be used for both A_diag * P_loc_off and A_offd * P_oth */ 1018 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 1019 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 1020 1021 /* Symbolic calc of A_loc_diag * P_loc_diag */ 1022 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(a->A, p->A, fill, &adpd);CHKERRQ(ierr); 1023 adpd_seq = (Mat_SeqAIJ*)((adpd)->data); 1024 adpdi = adpd_seq->i; adpdj = adpd_seq->j; 1025 p_off = (Mat_SeqAIJ*)((p->B)->data); 1026 poff_i = p_off->i; poff_j = p_off->j; 1027 1028 /* j_temp stores indices of a result row before they are added to the linked list */ 1029 ierr = PetscMalloc1(pN+2,&j_temp);CHKERRQ(ierr); 1030 1031 1032 /* Symbolic calc of the A_diag * p_loc_off */ 1033 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 1034 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space_diag);CHKERRQ(ierr); 1035 current_space = free_space_diag; 1036 1037 for (i=0; i<am; i++) { 1038 /* A_diag * P_loc_off */ 1039 nzi = adi[i+1] - adi[i]; 1040 for (j=0; j<nzi; j++) { 1041 row = *adj++; 1042 pnz = poff_i[row+1] - poff_i[row]; 1043 Jptr = poff_j + poff_i[row]; 1044 for(i1 = 0; i1 < pnz; i1++) { 1045 j_temp[i1] = p->garray[Jptr[i1]]; 1046 } 1047 /* add non-zero cols of P into the sorted linked list lnk */ 1048 ierr = PetscLLCondensedAddSorted(pnz,j_temp,lnk,lnkbt);CHKERRQ(ierr); 1049 } 1050 1051 adponz = lnk[0]; 1052 adpoi[i+1] = adpoi[i] + adponz; 1053 1054 /* if free space is not available, double the total space in the list */ 1055 if (current_space->local_remaining<adponz) { 1056 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(adponz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1057 nspacedouble++; 1058 } 1059 1060 /* Copy data into free space, then initialize lnk */ 1061 ierr = PetscLLCondensedClean(pN,adponz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1062 1063 current_space->array += adponz; 1064 current_space->local_used += adponz; 1065 current_space->local_remaining -= adponz; 1066 } 1067 1068 /* Symbolic calc of A_off * P_oth */ 1069 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(a->B, ptap->P_oth, fill, &aopoth);CHKERRQ(ierr); 1070 aopoth_seq = (Mat_SeqAIJ*)((aopoth)->data); 1071 aopothi = aopoth_seq->i; aopothj = aopoth_seq->j; 1072 1073 /* Allocate space for apj, adpj, aopj, ... */ 1074 /* destroy lists of free space and other temporary array(s) */ 1075 1076 ierr = PetscMalloc1(aopothi[am] + adpoi[am] + adpdi[am]+2, &ptap->apj);CHKERRQ(ierr); 1077 ierr = PetscMalloc1(adpoi[am]+2, &adpoj);CHKERRQ(ierr); 1078 1079 /* Copy from linked list to j-array */ 1080 ierr = PetscFreeSpaceContiguous(&free_space_diag,adpoj);CHKERRQ(ierr); 1081 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1082 1083 adpoJ = adpoj; 1084 adpdJ = adpdj; 1085 aopJ = aopothj; 1086 apj = ptap->apj; 1087 apJ = apj; /* still empty */ 1088 1089 /* Merge j-arrays of A_off * P, A_diag * P_loc_off, and */ 1090 /* A_diag * P_loc_diag to get A*P */ 1091 for (i = 0; i < am; i++) { 1092 aopnz = aopothi[i+1] - aopothi[i]; 1093 adponz = adpoi[i+1] - adpoi[i]; 1094 adpdnz = adpdi[i+1] - adpdi[i]; 1095 1096 /* Correct indices from A_diag*P_diag */ 1097 for(i1 = 0; i1 < adpdnz; i1++) { 1098 adpdJ[i1] += p_colstart; 1099 } 1100 /* Merge j-arrays of A_diag * P_loc_off and A_diag * P_loc_diag and A_off * P_oth */ 1101 Merge3SortedArrays(adponz, adpoJ, adpdnz, adpdJ, aopnz, aopJ, &apnz, apJ); 1102 ierr = MatPreallocateSet(i+rstart, apnz, apJ, dnz, onz); CHKERRQ(ierr); 1103 1104 aopJ += aopnz; 1105 adpoJ += adponz; 1106 adpdJ += adpdnz; 1107 apJ += apnz; 1108 api[i+1] = api[i] + apnz; 1109 } 1110 1111 /* malloc apa to store dense row A[i,:]*P */ 1112 ierr = PetscCalloc1(pN+2,&apa);CHKERRQ(ierr); 1113 1114 ptap->apa = apa; 1115 /* create and assemble symbolic parallel matrix Cmpi */ 1116 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 1117 ierr = MatSetSizes(Cmpi,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 1118 ierr = MatSetBlockSizesFromMats(Cmpi,A,P);CHKERRQ(ierr); 1119 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 1120 ierr = MatSetType(Cmpi,mtype);CHKERRQ(ierr); 1121 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 1122 1123 1124 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(Cmpi, apj, api);CHKERRQ(ierr); 1125 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1126 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1127 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 1128 1129 1130 ptap->destroy = Cmpi->ops->destroy; 1131 ptap->duplicate = Cmpi->ops->duplicate; 1132 Cmpi->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 1133 Cmpi->ops->destroy = MatDestroy_MPIAIJ_MatMatMult; 1134 1135 /* attach the supporting struct to Cmpi for reuse */ 1136 c = (Mat_MPIAIJ*)Cmpi->data; 1137 c->ap = ptap; 1138 *C = Cmpi; 1139 1140 /* set MatInfo */ 1141 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 1142 if (afill < 1.0) afill = 1.0; 1143 Cmpi->info.mallocs = nspacedouble; 1144 Cmpi->info.fill_ratio_given = fill; 1145 Cmpi->info.fill_ratio_needed = afill; 1146 1147 #if defined(PETSC_USE_INFO) 1148 if (api[am]) { 1149 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 1150 ierr = PetscInfo1(Cmpi,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1151 } else { 1152 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 1153 } 1154 #endif 1155 1156 ierr = MatDestroy(&aopoth);CHKERRQ(ierr); 1157 ierr = MatDestroy(&adpd);CHKERRQ(ierr); 1158 ierr = PetscFree(j_temp);CHKERRQ(ierr); 1159 ierr = PetscFree(adpoj);CHKERRQ(ierr); 1160 ierr = PetscFree(adpoi);CHKERRQ(ierr); 1161 PetscFunctionReturn(0); 1162 } 1163 1164 1165 /*-------------------------------------------------------------------------*/ 1166 PetscErrorCode MatTransposeMatMult_MPIAIJ_MPIAIJ(Mat P,Mat A,MatReuse scall,PetscReal fill,Mat *C) 1167 { 1168 PetscErrorCode ierr; 1169 const char *algTypes[3] = {"scalable","nonscalable","matmatmult"}; 1170 PetscInt aN=A->cmap->N,alg=1; /* set default algorithm */ 1171 PetscBool flg; 1172 1173 PetscFunctionBegin; 1174 if (scall == MAT_INITIAL_MATRIX) { 1175 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"MatTransposeMatMult","Mat");CHKERRQ(ierr); 1176 ierr = PetscOptionsEList("-mattransposematmult_via","Algorithmic approach","MatTransposeMatMult",algTypes,3,algTypes[1],&alg,&flg);CHKERRQ(ierr); 1177 ierr = PetscOptionsEnd();CHKERRQ(ierr); 1178 1179 ierr = PetscLogEventBegin(MAT_TransposeMatMultSymbolic,P,A,0,0);CHKERRQ(ierr); 1180 switch (alg) { 1181 case 1: 1182 if (!flg && aN > 100000) { /* may switch to scalable algorithm as default */ 1183 MatInfo Ainfo,Pinfo; 1184 PetscInt nz_local; 1185 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 1186 MPI_Comm comm; 1187 1188 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 1189 ierr = MatGetInfo(P,MAT_LOCAL,&Pinfo);CHKERRQ(ierr); 1190 nz_local = (PetscInt)(Ainfo.nz_allocated + Pinfo.nz_allocated); /* estimated local nonzero entries */ 1191 1192 if (aN > fill*nz_local) alg_scalable_loc = PETSC_TRUE; 1193 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1194 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 1195 1196 if (alg_scalable) { 1197 alg = 0; /* scalable algorithm would slower than nonscalable algorithm */ 1198 ierr = MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ(P,A,fill,C);CHKERRQ(ierr); 1199 break; 1200 } 1201 } 1202 ierr = MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(P,A,fill,C);CHKERRQ(ierr); 1203 break; 1204 case 2: 1205 { 1206 Mat Pt; 1207 Mat_APMPI *ptap; 1208 Mat_MPIAIJ *c; 1209 ierr = MatTranspose(P,MAT_INITIAL_MATRIX,&Pt);CHKERRQ(ierr); 1210 ierr = MatMatMult(Pt,A,MAT_INITIAL_MATRIX,fill,C);CHKERRQ(ierr); 1211 c = (Mat_MPIAIJ*)(*C)->data; 1212 ptap = c->ap; 1213 ptap->Pt = Pt; 1214 (*C)->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_matmatmult; 1215 (*C)->ops->freeintermediatedatastructures = MatFreeIntermediateDataStructures_MPIAIJ_AP; 1216 PetscFunctionReturn(0); 1217 } 1218 break; 1219 default: /* scalable algorithm */ 1220 ierr = MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ(P,A,fill,C);CHKERRQ(ierr); 1221 break; 1222 } 1223 ierr = PetscLogEventEnd(MAT_TransposeMatMultSymbolic,P,A,0,0);CHKERRQ(ierr); 1224 } 1225 ierr = PetscLogEventBegin(MAT_TransposeMatMultNumeric,P,A,0,0);CHKERRQ(ierr); 1226 ierr = (*(*C)->ops->mattransposemultnumeric)(P,A,*C);CHKERRQ(ierr); 1227 ierr = PetscLogEventEnd(MAT_TransposeMatMultNumeric,P,A,0,0);CHKERRQ(ierr); 1228 PetscFunctionReturn(0); 1229 } 1230 1231 /* This routine only works when scall=MAT_REUSE_MATRIX! */ 1232 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_matmatmult(Mat P,Mat A,Mat C) 1233 { 1234 PetscErrorCode ierr; 1235 Mat_MPIAIJ *c=(Mat_MPIAIJ*)C->data; 1236 Mat_APMPI *ptap= c->ap; 1237 Mat Pt; 1238 PetscBool freestruct; 1239 1240 PetscFunctionBegin; 1241 if (!ptap) { 1242 MPI_Comm comm; 1243 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1244 SETERRQ(comm,PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatFreeIntermediateDataStructures() or use '-mat_freeintermediatedatastructures'"); 1245 } 1246 1247 Pt=ptap->Pt; 1248 ierr = MatTranspose(P,MAT_REUSE_MATRIX,&Pt);CHKERRQ(ierr); 1249 ierr = MatMatMultNumeric(Pt,A,C);CHKERRQ(ierr); 1250 1251 /* supporting struct ptap consumes almost same amount of memory as C=PtAP, release it if C will not be updated by A and P */ 1252 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatFreeIntermediateDataStructures","Mat");CHKERRQ(ierr); 1253 freestruct = PETSC_FALSE; 1254 ierr = PetscOptionsBool("-mat_freeintermediatedatastructures","Free intermediate data structures", "MatFreeIntermediateDataStructures",freestruct, &freestruct, NULL);CHKERRQ(ierr); 1255 if (freestruct) { 1256 ierr = MatFreeIntermediateDataStructures(C);CHKERRQ(ierr); 1257 } 1258 ierr = PetscOptionsEnd();CHKERRQ(ierr); 1259 PetscFunctionReturn(0); 1260 } 1261 1262 /* This routine is modified from MatPtAPSymbolic_MPIAIJ_MPIAIJ() */ 1263 PetscErrorCode MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(Mat P,Mat A,PetscReal fill,Mat *C) 1264 { 1265 PetscErrorCode ierr; 1266 Mat_APMPI *ptap; 1267 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data,*c; 1268 MPI_Comm comm; 1269 PetscMPIInt size,rank; 1270 Mat Cmpi; 1271 PetscFreeSpaceList free_space=NULL,current_space=NULL; 1272 PetscInt pn=P->cmap->n,aN=A->cmap->N,an=A->cmap->n; 1273 PetscInt *lnk,i,k,nsend; 1274 PetscBT lnkbt; 1275 PetscMPIInt tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0,nrecv; 1276 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 1277 PetscInt len,proc,*dnz,*onz,*owners,nzi; 1278 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 1279 MPI_Request *swaits,*rwaits; 1280 MPI_Status *sstatus,rstatus; 1281 PetscLayout rowmap; 1282 PetscInt *owners_co,*coi,*coj; /* i and j array of (p->B)^T*A*P - used in the communication */ 1283 PetscMPIInt *len_r,*id_r; /* array of length of comm->size, store send/recv matrix values */ 1284 PetscInt *Jptr,*prmap=p->garray,con,j,Crmax; 1285 Mat_SeqAIJ *a_loc,*c_loc,*c_oth; 1286 PetscTable ta; 1287 MatType mtype; 1288 1289 PetscFunctionBegin; 1290 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1291 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1292 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1293 1294 /* create symbolic parallel matrix Cmpi */ 1295 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 1296 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 1297 ierr = MatSetType(Cmpi,mtype);CHKERRQ(ierr); 1298 1299 Cmpi->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 1300 1301 /* create struct Mat_APMPI and attached it to C later */ 1302 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1303 ptap->reuse = MAT_INITIAL_MATRIX; 1304 1305 /* (0) compute Rd = Pd^T, Ro = Po^T */ 1306 /* --------------------------------- */ 1307 ierr = MatTranspose_SeqAIJ(p->A,MAT_INITIAL_MATRIX,&ptap->Rd);CHKERRQ(ierr); 1308 ierr = MatTranspose_SeqAIJ(p->B,MAT_INITIAL_MATRIX,&ptap->Ro);CHKERRQ(ierr); 1309 1310 /* (1) compute symbolic A_loc */ 1311 /* ---------------------------*/ 1312 ierr = MatMPIAIJGetLocalMat(A,MAT_INITIAL_MATRIX,&ptap->A_loc);CHKERRQ(ierr); 1313 1314 /* (2-1) compute symbolic C_oth = Ro*A_loc */ 1315 /* ------------------------------------ */ 1316 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Ro,ptap->A_loc,fill,&ptap->C_oth);CHKERRQ(ierr); 1317 1318 /* (3) send coj of C_oth to other processors */ 1319 /* ------------------------------------------ */ 1320 /* determine row ownership */ 1321 ierr = PetscLayoutCreate(comm,&rowmap);CHKERRQ(ierr); 1322 rowmap->n = pn; 1323 rowmap->bs = 1; 1324 ierr = PetscLayoutSetUp(rowmap);CHKERRQ(ierr); 1325 owners = rowmap->range; 1326 1327 /* determine the number of messages to send, their lengths */ 1328 ierr = PetscMalloc4(size,&len_s,size,&len_si,size,&sstatus,size+2,&owners_co);CHKERRQ(ierr); 1329 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 1330 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 1331 1332 c_oth = (Mat_SeqAIJ*)ptap->C_oth->data; 1333 coi = c_oth->i; coj = c_oth->j; 1334 con = ptap->C_oth->rmap->n; 1335 proc = 0; 1336 for (i=0; i<con; i++) { 1337 while (prmap[i] >= owners[proc+1]) proc++; 1338 len_si[proc]++; /* num of rows in Co(=Pt*A) to be sent to [proc] */ 1339 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 1340 } 1341 1342 len = 0; /* max length of buf_si[], see (4) */ 1343 owners_co[0] = 0; 1344 nsend = 0; 1345 for (proc=0; proc<size; proc++) { 1346 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 1347 if (len_s[proc]) { 1348 nsend++; 1349 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 1350 len += len_si[proc]; 1351 } 1352 } 1353 1354 /* determine the number and length of messages to receive for coi and coj */ 1355 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&nrecv);CHKERRQ(ierr); 1356 ierr = PetscGatherMessageLengths2(comm,nsend,nrecv,len_s,len_si,&id_r,&len_r,&len_ri);CHKERRQ(ierr); 1357 1358 /* post the Irecv and Isend of coj */ 1359 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 1360 ierr = PetscPostIrecvInt(comm,tagj,nrecv,id_r,len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 1361 ierr = PetscMalloc1(nsend+1,&swaits);CHKERRQ(ierr); 1362 for (proc=0, k=0; proc<size; proc++) { 1363 if (!len_s[proc]) continue; 1364 i = owners_co[proc]; 1365 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 1366 k++; 1367 } 1368 1369 /* (2-2) compute symbolic C_loc = Rd*A_loc */ 1370 /* ---------------------------------------- */ 1371 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Rd,ptap->A_loc,fill,&ptap->C_loc);CHKERRQ(ierr); 1372 c_loc = (Mat_SeqAIJ*)ptap->C_loc->data; 1373 1374 /* receives coj are complete */ 1375 for (i=0; i<nrecv; i++) { 1376 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1377 } 1378 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1379 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 1380 1381 /* add received column indices into ta to update Crmax */ 1382 a_loc = (Mat_SeqAIJ*)(ptap->A_loc)->data; 1383 1384 /* create and initialize a linked list */ 1385 ierr = PetscTableCreate(an,aN,&ta);CHKERRQ(ierr); /* for compute Crmax */ 1386 MatRowMergeMax_SeqAIJ(a_loc,ptap->A_loc->rmap->N,ta); 1387 1388 for (k=0; k<nrecv; k++) {/* k-th received message */ 1389 Jptr = buf_rj[k]; 1390 for (j=0; j<len_r[k]; j++) { 1391 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 1392 } 1393 } 1394 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 1395 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 1396 1397 /* (4) send and recv coi */ 1398 /*-----------------------*/ 1399 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 1400 ierr = PetscPostIrecvInt(comm,tagi,nrecv,id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 1401 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 1402 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 1403 for (proc=0,k=0; proc<size; proc++) { 1404 if (!len_s[proc]) continue; 1405 /* form outgoing message for i-structure: 1406 buf_si[0]: nrows to be sent 1407 [1:nrows]: row index (global) 1408 [nrows+1:2*nrows+1]: i-structure index 1409 */ 1410 /*-------------------------------------------*/ 1411 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 1412 buf_si_i = buf_si + nrows+1; 1413 buf_si[0] = nrows; 1414 buf_si_i[0] = 0; 1415 nrows = 0; 1416 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 1417 nzi = coi[i+1] - coi[i]; 1418 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 1419 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 1420 nrows++; 1421 } 1422 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 1423 k++; 1424 buf_si += len_si[proc]; 1425 } 1426 for (i=0; i<nrecv; i++) { 1427 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1428 } 1429 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1430 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 1431 1432 ierr = PetscFree4(len_s,len_si,sstatus,owners_co);CHKERRQ(ierr); 1433 ierr = PetscFree(len_ri);CHKERRQ(ierr); 1434 ierr = PetscFree(swaits);CHKERRQ(ierr); 1435 ierr = PetscFree(buf_s);CHKERRQ(ierr); 1436 1437 /* (5) compute the local portion of Cmpi */ 1438 /* ------------------------------------------ */ 1439 /* set initial free space to be Crmax, sufficient for holding nozeros in each row of Cmpi */ 1440 ierr = PetscFreeSpaceGet(Crmax,&free_space);CHKERRQ(ierr); 1441 current_space = free_space; 1442 1443 ierr = PetscMalloc3(nrecv,&buf_ri_k,nrecv,&nextrow,nrecv,&nextci);CHKERRQ(ierr); 1444 for (k=0; k<nrecv; k++) { 1445 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1446 nrows = *buf_ri_k[k]; 1447 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 1448 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1449 } 1450 1451 ierr = MatPreallocateInitialize(comm,pn,an,dnz,onz);CHKERRQ(ierr); 1452 ierr = PetscLLCondensedCreate(Crmax,aN,&lnk,&lnkbt);CHKERRQ(ierr); 1453 for (i=0; i<pn; i++) { 1454 /* add C_loc into Cmpi */ 1455 nzi = c_loc->i[i+1] - c_loc->i[i]; 1456 Jptr = c_loc->j + c_loc->i[i]; 1457 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1458 1459 /* add received col data into lnk */ 1460 for (k=0; k<nrecv; k++) { /* k-th received message */ 1461 if (i == *nextrow[k]) { /* i-th row */ 1462 nzi = *(nextci[k]+1) - *nextci[k]; 1463 Jptr = buf_rj[k] + *nextci[k]; 1464 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1465 nextrow[k]++; nextci[k]++; 1466 } 1467 } 1468 nzi = lnk[0]; 1469 1470 /* copy data into free space, then initialize lnk */ 1471 ierr = PetscLLCondensedClean(aN,nzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1472 ierr = MatPreallocateSet(i+owners[rank],nzi,current_space->array,dnz,onz);CHKERRQ(ierr); 1473 } 1474 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1475 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1476 ierr = PetscFreeSpaceDestroy(free_space);CHKERRQ(ierr); 1477 1478 /* local sizes and preallocation */ 1479 ierr = MatSetSizes(Cmpi,pn,an,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 1480 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 1481 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 1482 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 1483 1484 /* members in merge */ 1485 ierr = PetscFree(id_r);CHKERRQ(ierr); 1486 ierr = PetscFree(len_r);CHKERRQ(ierr); 1487 ierr = PetscFree(buf_ri[0]);CHKERRQ(ierr); 1488 ierr = PetscFree(buf_ri);CHKERRQ(ierr); 1489 ierr = PetscFree(buf_rj[0]);CHKERRQ(ierr); 1490 ierr = PetscFree(buf_rj);CHKERRQ(ierr); 1491 ierr = PetscLayoutDestroy(&rowmap);CHKERRQ(ierr); 1492 1493 /* attach the supporting struct to Cmpi for reuse */ 1494 c = (Mat_MPIAIJ*)Cmpi->data; 1495 c->ap = ptap; 1496 ptap->destroy = Cmpi->ops->destroy; 1497 1498 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 1499 Cmpi->assembled = PETSC_FALSE; 1500 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 1501 Cmpi->ops->freeintermediatedatastructures = MatFreeIntermediateDataStructures_MPIAIJ_AP; 1502 1503 *C = Cmpi; 1504 PetscFunctionReturn(0); 1505 } 1506 1507 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable(Mat P,Mat A,Mat C) 1508 { 1509 PetscErrorCode ierr; 1510 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 1511 Mat_SeqAIJ *c_seq; 1512 Mat_APMPI *ptap = c->ap; 1513 Mat A_loc,C_loc,C_oth; 1514 PetscInt i,rstart,rend,cm,ncols,row; 1515 const PetscInt *cols; 1516 const PetscScalar *vals; 1517 PetscBool freestruct; 1518 1519 PetscFunctionBegin; 1520 if (!ptap) { 1521 MPI_Comm comm; 1522 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1523 SETERRQ(comm,PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatFreeIntermediateDataStructures() or use '-mat_freeintermediatedatastructures'"); 1524 } 1525 1526 ierr = MatZeroEntries(C);CHKERRQ(ierr); 1527 1528 if (ptap->reuse == MAT_REUSE_MATRIX) { 1529 /* These matrices are obtained in MatTransposeMatMultSymbolic() */ 1530 /* 1) get R = Pd^T, Ro = Po^T */ 1531 /*----------------------------*/ 1532 ierr = MatTranspose_SeqAIJ(p->A,MAT_REUSE_MATRIX,&ptap->Rd);CHKERRQ(ierr); 1533 ierr = MatTranspose_SeqAIJ(p->B,MAT_REUSE_MATRIX,&ptap->Ro);CHKERRQ(ierr); 1534 1535 /* 2) compute numeric A_loc */ 1536 /*--------------------------*/ 1537 ierr = MatMPIAIJGetLocalMat(A,MAT_REUSE_MATRIX,&ptap->A_loc);CHKERRQ(ierr); 1538 } 1539 1540 /* 3) C_loc = Rd*A_loc, C_oth = Ro*A_loc */ 1541 A_loc = ptap->A_loc; 1542 ierr = ((ptap->C_loc)->ops->matmultnumeric)(ptap->Rd,A_loc,ptap->C_loc);CHKERRQ(ierr); 1543 ierr = ((ptap->C_oth)->ops->matmultnumeric)(ptap->Ro,A_loc,ptap->C_oth);CHKERRQ(ierr); 1544 C_loc = ptap->C_loc; 1545 C_oth = ptap->C_oth; 1546 1547 /* add C_loc and Co to to C */ 1548 ierr = MatGetOwnershipRange(C,&rstart,&rend);CHKERRQ(ierr); 1549 1550 /* C_loc -> C */ 1551 cm = C_loc->rmap->N; 1552 c_seq = (Mat_SeqAIJ*)C_loc->data; 1553 cols = c_seq->j; 1554 vals = c_seq->a; 1555 for (i=0; i<cm; i++) { 1556 ncols = c_seq->i[i+1] - c_seq->i[i]; 1557 row = rstart + i; 1558 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1559 cols += ncols; vals += ncols; 1560 } 1561 1562 /* Co -> C, off-processor part */ 1563 cm = C_oth->rmap->N; 1564 c_seq = (Mat_SeqAIJ*)C_oth->data; 1565 cols = c_seq->j; 1566 vals = c_seq->a; 1567 for (i=0; i<cm; i++) { 1568 ncols = c_seq->i[i+1] - c_seq->i[i]; 1569 row = p->garray[i]; 1570 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1571 cols += ncols; vals += ncols; 1572 } 1573 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1574 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1575 1576 ptap->reuse = MAT_REUSE_MATRIX; 1577 1578 /* supporting struct ptap consumes almost same amount of memory as C=PtAP, release it if C will not be updated by A and P */ 1579 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatFreeIntermediateDataStructures","Mat");CHKERRQ(ierr); 1580 freestruct = PETSC_FALSE; 1581 ierr = PetscOptionsBool("-mat_freeintermediatedatastructures","Free intermediate data structures", "MatFreeIntermediateDataStructures",freestruct, &freestruct, NULL);CHKERRQ(ierr); 1582 if (freestruct) { 1583 ierr = MatFreeIntermediateDataStructures(C);CHKERRQ(ierr); 1584 } 1585 ierr = PetscOptionsEnd();CHKERRQ(ierr); 1586 PetscFunctionReturn(0); 1587 } 1588 1589 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ(Mat P,Mat A,Mat C) 1590 { 1591 PetscErrorCode ierr; 1592 Mat_Merge_SeqsToMPI *merge; 1593 Mat_MPIAIJ *p =(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 1594 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 1595 Mat_APMPI *ptap; 1596 PetscInt *adj; 1597 PetscInt i,j,k,anz,pnz,row,*cj,nexta; 1598 MatScalar *ada,*ca,valtmp; 1599 PetscInt am =A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 1600 MPI_Comm comm; 1601 PetscMPIInt size,rank,taga,*len_s; 1602 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 1603 PetscInt **buf_ri,**buf_rj; 1604 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 1605 MPI_Request *s_waits,*r_waits; 1606 MPI_Status *status; 1607 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 1608 PetscInt *ai,*aj,*coi,*coj,*poJ,*pdJ; 1609 Mat A_loc; 1610 Mat_SeqAIJ *a_loc; 1611 PetscBool freestruct; 1612 1613 PetscFunctionBegin; 1614 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1615 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1616 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1617 1618 ptap = c->ap; 1619 if (!ptap) SETERRQ(comm,PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatFreeIntermediateDataStructures() or use '-mat_freeintermediatedatastructures'"); 1620 merge = ptap->merge; 1621 1622 /* 2) compute numeric C_seq = P_loc^T*A_loc */ 1623 /*------------------------------------------*/ 1624 /* get data from symbolic products */ 1625 coi = merge->coi; coj = merge->coj; 1626 ierr = PetscCalloc1(coi[pon]+1,&coa);CHKERRQ(ierr); 1627 bi = merge->bi; bj = merge->bj; 1628 owners = merge->rowmap->range; 1629 ierr = PetscCalloc1(bi[cm]+1,&ba);CHKERRQ(ierr); 1630 1631 /* get A_loc by taking all local rows of A */ 1632 A_loc = ptap->A_loc; 1633 ierr = MatMPIAIJGetLocalMat(A,MAT_REUSE_MATRIX,&A_loc);CHKERRQ(ierr); 1634 a_loc = (Mat_SeqAIJ*)(A_loc)->data; 1635 ai = a_loc->i; 1636 aj = a_loc->j; 1637 1638 for (i=0; i<am; i++) { 1639 anz = ai[i+1] - ai[i]; 1640 adj = aj + ai[i]; 1641 ada = a_loc->a + ai[i]; 1642 1643 /* 2-b) Compute Cseq = P_loc[i,:]^T*A[i,:] using outer product */ 1644 /*-------------------------------------------------------------*/ 1645 /* put the value into Co=(p->B)^T*A (off-diagonal part, send to others) */ 1646 pnz = po->i[i+1] - po->i[i]; 1647 poJ = po->j + po->i[i]; 1648 pA = po->a + po->i[i]; 1649 for (j=0; j<pnz; j++) { 1650 row = poJ[j]; 1651 cj = coj + coi[row]; 1652 ca = coa + coi[row]; 1653 /* perform sparse axpy */ 1654 nexta = 0; 1655 valtmp = pA[j]; 1656 for (k=0; nexta<anz; k++) { 1657 if (cj[k] == adj[nexta]) { 1658 ca[k] += valtmp*ada[nexta]; 1659 nexta++; 1660 } 1661 } 1662 ierr = PetscLogFlops(2.0*anz);CHKERRQ(ierr); 1663 } 1664 1665 /* put the value into Cd (diagonal part) */ 1666 pnz = pd->i[i+1] - pd->i[i]; 1667 pdJ = pd->j + pd->i[i]; 1668 pA = pd->a + pd->i[i]; 1669 for (j=0; j<pnz; j++) { 1670 row = pdJ[j]; 1671 cj = bj + bi[row]; 1672 ca = ba + bi[row]; 1673 /* perform sparse axpy */ 1674 nexta = 0; 1675 valtmp = pA[j]; 1676 for (k=0; nexta<anz; k++) { 1677 if (cj[k] == adj[nexta]) { 1678 ca[k] += valtmp*ada[nexta]; 1679 nexta++; 1680 } 1681 } 1682 ierr = PetscLogFlops(2.0*anz);CHKERRQ(ierr); 1683 } 1684 } 1685 1686 /* 3) send and recv matrix values coa */ 1687 /*------------------------------------*/ 1688 buf_ri = merge->buf_ri; 1689 buf_rj = merge->buf_rj; 1690 len_s = merge->len_s; 1691 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 1692 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 1693 1694 ierr = PetscMalloc2(merge->nsend+1,&s_waits,size,&status);CHKERRQ(ierr); 1695 for (proc=0,k=0; proc<size; proc++) { 1696 if (!len_s[proc]) continue; 1697 i = merge->owners_co[proc]; 1698 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 1699 k++; 1700 } 1701 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 1702 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 1703 1704 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 1705 ierr = PetscFree(r_waits);CHKERRQ(ierr); 1706 ierr = PetscFree(coa);CHKERRQ(ierr); 1707 1708 /* 4) insert local Cseq and received values into Cmpi */ 1709 /*----------------------------------------------------*/ 1710 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 1711 for (k=0; k<merge->nrecv; k++) { 1712 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1713 nrows = *(buf_ri_k[k]); 1714 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 1715 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1716 } 1717 1718 for (i=0; i<cm; i++) { 1719 row = owners[rank] + i; /* global row index of C_seq */ 1720 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 1721 ba_i = ba + bi[i]; 1722 bnz = bi[i+1] - bi[i]; 1723 /* add received vals into ba */ 1724 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 1725 /* i-th row */ 1726 if (i == *nextrow[k]) { 1727 cnz = *(nextci[k]+1) - *nextci[k]; 1728 cj = buf_rj[k] + *(nextci[k]); 1729 ca = abuf_r[k] + *(nextci[k]); 1730 nextcj = 0; 1731 for (j=0; nextcj<cnz; j++) { 1732 if (bj_i[j] == cj[nextcj]) { /* bcol == ccol */ 1733 ba_i[j] += ca[nextcj++]; 1734 } 1735 } 1736 nextrow[k]++; nextci[k]++; 1737 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 1738 } 1739 } 1740 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 1741 } 1742 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1743 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1744 1745 ierr = PetscFree(ba);CHKERRQ(ierr); 1746 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 1747 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 1748 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1749 1750 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatFreeIntermediateDataStructures","Mat");CHKERRQ(ierr); 1751 freestruct = PETSC_FALSE; 1752 ierr = PetscOptionsBool("-mat_freeintermediatedatastructures","Free intermediate data structures", "MatFreeIntermediateDataStructures",freestruct, &freestruct, NULL);CHKERRQ(ierr); 1753 if (freestruct) { 1754 ierr = MatFreeIntermediateDataStructures(C);CHKERRQ(ierr); 1755 } 1756 ierr = PetscOptionsEnd();CHKERRQ(ierr); 1757 PetscFunctionReturn(0); 1758 } 1759 1760 PetscErrorCode MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ(Mat P,Mat A,PetscReal fill,Mat *C) 1761 { 1762 PetscErrorCode ierr; 1763 Mat Cmpi,A_loc,POt,PDt; 1764 Mat_APMPI *ptap; 1765 PetscFreeSpaceList free_space=NULL,current_space=NULL; 1766 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data,*a=(Mat_MPIAIJ*)A->data,*c; 1767 PetscInt *pdti,*pdtj,*poti,*potj,*ptJ; 1768 PetscInt nnz; 1769 PetscInt *lnk,*owners_co,*coi,*coj,i,k,pnz,row; 1770 PetscInt am =A->rmap->n,pn=P->cmap->n; 1771 MPI_Comm comm; 1772 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri; 1773 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 1774 PetscInt len,proc,*dnz,*onz,*owners; 1775 PetscInt nzi,*bi,*bj; 1776 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 1777 MPI_Request *swaits,*rwaits; 1778 MPI_Status *sstatus,rstatus; 1779 Mat_Merge_SeqsToMPI *merge; 1780 PetscInt *ai,*aj,*Jptr,anz,*prmap=p->garray,pon,nspacedouble=0,j; 1781 PetscReal afill =1.0,afill_tmp; 1782 PetscInt rstart = P->cmap->rstart,rmax,aN=A->cmap->N,Armax; 1783 PetscScalar *vals; 1784 Mat_SeqAIJ *a_loc,*pdt,*pot; 1785 PetscTable ta; 1786 MatType mtype; 1787 1788 PetscFunctionBegin; 1789 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1790 /* check if matrix local sizes are compatible */ 1791 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend) SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, A (%D, %D) != P (%D,%D)",A->rmap->rstart,A->rmap->rend,P->rmap->rstart,P->rmap->rend); 1792 1793 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1794 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1795 1796 /* create struct Mat_APMPI and attached it to C later */ 1797 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1798 1799 /* get A_loc by taking all local rows of A */ 1800 ierr = MatMPIAIJGetLocalMat(A,MAT_INITIAL_MATRIX,&A_loc);CHKERRQ(ierr); 1801 1802 ptap->A_loc = A_loc; 1803 a_loc = (Mat_SeqAIJ*)(A_loc)->data; 1804 ai = a_loc->i; 1805 aj = a_loc->j; 1806 1807 /* determine symbolic Co=(p->B)^T*A - send to others */ 1808 /*----------------------------------------------------*/ 1809 ierr = MatTransposeSymbolic_SeqAIJ(p->A,&PDt);CHKERRQ(ierr); 1810 pdt = (Mat_SeqAIJ*)PDt->data; 1811 pdti = pdt->i; pdtj = pdt->j; 1812 1813 ierr = MatTransposeSymbolic_SeqAIJ(p->B,&POt);CHKERRQ(ierr); 1814 pot = (Mat_SeqAIJ*)POt->data; 1815 poti = pot->i; potj = pot->j; 1816 1817 /* then, compute symbolic Co = (p->B)^T*A */ 1818 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 1819 >= (num of nonzero rows of C_seq) - pn */ 1820 ierr = PetscMalloc1(pon+1,&coi);CHKERRQ(ierr); 1821 coi[0] = 0; 1822 1823 /* set initial free space to be fill*(nnz(p->B) + nnz(A)) */ 1824 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(poti[pon],ai[am])); 1825 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1826 current_space = free_space; 1827 1828 /* create and initialize a linked list */ 1829 ierr = PetscTableCreate(A->cmap->n + a->B->cmap->N,aN,&ta);CHKERRQ(ierr); 1830 MatRowMergeMax_SeqAIJ(a_loc,am,ta); 1831 ierr = PetscTableGetCount(ta,&Armax);CHKERRQ(ierr); 1832 1833 ierr = PetscLLCondensedCreate_Scalable(Armax,&lnk);CHKERRQ(ierr); 1834 1835 for (i=0; i<pon; i++) { 1836 pnz = poti[i+1] - poti[i]; 1837 ptJ = potj + poti[i]; 1838 for (j=0; j<pnz; j++) { 1839 row = ptJ[j]; /* row of A_loc == col of Pot */ 1840 anz = ai[row+1] - ai[row]; 1841 Jptr = aj + ai[row]; 1842 /* add non-zero cols of AP into the sorted linked list lnk */ 1843 ierr = PetscLLCondensedAddSorted_Scalable(anz,Jptr,lnk);CHKERRQ(ierr); 1844 } 1845 nnz = lnk[0]; 1846 1847 /* If free space is not available, double the total space in the list */ 1848 if (current_space->local_remaining<nnz) { 1849 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1850 nspacedouble++; 1851 } 1852 1853 /* Copy data into free space, and zero out denserows */ 1854 ierr = PetscLLCondensedClean_Scalable(nnz,current_space->array,lnk);CHKERRQ(ierr); 1855 1856 current_space->array += nnz; 1857 current_space->local_used += nnz; 1858 current_space->local_remaining -= nnz; 1859 1860 coi[i+1] = coi[i] + nnz; 1861 } 1862 1863 ierr = PetscMalloc1(coi[pon]+1,&coj);CHKERRQ(ierr); 1864 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 1865 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); /* must destroy to get a new one for C */ 1866 1867 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + ai[am]+1); 1868 if (afill_tmp > afill) afill = afill_tmp; 1869 1870 /* send j-array (coj) of Co to other processors */ 1871 /*----------------------------------------------*/ 1872 /* determine row ownership */ 1873 ierr = PetscNew(&merge);CHKERRQ(ierr); 1874 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 1875 1876 merge->rowmap->n = pn; 1877 merge->rowmap->bs = 1; 1878 1879 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 1880 owners = merge->rowmap->range; 1881 1882 /* determine the number of messages to send, their lengths */ 1883 ierr = PetscCalloc1(size,&len_si);CHKERRQ(ierr); 1884 ierr = PetscMalloc1(size,&merge->len_s);CHKERRQ(ierr); 1885 1886 len_s = merge->len_s; 1887 merge->nsend = 0; 1888 1889 ierr = PetscMalloc1(size+2,&owners_co);CHKERRQ(ierr); 1890 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 1891 1892 proc = 0; 1893 for (i=0; i<pon; i++) { 1894 while (prmap[i] >= owners[proc+1]) proc++; 1895 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 1896 len_s[proc] += coi[i+1] - coi[i]; 1897 } 1898 1899 len = 0; /* max length of buf_si[] */ 1900 owners_co[0] = 0; 1901 for (proc=0; proc<size; proc++) { 1902 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 1903 if (len_si[proc]) { 1904 merge->nsend++; 1905 len_si[proc] = 2*(len_si[proc] + 1); 1906 len += len_si[proc]; 1907 } 1908 } 1909 1910 /* determine the number and length of messages to receive for coi and coj */ 1911 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 1912 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 1913 1914 /* post the Irecv and Isend of coj */ 1915 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 1916 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 1917 ierr = PetscMalloc1(merge->nsend+1,&swaits);CHKERRQ(ierr); 1918 for (proc=0, k=0; proc<size; proc++) { 1919 if (!len_s[proc]) continue; 1920 i = owners_co[proc]; 1921 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 1922 k++; 1923 } 1924 1925 /* receives and sends of coj are complete */ 1926 ierr = PetscMalloc1(size,&sstatus);CHKERRQ(ierr); 1927 for (i=0; i<merge->nrecv; i++) { 1928 PetscMPIInt icompleted; 1929 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1930 } 1931 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1932 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1933 1934 /* add received column indices into table to update Armax */ 1935 /* Armax can be as large as aN if a P[row,:] is dense, see src/ksp/ksp/examples/tutorials/ex56.c! */ 1936 for (k=0; k<merge->nrecv; k++) {/* k-th received message */ 1937 Jptr = buf_rj[k]; 1938 for (j=0; j<merge->len_r[k]; j++) { 1939 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 1940 } 1941 } 1942 ierr = PetscTableGetCount(ta,&Armax);CHKERRQ(ierr); 1943 /* printf("Armax %d, an %d + Bn %d = %d, aN %d\n",Armax,A->cmap->n,a->B->cmap->N,A->cmap->n+a->B->cmap->N,aN); */ 1944 1945 /* send and recv coi */ 1946 /*-------------------*/ 1947 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 1948 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 1949 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 1950 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 1951 for (proc=0,k=0; proc<size; proc++) { 1952 if (!len_s[proc]) continue; 1953 /* form outgoing message for i-structure: 1954 buf_si[0]: nrows to be sent 1955 [1:nrows]: row index (global) 1956 [nrows+1:2*nrows+1]: i-structure index 1957 */ 1958 /*-------------------------------------------*/ 1959 nrows = len_si[proc]/2 - 1; 1960 buf_si_i = buf_si + nrows+1; 1961 buf_si[0] = nrows; 1962 buf_si_i[0] = 0; 1963 nrows = 0; 1964 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 1965 nzi = coi[i+1] - coi[i]; 1966 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 1967 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 1968 nrows++; 1969 } 1970 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 1971 k++; 1972 buf_si += len_si[proc]; 1973 } 1974 i = merge->nrecv; 1975 while (i--) { 1976 PetscMPIInt icompleted; 1977 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1978 } 1979 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1980 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1981 ierr = PetscFree(len_si);CHKERRQ(ierr); 1982 ierr = PetscFree(len_ri);CHKERRQ(ierr); 1983 ierr = PetscFree(swaits);CHKERRQ(ierr); 1984 ierr = PetscFree(sstatus);CHKERRQ(ierr); 1985 ierr = PetscFree(buf_s);CHKERRQ(ierr); 1986 1987 /* compute the local portion of C (mpi mat) */ 1988 /*------------------------------------------*/ 1989 /* allocate bi array and free space for accumulating nonzero column info */ 1990 ierr = PetscMalloc1(pn+1,&bi);CHKERRQ(ierr); 1991 bi[0] = 0; 1992 1993 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 1994 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(pdti[pn],PetscIntSumTruncate(poti[pon],ai[am]))); 1995 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1996 current_space = free_space; 1997 1998 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 1999 for (k=0; k<merge->nrecv; k++) { 2000 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 2001 nrows = *buf_ri_k[k]; 2002 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 2003 nextci[k] = buf_ri_k[k] + (nrows + 1); /* points to the next i-structure of k-th recieved i-structure */ 2004 } 2005 2006 ierr = PetscLLCondensedCreate_Scalable(Armax,&lnk);CHKERRQ(ierr); 2007 ierr = MatPreallocateInitialize(comm,pn,A->cmap->n,dnz,onz);CHKERRQ(ierr); 2008 rmax = 0; 2009 for (i=0; i<pn; i++) { 2010 /* add pdt[i,:]*AP into lnk */ 2011 pnz = pdti[i+1] - pdti[i]; 2012 ptJ = pdtj + pdti[i]; 2013 for (j=0; j<pnz; j++) { 2014 row = ptJ[j]; /* row of AP == col of Pt */ 2015 anz = ai[row+1] - ai[row]; 2016 Jptr = aj + ai[row]; 2017 /* add non-zero cols of AP into the sorted linked list lnk */ 2018 ierr = PetscLLCondensedAddSorted_Scalable(anz,Jptr,lnk);CHKERRQ(ierr); 2019 } 2020 2021 /* add received col data into lnk */ 2022 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 2023 if (i == *nextrow[k]) { /* i-th row */ 2024 nzi = *(nextci[k]+1) - *nextci[k]; 2025 Jptr = buf_rj[k] + *nextci[k]; 2026 ierr = PetscLLCondensedAddSorted_Scalable(nzi,Jptr,lnk);CHKERRQ(ierr); 2027 nextrow[k]++; nextci[k]++; 2028 } 2029 } 2030 nnz = lnk[0]; 2031 2032 /* if free space is not available, make more free space */ 2033 if (current_space->local_remaining<nnz) { 2034 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 2035 nspacedouble++; 2036 } 2037 /* copy data into free space, then initialize lnk */ 2038 ierr = PetscLLCondensedClean_Scalable(nnz,current_space->array,lnk);CHKERRQ(ierr); 2039 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 2040 2041 current_space->array += nnz; 2042 current_space->local_used += nnz; 2043 current_space->local_remaining -= nnz; 2044 2045 bi[i+1] = bi[i] + nnz; 2046 if (nnz > rmax) rmax = nnz; 2047 } 2048 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 2049 2050 ierr = PetscMalloc1(bi[pn]+1,&bj);CHKERRQ(ierr); 2051 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 2052 afill_tmp = (PetscReal)bi[pn]/(pdti[pn] + poti[pon] + ai[am]+1); 2053 if (afill_tmp > afill) afill = afill_tmp; 2054 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 2055 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 2056 2057 ierr = MatDestroy(&POt);CHKERRQ(ierr); 2058 ierr = MatDestroy(&PDt);CHKERRQ(ierr); 2059 2060 /* create symbolic parallel matrix Cmpi - why cannot be assembled in Numeric part */ 2061 /*----------------------------------------------------------------------------------*/ 2062 ierr = PetscCalloc1(rmax+1,&vals);CHKERRQ(ierr); 2063 2064 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 2065 ierr = MatSetSizes(Cmpi,pn,A->cmap->n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2066 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(A->cmap->bs));CHKERRQ(ierr); 2067 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 2068 ierr = MatSetType(Cmpi,mtype);CHKERRQ(ierr); 2069 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 2070 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 2071 ierr = MatSetBlockSize(Cmpi,1);CHKERRQ(ierr); 2072 for (i=0; i<pn; i++) { 2073 row = i + rstart; 2074 nnz = bi[i+1] - bi[i]; 2075 Jptr = bj + bi[i]; 2076 ierr = MatSetValues(Cmpi,1,&row,nnz,Jptr,vals,INSERT_VALUES);CHKERRQ(ierr); 2077 } 2078 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2079 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2080 ierr = PetscFree(vals);CHKERRQ(ierr); 2081 2082 merge->bi = bi; 2083 merge->bj = bj; 2084 merge->coi = coi; 2085 merge->coj = coj; 2086 merge->buf_ri = buf_ri; 2087 merge->buf_rj = buf_rj; 2088 merge->owners_co = owners_co; 2089 2090 /* attach the supporting struct to Cmpi for reuse */ 2091 c = (Mat_MPIAIJ*)Cmpi->data; 2092 2093 c->ap = ptap; 2094 ptap->api = NULL; 2095 ptap->apj = NULL; 2096 ptap->merge = merge; 2097 ptap->apa = NULL; 2098 ptap->destroy = Cmpi->ops->destroy; 2099 ptap->duplicate = Cmpi->ops->duplicate; 2100 2101 Cmpi->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ; 2102 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 2103 Cmpi->ops->freeintermediatedatastructures = MatFreeIntermediateDataStructures_MPIAIJ_AP; 2104 2105 *C = Cmpi; 2106 #if defined(PETSC_USE_INFO) 2107 if (bi[pn] != 0) { 2108 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 2109 ierr = PetscInfo1(Cmpi,"Use MatTransposeMatMult(A,B,MatReuse,%g,&C) for best performance.\n",(double)afill);CHKERRQ(ierr); 2110 } else { 2111 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 2112 } 2113 #endif 2114 PetscFunctionReturn(0); 2115 } 2116