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 #include <petsc/private/vecscatterimpl.h> 13 14 #if defined(PETSC_HAVE_HYPRE) 15 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat); 16 #endif 17 18 PETSC_INTERN PetscErrorCode MatProductSymbolic_AB_MPIAIJ_MPIAIJ(Mat C) 19 { 20 PetscErrorCode ierr; 21 Mat_Product *product = C->product; 22 Mat A=product->A,B=product->B; 23 MatProductAlgorithm alg=product->alg; 24 PetscReal fill=product->fill; 25 PetscBool flg; 26 27 PetscFunctionBegin; 28 /* scalable */ 29 ierr = PetscStrcmp(alg,"scalable",&flg);CHKERRQ(ierr); 30 if (flg) { 31 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 32 PetscFunctionReturn(0); 33 } 34 35 /* nonscalable */ 36 ierr = PetscStrcmp(alg,"nonscalable",&flg);CHKERRQ(ierr); 37 if (flg) { 38 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(A,B,fill,C);CHKERRQ(ierr); 39 PetscFunctionReturn(0); 40 } 41 42 /* seqmpi */ 43 ierr = PetscStrcmp(alg,"seqmpi",&flg);CHKERRQ(ierr); 44 if (flg) { 45 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ_seqMPI(A,B,fill,C);CHKERRQ(ierr); 46 PetscFunctionReturn(0); 47 } 48 49 #if defined(PETSC_HAVE_HYPRE) 50 ierr = PetscStrcmp(alg,"hypre",&flg);CHKERRQ(ierr); 51 if (flg) { 52 ierr = MatMatMultSymbolic_AIJ_AIJ_wHYPRE(A,B,fill,C);CHKERRQ(ierr); 53 PetscFunctionReturn(0); 54 } 55 #endif 56 SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_SUP,"Mat Product Algorithm is not supported"); 57 } 58 59 PetscErrorCode MatDestroy_MPIAIJ_MatMatMult(void *data) 60 { 61 PetscErrorCode ierr; 62 Mat_APMPI *ptap = (Mat_APMPI*)data; 63 64 PetscFunctionBegin; 65 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 66 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 67 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 68 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 69 ierr = MatDestroy(&ptap->Pt);CHKERRQ(ierr); 70 ierr = PetscFree(ptap->api);CHKERRQ(ierr); 71 ierr = PetscFree(ptap->apj);CHKERRQ(ierr); 72 ierr = PetscFree(ptap->apa);CHKERRQ(ierr); 73 ierr = PetscFree(ptap);CHKERRQ(ierr); 74 PetscFunctionReturn(0); 75 } 76 77 PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable(Mat A,Mat P,Mat C) 78 { 79 PetscErrorCode ierr; 80 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*c=(Mat_MPIAIJ*)C->data; 81 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 82 Mat_SeqAIJ *cd =(Mat_SeqAIJ*)(c->A)->data,*co=(Mat_SeqAIJ*)(c->B)->data; 83 PetscScalar *cda=cd->a,*coa=co->a; 84 Mat_SeqAIJ *p_loc,*p_oth; 85 PetscScalar *apa,*ca; 86 PetscInt cm =C->rmap->n; 87 Mat_APMPI *ptap; 88 PetscInt *api,*apj,*apJ,i,k; 89 PetscInt cstart=C->cmap->rstart; 90 PetscInt cdnz,conz,k0,k1; 91 MPI_Comm comm; 92 PetscMPIInt size; 93 94 PetscFunctionBegin; 95 MatCheckProduct(C,3); 96 ptap = (Mat_APMPI*)C->product->data; 97 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtAP cannot be computed. Missing data"); 98 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 99 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 100 101 if (!ptap->P_oth && size>1) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"AP cannot be reused. Do not call MatProductClear()"); 102 103 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 104 /*-----------------------------------------------------*/ 105 /* update numerical values of P_oth and P_loc */ 106 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 107 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 108 109 /* 2) compute numeric C_loc = A_loc*P = Ad*P_loc + Ao*P_oth */ 110 /*----------------------------------------------------------*/ 111 /* get data from symbolic products */ 112 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 113 p_oth = NULL; 114 if (size >1) { 115 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 116 } 117 118 /* get apa for storing dense row A[i,:]*P */ 119 apa = ptap->apa; 120 121 api = ptap->api; 122 apj = ptap->apj; 123 for (i=0; i<cm; i++) { 124 /* compute apa = A[i,:]*P */ 125 AProw_nonscalable(i,ad,ao,p_loc,p_oth,apa); 126 127 /* set values in C */ 128 apJ = apj + api[i]; 129 cdnz = cd->i[i+1] - cd->i[i]; 130 conz = co->i[i+1] - co->i[i]; 131 132 /* 1st off-diagonal part of C */ 133 ca = coa + co->i[i]; 134 k = 0; 135 for (k0=0; k0<conz; k0++) { 136 if (apJ[k] >= cstart) break; 137 ca[k0] = apa[apJ[k]]; 138 apa[apJ[k++]] = 0.0; 139 } 140 141 /* diagonal part of C */ 142 ca = cda + cd->i[i]; 143 for (k1=0; k1<cdnz; k1++) { 144 ca[k1] = apa[apJ[k]]; 145 apa[apJ[k++]] = 0.0; 146 } 147 148 /* 2nd off-diagonal part of C */ 149 ca = coa + co->i[i]; 150 for (; k0<conz; k0++) { 151 ca[k0] = apa[apJ[k]]; 152 apa[apJ[k++]] = 0.0; 153 } 154 } 155 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 156 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 157 PetscFunctionReturn(0); 158 } 159 160 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(Mat A,Mat P,PetscReal fill,Mat C) 161 { 162 PetscErrorCode ierr; 163 MPI_Comm comm; 164 PetscMPIInt size; 165 Mat_APMPI *ptap; 166 PetscFreeSpaceList free_space=NULL,current_space=NULL; 167 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 168 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc,*p_oth; 169 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*dnz,*onz; 170 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,*aoj=ao->j,rstart=A->rmap->rstart; 171 PetscInt *lnk,i,pnz,row,*api,*apj,*Jptr,apnz,nspacedouble=0,j,nzi; 172 PetscInt am=A->rmap->n,pN=P->cmap->N,pn=P->cmap->n,pm=P->rmap->n; 173 PetscBT lnkbt; 174 PetscReal afill; 175 MatType mtype; 176 177 PetscFunctionBegin; 178 MatCheckProduct(C,4); 179 if (C->product->data) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 180 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 181 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 182 183 /* create struct Mat_APMPI and attached it to C later */ 184 ierr = PetscNew(&ptap);CHKERRQ(ierr); 185 186 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 187 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 188 189 /* get P_loc by taking all local rows of P */ 190 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 191 192 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 193 pi_loc = p_loc->i; pj_loc = p_loc->j; 194 if (size > 1) { 195 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 196 pi_oth = p_oth->i; pj_oth = p_oth->j; 197 } else { 198 p_oth = NULL; 199 pi_oth = NULL; pj_oth = NULL; 200 } 201 202 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 203 /*-------------------------------------------------------------------*/ 204 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 205 ptap->api = api; 206 api[0] = 0; 207 208 /* create and initialize a linked list */ 209 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 210 211 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 212 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space);CHKERRQ(ierr); 213 current_space = free_space; 214 215 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 216 for (i=0; i<am; i++) { 217 /* diagonal portion of A */ 218 nzi = adi[i+1] - adi[i]; 219 for (j=0; j<nzi; j++) { 220 row = *adj++; 221 pnz = pi_loc[row+1] - pi_loc[row]; 222 Jptr = pj_loc + pi_loc[row]; 223 /* add non-zero cols of P into the sorted linked list lnk */ 224 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 225 } 226 /* off-diagonal portion of A */ 227 nzi = aoi[i+1] - aoi[i]; 228 for (j=0; j<nzi; j++) { 229 row = *aoj++; 230 pnz = pi_oth[row+1] - pi_oth[row]; 231 Jptr = pj_oth + pi_oth[row]; 232 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 233 } 234 235 apnz = lnk[0]; 236 api[i+1] = api[i] + apnz; 237 238 /* if free space is not available, double the total space in the list */ 239 if (current_space->local_remaining<apnz) { 240 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 241 nspacedouble++; 242 } 243 244 /* Copy data into free space, then initialize lnk */ 245 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 246 ierr = MatPreallocateSet(i+rstart,apnz,current_space->array,dnz,onz);CHKERRQ(ierr); 247 248 current_space->array += apnz; 249 current_space->local_used += apnz; 250 current_space->local_remaining -= apnz; 251 } 252 253 /* Allocate space for apj, initialize apj, and */ 254 /* destroy list of free space and other temporary array(s) */ 255 ierr = PetscMalloc1(api[am]+1,&ptap->apj);CHKERRQ(ierr); 256 apj = ptap->apj; 257 ierr = PetscFreeSpaceContiguous(&free_space,ptap->apj);CHKERRQ(ierr); 258 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 259 260 /* malloc apa to store dense row A[i,:]*P */ 261 ierr = PetscCalloc1(pN,&ptap->apa);CHKERRQ(ierr); 262 263 /* set and assemble symbolic parallel matrix C */ 264 /*---------------------------------------------*/ 265 ierr = MatSetSizes(C,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 266 ierr = MatSetBlockSizesFromMats(C,A,P);CHKERRQ(ierr); 267 268 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 269 ierr = MatSetType(C,mtype);CHKERRQ(ierr); 270 ierr = MatMPIAIJSetPreallocation(C,0,dnz,0,onz);CHKERRQ(ierr); 271 272 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(C, apj, api);CHKERRQ(ierr); 273 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 274 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 275 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 276 277 C->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 278 C->ops->productnumeric = MatProductNumeric_AB; 279 280 /* attach the supporting struct to C for reuse */ 281 C->product->data = ptap; 282 C->product->destroy = MatDestroy_MPIAIJ_MatMatMult; 283 284 /* set MatInfo */ 285 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 286 if (afill < 1.0) afill = 1.0; 287 C->info.mallocs = nspacedouble; 288 C->info.fill_ratio_given = fill; 289 C->info.fill_ratio_needed = afill; 290 291 #if defined(PETSC_USE_INFO) 292 if (api[am]) { 293 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 294 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 295 } else { 296 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 297 } 298 #endif 299 PetscFunctionReturn(0); 300 } 301 302 /* ------------------------------------------------------- */ 303 static PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIDense(Mat,Mat,PetscReal,Mat); 304 static PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIDense(Mat,Mat,Mat); 305 306 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_MPIDense_AB(Mat C) 307 { 308 Mat_Product *product = C->product; 309 Mat A = product->A,B=product->B; 310 311 PetscFunctionBegin; 312 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) 313 SETERRQ4(PETSC_COMM_SELF,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); 314 315 C->ops->matmultsymbolic = MatMatMultSymbolic_MPIAIJ_MPIDense; 316 C->ops->productsymbolic = MatProductSymbolic_AB; 317 PetscFunctionReturn(0); 318 } 319 /* -------------------------------------------------------------------- */ 320 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_MPIDense_AtB(Mat C) 321 { 322 Mat_Product *product = C->product; 323 Mat A = product->A,B=product->B; 324 325 PetscFunctionBegin; 326 if (A->rmap->rstart != B->rmap->rstart || A->rmap->rend != B->rmap->rend) 327 SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, (%D, %D) != (%D,%D)",A->rmap->rstart,A->rmap->rend,B->rmap->rstart,B->rmap->rend); 328 329 C->ops->transposematmultsymbolic = MatTransposeMatMultSymbolic_MPIAIJ_MPIDense; 330 C->ops->productsymbolic = MatProductSymbolic_AtB; 331 PetscFunctionReturn(0); 332 } 333 334 /* --------------------------------------------------------------------- */ 335 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_MPIAIJ_MPIDense(Mat C) 336 { 337 PetscErrorCode ierr; 338 Mat_Product *product = C->product; 339 340 PetscFunctionBegin; 341 switch (product->type) { 342 case MATPRODUCT_AB: 343 ierr = MatProductSetFromOptions_MPIAIJ_MPIDense_AB(C);CHKERRQ(ierr); 344 break; 345 case MATPRODUCT_AtB: 346 ierr = MatProductSetFromOptions_MPIAIJ_MPIDense_AtB(C);CHKERRQ(ierr); 347 break; 348 default: 349 break; 350 } 351 PetscFunctionReturn(0); 352 } 353 /* ------------------------------------------------------- */ 354 355 typedef struct { 356 Mat workB,Bb,Cb,workB1,Bb1,Cb1; 357 MPI_Request *rwaits,*swaits; 358 PetscInt numBb; /* num of Bb matrices */ 359 PetscInt nsends,nrecvs; 360 MPI_Datatype *stype,*rtype; 361 PetscInt blda; 362 } MPIAIJ_MPIDense; 363 364 PetscErrorCode MatMPIAIJ_MPIDenseDestroy(void *ctx) 365 { 366 MPIAIJ_MPIDense *contents = (MPIAIJ_MPIDense*)ctx; 367 PetscErrorCode ierr; 368 PetscInt i; 369 370 PetscFunctionBegin; 371 ierr = MatDestroy(&contents->workB);CHKERRQ(ierr); 372 373 if (contents->numBb) { 374 ierr = MatDestroy(&contents->Bb);CHKERRQ(ierr); 375 ierr = MatDestroy(&contents->Cb);CHKERRQ(ierr); 376 377 ierr = MatDestroy(&contents->workB1);CHKERRQ(ierr); 378 ierr = MatDestroy(&contents->Bb1);CHKERRQ(ierr); 379 ierr = MatDestroy(&contents->Cb1);CHKERRQ(ierr); 380 } 381 for (i=0; i<contents->nsends; i++) { 382 ierr = MPI_Type_free(&contents->stype[i]);CHKERRQ(ierr); 383 } 384 for (i=0; i<contents->nrecvs; i++) { 385 ierr = MPI_Type_free(&contents->rtype[i]);CHKERRQ(ierr); 386 } 387 ierr = PetscFree4(contents->stype,contents->rtype,contents->rwaits,contents->swaits);CHKERRQ(ierr); 388 ierr = PetscFree(contents);CHKERRQ(ierr); 389 PetscFunctionReturn(0); 390 } 391 392 /* 393 Create Bb, Cb, Bb1 and Cb1 matrices to be used by MatMatMultSymbolic_MPIAIJ_MPIDense(). 394 These matrices are used as wrappers for sub-columns of B and C, thus their own matrix operations are not used. 395 Modified from MatCreateDense(). 396 */ 397 PETSC_STATIC_INLINE PetscErrorCode MatCreateSubMPIDense_private(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt rbs,PetscInt cbs,PetscInt lda,PetscScalar *data,Mat *A) 398 { 399 PetscErrorCode ierr; 400 401 PetscFunctionBegin; 402 ierr = MatCreate(comm,A);CHKERRQ(ierr); 403 ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr); 404 ierr = MatSetBlockSizes(*A,rbs,cbs);CHKERRQ(ierr); 405 ierr = MatSetType(*A,MATMPIDENSE);CHKERRQ(ierr); 406 ierr = MatMPIDenseSetPreallocation(*A,data);CHKERRQ(ierr); 407 ierr = MatDenseSetLDA(*A,lda);CHKERRQ(ierr); 408 (*A)->assembled = PETSC_TRUE; 409 PetscFunctionReturn(0); 410 } 411 412 static PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIDense(Mat A,Mat B,PetscReal fill,Mat C) 413 { 414 PetscErrorCode ierr; 415 Mat_MPIAIJ *aij=(Mat_MPIAIJ*)A->data; 416 PetscInt nz=aij->B->cmap->n,nsends,nrecvs,i,nrows_to,j,blda,clda; 417 MPIAIJ_MPIDense *contents; 418 VecScatter ctx=aij->Mvctx; 419 PetscInt Am=A->rmap->n,Bm=B->rmap->n,BN=B->cmap->N,Bbn,Bbn1,bs,nrows_from; 420 MPI_Comm comm; 421 MPI_Datatype type1,*stype,*rtype; 422 const PetscInt *sindices,*sstarts,*rstarts; 423 PetscMPIInt *disp; 424 PetscBool cisdense; 425 426 PetscFunctionBegin; 427 MatCheckProduct(C,4); 428 if (C->product->data) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_PLIB,"Product data not empty"); 429 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 430 ierr = PetscObjectTypeCompareAny((PetscObject)C,&cisdense,MATMPIDENSE,MATMPIDENSECUDA,"");CHKERRQ(ierr); 431 if (!cisdense) { 432 ierr = MatSetType(C,((PetscObject)B)->type_name);CHKERRQ(ierr); 433 } 434 ierr = MatSetSizes(C,Am,B->cmap->n,A->rmap->N,BN);CHKERRQ(ierr); 435 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 436 ierr = MatSetUp(C);CHKERRQ(ierr); 437 ierr = MatDenseGetLDA(B,&blda);CHKERRQ(ierr); 438 ierr = MatDenseGetLDA(C,&clda);CHKERRQ(ierr); 439 ierr = PetscNew(&contents);CHKERRQ(ierr); 440 contents->numBb = 0; 441 442 ierr = VecScatterGetRemote_Private(ctx,PETSC_TRUE/*send*/,&nsends,&sstarts,&sindices,NULL,NULL);CHKERRQ(ierr); 443 ierr = VecScatterGetRemoteOrdered_Private(ctx,PETSC_FALSE/*recv*/,&nrecvs,&rstarts,NULL,NULL,NULL);CHKERRQ(ierr); 444 445 /* Create column block of B and C for memory scalability when BN is too large */ 446 /* Estimate Bbn, column size of Bb */ 447 if (nz) { 448 Bbn1 = 2*Am*BN/nz; 449 } else Bbn1 = BN; 450 451 bs = PetscAbs(B->cmap->bs); 452 Bbn1 = Bbn1/bs *bs; /* Bbn1 is a multiple of bs */ 453 if (Bbn1 > BN) Bbn1 = BN; 454 ierr = MPI_Allreduce(&Bbn1,&Bbn,1,MPIU_INT,MPI_MAX,comm);CHKERRQ(ierr); 455 456 /* Enable runtime option for Bbn */ 457 ierr = PetscOptionsBegin(comm,((PetscObject)C)->prefix,"MatMatMult","Mat");CHKERRQ(ierr); 458 ierr = PetscOptionsInt("-matmatmult_Bbn","Number of columns in Bb","MatMatMult",Bbn,&Bbn,NULL);CHKERRQ(ierr); 459 ierr = PetscOptionsEnd();CHKERRQ(ierr); 460 Bbn = PetscMin(Bbn,BN); 461 462 if (Bbn > 0 && Bbn < BN) { 463 contents->numBb = BN/Bbn; 464 Bbn1 = BN - contents->numBb*Bbn; 465 } 466 467 if (contents->numBb) { 468 PetscScalar data[1]; /* fake array for Bb and Cb */ 469 470 ierr = PetscInfo3(C,"use Bb, BN=%D, Bbn=%D; numBb=%D\n",BN,Bbn,contents->numBb);CHKERRQ(ierr); 471 ierr = MatCreateSubMPIDense_private(comm,B->rmap->n,PETSC_DECIDE,A->rmap->N,Bbn,B->rmap->bs,B->cmap->bs,blda,data,&contents->Bb);CHKERRQ(ierr); 472 ierr = MatCreateSubMPIDense_private(comm,Am,PETSC_DECIDE,A->rmap->N,Bbn,C->rmap->bs,C->cmap->bs,clda,data,&contents->Cb);CHKERRQ(ierr); 473 474 if (Bbn1) { /* Create Bb1 and Cb1 for the remaining columns */ 475 ierr = PetscInfo2(C,"use Bb1, BN=%D, Bbn1=%D\n",BN,Bbn1);CHKERRQ(ierr); 476 ierr = MatCreateSubMPIDense_private(comm,B->rmap->n,PETSC_DECIDE,A->rmap->N,Bbn1,B->rmap->bs,B->cmap->bs,blda,data,&contents->Bb1);CHKERRQ(ierr); 477 ierr = MatCreateSubMPIDense_private(comm,Am,PETSC_DECIDE,A->rmap->N,Bbn1,C->rmap->bs,C->cmap->bs,clda,data,&contents->Cb1);CHKERRQ(ierr); 478 479 /* Create work matrix used to store off processor rows of B needed for local product */ 480 ierr = MatCreateSeqDense(PETSC_COMM_SELF,nz,Bbn1,NULL,&contents->workB1);CHKERRQ(ierr); 481 } 482 } 483 484 /* Create work matrix used to store off processor rows of B needed for local product */ 485 ierr = MatCreateSeqDense(PETSC_COMM_SELF,nz,Bbn,NULL,&contents->workB);CHKERRQ(ierr); 486 487 /* Use MPI derived data type to reduce memory required by the send/recv buffers */ 488 ierr = PetscMalloc4(nsends,&stype,nrecvs,&rtype,nrecvs,&contents->rwaits,nsends,&contents->swaits);CHKERRQ(ierr); 489 contents->stype = stype; 490 contents->nsends = nsends; 491 492 contents->rtype = rtype; 493 contents->nrecvs = nrecvs; 494 contents->blda = blda; 495 496 ierr = PetscMalloc1(Bm+1,&disp);CHKERRQ(ierr); 497 for (i=0; i<nsends; i++) { 498 nrows_to = sstarts[i+1]-sstarts[i]; 499 for (j=0; j<nrows_to; j++){ 500 disp[j] = sindices[sstarts[i]+j]; /* rowB to be sent */ 501 } 502 ierr = MPI_Type_create_indexed_block(nrows_to,1,(const PetscMPIInt *)disp,MPIU_SCALAR,&type1);CHKERRQ(ierr); 503 504 ierr = MPI_Type_create_resized(type1,0,blda*sizeof(PetscScalar),&stype[i]);CHKERRQ(ierr); 505 ierr = MPI_Type_commit(&stype[i]);CHKERRQ(ierr); 506 ierr = MPI_Type_free(&type1);CHKERRQ(ierr); 507 } 508 509 for (i=0; i<nrecvs; i++) { 510 /* received values from a process form a (nrows_from x Bbn) row block in workB (column-wise) */ 511 nrows_from = rstarts[i+1]-rstarts[i]; 512 disp[0] = 0; 513 ierr = MPI_Type_create_indexed_block(1, nrows_from, (const PetscMPIInt *)disp, MPIU_SCALAR, &type1);CHKERRQ(ierr); 514 ierr = MPI_Type_create_resized(type1, 0, nz*sizeof(PetscScalar), &rtype[i]);CHKERRQ(ierr); 515 ierr = MPI_Type_commit(&rtype[i]);CHKERRQ(ierr); 516 ierr = MPI_Type_free(&type1);CHKERRQ(ierr); 517 } 518 519 ierr = PetscFree(disp);CHKERRQ(ierr); 520 ierr = VecScatterRestoreRemote_Private(ctx,PETSC_TRUE/*send*/,&nsends,&sstarts,&sindices,NULL,NULL);CHKERRQ(ierr); 521 ierr = VecScatterRestoreRemoteOrdered_Private(ctx,PETSC_FALSE/*recv*/,&nrecvs,&rstarts,NULL,NULL,NULL);CHKERRQ(ierr); 522 ierr = MatSetOption(C,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE);CHKERRQ(ierr); 523 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 524 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 525 526 C->product->data = contents; 527 C->product->destroy = MatMPIAIJ_MPIDenseDestroy; 528 C->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIDense; 529 PetscFunctionReturn(0); 530 } 531 532 PETSC_INTERN PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat,Mat,Mat); 533 /* 534 Performs an efficient scatter on the rows of B needed by this process; this is 535 a modification of the VecScatterBegin_() routines. 536 537 Input: Bbidx = 0: B = Bb 538 = 1: B = Bb1, see MatMatMultSymbolic_MPIAIJ_MPIDense() 539 */ 540 PetscErrorCode MatMPIDenseScatter(Mat A,Mat B,PetscInt Bbidx,Mat C,Mat *outworkB) 541 { 542 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 543 PetscErrorCode ierr; 544 const PetscScalar *b; 545 PetscScalar *rvalues; 546 VecScatter ctx = aij->Mvctx; 547 const PetscInt *sindices,*sstarts,*rstarts; 548 const PetscMPIInt *sprocs,*rprocs; 549 PetscInt i,nsends,nrecvs; 550 MPI_Request *swaits,*rwaits; 551 MPI_Comm comm; 552 PetscMPIInt tag=((PetscObject)ctx)->tag,ncols=B->cmap->N,nrows=aij->B->cmap->n,nsends_mpi,nrecvs_mpi; 553 MPIAIJ_MPIDense *contents; 554 Mat workB; 555 MPI_Datatype *stype,*rtype; 556 PetscInt blda; 557 558 PetscFunctionBegin; 559 MatCheckProduct(C,4); 560 if (!C->product->data) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_PLIB,"Product data empty"); 561 contents = (MPIAIJ_MPIDense*)C->product->data; 562 ierr = VecScatterGetRemote_Private(ctx,PETSC_TRUE/*send*/,&nsends,&sstarts,&sindices,&sprocs,NULL/*bs*/);CHKERRQ(ierr); 563 ierr = VecScatterGetRemoteOrdered_Private(ctx,PETSC_FALSE/*recv*/,&nrecvs,&rstarts,NULL,&rprocs,NULL/*bs*/);CHKERRQ(ierr); 564 ierr = PetscMPIIntCast(nsends,&nsends_mpi);CHKERRQ(ierr); 565 ierr = PetscMPIIntCast(nrecvs,&nrecvs_mpi);CHKERRQ(ierr); 566 if (Bbidx == 0) { 567 workB = *outworkB = contents->workB; 568 } else { 569 workB = *outworkB = contents->workB1; 570 } 571 if (nrows != workB->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Number of rows of workB %D not equal to columns of aij->B %D",workB->cmap->n,nrows); 572 swaits = contents->swaits; 573 rwaits = contents->rwaits; 574 575 ierr = MatDenseGetArrayRead(B,&b);CHKERRQ(ierr); 576 ierr = MatDenseGetLDA(B,&blda);CHKERRQ(ierr); 577 if (blda != contents->blda) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot reuse an input matrix with lda %D != %D",blda,contents->blda); 578 ierr = MatDenseGetArray(workB,&rvalues);CHKERRQ(ierr); 579 580 /* Post recv, use MPI derived data type to save memory */ 581 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 582 rtype = contents->rtype; 583 for (i=0; i<nrecvs; i++) { 584 ierr = MPI_Irecv(rvalues+(rstarts[i]-rstarts[0]),ncols,rtype[i],rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 585 } 586 587 stype = contents->stype; 588 for (i=0; i<nsends; i++) { 589 ierr = MPI_Isend(b,ncols,stype[i],sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 590 } 591 592 if (nrecvs) {ierr = MPI_Waitall(nrecvs_mpi,rwaits,MPI_STATUSES_IGNORE);CHKERRQ(ierr);} 593 if (nsends) {ierr = MPI_Waitall(nsends_mpi,swaits,MPI_STATUSES_IGNORE);CHKERRQ(ierr);} 594 595 ierr = VecScatterRestoreRemote_Private(ctx,PETSC_TRUE/*send*/,&nsends,&sstarts,&sindices,&sprocs,NULL);CHKERRQ(ierr); 596 ierr = VecScatterRestoreRemoteOrdered_Private(ctx,PETSC_FALSE/*recv*/,&nrecvs,&rstarts,NULL,&rprocs,NULL);CHKERRQ(ierr); 597 ierr = MatDenseRestoreArrayRead(B,&b);CHKERRQ(ierr); 598 ierr = MatDenseRestoreArray(workB,&rvalues);CHKERRQ(ierr); 599 ierr = MatAssemblyBegin(workB,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 600 ierr = MatAssemblyEnd(workB,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 601 PetscFunctionReturn(0); 602 } 603 604 /* 605 Compute Cb = A*Bb 606 */ 607 static PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIDense_private(Mat A,Mat Bb,PetscInt Bbidx,Mat C,const PetscScalar *barray,PetscScalar *carray,Mat Cb) 608 { 609 PetscErrorCode ierr; 610 Mat workB; 611 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 612 Mat_MPIDense *cbdense = (Mat_MPIDense*)Cb->data; 613 614 PetscFunctionBegin; 615 /* Place barray to Bb */ 616 ierr = MatDensePlaceArray(Bb,barray);CHKERRQ(ierr); 617 618 /* get off processor parts of Bb needed to complete Cb=A*Bb */ 619 ierr = MatMPIDenseScatter(A,Bb,Bbidx,C,&workB);CHKERRQ(ierr); 620 ierr = MatDenseResetArray(Bb);CHKERRQ(ierr); 621 622 /* off-diagonal block of A times nonlocal rows of Bb */ 623 /* Place carray to Cb */ 624 ierr = MatDensePlaceArray(Cb,carray);CHKERRQ(ierr); 625 ierr = MatMatMultNumericAdd_SeqAIJ_SeqDense(aij->B,workB,cbdense->A);CHKERRQ(ierr); 626 ierr = MatDenseResetArray(Cb);CHKERRQ(ierr); 627 PetscFunctionReturn(0); 628 } 629 630 static PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIDense(Mat A,Mat B,Mat C) 631 { 632 PetscErrorCode ierr; 633 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 634 Mat_MPIDense *bdense = (Mat_MPIDense*)B->data; 635 Mat_MPIDense *cdense = (Mat_MPIDense*)C->data; 636 Mat workB; 637 MPIAIJ_MPIDense *contents; 638 PetscInt numBb; 639 640 PetscFunctionBegin; 641 MatCheckProduct(C,3); 642 if (!C->product->data) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_PLIB,"Product data empty"); 643 contents = (MPIAIJ_MPIDense*)C->product->data; 644 /* diagonal block of A times all local rows of B*/ 645 ierr = MatMatMult(aij->A,bdense->A,MAT_REUSE_MATRIX,PETSC_DEFAULT,&cdense->A);CHKERRQ(ierr); 646 numBb = contents->numBb; 647 if (!numBb) { 648 /* get off processor parts of B needed to complete C=A*B */ 649 ierr = MatMPIDenseScatter(A,B,0,C,&workB);CHKERRQ(ierr); 650 651 /* off-diagonal block of A times nonlocal rows of B */ 652 ierr = MatMatMultNumericAdd_SeqAIJ_SeqDense(aij->B,workB,cdense->A);CHKERRQ(ierr); 653 } else { 654 const PetscScalar *barray,*bptr; 655 PetscScalar *carray,*cptr; 656 PetscInt BbN=contents->Bb->cmap->N,i,blda,clda; 657 658 ierr = MatDenseGetArrayRead(B,&barray);CHKERRQ(ierr); 659 ierr = MatDenseGetArray(C,&carray);CHKERRQ(ierr); 660 ierr = MatDenseGetLDA(B,&blda);CHKERRQ(ierr); 661 ierr = MatDenseGetLDA(C,&clda);CHKERRQ(ierr); 662 bptr = barray; 663 cptr = carray; 664 for (i=0; i<numBb; i++) { 665 ierr = MatMatMultNumeric_MPIAIJ_MPIDense_private(A,contents->Bb,0,C,bptr,cptr,contents->Cb);CHKERRQ(ierr); 666 bptr += (size_t)blda*(size_t)BbN; 667 cptr += (size_t)clda*(size_t)BbN; 668 } 669 670 if (contents->Bb1) { 671 ierr = MatMatMultNumeric_MPIAIJ_MPIDense_private(A,contents->Bb1,1,C,bptr,cptr,contents->Cb1);CHKERRQ(ierr); 672 } 673 ierr = MatDenseRestoreArrayRead(B,&barray);CHKERRQ(ierr); 674 ierr = MatDenseRestoreArray(C,&carray);CHKERRQ(ierr); 675 } 676 PetscFunctionReturn(0); 677 } 678 679 PetscErrorCode MatMatMultNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 680 { 681 PetscErrorCode ierr; 682 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data,*c=(Mat_MPIAIJ*)C->data; 683 Mat_SeqAIJ *ad = (Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 684 Mat_SeqAIJ *cd = (Mat_SeqAIJ*)(c->A)->data,*co=(Mat_SeqAIJ*)(c->B)->data; 685 PetscInt *adi = ad->i,*adj,*aoi=ao->i,*aoj; 686 PetscScalar *ada,*aoa,*cda=cd->a,*coa=co->a; 687 Mat_SeqAIJ *p_loc,*p_oth; 688 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pj; 689 PetscScalar *pa_loc,*pa_oth,*pa,valtmp,*ca; 690 PetscInt cm = C->rmap->n,anz,pnz; 691 Mat_APMPI *ptap; 692 PetscScalar *apa_sparse; 693 PetscInt *api,*apj,*apJ,i,j,k,row; 694 PetscInt cstart = C->cmap->rstart; 695 PetscInt cdnz,conz,k0,k1,nextp; 696 MPI_Comm comm; 697 PetscMPIInt size; 698 699 PetscFunctionBegin; 700 MatCheckProduct(C,3); 701 ptap = (Mat_APMPI*)C->product->data; 702 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtAP cannot be computed. Missing data"); 703 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 704 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 705 if (!ptap->P_oth && size>1) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"AP cannot be reused. Do not call MatProductClear()"); 706 707 apa_sparse = ptap->apa; 708 709 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 710 /*-----------------------------------------------------*/ 711 /* update numerical values of P_oth and P_loc */ 712 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 713 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 714 715 /* 2) compute numeric C_loc = A_loc*P = Ad*P_loc + Ao*P_oth */ 716 /*----------------------------------------------------------*/ 717 /* get data from symbolic products */ 718 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 719 pi_loc = p_loc->i; pj_loc = p_loc->j; pa_loc = p_loc->a; 720 if (size >1) { 721 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 722 pi_oth = p_oth->i; pj_oth = p_oth->j; pa_oth = p_oth->a; 723 } else { 724 p_oth = NULL; pi_oth = NULL; pj_oth = NULL; pa_oth = NULL; 725 } 726 727 api = ptap->api; 728 apj = ptap->apj; 729 for (i=0; i<cm; i++) { 730 apJ = apj + api[i]; 731 732 /* diagonal portion of A */ 733 anz = adi[i+1] - adi[i]; 734 adj = ad->j + adi[i]; 735 ada = ad->a + adi[i]; 736 for (j=0; j<anz; j++) { 737 row = adj[j]; 738 pnz = pi_loc[row+1] - pi_loc[row]; 739 pj = pj_loc + pi_loc[row]; 740 pa = pa_loc + pi_loc[row]; 741 /* perform sparse axpy */ 742 valtmp = ada[j]; 743 nextp = 0; 744 for (k=0; nextp<pnz; k++) { 745 if (apJ[k] == pj[nextp]) { /* column of AP == column of P */ 746 apa_sparse[k] += valtmp*pa[nextp++]; 747 } 748 } 749 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 750 } 751 752 /* off-diagonal portion of A */ 753 anz = aoi[i+1] - aoi[i]; 754 aoj = ao->j + aoi[i]; 755 aoa = ao->a + aoi[i]; 756 for (j=0; j<anz; j++) { 757 row = aoj[j]; 758 pnz = pi_oth[row+1] - pi_oth[row]; 759 pj = pj_oth + pi_oth[row]; 760 pa = pa_oth + pi_oth[row]; 761 /* perform sparse axpy */ 762 valtmp = aoa[j]; 763 nextp = 0; 764 for (k=0; nextp<pnz; k++) { 765 if (apJ[k] == pj[nextp]) { /* column of AP == column of P */ 766 apa_sparse[k] += valtmp*pa[nextp++]; 767 } 768 } 769 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 770 } 771 772 /* set values in C */ 773 cdnz = cd->i[i+1] - cd->i[i]; 774 conz = co->i[i+1] - co->i[i]; 775 776 /* 1st off-diagonal part of C */ 777 ca = coa + co->i[i]; 778 k = 0; 779 for (k0=0; k0<conz; k0++) { 780 if (apJ[k] >= cstart) break; 781 ca[k0] = apa_sparse[k]; 782 apa_sparse[k] = 0.0; 783 k++; 784 } 785 786 /* diagonal part of C */ 787 ca = cda + cd->i[i]; 788 for (k1=0; k1<cdnz; k1++) { 789 ca[k1] = apa_sparse[k]; 790 apa_sparse[k] = 0.0; 791 k++; 792 } 793 794 /* 2nd off-diagonal part of C */ 795 ca = coa + co->i[i]; 796 for (; k0<conz; k0++) { 797 ca[k0] = apa_sparse[k]; 798 apa_sparse[k] = 0.0; 799 k++; 800 } 801 } 802 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 803 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 804 PetscFunctionReturn(0); 805 } 806 807 /* same as MatMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(), except using LLCondensed to avoid O(BN) memory requirement */ 808 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat C) 809 { 810 PetscErrorCode ierr; 811 MPI_Comm comm; 812 PetscMPIInt size; 813 Mat_APMPI *ptap; 814 PetscFreeSpaceList free_space = NULL,current_space=NULL; 815 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 816 Mat_SeqAIJ *ad = (Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc,*p_oth; 817 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*dnz,*onz; 818 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,*aoj=ao->j,rstart=A->rmap->rstart; 819 PetscInt i,pnz,row,*api,*apj,*Jptr,apnz,nspacedouble=0,j,nzi,*lnk,apnz_max=0; 820 PetscInt am=A->rmap->n,pn=P->cmap->n,pm=P->rmap->n,lsize=pn+20; 821 PetscReal afill; 822 MatType mtype; 823 824 PetscFunctionBegin; 825 MatCheckProduct(C,4); 826 if (C->product->data) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 827 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 828 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 829 830 /* create struct Mat_APMPI and attached it to C later */ 831 ierr = PetscNew(&ptap);CHKERRQ(ierr); 832 833 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 834 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 835 836 /* get P_loc by taking all local rows of P */ 837 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 838 839 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 840 pi_loc = p_loc->i; pj_loc = p_loc->j; 841 if (size > 1) { 842 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 843 pi_oth = p_oth->i; pj_oth = p_oth->j; 844 } else { 845 p_oth = NULL; 846 pi_oth = NULL; pj_oth = NULL; 847 } 848 849 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 850 /*-------------------------------------------------------------------*/ 851 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 852 ptap->api = api; 853 api[0] = 0; 854 855 ierr = PetscLLCondensedCreate_Scalable(lsize,&lnk);CHKERRQ(ierr); 856 857 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 858 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space);CHKERRQ(ierr); 859 current_space = free_space; 860 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 861 for (i=0; i<am; i++) { 862 /* diagonal portion of A */ 863 nzi = adi[i+1] - adi[i]; 864 for (j=0; j<nzi; j++) { 865 row = *adj++; 866 pnz = pi_loc[row+1] - pi_loc[row]; 867 Jptr = pj_loc + pi_loc[row]; 868 /* Expand list if it is not long enough */ 869 if (pnz+apnz_max > lsize) { 870 lsize = pnz+apnz_max; 871 ierr = PetscLLCondensedExpand_Scalable(lsize, &lnk);CHKERRQ(ierr); 872 } 873 /* add non-zero cols of P into the sorted linked list lnk */ 874 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 875 apnz = *lnk; /* The first element in the list is the number of items in the list */ 876 api[i+1] = api[i] + apnz; 877 if (apnz > apnz_max) apnz_max = apnz; 878 } 879 /* off-diagonal portion of A */ 880 nzi = aoi[i+1] - aoi[i]; 881 for (j=0; j<nzi; j++) { 882 row = *aoj++; 883 pnz = pi_oth[row+1] - pi_oth[row]; 884 Jptr = pj_oth + pi_oth[row]; 885 /* Expand list if it is not long enough */ 886 if (pnz+apnz_max > lsize) { 887 lsize = pnz + apnz_max; 888 ierr = PetscLLCondensedExpand_Scalable(lsize, &lnk);CHKERRQ(ierr); 889 } 890 /* add non-zero cols of P into the sorted linked list lnk */ 891 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 892 apnz = *lnk; /* The first element in the list is the number of items in the list */ 893 api[i+1] = api[i] + apnz; 894 if (apnz > apnz_max) apnz_max = apnz; 895 } 896 apnz = *lnk; 897 api[i+1] = api[i] + apnz; 898 if (apnz > apnz_max) apnz_max = apnz; 899 900 /* if free space is not available, double the total space in the list */ 901 if (current_space->local_remaining<apnz) { 902 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 903 nspacedouble++; 904 } 905 906 /* Copy data into free space, then initialize lnk */ 907 ierr = PetscLLCondensedClean_Scalable(apnz,current_space->array,lnk);CHKERRQ(ierr); 908 ierr = MatPreallocateSet(i+rstart,apnz,current_space->array,dnz,onz);CHKERRQ(ierr); 909 910 current_space->array += apnz; 911 current_space->local_used += apnz; 912 current_space->local_remaining -= apnz; 913 } 914 915 /* Allocate space for apj, initialize apj, and */ 916 /* destroy list of free space and other temporary array(s) */ 917 ierr = PetscMalloc1(api[am]+1,&ptap->apj);CHKERRQ(ierr); 918 apj = ptap->apj; 919 ierr = PetscFreeSpaceContiguous(&free_space,ptap->apj);CHKERRQ(ierr); 920 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 921 922 /* create and assemble symbolic parallel matrix C */ 923 /*----------------------------------------------------*/ 924 ierr = MatSetSizes(C,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 925 ierr = MatSetBlockSizesFromMats(C,A,P);CHKERRQ(ierr); 926 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 927 ierr = MatSetType(C,mtype);CHKERRQ(ierr); 928 ierr = MatMPIAIJSetPreallocation(C,0,dnz,0,onz);CHKERRQ(ierr); 929 930 /* malloc apa for assembly C */ 931 ierr = PetscCalloc1(apnz_max,&ptap->apa);CHKERRQ(ierr); 932 933 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(C, apj, api);CHKERRQ(ierr); 934 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 935 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 936 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 937 938 C->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ; 939 C->ops->productnumeric = MatProductNumeric_AB; 940 941 /* attach the supporting struct to C for reuse */ 942 C->product->data = ptap; 943 C->product->destroy = MatDestroy_MPIAIJ_MatMatMult; 944 945 /* set MatInfo */ 946 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 947 if (afill < 1.0) afill = 1.0; 948 C->info.mallocs = nspacedouble; 949 C->info.fill_ratio_given = fill; 950 C->info.fill_ratio_needed = afill; 951 952 #if defined(PETSC_USE_INFO) 953 if (api[am]) { 954 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 955 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 956 } else { 957 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 958 } 959 #endif 960 PetscFunctionReturn(0); 961 } 962 963 /* This function is needed for the seqMPI matrix-matrix multiplication. */ 964 /* Three input arrays are merged to one output array. The size of the */ 965 /* output array is also output. Duplicate entries only show up once. */ 966 static void Merge3SortedArrays(PetscInt size1, PetscInt *in1, 967 PetscInt size2, PetscInt *in2, 968 PetscInt size3, PetscInt *in3, 969 PetscInt *size4, PetscInt *out) 970 { 971 int i = 0, j = 0, k = 0, l = 0; 972 973 /* Traverse all three arrays */ 974 while (i<size1 && j<size2 && k<size3) { 975 if (in1[i] < in2[j] && in1[i] < in3[k]) { 976 out[l++] = in1[i++]; 977 } 978 else if(in2[j] < in1[i] && in2[j] < in3[k]) { 979 out[l++] = in2[j++]; 980 } 981 else if(in3[k] < in1[i] && in3[k] < in2[j]) { 982 out[l++] = in3[k++]; 983 } 984 else if(in1[i] == in2[j] && in1[i] < in3[k]) { 985 out[l++] = in1[i]; 986 i++, j++; 987 } 988 else if(in1[i] == in3[k] && in1[i] < in2[j]) { 989 out[l++] = in1[i]; 990 i++, k++; 991 } 992 else if(in3[k] == in2[j] && in2[j] < in1[i]) { 993 out[l++] = in2[j]; 994 k++, j++; 995 } 996 else if(in1[i] == in2[j] && in1[i] == in3[k]) { 997 out[l++] = in1[i]; 998 i++, j++, k++; 999 } 1000 } 1001 1002 /* Traverse two remaining arrays */ 1003 while (i<size1 && j<size2) { 1004 if (in1[i] < in2[j]) { 1005 out[l++] = in1[i++]; 1006 } 1007 else if(in1[i] > in2[j]) { 1008 out[l++] = in2[j++]; 1009 } 1010 else { 1011 out[l++] = in1[i]; 1012 i++, j++; 1013 } 1014 } 1015 1016 while (i<size1 && k<size3) { 1017 if (in1[i] < in3[k]) { 1018 out[l++] = in1[i++]; 1019 } 1020 else if(in1[i] > in3[k]) { 1021 out[l++] = in3[k++]; 1022 } 1023 else { 1024 out[l++] = in1[i]; 1025 i++, k++; 1026 } 1027 } 1028 1029 while (k<size3 && j<size2) { 1030 if (in3[k] < in2[j]) { 1031 out[l++] = in3[k++]; 1032 } 1033 else if(in3[k] > in2[j]) { 1034 out[l++] = in2[j++]; 1035 } 1036 else { 1037 out[l++] = in3[k]; 1038 k++, j++; 1039 } 1040 } 1041 1042 /* Traverse one remaining array */ 1043 while (i<size1) out[l++] = in1[i++]; 1044 while (j<size2) out[l++] = in2[j++]; 1045 while (k<size3) out[l++] = in3[k++]; 1046 1047 *size4 = l; 1048 } 1049 1050 /* This matrix-matrix multiplication algorithm divides the multiplication into three multiplications and */ 1051 /* adds up the products. Two of these three multiplications are performed with existing (sequential) */ 1052 /* matrix-matrix multiplications. */ 1053 PetscErrorCode MatMatMultSymbolic_MPIAIJ_MPIAIJ_seqMPI(Mat A, Mat P, PetscReal fill, Mat C) 1054 { 1055 PetscErrorCode ierr; 1056 MPI_Comm comm; 1057 PetscMPIInt size; 1058 Mat_APMPI *ptap; 1059 PetscFreeSpaceList free_space_diag=NULL, current_space=NULL; 1060 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data; 1061 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*p_loc; 1062 Mat_MPIAIJ *p =(Mat_MPIAIJ*)P->data; 1063 Mat_SeqAIJ *adpd_seq, *p_off, *aopoth_seq; 1064 PetscInt adponz, adpdnz; 1065 PetscInt *pi_loc,*dnz,*onz; 1066 PetscInt *adi=ad->i,*adj=ad->j,*aoi=ao->i,rstart=A->rmap->rstart; 1067 PetscInt *lnk,i, i1=0,pnz,row,*adpoi,*adpoj, *api, *adpoJ, *aopJ, *apJ,*Jptr, aopnz, nspacedouble=0,j,nzi, 1068 *apj,apnz, *adpdi, *adpdj, *adpdJ, *poff_i, *poff_j, *j_temp, *aopothi, *aopothj; 1069 PetscInt am=A->rmap->n,pN=P->cmap->N,pn=P->cmap->n,pm=P->rmap->n, p_colstart, p_colend; 1070 PetscBT lnkbt; 1071 PetscReal afill; 1072 PetscMPIInt rank; 1073 Mat adpd, aopoth; 1074 MatType mtype; 1075 const char *prefix; 1076 1077 PetscFunctionBegin; 1078 MatCheckProduct(C,4); 1079 if (C->product->data) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 1080 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1081 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1082 ierr = MPI_Comm_rank(comm, &rank);CHKERRQ(ierr); 1083 ierr = MatGetOwnershipRangeColumn(P, &p_colstart, &p_colend); CHKERRQ(ierr); 1084 1085 /* create struct Mat_APMPI and attached it to C later */ 1086 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1087 1088 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 1089 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 1090 1091 /* get P_loc by taking all local rows of P */ 1092 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 1093 1094 1095 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 1096 pi_loc = p_loc->i; 1097 1098 /* Allocate memory for the i arrays of the matrices A*P, A_diag*P_off and A_offd * P */ 1099 ierr = PetscMalloc1(am+2,&api);CHKERRQ(ierr); 1100 ierr = PetscMalloc1(am+2,&adpoi);CHKERRQ(ierr); 1101 1102 adpoi[0] = 0; 1103 ptap->api = api; 1104 api[0] = 0; 1105 1106 /* create and initialize a linked list, will be used for both A_diag * P_loc_off and A_offd * P_oth */ 1107 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 1108 ierr = MatPreallocateInitialize(comm,am,pn,dnz,onz);CHKERRQ(ierr); 1109 1110 /* Symbolic calc of A_loc_diag * P_loc_diag */ 1111 ierr = MatGetOptionsPrefix(A,&prefix);CHKERRQ(ierr); 1112 ierr = MatProductCreate(a->A,p->A,NULL,&adpd);CHKERRQ(ierr); 1113 ierr = MatGetOptionsPrefix(A,&prefix);CHKERRQ(ierr); 1114 ierr = MatSetOptionsPrefix(adpd,prefix);CHKERRQ(ierr); 1115 ierr = MatAppendOptionsPrefix(adpd,"inner_diag_");CHKERRQ(ierr); 1116 1117 ierr = MatProductSetType(adpd,MATPRODUCT_AB);CHKERRQ(ierr); 1118 ierr = MatProductSetAlgorithm(adpd,"sorted");CHKERRQ(ierr); 1119 ierr = MatProductSetFill(adpd,fill);CHKERRQ(ierr); 1120 ierr = MatProductSetFromOptions(adpd);CHKERRQ(ierr); 1121 ierr = MatProductSymbolic(adpd);CHKERRQ(ierr); 1122 1123 adpd_seq = (Mat_SeqAIJ*)((adpd)->data); 1124 adpdi = adpd_seq->i; adpdj = adpd_seq->j; 1125 p_off = (Mat_SeqAIJ*)((p->B)->data); 1126 poff_i = p_off->i; poff_j = p_off->j; 1127 1128 /* j_temp stores indices of a result row before they are added to the linked list */ 1129 ierr = PetscMalloc1(pN+2,&j_temp);CHKERRQ(ierr); 1130 1131 1132 /* Symbolic calc of the A_diag * p_loc_off */ 1133 /* Initial FreeSpace size is fill*(nnz(A)+nnz(P)) */ 1134 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space_diag);CHKERRQ(ierr); 1135 current_space = free_space_diag; 1136 1137 for (i=0; i<am; i++) { 1138 /* A_diag * P_loc_off */ 1139 nzi = adi[i+1] - adi[i]; 1140 for (j=0; j<nzi; j++) { 1141 row = *adj++; 1142 pnz = poff_i[row+1] - poff_i[row]; 1143 Jptr = poff_j + poff_i[row]; 1144 for(i1 = 0; i1 < pnz; i1++) { 1145 j_temp[i1] = p->garray[Jptr[i1]]; 1146 } 1147 /* add non-zero cols of P into the sorted linked list lnk */ 1148 ierr = PetscLLCondensedAddSorted(pnz,j_temp,lnk,lnkbt);CHKERRQ(ierr); 1149 } 1150 1151 adponz = lnk[0]; 1152 adpoi[i+1] = adpoi[i] + adponz; 1153 1154 /* if free space is not available, double the total space in the list */ 1155 if (current_space->local_remaining<adponz) { 1156 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(adponz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1157 nspacedouble++; 1158 } 1159 1160 /* Copy data into free space, then initialize lnk */ 1161 ierr = PetscLLCondensedClean(pN,adponz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1162 1163 current_space->array += adponz; 1164 current_space->local_used += adponz; 1165 current_space->local_remaining -= adponz; 1166 } 1167 1168 /* Symbolic calc of A_off * P_oth */ 1169 ierr = MatSetOptionsPrefix(a->B,prefix);CHKERRQ(ierr); 1170 ierr = MatAppendOptionsPrefix(a->B,"inner_offdiag_");CHKERRQ(ierr); 1171 ierr = MatCreate(PETSC_COMM_SELF,&aopoth);CHKERRQ(ierr); 1172 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(a->B, ptap->P_oth, fill, aopoth);CHKERRQ(ierr); 1173 aopoth_seq = (Mat_SeqAIJ*)((aopoth)->data); 1174 aopothi = aopoth_seq->i; aopothj = aopoth_seq->j; 1175 1176 /* Allocate space for apj, adpj, aopj, ... */ 1177 /* destroy lists of free space and other temporary array(s) */ 1178 1179 ierr = PetscMalloc1(aopothi[am] + adpoi[am] + adpdi[am]+2, &ptap->apj);CHKERRQ(ierr); 1180 ierr = PetscMalloc1(adpoi[am]+2, &adpoj);CHKERRQ(ierr); 1181 1182 /* Copy from linked list to j-array */ 1183 ierr = PetscFreeSpaceContiguous(&free_space_diag,adpoj);CHKERRQ(ierr); 1184 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1185 1186 adpoJ = adpoj; 1187 adpdJ = adpdj; 1188 aopJ = aopothj; 1189 apj = ptap->apj; 1190 apJ = apj; /* still empty */ 1191 1192 /* Merge j-arrays of A_off * P, A_diag * P_loc_off, and */ 1193 /* A_diag * P_loc_diag to get A*P */ 1194 for (i = 0; i < am; i++) { 1195 aopnz = aopothi[i+1] - aopothi[i]; 1196 adponz = adpoi[i+1] - adpoi[i]; 1197 adpdnz = adpdi[i+1] - adpdi[i]; 1198 1199 /* Correct indices from A_diag*P_diag */ 1200 for(i1 = 0; i1 < adpdnz; i1++) { 1201 adpdJ[i1] += p_colstart; 1202 } 1203 /* Merge j-arrays of A_diag * P_loc_off and A_diag * P_loc_diag and A_off * P_oth */ 1204 Merge3SortedArrays(adponz, adpoJ, adpdnz, adpdJ, aopnz, aopJ, &apnz, apJ); 1205 ierr = MatPreallocateSet(i+rstart, apnz, apJ, dnz, onz); CHKERRQ(ierr); 1206 1207 aopJ += aopnz; 1208 adpoJ += adponz; 1209 adpdJ += adpdnz; 1210 apJ += apnz; 1211 api[i+1] = api[i] + apnz; 1212 } 1213 1214 /* malloc apa to store dense row A[i,:]*P */ 1215 ierr = PetscCalloc1(pN+2,&ptap->apa);CHKERRQ(ierr); 1216 1217 /* create and assemble symbolic parallel matrix C */ 1218 ierr = MatSetSizes(C,am,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 1219 ierr = MatSetBlockSizesFromMats(C,A,P);CHKERRQ(ierr); 1220 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 1221 ierr = MatSetType(C,mtype);CHKERRQ(ierr); 1222 ierr = MatMPIAIJSetPreallocation(C,0,dnz,0,onz);CHKERRQ(ierr); 1223 1224 1225 ierr = MatSetValues_MPIAIJ_CopyFromCSRFormat_Symbolic(C, apj, api);CHKERRQ(ierr); 1226 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1227 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1228 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 1229 1230 1231 C->ops->matmultnumeric = MatMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 1232 C->ops->productnumeric = MatProductNumeric_AB; 1233 1234 /* attach the supporting struct to C for reuse */ 1235 C->product->data = ptap; 1236 C->product->destroy = MatDestroy_MPIAIJ_MatMatMult; 1237 1238 /* set MatInfo */ 1239 afill = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1) + 1.e-5; 1240 if (afill < 1.0) afill = 1.0; 1241 C->info.mallocs = nspacedouble; 1242 C->info.fill_ratio_given = fill; 1243 C->info.fill_ratio_needed = afill; 1244 1245 #if defined(PETSC_USE_INFO) 1246 if (api[am]) { 1247 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 1248 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1249 } else { 1250 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 1251 } 1252 #endif 1253 1254 ierr = MatDestroy(&aopoth);CHKERRQ(ierr); 1255 ierr = MatDestroy(&adpd);CHKERRQ(ierr); 1256 ierr = PetscFree(j_temp);CHKERRQ(ierr); 1257 ierr = PetscFree(adpoj);CHKERRQ(ierr); 1258 ierr = PetscFree(adpoi);CHKERRQ(ierr); 1259 PetscFunctionReturn(0); 1260 } 1261 1262 /*-------------------------------------------------------------------------*/ 1263 /* This routine only works when scall=MAT_REUSE_MATRIX! */ 1264 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_matmatmult(Mat P,Mat A,Mat C) 1265 { 1266 PetscErrorCode ierr; 1267 Mat_APMPI *ptap; 1268 Mat Pt; 1269 1270 PetscFunctionBegin; 1271 MatCheckProduct(C,3); 1272 ptap = (Mat_APMPI*)C->product->data; 1273 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtAP cannot be computed. Missing data"); 1274 if (!ptap->Pt) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatProductClear()"); 1275 1276 Pt = ptap->Pt; 1277 ierr = MatTranspose(P,MAT_REUSE_MATRIX,&Pt);CHKERRQ(ierr); 1278 ierr = MatMatMultNumeric_MPIAIJ_MPIAIJ(Pt,A,C);CHKERRQ(ierr); 1279 PetscFunctionReturn(0); 1280 } 1281 1282 /* This routine is modified from MatPtAPSymbolic_MPIAIJ_MPIAIJ() */ 1283 PetscErrorCode MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(Mat P,Mat A,PetscReal fill,Mat C) 1284 { 1285 PetscErrorCode ierr; 1286 Mat_APMPI *ptap; 1287 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data; 1288 MPI_Comm comm; 1289 PetscMPIInt size,rank; 1290 PetscFreeSpaceList free_space=NULL,current_space=NULL; 1291 PetscInt pn=P->cmap->n,aN=A->cmap->N,an=A->cmap->n; 1292 PetscInt *lnk,i,k,nsend; 1293 PetscBT lnkbt; 1294 PetscMPIInt tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0,nrecv; 1295 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 1296 PetscInt len,proc,*dnz,*onz,*owners,nzi; 1297 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 1298 MPI_Request *swaits,*rwaits; 1299 MPI_Status *sstatus,rstatus; 1300 PetscLayout rowmap; 1301 PetscInt *owners_co,*coi,*coj; /* i and j array of (p->B)^T*A*P - used in the communication */ 1302 PetscMPIInt *len_r,*id_r; /* array of length of comm->size, store send/recv matrix values */ 1303 PetscInt *Jptr,*prmap=p->garray,con,j,Crmax; 1304 Mat_SeqAIJ *a_loc,*c_loc,*c_oth; 1305 PetscTable ta; 1306 MatType mtype; 1307 const char *prefix; 1308 1309 PetscFunctionBegin; 1310 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1311 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1312 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1313 1314 /* create symbolic parallel matrix C */ 1315 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 1316 ierr = MatSetType(C,mtype);CHKERRQ(ierr); 1317 1318 C->ops->transposematmultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable; 1319 1320 /* create struct Mat_APMPI and attached it to C later */ 1321 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1322 ptap->reuse = MAT_INITIAL_MATRIX; 1323 1324 /* (0) compute Rd = Pd^T, Ro = Po^T */ 1325 /* --------------------------------- */ 1326 ierr = MatTranspose_SeqAIJ(p->A,MAT_INITIAL_MATRIX,&ptap->Rd);CHKERRQ(ierr); 1327 ierr = MatTranspose_SeqAIJ(p->B,MAT_INITIAL_MATRIX,&ptap->Ro);CHKERRQ(ierr); 1328 1329 /* (1) compute symbolic A_loc */ 1330 /* ---------------------------*/ 1331 ierr = MatMPIAIJGetLocalMat(A,MAT_INITIAL_MATRIX,&ptap->A_loc);CHKERRQ(ierr); 1332 1333 /* (2-1) compute symbolic C_oth = Ro*A_loc */ 1334 /* ------------------------------------ */ 1335 ierr = MatGetOptionsPrefix(A,&prefix);CHKERRQ(ierr); 1336 ierr = MatSetOptionsPrefix(ptap->Ro,prefix);CHKERRQ(ierr); 1337 ierr = MatAppendOptionsPrefix(ptap->Ro,"inner_offdiag_");CHKERRQ(ierr); 1338 ierr = MatCreate(PETSC_COMM_SELF,&ptap->C_oth);CHKERRQ(ierr); 1339 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Ro,ptap->A_loc,fill,ptap->C_oth);CHKERRQ(ierr); 1340 1341 /* (3) send coj of C_oth to other processors */ 1342 /* ------------------------------------------ */ 1343 /* determine row ownership */ 1344 ierr = PetscLayoutCreate(comm,&rowmap);CHKERRQ(ierr); 1345 rowmap->n = pn; 1346 rowmap->bs = 1; 1347 ierr = PetscLayoutSetUp(rowmap);CHKERRQ(ierr); 1348 owners = rowmap->range; 1349 1350 /* determine the number of messages to send, their lengths */ 1351 ierr = PetscMalloc4(size,&len_s,size,&len_si,size,&sstatus,size+2,&owners_co);CHKERRQ(ierr); 1352 ierr = PetscArrayzero(len_s,size);CHKERRQ(ierr); 1353 ierr = PetscArrayzero(len_si,size);CHKERRQ(ierr); 1354 1355 c_oth = (Mat_SeqAIJ*)ptap->C_oth->data; 1356 coi = c_oth->i; coj = c_oth->j; 1357 con = ptap->C_oth->rmap->n; 1358 proc = 0; 1359 for (i=0; i<con; i++) { 1360 while (prmap[i] >= owners[proc+1]) proc++; 1361 len_si[proc]++; /* num of rows in Co(=Pt*A) to be sent to [proc] */ 1362 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 1363 } 1364 1365 len = 0; /* max length of buf_si[], see (4) */ 1366 owners_co[0] = 0; 1367 nsend = 0; 1368 for (proc=0; proc<size; proc++) { 1369 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 1370 if (len_s[proc]) { 1371 nsend++; 1372 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 1373 len += len_si[proc]; 1374 } 1375 } 1376 1377 /* determine the number and length of messages to receive for coi and coj */ 1378 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&nrecv);CHKERRQ(ierr); 1379 ierr = PetscGatherMessageLengths2(comm,nsend,nrecv,len_s,len_si,&id_r,&len_r,&len_ri);CHKERRQ(ierr); 1380 1381 /* post the Irecv and Isend of coj */ 1382 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 1383 ierr = PetscPostIrecvInt(comm,tagj,nrecv,id_r,len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 1384 ierr = PetscMalloc1(nsend+1,&swaits);CHKERRQ(ierr); 1385 for (proc=0, k=0; proc<size; proc++) { 1386 if (!len_s[proc]) continue; 1387 i = owners_co[proc]; 1388 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 1389 k++; 1390 } 1391 1392 /* (2-2) compute symbolic C_loc = Rd*A_loc */ 1393 /* ---------------------------------------- */ 1394 ierr = MatSetOptionsPrefix(ptap->Rd,prefix);CHKERRQ(ierr); 1395 ierr = MatAppendOptionsPrefix(ptap->Rd,"inner_diag_");CHKERRQ(ierr); 1396 ierr = MatCreate(PETSC_COMM_SELF,&ptap->C_loc);CHKERRQ(ierr); 1397 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Rd,ptap->A_loc,fill,ptap->C_loc);CHKERRQ(ierr); 1398 c_loc = (Mat_SeqAIJ*)ptap->C_loc->data; 1399 1400 /* receives coj are complete */ 1401 for (i=0; i<nrecv; i++) { 1402 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1403 } 1404 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1405 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 1406 1407 /* add received column indices into ta to update Crmax */ 1408 a_loc = (Mat_SeqAIJ*)(ptap->A_loc)->data; 1409 1410 /* create and initialize a linked list */ 1411 ierr = PetscTableCreate(an,aN,&ta);CHKERRQ(ierr); /* for compute Crmax */ 1412 MatRowMergeMax_SeqAIJ(a_loc,ptap->A_loc->rmap->N,ta); 1413 1414 for (k=0; k<nrecv; k++) {/* k-th received message */ 1415 Jptr = buf_rj[k]; 1416 for (j=0; j<len_r[k]; j++) { 1417 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 1418 } 1419 } 1420 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 1421 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 1422 1423 /* (4) send and recv coi */ 1424 /*-----------------------*/ 1425 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 1426 ierr = PetscPostIrecvInt(comm,tagi,nrecv,id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 1427 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 1428 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 1429 for (proc=0,k=0; proc<size; proc++) { 1430 if (!len_s[proc]) continue; 1431 /* form outgoing message for i-structure: 1432 buf_si[0]: nrows to be sent 1433 [1:nrows]: row index (global) 1434 [nrows+1:2*nrows+1]: i-structure index 1435 */ 1436 /*-------------------------------------------*/ 1437 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 1438 buf_si_i = buf_si + nrows+1; 1439 buf_si[0] = nrows; 1440 buf_si_i[0] = 0; 1441 nrows = 0; 1442 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 1443 nzi = coi[i+1] - coi[i]; 1444 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 1445 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 1446 nrows++; 1447 } 1448 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 1449 k++; 1450 buf_si += len_si[proc]; 1451 } 1452 for (i=0; i<nrecv; i++) { 1453 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1454 } 1455 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1456 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 1457 1458 ierr = PetscFree4(len_s,len_si,sstatus,owners_co);CHKERRQ(ierr); 1459 ierr = PetscFree(len_ri);CHKERRQ(ierr); 1460 ierr = PetscFree(swaits);CHKERRQ(ierr); 1461 ierr = PetscFree(buf_s);CHKERRQ(ierr); 1462 1463 /* (5) compute the local portion of C */ 1464 /* ------------------------------------------ */ 1465 /* set initial free space to be Crmax, sufficient for holding nozeros in each row of C */ 1466 ierr = PetscFreeSpaceGet(Crmax,&free_space);CHKERRQ(ierr); 1467 current_space = free_space; 1468 1469 ierr = PetscMalloc3(nrecv,&buf_ri_k,nrecv,&nextrow,nrecv,&nextci);CHKERRQ(ierr); 1470 for (k=0; k<nrecv; k++) { 1471 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1472 nrows = *buf_ri_k[k]; 1473 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 1474 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1475 } 1476 1477 ierr = MatPreallocateInitialize(comm,pn,an,dnz,onz);CHKERRQ(ierr); 1478 ierr = PetscLLCondensedCreate(Crmax,aN,&lnk,&lnkbt);CHKERRQ(ierr); 1479 for (i=0; i<pn; i++) { 1480 /* add C_loc into C */ 1481 nzi = c_loc->i[i+1] - c_loc->i[i]; 1482 Jptr = c_loc->j + c_loc->i[i]; 1483 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1484 1485 /* add received col data into lnk */ 1486 for (k=0; k<nrecv; k++) { /* k-th received message */ 1487 if (i == *nextrow[k]) { /* i-th row */ 1488 nzi = *(nextci[k]+1) - *nextci[k]; 1489 Jptr = buf_rj[k] + *nextci[k]; 1490 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1491 nextrow[k]++; nextci[k]++; 1492 } 1493 } 1494 nzi = lnk[0]; 1495 1496 /* copy data into free space, then initialize lnk */ 1497 ierr = PetscLLCondensedClean(aN,nzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1498 ierr = MatPreallocateSet(i+owners[rank],nzi,current_space->array,dnz,onz);CHKERRQ(ierr); 1499 } 1500 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1501 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1502 ierr = PetscFreeSpaceDestroy(free_space);CHKERRQ(ierr); 1503 1504 /* local sizes and preallocation */ 1505 ierr = MatSetSizes(C,pn,an,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 1506 if (P->cmap->bs > 0) {ierr = PetscLayoutSetBlockSize(C->rmap,P->cmap->bs);CHKERRQ(ierr);} 1507 if (A->cmap->bs > 0) {ierr = PetscLayoutSetBlockSize(C->cmap,A->cmap->bs);CHKERRQ(ierr);} 1508 ierr = MatMPIAIJSetPreallocation(C,0,dnz,0,onz);CHKERRQ(ierr); 1509 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 1510 1511 /* members in merge */ 1512 ierr = PetscFree(id_r);CHKERRQ(ierr); 1513 ierr = PetscFree(len_r);CHKERRQ(ierr); 1514 ierr = PetscFree(buf_ri[0]);CHKERRQ(ierr); 1515 ierr = PetscFree(buf_ri);CHKERRQ(ierr); 1516 ierr = PetscFree(buf_rj[0]);CHKERRQ(ierr); 1517 ierr = PetscFree(buf_rj);CHKERRQ(ierr); 1518 ierr = PetscLayoutDestroy(&rowmap);CHKERRQ(ierr); 1519 1520 /* attach the supporting struct to C for reuse */ 1521 C->product->data = ptap; 1522 C->product->destroy = MatDestroy_MPIAIJ_PtAP; 1523 1524 /* C is not ready for use - assembly will be done by MatPtAPNumeric() */ 1525 C->assembled = PETSC_FALSE; 1526 PetscFunctionReturn(0); 1527 } 1528 1529 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_nonscalable(Mat P,Mat A,Mat C) 1530 { 1531 PetscErrorCode ierr; 1532 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data; 1533 Mat_SeqAIJ *c_seq; 1534 Mat_APMPI *ptap; 1535 Mat A_loc,C_loc,C_oth; 1536 PetscInt i,rstart,rend,cm,ncols,row; 1537 const PetscInt *cols; 1538 const PetscScalar *vals; 1539 1540 PetscFunctionBegin; 1541 MatCheckProduct(C,3); 1542 ptap = (Mat_APMPI*)C->product->data; 1543 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtAP cannot be computed. Missing data"); 1544 if (!ptap->A_loc) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatProductClear()"); 1545 ierr = MatZeroEntries(C);CHKERRQ(ierr); 1546 1547 if (ptap->reuse == MAT_REUSE_MATRIX) { 1548 /* These matrices are obtained in MatTransposeMatMultSymbolic() */ 1549 /* 1) get R = Pd^T, Ro = Po^T */ 1550 /*----------------------------*/ 1551 ierr = MatTranspose_SeqAIJ(p->A,MAT_REUSE_MATRIX,&ptap->Rd);CHKERRQ(ierr); 1552 ierr = MatTranspose_SeqAIJ(p->B,MAT_REUSE_MATRIX,&ptap->Ro);CHKERRQ(ierr); 1553 1554 /* 2) compute numeric A_loc */ 1555 /*--------------------------*/ 1556 ierr = MatMPIAIJGetLocalMat(A,MAT_REUSE_MATRIX,&ptap->A_loc);CHKERRQ(ierr); 1557 } 1558 1559 /* 3) C_loc = Rd*A_loc, C_oth = Ro*A_loc */ 1560 A_loc = ptap->A_loc; 1561 ierr = ((ptap->C_loc)->ops->matmultnumeric)(ptap->Rd,A_loc,ptap->C_loc);CHKERRQ(ierr); 1562 ierr = ((ptap->C_oth)->ops->matmultnumeric)(ptap->Ro,A_loc,ptap->C_oth);CHKERRQ(ierr); 1563 C_loc = ptap->C_loc; 1564 C_oth = ptap->C_oth; 1565 1566 /* add C_loc and Co to to C */ 1567 ierr = MatGetOwnershipRange(C,&rstart,&rend);CHKERRQ(ierr); 1568 1569 /* C_loc -> C */ 1570 cm = C_loc->rmap->N; 1571 c_seq = (Mat_SeqAIJ*)C_loc->data; 1572 cols = c_seq->j; 1573 vals = c_seq->a; 1574 for (i=0; i<cm; i++) { 1575 ncols = c_seq->i[i+1] - c_seq->i[i]; 1576 row = rstart + i; 1577 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1578 cols += ncols; vals += ncols; 1579 } 1580 1581 /* Co -> C, off-processor part */ 1582 cm = C_oth->rmap->N; 1583 c_seq = (Mat_SeqAIJ*)C_oth->data; 1584 cols = c_seq->j; 1585 vals = c_seq->a; 1586 for (i=0; i<cm; i++) { 1587 ncols = c_seq->i[i+1] - c_seq->i[i]; 1588 row = p->garray[i]; 1589 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1590 cols += ncols; vals += ncols; 1591 } 1592 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1593 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1594 1595 ptap->reuse = MAT_REUSE_MATRIX; 1596 PetscFunctionReturn(0); 1597 } 1598 1599 PetscErrorCode MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ(Mat P,Mat A,Mat C) 1600 { 1601 PetscErrorCode ierr; 1602 Mat_Merge_SeqsToMPI *merge; 1603 Mat_MPIAIJ *p =(Mat_MPIAIJ*)P->data; 1604 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 1605 Mat_APMPI *ptap; 1606 PetscInt *adj; 1607 PetscInt i,j,k,anz,pnz,row,*cj,nexta; 1608 MatScalar *ada,*ca,valtmp; 1609 PetscInt am=A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 1610 MPI_Comm comm; 1611 PetscMPIInt size,rank,taga,*len_s; 1612 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 1613 PetscInt **buf_ri,**buf_rj; 1614 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 1615 MPI_Request *s_waits,*r_waits; 1616 MPI_Status *status; 1617 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 1618 PetscInt *ai,*aj,*coi,*coj,*poJ,*pdJ; 1619 Mat A_loc; 1620 Mat_SeqAIJ *a_loc; 1621 1622 PetscFunctionBegin; 1623 MatCheckProduct(C,3); 1624 ptap = (Mat_APMPI*)C->product->data; 1625 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtAP cannot be computed. Missing data"); 1626 if (!ptap->A_loc) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"PtA cannot be reused. Do not call MatProductClear()"); 1627 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1628 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1629 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1630 1631 merge = ptap->merge; 1632 1633 /* 2) compute numeric C_seq = P_loc^T*A_loc */ 1634 /*------------------------------------------*/ 1635 /* get data from symbolic products */ 1636 coi = merge->coi; coj = merge->coj; 1637 ierr = PetscCalloc1(coi[pon]+1,&coa);CHKERRQ(ierr); 1638 bi = merge->bi; bj = merge->bj; 1639 owners = merge->rowmap->range; 1640 ierr = PetscCalloc1(bi[cm]+1,&ba);CHKERRQ(ierr); 1641 1642 /* get A_loc by taking all local rows of A */ 1643 A_loc = ptap->A_loc; 1644 ierr = MatMPIAIJGetLocalMat(A,MAT_REUSE_MATRIX,&A_loc);CHKERRQ(ierr); 1645 a_loc = (Mat_SeqAIJ*)(A_loc)->data; 1646 ai = a_loc->i; 1647 aj = a_loc->j; 1648 1649 for (i=0; i<am; i++) { 1650 anz = ai[i+1] - ai[i]; 1651 adj = aj + ai[i]; 1652 ada = a_loc->a + ai[i]; 1653 1654 /* 2-b) Compute Cseq = P_loc[i,:]^T*A[i,:] using outer product */ 1655 /*-------------------------------------------------------------*/ 1656 /* put the value into Co=(p->B)^T*A (off-diagonal part, send to others) */ 1657 pnz = po->i[i+1] - po->i[i]; 1658 poJ = po->j + po->i[i]; 1659 pA = po->a + po->i[i]; 1660 for (j=0; j<pnz; j++) { 1661 row = poJ[j]; 1662 cj = coj + coi[row]; 1663 ca = coa + coi[row]; 1664 /* perform sparse axpy */ 1665 nexta = 0; 1666 valtmp = pA[j]; 1667 for (k=0; nexta<anz; k++) { 1668 if (cj[k] == adj[nexta]) { 1669 ca[k] += valtmp*ada[nexta]; 1670 nexta++; 1671 } 1672 } 1673 ierr = PetscLogFlops(2.0*anz);CHKERRQ(ierr); 1674 } 1675 1676 /* put the value into Cd (diagonal part) */ 1677 pnz = pd->i[i+1] - pd->i[i]; 1678 pdJ = pd->j + pd->i[i]; 1679 pA = pd->a + pd->i[i]; 1680 for (j=0; j<pnz; j++) { 1681 row = pdJ[j]; 1682 cj = bj + bi[row]; 1683 ca = ba + bi[row]; 1684 /* perform sparse axpy */ 1685 nexta = 0; 1686 valtmp = pA[j]; 1687 for (k=0; nexta<anz; k++) { 1688 if (cj[k] == adj[nexta]) { 1689 ca[k] += valtmp*ada[nexta]; 1690 nexta++; 1691 } 1692 } 1693 ierr = PetscLogFlops(2.0*anz);CHKERRQ(ierr); 1694 } 1695 } 1696 1697 /* 3) send and recv matrix values coa */ 1698 /*------------------------------------*/ 1699 buf_ri = merge->buf_ri; 1700 buf_rj = merge->buf_rj; 1701 len_s = merge->len_s; 1702 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 1703 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 1704 1705 ierr = PetscMalloc2(merge->nsend+1,&s_waits,size,&status);CHKERRQ(ierr); 1706 for (proc=0,k=0; proc<size; proc++) { 1707 if (!len_s[proc]) continue; 1708 i = merge->owners_co[proc]; 1709 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 1710 k++; 1711 } 1712 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 1713 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 1714 1715 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 1716 ierr = PetscFree(r_waits);CHKERRQ(ierr); 1717 ierr = PetscFree(coa);CHKERRQ(ierr); 1718 1719 /* 4) insert local Cseq and received values into Cmpi */ 1720 /*----------------------------------------------------*/ 1721 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 1722 for (k=0; k<merge->nrecv; k++) { 1723 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1724 nrows = *(buf_ri_k[k]); 1725 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 1726 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1727 } 1728 1729 for (i=0; i<cm; i++) { 1730 row = owners[rank] + i; /* global row index of C_seq */ 1731 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 1732 ba_i = ba + bi[i]; 1733 bnz = bi[i+1] - bi[i]; 1734 /* add received vals into ba */ 1735 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 1736 /* i-th row */ 1737 if (i == *nextrow[k]) { 1738 cnz = *(nextci[k]+1) - *nextci[k]; 1739 cj = buf_rj[k] + *(nextci[k]); 1740 ca = abuf_r[k] + *(nextci[k]); 1741 nextcj = 0; 1742 for (j=0; nextcj<cnz; j++) { 1743 if (bj_i[j] == cj[nextcj]) { /* bcol == ccol */ 1744 ba_i[j] += ca[nextcj++]; 1745 } 1746 } 1747 nextrow[k]++; nextci[k]++; 1748 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 1749 } 1750 } 1751 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 1752 } 1753 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1754 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1755 1756 ierr = PetscFree(ba);CHKERRQ(ierr); 1757 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 1758 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 1759 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1760 PetscFunctionReturn(0); 1761 } 1762 1763 PetscErrorCode MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ(Mat P,Mat A,PetscReal fill,Mat C) 1764 { 1765 PetscErrorCode ierr; 1766 Mat A_loc,POt,PDt; 1767 Mat_APMPI *ptap; 1768 PetscFreeSpaceList free_space=NULL,current_space=NULL; 1769 Mat_MPIAIJ *p=(Mat_MPIAIJ*)P->data,*a=(Mat_MPIAIJ*)A->data; 1770 PetscInt *pdti,*pdtj,*poti,*potj,*ptJ; 1771 PetscInt nnz; 1772 PetscInt *lnk,*owners_co,*coi,*coj,i,k,pnz,row; 1773 PetscInt am =A->rmap->n,pn=P->cmap->n; 1774 MPI_Comm comm; 1775 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri; 1776 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 1777 PetscInt len,proc,*dnz,*onz,*owners; 1778 PetscInt nzi,*bi,*bj; 1779 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 1780 MPI_Request *swaits,*rwaits; 1781 MPI_Status *sstatus,rstatus; 1782 Mat_Merge_SeqsToMPI *merge; 1783 PetscInt *ai,*aj,*Jptr,anz,*prmap=p->garray,pon,nspacedouble=0,j; 1784 PetscReal afill =1.0,afill_tmp; 1785 PetscInt rstart = P->cmap->rstart,rmax,aN=A->cmap->N,Armax; 1786 Mat_SeqAIJ *a_loc,*pdt,*pot; 1787 PetscTable ta; 1788 MatType mtype; 1789 1790 PetscFunctionBegin; 1791 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1792 /* check if matrix local sizes are compatible */ 1793 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); 1794 1795 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1796 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1797 1798 /* create struct Mat_APMPI and attached it to C later */ 1799 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1800 1801 /* get A_loc by taking all local rows of A */ 1802 ierr = MatMPIAIJGetLocalMat(A,MAT_INITIAL_MATRIX,&A_loc);CHKERRQ(ierr); 1803 1804 ptap->A_loc = A_loc; 1805 a_loc = (Mat_SeqAIJ*)(A_loc)->data; 1806 ai = a_loc->i; 1807 aj = a_loc->j; 1808 1809 /* determine symbolic Co=(p->B)^T*A - send to others */ 1810 /*----------------------------------------------------*/ 1811 ierr = MatTransposeSymbolic_SeqAIJ(p->A,&PDt);CHKERRQ(ierr); 1812 pdt = (Mat_SeqAIJ*)PDt->data; 1813 pdti = pdt->i; pdtj = pdt->j; 1814 1815 ierr = MatTransposeSymbolic_SeqAIJ(p->B,&POt);CHKERRQ(ierr); 1816 pot = (Mat_SeqAIJ*)POt->data; 1817 poti = pot->i; potj = pot->j; 1818 1819 /* then, compute symbolic Co = (p->B)^T*A */ 1820 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 1821 >= (num of nonzero rows of C_seq) - pn */ 1822 ierr = PetscMalloc1(pon+1,&coi);CHKERRQ(ierr); 1823 coi[0] = 0; 1824 1825 /* set initial free space to be fill*(nnz(p->B) + nnz(A)) */ 1826 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(poti[pon],ai[am])); 1827 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1828 current_space = free_space; 1829 1830 /* create and initialize a linked list */ 1831 ierr = PetscTableCreate(A->cmap->n + a->B->cmap->N,aN,&ta);CHKERRQ(ierr); 1832 MatRowMergeMax_SeqAIJ(a_loc,am,ta); 1833 ierr = PetscTableGetCount(ta,&Armax);CHKERRQ(ierr); 1834 1835 ierr = PetscLLCondensedCreate_Scalable(Armax,&lnk);CHKERRQ(ierr); 1836 1837 for (i=0; i<pon; i++) { 1838 pnz = poti[i+1] - poti[i]; 1839 ptJ = potj + poti[i]; 1840 for (j=0; j<pnz; j++) { 1841 row = ptJ[j]; /* row of A_loc == col of Pot */ 1842 anz = ai[row+1] - ai[row]; 1843 Jptr = aj + ai[row]; 1844 /* add non-zero cols of AP into the sorted linked list lnk */ 1845 ierr = PetscLLCondensedAddSorted_Scalable(anz,Jptr,lnk);CHKERRQ(ierr); 1846 } 1847 nnz = lnk[0]; 1848 1849 /* If free space is not available, double the total space in the list */ 1850 if (current_space->local_remaining<nnz) { 1851 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1852 nspacedouble++; 1853 } 1854 1855 /* Copy data into free space, and zero out denserows */ 1856 ierr = PetscLLCondensedClean_Scalable(nnz,current_space->array,lnk);CHKERRQ(ierr); 1857 1858 current_space->array += nnz; 1859 current_space->local_used += nnz; 1860 current_space->local_remaining -= nnz; 1861 1862 coi[i+1] = coi[i] + nnz; 1863 } 1864 1865 ierr = PetscMalloc1(coi[pon]+1,&coj);CHKERRQ(ierr); 1866 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 1867 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); /* must destroy to get a new one for C */ 1868 1869 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + ai[am]+1); 1870 if (afill_tmp > afill) afill = afill_tmp; 1871 1872 /* send j-array (coj) of Co to other processors */ 1873 /*----------------------------------------------*/ 1874 /* determine row ownership */ 1875 ierr = PetscNew(&merge);CHKERRQ(ierr); 1876 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 1877 1878 merge->rowmap->n = pn; 1879 merge->rowmap->bs = 1; 1880 1881 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 1882 owners = merge->rowmap->range; 1883 1884 /* determine the number of messages to send, their lengths */ 1885 ierr = PetscCalloc1(size,&len_si);CHKERRQ(ierr); 1886 ierr = PetscCalloc1(size,&merge->len_s);CHKERRQ(ierr); 1887 1888 len_s = merge->len_s; 1889 merge->nsend = 0; 1890 1891 ierr = PetscMalloc1(size+2,&owners_co);CHKERRQ(ierr); 1892 1893 proc = 0; 1894 for (i=0; i<pon; i++) { 1895 while (prmap[i] >= owners[proc+1]) proc++; 1896 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 1897 len_s[proc] += coi[i+1] - coi[i]; 1898 } 1899 1900 len = 0; /* max length of buf_si[] */ 1901 owners_co[0] = 0; 1902 for (proc=0; proc<size; proc++) { 1903 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 1904 if (len_si[proc]) { 1905 merge->nsend++; 1906 len_si[proc] = 2*(len_si[proc] + 1); 1907 len += len_si[proc]; 1908 } 1909 } 1910 1911 /* determine the number and length of messages to receive for coi and coj */ 1912 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 1913 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 1914 1915 /* post the Irecv and Isend of coj */ 1916 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 1917 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 1918 ierr = PetscMalloc1(merge->nsend+1,&swaits);CHKERRQ(ierr); 1919 for (proc=0, k=0; proc<size; proc++) { 1920 if (!len_s[proc]) continue; 1921 i = owners_co[proc]; 1922 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 1923 k++; 1924 } 1925 1926 /* receives and sends of coj are complete */ 1927 ierr = PetscMalloc1(size,&sstatus);CHKERRQ(ierr); 1928 for (i=0; i<merge->nrecv; i++) { 1929 PetscMPIInt icompleted; 1930 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1931 } 1932 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1933 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1934 1935 /* add received column indices into table to update Armax */ 1936 /* Armax can be as large as aN if a P[row,:] is dense, see src/ksp/ksp/tutorials/ex56.c! */ 1937 for (k=0; k<merge->nrecv; k++) {/* k-th received message */ 1938 Jptr = buf_rj[k]; 1939 for (j=0; j<merge->len_r[k]; j++) { 1940 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 1941 } 1942 } 1943 ierr = PetscTableGetCount(ta,&Armax);CHKERRQ(ierr); 1944 /* 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); */ 1945 1946 /* send and recv coi */ 1947 /*-------------------*/ 1948 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 1949 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 1950 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 1951 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 1952 for (proc=0,k=0; proc<size; proc++) { 1953 if (!len_s[proc]) continue; 1954 /* form outgoing message for i-structure: 1955 buf_si[0]: nrows to be sent 1956 [1:nrows]: row index (global) 1957 [nrows+1:2*nrows+1]: i-structure index 1958 */ 1959 /*-------------------------------------------*/ 1960 nrows = len_si[proc]/2 - 1; 1961 buf_si_i = buf_si + nrows+1; 1962 buf_si[0] = nrows; 1963 buf_si_i[0] = 0; 1964 nrows = 0; 1965 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 1966 nzi = coi[i+1] - coi[i]; 1967 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 1968 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 1969 nrows++; 1970 } 1971 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 1972 k++; 1973 buf_si += len_si[proc]; 1974 } 1975 i = merge->nrecv; 1976 while (i--) { 1977 PetscMPIInt icompleted; 1978 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1979 } 1980 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1981 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1982 ierr = PetscFree(len_si);CHKERRQ(ierr); 1983 ierr = PetscFree(len_ri);CHKERRQ(ierr); 1984 ierr = PetscFree(swaits);CHKERRQ(ierr); 1985 ierr = PetscFree(sstatus);CHKERRQ(ierr); 1986 ierr = PetscFree(buf_s);CHKERRQ(ierr); 1987 1988 /* compute the local portion of C (mpi mat) */ 1989 /*------------------------------------------*/ 1990 /* allocate bi array and free space for accumulating nonzero column info */ 1991 ierr = PetscMalloc1(pn+1,&bi);CHKERRQ(ierr); 1992 bi[0] = 0; 1993 1994 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 1995 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(pdti[pn],PetscIntSumTruncate(poti[pon],ai[am]))); 1996 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1997 current_space = free_space; 1998 1999 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 2000 for (k=0; k<merge->nrecv; k++) { 2001 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 2002 nrows = *buf_ri_k[k]; 2003 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 2004 nextci[k] = buf_ri_k[k] + (nrows + 1); /* points to the next i-structure of k-th received i-structure */ 2005 } 2006 2007 ierr = PetscLLCondensedCreate_Scalable(Armax,&lnk);CHKERRQ(ierr); 2008 ierr = MatPreallocateInitialize(comm,pn,A->cmap->n,dnz,onz);CHKERRQ(ierr); 2009 rmax = 0; 2010 for (i=0; i<pn; i++) { 2011 /* add pdt[i,:]*AP into lnk */ 2012 pnz = pdti[i+1] - pdti[i]; 2013 ptJ = pdtj + pdti[i]; 2014 for (j=0; j<pnz; j++) { 2015 row = ptJ[j]; /* row of AP == col of Pt */ 2016 anz = ai[row+1] - ai[row]; 2017 Jptr = aj + ai[row]; 2018 /* add non-zero cols of AP into the sorted linked list lnk */ 2019 ierr = PetscLLCondensedAddSorted_Scalable(anz,Jptr,lnk);CHKERRQ(ierr); 2020 } 2021 2022 /* add received col data into lnk */ 2023 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 2024 if (i == *nextrow[k]) { /* i-th row */ 2025 nzi = *(nextci[k]+1) - *nextci[k]; 2026 Jptr = buf_rj[k] + *nextci[k]; 2027 ierr = PetscLLCondensedAddSorted_Scalable(nzi,Jptr,lnk);CHKERRQ(ierr); 2028 nextrow[k]++; nextci[k]++; 2029 } 2030 } 2031 nnz = lnk[0]; 2032 2033 /* if free space is not available, make more free space */ 2034 if (current_space->local_remaining<nnz) { 2035 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 2036 nspacedouble++; 2037 } 2038 /* copy data into free space, then initialize lnk */ 2039 ierr = PetscLLCondensedClean_Scalable(nnz,current_space->array,lnk);CHKERRQ(ierr); 2040 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 2041 2042 current_space->array += nnz; 2043 current_space->local_used += nnz; 2044 current_space->local_remaining -= nnz; 2045 2046 bi[i+1] = bi[i] + nnz; 2047 if (nnz > rmax) rmax = nnz; 2048 } 2049 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 2050 2051 ierr = PetscMalloc1(bi[pn]+1,&bj);CHKERRQ(ierr); 2052 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 2053 afill_tmp = (PetscReal)bi[pn]/(pdti[pn] + poti[pon] + ai[am]+1); 2054 if (afill_tmp > afill) afill = afill_tmp; 2055 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 2056 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 2057 2058 ierr = MatDestroy(&POt);CHKERRQ(ierr); 2059 ierr = MatDestroy(&PDt);CHKERRQ(ierr); 2060 2061 /* create symbolic parallel matrix C - why cannot be assembled in Numeric part */ 2062 /*-------------------------------------------------------------------------------*/ 2063 ierr = MatSetSizes(C,pn,A->cmap->n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2064 ierr = MatSetBlockSizes(C,PetscAbs(P->cmap->bs),PetscAbs(A->cmap->bs));CHKERRQ(ierr); 2065 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 2066 ierr = MatSetType(C,mtype);CHKERRQ(ierr); 2067 ierr = MatMPIAIJSetPreallocation(C,0,dnz,0,onz);CHKERRQ(ierr); 2068 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 2069 ierr = MatSetBlockSize(C,1);CHKERRQ(ierr); 2070 for (i=0; i<pn; i++) { 2071 row = i + rstart; 2072 nnz = bi[i+1] - bi[i]; 2073 Jptr = bj + bi[i]; 2074 ierr = MatSetValues(C,1,&row,nnz,Jptr,NULL,INSERT_VALUES);CHKERRQ(ierr); 2075 } 2076 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2077 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2078 merge->bi = bi; 2079 merge->bj = bj; 2080 merge->coi = coi; 2081 merge->coj = coj; 2082 merge->buf_ri = buf_ri; 2083 merge->buf_rj = buf_rj; 2084 merge->owners_co = owners_co; 2085 2086 /* attach the supporting struct to C for reuse */ 2087 C->product->data = ptap; 2088 C->product->destroy = MatDestroy_MPIAIJ_PtAP; 2089 ptap->merge = merge; 2090 2091 C->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ; 2092 2093 #if defined(PETSC_USE_INFO) 2094 if (bi[pn] != 0) { 2095 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 2096 ierr = PetscInfo1(C,"Use MatTransposeMatMult(A,B,MatReuse,%g,&C) for best performance.\n",(double)afill);CHKERRQ(ierr); 2097 } else { 2098 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 2099 } 2100 #endif 2101 PetscFunctionReturn(0); 2102 } 2103 2104 /* ---------------------------------------------------------------- */ 2105 static PetscErrorCode MatProductSymbolic_AtB_MPIAIJ_MPIAIJ(Mat C) 2106 { 2107 PetscErrorCode ierr; 2108 Mat_Product *product = C->product; 2109 Mat A=product->A,B=product->B; 2110 PetscReal fill=product->fill; 2111 PetscBool flg; 2112 2113 PetscFunctionBegin; 2114 /* scalable */ 2115 ierr = PetscStrcmp(product->alg,"scalable",&flg);CHKERRQ(ierr); 2116 if (flg) { 2117 ierr = MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 2118 goto next; 2119 } 2120 2121 /* nonscalable */ 2122 ierr = PetscStrcmp(product->alg,"nonscalable",&flg);CHKERRQ(ierr); 2123 if (flg) { 2124 ierr = MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ_nonscalable(A,B,fill,C);CHKERRQ(ierr); 2125 goto next; 2126 } 2127 2128 /* matmatmult */ 2129 ierr = PetscStrcmp(product->alg,"at*b",&flg);CHKERRQ(ierr); 2130 if (flg) { 2131 Mat At; 2132 Mat_APMPI *ptap; 2133 2134 ierr = MatTranspose(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 2135 ierr = MatMatMultSymbolic_MPIAIJ_MPIAIJ(At,B,fill,C);CHKERRQ(ierr); 2136 ptap = (Mat_APMPI*)C->product->data; 2137 if (ptap) { 2138 ptap->Pt = At; 2139 C->product->destroy = MatDestroy_MPIAIJ_PtAP; 2140 } 2141 C->ops->transposematmultnumeric = MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ_matmatmult; 2142 goto next; 2143 } 2144 2145 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatProduct type is not supported"); 2146 2147 next: 2148 C->ops->productnumeric = MatProductNumeric_AtB; 2149 PetscFunctionReturn(0); 2150 } 2151 2152 /* ---------------------------------------------------------------- */ 2153 /* Set options for MatMatMultxxx_MPIAIJ_MPIAIJ */ 2154 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_AB(Mat C) 2155 { 2156 PetscErrorCode ierr; 2157 Mat_Product *product = C->product; 2158 Mat A=product->A,B=product->B; 2159 #if defined(PETSC_HAVE_HYPRE) 2160 const char *algTypes[4] = {"scalable","nonscalable","seqmpi","hypre"}; 2161 PetscInt nalg = 4; 2162 #else 2163 const char *algTypes[3] = {"scalable","nonscalable","seqmpi"}; 2164 PetscInt nalg = 3; 2165 #endif 2166 PetscInt alg = 1; /* set nonscalable algorithm as default */ 2167 PetscBool flg; 2168 MPI_Comm comm; 2169 2170 PetscFunctionBegin; 2171 /* Check matrix local sizes */ 2172 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 2173 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) SETERRQ4(PETSC_COMM_SELF,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); 2174 2175 /* Set "nonscalable" as default algorithm */ 2176 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2177 if (flg) { 2178 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2179 2180 /* Set "scalable" as default if BN and local nonzeros of A and B are large */ 2181 if (B->cmap->N > 100000) { /* may switch to scalable algorithm as default */ 2182 MatInfo Ainfo,Binfo; 2183 PetscInt nz_local; 2184 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 2185 2186 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 2187 ierr = MatGetInfo(B,MAT_LOCAL,&Binfo);CHKERRQ(ierr); 2188 nz_local = (PetscInt)(Ainfo.nz_allocated + Binfo.nz_allocated); 2189 2190 if (B->cmap->N > product->fill*nz_local) alg_scalable_loc = PETSC_TRUE; 2191 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 2192 2193 if (alg_scalable) { 2194 alg = 0; /* scalable algorithm would 50% slower than nonscalable algorithm */ 2195 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2196 ierr = PetscInfo2(B,"Use scalable algorithm, BN %D, fill*nz_allocated %g\n",B->cmap->N,product->fill*nz_local);CHKERRQ(ierr); 2197 } 2198 } 2199 } 2200 2201 /* Get runtime option */ 2202 if (product->api_user) { 2203 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMult","Mat");CHKERRQ(ierr); 2204 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2205 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2206 } else { 2207 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AB","Mat");CHKERRQ(ierr); 2208 ierr = PetscOptionsEList("-matproduct_ab_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2209 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2210 } 2211 if (flg) { 2212 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2213 } 2214 2215 C->ops->productsymbolic = MatProductSymbolic_AB_MPIAIJ_MPIAIJ; 2216 PetscFunctionReturn(0); 2217 } 2218 2219 /* Set options for MatTransposeMatMultXXX_MPIAIJ_MPIAIJ */ 2220 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_AtB(Mat C) 2221 { 2222 PetscErrorCode ierr; 2223 Mat_Product *product = C->product; 2224 Mat A=product->A,B=product->B; 2225 const char *algTypes[3] = {"scalable","nonscalable","at*b"}; 2226 PetscInt nalg = 3; 2227 PetscInt alg = 1; /* set default algorithm */ 2228 PetscBool flg; 2229 MPI_Comm comm; 2230 2231 PetscFunctionBegin; 2232 /* Check matrix local sizes */ 2233 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 2234 if (A->rmap->rstart != B->rmap->rstart || A->rmap->rend != B->rmap->rend) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, A (%D, %D) != B (%D,%D)",A->rmap->rstart,A->rmap->rend,B->rmap->rstart,B->rmap->rend); 2235 2236 /* Set default algorithm */ 2237 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2238 if (flg) { 2239 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2240 } 2241 2242 /* Set "scalable" as default if BN and local nonzeros of A and B are large */ 2243 if (alg && B->cmap->N > 100000) { /* may switch to scalable algorithm as default */ 2244 MatInfo Ainfo,Binfo; 2245 PetscInt nz_local; 2246 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 2247 2248 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 2249 ierr = MatGetInfo(B,MAT_LOCAL,&Binfo);CHKERRQ(ierr); 2250 nz_local = (PetscInt)(Ainfo.nz_allocated + Binfo.nz_allocated); 2251 2252 if (B->cmap->N > product->fill*nz_local) alg_scalable_loc = PETSC_TRUE; 2253 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 2254 2255 if (alg_scalable) { 2256 alg = 0; /* scalable algorithm would 50% slower than nonscalable algorithm */ 2257 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2258 ierr = PetscInfo2(B,"Use scalable algorithm, BN %D, fill*nz_allocated %g\n",B->cmap->N,product->fill*nz_local);CHKERRQ(ierr); 2259 } 2260 } 2261 2262 /* Get runtime option */ 2263 if (product->api_user) { 2264 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatTransposeMatMult","Mat");CHKERRQ(ierr); 2265 ierr = PetscOptionsEList("-mattransposematmult_via","Algorithmic approach","MatTransposeMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2266 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2267 } else { 2268 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AtB","Mat");CHKERRQ(ierr); 2269 ierr = PetscOptionsEList("-matproduct_atb_via","Algorithmic approach","MatTransposeMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2270 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2271 } 2272 if (flg) { 2273 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2274 } 2275 2276 C->ops->productsymbolic = MatProductSymbolic_AtB_MPIAIJ_MPIAIJ; 2277 PetscFunctionReturn(0); 2278 } 2279 2280 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_PtAP(Mat C) 2281 { 2282 PetscErrorCode ierr; 2283 Mat_Product *product = C->product; 2284 Mat A=product->A,P=product->B; 2285 MPI_Comm comm; 2286 PetscBool flg; 2287 PetscInt alg=1; /* set default algorithm */ 2288 #if !defined(PETSC_HAVE_HYPRE) 2289 const char *algTypes[4] = {"scalable","nonscalable","allatonce","allatonce_merged"}; 2290 PetscInt nalg=4; 2291 #else 2292 const char *algTypes[5] = {"scalable","nonscalable","allatonce","allatonce_merged","hypre"}; 2293 PetscInt nalg=5; 2294 #endif 2295 PetscInt pN=P->cmap->N; 2296 2297 PetscFunctionBegin; 2298 /* Check matrix local sizes */ 2299 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 2300 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend) SETERRQ4(PETSC_COMM_SELF,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); 2301 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend) SETERRQ4(PETSC_COMM_SELF,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); 2302 2303 /* Set "nonscalable" as default algorithm */ 2304 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2305 if (flg) { 2306 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2307 2308 /* Set "scalable" as default if BN and local nonzeros of A and B are large */ 2309 if (pN > 100000) { 2310 MatInfo Ainfo,Pinfo; 2311 PetscInt nz_local; 2312 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 2313 2314 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 2315 ierr = MatGetInfo(P,MAT_LOCAL,&Pinfo);CHKERRQ(ierr); 2316 nz_local = (PetscInt)(Ainfo.nz_allocated + Pinfo.nz_allocated); 2317 2318 if (pN > product->fill*nz_local) alg_scalable_loc = PETSC_TRUE; 2319 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 2320 2321 if (alg_scalable) { 2322 alg = 0; /* scalable algorithm would 50% slower than nonscalable algorithm */ 2323 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2324 } 2325 } 2326 } 2327 2328 /* Get runtime option */ 2329 if (product->api_user) { 2330 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatPtAP","Mat");CHKERRQ(ierr); 2331 ierr = PetscOptionsEList("-matptap_via","Algorithmic approach","MatPtAP",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2332 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2333 } else { 2334 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_PtAP","Mat");CHKERRQ(ierr); 2335 ierr = PetscOptionsEList("-matproduct_ptap_via","Algorithmic approach","MatPtAP",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2336 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2337 } 2338 if (flg) { 2339 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2340 } 2341 2342 C->ops->productsymbolic = MatProductSymbolic_PtAP_MPIAIJ_MPIAIJ; 2343 PetscFunctionReturn(0); 2344 } 2345 2346 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_RARt(Mat C) 2347 { 2348 Mat_Product *product = C->product; 2349 Mat A = product->A,R=product->B; 2350 2351 PetscFunctionBegin; 2352 /* Check matrix local sizes */ 2353 if (A->cmap->n != R->cmap->n || A->rmap->n != R->cmap->n) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, A local (%D, %D), R local (%D,%D)",A->rmap->n,A->rmap->n,R->rmap->n,R->cmap->n); 2354 2355 C->ops->productsymbolic = MatProductSymbolic_RARt_MPIAIJ_MPIAIJ; 2356 PetscFunctionReturn(0); 2357 } 2358 2359 /* 2360 Set options for ABC = A*B*C = A*(B*C); ABC's algorithm must be chosen from AB's algorithm 2361 */ 2362 static PetscErrorCode MatProductSetFromOptions_MPIAIJ_ABC(Mat C) 2363 { 2364 PetscErrorCode ierr; 2365 Mat_Product *product = C->product; 2366 PetscBool flg = PETSC_FALSE; 2367 PetscInt alg = 1; /* default algorithm */ 2368 const char *algTypes[3] = {"scalable","nonscalable","seqmpi"}; 2369 PetscInt nalg = 3; 2370 2371 PetscFunctionBegin; 2372 /* Set default algorithm */ 2373 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2374 if (flg) { 2375 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2376 } 2377 2378 /* Get runtime option */ 2379 if (product->api_user) { 2380 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMatMult","Mat");CHKERRQ(ierr); 2381 ierr = PetscOptionsEList("-matmatmatmult_via","Algorithmic approach","MatMatMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2382 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2383 } else { 2384 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_ABC","Mat");CHKERRQ(ierr); 2385 ierr = PetscOptionsEList("-matproduct_abc_via","Algorithmic approach","MatProduct_ABC",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2386 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2387 } 2388 if (flg) { 2389 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2390 } 2391 2392 C->ops->matmatmultsymbolic = MatMatMatMultSymbolic_MPIAIJ_MPIAIJ_MPIAIJ; 2393 C->ops->productsymbolic = MatProductSymbolic_ABC; 2394 PetscFunctionReturn(0); 2395 } 2396 2397 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_MPIAIJ(Mat C) 2398 { 2399 PetscErrorCode ierr; 2400 Mat_Product *product = C->product; 2401 2402 PetscFunctionBegin; 2403 switch (product->type) { 2404 case MATPRODUCT_AB: 2405 ierr = MatProductSetFromOptions_MPIAIJ_AB(C);CHKERRQ(ierr); 2406 break; 2407 case MATPRODUCT_AtB: 2408 ierr = MatProductSetFromOptions_MPIAIJ_AtB(C);CHKERRQ(ierr); 2409 break; 2410 case MATPRODUCT_PtAP: 2411 ierr = MatProductSetFromOptions_MPIAIJ_PtAP(C);CHKERRQ(ierr); 2412 break; 2413 case MATPRODUCT_RARt: 2414 ierr = MatProductSetFromOptions_MPIAIJ_RARt(C);CHKERRQ(ierr); 2415 break; 2416 case MATPRODUCT_ABC: 2417 ierr = MatProductSetFromOptions_MPIAIJ_ABC(C);CHKERRQ(ierr); 2418 break; 2419 default: 2420 break; 2421 } 2422 PetscFunctionReturn(0); 2423 } 2424