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