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