1 2 /* 3 Defines matrix-matrix product routines for pairs of SeqAIJ matrices 4 C = A * B 5 */ 6 7 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 8 #include <../src/mat/utils/freespace.h> 9 #include <petscbt.h> 10 #include <petsc/private/isimpl.h> 11 #include <../src/mat/impls/dense/seq/dense.h> 12 13 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 14 { 15 PetscErrorCode ierr; 16 17 PetscFunctionBegin; 18 if (C->ops->matmultnumeric) { 19 PetscCheckFalse(C->ops->matmultnumeric == MatMatMultNumeric_SeqAIJ_SeqAIJ,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Recursive call"); 20 ierr = (*C->ops->matmultnumeric)(A,B,C);CHKERRQ(ierr); 21 } else { 22 ierr = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(A,B,C);CHKERRQ(ierr); 23 } 24 PetscFunctionReturn(0); 25 } 26 27 /* Modified from MatCreateSeqAIJWithArrays() */ 28 PETSC_INTERN PetscErrorCode MatSetSeqAIJWithArrays_private(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],MatType mtype,Mat mat) 29 { 30 PetscErrorCode ierr; 31 PetscInt ii; 32 Mat_SeqAIJ *aij; 33 PetscBool isseqaij, osingle, ofree_a, ofree_ij; 34 35 PetscFunctionBegin; 36 PetscCheckFalse(m > 0 && i[0],PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 37 ierr = MatSetSizes(mat,m,n,m,n);CHKERRQ(ierr); 38 39 if (!mtype) { 40 ierr = PetscObjectBaseTypeCompare((PetscObject)mat,MATSEQAIJ,&isseqaij);CHKERRQ(ierr); 41 if (!isseqaij) { ierr = MatSetType(mat,MATSEQAIJ);CHKERRQ(ierr); } 42 } else { 43 ierr = MatSetType(mat,mtype);CHKERRQ(ierr); 44 } 45 46 aij = (Mat_SeqAIJ*)(mat)->data; 47 osingle = aij->singlemalloc; 48 ofree_a = aij->free_a; 49 ofree_ij = aij->free_ij; 50 /* changes the free flags */ 51 ierr = MatSeqAIJSetPreallocation_SeqAIJ(mat,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr); 52 53 ierr = PetscFree(aij->ilen);CHKERRQ(ierr); 54 ierr = PetscFree(aij->imax);CHKERRQ(ierr); 55 ierr = PetscMalloc1(m,&aij->imax);CHKERRQ(ierr); 56 ierr = PetscMalloc1(m,&aij->ilen);CHKERRQ(ierr); 57 for (ii=0,aij->nonzerorowcnt=0,aij->rmax = 0; ii<m; ii++) { 58 const PetscInt rnz = i[ii+1] - i[ii]; 59 aij->nonzerorowcnt += !!rnz; 60 aij->rmax = PetscMax(aij->rmax,rnz); 61 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 62 } 63 aij->maxnz = i[m]; 64 aij->nz = i[m]; 65 66 if (osingle) { 67 ierr = PetscFree3(aij->a,aij->j,aij->i);CHKERRQ(ierr); 68 } else { 69 if (ofree_a) { ierr = PetscFree(aij->a);CHKERRQ(ierr); } 70 if (ofree_ij) { ierr = PetscFree(aij->j);CHKERRQ(ierr); } 71 if (ofree_ij) { ierr = PetscFree(aij->i);CHKERRQ(ierr); } 72 } 73 aij->i = i; 74 aij->j = j; 75 aij->a = a; 76 aij->nonew = -1; /* this indicates that inserting a new value in the matrix that generates a new nonzero is an error */ 77 /* default to not retain ownership */ 78 aij->singlemalloc = PETSC_FALSE; 79 aij->free_a = PETSC_FALSE; 80 aij->free_ij = PETSC_FALSE; 81 ierr = MatCheckCompressedRow(mat,aij->nonzerorowcnt,&aij->compressedrow,aij->i,m,0.6);CHKERRQ(ierr); 82 PetscFunctionReturn(0); 83 } 84 85 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 86 { 87 PetscErrorCode ierr; 88 Mat_Product *product = C->product; 89 MatProductAlgorithm alg; 90 PetscBool flg; 91 92 PetscFunctionBegin; 93 if (product) { 94 alg = product->alg; 95 } else { 96 alg = "sorted"; 97 } 98 /* sorted */ 99 ierr = PetscStrcmp(alg,"sorted",&flg);CHKERRQ(ierr); 100 if (flg) { 101 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(A,B,fill,C);CHKERRQ(ierr); 102 PetscFunctionReturn(0); 103 } 104 105 /* scalable */ 106 ierr = PetscStrcmp(alg,"scalable",&flg);CHKERRQ(ierr); 107 if (flg) { 108 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 109 PetscFunctionReturn(0); 110 } 111 112 /* scalable_fast */ 113 ierr = PetscStrcmp(alg,"scalable_fast",&flg);CHKERRQ(ierr); 114 if (flg) { 115 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 116 PetscFunctionReturn(0); 117 } 118 119 /* heap */ 120 ierr = PetscStrcmp(alg,"heap",&flg);CHKERRQ(ierr); 121 if (flg) { 122 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 123 PetscFunctionReturn(0); 124 } 125 126 /* btheap */ 127 ierr = PetscStrcmp(alg,"btheap",&flg);CHKERRQ(ierr); 128 if (flg) { 129 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 130 PetscFunctionReturn(0); 131 } 132 133 /* llcondensed */ 134 ierr = PetscStrcmp(alg,"llcondensed",&flg);CHKERRQ(ierr); 135 if (flg) { 136 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(A,B,fill,C);CHKERRQ(ierr); 137 PetscFunctionReturn(0); 138 } 139 140 /* rowmerge */ 141 ierr = PetscStrcmp(alg,"rowmerge",&flg);CHKERRQ(ierr); 142 if (flg) { 143 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(A,B,fill,C);CHKERRQ(ierr); 144 PetscFunctionReturn(0); 145 } 146 147 #if defined(PETSC_HAVE_HYPRE) 148 ierr = PetscStrcmp(alg,"hypre",&flg);CHKERRQ(ierr); 149 if (flg) { 150 ierr = MatMatMultSymbolic_AIJ_AIJ_wHYPRE(A,B,fill,C);CHKERRQ(ierr); 151 PetscFunctionReturn(0); 152 } 153 #endif 154 155 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Mat Product Algorithm is not supported"); 156 } 157 158 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat A,Mat B,PetscReal fill,Mat C) 159 { 160 PetscErrorCode ierr; 161 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 162 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 163 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 164 PetscReal afill; 165 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 166 PetscTable ta; 167 PetscBT lnkbt; 168 PetscFreeSpaceList free_space=NULL,current_space=NULL; 169 170 PetscFunctionBegin; 171 /* Get ci and cj */ 172 /*---------------*/ 173 /* Allocate ci array, arrays for fill computation and */ 174 /* free space for accumulating nonzero column info */ 175 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 176 ci[0] = 0; 177 178 /* create and initialize a linked list */ 179 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 180 MatRowMergeMax_SeqAIJ(b,bm,ta); 181 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 182 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 183 184 ierr = PetscLLCondensedCreate(Crmax,bn,&lnk,&lnkbt);CHKERRQ(ierr); 185 186 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 187 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 188 189 current_space = free_space; 190 191 /* Determine ci and cj */ 192 for (i=0; i<am; i++) { 193 anzi = ai[i+1] - ai[i]; 194 aj = a->j + ai[i]; 195 for (j=0; j<anzi; j++) { 196 brow = aj[j]; 197 bnzj = bi[brow+1] - bi[brow]; 198 bj = b->j + bi[brow]; 199 /* add non-zero cols of B into the sorted linked list lnk */ 200 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 201 } 202 /* add possible missing diagonal entry */ 203 if (C->force_diagonals) { 204 ierr = PetscLLCondensedAddSorted(1,&i,lnk,lnkbt);CHKERRQ(ierr); 205 } 206 cnzi = lnk[0]; 207 208 /* If free space is not available, make more free space */ 209 /* Double the amount of total space in the list */ 210 if (current_space->local_remaining<cnzi) { 211 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 212 ndouble++; 213 } 214 215 /* Copy data into free space, then initialize lnk */ 216 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 217 218 current_space->array += cnzi; 219 current_space->local_used += cnzi; 220 current_space->local_remaining -= cnzi; 221 222 ci[i+1] = ci[i] + cnzi; 223 } 224 225 /* Column indices are in the list of free space */ 226 /* Allocate space for cj, initialize cj, and */ 227 /* destroy list of free space and other temporary array(s) */ 228 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 229 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 230 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 231 232 /* put together the new symbolic matrix */ 233 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 234 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 235 236 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 237 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 238 c = (Mat_SeqAIJ*)(C->data); 239 c->free_a = PETSC_FALSE; 240 c->free_ij = PETSC_TRUE; 241 c->nonew = 0; 242 243 /* fast, needs non-scalable O(bn) array 'abdense' */ 244 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 245 246 /* set MatInfo */ 247 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 248 if (afill < 1.0) afill = 1.0; 249 C->info.mallocs = ndouble; 250 C->info.fill_ratio_given = fill; 251 C->info.fill_ratio_needed = afill; 252 253 #if defined(PETSC_USE_INFO) 254 if (ci[am]) { 255 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 256 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 257 } else { 258 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 259 } 260 #endif 261 PetscFunctionReturn(0); 262 } 263 264 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat A,Mat B,Mat C) 265 { 266 PetscErrorCode ierr; 267 PetscLogDouble flops=0.0; 268 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 269 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 270 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 271 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 272 PetscInt am =A->rmap->n,cm=C->rmap->n; 273 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 274 PetscScalar *ca,valtmp; 275 PetscScalar *ab_dense; 276 PetscContainer cab_dense; 277 const PetscScalar *aa,*ba,*baj; 278 279 PetscFunctionBegin; 280 ierr = MatSeqAIJGetArrayRead(A,&aa);CHKERRQ(ierr); 281 ierr = MatSeqAIJGetArrayRead(B,&ba);CHKERRQ(ierr); 282 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 283 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 284 c->a = ca; 285 c->free_a = PETSC_TRUE; 286 } else ca = c->a; 287 288 /* TODO this should be done in the symbolic phase */ 289 /* However, this function is so heavily used (sometimes in an hidden way through multnumeric function pointers 290 that is hard to eradicate) */ 291 ierr = PetscObjectQuery((PetscObject)C,"__PETSc__ab_dense",(PetscObject*)&cab_dense);CHKERRQ(ierr); 292 if (!cab_dense) { 293 ierr = PetscMalloc1(B->cmap->N,&ab_dense);CHKERRQ(ierr); 294 ierr = PetscContainerCreate(PETSC_COMM_SELF,&cab_dense);CHKERRQ(ierr); 295 ierr = PetscContainerSetPointer(cab_dense,ab_dense);CHKERRQ(ierr); 296 ierr = PetscContainerSetUserDestroy(cab_dense,PetscContainerUserDestroyDefault);CHKERRQ(ierr); 297 ierr = PetscObjectCompose((PetscObject)C,"__PETSc__ab_dense",(PetscObject)cab_dense);CHKERRQ(ierr); 298 ierr = PetscObjectDereference((PetscObject)cab_dense);CHKERRQ(ierr); 299 } 300 ierr = PetscContainerGetPointer(cab_dense,(void**)&ab_dense);CHKERRQ(ierr); 301 ierr = PetscArrayzero(ab_dense,B->cmap->N);CHKERRQ(ierr); 302 303 /* clean old values in C */ 304 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 305 /* Traverse A row-wise. */ 306 /* Build the ith row in C by summing over nonzero columns in A, */ 307 /* the rows of B corresponding to nonzeros of A. */ 308 for (i=0; i<am; i++) { 309 anzi = ai[i+1] - ai[i]; 310 for (j=0; j<anzi; j++) { 311 brow = aj[j]; 312 bnzi = bi[brow+1] - bi[brow]; 313 bjj = bj + bi[brow]; 314 baj = ba + bi[brow]; 315 /* perform dense axpy */ 316 valtmp = aa[j]; 317 for (k=0; k<bnzi; k++) { 318 ab_dense[bjj[k]] += valtmp*baj[k]; 319 } 320 flops += 2*bnzi; 321 } 322 aj += anzi; aa += anzi; 323 324 cnzi = ci[i+1] - ci[i]; 325 for (k=0; k<cnzi; k++) { 326 ca[k] += ab_dense[cj[k]]; 327 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 328 } 329 flops += cnzi; 330 cj += cnzi; ca += cnzi; 331 } 332 #if defined(PETSC_HAVE_DEVICE) 333 if (C->offloadmask != PETSC_OFFLOAD_UNALLOCATED) C->offloadmask = PETSC_OFFLOAD_CPU; 334 #endif 335 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 336 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 337 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 338 ierr = MatSeqAIJRestoreArrayRead(A,&aa);CHKERRQ(ierr); 339 ierr = MatSeqAIJRestoreArrayRead(B,&ba);CHKERRQ(ierr); 340 PetscFunctionReturn(0); 341 } 342 343 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 344 { 345 PetscErrorCode ierr; 346 PetscLogDouble flops=0.0; 347 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 348 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 349 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 350 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 351 PetscInt am = A->rmap->N,cm=C->rmap->N; 352 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 353 PetscScalar *ca=c->a,valtmp; 354 const PetscScalar *aa,*ba,*baj; 355 PetscInt nextb; 356 357 PetscFunctionBegin; 358 ierr = MatSeqAIJGetArrayRead(A,&aa);CHKERRQ(ierr); 359 ierr = MatSeqAIJGetArrayRead(B,&ba);CHKERRQ(ierr); 360 if (!ca) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 361 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 362 c->a = ca; 363 c->free_a = PETSC_TRUE; 364 } 365 366 /* clean old values in C */ 367 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 368 /* Traverse A row-wise. */ 369 /* Build the ith row in C by summing over nonzero columns in A, */ 370 /* the rows of B corresponding to nonzeros of A. */ 371 for (i=0; i<am; i++) { 372 anzi = ai[i+1] - ai[i]; 373 cnzi = ci[i+1] - ci[i]; 374 for (j=0; j<anzi; j++) { 375 brow = aj[j]; 376 bnzi = bi[brow+1] - bi[brow]; 377 bjj = bj + bi[brow]; 378 baj = ba + bi[brow]; 379 /* perform sparse axpy */ 380 valtmp = aa[j]; 381 nextb = 0; 382 for (k=0; nextb<bnzi; k++) { 383 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 384 ca[k] += valtmp*baj[nextb++]; 385 } 386 } 387 flops += 2*bnzi; 388 } 389 aj += anzi; aa += anzi; 390 cj += cnzi; ca += cnzi; 391 } 392 #if defined(PETSC_HAVE_DEVICE) 393 if (C->offloadmask != PETSC_OFFLOAD_UNALLOCATED) C->offloadmask = PETSC_OFFLOAD_CPU; 394 #endif 395 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 396 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 397 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 398 ierr = MatSeqAIJRestoreArrayRead(A,&aa);CHKERRQ(ierr); 399 ierr = MatSeqAIJRestoreArrayRead(B,&ba);CHKERRQ(ierr); 400 PetscFunctionReturn(0); 401 } 402 403 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat C) 404 { 405 PetscErrorCode ierr; 406 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 407 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 408 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 409 MatScalar *ca; 410 PetscReal afill; 411 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 412 PetscTable ta; 413 PetscFreeSpaceList free_space=NULL,current_space=NULL; 414 415 PetscFunctionBegin; 416 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 417 /*-----------------------------------------------------------------------------------------*/ 418 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 419 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 420 ci[0] = 0; 421 422 /* create and initialize a linked list */ 423 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 424 MatRowMergeMax_SeqAIJ(b,bm,ta); 425 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 426 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 427 428 ierr = PetscLLCondensedCreate_fast(Crmax,&lnk);CHKERRQ(ierr); 429 430 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 431 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 432 current_space = free_space; 433 434 /* Determine ci and cj */ 435 for (i=0; i<am; i++) { 436 anzi = ai[i+1] - ai[i]; 437 aj = a->j + ai[i]; 438 for (j=0; j<anzi; j++) { 439 brow = aj[j]; 440 bnzj = bi[brow+1] - bi[brow]; 441 bj = b->j + bi[brow]; 442 /* add non-zero cols of B into the sorted linked list lnk */ 443 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 444 } 445 /* add possible missing diagonal entry */ 446 if (C->force_diagonals) { 447 ierr = PetscLLCondensedAddSorted_fast(1,&i,lnk);CHKERRQ(ierr); 448 } 449 cnzi = lnk[1]; 450 451 /* If free space is not available, make more free space */ 452 /* Double the amount of total space in the list */ 453 if (current_space->local_remaining<cnzi) { 454 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 455 ndouble++; 456 } 457 458 /* Copy data into free space, then initialize lnk */ 459 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 460 461 current_space->array += cnzi; 462 current_space->local_used += cnzi; 463 current_space->local_remaining -= cnzi; 464 465 ci[i+1] = ci[i] + cnzi; 466 } 467 468 /* Column indices are in the list of free space */ 469 /* Allocate space for cj, initialize cj, and */ 470 /* destroy list of free space and other temporary array(s) */ 471 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 472 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 473 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 474 475 /* Allocate space for ca */ 476 ierr = PetscCalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 477 478 /* put together the new symbolic matrix */ 479 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 480 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 481 482 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 483 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 484 c = (Mat_SeqAIJ*)(C->data); 485 c->free_a = PETSC_TRUE; 486 c->free_ij = PETSC_TRUE; 487 c->nonew = 0; 488 489 /* slower, less memory */ 490 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; 491 492 /* set MatInfo */ 493 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 494 if (afill < 1.0) afill = 1.0; 495 C->info.mallocs = ndouble; 496 C->info.fill_ratio_given = fill; 497 C->info.fill_ratio_needed = afill; 498 499 #if defined(PETSC_USE_INFO) 500 if (ci[am]) { 501 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 502 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 503 } else { 504 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 505 } 506 #endif 507 PetscFunctionReturn(0); 508 } 509 510 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat C) 511 { 512 PetscErrorCode ierr; 513 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 514 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 515 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 516 MatScalar *ca; 517 PetscReal afill; 518 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 519 PetscTable ta; 520 PetscFreeSpaceList free_space=NULL,current_space=NULL; 521 522 PetscFunctionBegin; 523 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 524 /*---------------------------------------------------------------------------------------------*/ 525 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 526 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 527 ci[0] = 0; 528 529 /* create and initialize a linked list */ 530 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 531 MatRowMergeMax_SeqAIJ(b,bm,ta); 532 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 533 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 534 ierr = PetscLLCondensedCreate_Scalable(Crmax,&lnk);CHKERRQ(ierr); 535 536 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 537 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 538 current_space = free_space; 539 540 /* Determine ci and cj */ 541 for (i=0; i<am; i++) { 542 anzi = ai[i+1] - ai[i]; 543 aj = a->j + ai[i]; 544 for (j=0; j<anzi; j++) { 545 brow = aj[j]; 546 bnzj = bi[brow+1] - bi[brow]; 547 bj = b->j + bi[brow]; 548 /* add non-zero cols of B into the sorted linked list lnk */ 549 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 550 } 551 /* add possible missing diagonal entry */ 552 if (C->force_diagonals) { 553 ierr = PetscLLCondensedAddSorted_Scalable(1,&i,lnk);CHKERRQ(ierr); 554 } 555 556 cnzi = lnk[0]; 557 558 /* If free space is not available, make more free space */ 559 /* Double the amount of total space in the list */ 560 if (current_space->local_remaining<cnzi) { 561 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 562 ndouble++; 563 } 564 565 /* Copy data into free space, then initialize lnk */ 566 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 567 568 current_space->array += cnzi; 569 current_space->local_used += cnzi; 570 current_space->local_remaining -= cnzi; 571 572 ci[i+1] = ci[i] + cnzi; 573 } 574 575 /* Column indices are in the list of free space */ 576 /* Allocate space for cj, initialize cj, and */ 577 /* destroy list of free space and other temporary array(s) */ 578 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 579 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 580 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 581 582 /* Allocate space for ca */ 583 /*-----------------------*/ 584 ierr = PetscCalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 585 586 /* put together the new symbolic matrix */ 587 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 588 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 589 590 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 591 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 592 c = (Mat_SeqAIJ*)(C->data); 593 c->free_a = PETSC_TRUE; 594 c->free_ij = PETSC_TRUE; 595 c->nonew = 0; 596 597 /* slower, less memory */ 598 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; 599 600 /* set MatInfo */ 601 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 602 if (afill < 1.0) afill = 1.0; 603 C->info.mallocs = ndouble; 604 C->info.fill_ratio_given = fill; 605 C->info.fill_ratio_needed = afill; 606 607 #if defined(PETSC_USE_INFO) 608 if (ci[am]) { 609 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 610 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 611 } else { 612 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 613 } 614 #endif 615 PetscFunctionReturn(0); 616 } 617 618 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat C) 619 { 620 PetscErrorCode ierr; 621 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 622 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 623 PetscInt *ci,*cj,*bb; 624 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 625 PetscReal afill; 626 PetscInt i,j,col,ndouble = 0; 627 PetscFreeSpaceList free_space=NULL,current_space=NULL; 628 PetscHeap h; 629 630 PetscFunctionBegin; 631 /* Get ci and cj - by merging sorted rows using a heap */ 632 /*---------------------------------------------------------------------------------------------*/ 633 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 634 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 635 ci[0] = 0; 636 637 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 638 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 639 current_space = free_space; 640 641 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 642 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 643 644 /* Determine ci and cj */ 645 for (i=0; i<am; i++) { 646 const PetscInt anzi = ai[i+1] - ai[i]; /* number of nonzeros in this row of A, this is the number of rows of B that we merge */ 647 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 648 ci[i+1] = ci[i]; 649 /* Populate the min heap */ 650 for (j=0; j<anzi; j++) { 651 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 652 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 653 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 654 } 655 } 656 /* Pick off the min element, adding it to free space */ 657 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 658 while (j >= 0) { 659 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 660 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 661 ndouble++; 662 } 663 *(current_space->array++) = col; 664 current_space->local_used++; 665 current_space->local_remaining--; 666 ci[i+1]++; 667 668 /* stash if anything else remains in this row of B */ 669 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 670 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 671 PetscInt j2,col2; 672 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 673 if (col2 != col) break; 674 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 675 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 676 } 677 /* Put any stashed elements back into the min heap */ 678 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 679 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 680 } 681 } 682 ierr = PetscFree(bb);CHKERRQ(ierr); 683 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 684 685 /* Column indices are in the list of free space */ 686 /* Allocate space for cj, initialize cj, and */ 687 /* destroy list of free space and other temporary array(s) */ 688 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 689 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 690 691 /* put together the new symbolic matrix */ 692 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 693 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 694 695 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 696 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 697 c = (Mat_SeqAIJ*)(C->data); 698 c->free_a = PETSC_TRUE; 699 c->free_ij = PETSC_TRUE; 700 c->nonew = 0; 701 702 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 703 704 /* set MatInfo */ 705 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 706 if (afill < 1.0) afill = 1.0; 707 C->info.mallocs = ndouble; 708 C->info.fill_ratio_given = fill; 709 C->info.fill_ratio_needed = afill; 710 711 #if defined(PETSC_USE_INFO) 712 if (ci[am]) { 713 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 714 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 715 } else { 716 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 717 } 718 #endif 719 PetscFunctionReturn(0); 720 } 721 722 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat C) 723 { 724 PetscErrorCode ierr; 725 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 726 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 727 PetscInt *ci,*cj,*bb; 728 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 729 PetscReal afill; 730 PetscInt i,j,col,ndouble = 0; 731 PetscFreeSpaceList free_space=NULL,current_space=NULL; 732 PetscHeap h; 733 PetscBT bt; 734 735 PetscFunctionBegin; 736 /* Get ci and cj - using a heap for the sorted rows, but use BT so that each index is only added once */ 737 /*---------------------------------------------------------------------------------------------*/ 738 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 739 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 740 ci[0] = 0; 741 742 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 743 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 744 745 current_space = free_space; 746 747 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 748 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 749 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 750 751 /* Determine ci and cj */ 752 for (i=0; i<am; i++) { 753 const PetscInt anzi = ai[i+1] - ai[i]; /* number of nonzeros in this row of A, this is the number of rows of B that we merge */ 754 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 755 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 756 ci[i+1] = ci[i]; 757 /* Populate the min heap */ 758 for (j=0; j<anzi; j++) { 759 PetscInt brow = acol[j]; 760 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 761 PetscInt bcol = bj[bb[j]]; 762 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 763 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 764 bb[j]++; 765 break; 766 } 767 } 768 } 769 /* Pick off the min element, adding it to free space */ 770 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 771 while (j >= 0) { 772 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 773 fptr = NULL; /* need PetscBTMemzero */ 774 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 775 ndouble++; 776 } 777 *(current_space->array++) = col; 778 current_space->local_used++; 779 current_space->local_remaining--; 780 ci[i+1]++; 781 782 /* stash if anything else remains in this row of B */ 783 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 784 PetscInt bcol = bj[bb[j]]; 785 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 786 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 787 bb[j]++; 788 break; 789 } 790 } 791 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 792 } 793 if (fptr) { /* Clear the bits for this row */ 794 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 795 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 796 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 797 } 798 } 799 ierr = PetscFree(bb);CHKERRQ(ierr); 800 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 801 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 802 803 /* Column indices are in the list of free space */ 804 /* Allocate space for cj, initialize cj, and */ 805 /* destroy list of free space and other temporary array(s) */ 806 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 807 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 808 809 /* put together the new symbolic matrix */ 810 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 811 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 812 813 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 814 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 815 c = (Mat_SeqAIJ*)(C->data); 816 c->free_a = PETSC_TRUE; 817 c->free_ij = PETSC_TRUE; 818 c->nonew = 0; 819 820 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 821 822 /* set MatInfo */ 823 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 824 if (afill < 1.0) afill = 1.0; 825 C->info.mallocs = ndouble; 826 C->info.fill_ratio_given = fill; 827 C->info.fill_ratio_needed = afill; 828 829 #if defined(PETSC_USE_INFO) 830 if (ci[am]) { 831 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 832 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 833 } else { 834 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 835 } 836 #endif 837 PetscFunctionReturn(0); 838 } 839 840 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat A,Mat B,PetscReal fill,Mat C) 841 { 842 PetscErrorCode ierr; 843 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 844 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j,*inputi,*inputj,*inputcol,*inputcol_L1; 845 PetscInt *ci,*cj,*outputj,worki_L1[9],worki_L2[9]; 846 PetscInt c_maxmem,a_maxrownnz=0,a_rownnz; 847 const PetscInt workcol[8]={0,1,2,3,4,5,6,7}; 848 const PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 849 const PetscInt *brow_ptr[8],*brow_end[8]; 850 PetscInt window[8]; 851 PetscInt window_min,old_window_min,ci_nnz,outputi_nnz=0,L1_nrows,L2_nrows; 852 PetscInt i,k,ndouble=0,L1_rowsleft,rowsleft; 853 PetscReal afill; 854 PetscInt *workj_L1,*workj_L2,*workj_L3; 855 PetscInt L1_nnz,L2_nnz; 856 857 /* Step 1: Get upper bound on memory required for allocation. 858 Because of the way virtual memory works, 859 only the memory pages that are actually needed will be physically allocated. */ 860 PetscFunctionBegin; 861 ierr = PetscMalloc1(am+1,&ci);CHKERRQ(ierr); 862 for (i=0; i<am; i++) { 863 const PetscInt anzi = ai[i+1] - ai[i]; /* number of nonzeros in this row of A, this is the number of rows of B that we merge */ 864 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 865 a_rownnz = 0; 866 for (k=0; k<anzi; ++k) { 867 a_rownnz += bi[acol[k]+1] - bi[acol[k]]; 868 if (a_rownnz > bn) { 869 a_rownnz = bn; 870 break; 871 } 872 } 873 a_maxrownnz = PetscMax(a_maxrownnz, a_rownnz); 874 } 875 /* temporary work areas for merging rows */ 876 ierr = PetscMalloc1(a_maxrownnz*8,&workj_L1);CHKERRQ(ierr); 877 ierr = PetscMalloc1(a_maxrownnz*8,&workj_L2);CHKERRQ(ierr); 878 ierr = PetscMalloc1(a_maxrownnz,&workj_L3);CHKERRQ(ierr); 879 880 /* This should be enough for almost all matrices. If not, memory is reallocated later. */ 881 c_maxmem = 8*(ai[am]+bi[bm]); 882 /* Step 2: Populate pattern for C */ 883 ierr = PetscMalloc1(c_maxmem,&cj);CHKERRQ(ierr); 884 885 ci_nnz = 0; 886 ci[0] = 0; 887 worki_L1[0] = 0; 888 worki_L2[0] = 0; 889 for (i=0; i<am; i++) { 890 const PetscInt anzi = ai[i+1] - ai[i]; /* number of nonzeros in this row of A, this is the number of rows of B that we merge */ 891 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 892 rowsleft = anzi; 893 inputcol_L1 = acol; 894 L2_nnz = 0; 895 L2_nrows = 1; /* Number of rows to be merged on Level 3. output of L3 already exists -> initial value 1 */ 896 worki_L2[1] = 0; 897 outputi_nnz = 0; 898 899 /* If the number of indices in C so far + the max number of columns in the next row > c_maxmem -> allocate more memory */ 900 while (ci_nnz+a_maxrownnz > c_maxmem) { 901 c_maxmem *= 2; 902 ndouble++; 903 ierr = PetscRealloc(sizeof(PetscInt)*c_maxmem,&cj);CHKERRQ(ierr); 904 } 905 906 while (rowsleft) { 907 L1_rowsleft = PetscMin(64, rowsleft); /* In the inner loop max 64 rows of B can be merged */ 908 L1_nrows = 0; 909 L1_nnz = 0; 910 inputcol = inputcol_L1; 911 inputi = bi; 912 inputj = bj; 913 914 /* The following macro is used to specialize for small rows in A. 915 This helps with compiler unrolling, improving performance substantially. 916 Input: inputj inputi inputcol bn 917 Output: outputj outputi_nnz */ 918 #define MatMatMultSymbolic_RowMergeMacro(ANNZ) \ 919 window_min = bn; \ 920 outputi_nnz = 0; \ 921 for (k=0; k<ANNZ; ++k) { \ 922 brow_ptr[k] = inputj + inputi[inputcol[k]]; \ 923 brow_end[k] = inputj + inputi[inputcol[k]+1]; \ 924 window[k] = (brow_ptr[k] != brow_end[k]) ? *brow_ptr[k] : bn; \ 925 window_min = PetscMin(window[k], window_min); \ 926 } \ 927 while (window_min < bn) { \ 928 outputj[outputi_nnz++] = window_min; \ 929 /* advance front and compute new minimum */ \ 930 old_window_min = window_min; \ 931 window_min = bn; \ 932 for (k=0; k<ANNZ; ++k) { \ 933 if (window[k] == old_window_min) { \ 934 brow_ptr[k]++; \ 935 window[k] = (brow_ptr[k] != brow_end[k]) ? *brow_ptr[k] : bn; \ 936 } \ 937 window_min = PetscMin(window[k], window_min); \ 938 } \ 939 } 940 941 /************** L E V E L 1 ***************/ 942 /* Merge up to 8 rows of B to L1 work array*/ 943 while (L1_rowsleft) { 944 outputi_nnz = 0; 945 if (anzi > 8) outputj = workj_L1 + L1_nnz; /* Level 1 rowmerge*/ 946 else outputj = cj + ci_nnz; /* Merge directly to C */ 947 948 switch (L1_rowsleft) { 949 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 950 brow_end[0] = inputj + inputi[inputcol[0]+1]; 951 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 952 inputcol += L1_rowsleft; 953 rowsleft -= L1_rowsleft; 954 L1_rowsleft = 0; 955 break; 956 case 2: MatMatMultSymbolic_RowMergeMacro(2); 957 inputcol += L1_rowsleft; 958 rowsleft -= L1_rowsleft; 959 L1_rowsleft = 0; 960 break; 961 case 3: MatMatMultSymbolic_RowMergeMacro(3); 962 inputcol += L1_rowsleft; 963 rowsleft -= L1_rowsleft; 964 L1_rowsleft = 0; 965 break; 966 case 4: MatMatMultSymbolic_RowMergeMacro(4); 967 inputcol += L1_rowsleft; 968 rowsleft -= L1_rowsleft; 969 L1_rowsleft = 0; 970 break; 971 case 5: MatMatMultSymbolic_RowMergeMacro(5); 972 inputcol += L1_rowsleft; 973 rowsleft -= L1_rowsleft; 974 L1_rowsleft = 0; 975 break; 976 case 6: MatMatMultSymbolic_RowMergeMacro(6); 977 inputcol += L1_rowsleft; 978 rowsleft -= L1_rowsleft; 979 L1_rowsleft = 0; 980 break; 981 case 7: MatMatMultSymbolic_RowMergeMacro(7); 982 inputcol += L1_rowsleft; 983 rowsleft -= L1_rowsleft; 984 L1_rowsleft = 0; 985 break; 986 default: MatMatMultSymbolic_RowMergeMacro(8); 987 inputcol += 8; 988 rowsleft -= 8; 989 L1_rowsleft -= 8; 990 break; 991 } 992 inputcol_L1 = inputcol; 993 L1_nnz += outputi_nnz; 994 worki_L1[++L1_nrows] = L1_nnz; 995 } 996 997 /********************** L E V E L 2 ************************/ 998 /* Merge from L1 work array to either C or to L2 work array */ 999 if (anzi > 8) { 1000 inputi = worki_L1; 1001 inputj = workj_L1; 1002 inputcol = workcol; 1003 outputi_nnz = 0; 1004 1005 if (anzi <= 64) outputj = cj + ci_nnz; /* Merge from L1 work array to C */ 1006 else outputj = workj_L2 + L2_nnz; /* Merge from L1 work array to L2 work array */ 1007 1008 switch (L1_nrows) { 1009 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 1010 brow_end[0] = inputj + inputi[inputcol[0]+1]; 1011 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 1012 break; 1013 case 2: MatMatMultSymbolic_RowMergeMacro(2); break; 1014 case 3: MatMatMultSymbolic_RowMergeMacro(3); break; 1015 case 4: MatMatMultSymbolic_RowMergeMacro(4); break; 1016 case 5: MatMatMultSymbolic_RowMergeMacro(5); break; 1017 case 6: MatMatMultSymbolic_RowMergeMacro(6); break; 1018 case 7: MatMatMultSymbolic_RowMergeMacro(7); break; 1019 case 8: MatMatMultSymbolic_RowMergeMacro(8); break; 1020 default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatMult logic error: Not merging 1-8 rows from L1 work array!"); 1021 } 1022 L2_nnz += outputi_nnz; 1023 worki_L2[++L2_nrows] = L2_nnz; 1024 1025 /************************ L E V E L 3 **********************/ 1026 /* Merge from L2 work array to either C or to L2 work array */ 1027 if (anzi > 64 && (L2_nrows == 8 || rowsleft == 0)) { 1028 inputi = worki_L2; 1029 inputj = workj_L2; 1030 inputcol = workcol; 1031 outputi_nnz = 0; 1032 if (rowsleft) outputj = workj_L3; 1033 else outputj = cj + ci_nnz; 1034 switch (L2_nrows) { 1035 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 1036 brow_end[0] = inputj + inputi[inputcol[0]+1]; 1037 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 1038 break; 1039 case 2: MatMatMultSymbolic_RowMergeMacro(2); break; 1040 case 3: MatMatMultSymbolic_RowMergeMacro(3); break; 1041 case 4: MatMatMultSymbolic_RowMergeMacro(4); break; 1042 case 5: MatMatMultSymbolic_RowMergeMacro(5); break; 1043 case 6: MatMatMultSymbolic_RowMergeMacro(6); break; 1044 case 7: MatMatMultSymbolic_RowMergeMacro(7); break; 1045 case 8: MatMatMultSymbolic_RowMergeMacro(8); break; 1046 default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatMult logic error: Not merging 1-8 rows from L2 work array!"); 1047 } 1048 L2_nrows = 1; 1049 L2_nnz = outputi_nnz; 1050 worki_L2[1] = outputi_nnz; 1051 /* Copy to workj_L2 */ 1052 if (rowsleft) { 1053 for (k=0; k<outputi_nnz; ++k) workj_L2[k] = outputj[k]; 1054 } 1055 } 1056 } 1057 } /* while (rowsleft) */ 1058 #undef MatMatMultSymbolic_RowMergeMacro 1059 1060 /* terminate current row */ 1061 ci_nnz += outputi_nnz; 1062 ci[i+1] = ci_nnz; 1063 } 1064 1065 /* Step 3: Create the new symbolic matrix */ 1066 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 1067 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 1068 1069 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 1070 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 1071 c = (Mat_SeqAIJ*)(C->data); 1072 c->free_a = PETSC_TRUE; 1073 c->free_ij = PETSC_TRUE; 1074 c->nonew = 0; 1075 1076 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 1077 1078 /* set MatInfo */ 1079 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 1080 if (afill < 1.0) afill = 1.0; 1081 C->info.mallocs = ndouble; 1082 C->info.fill_ratio_given = fill; 1083 C->info.fill_ratio_needed = afill; 1084 1085 #if defined(PETSC_USE_INFO) 1086 if (ci[am]) { 1087 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 1088 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1089 } else { 1090 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 1091 } 1092 #endif 1093 1094 /* Step 4: Free temporary work areas */ 1095 ierr = PetscFree(workj_L1);CHKERRQ(ierr); 1096 ierr = PetscFree(workj_L2);CHKERRQ(ierr); 1097 ierr = PetscFree(workj_L3);CHKERRQ(ierr); 1098 PetscFunctionReturn(0); 1099 } 1100 1101 /* concatenate unique entries and then sort */ 1102 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat A,Mat B,PetscReal fill,Mat C) 1103 { 1104 PetscErrorCode ierr; 1105 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 1106 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 1107 PetscInt *ci,*cj,bcol; 1108 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 1109 PetscReal afill; 1110 PetscInt i,j,ndouble = 0; 1111 PetscSegBuffer seg,segrow; 1112 char *seen; 1113 1114 PetscFunctionBegin; 1115 ierr = PetscMalloc1(am+1,&ci);CHKERRQ(ierr); 1116 ci[0] = 0; 1117 1118 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 1119 ierr = PetscSegBufferCreate(sizeof(PetscInt),(PetscInt)(fill*(ai[am]+bi[bm])),&seg);CHKERRQ(ierr); 1120 ierr = PetscSegBufferCreate(sizeof(PetscInt),100,&segrow);CHKERRQ(ierr); 1121 ierr = PetscCalloc1(bn,&seen);CHKERRQ(ierr); 1122 1123 /* Determine ci and cj */ 1124 for (i=0; i<am; i++) { 1125 const PetscInt anzi = ai[i+1] - ai[i]; /* number of nonzeros in this row of A, this is the number of rows of B that we merge */ 1126 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 1127 PetscInt packlen = 0,*PETSC_RESTRICT crow; 1128 1129 /* Pack segrow */ 1130 for (j=0; j<anzi; j++) { 1131 PetscInt brow = acol[j],bjstart = bi[brow],bjend = bi[brow+1],k; 1132 for (k=bjstart; k<bjend; k++) { 1133 bcol = bj[k]; 1134 if (!seen[bcol]) { /* new entry */ 1135 PetscInt *PETSC_RESTRICT slot; 1136 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 1137 *slot = bcol; 1138 seen[bcol] = 1; 1139 packlen++; 1140 } 1141 } 1142 } 1143 1144 /* Check i-th diagonal entry */ 1145 if (C->force_diagonals && !seen[i]) { 1146 PetscInt *PETSC_RESTRICT slot; 1147 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 1148 *slot = i; 1149 seen[i] = 1; 1150 packlen++; 1151 } 1152 1153 ierr = PetscSegBufferGetInts(seg,packlen,&crow);CHKERRQ(ierr); 1154 ierr = PetscSegBufferExtractTo(segrow,crow);CHKERRQ(ierr); 1155 ierr = PetscSortInt(packlen,crow);CHKERRQ(ierr); 1156 ci[i+1] = ci[i] + packlen; 1157 for (j=0; j<packlen; j++) seen[crow[j]] = 0; 1158 } 1159 ierr = PetscSegBufferDestroy(&segrow);CHKERRQ(ierr); 1160 ierr = PetscFree(seen);CHKERRQ(ierr); 1161 1162 /* Column indices are in the segmented buffer */ 1163 ierr = PetscSegBufferExtractAlloc(seg,&cj);CHKERRQ(ierr); 1164 ierr = PetscSegBufferDestroy(&seg);CHKERRQ(ierr); 1165 1166 /* put together the new symbolic matrix */ 1167 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 1168 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 1169 1170 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 1171 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 1172 c = (Mat_SeqAIJ*)(C->data); 1173 c->free_a = PETSC_TRUE; 1174 c->free_ij = PETSC_TRUE; 1175 c->nonew = 0; 1176 1177 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 1178 1179 /* set MatInfo */ 1180 afill = (PetscReal)ci[am]/PetscMax(ai[am]+bi[bm],1) + 1.e-5; 1181 if (afill < 1.0) afill = 1.0; 1182 C->info.mallocs = ndouble; 1183 C->info.fill_ratio_given = fill; 1184 C->info.fill_ratio_needed = afill; 1185 1186 #if defined(PETSC_USE_INFO) 1187 if (ci[am]) { 1188 ierr = PetscInfo(C,"Reallocs %" PetscInt_FMT "; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 1189 ierr = PetscInfo(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1190 } else { 1191 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 1192 } 1193 #endif 1194 PetscFunctionReturn(0); 1195 } 1196 1197 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(void *data) 1198 { 1199 PetscErrorCode ierr; 1200 Mat_MatMatTransMult *abt=(Mat_MatMatTransMult *)data; 1201 1202 PetscFunctionBegin; 1203 ierr = MatTransposeColoringDestroy(&abt->matcoloring);CHKERRQ(ierr); 1204 ierr = MatDestroy(&abt->Bt_den);CHKERRQ(ierr); 1205 ierr = MatDestroy(&abt->ABt_den);CHKERRQ(ierr); 1206 ierr = PetscFree(abt);CHKERRQ(ierr); 1207 PetscFunctionReturn(0); 1208 } 1209 1210 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 1211 { 1212 PetscErrorCode ierr; 1213 Mat Bt; 1214 PetscInt *bti,*btj; 1215 Mat_MatMatTransMult *abt; 1216 Mat_Product *product = C->product; 1217 char *alg; 1218 1219 PetscFunctionBegin; 1220 PetscCheckFalse(!product,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1221 PetscCheckFalse(product->data,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 1222 1223 /* create symbolic Bt */ 1224 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 1225 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 1226 ierr = MatSetBlockSizes(Bt,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 1227 ierr = MatSetType(Bt,((PetscObject)A)->type_name);CHKERRQ(ierr); 1228 1229 /* get symbolic C=A*Bt */ 1230 ierr = PetscStrallocpy(product->alg,&alg);CHKERRQ(ierr); 1231 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); /* set algorithm for C = A*Bt */ 1232 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 1233 ierr = MatProductSetAlgorithm(C,alg);CHKERRQ(ierr); /* resume original algorithm for ABt product */ 1234 ierr = PetscFree(alg);CHKERRQ(ierr); 1235 1236 /* create a supporting struct for reuse intermediate dense matrices with matcoloring */ 1237 ierr = PetscNew(&abt);CHKERRQ(ierr); 1238 1239 product->data = abt; 1240 product->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 1241 1242 C->ops->mattransposemultnumeric = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ; 1243 1244 abt->usecoloring = PETSC_FALSE; 1245 ierr = PetscStrcmp(product->alg,"color",&abt->usecoloring);CHKERRQ(ierr); 1246 if (abt->usecoloring) { 1247 /* Create MatTransposeColoring from symbolic C=A*B^T */ 1248 MatTransposeColoring matcoloring; 1249 MatColoring coloring; 1250 ISColoring iscoloring; 1251 Mat Bt_dense,C_dense; 1252 1253 /* inode causes memory problem */ 1254 ierr = MatSetOption(C,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 1255 1256 ierr = MatColoringCreate(C,&coloring);CHKERRQ(ierr); 1257 ierr = MatColoringSetDistance(coloring,2);CHKERRQ(ierr); 1258 ierr = MatColoringSetType(coloring,MATCOLORINGSL);CHKERRQ(ierr); 1259 ierr = MatColoringSetFromOptions(coloring);CHKERRQ(ierr); 1260 ierr = MatColoringApply(coloring,&iscoloring);CHKERRQ(ierr); 1261 ierr = MatColoringDestroy(&coloring);CHKERRQ(ierr); 1262 ierr = MatTransposeColoringCreate(C,iscoloring,&matcoloring);CHKERRQ(ierr); 1263 1264 abt->matcoloring = matcoloring; 1265 1266 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 1267 1268 /* Create Bt_dense and C_dense = A*Bt_dense */ 1269 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 1270 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 1271 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 1272 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 1273 1274 Bt_dense->assembled = PETSC_TRUE; 1275 abt->Bt_den = Bt_dense; 1276 1277 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 1278 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 1279 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 1280 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 1281 1282 Bt_dense->assembled = PETSC_TRUE; 1283 abt->ABt_den = C_dense; 1284 1285 #if defined(PETSC_USE_INFO) 1286 { 1287 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 1288 ierr = PetscInfo(C,"Use coloring of C=A*B^T; B^T: %" PetscInt_FMT " %" PetscInt_FMT ", Bt_dense: %" PetscInt_FMT ",%" PetscInt_FMT "; Cnz %" PetscInt_FMT " / (cm*ncolors %" PetscInt_FMT ") = %g\n",B->cmap->n,B->rmap->n,Bt_dense->rmap->n,Bt_dense->cmap->n,c->nz,A->rmap->n*matcoloring->ncolors,(double)(((PetscReal)(c->nz))/((PetscReal)(A->rmap->n*matcoloring->ncolors))));CHKERRQ(ierr); 1289 } 1290 #endif 1291 } 1292 /* clean up */ 1293 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 1294 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 1295 PetscFunctionReturn(0); 1296 } 1297 1298 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1299 { 1300 PetscErrorCode ierr; 1301 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1302 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 1303 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 1304 PetscLogDouble flops=0.0; 1305 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 1306 Mat_MatMatTransMult *abt; 1307 Mat_Product *product = C->product; 1308 1309 PetscFunctionBegin; 1310 PetscCheckFalse(!product,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1311 abt = (Mat_MatMatTransMult *)product->data; 1312 PetscCheckFalse(!abt,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1313 /* clear old values in C */ 1314 if (!c->a) { 1315 ierr = PetscCalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 1316 c->a = ca; 1317 c->free_a = PETSC_TRUE; 1318 } else { 1319 ca = c->a; 1320 ierr = PetscArrayzero(ca,ci[cm]+1);CHKERRQ(ierr); 1321 } 1322 1323 if (abt->usecoloring) { 1324 MatTransposeColoring matcoloring = abt->matcoloring; 1325 Mat Bt_dense,C_dense = abt->ABt_den; 1326 1327 /* Get Bt_dense by Apply MatTransposeColoring to B */ 1328 Bt_dense = abt->Bt_den; 1329 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 1330 1331 /* C_dense = A*Bt_dense */ 1332 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 1333 1334 /* Recover C from C_dense */ 1335 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 1336 PetscFunctionReturn(0); 1337 } 1338 1339 for (i=0; i<cm; i++) { 1340 anzi = ai[i+1] - ai[i]; 1341 acol = aj + ai[i]; 1342 aval = aa + ai[i]; 1343 cnzi = ci[i+1] - ci[i]; 1344 ccol = cj + ci[i]; 1345 cval = ca + ci[i]; 1346 for (j=0; j<cnzi; j++) { 1347 brow = ccol[j]; 1348 bnzj = bi[brow+1] - bi[brow]; 1349 bcol = bj + bi[brow]; 1350 bval = ba + bi[brow]; 1351 1352 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 1353 nexta = 0; nextb = 0; 1354 while (nexta<anzi && nextb<bnzj) { 1355 while (nexta < anzi && acol[nexta] < bcol[nextb]) nexta++; 1356 if (nexta == anzi) break; 1357 while (nextb < bnzj && acol[nexta] > bcol[nextb]) nextb++; 1358 if (nextb == bnzj) break; 1359 if (acol[nexta] == bcol[nextb]) { 1360 cval[j] += aval[nexta]*bval[nextb]; 1361 nexta++; nextb++; 1362 flops += 2; 1363 } 1364 } 1365 } 1366 } 1367 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1368 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1369 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1370 PetscFunctionReturn(0); 1371 } 1372 1373 PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(void *data) 1374 { 1375 PetscErrorCode ierr; 1376 Mat_MatTransMatMult *atb = (Mat_MatTransMatMult*)data; 1377 1378 PetscFunctionBegin; 1379 ierr = MatDestroy(&atb->At);CHKERRQ(ierr); 1380 if (atb->destroy) { 1381 ierr = (*atb->destroy)(atb->data);CHKERRQ(ierr); 1382 } 1383 ierr = PetscFree(atb);CHKERRQ(ierr); 1384 PetscFunctionReturn(0); 1385 } 1386 1387 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 1388 { 1389 PetscErrorCode ierr; 1390 Mat At = NULL; 1391 PetscInt *ati,*atj; 1392 Mat_Product *product = C->product; 1393 PetscBool flg,def,square; 1394 1395 PetscFunctionBegin; 1396 MatCheckProduct(C,4); 1397 square = (PetscBool)(A == B && A->symmetric && A->symmetric_set); 1398 /* outerproduct */ 1399 ierr = PetscStrcmp(product->alg,"outerproduct",&flg);CHKERRQ(ierr); 1400 if (flg) { 1401 /* create symbolic At */ 1402 if (!square) { 1403 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1404 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1405 ierr = MatSetBlockSizes(At,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 1406 ierr = MatSetType(At,((PetscObject)A)->type_name);CHKERRQ(ierr); 1407 } 1408 /* get symbolic C=At*B */ 1409 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); 1410 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(square ? A : At,B,fill,C);CHKERRQ(ierr); 1411 1412 /* clean up */ 1413 if (!square) { 1414 ierr = MatDestroy(&At);CHKERRQ(ierr); 1415 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1416 } 1417 1418 C->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ; /* outerproduct */ 1419 ierr = MatProductSetAlgorithm(C,"outerproduct");CHKERRQ(ierr); 1420 PetscFunctionReturn(0); 1421 } 1422 1423 /* matmatmult */ 1424 ierr = PetscStrcmp(product->alg,"default",&def);CHKERRQ(ierr); 1425 ierr = PetscStrcmp(product->alg,"at*b",&flg);CHKERRQ(ierr); 1426 if (flg || def) { 1427 Mat_MatTransMatMult *atb; 1428 1429 PetscCheckFalse(product->data,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 1430 ierr = PetscNew(&atb);CHKERRQ(ierr); 1431 if (!square) { 1432 ierr = MatTranspose_SeqAIJ(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 1433 } 1434 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); 1435 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(square ? A : At,B,fill,C);CHKERRQ(ierr); 1436 ierr = MatProductSetAlgorithm(C,"at*b");CHKERRQ(ierr); 1437 product->data = atb; 1438 product->destroy = MatDestroy_SeqAIJ_MatTransMatMult; 1439 atb->At = At; 1440 atb->updateAt = PETSC_FALSE; /* because At is computed here */ 1441 1442 C->ops->mattransposemultnumeric = NULL; /* see MatProductNumeric_AtB_SeqAIJ_SeqAIJ */ 1443 PetscFunctionReturn(0); 1444 } 1445 1446 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Mat Product Algorithm is not supported"); 1447 } 1448 1449 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1450 { 1451 PetscErrorCode ierr; 1452 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1453 PetscInt am=A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1454 PetscInt cm=C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1455 PetscLogDouble flops=0.0; 1456 MatScalar *aa=a->a,*ba,*ca,*caj; 1457 1458 PetscFunctionBegin; 1459 if (!c->a) { 1460 ierr = PetscCalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 1461 1462 c->a = ca; 1463 c->free_a = PETSC_TRUE; 1464 } else { 1465 ca = c->a; 1466 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 1467 } 1468 1469 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1470 for (i=0; i<am; i++) { 1471 bj = b->j + bi[i]; 1472 ba = b->a + bi[i]; 1473 bnzi = bi[i+1] - bi[i]; 1474 anzi = ai[i+1] - ai[i]; 1475 for (j=0; j<anzi; j++) { 1476 nextb = 0; 1477 crow = *aj++; 1478 cjj = cj + ci[crow]; 1479 caj = ca + ci[crow]; 1480 /* perform sparse axpy operation. Note cjj includes bj. */ 1481 for (k=0; nextb<bnzi; k++) { 1482 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1483 caj[k] += (*aa)*(*(ba+nextb)); 1484 nextb++; 1485 } 1486 } 1487 flops += 2*bnzi; 1488 aa++; 1489 } 1490 } 1491 1492 /* Assemble the final matrix and clean up */ 1493 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1494 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1495 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1496 PetscFunctionReturn(0); 1497 } 1498 1499 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat C) 1500 { 1501 PetscErrorCode ierr; 1502 1503 PetscFunctionBegin; 1504 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1505 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1506 PetscFunctionReturn(0); 1507 } 1508 1509 PETSC_INTERN PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C,const PetscBool add) 1510 { 1511 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1512 PetscErrorCode ierr; 1513 PetscScalar *c,r1,r2,r3,r4,*c1,*c2,*c3,*c4; 1514 const PetscScalar *aa,*b,*b1,*b2,*b3,*b4,*av; 1515 const PetscInt *aj; 1516 PetscInt cm=C->rmap->n,cn=B->cmap->n,bm,am=A->rmap->n; 1517 PetscInt clda; 1518 PetscInt am4,bm4,col,i,j,n; 1519 1520 PetscFunctionBegin; 1521 if (!cm || !cn) PetscFunctionReturn(0); 1522 ierr = MatSeqAIJGetArrayRead(A,&av);CHKERRQ(ierr); 1523 if (add) { 1524 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1525 } else { 1526 ierr = MatDenseGetArrayWrite(C,&c);CHKERRQ(ierr); 1527 } 1528 ierr = MatDenseGetArrayRead(B,&b);CHKERRQ(ierr); 1529 ierr = MatDenseGetLDA(B,&bm);CHKERRQ(ierr); 1530 ierr = MatDenseGetLDA(C,&clda);CHKERRQ(ierr); 1531 am4 = 4*clda; 1532 bm4 = 4*bm; 1533 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1534 c1 = c; c2 = c1 + clda; c3 = c2 + clda; c4 = c3 + clda; 1535 for (col=0; col<(cn/4)*4; col += 4) { /* over columns of C */ 1536 for (i=0; i<am; i++) { /* over rows of A in those columns */ 1537 r1 = r2 = r3 = r4 = 0.0; 1538 n = a->i[i+1] - a->i[i]; 1539 aj = a->j + a->i[i]; 1540 aa = av + a->i[i]; 1541 for (j=0; j<n; j++) { 1542 const PetscScalar aatmp = aa[j]; 1543 const PetscInt ajtmp = aj[j]; 1544 r1 += aatmp*b1[ajtmp]; 1545 r2 += aatmp*b2[ajtmp]; 1546 r3 += aatmp*b3[ajtmp]; 1547 r4 += aatmp*b4[ajtmp]; 1548 } 1549 if (add) { 1550 c1[i] += r1; 1551 c2[i] += r2; 1552 c3[i] += r3; 1553 c4[i] += r4; 1554 } else { 1555 c1[i] = r1; 1556 c2[i] = r2; 1557 c3[i] = r3; 1558 c4[i] = r4; 1559 } 1560 } 1561 b1 += bm4; b2 += bm4; b3 += bm4; b4 += bm4; 1562 c1 += am4; c2 += am4; c3 += am4; c4 += am4; 1563 } 1564 /* process remaining columns */ 1565 if (col != cn) { 1566 PetscInt rc = cn-col; 1567 1568 if (rc == 1) { 1569 for (i=0; i<am; i++) { 1570 r1 = 0.0; 1571 n = a->i[i+1] - a->i[i]; 1572 aj = a->j + a->i[i]; 1573 aa = av + a->i[i]; 1574 for (j=0; j<n; j++) r1 += aa[j]*b1[aj[j]]; 1575 if (add) c1[i] += r1; 1576 else c1[i] = r1; 1577 } 1578 } else if (rc == 2) { 1579 for (i=0; i<am; i++) { 1580 r1 = r2 = 0.0; 1581 n = a->i[i+1] - a->i[i]; 1582 aj = a->j + a->i[i]; 1583 aa = av + a->i[i]; 1584 for (j=0; j<n; j++) { 1585 const PetscScalar aatmp = aa[j]; 1586 const PetscInt ajtmp = aj[j]; 1587 r1 += aatmp*b1[ajtmp]; 1588 r2 += aatmp*b2[ajtmp]; 1589 } 1590 if (add) { 1591 c1[i] += r1; 1592 c2[i] += r2; 1593 } else { 1594 c1[i] = r1; 1595 c2[i] = r2; 1596 } 1597 } 1598 } else { 1599 for (i=0; i<am; i++) { 1600 r1 = r2 = r3 = 0.0; 1601 n = a->i[i+1] - a->i[i]; 1602 aj = a->j + a->i[i]; 1603 aa = av + a->i[i]; 1604 for (j=0; j<n; j++) { 1605 const PetscScalar aatmp = aa[j]; 1606 const PetscInt ajtmp = aj[j]; 1607 r1 += aatmp*b1[ajtmp]; 1608 r2 += aatmp*b2[ajtmp]; 1609 r3 += aatmp*b3[ajtmp]; 1610 } 1611 if (add) { 1612 c1[i] += r1; 1613 c2[i] += r2; 1614 c3[i] += r3; 1615 } else { 1616 c1[i] = r1; 1617 c2[i] = r2; 1618 c3[i] = r3; 1619 } 1620 } 1621 } 1622 } 1623 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1624 if (add) { 1625 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1626 } else { 1627 ierr = MatDenseRestoreArrayWrite(C,&c);CHKERRQ(ierr); 1628 } 1629 ierr = MatDenseRestoreArrayRead(B,&b);CHKERRQ(ierr); 1630 ierr = MatSeqAIJRestoreArrayRead(A,&av);CHKERRQ(ierr); 1631 PetscFunctionReturn(0); 1632 } 1633 1634 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1635 { 1636 PetscErrorCode ierr; 1637 1638 PetscFunctionBegin; 1639 PetscCheckFalse(B->rmap->n != A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number columns in A %" PetscInt_FMT " not equal rows in B %" PetscInt_FMT,A->cmap->n,B->rmap->n); 1640 PetscCheckFalse(A->rmap->n != C->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number rows in C %" PetscInt_FMT " not equal rows in A %" PetscInt_FMT,C->rmap->n,A->rmap->n); 1641 PetscCheckFalse(B->cmap->n != C->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number columns in B %" PetscInt_FMT " not equal columns in C %" PetscInt_FMT,B->cmap->n,C->cmap->n); 1642 1643 ierr = MatMatMultNumericAdd_SeqAIJ_SeqDense(A,B,C,PETSC_FALSE);CHKERRQ(ierr); 1644 PetscFunctionReturn(0); 1645 } 1646 1647 /* ------------------------------------------------------- */ 1648 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_AB(Mat C) 1649 { 1650 PetscFunctionBegin; 1651 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqAIJ_SeqDense; 1652 C->ops->productsymbolic = MatProductSymbolic_AB; 1653 PetscFunctionReturn(0); 1654 } 1655 1656 PETSC_INTERN PetscErrorCode MatTMatTMultSymbolic_SeqAIJ_SeqDense(Mat,Mat,PetscReal,Mat); 1657 1658 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_AtB(Mat C) 1659 { 1660 PetscFunctionBegin; 1661 C->ops->transposematmultsymbolic = MatTMatTMultSymbolic_SeqAIJ_SeqDense; 1662 C->ops->productsymbolic = MatProductSymbolic_AtB; 1663 PetscFunctionReturn(0); 1664 } 1665 1666 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_ABt(Mat C) 1667 { 1668 PetscFunctionBegin; 1669 C->ops->mattransposemultsymbolic = MatTMatTMultSymbolic_SeqAIJ_SeqDense; 1670 C->ops->productsymbolic = MatProductSymbolic_ABt; 1671 PetscFunctionReturn(0); 1672 } 1673 1674 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense(Mat C) 1675 { 1676 PetscErrorCode ierr; 1677 Mat_Product *product = C->product; 1678 1679 PetscFunctionBegin; 1680 switch (product->type) { 1681 case MATPRODUCT_AB: 1682 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_AB(C);CHKERRQ(ierr); 1683 break; 1684 case MATPRODUCT_AtB: 1685 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_AtB(C);CHKERRQ(ierr); 1686 break; 1687 case MATPRODUCT_ABt: 1688 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_ABt(C);CHKERRQ(ierr); 1689 break; 1690 default: 1691 break; 1692 } 1693 PetscFunctionReturn(0); 1694 } 1695 /* ------------------------------------------------------- */ 1696 static PetscErrorCode MatProductSetFromOptions_SeqXBAIJ_SeqDense_AB(Mat C) 1697 { 1698 PetscErrorCode ierr; 1699 Mat_Product *product = C->product; 1700 Mat A = product->A; 1701 PetscBool baij; 1702 1703 PetscFunctionBegin; 1704 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&baij);CHKERRQ(ierr); 1705 if (!baij) { /* A is seqsbaij */ 1706 PetscBool sbaij; 1707 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&sbaij);CHKERRQ(ierr); 1708 PetscCheckFalse(!sbaij,PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"Mat must be either seqbaij or seqsbaij format"); 1709 1710 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqSBAIJ_SeqDense; 1711 } else { /* A is seqbaij */ 1712 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqBAIJ_SeqDense; 1713 } 1714 1715 C->ops->productsymbolic = MatProductSymbolic_AB; 1716 PetscFunctionReturn(0); 1717 } 1718 1719 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqXBAIJ_SeqDense(Mat C) 1720 { 1721 PetscErrorCode ierr; 1722 Mat_Product *product = C->product; 1723 1724 PetscFunctionBegin; 1725 MatCheckProduct(C,1); 1726 PetscCheckFalse(!product->A,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing A"); 1727 if (product->type == MATPRODUCT_AB || (product->type == MATPRODUCT_AtB && product->A->symmetric)) { 1728 ierr = MatProductSetFromOptions_SeqXBAIJ_SeqDense_AB(C);CHKERRQ(ierr); 1729 } 1730 PetscFunctionReturn(0); 1731 } 1732 1733 /* ------------------------------------------------------- */ 1734 static PetscErrorCode MatProductSetFromOptions_SeqDense_SeqAIJ_AB(Mat C) 1735 { 1736 PetscFunctionBegin; 1737 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqDense_SeqAIJ; 1738 C->ops->productsymbolic = MatProductSymbolic_AB; 1739 PetscFunctionReturn(0); 1740 } 1741 1742 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqDense_SeqAIJ(Mat C) 1743 { 1744 PetscErrorCode ierr; 1745 Mat_Product *product = C->product; 1746 1747 PetscFunctionBegin; 1748 if (product->type == MATPRODUCT_AB) { 1749 ierr = MatProductSetFromOptions_SeqDense_SeqAIJ_AB(C);CHKERRQ(ierr); 1750 } 1751 PetscFunctionReturn(0); 1752 } 1753 /* ------------------------------------------------------- */ 1754 1755 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1756 { 1757 PetscErrorCode ierr; 1758 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1759 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1760 PetscInt *bi = b->i,*bj=b->j; 1761 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1762 MatScalar *btval,*btval_den,*ba=b->a; 1763 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1764 1765 PetscFunctionBegin; 1766 btval_den=btdense->v; 1767 ierr = PetscArrayzero(btval_den,m*n);CHKERRQ(ierr); 1768 for (k=0; k<ncolors; k++) { 1769 ncolumns = coloring->ncolumns[k]; 1770 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1771 col = *(columns + colorforcol[k] + l); 1772 btcol = bj + bi[col]; 1773 btval = ba + bi[col]; 1774 anz = bi[col+1] - bi[col]; 1775 for (j=0; j<anz; j++) { 1776 brow = btcol[j]; 1777 btval_den[brow] = btval[j]; 1778 } 1779 } 1780 btval_den += m; 1781 } 1782 PetscFunctionReturn(0); 1783 } 1784 1785 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1786 { 1787 PetscErrorCode ierr; 1788 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1789 const PetscScalar *ca_den,*ca_den_ptr; 1790 PetscScalar *ca=csp->a; 1791 PetscInt k,l,m=Cden->rmap->n,ncolors=matcoloring->ncolors; 1792 PetscInt brows=matcoloring->brows,*den2sp=matcoloring->den2sp; 1793 PetscInt nrows,*row,*idx; 1794 PetscInt *rows=matcoloring->rows,*colorforrow=matcoloring->colorforrow; 1795 1796 PetscFunctionBegin; 1797 ierr = MatDenseGetArrayRead(Cden,&ca_den);CHKERRQ(ierr); 1798 1799 if (brows > 0) { 1800 PetscInt *lstart,row_end,row_start; 1801 lstart = matcoloring->lstart; 1802 ierr = PetscArrayzero(lstart,ncolors);CHKERRQ(ierr); 1803 1804 row_end = brows; 1805 if (row_end > m) row_end = m; 1806 for (row_start=0; row_start<m; row_start+=brows) { /* loop over row blocks of Csp */ 1807 ca_den_ptr = ca_den; 1808 for (k=0; k<ncolors; k++) { /* loop over colors (columns of Cden) */ 1809 nrows = matcoloring->nrows[k]; 1810 row = rows + colorforrow[k]; 1811 idx = den2sp + colorforrow[k]; 1812 for (l=lstart[k]; l<nrows; l++) { 1813 if (row[l] >= row_end) { 1814 lstart[k] = l; 1815 break; 1816 } else { 1817 ca[idx[l]] = ca_den_ptr[row[l]]; 1818 } 1819 } 1820 ca_den_ptr += m; 1821 } 1822 row_end += brows; 1823 if (row_end > m) row_end = m; 1824 } 1825 } else { /* non-blocked impl: loop over columns of Csp - slow if Csp is large */ 1826 ca_den_ptr = ca_den; 1827 for (k=0; k<ncolors; k++) { 1828 nrows = matcoloring->nrows[k]; 1829 row = rows + colorforrow[k]; 1830 idx = den2sp + colorforrow[k]; 1831 for (l=0; l<nrows; l++) { 1832 ca[idx[l]] = ca_den_ptr[row[l]]; 1833 } 1834 ca_den_ptr += m; 1835 } 1836 } 1837 1838 ierr = MatDenseRestoreArrayRead(Cden,&ca_den);CHKERRQ(ierr); 1839 #if defined(PETSC_USE_INFO) 1840 if (matcoloring->brows > 0) { 1841 ierr = PetscInfo(Csp,"Loop over %" PetscInt_FMT " row blocks for den2sp\n",brows);CHKERRQ(ierr); 1842 } else { 1843 ierr = PetscInfo(Csp,"Loop over colors/columns of Cden, inefficient for large sparse matrix product \n");CHKERRQ(ierr); 1844 } 1845 #endif 1846 PetscFunctionReturn(0); 1847 } 1848 1849 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1850 { 1851 PetscErrorCode ierr; 1852 PetscInt i,n,nrows,Nbs,j,k,m,ncols,col,cm; 1853 const PetscInt *is,*ci,*cj,*row_idx; 1854 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1855 IS *isa; 1856 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1857 PetscInt *colorforrow,*rows,*rows_i,*idxhit,*spidx,*den2sp,*den2sp_i; 1858 PetscInt *colorforcol,*columns,*columns_i,brows; 1859 PetscBool flg; 1860 1861 PetscFunctionBegin; 1862 ierr = ISColoringGetIS(iscoloring,PETSC_USE_POINTER,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1863 1864 /* bs >1 is not being tested yet! */ 1865 Nbs = mat->cmap->N/bs; 1866 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1867 c->N = Nbs; 1868 c->m = c->M; 1869 c->rstart = 0; 1870 c->brows = 100; 1871 1872 c->ncolors = nis; 1873 ierr = PetscMalloc3(nis,&c->ncolumns,nis,&c->nrows,nis+1,&colorforrow);CHKERRQ(ierr); 1874 ierr = PetscMalloc1(csp->nz+1,&rows);CHKERRQ(ierr); 1875 ierr = PetscMalloc1(csp->nz+1,&den2sp);CHKERRQ(ierr); 1876 1877 brows = c->brows; 1878 ierr = PetscOptionsGetInt(NULL,NULL,"-matden2sp_brows",&brows,&flg);CHKERRQ(ierr); 1879 if (flg) c->brows = brows; 1880 if (brows > 0) { 1881 ierr = PetscMalloc1(nis+1,&c->lstart);CHKERRQ(ierr); 1882 } 1883 1884 colorforrow[0] = 0; 1885 rows_i = rows; 1886 den2sp_i = den2sp; 1887 1888 ierr = PetscMalloc1(nis+1,&colorforcol);CHKERRQ(ierr); 1889 ierr = PetscMalloc1(Nbs+1,&columns);CHKERRQ(ierr); 1890 1891 colorforcol[0] = 0; 1892 columns_i = columns; 1893 1894 /* get column-wise storage of mat */ 1895 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1896 1897 cm = c->m; 1898 ierr = PetscMalloc1(cm+1,&rowhit);CHKERRQ(ierr); 1899 ierr = PetscMalloc1(cm+1,&idxhit);CHKERRQ(ierr); 1900 for (i=0; i<nis; i++) { /* loop over color */ 1901 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 1902 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 1903 1904 c->ncolumns[i] = n; 1905 if (n) { 1906 ierr = PetscArraycpy(columns_i,is,n);CHKERRQ(ierr); 1907 } 1908 colorforcol[i+1] = colorforcol[i] + n; 1909 columns_i += n; 1910 1911 /* fast, crude version requires O(N*N) work */ 1912 ierr = PetscArrayzero(rowhit,cm);CHKERRQ(ierr); 1913 1914 for (j=0; j<n; j++) { /* loop over columns*/ 1915 col = is[j]; 1916 row_idx = cj + ci[col]; 1917 m = ci[col+1] - ci[col]; 1918 for (k=0; k<m; k++) { /* loop over columns marking them in rowhit */ 1919 idxhit[*row_idx] = spidx[ci[col] + k]; 1920 rowhit[*row_idx++] = col + 1; 1921 } 1922 } 1923 /* count the number of hits */ 1924 nrows = 0; 1925 for (j=0; j<cm; j++) { 1926 if (rowhit[j]) nrows++; 1927 } 1928 c->nrows[i] = nrows; 1929 colorforrow[i+1] = colorforrow[i] + nrows; 1930 1931 nrows = 0; 1932 for (j=0; j<cm; j++) { /* loop over rows */ 1933 if (rowhit[j]) { 1934 rows_i[nrows] = j; 1935 den2sp_i[nrows] = idxhit[j]; 1936 nrows++; 1937 } 1938 } 1939 den2sp_i += nrows; 1940 1941 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1942 rows_i += nrows; 1943 } 1944 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1945 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1946 ierr = ISColoringRestoreIS(iscoloring,PETSC_USE_POINTER,&isa);CHKERRQ(ierr); 1947 PetscCheck(csp->nz == colorforrow[nis],PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %" PetscInt_FMT " != colorforrow[nis] %" PetscInt_FMT,csp->nz,colorforrow[nis]); 1948 1949 c->colorforrow = colorforrow; 1950 c->rows = rows; 1951 c->den2sp = den2sp; 1952 c->colorforcol = colorforcol; 1953 c->columns = columns; 1954 1955 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1956 PetscFunctionReturn(0); 1957 } 1958 1959 /* --------------------------------------------------------------- */ 1960 static PetscErrorCode MatProductNumeric_AtB_SeqAIJ_SeqAIJ(Mat C) 1961 { 1962 PetscErrorCode ierr; 1963 Mat_Product *product = C->product; 1964 Mat A=product->A,B=product->B; 1965 1966 PetscFunctionBegin; 1967 if (C->ops->mattransposemultnumeric) { 1968 /* Alg: "outerproduct" */ 1969 ierr = (*C->ops->mattransposemultnumeric)(A,B,C);CHKERRQ(ierr); 1970 } else { 1971 /* Alg: "matmatmult" -- C = At*B */ 1972 Mat_MatTransMatMult *atb = (Mat_MatTransMatMult *)product->data; 1973 Mat At; 1974 1975 PetscCheckFalse(!atb,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1976 At = atb->At; 1977 if (atb->updateAt && At) { /* At is computed in MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ() */ 1978 ierr = MatTranspose_SeqAIJ(A,MAT_REUSE_MATRIX,&At);CHKERRQ(ierr); 1979 } 1980 ierr = MatMatMultNumeric_SeqAIJ_SeqAIJ(At ? At : A,B,C);CHKERRQ(ierr); 1981 atb->updateAt = PETSC_TRUE; 1982 } 1983 PetscFunctionReturn(0); 1984 } 1985 1986 static PetscErrorCode MatProductSymbolic_AtB_SeqAIJ_SeqAIJ(Mat C) 1987 { 1988 PetscErrorCode ierr; 1989 Mat_Product *product = C->product; 1990 Mat A=product->A,B=product->B; 1991 PetscReal fill=product->fill; 1992 1993 PetscFunctionBegin; 1994 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1995 1996 C->ops->productnumeric = MatProductNumeric_AtB_SeqAIJ_SeqAIJ; 1997 PetscFunctionReturn(0); 1998 } 1999 2000 /* --------------------------------------------------------------- */ 2001 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_AB(Mat C) 2002 { 2003 PetscErrorCode ierr; 2004 Mat_Product *product = C->product; 2005 PetscInt alg = 0; /* default algorithm */ 2006 PetscBool flg = PETSC_FALSE; 2007 #if !defined(PETSC_HAVE_HYPRE) 2008 const char *algTypes[7] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge"}; 2009 PetscInt nalg = 7; 2010 #else 2011 const char *algTypes[8] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge","hypre"}; 2012 PetscInt nalg = 8; 2013 #endif 2014 2015 PetscFunctionBegin; 2016 /* Set default algorithm */ 2017 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2018 if (flg) { 2019 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2020 } 2021 2022 /* Get runtime option */ 2023 if (product->api_user) { 2024 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMult","Mat");CHKERRQ(ierr); 2025 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2026 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2027 } else { 2028 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AB","Mat");CHKERRQ(ierr); 2029 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_AB",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2030 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2031 } 2032 if (flg) { 2033 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2034 } 2035 2036 C->ops->productsymbolic = MatProductSymbolic_AB; 2037 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqAIJ_SeqAIJ; 2038 PetscFunctionReturn(0); 2039 } 2040 2041 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_AtB(Mat C) 2042 { 2043 PetscErrorCode ierr; 2044 Mat_Product *product = C->product; 2045 PetscInt alg = 0; /* default algorithm */ 2046 PetscBool flg = PETSC_FALSE; 2047 const char *algTypes[3] = {"default","at*b","outerproduct"}; 2048 PetscInt nalg = 3; 2049 2050 PetscFunctionBegin; 2051 /* Get runtime option */ 2052 if (product->api_user) { 2053 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatTransposeMatMult","Mat");CHKERRQ(ierr); 2054 ierr = PetscOptionsEList("-mattransposematmult_via","Algorithmic approach","MatTransposeMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2055 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2056 } else { 2057 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AtB","Mat");CHKERRQ(ierr); 2058 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_AtB",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2059 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2060 } 2061 if (flg) { 2062 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2063 } 2064 2065 C->ops->productsymbolic = MatProductSymbolic_AtB_SeqAIJ_SeqAIJ; 2066 PetscFunctionReturn(0); 2067 } 2068 2069 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_ABt(Mat C) 2070 { 2071 PetscErrorCode ierr; 2072 Mat_Product *product = C->product; 2073 PetscInt alg = 0; /* default algorithm */ 2074 PetscBool flg = PETSC_FALSE; 2075 const char *algTypes[2] = {"default","color"}; 2076 PetscInt nalg = 2; 2077 2078 PetscFunctionBegin; 2079 /* Set default algorithm */ 2080 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2081 if (!flg) { 2082 alg = 1; 2083 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2084 } 2085 2086 /* Get runtime option */ 2087 if (product->api_user) { 2088 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatTransposeMult","Mat");CHKERRQ(ierr); 2089 ierr = PetscOptionsEList("-matmattransmult_via","Algorithmic approach","MatMatTransposeMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2090 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2091 } else { 2092 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_ABt","Mat");CHKERRQ(ierr); 2093 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_ABt",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2094 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2095 } 2096 if (flg) { 2097 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2098 } 2099 2100 C->ops->mattransposemultsymbolic = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ; 2101 C->ops->productsymbolic = MatProductSymbolic_ABt; 2102 PetscFunctionReturn(0); 2103 } 2104 2105 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_PtAP(Mat C) 2106 { 2107 PetscErrorCode ierr; 2108 Mat_Product *product = C->product; 2109 PetscBool flg = PETSC_FALSE; 2110 PetscInt alg = 0; /* default algorithm -- alg=1 should be default!!! */ 2111 #if !defined(PETSC_HAVE_HYPRE) 2112 const char *algTypes[2] = {"scalable","rap"}; 2113 PetscInt nalg = 2; 2114 #else 2115 const char *algTypes[3] = {"scalable","rap","hypre"}; 2116 PetscInt nalg = 3; 2117 #endif 2118 2119 PetscFunctionBegin; 2120 /* Set default algorithm */ 2121 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2122 if (flg) { 2123 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2124 } 2125 2126 /* Get runtime option */ 2127 if (product->api_user) { 2128 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatPtAP","Mat");CHKERRQ(ierr); 2129 ierr = PetscOptionsEList("-matptap_via","Algorithmic approach","MatPtAP",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2130 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2131 } else { 2132 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_PtAP","Mat");CHKERRQ(ierr); 2133 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_PtAP",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2134 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2135 } 2136 if (flg) { 2137 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2138 } 2139 2140 C->ops->productsymbolic = MatProductSymbolic_PtAP_SeqAIJ_SeqAIJ; 2141 PetscFunctionReturn(0); 2142 } 2143 2144 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_RARt(Mat C) 2145 { 2146 PetscErrorCode ierr; 2147 Mat_Product *product = C->product; 2148 PetscBool flg = PETSC_FALSE; 2149 PetscInt alg = 0; /* default algorithm */ 2150 const char *algTypes[3] = {"r*a*rt","r*art","coloring_rart"}; 2151 PetscInt nalg = 3; 2152 2153 PetscFunctionBegin; 2154 /* Set default algorithm */ 2155 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2156 if (flg) { 2157 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2158 } 2159 2160 /* Get runtime option */ 2161 if (product->api_user) { 2162 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatRARt","Mat");CHKERRQ(ierr); 2163 ierr = PetscOptionsEList("-matrart_via","Algorithmic approach","MatRARt",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2164 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2165 } else { 2166 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_RARt","Mat");CHKERRQ(ierr); 2167 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_RARt",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2168 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2169 } 2170 if (flg) { 2171 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2172 } 2173 2174 C->ops->productsymbolic = MatProductSymbolic_RARt_SeqAIJ_SeqAIJ; 2175 PetscFunctionReturn(0); 2176 } 2177 2178 /* ABC = A*B*C = A*(B*C); ABC's algorithm must be chosen from AB's algorithm */ 2179 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_ABC(Mat C) 2180 { 2181 PetscErrorCode ierr; 2182 Mat_Product *product = C->product; 2183 PetscInt alg = 0; /* default algorithm */ 2184 PetscBool flg = PETSC_FALSE; 2185 const char *algTypes[7] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge"}; 2186 PetscInt nalg = 7; 2187 2188 PetscFunctionBegin; 2189 /* Set default algorithm */ 2190 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2191 if (flg) { 2192 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2193 } 2194 2195 /* Get runtime option */ 2196 if (product->api_user) { 2197 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMatMult","Mat");CHKERRQ(ierr); 2198 ierr = PetscOptionsEList("-matmatmatmult_via","Algorithmic approach","MatMatMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2199 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2200 } else { 2201 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_ABC","Mat");CHKERRQ(ierr); 2202 ierr = PetscOptionsEList("-mat_product_algorithm","Algorithmic approach","MatProduct_ABC",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2203 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2204 } 2205 if (flg) { 2206 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2207 } 2208 2209 C->ops->matmatmultsymbolic = MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ; 2210 C->ops->productsymbolic = MatProductSymbolic_ABC; 2211 PetscFunctionReturn(0); 2212 } 2213 2214 PetscErrorCode MatProductSetFromOptions_SeqAIJ(Mat C) 2215 { 2216 PetscErrorCode ierr; 2217 Mat_Product *product = C->product; 2218 2219 PetscFunctionBegin; 2220 switch (product->type) { 2221 case MATPRODUCT_AB: 2222 ierr = MatProductSetFromOptions_SeqAIJ_AB(C);CHKERRQ(ierr); 2223 break; 2224 case MATPRODUCT_AtB: 2225 ierr = MatProductSetFromOptions_SeqAIJ_AtB(C);CHKERRQ(ierr); 2226 break; 2227 case MATPRODUCT_ABt: 2228 ierr = MatProductSetFromOptions_SeqAIJ_ABt(C);CHKERRQ(ierr); 2229 break; 2230 case MATPRODUCT_PtAP: 2231 ierr = MatProductSetFromOptions_SeqAIJ_PtAP(C);CHKERRQ(ierr); 2232 break; 2233 case MATPRODUCT_RARt: 2234 ierr = MatProductSetFromOptions_SeqAIJ_RARt(C);CHKERRQ(ierr); 2235 break; 2236 case MATPRODUCT_ABC: 2237 ierr = MatProductSetFromOptions_SeqAIJ_ABC(C);CHKERRQ(ierr); 2238 break; 2239 default: 2240 break; 2241 } 2242 PetscFunctionReturn(0); 2243 } 2244