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 if (C->ops->matmultnumeric == MatMatMultNumeric_SeqAIJ_SeqAIJ) SETERRQ(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; 34 35 PetscFunctionBegin; 36 if (m > 0 && i[0]) SETERRQ(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 ierr = MatSeqAIJSetPreallocation_SeqAIJ(mat,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr); 46 aij = (Mat_SeqAIJ*)(mat)->data; 47 ierr = PetscMalloc1(m,&aij->imax);CHKERRQ(ierr); 48 ierr = PetscMalloc1(m,&aij->ilen);CHKERRQ(ierr); 49 50 aij->i = i; 51 aij->j = j; 52 aij->a = a; 53 aij->singlemalloc = PETSC_FALSE; 54 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 55 aij->free_a = PETSC_FALSE; 56 aij->free_ij = PETSC_FALSE; 57 58 for (ii=0; ii<m; ii++) { 59 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 60 } 61 62 PetscFunctionReturn(0); 63 } 64 65 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 66 { 67 PetscErrorCode ierr; 68 Mat_Product *product = C->product; 69 MatProductAlgorithm alg; 70 PetscBool flg; 71 72 PetscFunctionBegin; 73 if (product) { 74 alg = product->alg; 75 } else { 76 alg = "sorted"; 77 } 78 /* sorted */ 79 ierr = PetscStrcmp(alg,"sorted",&flg);CHKERRQ(ierr); 80 if (flg) { 81 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(A,B,fill,C);CHKERRQ(ierr); 82 PetscFunctionReturn(0); 83 } 84 85 /* scalable */ 86 ierr = PetscStrcmp(alg,"scalable",&flg);CHKERRQ(ierr); 87 if (flg) { 88 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 89 PetscFunctionReturn(0); 90 } 91 92 /* scalable_fast */ 93 ierr = PetscStrcmp(alg,"scalable_fast",&flg);CHKERRQ(ierr); 94 if (flg) { 95 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 96 PetscFunctionReturn(0); 97 } 98 99 /* heap */ 100 ierr = PetscStrcmp(alg,"heap",&flg);CHKERRQ(ierr); 101 if (flg) { 102 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 103 PetscFunctionReturn(0); 104 } 105 106 /* btheap */ 107 ierr = PetscStrcmp(alg,"btheap",&flg);CHKERRQ(ierr); 108 if (flg) { 109 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 110 PetscFunctionReturn(0); 111 } 112 113 /* llcondensed */ 114 ierr = PetscStrcmp(alg,"llcondensed",&flg);CHKERRQ(ierr); 115 if (flg) { 116 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(A,B,fill,C);CHKERRQ(ierr); 117 PetscFunctionReturn(0); 118 } 119 120 /* rowmerge */ 121 ierr = PetscStrcmp(alg,"rowmerge",&flg);CHKERRQ(ierr); 122 if (flg) { 123 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(A,B,fill,C);CHKERRQ(ierr); 124 PetscFunctionReturn(0); 125 } 126 127 #if defined(PETSC_HAVE_HYPRE) 128 ierr = PetscStrcmp(alg,"hypre",&flg);CHKERRQ(ierr); 129 if (flg) { 130 ierr = MatMatMultSymbolic_AIJ_AIJ_wHYPRE(A,B,fill,C);CHKERRQ(ierr); 131 PetscFunctionReturn(0); 132 } 133 #endif 134 135 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Mat Product Algorithm is not supported"); 136 } 137 138 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat A,Mat B,PetscReal fill,Mat C) 139 { 140 PetscErrorCode ierr; 141 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 142 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 143 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 144 PetscReal afill; 145 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 146 PetscTable ta; 147 PetscBT lnkbt; 148 PetscFreeSpaceList free_space=NULL,current_space=NULL; 149 150 PetscFunctionBegin; 151 /* Get ci and cj */ 152 /*---------------*/ 153 /* Allocate ci array, arrays for fill computation and */ 154 /* free space for accumulating nonzero column info */ 155 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 156 ci[0] = 0; 157 158 /* create and initialize a linked list */ 159 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 160 MatRowMergeMax_SeqAIJ(b,bm,ta); 161 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 162 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 163 164 ierr = PetscLLCondensedCreate(Crmax,bn,&lnk,&lnkbt);CHKERRQ(ierr); 165 166 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 167 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 168 169 current_space = free_space; 170 171 /* Determine ci and cj */ 172 for (i=0; i<am; i++) { 173 anzi = ai[i+1] - ai[i]; 174 aj = a->j + ai[i]; 175 for (j=0; j<anzi; j++) { 176 brow = aj[j]; 177 bnzj = bi[brow+1] - bi[brow]; 178 bj = b->j + bi[brow]; 179 /* add non-zero cols of B into the sorted linked list lnk */ 180 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 181 } 182 /* add possible missing diagonal entry */ 183 if (C->force_diagonals) { 184 ierr = PetscLLCondensedAddSorted(1,&i,lnk,lnkbt);CHKERRQ(ierr); 185 } 186 cnzi = lnk[0]; 187 188 /* If free space is not available, make more free space */ 189 /* Double the amount of total space in the list */ 190 if (current_space->local_remaining<cnzi) { 191 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 192 ndouble++; 193 } 194 195 /* Copy data into free space, then initialize lnk */ 196 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 197 198 current_space->array += cnzi; 199 current_space->local_used += cnzi; 200 current_space->local_remaining -= cnzi; 201 202 ci[i+1] = ci[i] + cnzi; 203 } 204 205 /* Column indices are in the list of free space */ 206 /* Allocate space for cj, initialize cj, and */ 207 /* destroy list of free space and other temporary array(s) */ 208 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 209 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 210 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 211 212 /* put together the new symbolic matrix */ 213 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 214 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 215 216 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 217 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 218 c = (Mat_SeqAIJ*)(C->data); 219 c->free_a = PETSC_FALSE; 220 c->free_ij = PETSC_TRUE; 221 c->nonew = 0; 222 223 /* fast, needs non-scalable O(bn) array 'abdense' */ 224 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 225 226 /* set MatInfo */ 227 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 228 if (afill < 1.0) afill = 1.0; 229 c->maxnz = ci[am]; 230 c->nz = ci[am]; 231 C->info.mallocs = ndouble; 232 C->info.fill_ratio_given = fill; 233 C->info.fill_ratio_needed = afill; 234 235 #if defined(PETSC_USE_INFO) 236 if (ci[am]) { 237 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 238 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 239 } else { 240 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 241 } 242 #endif 243 PetscFunctionReturn(0); 244 } 245 246 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat A,Mat B,Mat C) 247 { 248 PetscErrorCode ierr; 249 PetscLogDouble flops=0.0; 250 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 251 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 252 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 253 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 254 PetscInt am =A->rmap->n,cm=C->rmap->n; 255 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 256 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca,valtmp; 257 PetscScalar *ab_dense; 258 PetscContainer cab_dense; 259 260 PetscFunctionBegin; 261 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 262 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 263 c->a = ca; 264 c->free_a = PETSC_TRUE; 265 } else ca = c->a; 266 267 /* TODO this should be done in the symbolic phase */ 268 /* However, this function is so heavily used (sometimes in an hidden way through multnumeric function pointers 269 that is hard to eradicate) */ 270 ierr = PetscObjectQuery((PetscObject)C,"__PETSc__ab_dense",(PetscObject*)&cab_dense);CHKERRQ(ierr); 271 if (!cab_dense) { 272 ierr = PetscMalloc1(B->cmap->N,&ab_dense);CHKERRQ(ierr); 273 ierr = PetscContainerCreate(PETSC_COMM_SELF,&cab_dense);CHKERRQ(ierr); 274 ierr = PetscContainerSetPointer(cab_dense,ab_dense);CHKERRQ(ierr); 275 ierr = PetscContainerSetUserDestroy(cab_dense,PetscContainerUserDestroyDefault);CHKERRQ(ierr); 276 ierr = PetscObjectCompose((PetscObject)C,"__PETSc__ab_dense",(PetscObject)cab_dense);CHKERRQ(ierr); 277 ierr = PetscObjectDereference((PetscObject)cab_dense);CHKERRQ(ierr); 278 } 279 ierr = PetscContainerGetPointer(cab_dense,(void**)&ab_dense);CHKERRQ(ierr); 280 ierr = PetscArrayzero(ab_dense,B->cmap->N);CHKERRQ(ierr); 281 282 /* clean old values in C */ 283 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 284 /* Traverse A row-wise. */ 285 /* Build the ith row in C by summing over nonzero columns in A, */ 286 /* the rows of B corresponding to nonzeros of A. */ 287 for (i=0; i<am; i++) { 288 anzi = ai[i+1] - ai[i]; 289 for (j=0; j<anzi; j++) { 290 brow = aj[j]; 291 bnzi = bi[brow+1] - bi[brow]; 292 bjj = bj + bi[brow]; 293 baj = ba + bi[brow]; 294 /* perform dense axpy */ 295 valtmp = aa[j]; 296 for (k=0; k<bnzi; k++) { 297 ab_dense[bjj[k]] += valtmp*baj[k]; 298 } 299 flops += 2*bnzi; 300 } 301 aj += anzi; aa += anzi; 302 303 cnzi = ci[i+1] - ci[i]; 304 for (k=0; k<cnzi; k++) { 305 ca[k] += ab_dense[cj[k]]; 306 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 307 } 308 flops += cnzi; 309 cj += cnzi; ca += cnzi; 310 } 311 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 312 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 313 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 314 PetscFunctionReturn(0); 315 } 316 317 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 318 { 319 PetscErrorCode ierr; 320 PetscLogDouble flops=0.0; 321 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 322 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 323 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 324 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 325 PetscInt am = A->rmap->N,cm=C->rmap->N; 326 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 327 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca=c->a,valtmp; 328 PetscInt nextb; 329 330 PetscFunctionBegin; 331 if (!ca) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 332 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 333 c->a = ca; 334 c->free_a = PETSC_TRUE; 335 } 336 337 /* clean old values in C */ 338 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 339 /* Traverse A row-wise. */ 340 /* Build the ith row in C by summing over nonzero columns in A, */ 341 /* the rows of B corresponding to nonzeros of A. */ 342 for (i=0; i<am; i++) { 343 anzi = ai[i+1] - ai[i]; 344 cnzi = ci[i+1] - ci[i]; 345 for (j=0; j<anzi; j++) { 346 brow = aj[j]; 347 bnzi = bi[brow+1] - bi[brow]; 348 bjj = bj + bi[brow]; 349 baj = ba + bi[brow]; 350 /* perform sparse axpy */ 351 valtmp = aa[j]; 352 nextb = 0; 353 for (k=0; nextb<bnzi; k++) { 354 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 355 ca[k] += valtmp*baj[nextb++]; 356 } 357 } 358 flops += 2*bnzi; 359 } 360 aj += anzi; aa += anzi; 361 cj += cnzi; ca += cnzi; 362 } 363 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 364 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 365 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 366 PetscFunctionReturn(0); 367 } 368 369 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat C) 370 { 371 PetscErrorCode ierr; 372 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 373 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 374 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 375 MatScalar *ca; 376 PetscReal afill; 377 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 378 PetscTable ta; 379 PetscFreeSpaceList free_space=NULL,current_space=NULL; 380 381 PetscFunctionBegin; 382 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 383 /*-----------------------------------------------------------------------------------------*/ 384 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 385 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 386 ci[0] = 0; 387 388 /* create and initialize a linked list */ 389 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 390 MatRowMergeMax_SeqAIJ(b,bm,ta); 391 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 392 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 393 394 ierr = PetscLLCondensedCreate_fast(Crmax,&lnk);CHKERRQ(ierr); 395 396 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 397 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 398 current_space = free_space; 399 400 /* Determine ci and cj */ 401 for (i=0; i<am; i++) { 402 anzi = ai[i+1] - ai[i]; 403 aj = a->j + ai[i]; 404 for (j=0; j<anzi; j++) { 405 brow = aj[j]; 406 bnzj = bi[brow+1] - bi[brow]; 407 bj = b->j + bi[brow]; 408 /* add non-zero cols of B into the sorted linked list lnk */ 409 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 410 } 411 /* add possible missing diagonal entry */ 412 if (C->force_diagonals) { 413 ierr = PetscLLCondensedAddSorted_fast(1,&i,lnk);CHKERRQ(ierr); 414 } 415 cnzi = lnk[1]; 416 417 /* If free space is not available, make more free space */ 418 /* Double the amount of total space in the list */ 419 if (current_space->local_remaining<cnzi) { 420 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 421 ndouble++; 422 } 423 424 /* Copy data into free space, then initialize lnk */ 425 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 426 427 current_space->array += cnzi; 428 current_space->local_used += cnzi; 429 current_space->local_remaining -= cnzi; 430 431 ci[i+1] = ci[i] + cnzi; 432 } 433 434 /* Column indices are in the list of free space */ 435 /* Allocate space for cj, initialize cj, and */ 436 /* destroy list of free space and other temporary array(s) */ 437 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 438 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 439 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 440 441 /* Allocate space for ca */ 442 ierr = PetscCalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 443 444 /* put together the new symbolic matrix */ 445 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 446 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 447 448 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 449 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 450 c = (Mat_SeqAIJ*)(C->data); 451 c->free_a = PETSC_TRUE; 452 c->free_ij = PETSC_TRUE; 453 c->nonew = 0; 454 455 /* slower, less memory */ 456 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; 457 458 /* set MatInfo */ 459 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 460 if (afill < 1.0) afill = 1.0; 461 c->maxnz = ci[am]; 462 c->nz = ci[am]; 463 C->info.mallocs = ndouble; 464 C->info.fill_ratio_given = fill; 465 C->info.fill_ratio_needed = afill; 466 467 #if defined(PETSC_USE_INFO) 468 if (ci[am]) { 469 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 470 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 471 } else { 472 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 473 } 474 #endif 475 PetscFunctionReturn(0); 476 } 477 478 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat C) 479 { 480 PetscErrorCode ierr; 481 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 482 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 483 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 484 MatScalar *ca; 485 PetscReal afill; 486 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 487 PetscTable ta; 488 PetscFreeSpaceList free_space=NULL,current_space=NULL; 489 490 PetscFunctionBegin; 491 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 492 /*---------------------------------------------------------------------------------------------*/ 493 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 494 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 495 ci[0] = 0; 496 497 /* create and initialize a linked list */ 498 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 499 MatRowMergeMax_SeqAIJ(b,bm,ta); 500 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 501 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 502 ierr = PetscLLCondensedCreate_Scalable(Crmax,&lnk);CHKERRQ(ierr); 503 504 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 505 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 506 current_space = free_space; 507 508 /* Determine ci and cj */ 509 for (i=0; i<am; i++) { 510 anzi = ai[i+1] - ai[i]; 511 aj = a->j + ai[i]; 512 for (j=0; j<anzi; j++) { 513 brow = aj[j]; 514 bnzj = bi[brow+1] - bi[brow]; 515 bj = b->j + bi[brow]; 516 /* add non-zero cols of B into the sorted linked list lnk */ 517 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 518 } 519 /* add possible missing diagonal entry */ 520 if (C->force_diagonals) { 521 ierr = PetscLLCondensedAddSorted_Scalable(1,&i,lnk);CHKERRQ(ierr); 522 } 523 524 cnzi = lnk[0]; 525 526 /* If free space is not available, make more free space */ 527 /* Double the amount of total space in the list */ 528 if (current_space->local_remaining<cnzi) { 529 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 530 ndouble++; 531 } 532 533 /* Copy data into free space, then initialize lnk */ 534 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 535 536 current_space->array += cnzi; 537 current_space->local_used += cnzi; 538 current_space->local_remaining -= cnzi; 539 540 ci[i+1] = ci[i] + cnzi; 541 } 542 543 /* Column indices are in the list of free space */ 544 /* Allocate space for cj, initialize cj, and */ 545 /* destroy list of free space and other temporary array(s) */ 546 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 547 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 548 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 549 550 /* Allocate space for ca */ 551 /*-----------------------*/ 552 ierr = PetscCalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 553 554 /* put together the new symbolic matrix */ 555 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 556 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 557 558 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 559 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 560 c = (Mat_SeqAIJ*)(C->data); 561 c->free_a = PETSC_TRUE; 562 c->free_ij = PETSC_TRUE; 563 c->nonew = 0; 564 565 /* slower, less memory */ 566 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; 567 568 /* set MatInfo */ 569 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 570 if (afill < 1.0) afill = 1.0; 571 c->maxnz = ci[am]; 572 c->nz = ci[am]; 573 C->info.mallocs = ndouble; 574 C->info.fill_ratio_given = fill; 575 C->info.fill_ratio_needed = afill; 576 577 #if defined(PETSC_USE_INFO) 578 if (ci[am]) { 579 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 580 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 581 } else { 582 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 583 } 584 #endif 585 PetscFunctionReturn(0); 586 } 587 588 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat C) 589 { 590 PetscErrorCode ierr; 591 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 592 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 593 PetscInt *ci,*cj,*bb; 594 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 595 PetscReal afill; 596 PetscInt i,j,col,ndouble = 0; 597 PetscFreeSpaceList free_space=NULL,current_space=NULL; 598 PetscHeap h; 599 600 PetscFunctionBegin; 601 /* Get ci and cj - by merging sorted rows using a heap */ 602 /*---------------------------------------------------------------------------------------------*/ 603 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 604 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 605 ci[0] = 0; 606 607 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 608 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 609 current_space = free_space; 610 611 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 612 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 613 614 /* Determine ci and cj */ 615 for (i=0; i<am; i++) { 616 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 */ 617 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 618 ci[i+1] = ci[i]; 619 /* Populate the min heap */ 620 for (j=0; j<anzi; j++) { 621 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 622 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 623 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 624 } 625 } 626 /* Pick off the min element, adding it to free space */ 627 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 628 while (j >= 0) { 629 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 630 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 631 ndouble++; 632 } 633 *(current_space->array++) = col; 634 current_space->local_used++; 635 current_space->local_remaining--; 636 ci[i+1]++; 637 638 /* stash if anything else remains in this row of B */ 639 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 640 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 641 PetscInt j2,col2; 642 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 643 if (col2 != col) break; 644 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 645 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 646 } 647 /* Put any stashed elements back into the min heap */ 648 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 649 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 650 } 651 } 652 ierr = PetscFree(bb);CHKERRQ(ierr); 653 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 654 655 /* Column indices are in the list of free space */ 656 /* Allocate space for cj, initialize cj, and */ 657 /* destroy list of free space and other temporary array(s) */ 658 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 659 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 660 661 /* put together the new symbolic matrix */ 662 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 663 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 664 665 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 666 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 667 c = (Mat_SeqAIJ*)(C->data); 668 c->free_a = PETSC_TRUE; 669 c->free_ij = PETSC_TRUE; 670 c->nonew = 0; 671 672 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 673 674 /* set MatInfo */ 675 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 676 if (afill < 1.0) afill = 1.0; 677 c->maxnz = ci[am]; 678 c->nz = ci[am]; 679 C->info.mallocs = ndouble; 680 C->info.fill_ratio_given = fill; 681 C->info.fill_ratio_needed = afill; 682 683 #if defined(PETSC_USE_INFO) 684 if (ci[am]) { 685 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 686 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 687 } else { 688 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 689 } 690 #endif 691 PetscFunctionReturn(0); 692 } 693 694 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat C) 695 { 696 PetscErrorCode ierr; 697 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 698 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 699 PetscInt *ci,*cj,*bb; 700 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 701 PetscReal afill; 702 PetscInt i,j,col,ndouble = 0; 703 PetscFreeSpaceList free_space=NULL,current_space=NULL; 704 PetscHeap h; 705 PetscBT bt; 706 707 PetscFunctionBegin; 708 /* Get ci and cj - using a heap for the sorted rows, but use BT so that each index is only added once */ 709 /*---------------------------------------------------------------------------------------------*/ 710 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 711 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 712 ci[0] = 0; 713 714 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 715 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 716 717 current_space = free_space; 718 719 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 720 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 721 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 722 723 /* Determine ci and cj */ 724 for (i=0; i<am; i++) { 725 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 */ 726 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 727 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 728 ci[i+1] = ci[i]; 729 /* Populate the min heap */ 730 for (j=0; j<anzi; j++) { 731 PetscInt brow = acol[j]; 732 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 733 PetscInt bcol = bj[bb[j]]; 734 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 735 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 736 bb[j]++; 737 break; 738 } 739 } 740 } 741 /* Pick off the min element, adding it to free space */ 742 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 743 while (j >= 0) { 744 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 745 fptr = NULL; /* need PetscBTMemzero */ 746 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 747 ndouble++; 748 } 749 *(current_space->array++) = col; 750 current_space->local_used++; 751 current_space->local_remaining--; 752 ci[i+1]++; 753 754 /* stash if anything else remains in this row of B */ 755 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 756 PetscInt bcol = bj[bb[j]]; 757 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 758 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 759 bb[j]++; 760 break; 761 } 762 } 763 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 764 } 765 if (fptr) { /* Clear the bits for this row */ 766 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 767 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 768 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 769 } 770 } 771 ierr = PetscFree(bb);CHKERRQ(ierr); 772 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 773 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 774 775 /* Column indices are in the list of free space */ 776 /* Allocate space for cj, initialize cj, and */ 777 /* destroy list of free space and other temporary array(s) */ 778 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 779 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 780 781 /* put together the new symbolic matrix */ 782 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 783 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 784 785 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 786 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 787 c = (Mat_SeqAIJ*)(C->data); 788 c->free_a = PETSC_TRUE; 789 c->free_ij = PETSC_TRUE; 790 c->nonew = 0; 791 792 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 793 794 /* set MatInfo */ 795 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 796 if (afill < 1.0) afill = 1.0; 797 c->maxnz = ci[am]; 798 c->nz = ci[am]; 799 C->info.mallocs = ndouble; 800 C->info.fill_ratio_given = fill; 801 C->info.fill_ratio_needed = afill; 802 803 #if defined(PETSC_USE_INFO) 804 if (ci[am]) { 805 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 806 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 807 } else { 808 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 809 } 810 #endif 811 PetscFunctionReturn(0); 812 } 813 814 815 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat A,Mat B,PetscReal fill,Mat C) 816 { 817 PetscErrorCode ierr; 818 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 819 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j,*inputi,*inputj,*inputcol,*inputcol_L1; 820 PetscInt *ci,*cj,*outputj,worki_L1[9],worki_L2[9]; 821 PetscInt c_maxmem,a_maxrownnz=0,a_rownnz; 822 const PetscInt workcol[8]={0,1,2,3,4,5,6,7}; 823 const PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 824 const PetscInt *brow_ptr[8],*brow_end[8]; 825 PetscInt window[8]; 826 PetscInt window_min,old_window_min,ci_nnz,outputi_nnz=0,L1_nrows,L2_nrows; 827 PetscInt i,k,ndouble=0,L1_rowsleft,rowsleft; 828 PetscReal afill; 829 PetscInt *workj_L1,*workj_L2,*workj_L3; 830 PetscInt L1_nnz,L2_nnz; 831 832 /* Step 1: Get upper bound on memory required for allocation. 833 Because of the way virtual memory works, 834 only the memory pages that are actually needed will be physically allocated. */ 835 PetscFunctionBegin; 836 ierr = PetscMalloc1(am+1,&ci);CHKERRQ(ierr); 837 for (i=0; i<am; i++) { 838 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 */ 839 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 840 a_rownnz = 0; 841 for (k=0; k<anzi; ++k) { 842 a_rownnz += bi[acol[k]+1] - bi[acol[k]]; 843 if (a_rownnz > bn) { 844 a_rownnz = bn; 845 break; 846 } 847 } 848 a_maxrownnz = PetscMax(a_maxrownnz, a_rownnz); 849 } 850 /* temporary work areas for merging rows */ 851 ierr = PetscMalloc1(a_maxrownnz*8,&workj_L1);CHKERRQ(ierr); 852 ierr = PetscMalloc1(a_maxrownnz*8,&workj_L2);CHKERRQ(ierr); 853 ierr = PetscMalloc1(a_maxrownnz,&workj_L3);CHKERRQ(ierr); 854 855 /* This should be enough for almost all matrices. If not, memory is reallocated later. */ 856 c_maxmem = 8*(ai[am]+bi[bm]); 857 /* Step 2: Populate pattern for C */ 858 ierr = PetscMalloc1(c_maxmem,&cj);CHKERRQ(ierr); 859 860 ci_nnz = 0; 861 ci[0] = 0; 862 worki_L1[0] = 0; 863 worki_L2[0] = 0; 864 for (i=0; i<am; i++) { 865 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 */ 866 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 867 rowsleft = anzi; 868 inputcol_L1 = acol; 869 L2_nnz = 0; 870 L2_nrows = 1; /* Number of rows to be merged on Level 3. output of L3 already exists -> initial value 1 */ 871 worki_L2[1] = 0; 872 outputi_nnz = 0; 873 874 /* If the number of indices in C so far + the max number of columns in the next row > c_maxmem -> allocate more memory */ 875 while (ci_nnz+a_maxrownnz > c_maxmem) { 876 c_maxmem *= 2; 877 ndouble++; 878 ierr = PetscRealloc(sizeof(PetscInt)*c_maxmem,&cj);CHKERRQ(ierr); 879 } 880 881 while (rowsleft) { 882 L1_rowsleft = PetscMin(64, rowsleft); /* In the inner loop max 64 rows of B can be merged */ 883 L1_nrows = 0; 884 L1_nnz = 0; 885 inputcol = inputcol_L1; 886 inputi = bi; 887 inputj = bj; 888 889 /* The following macro is used to specialize for small rows in A. 890 This helps with compiler unrolling, improving performance substantially. 891 Input: inputj inputi inputcol bn 892 Output: outputj outputi_nnz */ 893 #define MatMatMultSymbolic_RowMergeMacro(ANNZ) \ 894 window_min = bn; \ 895 outputi_nnz = 0; \ 896 for (k=0; k<ANNZ; ++k) { \ 897 brow_ptr[k] = inputj + inputi[inputcol[k]]; \ 898 brow_end[k] = inputj + inputi[inputcol[k]+1]; \ 899 window[k] = (brow_ptr[k] != brow_end[k]) ? *brow_ptr[k] : bn; \ 900 window_min = PetscMin(window[k], window_min); \ 901 } \ 902 while (window_min < bn) { \ 903 outputj[outputi_nnz++] = window_min; \ 904 /* advance front and compute new minimum */ \ 905 old_window_min = window_min; \ 906 window_min = bn; \ 907 for (k=0; k<ANNZ; ++k) { \ 908 if (window[k] == old_window_min) { \ 909 brow_ptr[k]++; \ 910 window[k] = (brow_ptr[k] != brow_end[k]) ? *brow_ptr[k] : bn; \ 911 } \ 912 window_min = PetscMin(window[k], window_min); \ 913 } \ 914 } 915 916 /************** L E V E L 1 ***************/ 917 /* Merge up to 8 rows of B to L1 work array*/ 918 while (L1_rowsleft) { 919 outputi_nnz = 0; 920 if (anzi > 8) outputj = workj_L1 + L1_nnz; /* Level 1 rowmerge*/ 921 else outputj = cj + ci_nnz; /* Merge directly to C */ 922 923 switch (L1_rowsleft) { 924 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 925 brow_end[0] = inputj + inputi[inputcol[0]+1]; 926 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 927 inputcol += L1_rowsleft; 928 rowsleft -= L1_rowsleft; 929 L1_rowsleft = 0; 930 break; 931 case 2: MatMatMultSymbolic_RowMergeMacro(2); 932 inputcol += L1_rowsleft; 933 rowsleft -= L1_rowsleft; 934 L1_rowsleft = 0; 935 break; 936 case 3: MatMatMultSymbolic_RowMergeMacro(3); 937 inputcol += L1_rowsleft; 938 rowsleft -= L1_rowsleft; 939 L1_rowsleft = 0; 940 break; 941 case 4: MatMatMultSymbolic_RowMergeMacro(4); 942 inputcol += L1_rowsleft; 943 rowsleft -= L1_rowsleft; 944 L1_rowsleft = 0; 945 break; 946 case 5: MatMatMultSymbolic_RowMergeMacro(5); 947 inputcol += L1_rowsleft; 948 rowsleft -= L1_rowsleft; 949 L1_rowsleft = 0; 950 break; 951 case 6: MatMatMultSymbolic_RowMergeMacro(6); 952 inputcol += L1_rowsleft; 953 rowsleft -= L1_rowsleft; 954 L1_rowsleft = 0; 955 break; 956 case 7: MatMatMultSymbolic_RowMergeMacro(7); 957 inputcol += L1_rowsleft; 958 rowsleft -= L1_rowsleft; 959 L1_rowsleft = 0; 960 break; 961 default: MatMatMultSymbolic_RowMergeMacro(8); 962 inputcol += 8; 963 rowsleft -= 8; 964 L1_rowsleft -= 8; 965 break; 966 } 967 inputcol_L1 = inputcol; 968 L1_nnz += outputi_nnz; 969 worki_L1[++L1_nrows] = L1_nnz; 970 } 971 972 /********************** L E V E L 2 ************************/ 973 /* Merge from L1 work array to either C or to L2 work array */ 974 if (anzi > 8) { 975 inputi = worki_L1; 976 inputj = workj_L1; 977 inputcol = workcol; 978 outputi_nnz = 0; 979 980 if (anzi <= 64) outputj = cj + ci_nnz; /* Merge from L1 work array to C */ 981 else outputj = workj_L2 + L2_nnz; /* Merge from L1 work array to L2 work array */ 982 983 switch (L1_nrows) { 984 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 985 brow_end[0] = inputj + inputi[inputcol[0]+1]; 986 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 987 break; 988 case 2: MatMatMultSymbolic_RowMergeMacro(2); break; 989 case 3: MatMatMultSymbolic_RowMergeMacro(3); break; 990 case 4: MatMatMultSymbolic_RowMergeMacro(4); break; 991 case 5: MatMatMultSymbolic_RowMergeMacro(5); break; 992 case 6: MatMatMultSymbolic_RowMergeMacro(6); break; 993 case 7: MatMatMultSymbolic_RowMergeMacro(7); break; 994 case 8: MatMatMultSymbolic_RowMergeMacro(8); break; 995 default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatMult logic error: Not merging 1-8 rows from L1 work array!"); 996 } 997 L2_nnz += outputi_nnz; 998 worki_L2[++L2_nrows] = L2_nnz; 999 1000 /************************ L E V E L 3 **********************/ 1001 /* Merge from L2 work array to either C or to L2 work array */ 1002 if (anzi > 64 && (L2_nrows == 8 || rowsleft == 0)) { 1003 inputi = worki_L2; 1004 inputj = workj_L2; 1005 inputcol = workcol; 1006 outputi_nnz = 0; 1007 if (rowsleft) outputj = workj_L3; 1008 else outputj = cj + ci_nnz; 1009 switch (L2_nrows) { 1010 case 1: brow_ptr[0] = inputj + inputi[inputcol[0]]; 1011 brow_end[0] = inputj + inputi[inputcol[0]+1]; 1012 for (; brow_ptr[0] != brow_end[0]; ++brow_ptr[0]) outputj[outputi_nnz++] = *brow_ptr[0]; /* copy row in b over */ 1013 break; 1014 case 2: MatMatMultSymbolic_RowMergeMacro(2); break; 1015 case 3: MatMatMultSymbolic_RowMergeMacro(3); break; 1016 case 4: MatMatMultSymbolic_RowMergeMacro(4); break; 1017 case 5: MatMatMultSymbolic_RowMergeMacro(5); break; 1018 case 6: MatMatMultSymbolic_RowMergeMacro(6); break; 1019 case 7: MatMatMultSymbolic_RowMergeMacro(7); break; 1020 case 8: MatMatMultSymbolic_RowMergeMacro(8); break; 1021 default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatMult logic error: Not merging 1-8 rows from L2 work array!"); 1022 } 1023 L2_nrows = 1; 1024 L2_nnz = outputi_nnz; 1025 worki_L2[1] = outputi_nnz; 1026 /* Copy to workj_L2 */ 1027 if (rowsleft) { 1028 for (k=0; k<outputi_nnz; ++k) workj_L2[k] = outputj[k]; 1029 } 1030 } 1031 } 1032 } /* while (rowsleft) */ 1033 #undef MatMatMultSymbolic_RowMergeMacro 1034 1035 /* terminate current row */ 1036 ci_nnz += outputi_nnz; 1037 ci[i+1] = ci_nnz; 1038 } 1039 1040 /* Step 3: Create the new symbolic matrix */ 1041 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 1042 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 1043 1044 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 1045 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 1046 c = (Mat_SeqAIJ*)(C->data); 1047 c->free_a = PETSC_TRUE; 1048 c->free_ij = PETSC_TRUE; 1049 c->nonew = 0; 1050 1051 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 1052 1053 /* set MatInfo */ 1054 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 1055 if (afill < 1.0) afill = 1.0; 1056 c->maxnz = ci[am]; 1057 c->nz = ci[am]; 1058 C->info.mallocs = ndouble; 1059 C->info.fill_ratio_given = fill; 1060 C->info.fill_ratio_needed = afill; 1061 1062 #if defined(PETSC_USE_INFO) 1063 if (ci[am]) { 1064 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 1065 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1066 } else { 1067 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 1068 } 1069 #endif 1070 1071 /* Step 4: Free temporary work areas */ 1072 ierr = PetscFree(workj_L1);CHKERRQ(ierr); 1073 ierr = PetscFree(workj_L2);CHKERRQ(ierr); 1074 ierr = PetscFree(workj_L3);CHKERRQ(ierr); 1075 PetscFunctionReturn(0); 1076 } 1077 1078 /* concatenate unique entries and then sort */ 1079 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat A,Mat B,PetscReal fill,Mat C) 1080 { 1081 PetscErrorCode ierr; 1082 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 1083 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 1084 PetscInt *ci,*cj,bcol; 1085 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 1086 PetscReal afill; 1087 PetscInt i,j,ndouble = 0; 1088 PetscSegBuffer seg,segrow; 1089 char *seen; 1090 1091 PetscFunctionBegin; 1092 ierr = PetscMalloc1(am+1,&ci);CHKERRQ(ierr); 1093 ci[0] = 0; 1094 1095 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 1096 ierr = PetscSegBufferCreate(sizeof(PetscInt),(PetscInt)(fill*(ai[am]+bi[bm])),&seg);CHKERRQ(ierr); 1097 ierr = PetscSegBufferCreate(sizeof(PetscInt),100,&segrow);CHKERRQ(ierr); 1098 ierr = PetscCalloc1(bn,&seen);CHKERRQ(ierr); 1099 1100 /* Determine ci and cj */ 1101 for (i=0; i<am; i++) { 1102 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 */ 1103 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 1104 PetscInt packlen = 0,*PETSC_RESTRICT crow; 1105 1106 /* Pack segrow */ 1107 for (j=0; j<anzi; j++) { 1108 PetscInt brow = acol[j],bjstart = bi[brow],bjend = bi[brow+1],k; 1109 for (k=bjstart; k<bjend; k++) { 1110 bcol = bj[k]; 1111 if (!seen[bcol]) { /* new entry */ 1112 PetscInt *PETSC_RESTRICT slot; 1113 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 1114 *slot = bcol; 1115 seen[bcol] = 1; 1116 packlen++; 1117 } 1118 } 1119 } 1120 1121 /* Check i-th diagonal entry */ 1122 if (C->force_diagonals && !seen[i]) { 1123 PetscInt *PETSC_RESTRICT slot; 1124 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 1125 *slot = i; 1126 seen[i] = 1; 1127 packlen++; 1128 } 1129 1130 ierr = PetscSegBufferGetInts(seg,packlen,&crow);CHKERRQ(ierr); 1131 ierr = PetscSegBufferExtractTo(segrow,crow);CHKERRQ(ierr); 1132 ierr = PetscSortInt(packlen,crow);CHKERRQ(ierr); 1133 ci[i+1] = ci[i] + packlen; 1134 for (j=0; j<packlen; j++) seen[crow[j]] = 0; 1135 } 1136 ierr = PetscSegBufferDestroy(&segrow);CHKERRQ(ierr); 1137 ierr = PetscFree(seen);CHKERRQ(ierr); 1138 1139 /* Column indices are in the segmented buffer */ 1140 ierr = PetscSegBufferExtractAlloc(seg,&cj);CHKERRQ(ierr); 1141 ierr = PetscSegBufferDestroy(&seg);CHKERRQ(ierr); 1142 1143 /* put together the new symbolic matrix */ 1144 ierr = MatSetSeqAIJWithArrays_private(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,((PetscObject)A)->type_name,C);CHKERRQ(ierr); 1145 ierr = MatSetBlockSizesFromMats(C,A,B);CHKERRQ(ierr); 1146 1147 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 1148 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 1149 c = (Mat_SeqAIJ*)(C->data); 1150 c->free_a = PETSC_TRUE; 1151 c->free_ij = PETSC_TRUE; 1152 c->nonew = 0; 1153 1154 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted; 1155 1156 /* set MatInfo */ 1157 afill = (PetscReal)ci[am]/PetscMax(ai[am]+bi[bm],1) + 1.e-5; 1158 if (afill < 1.0) afill = 1.0; 1159 c->maxnz = ci[am]; 1160 c->nz = ci[am]; 1161 C->info.mallocs = ndouble; 1162 C->info.fill_ratio_given = fill; 1163 C->info.fill_ratio_needed = afill; 1164 1165 #if defined(PETSC_USE_INFO) 1166 if (ci[am]) { 1167 ierr = PetscInfo3(C,"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 1168 ierr = PetscInfo1(C,"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 1169 } else { 1170 ierr = PetscInfo(C,"Empty matrix product\n");CHKERRQ(ierr); 1171 } 1172 #endif 1173 PetscFunctionReturn(0); 1174 } 1175 1176 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(void *data) 1177 { 1178 PetscErrorCode ierr; 1179 Mat_MatMatTransMult *abt=(Mat_MatMatTransMult *)data; 1180 1181 PetscFunctionBegin; 1182 ierr = MatTransposeColoringDestroy(&abt->matcoloring);CHKERRQ(ierr); 1183 ierr = MatDestroy(&abt->Bt_den);CHKERRQ(ierr); 1184 ierr = MatDestroy(&abt->ABt_den);CHKERRQ(ierr); 1185 ierr = PetscFree(abt);CHKERRQ(ierr); 1186 PetscFunctionReturn(0); 1187 } 1188 1189 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 1190 { 1191 PetscErrorCode ierr; 1192 Mat Bt; 1193 PetscInt *bti,*btj; 1194 Mat_MatMatTransMult *abt; 1195 Mat_Product *product = C->product; 1196 char *alg; 1197 1198 PetscFunctionBegin; 1199 if (!product) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1200 if (product->data) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 1201 1202 /* create symbolic Bt */ 1203 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 1204 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 1205 ierr = MatSetBlockSizes(Bt,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 1206 ierr = MatSetType(Bt,((PetscObject)A)->type_name);CHKERRQ(ierr); 1207 1208 /* get symbolic C=A*Bt */ 1209 ierr = PetscStrallocpy(product->alg,&alg);CHKERRQ(ierr); 1210 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); /* set algorithm for C = A*Bt */ 1211 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 1212 ierr = MatProductSetAlgorithm(C,alg);CHKERRQ(ierr); /* resume original algorithm for ABt product */ 1213 ierr = PetscFree(alg);CHKERRQ(ierr); 1214 1215 /* create a supporting struct for reuse intermidiate dense matrices with matcoloring */ 1216 ierr = PetscNew(&abt);CHKERRQ(ierr); 1217 1218 product->data = abt; 1219 product->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 1220 1221 C->ops->mattransposemultnumeric = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ; 1222 1223 abt->usecoloring = PETSC_FALSE; 1224 ierr = PetscStrcmp(product->alg,"color",&abt->usecoloring);CHKERRQ(ierr); 1225 if (abt->usecoloring) { 1226 /* Create MatTransposeColoring from symbolic C=A*B^T */ 1227 MatTransposeColoring matcoloring; 1228 MatColoring coloring; 1229 ISColoring iscoloring; 1230 Mat Bt_dense,C_dense; 1231 1232 /* inode causes memory problem */ 1233 ierr = MatSetOption(C,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 1234 1235 ierr = MatColoringCreate(C,&coloring);CHKERRQ(ierr); 1236 ierr = MatColoringSetDistance(coloring,2);CHKERRQ(ierr); 1237 ierr = MatColoringSetType(coloring,MATCOLORINGSL);CHKERRQ(ierr); 1238 ierr = MatColoringSetFromOptions(coloring);CHKERRQ(ierr); 1239 ierr = MatColoringApply(coloring,&iscoloring);CHKERRQ(ierr); 1240 ierr = MatColoringDestroy(&coloring);CHKERRQ(ierr); 1241 ierr = MatTransposeColoringCreate(C,iscoloring,&matcoloring);CHKERRQ(ierr); 1242 1243 abt->matcoloring = matcoloring; 1244 1245 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 1246 1247 /* Create Bt_dense and C_dense = A*Bt_dense */ 1248 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 1249 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 1250 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 1251 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 1252 1253 Bt_dense->assembled = PETSC_TRUE; 1254 abt->Bt_den = Bt_dense; 1255 1256 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 1257 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 1258 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 1259 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 1260 1261 Bt_dense->assembled = PETSC_TRUE; 1262 abt->ABt_den = C_dense; 1263 1264 #if defined(PETSC_USE_INFO) 1265 { 1266 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 1267 ierr = PetscInfo7(C,"Use coloring of C=A*B^T; B^T: %D %D, Bt_dense: %D,%D; Cnz %D / (cm*ncolors %D) = %g\n",B->cmap->n,B->rmap->n,Bt_dense->rmap->n,Bt_dense->cmap->n,c->nz,A->rmap->n*matcoloring->ncolors,(PetscReal)(c->nz)/(A->rmap->n*matcoloring->ncolors));CHKERRQ(ierr); 1268 } 1269 #endif 1270 } 1271 /* clean up */ 1272 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 1273 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 1274 PetscFunctionReturn(0); 1275 } 1276 1277 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1278 { 1279 PetscErrorCode ierr; 1280 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1281 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 1282 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 1283 PetscLogDouble flops=0.0; 1284 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 1285 Mat_MatMatTransMult *abt; 1286 Mat_Product *product = C->product; 1287 1288 PetscFunctionBegin; 1289 if (!product) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1290 abt = (Mat_MatMatTransMult *)product->data; 1291 if (!abt) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1292 /* clear old values in C */ 1293 if (!c->a) { 1294 ierr = PetscCalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 1295 c->a = ca; 1296 c->free_a = PETSC_TRUE; 1297 } else { 1298 ca = c->a; 1299 ierr = PetscArrayzero(ca,ci[cm]+1);CHKERRQ(ierr); 1300 } 1301 1302 if (abt->usecoloring) { 1303 MatTransposeColoring matcoloring = abt->matcoloring; 1304 Mat Bt_dense,C_dense = abt->ABt_den; 1305 1306 /* Get Bt_dense by Apply MatTransposeColoring to B */ 1307 Bt_dense = abt->Bt_den; 1308 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 1309 1310 /* C_dense = A*Bt_dense */ 1311 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 1312 1313 /* Recover C from C_dense */ 1314 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 1315 PetscFunctionReturn(0); 1316 } 1317 1318 for (i=0; i<cm; i++) { 1319 anzi = ai[i+1] - ai[i]; 1320 acol = aj + ai[i]; 1321 aval = aa + ai[i]; 1322 cnzi = ci[i+1] - ci[i]; 1323 ccol = cj + ci[i]; 1324 cval = ca + ci[i]; 1325 for (j=0; j<cnzi; j++) { 1326 brow = ccol[j]; 1327 bnzj = bi[brow+1] - bi[brow]; 1328 bcol = bj + bi[brow]; 1329 bval = ba + bi[brow]; 1330 1331 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 1332 nexta = 0; nextb = 0; 1333 while (nexta<anzi && nextb<bnzj) { 1334 while (nexta < anzi && acol[nexta] < bcol[nextb]) nexta++; 1335 if (nexta == anzi) break; 1336 while (nextb < bnzj && acol[nexta] > bcol[nextb]) nextb++; 1337 if (nextb == bnzj) break; 1338 if (acol[nexta] == bcol[nextb]) { 1339 cval[j] += aval[nexta]*bval[nextb]; 1340 nexta++; nextb++; 1341 flops += 2; 1342 } 1343 } 1344 } 1345 } 1346 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1347 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1348 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1349 PetscFunctionReturn(0); 1350 } 1351 1352 PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(void *data) 1353 { 1354 PetscErrorCode ierr; 1355 Mat_MatTransMatMult *atb = (Mat_MatTransMatMult*)data; 1356 1357 PetscFunctionBegin; 1358 ierr = MatDestroy(&atb->At);CHKERRQ(ierr); 1359 if (atb->destroy) { 1360 ierr = (*atb->destroy)(atb->data);CHKERRQ(ierr); 1361 } 1362 ierr = PetscFree(atb);CHKERRQ(ierr); 1363 PetscFunctionReturn(0); 1364 } 1365 1366 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 1367 { 1368 PetscErrorCode ierr; 1369 Mat At = NULL; 1370 PetscInt *ati,*atj; 1371 Mat_Product *product = C->product; 1372 PetscBool flg,def,square; 1373 1374 PetscFunctionBegin; 1375 MatCheckProduct(C,4); 1376 square = (PetscBool)(A == B && A->symmetric && A->symmetric_set); 1377 /* outerproduct */ 1378 ierr = PetscStrcmp(product->alg,"outerproduct",&flg);CHKERRQ(ierr); 1379 if (flg) { 1380 /* create symbolic At */ 1381 if (!square) { 1382 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1383 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1384 ierr = MatSetBlockSizes(At,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 1385 ierr = MatSetType(At,((PetscObject)A)->type_name);CHKERRQ(ierr); 1386 } 1387 /* get symbolic C=At*B */ 1388 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); 1389 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(square ? A : At,B,fill,C);CHKERRQ(ierr); 1390 1391 /* clean up */ 1392 if (!square) { 1393 ierr = MatDestroy(&At);CHKERRQ(ierr); 1394 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1395 } 1396 1397 C->ops->mattransposemultnumeric = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ; /* outerproduct */ 1398 ierr = MatProductSetAlgorithm(C,"outerproduct");CHKERRQ(ierr); 1399 PetscFunctionReturn(0); 1400 } 1401 1402 /* matmatmult */ 1403 ierr = PetscStrcmp(product->alg,"default",&def);CHKERRQ(ierr); 1404 ierr = PetscStrcmp(product->alg,"at*b",&flg);CHKERRQ(ierr); 1405 if (flg || def) { 1406 Mat_MatTransMatMult *atb; 1407 1408 if (product->data) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Extra product struct not empty"); 1409 ierr = PetscNew(&atb);CHKERRQ(ierr); 1410 if (!square) { 1411 ierr = MatTranspose_SeqAIJ(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 1412 } 1413 ierr = MatProductSetAlgorithm(C,"sorted");CHKERRQ(ierr); 1414 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(square ? A : At,B,fill,C);CHKERRQ(ierr); 1415 ierr = MatProductSetAlgorithm(C,"at*b");CHKERRQ(ierr); 1416 product->data = atb; 1417 product->destroy = MatDestroy_SeqAIJ_MatTransMatMult; 1418 atb->At = At; 1419 atb->updateAt = PETSC_FALSE; /* because At is computed here */ 1420 1421 C->ops->mattransposemultnumeric = NULL; /* see MatProductNumeric_AtB_SeqAIJ_SeqAIJ */ 1422 PetscFunctionReturn(0); 1423 } 1424 1425 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Mat Product Algorithm is not supported"); 1426 } 1427 1428 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1429 { 1430 PetscErrorCode ierr; 1431 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1432 PetscInt am=A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1433 PetscInt cm=C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1434 PetscLogDouble flops=0.0; 1435 MatScalar *aa=a->a,*ba,*ca,*caj; 1436 1437 PetscFunctionBegin; 1438 if (!c->a) { 1439 ierr = PetscCalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 1440 1441 c->a = ca; 1442 c->free_a = PETSC_TRUE; 1443 } else { 1444 ca = c->a; 1445 ierr = PetscArrayzero(ca,ci[cm]);CHKERRQ(ierr); 1446 } 1447 1448 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1449 for (i=0; i<am; i++) { 1450 bj = b->j + bi[i]; 1451 ba = b->a + bi[i]; 1452 bnzi = bi[i+1] - bi[i]; 1453 anzi = ai[i+1] - ai[i]; 1454 for (j=0; j<anzi; j++) { 1455 nextb = 0; 1456 crow = *aj++; 1457 cjj = cj + ci[crow]; 1458 caj = ca + ci[crow]; 1459 /* perform sparse axpy operation. Note cjj includes bj. */ 1460 for (k=0; nextb<bnzi; k++) { 1461 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1462 caj[k] += (*aa)*(*(ba+nextb)); 1463 nextb++; 1464 } 1465 } 1466 flops += 2*bnzi; 1467 aa++; 1468 } 1469 } 1470 1471 /* Assemble the final matrix and clean up */ 1472 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1473 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1474 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1475 PetscFunctionReturn(0); 1476 } 1477 1478 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat C) 1479 { 1480 PetscErrorCode ierr; 1481 1482 PetscFunctionBegin; 1483 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1484 C->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1485 PetscFunctionReturn(0); 1486 } 1487 1488 PETSC_INTERN PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C,const PetscBool add) 1489 { 1490 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1491 Mat_SeqDense *bd=(Mat_SeqDense*)B->data; 1492 Mat_SeqDense *cd=(Mat_SeqDense*)C->data; 1493 PetscErrorCode ierr; 1494 PetscScalar *c,r1,r2,r3,r4,*c1,*c2,*c3,*c4; 1495 const PetscScalar *aa,*b,*b1,*b2,*b3,*b4,*av; 1496 const PetscInt *aj; 1497 PetscInt cm=C->rmap->n,cn=B->cmap->n,bm=bd->lda,am=A->rmap->n; 1498 PetscInt clda=cd->lda; 1499 PetscInt am4=4*clda,bm4=4*bm,col,i,j,n; 1500 1501 PetscFunctionBegin; 1502 if (!cm || !cn) PetscFunctionReturn(0); 1503 ierr = MatSeqAIJGetArrayRead(A,&av);CHKERRQ(ierr); 1504 if (add) { 1505 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1506 } else { 1507 ierr = MatDenseGetArrayWrite(C,&c);CHKERRQ(ierr); 1508 } 1509 ierr = MatDenseGetArrayRead(B,&b);CHKERRQ(ierr); 1510 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1511 c1 = c; c2 = c1 + clda; c3 = c2 + clda; c4 = c3 + clda; 1512 for (col=0; col<(cn/4)*4; col += 4) { /* over columns of C */ 1513 for (i=0; i<am; i++) { /* over rows of A in those columns */ 1514 r1 = r2 = r3 = r4 = 0.0; 1515 n = a->i[i+1] - a->i[i]; 1516 aj = a->j + a->i[i]; 1517 aa = av + a->i[i]; 1518 for (j=0; j<n; j++) { 1519 const PetscScalar aatmp = aa[j]; 1520 const PetscInt ajtmp = aj[j]; 1521 r1 += aatmp*b1[ajtmp]; 1522 r2 += aatmp*b2[ajtmp]; 1523 r3 += aatmp*b3[ajtmp]; 1524 r4 += aatmp*b4[ajtmp]; 1525 } 1526 if (add) { 1527 c1[i] += r1; 1528 c2[i] += r2; 1529 c3[i] += r3; 1530 c4[i] += r4; 1531 } else { 1532 c1[i] = r1; 1533 c2[i] = r2; 1534 c3[i] = r3; 1535 c4[i] = r4; 1536 } 1537 } 1538 b1 += bm4; b2 += bm4; b3 += bm4; b4 += bm4; 1539 c1 += am4; c2 += am4; c3 += am4; c4 += am4; 1540 } 1541 /* process remaining columns */ 1542 if (col != cn) { 1543 PetscInt rc = cn-col; 1544 1545 if (rc == 1) { 1546 for (i=0; i<am; i++) { 1547 r1 = 0.0; 1548 n = a->i[i+1] - a->i[i]; 1549 aj = a->j + a->i[i]; 1550 aa = av + a->i[i]; 1551 for (j=0; j<n; j++) r1 += aa[j]*b1[aj[j]]; 1552 if (add) c1[i] += r1; 1553 else c1[i] = r1; 1554 } 1555 } else if (rc == 2) { 1556 for (i=0; i<am; i++) { 1557 r1 = r2 = 0.0; 1558 n = a->i[i+1] - a->i[i]; 1559 aj = a->j + a->i[i]; 1560 aa = av + a->i[i]; 1561 for (j=0; j<n; j++) { 1562 const PetscScalar aatmp = aa[j]; 1563 const PetscInt ajtmp = aj[j]; 1564 r1 += aatmp*b1[ajtmp]; 1565 r2 += aatmp*b2[ajtmp]; 1566 } 1567 if (add) { 1568 c1[i] += r1; 1569 c2[i] += r2; 1570 } else { 1571 c1[i] = r1; 1572 c2[i] = r2; 1573 } 1574 } 1575 } else { 1576 for (i=0; i<am; i++) { 1577 r1 = r2 = r3 = 0.0; 1578 n = a->i[i+1] - a->i[i]; 1579 aj = a->j + a->i[i]; 1580 aa = av + a->i[i]; 1581 for (j=0; j<n; j++) { 1582 const PetscScalar aatmp = aa[j]; 1583 const PetscInt ajtmp = aj[j]; 1584 r1 += aatmp*b1[ajtmp]; 1585 r2 += aatmp*b2[ajtmp]; 1586 r3 += aatmp*b3[ajtmp]; 1587 } 1588 if (add) { 1589 c1[i] += r1; 1590 c2[i] += r2; 1591 c3[i] += r3; 1592 } else { 1593 c1[i] = r1; 1594 c2[i] = r2; 1595 c3[i] = r3; 1596 } 1597 } 1598 } 1599 } 1600 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1601 if (add) { 1602 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1603 } else { 1604 ierr = MatDenseRestoreArrayWrite(C,&c);CHKERRQ(ierr); 1605 } 1606 ierr = MatDenseRestoreArrayRead(B,&b);CHKERRQ(ierr); 1607 ierr = MatSeqAIJRestoreArrayRead(A,&av);CHKERRQ(ierr); 1608 PetscFunctionReturn(0); 1609 } 1610 1611 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1612 { 1613 PetscErrorCode ierr; 1614 1615 PetscFunctionBegin; 1616 if (B->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number columns in A %D not equal rows in B %D\n",A->cmap->n,B->rmap->n); 1617 if (A->rmap->n != C->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number rows in C %D not equal rows in A %D\n",C->rmap->n,A->rmap->n); 1618 if (B->cmap->n != C->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number columns in B %D not equal columns in C %D\n",B->cmap->n,C->cmap->n); 1619 1620 ierr = MatMatMultNumericAdd_SeqAIJ_SeqDense(A,B,C,PETSC_FALSE);CHKERRQ(ierr); 1621 PetscFunctionReturn(0); 1622 } 1623 1624 /* ------------------------------------------------------- */ 1625 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_AB(Mat C) 1626 { 1627 PetscFunctionBegin; 1628 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqAIJ_SeqDense; 1629 C->ops->productsymbolic = MatProductSymbolic_AB; 1630 PetscFunctionReturn(0); 1631 } 1632 1633 PETSC_INTERN PetscErrorCode MatTMatTMultSymbolic_SeqAIJ_SeqDense(Mat,Mat,PetscReal,Mat); 1634 1635 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_AtB(Mat C) 1636 { 1637 PetscFunctionBegin; 1638 C->ops->transposematmultsymbolic = MatTMatTMultSymbolic_SeqAIJ_SeqDense; 1639 C->ops->productsymbolic = MatProductSymbolic_AtB; 1640 PetscFunctionReturn(0); 1641 } 1642 1643 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense_ABt(Mat C) 1644 { 1645 PetscFunctionBegin; 1646 C->ops->mattransposemultsymbolic = MatTMatTMultSymbolic_SeqAIJ_SeqDense; 1647 C->ops->productsymbolic = MatProductSymbolic_ABt; 1648 PetscFunctionReturn(0); 1649 } 1650 1651 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqAIJ_SeqDense(Mat C) 1652 { 1653 PetscErrorCode ierr; 1654 Mat_Product *product = C->product; 1655 1656 PetscFunctionBegin; 1657 switch (product->type) { 1658 case MATPRODUCT_AB: 1659 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_AB(C);CHKERRQ(ierr); 1660 break; 1661 case MATPRODUCT_AtB: 1662 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_AtB(C);CHKERRQ(ierr); 1663 break; 1664 case MATPRODUCT_ABt: 1665 ierr = MatProductSetFromOptions_SeqAIJ_SeqDense_ABt(C);CHKERRQ(ierr); 1666 break; 1667 default: 1668 break; 1669 } 1670 PetscFunctionReturn(0); 1671 } 1672 /* ------------------------------------------------------- */ 1673 static PetscErrorCode MatProductSetFromOptions_SeqXBAIJ_SeqDense_AB(Mat C) 1674 { 1675 PetscErrorCode ierr; 1676 Mat_Product *product = C->product; 1677 Mat A = product->A; 1678 PetscBool baij; 1679 1680 PetscFunctionBegin; 1681 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&baij);CHKERRQ(ierr); 1682 if (!baij) { /* A is seqsbaij */ 1683 PetscBool sbaij; 1684 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&sbaij);CHKERRQ(ierr); 1685 if (!sbaij) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_WRONGSTATE,"Mat must be either seqbaij or seqsbaij format"); 1686 1687 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqSBAIJ_SeqDense; 1688 } else { /* A is seqbaij */ 1689 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqBAIJ_SeqDense; 1690 } 1691 1692 C->ops->productsymbolic = MatProductSymbolic_AB; 1693 PetscFunctionReturn(0); 1694 } 1695 1696 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqXBAIJ_SeqDense(Mat C) 1697 { 1698 PetscErrorCode ierr; 1699 Mat_Product *product = C->product; 1700 1701 PetscFunctionBegin; 1702 MatCheckProduct(C,1); 1703 if (!product->A) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing A"); 1704 if (product->type == MATPRODUCT_AB || (product->type == MATPRODUCT_AtB && product->A->symmetric)) { 1705 ierr = MatProductSetFromOptions_SeqXBAIJ_SeqDense_AB(C);CHKERRQ(ierr); 1706 } 1707 PetscFunctionReturn(0); 1708 } 1709 1710 /* ------------------------------------------------------- */ 1711 static PetscErrorCode MatProductSetFromOptions_SeqDense_SeqAIJ_AB(Mat C) 1712 { 1713 PetscFunctionBegin; 1714 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqDense_SeqAIJ; 1715 C->ops->productsymbolic = MatProductSymbolic_AB; 1716 PetscFunctionReturn(0); 1717 } 1718 1719 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqDense_SeqAIJ(Mat C) 1720 { 1721 PetscErrorCode ierr; 1722 Mat_Product *product = C->product; 1723 1724 PetscFunctionBegin; 1725 if (product->type == MATPRODUCT_AB) { 1726 ierr = MatProductSetFromOptions_SeqDense_SeqAIJ_AB(C);CHKERRQ(ierr); 1727 } 1728 PetscFunctionReturn(0); 1729 } 1730 /* ------------------------------------------------------- */ 1731 1732 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1733 { 1734 PetscErrorCode ierr; 1735 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1736 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1737 PetscInt *bi = b->i,*bj=b->j; 1738 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1739 MatScalar *btval,*btval_den,*ba=b->a; 1740 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1741 1742 PetscFunctionBegin; 1743 btval_den=btdense->v; 1744 ierr = PetscArrayzero(btval_den,m*n);CHKERRQ(ierr); 1745 for (k=0; k<ncolors; k++) { 1746 ncolumns = coloring->ncolumns[k]; 1747 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1748 col = *(columns + colorforcol[k] + l); 1749 btcol = bj + bi[col]; 1750 btval = ba + bi[col]; 1751 anz = bi[col+1] - bi[col]; 1752 for (j=0; j<anz; j++) { 1753 brow = btcol[j]; 1754 btval_den[brow] = btval[j]; 1755 } 1756 } 1757 btval_den += m; 1758 } 1759 PetscFunctionReturn(0); 1760 } 1761 1762 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1763 { 1764 PetscErrorCode ierr; 1765 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1766 const PetscScalar *ca_den,*ca_den_ptr; 1767 PetscScalar *ca=csp->a; 1768 PetscInt k,l,m=Cden->rmap->n,ncolors=matcoloring->ncolors; 1769 PetscInt brows=matcoloring->brows,*den2sp=matcoloring->den2sp; 1770 PetscInt nrows,*row,*idx; 1771 PetscInt *rows=matcoloring->rows,*colorforrow=matcoloring->colorforrow; 1772 1773 PetscFunctionBegin; 1774 ierr = MatDenseGetArrayRead(Cden,&ca_den);CHKERRQ(ierr); 1775 1776 if (brows > 0) { 1777 PetscInt *lstart,row_end,row_start; 1778 lstart = matcoloring->lstart; 1779 ierr = PetscArrayzero(lstart,ncolors);CHKERRQ(ierr); 1780 1781 row_end = brows; 1782 if (row_end > m) row_end = m; 1783 for (row_start=0; row_start<m; row_start+=brows) { /* loop over row blocks of Csp */ 1784 ca_den_ptr = ca_den; 1785 for (k=0; k<ncolors; k++) { /* loop over colors (columns of Cden) */ 1786 nrows = matcoloring->nrows[k]; 1787 row = rows + colorforrow[k]; 1788 idx = den2sp + colorforrow[k]; 1789 for (l=lstart[k]; l<nrows; l++) { 1790 if (row[l] >= row_end) { 1791 lstart[k] = l; 1792 break; 1793 } else { 1794 ca[idx[l]] = ca_den_ptr[row[l]]; 1795 } 1796 } 1797 ca_den_ptr += m; 1798 } 1799 row_end += brows; 1800 if (row_end > m) row_end = m; 1801 } 1802 } else { /* non-blocked impl: loop over columns of Csp - slow if Csp is large */ 1803 ca_den_ptr = ca_den; 1804 for (k=0; k<ncolors; k++) { 1805 nrows = matcoloring->nrows[k]; 1806 row = rows + colorforrow[k]; 1807 idx = den2sp + colorforrow[k]; 1808 for (l=0; l<nrows; l++) { 1809 ca[idx[l]] = ca_den_ptr[row[l]]; 1810 } 1811 ca_den_ptr += m; 1812 } 1813 } 1814 1815 ierr = MatDenseRestoreArrayRead(Cden,&ca_den);CHKERRQ(ierr); 1816 #if defined(PETSC_USE_INFO) 1817 if (matcoloring->brows > 0) { 1818 ierr = PetscInfo1(Csp,"Loop over %D row blocks for den2sp\n",brows);CHKERRQ(ierr); 1819 } else { 1820 ierr = PetscInfo(Csp,"Loop over colors/columns of Cden, inefficient for large sparse matrix product \n");CHKERRQ(ierr); 1821 } 1822 #endif 1823 PetscFunctionReturn(0); 1824 } 1825 1826 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1827 { 1828 PetscErrorCode ierr; 1829 PetscInt i,n,nrows,Nbs,j,k,m,ncols,col,cm; 1830 const PetscInt *is,*ci,*cj,*row_idx; 1831 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1832 IS *isa; 1833 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1834 PetscInt *colorforrow,*rows,*rows_i,*idxhit,*spidx,*den2sp,*den2sp_i; 1835 PetscInt *colorforcol,*columns,*columns_i,brows; 1836 PetscBool flg; 1837 1838 PetscFunctionBegin; 1839 ierr = ISColoringGetIS(iscoloring,PETSC_USE_POINTER,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1840 1841 /* bs >1 is not being tested yet! */ 1842 Nbs = mat->cmap->N/bs; 1843 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1844 c->N = Nbs; 1845 c->m = c->M; 1846 c->rstart = 0; 1847 c->brows = 100; 1848 1849 c->ncolors = nis; 1850 ierr = PetscMalloc3(nis,&c->ncolumns,nis,&c->nrows,nis+1,&colorforrow);CHKERRQ(ierr); 1851 ierr = PetscMalloc1(csp->nz+1,&rows);CHKERRQ(ierr); 1852 ierr = PetscMalloc1(csp->nz+1,&den2sp);CHKERRQ(ierr); 1853 1854 brows = c->brows; 1855 ierr = PetscOptionsGetInt(NULL,NULL,"-matden2sp_brows",&brows,&flg);CHKERRQ(ierr); 1856 if (flg) c->brows = brows; 1857 if (brows > 0) { 1858 ierr = PetscMalloc1(nis+1,&c->lstart);CHKERRQ(ierr); 1859 } 1860 1861 colorforrow[0] = 0; 1862 rows_i = rows; 1863 den2sp_i = den2sp; 1864 1865 ierr = PetscMalloc1(nis+1,&colorforcol);CHKERRQ(ierr); 1866 ierr = PetscMalloc1(Nbs+1,&columns);CHKERRQ(ierr); 1867 1868 colorforcol[0] = 0; 1869 columns_i = columns; 1870 1871 /* get column-wise storage of mat */ 1872 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1873 1874 cm = c->m; 1875 ierr = PetscMalloc1(cm+1,&rowhit);CHKERRQ(ierr); 1876 ierr = PetscMalloc1(cm+1,&idxhit);CHKERRQ(ierr); 1877 for (i=0; i<nis; i++) { /* loop over color */ 1878 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 1879 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 1880 1881 c->ncolumns[i] = n; 1882 if (n) { 1883 ierr = PetscArraycpy(columns_i,is,n);CHKERRQ(ierr); 1884 } 1885 colorforcol[i+1] = colorforcol[i] + n; 1886 columns_i += n; 1887 1888 /* fast, crude version requires O(N*N) work */ 1889 ierr = PetscArrayzero(rowhit,cm);CHKERRQ(ierr); 1890 1891 for (j=0; j<n; j++) { /* loop over columns*/ 1892 col = is[j]; 1893 row_idx = cj + ci[col]; 1894 m = ci[col+1] - ci[col]; 1895 for (k=0; k<m; k++) { /* loop over columns marking them in rowhit */ 1896 idxhit[*row_idx] = spidx[ci[col] + k]; 1897 rowhit[*row_idx++] = col + 1; 1898 } 1899 } 1900 /* count the number of hits */ 1901 nrows = 0; 1902 for (j=0; j<cm; j++) { 1903 if (rowhit[j]) nrows++; 1904 } 1905 c->nrows[i] = nrows; 1906 colorforrow[i+1] = colorforrow[i] + nrows; 1907 1908 nrows = 0; 1909 for (j=0; j<cm; j++) { /* loop over rows */ 1910 if (rowhit[j]) { 1911 rows_i[nrows] = j; 1912 den2sp_i[nrows] = idxhit[j]; 1913 nrows++; 1914 } 1915 } 1916 den2sp_i += nrows; 1917 1918 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1919 rows_i += nrows; 1920 } 1921 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1922 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1923 ierr = ISColoringRestoreIS(iscoloring,PETSC_USE_POINTER,&isa);CHKERRQ(ierr); 1924 if (csp->nz != colorforrow[nis]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %d != colorforrow[nis] %d",csp->nz,colorforrow[nis]); 1925 1926 c->colorforrow = colorforrow; 1927 c->rows = rows; 1928 c->den2sp = den2sp; 1929 c->colorforcol = colorforcol; 1930 c->columns = columns; 1931 1932 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1933 PetscFunctionReturn(0); 1934 } 1935 1936 /* --------------------------------------------------------------- */ 1937 static PetscErrorCode MatProductNumeric_AtB_SeqAIJ_SeqAIJ(Mat C) 1938 { 1939 PetscErrorCode ierr; 1940 Mat_Product *product = C->product; 1941 Mat A=product->A,B=product->B; 1942 1943 PetscFunctionBegin; 1944 if (C->ops->mattransposemultnumeric) { 1945 /* Alg: "outerproduct" */ 1946 ierr = (*C->ops->mattransposemultnumeric)(A,B,C);CHKERRQ(ierr); 1947 } else { 1948 /* Alg: "matmatmult" -- C = At*B */ 1949 Mat_MatTransMatMult *atb = (Mat_MatTransMatMult *)product->data; 1950 Mat At; 1951 1952 if (!atb) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Missing product struct"); 1953 At = atb->At; 1954 if (atb->updateAt && At) { /* At is computed in MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ() */ 1955 ierr = MatTranspose_SeqAIJ(A,MAT_REUSE_MATRIX,&At);CHKERRQ(ierr); 1956 } 1957 ierr = MatMatMultNumeric_SeqAIJ_SeqAIJ(At ? At : A,B,C);CHKERRQ(ierr); 1958 atb->updateAt = PETSC_TRUE; 1959 } 1960 PetscFunctionReturn(0); 1961 } 1962 1963 static PetscErrorCode MatProductSymbolic_AtB_SeqAIJ_SeqAIJ(Mat C) 1964 { 1965 PetscErrorCode ierr; 1966 Mat_Product *product = C->product; 1967 Mat A=product->A,B=product->B; 1968 PetscReal fill=product->fill; 1969 1970 PetscFunctionBegin; 1971 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1972 1973 C->ops->productnumeric = MatProductNumeric_AtB_SeqAIJ_SeqAIJ; 1974 PetscFunctionReturn(0); 1975 } 1976 1977 /* --------------------------------------------------------------- */ 1978 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_AB(Mat C) 1979 { 1980 PetscErrorCode ierr; 1981 Mat_Product *product = C->product; 1982 PetscInt alg = 0; /* default algorithm */ 1983 PetscBool flg = PETSC_FALSE; 1984 #if !defined(PETSC_HAVE_HYPRE) 1985 const char *algTypes[7] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge"}; 1986 PetscInt nalg = 7; 1987 #else 1988 const char *algTypes[8] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge","hypre"}; 1989 PetscInt nalg = 8; 1990 #endif 1991 1992 PetscFunctionBegin; 1993 /* Set default algorithm */ 1994 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 1995 if (flg) { 1996 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 1997 } 1998 1999 /* Get runtime option */ 2000 if (product->api_user) { 2001 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMult","Mat");CHKERRQ(ierr); 2002 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2003 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2004 } else { 2005 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AB","Mat");CHKERRQ(ierr); 2006 ierr = PetscOptionsEList("-matproduct_ab_via","Algorithmic approach","MatProduct_AB",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2007 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2008 } 2009 if (flg) { 2010 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2011 } 2012 2013 C->ops->productsymbolic = MatProductSymbolic_AB; 2014 C->ops->matmultsymbolic = MatMatMultSymbolic_SeqAIJ_SeqAIJ; 2015 PetscFunctionReturn(0); 2016 } 2017 2018 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_AtB(Mat C) 2019 { 2020 PetscErrorCode ierr; 2021 Mat_Product *product = C->product; 2022 PetscInt alg = 0; /* default algorithm */ 2023 PetscBool flg = PETSC_FALSE; 2024 const char *algTypes[3] = {"default","at*b","outerproduct"}; 2025 PetscInt nalg = 3; 2026 2027 PetscFunctionBegin; 2028 /* Get runtime option */ 2029 if (product->api_user) { 2030 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatTransposeMatMult","Mat");CHKERRQ(ierr); 2031 ierr = PetscOptionsEList("-mattransposematmult_via","Algorithmic approach","MatTransposeMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2032 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2033 } else { 2034 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_AtB","Mat");CHKERRQ(ierr); 2035 ierr = PetscOptionsEList("-matproduct_atb_via","Algorithmic approach","MatProduct_AtB",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2036 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2037 } 2038 if (flg) { 2039 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2040 } 2041 2042 C->ops->productsymbolic = MatProductSymbolic_AtB_SeqAIJ_SeqAIJ; 2043 PetscFunctionReturn(0); 2044 } 2045 2046 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_ABt(Mat C) 2047 { 2048 PetscErrorCode ierr; 2049 Mat_Product *product = C->product; 2050 PetscInt alg = 0; /* default algorithm */ 2051 PetscBool flg = PETSC_FALSE; 2052 const char *algTypes[2] = {"default","color"}; 2053 PetscInt nalg = 2; 2054 2055 PetscFunctionBegin; 2056 /* Set default algorithm */ 2057 ierr = PetscStrcmp(C->product->alg,"default",&flg);CHKERRQ(ierr); 2058 if (!flg) { 2059 alg = 1; 2060 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2061 } 2062 2063 /* Get runtime option */ 2064 if (product->api_user) { 2065 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatTransposeMult","Mat");CHKERRQ(ierr); 2066 ierr = PetscOptionsEList("-matmattransmult_via","Algorithmic approach","MatMatTransposeMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2067 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2068 } else { 2069 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_ABt","Mat");CHKERRQ(ierr); 2070 ierr = PetscOptionsEList("-matproduct_abt_via","Algorithmic approach","MatProduct_ABt",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2071 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2072 } 2073 if (flg) { 2074 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2075 } 2076 2077 C->ops->mattransposemultsymbolic = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ; 2078 C->ops->productsymbolic = MatProductSymbolic_ABt; 2079 PetscFunctionReturn(0); 2080 } 2081 2082 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_PtAP(Mat C) 2083 { 2084 PetscErrorCode ierr; 2085 Mat_Product *product = C->product; 2086 PetscBool flg = PETSC_FALSE; 2087 PetscInt alg = 0; /* default algorithm -- alg=1 should be default!!! */ 2088 #if !defined(PETSC_HAVE_HYPRE) 2089 const char *algTypes[2] = {"scalable","rap"}; 2090 PetscInt nalg = 2; 2091 #else 2092 const char *algTypes[3] = {"scalable","rap","hypre"}; 2093 PetscInt nalg = 3; 2094 #endif 2095 2096 PetscFunctionBegin; 2097 /* Set default algorithm */ 2098 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2099 if (flg) { 2100 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2101 } 2102 2103 /* Get runtime option */ 2104 if (product->api_user) { 2105 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatPtAP","Mat");CHKERRQ(ierr); 2106 ierr = PetscOptionsEList("-matptap_via","Algorithmic approach","MatPtAP",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2107 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2108 } else { 2109 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_PtAP","Mat");CHKERRQ(ierr); 2110 ierr = PetscOptionsEList("-matproduct_ptap_via","Algorithmic approach","MatProduct_PtAP",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2111 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2112 } 2113 if (flg) { 2114 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2115 } 2116 2117 C->ops->productsymbolic = MatProductSymbolic_PtAP_SeqAIJ_SeqAIJ; 2118 PetscFunctionReturn(0); 2119 } 2120 2121 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_RARt(Mat C) 2122 { 2123 PetscErrorCode ierr; 2124 Mat_Product *product = C->product; 2125 PetscBool flg = PETSC_FALSE; 2126 PetscInt alg = 0; /* default algorithm */ 2127 const char *algTypes[3] = {"r*a*rt","r*art","coloring_rart"}; 2128 PetscInt nalg = 3; 2129 2130 PetscFunctionBegin; 2131 /* Set default algorithm */ 2132 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2133 if (flg) { 2134 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2135 } 2136 2137 /* Get runtime option */ 2138 if (product->api_user) { 2139 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatRARt","Mat");CHKERRQ(ierr); 2140 ierr = PetscOptionsEList("-matrart_via","Algorithmic approach","MatRARt",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2141 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2142 } else { 2143 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_RARt","Mat");CHKERRQ(ierr); 2144 ierr = PetscOptionsEList("-matproduct_rart_via","Algorithmic approach","MatProduct_RARt",algTypes,nalg,algTypes[0],&alg,&flg);CHKERRQ(ierr); 2145 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2146 } 2147 if (flg) { 2148 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2149 } 2150 2151 C->ops->productsymbolic = MatProductSymbolic_RARt_SeqAIJ_SeqAIJ; 2152 PetscFunctionReturn(0); 2153 } 2154 2155 /* ABC = A*B*C = A*(B*C); ABC's algorithm must be chosen from AB's algorithm */ 2156 static PetscErrorCode MatProductSetFromOptions_SeqAIJ_ABC(Mat C) 2157 { 2158 PetscErrorCode ierr; 2159 Mat_Product *product = C->product; 2160 PetscInt alg = 0; /* default algorithm */ 2161 PetscBool flg = PETSC_FALSE; 2162 const char *algTypes[7] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","rowmerge"}; 2163 PetscInt nalg = 7; 2164 2165 PetscFunctionBegin; 2166 /* Set default algorithm */ 2167 ierr = PetscStrcmp(product->alg,"default",&flg);CHKERRQ(ierr); 2168 if (flg) { 2169 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2170 } 2171 2172 /* Get runtime option */ 2173 if (product->api_user) { 2174 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatMatMatMult","Mat");CHKERRQ(ierr); 2175 ierr = PetscOptionsEList("-matmatmatmult_via","Algorithmic approach","MatMatMatMult",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2176 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2177 } else { 2178 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)C),((PetscObject)C)->prefix,"MatProduct_ABC","Mat");CHKERRQ(ierr); 2179 ierr = PetscOptionsEList("-matproduct_abc_via","Algorithmic approach","MatProduct_ABC",algTypes,nalg,algTypes[alg],&alg,&flg);CHKERRQ(ierr); 2180 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2181 } 2182 if (flg) { 2183 ierr = MatProductSetAlgorithm(C,(MatProductAlgorithm)algTypes[alg]);CHKERRQ(ierr); 2184 } 2185 2186 C->ops->matmatmultsymbolic = MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ; 2187 C->ops->productsymbolic = MatProductSymbolic_ABC; 2188 PetscFunctionReturn(0); 2189 } 2190 2191 PetscErrorCode MatProductSetFromOptions_SeqAIJ(Mat C) 2192 { 2193 PetscErrorCode ierr; 2194 Mat_Product *product = C->product; 2195 2196 PetscFunctionBegin; 2197 switch (product->type) { 2198 case MATPRODUCT_AB: 2199 ierr = MatProductSetFromOptions_SeqAIJ_AB(C);CHKERRQ(ierr); 2200 break; 2201 case MATPRODUCT_AtB: 2202 ierr = MatProductSetFromOptions_SeqAIJ_AtB(C);CHKERRQ(ierr); 2203 break; 2204 case MATPRODUCT_ABt: 2205 ierr = MatProductSetFromOptions_SeqAIJ_ABt(C);CHKERRQ(ierr); 2206 break; 2207 case MATPRODUCT_PtAP: 2208 ierr = MatProductSetFromOptions_SeqAIJ_PtAP(C);CHKERRQ(ierr); 2209 break; 2210 case MATPRODUCT_RARt: 2211 ierr = MatProductSetFromOptions_SeqAIJ_RARt(C);CHKERRQ(ierr); 2212 break; 2213 case MATPRODUCT_ABC: 2214 ierr = MatProductSetFromOptions_SeqAIJ_ABC(C);CHKERRQ(ierr); 2215 break; 2216 default: 2217 break; 2218 } 2219 PetscFunctionReturn(0); 2220 } 2221