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 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat,Mat,PetscReal,Mat*); 14 15 #if defined(PETSC_HAVE_HYPRE) 16 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat*); 17 #endif 18 19 PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 20 { 21 PetscErrorCode ierr; 22 #if !defined(PETSC_HAVE_HYPRE) 23 const char *algTypes[6] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed"}; 24 PetscInt nalg = 6; 25 #else 26 const char *algTypes[7] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed","hypre"}; 27 PetscInt nalg = 7; 28 #endif 29 PetscInt alg = 0; /* set default algorithm */ 30 31 PetscFunctionBegin; 32 if (scall == MAT_INITIAL_MATRIX) { 33 ierr = PetscObjectOptionsBegin((PetscObject)A);CHKERRQ(ierr); 34 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,nalg,algTypes[0],&alg,NULL);CHKERRQ(ierr); 35 ierr = PetscOptionsEnd();CHKERRQ(ierr); 36 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 37 switch (alg) { 38 case 1: 39 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 40 break; 41 case 2: 42 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 43 break; 44 case 3: 45 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 46 break; 47 case 4: 48 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 49 break; 50 case 5: 51 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(A,B,fill,C);CHKERRQ(ierr); 52 break; 53 #if defined(PETSC_HAVE_HYPRE) 54 case 6: 55 ierr = MatMatMultSymbolic_AIJ_AIJ_wHYPRE(A,B,fill,C);CHKERRQ(ierr); 56 break; 57 #endif 58 default: 59 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 60 break; 61 } 62 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 63 } 64 65 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 66 ierr = (*(*C)->ops->matmultnumeric)(A,B,*C);CHKERRQ(ierr); 67 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 68 PetscFunctionReturn(0); 69 } 70 71 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat A,Mat B,PetscReal fill,Mat *C) 72 { 73 PetscErrorCode ierr; 74 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 75 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 76 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 77 PetscReal afill; 78 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 79 PetscTable ta; 80 PetscBT lnkbt; 81 PetscFreeSpaceList free_space=NULL,current_space=NULL; 82 83 PetscFunctionBegin; 84 /* Get ci and cj */ 85 /*---------------*/ 86 /* Allocate ci array, arrays for fill computation and */ 87 /* free space for accumulating nonzero column info */ 88 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 89 ci[0] = 0; 90 91 /* create and initialize a linked list */ 92 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 93 MatRowMergeMax_SeqAIJ(b,bm,ta); 94 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 95 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 96 97 ierr = PetscLLCondensedCreate(Crmax,bn,&lnk,&lnkbt);CHKERRQ(ierr); 98 99 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 100 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 101 102 current_space = free_space; 103 104 /* Determine ci and cj */ 105 for (i=0; i<am; i++) { 106 anzi = ai[i+1] - ai[i]; 107 aj = a->j + ai[i]; 108 for (j=0; j<anzi; j++) { 109 brow = aj[j]; 110 bnzj = bi[brow+1] - bi[brow]; 111 bj = b->j + bi[brow]; 112 /* add non-zero cols of B into the sorted linked list lnk */ 113 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 114 } 115 cnzi = lnk[0]; 116 117 /* If free space is not available, make more free space */ 118 /* Double the amount of total space in the list */ 119 if (current_space->local_remaining<cnzi) { 120 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 121 ndouble++; 122 } 123 124 /* Copy data into free space, then initialize lnk */ 125 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 126 127 current_space->array += cnzi; 128 current_space->local_used += cnzi; 129 current_space->local_remaining -= cnzi; 130 131 ci[i+1] = ci[i] + cnzi; 132 } 133 134 /* Column indices are in the list of free space */ 135 /* Allocate space for cj, initialize cj, and */ 136 /* destroy list of free space and other temporary array(s) */ 137 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 138 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 139 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 140 141 /* put together the new symbolic matrix */ 142 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 143 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 144 145 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 146 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 147 c = (Mat_SeqAIJ*)((*C)->data); 148 c->free_a = PETSC_FALSE; 149 c->free_ij = PETSC_TRUE; 150 c->nonew = 0; 151 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; /* fast, needs non-scalable O(bn) array 'abdense' */ 152 153 /* set MatInfo */ 154 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 155 if (afill < 1.0) afill = 1.0; 156 c->maxnz = ci[am]; 157 c->nz = ci[am]; 158 (*C)->info.mallocs = ndouble; 159 (*C)->info.fill_ratio_given = fill; 160 (*C)->info.fill_ratio_needed = afill; 161 162 #if defined(PETSC_USE_INFO) 163 if (ci[am]) { 164 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 165 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 166 } else { 167 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 168 } 169 #endif 170 PetscFunctionReturn(0); 171 } 172 173 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 174 { 175 PetscErrorCode ierr; 176 PetscLogDouble flops=0.0; 177 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 178 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 179 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 180 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 181 PetscInt am =A->rmap->n,cm=C->rmap->n; 182 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 183 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca,valtmp; 184 PetscScalar *ab_dense; 185 186 PetscFunctionBegin; 187 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 188 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 189 c->a = ca; 190 c->free_a = PETSC_TRUE; 191 } else { 192 ca = c->a; 193 } 194 if (!c->matmult_abdense) { 195 ierr = PetscCalloc1(B->cmap->N,&ab_dense);CHKERRQ(ierr); 196 c->matmult_abdense = ab_dense; 197 } else { 198 ab_dense = c->matmult_abdense; 199 } 200 201 /* clean old values in C */ 202 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 203 /* Traverse A row-wise. */ 204 /* Build the ith row in C by summing over nonzero columns in A, */ 205 /* the rows of B corresponding to nonzeros of A. */ 206 for (i=0; i<am; i++) { 207 anzi = ai[i+1] - ai[i]; 208 for (j=0; j<anzi; j++) { 209 brow = aj[j]; 210 bnzi = bi[brow+1] - bi[brow]; 211 bjj = bj + bi[brow]; 212 baj = ba + bi[brow]; 213 /* perform dense axpy */ 214 valtmp = aa[j]; 215 for (k=0; k<bnzi; k++) { 216 ab_dense[bjj[k]] += valtmp*baj[k]; 217 } 218 flops += 2*bnzi; 219 } 220 aj += anzi; aa += anzi; 221 222 cnzi = ci[i+1] - ci[i]; 223 for (k=0; k<cnzi; k++) { 224 ca[k] += ab_dense[cj[k]]; 225 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 226 } 227 flops += cnzi; 228 cj += cnzi; ca += cnzi; 229 } 230 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 231 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 232 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 233 PetscFunctionReturn(0); 234 } 235 236 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 237 { 238 PetscErrorCode ierr; 239 PetscLogDouble flops=0.0; 240 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 241 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 242 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 243 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 244 PetscInt am = A->rmap->N,cm=C->rmap->N; 245 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 246 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca=c->a,valtmp; 247 PetscInt nextb; 248 249 PetscFunctionBegin; 250 if (!ca) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 251 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 252 c->a = ca; 253 c->free_a = PETSC_TRUE; 254 } 255 256 /* clean old values in C */ 257 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 258 /* Traverse A row-wise. */ 259 /* Build the ith row in C by summing over nonzero columns in A, */ 260 /* the rows of B corresponding to nonzeros of A. */ 261 for (i=0; i<am; i++) { 262 anzi = ai[i+1] - ai[i]; 263 cnzi = ci[i+1] - ci[i]; 264 for (j=0; j<anzi; j++) { 265 brow = aj[j]; 266 bnzi = bi[brow+1] - bi[brow]; 267 bjj = bj + bi[brow]; 268 baj = ba + bi[brow]; 269 /* perform sparse axpy */ 270 valtmp = aa[j]; 271 nextb = 0; 272 for (k=0; nextb<bnzi; k++) { 273 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 274 ca[k] += valtmp*baj[nextb++]; 275 } 276 } 277 flops += 2*bnzi; 278 } 279 aj += anzi; aa += anzi; 280 cj += cnzi; ca += cnzi; 281 } 282 283 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 284 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 285 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 286 PetscFunctionReturn(0); 287 } 288 289 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat *C) 290 { 291 PetscErrorCode ierr; 292 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 293 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 294 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 295 MatScalar *ca; 296 PetscReal afill; 297 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 298 PetscTable ta; 299 PetscFreeSpaceList free_space=NULL,current_space=NULL; 300 301 PetscFunctionBegin; 302 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 303 /*-----------------------------------------------------------------------------------------*/ 304 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 305 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 306 ci[0] = 0; 307 308 /* create and initialize a linked list */ 309 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 310 MatRowMergeMax_SeqAIJ(b,bm,ta); 311 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 312 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 313 314 ierr = PetscLLCondensedCreate_fast(Crmax,&lnk);CHKERRQ(ierr); 315 316 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 317 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 318 current_space = free_space; 319 320 /* Determine ci and cj */ 321 for (i=0; i<am; i++) { 322 anzi = ai[i+1] - ai[i]; 323 aj = a->j + ai[i]; 324 for (j=0; j<anzi; j++) { 325 brow = aj[j]; 326 bnzj = bi[brow+1] - bi[brow]; 327 bj = b->j + bi[brow]; 328 /* add non-zero cols of B into the sorted linked list lnk */ 329 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 330 } 331 cnzi = lnk[1]; 332 333 /* If free space is not available, make more free space */ 334 /* Double the amount of total space in the list */ 335 if (current_space->local_remaining<cnzi) { 336 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 337 ndouble++; 338 } 339 340 /* Copy data into free space, then initialize lnk */ 341 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 342 343 current_space->array += cnzi; 344 current_space->local_used += cnzi; 345 current_space->local_remaining -= cnzi; 346 347 ci[i+1] = ci[i] + cnzi; 348 } 349 350 /* Column indices are in the list of free space */ 351 /* Allocate space for cj, initialize cj, and */ 352 /* destroy list of free space and other temporary array(s) */ 353 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 354 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 355 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 356 357 /* Allocate space for ca */ 358 ierr = PetscMalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 359 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 360 361 /* put together the new symbolic matrix */ 362 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 363 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 364 365 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 366 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 367 c = (Mat_SeqAIJ*)((*C)->data); 368 c->free_a = PETSC_TRUE; 369 c->free_ij = PETSC_TRUE; 370 c->nonew = 0; 371 372 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 373 374 /* set MatInfo */ 375 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 376 if (afill < 1.0) afill = 1.0; 377 c->maxnz = ci[am]; 378 c->nz = ci[am]; 379 (*C)->info.mallocs = ndouble; 380 (*C)->info.fill_ratio_given = fill; 381 (*C)->info.fill_ratio_needed = afill; 382 383 #if defined(PETSC_USE_INFO) 384 if (ci[am]) { 385 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 386 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 387 } else { 388 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 389 } 390 #endif 391 PetscFunctionReturn(0); 392 } 393 394 395 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat *C) 396 { 397 PetscErrorCode ierr; 398 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 399 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 400 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 401 MatScalar *ca; 402 PetscReal afill; 403 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,*lnk,ndouble=0,Crmax; 404 PetscTable ta; 405 PetscFreeSpaceList free_space=NULL,current_space=NULL; 406 407 PetscFunctionBegin; 408 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 409 /*---------------------------------------------------------------------------------------------*/ 410 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 411 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 412 ci[0] = 0; 413 414 /* create and initialize a linked list */ 415 ierr = PetscTableCreate(bn,bn,&ta);CHKERRQ(ierr); 416 MatRowMergeMax_SeqAIJ(b,bm,ta); 417 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 418 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 419 ierr = PetscLLCondensedCreate_Scalable(Crmax,&lnk);CHKERRQ(ierr); 420 421 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 422 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 423 current_space = free_space; 424 425 /* Determine ci and cj */ 426 for (i=0; i<am; i++) { 427 anzi = ai[i+1] - ai[i]; 428 aj = a->j + ai[i]; 429 for (j=0; j<anzi; j++) { 430 brow = aj[j]; 431 bnzj = bi[brow+1] - bi[brow]; 432 bj = b->j + bi[brow]; 433 /* add non-zero cols of B into the sorted linked list lnk */ 434 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 435 } 436 cnzi = lnk[0]; 437 438 /* If free space is not available, make more free space */ 439 /* Double the amount of total space in the list */ 440 if (current_space->local_remaining<cnzi) { 441 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(cnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 442 ndouble++; 443 } 444 445 /* Copy data into free space, then initialize lnk */ 446 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 447 448 current_space->array += cnzi; 449 current_space->local_used += cnzi; 450 current_space->local_remaining -= cnzi; 451 452 ci[i+1] = ci[i] + cnzi; 453 } 454 455 /* Column indices are in the list of free space */ 456 /* Allocate space for cj, initialize cj, and */ 457 /* destroy list of free space and other temporary array(s) */ 458 ierr = PetscMalloc1(ci[am]+1,&cj);CHKERRQ(ierr); 459 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 460 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 461 462 /* Allocate space for ca */ 463 /*-----------------------*/ 464 ierr = PetscMalloc1(ci[am]+1,&ca);CHKERRQ(ierr); 465 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 466 467 /* put together the new symbolic matrix */ 468 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 469 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 470 471 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 472 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 473 c = (Mat_SeqAIJ*)((*C)->data); 474 c->free_a = PETSC_TRUE; 475 c->free_ij = PETSC_TRUE; 476 c->nonew = 0; 477 478 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 479 480 /* set MatInfo */ 481 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 482 if (afill < 1.0) afill = 1.0; 483 c->maxnz = ci[am]; 484 c->nz = ci[am]; 485 (*C)->info.mallocs = ndouble; 486 (*C)->info.fill_ratio_given = fill; 487 (*C)->info.fill_ratio_needed = afill; 488 489 #if defined(PETSC_USE_INFO) 490 if (ci[am]) { 491 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 492 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 493 } else { 494 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 495 } 496 #endif 497 PetscFunctionReturn(0); 498 } 499 500 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat *C) 501 { 502 PetscErrorCode ierr; 503 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 504 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 505 PetscInt *ci,*cj,*bb; 506 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 507 PetscReal afill; 508 PetscInt i,j,col,ndouble = 0; 509 PetscFreeSpaceList free_space=NULL,current_space=NULL; 510 PetscHeap h; 511 512 PetscFunctionBegin; 513 /* Get ci and cj - by merging sorted rows using a heap */ 514 /*---------------------------------------------------------------------------------------------*/ 515 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 516 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 517 ci[0] = 0; 518 519 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 520 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 521 current_space = free_space; 522 523 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 524 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 525 526 /* Determine ci and cj */ 527 for (i=0; i<am; i++) { 528 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 */ 529 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 530 ci[i+1] = ci[i]; 531 /* Populate the min heap */ 532 for (j=0; j<anzi; j++) { 533 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 534 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 535 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 536 } 537 } 538 /* Pick off the min element, adding it to free space */ 539 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 540 while (j >= 0) { 541 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 542 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 543 ndouble++; 544 } 545 *(current_space->array++) = col; 546 current_space->local_used++; 547 current_space->local_remaining--; 548 ci[i+1]++; 549 550 /* stash if anything else remains in this row of B */ 551 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 552 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 553 PetscInt j2,col2; 554 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 555 if (col2 != col) break; 556 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 557 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 558 } 559 /* Put any stashed elements back into the min heap */ 560 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 561 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 562 } 563 } 564 ierr = PetscFree(bb);CHKERRQ(ierr); 565 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 566 567 /* Column indices are in the list of free space */ 568 /* Allocate space for cj, initialize cj, and */ 569 /* destroy list of free space and other temporary array(s) */ 570 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 571 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 572 573 /* put together the new symbolic matrix */ 574 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 575 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 576 577 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 578 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 579 c = (Mat_SeqAIJ*)((*C)->data); 580 c->free_a = PETSC_TRUE; 581 c->free_ij = PETSC_TRUE; 582 c->nonew = 0; 583 584 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 585 586 /* set MatInfo */ 587 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 588 if (afill < 1.0) afill = 1.0; 589 c->maxnz = ci[am]; 590 c->nz = ci[am]; 591 (*C)->info.mallocs = ndouble; 592 (*C)->info.fill_ratio_given = fill; 593 (*C)->info.fill_ratio_needed = afill; 594 595 #if defined(PETSC_USE_INFO) 596 if (ci[am]) { 597 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 598 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 599 } else { 600 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 601 } 602 #endif 603 PetscFunctionReturn(0); 604 } 605 606 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat *C) 607 { 608 PetscErrorCode ierr; 609 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 610 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 611 PetscInt *ci,*cj,*bb; 612 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 613 PetscReal afill; 614 PetscInt i,j,col,ndouble = 0; 615 PetscFreeSpaceList free_space=NULL,current_space=NULL; 616 PetscHeap h; 617 PetscBT bt; 618 619 PetscFunctionBegin; 620 /* Get ci and cj - using a heap for the sorted rows, but use BT so that each index is only added once */ 621 /*---------------------------------------------------------------------------------------------*/ 622 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 623 ierr = PetscMalloc1(am+2,&ci);CHKERRQ(ierr); 624 ci[0] = 0; 625 626 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 627 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ai[am],bi[bm])),&free_space);CHKERRQ(ierr); 628 629 current_space = free_space; 630 631 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 632 ierr = PetscMalloc1(a->rmax,&bb);CHKERRQ(ierr); 633 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 634 635 /* Determine ci and cj */ 636 for (i=0; i<am; i++) { 637 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 */ 638 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 639 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 640 ci[i+1] = ci[i]; 641 /* Populate the min heap */ 642 for (j=0; j<anzi; j++) { 643 PetscInt brow = acol[j]; 644 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 645 PetscInt bcol = bj[bb[j]]; 646 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 647 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 648 bb[j]++; 649 break; 650 } 651 } 652 } 653 /* Pick off the min element, adding it to free space */ 654 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 655 while (j >= 0) { 656 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 657 fptr = NULL; /* need PetscBTMemzero */ 658 ierr = PetscFreeSpaceGet(PetscMin(PetscIntMultTruncate(2,current_space->total_array_size),16 << 20),¤t_space);CHKERRQ(ierr); 659 ndouble++; 660 } 661 *(current_space->array++) = col; 662 current_space->local_used++; 663 current_space->local_remaining--; 664 ci[i+1]++; 665 666 /* stash if anything else remains in this row of B */ 667 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 668 PetscInt bcol = bj[bb[j]]; 669 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 670 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 671 bb[j]++; 672 break; 673 } 674 } 675 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 676 } 677 if (fptr) { /* Clear the bits for this row */ 678 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 679 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 680 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 681 } 682 } 683 ierr = PetscFree(bb);CHKERRQ(ierr); 684 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 685 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 686 687 /* Column indices are in the list of free space */ 688 /* Allocate space for cj, initialize cj, and */ 689 /* destroy list of free space and other temporary array(s) */ 690 ierr = PetscMalloc1(ci[am],&cj);CHKERRQ(ierr); 691 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 692 693 /* put together the new symbolic matrix */ 694 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 695 ierr = MatSetBlockSizesFromMats(*C,A,B);CHKERRQ(ierr); 696 697 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 698 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 699 c = (Mat_SeqAIJ*)((*C)->data); 700 c->free_a = PETSC_TRUE; 701 c->free_ij = PETSC_TRUE; 702 c->nonew = 0; 703 704 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 705 706 /* set MatInfo */ 707 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 708 if (afill < 1.0) afill = 1.0; 709 c->maxnz = ci[am]; 710 c->nz = ci[am]; 711 (*C)->info.mallocs = ndouble; 712 (*C)->info.fill_ratio_given = fill; 713 (*C)->info.fill_ratio_needed = afill; 714 715 #if defined(PETSC_USE_INFO) 716 if (ci[am]) { 717 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %g needed %g.\n",ndouble,(double)fill,(double)afill);CHKERRQ(ierr); 718 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%g,&C) for best performance.;\n",(double)afill);CHKERRQ(ierr); 719 } else { 720 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 721 } 722 #endif 723 PetscFunctionReturn(0); 724 } 725 726 /* concatenate unique entries and then sort */ 727 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 728 { 729 PetscErrorCode ierr; 730 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 731 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 732 PetscInt *ci,*cj; 733 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 734 PetscReal afill; 735 PetscInt i,j,ndouble = 0; 736 PetscSegBuffer seg,segrow; 737 char *seen; 738 739 PetscFunctionBegin; 740 ierr = PetscMalloc1(am+1,&ci);CHKERRQ(ierr); 741 ci[0] = 0; 742 743 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 744 ierr = PetscSegBufferCreate(sizeof(PetscInt),(PetscInt)(fill*(ai[am]+bi[bm])),&seg);CHKERRQ(ierr); 745 ierr = PetscSegBufferCreate(sizeof(PetscInt),100,&segrow);CHKERRQ(ierr); 746 ierr = PetscMalloc1(bn,&seen);CHKERRQ(ierr); 747 ierr = PetscMemzero(seen,bn*sizeof(char));CHKERRQ(ierr); 748 749 /* Determine ci and cj */ 750 for (i=0; i<am; i++) { 751 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 */ 752 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 753 PetscInt packlen = 0,*PETSC_RESTRICT crow; 754 /* Pack segrow */ 755 for (j=0; j<anzi; j++) { 756 PetscInt brow = acol[j],bjstart = bi[brow],bjend = bi[brow+1],k; 757 for (k=bjstart; k<bjend; k++) { 758 PetscInt bcol = bj[k]; 759 if (!seen[bcol]) { /* new entry */ 760 PetscInt *PETSC_RESTRICT slot; 761 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 762 *slot = bcol; 763 seen[bcol] = 1; 764 packlen++; 765 } 766 } 767 } 768 ierr = PetscSegBufferGetInts(seg,packlen,&crow);CHKERRQ(ierr); 769 ierr = PetscSegBufferExtractTo(segrow,crow);CHKERRQ(ierr); 770 ierr = PetscSortInt(packlen,crow);CHKERRQ(ierr); 771 ci[i+1] = ci[i] + packlen; 772 for (j=0; j<packlen; j++) seen[crow[j]] = 0; 773 } 774 ierr = PetscSegBufferDestroy(&segrow);CHKERRQ(ierr); 775 ierr = PetscFree(seen);CHKERRQ(ierr); 776 777 /* Column indices are in the segmented buffer */ 778 ierr = PetscSegBufferExtractAlloc(seg,&cj);CHKERRQ(ierr); 779 ierr = PetscSegBufferDestroy(&seg);CHKERRQ(ierr); 780 781 /* put together the new symbolic matrix */ 782 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,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; 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 /* This routine is not used. Should be removed! */ 815 PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 816 { 817 PetscErrorCode ierr; 818 819 PetscFunctionBegin; 820 if (scall == MAT_INITIAL_MATRIX) { 821 ierr = PetscLogEventBegin(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 822 ierr = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 823 ierr = PetscLogEventEnd(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 824 } 825 ierr = PetscLogEventBegin(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 826 ierr = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 827 ierr = PetscLogEventEnd(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 828 PetscFunctionReturn(0); 829 } 830 831 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(Mat A) 832 { 833 PetscErrorCode ierr; 834 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 835 Mat_MatMatTransMult *abt=a->abt; 836 837 PetscFunctionBegin; 838 ierr = (abt->destroy)(A);CHKERRQ(ierr); 839 ierr = MatTransposeColoringDestroy(&abt->matcoloring);CHKERRQ(ierr); 840 ierr = MatDestroy(&abt->Bt_den);CHKERRQ(ierr); 841 ierr = MatDestroy(&abt->ABt_den);CHKERRQ(ierr); 842 ierr = PetscFree(abt);CHKERRQ(ierr); 843 PetscFunctionReturn(0); 844 } 845 846 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 847 { 848 PetscErrorCode ierr; 849 Mat Bt; 850 PetscInt *bti,*btj; 851 Mat_MatMatTransMult *abt; 852 Mat_SeqAIJ *c; 853 854 PetscFunctionBegin; 855 /* create symbolic Bt */ 856 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 857 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 858 ierr = MatSetBlockSizes(Bt,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 859 860 /* get symbolic C=A*Bt */ 861 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 862 863 /* create a supporting struct for reuse intermidiate dense matrices with matcoloring */ 864 ierr = PetscNew(&abt);CHKERRQ(ierr); 865 c = (Mat_SeqAIJ*)(*C)->data; 866 c->abt = abt; 867 868 abt->usecoloring = PETSC_FALSE; 869 abt->destroy = (*C)->ops->destroy; 870 (*C)->ops->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 871 872 ierr = PetscOptionsGetBool(((PetscObject)A)->options,NULL,"-matmattransmult_color",&abt->usecoloring,NULL);CHKERRQ(ierr); 873 if (abt->usecoloring) { 874 /* Create MatTransposeColoring from symbolic C=A*B^T */ 875 MatTransposeColoring matcoloring; 876 MatColoring coloring; 877 ISColoring iscoloring; 878 Mat Bt_dense,C_dense; 879 Mat_SeqAIJ *c=(Mat_SeqAIJ*)(*C)->data; 880 /* inode causes memory problem, don't know why */ 881 if (c->inode.use) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MAT_USE_INODES is not supported. Use '-mat_no_inode'"); 882 883 ierr = MatColoringCreate(*C,&coloring);CHKERRQ(ierr); 884 ierr = MatColoringSetDistance(coloring,2);CHKERRQ(ierr); 885 ierr = MatColoringSetType(coloring,MATCOLORINGSL);CHKERRQ(ierr); 886 ierr = MatColoringSetFromOptions(coloring);CHKERRQ(ierr); 887 ierr = MatColoringApply(coloring,&iscoloring);CHKERRQ(ierr); 888 ierr = MatColoringDestroy(&coloring);CHKERRQ(ierr); 889 ierr = MatTransposeColoringCreate(*C,iscoloring,&matcoloring);CHKERRQ(ierr); 890 891 abt->matcoloring = matcoloring; 892 893 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 894 895 /* Create Bt_dense and C_dense = A*Bt_dense */ 896 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 897 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 898 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 899 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 900 901 Bt_dense->assembled = PETSC_TRUE; 902 abt->Bt_den = Bt_dense; 903 904 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 905 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 906 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 907 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 908 909 Bt_dense->assembled = PETSC_TRUE; 910 abt->ABt_den = C_dense; 911 912 #if defined(PETSC_USE_INFO) 913 { 914 Mat_SeqAIJ *c = (Mat_SeqAIJ*)(*C)->data; 915 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); 916 } 917 #endif 918 } 919 /* clean up */ 920 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 921 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 922 PetscFunctionReturn(0); 923 } 924 925 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 926 { 927 PetscErrorCode ierr; 928 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 929 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 930 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 931 PetscLogDouble flops=0.0; 932 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 933 Mat_MatMatTransMult *abt = c->abt; 934 935 PetscFunctionBegin; 936 /* clear old values in C */ 937 if (!c->a) { 938 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 939 c->a = ca; 940 c->free_a = PETSC_TRUE; 941 } else { 942 ca = c->a; 943 } 944 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 945 946 if (abt->usecoloring) { 947 MatTransposeColoring matcoloring = abt->matcoloring; 948 Mat Bt_dense,C_dense = abt->ABt_den; 949 950 /* Get Bt_dense by Apply MatTransposeColoring to B */ 951 Bt_dense = abt->Bt_den; 952 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 953 954 /* C_dense = A*Bt_dense */ 955 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 956 957 /* Recover C from C_dense */ 958 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 959 PetscFunctionReturn(0); 960 } 961 962 for (i=0; i<cm; i++) { 963 anzi = ai[i+1] - ai[i]; 964 acol = aj + ai[i]; 965 aval = aa + ai[i]; 966 cnzi = ci[i+1] - ci[i]; 967 ccol = cj + ci[i]; 968 cval = ca + ci[i]; 969 for (j=0; j<cnzi; j++) { 970 brow = ccol[j]; 971 bnzj = bi[brow+1] - bi[brow]; 972 bcol = bj + bi[brow]; 973 bval = ba + bi[brow]; 974 975 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 976 nexta = 0; nextb = 0; 977 while (nexta<anzi && nextb<bnzj) { 978 while (nexta < anzi && acol[nexta] < bcol[nextb]) nexta++; 979 if (nexta == anzi) break; 980 while (nextb < bnzj && acol[nexta] > bcol[nextb]) nextb++; 981 if (nextb == bnzj) break; 982 if (acol[nexta] == bcol[nextb]) { 983 cval[j] += aval[nexta]*bval[nextb]; 984 nexta++; nextb++; 985 flops += 2; 986 } 987 } 988 } 989 } 990 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 991 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 992 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 993 PetscFunctionReturn(0); 994 } 995 996 PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 997 { 998 PetscErrorCode ierr; 999 1000 PetscFunctionBegin; 1001 if (scall == MAT_INITIAL_MATRIX) { 1002 ierr = PetscLogEventBegin(MAT_TransposeMatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1003 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1004 ierr = PetscLogEventEnd(MAT_TransposeMatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1005 } 1006 ierr = PetscLogEventBegin(MAT_TransposeMatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1007 ierr = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 1008 ierr = PetscLogEventEnd(MAT_TransposeMatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1009 PetscFunctionReturn(0); 1010 } 1011 1012 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 1013 { 1014 PetscErrorCode ierr; 1015 Mat At; 1016 PetscInt *ati,*atj; 1017 1018 PetscFunctionBegin; 1019 /* create symbolic At */ 1020 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1021 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1022 ierr = MatSetBlockSizes(At,PetscAbs(A->cmap->bs),PetscAbs(B->cmap->bs));CHKERRQ(ierr); 1023 1024 /* get symbolic C=At*B */ 1025 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(At,B,fill,C);CHKERRQ(ierr); 1026 1027 /* clean up */ 1028 ierr = MatDestroy(&At);CHKERRQ(ierr); 1029 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1030 PetscFunctionReturn(0); 1031 } 1032 1033 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1034 { 1035 PetscErrorCode ierr; 1036 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1037 PetscInt am =A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1038 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1039 PetscLogDouble flops=0.0; 1040 MatScalar *aa =a->a,*ba,*ca,*caj; 1041 1042 PetscFunctionBegin; 1043 if (!c->a) { 1044 ierr = PetscMalloc1(ci[cm]+1,&ca);CHKERRQ(ierr); 1045 1046 c->a = ca; 1047 c->free_a = PETSC_TRUE; 1048 } else { 1049 ca = c->a; 1050 } 1051 /* clear old values in C */ 1052 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 1053 1054 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1055 for (i=0; i<am; i++) { 1056 bj = b->j + bi[i]; 1057 ba = b->a + bi[i]; 1058 bnzi = bi[i+1] - bi[i]; 1059 anzi = ai[i+1] - ai[i]; 1060 for (j=0; j<anzi; j++) { 1061 nextb = 0; 1062 crow = *aj++; 1063 cjj = cj + ci[crow]; 1064 caj = ca + ci[crow]; 1065 /* perform sparse axpy operation. Note cjj includes bj. */ 1066 for (k=0; nextb<bnzi; k++) { 1067 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1068 caj[k] += (*aa)*(*(ba+nextb)); 1069 nextb++; 1070 } 1071 } 1072 flops += 2*bnzi; 1073 aa++; 1074 } 1075 } 1076 1077 /* Assemble the final matrix and clean up */ 1078 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1079 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1080 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1081 PetscFunctionReturn(0); 1082 } 1083 1084 PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1085 { 1086 PetscErrorCode ierr; 1087 1088 PetscFunctionBegin; 1089 if (scall == MAT_INITIAL_MATRIX) { 1090 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1091 ierr = MatMatMultSymbolic_SeqAIJ_SeqDense(A,B,fill,C);CHKERRQ(ierr); 1092 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1093 } 1094 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1095 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,B,*C);CHKERRQ(ierr); 1096 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1097 PetscFunctionReturn(0); 1098 } 1099 1100 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C) 1101 { 1102 PetscErrorCode ierr; 1103 1104 PetscFunctionBegin; 1105 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1106 1107 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1108 PetscFunctionReturn(0); 1109 } 1110 1111 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1112 { 1113 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1114 Mat_SeqDense *bd = (Mat_SeqDense*)B->data; 1115 PetscErrorCode ierr; 1116 PetscScalar *c,*b,r1,r2,r3,r4,*c1,*c2,*c3,*c4,aatmp; 1117 const PetscScalar *aa,*b1,*b2,*b3,*b4; 1118 const PetscInt *aj; 1119 PetscInt cm=C->rmap->n,cn=B->cmap->n,bm=bd->lda,am=A->rmap->n; 1120 PetscInt am4=4*am,bm4=4*bm,col,i,j,n,ajtmp; 1121 1122 PetscFunctionBegin; 1123 if (!cm || !cn) PetscFunctionReturn(0); 1124 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); 1125 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); 1126 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); 1127 b = bd->v; 1128 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1129 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1130 c1 = c; c2 = c1 + am; c3 = c2 + am; c4 = c3 + am; 1131 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1132 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1133 r1 = r2 = r3 = r4 = 0.0; 1134 n = a->i[i+1] - a->i[i]; 1135 aj = a->j + a->i[i]; 1136 aa = a->a + a->i[i]; 1137 for (j=0; j<n; j++) { 1138 aatmp = aa[j]; ajtmp = aj[j]; 1139 r1 += aatmp*b1[ajtmp]; 1140 r2 += aatmp*b2[ajtmp]; 1141 r3 += aatmp*b3[ajtmp]; 1142 r4 += aatmp*b4[ajtmp]; 1143 } 1144 c1[i] = r1; 1145 c2[i] = r2; 1146 c3[i] = r3; 1147 c4[i] = r4; 1148 } 1149 b1 += bm4; b2 += bm4; b3 += bm4; b4 += bm4; 1150 c1 += am4; c2 += am4; c3 += am4; c4 += am4; 1151 } 1152 for (; col<cn; col++) { /* over extra columns of C */ 1153 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1154 r1 = 0.0; 1155 n = a->i[i+1] - a->i[i]; 1156 aj = a->j + a->i[i]; 1157 aa = a->a + a->i[i]; 1158 for (j=0; j<n; j++) { 1159 r1 += aa[j]*b1[aj[j]]; 1160 } 1161 c1[i] = r1; 1162 } 1163 b1 += bm; 1164 c1 += am; 1165 } 1166 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1167 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1168 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1169 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1170 PetscFunctionReturn(0); 1171 } 1172 1173 /* 1174 Note very similar to MatMult_SeqAIJ(), should generate both codes from same base 1175 */ 1176 PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1177 { 1178 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1179 Mat_SeqDense *bd = (Mat_SeqDense*)B->data; 1180 PetscErrorCode ierr; 1181 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1182 MatScalar *aa; 1183 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=bd->lda, col, i,j,n,*aj, am = A->rmap->n,*ii,arm; 1184 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam,*ridx; 1185 1186 PetscFunctionBegin; 1187 if (!cm || !cn) PetscFunctionReturn(0); 1188 b = bd->v; 1189 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1190 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1191 1192 if (a->compressedrow.use) { /* use compressed row format */ 1193 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1194 colam = col*am; 1195 arm = a->compressedrow.nrows; 1196 ii = a->compressedrow.i; 1197 ridx = a->compressedrow.rindex; 1198 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1199 r1 = r2 = r3 = r4 = 0.0; 1200 n = ii[i+1] - ii[i]; 1201 aj = a->j + ii[i]; 1202 aa = a->a + ii[i]; 1203 for (j=0; j<n; j++) { 1204 r1 += (*aa)*b1[*aj]; 1205 r2 += (*aa)*b2[*aj]; 1206 r3 += (*aa)*b3[*aj]; 1207 r4 += (*aa++)*b4[*aj++]; 1208 } 1209 c[colam + ridx[i]] += r1; 1210 c[colam + am + ridx[i]] += r2; 1211 c[colam + am2 + ridx[i]] += r3; 1212 c[colam + am3 + ridx[i]] += r4; 1213 } 1214 b1 += bm4; 1215 b2 += bm4; 1216 b3 += bm4; 1217 b4 += bm4; 1218 } 1219 for (; col<cn; col++) { /* over extra columns of C */ 1220 colam = col*am; 1221 arm = a->compressedrow.nrows; 1222 ii = a->compressedrow.i; 1223 ridx = a->compressedrow.rindex; 1224 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1225 r1 = 0.0; 1226 n = ii[i+1] - ii[i]; 1227 aj = a->j + ii[i]; 1228 aa = a->a + ii[i]; 1229 1230 for (j=0; j<n; j++) { 1231 r1 += (*aa++)*b1[*aj++]; 1232 } 1233 c[colam + ridx[i]] += r1; 1234 } 1235 b1 += bm; 1236 } 1237 } else { 1238 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1239 colam = col*am; 1240 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1241 r1 = r2 = r3 = r4 = 0.0; 1242 n = a->i[i+1] - a->i[i]; 1243 aj = a->j + a->i[i]; 1244 aa = a->a + a->i[i]; 1245 for (j=0; j<n; j++) { 1246 r1 += (*aa)*b1[*aj]; 1247 r2 += (*aa)*b2[*aj]; 1248 r3 += (*aa)*b3[*aj]; 1249 r4 += (*aa++)*b4[*aj++]; 1250 } 1251 c[colam + i] += r1; 1252 c[colam + am + i] += r2; 1253 c[colam + am2 + i] += r3; 1254 c[colam + am3 + i] += r4; 1255 } 1256 b1 += bm4; 1257 b2 += bm4; 1258 b3 += bm4; 1259 b4 += bm4; 1260 } 1261 for (; col<cn; col++) { /* over extra columns of C */ 1262 colam = col*am; 1263 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1264 r1 = 0.0; 1265 n = a->i[i+1] - a->i[i]; 1266 aj = a->j + a->i[i]; 1267 aa = a->a + a->i[i]; 1268 1269 for (j=0; j<n; j++) { 1270 r1 += (*aa++)*b1[*aj++]; 1271 } 1272 c[colam + i] += r1; 1273 } 1274 b1 += bm; 1275 } 1276 } 1277 ierr = PetscLogFlops(cn*2.0*a->nz);CHKERRQ(ierr); 1278 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1279 PetscFunctionReturn(0); 1280 } 1281 1282 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1283 { 1284 PetscErrorCode ierr; 1285 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1286 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1287 PetscInt *bi = b->i,*bj=b->j; 1288 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1289 MatScalar *btval,*btval_den,*ba=b->a; 1290 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1291 1292 PetscFunctionBegin; 1293 btval_den=btdense->v; 1294 ierr = PetscMemzero(btval_den,(m*n)*sizeof(MatScalar));CHKERRQ(ierr); 1295 for (k=0; k<ncolors; k++) { 1296 ncolumns = coloring->ncolumns[k]; 1297 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1298 col = *(columns + colorforcol[k] + l); 1299 btcol = bj + bi[col]; 1300 btval = ba + bi[col]; 1301 anz = bi[col+1] - bi[col]; 1302 for (j=0; j<anz; j++) { 1303 brow = btcol[j]; 1304 btval_den[brow] = btval[j]; 1305 } 1306 } 1307 btval_den += m; 1308 } 1309 PetscFunctionReturn(0); 1310 } 1311 1312 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1313 { 1314 PetscErrorCode ierr; 1315 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1316 PetscScalar *ca_den,*ca_den_ptr,*ca=csp->a; 1317 PetscInt k,l,m=Cden->rmap->n,ncolors=matcoloring->ncolors; 1318 PetscInt brows=matcoloring->brows,*den2sp=matcoloring->den2sp; 1319 PetscInt nrows,*row,*idx; 1320 PetscInt *rows=matcoloring->rows,*colorforrow=matcoloring->colorforrow; 1321 1322 PetscFunctionBegin; 1323 ierr = MatDenseGetArray(Cden,&ca_den);CHKERRQ(ierr); 1324 1325 if (brows > 0) { 1326 PetscInt *lstart,row_end,row_start; 1327 lstart = matcoloring->lstart; 1328 ierr = PetscMemzero(lstart,ncolors*sizeof(PetscInt));CHKERRQ(ierr); 1329 1330 row_end = brows; 1331 if (row_end > m) row_end = m; 1332 for (row_start=0; row_start<m; row_start+=brows) { /* loop over row blocks of Csp */ 1333 ca_den_ptr = ca_den; 1334 for (k=0; k<ncolors; k++) { /* loop over colors (columns of Cden) */ 1335 nrows = matcoloring->nrows[k]; 1336 row = rows + colorforrow[k]; 1337 idx = den2sp + colorforrow[k]; 1338 for (l=lstart[k]; l<nrows; l++) { 1339 if (row[l] >= row_end) { 1340 lstart[k] = l; 1341 break; 1342 } else { 1343 ca[idx[l]] = ca_den_ptr[row[l]]; 1344 } 1345 } 1346 ca_den_ptr += m; 1347 } 1348 row_end += brows; 1349 if (row_end > m) row_end = m; 1350 } 1351 } else { /* non-blocked impl: loop over columns of Csp - slow if Csp is large */ 1352 ca_den_ptr = ca_den; 1353 for (k=0; k<ncolors; k++) { 1354 nrows = matcoloring->nrows[k]; 1355 row = rows + colorforrow[k]; 1356 idx = den2sp + colorforrow[k]; 1357 for (l=0; l<nrows; l++) { 1358 ca[idx[l]] = ca_den_ptr[row[l]]; 1359 } 1360 ca_den_ptr += m; 1361 } 1362 } 1363 1364 ierr = MatDenseRestoreArray(Cden,&ca_den);CHKERRQ(ierr); 1365 #if defined(PETSC_USE_INFO) 1366 if (matcoloring->brows > 0) { 1367 ierr = PetscInfo1(Csp,"Loop over %D row blocks for den2sp\n",brows);CHKERRQ(ierr); 1368 } else { 1369 ierr = PetscInfo(Csp,"Loop over colors/columns of Cden, inefficient for large sparse matrix product \n");CHKERRQ(ierr); 1370 } 1371 #endif 1372 PetscFunctionReturn(0); 1373 } 1374 1375 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1376 { 1377 PetscErrorCode ierr; 1378 PetscInt i,n,nrows,Nbs,j,k,m,ncols,col,cm; 1379 const PetscInt *is,*ci,*cj,*row_idx; 1380 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1381 IS *isa; 1382 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1383 PetscInt *colorforrow,*rows,*rows_i,*idxhit,*spidx,*den2sp,*den2sp_i; 1384 PetscInt *colorforcol,*columns,*columns_i,brows; 1385 PetscBool flg; 1386 1387 PetscFunctionBegin; 1388 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1389 1390 /* bs >1 is not being tested yet! */ 1391 Nbs = mat->cmap->N/bs; 1392 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1393 c->N = Nbs; 1394 c->m = c->M; 1395 c->rstart = 0; 1396 c->brows = 100; 1397 1398 c->ncolors = nis; 1399 ierr = PetscMalloc3(nis,&c->ncolumns,nis,&c->nrows,nis+1,&colorforrow);CHKERRQ(ierr); 1400 ierr = PetscMalloc1(csp->nz+1,&rows);CHKERRQ(ierr); 1401 ierr = PetscMalloc1(csp->nz+1,&den2sp);CHKERRQ(ierr); 1402 1403 brows = c->brows; 1404 ierr = PetscOptionsGetInt(NULL,NULL,"-matden2sp_brows",&brows,&flg);CHKERRQ(ierr); 1405 if (flg) c->brows = brows; 1406 if (brows > 0) { 1407 ierr = PetscMalloc1(nis+1,&c->lstart);CHKERRQ(ierr); 1408 } 1409 1410 colorforrow[0] = 0; 1411 rows_i = rows; 1412 den2sp_i = den2sp; 1413 1414 ierr = PetscMalloc1(nis+1,&colorforcol);CHKERRQ(ierr); 1415 ierr = PetscMalloc1(Nbs+1,&columns);CHKERRQ(ierr); 1416 1417 colorforcol[0] = 0; 1418 columns_i = columns; 1419 1420 /* get column-wise storage of mat */ 1421 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1422 1423 cm = c->m; 1424 ierr = PetscMalloc1(cm+1,&rowhit);CHKERRQ(ierr); 1425 ierr = PetscMalloc1(cm+1,&idxhit);CHKERRQ(ierr); 1426 for (i=0; i<nis; i++) { /* loop over color */ 1427 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 1428 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 1429 1430 c->ncolumns[i] = n; 1431 if (n) { 1432 ierr = PetscMemcpy(columns_i,is,n*sizeof(PetscInt));CHKERRQ(ierr); 1433 } 1434 colorforcol[i+1] = colorforcol[i] + n; 1435 columns_i += n; 1436 1437 /* fast, crude version requires O(N*N) work */ 1438 ierr = PetscMemzero(rowhit,cm*sizeof(PetscInt));CHKERRQ(ierr); 1439 1440 for (j=0; j<n; j++) { /* loop over columns*/ 1441 col = is[j]; 1442 row_idx = cj + ci[col]; 1443 m = ci[col+1] - ci[col]; 1444 for (k=0; k<m; k++) { /* loop over columns marking them in rowhit */ 1445 idxhit[*row_idx] = spidx[ci[col] + k]; 1446 rowhit[*row_idx++] = col + 1; 1447 } 1448 } 1449 /* count the number of hits */ 1450 nrows = 0; 1451 for (j=0; j<cm; j++) { 1452 if (rowhit[j]) nrows++; 1453 } 1454 c->nrows[i] = nrows; 1455 colorforrow[i+1] = colorforrow[i] + nrows; 1456 1457 nrows = 0; 1458 for (j=0; j<cm; j++) { /* loop over rows */ 1459 if (rowhit[j]) { 1460 rows_i[nrows] = j; 1461 den2sp_i[nrows] = idxhit[j]; 1462 nrows++; 1463 } 1464 } 1465 den2sp_i += nrows; 1466 1467 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1468 rows_i += nrows; 1469 } 1470 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1471 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1472 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 1473 if (csp->nz != colorforrow[nis]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %d != colorforrow[nis] %d",csp->nz,colorforrow[nis]); 1474 1475 c->colorforrow = colorforrow; 1476 c->rows = rows; 1477 c->den2sp = den2sp; 1478 c->colorforcol = colorforcol; 1479 c->columns = columns; 1480 1481 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1482 PetscFunctionReturn(0); 1483 } 1484