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 <../src/mat/utils/petscheap.h> 10 #include <petscbt.h> 11 #include <../src/mat/impls/dense/seq/dense.h> 12 #include <petsctime.h> 13 14 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat,Mat,PetscReal,Mat*); 15 16 #undef __FUNCT__ 17 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqAIJ" 18 PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 19 { 20 PetscErrorCode ierr; 21 const char *algTypes[6] = {"sorted","scalable","scalable_fast","heap","btheap","llcondensed"}; 22 PetscInt alg=0; /* set default algorithm */ 23 24 PetscFunctionBegin; 25 if (scall == MAT_INITIAL_MATRIX) { 26 ierr = PetscObjectOptionsBegin((PetscObject)A);CHKERRQ(ierr); 27 ierr = PetscOptionsEList("-matmatmult_via","Algorithmic approach","MatMatMult",algTypes,6,algTypes[0],&alg,NULL);CHKERRQ(ierr); 28 ierr = PetscOptionsEnd();CHKERRQ(ierr); 29 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 30 switch (alg) { 31 case 1: 32 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 33 break; 34 case 2: 35 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 36 break; 37 case 3: 38 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 39 break; 40 case 4: 41 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 42 break; 43 case 5: 44 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(A,B,fill,C);CHKERRQ(ierr); 45 break; 46 default: 47 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 48 break; 49 } 50 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 51 } 52 53 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 54 ierr = (*(*C)->ops->matmultnumeric)(A,B,*C);CHKERRQ(ierr); 55 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 56 PetscFunctionReturn(0); 57 } 58 59 #undef __FUNCT__ 60 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed" 61 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat A,Mat B,PetscReal fill,Mat *C) 62 { 63 PetscErrorCode ierr; 64 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 65 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 66 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 67 PetscReal afill; 68 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 69 PetscBT lnkbt; 70 PetscFreeSpaceList free_space=NULL,current_space=NULL; 71 72 PetscFunctionBegin; 73 /* Get ci and cj */ 74 /*---------------*/ 75 /* Allocate ci array, arrays for fill computation and */ 76 /* free space for accumulating nonzero column info */ 77 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 78 ci[0] = 0; 79 80 /* create and initialize a linked list */ 81 nlnk_max = a->rmax*b->rmax; 82 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; 83 ierr = PetscLLCondensedCreate(nlnk_max,bn,&lnk,&lnkbt);CHKERRQ(ierr); 84 85 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 86 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 87 88 current_space = free_space; 89 90 /* Determine ci and cj */ 91 for (i=0; i<am; i++) { 92 anzi = ai[i+1] - ai[i]; 93 aj = a->j + ai[i]; 94 for (j=0; j<anzi; j++) { 95 brow = aj[j]; 96 bnzj = bi[brow+1] - bi[brow]; 97 bj = b->j + bi[brow]; 98 /* add non-zero cols of B into the sorted linked list lnk */ 99 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 100 } 101 cnzi = lnk[0]; 102 103 /* If free space is not available, make more free space */ 104 /* Double the amount of total space in the list */ 105 if (current_space->local_remaining<cnzi) { 106 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 107 ndouble++; 108 } 109 110 /* Copy data into free space, then initialize lnk */ 111 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 112 113 current_space->array += cnzi; 114 current_space->local_used += cnzi; 115 current_space->local_remaining -= cnzi; 116 117 ci[i+1] = ci[i] + cnzi; 118 } 119 120 /* Column indices are in the list of free space */ 121 /* Allocate space for cj, initialize cj, and */ 122 /* destroy list of free space and other temporary array(s) */ 123 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 124 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 125 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 126 127 /* put together the new symbolic matrix */ 128 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 129 130 (*C)->rmap->bs = A->rmap->bs; 131 (*C)->cmap->bs = B->cmap->bs; 132 133 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 134 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 135 c = (Mat_SeqAIJ*)((*C)->data); 136 c->free_a = PETSC_FALSE; 137 c->free_ij = PETSC_TRUE; 138 c->nonew = 0; 139 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; /* fast, needs non-scalable O(bn) array 'abdense' */ 140 141 /* set MatInfo */ 142 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 143 if (afill < 1.0) afill = 1.0; 144 c->maxnz = ci[am]; 145 c->nz = ci[am]; 146 (*C)->info.mallocs = ndouble; 147 (*C)->info.fill_ratio_given = fill; 148 (*C)->info.fill_ratio_needed = afill; 149 150 #if defined(PETSC_USE_INFO) 151 if (ci[am]) { 152 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 153 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 154 } else { 155 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 156 } 157 #endif 158 PetscFunctionReturn(0); 159 } 160 161 #undef __FUNCT__ 162 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ" 163 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 164 { 165 PetscErrorCode ierr; 166 PetscLogDouble flops=0.0; 167 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 168 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 169 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 170 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 171 PetscInt am =A->rmap->n,cm=C->rmap->n; 172 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 173 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca,valtmp; 174 PetscScalar *ab_dense; 175 176 PetscFunctionBegin; 177 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 178 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 179 c->a = ca; 180 c->free_a = PETSC_TRUE; 181 } else { 182 ca = c->a; 183 } 184 if (!c->matmult_abdense) { 185 ierr = PetscMalloc(B->cmap->N*sizeof(PetscScalar),&ab_dense);CHKERRQ(ierr); 186 ierr = PetscMemzero(ab_dense,B->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 187 c->matmult_abdense = ab_dense; 188 } else { 189 ab_dense = c->matmult_abdense; 190 } 191 192 /* clean old values in C */ 193 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 194 /* Traverse A row-wise. */ 195 /* Build the ith row in C by summing over nonzero columns in A, */ 196 /* the rows of B corresponding to nonzeros of A. */ 197 for (i=0; i<am; i++) { 198 anzi = ai[i+1] - ai[i]; 199 for (j=0; j<anzi; j++) { 200 brow = aj[j]; 201 bnzi = bi[brow+1] - bi[brow]; 202 bjj = bj + bi[brow]; 203 baj = ba + bi[brow]; 204 /* perform dense axpy */ 205 valtmp = aa[j]; 206 for (k=0; k<bnzi; k++) { 207 ab_dense[bjj[k]] += valtmp*baj[k]; 208 } 209 flops += 2*bnzi; 210 } 211 aj += anzi; aa += anzi; 212 213 cnzi = ci[i+1] - ci[i]; 214 for (k=0; k<cnzi; k++) { 215 ca[k] += ab_dense[cj[k]]; 216 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 217 } 218 flops += cnzi; 219 cj += cnzi; ca += cnzi; 220 } 221 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 222 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 223 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 224 PetscFunctionReturn(0); 225 } 226 227 #undef __FUNCT__ 228 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable" 229 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 230 { 231 PetscErrorCode ierr; 232 PetscLogDouble flops=0.0; 233 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 234 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 235 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 236 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 237 PetscInt am = A->rmap->N,cm=C->rmap->N; 238 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 239 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca=c->a,valtmp; 240 PetscInt nextb; 241 242 PetscFunctionBegin; 243 /* clean old values in C */ 244 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 245 /* Traverse A row-wise. */ 246 /* Build the ith row in C by summing over nonzero columns in A, */ 247 /* the rows of B corresponding to nonzeros of A. */ 248 for (i=0; i<am; i++) { 249 anzi = ai[i+1] - ai[i]; 250 cnzi = ci[i+1] - ci[i]; 251 for (j=0; j<anzi; j++) { 252 brow = aj[j]; 253 bnzi = bi[brow+1] - bi[brow]; 254 bjj = bj + bi[brow]; 255 baj = ba + bi[brow]; 256 /* perform sparse axpy */ 257 valtmp = aa[j]; 258 nextb = 0; 259 for (k=0; nextb<bnzi; k++) { 260 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 261 ca[k] += valtmp*baj[nextb++]; 262 } 263 } 264 flops += 2*bnzi; 265 } 266 aj += anzi; aa += anzi; 267 cj += cnzi; ca += cnzi; 268 } 269 270 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 271 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 272 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 273 PetscFunctionReturn(0); 274 } 275 276 #undef __FUNCT__ 277 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast" 278 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat *C) 279 { 280 PetscErrorCode ierr; 281 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 282 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 283 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 284 MatScalar *ca; 285 PetscReal afill; 286 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 287 PetscFreeSpaceList free_space=NULL,current_space=NULL; 288 289 PetscFunctionBegin; 290 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 291 /*-----------------------------------------------------------------------------------------*/ 292 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 293 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 294 ci[0] = 0; 295 296 /* create and initialize a linked list */ 297 nlnk_max = a->rmax*b->rmax; 298 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 299 ierr = PetscLLCondensedCreate_fast(nlnk_max,&lnk);CHKERRQ(ierr); 300 301 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 302 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 303 current_space = free_space; 304 305 /* Determine ci and cj */ 306 for (i=0; i<am; i++) { 307 anzi = ai[i+1] - ai[i]; 308 aj = a->j + ai[i]; 309 for (j=0; j<anzi; j++) { 310 brow = aj[j]; 311 bnzj = bi[brow+1] - bi[brow]; 312 bj = b->j + bi[brow]; 313 /* add non-zero cols of B into the sorted linked list lnk */ 314 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 315 } 316 cnzi = lnk[1]; 317 318 /* If free space is not available, make more free space */ 319 /* Double the amount of total space in the list */ 320 if (current_space->local_remaining<cnzi) { 321 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 322 ndouble++; 323 } 324 325 /* Copy data into free space, then initialize lnk */ 326 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 327 328 current_space->array += cnzi; 329 current_space->local_used += cnzi; 330 current_space->local_remaining -= cnzi; 331 332 ci[i+1] = ci[i] + cnzi; 333 } 334 335 /* Column indices are in the list of free space */ 336 /* Allocate space for cj, initialize cj, and */ 337 /* destroy list of free space and other temporary array(s) */ 338 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 339 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 340 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 341 342 /* Allocate space for ca */ 343 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 344 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 345 346 /* put together the new symbolic matrix */ 347 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 348 349 (*C)->rmap->bs = A->rmap->bs; 350 (*C)->cmap->bs = B->cmap->bs; 351 352 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 353 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 354 c = (Mat_SeqAIJ*)((*C)->data); 355 c->free_a = PETSC_TRUE; 356 c->free_ij = PETSC_TRUE; 357 c->nonew = 0; 358 359 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 360 361 /* set MatInfo */ 362 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 363 if (afill < 1.0) afill = 1.0; 364 c->maxnz = ci[am]; 365 c->nz = ci[am]; 366 (*C)->info.mallocs = ndouble; 367 (*C)->info.fill_ratio_given = fill; 368 (*C)->info.fill_ratio_needed = afill; 369 370 #if defined(PETSC_USE_INFO) 371 if (ci[am]) { 372 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 373 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 374 } else { 375 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 376 } 377 #endif 378 PetscFunctionReturn(0); 379 } 380 381 382 #undef __FUNCT__ 383 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable" 384 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat *C) 385 { 386 PetscErrorCode ierr; 387 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 388 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 389 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 390 MatScalar *ca; 391 PetscReal afill; 392 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 393 PetscFreeSpaceList free_space=NULL,current_space=NULL; 394 395 PetscFunctionBegin; 396 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 397 /*---------------------------------------------------------------------------------------------*/ 398 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 399 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 400 ci[0] = 0; 401 402 /* create and initialize a linked list */ 403 nlnk_max = a->rmax*b->rmax; 404 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 405 ierr = PetscLLCondensedCreate_Scalable(nlnk_max,&lnk);CHKERRQ(ierr); 406 407 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 408 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 409 current_space = free_space; 410 411 /* Determine ci and cj */ 412 for (i=0; i<am; i++) { 413 anzi = ai[i+1] - ai[i]; 414 aj = a->j + ai[i]; 415 for (j=0; j<anzi; j++) { 416 brow = aj[j]; 417 bnzj = bi[brow+1] - bi[brow]; 418 bj = b->j + bi[brow]; 419 /* add non-zero cols of B into the sorted linked list lnk */ 420 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 421 } 422 cnzi = lnk[0]; 423 424 /* If free space is not available, make more free space */ 425 /* Double the amount of total space in the list */ 426 if (current_space->local_remaining<cnzi) { 427 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 428 ndouble++; 429 } 430 431 /* Copy data into free space, then initialize lnk */ 432 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 433 434 current_space->array += cnzi; 435 current_space->local_used += cnzi; 436 current_space->local_remaining -= cnzi; 437 438 ci[i+1] = ci[i] + cnzi; 439 } 440 441 /* Column indices are in the list of free space */ 442 /* Allocate space for cj, initialize cj, and */ 443 /* destroy list of free space and other temporary array(s) */ 444 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 445 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 446 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 447 448 /* Allocate space for ca */ 449 /*-----------------------*/ 450 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 451 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 452 453 /* put together the new symbolic matrix */ 454 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 455 456 (*C)->rmap->bs = A->rmap->bs; 457 (*C)->cmap->bs = B->cmap->bs; 458 459 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 460 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 461 c = (Mat_SeqAIJ*)((*C)->data); 462 c->free_a = PETSC_TRUE; 463 c->free_ij = PETSC_TRUE; 464 c->nonew = 0; 465 466 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 467 468 /* set MatInfo */ 469 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 470 if (afill < 1.0) afill = 1.0; 471 c->maxnz = ci[am]; 472 c->nz = ci[am]; 473 (*C)->info.mallocs = ndouble; 474 (*C)->info.fill_ratio_given = fill; 475 (*C)->info.fill_ratio_needed = afill; 476 477 #if defined(PETSC_USE_INFO) 478 if (ci[am]) { 479 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 480 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 481 } else { 482 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 483 } 484 #endif 485 PetscFunctionReturn(0); 486 } 487 488 #undef __FUNCT__ 489 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap" 490 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat *C) 491 { 492 PetscErrorCode ierr; 493 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 494 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 495 PetscInt *ci,*cj,*bb; 496 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 497 PetscReal afill; 498 PetscInt i,j,col,ndouble = 0; 499 PetscFreeSpaceList free_space=NULL,current_space=NULL; 500 PetscHeap h; 501 502 PetscFunctionBegin; 503 /* Get ci and cj - by merging sorted rows using a heap */ 504 /*---------------------------------------------------------------------------------------------*/ 505 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 506 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 507 ci[0] = 0; 508 509 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 510 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 511 current_space = free_space; 512 513 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 514 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 515 516 /* Determine ci and cj */ 517 for (i=0; i<am; i++) { 518 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 */ 519 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 520 ci[i+1] = ci[i]; 521 /* Populate the min heap */ 522 for (j=0; j<anzi; j++) { 523 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 524 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 525 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 526 } 527 } 528 /* Pick off the min element, adding it to free space */ 529 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 530 while (j >= 0) { 531 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 532 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 533 ndouble++; 534 } 535 *(current_space->array++) = col; 536 current_space->local_used++; 537 current_space->local_remaining--; 538 ci[i+1]++; 539 540 /* stash if anything else remains in this row of B */ 541 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 542 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 543 PetscInt j2,col2; 544 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 545 if (col2 != col) break; 546 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 547 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 548 } 549 /* Put any stashed elements back into the min heap */ 550 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 551 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 552 } 553 } 554 ierr = PetscFree(bb);CHKERRQ(ierr); 555 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 556 557 /* Column indices are in the list of free space */ 558 /* Allocate space for cj, initialize cj, and */ 559 /* destroy list of free space and other temporary array(s) */ 560 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 561 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 562 563 /* put together the new symbolic matrix */ 564 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 565 566 (*C)->rmap->bs = A->rmap->bs; 567 (*C)->cmap->bs = B->cmap->bs; 568 569 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 570 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 571 c = (Mat_SeqAIJ*)((*C)->data); 572 c->free_a = PETSC_TRUE; 573 c->free_ij = PETSC_TRUE; 574 c->nonew = 0; 575 576 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 577 578 /* set MatInfo */ 579 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 580 if (afill < 1.0) afill = 1.0; 581 c->maxnz = ci[am]; 582 c->nz = ci[am]; 583 (*C)->info.mallocs = ndouble; 584 (*C)->info.fill_ratio_given = fill; 585 (*C)->info.fill_ratio_needed = afill; 586 587 #if defined(PETSC_USE_INFO) 588 if (ci[am]) { 589 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 590 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 591 } else { 592 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 593 } 594 #endif 595 PetscFunctionReturn(0); 596 } 597 598 #undef __FUNCT__ 599 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap" 600 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat *C) 601 { 602 PetscErrorCode ierr; 603 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 604 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 605 PetscInt *ci,*cj,*bb; 606 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 607 PetscReal afill; 608 PetscInt i,j,col,ndouble = 0; 609 PetscFreeSpaceList free_space=NULL,current_space=NULL; 610 PetscHeap h; 611 PetscBT bt; 612 613 PetscFunctionBegin; 614 /* Get ci and cj - using a heap for the sorted rows, but use BT so that each index is only added once */ 615 /*---------------------------------------------------------------------------------------------*/ 616 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 617 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 618 ci[0] = 0; 619 620 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 621 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 622 623 current_space = free_space; 624 625 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 626 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 627 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 628 629 /* Determine ci and cj */ 630 for (i=0; i<am; i++) { 631 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 */ 632 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 633 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 634 ci[i+1] = ci[i]; 635 /* Populate the min heap */ 636 for (j=0; j<anzi; j++) { 637 PetscInt brow = acol[j]; 638 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 639 PetscInt bcol = bj[bb[j]]; 640 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 641 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 642 bb[j]++; 643 break; 644 } 645 } 646 } 647 /* Pick off the min element, adding it to free space */ 648 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 649 while (j >= 0) { 650 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 651 fptr = NULL; /* need PetscBTMemzero */ 652 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 653 ndouble++; 654 } 655 *(current_space->array++) = col; 656 current_space->local_used++; 657 current_space->local_remaining--; 658 ci[i+1]++; 659 660 /* stash if anything else remains in this row of B */ 661 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 662 PetscInt bcol = bj[bb[j]]; 663 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 664 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 665 bb[j]++; 666 break; 667 } 668 } 669 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 670 } 671 if (fptr) { /* Clear the bits for this row */ 672 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 673 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 674 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 675 } 676 } 677 ierr = PetscFree(bb);CHKERRQ(ierr); 678 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 679 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 680 681 /* Column indices are in the list of free space */ 682 /* Allocate space for cj, initialize cj, and */ 683 /* destroy list of free space and other temporary array(s) */ 684 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 685 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 686 687 /* put together the new symbolic matrix */ 688 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 689 690 (*C)->rmap->bs = A->rmap->bs; 691 (*C)->cmap->bs = B->cmap->bs; 692 693 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 694 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 695 c = (Mat_SeqAIJ*)((*C)->data); 696 c->free_a = PETSC_TRUE; 697 c->free_ij = PETSC_TRUE; 698 c->nonew = 0; 699 700 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 701 702 /* set MatInfo */ 703 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 704 if (afill < 1.0) afill = 1.0; 705 c->maxnz = ci[am]; 706 c->nz = ci[am]; 707 (*C)->info.mallocs = ndouble; 708 (*C)->info.fill_ratio_given = fill; 709 (*C)->info.fill_ratio_needed = afill; 710 711 #if defined(PETSC_USE_INFO) 712 if (ci[am]) { 713 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 714 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 715 } else { 716 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 717 } 718 #endif 719 PetscFunctionReturn(0); 720 } 721 722 #undef __FUNCT__ 723 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ" 724 /* concatenate unique entries and then sort */ 725 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 726 { 727 PetscErrorCode ierr; 728 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 729 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 730 PetscInt *ci,*cj; 731 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 732 PetscReal afill; 733 PetscInt i,j,ndouble = 0; 734 PetscSegBuffer seg,segrow; 735 char *seen; 736 737 PetscFunctionBegin; 738 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 739 ci[0] = 0; 740 741 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 742 ierr = PetscSegBufferCreate(sizeof(PetscInt),(PetscInt)(fill*(ai[am]+bi[bm])),&seg);CHKERRQ(ierr); 743 ierr = PetscSegBufferCreate(sizeof(PetscInt),100,&segrow);CHKERRQ(ierr); 744 ierr = PetscMalloc(bn*sizeof(char),&seen);CHKERRQ(ierr); 745 ierr = PetscMemzero(seen,bn*sizeof(char));CHKERRQ(ierr); 746 747 /* Determine ci and cj */ 748 for (i=0; i<am; i++) { 749 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 */ 750 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 751 PetscInt packlen = 0,*PETSC_RESTRICT crow; 752 /* Pack segrow */ 753 for (j=0; j<anzi; j++) { 754 PetscInt brow = acol[j],bjstart = bi[brow],bjend = bi[brow+1],k; 755 for (k=bjstart; k<bjend; k++) { 756 PetscInt bcol = bj[k]; 757 if (!seen[bcol]) { /* new entry */ 758 PetscInt *PETSC_RESTRICT slot; 759 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 760 *slot = bcol; 761 seen[bcol] = 1; 762 packlen++; 763 } 764 } 765 } 766 ierr = PetscSegBufferGetInts(seg,packlen,&crow);CHKERRQ(ierr); 767 ierr = PetscSegBufferExtractTo(segrow,crow);CHKERRQ(ierr); 768 ierr = PetscSortInt(packlen,crow);CHKERRQ(ierr); 769 ci[i+1] = ci[i] + packlen; 770 for (j=0; j<packlen; j++) seen[crow[j]] = 0; 771 } 772 ierr = PetscSegBufferDestroy(&segrow);CHKERRQ(ierr); 773 ierr = PetscFree(seen);CHKERRQ(ierr); 774 775 /* Column indices are in the segmented buffer */ 776 ierr = PetscSegBufferExtractAlloc(seg,&cj);CHKERRQ(ierr); 777 ierr = PetscSegBufferDestroy(&seg);CHKERRQ(ierr); 778 779 /* put together the new symbolic matrix */ 780 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 781 782 (*C)->rmap->bs = A->rmap->bs; 783 (*C)->cmap->bs = B->cmap->bs; 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,fill,afill);CHKERRQ(ierr); 806 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",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 #undef __FUNCT__ 816 #define __FUNCT__ "MatMatTransposeMult_SeqAIJ_SeqAIJ" 817 PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 818 { 819 PetscErrorCode ierr; 820 821 PetscFunctionBegin; 822 if (scall == MAT_INITIAL_MATRIX) { 823 ierr = PetscLogEventBegin(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 824 ierr = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 825 ierr = PetscLogEventEnd(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 826 } 827 ierr = PetscLogEventBegin(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 828 ierr = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 829 ierr = PetscLogEventEnd(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 830 PetscFunctionReturn(0); 831 } 832 833 #undef __FUNCT__ 834 #define __FUNCT__ "MatDestroy_SeqAIJ_MatMatMultTrans" 835 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(Mat A) 836 { 837 PetscErrorCode ierr; 838 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 839 Mat_MatMatTransMult *abt=a->abt; 840 841 PetscFunctionBegin; 842 ierr = (abt->destroy)(A);CHKERRQ(ierr); 843 ierr = MatTransposeColoringDestroy(&abt->matcoloring);CHKERRQ(ierr); 844 ierr = MatDestroy(&abt->Bt_den);CHKERRQ(ierr); 845 ierr = MatDestroy(&abt->ABt_den);CHKERRQ(ierr); 846 ierr = PetscFree(abt);CHKERRQ(ierr); 847 PetscFunctionReturn(0); 848 } 849 850 #undef __FUNCT__ 851 #define __FUNCT__ "MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ" 852 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 853 { 854 PetscErrorCode ierr; 855 Mat Bt; 856 PetscInt *bti,*btj; 857 Mat_MatMatTransMult *abt; 858 Mat_SeqAIJ *c; 859 860 PetscFunctionBegin; 861 /* create symbolic Bt */ 862 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 863 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 864 865 Bt->rmap->bs = A->cmap->bs; 866 Bt->cmap->bs = B->cmap->bs; 867 868 /* get symbolic C=A*Bt */ 869 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 870 871 /* create a supporting struct for reuse intermidiate dense matrices with matcoloring */ 872 ierr = PetscNew(Mat_MatMatTransMult,&abt);CHKERRQ(ierr); 873 c = (Mat_SeqAIJ*)(*C)->data; 874 c->abt = abt; 875 876 abt->usecoloring = PETSC_FALSE; 877 abt->destroy = (*C)->ops->destroy; 878 (*C)->ops->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 879 880 ierr = PetscOptionsGetBool(NULL,"-matmattransmult_color",&abt->usecoloring,NULL);CHKERRQ(ierr); 881 if (abt->usecoloring) { 882 /* Create MatTransposeColoring from symbolic C=A*B^T */ 883 MatTransposeColoring matcoloring; 884 MatColoring coloring; 885 ISColoring iscoloring; 886 Mat Bt_dense,C_dense; 887 Mat_SeqAIJ *c=(Mat_SeqAIJ*)(*C)->data; 888 /* inode causes memory problem, don't know why */ 889 if (c->inode.use) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MAT_USE_INODES is not supported. Use '-mat_no_inode'"); 890 891 ierr = MatColoringCreate(*C,&coloring);CHKERRQ(ierr); 892 ierr = MatColoringSetDistance(coloring,2);CHKERRQ(ierr); 893 ierr = MatColoringSetType(coloring,MATCOLORINGSL);CHKERRQ(ierr); 894 ierr = MatColoringSetFromOptions(coloring);CHKERRQ(ierr); 895 ierr = MatColoringApply(coloring,&iscoloring);CHKERRQ(ierr); 896 ierr = MatColoringDestroy(&coloring);CHKERRQ(ierr); 897 ierr = MatTransposeColoringCreate(*C,iscoloring,&matcoloring);CHKERRQ(ierr); 898 899 abt->matcoloring = matcoloring; 900 901 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 902 903 /* Create Bt_dense and C_dense = A*Bt_dense */ 904 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 905 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 906 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 907 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 908 909 Bt_dense->assembled = PETSC_TRUE; 910 abt->Bt_den = Bt_dense; 911 912 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 913 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 914 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 915 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 916 917 Bt_dense->assembled = PETSC_TRUE; 918 abt->ABt_den = C_dense; 919 920 #if defined(PETSC_USE_INFO) 921 { 922 Mat_SeqAIJ *c = (Mat_SeqAIJ*)(*C)->data; 923 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); 924 } 925 #endif 926 } 927 /* clean up */ 928 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 929 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 930 PetscFunctionReturn(0); 931 } 932 933 #undef __FUNCT__ 934 #define __FUNCT__ "MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ" 935 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 936 { 937 PetscErrorCode ierr; 938 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 939 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 940 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 941 PetscLogDouble flops=0.0; 942 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 943 Mat_MatMatTransMult *abt = c->abt; 944 945 PetscFunctionBegin; 946 /* clear old values in C */ 947 if (!c->a) { 948 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 949 c->a = ca; 950 c->free_a = PETSC_TRUE; 951 } else { 952 ca = c->a; 953 } 954 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 955 956 if (abt->usecoloring) { 957 MatTransposeColoring matcoloring = abt->matcoloring; 958 Mat Bt_dense,C_dense = abt->ABt_den; 959 960 /* Get Bt_dense by Apply MatTransposeColoring to B */ 961 Bt_dense = abt->Bt_den; 962 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 963 964 /* C_dense = A*Bt_dense */ 965 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 966 967 /* Recover C from C_dense */ 968 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 969 PetscFunctionReturn(0); 970 } 971 972 for (i=0; i<cm; i++) { 973 anzi = ai[i+1] - ai[i]; 974 acol = aj + ai[i]; 975 aval = aa + ai[i]; 976 cnzi = ci[i+1] - ci[i]; 977 ccol = cj + ci[i]; 978 cval = ca + ci[i]; 979 for (j=0; j<cnzi; j++) { 980 brow = ccol[j]; 981 bnzj = bi[brow+1] - bi[brow]; 982 bcol = bj + bi[brow]; 983 bval = ba + bi[brow]; 984 985 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 986 nexta = 0; nextb = 0; 987 while (nexta<anzi && nextb<bnzj) { 988 while (nexta < anzi && acol[nexta] < bcol[nextb]) nexta++; 989 if (nexta == anzi) break; 990 while (nextb < bnzj && acol[nexta] > bcol[nextb]) nextb++; 991 if (nextb == bnzj) break; 992 if (acol[nexta] == bcol[nextb]) { 993 cval[j] += aval[nexta]*bval[nextb]; 994 nexta++; nextb++; 995 flops += 2; 996 } 997 } 998 } 999 } 1000 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1001 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1002 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1003 PetscFunctionReturn(0); 1004 } 1005 1006 #undef __FUNCT__ 1007 #define __FUNCT__ "MatTransposeMatMult_SeqAIJ_SeqAIJ" 1008 PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1009 { 1010 PetscErrorCode ierr; 1011 1012 PetscFunctionBegin; 1013 if (scall == MAT_INITIAL_MATRIX) { 1014 ierr = PetscLogEventBegin(MAT_TransposeMatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1015 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1016 ierr = PetscLogEventEnd(MAT_TransposeMatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1017 } 1018 ierr = PetscLogEventBegin(MAT_TransposeMatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1019 ierr = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 1020 ierr = PetscLogEventEnd(MAT_TransposeMatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1021 PetscFunctionReturn(0); 1022 } 1023 1024 #undef __FUNCT__ 1025 #define __FUNCT__ "MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ" 1026 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 1027 { 1028 PetscErrorCode ierr; 1029 Mat At; 1030 PetscInt *ati,*atj; 1031 1032 PetscFunctionBegin; 1033 /* create symbolic At */ 1034 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1035 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1036 1037 At->rmap->bs = A->cmap->bs; 1038 At->cmap->bs = B->cmap->bs; 1039 1040 /* get symbolic C=At*B */ 1041 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(At,B,fill,C);CHKERRQ(ierr); 1042 1043 /* clean up */ 1044 ierr = MatDestroy(&At);CHKERRQ(ierr); 1045 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1046 PetscFunctionReturn(0); 1047 } 1048 1049 #undef __FUNCT__ 1050 #define __FUNCT__ "MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ" 1051 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1052 { 1053 PetscErrorCode ierr; 1054 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1055 PetscInt am =A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1056 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1057 PetscLogDouble flops=0.0; 1058 MatScalar *aa =a->a,*ba,*ca,*caj; 1059 1060 PetscFunctionBegin; 1061 if (!c->a) { 1062 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 1063 1064 c->a = ca; 1065 c->free_a = PETSC_TRUE; 1066 } else { 1067 ca = c->a; 1068 } 1069 /* clear old values in C */ 1070 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 1071 1072 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1073 for (i=0; i<am; i++) { 1074 bj = b->j + bi[i]; 1075 ba = b->a + bi[i]; 1076 bnzi = bi[i+1] - bi[i]; 1077 anzi = ai[i+1] - ai[i]; 1078 for (j=0; j<anzi; j++) { 1079 nextb = 0; 1080 crow = *aj++; 1081 cjj = cj + ci[crow]; 1082 caj = ca + ci[crow]; 1083 /* perform sparse axpy operation. Note cjj includes bj. */ 1084 for (k=0; nextb<bnzi; k++) { 1085 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1086 caj[k] += (*aa)*(*(ba+nextb)); 1087 nextb++; 1088 } 1089 } 1090 flops += 2*bnzi; 1091 aa++; 1092 } 1093 } 1094 1095 /* Assemble the final matrix and clean up */ 1096 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1097 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1098 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1099 PetscFunctionReturn(0); 1100 } 1101 1102 #undef __FUNCT__ 1103 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqDense" 1104 PetscErrorCode MatMatMult_SeqAIJ_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1105 { 1106 PetscErrorCode ierr; 1107 1108 PetscFunctionBegin; 1109 if (scall == MAT_INITIAL_MATRIX) { 1110 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1111 ierr = MatMatMultSymbolic_SeqAIJ_SeqDense(A,B,fill,C);CHKERRQ(ierr); 1112 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1113 } 1114 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1115 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,B,*C);CHKERRQ(ierr); 1116 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1117 PetscFunctionReturn(0); 1118 } 1119 1120 #undef __FUNCT__ 1121 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqDense" 1122 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C) 1123 { 1124 PetscErrorCode ierr; 1125 1126 PetscFunctionBegin; 1127 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1128 1129 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1130 PetscFunctionReturn(0); 1131 } 1132 1133 #undef __FUNCT__ 1134 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqDense" 1135 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1136 { 1137 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1138 PetscErrorCode ierr; 1139 PetscScalar *c,*b,r1,r2,r3,r4,*c1,*c2,*c3,*c4,aatmp; 1140 const PetscScalar *aa,*b1,*b2,*b3,*b4; 1141 const PetscInt *aj; 1142 PetscInt cm=C->rmap->n,cn=B->cmap->n,bm=B->rmap->n,am=A->rmap->n; 1143 PetscInt am4=4*am,bm4=4*bm,col,i,j,n,ajtmp; 1144 1145 PetscFunctionBegin; 1146 if (!cm || !cn) PetscFunctionReturn(0); 1147 if (bm != 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,bm); 1148 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); 1149 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); 1150 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1151 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1152 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1153 c1 = c; c2 = c1 + am; c3 = c2 + am; c4 = c3 + am; 1154 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1155 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1156 r1 = r2 = r3 = r4 = 0.0; 1157 n = a->i[i+1] - a->i[i]; 1158 aj = a->j + a->i[i]; 1159 aa = a->a + a->i[i]; 1160 for (j=0; j<n; j++) { 1161 aatmp = aa[j]; ajtmp = aj[j]; 1162 r1 += aatmp*b1[ajtmp]; 1163 r2 += aatmp*b2[ajtmp]; 1164 r3 += aatmp*b3[ajtmp]; 1165 r4 += aatmp*b4[ajtmp]; 1166 } 1167 c1[i] = r1; 1168 c2[i] = r2; 1169 c3[i] = r3; 1170 c4[i] = r4; 1171 } 1172 b1 += bm4; b2 += bm4; b3 += bm4; b4 += bm4; 1173 c1 += am4; c2 += am4; c3 += am4; c4 += am4; 1174 } 1175 for (; col<cn; col++) { /* over extra columns of C */ 1176 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1177 r1 = 0.0; 1178 n = a->i[i+1] - a->i[i]; 1179 aj = a->j + a->i[i]; 1180 aa = a->a + a->i[i]; 1181 for (j=0; j<n; j++) { 1182 r1 += aa[j]*b1[aj[j]]; 1183 } 1184 c1[i] = r1; 1185 } 1186 b1 += bm; 1187 c1 += am; 1188 } 1189 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1190 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1191 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1192 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1193 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1194 PetscFunctionReturn(0); 1195 } 1196 1197 /* 1198 Note very similar to MatMult_SeqAIJ(), should generate both codes from same base 1199 */ 1200 #undef __FUNCT__ 1201 #define __FUNCT__ "MatMatMultNumericAdd_SeqAIJ_SeqDense" 1202 PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1203 { 1204 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1205 PetscErrorCode ierr; 1206 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1207 MatScalar *aa; 1208 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=B->rmap->n, col, i,j,n,*aj, am = A->rmap->n,*ii,arm; 1209 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam,*ridx; 1210 1211 PetscFunctionBegin; 1212 if (!cm || !cn) PetscFunctionReturn(0); 1213 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1214 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1215 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1216 1217 if (a->compressedrow.use) { /* use compressed row format */ 1218 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1219 colam = col*am; 1220 arm = a->compressedrow.nrows; 1221 ii = a->compressedrow.i; 1222 ridx = a->compressedrow.rindex; 1223 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1224 r1 = r2 = r3 = r4 = 0.0; 1225 n = ii[i+1] - ii[i]; 1226 aj = a->j + ii[i]; 1227 aa = a->a + ii[i]; 1228 for (j=0; j<n; j++) { 1229 r1 += (*aa)*b1[*aj]; 1230 r2 += (*aa)*b2[*aj]; 1231 r3 += (*aa)*b3[*aj]; 1232 r4 += (*aa++)*b4[*aj++]; 1233 } 1234 c[colam + ridx[i]] += r1; 1235 c[colam + am + ridx[i]] += r2; 1236 c[colam + am2 + ridx[i]] += r3; 1237 c[colam + am3 + ridx[i]] += r4; 1238 } 1239 b1 += bm4; 1240 b2 += bm4; 1241 b3 += bm4; 1242 b4 += bm4; 1243 } 1244 for (; col<cn; col++) { /* over extra columns of C */ 1245 colam = col*am; 1246 arm = a->compressedrow.nrows; 1247 ii = a->compressedrow.i; 1248 ridx = a->compressedrow.rindex; 1249 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1250 r1 = 0.0; 1251 n = ii[i+1] - ii[i]; 1252 aj = a->j + ii[i]; 1253 aa = a->a + ii[i]; 1254 1255 for (j=0; j<n; j++) { 1256 r1 += (*aa++)*b1[*aj++]; 1257 } 1258 c[colam + ridx[i]] += r1; 1259 } 1260 b1 += bm; 1261 } 1262 } else { 1263 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1264 colam = col*am; 1265 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1266 r1 = r2 = r3 = r4 = 0.0; 1267 n = a->i[i+1] - a->i[i]; 1268 aj = a->j + a->i[i]; 1269 aa = a->a + a->i[i]; 1270 for (j=0; j<n; j++) { 1271 r1 += (*aa)*b1[*aj]; 1272 r2 += (*aa)*b2[*aj]; 1273 r3 += (*aa)*b3[*aj]; 1274 r4 += (*aa++)*b4[*aj++]; 1275 } 1276 c[colam + i] += r1; 1277 c[colam + am + i] += r2; 1278 c[colam + am2 + i] += r3; 1279 c[colam + am3 + i] += r4; 1280 } 1281 b1 += bm4; 1282 b2 += bm4; 1283 b3 += bm4; 1284 b4 += bm4; 1285 } 1286 for (; col<cn; col++) { /* over extra columns of C */ 1287 colam = col*am; 1288 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1289 r1 = 0.0; 1290 n = a->i[i+1] - a->i[i]; 1291 aj = a->j + a->i[i]; 1292 aa = a->a + a->i[i]; 1293 1294 for (j=0; j<n; j++) { 1295 r1 += (*aa++)*b1[*aj++]; 1296 } 1297 c[colam + i] += r1; 1298 } 1299 b1 += bm; 1300 } 1301 } 1302 ierr = PetscLogFlops(cn*2.0*a->nz);CHKERRQ(ierr); 1303 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1304 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1305 PetscFunctionReturn(0); 1306 } 1307 1308 #undef __FUNCT__ 1309 #define __FUNCT__ "MatTransColoringApplySpToDen_SeqAIJ" 1310 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1311 { 1312 PetscErrorCode ierr; 1313 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1314 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1315 PetscInt *bi = b->i,*bj=b->j; 1316 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1317 MatScalar *btval,*btval_den,*ba=b->a; 1318 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1319 1320 PetscFunctionBegin; 1321 btval_den=btdense->v; 1322 ierr = PetscMemzero(btval_den,(m*n)*sizeof(MatScalar));CHKERRQ(ierr); 1323 for (k=0; k<ncolors; k++) { 1324 ncolumns = coloring->ncolumns[k]; 1325 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1326 col = *(columns + colorforcol[k] + l); 1327 btcol = bj + bi[col]; 1328 btval = ba + bi[col]; 1329 anz = bi[col+1] - bi[col]; 1330 for (j=0; j<anz; j++) { 1331 brow = btcol[j]; 1332 btval_den[brow] = btval[j]; 1333 } 1334 } 1335 btval_den += m; 1336 } 1337 PetscFunctionReturn(0); 1338 } 1339 1340 #undef __FUNCT__ 1341 #define __FUNCT__ "MatTransColoringApplyDenToSp_SeqAIJ" 1342 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1343 { 1344 PetscErrorCode ierr; 1345 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1346 PetscScalar *ca_den,*ca_den_ptr,*ca=csp->a; 1347 PetscInt k,l,m=Cden->rmap->n,ncolors=matcoloring->ncolors; 1348 PetscInt brows=matcoloring->brows,*den2sp=matcoloring->den2sp; 1349 PetscInt nrows,*row,*idx; 1350 PetscInt *rows=matcoloring->rows,*colorforrow=matcoloring->colorforrow; 1351 1352 PetscFunctionBegin; 1353 ierr = MatDenseGetArray(Cden,&ca_den);CHKERRQ(ierr); 1354 1355 if (brows > 0) { 1356 PetscInt *lstart,row_end,row_start; 1357 lstart = matcoloring->lstart; 1358 ierr = PetscMemzero(lstart,ncolors*sizeof(PetscInt));CHKERRQ(ierr); 1359 1360 row_end = brows; 1361 if (row_end > m) row_end = m; 1362 for (row_start=0; row_start<m; row_start+=brows) { /* loop over row blocks of Csp */ 1363 ca_den_ptr = ca_den; 1364 for (k=0; k<ncolors; k++) { /* loop over colors (columns of Cden) */ 1365 nrows = matcoloring->nrows[k]; 1366 row = rows + colorforrow[k]; 1367 idx = den2sp + colorforrow[k]; 1368 for (l=lstart[k]; l<nrows; l++) { 1369 if (row[l] >= row_end) { 1370 lstart[k] = l; 1371 break; 1372 } else { 1373 ca[idx[l]] = ca_den_ptr[row[l]]; 1374 } 1375 } 1376 ca_den_ptr += m; 1377 } 1378 row_end += brows; 1379 if (row_end > m) row_end = m; 1380 } 1381 } else { /* non-blocked impl: loop over columns of Csp - slow if Csp is large */ 1382 ca_den_ptr = ca_den; 1383 for (k=0; k<ncolors; k++) { 1384 nrows = matcoloring->nrows[k]; 1385 row = rows + colorforrow[k]; 1386 idx = den2sp + colorforrow[k]; 1387 for (l=0; l<nrows; l++) { 1388 ca[idx[l]] = ca_den_ptr[row[l]]; 1389 } 1390 ca_den_ptr += m; 1391 } 1392 } 1393 1394 ierr = MatDenseRestoreArray(Cden,&ca_den);CHKERRQ(ierr); 1395 #if defined(PETSC_USE_INFO) 1396 if (matcoloring->brows > 0) { 1397 ierr = PetscInfo1(Csp,"Loop over %D row blocks for den2sp\n",brows);CHKERRQ(ierr); 1398 } else { 1399 ierr = PetscInfo(Csp,"Loop over colors/columns of Cden, inefficient for large sparse matrix product \n");CHKERRQ(ierr); 1400 } 1401 #endif 1402 PetscFunctionReturn(0); 1403 } 1404 1405 /* 1406 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 1407 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 1408 spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ(). 1409 */ 1410 #undef __FUNCT__ 1411 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ_Color" 1412 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1413 { 1414 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1415 PetscErrorCode ierr; 1416 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 1417 PetscInt nz = a->i[m],row,*jj,mr,col; 1418 PetscInt *cspidx; 1419 1420 PetscFunctionBegin; 1421 *nn = n; 1422 if (!ia) PetscFunctionReturn(0); 1423 if (symmetric) { 1424 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MatGetColumnIJ_SeqAIJ_Color() not supported for the case symmetric"); 1425 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 1426 } else { 1427 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&collengths);CHKERRQ(ierr); 1428 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1429 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&cia);CHKERRQ(ierr); 1430 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cja);CHKERRQ(ierr); 1431 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cspidx);CHKERRQ(ierr); 1432 jj = a->j; 1433 for (i=0; i<nz; i++) { 1434 collengths[jj[i]]++; 1435 } 1436 cia[0] = oshift; 1437 for (i=0; i<n; i++) { 1438 cia[i+1] = cia[i] + collengths[i]; 1439 } 1440 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1441 jj = a->j; 1442 for (row=0; row<m; row++) { 1443 mr = a->i[row+1] - a->i[row]; 1444 for (i=0; i<mr; i++) { 1445 col = *jj++; 1446 1447 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 1448 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 1449 } 1450 } 1451 ierr = PetscFree(collengths);CHKERRQ(ierr); 1452 *ia = cia; *ja = cja; 1453 *spidx = cspidx; 1454 } 1455 PetscFunctionReturn(0); 1456 } 1457 1458 #undef __FUNCT__ 1459 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ_Color" 1460 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1461 { 1462 PetscErrorCode ierr; 1463 1464 PetscFunctionBegin; 1465 if (!ia) PetscFunctionReturn(0); 1466 1467 ierr = PetscFree(*ia);CHKERRQ(ierr); 1468 ierr = PetscFree(*ja);CHKERRQ(ierr); 1469 ierr = PetscFree(*spidx);CHKERRQ(ierr); 1470 PetscFunctionReturn(0); 1471 } 1472 1473 #undef __FUNCT__ 1474 #define __FUNCT__ "MatTransposeColoringCreate_SeqAIJ" 1475 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1476 { 1477 PetscErrorCode ierr; 1478 PetscInt i,n,nrows,Nbs,j,k,m,ncols,col,cm; 1479 const PetscInt *is,*ci,*cj,*row_idx; 1480 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1481 IS *isa; 1482 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1483 PetscInt *colorforrow,*rows,*rows_i,*idxhit,*spidx,*den2sp,*den2sp_i; 1484 PetscInt *colorforcol,*columns,*columns_i,brows; 1485 PetscBool flg; 1486 1487 PetscFunctionBegin; 1488 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1489 1490 /* bs >1 is not being tested yet! */ 1491 Nbs = mat->cmap->N/bs; 1492 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1493 c->N = Nbs; 1494 c->m = c->M; 1495 c->rstart = 0; 1496 c->brows = 100; 1497 1498 c->ncolors = nis; 1499 ierr = PetscMalloc3(nis,PetscInt,&c->ncolumns,nis,PetscInt,&c->nrows,nis+1,PetscInt,&colorforrow);CHKERRQ(ierr); 1500 ierr = PetscMalloc((csp->nz+1)*sizeof(PetscInt),&rows);CHKERRQ(ierr); 1501 ierr = PetscMalloc((csp->nz+1)*sizeof(PetscInt),&den2sp);CHKERRQ(ierr); 1502 1503 brows = c->brows; 1504 ierr = PetscOptionsGetInt(NULL,"-matden2sp_brows",&brows,&flg);CHKERRQ(ierr); 1505 if (flg) c->brows = brows; 1506 if (brows > 0) { 1507 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&c->lstart);CHKERRQ(ierr); 1508 } 1509 1510 colorforrow[0] = 0; 1511 rows_i = rows; 1512 den2sp_i = den2sp; 1513 1514 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&colorforcol);CHKERRQ(ierr); 1515 ierr = PetscMalloc((Nbs+1)*sizeof(PetscInt),&columns);CHKERRQ(ierr); 1516 1517 colorforcol[0] = 0; 1518 columns_i = columns; 1519 1520 /* get column-wise storage of mat */ 1521 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1522 1523 cm = c->m; 1524 ierr = PetscMalloc((cm+1)*sizeof(PetscInt),&rowhit);CHKERRQ(ierr); 1525 ierr = PetscMalloc((cm+1)*sizeof(PetscInt),&idxhit);CHKERRQ(ierr); 1526 for (i=0; i<nis; i++) { /* loop over color */ 1527 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 1528 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 1529 1530 c->ncolumns[i] = n; 1531 if (n) { 1532 ierr = PetscMemcpy(columns_i,is,n*sizeof(PetscInt));CHKERRQ(ierr); 1533 } 1534 colorforcol[i+1] = colorforcol[i] + n; 1535 columns_i += n; 1536 1537 /* fast, crude version requires O(N*N) work */ 1538 ierr = PetscMemzero(rowhit,cm*sizeof(PetscInt));CHKERRQ(ierr); 1539 1540 /* loop over columns*/ 1541 for (j=0; j<n; j++) { 1542 col = is[j]; 1543 row_idx = cj + ci[col]; 1544 m = ci[col+1] - ci[col]; 1545 /* loop over columns marking them in rowhit */ 1546 for (k=0; k<m; k++) { 1547 idxhit[*row_idx] = spidx[ci[col] + k]; 1548 rowhit[*row_idx++] = col + 1; 1549 } 1550 } 1551 /* count the number of hits */ 1552 nrows = 0; 1553 for (j=0; j<cm; j++) { 1554 if (rowhit[j]) nrows++; 1555 } 1556 c->nrows[i] = nrows; 1557 colorforrow[i+1] = colorforrow[i] + nrows; 1558 1559 nrows = 0; 1560 for (j=0; j<cm; j++) { 1561 if (rowhit[j]) { 1562 rows_i[nrows] = j; 1563 den2sp_i[nrows] = idxhit[j]; 1564 nrows++; 1565 } 1566 } 1567 den2sp_i += nrows; 1568 1569 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1570 rows_i += nrows; 1571 } 1572 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1573 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1574 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 1575 if (csp->nz != colorforrow[nis]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %d != colorforrow[nis] %d",csp->nz,colorforrow[nis]); 1576 1577 c->colorforrow = colorforrow; 1578 c->rows = rows; 1579 c->den2sp = den2sp; 1580 c->colorforcol = colorforcol; 1581 c->columns = columns; 1582 1583 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1584 PetscFunctionReturn(0); 1585 } 1586