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 13 #undef __FUNCT__ 14 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqAIJ" 15 PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 16 { 17 PetscErrorCode ierr; 18 PetscBool scalable=PETSC_FALSE,scalable_fast=PETSC_FALSE,heap = PETSC_FALSE,btheap = PETSC_FALSE; 19 20 PetscFunctionBegin; 21 if (scall == MAT_INITIAL_MATRIX) { 22 ierr = PetscObjectOptionsBegin((PetscObject)A);CHKERRQ(ierr); 23 ierr = PetscOptionsBool("-matmatmult_scalable","Use a scalable but slower C=A*B","",scalable,&scalable,NULL);CHKERRQ(ierr); 24 ierr = PetscOptionsBool("-matmatmult_scalable_fast","Use a scalable but slower C=A*B","",scalable_fast,&scalable_fast,NULL);CHKERRQ(ierr); 25 ierr = PetscOptionsBool("-matmatmult_heap","Use heap implementation of symbolic factorization C=A*B","",heap,&heap,NULL);CHKERRQ(ierr); 26 ierr = PetscOptionsBool("-matmatmult_btheap","Use btheap implementation of symbolic factorization C=A*B","",btheap,&btheap,NULL);CHKERRQ(ierr); 27 ierr = PetscOptionsEnd();CHKERRQ(ierr); 28 if (scalable_fast) { 29 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 30 } else if (scalable) { 31 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 32 } else if (heap) { 33 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 34 } else if (btheap) { 35 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 36 } else { 37 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 38 } 39 } 40 41 ierr = (*(*C)->ops->matmultnumeric)(A,B,*C);CHKERRQ(ierr); 42 PetscFunctionReturn(0); 43 } 44 45 #undef __FUNCT__ 46 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ" 47 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 48 { 49 PetscErrorCode ierr; 50 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 51 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 52 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 53 PetscReal afill; 54 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 55 PetscBT lnkbt; 56 PetscFreeSpaceList free_space=NULL,current_space=NULL; 57 58 PetscFunctionBegin; 59 /* Get ci and cj */ 60 /*---------------*/ 61 /* Allocate ci array, arrays for fill computation and */ 62 /* free space for accumulating nonzero column info */ 63 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 64 ci[0] = 0; 65 66 /* create and initialize a linked list */ 67 nlnk_max = a->rmax*b->rmax; 68 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; 69 ierr = PetscLLCondensedCreate(nlnk_max,bn,&lnk,&lnkbt);CHKERRQ(ierr); 70 71 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 72 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 73 74 current_space = free_space; 75 76 /* Determine ci and cj */ 77 for (i=0; i<am; i++) { 78 anzi = ai[i+1] - ai[i]; 79 aj = a->j + ai[i]; 80 for (j=0; j<anzi; j++) { 81 brow = aj[j]; 82 bnzj = bi[brow+1] - bi[brow]; 83 bj = b->j + bi[brow]; 84 /* add non-zero cols of B into the sorted linked list lnk */ 85 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 86 } 87 cnzi = lnk[0]; 88 89 /* If free space is not available, make more free space */ 90 /* Double the amount of total space in the list */ 91 if (current_space->local_remaining<cnzi) { 92 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 93 ndouble++; 94 } 95 96 /* Copy data into free space, then initialize lnk */ 97 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 98 99 current_space->array += cnzi; 100 current_space->local_used += cnzi; 101 current_space->local_remaining -= cnzi; 102 103 ci[i+1] = ci[i] + cnzi; 104 } 105 106 /* Column indices are in the list of free space */ 107 /* Allocate space for cj, initialize cj, and */ 108 /* destroy list of free space and other temporary array(s) */ 109 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 110 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 111 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 112 113 /* put together the new symbolic matrix */ 114 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 115 116 (*C)->rmap->bs = A->rmap->bs; 117 (*C)->cmap->bs = B->cmap->bs; 118 119 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 120 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 121 c = (Mat_SeqAIJ*)((*C)->data); 122 c->free_a = PETSC_FALSE; 123 c->free_ij = PETSC_TRUE; 124 c->nonew = 0; 125 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; /* fast, needs non-scalable O(bn) array 'abdense' */ 126 127 /* set MatInfo */ 128 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 129 if (afill < 1.0) afill = 1.0; 130 c->maxnz = ci[am]; 131 c->nz = ci[am]; 132 (*C)->info.mallocs = ndouble; 133 (*C)->info.fill_ratio_given = fill; 134 (*C)->info.fill_ratio_needed = afill; 135 136 #if defined(PETSC_USE_INFO) 137 if (ci[am]) { 138 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 139 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 140 } else { 141 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 142 } 143 #endif 144 PetscFunctionReturn(0); 145 } 146 147 #undef __FUNCT__ 148 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ" 149 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 150 { 151 PetscErrorCode ierr; 152 PetscLogDouble flops=0.0; 153 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 154 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 155 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 156 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 157 PetscInt am =A->rmap->n,cm=C->rmap->n; 158 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 159 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca,valtmp; 160 PetscScalar *ab_dense; 161 162 PetscFunctionBegin; 163 /* printf("MatMatMultNumeric_SeqAIJ_SeqAIJ...ca %p\n",c->a); */ 164 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 165 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 166 c->a = ca; 167 c->free_a = PETSC_TRUE; 168 169 ierr = PetscMalloc(B->cmap->N*sizeof(PetscScalar),&ab_dense);CHKERRQ(ierr); 170 ierr = PetscMemzero(ab_dense,B->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 171 172 c->matmult_abdense = ab_dense; 173 } else { 174 ca = c->a; 175 ab_dense = c->matmult_abdense; 176 } 177 178 /* clean old values in C */ 179 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 180 /* Traverse A row-wise. */ 181 /* Build the ith row in C by summing over nonzero columns in A, */ 182 /* the rows of B corresponding to nonzeros of A. */ 183 for (i=0; i<am; i++) { 184 anzi = ai[i+1] - ai[i]; 185 for (j=0; j<anzi; j++) { 186 brow = aj[j]; 187 bnzi = bi[brow+1] - bi[brow]; 188 bjj = bj + bi[brow]; 189 baj = ba + bi[brow]; 190 /* perform dense axpy */ 191 valtmp = aa[j]; 192 for (k=0; k<bnzi; k++) { 193 ab_dense[bjj[k]] += valtmp*baj[k]; 194 } 195 flops += 2*bnzi; 196 } 197 aj += anzi; aa += anzi; 198 199 cnzi = ci[i+1] - ci[i]; 200 for (k=0; k<cnzi; k++) { 201 ca[k] += ab_dense[cj[k]]; 202 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 203 } 204 flops += cnzi; 205 cj += cnzi; ca += cnzi; 206 } 207 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 208 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 209 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 210 PetscFunctionReturn(0); 211 } 212 213 #undef __FUNCT__ 214 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable" 215 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 216 { 217 PetscErrorCode ierr; 218 PetscLogDouble flops=0.0; 219 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 220 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 221 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 222 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 223 PetscInt am = A->rmap->N,cm=C->rmap->N; 224 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 225 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca=c->a,valtmp; 226 PetscInt nextb; 227 228 PetscFunctionBegin; 229 /* clean old values in C */ 230 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 231 /* Traverse A row-wise. */ 232 /* Build the ith row in C by summing over nonzero columns in A, */ 233 /* the rows of B corresponding to nonzeros of A. */ 234 for (i=0; i<am; i++) { 235 anzi = ai[i+1] - ai[i]; 236 cnzi = ci[i+1] - ci[i]; 237 for (j=0; j<anzi; j++) { 238 brow = aj[j]; 239 bnzi = bi[brow+1] - bi[brow]; 240 bjj = bj + bi[brow]; 241 baj = ba + bi[brow]; 242 /* perform sparse axpy */ 243 valtmp = aa[j]; 244 nextb = 0; 245 for (k=0; nextb<bnzi; k++) { 246 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 247 ca[k] += valtmp*baj[nextb++]; 248 } 249 } 250 flops += 2*bnzi; 251 } 252 aj += anzi; aa += anzi; 253 cj += cnzi; ca += cnzi; 254 } 255 256 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 257 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 258 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 259 PetscFunctionReturn(0); 260 } 261 262 #undef __FUNCT__ 263 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast" 264 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat *C) 265 { 266 PetscErrorCode ierr; 267 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 268 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 269 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 270 MatScalar *ca; 271 PetscReal afill; 272 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 273 PetscFreeSpaceList free_space=NULL,current_space=NULL; 274 275 PetscFunctionBegin; 276 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 277 /*-----------------------------------------------------------------------------------------*/ 278 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 279 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 280 ci[0] = 0; 281 282 /* create and initialize a linked list */ 283 nlnk_max = a->rmax*b->rmax; 284 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 285 ierr = PetscLLCondensedCreate_fast(nlnk_max,&lnk);CHKERRQ(ierr); 286 287 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 288 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 289 current_space = free_space; 290 291 /* Determine ci and cj */ 292 for (i=0; i<am; i++) { 293 anzi = ai[i+1] - ai[i]; 294 aj = a->j + ai[i]; 295 for (j=0; j<anzi; j++) { 296 brow = aj[j]; 297 bnzj = bi[brow+1] - bi[brow]; 298 bj = b->j + bi[brow]; 299 /* add non-zero cols of B into the sorted linked list lnk */ 300 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 301 } 302 cnzi = lnk[1]; 303 304 /* If free space is not available, make more free space */ 305 /* Double the amount of total space in the list */ 306 if (current_space->local_remaining<cnzi) { 307 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 308 ndouble++; 309 } 310 311 /* Copy data into free space, then initialize lnk */ 312 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 313 314 current_space->array += cnzi; 315 current_space->local_used += cnzi; 316 current_space->local_remaining -= cnzi; 317 318 ci[i+1] = ci[i] + cnzi; 319 } 320 321 /* Column indices are in the list of free space */ 322 /* Allocate space for cj, initialize cj, and */ 323 /* destroy list of free space and other temporary array(s) */ 324 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 325 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 326 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 327 328 /* Allocate space for ca */ 329 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 330 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 331 332 /* put together the new symbolic matrix */ 333 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 334 335 (*C)->rmap->bs = A->rmap->bs; 336 (*C)->cmap->bs = B->cmap->bs; 337 338 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 339 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 340 c = (Mat_SeqAIJ*)((*C)->data); 341 c->free_a = PETSC_TRUE; 342 c->free_ij = PETSC_TRUE; 343 c->nonew = 0; 344 345 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 346 347 /* set MatInfo */ 348 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 349 if (afill < 1.0) afill = 1.0; 350 c->maxnz = ci[am]; 351 c->nz = ci[am]; 352 (*C)->info.mallocs = ndouble; 353 (*C)->info.fill_ratio_given = fill; 354 (*C)->info.fill_ratio_needed = afill; 355 356 #if defined(PETSC_USE_INFO) 357 if (ci[am]) { 358 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 359 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 360 } else { 361 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 362 } 363 #endif 364 PetscFunctionReturn(0); 365 } 366 367 368 #undef __FUNCT__ 369 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable" 370 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat *C) 371 { 372 PetscErrorCode ierr; 373 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 374 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 375 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 376 MatScalar *ca; 377 PetscReal afill; 378 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 379 PetscFreeSpaceList free_space=NULL,current_space=NULL; 380 381 PetscFunctionBegin; 382 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 383 /*---------------------------------------------------------------------------------------------*/ 384 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 385 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 386 ci[0] = 0; 387 388 /* create and initialize a linked list */ 389 nlnk_max = a->rmax*b->rmax; 390 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 391 ierr = PetscLLCondensedCreate_Scalable(nlnk_max,&lnk);CHKERRQ(ierr); 392 393 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 394 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 395 current_space = free_space; 396 397 /* Determine ci and cj */ 398 for (i=0; i<am; i++) { 399 anzi = ai[i+1] - ai[i]; 400 aj = a->j + ai[i]; 401 for (j=0; j<anzi; j++) { 402 brow = aj[j]; 403 bnzj = bi[brow+1] - bi[brow]; 404 bj = b->j + bi[brow]; 405 /* add non-zero cols of B into the sorted linked list lnk */ 406 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 407 } 408 cnzi = lnk[0]; 409 410 /* If free space is not available, make more free space */ 411 /* Double the amount of total space in the list */ 412 if (current_space->local_remaining<cnzi) { 413 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 414 ndouble++; 415 } 416 417 /* Copy data into free space, then initialize lnk */ 418 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 419 420 current_space->array += cnzi; 421 current_space->local_used += cnzi; 422 current_space->local_remaining -= cnzi; 423 424 ci[i+1] = ci[i] + cnzi; 425 } 426 427 /* Column indices are in the list of free space */ 428 /* Allocate space for cj, initialize cj, and */ 429 /* destroy list of free space and other temporary array(s) */ 430 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 431 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 432 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 433 434 /* Allocate space for ca */ 435 /*-----------------------*/ 436 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 437 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 438 439 /* put together the new symbolic matrix */ 440 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 441 442 (*C)->rmap->bs = A->rmap->bs; 443 (*C)->cmap->bs = B->cmap->bs; 444 445 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 446 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 447 c = (Mat_SeqAIJ*)((*C)->data); 448 c->free_a = PETSC_TRUE; 449 c->free_ij = PETSC_TRUE; 450 c->nonew = 0; 451 452 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 453 454 /* set MatInfo */ 455 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 456 if (afill < 1.0) afill = 1.0; 457 c->maxnz = ci[am]; 458 c->nz = ci[am]; 459 (*C)->info.mallocs = ndouble; 460 (*C)->info.fill_ratio_given = fill; 461 (*C)->info.fill_ratio_needed = afill; 462 463 #if defined(PETSC_USE_INFO) 464 if (ci[am]) { 465 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 466 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 467 } else { 468 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 469 } 470 #endif 471 PetscFunctionReturn(0); 472 } 473 474 #undef __FUNCT__ 475 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap" 476 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat *C) 477 { 478 PetscErrorCode ierr; 479 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 480 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 481 PetscInt *ci,*cj,*bb; 482 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 483 PetscReal afill; 484 PetscInt i,j,col,ndouble = 0; 485 PetscFreeSpaceList free_space=NULL,current_space=NULL; 486 PetscHeap h; 487 488 PetscFunctionBegin; 489 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 490 /*---------------------------------------------------------------------------------------------*/ 491 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 492 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 493 ci[0] = 0; 494 495 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 496 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 497 current_space = free_space; 498 499 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 500 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 501 502 /* Determine ci and cj */ 503 for (i=0; i<am; i++) { 504 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 */ 505 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 506 ci[i+1] = ci[i]; 507 /* Populate the min heap */ 508 for (j=0; j<anzi; j++) { 509 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 510 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 511 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 512 } 513 } 514 /* Pick off the min element, adding it to free space */ 515 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 516 while (j >= 0) { 517 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 518 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 519 ndouble++; 520 } 521 *(current_space->array++) = col; 522 current_space->local_used++; 523 current_space->local_remaining--; 524 ci[i+1]++; 525 526 /* stash if anything else remains in this row of B */ 527 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 528 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 529 PetscInt j2,col2; 530 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 531 if (col2 != col) break; 532 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 533 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 534 } 535 /* Put any stashed elements back into the min heap */ 536 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 537 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 538 } 539 } 540 ierr = PetscFree(bb);CHKERRQ(ierr); 541 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 542 543 /* Column indices are in the list of free space */ 544 /* Allocate space for cj, initialize cj, and */ 545 /* destroy list of free space and other temporary array(s) */ 546 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 547 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 548 549 /* put together the new symbolic matrix */ 550 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 551 552 (*C)->rmap->bs = A->rmap->bs; 553 (*C)->cmap->bs = B->cmap->bs; 554 555 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 556 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 557 c = (Mat_SeqAIJ*)((*C)->data); 558 c->free_a = PETSC_TRUE; 559 c->free_ij = PETSC_TRUE; 560 c->nonew = 0; 561 562 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 563 564 /* set MatInfo */ 565 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 566 if (afill < 1.0) afill = 1.0; 567 c->maxnz = ci[am]; 568 c->nz = ci[am]; 569 (*C)->info.mallocs = ndouble; 570 (*C)->info.fill_ratio_given = fill; 571 (*C)->info.fill_ratio_needed = afill; 572 573 #if defined(PETSC_USE_INFO) 574 if (ci[am]) { 575 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 576 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 577 } else { 578 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 579 } 580 #endif 581 PetscFunctionReturn(0); 582 } 583 584 #undef __FUNCT__ 585 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap" 586 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat *C) 587 { 588 PetscErrorCode ierr; 589 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 590 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 591 PetscInt *ci,*cj,*bb; 592 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 593 PetscReal afill; 594 PetscInt i,j,col,ndouble = 0; 595 PetscFreeSpaceList free_space=NULL,current_space=NULL; 596 PetscHeap h; 597 PetscBT bt; 598 599 PetscFunctionBegin; 600 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 601 /*---------------------------------------------------------------------------------------------*/ 602 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 603 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 604 ci[0] = 0; 605 606 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 607 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 608 609 current_space = free_space; 610 611 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 612 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 613 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 614 615 /* Determine ci and cj */ 616 for (i=0; i<am; i++) { 617 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 */ 618 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 619 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 620 ci[i+1] = ci[i]; 621 /* Populate the min heap */ 622 for (j=0; j<anzi; j++) { 623 PetscInt brow = acol[j]; 624 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 625 PetscInt bcol = bj[bb[j]]; 626 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 627 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 628 bb[j]++; 629 break; 630 } 631 } 632 } 633 /* Pick off the min element, adding it to free space */ 634 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 635 while (j >= 0) { 636 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 637 fptr = NULL; /* need PetscBTMemzero */ 638 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 639 ndouble++; 640 } 641 *(current_space->array++) = col; 642 current_space->local_used++; 643 current_space->local_remaining--; 644 ci[i+1]++; 645 646 /* stash if anything else remains in this row of B */ 647 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 648 PetscInt bcol = bj[bb[j]]; 649 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 650 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 651 bb[j]++; 652 break; 653 } 654 } 655 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 656 } 657 if (fptr) { /* Clear the bits for this row */ 658 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 659 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 660 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 661 } 662 } 663 ierr = PetscFree(bb);CHKERRQ(ierr); 664 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 665 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 666 667 /* Column indices are in the list of free space */ 668 /* Allocate space for cj, initialize cj, and */ 669 /* destroy list of free space and other temporary array(s) */ 670 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 671 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 672 673 /* put together the new symbolic matrix */ 674 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 675 676 (*C)->rmap->bs = A->rmap->bs; 677 (*C)->cmap->bs = B->cmap->bs; 678 679 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 680 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 681 c = (Mat_SeqAIJ*)((*C)->data); 682 c->free_a = PETSC_TRUE; 683 c->free_ij = PETSC_TRUE; 684 c->nonew = 0; 685 686 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 687 688 /* set MatInfo */ 689 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 690 if (afill < 1.0) afill = 1.0; 691 c->maxnz = ci[am]; 692 c->nz = ci[am]; 693 (*C)->info.mallocs = ndouble; 694 (*C)->info.fill_ratio_given = fill; 695 (*C)->info.fill_ratio_needed = afill; 696 697 #if defined(PETSC_USE_INFO) 698 if (ci[am]) { 699 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 700 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 701 } else { 702 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 703 } 704 #endif 705 PetscFunctionReturn(0); 706 } 707 708 /* This routine is not used. Should be removed! */ 709 #undef __FUNCT__ 710 #define __FUNCT__ "MatMatTransposeMult_SeqAIJ_SeqAIJ" 711 PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 712 { 713 PetscErrorCode ierr; 714 715 PetscFunctionBegin; 716 if (scall == MAT_INITIAL_MATRIX) { 717 ierr = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 718 } 719 ierr = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 720 PetscFunctionReturn(0); 721 } 722 723 #undef __FUNCT__ 724 #define __FUNCT__ "PetscContainerDestroy_Mat_MatMatTransMult" 725 PetscErrorCode PetscContainerDestroy_Mat_MatMatTransMult(void *ptr) 726 { 727 PetscErrorCode ierr; 728 Mat_MatMatTransMult *multtrans=(Mat_MatMatTransMult*)ptr; 729 730 PetscFunctionBegin; 731 ierr = MatTransposeColoringDestroy(&multtrans->matcoloring);CHKERRQ(ierr); 732 ierr = MatDestroy(&multtrans->Bt_den);CHKERRQ(ierr); 733 ierr = MatDestroy(&multtrans->ABt_den);CHKERRQ(ierr); 734 ierr = PetscFree(multtrans);CHKERRQ(ierr); 735 PetscFunctionReturn(0); 736 } 737 738 #undef __FUNCT__ 739 #define __FUNCT__ "MatDestroy_SeqAIJ_MatMatMultTrans" 740 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(Mat A) 741 { 742 PetscErrorCode ierr; 743 PetscContainer container; 744 Mat_MatMatTransMult *multtrans=NULL; 745 746 PetscFunctionBegin; 747 ierr = PetscObjectQuery((PetscObject)A,"Mat_MatMatTransMult",(PetscObject*)&container);CHKERRQ(ierr); 748 if (!container) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 749 ierr = PetscContainerGetPointer(container,(void**)&multtrans);CHKERRQ(ierr); 750 751 A->ops->destroy = multtrans->destroy; 752 if (A->ops->destroy) { 753 ierr = (*A->ops->destroy)(A);CHKERRQ(ierr); 754 } 755 ierr = PetscObjectCompose((PetscObject)A,"Mat_MatMatTransMult",0);CHKERRQ(ierr); 756 PetscFunctionReturn(0); 757 } 758 759 #undef __FUNCT__ 760 #define __FUNCT__ "MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ" 761 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 762 { 763 PetscErrorCode ierr; 764 Mat Bt; 765 PetscInt *bti,*btj; 766 Mat_MatMatTransMult *multtrans; 767 PetscContainer container; 768 769 PetscFunctionBegin; 770 /* create symbolic Bt */ 771 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 772 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 773 774 Bt->rmap->bs = A->cmap->bs; 775 Bt->cmap->bs = B->cmap->bs; 776 777 /* get symbolic C=A*Bt */ 778 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 779 780 /* create a supporting struct for reuse intermidiate dense matrices with matcoloring */ 781 ierr = PetscNew(Mat_MatMatTransMult,&multtrans);CHKERRQ(ierr); 782 783 /* attach the supporting struct to C */ 784 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 785 ierr = PetscContainerSetPointer(container,multtrans);CHKERRQ(ierr); 786 ierr = PetscContainerSetUserDestroy(container,PetscContainerDestroy_Mat_MatMatTransMult);CHKERRQ(ierr); 787 ierr = PetscObjectCompose((PetscObject)(*C),"Mat_MatMatTransMult",(PetscObject)container);CHKERRQ(ierr); 788 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 789 790 multtrans->usecoloring = PETSC_FALSE; 791 multtrans->destroy = (*C)->ops->destroy; 792 (*C)->ops->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 793 794 ierr = PetscOptionsGetBool(NULL,"-matmattransmult_color",&multtrans->usecoloring,NULL);CHKERRQ(ierr); 795 if (multtrans->usecoloring) { 796 /* Create MatTransposeColoring from symbolic C=A*B^T */ 797 MatTransposeColoring matcoloring; 798 ISColoring iscoloring; 799 Mat Bt_dense,C_dense; 800 801 ierr = MatGetColoring(*C,MATCOLORINGLF,&iscoloring);CHKERRQ(ierr); 802 ierr = MatTransposeColoringCreate(*C,iscoloring,&matcoloring);CHKERRQ(ierr); 803 804 multtrans->matcoloring = matcoloring; 805 806 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 807 808 /* Create Bt_dense and C_dense = A*Bt_dense */ 809 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 810 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 811 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 812 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 813 814 Bt_dense->assembled = PETSC_TRUE; 815 multtrans->Bt_den = Bt_dense; 816 817 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 818 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 819 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 820 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 821 822 Bt_dense->assembled = PETSC_TRUE; 823 multtrans->ABt_den = C_dense; 824 825 #if defined(PETSC_USE_INFO) 826 { 827 Mat_SeqAIJ *c = (Mat_SeqAIJ*)(*C)->data; 828 ierr = PetscInfo5(*C,"Bt_dense: %D,%D; Cnz %D / (cm*ncolors %D) = %g\n",A->cmap->n,matcoloring->ncolors,c->nz,A->rmap->n*matcoloring->ncolors,(PetscReal)(c->nz)/(A->rmap->n*matcoloring->ncolors));CHKERRQ(ierr); 829 } 830 #endif 831 } 832 /* clean up */ 833 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 834 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 835 836 837 838 #if defined(INEFFICIENT_ALGORITHM) 839 /* The algorithm below computes am*bm sparse inner-product - inefficient! It will be deleted later. */ 840 PetscFreeSpaceList free_space=NULL,current_space=NULL; 841 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 842 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*ci,*cj,*acol,*bcol; 843 PetscInt am =A->rmap->N,bm=B->rmap->N; 844 PetscInt i,j,anzi,bnzj,cnzi,nlnk,*lnk,nspacedouble=0,ka,kb,index[1]; 845 MatScalar *ca; 846 PetscBT lnkbt; 847 PetscReal afill; 848 849 /* Allocate row pointer array ci */ 850 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 851 ci[0] = 0; 852 853 /* Create and initialize a linked list for C columns */ 854 nlnk = bm+1; 855 ierr = PetscLLCreate(bm,bm,nlnk,lnk,lnkbt);CHKERRQ(ierr); 856 857 /* Initial FreeSpace with size fill*(nnz(A)+nnz(B)) */ 858 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 859 current_space = free_space; 860 861 /* Determine symbolic info for each row of the product A*B^T: */ 862 for (i=0; i<am; i++) { 863 anzi = ai[i+1] - ai[i]; 864 cnzi = 0; 865 acol = aj + ai[i]; 866 for (j=0; j<bm; j++) { 867 bnzj = bi[j+1] - bi[j]; 868 bcol = bj + bi[j]; 869 /* sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 870 ka = 0; kb = 0; 871 while (ka < anzi && kb < bnzj) { 872 while (acol[ka] < bcol[kb] && ka < anzi) ka++; 873 if (ka == anzi) break; 874 while (acol[ka] > bcol[kb] && kb < bnzj) kb++; 875 if (kb == bnzj) break; 876 if (acol[ka] == bcol[kb]) { /* add nonzero c(i,j) to lnk */ 877 index[0] = j; 878 ierr = PetscLLAdd(1,index,bm,nlnk,lnk,lnkbt);CHKERRQ(ierr); 879 cnzi++; 880 break; 881 } 882 } 883 } 884 885 /* If free space is not available, make more free space */ 886 /* Double the amount of total space in the list */ 887 if (current_space->local_remaining<cnzi) { 888 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 889 nspacedouble++; 890 } 891 892 /* Copy data into free space, then initialize lnk */ 893 ierr = PetscLLClean(bm,bm,cnzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 894 895 current_space->array += cnzi; 896 current_space->local_used += cnzi; 897 current_space->local_remaining -= cnzi; 898 899 ci[i+1] = ci[i] + cnzi; 900 } 901 902 903 /* Column indices are in the list of free space. 904 Allocate array cj, copy column indices to cj, and destroy list of free space */ 905 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 906 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 907 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 908 909 /* Allocate space for ca */ 910 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 911 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 912 913 /* put together the new symbolic matrix */ 914 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bm,ci,cj,ca,C);CHKERRQ(ierr); 915 916 (*C)->rmap->bs = A->cmap->bs; 917 (*C)->cmap->bs = B->cmap->bs; 918 919 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 920 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 921 c = (Mat_SeqAIJ*)((*C)->data); 922 c->free_a = PETSC_TRUE; 923 c->free_ij = PETSC_TRUE; 924 c->nonew = 0; 925 926 /* set MatInfo */ 927 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 928 if (afill < 1.0) afill = 1.0; 929 c->maxnz = ci[am]; 930 c->nz = ci[am]; 931 (*C)->info.mallocs = nspacedouble; 932 (*C)->info.fill_ratio_given = fill; 933 (*C)->info.fill_ratio_needed = afill; 934 935 #if defined(PETSC_USE_INFO) 936 if (ci[am]) { 937 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",nspacedouble,fill,afill);CHKERRQ(ierr); 938 ierr = PetscInfo1((*C),"Use MatMatTransposeMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 939 } else { 940 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 941 } 942 #endif 943 #endif 944 PetscFunctionReturn(0); 945 } 946 947 /* #define USE_ARRAY - for sparse dot product. Slower than !USE_ARRAY */ 948 #undef __FUNCT__ 949 #define __FUNCT__ "MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ" 950 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 951 { 952 PetscErrorCode ierr; 953 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 954 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 955 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 956 PetscLogDouble flops=0.0; 957 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 958 Mat_MatMatTransMult *multtrans; 959 PetscContainer container; 960 #if defined(USE_ARRAY) 961 MatScalar *spdot; 962 #endif 963 964 PetscFunctionBegin; 965 /* clear old values in C */ 966 if (!c->a) { 967 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 968 c->a = ca; 969 c->free_a = PETSC_TRUE; 970 } else { 971 ca = c->a; 972 } 973 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 974 975 ierr = PetscObjectQuery((PetscObject)C,"Mat_MatMatTransMult",(PetscObject*)&container);CHKERRQ(ierr); 976 if (!container) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 977 ierr = PetscContainerGetPointer(container,(void**)&multtrans);CHKERRQ(ierr); 978 if (multtrans->usecoloring) { 979 MatTransposeColoring matcoloring = multtrans->matcoloring; 980 Mat Bt_dense; 981 PetscInt m,n; 982 Mat C_dense = multtrans->ABt_den; 983 984 Bt_dense = multtrans->Bt_den; 985 ierr = MatGetLocalSize(Bt_dense,&m,&n);CHKERRQ(ierr); 986 987 /* Get Bt_dense by Apply MatTransposeColoring to B */ 988 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 989 990 /* C_dense = A*Bt_dense */ 991 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 992 993 /* Recover C from C_dense */ 994 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 995 PetscFunctionReturn(0); 996 } 997 998 #if defined(USE_ARRAY) 999 /* allocate an array for implementing sparse inner-product */ 1000 ierr = PetscMalloc((A->cmap->n+1)*sizeof(MatScalar),&spdot);CHKERRQ(ierr); 1001 ierr = PetscMemzero(spdot,(A->cmap->n+1)*sizeof(MatScalar));CHKERRQ(ierr); 1002 #endif 1003 1004 for (i=0; i<cm; i++) { 1005 anzi = ai[i+1] - ai[i]; 1006 acol = aj + ai[i]; 1007 aval = aa + ai[i]; 1008 cnzi = ci[i+1] - ci[i]; 1009 ccol = cj + ci[i]; 1010 cval = ca + ci[i]; 1011 for (j=0; j<cnzi; j++) { 1012 brow = ccol[j]; 1013 bnzj = bi[brow+1] - bi[brow]; 1014 bcol = bj + bi[brow]; 1015 bval = ba + bi[brow]; 1016 1017 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 1018 #if defined(USE_ARRAY) 1019 /* put ba to spdot array */ 1020 for (nextb=0; nextb<bnzj; nextb++) spdot[bcol[nextb]] = bval[nextb]; 1021 /* c(i,j)=A[i,:]*B[j,:]^T */ 1022 for (nexta=0; nexta<anzi; nexta++) { 1023 cval[j] += spdot[acol[nexta]]*aval[nexta]; 1024 } 1025 /* zero spdot array */ 1026 for (nextb=0; nextb<bnzj; nextb++) spdot[bcol[nextb]] = 0.0; 1027 #else 1028 nexta = 0; nextb = 0; 1029 while (nexta<anzi && nextb<bnzj) { 1030 while (acol[nexta] < bcol[nextb] && nexta < anzi) nexta++; 1031 if (nexta == anzi) break; 1032 while (acol[nexta] > bcol[nextb] && nextb < bnzj) nextb++; 1033 if (nextb == bnzj) break; 1034 if (acol[nexta] == bcol[nextb]) { 1035 cval[j] += aval[nexta]*bval[nextb]; 1036 nexta++; nextb++; 1037 flops += 2; 1038 } 1039 } 1040 #endif 1041 } 1042 } 1043 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1044 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1045 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1046 #if defined(USE_ARRAY) 1047 ierr = PetscFree(spdot); 1048 #endif 1049 PetscFunctionReturn(0); 1050 } 1051 1052 #undef __FUNCT__ 1053 #define __FUNCT__ "MatTransposeMatMult_SeqAIJ_SeqAIJ" 1054 PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1055 { 1056 PetscErrorCode ierr; 1057 1058 PetscFunctionBegin; 1059 if (scall == MAT_INITIAL_MATRIX) { 1060 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1061 } 1062 ierr = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 1063 PetscFunctionReturn(0); 1064 } 1065 1066 #undef __FUNCT__ 1067 #define __FUNCT__ "MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ" 1068 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 1069 { 1070 PetscErrorCode ierr; 1071 Mat At; 1072 PetscInt *ati,*atj; 1073 1074 PetscFunctionBegin; 1075 /* create symbolic At */ 1076 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1077 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1078 1079 At->rmap->bs = A->cmap->bs; 1080 At->cmap->bs = B->cmap->bs; 1081 1082 /* get symbolic C=At*B */ 1083 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(At,B,fill,C);CHKERRQ(ierr); 1084 1085 /* clean up */ 1086 ierr = MatDestroy(&At);CHKERRQ(ierr); 1087 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1088 PetscFunctionReturn(0); 1089 } 1090 1091 #undef __FUNCT__ 1092 #define __FUNCT__ "MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ" 1093 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1094 { 1095 PetscErrorCode ierr; 1096 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1097 PetscInt am =A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1098 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1099 PetscLogDouble flops=0.0; 1100 MatScalar *aa =a->a,*ba,*ca,*caj; 1101 1102 PetscFunctionBegin; 1103 if (!c->a) { 1104 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 1105 1106 c->a = ca; 1107 c->free_a = PETSC_TRUE; 1108 } else { 1109 ca = c->a; 1110 } 1111 /* clear old values in C */ 1112 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 1113 1114 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1115 for (i=0; i<am; i++) { 1116 bj = b->j + bi[i]; 1117 ba = b->a + bi[i]; 1118 bnzi = bi[i+1] - bi[i]; 1119 anzi = ai[i+1] - ai[i]; 1120 for (j=0; j<anzi; j++) { 1121 nextb = 0; 1122 crow = *aj++; 1123 cjj = cj + ci[crow]; 1124 caj = ca + ci[crow]; 1125 /* perform sparse axpy operation. Note cjj includes bj. */ 1126 for (k=0; nextb<bnzi; k++) { 1127 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1128 caj[k] += (*aa)*(*(ba+nextb)); 1129 nextb++; 1130 } 1131 } 1132 flops += 2*bnzi; 1133 aa++; 1134 } 1135 } 1136 1137 /* Assemble the final matrix and clean up */ 1138 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1139 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1140 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1141 PetscFunctionReturn(0); 1142 } 1143 1144 EXTERN_C_BEGIN 1145 #undef __FUNCT__ 1146 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqDense" 1147 PetscErrorCode MatMatMult_SeqAIJ_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1148 { 1149 PetscErrorCode ierr; 1150 1151 PetscFunctionBegin; 1152 if (scall == MAT_INITIAL_MATRIX) { 1153 ierr = MatMatMultSymbolic_SeqAIJ_SeqDense(A,B,fill,C);CHKERRQ(ierr); 1154 } 1155 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,B,*C);CHKERRQ(ierr); 1156 PetscFunctionReturn(0); 1157 } 1158 EXTERN_C_END 1159 1160 #undef __FUNCT__ 1161 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqDense" 1162 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C) 1163 { 1164 PetscErrorCode ierr; 1165 1166 PetscFunctionBegin; 1167 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1168 1169 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1170 PetscFunctionReturn(0); 1171 } 1172 1173 #undef __FUNCT__ 1174 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqDense" 1175 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1176 { 1177 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1178 PetscErrorCode ierr; 1179 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1180 MatScalar *aa; 1181 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=B->rmap->n, col, i,j,n,*aj, am = A->rmap->n; 1182 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam; 1183 1184 PetscFunctionBegin; 1185 if (!cm || !cn) PetscFunctionReturn(0); 1186 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); 1187 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); 1188 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); 1189 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1190 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1191 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1192 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1193 colam = col*am; 1194 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1195 r1 = r2 = r3 = r4 = 0.0; 1196 n = a->i[i+1] - a->i[i]; 1197 aj = a->j + a->i[i]; 1198 aa = a->a + a->i[i]; 1199 for (j=0; j<n; j++) { 1200 r1 += (*aa)*b1[*aj]; 1201 r2 += (*aa)*b2[*aj]; 1202 r3 += (*aa)*b3[*aj]; 1203 r4 += (*aa++)*b4[*aj++]; 1204 } 1205 c[colam + i] = r1; 1206 c[colam + am + i] = r2; 1207 c[colam + am2 + i] = r3; 1208 c[colam + am3 + i] = r4; 1209 } 1210 b1 += bm4; 1211 b2 += bm4; 1212 b3 += bm4; 1213 b4 += bm4; 1214 } 1215 for (; col<cn; col++) { /* over extra columns of C */ 1216 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1217 r1 = 0.0; 1218 n = a->i[i+1] - a->i[i]; 1219 aj = a->j + a->i[i]; 1220 aa = a->a + a->i[i]; 1221 1222 for (j=0; j<n; j++) { 1223 r1 += (*aa++)*b1[*aj++]; 1224 } 1225 c[col*am + i] = r1; 1226 } 1227 b1 += bm; 1228 } 1229 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1230 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1231 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1232 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1233 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1234 PetscFunctionReturn(0); 1235 } 1236 1237 /* 1238 Note very similar to MatMult_SeqAIJ(), should generate both codes from same base 1239 */ 1240 #undef __FUNCT__ 1241 #define __FUNCT__ "MatMatMultNumericAdd_SeqAIJ_SeqDense" 1242 PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1243 { 1244 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1245 PetscErrorCode ierr; 1246 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1247 MatScalar *aa; 1248 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=B->rmap->n, col, i,j,n,*aj, am = A->rmap->n,*ii,arm; 1249 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam,*ridx; 1250 1251 PetscFunctionBegin; 1252 if (!cm || !cn) PetscFunctionReturn(0); 1253 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1254 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1255 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1256 1257 if (a->compressedrow.use) { /* use compressed row format */ 1258 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1259 colam = col*am; 1260 arm = a->compressedrow.nrows; 1261 ii = a->compressedrow.i; 1262 ridx = a->compressedrow.rindex; 1263 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1264 r1 = r2 = r3 = r4 = 0.0; 1265 n = ii[i+1] - ii[i]; 1266 aj = a->j + ii[i]; 1267 aa = a->a + ii[i]; 1268 for (j=0; j<n; j++) { 1269 r1 += (*aa)*b1[*aj]; 1270 r2 += (*aa)*b2[*aj]; 1271 r3 += (*aa)*b3[*aj]; 1272 r4 += (*aa++)*b4[*aj++]; 1273 } 1274 c[colam + ridx[i]] += r1; 1275 c[colam + am + ridx[i]] += r2; 1276 c[colam + am2 + ridx[i]] += r3; 1277 c[colam + am3 + ridx[i]] += r4; 1278 } 1279 b1 += bm4; 1280 b2 += bm4; 1281 b3 += bm4; 1282 b4 += bm4; 1283 } 1284 for (; col<cn; col++) { /* over extra columns of C */ 1285 colam = col*am; 1286 arm = a->compressedrow.nrows; 1287 ii = a->compressedrow.i; 1288 ridx = a->compressedrow.rindex; 1289 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1290 r1 = 0.0; 1291 n = ii[i+1] - ii[i]; 1292 aj = a->j + ii[i]; 1293 aa = a->a + ii[i]; 1294 1295 for (j=0; j<n; j++) { 1296 r1 += (*aa++)*b1[*aj++]; 1297 } 1298 c[colam + ridx[i]] += r1; 1299 } 1300 b1 += bm; 1301 } 1302 } else { 1303 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1304 colam = col*am; 1305 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1306 r1 = r2 = r3 = r4 = 0.0; 1307 n = a->i[i+1] - a->i[i]; 1308 aj = a->j + a->i[i]; 1309 aa = a->a + a->i[i]; 1310 for (j=0; j<n; j++) { 1311 r1 += (*aa)*b1[*aj]; 1312 r2 += (*aa)*b2[*aj]; 1313 r3 += (*aa)*b3[*aj]; 1314 r4 += (*aa++)*b4[*aj++]; 1315 } 1316 c[colam + i] += r1; 1317 c[colam + am + i] += r2; 1318 c[colam + am2 + i] += r3; 1319 c[colam + am3 + i] += r4; 1320 } 1321 b1 += bm4; 1322 b2 += bm4; 1323 b3 += bm4; 1324 b4 += bm4; 1325 } 1326 for (; col<cn; col++) { /* over extra columns of C */ 1327 colam = col*am; 1328 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1329 r1 = 0.0; 1330 n = a->i[i+1] - a->i[i]; 1331 aj = a->j + a->i[i]; 1332 aa = a->a + a->i[i]; 1333 1334 for (j=0; j<n; j++) { 1335 r1 += (*aa++)*b1[*aj++]; 1336 } 1337 c[colam + i] += r1; 1338 } 1339 b1 += bm; 1340 } 1341 } 1342 ierr = PetscLogFlops(cn*2.0*a->nz);CHKERRQ(ierr); 1343 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1344 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1345 PetscFunctionReturn(0); 1346 } 1347 1348 #undef __FUNCT__ 1349 #define __FUNCT__ "MatTransColoringApplySpToDen_SeqAIJ" 1350 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1351 { 1352 PetscErrorCode ierr; 1353 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1354 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1355 PetscInt *bi = b->i,*bj=b->j; 1356 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1357 MatScalar *btval,*btval_den,*ba=b->a; 1358 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1359 1360 PetscFunctionBegin; 1361 btval_den=btdense->v; 1362 ierr = PetscMemzero(btval_den,(m*n)*sizeof(MatScalar));CHKERRQ(ierr); 1363 for (k=0; k<ncolors; k++) { 1364 ncolumns = coloring->ncolumns[k]; 1365 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1366 col = *(columns + colorforcol[k] + l); 1367 btcol = bj + bi[col]; 1368 btval = ba + bi[col]; 1369 anz = bi[col+1] - bi[col]; 1370 for (j=0; j<anz; j++) { 1371 brow = btcol[j]; 1372 btval_den[brow] = btval[j]; 1373 } 1374 } 1375 btval_den += m; 1376 } 1377 PetscFunctionReturn(0); 1378 } 1379 1380 #undef __FUNCT__ 1381 #define __FUNCT__ "MatTransColoringApplyDenToSp_SeqAIJ" 1382 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1383 { 1384 PetscErrorCode ierr; 1385 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1386 PetscInt k,l,*row,*idx,m,ncolors=matcoloring->ncolors,nrows; 1387 PetscScalar *ca_den,*cp_den,*ca=csp->a; 1388 PetscInt *rows=matcoloring->rows,*spidx=matcoloring->columnsforspidx,*colorforrow=matcoloring->colorforrow; 1389 1390 PetscFunctionBegin; 1391 ierr = MatGetLocalSize(Csp,&m,NULL);CHKERRQ(ierr); 1392 ierr = MatDenseGetArray(Cden,&ca_den);CHKERRQ(ierr); 1393 cp_den = ca_den; 1394 for (k=0; k<ncolors; k++) { 1395 nrows = matcoloring->nrows[k]; 1396 row = rows + colorforrow[k]; 1397 idx = spidx + colorforrow[k]; 1398 for (l=0; l<nrows; l++) { 1399 ca[idx[l]] = cp_den[row[l]]; 1400 } 1401 cp_den += m; 1402 } 1403 ierr = MatDenseRestoreArray(Cden,&ca_den);CHKERRQ(ierr); 1404 PetscFunctionReturn(0); 1405 } 1406 1407 /* 1408 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 1409 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 1410 spidx[], index of a->j, to be used for setting 'columnsforspidx' in MatTransposeColoringCreate_SeqAIJ(). 1411 */ 1412 #undef __FUNCT__ 1413 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ_Color" 1414 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1415 { 1416 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1417 PetscErrorCode ierr; 1418 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 1419 PetscInt nz = a->i[m],row,*jj,mr,col; 1420 PetscInt *cspidx; 1421 1422 PetscFunctionBegin; 1423 *nn = n; 1424 if (!ia) PetscFunctionReturn(0); 1425 if (symmetric) { 1426 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MatGetColumnIJ_SeqAIJ_Color() not supported for the case symmetric"); 1427 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 1428 } else { 1429 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&collengths);CHKERRQ(ierr); 1430 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1431 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&cia);CHKERRQ(ierr); 1432 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cja);CHKERRQ(ierr); 1433 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cspidx);CHKERRQ(ierr); 1434 jj = a->j; 1435 for (i=0; i<nz; i++) { 1436 collengths[jj[i]]++; 1437 } 1438 cia[0] = oshift; 1439 for (i=0; i<n; i++) { 1440 cia[i+1] = cia[i] + collengths[i]; 1441 } 1442 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1443 jj = a->j; 1444 for (row=0; row<m; row++) { 1445 mr = a->i[row+1] - a->i[row]; 1446 for (i=0; i<mr; i++) { 1447 col = *jj++; 1448 1449 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 1450 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 1451 } 1452 } 1453 ierr = PetscFree(collengths);CHKERRQ(ierr); 1454 *ia = cia; *ja = cja; 1455 *spidx = cspidx; 1456 } 1457 PetscFunctionReturn(0); 1458 } 1459 1460 #undef __FUNCT__ 1461 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ_Color" 1462 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1463 { 1464 PetscErrorCode ierr; 1465 1466 PetscFunctionBegin; 1467 if (!ia) PetscFunctionReturn(0); 1468 1469 ierr = PetscFree(*ia);CHKERRQ(ierr); 1470 ierr = PetscFree(*ja);CHKERRQ(ierr); 1471 ierr = PetscFree(*spidx);CHKERRQ(ierr); 1472 PetscFunctionReturn(0); 1473 } 1474 1475 #undef __FUNCT__ 1476 #define __FUNCT__ "MatTransposeColoringCreate_SeqAIJ" 1477 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1478 { 1479 PetscErrorCode ierr; 1480 PetscInt i,n,nrows,N,j,k,m,ncols,col,cm; 1481 const PetscInt *is,*ci,*cj,*row_idx; 1482 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1483 IS *isa; 1484 PetscBool flg1,flg2; 1485 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1486 PetscInt *colorforrow,*rows,*rows_i,*columnsforspidx,*columnsforspidx_i,*idxhit,*spidx; 1487 PetscInt *colorforcol,*columns,*columns_i; 1488 1489 PetscFunctionBegin; 1490 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1491 1492 /* this is ugly way to get blocksize but cannot call MatGetBlockSize() because AIJ can have bs > 1 */ 1493 ierr = PetscObjectTypeCompare((PetscObject)mat,MATSEQBAIJ,&flg1);CHKERRQ(ierr); 1494 ierr = PetscObjectTypeCompare((PetscObject)mat,MATMPIBAIJ,&flg2);CHKERRQ(ierr); 1495 if (flg1 || flg2) { 1496 ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr); 1497 } 1498 1499 N = mat->cmap->N/bs; 1500 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1501 c->N = mat->cmap->N/bs; 1502 c->m = mat->rmap->N/bs; 1503 c->rstart = 0; 1504 1505 c->ncolors = nis; 1506 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 1507 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 1508 ierr = PetscMalloc2(csp->nz+1,PetscInt,&rows,csp->nz+1,PetscInt,&columnsforspidx);CHKERRQ(ierr); 1509 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&colorforrow);CHKERRQ(ierr); 1510 1511 colorforrow[0] = 0; 1512 rows_i = rows; 1513 columnsforspidx_i = columnsforspidx; 1514 1515 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&colorforcol);CHKERRQ(ierr); 1516 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&columns);CHKERRQ(ierr); 1517 1518 colorforcol[0] = 0; 1519 columns_i = columns; 1520 1521 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); /* column-wise storage of mat */ 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++) { 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 columnsforspidx_i[nrows] = idxhit[j]; 1564 nrows++; 1565 } 1566 } 1567 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1568 rows_i += nrows; columnsforspidx_i += nrows; 1569 } 1570 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1571 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1572 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 1573 #if defined(PETSC_USE_DEBUG) 1574 if (csp->nz != colorforrow[nis]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %d != colorforrow[nis] %d",csp->nz,colorforrow[nis]); 1575 #endif 1576 1577 c->colorforrow = colorforrow; 1578 c->rows = rows; 1579 c->columnsforspidx = columnsforspidx; 1580 c->colorforcol = colorforcol; 1581 c->columns = columns; 1582 1583 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1584 PetscFunctionReturn(0); 1585 } 1586