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 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat,Mat,PetscReal,Mat*); 14 15 #undef __FUNCT__ 16 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqAIJ" 17 PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 18 { 19 PetscErrorCode ierr; 20 PetscBool scalable=PETSC_FALSE,scalable_fast=PETSC_FALSE,heap = PETSC_FALSE,btheap = PETSC_FALSE,llcondensed = PETSC_FALSE; 21 22 PetscFunctionBegin; 23 if (scall == MAT_INITIAL_MATRIX) { 24 ierr = PetscObjectOptionsBegin((PetscObject)A);CHKERRQ(ierr); 25 ierr = PetscOptionsBool("-matmatmult_scalable","Use a scalable but slower C=A*B","",scalable,&scalable,NULL);CHKERRQ(ierr); 26 ierr = PetscOptionsBool("-matmatmult_scalable_fast","Use a scalable but slower C=A*B","",scalable_fast,&scalable_fast,NULL);CHKERRQ(ierr); 27 ierr = PetscOptionsBool("-matmatmult_heap","Use heap implementation of symbolic factorization C=A*B","",heap,&heap,NULL);CHKERRQ(ierr); 28 ierr = PetscOptionsBool("-matmatmult_btheap","Use btheap implementation of symbolic factorization C=A*B","",btheap,&btheap,NULL);CHKERRQ(ierr); 29 ierr = PetscOptionsBool("-matmatmult_llcondensed","Use LLCondensed to for symbolic C=A*B","",llcondensed,&llcondensed,NULL);CHKERRQ(ierr); 30 ierr = PetscOptionsEnd();CHKERRQ(ierr); 31 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 32 if (scalable_fast) { 33 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(A,B,fill,C);CHKERRQ(ierr); 34 } else if (scalable) { 35 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(A,B,fill,C);CHKERRQ(ierr); 36 } else if (heap) { 37 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(A,B,fill,C);CHKERRQ(ierr); 38 } else if (btheap) { 39 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(A,B,fill,C);CHKERRQ(ierr); 40 } else if (llcondensed) { 41 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(A,B,fill,C);CHKERRQ(ierr); 42 } else { 43 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 44 } 45 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 46 } 47 48 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 49 ierr = (*(*C)->ops->matmultnumeric)(A,B,*C);CHKERRQ(ierr); 50 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 51 PetscFunctionReturn(0); 52 } 53 54 #undef __FUNCT__ 55 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed" 56 static PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat A,Mat B,PetscReal fill,Mat *C) 57 { 58 PetscErrorCode ierr; 59 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 60 PetscInt *ai=a->i,*bi=b->i,*ci,*cj; 61 PetscInt am =A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 62 PetscReal afill; 63 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 64 PetscBT lnkbt; 65 PetscFreeSpaceList free_space=NULL,current_space=NULL; 66 67 PetscFunctionBegin; 68 /* Get ci and cj */ 69 /*---------------*/ 70 /* Allocate ci array, arrays for fill computation and */ 71 /* free space for accumulating nonzero column info */ 72 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 73 ci[0] = 0; 74 75 /* create and initialize a linked list */ 76 nlnk_max = a->rmax*b->rmax; 77 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; 78 ierr = PetscLLCondensedCreate(nlnk_max,bn,&lnk,&lnkbt);CHKERRQ(ierr); 79 80 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 81 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 82 83 current_space = free_space; 84 85 /* Determine ci and cj */ 86 for (i=0; i<am; i++) { 87 anzi = ai[i+1] - ai[i]; 88 aj = a->j + ai[i]; 89 for (j=0; j<anzi; j++) { 90 brow = aj[j]; 91 bnzj = bi[brow+1] - bi[brow]; 92 bj = b->j + bi[brow]; 93 /* add non-zero cols of B into the sorted linked list lnk */ 94 ierr = PetscLLCondensedAddSorted(bnzj,bj,lnk,lnkbt);CHKERRQ(ierr); 95 } 96 cnzi = lnk[0]; 97 98 /* If free space is not available, make more free space */ 99 /* Double the amount of total space in the list */ 100 if (current_space->local_remaining<cnzi) { 101 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 102 ndouble++; 103 } 104 105 /* Copy data into free space, then initialize lnk */ 106 ierr = PetscLLCondensedClean(bn,cnzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 107 108 current_space->array += cnzi; 109 current_space->local_used += cnzi; 110 current_space->local_remaining -= cnzi; 111 112 ci[i+1] = ci[i] + cnzi; 113 } 114 115 /* Column indices are in the list of free space */ 116 /* Allocate space for cj, initialize cj, and */ 117 /* destroy list of free space and other temporary array(s) */ 118 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 119 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 120 ierr = PetscLLCondensedDestroy(lnk,lnkbt);CHKERRQ(ierr); 121 122 /* put together the new symbolic matrix */ 123 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 124 125 (*C)->rmap->bs = A->rmap->bs; 126 (*C)->cmap->bs = B->cmap->bs; 127 128 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 129 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 130 c = (Mat_SeqAIJ*)((*C)->data); 131 c->free_a = PETSC_FALSE; 132 c->free_ij = PETSC_TRUE; 133 c->nonew = 0; 134 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; /* fast, needs non-scalable O(bn) array 'abdense' */ 135 136 /* set MatInfo */ 137 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 138 if (afill < 1.0) afill = 1.0; 139 c->maxnz = ci[am]; 140 c->nz = ci[am]; 141 (*C)->info.mallocs = ndouble; 142 (*C)->info.fill_ratio_given = fill; 143 (*C)->info.fill_ratio_needed = afill; 144 145 #if defined(PETSC_USE_INFO) 146 if (ci[am]) { 147 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 148 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 149 } else { 150 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 151 } 152 #endif 153 PetscFunctionReturn(0); 154 } 155 156 #undef __FUNCT__ 157 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ" 158 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 159 { 160 PetscErrorCode ierr; 161 PetscLogDouble flops=0.0; 162 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 163 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 164 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 165 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 166 PetscInt am =A->rmap->n,cm=C->rmap->n; 167 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 168 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca,valtmp; 169 PetscScalar *ab_dense; 170 171 PetscFunctionBegin; 172 if (!c->a) { /* first call of MatMatMultNumeric_SeqAIJ_SeqAIJ, allocate ca and matmult_abdense */ 173 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 174 c->a = ca; 175 c->free_a = PETSC_TRUE; 176 177 ierr = PetscMalloc(B->cmap->N*sizeof(PetscScalar),&ab_dense);CHKERRQ(ierr); 178 ierr = PetscMemzero(ab_dense,B->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 179 180 c->matmult_abdense = ab_dense; 181 } else { 182 ca = c->a; 183 ab_dense = c->matmult_abdense; 184 } 185 186 /* clean old values in C */ 187 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 188 /* Traverse A row-wise. */ 189 /* Build the ith row in C by summing over nonzero columns in A, */ 190 /* the rows of B corresponding to nonzeros of A. */ 191 for (i=0; i<am; i++) { 192 anzi = ai[i+1] - ai[i]; 193 for (j=0; j<anzi; j++) { 194 brow = aj[j]; 195 bnzi = bi[brow+1] - bi[brow]; 196 bjj = bj + bi[brow]; 197 baj = ba + bi[brow]; 198 /* perform dense axpy */ 199 valtmp = aa[j]; 200 for (k=0; k<bnzi; k++) { 201 ab_dense[bjj[k]] += valtmp*baj[k]; 202 } 203 flops += 2*bnzi; 204 } 205 aj += anzi; aa += anzi; 206 207 cnzi = ci[i+1] - ci[i]; 208 for (k=0; k<cnzi; k++) { 209 ca[k] += ab_dense[cj[k]]; 210 ab_dense[cj[k]] = 0.0; /* zero ab_dense */ 211 } 212 flops += cnzi; 213 cj += cnzi; ca += cnzi; 214 } 215 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 216 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 217 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 218 PetscFunctionReturn(0); 219 } 220 221 #undef __FUNCT__ 222 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable" 223 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,Mat C) 224 { 225 PetscErrorCode ierr; 226 PetscLogDouble flops=0.0; 227 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 228 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 229 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 230 PetscInt *ai = a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bjj,*ci=c->i,*cj=c->j; 231 PetscInt am = A->rmap->N,cm=C->rmap->N; 232 PetscInt i,j,k,anzi,bnzi,cnzi,brow; 233 PetscScalar *aa=a->a,*ba=b->a,*baj,*ca=c->a,valtmp; 234 PetscInt nextb; 235 236 PetscFunctionBegin; 237 /* clean old values in C */ 238 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 239 /* Traverse A row-wise. */ 240 /* Build the ith row in C by summing over nonzero columns in A, */ 241 /* the rows of B corresponding to nonzeros of A. */ 242 for (i=0; i<am; i++) { 243 anzi = ai[i+1] - ai[i]; 244 cnzi = ci[i+1] - ci[i]; 245 for (j=0; j<anzi; j++) { 246 brow = aj[j]; 247 bnzi = bi[brow+1] - bi[brow]; 248 bjj = bj + bi[brow]; 249 baj = ba + bi[brow]; 250 /* perform sparse axpy */ 251 valtmp = aa[j]; 252 nextb = 0; 253 for (k=0; nextb<bnzi; k++) { 254 if (cj[k] == bjj[nextb]) { /* ccol == bcol */ 255 ca[k] += valtmp*baj[nextb++]; 256 } 257 } 258 flops += 2*bnzi; 259 } 260 aj += anzi; aa += anzi; 261 cj += cnzi; ca += cnzi; 262 } 263 264 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 265 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 266 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 267 PetscFunctionReturn(0); 268 } 269 270 #undef __FUNCT__ 271 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast" 272 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat A,Mat B,PetscReal fill,Mat *C) 273 { 274 PetscErrorCode ierr; 275 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 276 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 277 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 278 MatScalar *ca; 279 PetscReal afill; 280 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 281 PetscFreeSpaceList free_space=NULL,current_space=NULL; 282 283 PetscFunctionBegin; 284 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_fast() */ 285 /*-----------------------------------------------------------------------------------------*/ 286 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 287 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 288 ci[0] = 0; 289 290 /* create and initialize a linked list */ 291 nlnk_max = a->rmax*b->rmax; 292 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 293 ierr = PetscLLCondensedCreate_fast(nlnk_max,&lnk);CHKERRQ(ierr); 294 295 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 296 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 297 current_space = free_space; 298 299 /* Determine ci and cj */ 300 for (i=0; i<am; i++) { 301 anzi = ai[i+1] - ai[i]; 302 aj = a->j + ai[i]; 303 for (j=0; j<anzi; j++) { 304 brow = aj[j]; 305 bnzj = bi[brow+1] - bi[brow]; 306 bj = b->j + bi[brow]; 307 /* add non-zero cols of B into the sorted linked list lnk */ 308 ierr = PetscLLCondensedAddSorted_fast(bnzj,bj,lnk);CHKERRQ(ierr); 309 } 310 cnzi = lnk[1]; 311 312 /* If free space is not available, make more free space */ 313 /* Double the amount of total space in the list */ 314 if (current_space->local_remaining<cnzi) { 315 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 316 ndouble++; 317 } 318 319 /* Copy data into free space, then initialize lnk */ 320 ierr = PetscLLCondensedClean_fast(cnzi,current_space->array,lnk);CHKERRQ(ierr); 321 322 current_space->array += cnzi; 323 current_space->local_used += cnzi; 324 current_space->local_remaining -= cnzi; 325 326 ci[i+1] = ci[i] + cnzi; 327 } 328 329 /* Column indices are in the list of free space */ 330 /* Allocate space for cj, initialize cj, and */ 331 /* destroy list of free space and other temporary array(s) */ 332 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 333 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 334 ierr = PetscLLCondensedDestroy_fast(lnk);CHKERRQ(ierr); 335 336 /* Allocate space for ca */ 337 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 338 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 339 340 /* put together the new symbolic matrix */ 341 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 342 343 (*C)->rmap->bs = A->rmap->bs; 344 (*C)->cmap->bs = B->cmap->bs; 345 346 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 347 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 348 c = (Mat_SeqAIJ*)((*C)->data); 349 c->free_a = PETSC_TRUE; 350 c->free_ij = PETSC_TRUE; 351 c->nonew = 0; 352 353 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 354 355 /* set MatInfo */ 356 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 357 if (afill < 1.0) afill = 1.0; 358 c->maxnz = ci[am]; 359 c->nz = ci[am]; 360 (*C)->info.mallocs = ndouble; 361 (*C)->info.fill_ratio_given = fill; 362 (*C)->info.fill_ratio_needed = afill; 363 364 #if defined(PETSC_USE_INFO) 365 if (ci[am]) { 366 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 367 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 368 } else { 369 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 370 } 371 #endif 372 PetscFunctionReturn(0); 373 } 374 375 376 #undef __FUNCT__ 377 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable" 378 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat A,Mat B,PetscReal fill,Mat *C) 379 { 380 PetscErrorCode ierr; 381 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 382 PetscInt *ai = a->i,*bi=b->i,*ci,*cj; 383 PetscInt am = A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 384 MatScalar *ca; 385 PetscReal afill; 386 PetscInt i,j,anzi,brow,bnzj,cnzi,*bj,*aj,nlnk_max,*lnk,ndouble=0; 387 PetscFreeSpaceList free_space=NULL,current_space=NULL; 388 389 PetscFunctionBegin; 390 /* Get ci and cj - same as MatMatMultSymbolic_SeqAIJ_SeqAIJ except using PetscLLxxx_Scalalbe() */ 391 /*---------------------------------------------------------------------------------------------*/ 392 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 393 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 394 ci[0] = 0; 395 396 /* create and initialize a linked list */ 397 nlnk_max = a->rmax*b->rmax; 398 if (!nlnk_max || nlnk_max > bn) nlnk_max = bn; /* in case rmax is not defined for A or B */ 399 ierr = PetscLLCondensedCreate_Scalable(nlnk_max,&lnk);CHKERRQ(ierr); 400 401 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 402 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 403 current_space = free_space; 404 405 /* Determine ci and cj */ 406 for (i=0; i<am; i++) { 407 anzi = ai[i+1] - ai[i]; 408 aj = a->j + ai[i]; 409 for (j=0; j<anzi; j++) { 410 brow = aj[j]; 411 bnzj = bi[brow+1] - bi[brow]; 412 bj = b->j + bi[brow]; 413 /* add non-zero cols of B into the sorted linked list lnk */ 414 ierr = PetscLLCondensedAddSorted_Scalable(bnzj,bj,lnk);CHKERRQ(ierr); 415 } 416 cnzi = lnk[0]; 417 418 /* If free space is not available, make more free space */ 419 /* Double the amount of total space in the list */ 420 if (current_space->local_remaining<cnzi) { 421 ierr = PetscFreeSpaceGet(cnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 422 ndouble++; 423 } 424 425 /* Copy data into free space, then initialize lnk */ 426 ierr = PetscLLCondensedClean_Scalable(cnzi,current_space->array,lnk);CHKERRQ(ierr); 427 428 current_space->array += cnzi; 429 current_space->local_used += cnzi; 430 current_space->local_remaining -= cnzi; 431 432 ci[i+1] = ci[i] + cnzi; 433 } 434 435 /* Column indices are in the list of free space */ 436 /* Allocate space for cj, initialize cj, and */ 437 /* destroy list of free space and other temporary array(s) */ 438 ierr = PetscMalloc((ci[am]+1)*sizeof(PetscInt),&cj);CHKERRQ(ierr); 439 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 440 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 441 442 /* Allocate space for ca */ 443 /*-----------------------*/ 444 ierr = PetscMalloc((ci[am]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 445 ierr = PetscMemzero(ca,(ci[am]+1)*sizeof(MatScalar));CHKERRQ(ierr); 446 447 /* put together the new symbolic matrix */ 448 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,ca,C);CHKERRQ(ierr); 449 450 (*C)->rmap->bs = A->rmap->bs; 451 (*C)->cmap->bs = B->cmap->bs; 452 453 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 454 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 455 c = (Mat_SeqAIJ*)((*C)->data); 456 c->free_a = PETSC_TRUE; 457 c->free_ij = PETSC_TRUE; 458 c->nonew = 0; 459 460 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable; /* slower, less memory */ 461 462 /* set MatInfo */ 463 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 464 if (afill < 1.0) afill = 1.0; 465 c->maxnz = ci[am]; 466 c->nz = ci[am]; 467 (*C)->info.mallocs = ndouble; 468 (*C)->info.fill_ratio_given = fill; 469 (*C)->info.fill_ratio_needed = afill; 470 471 #if defined(PETSC_USE_INFO) 472 if (ci[am]) { 473 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 474 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 475 } else { 476 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 477 } 478 #endif 479 PetscFunctionReturn(0); 480 } 481 482 #undef __FUNCT__ 483 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap" 484 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat A,Mat B,PetscReal fill,Mat *C) 485 { 486 PetscErrorCode ierr; 487 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 488 const PetscInt *ai=a->i,*bi=b->i,*aj=a->j,*bj=b->j; 489 PetscInt *ci,*cj,*bb; 490 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 491 PetscReal afill; 492 PetscInt i,j,col,ndouble = 0; 493 PetscFreeSpaceList free_space=NULL,current_space=NULL; 494 PetscHeap h; 495 496 PetscFunctionBegin; 497 /* Get ci and cj - by merging sorted rows using a heap */ 498 /*---------------------------------------------------------------------------------------------*/ 499 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 500 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 501 ci[0] = 0; 502 503 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 504 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 505 current_space = free_space; 506 507 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 508 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 509 510 /* Determine ci and cj */ 511 for (i=0; i<am; i++) { 512 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 */ 513 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 514 ci[i+1] = ci[i]; 515 /* Populate the min heap */ 516 for (j=0; j<anzi; j++) { 517 bb[j] = bi[acol[j]]; /* bb points at the start of the row */ 518 if (bb[j] < bi[acol[j]+1]) { /* Add if row is nonempty */ 519 ierr = PetscHeapAdd(h,j,bj[bb[j]++]);CHKERRQ(ierr); 520 } 521 } 522 /* Pick off the min element, adding it to free space */ 523 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 524 while (j >= 0) { 525 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 526 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 527 ndouble++; 528 } 529 *(current_space->array++) = col; 530 current_space->local_used++; 531 current_space->local_remaining--; 532 ci[i+1]++; 533 534 /* stash if anything else remains in this row of B */ 535 if (bb[j] < bi[acol[j]+1]) {ierr = PetscHeapStash(h,j,bj[bb[j]++]);CHKERRQ(ierr);} 536 while (1) { /* pop and stash any other rows of B that also had an entry in this column */ 537 PetscInt j2,col2; 538 ierr = PetscHeapPeek(h,&j2,&col2);CHKERRQ(ierr); 539 if (col2 != col) break; 540 ierr = PetscHeapPop(h,&j2,&col2);CHKERRQ(ierr); 541 if (bb[j2] < bi[acol[j2]+1]) {ierr = PetscHeapStash(h,j2,bj[bb[j2]++]);CHKERRQ(ierr);} 542 } 543 /* Put any stashed elements back into the min heap */ 544 ierr = PetscHeapUnstash(h);CHKERRQ(ierr); 545 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 546 } 547 } 548 ierr = PetscFree(bb);CHKERRQ(ierr); 549 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 550 551 /* Column indices are in the list of free space */ 552 /* Allocate space for cj, initialize cj, and */ 553 /* destroy list of free space and other temporary array(s) */ 554 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 555 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 556 557 /* put together the new symbolic matrix */ 558 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 559 560 (*C)->rmap->bs = A->rmap->bs; 561 (*C)->cmap->bs = B->cmap->bs; 562 563 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 564 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 565 c = (Mat_SeqAIJ*)((*C)->data); 566 c->free_a = PETSC_TRUE; 567 c->free_ij = PETSC_TRUE; 568 c->nonew = 0; 569 570 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 571 572 /* set MatInfo */ 573 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 574 if (afill < 1.0) afill = 1.0; 575 c->maxnz = ci[am]; 576 c->nz = ci[am]; 577 (*C)->info.mallocs = ndouble; 578 (*C)->info.fill_ratio_given = fill; 579 (*C)->info.fill_ratio_needed = afill; 580 581 #if defined(PETSC_USE_INFO) 582 if (ci[am]) { 583 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 584 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 585 } else { 586 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 587 } 588 #endif 589 PetscFunctionReturn(0); 590 } 591 592 #undef __FUNCT__ 593 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap" 594 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat A,Mat B,PetscReal fill,Mat *C) 595 { 596 PetscErrorCode ierr; 597 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 598 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 599 PetscInt *ci,*cj,*bb; 600 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 601 PetscReal afill; 602 PetscInt i,j,col,ndouble = 0; 603 PetscFreeSpaceList free_space=NULL,current_space=NULL; 604 PetscHeap h; 605 PetscBT bt; 606 607 PetscFunctionBegin; 608 /* Get ci and cj - using a heap for the sorted rows, but use BT so that each index is only added once */ 609 /*---------------------------------------------------------------------------------------------*/ 610 /* Allocate arrays for fill computation and free space for accumulating nonzero column */ 611 ierr = PetscMalloc(((am+1)+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 612 ci[0] = 0; 613 614 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 615 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+bi[bm])),&free_space);CHKERRQ(ierr); 616 617 current_space = free_space; 618 619 ierr = PetscHeapCreate(a->rmax,&h);CHKERRQ(ierr); 620 ierr = PetscMalloc(a->rmax*sizeof(PetscInt),&bb);CHKERRQ(ierr); 621 ierr = PetscBTCreate(bn,&bt);CHKERRQ(ierr); 622 623 /* Determine ci and cj */ 624 for (i=0; i<am; i++) { 625 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 */ 626 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 627 const PetscInt *fptr = current_space->array; /* Save beginning of the row so we can clear the BT later */ 628 ci[i+1] = ci[i]; 629 /* Populate the min heap */ 630 for (j=0; j<anzi; j++) { 631 PetscInt brow = acol[j]; 632 for (bb[j] = bi[brow]; bb[j] < bi[brow+1]; bb[j]++) { 633 PetscInt bcol = bj[bb[j]]; 634 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 635 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 636 bb[j]++; 637 break; 638 } 639 } 640 } 641 /* Pick off the min element, adding it to free space */ 642 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 643 while (j >= 0) { 644 if (current_space->local_remaining < 1) { /* double the size, but don't exceed 16 MiB */ 645 fptr = NULL; /* need PetscBTMemzero */ 646 ierr = PetscFreeSpaceGet(PetscMin(2*current_space->total_array_size,16 << 20),¤t_space);CHKERRQ(ierr); 647 ndouble++; 648 } 649 *(current_space->array++) = col; 650 current_space->local_used++; 651 current_space->local_remaining--; 652 ci[i+1]++; 653 654 /* stash if anything else remains in this row of B */ 655 for (; bb[j] < bi[acol[j]+1]; bb[j]++) { 656 PetscInt bcol = bj[bb[j]]; 657 if (!PetscBTLookupSet(bt,bcol)) { /* new entry */ 658 ierr = PetscHeapAdd(h,j,bcol);CHKERRQ(ierr); 659 bb[j]++; 660 break; 661 } 662 } 663 ierr = PetscHeapPop(h,&j,&col);CHKERRQ(ierr); 664 } 665 if (fptr) { /* Clear the bits for this row */ 666 for (; fptr<current_space->array; fptr++) {ierr = PetscBTClear(bt,*fptr);CHKERRQ(ierr);} 667 } else { /* We reallocated so we don't remember (easily) how to clear only the bits we changed */ 668 ierr = PetscBTMemzero(bn,bt);CHKERRQ(ierr); 669 } 670 } 671 ierr = PetscFree(bb);CHKERRQ(ierr); 672 ierr = PetscHeapDestroy(&h);CHKERRQ(ierr); 673 ierr = PetscBTDestroy(&bt);CHKERRQ(ierr); 674 675 /* Column indices are in the list of free space */ 676 /* Allocate space for cj, initialize cj, and */ 677 /* destroy list of free space and other temporary array(s) */ 678 ierr = PetscMalloc(ci[am]*sizeof(PetscInt),&cj);CHKERRQ(ierr); 679 ierr = PetscFreeSpaceContiguous(&free_space,cj);CHKERRQ(ierr); 680 681 /* put together the new symbolic matrix */ 682 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 683 684 (*C)->rmap->bs = A->rmap->bs; 685 (*C)->cmap->bs = B->cmap->bs; 686 687 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 688 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 689 c = (Mat_SeqAIJ*)((*C)->data); 690 c->free_a = PETSC_TRUE; 691 c->free_ij = PETSC_TRUE; 692 c->nonew = 0; 693 694 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 695 696 /* set MatInfo */ 697 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 698 if (afill < 1.0) afill = 1.0; 699 c->maxnz = ci[am]; 700 c->nz = ci[am]; 701 (*C)->info.mallocs = ndouble; 702 (*C)->info.fill_ratio_given = fill; 703 (*C)->info.fill_ratio_needed = afill; 704 705 #if defined(PETSC_USE_INFO) 706 if (ci[am]) { 707 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 708 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 709 } else { 710 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 711 } 712 #endif 713 PetscFunctionReturn(0); 714 } 715 716 #undef __FUNCT__ 717 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqAIJ" 718 /* concatenate unique entries and then sort */ 719 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 720 { 721 PetscErrorCode ierr; 722 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c; 723 const PetscInt *ai = a->i,*bi=b->i,*aj=a->j,*bj=b->j; 724 PetscInt *ci,*cj; 725 PetscInt am=A->rmap->N,bn=B->cmap->N,bm=B->rmap->N; 726 PetscReal afill; 727 PetscInt i,j,ndouble = 0; 728 PetscSegBuffer seg,segrow; 729 char *seen; 730 731 PetscFunctionBegin; 732 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 733 ci[0] = 0; 734 735 /* Initial FreeSpace size is fill*(nnz(A)+nnz(B)) */ 736 ierr = PetscSegBufferCreate(sizeof(PetscInt),(PetscInt)(fill*(ai[am]+bi[bm])),&seg);CHKERRQ(ierr); 737 ierr = PetscSegBufferCreate(sizeof(PetscInt),100,&segrow);CHKERRQ(ierr); 738 ierr = PetscMalloc(bn*sizeof(char),&seen);CHKERRQ(ierr); 739 ierr = PetscMemzero(seen,bn*sizeof(char));CHKERRQ(ierr); 740 741 /* Determine ci and cj */ 742 for (i=0; i<am; i++) { 743 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 */ 744 const PetscInt *acol = aj + ai[i]; /* column indices of nonzero entries in this row */ 745 PetscInt packlen = 0,*PETSC_RESTRICT crow; 746 /* Pack segrow */ 747 for (j=0; j<anzi; j++) { 748 PetscInt brow = acol[j],bjstart = bi[brow],bjend = bi[brow+1],k; 749 for (k=bjstart; k<bjend; k++) { 750 PetscInt bcol = bj[k]; 751 if (!seen[bcol]) { /* new entry */ 752 PetscInt *PETSC_RESTRICT slot; 753 ierr = PetscSegBufferGetInts(segrow,1,&slot);CHKERRQ(ierr); 754 *slot = bcol; 755 seen[bcol] = 1; 756 packlen++; 757 } 758 } 759 } 760 ierr = PetscSegBufferGetInts(seg,packlen,&crow);CHKERRQ(ierr); 761 ierr = PetscSegBufferExtractTo(segrow,crow);CHKERRQ(ierr); 762 ierr = PetscSortInt(packlen,crow);CHKERRQ(ierr); 763 ci[i+1] = ci[i] + packlen; 764 for (j=0; j<packlen; j++) seen[crow[j]] = 0; 765 } 766 ierr = PetscSegBufferDestroy(&segrow);CHKERRQ(ierr); 767 ierr = PetscFree(seen);CHKERRQ(ierr); 768 769 /* Column indices are in the segmented buffer */ 770 ierr = PetscSegBufferExtractAlloc(seg,&cj);CHKERRQ(ierr); 771 ierr = PetscSegBufferDestroy(&seg);CHKERRQ(ierr); 772 773 /* put together the new symbolic matrix */ 774 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),am,bn,ci,cj,NULL,C);CHKERRQ(ierr); 775 776 (*C)->rmap->bs = A->rmap->bs; 777 (*C)->cmap->bs = B->cmap->bs; 778 779 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 780 /* These are PETSc arrays, so change flags so arrays can be deleted by PETSc */ 781 c = (Mat_SeqAIJ*)((*C)->data); 782 c->free_a = PETSC_TRUE; 783 c->free_ij = PETSC_TRUE; 784 c->nonew = 0; 785 786 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqAIJ; 787 788 /* set MatInfo */ 789 afill = (PetscReal)ci[am]/(ai[am]+bi[bm]) + 1.e-5; 790 if (afill < 1.0) afill = 1.0; 791 c->maxnz = ci[am]; 792 c->nz = ci[am]; 793 (*C)->info.mallocs = ndouble; 794 (*C)->info.fill_ratio_given = fill; 795 (*C)->info.fill_ratio_needed = afill; 796 797 #if defined(PETSC_USE_INFO) 798 if (ci[am]) { 799 ierr = PetscInfo3((*C),"Reallocs %D; Fill ratio: given %G needed %G.\n",ndouble,fill,afill);CHKERRQ(ierr); 800 ierr = PetscInfo1((*C),"Use MatMatMult(A,B,MatReuse,%G,&C) for best performance.;\n",afill);CHKERRQ(ierr); 801 } else { 802 ierr = PetscInfo((*C),"Empty matrix product\n");CHKERRQ(ierr); 803 } 804 #endif 805 PetscFunctionReturn(0); 806 } 807 808 /* This routine is not used. Should be removed! */ 809 #undef __FUNCT__ 810 #define __FUNCT__ "MatMatTransposeMult_SeqAIJ_SeqAIJ" 811 PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 812 { 813 PetscErrorCode ierr; 814 815 PetscFunctionBegin; 816 if (scall == MAT_INITIAL_MATRIX) { 817 ierr = PetscLogEventBegin(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 818 ierr = MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 819 ierr = PetscLogEventEnd(MAT_MatTransposeMultSymbolic,A,B,0,0);CHKERRQ(ierr); 820 } 821 ierr = PetscLogEventBegin(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 822 ierr = MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 823 ierr = PetscLogEventEnd(MAT_MatTransposeMultNumeric,A,B,0,0);CHKERRQ(ierr); 824 PetscFunctionReturn(0); 825 } 826 827 #undef __FUNCT__ 828 #define __FUNCT__ "PetscContainerDestroy_Mat_MatMatTransMult" 829 PetscErrorCode PetscContainerDestroy_Mat_MatMatTransMult(void *ptr) 830 { 831 PetscErrorCode ierr; 832 Mat_MatMatTransMult *multtrans=(Mat_MatMatTransMult*)ptr; 833 834 PetscFunctionBegin; 835 ierr = MatTransposeColoringDestroy(&multtrans->matcoloring);CHKERRQ(ierr); 836 ierr = MatDestroy(&multtrans->Bt_den);CHKERRQ(ierr); 837 ierr = MatDestroy(&multtrans->ABt_den);CHKERRQ(ierr); 838 ierr = PetscFree(multtrans);CHKERRQ(ierr); 839 PetscFunctionReturn(0); 840 } 841 842 #undef __FUNCT__ 843 #define __FUNCT__ "MatDestroy_SeqAIJ_MatMatMultTrans" 844 PetscErrorCode MatDestroy_SeqAIJ_MatMatMultTrans(Mat A) 845 { 846 PetscErrorCode ierr; 847 PetscContainer container; 848 Mat_MatMatTransMult *multtrans=NULL; 849 850 PetscFunctionBegin; 851 ierr = PetscObjectQuery((PetscObject)A,"Mat_MatMatTransMult",(PetscObject*)&container);CHKERRQ(ierr); 852 if (!container) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 853 ierr = PetscContainerGetPointer(container,(void**)&multtrans);CHKERRQ(ierr); 854 855 A->ops->destroy = multtrans->destroy; 856 if (A->ops->destroy) { 857 ierr = (*A->ops->destroy)(A);CHKERRQ(ierr); 858 } 859 ierr = PetscObjectCompose((PetscObject)A,"Mat_MatMatTransMult",0);CHKERRQ(ierr); 860 PetscFunctionReturn(0); 861 } 862 863 #undef __FUNCT__ 864 #define __FUNCT__ "MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ" 865 PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 866 { 867 PetscErrorCode ierr; 868 Mat Bt; 869 PetscInt *bti,*btj; 870 Mat_MatMatTransMult *multtrans; 871 PetscContainer container; 872 873 PetscFunctionBegin; 874 /* create symbolic Bt */ 875 ierr = MatGetSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 876 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,B->cmap->n,B->rmap->n,bti,btj,NULL,&Bt);CHKERRQ(ierr); 877 878 Bt->rmap->bs = A->cmap->bs; 879 Bt->cmap->bs = B->cmap->bs; 880 881 /* get symbolic C=A*Bt */ 882 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(A,Bt,fill,C);CHKERRQ(ierr); 883 884 /* create a supporting struct for reuse intermidiate dense matrices with matcoloring */ 885 ierr = PetscNew(Mat_MatMatTransMult,&multtrans);CHKERRQ(ierr); 886 887 /* attach the supporting struct to C */ 888 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 889 ierr = PetscContainerSetPointer(container,multtrans);CHKERRQ(ierr); 890 ierr = PetscContainerSetUserDestroy(container,PetscContainerDestroy_Mat_MatMatTransMult);CHKERRQ(ierr); 891 ierr = PetscObjectCompose((PetscObject)(*C),"Mat_MatMatTransMult",(PetscObject)container);CHKERRQ(ierr); 892 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 893 894 multtrans->usecoloring = PETSC_FALSE; 895 multtrans->destroy = (*C)->ops->destroy; 896 (*C)->ops->destroy = MatDestroy_SeqAIJ_MatMatMultTrans; 897 898 ierr = PetscOptionsGetBool(NULL,"-matmattransmult_color",&multtrans->usecoloring,NULL);CHKERRQ(ierr); 899 if (multtrans->usecoloring) { 900 /* Create MatTransposeColoring from symbolic C=A*B^T */ 901 MatTransposeColoring matcoloring; 902 ISColoring iscoloring; 903 Mat Bt_dense,C_dense; 904 905 ierr = MatGetColoring(*C,MATCOLORINGLF,&iscoloring);CHKERRQ(ierr); 906 ierr = MatTransposeColoringCreate(*C,iscoloring,&matcoloring);CHKERRQ(ierr); 907 908 multtrans->matcoloring = matcoloring; 909 910 ierr = ISColoringDestroy(&iscoloring);CHKERRQ(ierr); 911 912 /* Create Bt_dense and C_dense = A*Bt_dense */ 913 ierr = MatCreate(PETSC_COMM_SELF,&Bt_dense);CHKERRQ(ierr); 914 ierr = MatSetSizes(Bt_dense,A->cmap->n,matcoloring->ncolors,A->cmap->n,matcoloring->ncolors);CHKERRQ(ierr); 915 ierr = MatSetType(Bt_dense,MATSEQDENSE);CHKERRQ(ierr); 916 ierr = MatSeqDenseSetPreallocation(Bt_dense,NULL);CHKERRQ(ierr); 917 918 Bt_dense->assembled = PETSC_TRUE; 919 multtrans->Bt_den = Bt_dense; 920 921 ierr = MatCreate(PETSC_COMM_SELF,&C_dense);CHKERRQ(ierr); 922 ierr = MatSetSizes(C_dense,A->rmap->n,matcoloring->ncolors,A->rmap->n,matcoloring->ncolors);CHKERRQ(ierr); 923 ierr = MatSetType(C_dense,MATSEQDENSE);CHKERRQ(ierr); 924 ierr = MatSeqDenseSetPreallocation(C_dense,NULL);CHKERRQ(ierr); 925 926 Bt_dense->assembled = PETSC_TRUE; 927 multtrans->ABt_den = C_dense; 928 929 #if defined(PETSC_USE_INFO) 930 { 931 Mat_SeqAIJ *c = (Mat_SeqAIJ*)(*C)->data; 932 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); 933 } 934 #endif 935 } 936 /* clean up */ 937 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 938 ierr = MatRestoreSymbolicTranspose_SeqAIJ(B,&bti,&btj);CHKERRQ(ierr); 939 PetscFunctionReturn(0); 940 } 941 942 /* #define USE_ARRAY - for sparse dot product. Slower than !USE_ARRAY */ 943 #undef __FUNCT__ 944 #define __FUNCT__ "MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ" 945 PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 946 { 947 PetscErrorCode ierr; 948 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 949 PetscInt *ai =a->i,*aj=a->j,*bi=b->i,*bj=b->j,anzi,bnzj,nexta,nextb,*acol,*bcol,brow; 950 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,i,j,cnzi,*ccol; 951 PetscLogDouble flops=0.0; 952 MatScalar *aa =a->a,*aval,*ba=b->a,*bval,*ca,*cval; 953 Mat_MatMatTransMult *multtrans; 954 PetscContainer container; 955 #if defined(USE_ARRAY) 956 MatScalar *spdot; 957 #endif 958 959 PetscFunctionBegin; 960 /* clear old values in C */ 961 if (!c->a) { 962 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 963 c->a = ca; 964 c->free_a = PETSC_TRUE; 965 } else { 966 ca = c->a; 967 } 968 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 969 970 ierr = PetscObjectQuery((PetscObject)C,"Mat_MatMatTransMult",(PetscObject*)&container);CHKERRQ(ierr); 971 if (!container) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 972 ierr = PetscContainerGetPointer(container,(void**)&multtrans);CHKERRQ(ierr); 973 if (multtrans->usecoloring) { 974 MatTransposeColoring matcoloring = multtrans->matcoloring; 975 Mat Bt_dense; 976 PetscInt m,n; 977 Mat C_dense = multtrans->ABt_den; 978 PetscLogDouble t0,t1,t2,t3,Bt_den,C_den,C_sp; 979 980 /* Get Bt_dense by Apply MatTransposeColoring to B */ 981 ierr = PetscGetTime(&t0);CHKERRQ(ierr); 982 Bt_dense = multtrans->Bt_den; 983 ierr = MatTransColoringApplySpToDen(matcoloring,B,Bt_dense);CHKERRQ(ierr); 984 ierr = PetscGetTime(&t1);CHKERRQ(ierr); 985 Bt_den = t1 - t0; 986 987 /* C_dense = A*Bt_dense */ 988 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,Bt_dense,C_dense);CHKERRQ(ierr); 989 ierr = PetscGetTime(&t2);CHKERRQ(ierr); 990 C_den = t2 - t1; 991 992 /* Recover C from C_dense */ 993 ierr = MatTransColoringApplyDenToSp(matcoloring,C_dense,C);CHKERRQ(ierr); 994 ierr = PetscGetTime(&t3);CHKERRQ(ierr); 995 C_sp = t3 - t2; 996 #if defined(PETSC_USE_INFO) 997 ierr = PetscInfo4(C,"Coloring A*B^T = Bt_den %g + C_den %g + C_sp %g = %g;\n",Bt_den,C_den,C_sp,Bt_den+C_den+C_sp);CHKERRQ(ierr); 998 ierr = MatGetLocalSize(Bt_dense,&m,&n);CHKERRQ(ierr); 999 ierr = PetscInfo4(C,"Bt_den: %d x %d, B: %d x %d\n",m,n,m,C->cmap->n);CHKERRQ(ierr); 1000 #endif 1001 PetscFunctionReturn(0); 1002 } 1003 1004 #if defined(USE_ARRAY) 1005 /* allocate an array for implementing sparse inner-product */ 1006 ierr = PetscMalloc((A->cmap->n+1)*sizeof(MatScalar),&spdot);CHKERRQ(ierr); 1007 ierr = PetscMemzero(spdot,(A->cmap->n+1)*sizeof(MatScalar));CHKERRQ(ierr); 1008 #endif 1009 1010 for (i=0; i<cm; i++) { 1011 anzi = ai[i+1] - ai[i]; 1012 acol = aj + ai[i]; 1013 aval = aa + ai[i]; 1014 cnzi = ci[i+1] - ci[i]; 1015 ccol = cj + ci[i]; 1016 cval = ca + ci[i]; 1017 for (j=0; j<cnzi; j++) { 1018 brow = ccol[j]; 1019 bnzj = bi[brow+1] - bi[brow]; 1020 bcol = bj + bi[brow]; 1021 bval = ba + bi[brow]; 1022 1023 /* perform sparse inner-product c(i,j)=A[i,:]*B[j,:]^T */ 1024 #if defined(USE_ARRAY) 1025 /* put ba to spdot array */ 1026 for (nextb=0; nextb<bnzj; nextb++) spdot[bcol[nextb]] = bval[nextb]; 1027 /* c(i,j)=A[i,:]*B[j,:]^T */ 1028 for (nexta=0; nexta<anzi; nexta++) { 1029 cval[j] += spdot[acol[nexta]]*aval[nexta]; 1030 } 1031 /* zero spdot array */ 1032 for (nextb=0; nextb<bnzj; nextb++) spdot[bcol[nextb]] = 0.0; 1033 #else 1034 nexta = 0; nextb = 0; 1035 while (nexta<anzi && nextb<bnzj) { 1036 while (nexta < anzi && acol[nexta] < bcol[nextb]) nexta++; 1037 if (nexta == anzi) break; 1038 while (nextb < bnzj && acol[nexta] > bcol[nextb]) nextb++; 1039 if (nextb == bnzj) break; 1040 if (acol[nexta] == bcol[nextb]) { 1041 cval[j] += aval[nexta]*bval[nextb]; 1042 nexta++; nextb++; 1043 flops += 2; 1044 } 1045 } 1046 #endif 1047 } 1048 } 1049 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1050 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1051 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1052 #if defined(USE_ARRAY) 1053 ierr = PetscFree(spdot); 1054 #endif 1055 PetscFunctionReturn(0); 1056 } 1057 1058 #undef __FUNCT__ 1059 #define __FUNCT__ "MatTransposeMatMult_SeqAIJ_SeqAIJ" 1060 PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1061 { 1062 PetscErrorCode ierr; 1063 1064 PetscFunctionBegin; 1065 if (scall == MAT_INITIAL_MATRIX) { 1066 ierr = MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 1067 } 1068 ierr = MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(A,B,*C);CHKERRQ(ierr); 1069 PetscFunctionReturn(0); 1070 } 1071 1072 #undef __FUNCT__ 1073 #define __FUNCT__ "MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ" 1074 PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 1075 { 1076 PetscErrorCode ierr; 1077 Mat At; 1078 PetscInt *ati,*atj; 1079 1080 PetscFunctionBegin; 1081 /* create symbolic At */ 1082 ierr = MatGetSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1083 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,A->cmap->n,A->rmap->n,ati,atj,NULL,&At);CHKERRQ(ierr); 1084 1085 At->rmap->bs = A->cmap->bs; 1086 At->cmap->bs = B->cmap->bs; 1087 1088 /* get symbolic C=At*B */ 1089 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(At,B,fill,C);CHKERRQ(ierr); 1090 1091 /* clean up */ 1092 ierr = MatDestroy(&At);CHKERRQ(ierr); 1093 ierr = MatRestoreSymbolicTranspose_SeqAIJ(A,&ati,&atj);CHKERRQ(ierr); 1094 PetscFunctionReturn(0); 1095 } 1096 1097 #undef __FUNCT__ 1098 #define __FUNCT__ "MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ" 1099 PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat A,Mat B,Mat C) 1100 { 1101 PetscErrorCode ierr; 1102 Mat_SeqAIJ *a =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data,*c=(Mat_SeqAIJ*)C->data; 1103 PetscInt am =A->rmap->n,anzi,*ai=a->i,*aj=a->j,*bi=b->i,*bj,bnzi,nextb; 1104 PetscInt cm =C->rmap->n,*ci=c->i,*cj=c->j,crow,*cjj,i,j,k; 1105 PetscLogDouble flops=0.0; 1106 MatScalar *aa =a->a,*ba,*ca,*caj; 1107 1108 PetscFunctionBegin; 1109 if (!c->a) { 1110 ierr = PetscMalloc((ci[cm]+1)*sizeof(MatScalar),&ca);CHKERRQ(ierr); 1111 1112 c->a = ca; 1113 c->free_a = PETSC_TRUE; 1114 } else { 1115 ca = c->a; 1116 } 1117 /* clear old values in C */ 1118 ierr = PetscMemzero(ca,ci[cm]*sizeof(MatScalar));CHKERRQ(ierr); 1119 1120 /* compute A^T*B using outer product (A^T)[:,i]*B[i,:] */ 1121 for (i=0; i<am; i++) { 1122 bj = b->j + bi[i]; 1123 ba = b->a + bi[i]; 1124 bnzi = bi[i+1] - bi[i]; 1125 anzi = ai[i+1] - ai[i]; 1126 for (j=0; j<anzi; j++) { 1127 nextb = 0; 1128 crow = *aj++; 1129 cjj = cj + ci[crow]; 1130 caj = ca + ci[crow]; 1131 /* perform sparse axpy operation. Note cjj includes bj. */ 1132 for (k=0; nextb<bnzi; k++) { 1133 if (cjj[k] == *(bj+nextb)) { /* ccol == bcol */ 1134 caj[k] += (*aa)*(*(ba+nextb)); 1135 nextb++; 1136 } 1137 } 1138 flops += 2*bnzi; 1139 aa++; 1140 } 1141 } 1142 1143 /* Assemble the final matrix and clean up */ 1144 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1145 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1146 ierr = PetscLogFlops(flops);CHKERRQ(ierr); 1147 PetscFunctionReturn(0); 1148 } 1149 1150 #undef __FUNCT__ 1151 #define __FUNCT__ "MatMatMult_SeqAIJ_SeqDense" 1152 PetscErrorCode MatMatMult_SeqAIJ_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 1153 { 1154 PetscErrorCode ierr; 1155 1156 PetscFunctionBegin; 1157 if (scall == MAT_INITIAL_MATRIX) { 1158 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1159 ierr = MatMatMultSymbolic_SeqAIJ_SeqDense(A,B,fill,C);CHKERRQ(ierr); 1160 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 1161 } 1162 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1163 ierr = MatMatMultNumeric_SeqAIJ_SeqDense(A,B,*C);CHKERRQ(ierr); 1164 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 1165 PetscFunctionReturn(0); 1166 } 1167 1168 #undef __FUNCT__ 1169 #define __FUNCT__ "MatMatMultSymbolic_SeqAIJ_SeqDense" 1170 PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C) 1171 { 1172 PetscErrorCode ierr; 1173 1174 PetscFunctionBegin; 1175 ierr = MatMatMultSymbolic_SeqDense_SeqDense(A,B,0.0,C);CHKERRQ(ierr); 1176 1177 (*C)->ops->matmultnumeric = MatMatMultNumeric_SeqAIJ_SeqDense; 1178 PetscFunctionReturn(0); 1179 } 1180 1181 #undef __FUNCT__ 1182 #define __FUNCT__ "MatMatMultNumeric_SeqAIJ_SeqDense" 1183 PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1184 { 1185 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1186 PetscErrorCode ierr; 1187 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1188 MatScalar *aa; 1189 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=B->rmap->n, col, i,j,n,*aj, am = A->rmap->n; 1190 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam; 1191 1192 PetscFunctionBegin; 1193 if (!cm || !cn) PetscFunctionReturn(0); 1194 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); 1195 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); 1196 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); 1197 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1198 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1199 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1200 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1201 colam = col*am; 1202 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1203 r1 = r2 = r3 = r4 = 0.0; 1204 n = a->i[i+1] - a->i[i]; 1205 aj = a->j + a->i[i]; 1206 aa = a->a + a->i[i]; 1207 for (j=0; j<n; j++) { 1208 r1 += (*aa)*b1[*aj]; 1209 r2 += (*aa)*b2[*aj]; 1210 r3 += (*aa)*b3[*aj]; 1211 r4 += (*aa++)*b4[*aj++]; 1212 } 1213 c[colam + i] = r1; 1214 c[colam + am + i] = r2; 1215 c[colam + am2 + i] = r3; 1216 c[colam + am3 + i] = r4; 1217 } 1218 b1 += bm4; 1219 b2 += bm4; 1220 b3 += bm4; 1221 b4 += bm4; 1222 } 1223 for (; col<cn; col++) { /* over extra columns of C */ 1224 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1225 r1 = 0.0; 1226 n = a->i[i+1] - a->i[i]; 1227 aj = a->j + a->i[i]; 1228 aa = a->a + a->i[i]; 1229 1230 for (j=0; j<n; j++) { 1231 r1 += (*aa++)*b1[*aj++]; 1232 } 1233 c[col*am + i] = r1; 1234 } 1235 b1 += bm; 1236 } 1237 ierr = PetscLogFlops(cn*(2.0*a->nz));CHKERRQ(ierr); 1238 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1239 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1240 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1241 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1242 PetscFunctionReturn(0); 1243 } 1244 1245 /* 1246 Note very similar to MatMult_SeqAIJ(), should generate both codes from same base 1247 */ 1248 #undef __FUNCT__ 1249 #define __FUNCT__ "MatMatMultNumericAdd_SeqAIJ_SeqDense" 1250 PetscErrorCode MatMatMultNumericAdd_SeqAIJ_SeqDense(Mat A,Mat B,Mat C) 1251 { 1252 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 1253 PetscErrorCode ierr; 1254 PetscScalar *b,*c,r1,r2,r3,r4,*b1,*b2,*b3,*b4; 1255 MatScalar *aa; 1256 PetscInt cm = C->rmap->n, cn=B->cmap->n, bm=B->rmap->n, col, i,j,n,*aj, am = A->rmap->n,*ii,arm; 1257 PetscInt am2 = 2*am, am3 = 3*am, bm4 = 4*bm,colam,*ridx; 1258 1259 PetscFunctionBegin; 1260 if (!cm || !cn) PetscFunctionReturn(0); 1261 ierr = MatDenseGetArray(B,&b);CHKERRQ(ierr); 1262 ierr = MatDenseGetArray(C,&c);CHKERRQ(ierr); 1263 b1 = b; b2 = b1 + bm; b3 = b2 + bm; b4 = b3 + bm; 1264 1265 if (a->compressedrow.use) { /* use compressed row format */ 1266 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1267 colam = col*am; 1268 arm = a->compressedrow.nrows; 1269 ii = a->compressedrow.i; 1270 ridx = a->compressedrow.rindex; 1271 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1272 r1 = r2 = r3 = r4 = 0.0; 1273 n = ii[i+1] - ii[i]; 1274 aj = a->j + ii[i]; 1275 aa = a->a + ii[i]; 1276 for (j=0; j<n; j++) { 1277 r1 += (*aa)*b1[*aj]; 1278 r2 += (*aa)*b2[*aj]; 1279 r3 += (*aa)*b3[*aj]; 1280 r4 += (*aa++)*b4[*aj++]; 1281 } 1282 c[colam + ridx[i]] += r1; 1283 c[colam + am + ridx[i]] += r2; 1284 c[colam + am2 + ridx[i]] += r3; 1285 c[colam + am3 + ridx[i]] += r4; 1286 } 1287 b1 += bm4; 1288 b2 += bm4; 1289 b3 += bm4; 1290 b4 += bm4; 1291 } 1292 for (; col<cn; col++) { /* over extra columns of C */ 1293 colam = col*am; 1294 arm = a->compressedrow.nrows; 1295 ii = a->compressedrow.i; 1296 ridx = a->compressedrow.rindex; 1297 for (i=0; i<arm; i++) { /* over rows of C in those columns */ 1298 r1 = 0.0; 1299 n = ii[i+1] - ii[i]; 1300 aj = a->j + ii[i]; 1301 aa = a->a + ii[i]; 1302 1303 for (j=0; j<n; j++) { 1304 r1 += (*aa++)*b1[*aj++]; 1305 } 1306 c[colam + ridx[i]] += r1; 1307 } 1308 b1 += bm; 1309 } 1310 } else { 1311 for (col=0; col<cn-4; col += 4) { /* over columns of C */ 1312 colam = col*am; 1313 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1314 r1 = r2 = r3 = r4 = 0.0; 1315 n = a->i[i+1] - a->i[i]; 1316 aj = a->j + a->i[i]; 1317 aa = a->a + a->i[i]; 1318 for (j=0; j<n; j++) { 1319 r1 += (*aa)*b1[*aj]; 1320 r2 += (*aa)*b2[*aj]; 1321 r3 += (*aa)*b3[*aj]; 1322 r4 += (*aa++)*b4[*aj++]; 1323 } 1324 c[colam + i] += r1; 1325 c[colam + am + i] += r2; 1326 c[colam + am2 + i] += r3; 1327 c[colam + am3 + i] += r4; 1328 } 1329 b1 += bm4; 1330 b2 += bm4; 1331 b3 += bm4; 1332 b4 += bm4; 1333 } 1334 for (; col<cn; col++) { /* over extra columns of C */ 1335 colam = col*am; 1336 for (i=0; i<am; i++) { /* over rows of C in those columns */ 1337 r1 = 0.0; 1338 n = a->i[i+1] - a->i[i]; 1339 aj = a->j + a->i[i]; 1340 aa = a->a + a->i[i]; 1341 1342 for (j=0; j<n; j++) { 1343 r1 += (*aa++)*b1[*aj++]; 1344 } 1345 c[colam + i] += r1; 1346 } 1347 b1 += bm; 1348 } 1349 } 1350 ierr = PetscLogFlops(cn*2.0*a->nz);CHKERRQ(ierr); 1351 ierr = MatDenseRestoreArray(B,&b);CHKERRQ(ierr); 1352 ierr = MatDenseRestoreArray(C,&c);CHKERRQ(ierr); 1353 PetscFunctionReturn(0); 1354 } 1355 1356 #undef __FUNCT__ 1357 #define __FUNCT__ "MatTransColoringApplySpToDen_SeqAIJ" 1358 PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring coloring,Mat B,Mat Btdense) 1359 { 1360 PetscErrorCode ierr; 1361 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 1362 Mat_SeqDense *btdense = (Mat_SeqDense*)Btdense->data; 1363 PetscInt *bi = b->i,*bj=b->j; 1364 PetscInt m = Btdense->rmap->n,n=Btdense->cmap->n,j,k,l,col,anz,*btcol,brow,ncolumns; 1365 MatScalar *btval,*btval_den,*ba=b->a; 1366 PetscInt *columns=coloring->columns,*colorforcol=coloring->colorforcol,ncolors=coloring->ncolors; 1367 1368 PetscFunctionBegin; 1369 btval_den=btdense->v; 1370 ierr = PetscMemzero(btval_den,(m*n)*sizeof(MatScalar));CHKERRQ(ierr); 1371 for (k=0; k<ncolors; k++) { 1372 ncolumns = coloring->ncolumns[k]; 1373 for (l=0; l<ncolumns; l++) { /* insert a row of B to a column of Btdense */ 1374 col = *(columns + colorforcol[k] + l); 1375 btcol = bj + bi[col]; 1376 btval = ba + bi[col]; 1377 anz = bi[col+1] - bi[col]; 1378 for (j=0; j<anz; j++) { 1379 brow = btcol[j]; 1380 btval_den[brow] = btval[j]; 1381 } 1382 } 1383 btval_den += m; 1384 } 1385 PetscFunctionReturn(0); 1386 } 1387 1388 #undef __FUNCT__ 1389 #define __FUNCT__ "MatTransColoringApplyDenToSp_SeqAIJ" 1390 PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring matcoloring,Mat Cden,Mat Csp) 1391 { 1392 PetscErrorCode ierr; 1393 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)Csp->data; 1394 PetscInt k,l,*row,*idx,m,ncolors=matcoloring->ncolors,nrows; 1395 PetscScalar *ca_den,*cp_den,*ca=csp->a; 1396 PetscInt *rows=matcoloring->rows,*spidx=matcoloring->columnsforspidx,*colorforrow=matcoloring->colorforrow; 1397 1398 PetscFunctionBegin; 1399 ierr = MatGetLocalSize(Csp,&m,NULL);CHKERRQ(ierr); 1400 ierr = MatDenseGetArray(Cden,&ca_den);CHKERRQ(ierr); 1401 cp_den = ca_den; 1402 for (k=0; k<ncolors; k++) { 1403 nrows = matcoloring->nrows[k]; 1404 row = rows + colorforrow[k]; 1405 idx = spidx + colorforrow[k]; 1406 for (l=0; l<nrows; l++) { 1407 ca[idx[l]] = cp_den[row[l]]; 1408 } 1409 cp_den += m; 1410 } 1411 ierr = MatDenseRestoreArray(Cden,&ca_den);CHKERRQ(ierr); 1412 PetscFunctionReturn(0); 1413 } 1414 1415 /* 1416 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 1417 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 1418 spidx[], index of a->j, to be used for setting 'columnsforspidx' in MatTransposeColoringCreate_SeqAIJ(). 1419 */ 1420 #undef __FUNCT__ 1421 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ_Color" 1422 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1423 { 1424 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1425 PetscErrorCode ierr; 1426 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 1427 PetscInt nz = a->i[m],row,*jj,mr,col; 1428 PetscInt *cspidx; 1429 1430 PetscFunctionBegin; 1431 *nn = n; 1432 if (!ia) PetscFunctionReturn(0); 1433 if (symmetric) { 1434 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MatGetColumnIJ_SeqAIJ_Color() not supported for the case symmetric"); 1435 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 1436 } else { 1437 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&collengths);CHKERRQ(ierr); 1438 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1439 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&cia);CHKERRQ(ierr); 1440 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cja);CHKERRQ(ierr); 1441 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cspidx);CHKERRQ(ierr); 1442 jj = a->j; 1443 for (i=0; i<nz; i++) { 1444 collengths[jj[i]]++; 1445 } 1446 cia[0] = oshift; 1447 for (i=0; i<n; i++) { 1448 cia[i+1] = cia[i] + collengths[i]; 1449 } 1450 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 1451 jj = a->j; 1452 for (row=0; row<m; row++) { 1453 mr = a->i[row+1] - a->i[row]; 1454 for (i=0; i<mr; i++) { 1455 col = *jj++; 1456 1457 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 1458 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 1459 } 1460 } 1461 ierr = PetscFree(collengths);CHKERRQ(ierr); 1462 *ia = cia; *ja = cja; 1463 *spidx = cspidx; 1464 } 1465 PetscFunctionReturn(0); 1466 } 1467 1468 #undef __FUNCT__ 1469 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ_Color" 1470 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 1471 { 1472 PetscErrorCode ierr; 1473 1474 PetscFunctionBegin; 1475 if (!ia) PetscFunctionReturn(0); 1476 1477 ierr = PetscFree(*ia);CHKERRQ(ierr); 1478 ierr = PetscFree(*ja);CHKERRQ(ierr); 1479 ierr = PetscFree(*spidx);CHKERRQ(ierr); 1480 PetscFunctionReturn(0); 1481 } 1482 1483 #undef __FUNCT__ 1484 #define __FUNCT__ "MatTransposeColoringCreate_SeqAIJ" 1485 PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat mat,ISColoring iscoloring,MatTransposeColoring c) 1486 { 1487 PetscErrorCode ierr; 1488 PetscInt i,n,nrows,N,j,k,m,ncols,col,cm; 1489 const PetscInt *is,*ci,*cj,*row_idx; 1490 PetscInt nis = iscoloring->n,*rowhit,bs = 1; 1491 IS *isa; 1492 PetscBool flg1,flg2; 1493 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)mat->data; 1494 PetscInt *colorforrow,*rows,*rows_i,*columnsforspidx,*columnsforspidx_i,*idxhit,*spidx; 1495 PetscInt *colorforcol,*columns,*columns_i; 1496 1497 PetscFunctionBegin; 1498 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 1499 1500 /* this is ugly way to get blocksize but cannot call MatGetBlockSize() because AIJ can have bs > 1 */ 1501 ierr = PetscObjectTypeCompare((PetscObject)mat,MATSEQBAIJ,&flg1);CHKERRQ(ierr); 1502 ierr = PetscObjectTypeCompare((PetscObject)mat,MATMPIBAIJ,&flg2);CHKERRQ(ierr); 1503 if (flg1 || flg2) { 1504 ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr); 1505 } 1506 1507 N = mat->cmap->N/bs; 1508 c->M = mat->rmap->N/bs; /* set total rows, columns and local rows */ 1509 c->N = mat->cmap->N/bs; 1510 c->m = mat->rmap->N/bs; 1511 c->rstart = 0; 1512 1513 c->ncolors = nis; 1514 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 1515 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 1516 ierr = PetscMalloc2(csp->nz+1,PetscInt,&rows,csp->nz+1,PetscInt,&columnsforspidx);CHKERRQ(ierr); 1517 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&colorforrow);CHKERRQ(ierr); 1518 1519 colorforrow[0] = 0; 1520 rows_i = rows; 1521 columnsforspidx_i = columnsforspidx; 1522 1523 ierr = PetscMalloc((nis+1)*sizeof(PetscInt),&colorforcol);CHKERRQ(ierr); 1524 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&columns);CHKERRQ(ierr); 1525 1526 colorforcol[0] = 0; 1527 columns_i = columns; 1528 1529 ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); /* column-wise storage of mat */ 1530 1531 cm = c->m; 1532 ierr = PetscMalloc((cm+1)*sizeof(PetscInt),&rowhit);CHKERRQ(ierr); 1533 ierr = PetscMalloc((cm+1)*sizeof(PetscInt),&idxhit);CHKERRQ(ierr); 1534 for (i=0; i<nis; i++) { 1535 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 1536 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 1537 1538 c->ncolumns[i] = n; 1539 if (n) { 1540 ierr = PetscMemcpy(columns_i,is,n*sizeof(PetscInt));CHKERRQ(ierr); 1541 } 1542 colorforcol[i+1] = colorforcol[i] + n; 1543 columns_i += n; 1544 1545 /* fast, crude version requires O(N*N) work */ 1546 ierr = PetscMemzero(rowhit,cm*sizeof(PetscInt));CHKERRQ(ierr); 1547 1548 /* loop over columns*/ 1549 for (j=0; j<n; j++) { 1550 col = is[j]; 1551 row_idx = cj + ci[col]; 1552 m = ci[col+1] - ci[col]; 1553 /* loop over columns marking them in rowhit */ 1554 for (k=0; k<m; k++) { 1555 idxhit[*row_idx] = spidx[ci[col] + k]; 1556 rowhit[*row_idx++] = col + 1; 1557 } 1558 } 1559 /* count the number of hits */ 1560 nrows = 0; 1561 for (j=0; j<cm; j++) { 1562 if (rowhit[j]) nrows++; 1563 } 1564 c->nrows[i] = nrows; 1565 colorforrow[i+1] = colorforrow[i] + nrows; 1566 1567 nrows = 0; 1568 for (j=0; j<cm; j++) { 1569 if (rowhit[j]) { 1570 rows_i[nrows] = j; 1571 columnsforspidx_i[nrows] = idxhit[j]; 1572 nrows++; 1573 } 1574 } 1575 ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr); 1576 rows_i += nrows; columnsforspidx_i += nrows; 1577 } 1578 ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr); 1579 ierr = PetscFree(rowhit);CHKERRQ(ierr); 1580 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 1581 #if defined(PETSC_USE_DEBUG) 1582 if (csp->nz != colorforrow[nis]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csp->nz %d != colorforrow[nis] %d",csp->nz,colorforrow[nis]); 1583 #endif 1584 1585 c->colorforrow = colorforrow; 1586 c->rows = rows; 1587 c->columnsforspidx = columnsforspidx; 1588 c->colorforcol = colorforcol; 1589 c->columns = columns; 1590 1591 ierr = PetscFree(idxhit);CHKERRQ(ierr); 1592 PetscFunctionReturn(0); 1593 } 1594