1 2 3 #include <../src/mat/impls/aij/mpi/mpiaij.h> /*I "petscmat.h" I*/ 4 #include <petsc/private/vecimpl.h> 5 #include <petsc/private/isimpl.h> 6 #include <petscblaslapack.h> 7 #include <petscsf.h> 8 9 /*MC 10 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 11 12 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 13 and MATMPIAIJ otherwise. As a result, for single process communicators, 14 MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported 15 for communicators controlling multiple processes. It is recommended that you call both of 16 the above preallocation routines for simplicity. 17 18 Options Database Keys: 19 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 20 21 Developer Notes: Subclasses include MATAIJCUSP, MATAIJCUSPARSE, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when 22 enough exist. 23 24 Level: beginner 25 26 .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ, MATMPIAIJ 27 M*/ 28 29 /*MC 30 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 31 32 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 33 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 34 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 35 for communicators controlling multiple processes. It is recommended that you call both of 36 the above preallocation routines for simplicity. 37 38 Options Database Keys: 39 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 40 41 Level: beginner 42 43 .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL 44 M*/ 45 46 PetscErrorCode MatSetBlockSizes_MPIAIJ(Mat M, PetscInt rbs, PetscInt cbs) 47 { 48 PetscErrorCode ierr; 49 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)M->data; 50 51 PetscFunctionBegin; 52 if (mat->A) { 53 ierr = MatSetBlockSizes(mat->A,rbs,cbs);CHKERRQ(ierr); 54 ierr = MatSetBlockSizes(mat->B,rbs,1);CHKERRQ(ierr); 55 } 56 PetscFunctionReturn(0); 57 } 58 59 PetscErrorCode MatFindNonzeroRows_MPIAIJ(Mat M,IS *keptrows) 60 { 61 PetscErrorCode ierr; 62 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)M->data; 63 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data; 64 Mat_SeqAIJ *b = (Mat_SeqAIJ*)mat->B->data; 65 const PetscInt *ia,*ib; 66 const MatScalar *aa,*bb; 67 PetscInt na,nb,i,j,*rows,cnt=0,n0rows; 68 PetscInt m = M->rmap->n,rstart = M->rmap->rstart; 69 70 PetscFunctionBegin; 71 *keptrows = 0; 72 ia = a->i; 73 ib = b->i; 74 for (i=0; i<m; i++) { 75 na = ia[i+1] - ia[i]; 76 nb = ib[i+1] - ib[i]; 77 if (!na && !nb) { 78 cnt++; 79 goto ok1; 80 } 81 aa = a->a + ia[i]; 82 for (j=0; j<na; j++) { 83 if (aa[j] != 0.0) goto ok1; 84 } 85 bb = b->a + ib[i]; 86 for (j=0; j <nb; j++) { 87 if (bb[j] != 0.0) goto ok1; 88 } 89 cnt++; 90 ok1:; 91 } 92 ierr = MPIU_Allreduce(&cnt,&n0rows,1,MPIU_INT,MPI_SUM,PetscObjectComm((PetscObject)M));CHKERRQ(ierr); 93 if (!n0rows) PetscFunctionReturn(0); 94 ierr = PetscMalloc1(M->rmap->n-cnt,&rows);CHKERRQ(ierr); 95 cnt = 0; 96 for (i=0; i<m; i++) { 97 na = ia[i+1] - ia[i]; 98 nb = ib[i+1] - ib[i]; 99 if (!na && !nb) continue; 100 aa = a->a + ia[i]; 101 for (j=0; j<na;j++) { 102 if (aa[j] != 0.0) { 103 rows[cnt++] = rstart + i; 104 goto ok2; 105 } 106 } 107 bb = b->a + ib[i]; 108 for (j=0; j<nb; j++) { 109 if (bb[j] != 0.0) { 110 rows[cnt++] = rstart + i; 111 goto ok2; 112 } 113 } 114 ok2:; 115 } 116 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)M),cnt,rows,PETSC_OWN_POINTER,keptrows);CHKERRQ(ierr); 117 PetscFunctionReturn(0); 118 } 119 120 PetscErrorCode MatDiagonalSet_MPIAIJ(Mat Y,Vec D,InsertMode is) 121 { 122 PetscErrorCode ierr; 123 Mat_MPIAIJ *aij = (Mat_MPIAIJ*) Y->data; 124 125 PetscFunctionBegin; 126 if (Y->assembled && Y->rmap->rstart == Y->cmap->rstart && Y->rmap->rend == Y->cmap->rend) { 127 ierr = MatDiagonalSet(aij->A,D,is);CHKERRQ(ierr); 128 } else { 129 ierr = MatDiagonalSet_Default(Y,D,is);CHKERRQ(ierr); 130 } 131 PetscFunctionReturn(0); 132 } 133 134 135 PetscErrorCode MatFindZeroDiagonals_MPIAIJ(Mat M,IS *zrows) 136 { 137 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)M->data; 138 PetscErrorCode ierr; 139 PetscInt i,rstart,nrows,*rows; 140 141 PetscFunctionBegin; 142 *zrows = NULL; 143 ierr = MatFindZeroDiagonals_SeqAIJ_Private(aij->A,&nrows,&rows);CHKERRQ(ierr); 144 ierr = MatGetOwnershipRange(M,&rstart,NULL);CHKERRQ(ierr); 145 for (i=0; i<nrows; i++) rows[i] += rstart; 146 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)M),nrows,rows,PETSC_OWN_POINTER,zrows);CHKERRQ(ierr); 147 PetscFunctionReturn(0); 148 } 149 150 PetscErrorCode MatGetColumnNorms_MPIAIJ(Mat A,NormType type,PetscReal *norms) 151 { 152 PetscErrorCode ierr; 153 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 154 PetscInt i,n,*garray = aij->garray; 155 Mat_SeqAIJ *a_aij = (Mat_SeqAIJ*) aij->A->data; 156 Mat_SeqAIJ *b_aij = (Mat_SeqAIJ*) aij->B->data; 157 PetscReal *work; 158 159 PetscFunctionBegin; 160 ierr = MatGetSize(A,NULL,&n);CHKERRQ(ierr); 161 ierr = PetscCalloc1(n,&work);CHKERRQ(ierr); 162 if (type == NORM_2) { 163 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 164 work[A->cmap->rstart + a_aij->j[i]] += PetscAbsScalar(a_aij->a[i]*a_aij->a[i]); 165 } 166 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 167 work[garray[b_aij->j[i]]] += PetscAbsScalar(b_aij->a[i]*b_aij->a[i]); 168 } 169 } else if (type == NORM_1) { 170 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 171 work[A->cmap->rstart + a_aij->j[i]] += PetscAbsScalar(a_aij->a[i]); 172 } 173 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 174 work[garray[b_aij->j[i]]] += PetscAbsScalar(b_aij->a[i]); 175 } 176 } else if (type == NORM_INFINITY) { 177 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 178 work[A->cmap->rstart + a_aij->j[i]] = PetscMax(PetscAbsScalar(a_aij->a[i]), work[A->cmap->rstart + a_aij->j[i]]); 179 } 180 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 181 work[garray[b_aij->j[i]]] = PetscMax(PetscAbsScalar(b_aij->a[i]),work[garray[b_aij->j[i]]]); 182 } 183 184 } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Unknown NormType"); 185 if (type == NORM_INFINITY) { 186 ierr = MPIU_Allreduce(work,norms,n,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 187 } else { 188 ierr = MPIU_Allreduce(work,norms,n,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 189 } 190 ierr = PetscFree(work);CHKERRQ(ierr); 191 if (type == NORM_2) { 192 for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]); 193 } 194 PetscFunctionReturn(0); 195 } 196 197 PetscErrorCode MatFindOffBlockDiagonalEntries_MPIAIJ(Mat A,IS *is) 198 { 199 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 200 IS sis,gis; 201 PetscErrorCode ierr; 202 const PetscInt *isis,*igis; 203 PetscInt n,*iis,nsis,ngis,rstart,i; 204 205 PetscFunctionBegin; 206 ierr = MatFindOffBlockDiagonalEntries(a->A,&sis);CHKERRQ(ierr); 207 ierr = MatFindNonzeroRows(a->B,&gis);CHKERRQ(ierr); 208 ierr = ISGetSize(gis,&ngis);CHKERRQ(ierr); 209 ierr = ISGetSize(sis,&nsis);CHKERRQ(ierr); 210 ierr = ISGetIndices(sis,&isis);CHKERRQ(ierr); 211 ierr = ISGetIndices(gis,&igis);CHKERRQ(ierr); 212 213 ierr = PetscMalloc1(ngis+nsis,&iis);CHKERRQ(ierr); 214 ierr = PetscMemcpy(iis,igis,ngis*sizeof(PetscInt));CHKERRQ(ierr); 215 ierr = PetscMemcpy(iis+ngis,isis,nsis*sizeof(PetscInt));CHKERRQ(ierr); 216 n = ngis + nsis; 217 ierr = PetscSortRemoveDupsInt(&n,iis);CHKERRQ(ierr); 218 ierr = MatGetOwnershipRange(A,&rstart,NULL);CHKERRQ(ierr); 219 for (i=0; i<n; i++) iis[i] += rstart; 220 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)A),n,iis,PETSC_OWN_POINTER,is);CHKERRQ(ierr); 221 222 ierr = ISRestoreIndices(sis,&isis);CHKERRQ(ierr); 223 ierr = ISRestoreIndices(gis,&igis);CHKERRQ(ierr); 224 ierr = ISDestroy(&sis);CHKERRQ(ierr); 225 ierr = ISDestroy(&gis);CHKERRQ(ierr); 226 PetscFunctionReturn(0); 227 } 228 229 /* 230 Distributes a SeqAIJ matrix across a set of processes. Code stolen from 231 MatLoad_MPIAIJ(). Horrible lack of reuse. Should be a routine for each matrix type. 232 233 Only for square matrices 234 235 Used by a preconditioner, hence PETSC_EXTERN 236 */ 237 PETSC_EXTERN PetscErrorCode MatDistribute_MPIAIJ(MPI_Comm comm,Mat gmat,PetscInt m,MatReuse reuse,Mat *inmat) 238 { 239 PetscMPIInt rank,size; 240 PetscInt *rowners,*dlens,*olens,i,rstart,rend,j,jj,nz = 0,*gmataj,cnt,row,*ld,bses[2]; 241 PetscErrorCode ierr; 242 Mat mat; 243 Mat_SeqAIJ *gmata; 244 PetscMPIInt tag; 245 MPI_Status status; 246 PetscBool aij; 247 MatScalar *gmataa,*ao,*ad,*gmataarestore=0; 248 249 PetscFunctionBegin; 250 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 251 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 252 if (!rank) { 253 ierr = PetscObjectTypeCompare((PetscObject)gmat,MATSEQAIJ,&aij);CHKERRQ(ierr); 254 if (!aij) SETERRQ1(PetscObjectComm((PetscObject)gmat),PETSC_ERR_SUP,"Currently no support for input matrix of type %s\n",((PetscObject)gmat)->type_name); 255 } 256 if (reuse == MAT_INITIAL_MATRIX) { 257 ierr = MatCreate(comm,&mat);CHKERRQ(ierr); 258 ierr = MatSetSizes(mat,m,m,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 259 ierr = MatGetBlockSizes(gmat,&bses[0],&bses[1]);CHKERRQ(ierr); 260 ierr = MPI_Bcast(bses,2,MPIU_INT,0,comm);CHKERRQ(ierr); 261 ierr = MatSetBlockSizes(mat,bses[0],bses[1]);CHKERRQ(ierr); 262 ierr = MatSetType(mat,MATAIJ);CHKERRQ(ierr); 263 ierr = PetscMalloc1(size+1,&rowners);CHKERRQ(ierr); 264 ierr = PetscMalloc2(m,&dlens,m,&olens);CHKERRQ(ierr); 265 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 266 267 rowners[0] = 0; 268 for (i=2; i<=size; i++) rowners[i] += rowners[i-1]; 269 rstart = rowners[rank]; 270 rend = rowners[rank+1]; 271 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 272 if (!rank) { 273 gmata = (Mat_SeqAIJ*) gmat->data; 274 /* send row lengths to all processors */ 275 for (i=0; i<m; i++) dlens[i] = gmata->ilen[i]; 276 for (i=1; i<size; i++) { 277 ierr = MPI_Send(gmata->ilen + rowners[i],rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 278 } 279 /* determine number diagonal and off-diagonal counts */ 280 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 281 ierr = PetscCalloc1(m,&ld);CHKERRQ(ierr); 282 jj = 0; 283 for (i=0; i<m; i++) { 284 for (j=0; j<dlens[i]; j++) { 285 if (gmata->j[jj] < rstart) ld[i]++; 286 if (gmata->j[jj] < rstart || gmata->j[jj] >= rend) olens[i]++; 287 jj++; 288 } 289 } 290 /* send column indices to other processes */ 291 for (i=1; i<size; i++) { 292 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 293 ierr = MPI_Send(&nz,1,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 294 ierr = MPI_Send(gmata->j + gmata->i[rowners[i]],nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 295 } 296 297 /* send numerical values to other processes */ 298 for (i=1; i<size; i++) { 299 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 300 ierr = MPI_Send(gmata->a + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 301 } 302 gmataa = gmata->a; 303 gmataj = gmata->j; 304 305 } else { 306 /* receive row lengths */ 307 ierr = MPI_Recv(dlens,m,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 308 /* receive column indices */ 309 ierr = MPI_Recv(&nz,1,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 310 ierr = PetscMalloc2(nz,&gmataa,nz,&gmataj);CHKERRQ(ierr); 311 ierr = MPI_Recv(gmataj,nz,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 312 /* determine number diagonal and off-diagonal counts */ 313 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 314 ierr = PetscCalloc1(m,&ld);CHKERRQ(ierr); 315 jj = 0; 316 for (i=0; i<m; i++) { 317 for (j=0; j<dlens[i]; j++) { 318 if (gmataj[jj] < rstart) ld[i]++; 319 if (gmataj[jj] < rstart || gmataj[jj] >= rend) olens[i]++; 320 jj++; 321 } 322 } 323 /* receive numerical values */ 324 ierr = PetscMemzero(gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); 325 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 326 } 327 /* set preallocation */ 328 for (i=0; i<m; i++) { 329 dlens[i] -= olens[i]; 330 } 331 ierr = MatSeqAIJSetPreallocation(mat,0,dlens);CHKERRQ(ierr); 332 ierr = MatMPIAIJSetPreallocation(mat,0,dlens,0,olens);CHKERRQ(ierr); 333 334 for (i=0; i<m; i++) { 335 dlens[i] += olens[i]; 336 } 337 cnt = 0; 338 for (i=0; i<m; i++) { 339 row = rstart + i; 340 ierr = MatSetValues(mat,1,&row,dlens[i],gmataj+cnt,gmataa+cnt,INSERT_VALUES);CHKERRQ(ierr); 341 cnt += dlens[i]; 342 } 343 if (rank) { 344 ierr = PetscFree2(gmataa,gmataj);CHKERRQ(ierr); 345 } 346 ierr = PetscFree2(dlens,olens);CHKERRQ(ierr); 347 ierr = PetscFree(rowners);CHKERRQ(ierr); 348 349 ((Mat_MPIAIJ*)(mat->data))->ld = ld; 350 351 *inmat = mat; 352 } else { /* column indices are already set; only need to move over numerical values from process 0 */ 353 Mat_SeqAIJ *Ad = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->A->data; 354 Mat_SeqAIJ *Ao = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->B->data; 355 mat = *inmat; 356 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 357 if (!rank) { 358 /* send numerical values to other processes */ 359 gmata = (Mat_SeqAIJ*) gmat->data; 360 ierr = MatGetOwnershipRanges(mat,(const PetscInt**)&rowners);CHKERRQ(ierr); 361 gmataa = gmata->a; 362 for (i=1; i<size; i++) { 363 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 364 ierr = MPI_Send(gmataa + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 365 } 366 nz = gmata->i[rowners[1]]-gmata->i[rowners[0]]; 367 } else { 368 /* receive numerical values from process 0*/ 369 nz = Ad->nz + Ao->nz; 370 ierr = PetscMalloc1(nz,&gmataa);CHKERRQ(ierr); gmataarestore = gmataa; 371 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 372 } 373 /* transfer numerical values into the diagonal A and off diagonal B parts of mat */ 374 ld = ((Mat_MPIAIJ*)(mat->data))->ld; 375 ad = Ad->a; 376 ao = Ao->a; 377 if (mat->rmap->n) { 378 i = 0; 379 nz = ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ao += nz; gmataa += nz; 380 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 381 } 382 for (i=1; i<mat->rmap->n; i++) { 383 nz = Ao->i[i] - Ao->i[i-1] - ld[i-1] + ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ao += nz; gmataa += nz; 384 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 385 } 386 i--; 387 if (mat->rmap->n) { 388 nz = Ao->i[i+1] - Ao->i[i] - ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); 389 } 390 if (rank) { 391 ierr = PetscFree(gmataarestore);CHKERRQ(ierr); 392 } 393 } 394 ierr = MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 395 ierr = MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 396 PetscFunctionReturn(0); 397 } 398 399 /* 400 Local utility routine that creates a mapping from the global column 401 number to the local number in the off-diagonal part of the local 402 storage of the matrix. When PETSC_USE_CTABLE is used this is scalable at 403 a slightly higher hash table cost; without it it is not scalable (each processor 404 has an order N integer array but is fast to acess. 405 */ 406 PetscErrorCode MatCreateColmap_MPIAIJ_Private(Mat mat) 407 { 408 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 409 PetscErrorCode ierr; 410 PetscInt n = aij->B->cmap->n,i; 411 412 PetscFunctionBegin; 413 if (!aij->garray) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"MPIAIJ Matrix was assembled but is missing garray"); 414 #if defined(PETSC_USE_CTABLE) 415 ierr = PetscTableCreate(n,mat->cmap->N+1,&aij->colmap);CHKERRQ(ierr); 416 for (i=0; i<n; i++) { 417 ierr = PetscTableAdd(aij->colmap,aij->garray[i]+1,i+1,INSERT_VALUES);CHKERRQ(ierr); 418 } 419 #else 420 ierr = PetscCalloc1(mat->cmap->N+1,&aij->colmap);CHKERRQ(ierr); 421 ierr = PetscLogObjectMemory((PetscObject)mat,(mat->cmap->N+1)*sizeof(PetscInt));CHKERRQ(ierr); 422 for (i=0; i<n; i++) aij->colmap[aij->garray[i]] = i+1; 423 #endif 424 PetscFunctionReturn(0); 425 } 426 427 #define MatSetValues_SeqAIJ_A_Private(row,col,value,addv,orow,ocol) \ 428 { \ 429 if (col <= lastcol1) low1 = 0; \ 430 else high1 = nrow1; \ 431 lastcol1 = col;\ 432 while (high1-low1 > 5) { \ 433 t = (low1+high1)/2; \ 434 if (rp1[t] > col) high1 = t; \ 435 else low1 = t; \ 436 } \ 437 for (_i=low1; _i<high1; _i++) { \ 438 if (rp1[_i] > col) break; \ 439 if (rp1[_i] == col) { \ 440 if (addv == ADD_VALUES) ap1[_i] += value; \ 441 else ap1[_i] = value; \ 442 goto a_noinsert; \ 443 } \ 444 } \ 445 if (value == 0.0 && ignorezeroentries && row != col) {low1 = 0; high1 = nrow1;goto a_noinsert;} \ 446 if (nonew == 1) {low1 = 0; high1 = nrow1; goto a_noinsert;} \ 447 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at global row/column (%D, %D) into matrix", orow, ocol); \ 448 MatSeqXAIJReallocateAIJ(A,am,1,nrow1,row,col,rmax1,aa,ai,aj,rp1,ap1,aimax,nonew,MatScalar); \ 449 N = nrow1++ - 1; a->nz++; high1++; \ 450 /* shift up all the later entries in this row */ \ 451 for (ii=N; ii>=_i; ii--) { \ 452 rp1[ii+1] = rp1[ii]; \ 453 ap1[ii+1] = ap1[ii]; \ 454 } \ 455 rp1[_i] = col; \ 456 ap1[_i] = value; \ 457 A->nonzerostate++;\ 458 a_noinsert: ; \ 459 ailen[row] = nrow1; \ 460 } 461 462 463 #define MatSetValues_SeqAIJ_B_Private(row,col,value,addv,orow,ocol) \ 464 { \ 465 if (col <= lastcol2) low2 = 0; \ 466 else high2 = nrow2; \ 467 lastcol2 = col; \ 468 while (high2-low2 > 5) { \ 469 t = (low2+high2)/2; \ 470 if (rp2[t] > col) high2 = t; \ 471 else low2 = t; \ 472 } \ 473 for (_i=low2; _i<high2; _i++) { \ 474 if (rp2[_i] > col) break; \ 475 if (rp2[_i] == col) { \ 476 if (addv == ADD_VALUES) ap2[_i] += value; \ 477 else ap2[_i] = value; \ 478 goto b_noinsert; \ 479 } \ 480 } \ 481 if (value == 0.0 && ignorezeroentries) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 482 if (nonew == 1) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 483 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at global row/column (%D, %D) into matrix", orow, ocol); \ 484 MatSeqXAIJReallocateAIJ(B,bm,1,nrow2,row,col,rmax2,ba,bi,bj,rp2,ap2,bimax,nonew,MatScalar); \ 485 N = nrow2++ - 1; b->nz++; high2++; \ 486 /* shift up all the later entries in this row */ \ 487 for (ii=N; ii>=_i; ii--) { \ 488 rp2[ii+1] = rp2[ii]; \ 489 ap2[ii+1] = ap2[ii]; \ 490 } \ 491 rp2[_i] = col; \ 492 ap2[_i] = value; \ 493 B->nonzerostate++; \ 494 b_noinsert: ; \ 495 bilen[row] = nrow2; \ 496 } 497 498 PetscErrorCode MatSetValuesRow_MPIAIJ(Mat A,PetscInt row,const PetscScalar v[]) 499 { 500 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 501 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data,*b = (Mat_SeqAIJ*)mat->B->data; 502 PetscErrorCode ierr; 503 PetscInt l,*garray = mat->garray,diag; 504 505 PetscFunctionBegin; 506 /* code only works for square matrices A */ 507 508 /* find size of row to the left of the diagonal part */ 509 ierr = MatGetOwnershipRange(A,&diag,0);CHKERRQ(ierr); 510 row = row - diag; 511 for (l=0; l<b->i[row+1]-b->i[row]; l++) { 512 if (garray[b->j[b->i[row]+l]] > diag) break; 513 } 514 ierr = PetscMemcpy(b->a+b->i[row],v,l*sizeof(PetscScalar));CHKERRQ(ierr); 515 516 /* diagonal part */ 517 ierr = PetscMemcpy(a->a+a->i[row],v+l,(a->i[row+1]-a->i[row])*sizeof(PetscScalar));CHKERRQ(ierr); 518 519 /* right of diagonal part */ 520 ierr = PetscMemcpy(b->a+b->i[row]+l,v+l+a->i[row+1]-a->i[row],(b->i[row+1]-b->i[row]-l)*sizeof(PetscScalar));CHKERRQ(ierr); 521 PetscFunctionReturn(0); 522 } 523 524 PetscErrorCode MatSetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv) 525 { 526 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 527 PetscScalar value; 528 PetscErrorCode ierr; 529 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 530 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 531 PetscBool roworiented = aij->roworiented; 532 533 /* Some Variables required in the macro */ 534 Mat A = aij->A; 535 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 536 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 537 MatScalar *aa = a->a; 538 PetscBool ignorezeroentries = a->ignorezeroentries; 539 Mat B = aij->B; 540 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 541 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 542 MatScalar *ba = b->a; 543 544 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 545 PetscInt nonew; 546 MatScalar *ap1,*ap2; 547 548 PetscFunctionBegin; 549 for (i=0; i<m; i++) { 550 if (im[i] < 0) continue; 551 #if defined(PETSC_USE_DEBUG) 552 if (im[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap->N-1); 553 #endif 554 if (im[i] >= rstart && im[i] < rend) { 555 row = im[i] - rstart; 556 lastcol1 = -1; 557 rp1 = aj + ai[row]; 558 ap1 = aa + ai[row]; 559 rmax1 = aimax[row]; 560 nrow1 = ailen[row]; 561 low1 = 0; 562 high1 = nrow1; 563 lastcol2 = -1; 564 rp2 = bj + bi[row]; 565 ap2 = ba + bi[row]; 566 rmax2 = bimax[row]; 567 nrow2 = bilen[row]; 568 low2 = 0; 569 high2 = nrow2; 570 571 for (j=0; j<n; j++) { 572 if (roworiented) value = v[i*n+j]; 573 else value = v[i+j*m]; 574 if (in[j] >= cstart && in[j] < cend) { 575 col = in[j] - cstart; 576 nonew = a->nonew; 577 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 578 MatSetValues_SeqAIJ_A_Private(row,col,value,addv,im[i],in[j]); 579 } else if (in[j] < 0) continue; 580 #if defined(PETSC_USE_DEBUG) 581 else if (in[j] >= mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap->N-1); 582 #endif 583 else { 584 if (mat->was_assembled) { 585 if (!aij->colmap) { 586 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 587 } 588 #if defined(PETSC_USE_CTABLE) 589 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 590 col--; 591 #else 592 col = aij->colmap[in[j]] - 1; 593 #endif 594 if (col < 0 && !((Mat_SeqAIJ*)(aij->B->data))->nonew) { 595 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 596 col = in[j]; 597 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 598 B = aij->B; 599 b = (Mat_SeqAIJ*)B->data; 600 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; ba = b->a; 601 rp2 = bj + bi[row]; 602 ap2 = ba + bi[row]; 603 rmax2 = bimax[row]; 604 nrow2 = bilen[row]; 605 low2 = 0; 606 high2 = nrow2; 607 bm = aij->B->rmap->n; 608 ba = b->a; 609 } else if (col < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at global row/column (%D, %D) into matrix", im[i], in[j]); 610 } else col = in[j]; 611 nonew = b->nonew; 612 MatSetValues_SeqAIJ_B_Private(row,col,value,addv,im[i],in[j]); 613 } 614 } 615 } else { 616 if (mat->nooffprocentries) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Setting off process row %D even though MatSetOption(,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) was set",im[i]); 617 if (!aij->donotstash) { 618 mat->assembled = PETSC_FALSE; 619 if (roworiented) { 620 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 621 } else { 622 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 623 } 624 } 625 } 626 } 627 PetscFunctionReturn(0); 628 } 629 630 PetscErrorCode MatGetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[]) 631 { 632 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 633 PetscErrorCode ierr; 634 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 635 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 636 637 PetscFunctionBegin; 638 for (i=0; i<m; i++) { 639 if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]);*/ 640 if (idxm[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",idxm[i],mat->rmap->N-1); 641 if (idxm[i] >= rstart && idxm[i] < rend) { 642 row = idxm[i] - rstart; 643 for (j=0; j<n; j++) { 644 if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",idxn[j]); */ 645 if (idxn[j] >= mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",idxn[j],mat->cmap->N-1); 646 if (idxn[j] >= cstart && idxn[j] < cend) { 647 col = idxn[j] - cstart; 648 ierr = MatGetValues(aij->A,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 649 } else { 650 if (!aij->colmap) { 651 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 652 } 653 #if defined(PETSC_USE_CTABLE) 654 ierr = PetscTableFind(aij->colmap,idxn[j]+1,&col);CHKERRQ(ierr); 655 col--; 656 #else 657 col = aij->colmap[idxn[j]] - 1; 658 #endif 659 if ((col < 0) || (aij->garray[col] != idxn[j])) *(v+i*n+j) = 0.0; 660 else { 661 ierr = MatGetValues(aij->B,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 662 } 663 } 664 } 665 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only local values currently supported"); 666 } 667 PetscFunctionReturn(0); 668 } 669 670 extern PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat,Vec,Vec); 671 672 PetscErrorCode MatAssemblyBegin_MPIAIJ(Mat mat,MatAssemblyType mode) 673 { 674 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 675 PetscErrorCode ierr; 676 PetscInt nstash,reallocs; 677 678 PetscFunctionBegin; 679 if (aij->donotstash || mat->nooffprocentries) PetscFunctionReturn(0); 680 681 ierr = MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap->range);CHKERRQ(ierr); 682 ierr = MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);CHKERRQ(ierr); 683 ierr = PetscInfo2(aij->A,"Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);CHKERRQ(ierr); 684 PetscFunctionReturn(0); 685 } 686 687 PetscErrorCode MatAssemblyEnd_MPIAIJ(Mat mat,MatAssemblyType mode) 688 { 689 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 690 Mat_SeqAIJ *a = (Mat_SeqAIJ*)aij->A->data; 691 PetscErrorCode ierr; 692 PetscMPIInt n; 693 PetscInt i,j,rstart,ncols,flg; 694 PetscInt *row,*col; 695 PetscBool other_disassembled; 696 PetscScalar *val; 697 698 /* do not use 'b = (Mat_SeqAIJ*)aij->B->data' as B can be reset in disassembly */ 699 700 PetscFunctionBegin; 701 if (!aij->donotstash && !mat->nooffprocentries) { 702 while (1) { 703 ierr = MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);CHKERRQ(ierr); 704 if (!flg) break; 705 706 for (i=0; i<n; ) { 707 /* Now identify the consecutive vals belonging to the same row */ 708 for (j=i,rstart=row[j]; j<n; j++) { 709 if (row[j] != rstart) break; 710 } 711 if (j < n) ncols = j-i; 712 else ncols = n-i; 713 /* Now assemble all these values with a single function call */ 714 ierr = MatSetValues_MPIAIJ(mat,1,row+i,ncols,col+i,val+i,mat->insertmode);CHKERRQ(ierr); 715 716 i = j; 717 } 718 } 719 ierr = MatStashScatterEnd_Private(&mat->stash);CHKERRQ(ierr); 720 } 721 ierr = MatAssemblyBegin(aij->A,mode);CHKERRQ(ierr); 722 ierr = MatAssemblyEnd(aij->A,mode);CHKERRQ(ierr); 723 724 /* determine if any processor has disassembled, if so we must 725 also disassemble ourselfs, in order that we may reassemble. */ 726 /* 727 if nonzero structure of submatrix B cannot change then we know that 728 no processor disassembled thus we can skip this stuff 729 */ 730 if (!((Mat_SeqAIJ*)aij->B->data)->nonew) { 731 ierr = MPIU_Allreduce(&mat->was_assembled,&other_disassembled,1,MPIU_BOOL,MPI_PROD,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 732 if (mat->was_assembled && !other_disassembled) { 733 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 734 } 735 } 736 if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) { 737 ierr = MatSetUpMultiply_MPIAIJ(mat);CHKERRQ(ierr); 738 } 739 ierr = MatSetOption(aij->B,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 740 ierr = MatAssemblyBegin(aij->B,mode);CHKERRQ(ierr); 741 ierr = MatAssemblyEnd(aij->B,mode);CHKERRQ(ierr); 742 743 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 744 745 aij->rowvalues = 0; 746 747 ierr = VecDestroy(&aij->diag);CHKERRQ(ierr); 748 if (a->inode.size) mat->ops->multdiagonalblock = MatMultDiagonalBlock_MPIAIJ; 749 750 /* if no new nonzero locations are allowed in matrix then only set the matrix state the first time through */ 751 if ((!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) || !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 752 PetscObjectState state = aij->A->nonzerostate + aij->B->nonzerostate; 753 ierr = MPIU_Allreduce(&state,&mat->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 754 } 755 PetscFunctionReturn(0); 756 } 757 758 PetscErrorCode MatZeroEntries_MPIAIJ(Mat A) 759 { 760 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 761 PetscErrorCode ierr; 762 763 PetscFunctionBegin; 764 ierr = MatZeroEntries(l->A);CHKERRQ(ierr); 765 ierr = MatZeroEntries(l->B);CHKERRQ(ierr); 766 PetscFunctionReturn(0); 767 } 768 769 PetscErrorCode MatZeroRows_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 770 { 771 Mat_MPIAIJ *mat = (Mat_MPIAIJ *) A->data; 772 PetscInt *lrows; 773 PetscInt r, len; 774 PetscErrorCode ierr; 775 776 PetscFunctionBegin; 777 /* get locally owned rows */ 778 ierr = MatZeroRowsMapLocal_Private(A,N,rows,&len,&lrows);CHKERRQ(ierr); 779 /* fix right hand side if needed */ 780 if (x && b) { 781 const PetscScalar *xx; 782 PetscScalar *bb; 783 784 ierr = VecGetArrayRead(x, &xx);CHKERRQ(ierr); 785 ierr = VecGetArray(b, &bb);CHKERRQ(ierr); 786 for (r = 0; r < len; ++r) bb[lrows[r]] = diag*xx[lrows[r]]; 787 ierr = VecRestoreArrayRead(x, &xx);CHKERRQ(ierr); 788 ierr = VecRestoreArray(b, &bb);CHKERRQ(ierr); 789 } 790 /* Must zero l->B before l->A because the (diag) case below may put values into l->B*/ 791 ierr = MatZeroRows(mat->B, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 792 if (A->congruentlayouts == -1) { /* first time we compare rows and cols layouts */ 793 PetscBool cong; 794 ierr = PetscLayoutCompare(A->rmap,A->cmap,&cong);CHKERRQ(ierr); 795 if (cong) A->congruentlayouts = 1; 796 else A->congruentlayouts = 0; 797 } 798 if ((diag != 0.0) && A->congruentlayouts) { 799 ierr = MatZeroRows(mat->A, len, lrows, diag, NULL, NULL);CHKERRQ(ierr); 800 } else if (diag != 0.0) { 801 ierr = MatZeroRows(mat->A, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 802 if (((Mat_SeqAIJ *) mat->A->data)->nonew) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "MatZeroRows() on rectangular matrices cannot be used with the Mat options\nMAT_NEW_NONZERO_LOCATIONS,MAT_NEW_NONZERO_LOCATION_ERR,MAT_NEW_NONZERO_ALLOCATION_ERR"); 803 for (r = 0; r < len; ++r) { 804 const PetscInt row = lrows[r] + A->rmap->rstart; 805 ierr = MatSetValues(A, 1, &row, 1, &row, &diag, INSERT_VALUES);CHKERRQ(ierr); 806 } 807 ierr = MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 808 ierr = MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 809 } else { 810 ierr = MatZeroRows(mat->A, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 811 } 812 ierr = PetscFree(lrows);CHKERRQ(ierr); 813 814 /* only change matrix nonzero state if pattern was allowed to be changed */ 815 if (!((Mat_SeqAIJ*)(mat->A->data))->keepnonzeropattern) { 816 PetscObjectState state = mat->A->nonzerostate + mat->B->nonzerostate; 817 ierr = MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 818 } 819 PetscFunctionReturn(0); 820 } 821 822 PetscErrorCode MatZeroRowsColumns_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 823 { 824 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 825 PetscErrorCode ierr; 826 PetscMPIInt n = A->rmap->n; 827 PetscInt i,j,r,m,p = 0,len = 0; 828 PetscInt *lrows,*owners = A->rmap->range; 829 PetscSFNode *rrows; 830 PetscSF sf; 831 const PetscScalar *xx; 832 PetscScalar *bb,*mask; 833 Vec xmask,lmask; 834 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)l->B->data; 835 const PetscInt *aj, *ii,*ridx; 836 PetscScalar *aa; 837 838 PetscFunctionBegin; 839 /* Create SF where leaves are input rows and roots are owned rows */ 840 ierr = PetscMalloc1(n, &lrows);CHKERRQ(ierr); 841 for (r = 0; r < n; ++r) lrows[r] = -1; 842 ierr = PetscMalloc1(N, &rrows);CHKERRQ(ierr); 843 for (r = 0; r < N; ++r) { 844 const PetscInt idx = rows[r]; 845 if (idx < 0 || A->rmap->N <= idx) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range [0,%D)",idx,A->rmap->N); 846 if (idx < owners[p] || owners[p+1] <= idx) { /* short-circuit the search if the last p owns this row too */ 847 ierr = PetscLayoutFindOwner(A->rmap,idx,&p);CHKERRQ(ierr); 848 } 849 rrows[r].rank = p; 850 rrows[r].index = rows[r] - owners[p]; 851 } 852 ierr = PetscSFCreate(PetscObjectComm((PetscObject) A), &sf);CHKERRQ(ierr); 853 ierr = PetscSFSetGraph(sf, n, N, NULL, PETSC_OWN_POINTER, rrows, PETSC_OWN_POINTER);CHKERRQ(ierr); 854 /* Collect flags for rows to be zeroed */ 855 ierr = PetscSFReduceBegin(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);CHKERRQ(ierr); 856 ierr = PetscSFReduceEnd(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);CHKERRQ(ierr); 857 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 858 /* Compress and put in row numbers */ 859 for (r = 0; r < n; ++r) if (lrows[r] >= 0) lrows[len++] = r; 860 /* zero diagonal part of matrix */ 861 ierr = MatZeroRowsColumns(l->A,len,lrows,diag,x,b);CHKERRQ(ierr); 862 /* handle off diagonal part of matrix */ 863 ierr = MatCreateVecs(A,&xmask,NULL);CHKERRQ(ierr); 864 ierr = VecDuplicate(l->lvec,&lmask);CHKERRQ(ierr); 865 ierr = VecGetArray(xmask,&bb);CHKERRQ(ierr); 866 for (i=0; i<len; i++) bb[lrows[i]] = 1; 867 ierr = VecRestoreArray(xmask,&bb);CHKERRQ(ierr); 868 ierr = VecScatterBegin(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 869 ierr = VecScatterEnd(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 870 ierr = VecDestroy(&xmask);CHKERRQ(ierr); 871 if (x) { 872 ierr = VecScatterBegin(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 873 ierr = VecScatterEnd(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 874 ierr = VecGetArrayRead(l->lvec,&xx);CHKERRQ(ierr); 875 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 876 } 877 ierr = VecGetArray(lmask,&mask);CHKERRQ(ierr); 878 /* remove zeroed rows of off diagonal matrix */ 879 ii = aij->i; 880 for (i=0; i<len; i++) { 881 ierr = PetscMemzero(aij->a + ii[lrows[i]],(ii[lrows[i]+1] - ii[lrows[i]])*sizeof(PetscScalar));CHKERRQ(ierr); 882 } 883 /* loop over all elements of off process part of matrix zeroing removed columns*/ 884 if (aij->compressedrow.use) { 885 m = aij->compressedrow.nrows; 886 ii = aij->compressedrow.i; 887 ridx = aij->compressedrow.rindex; 888 for (i=0; i<m; i++) { 889 n = ii[i+1] - ii[i]; 890 aj = aij->j + ii[i]; 891 aa = aij->a + ii[i]; 892 893 for (j=0; j<n; j++) { 894 if (PetscAbsScalar(mask[*aj])) { 895 if (b) bb[*ridx] -= *aa*xx[*aj]; 896 *aa = 0.0; 897 } 898 aa++; 899 aj++; 900 } 901 ridx++; 902 } 903 } else { /* do not use compressed row format */ 904 m = l->B->rmap->n; 905 for (i=0; i<m; i++) { 906 n = ii[i+1] - ii[i]; 907 aj = aij->j + ii[i]; 908 aa = aij->a + ii[i]; 909 for (j=0; j<n; j++) { 910 if (PetscAbsScalar(mask[*aj])) { 911 if (b) bb[i] -= *aa*xx[*aj]; 912 *aa = 0.0; 913 } 914 aa++; 915 aj++; 916 } 917 } 918 } 919 if (x) { 920 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 921 ierr = VecRestoreArrayRead(l->lvec,&xx);CHKERRQ(ierr); 922 } 923 ierr = VecRestoreArray(lmask,&mask);CHKERRQ(ierr); 924 ierr = VecDestroy(&lmask);CHKERRQ(ierr); 925 ierr = PetscFree(lrows);CHKERRQ(ierr); 926 927 /* only change matrix nonzero state if pattern was allowed to be changed */ 928 if (!((Mat_SeqAIJ*)(l->A->data))->keepnonzeropattern) { 929 PetscObjectState state = l->A->nonzerostate + l->B->nonzerostate; 930 ierr = MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 931 } 932 PetscFunctionReturn(0); 933 } 934 935 PetscErrorCode MatMult_MPIAIJ(Mat A,Vec xx,Vec yy) 936 { 937 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 938 PetscErrorCode ierr; 939 PetscInt nt; 940 941 PetscFunctionBegin; 942 ierr = VecGetLocalSize(xx,&nt);CHKERRQ(ierr); 943 if (nt != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Incompatible partition of A (%D) and xx (%D)",A->cmap->n,nt); 944 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 945 ierr = (*a->A->ops->mult)(a->A,xx,yy);CHKERRQ(ierr); 946 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 947 ierr = (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);CHKERRQ(ierr); 948 PetscFunctionReturn(0); 949 } 950 951 PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat A,Vec bb,Vec xx) 952 { 953 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 954 PetscErrorCode ierr; 955 956 PetscFunctionBegin; 957 ierr = MatMultDiagonalBlock(a->A,bb,xx);CHKERRQ(ierr); 958 PetscFunctionReturn(0); 959 } 960 961 PetscErrorCode MatMultAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 962 { 963 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 964 PetscErrorCode ierr; 965 966 PetscFunctionBegin; 967 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 968 ierr = (*a->A->ops->multadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 969 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 970 ierr = (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);CHKERRQ(ierr); 971 PetscFunctionReturn(0); 972 } 973 974 PetscErrorCode MatMultTranspose_MPIAIJ(Mat A,Vec xx,Vec yy) 975 { 976 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 977 PetscErrorCode ierr; 978 PetscBool merged; 979 980 PetscFunctionBegin; 981 ierr = VecScatterGetMerged(a->Mvctx,&merged);CHKERRQ(ierr); 982 /* do nondiagonal part */ 983 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 984 if (!merged) { 985 /* send it on its way */ 986 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 987 /* do local part */ 988 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 989 /* receive remote parts: note this assumes the values are not actually */ 990 /* added in yy until the next line, */ 991 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 992 } else { 993 /* do local part */ 994 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 995 /* send it on its way */ 996 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 997 /* values actually were received in the Begin() but we need to call this nop */ 998 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 999 } 1000 PetscFunctionReturn(0); 1001 } 1002 1003 PetscErrorCode MatIsTranspose_MPIAIJ(Mat Amat,Mat Bmat,PetscReal tol,PetscBool *f) 1004 { 1005 MPI_Comm comm; 1006 Mat_MPIAIJ *Aij = (Mat_MPIAIJ*) Amat->data, *Bij; 1007 Mat Adia = Aij->A, Bdia, Aoff,Boff,*Aoffs,*Boffs; 1008 IS Me,Notme; 1009 PetscErrorCode ierr; 1010 PetscInt M,N,first,last,*notme,i; 1011 PetscMPIInt size; 1012 1013 PetscFunctionBegin; 1014 /* Easy test: symmetric diagonal block */ 1015 Bij = (Mat_MPIAIJ*) Bmat->data; Bdia = Bij->A; 1016 ierr = MatIsTranspose(Adia,Bdia,tol,f);CHKERRQ(ierr); 1017 if (!*f) PetscFunctionReturn(0); 1018 ierr = PetscObjectGetComm((PetscObject)Amat,&comm);CHKERRQ(ierr); 1019 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1020 if (size == 1) PetscFunctionReturn(0); 1021 1022 /* Hard test: off-diagonal block. This takes a MatCreateSubMatrix. */ 1023 ierr = MatGetSize(Amat,&M,&N);CHKERRQ(ierr); 1024 ierr = MatGetOwnershipRange(Amat,&first,&last);CHKERRQ(ierr); 1025 ierr = PetscMalloc1(N-last+first,¬me);CHKERRQ(ierr); 1026 for (i=0; i<first; i++) notme[i] = i; 1027 for (i=last; i<M; i++) notme[i-last+first] = i; 1028 ierr = ISCreateGeneral(MPI_COMM_SELF,N-last+first,notme,PETSC_COPY_VALUES,&Notme);CHKERRQ(ierr); 1029 ierr = ISCreateStride(MPI_COMM_SELF,last-first,first,1,&Me);CHKERRQ(ierr); 1030 ierr = MatCreateSubMatrices(Amat,1,&Me,&Notme,MAT_INITIAL_MATRIX,&Aoffs);CHKERRQ(ierr); 1031 Aoff = Aoffs[0]; 1032 ierr = MatCreateSubMatrices(Bmat,1,&Notme,&Me,MAT_INITIAL_MATRIX,&Boffs);CHKERRQ(ierr); 1033 Boff = Boffs[0]; 1034 ierr = MatIsTranspose(Aoff,Boff,tol,f);CHKERRQ(ierr); 1035 ierr = MatDestroyMatrices(1,&Aoffs);CHKERRQ(ierr); 1036 ierr = MatDestroyMatrices(1,&Boffs);CHKERRQ(ierr); 1037 ierr = ISDestroy(&Me);CHKERRQ(ierr); 1038 ierr = ISDestroy(&Notme);CHKERRQ(ierr); 1039 ierr = PetscFree(notme);CHKERRQ(ierr); 1040 PetscFunctionReturn(0); 1041 } 1042 1043 PetscErrorCode MatMultTransposeAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1044 { 1045 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1046 PetscErrorCode ierr; 1047 1048 PetscFunctionBegin; 1049 /* do nondiagonal part */ 1050 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 1051 /* send it on its way */ 1052 ierr = VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1053 /* do local part */ 1054 ierr = (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 1055 /* receive remote parts */ 1056 ierr = VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1057 PetscFunctionReturn(0); 1058 } 1059 1060 /* 1061 This only works correctly for square matrices where the subblock A->A is the 1062 diagonal block 1063 */ 1064 PetscErrorCode MatGetDiagonal_MPIAIJ(Mat A,Vec v) 1065 { 1066 PetscErrorCode ierr; 1067 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1068 1069 PetscFunctionBegin; 1070 if (A->rmap->N != A->cmap->N) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); 1071 if (A->rmap->rstart != A->cmap->rstart || A->rmap->rend != A->cmap->rend) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"row partition must equal col partition"); 1072 ierr = MatGetDiagonal(a->A,v);CHKERRQ(ierr); 1073 PetscFunctionReturn(0); 1074 } 1075 1076 PetscErrorCode MatScale_MPIAIJ(Mat A,PetscScalar aa) 1077 { 1078 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1079 PetscErrorCode ierr; 1080 1081 PetscFunctionBegin; 1082 ierr = MatScale(a->A,aa);CHKERRQ(ierr); 1083 ierr = MatScale(a->B,aa);CHKERRQ(ierr); 1084 PetscFunctionReturn(0); 1085 } 1086 1087 PetscErrorCode MatDestroy_MPIAIJ(Mat mat) 1088 { 1089 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1090 PetscErrorCode ierr; 1091 1092 PetscFunctionBegin; 1093 #if defined(PETSC_USE_LOG) 1094 PetscLogObjectState((PetscObject)mat,"Rows=%D, Cols=%D",mat->rmap->N,mat->cmap->N); 1095 #endif 1096 ierr = MatStashDestroy_Private(&mat->stash);CHKERRQ(ierr); 1097 ierr = VecDestroy(&aij->diag);CHKERRQ(ierr); 1098 ierr = MatDestroy(&aij->A);CHKERRQ(ierr); 1099 ierr = MatDestroy(&aij->B);CHKERRQ(ierr); 1100 #if defined(PETSC_USE_CTABLE) 1101 ierr = PetscTableDestroy(&aij->colmap);CHKERRQ(ierr); 1102 #else 1103 ierr = PetscFree(aij->colmap);CHKERRQ(ierr); 1104 #endif 1105 ierr = PetscFree(aij->garray);CHKERRQ(ierr); 1106 ierr = VecDestroy(&aij->lvec);CHKERRQ(ierr); 1107 ierr = VecScatterDestroy(&aij->Mvctx);CHKERRQ(ierr); 1108 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 1109 ierr = PetscFree(aij->ld);CHKERRQ(ierr); 1110 ierr = PetscFree(mat->data);CHKERRQ(ierr); 1111 1112 ierr = PetscObjectChangeTypeName((PetscObject)mat,0);CHKERRQ(ierr); 1113 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C",NULL);CHKERRQ(ierr); 1114 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C",NULL);CHKERRQ(ierr); 1115 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatIsTranspose_C",NULL);CHKERRQ(ierr); 1116 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocation_C",NULL);CHKERRQ(ierr); 1117 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocationCSR_C",NULL);CHKERRQ(ierr); 1118 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatDiagonalScaleLocal_C",NULL);CHKERRQ(ierr); 1119 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_mpisbaij_C",NULL);CHKERRQ(ierr); 1120 #if defined(PETSC_HAVE_ELEMENTAL) 1121 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_elemental_C",NULL);CHKERRQ(ierr); 1122 #endif 1123 #if defined(PETSC_HAVE_HYPRE) 1124 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_hypre_C",NULL);CHKERRQ(ierr); 1125 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMatMatMult_transpose_mpiaij_mpiaij_C",NULL);CHKERRQ(ierr); 1126 #endif 1127 PetscFunctionReturn(0); 1128 } 1129 1130 PetscErrorCode MatView_MPIAIJ_Binary(Mat mat,PetscViewer viewer) 1131 { 1132 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1133 Mat_SeqAIJ *A = (Mat_SeqAIJ*)aij->A->data; 1134 Mat_SeqAIJ *B = (Mat_SeqAIJ*)aij->B->data; 1135 PetscErrorCode ierr; 1136 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 1137 int fd; 1138 PetscInt nz,header[4],*row_lengths,*range=0,rlen,i; 1139 PetscInt nzmax,*column_indices,j,k,col,*garray = aij->garray,cnt,cstart = mat->cmap->rstart,rnz = 0; 1140 PetscScalar *column_values; 1141 PetscInt message_count,flowcontrolcount; 1142 FILE *file; 1143 1144 PetscFunctionBegin; 1145 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);CHKERRQ(ierr); 1146 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr); 1147 nz = A->nz + B->nz; 1148 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 1149 if (!rank) { 1150 header[0] = MAT_FILE_CLASSID; 1151 header[1] = mat->rmap->N; 1152 header[2] = mat->cmap->N; 1153 1154 ierr = MPI_Reduce(&nz,&header[3],1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1155 ierr = PetscBinaryWrite(fd,header,4,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1156 /* get largest number of rows any processor has */ 1157 rlen = mat->rmap->n; 1158 range = mat->rmap->range; 1159 for (i=1; i<size; i++) rlen = PetscMax(rlen,range[i+1] - range[i]); 1160 } else { 1161 ierr = MPI_Reduce(&nz,0,1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1162 rlen = mat->rmap->n; 1163 } 1164 1165 /* load up the local row counts */ 1166 ierr = PetscMalloc1(rlen+1,&row_lengths);CHKERRQ(ierr); 1167 for (i=0; i<mat->rmap->n; i++) row_lengths[i] = A->i[i+1] - A->i[i] + B->i[i+1] - B->i[i]; 1168 1169 /* store the row lengths to the file */ 1170 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1171 if (!rank) { 1172 ierr = PetscBinaryWrite(fd,row_lengths,mat->rmap->n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1173 for (i=1; i<size; i++) { 1174 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1175 rlen = range[i+1] - range[i]; 1176 ierr = MPIULong_Recv(row_lengths,rlen,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1177 ierr = PetscBinaryWrite(fd,row_lengths,rlen,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1178 } 1179 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1180 } else { 1181 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1182 ierr = MPIULong_Send(row_lengths,mat->rmap->n,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1183 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1184 } 1185 ierr = PetscFree(row_lengths);CHKERRQ(ierr); 1186 1187 /* load up the local column indices */ 1188 nzmax = nz; /* th processor needs space a largest processor needs */ 1189 ierr = MPI_Reduce(&nz,&nzmax,1,MPIU_INT,MPI_MAX,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1190 ierr = PetscMalloc1(nzmax+1,&column_indices);CHKERRQ(ierr); 1191 cnt = 0; 1192 for (i=0; i<mat->rmap->n; i++) { 1193 for (j=B->i[i]; j<B->i[i+1]; j++) { 1194 if ((col = garray[B->j[j]]) > cstart) break; 1195 column_indices[cnt++] = col; 1196 } 1197 for (k=A->i[i]; k<A->i[i+1]; k++) column_indices[cnt++] = A->j[k] + cstart; 1198 for (; j<B->i[i+1]; j++) column_indices[cnt++] = garray[B->j[j]]; 1199 } 1200 if (cnt != A->nz + B->nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: cnt = %D nz = %D",cnt,A->nz+B->nz); 1201 1202 /* store the column indices to the file */ 1203 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1204 if (!rank) { 1205 MPI_Status status; 1206 ierr = PetscBinaryWrite(fd,column_indices,nz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1207 for (i=1; i<size; i++) { 1208 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1209 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);CHKERRQ(ierr); 1210 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1211 ierr = MPIULong_Recv(column_indices,rnz,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1212 ierr = PetscBinaryWrite(fd,column_indices,rnz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1213 } 1214 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1215 } else { 1216 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1217 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1218 ierr = MPIULong_Send(column_indices,nz,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1219 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1220 } 1221 ierr = PetscFree(column_indices);CHKERRQ(ierr); 1222 1223 /* load up the local column values */ 1224 ierr = PetscMalloc1(nzmax+1,&column_values);CHKERRQ(ierr); 1225 cnt = 0; 1226 for (i=0; i<mat->rmap->n; i++) { 1227 for (j=B->i[i]; j<B->i[i+1]; j++) { 1228 if (garray[B->j[j]] > cstart) break; 1229 column_values[cnt++] = B->a[j]; 1230 } 1231 for (k=A->i[i]; k<A->i[i+1]; k++) column_values[cnt++] = A->a[k]; 1232 for (; j<B->i[i+1]; j++) column_values[cnt++] = B->a[j]; 1233 } 1234 if (cnt != A->nz + B->nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Internal PETSc error: cnt = %D nz = %D",cnt,A->nz+B->nz); 1235 1236 /* store the column values to the file */ 1237 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1238 if (!rank) { 1239 MPI_Status status; 1240 ierr = PetscBinaryWrite(fd,column_values,nz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1241 for (i=1; i<size; i++) { 1242 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1243 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);CHKERRQ(ierr); 1244 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1245 ierr = MPIULong_Recv(column_values,rnz,MPIU_SCALAR,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1246 ierr = PetscBinaryWrite(fd,column_values,rnz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1247 } 1248 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1249 } else { 1250 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1251 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1252 ierr = MPIULong_Send(column_values,nz,MPIU_SCALAR,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1253 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1254 } 1255 ierr = PetscFree(column_values);CHKERRQ(ierr); 1256 1257 ierr = PetscViewerBinaryGetInfoPointer(viewer,&file);CHKERRQ(ierr); 1258 if (file) fprintf(file,"-matload_block_size %d\n",(int)PetscAbs(mat->rmap->bs)); 1259 PetscFunctionReturn(0); 1260 } 1261 1262 #include <petscdraw.h> 1263 PetscErrorCode MatView_MPIAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer) 1264 { 1265 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1266 PetscErrorCode ierr; 1267 PetscMPIInt rank = aij->rank,size = aij->size; 1268 PetscBool isdraw,iascii,isbinary; 1269 PetscViewer sviewer; 1270 PetscViewerFormat format; 1271 1272 PetscFunctionBegin; 1273 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1274 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1275 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1276 if (iascii) { 1277 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 1278 if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 1279 MatInfo info; 1280 PetscBool inodes; 1281 1282 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);CHKERRQ(ierr); 1283 ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr); 1284 ierr = MatInodeGetInodeSizes(aij->A,NULL,(PetscInt**)&inodes,NULL);CHKERRQ(ierr); 1285 ierr = PetscViewerASCIIPushSynchronized(viewer);CHKERRQ(ierr); 1286 if (!inodes) { 1287 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, not using I-node routines\n", 1288 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1289 } else { 1290 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, using I-node routines\n", 1291 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1292 } 1293 ierr = MatGetInfo(aij->A,MAT_LOCAL,&info);CHKERRQ(ierr); 1294 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1295 ierr = MatGetInfo(aij->B,MAT_LOCAL,&info);CHKERRQ(ierr); 1296 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1297 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 1298 ierr = PetscViewerASCIIPopSynchronized(viewer);CHKERRQ(ierr); 1299 ierr = PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");CHKERRQ(ierr); 1300 ierr = VecScatterView(aij->Mvctx,viewer);CHKERRQ(ierr); 1301 PetscFunctionReturn(0); 1302 } else if (format == PETSC_VIEWER_ASCII_INFO) { 1303 PetscInt inodecount,inodelimit,*inodes; 1304 ierr = MatInodeGetInodeSizes(aij->A,&inodecount,&inodes,&inodelimit);CHKERRQ(ierr); 1305 if (inodes) { 1306 ierr = PetscViewerASCIIPrintf(viewer,"using I-node (on process 0) routines: found %D nodes, limit used is %D\n",inodecount,inodelimit);CHKERRQ(ierr); 1307 } else { 1308 ierr = PetscViewerASCIIPrintf(viewer,"not using I-node (on process 0) routines\n");CHKERRQ(ierr); 1309 } 1310 PetscFunctionReturn(0); 1311 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) { 1312 PetscFunctionReturn(0); 1313 } 1314 } else if (isbinary) { 1315 if (size == 1) { 1316 ierr = PetscObjectSetName((PetscObject)aij->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1317 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 1318 } else { 1319 ierr = MatView_MPIAIJ_Binary(mat,viewer);CHKERRQ(ierr); 1320 } 1321 PetscFunctionReturn(0); 1322 } else if (isdraw) { 1323 PetscDraw draw; 1324 PetscBool isnull; 1325 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 1326 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); 1327 if (isnull) PetscFunctionReturn(0); 1328 } 1329 1330 { 1331 /* assemble the entire matrix onto first processor. */ 1332 Mat A; 1333 Mat_SeqAIJ *Aloc; 1334 PetscInt M = mat->rmap->N,N = mat->cmap->N,m,*ai,*aj,row,*cols,i,*ct; 1335 MatScalar *a; 1336 1337 ierr = MatCreate(PetscObjectComm((PetscObject)mat),&A);CHKERRQ(ierr); 1338 if (!rank) { 1339 ierr = MatSetSizes(A,M,N,M,N);CHKERRQ(ierr); 1340 } else { 1341 ierr = MatSetSizes(A,0,0,M,N);CHKERRQ(ierr); 1342 } 1343 /* This is just a temporary matrix, so explicitly using MATMPIAIJ is probably best */ 1344 ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr); 1345 ierr = MatMPIAIJSetPreallocation(A,0,NULL,0,NULL);CHKERRQ(ierr); 1346 ierr = MatSetOption(A,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr); 1347 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)A);CHKERRQ(ierr); 1348 1349 /* copy over the A part */ 1350 Aloc = (Mat_SeqAIJ*)aij->A->data; 1351 m = aij->A->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1352 row = mat->rmap->rstart; 1353 for (i=0; i<ai[m]; i++) aj[i] += mat->cmap->rstart; 1354 for (i=0; i<m; i++) { 1355 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],aj,a,INSERT_VALUES);CHKERRQ(ierr); 1356 row++; 1357 a += ai[i+1]-ai[i]; aj += ai[i+1]-ai[i]; 1358 } 1359 aj = Aloc->j; 1360 for (i=0; i<ai[m]; i++) aj[i] -= mat->cmap->rstart; 1361 1362 /* copy over the B part */ 1363 Aloc = (Mat_SeqAIJ*)aij->B->data; 1364 m = aij->B->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1365 row = mat->rmap->rstart; 1366 ierr = PetscMalloc1(ai[m]+1,&cols);CHKERRQ(ierr); 1367 ct = cols; 1368 for (i=0; i<ai[m]; i++) cols[i] = aij->garray[aj[i]]; 1369 for (i=0; i<m; i++) { 1370 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],cols,a,INSERT_VALUES);CHKERRQ(ierr); 1371 row++; 1372 a += ai[i+1]-ai[i]; cols += ai[i+1]-ai[i]; 1373 } 1374 ierr = PetscFree(ct);CHKERRQ(ierr); 1375 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1376 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1377 /* 1378 Everyone has to call to draw the matrix since the graphics waits are 1379 synchronized across all processors that share the PetscDraw object 1380 */ 1381 ierr = PetscViewerGetSubViewer(viewer,PETSC_COMM_SELF,&sviewer);CHKERRQ(ierr); 1382 if (!rank) { 1383 ierr = PetscObjectSetName((PetscObject)((Mat_MPIAIJ*)(A->data))->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1384 ierr = MatView_SeqAIJ(((Mat_MPIAIJ*)(A->data))->A,sviewer);CHKERRQ(ierr); 1385 } 1386 ierr = PetscViewerRestoreSubViewer(viewer,PETSC_COMM_SELF,&sviewer);CHKERRQ(ierr); 1387 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 1388 ierr = MatDestroy(&A);CHKERRQ(ierr); 1389 } 1390 PetscFunctionReturn(0); 1391 } 1392 1393 PetscErrorCode MatView_MPIAIJ(Mat mat,PetscViewer viewer) 1394 { 1395 PetscErrorCode ierr; 1396 PetscBool iascii,isdraw,issocket,isbinary; 1397 1398 PetscFunctionBegin; 1399 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1400 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1401 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1402 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERSOCKET,&issocket);CHKERRQ(ierr); 1403 if (iascii || isdraw || isbinary || issocket) { 1404 ierr = MatView_MPIAIJ_ASCIIorDraworSocket(mat,viewer);CHKERRQ(ierr); 1405 } 1406 PetscFunctionReturn(0); 1407 } 1408 1409 PetscErrorCode MatSOR_MPIAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1410 { 1411 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1412 PetscErrorCode ierr; 1413 Vec bb1 = 0; 1414 PetscBool hasop; 1415 1416 PetscFunctionBegin; 1417 if (flag == SOR_APPLY_UPPER) { 1418 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1419 PetscFunctionReturn(0); 1420 } 1421 1422 if (its > 1 || ~flag & SOR_ZERO_INITIAL_GUESS || flag & SOR_EISENSTAT) { 1423 ierr = VecDuplicate(bb,&bb1);CHKERRQ(ierr); 1424 } 1425 1426 if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP) { 1427 if (flag & SOR_ZERO_INITIAL_GUESS) { 1428 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1429 its--; 1430 } 1431 1432 while (its--) { 1433 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1434 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1435 1436 /* update rhs: bb1 = bb - B*x */ 1437 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1438 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1439 1440 /* local sweep */ 1441 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1442 } 1443 } else if (flag & SOR_LOCAL_FORWARD_SWEEP) { 1444 if (flag & SOR_ZERO_INITIAL_GUESS) { 1445 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1446 its--; 1447 } 1448 while (its--) { 1449 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1450 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1451 1452 /* update rhs: bb1 = bb - B*x */ 1453 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1454 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1455 1456 /* local sweep */ 1457 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_FORWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1458 } 1459 } else if (flag & SOR_LOCAL_BACKWARD_SWEEP) { 1460 if (flag & SOR_ZERO_INITIAL_GUESS) { 1461 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1462 its--; 1463 } 1464 while (its--) { 1465 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1466 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1467 1468 /* update rhs: bb1 = bb - B*x */ 1469 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1470 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1471 1472 /* local sweep */ 1473 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_BACKWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1474 } 1475 } else if (flag & SOR_EISENSTAT) { 1476 Vec xx1; 1477 1478 ierr = VecDuplicate(bb,&xx1);CHKERRQ(ierr); 1479 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_BACKWARD_SWEEP),fshift,lits,1,xx);CHKERRQ(ierr); 1480 1481 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1482 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1483 if (!mat->diag) { 1484 ierr = MatCreateVecs(matin,&mat->diag,NULL);CHKERRQ(ierr); 1485 ierr = MatGetDiagonal(matin,mat->diag);CHKERRQ(ierr); 1486 } 1487 ierr = MatHasOperation(matin,MATOP_MULT_DIAGONAL_BLOCK,&hasop);CHKERRQ(ierr); 1488 if (hasop) { 1489 ierr = MatMultDiagonalBlock(matin,xx,bb1);CHKERRQ(ierr); 1490 } else { 1491 ierr = VecPointwiseMult(bb1,mat->diag,xx);CHKERRQ(ierr); 1492 } 1493 ierr = VecAYPX(bb1,(omega-2.0)/omega,bb);CHKERRQ(ierr); 1494 1495 ierr = MatMultAdd(mat->B,mat->lvec,bb1,bb1);CHKERRQ(ierr); 1496 1497 /* local sweep */ 1498 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_FORWARD_SWEEP),fshift,lits,1,xx1);CHKERRQ(ierr); 1499 ierr = VecAXPY(xx,1.0,xx1);CHKERRQ(ierr); 1500 ierr = VecDestroy(&xx1);CHKERRQ(ierr); 1501 } else SETERRQ(PetscObjectComm((PetscObject)matin),PETSC_ERR_SUP,"Parallel SOR not supported"); 1502 1503 ierr = VecDestroy(&bb1);CHKERRQ(ierr); 1504 1505 matin->factorerrortype = mat->A->factorerrortype; 1506 PetscFunctionReturn(0); 1507 } 1508 1509 PetscErrorCode MatPermute_MPIAIJ(Mat A,IS rowp,IS colp,Mat *B) 1510 { 1511 Mat aA,aB,Aperm; 1512 const PetscInt *rwant,*cwant,*gcols,*ai,*bi,*aj,*bj; 1513 PetscScalar *aa,*ba; 1514 PetscInt i,j,m,n,ng,anz,bnz,*dnnz,*onnz,*tdnnz,*tonnz,*rdest,*cdest,*work,*gcdest; 1515 PetscSF rowsf,sf; 1516 IS parcolp = NULL; 1517 PetscBool done; 1518 PetscErrorCode ierr; 1519 1520 PetscFunctionBegin; 1521 ierr = MatGetLocalSize(A,&m,&n);CHKERRQ(ierr); 1522 ierr = ISGetIndices(rowp,&rwant);CHKERRQ(ierr); 1523 ierr = ISGetIndices(colp,&cwant);CHKERRQ(ierr); 1524 ierr = PetscMalloc3(PetscMax(m,n),&work,m,&rdest,n,&cdest);CHKERRQ(ierr); 1525 1526 /* Invert row permutation to find out where my rows should go */ 1527 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&rowsf);CHKERRQ(ierr); 1528 ierr = PetscSFSetGraphLayout(rowsf,A->rmap,A->rmap->n,NULL,PETSC_OWN_POINTER,rwant);CHKERRQ(ierr); 1529 ierr = PetscSFSetFromOptions(rowsf);CHKERRQ(ierr); 1530 for (i=0; i<m; i++) work[i] = A->rmap->rstart + i; 1531 ierr = PetscSFReduceBegin(rowsf,MPIU_INT,work,rdest,MPIU_REPLACE);CHKERRQ(ierr); 1532 ierr = PetscSFReduceEnd(rowsf,MPIU_INT,work,rdest,MPIU_REPLACE);CHKERRQ(ierr); 1533 1534 /* Invert column permutation to find out where my columns should go */ 1535 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1536 ierr = PetscSFSetGraphLayout(sf,A->cmap,A->cmap->n,NULL,PETSC_OWN_POINTER,cwant);CHKERRQ(ierr); 1537 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1538 for (i=0; i<n; i++) work[i] = A->cmap->rstart + i; 1539 ierr = PetscSFReduceBegin(sf,MPIU_INT,work,cdest,MPIU_REPLACE);CHKERRQ(ierr); 1540 ierr = PetscSFReduceEnd(sf,MPIU_INT,work,cdest,MPIU_REPLACE);CHKERRQ(ierr); 1541 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1542 1543 ierr = ISRestoreIndices(rowp,&rwant);CHKERRQ(ierr); 1544 ierr = ISRestoreIndices(colp,&cwant);CHKERRQ(ierr); 1545 ierr = MatMPIAIJGetSeqAIJ(A,&aA,&aB,&gcols);CHKERRQ(ierr); 1546 1547 /* Find out where my gcols should go */ 1548 ierr = MatGetSize(aB,NULL,&ng);CHKERRQ(ierr); 1549 ierr = PetscMalloc1(ng,&gcdest);CHKERRQ(ierr); 1550 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1551 ierr = PetscSFSetGraphLayout(sf,A->cmap,ng,NULL,PETSC_OWN_POINTER,gcols);CHKERRQ(ierr); 1552 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1553 ierr = PetscSFBcastBegin(sf,MPIU_INT,cdest,gcdest);CHKERRQ(ierr); 1554 ierr = PetscSFBcastEnd(sf,MPIU_INT,cdest,gcdest);CHKERRQ(ierr); 1555 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1556 1557 ierr = PetscCalloc4(m,&dnnz,m,&onnz,m,&tdnnz,m,&tonnz);CHKERRQ(ierr); 1558 ierr = MatGetRowIJ(aA,0,PETSC_FALSE,PETSC_FALSE,&anz,&ai,&aj,&done);CHKERRQ(ierr); 1559 ierr = MatGetRowIJ(aB,0,PETSC_FALSE,PETSC_FALSE,&bnz,&bi,&bj,&done);CHKERRQ(ierr); 1560 for (i=0; i<m; i++) { 1561 PetscInt row = rdest[i],rowner; 1562 ierr = PetscLayoutFindOwner(A->rmap,row,&rowner);CHKERRQ(ierr); 1563 for (j=ai[i]; j<ai[i+1]; j++) { 1564 PetscInt cowner,col = cdest[aj[j]]; 1565 ierr = PetscLayoutFindOwner(A->cmap,col,&cowner);CHKERRQ(ierr); /* Could build an index for the columns to eliminate this search */ 1566 if (rowner == cowner) dnnz[i]++; 1567 else onnz[i]++; 1568 } 1569 for (j=bi[i]; j<bi[i+1]; j++) { 1570 PetscInt cowner,col = gcdest[bj[j]]; 1571 ierr = PetscLayoutFindOwner(A->cmap,col,&cowner);CHKERRQ(ierr); 1572 if (rowner == cowner) dnnz[i]++; 1573 else onnz[i]++; 1574 } 1575 } 1576 ierr = PetscSFBcastBegin(rowsf,MPIU_INT,dnnz,tdnnz);CHKERRQ(ierr); 1577 ierr = PetscSFBcastEnd(rowsf,MPIU_INT,dnnz,tdnnz);CHKERRQ(ierr); 1578 ierr = PetscSFBcastBegin(rowsf,MPIU_INT,onnz,tonnz);CHKERRQ(ierr); 1579 ierr = PetscSFBcastEnd(rowsf,MPIU_INT,onnz,tonnz);CHKERRQ(ierr); 1580 ierr = PetscSFDestroy(&rowsf);CHKERRQ(ierr); 1581 1582 ierr = MatCreateAIJ(PetscObjectComm((PetscObject)A),A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N,0,tdnnz,0,tonnz,&Aperm);CHKERRQ(ierr); 1583 ierr = MatSeqAIJGetArray(aA,&aa);CHKERRQ(ierr); 1584 ierr = MatSeqAIJGetArray(aB,&ba);CHKERRQ(ierr); 1585 for (i=0; i<m; i++) { 1586 PetscInt *acols = dnnz,*bcols = onnz; /* Repurpose now-unneeded arrays */ 1587 PetscInt j0,rowlen; 1588 rowlen = ai[i+1] - ai[i]; 1589 for (j0=j=0; j<rowlen; j0=j) { /* rowlen could be larger than number of rows m, so sum in batches */ 1590 for ( ; j<PetscMin(rowlen,j0+m); j++) acols[j-j0] = cdest[aj[ai[i]+j]]; 1591 ierr = MatSetValues(Aperm,1,&rdest[i],j-j0,acols,aa+ai[i]+j0,INSERT_VALUES);CHKERRQ(ierr); 1592 } 1593 rowlen = bi[i+1] - bi[i]; 1594 for (j0=j=0; j<rowlen; j0=j) { 1595 for ( ; j<PetscMin(rowlen,j0+m); j++) bcols[j-j0] = gcdest[bj[bi[i]+j]]; 1596 ierr = MatSetValues(Aperm,1,&rdest[i],j-j0,bcols,ba+bi[i]+j0,INSERT_VALUES);CHKERRQ(ierr); 1597 } 1598 } 1599 ierr = MatAssemblyBegin(Aperm,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1600 ierr = MatAssemblyEnd(Aperm,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1601 ierr = MatRestoreRowIJ(aA,0,PETSC_FALSE,PETSC_FALSE,&anz,&ai,&aj,&done);CHKERRQ(ierr); 1602 ierr = MatRestoreRowIJ(aB,0,PETSC_FALSE,PETSC_FALSE,&bnz,&bi,&bj,&done);CHKERRQ(ierr); 1603 ierr = MatSeqAIJRestoreArray(aA,&aa);CHKERRQ(ierr); 1604 ierr = MatSeqAIJRestoreArray(aB,&ba);CHKERRQ(ierr); 1605 ierr = PetscFree4(dnnz,onnz,tdnnz,tonnz);CHKERRQ(ierr); 1606 ierr = PetscFree3(work,rdest,cdest);CHKERRQ(ierr); 1607 ierr = PetscFree(gcdest);CHKERRQ(ierr); 1608 if (parcolp) {ierr = ISDestroy(&colp);CHKERRQ(ierr);} 1609 *B = Aperm; 1610 PetscFunctionReturn(0); 1611 } 1612 1613 PetscErrorCode MatGetGhosts_MPIAIJ(Mat mat,PetscInt *nghosts,const PetscInt *ghosts[]) 1614 { 1615 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1616 PetscErrorCode ierr; 1617 1618 PetscFunctionBegin; 1619 ierr = MatGetSize(aij->B,NULL,nghosts);CHKERRQ(ierr); 1620 if (ghosts) *ghosts = aij->garray; 1621 PetscFunctionReturn(0); 1622 } 1623 1624 PetscErrorCode MatGetInfo_MPIAIJ(Mat matin,MatInfoType flag,MatInfo *info) 1625 { 1626 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1627 Mat A = mat->A,B = mat->B; 1628 PetscErrorCode ierr; 1629 PetscReal isend[5],irecv[5]; 1630 1631 PetscFunctionBegin; 1632 info->block_size = 1.0; 1633 ierr = MatGetInfo(A,MAT_LOCAL,info);CHKERRQ(ierr); 1634 1635 isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded; 1636 isend[3] = info->memory; isend[4] = info->mallocs; 1637 1638 ierr = MatGetInfo(B,MAT_LOCAL,info);CHKERRQ(ierr); 1639 1640 isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded; 1641 isend[3] += info->memory; isend[4] += info->mallocs; 1642 if (flag == MAT_LOCAL) { 1643 info->nz_used = isend[0]; 1644 info->nz_allocated = isend[1]; 1645 info->nz_unneeded = isend[2]; 1646 info->memory = isend[3]; 1647 info->mallocs = isend[4]; 1648 } else if (flag == MAT_GLOBAL_MAX) { 1649 ierr = MPIU_Allreduce(isend,irecv,5,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)matin));CHKERRQ(ierr); 1650 1651 info->nz_used = irecv[0]; 1652 info->nz_allocated = irecv[1]; 1653 info->nz_unneeded = irecv[2]; 1654 info->memory = irecv[3]; 1655 info->mallocs = irecv[4]; 1656 } else if (flag == MAT_GLOBAL_SUM) { 1657 ierr = MPIU_Allreduce(isend,irecv,5,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)matin));CHKERRQ(ierr); 1658 1659 info->nz_used = irecv[0]; 1660 info->nz_allocated = irecv[1]; 1661 info->nz_unneeded = irecv[2]; 1662 info->memory = irecv[3]; 1663 info->mallocs = irecv[4]; 1664 } 1665 info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */ 1666 info->fill_ratio_needed = 0; 1667 info->factor_mallocs = 0; 1668 PetscFunctionReturn(0); 1669 } 1670 1671 PetscErrorCode MatSetOption_MPIAIJ(Mat A,MatOption op,PetscBool flg) 1672 { 1673 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1674 PetscErrorCode ierr; 1675 1676 PetscFunctionBegin; 1677 switch (op) { 1678 case MAT_NEW_NONZERO_LOCATIONS: 1679 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1680 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1681 case MAT_KEEP_NONZERO_PATTERN: 1682 case MAT_NEW_NONZERO_LOCATION_ERR: 1683 case MAT_USE_INODES: 1684 case MAT_IGNORE_ZERO_ENTRIES: 1685 MatCheckPreallocated(A,1); 1686 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1687 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1688 break; 1689 case MAT_ROW_ORIENTED: 1690 MatCheckPreallocated(A,1); 1691 a->roworiented = flg; 1692 1693 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1694 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1695 break; 1696 case MAT_NEW_DIAGONALS: 1697 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1698 break; 1699 case MAT_IGNORE_OFF_PROC_ENTRIES: 1700 a->donotstash = flg; 1701 break; 1702 case MAT_SPD: 1703 A->spd_set = PETSC_TRUE; 1704 A->spd = flg; 1705 if (flg) { 1706 A->symmetric = PETSC_TRUE; 1707 A->structurally_symmetric = PETSC_TRUE; 1708 A->symmetric_set = PETSC_TRUE; 1709 A->structurally_symmetric_set = PETSC_TRUE; 1710 } 1711 break; 1712 case MAT_SYMMETRIC: 1713 MatCheckPreallocated(A,1); 1714 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1715 break; 1716 case MAT_STRUCTURALLY_SYMMETRIC: 1717 MatCheckPreallocated(A,1); 1718 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1719 break; 1720 case MAT_HERMITIAN: 1721 MatCheckPreallocated(A,1); 1722 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1723 break; 1724 case MAT_SYMMETRY_ETERNAL: 1725 MatCheckPreallocated(A,1); 1726 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1727 break; 1728 case MAT_SUBMAT_SINGLEIS: 1729 A->submat_singleis = flg; 1730 break; 1731 default: 1732 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1733 } 1734 PetscFunctionReturn(0); 1735 } 1736 1737 PetscErrorCode MatGetRow_MPIAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1738 { 1739 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1740 PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p; 1741 PetscErrorCode ierr; 1742 PetscInt i,*cworkA,*cworkB,**pcA,**pcB,cstart = matin->cmap->rstart; 1743 PetscInt nztot,nzA,nzB,lrow,rstart = matin->rmap->rstart,rend = matin->rmap->rend; 1744 PetscInt *cmap,*idx_p; 1745 1746 PetscFunctionBegin; 1747 if (mat->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Already active"); 1748 mat->getrowactive = PETSC_TRUE; 1749 1750 if (!mat->rowvalues && (idx || v)) { 1751 /* 1752 allocate enough space to hold information from the longest row. 1753 */ 1754 Mat_SeqAIJ *Aa = (Mat_SeqAIJ*)mat->A->data,*Ba = (Mat_SeqAIJ*)mat->B->data; 1755 PetscInt max = 1,tmp; 1756 for (i=0; i<matin->rmap->n; i++) { 1757 tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; 1758 if (max < tmp) max = tmp; 1759 } 1760 ierr = PetscMalloc2(max,&mat->rowvalues,max,&mat->rowindices);CHKERRQ(ierr); 1761 } 1762 1763 if (row < rstart || row >= rend) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Only local rows"); 1764 lrow = row - rstart; 1765 1766 pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB; 1767 if (!v) {pvA = 0; pvB = 0;} 1768 if (!idx) {pcA = 0; if (!v) pcB = 0;} 1769 ierr = (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1770 ierr = (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1771 nztot = nzA + nzB; 1772 1773 cmap = mat->garray; 1774 if (v || idx) { 1775 if (nztot) { 1776 /* Sort by increasing column numbers, assuming A and B already sorted */ 1777 PetscInt imark = -1; 1778 if (v) { 1779 *v = v_p = mat->rowvalues; 1780 for (i=0; i<nzB; i++) { 1781 if (cmap[cworkB[i]] < cstart) v_p[i] = vworkB[i]; 1782 else break; 1783 } 1784 imark = i; 1785 for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i]; 1786 for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i]; 1787 } 1788 if (idx) { 1789 *idx = idx_p = mat->rowindices; 1790 if (imark > -1) { 1791 for (i=0; i<imark; i++) { 1792 idx_p[i] = cmap[cworkB[i]]; 1793 } 1794 } else { 1795 for (i=0; i<nzB; i++) { 1796 if (cmap[cworkB[i]] < cstart) idx_p[i] = cmap[cworkB[i]]; 1797 else break; 1798 } 1799 imark = i; 1800 } 1801 for (i=0; i<nzA; i++) idx_p[imark+i] = cstart + cworkA[i]; 1802 for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]]; 1803 } 1804 } else { 1805 if (idx) *idx = 0; 1806 if (v) *v = 0; 1807 } 1808 } 1809 *nz = nztot; 1810 ierr = (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1811 ierr = (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1812 PetscFunctionReturn(0); 1813 } 1814 1815 PetscErrorCode MatRestoreRow_MPIAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1816 { 1817 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1818 1819 PetscFunctionBegin; 1820 if (!aij->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"MatGetRow() must be called first"); 1821 aij->getrowactive = PETSC_FALSE; 1822 PetscFunctionReturn(0); 1823 } 1824 1825 PetscErrorCode MatNorm_MPIAIJ(Mat mat,NormType type,PetscReal *norm) 1826 { 1827 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1828 Mat_SeqAIJ *amat = (Mat_SeqAIJ*)aij->A->data,*bmat = (Mat_SeqAIJ*)aij->B->data; 1829 PetscErrorCode ierr; 1830 PetscInt i,j,cstart = mat->cmap->rstart; 1831 PetscReal sum = 0.0; 1832 MatScalar *v; 1833 1834 PetscFunctionBegin; 1835 if (aij->size == 1) { 1836 ierr = MatNorm(aij->A,type,norm);CHKERRQ(ierr); 1837 } else { 1838 if (type == NORM_FROBENIUS) { 1839 v = amat->a; 1840 for (i=0; i<amat->nz; i++) { 1841 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1842 } 1843 v = bmat->a; 1844 for (i=0; i<bmat->nz; i++) { 1845 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1846 } 1847 ierr = MPIU_Allreduce(&sum,norm,1,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1848 *norm = PetscSqrtReal(*norm); 1849 ierr = PetscLogFlops(2*amat->nz+2*bmat->nz);CHKERRQ(ierr); 1850 } else if (type == NORM_1) { /* max column norm */ 1851 PetscReal *tmp,*tmp2; 1852 PetscInt *jj,*garray = aij->garray; 1853 ierr = PetscCalloc1(mat->cmap->N+1,&tmp);CHKERRQ(ierr); 1854 ierr = PetscMalloc1(mat->cmap->N+1,&tmp2);CHKERRQ(ierr); 1855 *norm = 0.0; 1856 v = amat->a; jj = amat->j; 1857 for (j=0; j<amat->nz; j++) { 1858 tmp[cstart + *jj++] += PetscAbsScalar(*v); v++; 1859 } 1860 v = bmat->a; jj = bmat->j; 1861 for (j=0; j<bmat->nz; j++) { 1862 tmp[garray[*jj++]] += PetscAbsScalar(*v); v++; 1863 } 1864 ierr = MPIU_Allreduce(tmp,tmp2,mat->cmap->N,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1865 for (j=0; j<mat->cmap->N; j++) { 1866 if (tmp2[j] > *norm) *norm = tmp2[j]; 1867 } 1868 ierr = PetscFree(tmp);CHKERRQ(ierr); 1869 ierr = PetscFree(tmp2);CHKERRQ(ierr); 1870 ierr = PetscLogFlops(PetscMax(amat->nz+bmat->nz-1,0));CHKERRQ(ierr); 1871 } else if (type == NORM_INFINITY) { /* max row norm */ 1872 PetscReal ntemp = 0.0; 1873 for (j=0; j<aij->A->rmap->n; j++) { 1874 v = amat->a + amat->i[j]; 1875 sum = 0.0; 1876 for (i=0; i<amat->i[j+1]-amat->i[j]; i++) { 1877 sum += PetscAbsScalar(*v); v++; 1878 } 1879 v = bmat->a + bmat->i[j]; 1880 for (i=0; i<bmat->i[j+1]-bmat->i[j]; i++) { 1881 sum += PetscAbsScalar(*v); v++; 1882 } 1883 if (sum > ntemp) ntemp = sum; 1884 } 1885 ierr = MPIU_Allreduce(&ntemp,norm,1,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1886 ierr = PetscLogFlops(PetscMax(amat->nz+bmat->nz-1,0));CHKERRQ(ierr); 1887 } else SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"No support for two norm"); 1888 } 1889 PetscFunctionReturn(0); 1890 } 1891 1892 PetscErrorCode MatTranspose_MPIAIJ(Mat A,MatReuse reuse,Mat *matout) 1893 { 1894 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1895 Mat_SeqAIJ *Aloc=(Mat_SeqAIJ*)a->A->data,*Bloc=(Mat_SeqAIJ*)a->B->data; 1896 PetscErrorCode ierr; 1897 PetscInt M = A->rmap->N,N = A->cmap->N,ma,na,mb,nb,*ai,*aj,*bi,*bj,row,*cols,*cols_tmp,i; 1898 PetscInt cstart = A->cmap->rstart,ncol; 1899 Mat B; 1900 MatScalar *array; 1901 1902 PetscFunctionBegin; 1903 if (reuse == MAT_INPLACE_MATRIX && M != N) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_SIZ,"Square matrix only for in-place"); 1904 1905 ma = A->rmap->n; na = A->cmap->n; mb = a->B->rmap->n; nb = a->B->cmap->n; 1906 ai = Aloc->i; aj = Aloc->j; 1907 bi = Bloc->i; bj = Bloc->j; 1908 if (reuse == MAT_INITIAL_MATRIX || *matout == A) { 1909 PetscInt *d_nnz,*g_nnz,*o_nnz; 1910 PetscSFNode *oloc; 1911 PETSC_UNUSED PetscSF sf; 1912 1913 ierr = PetscMalloc4(na,&d_nnz,na,&o_nnz,nb,&g_nnz,nb,&oloc);CHKERRQ(ierr); 1914 /* compute d_nnz for preallocation */ 1915 ierr = PetscMemzero(d_nnz,na*sizeof(PetscInt));CHKERRQ(ierr); 1916 for (i=0; i<ai[ma]; i++) { 1917 d_nnz[aj[i]]++; 1918 aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1919 } 1920 /* compute local off-diagonal contributions */ 1921 ierr = PetscMemzero(g_nnz,nb*sizeof(PetscInt));CHKERRQ(ierr); 1922 for (i=0; i<bi[ma]; i++) g_nnz[bj[i]]++; 1923 /* map those to global */ 1924 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1925 ierr = PetscSFSetGraphLayout(sf,A->cmap,nb,NULL,PETSC_USE_POINTER,a->garray);CHKERRQ(ierr); 1926 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1927 ierr = PetscMemzero(o_nnz,na*sizeof(PetscInt));CHKERRQ(ierr); 1928 ierr = PetscSFReduceBegin(sf,MPIU_INT,g_nnz,o_nnz,MPIU_SUM);CHKERRQ(ierr); 1929 ierr = PetscSFReduceEnd(sf,MPIU_INT,g_nnz,o_nnz,MPIU_SUM);CHKERRQ(ierr); 1930 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1931 1932 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 1933 ierr = MatSetSizes(B,A->cmap->n,A->rmap->n,N,M);CHKERRQ(ierr); 1934 ierr = MatSetBlockSizes(B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));CHKERRQ(ierr); 1935 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 1936 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 1937 ierr = PetscFree4(d_nnz,o_nnz,g_nnz,oloc);CHKERRQ(ierr); 1938 } else { 1939 B = *matout; 1940 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 1941 for (i=0; i<ai[ma]; i++) aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1942 } 1943 1944 /* copy over the A part */ 1945 array = Aloc->a; 1946 row = A->rmap->rstart; 1947 for (i=0; i<ma; i++) { 1948 ncol = ai[i+1]-ai[i]; 1949 ierr = MatSetValues(B,ncol,aj,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1950 row++; 1951 array += ncol; aj += ncol; 1952 } 1953 aj = Aloc->j; 1954 for (i=0; i<ai[ma]; i++) aj[i] -= cstart; /* resume local col index */ 1955 1956 /* copy over the B part */ 1957 ierr = PetscCalloc1(bi[mb],&cols);CHKERRQ(ierr); 1958 array = Bloc->a; 1959 row = A->rmap->rstart; 1960 for (i=0; i<bi[mb]; i++) cols[i] = a->garray[bj[i]]; 1961 cols_tmp = cols; 1962 for (i=0; i<mb; i++) { 1963 ncol = bi[i+1]-bi[i]; 1964 ierr = MatSetValues(B,ncol,cols_tmp,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1965 row++; 1966 array += ncol; cols_tmp += ncol; 1967 } 1968 ierr = PetscFree(cols);CHKERRQ(ierr); 1969 1970 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1971 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1972 if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) { 1973 *matout = B; 1974 } else { 1975 ierr = MatHeaderMerge(A,&B);CHKERRQ(ierr); 1976 } 1977 PetscFunctionReturn(0); 1978 } 1979 1980 PetscErrorCode MatDiagonalScale_MPIAIJ(Mat mat,Vec ll,Vec rr) 1981 { 1982 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1983 Mat a = aij->A,b = aij->B; 1984 PetscErrorCode ierr; 1985 PetscInt s1,s2,s3; 1986 1987 PetscFunctionBegin; 1988 ierr = MatGetLocalSize(mat,&s2,&s3);CHKERRQ(ierr); 1989 if (rr) { 1990 ierr = VecGetLocalSize(rr,&s1);CHKERRQ(ierr); 1991 if (s1!=s3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"right vector non-conforming local size"); 1992 /* Overlap communication with computation. */ 1993 ierr = VecScatterBegin(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1994 } 1995 if (ll) { 1996 ierr = VecGetLocalSize(ll,&s1);CHKERRQ(ierr); 1997 if (s1!=s2) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"left vector non-conforming local size"); 1998 ierr = (*b->ops->diagonalscale)(b,ll,0);CHKERRQ(ierr); 1999 } 2000 /* scale the diagonal block */ 2001 ierr = (*a->ops->diagonalscale)(a,ll,rr);CHKERRQ(ierr); 2002 2003 if (rr) { 2004 /* Do a scatter end and then right scale the off-diagonal block */ 2005 ierr = VecScatterEnd(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2006 ierr = (*b->ops->diagonalscale)(b,0,aij->lvec);CHKERRQ(ierr); 2007 } 2008 PetscFunctionReturn(0); 2009 } 2010 2011 PetscErrorCode MatSetUnfactored_MPIAIJ(Mat A) 2012 { 2013 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2014 PetscErrorCode ierr; 2015 2016 PetscFunctionBegin; 2017 ierr = MatSetUnfactored(a->A);CHKERRQ(ierr); 2018 PetscFunctionReturn(0); 2019 } 2020 2021 PetscErrorCode MatEqual_MPIAIJ(Mat A,Mat B,PetscBool *flag) 2022 { 2023 Mat_MPIAIJ *matB = (Mat_MPIAIJ*)B->data,*matA = (Mat_MPIAIJ*)A->data; 2024 Mat a,b,c,d; 2025 PetscBool flg; 2026 PetscErrorCode ierr; 2027 2028 PetscFunctionBegin; 2029 a = matA->A; b = matA->B; 2030 c = matB->A; d = matB->B; 2031 2032 ierr = MatEqual(a,c,&flg);CHKERRQ(ierr); 2033 if (flg) { 2034 ierr = MatEqual(b,d,&flg);CHKERRQ(ierr); 2035 } 2036 ierr = MPIU_Allreduce(&flg,flag,1,MPIU_BOOL,MPI_LAND,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 2037 PetscFunctionReturn(0); 2038 } 2039 2040 PetscErrorCode MatCopy_MPIAIJ(Mat A,Mat B,MatStructure str) 2041 { 2042 PetscErrorCode ierr; 2043 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2044 Mat_MPIAIJ *b = (Mat_MPIAIJ*)B->data; 2045 2046 PetscFunctionBegin; 2047 /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */ 2048 if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) { 2049 /* because of the column compression in the off-processor part of the matrix a->B, 2050 the number of columns in a->B and b->B may be different, hence we cannot call 2051 the MatCopy() directly on the two parts. If need be, we can provide a more 2052 efficient copy than the MatCopy_Basic() by first uncompressing the a->B matrices 2053 then copying the submatrices */ 2054 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2055 } else { 2056 ierr = MatCopy(a->A,b->A,str);CHKERRQ(ierr); 2057 ierr = MatCopy(a->B,b->B,str);CHKERRQ(ierr); 2058 } 2059 PetscFunctionReturn(0); 2060 } 2061 2062 PetscErrorCode MatSetUp_MPIAIJ(Mat A) 2063 { 2064 PetscErrorCode ierr; 2065 2066 PetscFunctionBegin; 2067 ierr = MatMPIAIJSetPreallocation(A,PETSC_DEFAULT,0,PETSC_DEFAULT,0);CHKERRQ(ierr); 2068 PetscFunctionReturn(0); 2069 } 2070 2071 /* 2072 Computes the number of nonzeros per row needed for preallocation when X and Y 2073 have different nonzero structure. 2074 */ 2075 PetscErrorCode MatAXPYGetPreallocation_MPIX_private(PetscInt m,const PetscInt *xi,const PetscInt *xj,const PetscInt *xltog,const PetscInt *yi,const PetscInt *yj,const PetscInt *yltog,PetscInt *nnz) 2076 { 2077 PetscInt i,j,k,nzx,nzy; 2078 2079 PetscFunctionBegin; 2080 /* Set the number of nonzeros in the new matrix */ 2081 for (i=0; i<m; i++) { 2082 const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i]; 2083 nzx = xi[i+1] - xi[i]; 2084 nzy = yi[i+1] - yi[i]; 2085 nnz[i] = 0; 2086 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 2087 for (; k<nzy && yltog[yjj[k]]<xltog[xjj[j]]; k++) nnz[i]++; /* Catch up to X */ 2088 if (k<nzy && yltog[yjj[k]]==xltog[xjj[j]]) k++; /* Skip duplicate */ 2089 nnz[i]++; 2090 } 2091 for (; k<nzy; k++) nnz[i]++; 2092 } 2093 PetscFunctionReturn(0); 2094 } 2095 2096 /* This is the same as MatAXPYGetPreallocation_SeqAIJ, except that the local-to-global map is provided */ 2097 static PetscErrorCode MatAXPYGetPreallocation_MPIAIJ(Mat Y,const PetscInt *yltog,Mat X,const PetscInt *xltog,PetscInt *nnz) 2098 { 2099 PetscErrorCode ierr; 2100 PetscInt m = Y->rmap->N; 2101 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 2102 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 2103 2104 PetscFunctionBegin; 2105 ierr = MatAXPYGetPreallocation_MPIX_private(m,x->i,x->j,xltog,y->i,y->j,yltog,nnz);CHKERRQ(ierr); 2106 PetscFunctionReturn(0); 2107 } 2108 2109 PetscErrorCode MatAXPY_MPIAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 2110 { 2111 PetscErrorCode ierr; 2112 Mat_MPIAIJ *xx = (Mat_MPIAIJ*)X->data,*yy = (Mat_MPIAIJ*)Y->data; 2113 PetscBLASInt bnz,one=1; 2114 Mat_SeqAIJ *x,*y; 2115 2116 PetscFunctionBegin; 2117 if (str == SAME_NONZERO_PATTERN) { 2118 PetscScalar alpha = a; 2119 x = (Mat_SeqAIJ*)xx->A->data; 2120 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 2121 y = (Mat_SeqAIJ*)yy->A->data; 2122 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 2123 x = (Mat_SeqAIJ*)xx->B->data; 2124 y = (Mat_SeqAIJ*)yy->B->data; 2125 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 2126 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 2127 ierr = PetscObjectStateIncrease((PetscObject)Y);CHKERRQ(ierr); 2128 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 2129 ierr = MatAXPY_Basic(Y,a,X,str);CHKERRQ(ierr); 2130 } else { 2131 Mat B; 2132 PetscInt *nnz_d,*nnz_o; 2133 ierr = PetscMalloc1(yy->A->rmap->N,&nnz_d);CHKERRQ(ierr); 2134 ierr = PetscMalloc1(yy->B->rmap->N,&nnz_o);CHKERRQ(ierr); 2135 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 2136 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 2137 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 2138 ierr = MatSetBlockSizesFromMats(B,Y,Y);CHKERRQ(ierr); 2139 ierr = MatSetType(B,MATMPIAIJ);CHKERRQ(ierr); 2140 ierr = MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); 2141 ierr = MatAXPYGetPreallocation_MPIAIJ(yy->B,yy->garray,xx->B,xx->garray,nnz_o);CHKERRQ(ierr); 2142 ierr = MatMPIAIJSetPreallocation(B,0,nnz_d,0,nnz_o);CHKERRQ(ierr); 2143 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 2144 ierr = MatHeaderReplace(Y,&B);CHKERRQ(ierr); 2145 ierr = PetscFree(nnz_d);CHKERRQ(ierr); 2146 ierr = PetscFree(nnz_o);CHKERRQ(ierr); 2147 } 2148 PetscFunctionReturn(0); 2149 } 2150 2151 extern PetscErrorCode MatConjugate_SeqAIJ(Mat); 2152 2153 PetscErrorCode MatConjugate_MPIAIJ(Mat mat) 2154 { 2155 #if defined(PETSC_USE_COMPLEX) 2156 PetscErrorCode ierr; 2157 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2158 2159 PetscFunctionBegin; 2160 ierr = MatConjugate_SeqAIJ(aij->A);CHKERRQ(ierr); 2161 ierr = MatConjugate_SeqAIJ(aij->B);CHKERRQ(ierr); 2162 #else 2163 PetscFunctionBegin; 2164 #endif 2165 PetscFunctionReturn(0); 2166 } 2167 2168 PetscErrorCode MatRealPart_MPIAIJ(Mat A) 2169 { 2170 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2171 PetscErrorCode ierr; 2172 2173 PetscFunctionBegin; 2174 ierr = MatRealPart(a->A);CHKERRQ(ierr); 2175 ierr = MatRealPart(a->B);CHKERRQ(ierr); 2176 PetscFunctionReturn(0); 2177 } 2178 2179 PetscErrorCode MatImaginaryPart_MPIAIJ(Mat A) 2180 { 2181 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2182 PetscErrorCode ierr; 2183 2184 PetscFunctionBegin; 2185 ierr = MatImaginaryPart(a->A);CHKERRQ(ierr); 2186 ierr = MatImaginaryPart(a->B);CHKERRQ(ierr); 2187 PetscFunctionReturn(0); 2188 } 2189 2190 PetscErrorCode MatGetRowMaxAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2191 { 2192 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2193 PetscErrorCode ierr; 2194 PetscInt i,*idxb = 0; 2195 PetscScalar *va,*vb; 2196 Vec vtmp; 2197 2198 PetscFunctionBegin; 2199 ierr = MatGetRowMaxAbs(a->A,v,idx);CHKERRQ(ierr); 2200 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2201 if (idx) { 2202 for (i=0; i<A->rmap->n; i++) { 2203 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2204 } 2205 } 2206 2207 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2208 if (idx) { 2209 ierr = PetscMalloc1(A->rmap->n,&idxb);CHKERRQ(ierr); 2210 } 2211 ierr = MatGetRowMaxAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2212 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2213 2214 for (i=0; i<A->rmap->n; i++) { 2215 if (PetscAbsScalar(va[i]) < PetscAbsScalar(vb[i])) { 2216 va[i] = vb[i]; 2217 if (idx) idx[i] = a->garray[idxb[i]]; 2218 } 2219 } 2220 2221 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2222 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2223 ierr = PetscFree(idxb);CHKERRQ(ierr); 2224 ierr = VecDestroy(&vtmp);CHKERRQ(ierr); 2225 PetscFunctionReturn(0); 2226 } 2227 2228 PetscErrorCode MatGetRowMinAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2229 { 2230 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2231 PetscErrorCode ierr; 2232 PetscInt i,*idxb = 0; 2233 PetscScalar *va,*vb; 2234 Vec vtmp; 2235 2236 PetscFunctionBegin; 2237 ierr = MatGetRowMinAbs(a->A,v,idx);CHKERRQ(ierr); 2238 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2239 if (idx) { 2240 for (i=0; i<A->cmap->n; i++) { 2241 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2242 } 2243 } 2244 2245 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2246 if (idx) { 2247 ierr = PetscMalloc1(A->rmap->n,&idxb);CHKERRQ(ierr); 2248 } 2249 ierr = MatGetRowMinAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2250 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2251 2252 for (i=0; i<A->rmap->n; i++) { 2253 if (PetscAbsScalar(va[i]) > PetscAbsScalar(vb[i])) { 2254 va[i] = vb[i]; 2255 if (idx) idx[i] = a->garray[idxb[i]]; 2256 } 2257 } 2258 2259 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2260 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2261 ierr = PetscFree(idxb);CHKERRQ(ierr); 2262 ierr = VecDestroy(&vtmp);CHKERRQ(ierr); 2263 PetscFunctionReturn(0); 2264 } 2265 2266 PetscErrorCode MatGetRowMin_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2267 { 2268 Mat_MPIAIJ *mat = (Mat_MPIAIJ*) A->data; 2269 PetscInt n = A->rmap->n; 2270 PetscInt cstart = A->cmap->rstart; 2271 PetscInt *cmap = mat->garray; 2272 PetscInt *diagIdx, *offdiagIdx; 2273 Vec diagV, offdiagV; 2274 PetscScalar *a, *diagA, *offdiagA; 2275 PetscInt r; 2276 PetscErrorCode ierr; 2277 2278 PetscFunctionBegin; 2279 ierr = PetscMalloc2(n,&diagIdx,n,&offdiagIdx);CHKERRQ(ierr); 2280 ierr = VecCreateSeq(PetscObjectComm((PetscObject)A), n, &diagV);CHKERRQ(ierr); 2281 ierr = VecCreateSeq(PetscObjectComm((PetscObject)A), n, &offdiagV);CHKERRQ(ierr); 2282 ierr = MatGetRowMin(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2283 ierr = MatGetRowMin(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2284 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2285 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2286 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2287 for (r = 0; r < n; ++r) { 2288 if (PetscAbsScalar(diagA[r]) <= PetscAbsScalar(offdiagA[r])) { 2289 a[r] = diagA[r]; 2290 idx[r] = cstart + diagIdx[r]; 2291 } else { 2292 a[r] = offdiagA[r]; 2293 idx[r] = cmap[offdiagIdx[r]]; 2294 } 2295 } 2296 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2297 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2298 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2299 ierr = VecDestroy(&diagV);CHKERRQ(ierr); 2300 ierr = VecDestroy(&offdiagV);CHKERRQ(ierr); 2301 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2302 PetscFunctionReturn(0); 2303 } 2304 2305 PetscErrorCode MatGetRowMax_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2306 { 2307 Mat_MPIAIJ *mat = (Mat_MPIAIJ*) A->data; 2308 PetscInt n = A->rmap->n; 2309 PetscInt cstart = A->cmap->rstart; 2310 PetscInt *cmap = mat->garray; 2311 PetscInt *diagIdx, *offdiagIdx; 2312 Vec diagV, offdiagV; 2313 PetscScalar *a, *diagA, *offdiagA; 2314 PetscInt r; 2315 PetscErrorCode ierr; 2316 2317 PetscFunctionBegin; 2318 ierr = PetscMalloc2(n,&diagIdx,n,&offdiagIdx);CHKERRQ(ierr); 2319 ierr = VecCreateSeq(PETSC_COMM_SELF, n, &diagV);CHKERRQ(ierr); 2320 ierr = VecCreateSeq(PETSC_COMM_SELF, n, &offdiagV);CHKERRQ(ierr); 2321 ierr = MatGetRowMax(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2322 ierr = MatGetRowMax(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2323 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2324 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2325 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2326 for (r = 0; r < n; ++r) { 2327 if (PetscAbsScalar(diagA[r]) >= PetscAbsScalar(offdiagA[r])) { 2328 a[r] = diagA[r]; 2329 idx[r] = cstart + diagIdx[r]; 2330 } else { 2331 a[r] = offdiagA[r]; 2332 idx[r] = cmap[offdiagIdx[r]]; 2333 } 2334 } 2335 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2336 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2337 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2338 ierr = VecDestroy(&diagV);CHKERRQ(ierr); 2339 ierr = VecDestroy(&offdiagV);CHKERRQ(ierr); 2340 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2341 PetscFunctionReturn(0); 2342 } 2343 2344 PetscErrorCode MatGetSeqNonzeroStructure_MPIAIJ(Mat mat,Mat *newmat) 2345 { 2346 PetscErrorCode ierr; 2347 Mat *dummy; 2348 2349 PetscFunctionBegin; 2350 ierr = MatCreateSubMatrix_MPIAIJ_All(mat,MAT_DO_NOT_GET_VALUES,MAT_INITIAL_MATRIX,&dummy);CHKERRQ(ierr); 2351 *newmat = *dummy; 2352 ierr = PetscFree(dummy);CHKERRQ(ierr); 2353 PetscFunctionReturn(0); 2354 } 2355 2356 PetscErrorCode MatInvertBlockDiagonal_MPIAIJ(Mat A,const PetscScalar **values) 2357 { 2358 Mat_MPIAIJ *a = (Mat_MPIAIJ*) A->data; 2359 PetscErrorCode ierr; 2360 2361 PetscFunctionBegin; 2362 ierr = MatInvertBlockDiagonal(a->A,values);CHKERRQ(ierr); 2363 A->factorerrortype = a->A->factorerrortype; 2364 PetscFunctionReturn(0); 2365 } 2366 2367 static PetscErrorCode MatSetRandom_MPIAIJ(Mat x,PetscRandom rctx) 2368 { 2369 PetscErrorCode ierr; 2370 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)x->data; 2371 2372 PetscFunctionBegin; 2373 ierr = MatSetRandom(aij->A,rctx);CHKERRQ(ierr); 2374 ierr = MatSetRandom(aij->B,rctx);CHKERRQ(ierr); 2375 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2376 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2377 PetscFunctionReturn(0); 2378 } 2379 2380 PetscErrorCode MatMPIAIJSetUseScalableIncreaseOverlap_MPIAIJ(Mat A,PetscBool sc) 2381 { 2382 PetscFunctionBegin; 2383 if (sc) A->ops->increaseoverlap = MatIncreaseOverlap_MPIAIJ_Scalable; 2384 else A->ops->increaseoverlap = MatIncreaseOverlap_MPIAIJ; 2385 PetscFunctionReturn(0); 2386 } 2387 2388 /*@ 2389 MatMPIAIJSetUseScalableIncreaseOverlap - Determine if the matrix uses a scalable algorithm to compute the overlap 2390 2391 Collective on Mat 2392 2393 Input Parameters: 2394 + A - the matrix 2395 - sc - PETSC_TRUE indicates use the scalable algorithm (default is not to use the scalable algorithm) 2396 2397 Level: advanced 2398 2399 @*/ 2400 PetscErrorCode MatMPIAIJSetUseScalableIncreaseOverlap(Mat A,PetscBool sc) 2401 { 2402 PetscErrorCode ierr; 2403 2404 PetscFunctionBegin; 2405 ierr = PetscTryMethod(A,"MatMPIAIJSetUseScalableIncreaseOverlap_C",(Mat,PetscBool),(A,sc));CHKERRQ(ierr); 2406 PetscFunctionReturn(0); 2407 } 2408 2409 PetscErrorCode MatSetFromOptions_MPIAIJ(PetscOptionItems *PetscOptionsObject,Mat A) 2410 { 2411 PetscErrorCode ierr; 2412 PetscBool sc = PETSC_FALSE,flg; 2413 2414 PetscFunctionBegin; 2415 ierr = PetscOptionsHead(PetscOptionsObject,"MPIAIJ options");CHKERRQ(ierr); 2416 ierr = PetscObjectOptionsBegin((PetscObject)A); 2417 if (A->ops->increaseoverlap == MatIncreaseOverlap_MPIAIJ_Scalable) sc = PETSC_TRUE; 2418 ierr = PetscOptionsBool("-mat_increase_overlap_scalable","Use a scalable algorithm to compute the overlap","MatIncreaseOverlap",sc,&sc,&flg);CHKERRQ(ierr); 2419 if (flg) { 2420 ierr = MatMPIAIJSetUseScalableIncreaseOverlap(A,sc);CHKERRQ(ierr); 2421 } 2422 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2423 PetscFunctionReturn(0); 2424 } 2425 2426 PetscErrorCode MatShift_MPIAIJ(Mat Y,PetscScalar a) 2427 { 2428 PetscErrorCode ierr; 2429 Mat_MPIAIJ *maij = (Mat_MPIAIJ*)Y->data; 2430 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)maij->A->data; 2431 2432 PetscFunctionBegin; 2433 if (!Y->preallocated) { 2434 ierr = MatMPIAIJSetPreallocation(Y,1,NULL,0,NULL);CHKERRQ(ierr); 2435 } else if (!aij->nz) { 2436 PetscInt nonew = aij->nonew; 2437 ierr = MatSeqAIJSetPreallocation(maij->A,1,NULL);CHKERRQ(ierr); 2438 aij->nonew = nonew; 2439 } 2440 ierr = MatShift_Basic(Y,a);CHKERRQ(ierr); 2441 PetscFunctionReturn(0); 2442 } 2443 2444 PetscErrorCode MatMissingDiagonal_MPIAIJ(Mat A,PetscBool *missing,PetscInt *d) 2445 { 2446 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2447 PetscErrorCode ierr; 2448 2449 PetscFunctionBegin; 2450 if (A->rmap->n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only works for square matrices"); 2451 ierr = MatMissingDiagonal(a->A,missing,d);CHKERRQ(ierr); 2452 if (d) { 2453 PetscInt rstart; 2454 ierr = MatGetOwnershipRange(A,&rstart,NULL);CHKERRQ(ierr); 2455 *d += rstart; 2456 2457 } 2458 PetscFunctionReturn(0); 2459 } 2460 2461 2462 /* -------------------------------------------------------------------*/ 2463 static struct _MatOps MatOps_Values = {MatSetValues_MPIAIJ, 2464 MatGetRow_MPIAIJ, 2465 MatRestoreRow_MPIAIJ, 2466 MatMult_MPIAIJ, 2467 /* 4*/ MatMultAdd_MPIAIJ, 2468 MatMultTranspose_MPIAIJ, 2469 MatMultTransposeAdd_MPIAIJ, 2470 0, 2471 0, 2472 0, 2473 /*10*/ 0, 2474 0, 2475 0, 2476 MatSOR_MPIAIJ, 2477 MatTranspose_MPIAIJ, 2478 /*15*/ MatGetInfo_MPIAIJ, 2479 MatEqual_MPIAIJ, 2480 MatGetDiagonal_MPIAIJ, 2481 MatDiagonalScale_MPIAIJ, 2482 MatNorm_MPIAIJ, 2483 /*20*/ MatAssemblyBegin_MPIAIJ, 2484 MatAssemblyEnd_MPIAIJ, 2485 MatSetOption_MPIAIJ, 2486 MatZeroEntries_MPIAIJ, 2487 /*24*/ MatZeroRows_MPIAIJ, 2488 0, 2489 0, 2490 0, 2491 0, 2492 /*29*/ MatSetUp_MPIAIJ, 2493 0, 2494 0, 2495 MatGetDiagonalBlock_MPIAIJ, 2496 0, 2497 /*34*/ MatDuplicate_MPIAIJ, 2498 0, 2499 0, 2500 0, 2501 0, 2502 /*39*/ MatAXPY_MPIAIJ, 2503 MatCreateSubMatrices_MPIAIJ, 2504 MatIncreaseOverlap_MPIAIJ, 2505 MatGetValues_MPIAIJ, 2506 MatCopy_MPIAIJ, 2507 /*44*/ MatGetRowMax_MPIAIJ, 2508 MatScale_MPIAIJ, 2509 MatShift_MPIAIJ, 2510 MatDiagonalSet_MPIAIJ, 2511 MatZeroRowsColumns_MPIAIJ, 2512 /*49*/ MatSetRandom_MPIAIJ, 2513 0, 2514 0, 2515 0, 2516 0, 2517 /*54*/ MatFDColoringCreate_MPIXAIJ, 2518 0, 2519 MatSetUnfactored_MPIAIJ, 2520 MatPermute_MPIAIJ, 2521 0, 2522 /*59*/ MatCreateSubMatrix_MPIAIJ, 2523 MatDestroy_MPIAIJ, 2524 MatView_MPIAIJ, 2525 0, 2526 MatMatMatMult_MPIAIJ_MPIAIJ_MPIAIJ, 2527 /*64*/ MatMatMatMultSymbolic_MPIAIJ_MPIAIJ_MPIAIJ, 2528 MatMatMatMultNumeric_MPIAIJ_MPIAIJ_MPIAIJ, 2529 0, 2530 0, 2531 0, 2532 /*69*/ MatGetRowMaxAbs_MPIAIJ, 2533 MatGetRowMinAbs_MPIAIJ, 2534 0, 2535 0, 2536 0, 2537 0, 2538 /*75*/ MatFDColoringApply_AIJ, 2539 MatSetFromOptions_MPIAIJ, 2540 0, 2541 0, 2542 MatFindZeroDiagonals_MPIAIJ, 2543 /*80*/ 0, 2544 0, 2545 0, 2546 /*83*/ MatLoad_MPIAIJ, 2547 0, 2548 0, 2549 0, 2550 0, 2551 0, 2552 /*89*/ MatMatMult_MPIAIJ_MPIAIJ, 2553 MatMatMultSymbolic_MPIAIJ_MPIAIJ, 2554 MatMatMultNumeric_MPIAIJ_MPIAIJ, 2555 MatPtAP_MPIAIJ_MPIAIJ, 2556 MatPtAPSymbolic_MPIAIJ_MPIAIJ, 2557 /*94*/ MatPtAPNumeric_MPIAIJ_MPIAIJ, 2558 0, 2559 0, 2560 0, 2561 0, 2562 /*99*/ 0, 2563 0, 2564 0, 2565 MatConjugate_MPIAIJ, 2566 0, 2567 /*104*/MatSetValuesRow_MPIAIJ, 2568 MatRealPart_MPIAIJ, 2569 MatImaginaryPart_MPIAIJ, 2570 0, 2571 0, 2572 /*109*/0, 2573 0, 2574 MatGetRowMin_MPIAIJ, 2575 0, 2576 MatMissingDiagonal_MPIAIJ, 2577 /*114*/MatGetSeqNonzeroStructure_MPIAIJ, 2578 0, 2579 MatGetGhosts_MPIAIJ, 2580 0, 2581 0, 2582 /*119*/0, 2583 0, 2584 0, 2585 0, 2586 MatGetMultiProcBlock_MPIAIJ, 2587 /*124*/MatFindNonzeroRows_MPIAIJ, 2588 MatGetColumnNorms_MPIAIJ, 2589 MatInvertBlockDiagonal_MPIAIJ, 2590 0, 2591 MatCreateSubMatricesMPI_MPIAIJ, 2592 /*129*/0, 2593 MatTransposeMatMult_MPIAIJ_MPIAIJ, 2594 MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ, 2595 MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ, 2596 0, 2597 /*134*/0, 2598 0, 2599 0, 2600 0, 2601 0, 2602 /*139*/MatSetBlockSizes_MPIAIJ, 2603 0, 2604 0, 2605 MatFDColoringSetUp_MPIXAIJ, 2606 MatFindOffBlockDiagonalEntries_MPIAIJ, 2607 /*144*/MatCreateMPIMatConcatenateSeqMat_MPIAIJ 2608 }; 2609 2610 /* ----------------------------------------------------------------------------------------*/ 2611 2612 PetscErrorCode MatStoreValues_MPIAIJ(Mat mat) 2613 { 2614 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2615 PetscErrorCode ierr; 2616 2617 PetscFunctionBegin; 2618 ierr = MatStoreValues(aij->A);CHKERRQ(ierr); 2619 ierr = MatStoreValues(aij->B);CHKERRQ(ierr); 2620 PetscFunctionReturn(0); 2621 } 2622 2623 PetscErrorCode MatRetrieveValues_MPIAIJ(Mat mat) 2624 { 2625 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2626 PetscErrorCode ierr; 2627 2628 PetscFunctionBegin; 2629 ierr = MatRetrieveValues(aij->A);CHKERRQ(ierr); 2630 ierr = MatRetrieveValues(aij->B);CHKERRQ(ierr); 2631 PetscFunctionReturn(0); 2632 } 2633 2634 PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJ(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 2635 { 2636 Mat_MPIAIJ *b; 2637 PetscErrorCode ierr; 2638 2639 PetscFunctionBegin; 2640 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 2641 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 2642 b = (Mat_MPIAIJ*)B->data; 2643 2644 #if defined(PETSC_USE_CTABLE) 2645 ierr = PetscTableDestroy(&b->colmap);CHKERRQ(ierr); 2646 #else 2647 ierr = PetscFree(b->colmap);CHKERRQ(ierr); 2648 #endif 2649 ierr = PetscFree(b->garray);CHKERRQ(ierr); 2650 ierr = VecDestroy(&b->lvec);CHKERRQ(ierr); 2651 ierr = VecScatterDestroy(&b->Mvctx);CHKERRQ(ierr); 2652 2653 /* Because the B will have been resized we simply destroy it and create a new one each time */ 2654 ierr = MatDestroy(&b->B);CHKERRQ(ierr); 2655 ierr = MatCreate(PETSC_COMM_SELF,&b->B);CHKERRQ(ierr); 2656 ierr = MatSetSizes(b->B,B->rmap->n,B->cmap->N,B->rmap->n,B->cmap->N);CHKERRQ(ierr); 2657 ierr = MatSetBlockSizesFromMats(b->B,B,B);CHKERRQ(ierr); 2658 ierr = MatSetType(b->B,MATSEQAIJ);CHKERRQ(ierr); 2659 ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)b->B);CHKERRQ(ierr); 2660 2661 if (!B->preallocated) { 2662 ierr = MatCreate(PETSC_COMM_SELF,&b->A);CHKERRQ(ierr); 2663 ierr = MatSetSizes(b->A,B->rmap->n,B->cmap->n,B->rmap->n,B->cmap->n);CHKERRQ(ierr); 2664 ierr = MatSetBlockSizesFromMats(b->A,B,B);CHKERRQ(ierr); 2665 ierr = MatSetType(b->A,MATSEQAIJ);CHKERRQ(ierr); 2666 ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)b->A);CHKERRQ(ierr); 2667 } 2668 2669 ierr = MatSeqAIJSetPreallocation(b->A,d_nz,d_nnz);CHKERRQ(ierr); 2670 ierr = MatSeqAIJSetPreallocation(b->B,o_nz,o_nnz);CHKERRQ(ierr); 2671 B->preallocated = PETSC_TRUE; 2672 B->was_assembled = PETSC_FALSE; 2673 B->assembled = PETSC_FALSE;; 2674 PetscFunctionReturn(0); 2675 } 2676 2677 PetscErrorCode MatDuplicate_MPIAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat) 2678 { 2679 Mat mat; 2680 Mat_MPIAIJ *a,*oldmat = (Mat_MPIAIJ*)matin->data; 2681 PetscErrorCode ierr; 2682 2683 PetscFunctionBegin; 2684 *newmat = 0; 2685 ierr = MatCreate(PetscObjectComm((PetscObject)matin),&mat);CHKERRQ(ierr); 2686 ierr = MatSetSizes(mat,matin->rmap->n,matin->cmap->n,matin->rmap->N,matin->cmap->N);CHKERRQ(ierr); 2687 ierr = MatSetBlockSizesFromMats(mat,matin,matin);CHKERRQ(ierr); 2688 ierr = MatSetType(mat,((PetscObject)matin)->type_name);CHKERRQ(ierr); 2689 ierr = PetscMemcpy(mat->ops,matin->ops,sizeof(struct _MatOps));CHKERRQ(ierr); 2690 a = (Mat_MPIAIJ*)mat->data; 2691 2692 mat->factortype = matin->factortype; 2693 mat->assembled = PETSC_TRUE; 2694 mat->insertmode = NOT_SET_VALUES; 2695 mat->preallocated = PETSC_TRUE; 2696 2697 a->size = oldmat->size; 2698 a->rank = oldmat->rank; 2699 a->donotstash = oldmat->donotstash; 2700 a->roworiented = oldmat->roworiented; 2701 a->rowindices = 0; 2702 a->rowvalues = 0; 2703 a->getrowactive = PETSC_FALSE; 2704 2705 ierr = PetscLayoutReference(matin->rmap,&mat->rmap);CHKERRQ(ierr); 2706 ierr = PetscLayoutReference(matin->cmap,&mat->cmap);CHKERRQ(ierr); 2707 2708 if (oldmat->colmap) { 2709 #if defined(PETSC_USE_CTABLE) 2710 ierr = PetscTableCreateCopy(oldmat->colmap,&a->colmap);CHKERRQ(ierr); 2711 #else 2712 ierr = PetscMalloc1(mat->cmap->N,&a->colmap);CHKERRQ(ierr); 2713 ierr = PetscLogObjectMemory((PetscObject)mat,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 2714 ierr = PetscMemcpy(a->colmap,oldmat->colmap,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 2715 #endif 2716 } else a->colmap = 0; 2717 if (oldmat->garray) { 2718 PetscInt len; 2719 len = oldmat->B->cmap->n; 2720 ierr = PetscMalloc1(len+1,&a->garray);CHKERRQ(ierr); 2721 ierr = PetscLogObjectMemory((PetscObject)mat,len*sizeof(PetscInt));CHKERRQ(ierr); 2722 if (len) { ierr = PetscMemcpy(a->garray,oldmat->garray,len*sizeof(PetscInt));CHKERRQ(ierr); } 2723 } else a->garray = 0; 2724 2725 ierr = VecDuplicate(oldmat->lvec,&a->lvec);CHKERRQ(ierr); 2726 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->lvec);CHKERRQ(ierr); 2727 ierr = VecScatterCopy(oldmat->Mvctx,&a->Mvctx);CHKERRQ(ierr); 2728 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->Mvctx);CHKERRQ(ierr); 2729 ierr = MatDuplicate(oldmat->A,cpvalues,&a->A);CHKERRQ(ierr); 2730 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->A);CHKERRQ(ierr); 2731 ierr = MatDuplicate(oldmat->B,cpvalues,&a->B);CHKERRQ(ierr); 2732 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->B);CHKERRQ(ierr); 2733 ierr = PetscFunctionListDuplicate(((PetscObject)matin)->qlist,&((PetscObject)mat)->qlist);CHKERRQ(ierr); 2734 *newmat = mat; 2735 PetscFunctionReturn(0); 2736 } 2737 2738 2739 2740 PetscErrorCode MatLoad_MPIAIJ(Mat newMat, PetscViewer viewer) 2741 { 2742 PetscScalar *vals,*svals; 2743 MPI_Comm comm; 2744 PetscErrorCode ierr; 2745 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 2746 PetscInt i,nz,j,rstart,rend,mmax,maxnz = 0; 2747 PetscInt header[4],*rowlengths = 0,M,N,m,*cols; 2748 PetscInt *ourlens = NULL,*procsnz = NULL,*offlens = NULL,jj,*mycols,*smycols; 2749 PetscInt cend,cstart,n,*rowners; 2750 int fd; 2751 PetscInt bs = newMat->rmap->bs; 2752 2753 PetscFunctionBegin; 2754 /* force binary viewer to load .info file if it has not yet done so */ 2755 ierr = PetscViewerSetUp(viewer);CHKERRQ(ierr); 2756 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 2757 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 2758 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 2759 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 2760 if (!rank) { 2761 ierr = PetscBinaryRead(fd,(char*)header,4,PETSC_INT);CHKERRQ(ierr); 2762 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object"); 2763 if (header[3] < 0) SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as MATMPIAIJ"); 2764 } 2765 2766 ierr = PetscOptionsBegin(comm,NULL,"Options for loading MATMPIAIJ matrix","Mat");CHKERRQ(ierr); 2767 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 2768 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2769 if (bs < 0) bs = 1; 2770 2771 ierr = MPI_Bcast(header+1,3,MPIU_INT,0,comm);CHKERRQ(ierr); 2772 M = header[1]; N = header[2]; 2773 2774 /* If global sizes are set, check if they are consistent with that given in the file */ 2775 if (newMat->rmap->N >= 0 && newMat->rmap->N != M) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent # of rows:Matrix in file has (%D) and input matrix has (%D)",newMat->rmap->N,M); 2776 if (newMat->cmap->N >=0 && newMat->cmap->N != N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent # of cols:Matrix in file has (%D) and input matrix has (%D)",newMat->cmap->N,N); 2777 2778 /* determine ownership of all (block) rows */ 2779 if (M%bs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Inconsistent # of rows (%d) and block size (%d)",M,bs); 2780 if (newMat->rmap->n < 0) m = bs*((M/bs)/size + (((M/bs) % size) > rank)); /* PETSC_DECIDE */ 2781 else m = newMat->rmap->n; /* Set by user */ 2782 2783 ierr = PetscMalloc1(size+1,&rowners);CHKERRQ(ierr); 2784 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 2785 2786 /* First process needs enough room for process with most rows */ 2787 if (!rank) { 2788 mmax = rowners[1]; 2789 for (i=2; i<=size; i++) { 2790 mmax = PetscMax(mmax, rowners[i]); 2791 } 2792 } else mmax = -1; /* unused, but compilers complain */ 2793 2794 rowners[0] = 0; 2795 for (i=2; i<=size; i++) { 2796 rowners[i] += rowners[i-1]; 2797 } 2798 rstart = rowners[rank]; 2799 rend = rowners[rank+1]; 2800 2801 /* distribute row lengths to all processors */ 2802 ierr = PetscMalloc2(m,&ourlens,m,&offlens);CHKERRQ(ierr); 2803 if (!rank) { 2804 ierr = PetscBinaryRead(fd,ourlens,m,PETSC_INT);CHKERRQ(ierr); 2805 ierr = PetscMalloc1(mmax,&rowlengths);CHKERRQ(ierr); 2806 ierr = PetscCalloc1(size,&procsnz);CHKERRQ(ierr); 2807 for (j=0; j<m; j++) { 2808 procsnz[0] += ourlens[j]; 2809 } 2810 for (i=1; i<size; i++) { 2811 ierr = PetscBinaryRead(fd,rowlengths,rowners[i+1]-rowners[i],PETSC_INT);CHKERRQ(ierr); 2812 /* calculate the number of nonzeros on each processor */ 2813 for (j=0; j<rowners[i+1]-rowners[i]; j++) { 2814 procsnz[i] += rowlengths[j]; 2815 } 2816 ierr = MPIULong_Send(rowlengths,rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2817 } 2818 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 2819 } else { 2820 ierr = MPIULong_Recv(ourlens,m,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 2821 } 2822 2823 if (!rank) { 2824 /* determine max buffer needed and allocate it */ 2825 maxnz = 0; 2826 for (i=0; i<size; i++) { 2827 maxnz = PetscMax(maxnz,procsnz[i]); 2828 } 2829 ierr = PetscMalloc1(maxnz,&cols);CHKERRQ(ierr); 2830 2831 /* read in my part of the matrix column indices */ 2832 nz = procsnz[0]; 2833 ierr = PetscMalloc1(nz,&mycols);CHKERRQ(ierr); 2834 ierr = PetscBinaryRead(fd,mycols,nz,PETSC_INT);CHKERRQ(ierr); 2835 2836 /* read in every one elses and ship off */ 2837 for (i=1; i<size; i++) { 2838 nz = procsnz[i]; 2839 ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr); 2840 ierr = MPIULong_Send(cols,nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2841 } 2842 ierr = PetscFree(cols);CHKERRQ(ierr); 2843 } else { 2844 /* determine buffer space needed for message */ 2845 nz = 0; 2846 for (i=0; i<m; i++) { 2847 nz += ourlens[i]; 2848 } 2849 ierr = PetscMalloc1(nz,&mycols);CHKERRQ(ierr); 2850 2851 /* receive message of column indices*/ 2852 ierr = MPIULong_Recv(mycols,nz,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 2853 } 2854 2855 /* determine column ownership if matrix is not square */ 2856 if (N != M) { 2857 if (newMat->cmap->n < 0) n = N/size + ((N % size) > rank); 2858 else n = newMat->cmap->n; 2859 ierr = MPI_Scan(&n,&cend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 2860 cstart = cend - n; 2861 } else { 2862 cstart = rstart; 2863 cend = rend; 2864 n = cend - cstart; 2865 } 2866 2867 /* loop over local rows, determining number of off diagonal entries */ 2868 ierr = PetscMemzero(offlens,m*sizeof(PetscInt));CHKERRQ(ierr); 2869 jj = 0; 2870 for (i=0; i<m; i++) { 2871 for (j=0; j<ourlens[i]; j++) { 2872 if (mycols[jj] < cstart || mycols[jj] >= cend) offlens[i]++; 2873 jj++; 2874 } 2875 } 2876 2877 for (i=0; i<m; i++) { 2878 ourlens[i] -= offlens[i]; 2879 } 2880 ierr = MatSetSizes(newMat,m,n,M,N);CHKERRQ(ierr); 2881 2882 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 2883 2884 ierr = MatMPIAIJSetPreallocation(newMat,0,ourlens,0,offlens);CHKERRQ(ierr); 2885 2886 for (i=0; i<m; i++) { 2887 ourlens[i] += offlens[i]; 2888 } 2889 2890 if (!rank) { 2891 ierr = PetscMalloc1(maxnz+1,&vals);CHKERRQ(ierr); 2892 2893 /* read in my part of the matrix numerical values */ 2894 nz = procsnz[0]; 2895 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2896 2897 /* insert into matrix */ 2898 jj = rstart; 2899 smycols = mycols; 2900 svals = vals; 2901 for (i=0; i<m; i++) { 2902 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2903 smycols += ourlens[i]; 2904 svals += ourlens[i]; 2905 jj++; 2906 } 2907 2908 /* read in other processors and ship out */ 2909 for (i=1; i<size; i++) { 2910 nz = procsnz[i]; 2911 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2912 ierr = MPIULong_Send(vals,nz,MPIU_SCALAR,i,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 2913 } 2914 ierr = PetscFree(procsnz);CHKERRQ(ierr); 2915 } else { 2916 /* receive numeric values */ 2917 ierr = PetscMalloc1(nz+1,&vals);CHKERRQ(ierr); 2918 2919 /* receive message of values*/ 2920 ierr = MPIULong_Recv(vals,nz,MPIU_SCALAR,0,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 2921 2922 /* insert into matrix */ 2923 jj = rstart; 2924 smycols = mycols; 2925 svals = vals; 2926 for (i=0; i<m; i++) { 2927 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2928 smycols += ourlens[i]; 2929 svals += ourlens[i]; 2930 jj++; 2931 } 2932 } 2933 ierr = PetscFree2(ourlens,offlens);CHKERRQ(ierr); 2934 ierr = PetscFree(vals);CHKERRQ(ierr); 2935 ierr = PetscFree(mycols);CHKERRQ(ierr); 2936 ierr = PetscFree(rowners);CHKERRQ(ierr); 2937 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2938 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2939 PetscFunctionReturn(0); 2940 } 2941 2942 /* TODO: Not scalable because of ISAllGather() unless getting all columns. */ 2943 PetscErrorCode MatCreateSubMatrix_MPIAIJ(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *newmat) 2944 { 2945 PetscErrorCode ierr; 2946 IS iscol_local; 2947 PetscInt csize; 2948 2949 PetscFunctionBegin; 2950 ierr = ISGetLocalSize(iscol,&csize);CHKERRQ(ierr); 2951 if (call == MAT_REUSE_MATRIX) { 2952 ierr = PetscObjectQuery((PetscObject)*newmat,"ISAllGather",(PetscObject*)&iscol_local);CHKERRQ(ierr); 2953 if (!iscol_local) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 2954 } else { 2955 /* check if we are grabbing all columns*/ 2956 PetscBool isstride; 2957 PetscMPIInt lisstride = 0,gisstride; 2958 ierr = PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&isstride);CHKERRQ(ierr); 2959 if (isstride) { 2960 PetscInt start,len,mstart,mlen; 2961 ierr = ISStrideGetInfo(iscol,&start,NULL);CHKERRQ(ierr); 2962 ierr = ISGetLocalSize(iscol,&len);CHKERRQ(ierr); 2963 ierr = MatGetOwnershipRangeColumn(mat,&mstart,&mlen);CHKERRQ(ierr); 2964 if (mstart == start && mlen-mstart == len) lisstride = 1; 2965 } 2966 ierr = MPIU_Allreduce(&lisstride,&gisstride,1,MPI_INT,MPI_MIN,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 2967 if (gisstride) { 2968 PetscInt N; 2969 ierr = MatGetSize(mat,NULL,&N);CHKERRQ(ierr); 2970 ierr = ISCreateStride(PetscObjectComm((PetscObject)mat),N,0,1,&iscol_local);CHKERRQ(ierr); 2971 ierr = ISSetIdentity(iscol_local);CHKERRQ(ierr); 2972 ierr = PetscInfo(mat,"Optimizing for obtaining all columns of the matrix; skipping ISAllGather()\n");CHKERRQ(ierr); 2973 } else { 2974 PetscInt cbs; 2975 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2976 ierr = ISAllGather(iscol,&iscol_local);CHKERRQ(ierr); 2977 ierr = ISSetBlockSize(iscol_local,cbs);CHKERRQ(ierr); 2978 } 2979 } 2980 2981 ierr = MatCreateSubMatrix_MPIAIJ_Private(mat,isrow,iscol_local,csize,call,newmat);CHKERRQ(ierr); 2982 if (call == MAT_INITIAL_MATRIX) { 2983 ierr = PetscObjectCompose((PetscObject)*newmat,"ISAllGather",(PetscObject)iscol_local);CHKERRQ(ierr); 2984 ierr = ISDestroy(&iscol_local);CHKERRQ(ierr); 2985 } 2986 PetscFunctionReturn(0); 2987 } 2988 2989 extern PetscErrorCode MatCreateSubMatrices_MPIAIJ_Local(Mat,PetscInt,const IS[],const IS[],MatReuse,Mat*); 2990 extern PetscErrorCode MatCreateSubMatrices_MPIAIJ_SingleIS_Local(Mat,PetscInt,const IS[],const IS[],MatReuse,PetscBool,Mat*); 2991 2992 PetscErrorCode MatCreateSubMatrix_MPIAIJ_Private_SameDist(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat) 2993 { 2994 PetscErrorCode ierr; 2995 PetscMPIInt rank,size; 2996 PetscInt i,m,n,rstart,row,rend,nz,*cwork,j,bs,cbs; 2997 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 2998 Mat M,Mreuse; 2999 MatScalar *vwork,*aa; 3000 MPI_Comm comm; 3001 Mat_SeqAIJ *aij; 3002 //PetscBool colflag,allcolumns=PETSC_FALSE; 3003 3004 PetscFunctionBegin; 3005 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3006 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3007 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3008 if (!rank) printf("MatCreateSubMatrix_MPIAIJ_Private_SameDist...\n"); 3009 3010 //ierr = ISView(isrow,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); 3011 //ierr = ISSorted(iscol,&colflag);CHKERRQ(ierr); 3012 //if (!colflag) printf("[%d] iscol is not sorted!\n",rank); 3013 3014 /* Check for special case: each processor gets entire matrix columns */ 3015 //ierr = ISIdentity(iscol,&colflag);CHKERRQ(ierr); 3016 ierr = ISGetLocalSize(iscol,&nlocal);CHKERRQ(ierr); 3017 //if (colflag && nlocal == mat->cmap->N) allcolumns = PETSC_TRUE; //not working yet!!! 3018 3019 //==================== new ==================== 3020 Mat_MPIAIJ *a = (Mat_MPIAIJ*)mat->data; 3021 Mat A = a->A,B = a->B; 3022 PetscInt *garray = a->garray; 3023 PetscInt count = 0,An=A->cmap->N,Bn=B->cmap->N,cstart=mat->cmap->rstart,cend=mat->cmap->rend; 3024 const PetscInt *is_idx; 3025 IS iscol_local = iscol; 3026 3027 if (rank==-1) { 3028 printf("[%d] iscol_local:\n",rank); 3029 ierr = ISView(iscol_local,0);CHKERRQ(ierr); 3030 } 3031 3032 //printf("[%d] cstart/end: %d - %d; cstart/end: %d - %d; nlocal %d; mat: %d %d\n",rank,mat->rmap->n,mat->cmap->n,cstart,cend,nlocal,mat->rmap->N,mat->cmap->N); 3033 ierr = ISGetIndices(iscol_local,&is_idx);CHKERRQ(ierr); 3034 3035 /* (1) create a reduced scalable iscol_loc */ 3036 IS iscol_loc; 3037 PetscInt idx[nlocal],cmap[nlocal]; 3038 count = 0; 3039 3040 /* A part */ 3041 j = cstart; 3042 for (i=0; i<nlocal; i++) { 3043 if (j >= cend) break; 3044 if (is_idx[i] == j) { 3045 idx[count] = j; 3046 cmap[count] = i; /* column index in submat */ 3047 count++; j++; 3048 } else if (is_idx[i] > j) { 3049 while (is_idx[i] > j && j < cend-1) j++; 3050 if (is_idx[i] == j) { 3051 idx[count] = j; 3052 cmap[count] = i; /* column index in submat */ 3053 count++; j++; 3054 } 3055 } 3056 } 3057 3058 /* B part */ 3059 j = 0; 3060 for (i=0; i<nlocal; i++) { 3061 if (j >= Bn) break; 3062 if (is_idx[i] == garray[j]) { 3063 idx[count] = garray[j]; 3064 cmap[count] = i; /* column index in submat */ 3065 count++; j++; 3066 } else if (is_idx[i] > garray[j]) { 3067 while (is_idx[i] > garray[j] && j < Bn-1) j++; 3068 if (is_idx[i] == garray[j]) { 3069 idx[count] = garray[j]; 3070 cmap[count] = i; /* column index in submat */ 3071 count++; j++; 3072 } 3073 } 3074 } 3075 ierr = ISRestoreIndices(iscol_local,&is_idx);CHKERRQ(ierr); 3076 //printf("[%d] nlocal %d, N %d, Bn %d + An %d = %d, count %d\n",rank,nlocal,mat->cmap->N,Bn,An,Bn+An,count); 3077 3078 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)iscol_local),count,idx,PETSC_COPY_VALUES,&iscol_loc);CHKERRQ(ierr); 3079 ierr = ISSort(iscol_loc);CHKERRQ(ierr); 3080 ierr = PetscSortInt(count,cmap);CHKERRQ(ierr); 3081 if (rank==-1) { 3082 printf("[%d] iscol_loc:\n",rank); 3083 ierr = ISView(iscol_loc,0);CHKERRQ(ierr); 3084 for (i=0; i<count; i++) printf("cmap[%d] = %d\n",i,cmap[i]); 3085 } 3086 3087 Mat Msub; 3088 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol_loc,MAT_INITIAL_MATRIX,PETSC_FALSE,&Msub);CHKERRQ(ierr); 3089 if (rank==-1) { 3090 printf("[%d] Msub: %d %d \n",rank,Msub->rmap->N,Msub->cmap->N); 3091 ierr = MatView(Msub,0);CHKERRQ(ierr); 3092 } 3093 ierr = ISDestroy(&iscol_loc);CHKERRQ(ierr); 3094 //========================================================== 3095 3096 if (call == MAT_REUSE_MATRIX) { 3097 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject*)&Mreuse);CHKERRQ(ierr); 3098 if (!Mreuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3099 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol,MAT_REUSE_MATRIX,PETSC_FALSE,&Mreuse);CHKERRQ(ierr); 3100 } else { 3101 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol,MAT_INITIAL_MATRIX,PETSC_FALSE,&Mreuse);CHKERRQ(ierr); 3102 if (rank==-11) { 3103 printf("[%d] Msub: %d %d \n",rank,Msub->rmap->N,Msub->cmap->N); 3104 ierr = MatView(Mreuse,0);CHKERRQ(ierr); 3105 } 3106 } 3107 3108 /* 3109 m - number of local rows 3110 n - number of columns (same on all processors) 3111 rstart - first row in new global matrix generated 3112 */ 3113 ierr = MatGetSize(Mreuse,&m,&n);CHKERRQ(ierr); 3114 ierr = MatGetBlockSizes(Mreuse,&bs,&cbs);CHKERRQ(ierr); 3115 if (call == MAT_INITIAL_MATRIX) { 3116 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3117 ii = aij->i; 3118 jj = aij->j; 3119 3120 /* 3121 Determine the number of non-zeros in the diagonal and off-diagonal 3122 portions of the matrix in order to do correct preallocation 3123 */ 3124 3125 /* first get start and end of "diagonal" columns */ 3126 if (csize == PETSC_DECIDE) { 3127 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 3128 if (mglobal == n) { /* square matrix */ 3129 nlocal = m; 3130 } else { 3131 nlocal = n/size + ((n % size) > rank); 3132 } 3133 } else { 3134 nlocal = csize; 3135 } 3136 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3137 rstart = rend - nlocal; 3138 if (rank == size - 1 && rend != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Local column sizes %D do not add up to total number of columns %D",rend,n); 3139 3140 /* next, compute all the lengths */ 3141 ierr = PetscMalloc1(2*m+1,&dlens);CHKERRQ(ierr); 3142 olens = dlens + m; 3143 for (i=0; i<m; i++) { 3144 jend = ii[i+1] - ii[i]; 3145 olen = 0; 3146 dlen = 0; 3147 for (j=0; j<jend; j++) { 3148 if (*jj < rstart || *jj >= rend) olen++; 3149 else dlen++; 3150 jj++; 3151 } 3152 olens[i] = olen; 3153 dlens[i] = dlen; 3154 } 3155 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 3156 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,n);CHKERRQ(ierr); 3157 ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); 3158 ierr = MatSetType(M,((PetscObject)mat)->type_name);CHKERRQ(ierr); 3159 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 3160 ierr = PetscFree(dlens);CHKERRQ(ierr); 3161 } else { 3162 PetscInt ml,nl; 3163 3164 M = *newmat; 3165 ierr = MatGetLocalSize(M,&ml,&nl);CHKERRQ(ierr); 3166 if (ml != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 3167 ierr = MatZeroEntries(M);CHKERRQ(ierr); 3168 /* 3169 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 3170 rather than the slower MatSetValues(). 3171 */ 3172 M->was_assembled = PETSC_TRUE; 3173 M->assembled = PETSC_FALSE; 3174 } 3175 3176 //========================== 3177 Mat_SeqAIJ *aijsub = (Mat_SeqAIJ*)(Msub)->data; 3178 PetscInt *iisub = aijsub->i,*jjsub = aijsub->j,nzsub,colsub[count]; 3179 //=========================== 3180 3181 ierr = MatGetOwnershipRange(M,&rstart,&rend);CHKERRQ(ierr); 3182 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3183 ii = aij->i; 3184 jj = aij->j; 3185 aa = aijsub->a; 3186 for (i=0; i<m; i++) { 3187 row = rstart + i; 3188 nz = ii[i+1] - ii[i]; 3189 cwork = jj; jj += nz; 3190 vwork = aa; aa += nz; 3191 3192 nzsub = iisub[i+1] - iisub[i]; 3193 if (nzsub != nz) printf(" nzsub != nz\n"); 3194 3195 for (j=0; j<nz; j++) { 3196 colsub[j] = cmap[jjsub[iisub[i]+j]]; 3197 if (colsub[j] != cwork[j]) { 3198 printf("[%d] row %d, colsub %d != %d\n",rank,i,colsub[j],cwork[j]); 3199 } 3200 } 3201 3202 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,colsub,vwork,INSERT_VALUES);CHKERRQ(ierr); 3203 } 3204 3205 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3206 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3207 *newmat = M; 3208 //ierr = MatView(M,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); 3209 3210 /* save submatrix used in processor for next request */ 3211 if (call == MAT_INITIAL_MATRIX) { 3212 ierr = PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);CHKERRQ(ierr); 3213 ierr = MatDestroy(&Mreuse);CHKERRQ(ierr); 3214 } 3215 3216 ierr = MatDestroy(&Msub);CHKERRQ(ierr); 3217 PetscFunctionReturn(0); 3218 } 3219 3220 /* 3221 Not great since it makes two copies of the submatrix, first an SeqAIJ 3222 in local and then by concatenating the local matrices the end result. 3223 Writing it directly would be much like MatCreateSubMatrices_MPIAIJ() 3224 3225 Note: This requires a sequential iscol with all indices. 3226 */ 3227 PetscErrorCode MatCreateSubMatrix_MPIAIJ_Private(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat) 3228 { 3229 PetscErrorCode ierr; 3230 PetscMPIInt rank,size; 3231 PetscInt i,m,n,rstart,row,rend,nz,*cwork,j,bs,cbs; 3232 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 3233 Mat M,Mreuse; 3234 MatScalar *vwork,*aa; 3235 MPI_Comm comm; 3236 Mat_SeqAIJ *aij; 3237 PetscBool sameDist=PETSC_FALSE,tsameDist; 3238 3239 PetscFunctionBegin; 3240 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3241 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3242 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3243 3244 /* If isrow has same processor distribution as mat, then use a scalable routine */ 3245 ierr = ISGetLocalSize(isrow,&n);CHKERRQ(ierr); 3246 if (!n) { 3247 sameDist=PETSC_TRUE; 3248 } else { 3249 ierr = ISGetMinMax(isrow,&i,&j);CHKERRQ(ierr); 3250 ierr = MatGetOwnershipRange(mat,&rstart,&rend);CHKERRQ(ierr); 3251 if (i >= rstart && j < rend) sameDist=PETSC_TRUE; 3252 } 3253 ierr = MPIU_Allreduce(&sameDist,&tsameDist,1,MPIU_BOOL,MPI_LAND,comm);CHKERRQ(ierr); 3254 if (tsameDist) { 3255 ierr = MatCreateSubMatrix_MPIAIJ_Private_SameDist(mat,isrow,iscol,csize,call,newmat);CHKERRQ(ierr); 3256 PetscFunctionReturn(0); 3257 } 3258 3259 if (call == MAT_REUSE_MATRIX) { 3260 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject*)&Mreuse);CHKERRQ(ierr); 3261 if (!Mreuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3262 ierr = MatCreateSubMatrices_MPIAIJ_Local(mat,1,&isrow,&iscol,MAT_REUSE_MATRIX,&Mreuse);CHKERRQ(ierr); 3263 } else { 3264 ierr = MatCreateSubMatrices_MPIAIJ_Local(mat,1,&isrow,&iscol,MAT_INITIAL_MATRIX,&Mreuse);CHKERRQ(ierr); 3265 } 3266 3267 /* 3268 m - number of local rows 3269 n - number of columns (same on all processors) 3270 rstart - first row in new global matrix generated 3271 */ 3272 ierr = MatGetSize(Mreuse,&m,&n);CHKERRQ(ierr); 3273 ierr = MatGetBlockSizes(Mreuse,&bs,&cbs);CHKERRQ(ierr); 3274 if (call == MAT_INITIAL_MATRIX) { 3275 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3276 ii = aij->i; 3277 jj = aij->j; 3278 3279 /* 3280 Determine the number of non-zeros in the diagonal and off-diagonal 3281 portions of the matrix in order to do correct preallocation 3282 */ 3283 3284 /* first get start and end of "diagonal" columns */ 3285 if (csize == PETSC_DECIDE) { 3286 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 3287 if (mglobal == n) { /* square matrix */ 3288 nlocal = m; 3289 } else { 3290 nlocal = n/size + ((n % size) > rank); 3291 } 3292 } else { 3293 nlocal = csize; 3294 } 3295 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3296 rstart = rend - nlocal; 3297 if (rank == size - 1 && rend != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Local column sizes %D do not add up to total number of columns %D",rend,n); 3298 3299 /* next, compute all the lengths */ 3300 ierr = PetscMalloc1(2*m+1,&dlens);CHKERRQ(ierr); 3301 olens = dlens + m; 3302 for (i=0; i<m; i++) { 3303 jend = ii[i+1] - ii[i]; 3304 olen = 0; 3305 dlen = 0; 3306 for (j=0; j<jend; j++) { 3307 if (*jj < rstart || *jj >= rend) olen++; 3308 else dlen++; 3309 jj++; 3310 } 3311 olens[i] = olen; 3312 dlens[i] = dlen; 3313 } 3314 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 3315 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,n);CHKERRQ(ierr); 3316 ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); 3317 ierr = MatSetType(M,((PetscObject)mat)->type_name);CHKERRQ(ierr); 3318 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 3319 ierr = PetscFree(dlens);CHKERRQ(ierr); 3320 } else { 3321 PetscInt ml,nl; 3322 3323 M = *newmat; 3324 ierr = MatGetLocalSize(M,&ml,&nl);CHKERRQ(ierr); 3325 if (ml != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 3326 ierr = MatZeroEntries(M);CHKERRQ(ierr); 3327 /* 3328 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 3329 rather than the slower MatSetValues(). 3330 */ 3331 M->was_assembled = PETSC_TRUE; 3332 M->assembled = PETSC_FALSE; 3333 } 3334 ierr = MatGetOwnershipRange(M,&rstart,&rend);CHKERRQ(ierr); 3335 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3336 ii = aij->i; 3337 jj = aij->j; 3338 aa = aij->a; 3339 for (i=0; i<m; i++) { 3340 row = rstart + i; 3341 nz = ii[i+1] - ii[i]; 3342 cwork = jj; jj += nz; 3343 vwork = aa; aa += nz; 3344 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,cwork,vwork,INSERT_VALUES);CHKERRQ(ierr); 3345 } 3346 3347 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3348 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3349 *newmat = M; 3350 3351 /* save submatrix used in processor for next request */ 3352 if (call == MAT_INITIAL_MATRIX) { 3353 ierr = PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);CHKERRQ(ierr); 3354 ierr = MatDestroy(&Mreuse);CHKERRQ(ierr); 3355 } 3356 PetscFunctionReturn(0); 3357 } 3358 3359 PetscErrorCode MatMPIAIJSetPreallocationCSR_MPIAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3360 { 3361 PetscInt m,cstart, cend,j,nnz,i,d; 3362 PetscInt *d_nnz,*o_nnz,nnz_max = 0,rstart,ii; 3363 const PetscInt *JJ; 3364 PetscScalar *values; 3365 PetscErrorCode ierr; 3366 PetscBool nooffprocentries; 3367 3368 PetscFunctionBegin; 3369 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Ii[0] must be 0 it is %D",Ii[0]); 3370 3371 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3372 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3373 m = B->rmap->n; 3374 cstart = B->cmap->rstart; 3375 cend = B->cmap->rend; 3376 rstart = B->rmap->rstart; 3377 3378 ierr = PetscMalloc2(m,&d_nnz,m,&o_nnz);CHKERRQ(ierr); 3379 3380 #if defined(PETSC_USE_DEBUGGING) 3381 for (i=0; i<m; i++) { 3382 nnz = Ii[i+1]- Ii[i]; 3383 JJ = J + Ii[i]; 3384 if (nnz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Local row %D has a negative %D number of columns",i,nnz); 3385 if (nnz && (JJ[0] < 0)) SETERRRQ1(PETSC_ERR_ARG_WRONGSTATE,"Row %D starts with negative column index",i,j); 3386 if (nnz && (JJ[nnz-1] >= B->cmap->N) SETERRRQ3(PETSC_ERR_ARG_WRONGSTATE,"Row %D ends with too large a column index %D (max allowed %D)",i,JJ[nnz-1],B->cmap->N); 3387 } 3388 #endif 3389 3390 for (i=0; i<m; i++) { 3391 nnz = Ii[i+1]- Ii[i]; 3392 JJ = J + Ii[i]; 3393 nnz_max = PetscMax(nnz_max,nnz); 3394 d = 0; 3395 for (j=0; j<nnz; j++) { 3396 if (cstart <= JJ[j] && JJ[j] < cend) d++; 3397 } 3398 d_nnz[i] = d; 3399 o_nnz[i] = nnz - d; 3400 } 3401 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 3402 ierr = PetscFree2(d_nnz,o_nnz);CHKERRQ(ierr); 3403 3404 if (v) values = (PetscScalar*)v; 3405 else { 3406 ierr = PetscCalloc1(nnz_max+1,&values);CHKERRQ(ierr); 3407 } 3408 3409 for (i=0; i<m; i++) { 3410 ii = i + rstart; 3411 nnz = Ii[i+1]- Ii[i]; 3412 ierr = MatSetValues_MPIAIJ(B,1,&ii,nnz,J+Ii[i],values+(v ? Ii[i] : 0),INSERT_VALUES);CHKERRQ(ierr); 3413 } 3414 nooffprocentries = B->nooffprocentries; 3415 B->nooffprocentries = PETSC_TRUE; 3416 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3417 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3418 B->nooffprocentries = nooffprocentries; 3419 3420 if (!v) { 3421 ierr = PetscFree(values);CHKERRQ(ierr); 3422 } 3423 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3424 PetscFunctionReturn(0); 3425 } 3426 3427 /*@ 3428 MatMPIAIJSetPreallocationCSR - Allocates memory for a sparse parallel matrix in AIJ format 3429 (the default parallel PETSc format). 3430 3431 Collective on MPI_Comm 3432 3433 Input Parameters: 3434 + B - the matrix 3435 . i - the indices into j for the start of each local row (starts with zero) 3436 . j - the column indices for each local row (starts with zero) 3437 - v - optional values in the matrix 3438 3439 Level: developer 3440 3441 Notes: 3442 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3443 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3444 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3445 3446 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3447 3448 The format which is used for the sparse matrix input, is equivalent to a 3449 row-major ordering.. i.e for the following matrix, the input data expected is 3450 as shown 3451 3452 $ 1 0 0 3453 $ 2 0 3 P0 3454 $ ------- 3455 $ 4 5 6 P1 3456 $ 3457 $ Process0 [P0]: rows_owned=[0,1] 3458 $ i = {0,1,3} [size = nrow+1 = 2+1] 3459 $ j = {0,0,2} [size = 3] 3460 $ v = {1,2,3} [size = 3] 3461 $ 3462 $ Process1 [P1]: rows_owned=[2] 3463 $ i = {0,3} [size = nrow+1 = 1+1] 3464 $ j = {0,1,2} [size = 3] 3465 $ v = {4,5,6} [size = 3] 3466 3467 .keywords: matrix, aij, compressed row, sparse, parallel 3468 3469 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatCreateAIJ(), MATMPIAIJ, 3470 MatCreateSeqAIJWithArrays(), MatCreateMPIAIJWithSplitArrays() 3471 @*/ 3472 PetscErrorCode MatMPIAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[], const PetscScalar v[]) 3473 { 3474 PetscErrorCode ierr; 3475 3476 PetscFunctionBegin; 3477 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3478 PetscFunctionReturn(0); 3479 } 3480 3481 /*@C 3482 MatMPIAIJSetPreallocation - Preallocates memory for a sparse parallel matrix in AIJ format 3483 (the default parallel PETSc format). For good matrix assembly performance 3484 the user should preallocate the matrix storage by setting the parameters 3485 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3486 performance can be increased by more than a factor of 50. 3487 3488 Collective on MPI_Comm 3489 3490 Input Parameters: 3491 + B - the matrix 3492 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3493 (same value is used for all local rows) 3494 . d_nnz - array containing the number of nonzeros in the various rows of the 3495 DIAGONAL portion of the local submatrix (possibly different for each row) 3496 or NULL (PETSC_NULL_INTEGER in Fortran), if d_nz is used to specify the nonzero structure. 3497 The size of this array is equal to the number of local rows, i.e 'm'. 3498 For matrices that will be factored, you must leave room for (and set) 3499 the diagonal entry even if it is zero. 3500 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3501 submatrix (same value is used for all local rows). 3502 - o_nnz - array containing the number of nonzeros in the various rows of the 3503 OFF-DIAGONAL portion of the local submatrix (possibly different for 3504 each row) or NULL (PETSC_NULL_INTEGER in Fortran), if o_nz is used to specify the nonzero 3505 structure. The size of this array is equal to the number 3506 of local rows, i.e 'm'. 3507 3508 If the *_nnz parameter is given then the *_nz parameter is ignored 3509 3510 The AIJ format (also called the Yale sparse matrix format or 3511 compressed row storage (CSR)), is fully compatible with standard Fortran 77 3512 storage. The stored row and column indices begin with zero. 3513 See Users-Manual: ch_mat for details. 3514 3515 The parallel matrix is partitioned such that the first m0 rows belong to 3516 process 0, the next m1 rows belong to process 1, the next m2 rows belong 3517 to process 2 etc.. where m0,m1,m2... are the input parameter 'm'. 3518 3519 The DIAGONAL portion of the local submatrix of a processor can be defined 3520 as the submatrix which is obtained by extraction the part corresponding to 3521 the rows r1-r2 and columns c1-c2 of the global matrix, where r1 is the 3522 first row that belongs to the processor, r2 is the last row belonging to 3523 the this processor, and c1-c2 is range of indices of the local part of a 3524 vector suitable for applying the matrix to. This is an mxn matrix. In the 3525 common case of a square matrix, the row and column ranges are the same and 3526 the DIAGONAL part is also square. The remaining portion of the local 3527 submatrix (mxN) constitute the OFF-DIAGONAL portion. 3528 3529 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3530 3531 You can call MatGetInfo() to get information on how effective the preallocation was; 3532 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3533 You can also run with the option -info and look for messages with the string 3534 malloc in them to see if additional memory allocation was needed. 3535 3536 Example usage: 3537 3538 Consider the following 8x8 matrix with 34 non-zero values, that is 3539 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 3540 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 3541 as follows: 3542 3543 .vb 3544 1 2 0 | 0 3 0 | 0 4 3545 Proc0 0 5 6 | 7 0 0 | 8 0 3546 9 0 10 | 11 0 0 | 12 0 3547 ------------------------------------- 3548 13 0 14 | 15 16 17 | 0 0 3549 Proc1 0 18 0 | 19 20 21 | 0 0 3550 0 0 0 | 22 23 0 | 24 0 3551 ------------------------------------- 3552 Proc2 25 26 27 | 0 0 28 | 29 0 3553 30 0 0 | 31 32 33 | 0 34 3554 .ve 3555 3556 This can be represented as a collection of submatrices as: 3557 3558 .vb 3559 A B C 3560 D E F 3561 G H I 3562 .ve 3563 3564 Where the submatrices A,B,C are owned by proc0, D,E,F are 3565 owned by proc1, G,H,I are owned by proc2. 3566 3567 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3568 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3569 The 'M','N' parameters are 8,8, and have the same values on all procs. 3570 3571 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 3572 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 3573 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 3574 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 3575 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 3576 matrix, ans [DF] as another SeqAIJ matrix. 3577 3578 When d_nz, o_nz parameters are specified, d_nz storage elements are 3579 allocated for every row of the local diagonal submatrix, and o_nz 3580 storage locations are allocated for every row of the OFF-DIAGONAL submat. 3581 One way to choose d_nz and o_nz is to use the max nonzerors per local 3582 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 3583 In this case, the values of d_nz,o_nz are: 3584 .vb 3585 proc0 : dnz = 2, o_nz = 2 3586 proc1 : dnz = 3, o_nz = 2 3587 proc2 : dnz = 1, o_nz = 4 3588 .ve 3589 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3590 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3591 for proc3. i.e we are using 12+15+10=37 storage locations to store 3592 34 values. 3593 3594 When d_nnz, o_nnz parameters are specified, the storage is specified 3595 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3596 In the above case the values for d_nnz,o_nnz are: 3597 .vb 3598 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3599 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3600 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3601 .ve 3602 Here the space allocated is sum of all the above values i.e 34, and 3603 hence pre-allocation is perfect. 3604 3605 Level: intermediate 3606 3607 .keywords: matrix, aij, compressed row, sparse, parallel 3608 3609 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateAIJ(), MatMPIAIJSetPreallocationCSR(), 3610 MATMPIAIJ, MatGetInfo(), PetscSplitOwnership() 3611 @*/ 3612 PetscErrorCode MatMPIAIJSetPreallocation(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 3613 { 3614 PetscErrorCode ierr; 3615 3616 PetscFunctionBegin; 3617 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3618 PetscValidType(B,1); 3619 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[]),(B,d_nz,d_nnz,o_nz,o_nnz));CHKERRQ(ierr); 3620 PetscFunctionReturn(0); 3621 } 3622 3623 /*@ 3624 MatCreateMPIAIJWithArrays - creates a MPI AIJ matrix using arrays that contain in standard 3625 CSR format the local rows. 3626 3627 Collective on MPI_Comm 3628 3629 Input Parameters: 3630 + comm - MPI communicator 3631 . m - number of local rows (Cannot be PETSC_DECIDE) 3632 . n - This value should be the same as the local size used in creating the 3633 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3634 calculated if N is given) For square matrices n is almost always m. 3635 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3636 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3637 . i - row indices 3638 . j - column indices 3639 - a - matrix values 3640 3641 Output Parameter: 3642 . mat - the matrix 3643 3644 Level: intermediate 3645 3646 Notes: 3647 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3648 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3649 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3650 3651 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3652 3653 The format which is used for the sparse matrix input, is equivalent to a 3654 row-major ordering.. i.e for the following matrix, the input data expected is 3655 as shown 3656 3657 $ 1 0 0 3658 $ 2 0 3 P0 3659 $ ------- 3660 $ 4 5 6 P1 3661 $ 3662 $ Process0 [P0]: rows_owned=[0,1] 3663 $ i = {0,1,3} [size = nrow+1 = 2+1] 3664 $ j = {0,0,2} [size = 3] 3665 $ v = {1,2,3} [size = 3] 3666 $ 3667 $ Process1 [P1]: rows_owned=[2] 3668 $ i = {0,3} [size = nrow+1 = 1+1] 3669 $ j = {0,1,2} [size = 3] 3670 $ v = {4,5,6} [size = 3] 3671 3672 .keywords: matrix, aij, compressed row, sparse, parallel 3673 3674 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 3675 MATMPIAIJ, MatCreateAIJ(), MatCreateMPIAIJWithSplitArrays() 3676 @*/ 3677 PetscErrorCode MatCreateMPIAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,const PetscInt i[],const PetscInt j[],const PetscScalar a[],Mat *mat) 3678 { 3679 PetscErrorCode ierr; 3680 3681 PetscFunctionBegin; 3682 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 3683 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 3684 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 3685 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 3686 /* ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); */ 3687 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 3688 ierr = MatMPIAIJSetPreallocationCSR(*mat,i,j,a);CHKERRQ(ierr); 3689 PetscFunctionReturn(0); 3690 } 3691 3692 /*@C 3693 MatCreateAIJ - Creates a sparse parallel matrix in AIJ format 3694 (the default parallel PETSc format). For good matrix assembly performance 3695 the user should preallocate the matrix storage by setting the parameters 3696 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3697 performance can be increased by more than a factor of 50. 3698 3699 Collective on MPI_Comm 3700 3701 Input Parameters: 3702 + comm - MPI communicator 3703 . m - number of local rows (or PETSC_DECIDE to have calculated if M is given) 3704 This value should be the same as the local size used in creating the 3705 y vector for the matrix-vector product y = Ax. 3706 . n - This value should be the same as the local size used in creating the 3707 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3708 calculated if N is given) For square matrices n is almost always m. 3709 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3710 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3711 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3712 (same value is used for all local rows) 3713 . d_nnz - array containing the number of nonzeros in the various rows of the 3714 DIAGONAL portion of the local submatrix (possibly different for each row) 3715 or NULL, if d_nz is used to specify the nonzero structure. 3716 The size of this array is equal to the number of local rows, i.e 'm'. 3717 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3718 submatrix (same value is used for all local rows). 3719 - o_nnz - array containing the number of nonzeros in the various rows of the 3720 OFF-DIAGONAL portion of the local submatrix (possibly different for 3721 each row) or NULL, if o_nz is used to specify the nonzero 3722 structure. The size of this array is equal to the number 3723 of local rows, i.e 'm'. 3724 3725 Output Parameter: 3726 . A - the matrix 3727 3728 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3729 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3730 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3731 3732 Notes: 3733 If the *_nnz parameter is given then the *_nz parameter is ignored 3734 3735 m,n,M,N parameters specify the size of the matrix, and its partitioning across 3736 processors, while d_nz,d_nnz,o_nz,o_nnz parameters specify the approximate 3737 storage requirements for this matrix. 3738 3739 If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one 3740 processor than it must be used on all processors that share the object for 3741 that argument. 3742 3743 The user MUST specify either the local or global matrix dimensions 3744 (possibly both). 3745 3746 The parallel matrix is partitioned across processors such that the 3747 first m0 rows belong to process 0, the next m1 rows belong to 3748 process 1, the next m2 rows belong to process 2 etc.. where 3749 m0,m1,m2,.. are the input parameter 'm'. i.e each processor stores 3750 values corresponding to [m x N] submatrix. 3751 3752 The columns are logically partitioned with the n0 columns belonging 3753 to 0th partition, the next n1 columns belonging to the next 3754 partition etc.. where n0,n1,n2... are the input parameter 'n'. 3755 3756 The DIAGONAL portion of the local submatrix on any given processor 3757 is the submatrix corresponding to the rows and columns m,n 3758 corresponding to the given processor. i.e diagonal matrix on 3759 process 0 is [m0 x n0], diagonal matrix on process 1 is [m1 x n1] 3760 etc. The remaining portion of the local submatrix [m x (N-n)] 3761 constitute the OFF-DIAGONAL portion. The example below better 3762 illustrates this concept. 3763 3764 For a square global matrix we define each processor's diagonal portion 3765 to be its local rows and the corresponding columns (a square submatrix); 3766 each processor's off-diagonal portion encompasses the remainder of the 3767 local matrix (a rectangular submatrix). 3768 3769 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3770 3771 When calling this routine with a single process communicator, a matrix of 3772 type SEQAIJ is returned. If a matrix of type MATMPIAIJ is desired for this 3773 type of communicator, use the construction mechanism: 3774 MatCreate(...,&A); MatSetType(A,MATMPIAIJ); MatSetSizes(A, m,n,M,N); MatMPIAIJSetPreallocation(A,...); 3775 3776 By default, this format uses inodes (identical nodes) when possible. 3777 We search for consecutive rows with the same nonzero structure, thereby 3778 reusing matrix information to achieve increased efficiency. 3779 3780 Options Database Keys: 3781 + -mat_no_inode - Do not use inodes 3782 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3783 - -mat_aij_oneindex - Internally use indexing starting at 1 3784 rather than 0. Note that when calling MatSetValues(), 3785 the user still MUST index entries starting at 0! 3786 3787 3788 Example usage: 3789 3790 Consider the following 8x8 matrix with 34 non-zero values, that is 3791 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 3792 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 3793 as follows: 3794 3795 .vb 3796 1 2 0 | 0 3 0 | 0 4 3797 Proc0 0 5 6 | 7 0 0 | 8 0 3798 9 0 10 | 11 0 0 | 12 0 3799 ------------------------------------- 3800 13 0 14 | 15 16 17 | 0 0 3801 Proc1 0 18 0 | 19 20 21 | 0 0 3802 0 0 0 | 22 23 0 | 24 0 3803 ------------------------------------- 3804 Proc2 25 26 27 | 0 0 28 | 29 0 3805 30 0 0 | 31 32 33 | 0 34 3806 .ve 3807 3808 This can be represented as a collection of submatrices as: 3809 3810 .vb 3811 A B C 3812 D E F 3813 G H I 3814 .ve 3815 3816 Where the submatrices A,B,C are owned by proc0, D,E,F are 3817 owned by proc1, G,H,I are owned by proc2. 3818 3819 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3820 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3821 The 'M','N' parameters are 8,8, and have the same values on all procs. 3822 3823 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 3824 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 3825 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 3826 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 3827 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 3828 matrix, ans [DF] as another SeqAIJ matrix. 3829 3830 When d_nz, o_nz parameters are specified, d_nz storage elements are 3831 allocated for every row of the local diagonal submatrix, and o_nz 3832 storage locations are allocated for every row of the OFF-DIAGONAL submat. 3833 One way to choose d_nz and o_nz is to use the max nonzerors per local 3834 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 3835 In this case, the values of d_nz,o_nz are: 3836 .vb 3837 proc0 : dnz = 2, o_nz = 2 3838 proc1 : dnz = 3, o_nz = 2 3839 proc2 : dnz = 1, o_nz = 4 3840 .ve 3841 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3842 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3843 for proc3. i.e we are using 12+15+10=37 storage locations to store 3844 34 values. 3845 3846 When d_nnz, o_nnz parameters are specified, the storage is specified 3847 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3848 In the above case the values for d_nnz,o_nnz are: 3849 .vb 3850 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3851 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3852 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3853 .ve 3854 Here the space allocated is sum of all the above values i.e 34, and 3855 hence pre-allocation is perfect. 3856 3857 Level: intermediate 3858 3859 .keywords: matrix, aij, compressed row, sparse, parallel 3860 3861 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 3862 MATMPIAIJ, MatCreateMPIAIJWithArrays() 3863 @*/ 3864 PetscErrorCode MatCreateAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[],Mat *A) 3865 { 3866 PetscErrorCode ierr; 3867 PetscMPIInt size; 3868 3869 PetscFunctionBegin; 3870 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3871 ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr); 3872 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3873 if (size > 1) { 3874 ierr = MatSetType(*A,MATMPIAIJ);CHKERRQ(ierr); 3875 ierr = MatMPIAIJSetPreallocation(*A,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr); 3876 } else { 3877 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3878 ierr = MatSeqAIJSetPreallocation(*A,d_nz,d_nnz);CHKERRQ(ierr); 3879 } 3880 PetscFunctionReturn(0); 3881 } 3882 3883 PetscErrorCode MatMPIAIJGetSeqAIJ(Mat A,Mat *Ad,Mat *Ao,const PetscInt *colmap[]) 3884 { 3885 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 3886 PetscBool flg; 3887 PetscErrorCode ierr; 3888 3889 PetscFunctionBegin; 3890 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&flg);CHKERRQ(ierr); 3891 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"This function requires a MATMPIAIJ matrix as input"); 3892 if (Ad) *Ad = a->A; 3893 if (Ao) *Ao = a->B; 3894 if (colmap) *colmap = a->garray; 3895 PetscFunctionReturn(0); 3896 } 3897 3898 PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat) 3899 { 3900 PetscErrorCode ierr; 3901 PetscInt m,N,i,rstart,nnz,Ii; 3902 PetscInt *indx; 3903 PetscScalar *values; 3904 3905 PetscFunctionBegin; 3906 ierr = MatGetSize(inmat,&m,&N);CHKERRQ(ierr); 3907 if (scall == MAT_INITIAL_MATRIX) { /* symbolic phase */ 3908 PetscInt *dnz,*onz,sum,bs,cbs; 3909 3910 if (n == PETSC_DECIDE) { 3911 ierr = PetscSplitOwnership(comm,&n,&N);CHKERRQ(ierr); 3912 } 3913 /* Check sum(n) = N */ 3914 ierr = MPIU_Allreduce(&n,&sum,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3915 if (sum != N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Sum of local columns != global columns %d",N); 3916 3917 ierr = MPI_Scan(&m, &rstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3918 rstart -= m; 3919 3920 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 3921 for (i=0; i<m; i++) { 3922 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,NULL);CHKERRQ(ierr); 3923 ierr = MatPreallocateSet(i+rstart,nnz,indx,dnz,onz);CHKERRQ(ierr); 3924 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,NULL);CHKERRQ(ierr); 3925 } 3926 3927 ierr = MatCreate(comm,outmat);CHKERRQ(ierr); 3928 ierr = MatSetSizes(*outmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 3929 ierr = MatGetBlockSizes(inmat,&bs,&cbs);CHKERRQ(ierr); 3930 ierr = MatSetBlockSizes(*outmat,bs,cbs);CHKERRQ(ierr); 3931 ierr = MatSetType(*outmat,MATAIJ);CHKERRQ(ierr); 3932 ierr = MatSeqAIJSetPreallocation(*outmat,0,dnz);CHKERRQ(ierr); 3933 ierr = MatMPIAIJSetPreallocation(*outmat,0,dnz,0,onz);CHKERRQ(ierr); 3934 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 3935 } 3936 3937 /* numeric phase */ 3938 ierr = MatGetOwnershipRange(*outmat,&rstart,NULL);CHKERRQ(ierr); 3939 for (i=0; i<m; i++) { 3940 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 3941 Ii = i + rstart; 3942 ierr = MatSetValues(*outmat,1,&Ii,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 3943 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 3944 } 3945 ierr = MatAssemblyBegin(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3946 ierr = MatAssemblyEnd(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3947 PetscFunctionReturn(0); 3948 } 3949 3950 PetscErrorCode MatFileSplit(Mat A,char *outfile) 3951 { 3952 PetscErrorCode ierr; 3953 PetscMPIInt rank; 3954 PetscInt m,N,i,rstart,nnz; 3955 size_t len; 3956 const PetscInt *indx; 3957 PetscViewer out; 3958 char *name; 3959 Mat B; 3960 const PetscScalar *values; 3961 3962 PetscFunctionBegin; 3963 ierr = MatGetLocalSize(A,&m,0);CHKERRQ(ierr); 3964 ierr = MatGetSize(A,0,&N);CHKERRQ(ierr); 3965 /* Should this be the type of the diagonal block of A? */ 3966 ierr = MatCreate(PETSC_COMM_SELF,&B);CHKERRQ(ierr); 3967 ierr = MatSetSizes(B,m,N,m,N);CHKERRQ(ierr); 3968 ierr = MatSetBlockSizesFromMats(B,A,A);CHKERRQ(ierr); 3969 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 3970 ierr = MatSeqAIJSetPreallocation(B,0,NULL);CHKERRQ(ierr); 3971 ierr = MatGetOwnershipRange(A,&rstart,0);CHKERRQ(ierr); 3972 for (i=0; i<m; i++) { 3973 ierr = MatGetRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 3974 ierr = MatSetValues(B,1,&i,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 3975 ierr = MatRestoreRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 3976 } 3977 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3978 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3979 3980 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)A),&rank);CHKERRQ(ierr); 3981 ierr = PetscStrlen(outfile,&len);CHKERRQ(ierr); 3982 ierr = PetscMalloc1(len+5,&name);CHKERRQ(ierr); 3983 sprintf(name,"%s.%d",outfile,rank); 3984 ierr = PetscViewerBinaryOpen(PETSC_COMM_SELF,name,FILE_MODE_APPEND,&out);CHKERRQ(ierr); 3985 ierr = PetscFree(name);CHKERRQ(ierr); 3986 ierr = MatView(B,out);CHKERRQ(ierr); 3987 ierr = PetscViewerDestroy(&out);CHKERRQ(ierr); 3988 ierr = MatDestroy(&B);CHKERRQ(ierr); 3989 PetscFunctionReturn(0); 3990 } 3991 3992 PetscErrorCode MatDestroy_MPIAIJ_SeqsToMPI(Mat A) 3993 { 3994 PetscErrorCode ierr; 3995 Mat_Merge_SeqsToMPI *merge; 3996 PetscContainer container; 3997 3998 PetscFunctionBegin; 3999 ierr = PetscObjectQuery((PetscObject)A,"MatMergeSeqsToMPI",(PetscObject*)&container);CHKERRQ(ierr); 4000 if (container) { 4001 ierr = PetscContainerGetPointer(container,(void**)&merge);CHKERRQ(ierr); 4002 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 4003 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 4004 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 4005 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 4006 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 4007 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 4008 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 4009 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 4010 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 4011 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 4012 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 4013 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 4014 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 4015 ierr = PetscFree(merge);CHKERRQ(ierr); 4016 ierr = PetscObjectCompose((PetscObject)A,"MatMergeSeqsToMPI",0);CHKERRQ(ierr); 4017 } 4018 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 4019 PetscFunctionReturn(0); 4020 } 4021 4022 #include <../src/mat/utils/freespace.h> 4023 #include <petscbt.h> 4024 4025 PetscErrorCode MatCreateMPIAIJSumSeqAIJNumeric(Mat seqmat,Mat mpimat) 4026 { 4027 PetscErrorCode ierr; 4028 MPI_Comm comm; 4029 Mat_SeqAIJ *a =(Mat_SeqAIJ*)seqmat->data; 4030 PetscMPIInt size,rank,taga,*len_s; 4031 PetscInt N=mpimat->cmap->N,i,j,*owners,*ai=a->i,*aj; 4032 PetscInt proc,m; 4033 PetscInt **buf_ri,**buf_rj; 4034 PetscInt k,anzi,*bj_i,*bi,*bj,arow,bnzi,nextaj; 4035 PetscInt nrows,**buf_ri_k,**nextrow,**nextai; 4036 MPI_Request *s_waits,*r_waits; 4037 MPI_Status *status; 4038 MatScalar *aa=a->a; 4039 MatScalar **abuf_r,*ba_i; 4040 Mat_Merge_SeqsToMPI *merge; 4041 PetscContainer container; 4042 4043 PetscFunctionBegin; 4044 ierr = PetscObjectGetComm((PetscObject)mpimat,&comm);CHKERRQ(ierr); 4045 ierr = PetscLogEventBegin(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4046 4047 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4048 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4049 4050 ierr = PetscObjectQuery((PetscObject)mpimat,"MatMergeSeqsToMPI",(PetscObject*)&container);CHKERRQ(ierr); 4051 ierr = PetscContainerGetPointer(container,(void**)&merge);CHKERRQ(ierr); 4052 4053 bi = merge->bi; 4054 bj = merge->bj; 4055 buf_ri = merge->buf_ri; 4056 buf_rj = merge->buf_rj; 4057 4058 ierr = PetscMalloc1(size,&status);CHKERRQ(ierr); 4059 owners = merge->rowmap->range; 4060 len_s = merge->len_s; 4061 4062 /* send and recv matrix values */ 4063 /*-----------------------------*/ 4064 ierr = PetscObjectGetNewTag((PetscObject)mpimat,&taga);CHKERRQ(ierr); 4065 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 4066 4067 ierr = PetscMalloc1(merge->nsend+1,&s_waits);CHKERRQ(ierr); 4068 for (proc=0,k=0; proc<size; proc++) { 4069 if (!len_s[proc]) continue; 4070 i = owners[proc]; 4071 ierr = MPI_Isend(aa+ai[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 4072 k++; 4073 } 4074 4075 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 4076 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 4077 ierr = PetscFree(status);CHKERRQ(ierr); 4078 4079 ierr = PetscFree(s_waits);CHKERRQ(ierr); 4080 ierr = PetscFree(r_waits);CHKERRQ(ierr); 4081 4082 /* insert mat values of mpimat */ 4083 /*----------------------------*/ 4084 ierr = PetscMalloc1(N,&ba_i);CHKERRQ(ierr); 4085 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextai);CHKERRQ(ierr); 4086 4087 for (k=0; k<merge->nrecv; k++) { 4088 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4089 nrows = *(buf_ri_k[k]); 4090 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 4091 nextai[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 4092 } 4093 4094 /* set values of ba */ 4095 m = merge->rowmap->n; 4096 for (i=0; i<m; i++) { 4097 arow = owners[rank] + i; 4098 bj_i = bj+bi[i]; /* col indices of the i-th row of mpimat */ 4099 bnzi = bi[i+1] - bi[i]; 4100 ierr = PetscMemzero(ba_i,bnzi*sizeof(PetscScalar));CHKERRQ(ierr); 4101 4102 /* add local non-zero vals of this proc's seqmat into ba */ 4103 anzi = ai[arow+1] - ai[arow]; 4104 aj = a->j + ai[arow]; 4105 aa = a->a + ai[arow]; 4106 nextaj = 0; 4107 for (j=0; nextaj<anzi; j++) { 4108 if (*(bj_i + j) == aj[nextaj]) { /* bcol == acol */ 4109 ba_i[j] += aa[nextaj++]; 4110 } 4111 } 4112 4113 /* add received vals into ba */ 4114 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 4115 /* i-th row */ 4116 if (i == *nextrow[k]) { 4117 anzi = *(nextai[k]+1) - *nextai[k]; 4118 aj = buf_rj[k] + *(nextai[k]); 4119 aa = abuf_r[k] + *(nextai[k]); 4120 nextaj = 0; 4121 for (j=0; nextaj<anzi; j++) { 4122 if (*(bj_i + j) == aj[nextaj]) { /* bcol == acol */ 4123 ba_i[j] += aa[nextaj++]; 4124 } 4125 } 4126 nextrow[k]++; nextai[k]++; 4127 } 4128 } 4129 ierr = MatSetValues(mpimat,1,&arow,bnzi,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 4130 } 4131 ierr = MatAssemblyBegin(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4132 ierr = MatAssemblyEnd(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4133 4134 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 4135 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 4136 ierr = PetscFree(ba_i);CHKERRQ(ierr); 4137 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4138 ierr = PetscLogEventEnd(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4139 PetscFunctionReturn(0); 4140 } 4141 4142 PetscErrorCode MatCreateMPIAIJSumSeqAIJSymbolic(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,Mat *mpimat) 4143 { 4144 PetscErrorCode ierr; 4145 Mat B_mpi; 4146 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 4147 PetscMPIInt size,rank,tagi,tagj,*len_s,*len_si,*len_ri; 4148 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 4149 PetscInt M=seqmat->rmap->n,N=seqmat->cmap->n,i,*owners,*ai=a->i,*aj=a->j; 4150 PetscInt len,proc,*dnz,*onz,bs,cbs; 4151 PetscInt k,anzi,*bi,*bj,*lnk,nlnk,arow,bnzi,nspacedouble=0; 4152 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextai; 4153 MPI_Request *si_waits,*sj_waits,*ri_waits,*rj_waits; 4154 MPI_Status *status; 4155 PetscFreeSpaceList free_space=NULL,current_space=NULL; 4156 PetscBT lnkbt; 4157 Mat_Merge_SeqsToMPI *merge; 4158 PetscContainer container; 4159 4160 PetscFunctionBegin; 4161 ierr = PetscLogEventBegin(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4162 4163 /* make sure it is a PETSc comm */ 4164 ierr = PetscCommDuplicate(comm,&comm,NULL);CHKERRQ(ierr); 4165 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4166 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4167 4168 ierr = PetscNew(&merge);CHKERRQ(ierr); 4169 ierr = PetscMalloc1(size,&status);CHKERRQ(ierr); 4170 4171 /* determine row ownership */ 4172 /*---------------------------------------------------------*/ 4173 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 4174 ierr = PetscLayoutSetLocalSize(merge->rowmap,m);CHKERRQ(ierr); 4175 ierr = PetscLayoutSetSize(merge->rowmap,M);CHKERRQ(ierr); 4176 ierr = PetscLayoutSetBlockSize(merge->rowmap,1);CHKERRQ(ierr); 4177 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 4178 ierr = PetscMalloc1(size,&len_si);CHKERRQ(ierr); 4179 ierr = PetscMalloc1(size,&merge->len_s);CHKERRQ(ierr); 4180 4181 m = merge->rowmap->n; 4182 owners = merge->rowmap->range; 4183 4184 /* determine the number of messages to send, their lengths */ 4185 /*---------------------------------------------------------*/ 4186 len_s = merge->len_s; 4187 4188 len = 0; /* length of buf_si[] */ 4189 merge->nsend = 0; 4190 for (proc=0; proc<size; proc++) { 4191 len_si[proc] = 0; 4192 if (proc == rank) { 4193 len_s[proc] = 0; 4194 } else { 4195 len_si[proc] = owners[proc+1] - owners[proc] + 1; 4196 len_s[proc] = ai[owners[proc+1]] - ai[owners[proc]]; /* num of rows to be sent to [proc] */ 4197 } 4198 if (len_s[proc]) { 4199 merge->nsend++; 4200 nrows = 0; 4201 for (i=owners[proc]; i<owners[proc+1]; i++) { 4202 if (ai[i+1] > ai[i]) nrows++; 4203 } 4204 len_si[proc] = 2*(nrows+1); 4205 len += len_si[proc]; 4206 } 4207 } 4208 4209 /* determine the number and length of messages to receive for ij-structure */ 4210 /*-------------------------------------------------------------------------*/ 4211 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 4212 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 4213 4214 /* post the Irecv of j-structure */ 4215 /*-------------------------------*/ 4216 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 4217 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rj_waits);CHKERRQ(ierr); 4218 4219 /* post the Isend of j-structure */ 4220 /*--------------------------------*/ 4221 ierr = PetscMalloc2(merge->nsend,&si_waits,merge->nsend,&sj_waits);CHKERRQ(ierr); 4222 4223 for (proc=0, k=0; proc<size; proc++) { 4224 if (!len_s[proc]) continue; 4225 i = owners[proc]; 4226 ierr = MPI_Isend(aj+ai[i],len_s[proc],MPIU_INT,proc,tagj,comm,sj_waits+k);CHKERRQ(ierr); 4227 k++; 4228 } 4229 4230 /* receives and sends of j-structure are complete */ 4231 /*------------------------------------------------*/ 4232 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,rj_waits,status);CHKERRQ(ierr);} 4233 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,sj_waits,status);CHKERRQ(ierr);} 4234 4235 /* send and recv i-structure */ 4236 /*---------------------------*/ 4237 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 4238 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&ri_waits);CHKERRQ(ierr); 4239 4240 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 4241 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 4242 for (proc=0,k=0; proc<size; proc++) { 4243 if (!len_s[proc]) continue; 4244 /* form outgoing message for i-structure: 4245 buf_si[0]: nrows to be sent 4246 [1:nrows]: row index (global) 4247 [nrows+1:2*nrows+1]: i-structure index 4248 */ 4249 /*-------------------------------------------*/ 4250 nrows = len_si[proc]/2 - 1; 4251 buf_si_i = buf_si + nrows+1; 4252 buf_si[0] = nrows; 4253 buf_si_i[0] = 0; 4254 nrows = 0; 4255 for (i=owners[proc]; i<owners[proc+1]; i++) { 4256 anzi = ai[i+1] - ai[i]; 4257 if (anzi) { 4258 buf_si_i[nrows+1] = buf_si_i[nrows] + anzi; /* i-structure */ 4259 buf_si[nrows+1] = i-owners[proc]; /* local row index */ 4260 nrows++; 4261 } 4262 } 4263 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,si_waits+k);CHKERRQ(ierr); 4264 k++; 4265 buf_si += len_si[proc]; 4266 } 4267 4268 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,ri_waits,status);CHKERRQ(ierr);} 4269 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,si_waits,status);CHKERRQ(ierr);} 4270 4271 ierr = PetscInfo2(seqmat,"nsend: %D, nrecv: %D\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 4272 for (i=0; i<merge->nrecv; i++) { 4273 ierr = PetscInfo3(seqmat,"recv len_ri=%D, len_rj=%D from [%D]\n",len_ri[i],merge->len_r[i],merge->id_r[i]);CHKERRQ(ierr); 4274 } 4275 4276 ierr = PetscFree(len_si);CHKERRQ(ierr); 4277 ierr = PetscFree(len_ri);CHKERRQ(ierr); 4278 ierr = PetscFree(rj_waits);CHKERRQ(ierr); 4279 ierr = PetscFree2(si_waits,sj_waits);CHKERRQ(ierr); 4280 ierr = PetscFree(ri_waits);CHKERRQ(ierr); 4281 ierr = PetscFree(buf_s);CHKERRQ(ierr); 4282 ierr = PetscFree(status);CHKERRQ(ierr); 4283 4284 /* compute a local seq matrix in each processor */ 4285 /*----------------------------------------------*/ 4286 /* allocate bi array and free space for accumulating nonzero column info */ 4287 ierr = PetscMalloc1(m+1,&bi);CHKERRQ(ierr); 4288 bi[0] = 0; 4289 4290 /* create and initialize a linked list */ 4291 nlnk = N+1; 4292 ierr = PetscLLCreate(N,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4293 4294 /* initial FreeSpace size is 2*(num of local nnz(seqmat)) */ 4295 len = ai[owners[rank+1]] - ai[owners[rank]]; 4296 ierr = PetscFreeSpaceGet(PetscIntMultTruncate(2,len)+1,&free_space);CHKERRQ(ierr); 4297 4298 current_space = free_space; 4299 4300 /* determine symbolic info for each local row */ 4301 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextai);CHKERRQ(ierr); 4302 4303 for (k=0; k<merge->nrecv; k++) { 4304 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4305 nrows = *buf_ri_k[k]; 4306 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 4307 nextai[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 4308 } 4309 4310 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 4311 len = 0; 4312 for (i=0; i<m; i++) { 4313 bnzi = 0; 4314 /* add local non-zero cols of this proc's seqmat into lnk */ 4315 arow = owners[rank] + i; 4316 anzi = ai[arow+1] - ai[arow]; 4317 aj = a->j + ai[arow]; 4318 ierr = PetscLLAddSorted(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4319 bnzi += nlnk; 4320 /* add received col data into lnk */ 4321 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 4322 if (i == *nextrow[k]) { /* i-th row */ 4323 anzi = *(nextai[k]+1) - *nextai[k]; 4324 aj = buf_rj[k] + *nextai[k]; 4325 ierr = PetscLLAddSorted(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4326 bnzi += nlnk; 4327 nextrow[k]++; nextai[k]++; 4328 } 4329 } 4330 if (len < bnzi) len = bnzi; /* =max(bnzi) */ 4331 4332 /* if free space is not available, make more free space */ 4333 if (current_space->local_remaining<bnzi) { 4334 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(bnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 4335 nspacedouble++; 4336 } 4337 /* copy data into free space, then initialize lnk */ 4338 ierr = PetscLLClean(N,N,bnzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 4339 ierr = MatPreallocateSet(i+owners[rank],bnzi,current_space->array,dnz,onz);CHKERRQ(ierr); 4340 4341 current_space->array += bnzi; 4342 current_space->local_used += bnzi; 4343 current_space->local_remaining -= bnzi; 4344 4345 bi[i+1] = bi[i] + bnzi; 4346 } 4347 4348 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4349 4350 ierr = PetscMalloc1(bi[m]+1,&bj);CHKERRQ(ierr); 4351 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 4352 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 4353 4354 /* create symbolic parallel matrix B_mpi */ 4355 /*---------------------------------------*/ 4356 ierr = MatGetBlockSizes(seqmat,&bs,&cbs);CHKERRQ(ierr); 4357 ierr = MatCreate(comm,&B_mpi);CHKERRQ(ierr); 4358 if (n==PETSC_DECIDE) { 4359 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,N);CHKERRQ(ierr); 4360 } else { 4361 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 4362 } 4363 ierr = MatSetBlockSizes(B_mpi,bs,cbs);CHKERRQ(ierr); 4364 ierr = MatSetType(B_mpi,MATMPIAIJ);CHKERRQ(ierr); 4365 ierr = MatMPIAIJSetPreallocation(B_mpi,0,dnz,0,onz);CHKERRQ(ierr); 4366 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 4367 ierr = MatSetOption(B_mpi,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr); 4368 4369 /* B_mpi is not ready for use - assembly will be done by MatCreateMPIAIJSumSeqAIJNumeric() */ 4370 B_mpi->assembled = PETSC_FALSE; 4371 B_mpi->ops->destroy = MatDestroy_MPIAIJ_SeqsToMPI; 4372 merge->bi = bi; 4373 merge->bj = bj; 4374 merge->buf_ri = buf_ri; 4375 merge->buf_rj = buf_rj; 4376 merge->coi = NULL; 4377 merge->coj = NULL; 4378 merge->owners_co = NULL; 4379 4380 ierr = PetscCommDestroy(&comm);CHKERRQ(ierr); 4381 4382 /* attach the supporting struct to B_mpi for reuse */ 4383 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 4384 ierr = PetscContainerSetPointer(container,merge);CHKERRQ(ierr); 4385 ierr = PetscObjectCompose((PetscObject)B_mpi,"MatMergeSeqsToMPI",(PetscObject)container);CHKERRQ(ierr); 4386 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 4387 *mpimat = B_mpi; 4388 4389 ierr = PetscLogEventEnd(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4390 PetscFunctionReturn(0); 4391 } 4392 4393 /*@C 4394 MatCreateMPIAIJSumSeqAIJ - Creates a MATMPIAIJ matrix by adding sequential 4395 matrices from each processor 4396 4397 Collective on MPI_Comm 4398 4399 Input Parameters: 4400 + comm - the communicators the parallel matrix will live on 4401 . seqmat - the input sequential matrices 4402 . m - number of local rows (or PETSC_DECIDE) 4403 . n - number of local columns (or PETSC_DECIDE) 4404 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4405 4406 Output Parameter: 4407 . mpimat - the parallel matrix generated 4408 4409 Level: advanced 4410 4411 Notes: 4412 The dimensions of the sequential matrix in each processor MUST be the same. 4413 The input seqmat is included into the container "Mat_Merge_SeqsToMPI", and will be 4414 destroyed when mpimat is destroyed. Call PetscObjectQuery() to access seqmat. 4415 @*/ 4416 PetscErrorCode MatCreateMPIAIJSumSeqAIJ(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,MatReuse scall,Mat *mpimat) 4417 { 4418 PetscErrorCode ierr; 4419 PetscMPIInt size; 4420 4421 PetscFunctionBegin; 4422 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4423 if (size == 1) { 4424 ierr = PetscLogEventBegin(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4425 if (scall == MAT_INITIAL_MATRIX) { 4426 ierr = MatDuplicate(seqmat,MAT_COPY_VALUES,mpimat);CHKERRQ(ierr); 4427 } else { 4428 ierr = MatCopy(seqmat,*mpimat,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 4429 } 4430 ierr = PetscLogEventEnd(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4431 PetscFunctionReturn(0); 4432 } 4433 ierr = PetscLogEventBegin(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4434 if (scall == MAT_INITIAL_MATRIX) { 4435 ierr = MatCreateMPIAIJSumSeqAIJSymbolic(comm,seqmat,m,n,mpimat);CHKERRQ(ierr); 4436 } 4437 ierr = MatCreateMPIAIJSumSeqAIJNumeric(seqmat,*mpimat);CHKERRQ(ierr); 4438 ierr = PetscLogEventEnd(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4439 PetscFunctionReturn(0); 4440 } 4441 4442 /*@ 4443 MatMPIAIJGetLocalMat - Creates a SeqAIJ from a MATMPIAIJ matrix by taking all its local rows and putting them into a sequential vector with 4444 mlocal rows and n columns. Where mlocal is the row count obtained with MatGetLocalSize() and n is the global column count obtained 4445 with MatGetSize() 4446 4447 Not Collective 4448 4449 Input Parameters: 4450 + A - the matrix 4451 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4452 4453 Output Parameter: 4454 . A_loc - the local sequential matrix generated 4455 4456 Level: developer 4457 4458 .seealso: MatGetOwnerShipRange(), MatMPIAIJGetLocalMatCondensed() 4459 4460 @*/ 4461 PetscErrorCode MatMPIAIJGetLocalMat(Mat A,MatReuse scall,Mat *A_loc) 4462 { 4463 PetscErrorCode ierr; 4464 Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data; 4465 Mat_SeqAIJ *mat,*a,*b; 4466 PetscInt *ai,*aj,*bi,*bj,*cmap=mpimat->garray; 4467 MatScalar *aa,*ba,*cam; 4468 PetscScalar *ca; 4469 PetscInt am=A->rmap->n,i,j,k,cstart=A->cmap->rstart; 4470 PetscInt *ci,*cj,col,ncols_d,ncols_o,jo; 4471 PetscBool match; 4472 MPI_Comm comm; 4473 PetscMPIInt size; 4474 4475 PetscFunctionBegin; 4476 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&match);CHKERRQ(ierr); 4477 if (!match) SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_SUP,"Requires MATMPIAIJ matrix as input"); 4478 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 4479 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4480 if (size == 1 && scall == MAT_REUSE_MATRIX) PetscFunctionReturn(0); 4481 4482 ierr = PetscLogEventBegin(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4483 a = (Mat_SeqAIJ*)(mpimat->A)->data; 4484 b = (Mat_SeqAIJ*)(mpimat->B)->data; 4485 ai = a->i; aj = a->j; bi = b->i; bj = b->j; 4486 aa = a->a; ba = b->a; 4487 if (scall == MAT_INITIAL_MATRIX) { 4488 if (size == 1) { 4489 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap->N,ai,aj,aa,A_loc);CHKERRQ(ierr); 4490 PetscFunctionReturn(0); 4491 } 4492 4493 ierr = PetscMalloc1(1+am,&ci);CHKERRQ(ierr); 4494 ci[0] = 0; 4495 for (i=0; i<am; i++) { 4496 ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]); 4497 } 4498 ierr = PetscMalloc1(1+ci[am],&cj);CHKERRQ(ierr); 4499 ierr = PetscMalloc1(1+ci[am],&ca);CHKERRQ(ierr); 4500 k = 0; 4501 for (i=0; i<am; i++) { 4502 ncols_o = bi[i+1] - bi[i]; 4503 ncols_d = ai[i+1] - ai[i]; 4504 /* off-diagonal portion of A */ 4505 for (jo=0; jo<ncols_o; jo++) { 4506 col = cmap[*bj]; 4507 if (col >= cstart) break; 4508 cj[k] = col; bj++; 4509 ca[k++] = *ba++; 4510 } 4511 /* diagonal portion of A */ 4512 for (j=0; j<ncols_d; j++) { 4513 cj[k] = cstart + *aj++; 4514 ca[k++] = *aa++; 4515 } 4516 /* off-diagonal portion of A */ 4517 for (j=jo; j<ncols_o; j++) { 4518 cj[k] = cmap[*bj++]; 4519 ca[k++] = *ba++; 4520 } 4521 } 4522 /* put together the new matrix */ 4523 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap->N,ci,cj,ca,A_loc);CHKERRQ(ierr); 4524 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4525 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4526 mat = (Mat_SeqAIJ*)(*A_loc)->data; 4527 mat->free_a = PETSC_TRUE; 4528 mat->free_ij = PETSC_TRUE; 4529 mat->nonew = 0; 4530 } else if (scall == MAT_REUSE_MATRIX) { 4531 mat=(Mat_SeqAIJ*)(*A_loc)->data; 4532 ci = mat->i; cj = mat->j; cam = mat->a; 4533 for (i=0; i<am; i++) { 4534 /* off-diagonal portion of A */ 4535 ncols_o = bi[i+1] - bi[i]; 4536 for (jo=0; jo<ncols_o; jo++) { 4537 col = cmap[*bj]; 4538 if (col >= cstart) break; 4539 *cam++ = *ba++; bj++; 4540 } 4541 /* diagonal portion of A */ 4542 ncols_d = ai[i+1] - ai[i]; 4543 for (j=0; j<ncols_d; j++) *cam++ = *aa++; 4544 /* off-diagonal portion of A */ 4545 for (j=jo; j<ncols_o; j++) { 4546 *cam++ = *ba++; bj++; 4547 } 4548 } 4549 } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 4550 ierr = PetscLogEventEnd(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4551 PetscFunctionReturn(0); 4552 } 4553 4554 /*@C 4555 MatMPIAIJGetLocalMatCondensed - Creates a SeqAIJ matrix from an MATMPIAIJ matrix by taking all its local rows and NON-ZERO columns 4556 4557 Not Collective 4558 4559 Input Parameters: 4560 + A - the matrix 4561 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4562 - row, col - index sets of rows and columns to extract (or NULL) 4563 4564 Output Parameter: 4565 . A_loc - the local sequential matrix generated 4566 4567 Level: developer 4568 4569 .seealso: MatGetOwnershipRange(), MatMPIAIJGetLocalMat() 4570 4571 @*/ 4572 PetscErrorCode MatMPIAIJGetLocalMatCondensed(Mat A,MatReuse scall,IS *row,IS *col,Mat *A_loc) 4573 { 4574 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4575 PetscErrorCode ierr; 4576 PetscInt i,start,end,ncols,nzA,nzB,*cmap,imark,*idx; 4577 IS isrowa,iscola; 4578 Mat *aloc; 4579 PetscBool match; 4580 4581 PetscFunctionBegin; 4582 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&match);CHKERRQ(ierr); 4583 if (!match) SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_SUP,"Requires MATMPIAIJ matrix as input"); 4584 ierr = PetscLogEventBegin(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4585 if (!row) { 4586 start = A->rmap->rstart; end = A->rmap->rend; 4587 ierr = ISCreateStride(PETSC_COMM_SELF,end-start,start,1,&isrowa);CHKERRQ(ierr); 4588 } else { 4589 isrowa = *row; 4590 } 4591 if (!col) { 4592 start = A->cmap->rstart; 4593 cmap = a->garray; 4594 nzA = a->A->cmap->n; 4595 nzB = a->B->cmap->n; 4596 ierr = PetscMalloc1(nzA+nzB, &idx);CHKERRQ(ierr); 4597 ncols = 0; 4598 for (i=0; i<nzB; i++) { 4599 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4600 else break; 4601 } 4602 imark = i; 4603 for (i=0; i<nzA; i++) idx[ncols++] = start + i; 4604 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; 4605 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&iscola);CHKERRQ(ierr); 4606 } else { 4607 iscola = *col; 4608 } 4609 if (scall != MAT_INITIAL_MATRIX) { 4610 ierr = PetscMalloc1(1,&aloc);CHKERRQ(ierr); 4611 aloc[0] = *A_loc; 4612 } 4613 ierr = MatCreateSubMatrices(A,1,&isrowa,&iscola,scall,&aloc);CHKERRQ(ierr); 4614 *A_loc = aloc[0]; 4615 ierr = PetscFree(aloc);CHKERRQ(ierr); 4616 if (!row) { 4617 ierr = ISDestroy(&isrowa);CHKERRQ(ierr); 4618 } 4619 if (!col) { 4620 ierr = ISDestroy(&iscola);CHKERRQ(ierr); 4621 } 4622 ierr = PetscLogEventEnd(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4623 PetscFunctionReturn(0); 4624 } 4625 4626 /*@C 4627 MatGetBrowsOfAcols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns of local A 4628 4629 Collective on Mat 4630 4631 Input Parameters: 4632 + A,B - the matrices in mpiaij format 4633 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4634 - rowb, colb - index sets of rows and columns of B to extract (or NULL) 4635 4636 Output Parameter: 4637 + rowb, colb - index sets of rows and columns of B to extract 4638 - B_seq - the sequential matrix generated 4639 4640 Level: developer 4641 4642 @*/ 4643 PetscErrorCode MatGetBrowsOfAcols(Mat A,Mat B,MatReuse scall,IS *rowb,IS *colb,Mat *B_seq) 4644 { 4645 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4646 PetscErrorCode ierr; 4647 PetscInt *idx,i,start,ncols,nzA,nzB,*cmap,imark; 4648 IS isrowb,iscolb; 4649 Mat *bseq=NULL; 4650 4651 PetscFunctionBegin; 4652 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) { 4653 SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, (%D, %D) != (%D,%D)",A->cmap->rstart,A->cmap->rend,B->rmap->rstart,B->rmap->rend); 4654 } 4655 ierr = PetscLogEventBegin(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4656 4657 if (scall == MAT_INITIAL_MATRIX) { 4658 start = A->cmap->rstart; 4659 cmap = a->garray; 4660 nzA = a->A->cmap->n; 4661 nzB = a->B->cmap->n; 4662 ierr = PetscMalloc1(nzA+nzB, &idx);CHKERRQ(ierr); 4663 ncols = 0; 4664 for (i=0; i<nzB; i++) { /* row < local row index */ 4665 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4666 else break; 4667 } 4668 imark = i; 4669 for (i=0; i<nzA; i++) idx[ncols++] = start + i; /* local rows */ 4670 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; /* row > local row index */ 4671 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&isrowb);CHKERRQ(ierr); 4672 ierr = ISCreateStride(PETSC_COMM_SELF,B->cmap->N,0,1,&iscolb);CHKERRQ(ierr); 4673 } else { 4674 if (!rowb || !colb) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"IS rowb and colb must be provided for MAT_REUSE_MATRIX"); 4675 isrowb = *rowb; iscolb = *colb; 4676 ierr = PetscMalloc1(1,&bseq);CHKERRQ(ierr); 4677 bseq[0] = *B_seq; 4678 } 4679 ierr = MatCreateSubMatrices(B,1,&isrowb,&iscolb,scall,&bseq);CHKERRQ(ierr); 4680 *B_seq = bseq[0]; 4681 ierr = PetscFree(bseq);CHKERRQ(ierr); 4682 if (!rowb) { 4683 ierr = ISDestroy(&isrowb);CHKERRQ(ierr); 4684 } else { 4685 *rowb = isrowb; 4686 } 4687 if (!colb) { 4688 ierr = ISDestroy(&iscolb);CHKERRQ(ierr); 4689 } else { 4690 *colb = iscolb; 4691 } 4692 ierr = PetscLogEventEnd(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4693 PetscFunctionReturn(0); 4694 } 4695 4696 /* 4697 MatGetBrowsOfAoCols_MPIAIJ - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns 4698 of the OFF-DIAGONAL portion of local A 4699 4700 Collective on Mat 4701 4702 Input Parameters: 4703 + A,B - the matrices in mpiaij format 4704 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4705 4706 Output Parameter: 4707 + startsj_s - starting point in B's sending j-arrays, saved for MAT_REUSE (or NULL) 4708 . startsj_r - starting point in B's receiving j-arrays, saved for MAT_REUSE (or NULL) 4709 . bufa_ptr - array for sending matrix values, saved for MAT_REUSE (or NULL) 4710 - B_oth - the sequential matrix generated with size aBn=a->B->cmap->n by B->cmap->N 4711 4712 Level: developer 4713 4714 */ 4715 PetscErrorCode MatGetBrowsOfAoCols_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscInt **startsj_s,PetscInt **startsj_r,MatScalar **bufa_ptr,Mat *B_oth) 4716 { 4717 VecScatter_MPI_General *gen_to,*gen_from; 4718 PetscErrorCode ierr; 4719 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4720 Mat_SeqAIJ *b_oth; 4721 VecScatter ctx =a->Mvctx; 4722 MPI_Comm comm; 4723 PetscMPIInt *rprocs,*sprocs,tag=((PetscObject)ctx)->tag,rank; 4724 PetscInt *rowlen,*bufj,*bufJ,ncols,aBn=a->B->cmap->n,row,*b_othi,*b_othj; 4725 PetscInt *rvalues,*svalues; 4726 MatScalar *b_otha,*bufa,*bufA; 4727 PetscInt i,j,k,l,ll,nrecvs,nsends,nrows,*srow,*rstarts,*rstartsj = 0,*sstarts,*sstartsj,len; 4728 MPI_Request *rwaits = NULL,*swaits = NULL; 4729 MPI_Status *sstatus,rstatus; 4730 PetscMPIInt jj,size; 4731 PetscInt *cols,sbs,rbs; 4732 PetscScalar *vals; 4733 4734 PetscFunctionBegin; 4735 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 4736 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4737 4738 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) { 4739 SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, (%d, %d) != (%d,%d)",A->cmap->rstart,A->cmap->rend,B->rmap->rstart,B->rmap->rend); 4740 } 4741 ierr = PetscLogEventBegin(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 4742 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4743 4744 if (size == 1) { 4745 startsj_s = NULL; 4746 bufa_ptr = NULL; 4747 *B_oth = NULL; 4748 PetscFunctionReturn(0); 4749 } 4750 4751 gen_to = (VecScatter_MPI_General*)ctx->todata; 4752 gen_from = (VecScatter_MPI_General*)ctx->fromdata; 4753 nrecvs = gen_from->n; 4754 nsends = gen_to->n; 4755 4756 ierr = PetscMalloc2(nrecvs,&rwaits,nsends,&swaits);CHKERRQ(ierr); 4757 srow = gen_to->indices; /* local row index to be sent */ 4758 sstarts = gen_to->starts; 4759 sprocs = gen_to->procs; 4760 sstatus = gen_to->sstatus; 4761 sbs = gen_to->bs; 4762 rstarts = gen_from->starts; 4763 rprocs = gen_from->procs; 4764 rbs = gen_from->bs; 4765 4766 if (!startsj_s || !bufa_ptr) scall = MAT_INITIAL_MATRIX; 4767 if (scall == MAT_INITIAL_MATRIX) { 4768 /* i-array */ 4769 /*---------*/ 4770 /* post receives */ 4771 ierr = PetscMalloc1(rbs*(rstarts[nrecvs] - rstarts[0]),&rvalues);CHKERRQ(ierr); 4772 for (i=0; i<nrecvs; i++) { 4773 rowlen = rvalues + rstarts[i]*rbs; 4774 nrows = (rstarts[i+1]-rstarts[i])*rbs; /* num of indices to be received */ 4775 ierr = MPI_Irecv(rowlen,nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4776 } 4777 4778 /* pack the outgoing message */ 4779 ierr = PetscMalloc2(nsends+1,&sstartsj,nrecvs+1,&rstartsj);CHKERRQ(ierr); 4780 4781 sstartsj[0] = 0; 4782 rstartsj[0] = 0; 4783 len = 0; /* total length of j or a array to be sent */ 4784 k = 0; 4785 ierr = PetscMalloc1(sbs*(sstarts[nsends] - sstarts[0]),&svalues);CHKERRQ(ierr); 4786 for (i=0; i<nsends; i++) { 4787 rowlen = svalues + sstarts[i]*sbs; 4788 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4789 for (j=0; j<nrows; j++) { 4790 row = srow[k] + B->rmap->range[rank]; /* global row idx */ 4791 for (l=0; l<sbs; l++) { 4792 ierr = MatGetRow_MPIAIJ(B,row+l,&ncols,NULL,NULL);CHKERRQ(ierr); /* rowlength */ 4793 4794 rowlen[j*sbs+l] = ncols; 4795 4796 len += ncols; 4797 ierr = MatRestoreRow_MPIAIJ(B,row+l,&ncols,NULL,NULL);CHKERRQ(ierr); 4798 } 4799 k++; 4800 } 4801 ierr = MPI_Isend(rowlen,nrows*sbs,MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4802 4803 sstartsj[i+1] = len; /* starting point of (i+1)-th outgoing msg in bufj and bufa */ 4804 } 4805 /* recvs and sends of i-array are completed */ 4806 i = nrecvs; 4807 while (i--) { 4808 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4809 } 4810 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4811 ierr = PetscFree(svalues);CHKERRQ(ierr); 4812 4813 /* allocate buffers for sending j and a arrays */ 4814 ierr = PetscMalloc1(len+1,&bufj);CHKERRQ(ierr); 4815 ierr = PetscMalloc1(len+1,&bufa);CHKERRQ(ierr); 4816 4817 /* create i-array of B_oth */ 4818 ierr = PetscMalloc1(aBn+2,&b_othi);CHKERRQ(ierr); 4819 4820 b_othi[0] = 0; 4821 len = 0; /* total length of j or a array to be received */ 4822 k = 0; 4823 for (i=0; i<nrecvs; i++) { 4824 rowlen = rvalues + rstarts[i]*rbs; 4825 nrows = rbs*(rstarts[i+1]-rstarts[i]); /* num of rows to be received */ 4826 for (j=0; j<nrows; j++) { 4827 b_othi[k+1] = b_othi[k] + rowlen[j]; 4828 ierr = PetscIntSumError(rowlen[j],len,&len);CHKERRQ(ierr); 4829 k++; 4830 } 4831 rstartsj[i+1] = len; /* starting point of (i+1)-th incoming msg in bufj and bufa */ 4832 } 4833 ierr = PetscFree(rvalues);CHKERRQ(ierr); 4834 4835 /* allocate space for j and a arrrays of B_oth */ 4836 ierr = PetscMalloc1(b_othi[aBn]+1,&b_othj);CHKERRQ(ierr); 4837 ierr = PetscMalloc1(b_othi[aBn]+1,&b_otha);CHKERRQ(ierr); 4838 4839 /* j-array */ 4840 /*---------*/ 4841 /* post receives of j-array */ 4842 for (i=0; i<nrecvs; i++) { 4843 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 4844 ierr = MPI_Irecv(b_othj+rstartsj[i],nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4845 } 4846 4847 /* pack the outgoing message j-array */ 4848 k = 0; 4849 for (i=0; i<nsends; i++) { 4850 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4851 bufJ = bufj+sstartsj[i]; 4852 for (j=0; j<nrows; j++) { 4853 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 4854 for (ll=0; ll<sbs; ll++) { 4855 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,&cols,NULL);CHKERRQ(ierr); 4856 for (l=0; l<ncols; l++) { 4857 *bufJ++ = cols[l]; 4858 } 4859 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,&cols,NULL);CHKERRQ(ierr); 4860 } 4861 } 4862 ierr = MPI_Isend(bufj+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4863 } 4864 4865 /* recvs and sends of j-array are completed */ 4866 i = nrecvs; 4867 while (i--) { 4868 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4869 } 4870 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4871 } else if (scall == MAT_REUSE_MATRIX) { 4872 sstartsj = *startsj_s; 4873 rstartsj = *startsj_r; 4874 bufa = *bufa_ptr; 4875 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 4876 b_otha = b_oth->a; 4877 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE, "Matrix P does not posses an object container"); 4878 4879 /* a-array */ 4880 /*---------*/ 4881 /* post receives of a-array */ 4882 for (i=0; i<nrecvs; i++) { 4883 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 4884 ierr = MPI_Irecv(b_otha+rstartsj[i],nrows,MPIU_SCALAR,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4885 } 4886 4887 /* pack the outgoing message a-array */ 4888 k = 0; 4889 for (i=0; i<nsends; i++) { 4890 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4891 bufA = bufa+sstartsj[i]; 4892 for (j=0; j<nrows; j++) { 4893 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 4894 for (ll=0; ll<sbs; ll++) { 4895 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,NULL,&vals);CHKERRQ(ierr); 4896 for (l=0; l<ncols; l++) { 4897 *bufA++ = vals[l]; 4898 } 4899 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,NULL,&vals);CHKERRQ(ierr); 4900 } 4901 } 4902 ierr = MPI_Isend(bufa+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_SCALAR,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4903 } 4904 /* recvs and sends of a-array are completed */ 4905 i = nrecvs; 4906 while (i--) { 4907 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4908 } 4909 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4910 ierr = PetscFree2(rwaits,swaits);CHKERRQ(ierr); 4911 4912 if (scall == MAT_INITIAL_MATRIX) { 4913 /* put together the new matrix */ 4914 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,aBn,B->cmap->N,b_othi,b_othj,b_otha,B_oth);CHKERRQ(ierr); 4915 4916 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4917 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4918 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 4919 b_oth->free_a = PETSC_TRUE; 4920 b_oth->free_ij = PETSC_TRUE; 4921 b_oth->nonew = 0; 4922 4923 ierr = PetscFree(bufj);CHKERRQ(ierr); 4924 if (!startsj_s || !bufa_ptr) { 4925 ierr = PetscFree2(sstartsj,rstartsj);CHKERRQ(ierr); 4926 ierr = PetscFree(bufa_ptr);CHKERRQ(ierr); 4927 } else { 4928 *startsj_s = sstartsj; 4929 *startsj_r = rstartsj; 4930 *bufa_ptr = bufa; 4931 } 4932 } 4933 ierr = PetscLogEventEnd(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 4934 PetscFunctionReturn(0); 4935 } 4936 4937 /*@C 4938 MatGetCommunicationStructs - Provides access to the communication structures used in matrix-vector multiplication. 4939 4940 Not Collective 4941 4942 Input Parameters: 4943 . A - The matrix in mpiaij format 4944 4945 Output Parameter: 4946 + lvec - The local vector holding off-process values from the argument to a matrix-vector product 4947 . colmap - A map from global column index to local index into lvec 4948 - multScatter - A scatter from the argument of a matrix-vector product to lvec 4949 4950 Level: developer 4951 4952 @*/ 4953 #if defined(PETSC_USE_CTABLE) 4954 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscTable *colmap, VecScatter *multScatter) 4955 #else 4956 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscInt *colmap[], VecScatter *multScatter) 4957 #endif 4958 { 4959 Mat_MPIAIJ *a; 4960 4961 PetscFunctionBegin; 4962 PetscValidHeaderSpecific(A, MAT_CLASSID, 1); 4963 PetscValidPointer(lvec, 2); 4964 PetscValidPointer(colmap, 3); 4965 PetscValidPointer(multScatter, 4); 4966 a = (Mat_MPIAIJ*) A->data; 4967 if (lvec) *lvec = a->lvec; 4968 if (colmap) *colmap = a->colmap; 4969 if (multScatter) *multScatter = a->Mvctx; 4970 PetscFunctionReturn(0); 4971 } 4972 4973 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJCRL(Mat,MatType,MatReuse,Mat*); 4974 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJPERM(Mat,MatType,MatReuse,Mat*); 4975 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPISBAIJ(Mat,MatType,MatReuse,Mat*); 4976 #if defined(PETSC_HAVE_ELEMENTAL) 4977 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_Elemental(Mat,MatType,MatReuse,Mat*); 4978 #endif 4979 #if defined(PETSC_HAVE_HYPRE) 4980 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*); 4981 PETSC_INTERN PetscErrorCode MatMatMatMult_Transpose_AIJ_AIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*); 4982 #endif 4983 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_IS(Mat,MatType,MatReuse,Mat*); 4984 4985 /* 4986 Computes (B'*A')' since computing B*A directly is untenable 4987 4988 n p p 4989 ( ) ( ) ( ) 4990 m ( A ) * n ( B ) = m ( C ) 4991 ( ) ( ) ( ) 4992 4993 */ 4994 PetscErrorCode MatMatMultNumeric_MPIDense_MPIAIJ(Mat A,Mat B,Mat C) 4995 { 4996 PetscErrorCode ierr; 4997 Mat At,Bt,Ct; 4998 4999 PetscFunctionBegin; 5000 ierr = MatTranspose(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 5001 ierr = MatTranspose(B,MAT_INITIAL_MATRIX,&Bt);CHKERRQ(ierr); 5002 ierr = MatMatMult(Bt,At,MAT_INITIAL_MATRIX,1.0,&Ct);CHKERRQ(ierr); 5003 ierr = MatDestroy(&At);CHKERRQ(ierr); 5004 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 5005 ierr = MatTranspose(Ct,MAT_REUSE_MATRIX,&C);CHKERRQ(ierr); 5006 ierr = MatDestroy(&Ct);CHKERRQ(ierr); 5007 PetscFunctionReturn(0); 5008 } 5009 5010 PetscErrorCode MatMatMultSymbolic_MPIDense_MPIAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 5011 { 5012 PetscErrorCode ierr; 5013 PetscInt m=A->rmap->n,n=B->cmap->n; 5014 Mat Cmat; 5015 5016 PetscFunctionBegin; 5017 if (A->cmap->n != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->cmap->n %d != B->rmap->n %d\n",A->cmap->n,B->rmap->n); 5018 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 5019 ierr = MatSetSizes(Cmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 5020 ierr = MatSetBlockSizesFromMats(Cmat,A,B);CHKERRQ(ierr); 5021 ierr = MatSetType(Cmat,MATMPIDENSE);CHKERRQ(ierr); 5022 ierr = MatMPIDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 5023 ierr = MatAssemblyBegin(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5024 ierr = MatAssemblyEnd(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5025 5026 Cmat->ops->matmultnumeric = MatMatMultNumeric_MPIDense_MPIAIJ; 5027 5028 *C = Cmat; 5029 PetscFunctionReturn(0); 5030 } 5031 5032 /* ----------------------------------------------------------------*/ 5033 PETSC_INTERN PetscErrorCode MatMatMult_MPIDense_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 5034 { 5035 PetscErrorCode ierr; 5036 5037 PetscFunctionBegin; 5038 if (scall == MAT_INITIAL_MATRIX) { 5039 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 5040 ierr = MatMatMultSymbolic_MPIDense_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 5041 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 5042 } 5043 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 5044 ierr = MatMatMultNumeric_MPIDense_MPIAIJ(A,B,*C);CHKERRQ(ierr); 5045 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 5046 PetscFunctionReturn(0); 5047 } 5048 5049 /*MC 5050 MATMPIAIJ - MATMPIAIJ = "mpiaij" - A matrix type to be used for parallel sparse matrices. 5051 5052 Options Database Keys: 5053 . -mat_type mpiaij - sets the matrix type to "mpiaij" during a call to MatSetFromOptions() 5054 5055 Level: beginner 5056 5057 .seealso: MatCreateAIJ() 5058 M*/ 5059 5060 PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJ(Mat B) 5061 { 5062 Mat_MPIAIJ *b; 5063 PetscErrorCode ierr; 5064 PetscMPIInt size; 5065 5066 PetscFunctionBegin; 5067 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 5068 5069 ierr = PetscNewLog(B,&b);CHKERRQ(ierr); 5070 B->data = (void*)b; 5071 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 5072 B->assembled = PETSC_FALSE; 5073 B->insertmode = NOT_SET_VALUES; 5074 b->size = size; 5075 5076 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)B),&b->rank);CHKERRQ(ierr); 5077 5078 /* build cache for off array entries formed */ 5079 ierr = MatStashCreate_Private(PetscObjectComm((PetscObject)B),1,&B->stash);CHKERRQ(ierr); 5080 5081 b->donotstash = PETSC_FALSE; 5082 b->colmap = 0; 5083 b->garray = 0; 5084 b->roworiented = PETSC_TRUE; 5085 5086 /* stuff used for matrix vector multiply */ 5087 b->lvec = NULL; 5088 b->Mvctx = NULL; 5089 5090 /* stuff for MatGetRow() */ 5091 b->rowindices = 0; 5092 b->rowvalues = 0; 5093 b->getrowactive = PETSC_FALSE; 5094 5095 /* flexible pointer used in CUSP/CUSPARSE classes */ 5096 b->spptr = NULL; 5097 5098 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetUseScalableIncreaseOverlap_C",MatMPIAIJSetUseScalableIncreaseOverlap_MPIAIJ);CHKERRQ(ierr); 5099 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_MPIAIJ);CHKERRQ(ierr); 5100 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_MPIAIJ);CHKERRQ(ierr); 5101 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_MPIAIJ);CHKERRQ(ierr); 5102 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetPreallocation_C",MatMPIAIJSetPreallocation_MPIAIJ);CHKERRQ(ierr); 5103 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetPreallocationCSR_C",MatMPIAIJSetPreallocationCSR_MPIAIJ);CHKERRQ(ierr); 5104 ierr = PetscObjectComposeFunction((PetscObject)B,"MatDiagonalScaleLocal_C",MatDiagonalScaleLocal_MPIAIJ);CHKERRQ(ierr); 5105 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpiaijperm_C",MatConvert_MPIAIJ_MPIAIJPERM);CHKERRQ(ierr); 5106 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpiaijcrl_C",MatConvert_MPIAIJ_MPIAIJCRL);CHKERRQ(ierr); 5107 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpisbaij_C",MatConvert_MPIAIJ_MPISBAIJ);CHKERRQ(ierr); 5108 #if defined(PETSC_HAVE_ELEMENTAL) 5109 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_elemental_C",MatConvert_MPIAIJ_Elemental);CHKERRQ(ierr); 5110 #endif 5111 #if defined(PETSC_HAVE_HYPRE) 5112 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_hypre_C",MatConvert_AIJ_HYPRE);CHKERRQ(ierr); 5113 #endif 5114 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_is_C",MatConvert_MPIAIJ_IS);CHKERRQ(ierr); 5115 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_mpidense_mpiaij_C",MatMatMult_MPIDense_MPIAIJ);CHKERRQ(ierr); 5116 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_mpidense_mpiaij_C",MatMatMultSymbolic_MPIDense_MPIAIJ);CHKERRQ(ierr); 5117 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_mpidense_mpiaij_C",MatMatMultNumeric_MPIDense_MPIAIJ);CHKERRQ(ierr); 5118 #if defined(PETSC_HAVE_HYPRE) 5119 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMatMult_transpose_mpiaij_mpiaij_C",MatMatMatMult_Transpose_AIJ_AIJ);CHKERRQ(ierr); 5120 #endif 5121 ierr = PetscObjectChangeTypeName((PetscObject)B,MATMPIAIJ);CHKERRQ(ierr); 5122 PetscFunctionReturn(0); 5123 } 5124 5125 /*@C 5126 MatCreateMPIAIJWithSplitArrays - creates a MPI AIJ matrix using arrays that contain the "diagonal" 5127 and "off-diagonal" part of the matrix in CSR format. 5128 5129 Collective on MPI_Comm 5130 5131 Input Parameters: 5132 + comm - MPI communicator 5133 . m - number of local rows (Cannot be PETSC_DECIDE) 5134 . n - This value should be the same as the local size used in creating the 5135 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 5136 calculated if N is given) For square matrices n is almost always m. 5137 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 5138 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 5139 . i - row indices for "diagonal" portion of matrix 5140 . j - column indices 5141 . a - matrix values 5142 . oi - row indices for "off-diagonal" portion of matrix 5143 . oj - column indices 5144 - oa - matrix values 5145 5146 Output Parameter: 5147 . mat - the matrix 5148 5149 Level: advanced 5150 5151 Notes: 5152 The i, j, and a arrays ARE NOT copied by this routine into the internal format used by PETSc. The user 5153 must free the arrays once the matrix has been destroyed and not before. 5154 5155 The i and j indices are 0 based 5156 5157 See MatCreateAIJ() for the definition of "diagonal" and "off-diagonal" portion of the matrix 5158 5159 This sets local rows and cannot be used to set off-processor values. 5160 5161 Use of this routine is discouraged because it is inflexible and cumbersome to use. It is extremely rare that a 5162 legacy application natively assembles into exactly this split format. The code to do so is nontrivial and does 5163 not easily support in-place reassembly. It is recommended to use MatSetValues() (or a variant thereof) because 5164 the resulting assembly is easier to implement, will work with any matrix format, and the user does not have to 5165 keep track of the underlying array. Use MatSetOption(A,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) to disable all 5166 communication if it is known that only local entries will be set. 5167 5168 .keywords: matrix, aij, compressed row, sparse, parallel 5169 5170 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 5171 MATMPIAIJ, MatCreateAIJ(), MatCreateMPIAIJWithArrays() 5172 @*/ 5173 PetscErrorCode MatCreateMPIAIJWithSplitArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt i[],PetscInt j[],PetscScalar a[],PetscInt oi[], PetscInt oj[],PetscScalar oa[],Mat *mat) 5174 { 5175 PetscErrorCode ierr; 5176 Mat_MPIAIJ *maij; 5177 5178 PetscFunctionBegin; 5179 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 5180 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 5181 if (oi[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"oi (row indices) must start with 0"); 5182 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 5183 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 5184 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 5185 maij = (Mat_MPIAIJ*) (*mat)->data; 5186 5187 (*mat)->preallocated = PETSC_TRUE; 5188 5189 ierr = PetscLayoutSetUp((*mat)->rmap);CHKERRQ(ierr); 5190 ierr = PetscLayoutSetUp((*mat)->cmap);CHKERRQ(ierr); 5191 5192 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,i,j,a,&maij->A);CHKERRQ(ierr); 5193 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,(*mat)->cmap->N,oi,oj,oa,&maij->B);CHKERRQ(ierr); 5194 5195 ierr = MatAssemblyBegin(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5196 ierr = MatAssemblyEnd(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5197 ierr = MatAssemblyBegin(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5198 ierr = MatAssemblyEnd(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5199 5200 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE);CHKERRQ(ierr); 5201 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5202 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5203 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_FALSE);CHKERRQ(ierr); 5204 ierr = MatSetOption(*mat,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 5205 PetscFunctionReturn(0); 5206 } 5207 5208 /* 5209 Special version for direct calls from Fortran 5210 */ 5211 #include <petsc/private/fortranimpl.h> 5212 5213 /* Change these macros so can be used in void function */ 5214 #undef CHKERRQ 5215 #define CHKERRQ(ierr) CHKERRABORT(PETSC_COMM_WORLD,ierr) 5216 #undef SETERRQ2 5217 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 5218 #undef SETERRQ3 5219 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 5220 #undef SETERRQ 5221 #define SETERRQ(c,ierr,b) CHKERRABORT(c,ierr) 5222 5223 #if defined(PETSC_HAVE_FORTRAN_CAPS) 5224 #define matsetvaluesmpiaij_ MATSETVALUESMPIAIJ 5225 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 5226 #define matsetvaluesmpiaij_ matsetvaluesmpiaij 5227 #else 5228 #endif 5229 PETSC_EXTERN void PETSC_STDCALL matsetvaluesmpiaij_(Mat *mmat,PetscInt *mm,const PetscInt im[],PetscInt *mn,const PetscInt in[],const PetscScalar v[],InsertMode *maddv,PetscErrorCode *_ierr) 5230 { 5231 Mat mat = *mmat; 5232 PetscInt m = *mm, n = *mn; 5233 InsertMode addv = *maddv; 5234 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 5235 PetscScalar value; 5236 PetscErrorCode ierr; 5237 5238 MatCheckPreallocated(mat,1); 5239 if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv; 5240 5241 #if defined(PETSC_USE_DEBUG) 5242 else if (mat->insertmode != addv) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot mix add values and insert values"); 5243 #endif 5244 { 5245 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 5246 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 5247 PetscBool roworiented = aij->roworiented; 5248 5249 /* Some Variables required in the macro */ 5250 Mat A = aij->A; 5251 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 5252 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 5253 MatScalar *aa = a->a; 5254 PetscBool ignorezeroentries = (((a->ignorezeroentries)&&(addv==ADD_VALUES)) ? PETSC_TRUE : PETSC_FALSE); 5255 Mat B = aij->B; 5256 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 5257 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 5258 MatScalar *ba = b->a; 5259 5260 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 5261 PetscInt nonew = a->nonew; 5262 MatScalar *ap1,*ap2; 5263 5264 PetscFunctionBegin; 5265 for (i=0; i<m; i++) { 5266 if (im[i] < 0) continue; 5267 #if defined(PETSC_USE_DEBUG) 5268 if (im[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap->N-1); 5269 #endif 5270 if (im[i] >= rstart && im[i] < rend) { 5271 row = im[i] - rstart; 5272 lastcol1 = -1; 5273 rp1 = aj + ai[row]; 5274 ap1 = aa + ai[row]; 5275 rmax1 = aimax[row]; 5276 nrow1 = ailen[row]; 5277 low1 = 0; 5278 high1 = nrow1; 5279 lastcol2 = -1; 5280 rp2 = bj + bi[row]; 5281 ap2 = ba + bi[row]; 5282 rmax2 = bimax[row]; 5283 nrow2 = bilen[row]; 5284 low2 = 0; 5285 high2 = nrow2; 5286 5287 for (j=0; j<n; j++) { 5288 if (roworiented) value = v[i*n+j]; 5289 else value = v[i+j*m]; 5290 if (in[j] >= cstart && in[j] < cend) { 5291 col = in[j] - cstart; 5292 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 5293 MatSetValues_SeqAIJ_A_Private(row,col,value,addv,im[i],in[j]); 5294 } else if (in[j] < 0) continue; 5295 #if defined(PETSC_USE_DEBUG) 5296 else if (in[j] >= mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap->N-1); 5297 #endif 5298 else { 5299 if (mat->was_assembled) { 5300 if (!aij->colmap) { 5301 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 5302 } 5303 #if defined(PETSC_USE_CTABLE) 5304 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 5305 col--; 5306 #else 5307 col = aij->colmap[in[j]] - 1; 5308 #endif 5309 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 5310 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 5311 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 5312 col = in[j]; 5313 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 5314 B = aij->B; 5315 b = (Mat_SeqAIJ*)B->data; 5316 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; 5317 rp2 = bj + bi[row]; 5318 ap2 = ba + bi[row]; 5319 rmax2 = bimax[row]; 5320 nrow2 = bilen[row]; 5321 low2 = 0; 5322 high2 = nrow2; 5323 bm = aij->B->rmap->n; 5324 ba = b->a; 5325 } 5326 } else col = in[j]; 5327 MatSetValues_SeqAIJ_B_Private(row,col,value,addv,im[i],in[j]); 5328 } 5329 } 5330 } else if (!aij->donotstash) { 5331 if (roworiented) { 5332 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5333 } else { 5334 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5335 } 5336 } 5337 } 5338 } 5339 PetscFunctionReturnVoid(); 5340 } 5341 5342