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 PetscErrorCode MatFindZeroDiagonals_MPIAIJ(Mat M,IS *zrows) 135 { 136 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)M->data; 137 PetscErrorCode ierr; 138 PetscInt i,rstart,nrows,*rows; 139 140 PetscFunctionBegin; 141 *zrows = NULL; 142 ierr = MatFindZeroDiagonals_SeqAIJ_Private(aij->A,&nrows,&rows);CHKERRQ(ierr); 143 ierr = MatGetOwnershipRange(M,&rstart,NULL);CHKERRQ(ierr); 144 for (i=0; i<nrows; i++) rows[i] += rstart; 145 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)M),nrows,rows,PETSC_OWN_POINTER,zrows);CHKERRQ(ierr); 146 PetscFunctionReturn(0); 147 } 148 149 PetscErrorCode MatGetColumnNorms_MPIAIJ(Mat A,NormType type,PetscReal *norms) 150 { 151 PetscErrorCode ierr; 152 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)A->data; 153 PetscInt i,n,*garray = aij->garray; 154 Mat_SeqAIJ *a_aij = (Mat_SeqAIJ*) aij->A->data; 155 Mat_SeqAIJ *b_aij = (Mat_SeqAIJ*) aij->B->data; 156 PetscReal *work; 157 158 PetscFunctionBegin; 159 ierr = MatGetSize(A,NULL,&n);CHKERRQ(ierr); 160 ierr = PetscCalloc1(n,&work);CHKERRQ(ierr); 161 if (type == NORM_2) { 162 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 163 work[A->cmap->rstart + a_aij->j[i]] += PetscAbsScalar(a_aij->a[i]*a_aij->a[i]); 164 } 165 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 166 work[garray[b_aij->j[i]]] += PetscAbsScalar(b_aij->a[i]*b_aij->a[i]); 167 } 168 } else if (type == NORM_1) { 169 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 170 work[A->cmap->rstart + a_aij->j[i]] += PetscAbsScalar(a_aij->a[i]); 171 } 172 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 173 work[garray[b_aij->j[i]]] += PetscAbsScalar(b_aij->a[i]); 174 } 175 } else if (type == NORM_INFINITY) { 176 for (i=0; i<a_aij->i[aij->A->rmap->n]; i++) { 177 work[A->cmap->rstart + a_aij->j[i]] = PetscMax(PetscAbsScalar(a_aij->a[i]), work[A->cmap->rstart + a_aij->j[i]]); 178 } 179 for (i=0; i<b_aij->i[aij->B->rmap->n]; i++) { 180 work[garray[b_aij->j[i]]] = PetscMax(PetscAbsScalar(b_aij->a[i]),work[garray[b_aij->j[i]]]); 181 } 182 183 } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Unknown NormType"); 184 if (type == NORM_INFINITY) { 185 ierr = MPIU_Allreduce(work,norms,n,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 186 } else { 187 ierr = MPIU_Allreduce(work,norms,n,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 188 } 189 ierr = PetscFree(work);CHKERRQ(ierr); 190 if (type == NORM_2) { 191 for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]); 192 } 193 PetscFunctionReturn(0); 194 } 195 196 PetscErrorCode MatFindOffBlockDiagonalEntries_MPIAIJ(Mat A,IS *is) 197 { 198 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 199 IS sis,gis; 200 PetscErrorCode ierr; 201 const PetscInt *isis,*igis; 202 PetscInt n,*iis,nsis,ngis,rstart,i; 203 204 PetscFunctionBegin; 205 ierr = MatFindOffBlockDiagonalEntries(a->A,&sis);CHKERRQ(ierr); 206 ierr = MatFindNonzeroRows(a->B,&gis);CHKERRQ(ierr); 207 ierr = ISGetSize(gis,&ngis);CHKERRQ(ierr); 208 ierr = ISGetSize(sis,&nsis);CHKERRQ(ierr); 209 ierr = ISGetIndices(sis,&isis);CHKERRQ(ierr); 210 ierr = ISGetIndices(gis,&igis);CHKERRQ(ierr); 211 212 ierr = PetscMalloc1(ngis+nsis,&iis);CHKERRQ(ierr); 213 ierr = PetscMemcpy(iis,igis,ngis*sizeof(PetscInt));CHKERRQ(ierr); 214 ierr = PetscMemcpy(iis+ngis,isis,nsis*sizeof(PetscInt));CHKERRQ(ierr); 215 n = ngis + nsis; 216 ierr = PetscSortRemoveDupsInt(&n,iis);CHKERRQ(ierr); 217 ierr = MatGetOwnershipRange(A,&rstart,NULL);CHKERRQ(ierr); 218 for (i=0; i<n; i++) iis[i] += rstart; 219 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)A),n,iis,PETSC_OWN_POINTER,is);CHKERRQ(ierr); 220 221 ierr = ISRestoreIndices(sis,&isis);CHKERRQ(ierr); 222 ierr = ISRestoreIndices(gis,&igis);CHKERRQ(ierr); 223 ierr = ISDestroy(&sis);CHKERRQ(ierr); 224 ierr = ISDestroy(&gis);CHKERRQ(ierr); 225 PetscFunctionReturn(0); 226 } 227 228 /* 229 Distributes a SeqAIJ matrix across a set of processes. Code stolen from 230 MatLoad_MPIAIJ(). Horrible lack of reuse. Should be a routine for each matrix type. 231 232 Only for square matrices 233 234 Used by a preconditioner, hence PETSC_EXTERN 235 */ 236 PETSC_EXTERN PetscErrorCode MatDistribute_MPIAIJ(MPI_Comm comm,Mat gmat,PetscInt m,MatReuse reuse,Mat *inmat) 237 { 238 PetscMPIInt rank,size; 239 PetscInt *rowners,*dlens,*olens,i,rstart,rend,j,jj,nz = 0,*gmataj,cnt,row,*ld,bses[2]; 240 PetscErrorCode ierr; 241 Mat mat; 242 Mat_SeqAIJ *gmata; 243 PetscMPIInt tag; 244 MPI_Status status; 245 PetscBool aij; 246 MatScalar *gmataa,*ao,*ad,*gmataarestore=0; 247 248 PetscFunctionBegin; 249 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 250 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 251 if (!rank) { 252 ierr = PetscObjectTypeCompare((PetscObject)gmat,MATSEQAIJ,&aij);CHKERRQ(ierr); 253 if (!aij) SETERRQ1(PetscObjectComm((PetscObject)gmat),PETSC_ERR_SUP,"Currently no support for input matrix of type %s\n",((PetscObject)gmat)->type_name); 254 } 255 if (reuse == MAT_INITIAL_MATRIX) { 256 ierr = MatCreate(comm,&mat);CHKERRQ(ierr); 257 ierr = MatSetSizes(mat,m,m,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 258 ierr = MatGetBlockSizes(gmat,&bses[0],&bses[1]);CHKERRQ(ierr); 259 ierr = MPI_Bcast(bses,2,MPIU_INT,0,comm);CHKERRQ(ierr); 260 ierr = MatSetBlockSizes(mat,bses[0],bses[1]);CHKERRQ(ierr); 261 ierr = MatSetType(mat,MATAIJ);CHKERRQ(ierr); 262 ierr = PetscMalloc1(size+1,&rowners);CHKERRQ(ierr); 263 ierr = PetscMalloc2(m,&dlens,m,&olens);CHKERRQ(ierr); 264 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 265 266 rowners[0] = 0; 267 for (i=2; i<=size; i++) rowners[i] += rowners[i-1]; 268 rstart = rowners[rank]; 269 rend = rowners[rank+1]; 270 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 271 if (!rank) { 272 gmata = (Mat_SeqAIJ*) gmat->data; 273 /* send row lengths to all processors */ 274 for (i=0; i<m; i++) dlens[i] = gmata->ilen[i]; 275 for (i=1; i<size; i++) { 276 ierr = MPI_Send(gmata->ilen + rowners[i],rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 277 } 278 /* determine number diagonal and off-diagonal counts */ 279 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 280 ierr = PetscCalloc1(m,&ld);CHKERRQ(ierr); 281 jj = 0; 282 for (i=0; i<m; i++) { 283 for (j=0; j<dlens[i]; j++) { 284 if (gmata->j[jj] < rstart) ld[i]++; 285 if (gmata->j[jj] < rstart || gmata->j[jj] >= rend) olens[i]++; 286 jj++; 287 } 288 } 289 /* send column indices to other processes */ 290 for (i=1; i<size; i++) { 291 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 292 ierr = MPI_Send(&nz,1,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 293 ierr = MPI_Send(gmata->j + gmata->i[rowners[i]],nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 294 } 295 296 /* send numerical values to other processes */ 297 for (i=1; i<size; i++) { 298 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 299 ierr = MPI_Send(gmata->a + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 300 } 301 gmataa = gmata->a; 302 gmataj = gmata->j; 303 304 } else { 305 /* receive row lengths */ 306 ierr = MPI_Recv(dlens,m,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 307 /* receive column indices */ 308 ierr = MPI_Recv(&nz,1,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 309 ierr = PetscMalloc2(nz,&gmataa,nz,&gmataj);CHKERRQ(ierr); 310 ierr = MPI_Recv(gmataj,nz,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 311 /* determine number diagonal and off-diagonal counts */ 312 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 313 ierr = PetscCalloc1(m,&ld);CHKERRQ(ierr); 314 jj = 0; 315 for (i=0; i<m; i++) { 316 for (j=0; j<dlens[i]; j++) { 317 if (gmataj[jj] < rstart) ld[i]++; 318 if (gmataj[jj] < rstart || gmataj[jj] >= rend) olens[i]++; 319 jj++; 320 } 321 } 322 /* receive numerical values */ 323 ierr = PetscMemzero(gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); 324 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 325 } 326 /* set preallocation */ 327 for (i=0; i<m; i++) { 328 dlens[i] -= olens[i]; 329 } 330 ierr = MatSeqAIJSetPreallocation(mat,0,dlens);CHKERRQ(ierr); 331 ierr = MatMPIAIJSetPreallocation(mat,0,dlens,0,olens);CHKERRQ(ierr); 332 333 for (i=0; i<m; i++) { 334 dlens[i] += olens[i]; 335 } 336 cnt = 0; 337 for (i=0; i<m; i++) { 338 row = rstart + i; 339 ierr = MatSetValues(mat,1,&row,dlens[i],gmataj+cnt,gmataa+cnt,INSERT_VALUES);CHKERRQ(ierr); 340 cnt += dlens[i]; 341 } 342 if (rank) { 343 ierr = PetscFree2(gmataa,gmataj);CHKERRQ(ierr); 344 } 345 ierr = PetscFree2(dlens,olens);CHKERRQ(ierr); 346 ierr = PetscFree(rowners);CHKERRQ(ierr); 347 348 ((Mat_MPIAIJ*)(mat->data))->ld = ld; 349 350 *inmat = mat; 351 } else { /* column indices are already set; only need to move over numerical values from process 0 */ 352 Mat_SeqAIJ *Ad = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->A->data; 353 Mat_SeqAIJ *Ao = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->B->data; 354 mat = *inmat; 355 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 356 if (!rank) { 357 /* send numerical values to other processes */ 358 gmata = (Mat_SeqAIJ*) gmat->data; 359 ierr = MatGetOwnershipRanges(mat,(const PetscInt**)&rowners);CHKERRQ(ierr); 360 gmataa = gmata->a; 361 for (i=1; i<size; i++) { 362 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 363 ierr = MPI_Send(gmataa + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 364 } 365 nz = gmata->i[rowners[1]]-gmata->i[rowners[0]]; 366 } else { 367 /* receive numerical values from process 0*/ 368 nz = Ad->nz + Ao->nz; 369 ierr = PetscMalloc1(nz,&gmataa);CHKERRQ(ierr); gmataarestore = gmataa; 370 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 371 } 372 /* transfer numerical values into the diagonal A and off diagonal B parts of mat */ 373 ld = ((Mat_MPIAIJ*)(mat->data))->ld; 374 ad = Ad->a; 375 ao = Ao->a; 376 if (mat->rmap->n) { 377 i = 0; 378 nz = ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ao += nz; gmataa += nz; 379 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 380 } 381 for (i=1; i<mat->rmap->n; i++) { 382 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; 383 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 384 } 385 i--; 386 if (mat->rmap->n) { 387 nz = Ao->i[i+1] - Ao->i[i] - ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); 388 } 389 if (rank) { 390 ierr = PetscFree(gmataarestore);CHKERRQ(ierr); 391 } 392 } 393 ierr = MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 394 ierr = MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 395 PetscFunctionReturn(0); 396 } 397 398 /* 399 Local utility routine that creates a mapping from the global column 400 number to the local number in the off-diagonal part of the local 401 storage of the matrix. When PETSC_USE_CTABLE is used this is scalable at 402 a slightly higher hash table cost; without it it is not scalable (each processor 403 has an order N integer array but is fast to acess. 404 */ 405 PetscErrorCode MatCreateColmap_MPIAIJ_Private(Mat mat) 406 { 407 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 408 PetscErrorCode ierr; 409 PetscInt n = aij->B->cmap->n,i; 410 411 PetscFunctionBegin; 412 if (!aij->garray) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"MPIAIJ Matrix was assembled but is missing garray"); 413 #if defined(PETSC_USE_CTABLE) 414 ierr = PetscTableCreate(n,mat->cmap->N+1,&aij->colmap);CHKERRQ(ierr); 415 for (i=0; i<n; i++) { 416 ierr = PetscTableAdd(aij->colmap,aij->garray[i]+1,i+1,INSERT_VALUES);CHKERRQ(ierr); 417 } 418 #else 419 ierr = PetscCalloc1(mat->cmap->N+1,&aij->colmap);CHKERRQ(ierr); 420 ierr = PetscLogObjectMemory((PetscObject)mat,(mat->cmap->N+1)*sizeof(PetscInt));CHKERRQ(ierr); 421 for (i=0; i<n; i++) aij->colmap[aij->garray[i]] = i+1; 422 #endif 423 PetscFunctionReturn(0); 424 } 425 426 #define MatSetValues_SeqAIJ_A_Private(row,col,value,addv,orow,ocol) \ 427 { \ 428 if (col <= lastcol1) low1 = 0; \ 429 else high1 = nrow1; \ 430 lastcol1 = col;\ 431 while (high1-low1 > 5) { \ 432 t = (low1+high1)/2; \ 433 if (rp1[t] > col) high1 = t; \ 434 else low1 = t; \ 435 } \ 436 for (_i=low1; _i<high1; _i++) { \ 437 if (rp1[_i] > col) break; \ 438 if (rp1[_i] == col) { \ 439 if (addv == ADD_VALUES) ap1[_i] += value; \ 440 else ap1[_i] = value; \ 441 goto a_noinsert; \ 442 } \ 443 } \ 444 if (value == 0.0 && ignorezeroentries && row != col) {low1 = 0; high1 = nrow1;goto a_noinsert;} \ 445 if (nonew == 1) {low1 = 0; high1 = nrow1; goto a_noinsert;} \ 446 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); \ 447 MatSeqXAIJReallocateAIJ(A,am,1,nrow1,row,col,rmax1,aa,ai,aj,rp1,ap1,aimax,nonew,MatScalar); \ 448 N = nrow1++ - 1; a->nz++; high1++; \ 449 /* shift up all the later entries in this row */ \ 450 for (ii=N; ii>=_i; ii--) { \ 451 rp1[ii+1] = rp1[ii]; \ 452 ap1[ii+1] = ap1[ii]; \ 453 } \ 454 rp1[_i] = col; \ 455 ap1[_i] = value; \ 456 A->nonzerostate++;\ 457 a_noinsert: ; \ 458 ailen[row] = nrow1; \ 459 } 460 461 #define MatSetValues_SeqAIJ_B_Private(row,col,value,addv,orow,ocol) \ 462 { \ 463 if (col <= lastcol2) low2 = 0; \ 464 else high2 = nrow2; \ 465 lastcol2 = col; \ 466 while (high2-low2 > 5) { \ 467 t = (low2+high2)/2; \ 468 if (rp2[t] > col) high2 = t; \ 469 else low2 = t; \ 470 } \ 471 for (_i=low2; _i<high2; _i++) { \ 472 if (rp2[_i] > col) break; \ 473 if (rp2[_i] == col) { \ 474 if (addv == ADD_VALUES) ap2[_i] += value; \ 475 else ap2[_i] = value; \ 476 goto b_noinsert; \ 477 } \ 478 } \ 479 if (value == 0.0 && ignorezeroentries) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 480 if (nonew == 1) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 481 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); \ 482 MatSeqXAIJReallocateAIJ(B,bm,1,nrow2,row,col,rmax2,ba,bi,bj,rp2,ap2,bimax,nonew,MatScalar); \ 483 N = nrow2++ - 1; b->nz++; high2++; \ 484 /* shift up all the later entries in this row */ \ 485 for (ii=N; ii>=_i; ii--) { \ 486 rp2[ii+1] = rp2[ii]; \ 487 ap2[ii+1] = ap2[ii]; \ 488 } \ 489 rp2[_i] = col; \ 490 ap2[_i] = value; \ 491 B->nonzerostate++; \ 492 b_noinsert: ; \ 493 bilen[row] = nrow2; \ 494 } 495 496 PetscErrorCode MatSetValuesRow_MPIAIJ(Mat A,PetscInt row,const PetscScalar v[]) 497 { 498 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 499 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data,*b = (Mat_SeqAIJ*)mat->B->data; 500 PetscErrorCode ierr; 501 PetscInt l,*garray = mat->garray,diag; 502 503 PetscFunctionBegin; 504 /* code only works for square matrices A */ 505 506 /* find size of row to the left of the diagonal part */ 507 ierr = MatGetOwnershipRange(A,&diag,0);CHKERRQ(ierr); 508 row = row - diag; 509 for (l=0; l<b->i[row+1]-b->i[row]; l++) { 510 if (garray[b->j[b->i[row]+l]] > diag) break; 511 } 512 ierr = PetscMemcpy(b->a+b->i[row],v,l*sizeof(PetscScalar));CHKERRQ(ierr); 513 514 /* diagonal part */ 515 ierr = PetscMemcpy(a->a+a->i[row],v+l,(a->i[row+1]-a->i[row])*sizeof(PetscScalar));CHKERRQ(ierr); 516 517 /* right of diagonal part */ 518 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); 519 PetscFunctionReturn(0); 520 } 521 522 PetscErrorCode MatSetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv) 523 { 524 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 525 PetscScalar value; 526 PetscErrorCode ierr; 527 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 528 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 529 PetscBool roworiented = aij->roworiented; 530 531 /* Some Variables required in the macro */ 532 Mat A = aij->A; 533 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 534 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 535 MatScalar *aa = a->a; 536 PetscBool ignorezeroentries = a->ignorezeroentries; 537 Mat B = aij->B; 538 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 539 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 540 MatScalar *ba = b->a; 541 542 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 543 PetscInt nonew; 544 MatScalar *ap1,*ap2; 545 546 PetscFunctionBegin; 547 for (i=0; i<m; i++) { 548 if (im[i] < 0) continue; 549 #if defined(PETSC_USE_DEBUG) 550 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); 551 #endif 552 if (im[i] >= rstart && im[i] < rend) { 553 row = im[i] - rstart; 554 lastcol1 = -1; 555 rp1 = aj + ai[row]; 556 ap1 = aa + ai[row]; 557 rmax1 = aimax[row]; 558 nrow1 = ailen[row]; 559 low1 = 0; 560 high1 = nrow1; 561 lastcol2 = -1; 562 rp2 = bj + bi[row]; 563 ap2 = ba + bi[row]; 564 rmax2 = bimax[row]; 565 nrow2 = bilen[row]; 566 low2 = 0; 567 high2 = nrow2; 568 569 for (j=0; j<n; j++) { 570 if (roworiented) value = v[i*n+j]; 571 else value = v[i+j*m]; 572 if (in[j] >= cstart && in[j] < cend) { 573 col = in[j] - cstart; 574 nonew = a->nonew; 575 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 576 MatSetValues_SeqAIJ_A_Private(row,col,value,addv,im[i],in[j]); 577 } else if (in[j] < 0) continue; 578 #if defined(PETSC_USE_DEBUG) 579 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); 580 #endif 581 else { 582 if (mat->was_assembled) { 583 if (!aij->colmap) { 584 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 585 } 586 #if defined(PETSC_USE_CTABLE) 587 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 588 col--; 589 #else 590 col = aij->colmap[in[j]] - 1; 591 #endif 592 if (col < 0 && !((Mat_SeqAIJ*)(aij->B->data))->nonew) { 593 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 594 col = in[j]; 595 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 596 B = aij->B; 597 b = (Mat_SeqAIJ*)B->data; 598 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; ba = b->a; 599 rp2 = bj + bi[row]; 600 ap2 = ba + bi[row]; 601 rmax2 = bimax[row]; 602 nrow2 = bilen[row]; 603 low2 = 0; 604 high2 = nrow2; 605 bm = aij->B->rmap->n; 606 ba = b->a; 607 } 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]); 608 } else col = in[j]; 609 nonew = b->nonew; 610 MatSetValues_SeqAIJ_B_Private(row,col,value,addv,im[i],in[j]); 611 } 612 } 613 } else { 614 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]); 615 if (!aij->donotstash) { 616 mat->assembled = PETSC_FALSE; 617 if (roworiented) { 618 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 619 } else { 620 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 621 } 622 } 623 } 624 } 625 PetscFunctionReturn(0); 626 } 627 628 PetscErrorCode MatGetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[]) 629 { 630 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 631 PetscErrorCode ierr; 632 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 633 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 634 635 PetscFunctionBegin; 636 for (i=0; i<m; i++) { 637 if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]);*/ 638 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); 639 if (idxm[i] >= rstart && idxm[i] < rend) { 640 row = idxm[i] - rstart; 641 for (j=0; j<n; j++) { 642 if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",idxn[j]); */ 643 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); 644 if (idxn[j] >= cstart && idxn[j] < cend) { 645 col = idxn[j] - cstart; 646 ierr = MatGetValues(aij->A,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 647 } else { 648 if (!aij->colmap) { 649 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 650 } 651 #if defined(PETSC_USE_CTABLE) 652 ierr = PetscTableFind(aij->colmap,idxn[j]+1,&col);CHKERRQ(ierr); 653 col--; 654 #else 655 col = aij->colmap[idxn[j]] - 1; 656 #endif 657 if ((col < 0) || (aij->garray[col] != idxn[j])) *(v+i*n+j) = 0.0; 658 else { 659 ierr = MatGetValues(aij->B,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 660 } 661 } 662 } 663 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only local values currently supported"); 664 } 665 PetscFunctionReturn(0); 666 } 667 668 extern PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat,Vec,Vec); 669 670 PetscErrorCode MatAssemblyBegin_MPIAIJ(Mat mat,MatAssemblyType mode) 671 { 672 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 673 PetscErrorCode ierr; 674 PetscInt nstash,reallocs; 675 676 PetscFunctionBegin; 677 if (aij->donotstash || mat->nooffprocentries) PetscFunctionReturn(0); 678 679 ierr = MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap->range);CHKERRQ(ierr); 680 ierr = MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);CHKERRQ(ierr); 681 ierr = PetscInfo2(aij->A,"Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);CHKERRQ(ierr); 682 PetscFunctionReturn(0); 683 } 684 685 PetscErrorCode MatAssemblyEnd_MPIAIJ(Mat mat,MatAssemblyType mode) 686 { 687 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 688 Mat_SeqAIJ *a = (Mat_SeqAIJ*)aij->A->data; 689 PetscErrorCode ierr; 690 PetscMPIInt n; 691 PetscInt i,j,rstart,ncols,flg; 692 PetscInt *row,*col; 693 PetscBool other_disassembled; 694 PetscScalar *val; 695 696 /* do not use 'b = (Mat_SeqAIJ*)aij->B->data' as B can be reset in disassembly */ 697 698 PetscFunctionBegin; 699 if (!aij->donotstash && !mat->nooffprocentries) { 700 while (1) { 701 ierr = MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);CHKERRQ(ierr); 702 if (!flg) break; 703 704 for (i=0; i<n; ) { 705 /* Now identify the consecutive vals belonging to the same row */ 706 for (j=i,rstart=row[j]; j<n; j++) { 707 if (row[j] != rstart) break; 708 } 709 if (j < n) ncols = j-i; 710 else ncols = n-i; 711 /* Now assemble all these values with a single function call */ 712 ierr = MatSetValues_MPIAIJ(mat,1,row+i,ncols,col+i,val+i,mat->insertmode);CHKERRQ(ierr); 713 714 i = j; 715 } 716 } 717 ierr = MatStashScatterEnd_Private(&mat->stash);CHKERRQ(ierr); 718 } 719 ierr = MatAssemblyBegin(aij->A,mode);CHKERRQ(ierr); 720 ierr = MatAssemblyEnd(aij->A,mode);CHKERRQ(ierr); 721 722 /* determine if any processor has disassembled, if so we must 723 also disassemble ourselfs, in order that we may reassemble. */ 724 /* 725 if nonzero structure of submatrix B cannot change then we know that 726 no processor disassembled thus we can skip this stuff 727 */ 728 if (!((Mat_SeqAIJ*)aij->B->data)->nonew) { 729 ierr = MPIU_Allreduce(&mat->was_assembled,&other_disassembled,1,MPIU_BOOL,MPI_PROD,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 730 if (mat->was_assembled && !other_disassembled) { 731 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 732 } 733 } 734 if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) { 735 ierr = MatSetUpMultiply_MPIAIJ(mat);CHKERRQ(ierr); 736 } 737 ierr = MatSetOption(aij->B,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 738 ierr = MatAssemblyBegin(aij->B,mode);CHKERRQ(ierr); 739 ierr = MatAssemblyEnd(aij->B,mode);CHKERRQ(ierr); 740 741 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 742 743 aij->rowvalues = 0; 744 745 ierr = VecDestroy(&aij->diag);CHKERRQ(ierr); 746 if (a->inode.size) mat->ops->multdiagonalblock = MatMultDiagonalBlock_MPIAIJ; 747 748 /* if no new nonzero locations are allowed in matrix then only set the matrix state the first time through */ 749 if ((!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) || !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 750 PetscObjectState state = aij->A->nonzerostate + aij->B->nonzerostate; 751 ierr = MPIU_Allreduce(&state,&mat->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 752 } 753 PetscFunctionReturn(0); 754 } 755 756 PetscErrorCode MatZeroEntries_MPIAIJ(Mat A) 757 { 758 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 759 PetscErrorCode ierr; 760 761 PetscFunctionBegin; 762 ierr = MatZeroEntries(l->A);CHKERRQ(ierr); 763 ierr = MatZeroEntries(l->B);CHKERRQ(ierr); 764 PetscFunctionReturn(0); 765 } 766 767 PetscErrorCode MatZeroRows_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 768 { 769 Mat_MPIAIJ *mat = (Mat_MPIAIJ *) A->data; 770 PetscInt *lrows; 771 PetscInt r, len; 772 PetscErrorCode ierr; 773 774 PetscFunctionBegin; 775 /* get locally owned rows */ 776 ierr = MatZeroRowsMapLocal_Private(A,N,rows,&len,&lrows);CHKERRQ(ierr); 777 /* fix right hand side if needed */ 778 if (x && b) { 779 const PetscScalar *xx; 780 PetscScalar *bb; 781 782 ierr = VecGetArrayRead(x, &xx);CHKERRQ(ierr); 783 ierr = VecGetArray(b, &bb);CHKERRQ(ierr); 784 for (r = 0; r < len; ++r) bb[lrows[r]] = diag*xx[lrows[r]]; 785 ierr = VecRestoreArrayRead(x, &xx);CHKERRQ(ierr); 786 ierr = VecRestoreArray(b, &bb);CHKERRQ(ierr); 787 } 788 /* Must zero l->B before l->A because the (diag) case below may put values into l->B*/ 789 ierr = MatZeroRows(mat->B, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 790 if (A->congruentlayouts == -1) { /* first time we compare rows and cols layouts */ 791 PetscBool cong; 792 ierr = PetscLayoutCompare(A->rmap,A->cmap,&cong);CHKERRQ(ierr); 793 if (cong) A->congruentlayouts = 1; 794 else A->congruentlayouts = 0; 795 } 796 if ((diag != 0.0) && A->congruentlayouts) { 797 ierr = MatZeroRows(mat->A, len, lrows, diag, NULL, NULL);CHKERRQ(ierr); 798 } else if (diag != 0.0) { 799 ierr = MatZeroRows(mat->A, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 800 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"); 801 for (r = 0; r < len; ++r) { 802 const PetscInt row = lrows[r] + A->rmap->rstart; 803 ierr = MatSetValues(A, 1, &row, 1, &row, &diag, INSERT_VALUES);CHKERRQ(ierr); 804 } 805 ierr = MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 806 ierr = MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 807 } else { 808 ierr = MatZeroRows(mat->A, len, lrows, 0.0, NULL, NULL);CHKERRQ(ierr); 809 } 810 ierr = PetscFree(lrows);CHKERRQ(ierr); 811 812 /* only change matrix nonzero state if pattern was allowed to be changed */ 813 if (!((Mat_SeqAIJ*)(mat->A->data))->keepnonzeropattern) { 814 PetscObjectState state = mat->A->nonzerostate + mat->B->nonzerostate; 815 ierr = MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 816 } 817 PetscFunctionReturn(0); 818 } 819 820 PetscErrorCode MatZeroRowsColumns_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 821 { 822 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 823 PetscErrorCode ierr; 824 PetscMPIInt n = A->rmap->n; 825 PetscInt i,j,r,m,p = 0,len = 0; 826 PetscInt *lrows,*owners = A->rmap->range; 827 PetscSFNode *rrows; 828 PetscSF sf; 829 const PetscScalar *xx; 830 PetscScalar *bb,*mask; 831 Vec xmask,lmask; 832 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)l->B->data; 833 const PetscInt *aj, *ii,*ridx; 834 PetscScalar *aa; 835 836 PetscFunctionBegin; 837 /* Create SF where leaves are input rows and roots are owned rows */ 838 ierr = PetscMalloc1(n, &lrows);CHKERRQ(ierr); 839 for (r = 0; r < n; ++r) lrows[r] = -1; 840 ierr = PetscMalloc1(N, &rrows);CHKERRQ(ierr); 841 for (r = 0; r < N; ++r) { 842 const PetscInt idx = rows[r]; 843 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); 844 if (idx < owners[p] || owners[p+1] <= idx) { /* short-circuit the search if the last p owns this row too */ 845 ierr = PetscLayoutFindOwner(A->rmap,idx,&p);CHKERRQ(ierr); 846 } 847 rrows[r].rank = p; 848 rrows[r].index = rows[r] - owners[p]; 849 } 850 ierr = PetscSFCreate(PetscObjectComm((PetscObject) A), &sf);CHKERRQ(ierr); 851 ierr = PetscSFSetGraph(sf, n, N, NULL, PETSC_OWN_POINTER, rrows, PETSC_OWN_POINTER);CHKERRQ(ierr); 852 /* Collect flags for rows to be zeroed */ 853 ierr = PetscSFReduceBegin(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);CHKERRQ(ierr); 854 ierr = PetscSFReduceEnd(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);CHKERRQ(ierr); 855 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 856 /* Compress and put in row numbers */ 857 for (r = 0; r < n; ++r) if (lrows[r] >= 0) lrows[len++] = r; 858 /* zero diagonal part of matrix */ 859 ierr = MatZeroRowsColumns(l->A,len,lrows,diag,x,b);CHKERRQ(ierr); 860 /* handle off diagonal part of matrix */ 861 ierr = MatCreateVecs(A,&xmask,NULL);CHKERRQ(ierr); 862 ierr = VecDuplicate(l->lvec,&lmask);CHKERRQ(ierr); 863 ierr = VecGetArray(xmask,&bb);CHKERRQ(ierr); 864 for (i=0; i<len; i++) bb[lrows[i]] = 1; 865 ierr = VecRestoreArray(xmask,&bb);CHKERRQ(ierr); 866 ierr = VecScatterBegin(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 867 ierr = VecScatterEnd(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 868 ierr = VecDestroy(&xmask);CHKERRQ(ierr); 869 if (x) { 870 ierr = VecScatterBegin(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 871 ierr = VecScatterEnd(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 872 ierr = VecGetArrayRead(l->lvec,&xx);CHKERRQ(ierr); 873 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 874 } 875 ierr = VecGetArray(lmask,&mask);CHKERRQ(ierr); 876 /* remove zeroed rows of off diagonal matrix */ 877 ii = aij->i; 878 for (i=0; i<len; i++) { 879 ierr = PetscMemzero(aij->a + ii[lrows[i]],(ii[lrows[i]+1] - ii[lrows[i]])*sizeof(PetscScalar));CHKERRQ(ierr); 880 } 881 /* loop over all elements of off process part of matrix zeroing removed columns*/ 882 if (aij->compressedrow.use) { 883 m = aij->compressedrow.nrows; 884 ii = aij->compressedrow.i; 885 ridx = aij->compressedrow.rindex; 886 for (i=0; i<m; i++) { 887 n = ii[i+1] - ii[i]; 888 aj = aij->j + ii[i]; 889 aa = aij->a + ii[i]; 890 891 for (j=0; j<n; j++) { 892 if (PetscAbsScalar(mask[*aj])) { 893 if (b) bb[*ridx] -= *aa*xx[*aj]; 894 *aa = 0.0; 895 } 896 aa++; 897 aj++; 898 } 899 ridx++; 900 } 901 } else { /* do not use compressed row format */ 902 m = l->B->rmap->n; 903 for (i=0; i<m; i++) { 904 n = ii[i+1] - ii[i]; 905 aj = aij->j + ii[i]; 906 aa = aij->a + ii[i]; 907 for (j=0; j<n; j++) { 908 if (PetscAbsScalar(mask[*aj])) { 909 if (b) bb[i] -= *aa*xx[*aj]; 910 *aa = 0.0; 911 } 912 aa++; 913 aj++; 914 } 915 } 916 } 917 if (x) { 918 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 919 ierr = VecRestoreArrayRead(l->lvec,&xx);CHKERRQ(ierr); 920 } 921 ierr = VecRestoreArray(lmask,&mask);CHKERRQ(ierr); 922 ierr = VecDestroy(&lmask);CHKERRQ(ierr); 923 ierr = PetscFree(lrows);CHKERRQ(ierr); 924 925 /* only change matrix nonzero state if pattern was allowed to be changed */ 926 if (!((Mat_SeqAIJ*)(l->A->data))->keepnonzeropattern) { 927 PetscObjectState state = l->A->nonzerostate + l->B->nonzerostate; 928 ierr = MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 929 } 930 PetscFunctionReturn(0); 931 } 932 933 PetscErrorCode MatMult_MPIAIJ(Mat A,Vec xx,Vec yy) 934 { 935 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 936 PetscErrorCode ierr; 937 PetscInt nt; 938 939 PetscFunctionBegin; 940 ierr = VecGetLocalSize(xx,&nt);CHKERRQ(ierr); 941 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); 942 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 943 ierr = (*a->A->ops->mult)(a->A,xx,yy);CHKERRQ(ierr); 944 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 945 ierr = (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);CHKERRQ(ierr); 946 PetscFunctionReturn(0); 947 } 948 949 PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat A,Vec bb,Vec xx) 950 { 951 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 952 PetscErrorCode ierr; 953 954 PetscFunctionBegin; 955 ierr = MatMultDiagonalBlock(a->A,bb,xx);CHKERRQ(ierr); 956 PetscFunctionReturn(0); 957 } 958 959 PetscErrorCode MatMultAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 960 { 961 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 962 PetscErrorCode ierr; 963 964 PetscFunctionBegin; 965 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 966 ierr = (*a->A->ops->multadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 967 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 968 ierr = (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);CHKERRQ(ierr); 969 PetscFunctionReturn(0); 970 } 971 972 PetscErrorCode MatMultTranspose_MPIAIJ(Mat A,Vec xx,Vec yy) 973 { 974 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 975 PetscErrorCode ierr; 976 PetscBool merged; 977 978 PetscFunctionBegin; 979 ierr = VecScatterGetMerged(a->Mvctx,&merged);CHKERRQ(ierr); 980 /* do nondiagonal part */ 981 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 982 if (!merged) { 983 /* send it on its way */ 984 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 985 /* do local part */ 986 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 987 /* receive remote parts: note this assumes the values are not actually */ 988 /* added in yy until the next line, */ 989 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 990 } else { 991 /* do local part */ 992 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 993 /* send it on its way */ 994 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 995 /* values actually were received in the Begin() but we need to call this nop */ 996 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 997 } 998 PetscFunctionReturn(0); 999 } 1000 1001 PetscErrorCode MatIsTranspose_MPIAIJ(Mat Amat,Mat Bmat,PetscReal tol,PetscBool *f) 1002 { 1003 MPI_Comm comm; 1004 Mat_MPIAIJ *Aij = (Mat_MPIAIJ*) Amat->data, *Bij; 1005 Mat Adia = Aij->A, Bdia, Aoff,Boff,*Aoffs,*Boffs; 1006 IS Me,Notme; 1007 PetscErrorCode ierr; 1008 PetscInt M,N,first,last,*notme,i; 1009 PetscMPIInt size; 1010 1011 PetscFunctionBegin; 1012 /* Easy test: symmetric diagonal block */ 1013 Bij = (Mat_MPIAIJ*) Bmat->data; Bdia = Bij->A; 1014 ierr = MatIsTranspose(Adia,Bdia,tol,f);CHKERRQ(ierr); 1015 if (!*f) PetscFunctionReturn(0); 1016 ierr = PetscObjectGetComm((PetscObject)Amat,&comm);CHKERRQ(ierr); 1017 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1018 if (size == 1) PetscFunctionReturn(0); 1019 1020 /* Hard test: off-diagonal block. This takes a MatCreateSubMatrix. */ 1021 ierr = MatGetSize(Amat,&M,&N);CHKERRQ(ierr); 1022 ierr = MatGetOwnershipRange(Amat,&first,&last);CHKERRQ(ierr); 1023 ierr = PetscMalloc1(N-last+first,¬me);CHKERRQ(ierr); 1024 for (i=0; i<first; i++) notme[i] = i; 1025 for (i=last; i<M; i++) notme[i-last+first] = i; 1026 ierr = ISCreateGeneral(MPI_COMM_SELF,N-last+first,notme,PETSC_COPY_VALUES,&Notme);CHKERRQ(ierr); 1027 ierr = ISCreateStride(MPI_COMM_SELF,last-first,first,1,&Me);CHKERRQ(ierr); 1028 ierr = MatCreateSubMatrices(Amat,1,&Me,&Notme,MAT_INITIAL_MATRIX,&Aoffs);CHKERRQ(ierr); 1029 Aoff = Aoffs[0]; 1030 ierr = MatCreateSubMatrices(Bmat,1,&Notme,&Me,MAT_INITIAL_MATRIX,&Boffs);CHKERRQ(ierr); 1031 Boff = Boffs[0]; 1032 ierr = MatIsTranspose(Aoff,Boff,tol,f);CHKERRQ(ierr); 1033 ierr = MatDestroyMatrices(1,&Aoffs);CHKERRQ(ierr); 1034 ierr = MatDestroyMatrices(1,&Boffs);CHKERRQ(ierr); 1035 ierr = ISDestroy(&Me);CHKERRQ(ierr); 1036 ierr = ISDestroy(&Notme);CHKERRQ(ierr); 1037 ierr = PetscFree(notme);CHKERRQ(ierr); 1038 PetscFunctionReturn(0); 1039 } 1040 1041 PetscErrorCode MatMultTransposeAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1042 { 1043 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1044 PetscErrorCode ierr; 1045 1046 PetscFunctionBegin; 1047 /* do nondiagonal part */ 1048 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 1049 /* send it on its way */ 1050 ierr = VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1051 /* do local part */ 1052 ierr = (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 1053 /* receive remote parts */ 1054 ierr = VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1055 PetscFunctionReturn(0); 1056 } 1057 1058 /* 1059 This only works correctly for square matrices where the subblock A->A is the 1060 diagonal block 1061 */ 1062 PetscErrorCode MatGetDiagonal_MPIAIJ(Mat A,Vec v) 1063 { 1064 PetscErrorCode ierr; 1065 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1066 1067 PetscFunctionBegin; 1068 if (A->rmap->N != A->cmap->N) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); 1069 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"); 1070 ierr = MatGetDiagonal(a->A,v);CHKERRQ(ierr); 1071 PetscFunctionReturn(0); 1072 } 1073 1074 PetscErrorCode MatScale_MPIAIJ(Mat A,PetscScalar aa) 1075 { 1076 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1077 PetscErrorCode ierr; 1078 1079 PetscFunctionBegin; 1080 ierr = MatScale(a->A,aa);CHKERRQ(ierr); 1081 ierr = MatScale(a->B,aa);CHKERRQ(ierr); 1082 PetscFunctionReturn(0); 1083 } 1084 1085 PetscErrorCode MatDestroy_MPIAIJ(Mat mat) 1086 { 1087 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1088 PetscErrorCode ierr; 1089 1090 PetscFunctionBegin; 1091 #if defined(PETSC_USE_LOG) 1092 PetscLogObjectState((PetscObject)mat,"Rows=%D, Cols=%D",mat->rmap->N,mat->cmap->N); 1093 #endif 1094 ierr = MatStashDestroy_Private(&mat->stash);CHKERRQ(ierr); 1095 ierr = VecDestroy(&aij->diag);CHKERRQ(ierr); 1096 ierr = MatDestroy(&aij->A);CHKERRQ(ierr); 1097 ierr = MatDestroy(&aij->B);CHKERRQ(ierr); 1098 #if defined(PETSC_USE_CTABLE) 1099 ierr = PetscTableDestroy(&aij->colmap);CHKERRQ(ierr); 1100 #else 1101 ierr = PetscFree(aij->colmap);CHKERRQ(ierr); 1102 #endif 1103 ierr = PetscFree(aij->garray);CHKERRQ(ierr); 1104 ierr = VecDestroy(&aij->lvec);CHKERRQ(ierr); 1105 ierr = VecScatterDestroy(&aij->Mvctx);CHKERRQ(ierr); 1106 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 1107 ierr = PetscFree(aij->ld);CHKERRQ(ierr); 1108 ierr = PetscFree(mat->data);CHKERRQ(ierr); 1109 1110 ierr = PetscObjectChangeTypeName((PetscObject)mat,0);CHKERRQ(ierr); 1111 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C",NULL);CHKERRQ(ierr); 1112 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C",NULL);CHKERRQ(ierr); 1113 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatIsTranspose_C",NULL);CHKERRQ(ierr); 1114 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocation_C",NULL);CHKERRQ(ierr); 1115 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocationCSR_C",NULL);CHKERRQ(ierr); 1116 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatDiagonalScaleLocal_C",NULL);CHKERRQ(ierr); 1117 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_mpisbaij_C",NULL);CHKERRQ(ierr); 1118 #if defined(PETSC_HAVE_ELEMENTAL) 1119 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_elemental_C",NULL);CHKERRQ(ierr); 1120 #endif 1121 #if defined(PETSC_HAVE_HYPRE) 1122 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_hypre_C",NULL);CHKERRQ(ierr); 1123 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMatMatMult_transpose_mpiaij_mpiaij_C",NULL);CHKERRQ(ierr); 1124 #endif 1125 PetscFunctionReturn(0); 1126 } 1127 1128 PetscErrorCode MatView_MPIAIJ_Binary(Mat mat,PetscViewer viewer) 1129 { 1130 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1131 Mat_SeqAIJ *A = (Mat_SeqAIJ*)aij->A->data; 1132 Mat_SeqAIJ *B = (Mat_SeqAIJ*)aij->B->data; 1133 PetscErrorCode ierr; 1134 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 1135 int fd; 1136 PetscInt nz,header[4],*row_lengths,*range=0,rlen,i; 1137 PetscInt nzmax,*column_indices,j,k,col,*garray = aij->garray,cnt,cstart = mat->cmap->rstart,rnz = 0; 1138 PetscScalar *column_values; 1139 PetscInt message_count,flowcontrolcount; 1140 FILE *file; 1141 1142 PetscFunctionBegin; 1143 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);CHKERRQ(ierr); 1144 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr); 1145 nz = A->nz + B->nz; 1146 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 1147 if (!rank) { 1148 header[0] = MAT_FILE_CLASSID; 1149 header[1] = mat->rmap->N; 1150 header[2] = mat->cmap->N; 1151 1152 ierr = MPI_Reduce(&nz,&header[3],1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1153 ierr = PetscBinaryWrite(fd,header,4,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1154 /* get largest number of rows any processor has */ 1155 rlen = mat->rmap->n; 1156 range = mat->rmap->range; 1157 for (i=1; i<size; i++) rlen = PetscMax(rlen,range[i+1] - range[i]); 1158 } else { 1159 ierr = MPI_Reduce(&nz,0,1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1160 rlen = mat->rmap->n; 1161 } 1162 1163 /* load up the local row counts */ 1164 ierr = PetscMalloc1(rlen+1,&row_lengths);CHKERRQ(ierr); 1165 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]; 1166 1167 /* store the row lengths to the file */ 1168 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1169 if (!rank) { 1170 ierr = PetscBinaryWrite(fd,row_lengths,mat->rmap->n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1171 for (i=1; i<size; i++) { 1172 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1173 rlen = range[i+1] - range[i]; 1174 ierr = MPIULong_Recv(row_lengths,rlen,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1175 ierr = PetscBinaryWrite(fd,row_lengths,rlen,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1176 } 1177 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1178 } else { 1179 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1180 ierr = MPIULong_Send(row_lengths,mat->rmap->n,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1181 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1182 } 1183 ierr = PetscFree(row_lengths);CHKERRQ(ierr); 1184 1185 /* load up the local column indices */ 1186 nzmax = nz; /* th processor needs space a largest processor needs */ 1187 ierr = MPI_Reduce(&nz,&nzmax,1,MPIU_INT,MPI_MAX,0,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1188 ierr = PetscMalloc1(nzmax+1,&column_indices);CHKERRQ(ierr); 1189 cnt = 0; 1190 for (i=0; i<mat->rmap->n; i++) { 1191 for (j=B->i[i]; j<B->i[i+1]; j++) { 1192 if ((col = garray[B->j[j]]) > cstart) break; 1193 column_indices[cnt++] = col; 1194 } 1195 for (k=A->i[i]; k<A->i[i+1]; k++) column_indices[cnt++] = A->j[k] + cstart; 1196 for (; j<B->i[i+1]; j++) column_indices[cnt++] = garray[B->j[j]]; 1197 } 1198 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); 1199 1200 /* store the column indices to the file */ 1201 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1202 if (!rank) { 1203 MPI_Status status; 1204 ierr = PetscBinaryWrite(fd,column_indices,nz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1205 for (i=1; i<size; i++) { 1206 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1207 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);CHKERRQ(ierr); 1208 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1209 ierr = MPIULong_Recv(column_indices,rnz,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1210 ierr = PetscBinaryWrite(fd,column_indices,rnz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1211 } 1212 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1213 } else { 1214 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1215 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1216 ierr = MPIULong_Send(column_indices,nz,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1217 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1218 } 1219 ierr = PetscFree(column_indices);CHKERRQ(ierr); 1220 1221 /* load up the local column values */ 1222 ierr = PetscMalloc1(nzmax+1,&column_values);CHKERRQ(ierr); 1223 cnt = 0; 1224 for (i=0; i<mat->rmap->n; i++) { 1225 for (j=B->i[i]; j<B->i[i+1]; j++) { 1226 if (garray[B->j[j]] > cstart) break; 1227 column_values[cnt++] = B->a[j]; 1228 } 1229 for (k=A->i[i]; k<A->i[i+1]; k++) column_values[cnt++] = A->a[k]; 1230 for (; j<B->i[i+1]; j++) column_values[cnt++] = B->a[j]; 1231 } 1232 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); 1233 1234 /* store the column values to the file */ 1235 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1236 if (!rank) { 1237 MPI_Status status; 1238 ierr = PetscBinaryWrite(fd,column_values,nz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1239 for (i=1; i<size; i++) { 1240 ierr = PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);CHKERRQ(ierr); 1241 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);CHKERRQ(ierr); 1242 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1243 ierr = MPIULong_Recv(column_values,rnz,MPIU_SCALAR,i,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1244 ierr = PetscBinaryWrite(fd,column_values,rnz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1245 } 1246 ierr = PetscViewerFlowControlEndMaster(viewer,&message_count);CHKERRQ(ierr); 1247 } else { 1248 ierr = PetscViewerFlowControlStepWorker(viewer,rank,&message_count);CHKERRQ(ierr); 1249 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1250 ierr = MPIULong_Send(column_values,nz,MPIU_SCALAR,0,tag,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1251 ierr = PetscViewerFlowControlEndWorker(viewer,&message_count);CHKERRQ(ierr); 1252 } 1253 ierr = PetscFree(column_values);CHKERRQ(ierr); 1254 1255 ierr = PetscViewerBinaryGetInfoPointer(viewer,&file);CHKERRQ(ierr); 1256 if (file) fprintf(file,"-matload_block_size %d\n",(int)PetscAbs(mat->rmap->bs)); 1257 PetscFunctionReturn(0); 1258 } 1259 1260 #include <petscdraw.h> 1261 PetscErrorCode MatView_MPIAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer) 1262 { 1263 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1264 PetscErrorCode ierr; 1265 PetscMPIInt rank = aij->rank,size = aij->size; 1266 PetscBool isdraw,iascii,isbinary; 1267 PetscViewer sviewer; 1268 PetscViewerFormat format; 1269 1270 PetscFunctionBegin; 1271 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1272 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1273 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1274 if (iascii) { 1275 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 1276 if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 1277 MatInfo info; 1278 PetscBool inodes; 1279 1280 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);CHKERRQ(ierr); 1281 ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr); 1282 ierr = MatInodeGetInodeSizes(aij->A,NULL,(PetscInt**)&inodes,NULL);CHKERRQ(ierr); 1283 ierr = PetscViewerASCIIPushSynchronized(viewer);CHKERRQ(ierr); 1284 if (!inodes) { 1285 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, not using I-node routines\n", 1286 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1287 } else { 1288 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, using I-node routines\n", 1289 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1290 } 1291 ierr = MatGetInfo(aij->A,MAT_LOCAL,&info);CHKERRQ(ierr); 1292 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1293 ierr = MatGetInfo(aij->B,MAT_LOCAL,&info);CHKERRQ(ierr); 1294 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1295 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 1296 ierr = PetscViewerASCIIPopSynchronized(viewer);CHKERRQ(ierr); 1297 ierr = PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");CHKERRQ(ierr); 1298 ierr = VecScatterView(aij->Mvctx,viewer);CHKERRQ(ierr); 1299 PetscFunctionReturn(0); 1300 } else if (format == PETSC_VIEWER_ASCII_INFO) { 1301 PetscInt inodecount,inodelimit,*inodes; 1302 ierr = MatInodeGetInodeSizes(aij->A,&inodecount,&inodes,&inodelimit);CHKERRQ(ierr); 1303 if (inodes) { 1304 ierr = PetscViewerASCIIPrintf(viewer,"using I-node (on process 0) routines: found %D nodes, limit used is %D\n",inodecount,inodelimit);CHKERRQ(ierr); 1305 } else { 1306 ierr = PetscViewerASCIIPrintf(viewer,"not using I-node (on process 0) routines\n");CHKERRQ(ierr); 1307 } 1308 PetscFunctionReturn(0); 1309 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) { 1310 PetscFunctionReturn(0); 1311 } 1312 } else if (isbinary) { 1313 if (size == 1) { 1314 ierr = PetscObjectSetName((PetscObject)aij->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1315 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 1316 } else { 1317 ierr = MatView_MPIAIJ_Binary(mat,viewer);CHKERRQ(ierr); 1318 } 1319 PetscFunctionReturn(0); 1320 } else if (isdraw) { 1321 PetscDraw draw; 1322 PetscBool isnull; 1323 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 1324 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); 1325 if (isnull) PetscFunctionReturn(0); 1326 } 1327 1328 { 1329 /* assemble the entire matrix onto first processor. */ 1330 Mat A; 1331 Mat_SeqAIJ *Aloc; 1332 PetscInt M = mat->rmap->N,N = mat->cmap->N,m,*ai,*aj,row,*cols,i,*ct; 1333 MatScalar *a; 1334 1335 ierr = MatCreate(PetscObjectComm((PetscObject)mat),&A);CHKERRQ(ierr); 1336 if (!rank) { 1337 ierr = MatSetSizes(A,M,N,M,N);CHKERRQ(ierr); 1338 } else { 1339 ierr = MatSetSizes(A,0,0,M,N);CHKERRQ(ierr); 1340 } 1341 /* This is just a temporary matrix, so explicitly using MATMPIAIJ is probably best */ 1342 ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr); 1343 ierr = MatMPIAIJSetPreallocation(A,0,NULL,0,NULL);CHKERRQ(ierr); 1344 ierr = MatSetOption(A,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr); 1345 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)A);CHKERRQ(ierr); 1346 1347 /* copy over the A part */ 1348 Aloc = (Mat_SeqAIJ*)aij->A->data; 1349 m = aij->A->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1350 row = mat->rmap->rstart; 1351 for (i=0; i<ai[m]; i++) aj[i] += mat->cmap->rstart; 1352 for (i=0; i<m; i++) { 1353 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],aj,a,INSERT_VALUES);CHKERRQ(ierr); 1354 row++; 1355 a += ai[i+1]-ai[i]; aj += ai[i+1]-ai[i]; 1356 } 1357 aj = Aloc->j; 1358 for (i=0; i<ai[m]; i++) aj[i] -= mat->cmap->rstart; 1359 1360 /* copy over the B part */ 1361 Aloc = (Mat_SeqAIJ*)aij->B->data; 1362 m = aij->B->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1363 row = mat->rmap->rstart; 1364 ierr = PetscMalloc1(ai[m]+1,&cols);CHKERRQ(ierr); 1365 ct = cols; 1366 for (i=0; i<ai[m]; i++) cols[i] = aij->garray[aj[i]]; 1367 for (i=0; i<m; i++) { 1368 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],cols,a,INSERT_VALUES);CHKERRQ(ierr); 1369 row++; 1370 a += ai[i+1]-ai[i]; cols += ai[i+1]-ai[i]; 1371 } 1372 ierr = PetscFree(ct);CHKERRQ(ierr); 1373 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1374 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1375 /* 1376 Everyone has to call to draw the matrix since the graphics waits are 1377 synchronized across all processors that share the PetscDraw object 1378 */ 1379 ierr = PetscViewerGetSubViewer(viewer,PETSC_COMM_SELF,&sviewer);CHKERRQ(ierr); 1380 if (!rank) { 1381 ierr = PetscObjectSetName((PetscObject)((Mat_MPIAIJ*)(A->data))->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1382 ierr = MatView_SeqAIJ(((Mat_MPIAIJ*)(A->data))->A,sviewer);CHKERRQ(ierr); 1383 } 1384 ierr = PetscViewerRestoreSubViewer(viewer,PETSC_COMM_SELF,&sviewer);CHKERRQ(ierr); 1385 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 1386 ierr = MatDestroy(&A);CHKERRQ(ierr); 1387 } 1388 PetscFunctionReturn(0); 1389 } 1390 1391 PetscErrorCode MatView_MPIAIJ(Mat mat,PetscViewer viewer) 1392 { 1393 PetscErrorCode ierr; 1394 PetscBool iascii,isdraw,issocket,isbinary; 1395 1396 PetscFunctionBegin; 1397 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1398 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1399 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1400 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERSOCKET,&issocket);CHKERRQ(ierr); 1401 if (iascii || isdraw || isbinary || issocket) { 1402 ierr = MatView_MPIAIJ_ASCIIorDraworSocket(mat,viewer);CHKERRQ(ierr); 1403 } 1404 PetscFunctionReturn(0); 1405 } 1406 1407 PetscErrorCode MatSOR_MPIAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1408 { 1409 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1410 PetscErrorCode ierr; 1411 Vec bb1 = 0; 1412 PetscBool hasop; 1413 1414 PetscFunctionBegin; 1415 if (flag == SOR_APPLY_UPPER) { 1416 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1417 PetscFunctionReturn(0); 1418 } 1419 1420 if (its > 1 || ~flag & SOR_ZERO_INITIAL_GUESS || flag & SOR_EISENSTAT) { 1421 ierr = VecDuplicate(bb,&bb1);CHKERRQ(ierr); 1422 } 1423 1424 if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP) { 1425 if (flag & SOR_ZERO_INITIAL_GUESS) { 1426 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1427 its--; 1428 } 1429 1430 while (its--) { 1431 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1432 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1433 1434 /* update rhs: bb1 = bb - B*x */ 1435 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1436 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1437 1438 /* local sweep */ 1439 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1440 } 1441 } else if (flag & SOR_LOCAL_FORWARD_SWEEP) { 1442 if (flag & SOR_ZERO_INITIAL_GUESS) { 1443 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1444 its--; 1445 } 1446 while (its--) { 1447 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1448 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1449 1450 /* update rhs: bb1 = bb - B*x */ 1451 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1452 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1453 1454 /* local sweep */ 1455 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_FORWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1456 } 1457 } else if (flag & SOR_LOCAL_BACKWARD_SWEEP) { 1458 if (flag & SOR_ZERO_INITIAL_GUESS) { 1459 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1460 its--; 1461 } 1462 while (its--) { 1463 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1464 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1465 1466 /* update rhs: bb1 = bb - B*x */ 1467 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1468 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1469 1470 /* local sweep */ 1471 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_BACKWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1472 } 1473 } else if (flag & SOR_EISENSTAT) { 1474 Vec xx1; 1475 1476 ierr = VecDuplicate(bb,&xx1);CHKERRQ(ierr); 1477 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_BACKWARD_SWEEP),fshift,lits,1,xx);CHKERRQ(ierr); 1478 1479 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1480 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1481 if (!mat->diag) { 1482 ierr = MatCreateVecs(matin,&mat->diag,NULL);CHKERRQ(ierr); 1483 ierr = MatGetDiagonal(matin,mat->diag);CHKERRQ(ierr); 1484 } 1485 ierr = MatHasOperation(matin,MATOP_MULT_DIAGONAL_BLOCK,&hasop);CHKERRQ(ierr); 1486 if (hasop) { 1487 ierr = MatMultDiagonalBlock(matin,xx,bb1);CHKERRQ(ierr); 1488 } else { 1489 ierr = VecPointwiseMult(bb1,mat->diag,xx);CHKERRQ(ierr); 1490 } 1491 ierr = VecAYPX(bb1,(omega-2.0)/omega,bb);CHKERRQ(ierr); 1492 1493 ierr = MatMultAdd(mat->B,mat->lvec,bb1,bb1);CHKERRQ(ierr); 1494 1495 /* local sweep */ 1496 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_FORWARD_SWEEP),fshift,lits,1,xx1);CHKERRQ(ierr); 1497 ierr = VecAXPY(xx,1.0,xx1);CHKERRQ(ierr); 1498 ierr = VecDestroy(&xx1);CHKERRQ(ierr); 1499 } else SETERRQ(PetscObjectComm((PetscObject)matin),PETSC_ERR_SUP,"Parallel SOR not supported"); 1500 1501 ierr = VecDestroy(&bb1);CHKERRQ(ierr); 1502 1503 matin->factorerrortype = mat->A->factorerrortype; 1504 PetscFunctionReturn(0); 1505 } 1506 1507 PetscErrorCode MatPermute_MPIAIJ(Mat A,IS rowp,IS colp,Mat *B) 1508 { 1509 Mat aA,aB,Aperm; 1510 const PetscInt *rwant,*cwant,*gcols,*ai,*bi,*aj,*bj; 1511 PetscScalar *aa,*ba; 1512 PetscInt i,j,m,n,ng,anz,bnz,*dnnz,*onnz,*tdnnz,*tonnz,*rdest,*cdest,*work,*gcdest; 1513 PetscSF rowsf,sf; 1514 IS parcolp = NULL; 1515 PetscBool done; 1516 PetscErrorCode ierr; 1517 1518 PetscFunctionBegin; 1519 ierr = MatGetLocalSize(A,&m,&n);CHKERRQ(ierr); 1520 ierr = ISGetIndices(rowp,&rwant);CHKERRQ(ierr); 1521 ierr = ISGetIndices(colp,&cwant);CHKERRQ(ierr); 1522 ierr = PetscMalloc3(PetscMax(m,n),&work,m,&rdest,n,&cdest);CHKERRQ(ierr); 1523 1524 /* Invert row permutation to find out where my rows should go */ 1525 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&rowsf);CHKERRQ(ierr); 1526 ierr = PetscSFSetGraphLayout(rowsf,A->rmap,A->rmap->n,NULL,PETSC_OWN_POINTER,rwant);CHKERRQ(ierr); 1527 ierr = PetscSFSetFromOptions(rowsf);CHKERRQ(ierr); 1528 for (i=0; i<m; i++) work[i] = A->rmap->rstart + i; 1529 ierr = PetscSFReduceBegin(rowsf,MPIU_INT,work,rdest,MPIU_REPLACE);CHKERRQ(ierr); 1530 ierr = PetscSFReduceEnd(rowsf,MPIU_INT,work,rdest,MPIU_REPLACE);CHKERRQ(ierr); 1531 1532 /* Invert column permutation to find out where my columns should go */ 1533 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1534 ierr = PetscSFSetGraphLayout(sf,A->cmap,A->cmap->n,NULL,PETSC_OWN_POINTER,cwant);CHKERRQ(ierr); 1535 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1536 for (i=0; i<n; i++) work[i] = A->cmap->rstart + i; 1537 ierr = PetscSFReduceBegin(sf,MPIU_INT,work,cdest,MPIU_REPLACE);CHKERRQ(ierr); 1538 ierr = PetscSFReduceEnd(sf,MPIU_INT,work,cdest,MPIU_REPLACE);CHKERRQ(ierr); 1539 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1540 1541 ierr = ISRestoreIndices(rowp,&rwant);CHKERRQ(ierr); 1542 ierr = ISRestoreIndices(colp,&cwant);CHKERRQ(ierr); 1543 ierr = MatMPIAIJGetSeqAIJ(A,&aA,&aB,&gcols);CHKERRQ(ierr); 1544 1545 /* Find out where my gcols should go */ 1546 ierr = MatGetSize(aB,NULL,&ng);CHKERRQ(ierr); 1547 ierr = PetscMalloc1(ng,&gcdest);CHKERRQ(ierr); 1548 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1549 ierr = PetscSFSetGraphLayout(sf,A->cmap,ng,NULL,PETSC_OWN_POINTER,gcols);CHKERRQ(ierr); 1550 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1551 ierr = PetscSFBcastBegin(sf,MPIU_INT,cdest,gcdest);CHKERRQ(ierr); 1552 ierr = PetscSFBcastEnd(sf,MPIU_INT,cdest,gcdest);CHKERRQ(ierr); 1553 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1554 1555 ierr = PetscCalloc4(m,&dnnz,m,&onnz,m,&tdnnz,m,&tonnz);CHKERRQ(ierr); 1556 ierr = MatGetRowIJ(aA,0,PETSC_FALSE,PETSC_FALSE,&anz,&ai,&aj,&done);CHKERRQ(ierr); 1557 ierr = MatGetRowIJ(aB,0,PETSC_FALSE,PETSC_FALSE,&bnz,&bi,&bj,&done);CHKERRQ(ierr); 1558 for (i=0; i<m; i++) { 1559 PetscInt row = rdest[i],rowner; 1560 ierr = PetscLayoutFindOwner(A->rmap,row,&rowner);CHKERRQ(ierr); 1561 for (j=ai[i]; j<ai[i+1]; j++) { 1562 PetscInt cowner,col = cdest[aj[j]]; 1563 ierr = PetscLayoutFindOwner(A->cmap,col,&cowner);CHKERRQ(ierr); /* Could build an index for the columns to eliminate this search */ 1564 if (rowner == cowner) dnnz[i]++; 1565 else onnz[i]++; 1566 } 1567 for (j=bi[i]; j<bi[i+1]; j++) { 1568 PetscInt cowner,col = gcdest[bj[j]]; 1569 ierr = PetscLayoutFindOwner(A->cmap,col,&cowner);CHKERRQ(ierr); 1570 if (rowner == cowner) dnnz[i]++; 1571 else onnz[i]++; 1572 } 1573 } 1574 ierr = PetscSFBcastBegin(rowsf,MPIU_INT,dnnz,tdnnz);CHKERRQ(ierr); 1575 ierr = PetscSFBcastEnd(rowsf,MPIU_INT,dnnz,tdnnz);CHKERRQ(ierr); 1576 ierr = PetscSFBcastBegin(rowsf,MPIU_INT,onnz,tonnz);CHKERRQ(ierr); 1577 ierr = PetscSFBcastEnd(rowsf,MPIU_INT,onnz,tonnz);CHKERRQ(ierr); 1578 ierr = PetscSFDestroy(&rowsf);CHKERRQ(ierr); 1579 1580 ierr = MatCreateAIJ(PetscObjectComm((PetscObject)A),A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N,0,tdnnz,0,tonnz,&Aperm);CHKERRQ(ierr); 1581 ierr = MatSeqAIJGetArray(aA,&aa);CHKERRQ(ierr); 1582 ierr = MatSeqAIJGetArray(aB,&ba);CHKERRQ(ierr); 1583 for (i=0; i<m; i++) { 1584 PetscInt *acols = dnnz,*bcols = onnz; /* Repurpose now-unneeded arrays */ 1585 PetscInt j0,rowlen; 1586 rowlen = ai[i+1] - ai[i]; 1587 for (j0=j=0; j<rowlen; j0=j) { /* rowlen could be larger than number of rows m, so sum in batches */ 1588 for ( ; j<PetscMin(rowlen,j0+m); j++) acols[j-j0] = cdest[aj[ai[i]+j]]; 1589 ierr = MatSetValues(Aperm,1,&rdest[i],j-j0,acols,aa+ai[i]+j0,INSERT_VALUES);CHKERRQ(ierr); 1590 } 1591 rowlen = bi[i+1] - bi[i]; 1592 for (j0=j=0; j<rowlen; j0=j) { 1593 for ( ; j<PetscMin(rowlen,j0+m); j++) bcols[j-j0] = gcdest[bj[bi[i]+j]]; 1594 ierr = MatSetValues(Aperm,1,&rdest[i],j-j0,bcols,ba+bi[i]+j0,INSERT_VALUES);CHKERRQ(ierr); 1595 } 1596 } 1597 ierr = MatAssemblyBegin(Aperm,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1598 ierr = MatAssemblyEnd(Aperm,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1599 ierr = MatRestoreRowIJ(aA,0,PETSC_FALSE,PETSC_FALSE,&anz,&ai,&aj,&done);CHKERRQ(ierr); 1600 ierr = MatRestoreRowIJ(aB,0,PETSC_FALSE,PETSC_FALSE,&bnz,&bi,&bj,&done);CHKERRQ(ierr); 1601 ierr = MatSeqAIJRestoreArray(aA,&aa);CHKERRQ(ierr); 1602 ierr = MatSeqAIJRestoreArray(aB,&ba);CHKERRQ(ierr); 1603 ierr = PetscFree4(dnnz,onnz,tdnnz,tonnz);CHKERRQ(ierr); 1604 ierr = PetscFree3(work,rdest,cdest);CHKERRQ(ierr); 1605 ierr = PetscFree(gcdest);CHKERRQ(ierr); 1606 if (parcolp) {ierr = ISDestroy(&colp);CHKERRQ(ierr);} 1607 *B = Aperm; 1608 PetscFunctionReturn(0); 1609 } 1610 1611 PetscErrorCode MatGetGhosts_MPIAIJ(Mat mat,PetscInt *nghosts,const PetscInt *ghosts[]) 1612 { 1613 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1614 PetscErrorCode ierr; 1615 1616 PetscFunctionBegin; 1617 ierr = MatGetSize(aij->B,NULL,nghosts);CHKERRQ(ierr); 1618 if (ghosts) *ghosts = aij->garray; 1619 PetscFunctionReturn(0); 1620 } 1621 1622 PetscErrorCode MatGetInfo_MPIAIJ(Mat matin,MatInfoType flag,MatInfo *info) 1623 { 1624 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1625 Mat A = mat->A,B = mat->B; 1626 PetscErrorCode ierr; 1627 PetscReal isend[5],irecv[5]; 1628 1629 PetscFunctionBegin; 1630 info->block_size = 1.0; 1631 ierr = MatGetInfo(A,MAT_LOCAL,info);CHKERRQ(ierr); 1632 1633 isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded; 1634 isend[3] = info->memory; isend[4] = info->mallocs; 1635 1636 ierr = MatGetInfo(B,MAT_LOCAL,info);CHKERRQ(ierr); 1637 1638 isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded; 1639 isend[3] += info->memory; isend[4] += info->mallocs; 1640 if (flag == MAT_LOCAL) { 1641 info->nz_used = isend[0]; 1642 info->nz_allocated = isend[1]; 1643 info->nz_unneeded = isend[2]; 1644 info->memory = isend[3]; 1645 info->mallocs = isend[4]; 1646 } else if (flag == MAT_GLOBAL_MAX) { 1647 ierr = MPIU_Allreduce(isend,irecv,5,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)matin));CHKERRQ(ierr); 1648 1649 info->nz_used = irecv[0]; 1650 info->nz_allocated = irecv[1]; 1651 info->nz_unneeded = irecv[2]; 1652 info->memory = irecv[3]; 1653 info->mallocs = irecv[4]; 1654 } else if (flag == MAT_GLOBAL_SUM) { 1655 ierr = MPIU_Allreduce(isend,irecv,5,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)matin));CHKERRQ(ierr); 1656 1657 info->nz_used = irecv[0]; 1658 info->nz_allocated = irecv[1]; 1659 info->nz_unneeded = irecv[2]; 1660 info->memory = irecv[3]; 1661 info->mallocs = irecv[4]; 1662 } 1663 info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */ 1664 info->fill_ratio_needed = 0; 1665 info->factor_mallocs = 0; 1666 PetscFunctionReturn(0); 1667 } 1668 1669 PetscErrorCode MatSetOption_MPIAIJ(Mat A,MatOption op,PetscBool flg) 1670 { 1671 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1672 PetscErrorCode ierr; 1673 1674 PetscFunctionBegin; 1675 switch (op) { 1676 case MAT_NEW_NONZERO_LOCATIONS: 1677 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1678 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1679 case MAT_KEEP_NONZERO_PATTERN: 1680 case MAT_NEW_NONZERO_LOCATION_ERR: 1681 case MAT_USE_INODES: 1682 case MAT_IGNORE_ZERO_ENTRIES: 1683 MatCheckPreallocated(A,1); 1684 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1685 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1686 break; 1687 case MAT_ROW_ORIENTED: 1688 MatCheckPreallocated(A,1); 1689 a->roworiented = flg; 1690 1691 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1692 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1693 break; 1694 case MAT_NEW_DIAGONALS: 1695 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1696 break; 1697 case MAT_IGNORE_OFF_PROC_ENTRIES: 1698 a->donotstash = flg; 1699 break; 1700 case MAT_SPD: 1701 A->spd_set = PETSC_TRUE; 1702 A->spd = flg; 1703 if (flg) { 1704 A->symmetric = PETSC_TRUE; 1705 A->structurally_symmetric = PETSC_TRUE; 1706 A->symmetric_set = PETSC_TRUE; 1707 A->structurally_symmetric_set = PETSC_TRUE; 1708 } 1709 break; 1710 case MAT_SYMMETRIC: 1711 MatCheckPreallocated(A,1); 1712 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1713 break; 1714 case MAT_STRUCTURALLY_SYMMETRIC: 1715 MatCheckPreallocated(A,1); 1716 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1717 break; 1718 case MAT_HERMITIAN: 1719 MatCheckPreallocated(A,1); 1720 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1721 break; 1722 case MAT_SYMMETRY_ETERNAL: 1723 MatCheckPreallocated(A,1); 1724 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1725 break; 1726 case MAT_SUBMAT_SINGLEIS: 1727 A->submat_singleis = flg; 1728 break; 1729 case MAT_STRUCTURE_ONLY: 1730 /* The option is handled directly by MatSetOption() */ 1731 break; 1732 default: 1733 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1734 } 1735 PetscFunctionReturn(0); 1736 } 1737 1738 PetscErrorCode MatGetRow_MPIAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1739 { 1740 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1741 PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p; 1742 PetscErrorCode ierr; 1743 PetscInt i,*cworkA,*cworkB,**pcA,**pcB,cstart = matin->cmap->rstart; 1744 PetscInt nztot,nzA,nzB,lrow,rstart = matin->rmap->rstart,rend = matin->rmap->rend; 1745 PetscInt *cmap,*idx_p; 1746 1747 PetscFunctionBegin; 1748 if (mat->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Already active"); 1749 mat->getrowactive = PETSC_TRUE; 1750 1751 if (!mat->rowvalues && (idx || v)) { 1752 /* 1753 allocate enough space to hold information from the longest row. 1754 */ 1755 Mat_SeqAIJ *Aa = (Mat_SeqAIJ*)mat->A->data,*Ba = (Mat_SeqAIJ*)mat->B->data; 1756 PetscInt max = 1,tmp; 1757 for (i=0; i<matin->rmap->n; i++) { 1758 tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; 1759 if (max < tmp) max = tmp; 1760 } 1761 ierr = PetscMalloc2(max,&mat->rowvalues,max,&mat->rowindices);CHKERRQ(ierr); 1762 } 1763 1764 if (row < rstart || row >= rend) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Only local rows"); 1765 lrow = row - rstart; 1766 1767 pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB; 1768 if (!v) {pvA = 0; pvB = 0;} 1769 if (!idx) {pcA = 0; if (!v) pcB = 0;} 1770 ierr = (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1771 ierr = (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1772 nztot = nzA + nzB; 1773 1774 cmap = mat->garray; 1775 if (v || idx) { 1776 if (nztot) { 1777 /* Sort by increasing column numbers, assuming A and B already sorted */ 1778 PetscInt imark = -1; 1779 if (v) { 1780 *v = v_p = mat->rowvalues; 1781 for (i=0; i<nzB; i++) { 1782 if (cmap[cworkB[i]] < cstart) v_p[i] = vworkB[i]; 1783 else break; 1784 } 1785 imark = i; 1786 for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i]; 1787 for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i]; 1788 } 1789 if (idx) { 1790 *idx = idx_p = mat->rowindices; 1791 if (imark > -1) { 1792 for (i=0; i<imark; i++) { 1793 idx_p[i] = cmap[cworkB[i]]; 1794 } 1795 } else { 1796 for (i=0; i<nzB; i++) { 1797 if (cmap[cworkB[i]] < cstart) idx_p[i] = cmap[cworkB[i]]; 1798 else break; 1799 } 1800 imark = i; 1801 } 1802 for (i=0; i<nzA; i++) idx_p[imark+i] = cstart + cworkA[i]; 1803 for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]]; 1804 } 1805 } else { 1806 if (idx) *idx = 0; 1807 if (v) *v = 0; 1808 } 1809 } 1810 *nz = nztot; 1811 ierr = (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1812 ierr = (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1813 PetscFunctionReturn(0); 1814 } 1815 1816 PetscErrorCode MatRestoreRow_MPIAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1817 { 1818 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1819 1820 PetscFunctionBegin; 1821 if (!aij->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"MatGetRow() must be called first"); 1822 aij->getrowactive = PETSC_FALSE; 1823 PetscFunctionReturn(0); 1824 } 1825 1826 PetscErrorCode MatNorm_MPIAIJ(Mat mat,NormType type,PetscReal *norm) 1827 { 1828 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1829 Mat_SeqAIJ *amat = (Mat_SeqAIJ*)aij->A->data,*bmat = (Mat_SeqAIJ*)aij->B->data; 1830 PetscErrorCode ierr; 1831 PetscInt i,j,cstart = mat->cmap->rstart; 1832 PetscReal sum = 0.0; 1833 MatScalar *v; 1834 1835 PetscFunctionBegin; 1836 if (aij->size == 1) { 1837 ierr = MatNorm(aij->A,type,norm);CHKERRQ(ierr); 1838 } else { 1839 if (type == NORM_FROBENIUS) { 1840 v = amat->a; 1841 for (i=0; i<amat->nz; i++) { 1842 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1843 } 1844 v = bmat->a; 1845 for (i=0; i<bmat->nz; i++) { 1846 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1847 } 1848 ierr = MPIU_Allreduce(&sum,norm,1,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1849 *norm = PetscSqrtReal(*norm); 1850 ierr = PetscLogFlops(2*amat->nz+2*bmat->nz);CHKERRQ(ierr); 1851 } else if (type == NORM_1) { /* max column norm */ 1852 PetscReal *tmp,*tmp2; 1853 PetscInt *jj,*garray = aij->garray; 1854 ierr = PetscCalloc1(mat->cmap->N+1,&tmp);CHKERRQ(ierr); 1855 ierr = PetscMalloc1(mat->cmap->N+1,&tmp2);CHKERRQ(ierr); 1856 *norm = 0.0; 1857 v = amat->a; jj = amat->j; 1858 for (j=0; j<amat->nz; j++) { 1859 tmp[cstart + *jj++] += PetscAbsScalar(*v); v++; 1860 } 1861 v = bmat->a; jj = bmat->j; 1862 for (j=0; j<bmat->nz; j++) { 1863 tmp[garray[*jj++]] += PetscAbsScalar(*v); v++; 1864 } 1865 ierr = MPIU_Allreduce(tmp,tmp2,mat->cmap->N,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1866 for (j=0; j<mat->cmap->N; j++) { 1867 if (tmp2[j] > *norm) *norm = tmp2[j]; 1868 } 1869 ierr = PetscFree(tmp);CHKERRQ(ierr); 1870 ierr = PetscFree(tmp2);CHKERRQ(ierr); 1871 ierr = PetscLogFlops(PetscMax(amat->nz+bmat->nz-1,0));CHKERRQ(ierr); 1872 } else if (type == NORM_INFINITY) { /* max row norm */ 1873 PetscReal ntemp = 0.0; 1874 for (j=0; j<aij->A->rmap->n; j++) { 1875 v = amat->a + amat->i[j]; 1876 sum = 0.0; 1877 for (i=0; i<amat->i[j+1]-amat->i[j]; i++) { 1878 sum += PetscAbsScalar(*v); v++; 1879 } 1880 v = bmat->a + bmat->i[j]; 1881 for (i=0; i<bmat->i[j+1]-bmat->i[j]; i++) { 1882 sum += PetscAbsScalar(*v); v++; 1883 } 1884 if (sum > ntemp) ntemp = sum; 1885 } 1886 ierr = MPIU_Allreduce(&ntemp,norm,1,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 1887 ierr = PetscLogFlops(PetscMax(amat->nz+bmat->nz-1,0));CHKERRQ(ierr); 1888 } else SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"No support for two norm"); 1889 } 1890 PetscFunctionReturn(0); 1891 } 1892 1893 PetscErrorCode MatTranspose_MPIAIJ(Mat A,MatReuse reuse,Mat *matout) 1894 { 1895 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1896 Mat_SeqAIJ *Aloc=(Mat_SeqAIJ*)a->A->data,*Bloc=(Mat_SeqAIJ*)a->B->data; 1897 PetscErrorCode ierr; 1898 PetscInt M = A->rmap->N,N = A->cmap->N,ma,na,mb,nb,*ai,*aj,*bi,*bj,row,*cols,*cols_tmp,i; 1899 PetscInt cstart = A->cmap->rstart,ncol; 1900 Mat B; 1901 MatScalar *array; 1902 1903 PetscFunctionBegin; 1904 if (reuse == MAT_INPLACE_MATRIX && M != N) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_SIZ,"Square matrix only for in-place"); 1905 1906 ma = A->rmap->n; na = A->cmap->n; mb = a->B->rmap->n; nb = a->B->cmap->n; 1907 ai = Aloc->i; aj = Aloc->j; 1908 bi = Bloc->i; bj = Bloc->j; 1909 if (reuse == MAT_INITIAL_MATRIX || *matout == A) { 1910 PetscInt *d_nnz,*g_nnz,*o_nnz; 1911 PetscSFNode *oloc; 1912 PETSC_UNUSED PetscSF sf; 1913 1914 ierr = PetscMalloc4(na,&d_nnz,na,&o_nnz,nb,&g_nnz,nb,&oloc);CHKERRQ(ierr); 1915 /* compute d_nnz for preallocation */ 1916 ierr = PetscMemzero(d_nnz,na*sizeof(PetscInt));CHKERRQ(ierr); 1917 for (i=0; i<ai[ma]; i++) { 1918 d_nnz[aj[i]]++; 1919 aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1920 } 1921 /* compute local off-diagonal contributions */ 1922 ierr = PetscMemzero(g_nnz,nb*sizeof(PetscInt));CHKERRQ(ierr); 1923 for (i=0; i<bi[ma]; i++) g_nnz[bj[i]]++; 1924 /* map those to global */ 1925 ierr = PetscSFCreate(PetscObjectComm((PetscObject)A),&sf);CHKERRQ(ierr); 1926 ierr = PetscSFSetGraphLayout(sf,A->cmap,nb,NULL,PETSC_USE_POINTER,a->garray);CHKERRQ(ierr); 1927 ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 1928 ierr = PetscMemzero(o_nnz,na*sizeof(PetscInt));CHKERRQ(ierr); 1929 ierr = PetscSFReduceBegin(sf,MPIU_INT,g_nnz,o_nnz,MPIU_SUM);CHKERRQ(ierr); 1930 ierr = PetscSFReduceEnd(sf,MPIU_INT,g_nnz,o_nnz,MPIU_SUM);CHKERRQ(ierr); 1931 ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 1932 1933 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 1934 ierr = MatSetSizes(B,A->cmap->n,A->rmap->n,N,M);CHKERRQ(ierr); 1935 ierr = MatSetBlockSizes(B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));CHKERRQ(ierr); 1936 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 1937 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 1938 ierr = PetscFree4(d_nnz,o_nnz,g_nnz,oloc);CHKERRQ(ierr); 1939 } else { 1940 B = *matout; 1941 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 1942 for (i=0; i<ai[ma]; i++) aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1943 } 1944 1945 /* copy over the A part */ 1946 array = Aloc->a; 1947 row = A->rmap->rstart; 1948 for (i=0; i<ma; i++) { 1949 ncol = ai[i+1]-ai[i]; 1950 ierr = MatSetValues(B,ncol,aj,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1951 row++; 1952 array += ncol; aj += ncol; 1953 } 1954 aj = Aloc->j; 1955 for (i=0; i<ai[ma]; i++) aj[i] -= cstart; /* resume local col index */ 1956 1957 /* copy over the B part */ 1958 ierr = PetscCalloc1(bi[mb],&cols);CHKERRQ(ierr); 1959 array = Bloc->a; 1960 row = A->rmap->rstart; 1961 for (i=0; i<bi[mb]; i++) cols[i] = a->garray[bj[i]]; 1962 cols_tmp = cols; 1963 for (i=0; i<mb; i++) { 1964 ncol = bi[i+1]-bi[i]; 1965 ierr = MatSetValues(B,ncol,cols_tmp,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1966 row++; 1967 array += ncol; cols_tmp += ncol; 1968 } 1969 ierr = PetscFree(cols);CHKERRQ(ierr); 1970 1971 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1972 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1973 if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) { 1974 *matout = B; 1975 } else { 1976 ierr = MatHeaderMerge(A,&B);CHKERRQ(ierr); 1977 } 1978 PetscFunctionReturn(0); 1979 } 1980 1981 PetscErrorCode MatDiagonalScale_MPIAIJ(Mat mat,Vec ll,Vec rr) 1982 { 1983 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1984 Mat a = aij->A,b = aij->B; 1985 PetscErrorCode ierr; 1986 PetscInt s1,s2,s3; 1987 1988 PetscFunctionBegin; 1989 ierr = MatGetLocalSize(mat,&s2,&s3);CHKERRQ(ierr); 1990 if (rr) { 1991 ierr = VecGetLocalSize(rr,&s1);CHKERRQ(ierr); 1992 if (s1!=s3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"right vector non-conforming local size"); 1993 /* Overlap communication with computation. */ 1994 ierr = VecScatterBegin(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1995 } 1996 if (ll) { 1997 ierr = VecGetLocalSize(ll,&s1);CHKERRQ(ierr); 1998 if (s1!=s2) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"left vector non-conforming local size"); 1999 ierr = (*b->ops->diagonalscale)(b,ll,0);CHKERRQ(ierr); 2000 } 2001 /* scale the diagonal block */ 2002 ierr = (*a->ops->diagonalscale)(a,ll,rr);CHKERRQ(ierr); 2003 2004 if (rr) { 2005 /* Do a scatter end and then right scale the off-diagonal block */ 2006 ierr = VecScatterEnd(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2007 ierr = (*b->ops->diagonalscale)(b,0,aij->lvec);CHKERRQ(ierr); 2008 } 2009 PetscFunctionReturn(0); 2010 } 2011 2012 PetscErrorCode MatSetUnfactored_MPIAIJ(Mat A) 2013 { 2014 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2015 PetscErrorCode ierr; 2016 2017 PetscFunctionBegin; 2018 ierr = MatSetUnfactored(a->A);CHKERRQ(ierr); 2019 PetscFunctionReturn(0); 2020 } 2021 2022 PetscErrorCode MatEqual_MPIAIJ(Mat A,Mat B,PetscBool *flag) 2023 { 2024 Mat_MPIAIJ *matB = (Mat_MPIAIJ*)B->data,*matA = (Mat_MPIAIJ*)A->data; 2025 Mat a,b,c,d; 2026 PetscBool flg; 2027 PetscErrorCode ierr; 2028 2029 PetscFunctionBegin; 2030 a = matA->A; b = matA->B; 2031 c = matB->A; d = matB->B; 2032 2033 ierr = MatEqual(a,c,&flg);CHKERRQ(ierr); 2034 if (flg) { 2035 ierr = MatEqual(b,d,&flg);CHKERRQ(ierr); 2036 } 2037 ierr = MPIU_Allreduce(&flg,flag,1,MPIU_BOOL,MPI_LAND,PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 2038 PetscFunctionReturn(0); 2039 } 2040 2041 PetscErrorCode MatCopy_MPIAIJ(Mat A,Mat B,MatStructure str) 2042 { 2043 PetscErrorCode ierr; 2044 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2045 Mat_MPIAIJ *b = (Mat_MPIAIJ*)B->data; 2046 2047 PetscFunctionBegin; 2048 /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */ 2049 if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) { 2050 /* because of the column compression in the off-processor part of the matrix a->B, 2051 the number of columns in a->B and b->B may be different, hence we cannot call 2052 the MatCopy() directly on the two parts. If need be, we can provide a more 2053 efficient copy than the MatCopy_Basic() by first uncompressing the a->B matrices 2054 then copying the submatrices */ 2055 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2056 } else { 2057 ierr = MatCopy(a->A,b->A,str);CHKERRQ(ierr); 2058 ierr = MatCopy(a->B,b->B,str);CHKERRQ(ierr); 2059 } 2060 PetscFunctionReturn(0); 2061 } 2062 2063 PetscErrorCode MatSetUp_MPIAIJ(Mat A) 2064 { 2065 PetscErrorCode ierr; 2066 2067 PetscFunctionBegin; 2068 ierr = MatMPIAIJSetPreallocation(A,PETSC_DEFAULT,0,PETSC_DEFAULT,0);CHKERRQ(ierr); 2069 PetscFunctionReturn(0); 2070 } 2071 2072 /* 2073 Computes the number of nonzeros per row needed for preallocation when X and Y 2074 have different nonzero structure. 2075 */ 2076 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) 2077 { 2078 PetscInt i,j,k,nzx,nzy; 2079 2080 PetscFunctionBegin; 2081 /* Set the number of nonzeros in the new matrix */ 2082 for (i=0; i<m; i++) { 2083 const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i]; 2084 nzx = xi[i+1] - xi[i]; 2085 nzy = yi[i+1] - yi[i]; 2086 nnz[i] = 0; 2087 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 2088 for (; k<nzy && yltog[yjj[k]]<xltog[xjj[j]]; k++) nnz[i]++; /* Catch up to X */ 2089 if (k<nzy && yltog[yjj[k]]==xltog[xjj[j]]) k++; /* Skip duplicate */ 2090 nnz[i]++; 2091 } 2092 for (; k<nzy; k++) nnz[i]++; 2093 } 2094 PetscFunctionReturn(0); 2095 } 2096 2097 /* This is the same as MatAXPYGetPreallocation_SeqAIJ, except that the local-to-global map is provided */ 2098 static PetscErrorCode MatAXPYGetPreallocation_MPIAIJ(Mat Y,const PetscInt *yltog,Mat X,const PetscInt *xltog,PetscInt *nnz) 2099 { 2100 PetscErrorCode ierr; 2101 PetscInt m = Y->rmap->N; 2102 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 2103 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 2104 2105 PetscFunctionBegin; 2106 ierr = MatAXPYGetPreallocation_MPIX_private(m,x->i,x->j,xltog,y->i,y->j,yltog,nnz);CHKERRQ(ierr); 2107 PetscFunctionReturn(0); 2108 } 2109 2110 PetscErrorCode MatAXPY_MPIAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 2111 { 2112 PetscErrorCode ierr; 2113 Mat_MPIAIJ *xx = (Mat_MPIAIJ*)X->data,*yy = (Mat_MPIAIJ*)Y->data; 2114 PetscBLASInt bnz,one=1; 2115 Mat_SeqAIJ *x,*y; 2116 2117 PetscFunctionBegin; 2118 if (str == SAME_NONZERO_PATTERN) { 2119 PetscScalar alpha = a; 2120 x = (Mat_SeqAIJ*)xx->A->data; 2121 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 2122 y = (Mat_SeqAIJ*)yy->A->data; 2123 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 2124 x = (Mat_SeqAIJ*)xx->B->data; 2125 y = (Mat_SeqAIJ*)yy->B->data; 2126 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 2127 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 2128 ierr = PetscObjectStateIncrease((PetscObject)Y);CHKERRQ(ierr); 2129 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 2130 ierr = MatAXPY_Basic(Y,a,X,str);CHKERRQ(ierr); 2131 } else { 2132 Mat B; 2133 PetscInt *nnz_d,*nnz_o; 2134 ierr = PetscMalloc1(yy->A->rmap->N,&nnz_d);CHKERRQ(ierr); 2135 ierr = PetscMalloc1(yy->B->rmap->N,&nnz_o);CHKERRQ(ierr); 2136 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 2137 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 2138 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 2139 ierr = MatSetBlockSizesFromMats(B,Y,Y);CHKERRQ(ierr); 2140 ierr = MatSetType(B,MATMPIAIJ);CHKERRQ(ierr); 2141 ierr = MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); 2142 ierr = MatAXPYGetPreallocation_MPIAIJ(yy->B,yy->garray,xx->B,xx->garray,nnz_o);CHKERRQ(ierr); 2143 ierr = MatMPIAIJSetPreallocation(B,0,nnz_d,0,nnz_o);CHKERRQ(ierr); 2144 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 2145 ierr = MatHeaderReplace(Y,&B);CHKERRQ(ierr); 2146 ierr = PetscFree(nnz_d);CHKERRQ(ierr); 2147 ierr = PetscFree(nnz_o);CHKERRQ(ierr); 2148 } 2149 PetscFunctionReturn(0); 2150 } 2151 2152 extern PetscErrorCode MatConjugate_SeqAIJ(Mat); 2153 2154 PetscErrorCode MatConjugate_MPIAIJ(Mat mat) 2155 { 2156 #if defined(PETSC_USE_COMPLEX) 2157 PetscErrorCode ierr; 2158 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2159 2160 PetscFunctionBegin; 2161 ierr = MatConjugate_SeqAIJ(aij->A);CHKERRQ(ierr); 2162 ierr = MatConjugate_SeqAIJ(aij->B);CHKERRQ(ierr); 2163 #else 2164 PetscFunctionBegin; 2165 #endif 2166 PetscFunctionReturn(0); 2167 } 2168 2169 PetscErrorCode MatRealPart_MPIAIJ(Mat A) 2170 { 2171 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2172 PetscErrorCode ierr; 2173 2174 PetscFunctionBegin; 2175 ierr = MatRealPart(a->A);CHKERRQ(ierr); 2176 ierr = MatRealPart(a->B);CHKERRQ(ierr); 2177 PetscFunctionReturn(0); 2178 } 2179 2180 PetscErrorCode MatImaginaryPart_MPIAIJ(Mat A) 2181 { 2182 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2183 PetscErrorCode ierr; 2184 2185 PetscFunctionBegin; 2186 ierr = MatImaginaryPart(a->A);CHKERRQ(ierr); 2187 ierr = MatImaginaryPart(a->B);CHKERRQ(ierr); 2188 PetscFunctionReturn(0); 2189 } 2190 2191 PetscErrorCode MatGetRowMaxAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2192 { 2193 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2194 PetscErrorCode ierr; 2195 PetscInt i,*idxb = 0; 2196 PetscScalar *va,*vb; 2197 Vec vtmp; 2198 2199 PetscFunctionBegin; 2200 ierr = MatGetRowMaxAbs(a->A,v,idx);CHKERRQ(ierr); 2201 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2202 if (idx) { 2203 for (i=0; i<A->rmap->n; i++) { 2204 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2205 } 2206 } 2207 2208 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2209 if (idx) { 2210 ierr = PetscMalloc1(A->rmap->n,&idxb);CHKERRQ(ierr); 2211 } 2212 ierr = MatGetRowMaxAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2213 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2214 2215 for (i=0; i<A->rmap->n; i++) { 2216 if (PetscAbsScalar(va[i]) < PetscAbsScalar(vb[i])) { 2217 va[i] = vb[i]; 2218 if (idx) idx[i] = a->garray[idxb[i]]; 2219 } 2220 } 2221 2222 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2223 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2224 ierr = PetscFree(idxb);CHKERRQ(ierr); 2225 ierr = VecDestroy(&vtmp);CHKERRQ(ierr); 2226 PetscFunctionReturn(0); 2227 } 2228 2229 PetscErrorCode MatGetRowMinAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2230 { 2231 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2232 PetscErrorCode ierr; 2233 PetscInt i,*idxb = 0; 2234 PetscScalar *va,*vb; 2235 Vec vtmp; 2236 2237 PetscFunctionBegin; 2238 ierr = MatGetRowMinAbs(a->A,v,idx);CHKERRQ(ierr); 2239 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2240 if (idx) { 2241 for (i=0; i<A->cmap->n; i++) { 2242 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2243 } 2244 } 2245 2246 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2247 if (idx) { 2248 ierr = PetscMalloc1(A->rmap->n,&idxb);CHKERRQ(ierr); 2249 } 2250 ierr = MatGetRowMinAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2251 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2252 2253 for (i=0; i<A->rmap->n; i++) { 2254 if (PetscAbsScalar(va[i]) > PetscAbsScalar(vb[i])) { 2255 va[i] = vb[i]; 2256 if (idx) idx[i] = a->garray[idxb[i]]; 2257 } 2258 } 2259 2260 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2261 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2262 ierr = PetscFree(idxb);CHKERRQ(ierr); 2263 ierr = VecDestroy(&vtmp);CHKERRQ(ierr); 2264 PetscFunctionReturn(0); 2265 } 2266 2267 PetscErrorCode MatGetRowMin_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2268 { 2269 Mat_MPIAIJ *mat = (Mat_MPIAIJ*) A->data; 2270 PetscInt n = A->rmap->n; 2271 PetscInt cstart = A->cmap->rstart; 2272 PetscInt *cmap = mat->garray; 2273 PetscInt *diagIdx, *offdiagIdx; 2274 Vec diagV, offdiagV; 2275 PetscScalar *a, *diagA, *offdiagA; 2276 PetscInt r; 2277 PetscErrorCode ierr; 2278 2279 PetscFunctionBegin; 2280 ierr = PetscMalloc2(n,&diagIdx,n,&offdiagIdx);CHKERRQ(ierr); 2281 ierr = VecCreateSeq(PetscObjectComm((PetscObject)A), n, &diagV);CHKERRQ(ierr); 2282 ierr = VecCreateSeq(PetscObjectComm((PetscObject)A), n, &offdiagV);CHKERRQ(ierr); 2283 ierr = MatGetRowMin(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2284 ierr = MatGetRowMin(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2285 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2286 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2287 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2288 for (r = 0; r < n; ++r) { 2289 if (PetscAbsScalar(diagA[r]) <= PetscAbsScalar(offdiagA[r])) { 2290 a[r] = diagA[r]; 2291 idx[r] = cstart + diagIdx[r]; 2292 } else { 2293 a[r] = offdiagA[r]; 2294 idx[r] = cmap[offdiagIdx[r]]; 2295 } 2296 } 2297 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2298 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2299 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2300 ierr = VecDestroy(&diagV);CHKERRQ(ierr); 2301 ierr = VecDestroy(&offdiagV);CHKERRQ(ierr); 2302 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2303 PetscFunctionReturn(0); 2304 } 2305 2306 PetscErrorCode MatGetRowMax_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2307 { 2308 Mat_MPIAIJ *mat = (Mat_MPIAIJ*) A->data; 2309 PetscInt n = A->rmap->n; 2310 PetscInt cstart = A->cmap->rstart; 2311 PetscInt *cmap = mat->garray; 2312 PetscInt *diagIdx, *offdiagIdx; 2313 Vec diagV, offdiagV; 2314 PetscScalar *a, *diagA, *offdiagA; 2315 PetscInt r; 2316 PetscErrorCode ierr; 2317 2318 PetscFunctionBegin; 2319 ierr = PetscMalloc2(n,&diagIdx,n,&offdiagIdx);CHKERRQ(ierr); 2320 ierr = VecCreateSeq(PETSC_COMM_SELF, n, &diagV);CHKERRQ(ierr); 2321 ierr = VecCreateSeq(PETSC_COMM_SELF, n, &offdiagV);CHKERRQ(ierr); 2322 ierr = MatGetRowMax(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2323 ierr = MatGetRowMax(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2324 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2325 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2326 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2327 for (r = 0; r < n; ++r) { 2328 if (PetscAbsScalar(diagA[r]) >= PetscAbsScalar(offdiagA[r])) { 2329 a[r] = diagA[r]; 2330 idx[r] = cstart + diagIdx[r]; 2331 } else { 2332 a[r] = offdiagA[r]; 2333 idx[r] = cmap[offdiagIdx[r]]; 2334 } 2335 } 2336 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2337 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2338 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2339 ierr = VecDestroy(&diagV);CHKERRQ(ierr); 2340 ierr = VecDestroy(&offdiagV);CHKERRQ(ierr); 2341 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2342 PetscFunctionReturn(0); 2343 } 2344 2345 PetscErrorCode MatGetSeqNonzeroStructure_MPIAIJ(Mat mat,Mat *newmat) 2346 { 2347 PetscErrorCode ierr; 2348 Mat *dummy; 2349 2350 PetscFunctionBegin; 2351 ierr = MatCreateSubMatrix_MPIAIJ_All(mat,MAT_DO_NOT_GET_VALUES,MAT_INITIAL_MATRIX,&dummy);CHKERRQ(ierr); 2352 *newmat = *dummy; 2353 ierr = PetscFree(dummy);CHKERRQ(ierr); 2354 PetscFunctionReturn(0); 2355 } 2356 2357 PetscErrorCode MatInvertBlockDiagonal_MPIAIJ(Mat A,const PetscScalar **values) 2358 { 2359 Mat_MPIAIJ *a = (Mat_MPIAIJ*) A->data; 2360 PetscErrorCode ierr; 2361 2362 PetscFunctionBegin; 2363 ierr = MatInvertBlockDiagonal(a->A,values);CHKERRQ(ierr); 2364 A->factorerrortype = a->A->factorerrortype; 2365 PetscFunctionReturn(0); 2366 } 2367 2368 static PetscErrorCode MatSetRandom_MPIAIJ(Mat x,PetscRandom rctx) 2369 { 2370 PetscErrorCode ierr; 2371 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)x->data; 2372 2373 PetscFunctionBegin; 2374 ierr = MatSetRandom(aij->A,rctx);CHKERRQ(ierr); 2375 ierr = MatSetRandom(aij->B,rctx);CHKERRQ(ierr); 2376 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2377 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2378 PetscFunctionReturn(0); 2379 } 2380 2381 PetscErrorCode MatMPIAIJSetUseScalableIncreaseOverlap_MPIAIJ(Mat A,PetscBool sc) 2382 { 2383 PetscFunctionBegin; 2384 if (sc) A->ops->increaseoverlap = MatIncreaseOverlap_MPIAIJ_Scalable; 2385 else A->ops->increaseoverlap = MatIncreaseOverlap_MPIAIJ; 2386 PetscFunctionReturn(0); 2387 } 2388 2389 /*@ 2390 MatMPIAIJSetUseScalableIncreaseOverlap - Determine if the matrix uses a scalable algorithm to compute the overlap 2391 2392 Collective on Mat 2393 2394 Input Parameters: 2395 + A - the matrix 2396 - sc - PETSC_TRUE indicates use the scalable algorithm (default is not to use the scalable algorithm) 2397 2398 Level: advanced 2399 2400 @*/ 2401 PetscErrorCode MatMPIAIJSetUseScalableIncreaseOverlap(Mat A,PetscBool sc) 2402 { 2403 PetscErrorCode ierr; 2404 2405 PetscFunctionBegin; 2406 ierr = PetscTryMethod(A,"MatMPIAIJSetUseScalableIncreaseOverlap_C",(Mat,PetscBool),(A,sc));CHKERRQ(ierr); 2407 PetscFunctionReturn(0); 2408 } 2409 2410 PetscErrorCode MatSetFromOptions_MPIAIJ(PetscOptionItems *PetscOptionsObject,Mat A) 2411 { 2412 PetscErrorCode ierr; 2413 PetscBool sc = PETSC_FALSE,flg; 2414 2415 PetscFunctionBegin; 2416 ierr = PetscOptionsHead(PetscOptionsObject,"MPIAIJ options");CHKERRQ(ierr); 2417 ierr = PetscObjectOptionsBegin((PetscObject)A); 2418 if (A->ops->increaseoverlap == MatIncreaseOverlap_MPIAIJ_Scalable) sc = PETSC_TRUE; 2419 ierr = PetscOptionsBool("-mat_increase_overlap_scalable","Use a scalable algorithm to compute the overlap","MatIncreaseOverlap",sc,&sc,&flg);CHKERRQ(ierr); 2420 if (flg) { 2421 ierr = MatMPIAIJSetUseScalableIncreaseOverlap(A,sc);CHKERRQ(ierr); 2422 } 2423 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2424 PetscFunctionReturn(0); 2425 } 2426 2427 PetscErrorCode MatShift_MPIAIJ(Mat Y,PetscScalar a) 2428 { 2429 PetscErrorCode ierr; 2430 Mat_MPIAIJ *maij = (Mat_MPIAIJ*)Y->data; 2431 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)maij->A->data; 2432 2433 PetscFunctionBegin; 2434 if (!Y->preallocated) { 2435 ierr = MatMPIAIJSetPreallocation(Y,1,NULL,0,NULL);CHKERRQ(ierr); 2436 } else if (!aij->nz) { 2437 PetscInt nonew = aij->nonew; 2438 ierr = MatSeqAIJSetPreallocation(maij->A,1,NULL);CHKERRQ(ierr); 2439 aij->nonew = nonew; 2440 } 2441 ierr = MatShift_Basic(Y,a);CHKERRQ(ierr); 2442 PetscFunctionReturn(0); 2443 } 2444 2445 PetscErrorCode MatMissingDiagonal_MPIAIJ(Mat A,PetscBool *missing,PetscInt *d) 2446 { 2447 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2448 PetscErrorCode ierr; 2449 2450 PetscFunctionBegin; 2451 if (A->rmap->n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only works for square matrices"); 2452 ierr = MatMissingDiagonal(a->A,missing,d);CHKERRQ(ierr); 2453 if (d) { 2454 PetscInt rstart; 2455 ierr = MatGetOwnershipRange(A,&rstart,NULL);CHKERRQ(ierr); 2456 *d += rstart; 2457 2458 } 2459 PetscFunctionReturn(0); 2460 } 2461 2462 2463 /* -------------------------------------------------------------------*/ 2464 static struct _MatOps MatOps_Values = {MatSetValues_MPIAIJ, 2465 MatGetRow_MPIAIJ, 2466 MatRestoreRow_MPIAIJ, 2467 MatMult_MPIAIJ, 2468 /* 4*/ MatMultAdd_MPIAIJ, 2469 MatMultTranspose_MPIAIJ, 2470 MatMultTransposeAdd_MPIAIJ, 2471 0, 2472 0, 2473 0, 2474 /*10*/ 0, 2475 0, 2476 0, 2477 MatSOR_MPIAIJ, 2478 MatTranspose_MPIAIJ, 2479 /*15*/ MatGetInfo_MPIAIJ, 2480 MatEqual_MPIAIJ, 2481 MatGetDiagonal_MPIAIJ, 2482 MatDiagonalScale_MPIAIJ, 2483 MatNorm_MPIAIJ, 2484 /*20*/ MatAssemblyBegin_MPIAIJ, 2485 MatAssemblyEnd_MPIAIJ, 2486 MatSetOption_MPIAIJ, 2487 MatZeroEntries_MPIAIJ, 2488 /*24*/ MatZeroRows_MPIAIJ, 2489 0, 2490 0, 2491 0, 2492 0, 2493 /*29*/ MatSetUp_MPIAIJ, 2494 0, 2495 0, 2496 MatGetDiagonalBlock_MPIAIJ, 2497 0, 2498 /*34*/ MatDuplicate_MPIAIJ, 2499 0, 2500 0, 2501 0, 2502 0, 2503 /*39*/ MatAXPY_MPIAIJ, 2504 MatCreateSubMatrices_MPIAIJ, 2505 MatIncreaseOverlap_MPIAIJ, 2506 MatGetValues_MPIAIJ, 2507 MatCopy_MPIAIJ, 2508 /*44*/ MatGetRowMax_MPIAIJ, 2509 MatScale_MPIAIJ, 2510 MatShift_MPIAIJ, 2511 MatDiagonalSet_MPIAIJ, 2512 MatZeroRowsColumns_MPIAIJ, 2513 /*49*/ MatSetRandom_MPIAIJ, 2514 0, 2515 0, 2516 0, 2517 0, 2518 /*54*/ MatFDColoringCreate_MPIXAIJ, 2519 0, 2520 MatSetUnfactored_MPIAIJ, 2521 MatPermute_MPIAIJ, 2522 0, 2523 /*59*/ MatCreateSubMatrix_MPIAIJ, 2524 MatDestroy_MPIAIJ, 2525 MatView_MPIAIJ, 2526 0, 2527 MatMatMatMult_MPIAIJ_MPIAIJ_MPIAIJ, 2528 /*64*/ MatMatMatMultSymbolic_MPIAIJ_MPIAIJ_MPIAIJ, 2529 MatMatMatMultNumeric_MPIAIJ_MPIAIJ_MPIAIJ, 2530 0, 2531 0, 2532 0, 2533 /*69*/ MatGetRowMaxAbs_MPIAIJ, 2534 MatGetRowMinAbs_MPIAIJ, 2535 0, 2536 0, 2537 0, 2538 0, 2539 /*75*/ MatFDColoringApply_AIJ, 2540 MatSetFromOptions_MPIAIJ, 2541 0, 2542 0, 2543 MatFindZeroDiagonals_MPIAIJ, 2544 /*80*/ 0, 2545 0, 2546 0, 2547 /*83*/ MatLoad_MPIAIJ, 2548 0, 2549 0, 2550 0, 2551 0, 2552 0, 2553 /*89*/ MatMatMult_MPIAIJ_MPIAIJ, 2554 MatMatMultSymbolic_MPIAIJ_MPIAIJ, 2555 MatMatMultNumeric_MPIAIJ_MPIAIJ, 2556 MatPtAP_MPIAIJ_MPIAIJ, 2557 MatPtAPSymbolic_MPIAIJ_MPIAIJ, 2558 /*94*/ MatPtAPNumeric_MPIAIJ_MPIAIJ, 2559 0, 2560 0, 2561 0, 2562 0, 2563 /*99*/ 0, 2564 0, 2565 0, 2566 MatConjugate_MPIAIJ, 2567 0, 2568 /*104*/MatSetValuesRow_MPIAIJ, 2569 MatRealPart_MPIAIJ, 2570 MatImaginaryPart_MPIAIJ, 2571 0, 2572 0, 2573 /*109*/0, 2574 0, 2575 MatGetRowMin_MPIAIJ, 2576 0, 2577 MatMissingDiagonal_MPIAIJ, 2578 /*114*/MatGetSeqNonzeroStructure_MPIAIJ, 2579 0, 2580 MatGetGhosts_MPIAIJ, 2581 0, 2582 0, 2583 /*119*/0, 2584 0, 2585 0, 2586 0, 2587 MatGetMultiProcBlock_MPIAIJ, 2588 /*124*/MatFindNonzeroRows_MPIAIJ, 2589 MatGetColumnNorms_MPIAIJ, 2590 MatInvertBlockDiagonal_MPIAIJ, 2591 0, 2592 MatCreateSubMatricesMPI_MPIAIJ, 2593 /*129*/0, 2594 MatTransposeMatMult_MPIAIJ_MPIAIJ, 2595 MatTransposeMatMultSymbolic_MPIAIJ_MPIAIJ, 2596 MatTransposeMatMultNumeric_MPIAIJ_MPIAIJ, 2597 0, 2598 /*134*/0, 2599 0, 2600 0, 2601 0, 2602 0, 2603 /*139*/MatSetBlockSizes_MPIAIJ, 2604 0, 2605 0, 2606 MatFDColoringSetUp_MPIXAIJ, 2607 MatFindOffBlockDiagonalEntries_MPIAIJ, 2608 /*144*/MatCreateMPIMatConcatenateSeqMat_MPIAIJ 2609 }; 2610 2611 /* ----------------------------------------------------------------------------------------*/ 2612 2613 PetscErrorCode MatStoreValues_MPIAIJ(Mat mat) 2614 { 2615 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2616 PetscErrorCode ierr; 2617 2618 PetscFunctionBegin; 2619 ierr = MatStoreValues(aij->A);CHKERRQ(ierr); 2620 ierr = MatStoreValues(aij->B);CHKERRQ(ierr); 2621 PetscFunctionReturn(0); 2622 } 2623 2624 PetscErrorCode MatRetrieveValues_MPIAIJ(Mat mat) 2625 { 2626 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2627 PetscErrorCode ierr; 2628 2629 PetscFunctionBegin; 2630 ierr = MatRetrieveValues(aij->A);CHKERRQ(ierr); 2631 ierr = MatRetrieveValues(aij->B);CHKERRQ(ierr); 2632 PetscFunctionReturn(0); 2633 } 2634 2635 PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJ(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 2636 { 2637 Mat_MPIAIJ *b; 2638 PetscErrorCode ierr; 2639 2640 PetscFunctionBegin; 2641 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 2642 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 2643 b = (Mat_MPIAIJ*)B->data; 2644 2645 #if defined(PETSC_USE_CTABLE) 2646 ierr = PetscTableDestroy(&b->colmap);CHKERRQ(ierr); 2647 #else 2648 ierr = PetscFree(b->colmap);CHKERRQ(ierr); 2649 #endif 2650 ierr = PetscFree(b->garray);CHKERRQ(ierr); 2651 ierr = VecDestroy(&b->lvec);CHKERRQ(ierr); 2652 ierr = VecScatterDestroy(&b->Mvctx);CHKERRQ(ierr); 2653 2654 /* Because the B will have been resized we simply destroy it and create a new one each time */ 2655 ierr = MatDestroy(&b->B);CHKERRQ(ierr); 2656 ierr = MatCreate(PETSC_COMM_SELF,&b->B);CHKERRQ(ierr); 2657 ierr = MatSetSizes(b->B,B->rmap->n,B->cmap->N,B->rmap->n,B->cmap->N);CHKERRQ(ierr); 2658 ierr = MatSetBlockSizesFromMats(b->B,B,B);CHKERRQ(ierr); 2659 ierr = MatSetType(b->B,MATSEQAIJ);CHKERRQ(ierr); 2660 ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)b->B);CHKERRQ(ierr); 2661 2662 if (!B->preallocated) { 2663 ierr = MatCreate(PETSC_COMM_SELF,&b->A);CHKERRQ(ierr); 2664 ierr = MatSetSizes(b->A,B->rmap->n,B->cmap->n,B->rmap->n,B->cmap->n);CHKERRQ(ierr); 2665 ierr = MatSetBlockSizesFromMats(b->A,B,B);CHKERRQ(ierr); 2666 ierr = MatSetType(b->A,MATSEQAIJ);CHKERRQ(ierr); 2667 ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)b->A);CHKERRQ(ierr); 2668 } 2669 2670 ierr = MatSeqAIJSetPreallocation(b->A,d_nz,d_nnz);CHKERRQ(ierr); 2671 ierr = MatSeqAIJSetPreallocation(b->B,o_nz,o_nnz);CHKERRQ(ierr); 2672 B->preallocated = PETSC_TRUE; 2673 B->was_assembled = PETSC_FALSE; 2674 B->assembled = PETSC_FALSE;; 2675 PetscFunctionReturn(0); 2676 } 2677 2678 PetscErrorCode MatDuplicate_MPIAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat) 2679 { 2680 Mat mat; 2681 Mat_MPIAIJ *a,*oldmat = (Mat_MPIAIJ*)matin->data; 2682 PetscErrorCode ierr; 2683 2684 PetscFunctionBegin; 2685 *newmat = 0; 2686 ierr = MatCreate(PetscObjectComm((PetscObject)matin),&mat);CHKERRQ(ierr); 2687 ierr = MatSetSizes(mat,matin->rmap->n,matin->cmap->n,matin->rmap->N,matin->cmap->N);CHKERRQ(ierr); 2688 ierr = MatSetBlockSizesFromMats(mat,matin,matin);CHKERRQ(ierr); 2689 ierr = MatSetType(mat,((PetscObject)matin)->type_name);CHKERRQ(ierr); 2690 ierr = PetscMemcpy(mat->ops,matin->ops,sizeof(struct _MatOps));CHKERRQ(ierr); 2691 a = (Mat_MPIAIJ*)mat->data; 2692 2693 mat->factortype = matin->factortype; 2694 mat->assembled = PETSC_TRUE; 2695 mat->insertmode = NOT_SET_VALUES; 2696 mat->preallocated = PETSC_TRUE; 2697 2698 a->size = oldmat->size; 2699 a->rank = oldmat->rank; 2700 a->donotstash = oldmat->donotstash; 2701 a->roworiented = oldmat->roworiented; 2702 a->rowindices = 0; 2703 a->rowvalues = 0; 2704 a->getrowactive = PETSC_FALSE; 2705 2706 ierr = PetscLayoutReference(matin->rmap,&mat->rmap);CHKERRQ(ierr); 2707 ierr = PetscLayoutReference(matin->cmap,&mat->cmap);CHKERRQ(ierr); 2708 2709 if (oldmat->colmap) { 2710 #if defined(PETSC_USE_CTABLE) 2711 ierr = PetscTableCreateCopy(oldmat->colmap,&a->colmap);CHKERRQ(ierr); 2712 #else 2713 ierr = PetscMalloc1(mat->cmap->N,&a->colmap);CHKERRQ(ierr); 2714 ierr = PetscLogObjectMemory((PetscObject)mat,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 2715 ierr = PetscMemcpy(a->colmap,oldmat->colmap,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 2716 #endif 2717 } else a->colmap = 0; 2718 if (oldmat->garray) { 2719 PetscInt len; 2720 len = oldmat->B->cmap->n; 2721 ierr = PetscMalloc1(len+1,&a->garray);CHKERRQ(ierr); 2722 ierr = PetscLogObjectMemory((PetscObject)mat,len*sizeof(PetscInt));CHKERRQ(ierr); 2723 if (len) { ierr = PetscMemcpy(a->garray,oldmat->garray,len*sizeof(PetscInt));CHKERRQ(ierr); } 2724 } else a->garray = 0; 2725 2726 ierr = VecDuplicate(oldmat->lvec,&a->lvec);CHKERRQ(ierr); 2727 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->lvec);CHKERRQ(ierr); 2728 ierr = VecScatterCopy(oldmat->Mvctx,&a->Mvctx);CHKERRQ(ierr); 2729 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->Mvctx);CHKERRQ(ierr); 2730 ierr = MatDuplicate(oldmat->A,cpvalues,&a->A);CHKERRQ(ierr); 2731 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->A);CHKERRQ(ierr); 2732 ierr = MatDuplicate(oldmat->B,cpvalues,&a->B);CHKERRQ(ierr); 2733 ierr = PetscLogObjectParent((PetscObject)mat,(PetscObject)a->B);CHKERRQ(ierr); 2734 ierr = PetscFunctionListDuplicate(((PetscObject)matin)->qlist,&((PetscObject)mat)->qlist);CHKERRQ(ierr); 2735 *newmat = mat; 2736 PetscFunctionReturn(0); 2737 } 2738 2739 PetscErrorCode MatLoad_MPIAIJ(Mat newMat, PetscViewer viewer) 2740 { 2741 PetscScalar *vals,*svals; 2742 MPI_Comm comm; 2743 PetscErrorCode ierr; 2744 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 2745 PetscInt i,nz,j,rstart,rend,mmax,maxnz = 0; 2746 PetscInt header[4],*rowlengths = 0,M,N,m,*cols; 2747 PetscInt *ourlens = NULL,*procsnz = NULL,*offlens = NULL,jj,*mycols,*smycols; 2748 PetscInt cend,cstart,n,*rowners; 2749 int fd; 2750 PetscInt bs = newMat->rmap->bs; 2751 2752 PetscFunctionBegin; 2753 /* force binary viewer to load .info file if it has not yet done so */ 2754 ierr = PetscViewerSetUp(viewer);CHKERRQ(ierr); 2755 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 2756 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 2757 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 2758 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 2759 if (!rank) { 2760 ierr = PetscBinaryRead(fd,(char*)header,4,PETSC_INT);CHKERRQ(ierr); 2761 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object"); 2762 if (header[3] < 0) SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as MATMPIAIJ"); 2763 } 2764 2765 ierr = PetscOptionsBegin(comm,NULL,"Options for loading MATMPIAIJ matrix","Mat");CHKERRQ(ierr); 2766 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 2767 ierr = PetscOptionsEnd();CHKERRQ(ierr); 2768 if (bs < 0) bs = 1; 2769 2770 ierr = MPI_Bcast(header+1,3,MPIU_INT,0,comm);CHKERRQ(ierr); 2771 M = header[1]; N = header[2]; 2772 2773 /* If global sizes are set, check if they are consistent with that given in the file */ 2774 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); 2775 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); 2776 2777 /* determine ownership of all (block) rows */ 2778 if (M%bs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Inconsistent # of rows (%d) and block size (%d)",M,bs); 2779 if (newMat->rmap->n < 0) m = bs*((M/bs)/size + (((M/bs) % size) > rank)); /* PETSC_DECIDE */ 2780 else m = newMat->rmap->n; /* Set by user */ 2781 2782 ierr = PetscMalloc1(size+1,&rowners);CHKERRQ(ierr); 2783 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 2784 2785 /* First process needs enough room for process with most rows */ 2786 if (!rank) { 2787 mmax = rowners[1]; 2788 for (i=2; i<=size; i++) { 2789 mmax = PetscMax(mmax, rowners[i]); 2790 } 2791 } else mmax = -1; /* unused, but compilers complain */ 2792 2793 rowners[0] = 0; 2794 for (i=2; i<=size; i++) { 2795 rowners[i] += rowners[i-1]; 2796 } 2797 rstart = rowners[rank]; 2798 rend = rowners[rank+1]; 2799 2800 /* distribute row lengths to all processors */ 2801 ierr = PetscMalloc2(m,&ourlens,m,&offlens);CHKERRQ(ierr); 2802 if (!rank) { 2803 ierr = PetscBinaryRead(fd,ourlens,m,PETSC_INT);CHKERRQ(ierr); 2804 ierr = PetscMalloc1(mmax,&rowlengths);CHKERRQ(ierr); 2805 ierr = PetscCalloc1(size,&procsnz);CHKERRQ(ierr); 2806 for (j=0; j<m; j++) { 2807 procsnz[0] += ourlens[j]; 2808 } 2809 for (i=1; i<size; i++) { 2810 ierr = PetscBinaryRead(fd,rowlengths,rowners[i+1]-rowners[i],PETSC_INT);CHKERRQ(ierr); 2811 /* calculate the number of nonzeros on each processor */ 2812 for (j=0; j<rowners[i+1]-rowners[i]; j++) { 2813 procsnz[i] += rowlengths[j]; 2814 } 2815 ierr = MPIULong_Send(rowlengths,rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2816 } 2817 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 2818 } else { 2819 ierr = MPIULong_Recv(ourlens,m,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 2820 } 2821 2822 if (!rank) { 2823 /* determine max buffer needed and allocate it */ 2824 maxnz = 0; 2825 for (i=0; i<size; i++) { 2826 maxnz = PetscMax(maxnz,procsnz[i]); 2827 } 2828 ierr = PetscMalloc1(maxnz,&cols);CHKERRQ(ierr); 2829 2830 /* read in my part of the matrix column indices */ 2831 nz = procsnz[0]; 2832 ierr = PetscMalloc1(nz,&mycols);CHKERRQ(ierr); 2833 ierr = PetscBinaryRead(fd,mycols,nz,PETSC_INT);CHKERRQ(ierr); 2834 2835 /* read in every one elses and ship off */ 2836 for (i=1; i<size; i++) { 2837 nz = procsnz[i]; 2838 ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr); 2839 ierr = MPIULong_Send(cols,nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2840 } 2841 ierr = PetscFree(cols);CHKERRQ(ierr); 2842 } else { 2843 /* determine buffer space needed for message */ 2844 nz = 0; 2845 for (i=0; i<m; i++) { 2846 nz += ourlens[i]; 2847 } 2848 ierr = PetscMalloc1(nz,&mycols);CHKERRQ(ierr); 2849 2850 /* receive message of column indices*/ 2851 ierr = MPIULong_Recv(mycols,nz,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 2852 } 2853 2854 /* determine column ownership if matrix is not square */ 2855 if (N != M) { 2856 if (newMat->cmap->n < 0) n = N/size + ((N % size) > rank); 2857 else n = newMat->cmap->n; 2858 ierr = MPI_Scan(&n,&cend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 2859 cstart = cend - n; 2860 } else { 2861 cstart = rstart; 2862 cend = rend; 2863 n = cend - cstart; 2864 } 2865 2866 /* loop over local rows, determining number of off diagonal entries */ 2867 ierr = PetscMemzero(offlens,m*sizeof(PetscInt));CHKERRQ(ierr); 2868 jj = 0; 2869 for (i=0; i<m; i++) { 2870 for (j=0; j<ourlens[i]; j++) { 2871 if (mycols[jj] < cstart || mycols[jj] >= cend) offlens[i]++; 2872 jj++; 2873 } 2874 } 2875 2876 for (i=0; i<m; i++) { 2877 ourlens[i] -= offlens[i]; 2878 } 2879 ierr = MatSetSizes(newMat,m,n,M,N);CHKERRQ(ierr); 2880 2881 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 2882 2883 ierr = MatMPIAIJSetPreallocation(newMat,0,ourlens,0,offlens);CHKERRQ(ierr); 2884 2885 for (i=0; i<m; i++) { 2886 ourlens[i] += offlens[i]; 2887 } 2888 2889 if (!rank) { 2890 ierr = PetscMalloc1(maxnz+1,&vals);CHKERRQ(ierr); 2891 2892 /* read in my part of the matrix numerical values */ 2893 nz = procsnz[0]; 2894 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2895 2896 /* insert into matrix */ 2897 jj = rstart; 2898 smycols = mycols; 2899 svals = vals; 2900 for (i=0; i<m; i++) { 2901 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2902 smycols += ourlens[i]; 2903 svals += ourlens[i]; 2904 jj++; 2905 } 2906 2907 /* read in other processors and ship out */ 2908 for (i=1; i<size; i++) { 2909 nz = procsnz[i]; 2910 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2911 ierr = MPIULong_Send(vals,nz,MPIU_SCALAR,i,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 2912 } 2913 ierr = PetscFree(procsnz);CHKERRQ(ierr); 2914 } else { 2915 /* receive numeric values */ 2916 ierr = PetscMalloc1(nz+1,&vals);CHKERRQ(ierr); 2917 2918 /* receive message of values*/ 2919 ierr = MPIULong_Recv(vals,nz,MPIU_SCALAR,0,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 2920 2921 /* insert into matrix */ 2922 jj = rstart; 2923 smycols = mycols; 2924 svals = vals; 2925 for (i=0; i<m; i++) { 2926 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2927 smycols += ourlens[i]; 2928 svals += ourlens[i]; 2929 jj++; 2930 } 2931 } 2932 ierr = PetscFree2(ourlens,offlens);CHKERRQ(ierr); 2933 ierr = PetscFree(vals);CHKERRQ(ierr); 2934 ierr = PetscFree(mycols);CHKERRQ(ierr); 2935 ierr = PetscFree(rowners);CHKERRQ(ierr); 2936 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2937 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2938 PetscFunctionReturn(0); 2939 } 2940 2941 /* Not scalable because of ISAllGather() unless getting all columns. */ 2942 PetscErrorCode ISGetSeqIS_Private(Mat mat,IS iscol,IS *isseq) 2943 { 2944 PetscErrorCode ierr; 2945 IS iscol_local; 2946 PetscBool isstride; 2947 PetscMPIInt lisstride=0,gisstride; 2948 2949 PetscFunctionBegin; 2950 /* check if we are grabbing all columns*/ 2951 ierr = PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&isstride);CHKERRQ(ierr); 2952 2953 if (isstride) { 2954 PetscInt start,len,mstart,mlen; 2955 ierr = ISStrideGetInfo(iscol,&start,NULL);CHKERRQ(ierr); 2956 ierr = ISGetLocalSize(iscol,&len);CHKERRQ(ierr); 2957 ierr = MatGetOwnershipRangeColumn(mat,&mstart,&mlen);CHKERRQ(ierr); 2958 if (mstart == start && mlen-mstart == len) lisstride = 1; 2959 } 2960 2961 ierr = MPIU_Allreduce(&lisstride,&gisstride,1,MPI_INT,MPI_MIN,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 2962 if (gisstride) { 2963 PetscInt N; 2964 ierr = MatGetSize(mat,NULL,&N);CHKERRQ(ierr); 2965 ierr = ISCreateStride(PetscObjectComm((PetscObject)mat),N,0,1,&iscol_local);CHKERRQ(ierr); 2966 ierr = ISSetIdentity(iscol_local);CHKERRQ(ierr); 2967 ierr = PetscInfo(mat,"Optimizing for obtaining all columns of the matrix; skipping ISAllGather()\n");CHKERRQ(ierr); 2968 } else { 2969 PetscInt cbs; 2970 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2971 ierr = ISAllGather(iscol,&iscol_local);CHKERRQ(ierr); 2972 ierr = ISSetBlockSize(iscol_local,cbs);CHKERRQ(ierr); 2973 } 2974 2975 *isseq = iscol_local; 2976 PetscFunctionReturn(0); 2977 } 2978 2979 /* 2980 Used by MatCreateSubMatrix_MPIAIJ_SameRowColDist() to avoid ISAllGather() and global size of iscol_local 2981 (see MatCreateSubMatrix_MPIAIJ_nonscalable) 2982 2983 Input Parameters: 2984 mat - matrix 2985 isrow - parallel row index set; its local indices are a subset of local columns of mat, 2986 i.e., mat->rstart <= isrow[i] < mat->rend 2987 iscol - parallel column index set; its local indices are a subset of local columns of mat, 2988 i.e., mat->cstart <= iscol[i] < mat->cend 2989 Output Parameter: 2990 isrow_d,iscol_d - sequential row and column index sets for retrieving mat->A 2991 iscol_o - sequential column index set for retrieving mat->B 2992 garray - column map; garray[i] indicates global location of iscol_o[i] in iscol 2993 */ 2994 PetscErrorCode ISGetSeqIS_SameColDist_Private(Mat mat,IS isrow,IS iscol,IS *isrow_d,IS *iscol_d,IS *iscol_o,const PetscInt *garray[]) 2995 { 2996 PetscErrorCode ierr; 2997 Vec x,cmap; 2998 const PetscInt *is_idx; 2999 PetscScalar *xarray,*cmaparray; 3000 PetscInt ncols,isstart,*idx,m,rstart,*cmap1,count; 3001 Mat_MPIAIJ *a=(Mat_MPIAIJ*)mat->data; 3002 Mat B=a->B; 3003 Vec lvec=a->lvec,lcmap; 3004 PetscInt i,cstart,cend,Bn=B->cmap->N; 3005 MPI_Comm comm; 3006 3007 PetscFunctionBegin; 3008 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3009 ierr = ISGetLocalSize(iscol,&ncols);CHKERRQ(ierr); 3010 3011 /* (1) iscol is a sub-column vector of mat, pad it with '-1.' to form a full vector x */ 3012 ierr = MatCreateVecs(mat,&x,NULL);CHKERRQ(ierr); 3013 ierr = VecDuplicate(x,&cmap);CHKERRQ(ierr); 3014 ierr = VecSet(x,-1.0);CHKERRQ(ierr); 3015 3016 /* Get start indices */ 3017 ierr = MPI_Scan(&ncols,&isstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3018 isstart -= ncols; 3019 ierr = MatGetOwnershipRangeColumn(mat,&cstart,&cend);CHKERRQ(ierr); 3020 3021 ierr = ISGetIndices(iscol,&is_idx);CHKERRQ(ierr); 3022 ierr = VecGetArray(x,&xarray);CHKERRQ(ierr); 3023 ierr = VecGetArray(cmap,&cmaparray);CHKERRQ(ierr); 3024 ierr = PetscMalloc1(ncols,&idx);CHKERRQ(ierr); 3025 for (i=0; i<ncols; i++) { 3026 xarray[is_idx[i]-cstart] = (PetscScalar)is_idx[i]; 3027 cmaparray[is_idx[i]-cstart] = i + isstart; /* global index of iscol[i] */ 3028 idx[i] = is_idx[i]-cstart; /* local index of iscol[i] */ 3029 } 3030 ierr = VecRestoreArray(x,&xarray);CHKERRQ(ierr); 3031 ierr = VecRestoreArray(cmap,&cmaparray);CHKERRQ(ierr); 3032 ierr = ISRestoreIndices(iscol,&is_idx);CHKERRQ(ierr); 3033 3034 /* Get iscol_d */ 3035 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,iscol_d);CHKERRQ(ierr); 3036 3037 /* Get isrow_d */ 3038 ierr = ISGetLocalSize(isrow,&m);CHKERRQ(ierr); 3039 rstart = mat->rmap->rstart; 3040 ierr = PetscMalloc1(m,&idx);CHKERRQ(ierr); 3041 ierr = ISGetIndices(isrow,&is_idx);CHKERRQ(ierr); 3042 for (i=0; i<m; i++) idx[i] = is_idx[i]-rstart; 3043 ierr = ISRestoreIndices(isrow,&is_idx);CHKERRQ(ierr); 3044 3045 ierr = ISCreateGeneral(PETSC_COMM_SELF,m,idx,PETSC_OWN_POINTER,isrow_d);CHKERRQ(ierr); 3046 ierr = ISGetBlockSize(isrow,&i);CHKERRQ(ierr); 3047 ierr = ISSetBlockSize(*isrow_d,i);CHKERRQ(ierr); 3048 3049 /* (2) Scatter x and cmap using aij->Mvctx to get their off-process portions (see MatMult_MPIAIJ) */ 3050 ierr = VecScatterBegin(a->Mvctx,x,lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 3051 3052 ierr = VecDuplicate(lvec,&lcmap);CHKERRQ(ierr); 3053 3054 ierr = VecScatterEnd(a->Mvctx,x,lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 3055 ierr = VecScatterBegin(a->Mvctx,cmap,lcmap,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 3056 ierr = VecScatterEnd(a->Mvctx,cmap,lcmap,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 3057 3058 /* (3) create sequential iscol_o (a subset of iscol) and isgarray */ 3059 /* off-process column indices */ 3060 count = 0; 3061 ierr = PetscMalloc1(Bn,&idx);CHKERRQ(ierr); 3062 ierr = PetscMalloc1(Bn,&cmap1);CHKERRQ(ierr); 3063 3064 ierr = VecGetArray(lvec,&xarray);CHKERRQ(ierr); 3065 ierr = VecGetArray(lcmap,&cmaparray);CHKERRQ(ierr); 3066 for (i=0; i<Bn; i++) { 3067 if (PetscRealPart(xarray[i]) > -1.0) { 3068 idx[count] = i; /* local column index in off-diagonal part B */ 3069 cmap1[count++] = (PetscInt)PetscRealPart(cmaparray[i]); /* column index in submat */ 3070 } 3071 } 3072 ierr = VecRestoreArray(lvec,&xarray);CHKERRQ(ierr); 3073 ierr = VecRestoreArray(lcmap,&cmaparray);CHKERRQ(ierr); 3074 3075 ierr = ISCreateGeneral(PETSC_COMM_SELF,count,idx,PETSC_COPY_VALUES,iscol_o);CHKERRQ(ierr); 3076 ierr = PetscFree(idx);CHKERRQ(ierr); 3077 3078 *garray = cmap1; 3079 3080 ierr = VecDestroy(&x);CHKERRQ(ierr); 3081 ierr = VecDestroy(&cmap);CHKERRQ(ierr); 3082 ierr = VecDestroy(&lcmap);CHKERRQ(ierr); 3083 PetscFunctionReturn(0); 3084 } 3085 3086 /* isrow and iscol have same processor distribution as mat, output *submat is a submatrix of local mat */ 3087 PetscErrorCode MatCreateSubMatrix_MPIAIJ_SameRowColDist(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *submat) 3088 { 3089 PetscErrorCode ierr; 3090 Mat_MPIAIJ *a = (Mat_MPIAIJ*)mat->data,*asub; 3091 Mat M = NULL; 3092 MPI_Comm comm; 3093 IS iscol_d,isrow_d,iscol_o; 3094 Mat Asub = NULL,Bsub = NULL; 3095 PetscInt n; 3096 3097 PetscFunctionBegin; 3098 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3099 3100 if (call == MAT_REUSE_MATRIX) { 3101 /* Retrieve isrow_d, iscol_d and iscol_o from submat */ 3102 ierr = PetscObjectQuery((PetscObject)*submat,"isrow_d",(PetscObject*)&isrow_d);CHKERRQ(ierr); 3103 if (!isrow_d) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"isrow_d passed in was not used before, cannot reuse"); 3104 3105 ierr = PetscObjectQuery((PetscObject)*submat,"iscol_d",(PetscObject*)&iscol_d);CHKERRQ(ierr); 3106 if (!iscol_d) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"iscol_d passed in was not used before, cannot reuse"); 3107 3108 ierr = PetscObjectQuery((PetscObject)*submat,"iscol_o",(PetscObject*)&iscol_o);CHKERRQ(ierr); 3109 if (!iscol_o) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"iscol_o passed in was not used before, cannot reuse"); 3110 3111 /* Update diagonal and off-diagonal portions of submat */ 3112 asub = (Mat_MPIAIJ*)(*submat)->data; 3113 ierr = MatCreateSubMatrix_SeqAIJ(a->A,isrow_d,iscol_d,PETSC_DECIDE,MAT_REUSE_MATRIX,&asub->A);CHKERRQ(ierr); 3114 ierr = ISGetLocalSize(iscol_o,&n);CHKERRQ(ierr); 3115 if (n) { 3116 ierr = MatCreateSubMatrix_SeqAIJ(a->B,isrow_d,iscol_o,PETSC_DECIDE,MAT_REUSE_MATRIX,&asub->B);CHKERRQ(ierr); 3117 } 3118 ierr = MatAssemblyBegin(*submat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3119 ierr = MatAssemblyEnd(*submat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3120 3121 } else { /* call == MAT_INITIAL_MATRIX) */ 3122 const PetscInt *garray; 3123 PetscInt BsubN; 3124 3125 /* Create isrow_d, iscol_d, iscol_o and isgarray (replace isgarray with array?) */ 3126 ierr = ISGetSeqIS_SameColDist_Private(mat,isrow,iscol,&isrow_d,&iscol_d,&iscol_o,&garray);CHKERRQ(ierr); 3127 3128 /* Create local submatrices Asub and Bsub */ 3129 ierr = MatCreateSubMatrix_SeqAIJ(a->A,isrow_d,iscol_d,PETSC_DECIDE,MAT_INITIAL_MATRIX,&Asub);CHKERRQ(ierr); 3130 ierr = MatCreateSubMatrix_SeqAIJ(a->B,isrow_d,iscol_o,PETSC_DECIDE,MAT_INITIAL_MATRIX,&Bsub);CHKERRQ(ierr); 3131 3132 /* Create submatrix M */ 3133 ierr = MatCreateMPIAIJWithSeqAIJ(comm,Asub,Bsub,garray,&M);CHKERRQ(ierr); 3134 3135 /* If Bsub has empty columns, compress iscol_o such that it will retrieve condensed Bsub from a->B during reuse */ 3136 asub = (Mat_MPIAIJ*)M->data; 3137 3138 ierr = ISGetLocalSize(iscol_o,&BsubN);CHKERRQ(ierr); 3139 n = asub->B->cmap->N; 3140 if (BsubN > n) { 3141 /* This case can be tested using ~petsc/src/tao/bound/examples/tutorials/runplate2_3 */ 3142 const PetscInt *idx; 3143 PetscInt i,j,*idx_new,*subgarray = asub->garray; 3144 ierr = PetscInfo2(M,"submatrix Bn %D != BsubN %D, update iscol_o\n",n,BsubN);CHKERRQ(ierr); 3145 3146 ierr = PetscMalloc1(n,&idx_new);CHKERRQ(ierr); 3147 j = 0; 3148 ierr = ISGetIndices(iscol_o,&idx);CHKERRQ(ierr); 3149 for (i=0; i<n; i++) { 3150 if (j >= BsubN) break; 3151 while (subgarray[i] > garray[j]) j++; 3152 3153 if (subgarray[i] == garray[j]) { 3154 idx_new[i] = idx[j++]; 3155 } else SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"subgarray[%D]=%D cannot < garray[%D]=%D",i,subgarray[i],j,garray[j]); 3156 } 3157 ierr = ISRestoreIndices(iscol_o,&idx);CHKERRQ(ierr); 3158 3159 ierr = ISDestroy(&iscol_o);CHKERRQ(ierr); 3160 ierr = ISCreateGeneral(PETSC_COMM_SELF,n,idx_new,PETSC_OWN_POINTER,&iscol_o);CHKERRQ(ierr); 3161 3162 } else if (BsubN < n) { 3163 SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Columns of Bsub cannot be smaller than B's",BsubN,asub->B->cmap->N); 3164 } 3165 3166 ierr = PetscFree(garray);CHKERRQ(ierr); 3167 *submat = M; 3168 3169 /* Save isrow_d, iscol_d and iscol_o used in processor for next request */ 3170 ierr = PetscObjectCompose((PetscObject)M,"isrow_d",(PetscObject)isrow_d);CHKERRQ(ierr); 3171 ierr = ISDestroy(&isrow_d);CHKERRQ(ierr); 3172 3173 ierr = PetscObjectCompose((PetscObject)M,"iscol_d",(PetscObject)iscol_d);CHKERRQ(ierr); 3174 ierr = ISDestroy(&iscol_d);CHKERRQ(ierr); 3175 3176 ierr = PetscObjectCompose((PetscObject)M,"iscol_o",(PetscObject)iscol_o);CHKERRQ(ierr); 3177 ierr = ISDestroy(&iscol_o);CHKERRQ(ierr); 3178 } 3179 PetscFunctionReturn(0); 3180 } 3181 3182 PetscErrorCode MatCreateSubMatrix_MPIAIJ(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *newmat) 3183 { 3184 PetscErrorCode ierr; 3185 IS iscol_local,isrow_d; 3186 PetscInt csize; 3187 PetscInt n,i,j,start,end; 3188 PetscBool sameRowDist=PETSC_FALSE,sameDist[2],tsameDist[2]; 3189 MPI_Comm comm; 3190 3191 PetscFunctionBegin; 3192 /* If isrow has same processor distribution as mat, 3193 call MatCreateSubMatrix_MPIAIJ_SameRowDist() to avoid using a hash table with global size of iscol */ 3194 if (call == MAT_REUSE_MATRIX) { 3195 ierr = PetscObjectQuery((PetscObject)*newmat,"isrow_d",(PetscObject*)&isrow_d);CHKERRQ(ierr); 3196 if (isrow_d) { 3197 sameRowDist = PETSC_TRUE; 3198 tsameDist[1] = PETSC_TRUE; /* sameColDist */ 3199 } else { 3200 ierr = PetscObjectQuery((PetscObject)*newmat,"SubIScol",(PetscObject*)&iscol_local);CHKERRQ(ierr); 3201 if (iscol_local) { 3202 sameRowDist = PETSC_TRUE; 3203 tsameDist[1] = PETSC_FALSE; /* !sameColDist */ 3204 } 3205 } 3206 } else { 3207 /* Check if isrow has same processor distribution as mat */ 3208 sameDist[0] = PETSC_FALSE; 3209 ierr = ISGetLocalSize(isrow,&n);CHKERRQ(ierr); 3210 if (!n) { 3211 sameDist[0] = PETSC_TRUE; 3212 } else { 3213 ierr = ISGetMinMax(isrow,&i,&j);CHKERRQ(ierr); 3214 ierr = MatGetOwnershipRange(mat,&start,&end);CHKERRQ(ierr); 3215 if (i >= start && j < end) { 3216 sameDist[0] = PETSC_TRUE; 3217 } 3218 } 3219 3220 /* Check if iscol has same processor distribution as mat */ 3221 sameDist[1] = PETSC_FALSE; 3222 ierr = ISGetLocalSize(iscol,&n);CHKERRQ(ierr); 3223 if (!n) { 3224 sameDist[1] = PETSC_TRUE; 3225 } else { 3226 ierr = ISGetMinMax(iscol,&i,&j);CHKERRQ(ierr); 3227 ierr = MatGetOwnershipRangeColumn(mat,&start,&end);CHKERRQ(ierr); 3228 if (i >= start && j < end) sameDist[1] = PETSC_TRUE; 3229 } 3230 3231 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3232 ierr = MPIU_Allreduce(&sameDist,&tsameDist,2,MPIU_BOOL,MPI_LAND,comm);CHKERRQ(ierr); 3233 sameRowDist = tsameDist[0]; 3234 } 3235 3236 if (sameRowDist) { 3237 if (tsameDist[1]) { /* sameRowDist & sameColDist */ 3238 /* isrow and iscol have same processor distribution as mat */ 3239 ierr = MatCreateSubMatrix_MPIAIJ_SameRowColDist(mat,isrow,iscol,call,newmat);CHKERRQ(ierr); 3240 } else { /* sameRowDist */ 3241 /* isrow has same processor distribution as mat */ 3242 ierr = MatCreateSubMatrix_MPIAIJ_SameRowDist(mat,isrow,iscol,call,newmat);CHKERRQ(ierr); 3243 } 3244 PetscFunctionReturn(0); 3245 } 3246 3247 /* General case: iscol -> iscol_local which has global size of iscol */ 3248 if (call == MAT_REUSE_MATRIX) { 3249 ierr = PetscObjectQuery((PetscObject)*newmat,"ISAllGather",(PetscObject*)&iscol_local);CHKERRQ(ierr); 3250 if (!iscol_local) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3251 } else { 3252 ierr = ISGetSeqIS_Private(mat,iscol,&iscol_local);CHKERRQ(ierr); 3253 } 3254 3255 ierr = ISGetLocalSize(iscol,&csize);CHKERRQ(ierr); 3256 ierr = MatCreateSubMatrix_MPIAIJ_nonscalable(mat,isrow,iscol_local,csize,call,newmat);CHKERRQ(ierr); 3257 3258 if (call == MAT_INITIAL_MATRIX) { 3259 ierr = PetscObjectCompose((PetscObject)*newmat,"ISAllGather",(PetscObject)iscol_local);CHKERRQ(ierr); 3260 ierr = ISDestroy(&iscol_local);CHKERRQ(ierr); 3261 } 3262 PetscFunctionReturn(0); 3263 } 3264 3265 /*@C 3266 MatCreateMPIAIJWithSeqAIJ - creates a MPIAIJ matrix using SeqAIJ matrices that contain the "diagonal" 3267 and "off-diagonal" part of the matrix in CSR format. 3268 3269 Collective on MPI_Comm 3270 3271 Input Parameters: 3272 + comm - MPI communicator 3273 . A - "diagonal" portion of matrix 3274 . B - "off-diagonal" portion of matrix, may have empty columns, will be destroyed by this routine 3275 - garray - global index of B columns 3276 3277 Output Parameter: 3278 . mat - the matrix, with input A as its local diagonal matrix 3279 Level: advanced 3280 3281 Notes: 3282 See MatCreateAIJ() for the definition of "diagonal" and "off-diagonal" portion of the matrix. 3283 A becomes part of output mat, B is destroyed by this routine. The user cannot use A and B anymore. 3284 3285 .seealso: MatCreateMPIAIJWithSplitArrays() 3286 @*/ 3287 PetscErrorCode MatCreateMPIAIJWithSeqAIJ(MPI_Comm comm,Mat A,Mat B,const PetscInt garray[],Mat *mat) 3288 { 3289 PetscErrorCode ierr; 3290 Mat_MPIAIJ *maij; 3291 Mat_SeqAIJ *b=(Mat_SeqAIJ*)B->data,*bnew; 3292 PetscInt *oi=b->i,*oj=b->j,i,nz,col; 3293 PetscScalar *oa=b->a; 3294 Mat Bnew; 3295 PetscInt m,n,N; 3296 3297 PetscFunctionBegin; 3298 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 3299 ierr = MatGetSize(A,&m,&n);CHKERRQ(ierr); 3300 if (m != B->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Am %D != Bm %D",m,B->rmap->N); 3301 if (A->rmap->bs != B->rmap->bs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"A row bs %D != B row bs %D",A->rmap->bs,B->rmap->bs); 3302 if (A->cmap->bs != B->cmap->bs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"A column bs %D != B column bs %D",A->cmap->bs,B->cmap->bs); 3303 3304 /* Get global columns of mat */ 3305 ierr = MPIU_Allreduce(&n,&N,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3306 3307 ierr = MatSetSizes(*mat,m,n,PETSC_DECIDE,N);CHKERRQ(ierr); 3308 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 3309 ierr = MatSetBlockSizes(*mat,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 3310 maij = (Mat_MPIAIJ*)(*mat)->data; 3311 3312 (*mat)->preallocated = PETSC_TRUE; 3313 3314 ierr = PetscLayoutSetUp((*mat)->rmap);CHKERRQ(ierr); 3315 ierr = PetscLayoutSetUp((*mat)->cmap);CHKERRQ(ierr); 3316 3317 /* Set A as diagonal portion of *mat */ 3318 maij->A = A; 3319 3320 nz = oi[m]; 3321 for (i=0; i<nz; i++) { 3322 col = oj[i]; 3323 oj[i] = garray[col]; 3324 } 3325 3326 /* Set Bnew as off-diagonal portion of *mat */ 3327 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,N,oi,oj,oa,&Bnew);CHKERRQ(ierr); 3328 bnew = (Mat_SeqAIJ*)Bnew->data; 3329 bnew->maxnz = b->maxnz; /* allocated nonzeros of B */ 3330 maij->B = Bnew; 3331 3332 if (B->rmap->N != Bnew->rmap->N) SETERRQ2(PETSC_COMM_SELF,0,"BN %d != BnewN %d",B->rmap->N,Bnew->rmap->N); 3333 3334 b->singlemalloc = PETSC_FALSE; /* B arrays are shared by Bnew */ 3335 b->free_a = PETSC_FALSE; 3336 b->free_ij = PETSC_FALSE; 3337 ierr = MatDestroy(&B);CHKERRQ(ierr); 3338 3339 bnew->singlemalloc = PETSC_TRUE; /* arrays will be freed by MatDestroy(&Bnew) */ 3340 bnew->free_a = PETSC_TRUE; 3341 bnew->free_ij = PETSC_TRUE; 3342 3343 /* condense columns of maij->B */ 3344 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE);CHKERRQ(ierr); 3345 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3346 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3347 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_FALSE);CHKERRQ(ierr); 3348 ierr = MatSetOption(*mat,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3349 PetscFunctionReturn(0); 3350 } 3351 3352 extern PetscErrorCode MatCreateSubMatrices_MPIAIJ_SingleIS_Local(Mat,PetscInt,const IS[],const IS[],MatReuse,PetscBool,Mat*); 3353 3354 PetscErrorCode MatCreateSubMatrix_MPIAIJ_SameRowDist(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *newmat) 3355 { 3356 PetscErrorCode ierr; 3357 PetscInt i,m,n,rstart,row,rend,nz,j,bs,cbs; 3358 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 3359 Mat_MPIAIJ *a=(Mat_MPIAIJ*)mat->data; 3360 Mat M,Msub,B=a->B; 3361 MatScalar *aa; 3362 Mat_SeqAIJ *aij; 3363 PetscInt *garray = a->garray,*colsub,Ncols; 3364 PetscInt count,Bn=B->cmap->N,cstart=mat->cmap->rstart,cend=mat->cmap->rend; 3365 IS iscol_sub,iscmap; 3366 const PetscInt *is_idx,*cmap; 3367 PetscBool allcolumns=PETSC_FALSE; 3368 IS iscol_local=NULL; 3369 MPI_Comm comm; 3370 3371 PetscFunctionBegin; 3372 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3373 3374 if (call == MAT_REUSE_MATRIX) { 3375 ierr = PetscObjectQuery((PetscObject)*newmat,"SubIScol",(PetscObject*)&iscol_sub);CHKERRQ(ierr); 3376 if (!iscol_sub) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"SubIScol passed in was not used before, cannot reuse"); 3377 ierr = ISGetLocalSize(iscol_sub,&count);CHKERRQ(ierr); 3378 3379 ierr = PetscObjectQuery((PetscObject)*newmat,"Subcmap",(PetscObject*)&iscmap);CHKERRQ(ierr); 3380 if (!iscmap) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Subcmap passed in was not used before, cannot reuse"); 3381 3382 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject*)&Msub);CHKERRQ(ierr); 3383 if (!Msub) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3384 3385 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol_sub,MAT_REUSE_MATRIX,PETSC_FALSE,&Msub);CHKERRQ(ierr); 3386 3387 } else { /* call == MAT_INITIAL_MATRIX) */ 3388 PetscBool flg; 3389 3390 ierr = ISGetLocalSize(iscol,&n);CHKERRQ(ierr); 3391 ierr = ISGetSize(iscol,&Ncols);CHKERRQ(ierr); 3392 3393 /* (1) iscol -> nonscalable iscol_local */ 3394 ierr = ISGetSeqIS_Private(mat,iscol,&iscol_local);CHKERRQ(ierr); 3395 ierr = ISGetLocalSize(iscol_local,&n);CHKERRQ(ierr); /* local size of iscol_local = global columns of newmat */ 3396 if (n != Ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"n %d != Ncols %d",n,Ncols); 3397 3398 /* Check for special case: each processor gets entire matrix columns */ 3399 ierr = ISIdentity(iscol_local,&flg);CHKERRQ(ierr); 3400 if (flg && n == mat->cmap->N) allcolumns = PETSC_TRUE; 3401 if (allcolumns) { 3402 iscol_sub = iscol_local; 3403 ierr = PetscObjectReference((PetscObject)iscol_local);CHKERRQ(ierr); 3404 ierr = ISCreateStride(PETSC_COMM_SELF,n,0,1,&iscmap);CHKERRQ(ierr); 3405 3406 } else { 3407 /* (2) iscol_local -> iscol_sub and iscmap */ 3408 PetscInt *idx,*cmap1,k,cbs; 3409 3410 /* implementation below requires iscol_local be sorted, it can have duplicate indices */ 3411 ierr = ISSorted(iscol_local,&flg);CHKERRQ(ierr); 3412 if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"unsorted iscol_local is not implemented yet"); 3413 3414 ierr = PetscMalloc1(Ncols,&idx);CHKERRQ(ierr); 3415 ierr = PetscMalloc1(Ncols,&cmap1);CHKERRQ(ierr); 3416 ierr = ISGetIndices(iscol_local,&is_idx);CHKERRQ(ierr); 3417 count = 0; 3418 k = 0; 3419 for (i=0; i<Ncols; i++) { 3420 j = is_idx[i]; 3421 if (j >= cstart && j < cend) { 3422 /* diagonal part of mat */ 3423 idx[count] = j; 3424 cmap1[count++] = i; /* column index in submat */ 3425 } else if (Bn) { 3426 /* off-diagonal part of mat */ 3427 if (j == garray[k]) { 3428 idx[count] = j; 3429 cmap1[count++] = i; /* column index in submat */ 3430 } else if (j > garray[k]) { 3431 while (j > garray[k] && k < Bn-1) k++; 3432 if (j == garray[k]) { 3433 idx[count] = j; 3434 cmap1[count++] = i; /* column index in submat */ 3435 } 3436 } 3437 } 3438 } 3439 ierr = ISRestoreIndices(iscol_local,&is_idx);CHKERRQ(ierr); 3440 3441 ierr = ISCreateGeneral(PETSC_COMM_SELF,count,idx,PETSC_OWN_POINTER,&iscol_sub);CHKERRQ(ierr); 3442 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)iscol_local),count,cmap1,PETSC_OWN_POINTER,&iscmap);CHKERRQ(ierr); 3443 } 3444 3445 /* (3) Create sequential Msub */ 3446 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol_sub,MAT_INITIAL_MATRIX,allcolumns,&Msub);CHKERRQ(ierr); 3447 } 3448 3449 ierr = ISGetLocalSize(iscol_sub,&count);CHKERRQ(ierr); 3450 aij = (Mat_SeqAIJ*)(Msub)->data; 3451 ii = aij->i; 3452 ierr = ISGetIndices(iscmap,&cmap);CHKERRQ(ierr); 3453 3454 /* 3455 m - number of local rows 3456 Ncols - number of columns (same on all processors) 3457 rstart - first row in new global matrix generated 3458 */ 3459 ierr = MatGetSize(Msub,&m,NULL);CHKERRQ(ierr); 3460 3461 if (call == MAT_INITIAL_MATRIX) { 3462 /* (4) Create parallel newmat */ 3463 PetscMPIInt rank,size; 3464 PetscInt csize; 3465 3466 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3467 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3468 3469 /* 3470 Determine the number of non-zeros in the diagonal and off-diagonal 3471 portions of the matrix in order to do correct preallocation 3472 */ 3473 3474 /* first get start and end of "diagonal" columns */ 3475 ierr = ISGetLocalSize(iscol,&csize);CHKERRQ(ierr); 3476 if (csize == PETSC_DECIDE) { 3477 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 3478 if (mglobal == Ncols) { /* square matrix */ 3479 nlocal = m; 3480 } else { 3481 nlocal = Ncols/size + ((Ncols % size) > rank); 3482 } 3483 } else { 3484 nlocal = csize; 3485 } 3486 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3487 rstart = rend - nlocal; 3488 if (rank == size - 1 && rend != Ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Local column sizes %D do not add up to total number of columns %D",rend,Ncols); 3489 3490 /* next, compute all the lengths */ 3491 jj = aij->j; 3492 ierr = PetscMalloc1(2*m+1,&dlens);CHKERRQ(ierr); 3493 olens = dlens + m; 3494 for (i=0; i<m; i++) { 3495 jend = ii[i+1] - ii[i]; 3496 olen = 0; 3497 dlen = 0; 3498 for (j=0; j<jend; j++) { 3499 if (cmap[*jj] < rstart || cmap[*jj] >= rend) olen++; 3500 else dlen++; 3501 jj++; 3502 } 3503 olens[i] = olen; 3504 dlens[i] = dlen; 3505 } 3506 ierr = MatGetBlockSizes(Msub,&bs,&cbs);CHKERRQ(ierr); 3507 3508 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 3509 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,Ncols);CHKERRQ(ierr); 3510 ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); 3511 ierr = MatSetType(M,((PetscObject)mat)->type_name);CHKERRQ(ierr); 3512 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 3513 ierr = PetscFree(dlens);CHKERRQ(ierr); 3514 3515 } else { /* call == MAT_REUSE_MATRIX */ 3516 M = *newmat; 3517 ierr = MatGetLocalSize(M,&i,NULL);CHKERRQ(ierr); 3518 if (i != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 3519 ierr = MatZeroEntries(M);CHKERRQ(ierr); 3520 /* 3521 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 3522 rather than the slower MatSetValues(). 3523 */ 3524 M->was_assembled = PETSC_TRUE; 3525 M->assembled = PETSC_FALSE; 3526 } 3527 3528 /* (5) Set values of Msub to *newmat */ 3529 ierr = PetscMalloc1(count,&colsub);CHKERRQ(ierr); 3530 ierr = MatGetOwnershipRange(M,&rstart,NULL);CHKERRQ(ierr); 3531 3532 jj = aij->j; 3533 aa = aij->a; 3534 for (i=0; i<m; i++) { 3535 row = rstart + i; 3536 nz = ii[i+1] - ii[i]; 3537 for (j=0; j<nz; j++) colsub[j] = cmap[jj[j]]; 3538 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,colsub,aa,INSERT_VALUES);CHKERRQ(ierr); 3539 jj += nz; aa += nz; 3540 } 3541 ierr = ISRestoreIndices(iscmap,&cmap);CHKERRQ(ierr); 3542 3543 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3544 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3545 3546 ierr = PetscFree(colsub);CHKERRQ(ierr); 3547 3548 /* save Msub, iscol_sub and iscmap used in processor for next request */ 3549 if (call == MAT_INITIAL_MATRIX) { 3550 *newmat = M; 3551 ierr = PetscObjectCompose((PetscObject)(*newmat),"SubMatrix",(PetscObject)Msub);CHKERRQ(ierr); 3552 ierr = MatDestroy(&Msub);CHKERRQ(ierr); 3553 3554 ierr = PetscObjectCompose((PetscObject)(*newmat),"SubIScol",(PetscObject)iscol_sub);CHKERRQ(ierr); 3555 ierr = ISDestroy(&iscol_sub);CHKERRQ(ierr); 3556 3557 ierr = PetscObjectCompose((PetscObject)(*newmat),"Subcmap",(PetscObject)iscmap);CHKERRQ(ierr); 3558 ierr = ISDestroy(&iscmap);CHKERRQ(ierr); 3559 3560 if (iscol_local) { 3561 ierr = PetscObjectCompose((PetscObject)(*newmat),"ISAllGather",(PetscObject)iscol_local);CHKERRQ(ierr); 3562 ierr = ISDestroy(&iscol_local);CHKERRQ(ierr); 3563 } 3564 } 3565 PetscFunctionReturn(0); 3566 } 3567 3568 /* 3569 Not great since it makes two copies of the submatrix, first an SeqAIJ 3570 in local and then by concatenating the local matrices the end result. 3571 Writing it directly would be much like MatCreateSubMatrices_MPIAIJ() 3572 3573 Note: This requires a sequential iscol with all indices. 3574 */ 3575 PetscErrorCode MatCreateSubMatrix_MPIAIJ_nonscalable(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat) 3576 { 3577 PetscErrorCode ierr; 3578 PetscMPIInt rank,size; 3579 PetscInt i,m,n,rstart,row,rend,nz,*cwork,j,bs,cbs; 3580 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 3581 Mat M,Mreuse; 3582 MatScalar *aa,*vwork; 3583 MPI_Comm comm; 3584 Mat_SeqAIJ *aij; 3585 PetscBool colflag,allcolumns=PETSC_FALSE; 3586 3587 PetscFunctionBegin; 3588 ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr); 3589 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3590 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3591 3592 /* Check for special case: each processor gets entire matrix columns */ 3593 ierr = ISIdentity(iscol,&colflag);CHKERRQ(ierr); 3594 ierr = ISGetLocalSize(iscol,&n);CHKERRQ(ierr); 3595 if (colflag && n == mat->cmap->N) allcolumns = PETSC_TRUE; 3596 3597 if (call == MAT_REUSE_MATRIX) { 3598 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject*)&Mreuse);CHKERRQ(ierr); 3599 if (!Mreuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3600 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol,MAT_REUSE_MATRIX,allcolumns,&Mreuse);CHKERRQ(ierr); 3601 } else { 3602 ierr = MatCreateSubMatrices_MPIAIJ_SingleIS_Local(mat,1,&isrow,&iscol,MAT_INITIAL_MATRIX,allcolumns,&Mreuse);CHKERRQ(ierr); 3603 } 3604 3605 /* 3606 m - number of local rows 3607 n - number of columns (same on all processors) 3608 rstart - first row in new global matrix generated 3609 */ 3610 ierr = MatGetSize(Mreuse,&m,&n);CHKERRQ(ierr); 3611 ierr = MatGetBlockSizes(Mreuse,&bs,&cbs);CHKERRQ(ierr); 3612 if (call == MAT_INITIAL_MATRIX) { 3613 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3614 ii = aij->i; 3615 jj = aij->j; 3616 3617 /* 3618 Determine the number of non-zeros in the diagonal and off-diagonal 3619 portions of the matrix in order to do correct preallocation 3620 */ 3621 3622 /* first get start and end of "diagonal" columns */ 3623 if (csize == PETSC_DECIDE) { 3624 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 3625 if (mglobal == n) { /* square matrix */ 3626 nlocal = m; 3627 } else { 3628 nlocal = n/size + ((n % size) > rank); 3629 } 3630 } else { 3631 nlocal = csize; 3632 } 3633 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3634 rstart = rend - nlocal; 3635 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); 3636 3637 /* next, compute all the lengths */ 3638 ierr = PetscMalloc1(2*m+1,&dlens);CHKERRQ(ierr); 3639 olens = dlens + m; 3640 for (i=0; i<m; i++) { 3641 jend = ii[i+1] - ii[i]; 3642 olen = 0; 3643 dlen = 0; 3644 for (j=0; j<jend; j++) { 3645 if (*jj < rstart || *jj >= rend) olen++; 3646 else dlen++; 3647 jj++; 3648 } 3649 olens[i] = olen; 3650 dlens[i] = dlen; 3651 } 3652 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 3653 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,n);CHKERRQ(ierr); 3654 ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); 3655 ierr = MatSetType(M,((PetscObject)mat)->type_name);CHKERRQ(ierr); 3656 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 3657 ierr = PetscFree(dlens);CHKERRQ(ierr); 3658 } else { 3659 PetscInt ml,nl; 3660 3661 M = *newmat; 3662 ierr = MatGetLocalSize(M,&ml,&nl);CHKERRQ(ierr); 3663 if (ml != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 3664 ierr = MatZeroEntries(M);CHKERRQ(ierr); 3665 /* 3666 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 3667 rather than the slower MatSetValues(). 3668 */ 3669 M->was_assembled = PETSC_TRUE; 3670 M->assembled = PETSC_FALSE; 3671 } 3672 ierr = MatGetOwnershipRange(M,&rstart,&rend);CHKERRQ(ierr); 3673 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3674 ii = aij->i; 3675 jj = aij->j; 3676 aa = aij->a; 3677 for (i=0; i<m; i++) { 3678 row = rstart + i; 3679 nz = ii[i+1] - ii[i]; 3680 cwork = jj; jj += nz; 3681 vwork = aa; aa += nz; 3682 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,cwork,vwork,INSERT_VALUES);CHKERRQ(ierr); 3683 } 3684 3685 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3686 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3687 *newmat = M; 3688 3689 /* save submatrix used in processor for next request */ 3690 if (call == MAT_INITIAL_MATRIX) { 3691 ierr = PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);CHKERRQ(ierr); 3692 ierr = MatDestroy(&Mreuse);CHKERRQ(ierr); 3693 } 3694 PetscFunctionReturn(0); 3695 } 3696 3697 PetscErrorCode MatMPIAIJSetPreallocationCSR_MPIAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3698 { 3699 PetscInt m,cstart, cend,j,nnz,i,d; 3700 PetscInt *d_nnz,*o_nnz,nnz_max = 0,rstart,ii; 3701 const PetscInt *JJ; 3702 PetscScalar *values; 3703 PetscErrorCode ierr; 3704 PetscBool nooffprocentries; 3705 3706 PetscFunctionBegin; 3707 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Ii[0] must be 0 it is %D",Ii[0]); 3708 3709 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3710 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3711 m = B->rmap->n; 3712 cstart = B->cmap->rstart; 3713 cend = B->cmap->rend; 3714 rstart = B->rmap->rstart; 3715 3716 ierr = PetscMalloc2(m,&d_nnz,m,&o_nnz);CHKERRQ(ierr); 3717 3718 #if defined(PETSC_USE_DEBUGGING) 3719 for (i=0; i<m; i++) { 3720 nnz = Ii[i+1]- Ii[i]; 3721 JJ = J + Ii[i]; 3722 if (nnz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Local row %D has a negative %D number of columns",i,nnz); 3723 if (nnz && (JJ[0] < 0)) SETERRRQ1(PETSC_ERR_ARG_WRONGSTATE,"Row %D starts with negative column index",i,j); 3724 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); 3725 } 3726 #endif 3727 3728 for (i=0; i<m; i++) { 3729 nnz = Ii[i+1]- Ii[i]; 3730 JJ = J + Ii[i]; 3731 nnz_max = PetscMax(nnz_max,nnz); 3732 d = 0; 3733 for (j=0; j<nnz; j++) { 3734 if (cstart <= JJ[j] && JJ[j] < cend) d++; 3735 } 3736 d_nnz[i] = d; 3737 o_nnz[i] = nnz - d; 3738 } 3739 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 3740 ierr = PetscFree2(d_nnz,o_nnz);CHKERRQ(ierr); 3741 3742 if (v) values = (PetscScalar*)v; 3743 else { 3744 ierr = PetscCalloc1(nnz_max+1,&values);CHKERRQ(ierr); 3745 } 3746 3747 for (i=0; i<m; i++) { 3748 ii = i + rstart; 3749 nnz = Ii[i+1]- Ii[i]; 3750 ierr = MatSetValues_MPIAIJ(B,1,&ii,nnz,J+Ii[i],values+(v ? Ii[i] : 0),INSERT_VALUES);CHKERRQ(ierr); 3751 } 3752 nooffprocentries = B->nooffprocentries; 3753 B->nooffprocentries = PETSC_TRUE; 3754 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3755 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3756 B->nooffprocentries = nooffprocentries; 3757 3758 if (!v) { 3759 ierr = PetscFree(values);CHKERRQ(ierr); 3760 } 3761 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3762 PetscFunctionReturn(0); 3763 } 3764 3765 /*@ 3766 MatMPIAIJSetPreallocationCSR - Allocates memory for a sparse parallel matrix in AIJ format 3767 (the default parallel PETSc format). 3768 3769 Collective on MPI_Comm 3770 3771 Input Parameters: 3772 + B - the matrix 3773 . i - the indices into j for the start of each local row (starts with zero) 3774 . j - the column indices for each local row (starts with zero) 3775 - v - optional values in the matrix 3776 3777 Level: developer 3778 3779 Notes: 3780 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3781 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3782 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3783 3784 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3785 3786 The format which is used for the sparse matrix input, is equivalent to a 3787 row-major ordering.. i.e for the following matrix, the input data expected is 3788 as shown 3789 3790 $ 1 0 0 3791 $ 2 0 3 P0 3792 $ ------- 3793 $ 4 5 6 P1 3794 $ 3795 $ Process0 [P0]: rows_owned=[0,1] 3796 $ i = {0,1,3} [size = nrow+1 = 2+1] 3797 $ j = {0,0,2} [size = 3] 3798 $ v = {1,2,3} [size = 3] 3799 $ 3800 $ Process1 [P1]: rows_owned=[2] 3801 $ i = {0,3} [size = nrow+1 = 1+1] 3802 $ j = {0,1,2} [size = 3] 3803 $ v = {4,5,6} [size = 3] 3804 3805 .keywords: matrix, aij, compressed row, sparse, parallel 3806 3807 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatCreateAIJ(), MATMPIAIJ, 3808 MatCreateSeqAIJWithArrays(), MatCreateMPIAIJWithSplitArrays() 3809 @*/ 3810 PetscErrorCode MatMPIAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[], const PetscScalar v[]) 3811 { 3812 PetscErrorCode ierr; 3813 3814 PetscFunctionBegin; 3815 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3816 PetscFunctionReturn(0); 3817 } 3818 3819 /*@C 3820 MatMPIAIJSetPreallocation - Preallocates memory for a sparse parallel matrix in AIJ format 3821 (the default parallel PETSc format). For good matrix assembly performance 3822 the user should preallocate the matrix storage by setting the parameters 3823 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3824 performance can be increased by more than a factor of 50. 3825 3826 Collective on MPI_Comm 3827 3828 Input Parameters: 3829 + B - the matrix 3830 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3831 (same value is used for all local rows) 3832 . d_nnz - array containing the number of nonzeros in the various rows of the 3833 DIAGONAL portion of the local submatrix (possibly different for each row) 3834 or NULL (PETSC_NULL_INTEGER in Fortran), if d_nz is used to specify the nonzero structure. 3835 The size of this array is equal to the number of local rows, i.e 'm'. 3836 For matrices that will be factored, you must leave room for (and set) 3837 the diagonal entry even if it is zero. 3838 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3839 submatrix (same value is used for all local rows). 3840 - o_nnz - array containing the number of nonzeros in the various rows of the 3841 OFF-DIAGONAL portion of the local submatrix (possibly different for 3842 each row) or NULL (PETSC_NULL_INTEGER in Fortran), if o_nz is used to specify the nonzero 3843 structure. The size of this array is equal to the number 3844 of local rows, i.e 'm'. 3845 3846 If the *_nnz parameter is given then the *_nz parameter is ignored 3847 3848 The AIJ format (also called the Yale sparse matrix format or 3849 compressed row storage (CSR)), is fully compatible with standard Fortran 77 3850 storage. The stored row and column indices begin with zero. 3851 See Users-Manual: ch_mat for details. 3852 3853 The parallel matrix is partitioned such that the first m0 rows belong to 3854 process 0, the next m1 rows belong to process 1, the next m2 rows belong 3855 to process 2 etc.. where m0,m1,m2... are the input parameter 'm'. 3856 3857 The DIAGONAL portion of the local submatrix of a processor can be defined 3858 as the submatrix which is obtained by extraction the part corresponding to 3859 the rows r1-r2 and columns c1-c2 of the global matrix, where r1 is the 3860 first row that belongs to the processor, r2 is the last row belonging to 3861 the this processor, and c1-c2 is range of indices of the local part of a 3862 vector suitable for applying the matrix to. This is an mxn matrix. In the 3863 common case of a square matrix, the row and column ranges are the same and 3864 the DIAGONAL part is also square. The remaining portion of the local 3865 submatrix (mxN) constitute the OFF-DIAGONAL portion. 3866 3867 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3868 3869 You can call MatGetInfo() to get information on how effective the preallocation was; 3870 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3871 You can also run with the option -info and look for messages with the string 3872 malloc in them to see if additional memory allocation was needed. 3873 3874 Example usage: 3875 3876 Consider the following 8x8 matrix with 34 non-zero values, that is 3877 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 3878 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 3879 as follows: 3880 3881 .vb 3882 1 2 0 | 0 3 0 | 0 4 3883 Proc0 0 5 6 | 7 0 0 | 8 0 3884 9 0 10 | 11 0 0 | 12 0 3885 ------------------------------------- 3886 13 0 14 | 15 16 17 | 0 0 3887 Proc1 0 18 0 | 19 20 21 | 0 0 3888 0 0 0 | 22 23 0 | 24 0 3889 ------------------------------------- 3890 Proc2 25 26 27 | 0 0 28 | 29 0 3891 30 0 0 | 31 32 33 | 0 34 3892 .ve 3893 3894 This can be represented as a collection of submatrices as: 3895 3896 .vb 3897 A B C 3898 D E F 3899 G H I 3900 .ve 3901 3902 Where the submatrices A,B,C are owned by proc0, D,E,F are 3903 owned by proc1, G,H,I are owned by proc2. 3904 3905 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3906 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3907 The 'M','N' parameters are 8,8, and have the same values on all procs. 3908 3909 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 3910 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 3911 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 3912 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 3913 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 3914 matrix, ans [DF] as another SeqAIJ matrix. 3915 3916 When d_nz, o_nz parameters are specified, d_nz storage elements are 3917 allocated for every row of the local diagonal submatrix, and o_nz 3918 storage locations are allocated for every row of the OFF-DIAGONAL submat. 3919 One way to choose d_nz and o_nz is to use the max nonzerors per local 3920 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 3921 In this case, the values of d_nz,o_nz are: 3922 .vb 3923 proc0 : dnz = 2, o_nz = 2 3924 proc1 : dnz = 3, o_nz = 2 3925 proc2 : dnz = 1, o_nz = 4 3926 .ve 3927 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3928 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3929 for proc3. i.e we are using 12+15+10=37 storage locations to store 3930 34 values. 3931 3932 When d_nnz, o_nnz parameters are specified, the storage is specified 3933 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3934 In the above case the values for d_nnz,o_nnz are: 3935 .vb 3936 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3937 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3938 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3939 .ve 3940 Here the space allocated is sum of all the above values i.e 34, and 3941 hence pre-allocation is perfect. 3942 3943 Level: intermediate 3944 3945 .keywords: matrix, aij, compressed row, sparse, parallel 3946 3947 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateAIJ(), MatMPIAIJSetPreallocationCSR(), 3948 MATMPIAIJ, MatGetInfo(), PetscSplitOwnership() 3949 @*/ 3950 PetscErrorCode MatMPIAIJSetPreallocation(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 3951 { 3952 PetscErrorCode ierr; 3953 3954 PetscFunctionBegin; 3955 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3956 PetscValidType(B,1); 3957 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[]),(B,d_nz,d_nnz,o_nz,o_nnz));CHKERRQ(ierr); 3958 PetscFunctionReturn(0); 3959 } 3960 3961 /*@ 3962 MatCreateMPIAIJWithArrays - creates a MPI AIJ matrix using arrays that contain in standard 3963 CSR format the local rows. 3964 3965 Collective on MPI_Comm 3966 3967 Input Parameters: 3968 + comm - MPI communicator 3969 . m - number of local rows (Cannot be PETSC_DECIDE) 3970 . n - This value should be the same as the local size used in creating the 3971 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3972 calculated if N is given) For square matrices n is almost always m. 3973 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3974 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3975 . i - row indices 3976 . j - column indices 3977 - a - matrix values 3978 3979 Output Parameter: 3980 . mat - the matrix 3981 3982 Level: intermediate 3983 3984 Notes: 3985 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3986 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3987 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3988 3989 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3990 3991 The format which is used for the sparse matrix input, is equivalent to a 3992 row-major ordering.. i.e for the following matrix, the input data expected is 3993 as shown 3994 3995 $ 1 0 0 3996 $ 2 0 3 P0 3997 $ ------- 3998 $ 4 5 6 P1 3999 $ 4000 $ Process0 [P0]: rows_owned=[0,1] 4001 $ i = {0,1,3} [size = nrow+1 = 2+1] 4002 $ j = {0,0,2} [size = 3] 4003 $ v = {1,2,3} [size = 3] 4004 $ 4005 $ Process1 [P1]: rows_owned=[2] 4006 $ i = {0,3} [size = nrow+1 = 1+1] 4007 $ j = {0,1,2} [size = 3] 4008 $ v = {4,5,6} [size = 3] 4009 4010 .keywords: matrix, aij, compressed row, sparse, parallel 4011 4012 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 4013 MATMPIAIJ, MatCreateAIJ(), MatCreateMPIAIJWithSplitArrays() 4014 @*/ 4015 PetscErrorCode MatCreateMPIAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,const PetscInt i[],const PetscInt j[],const PetscScalar a[],Mat *mat) 4016 { 4017 PetscErrorCode ierr; 4018 4019 PetscFunctionBegin; 4020 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4021 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 4022 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4023 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 4024 /* ierr = MatSetBlockSizes(M,bs,cbs);CHKERRQ(ierr); */ 4025 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 4026 ierr = MatMPIAIJSetPreallocationCSR(*mat,i,j,a);CHKERRQ(ierr); 4027 PetscFunctionReturn(0); 4028 } 4029 4030 /*@C 4031 MatCreateAIJ - Creates a sparse parallel matrix in AIJ format 4032 (the default parallel PETSc format). For good matrix assembly performance 4033 the user should preallocate the matrix storage by setting the parameters 4034 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 4035 performance can be increased by more than a factor of 50. 4036 4037 Collective on MPI_Comm 4038 4039 Input Parameters: 4040 + comm - MPI communicator 4041 . m - number of local rows (or PETSC_DECIDE to have calculated if M is given) 4042 This value should be the same as the local size used in creating the 4043 y vector for the matrix-vector product y = Ax. 4044 . n - This value should be the same as the local size used in creating the 4045 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 4046 calculated if N is given) For square matrices n is almost always m. 4047 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 4048 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 4049 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 4050 (same value is used for all local rows) 4051 . d_nnz - array containing the number of nonzeros in the various rows of the 4052 DIAGONAL portion of the local submatrix (possibly different for each row) 4053 or NULL, if d_nz is used to specify the nonzero structure. 4054 The size of this array is equal to the number of local rows, i.e 'm'. 4055 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 4056 submatrix (same value is used for all local rows). 4057 - o_nnz - array containing the number of nonzeros in the various rows of the 4058 OFF-DIAGONAL portion of the local submatrix (possibly different for 4059 each row) or NULL, if o_nz is used to specify the nonzero 4060 structure. The size of this array is equal to the number 4061 of local rows, i.e 'm'. 4062 4063 Output Parameter: 4064 . A - the matrix 4065 4066 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 4067 MatXXXXSetPreallocation() paradgm instead of this routine directly. 4068 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 4069 4070 Notes: 4071 If the *_nnz parameter is given then the *_nz parameter is ignored 4072 4073 m,n,M,N parameters specify the size of the matrix, and its partitioning across 4074 processors, while d_nz,d_nnz,o_nz,o_nnz parameters specify the approximate 4075 storage requirements for this matrix. 4076 4077 If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one 4078 processor than it must be used on all processors that share the object for 4079 that argument. 4080 4081 The user MUST specify either the local or global matrix dimensions 4082 (possibly both). 4083 4084 The parallel matrix is partitioned across processors such that the 4085 first m0 rows belong to process 0, the next m1 rows belong to 4086 process 1, the next m2 rows belong to process 2 etc.. where 4087 m0,m1,m2,.. are the input parameter 'm'. i.e each processor stores 4088 values corresponding to [m x N] submatrix. 4089 4090 The columns are logically partitioned with the n0 columns belonging 4091 to 0th partition, the next n1 columns belonging to the next 4092 partition etc.. where n0,n1,n2... are the input parameter 'n'. 4093 4094 The DIAGONAL portion of the local submatrix on any given processor 4095 is the submatrix corresponding to the rows and columns m,n 4096 corresponding to the given processor. i.e diagonal matrix on 4097 process 0 is [m0 x n0], diagonal matrix on process 1 is [m1 x n1] 4098 etc. The remaining portion of the local submatrix [m x (N-n)] 4099 constitute the OFF-DIAGONAL portion. The example below better 4100 illustrates this concept. 4101 4102 For a square global matrix we define each processor's diagonal portion 4103 to be its local rows and the corresponding columns (a square submatrix); 4104 each processor's off-diagonal portion encompasses the remainder of the 4105 local matrix (a rectangular submatrix). 4106 4107 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 4108 4109 When calling this routine with a single process communicator, a matrix of 4110 type SEQAIJ is returned. If a matrix of type MATMPIAIJ is desired for this 4111 type of communicator, use the construction mechanism: 4112 MatCreate(...,&A); MatSetType(A,MATMPIAIJ); MatSetSizes(A, m,n,M,N); MatMPIAIJSetPreallocation(A,...); 4113 4114 By default, this format uses inodes (identical nodes) when possible. 4115 We search for consecutive rows with the same nonzero structure, thereby 4116 reusing matrix information to achieve increased efficiency. 4117 4118 Options Database Keys: 4119 + -mat_no_inode - Do not use inodes 4120 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 4121 - -mat_aij_oneindex - Internally use indexing starting at 1 4122 rather than 0. Note that when calling MatSetValues(), 4123 the user still MUST index entries starting at 0! 4124 4125 4126 Example usage: 4127 4128 Consider the following 8x8 matrix with 34 non-zero values, that is 4129 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 4130 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 4131 as follows: 4132 4133 .vb 4134 1 2 0 | 0 3 0 | 0 4 4135 Proc0 0 5 6 | 7 0 0 | 8 0 4136 9 0 10 | 11 0 0 | 12 0 4137 ------------------------------------- 4138 13 0 14 | 15 16 17 | 0 0 4139 Proc1 0 18 0 | 19 20 21 | 0 0 4140 0 0 0 | 22 23 0 | 24 0 4141 ------------------------------------- 4142 Proc2 25 26 27 | 0 0 28 | 29 0 4143 30 0 0 | 31 32 33 | 0 34 4144 .ve 4145 4146 This can be represented as a collection of submatrices as: 4147 4148 .vb 4149 A B C 4150 D E F 4151 G H I 4152 .ve 4153 4154 Where the submatrices A,B,C are owned by proc0, D,E,F are 4155 owned by proc1, G,H,I are owned by proc2. 4156 4157 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 4158 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 4159 The 'M','N' parameters are 8,8, and have the same values on all procs. 4160 4161 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 4162 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 4163 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 4164 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 4165 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 4166 matrix, ans [DF] as another SeqAIJ matrix. 4167 4168 When d_nz, o_nz parameters are specified, d_nz storage elements are 4169 allocated for every row of the local diagonal submatrix, and o_nz 4170 storage locations are allocated for every row of the OFF-DIAGONAL submat. 4171 One way to choose d_nz and o_nz is to use the max nonzerors per local 4172 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 4173 In this case, the values of d_nz,o_nz are: 4174 .vb 4175 proc0 : dnz = 2, o_nz = 2 4176 proc1 : dnz = 3, o_nz = 2 4177 proc2 : dnz = 1, o_nz = 4 4178 .ve 4179 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 4180 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 4181 for proc3. i.e we are using 12+15+10=37 storage locations to store 4182 34 values. 4183 4184 When d_nnz, o_nnz parameters are specified, the storage is specified 4185 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 4186 In the above case the values for d_nnz,o_nnz are: 4187 .vb 4188 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 4189 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 4190 proc2: d_nnz = [1,1] and o_nnz = [4,4] 4191 .ve 4192 Here the space allocated is sum of all the above values i.e 34, and 4193 hence pre-allocation is perfect. 4194 4195 Level: intermediate 4196 4197 .keywords: matrix, aij, compressed row, sparse, parallel 4198 4199 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 4200 MATMPIAIJ, MatCreateMPIAIJWithArrays() 4201 @*/ 4202 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) 4203 { 4204 PetscErrorCode ierr; 4205 PetscMPIInt size; 4206 4207 PetscFunctionBegin; 4208 ierr = MatCreate(comm,A);CHKERRQ(ierr); 4209 ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr); 4210 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4211 if (size > 1) { 4212 ierr = MatSetType(*A,MATMPIAIJ);CHKERRQ(ierr); 4213 ierr = MatMPIAIJSetPreallocation(*A,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr); 4214 } else { 4215 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 4216 ierr = MatSeqAIJSetPreallocation(*A,d_nz,d_nnz);CHKERRQ(ierr); 4217 } 4218 PetscFunctionReturn(0); 4219 } 4220 4221 PetscErrorCode MatMPIAIJGetSeqAIJ(Mat A,Mat *Ad,Mat *Ao,const PetscInt *colmap[]) 4222 { 4223 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 4224 PetscBool flg; 4225 PetscErrorCode ierr; 4226 4227 PetscFunctionBegin; 4228 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&flg);CHKERRQ(ierr); 4229 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"This function requires a MATMPIAIJ matrix as input"); 4230 if (Ad) *Ad = a->A; 4231 if (Ao) *Ao = a->B; 4232 if (colmap) *colmap = a->garray; 4233 PetscFunctionReturn(0); 4234 } 4235 4236 PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat) 4237 { 4238 PetscErrorCode ierr; 4239 PetscInt m,N,i,rstart,nnz,Ii; 4240 PetscInt *indx; 4241 PetscScalar *values; 4242 4243 PetscFunctionBegin; 4244 ierr = MatGetSize(inmat,&m,&N);CHKERRQ(ierr); 4245 if (scall == MAT_INITIAL_MATRIX) { /* symbolic phase */ 4246 PetscInt *dnz,*onz,sum,bs,cbs; 4247 4248 if (n == PETSC_DECIDE) { 4249 ierr = PetscSplitOwnership(comm,&n,&N);CHKERRQ(ierr); 4250 } 4251 /* Check sum(n) = N */ 4252 ierr = MPIU_Allreduce(&n,&sum,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 4253 if (sum != N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Sum of local columns != global columns %d",N); 4254 4255 ierr = MPI_Scan(&m, &rstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 4256 rstart -= m; 4257 4258 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 4259 for (i=0; i<m; i++) { 4260 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,NULL);CHKERRQ(ierr); 4261 ierr = MatPreallocateSet(i+rstart,nnz,indx,dnz,onz);CHKERRQ(ierr); 4262 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,NULL);CHKERRQ(ierr); 4263 } 4264 4265 ierr = MatCreate(comm,outmat);CHKERRQ(ierr); 4266 ierr = MatSetSizes(*outmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 4267 ierr = MatGetBlockSizes(inmat,&bs,&cbs);CHKERRQ(ierr); 4268 ierr = MatSetBlockSizes(*outmat,bs,cbs);CHKERRQ(ierr); 4269 ierr = MatSetType(*outmat,MATAIJ);CHKERRQ(ierr); 4270 ierr = MatSeqAIJSetPreallocation(*outmat,0,dnz);CHKERRQ(ierr); 4271 ierr = MatMPIAIJSetPreallocation(*outmat,0,dnz,0,onz);CHKERRQ(ierr); 4272 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 4273 } 4274 4275 /* numeric phase */ 4276 ierr = MatGetOwnershipRange(*outmat,&rstart,NULL);CHKERRQ(ierr); 4277 for (i=0; i<m; i++) { 4278 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 4279 Ii = i + rstart; 4280 ierr = MatSetValues(*outmat,1,&Ii,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 4281 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 4282 } 4283 ierr = MatAssemblyBegin(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4284 ierr = MatAssemblyEnd(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4285 PetscFunctionReturn(0); 4286 } 4287 4288 PetscErrorCode MatFileSplit(Mat A,char *outfile) 4289 { 4290 PetscErrorCode ierr; 4291 PetscMPIInt rank; 4292 PetscInt m,N,i,rstart,nnz; 4293 size_t len; 4294 const PetscInt *indx; 4295 PetscViewer out; 4296 char *name; 4297 Mat B; 4298 const PetscScalar *values; 4299 4300 PetscFunctionBegin; 4301 ierr = MatGetLocalSize(A,&m,0);CHKERRQ(ierr); 4302 ierr = MatGetSize(A,0,&N);CHKERRQ(ierr); 4303 /* Should this be the type of the diagonal block of A? */ 4304 ierr = MatCreate(PETSC_COMM_SELF,&B);CHKERRQ(ierr); 4305 ierr = MatSetSizes(B,m,N,m,N);CHKERRQ(ierr); 4306 ierr = MatSetBlockSizesFromMats(B,A,A);CHKERRQ(ierr); 4307 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 4308 ierr = MatSeqAIJSetPreallocation(B,0,NULL);CHKERRQ(ierr); 4309 ierr = MatGetOwnershipRange(A,&rstart,0);CHKERRQ(ierr); 4310 for (i=0; i<m; i++) { 4311 ierr = MatGetRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 4312 ierr = MatSetValues(B,1,&i,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 4313 ierr = MatRestoreRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 4314 } 4315 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4316 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4317 4318 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)A),&rank);CHKERRQ(ierr); 4319 ierr = PetscStrlen(outfile,&len);CHKERRQ(ierr); 4320 ierr = PetscMalloc1(len+5,&name);CHKERRQ(ierr); 4321 sprintf(name,"%s.%d",outfile,rank); 4322 ierr = PetscViewerBinaryOpen(PETSC_COMM_SELF,name,FILE_MODE_APPEND,&out);CHKERRQ(ierr); 4323 ierr = PetscFree(name);CHKERRQ(ierr); 4324 ierr = MatView(B,out);CHKERRQ(ierr); 4325 ierr = PetscViewerDestroy(&out);CHKERRQ(ierr); 4326 ierr = MatDestroy(&B);CHKERRQ(ierr); 4327 PetscFunctionReturn(0); 4328 } 4329 4330 PetscErrorCode MatDestroy_MPIAIJ_SeqsToMPI(Mat A) 4331 { 4332 PetscErrorCode ierr; 4333 Mat_Merge_SeqsToMPI *merge; 4334 PetscContainer container; 4335 4336 PetscFunctionBegin; 4337 ierr = PetscObjectQuery((PetscObject)A,"MatMergeSeqsToMPI",(PetscObject*)&container);CHKERRQ(ierr); 4338 if (container) { 4339 ierr = PetscContainerGetPointer(container,(void**)&merge);CHKERRQ(ierr); 4340 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 4341 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 4342 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 4343 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 4344 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 4345 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 4346 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 4347 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 4348 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 4349 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 4350 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 4351 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 4352 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 4353 ierr = PetscFree(merge);CHKERRQ(ierr); 4354 ierr = PetscObjectCompose((PetscObject)A,"MatMergeSeqsToMPI",0);CHKERRQ(ierr); 4355 } 4356 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 4357 PetscFunctionReturn(0); 4358 } 4359 4360 #include <../src/mat/utils/freespace.h> 4361 #include <petscbt.h> 4362 4363 PetscErrorCode MatCreateMPIAIJSumSeqAIJNumeric(Mat seqmat,Mat mpimat) 4364 { 4365 PetscErrorCode ierr; 4366 MPI_Comm comm; 4367 Mat_SeqAIJ *a =(Mat_SeqAIJ*)seqmat->data; 4368 PetscMPIInt size,rank,taga,*len_s; 4369 PetscInt N=mpimat->cmap->N,i,j,*owners,*ai=a->i,*aj; 4370 PetscInt proc,m; 4371 PetscInt **buf_ri,**buf_rj; 4372 PetscInt k,anzi,*bj_i,*bi,*bj,arow,bnzi,nextaj; 4373 PetscInt nrows,**buf_ri_k,**nextrow,**nextai; 4374 MPI_Request *s_waits,*r_waits; 4375 MPI_Status *status; 4376 MatScalar *aa=a->a; 4377 MatScalar **abuf_r,*ba_i; 4378 Mat_Merge_SeqsToMPI *merge; 4379 PetscContainer container; 4380 4381 PetscFunctionBegin; 4382 ierr = PetscObjectGetComm((PetscObject)mpimat,&comm);CHKERRQ(ierr); 4383 ierr = PetscLogEventBegin(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4384 4385 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4386 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4387 4388 ierr = PetscObjectQuery((PetscObject)mpimat,"MatMergeSeqsToMPI",(PetscObject*)&container);CHKERRQ(ierr); 4389 ierr = PetscContainerGetPointer(container,(void**)&merge);CHKERRQ(ierr); 4390 4391 bi = merge->bi; 4392 bj = merge->bj; 4393 buf_ri = merge->buf_ri; 4394 buf_rj = merge->buf_rj; 4395 4396 ierr = PetscMalloc1(size,&status);CHKERRQ(ierr); 4397 owners = merge->rowmap->range; 4398 len_s = merge->len_s; 4399 4400 /* send and recv matrix values */ 4401 /*-----------------------------*/ 4402 ierr = PetscObjectGetNewTag((PetscObject)mpimat,&taga);CHKERRQ(ierr); 4403 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 4404 4405 ierr = PetscMalloc1(merge->nsend+1,&s_waits);CHKERRQ(ierr); 4406 for (proc=0,k=0; proc<size; proc++) { 4407 if (!len_s[proc]) continue; 4408 i = owners[proc]; 4409 ierr = MPI_Isend(aa+ai[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 4410 k++; 4411 } 4412 4413 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 4414 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 4415 ierr = PetscFree(status);CHKERRQ(ierr); 4416 4417 ierr = PetscFree(s_waits);CHKERRQ(ierr); 4418 ierr = PetscFree(r_waits);CHKERRQ(ierr); 4419 4420 /* insert mat values of mpimat */ 4421 /*----------------------------*/ 4422 ierr = PetscMalloc1(N,&ba_i);CHKERRQ(ierr); 4423 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextai);CHKERRQ(ierr); 4424 4425 for (k=0; k<merge->nrecv; k++) { 4426 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4427 nrows = *(buf_ri_k[k]); 4428 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 4429 nextai[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 4430 } 4431 4432 /* set values of ba */ 4433 m = merge->rowmap->n; 4434 for (i=0; i<m; i++) { 4435 arow = owners[rank] + i; 4436 bj_i = bj+bi[i]; /* col indices of the i-th row of mpimat */ 4437 bnzi = bi[i+1] - bi[i]; 4438 ierr = PetscMemzero(ba_i,bnzi*sizeof(PetscScalar));CHKERRQ(ierr); 4439 4440 /* add local non-zero vals of this proc's seqmat into ba */ 4441 anzi = ai[arow+1] - ai[arow]; 4442 aj = a->j + ai[arow]; 4443 aa = a->a + ai[arow]; 4444 nextaj = 0; 4445 for (j=0; nextaj<anzi; j++) { 4446 if (*(bj_i + j) == aj[nextaj]) { /* bcol == acol */ 4447 ba_i[j] += aa[nextaj++]; 4448 } 4449 } 4450 4451 /* add received vals into ba */ 4452 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 4453 /* i-th row */ 4454 if (i == *nextrow[k]) { 4455 anzi = *(nextai[k]+1) - *nextai[k]; 4456 aj = buf_rj[k] + *(nextai[k]); 4457 aa = abuf_r[k] + *(nextai[k]); 4458 nextaj = 0; 4459 for (j=0; nextaj<anzi; j++) { 4460 if (*(bj_i + j) == aj[nextaj]) { /* bcol == acol */ 4461 ba_i[j] += aa[nextaj++]; 4462 } 4463 } 4464 nextrow[k]++; nextai[k]++; 4465 } 4466 } 4467 ierr = MatSetValues(mpimat,1,&arow,bnzi,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 4468 } 4469 ierr = MatAssemblyBegin(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4470 ierr = MatAssemblyEnd(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4471 4472 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 4473 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 4474 ierr = PetscFree(ba_i);CHKERRQ(ierr); 4475 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4476 ierr = PetscLogEventEnd(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4477 PetscFunctionReturn(0); 4478 } 4479 4480 PetscErrorCode MatCreateMPIAIJSumSeqAIJSymbolic(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,Mat *mpimat) 4481 { 4482 PetscErrorCode ierr; 4483 Mat B_mpi; 4484 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 4485 PetscMPIInt size,rank,tagi,tagj,*len_s,*len_si,*len_ri; 4486 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 4487 PetscInt M=seqmat->rmap->n,N=seqmat->cmap->n,i,*owners,*ai=a->i,*aj=a->j; 4488 PetscInt len,proc,*dnz,*onz,bs,cbs; 4489 PetscInt k,anzi,*bi,*bj,*lnk,nlnk,arow,bnzi,nspacedouble=0; 4490 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextai; 4491 MPI_Request *si_waits,*sj_waits,*ri_waits,*rj_waits; 4492 MPI_Status *status; 4493 PetscFreeSpaceList free_space=NULL,current_space=NULL; 4494 PetscBT lnkbt; 4495 Mat_Merge_SeqsToMPI *merge; 4496 PetscContainer container; 4497 4498 PetscFunctionBegin; 4499 ierr = PetscLogEventBegin(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4500 4501 /* make sure it is a PETSc comm */ 4502 ierr = PetscCommDuplicate(comm,&comm,NULL);CHKERRQ(ierr); 4503 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4504 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4505 4506 ierr = PetscNew(&merge);CHKERRQ(ierr); 4507 ierr = PetscMalloc1(size,&status);CHKERRQ(ierr); 4508 4509 /* determine row ownership */ 4510 /*---------------------------------------------------------*/ 4511 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 4512 ierr = PetscLayoutSetLocalSize(merge->rowmap,m);CHKERRQ(ierr); 4513 ierr = PetscLayoutSetSize(merge->rowmap,M);CHKERRQ(ierr); 4514 ierr = PetscLayoutSetBlockSize(merge->rowmap,1);CHKERRQ(ierr); 4515 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 4516 ierr = PetscMalloc1(size,&len_si);CHKERRQ(ierr); 4517 ierr = PetscMalloc1(size,&merge->len_s);CHKERRQ(ierr); 4518 4519 m = merge->rowmap->n; 4520 owners = merge->rowmap->range; 4521 4522 /* determine the number of messages to send, their lengths */ 4523 /*---------------------------------------------------------*/ 4524 len_s = merge->len_s; 4525 4526 len = 0; /* length of buf_si[] */ 4527 merge->nsend = 0; 4528 for (proc=0; proc<size; proc++) { 4529 len_si[proc] = 0; 4530 if (proc == rank) { 4531 len_s[proc] = 0; 4532 } else { 4533 len_si[proc] = owners[proc+1] - owners[proc] + 1; 4534 len_s[proc] = ai[owners[proc+1]] - ai[owners[proc]]; /* num of rows to be sent to [proc] */ 4535 } 4536 if (len_s[proc]) { 4537 merge->nsend++; 4538 nrows = 0; 4539 for (i=owners[proc]; i<owners[proc+1]; i++) { 4540 if (ai[i+1] > ai[i]) nrows++; 4541 } 4542 len_si[proc] = 2*(nrows+1); 4543 len += len_si[proc]; 4544 } 4545 } 4546 4547 /* determine the number and length of messages to receive for ij-structure */ 4548 /*-------------------------------------------------------------------------*/ 4549 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 4550 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 4551 4552 /* post the Irecv of j-structure */ 4553 /*-------------------------------*/ 4554 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 4555 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rj_waits);CHKERRQ(ierr); 4556 4557 /* post the Isend of j-structure */ 4558 /*--------------------------------*/ 4559 ierr = PetscMalloc2(merge->nsend,&si_waits,merge->nsend,&sj_waits);CHKERRQ(ierr); 4560 4561 for (proc=0, k=0; proc<size; proc++) { 4562 if (!len_s[proc]) continue; 4563 i = owners[proc]; 4564 ierr = MPI_Isend(aj+ai[i],len_s[proc],MPIU_INT,proc,tagj,comm,sj_waits+k);CHKERRQ(ierr); 4565 k++; 4566 } 4567 4568 /* receives and sends of j-structure are complete */ 4569 /*------------------------------------------------*/ 4570 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,rj_waits,status);CHKERRQ(ierr);} 4571 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,sj_waits,status);CHKERRQ(ierr);} 4572 4573 /* send and recv i-structure */ 4574 /*---------------------------*/ 4575 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 4576 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&ri_waits);CHKERRQ(ierr); 4577 4578 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 4579 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 4580 for (proc=0,k=0; proc<size; proc++) { 4581 if (!len_s[proc]) continue; 4582 /* form outgoing message for i-structure: 4583 buf_si[0]: nrows to be sent 4584 [1:nrows]: row index (global) 4585 [nrows+1:2*nrows+1]: i-structure index 4586 */ 4587 /*-------------------------------------------*/ 4588 nrows = len_si[proc]/2 - 1; 4589 buf_si_i = buf_si + nrows+1; 4590 buf_si[0] = nrows; 4591 buf_si_i[0] = 0; 4592 nrows = 0; 4593 for (i=owners[proc]; i<owners[proc+1]; i++) { 4594 anzi = ai[i+1] - ai[i]; 4595 if (anzi) { 4596 buf_si_i[nrows+1] = buf_si_i[nrows] + anzi; /* i-structure */ 4597 buf_si[nrows+1] = i-owners[proc]; /* local row index */ 4598 nrows++; 4599 } 4600 } 4601 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,si_waits+k);CHKERRQ(ierr); 4602 k++; 4603 buf_si += len_si[proc]; 4604 } 4605 4606 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,ri_waits,status);CHKERRQ(ierr);} 4607 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,si_waits,status);CHKERRQ(ierr);} 4608 4609 ierr = PetscInfo2(seqmat,"nsend: %D, nrecv: %D\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 4610 for (i=0; i<merge->nrecv; i++) { 4611 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); 4612 } 4613 4614 ierr = PetscFree(len_si);CHKERRQ(ierr); 4615 ierr = PetscFree(len_ri);CHKERRQ(ierr); 4616 ierr = PetscFree(rj_waits);CHKERRQ(ierr); 4617 ierr = PetscFree2(si_waits,sj_waits);CHKERRQ(ierr); 4618 ierr = PetscFree(ri_waits);CHKERRQ(ierr); 4619 ierr = PetscFree(buf_s);CHKERRQ(ierr); 4620 ierr = PetscFree(status);CHKERRQ(ierr); 4621 4622 /* compute a local seq matrix in each processor */ 4623 /*----------------------------------------------*/ 4624 /* allocate bi array and free space for accumulating nonzero column info */ 4625 ierr = PetscMalloc1(m+1,&bi);CHKERRQ(ierr); 4626 bi[0] = 0; 4627 4628 /* create and initialize a linked list */ 4629 nlnk = N+1; 4630 ierr = PetscLLCreate(N,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4631 4632 /* initial FreeSpace size is 2*(num of local nnz(seqmat)) */ 4633 len = ai[owners[rank+1]] - ai[owners[rank]]; 4634 ierr = PetscFreeSpaceGet(PetscIntMultTruncate(2,len)+1,&free_space);CHKERRQ(ierr); 4635 4636 current_space = free_space; 4637 4638 /* determine symbolic info for each local row */ 4639 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextai);CHKERRQ(ierr); 4640 4641 for (k=0; k<merge->nrecv; k++) { 4642 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4643 nrows = *buf_ri_k[k]; 4644 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 4645 nextai[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 4646 } 4647 4648 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 4649 len = 0; 4650 for (i=0; i<m; i++) { 4651 bnzi = 0; 4652 /* add local non-zero cols of this proc's seqmat into lnk */ 4653 arow = owners[rank] + i; 4654 anzi = ai[arow+1] - ai[arow]; 4655 aj = a->j + ai[arow]; 4656 ierr = PetscLLAddSorted(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4657 bnzi += nlnk; 4658 /* add received col data into lnk */ 4659 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 4660 if (i == *nextrow[k]) { /* i-th row */ 4661 anzi = *(nextai[k]+1) - *nextai[k]; 4662 aj = buf_rj[k] + *nextai[k]; 4663 ierr = PetscLLAddSorted(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4664 bnzi += nlnk; 4665 nextrow[k]++; nextai[k]++; 4666 } 4667 } 4668 if (len < bnzi) len = bnzi; /* =max(bnzi) */ 4669 4670 /* if free space is not available, make more free space */ 4671 if (current_space->local_remaining<bnzi) { 4672 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(bnzi,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 4673 nspacedouble++; 4674 } 4675 /* copy data into free space, then initialize lnk */ 4676 ierr = PetscLLClean(N,N,bnzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 4677 ierr = MatPreallocateSet(i+owners[rank],bnzi,current_space->array,dnz,onz);CHKERRQ(ierr); 4678 4679 current_space->array += bnzi; 4680 current_space->local_used += bnzi; 4681 current_space->local_remaining -= bnzi; 4682 4683 bi[i+1] = bi[i] + bnzi; 4684 } 4685 4686 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4687 4688 ierr = PetscMalloc1(bi[m]+1,&bj);CHKERRQ(ierr); 4689 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 4690 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 4691 4692 /* create symbolic parallel matrix B_mpi */ 4693 /*---------------------------------------*/ 4694 ierr = MatGetBlockSizes(seqmat,&bs,&cbs);CHKERRQ(ierr); 4695 ierr = MatCreate(comm,&B_mpi);CHKERRQ(ierr); 4696 if (n==PETSC_DECIDE) { 4697 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,N);CHKERRQ(ierr); 4698 } else { 4699 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 4700 } 4701 ierr = MatSetBlockSizes(B_mpi,bs,cbs);CHKERRQ(ierr); 4702 ierr = MatSetType(B_mpi,MATMPIAIJ);CHKERRQ(ierr); 4703 ierr = MatMPIAIJSetPreallocation(B_mpi,0,dnz,0,onz);CHKERRQ(ierr); 4704 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 4705 ierr = MatSetOption(B_mpi,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr); 4706 4707 /* B_mpi is not ready for use - assembly will be done by MatCreateMPIAIJSumSeqAIJNumeric() */ 4708 B_mpi->assembled = PETSC_FALSE; 4709 B_mpi->ops->destroy = MatDestroy_MPIAIJ_SeqsToMPI; 4710 merge->bi = bi; 4711 merge->bj = bj; 4712 merge->buf_ri = buf_ri; 4713 merge->buf_rj = buf_rj; 4714 merge->coi = NULL; 4715 merge->coj = NULL; 4716 merge->owners_co = NULL; 4717 4718 ierr = PetscCommDestroy(&comm);CHKERRQ(ierr); 4719 4720 /* attach the supporting struct to B_mpi for reuse */ 4721 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 4722 ierr = PetscContainerSetPointer(container,merge);CHKERRQ(ierr); 4723 ierr = PetscObjectCompose((PetscObject)B_mpi,"MatMergeSeqsToMPI",(PetscObject)container);CHKERRQ(ierr); 4724 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 4725 *mpimat = B_mpi; 4726 4727 ierr = PetscLogEventEnd(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4728 PetscFunctionReturn(0); 4729 } 4730 4731 /*@C 4732 MatCreateMPIAIJSumSeqAIJ - Creates a MATMPIAIJ matrix by adding sequential 4733 matrices from each processor 4734 4735 Collective on MPI_Comm 4736 4737 Input Parameters: 4738 + comm - the communicators the parallel matrix will live on 4739 . seqmat - the input sequential matrices 4740 . m - number of local rows (or PETSC_DECIDE) 4741 . n - number of local columns (or PETSC_DECIDE) 4742 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4743 4744 Output Parameter: 4745 . mpimat - the parallel matrix generated 4746 4747 Level: advanced 4748 4749 Notes: 4750 The dimensions of the sequential matrix in each processor MUST be the same. 4751 The input seqmat is included into the container "Mat_Merge_SeqsToMPI", and will be 4752 destroyed when mpimat is destroyed. Call PetscObjectQuery() to access seqmat. 4753 @*/ 4754 PetscErrorCode MatCreateMPIAIJSumSeqAIJ(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,MatReuse scall,Mat *mpimat) 4755 { 4756 PetscErrorCode ierr; 4757 PetscMPIInt size; 4758 4759 PetscFunctionBegin; 4760 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4761 if (size == 1) { 4762 ierr = PetscLogEventBegin(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4763 if (scall == MAT_INITIAL_MATRIX) { 4764 ierr = MatDuplicate(seqmat,MAT_COPY_VALUES,mpimat);CHKERRQ(ierr); 4765 } else { 4766 ierr = MatCopy(seqmat,*mpimat,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 4767 } 4768 ierr = PetscLogEventEnd(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4769 PetscFunctionReturn(0); 4770 } 4771 ierr = PetscLogEventBegin(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4772 if (scall == MAT_INITIAL_MATRIX) { 4773 ierr = MatCreateMPIAIJSumSeqAIJSymbolic(comm,seqmat,m,n,mpimat);CHKERRQ(ierr); 4774 } 4775 ierr = MatCreateMPIAIJSumSeqAIJNumeric(seqmat,*mpimat);CHKERRQ(ierr); 4776 ierr = PetscLogEventEnd(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4777 PetscFunctionReturn(0); 4778 } 4779 4780 /*@ 4781 MatMPIAIJGetLocalMat - Creates a SeqAIJ from a MATMPIAIJ matrix by taking all its local rows and putting them into a sequential vector with 4782 mlocal rows and n columns. Where mlocal is the row count obtained with MatGetLocalSize() and n is the global column count obtained 4783 with MatGetSize() 4784 4785 Not Collective 4786 4787 Input Parameters: 4788 + A - the matrix 4789 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4790 4791 Output Parameter: 4792 . A_loc - the local sequential matrix generated 4793 4794 Level: developer 4795 4796 .seealso: MatGetOwnerShipRange(), MatMPIAIJGetLocalMatCondensed() 4797 4798 @*/ 4799 PetscErrorCode MatMPIAIJGetLocalMat(Mat A,MatReuse scall,Mat *A_loc) 4800 { 4801 PetscErrorCode ierr; 4802 Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data; 4803 Mat_SeqAIJ *mat,*a,*b; 4804 PetscInt *ai,*aj,*bi,*bj,*cmap=mpimat->garray; 4805 MatScalar *aa,*ba,*cam; 4806 PetscScalar *ca; 4807 PetscInt am=A->rmap->n,i,j,k,cstart=A->cmap->rstart; 4808 PetscInt *ci,*cj,col,ncols_d,ncols_o,jo; 4809 PetscBool match; 4810 MPI_Comm comm; 4811 PetscMPIInt size; 4812 4813 PetscFunctionBegin; 4814 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&match);CHKERRQ(ierr); 4815 if (!match) SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_SUP,"Requires MATMPIAIJ matrix as input"); 4816 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 4817 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4818 if (size == 1 && scall == MAT_REUSE_MATRIX) PetscFunctionReturn(0); 4819 4820 ierr = PetscLogEventBegin(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4821 a = (Mat_SeqAIJ*)(mpimat->A)->data; 4822 b = (Mat_SeqAIJ*)(mpimat->B)->data; 4823 ai = a->i; aj = a->j; bi = b->i; bj = b->j; 4824 aa = a->a; ba = b->a; 4825 if (scall == MAT_INITIAL_MATRIX) { 4826 if (size == 1) { 4827 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap->N,ai,aj,aa,A_loc);CHKERRQ(ierr); 4828 PetscFunctionReturn(0); 4829 } 4830 4831 ierr = PetscMalloc1(1+am,&ci);CHKERRQ(ierr); 4832 ci[0] = 0; 4833 for (i=0; i<am; i++) { 4834 ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]); 4835 } 4836 ierr = PetscMalloc1(1+ci[am],&cj);CHKERRQ(ierr); 4837 ierr = PetscMalloc1(1+ci[am],&ca);CHKERRQ(ierr); 4838 k = 0; 4839 for (i=0; i<am; i++) { 4840 ncols_o = bi[i+1] - bi[i]; 4841 ncols_d = ai[i+1] - ai[i]; 4842 /* off-diagonal portion of A */ 4843 for (jo=0; jo<ncols_o; jo++) { 4844 col = cmap[*bj]; 4845 if (col >= cstart) break; 4846 cj[k] = col; bj++; 4847 ca[k++] = *ba++; 4848 } 4849 /* diagonal portion of A */ 4850 for (j=0; j<ncols_d; j++) { 4851 cj[k] = cstart + *aj++; 4852 ca[k++] = *aa++; 4853 } 4854 /* off-diagonal portion of A */ 4855 for (j=jo; j<ncols_o; j++) { 4856 cj[k] = cmap[*bj++]; 4857 ca[k++] = *ba++; 4858 } 4859 } 4860 /* put together the new matrix */ 4861 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap->N,ci,cj,ca,A_loc);CHKERRQ(ierr); 4862 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4863 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4864 mat = (Mat_SeqAIJ*)(*A_loc)->data; 4865 mat->free_a = PETSC_TRUE; 4866 mat->free_ij = PETSC_TRUE; 4867 mat->nonew = 0; 4868 } else if (scall == MAT_REUSE_MATRIX) { 4869 mat=(Mat_SeqAIJ*)(*A_loc)->data; 4870 ci = mat->i; cj = mat->j; cam = mat->a; 4871 for (i=0; i<am; i++) { 4872 /* off-diagonal portion of A */ 4873 ncols_o = bi[i+1] - bi[i]; 4874 for (jo=0; jo<ncols_o; jo++) { 4875 col = cmap[*bj]; 4876 if (col >= cstart) break; 4877 *cam++ = *ba++; bj++; 4878 } 4879 /* diagonal portion of A */ 4880 ncols_d = ai[i+1] - ai[i]; 4881 for (j=0; j<ncols_d; j++) *cam++ = *aa++; 4882 /* off-diagonal portion of A */ 4883 for (j=jo; j<ncols_o; j++) { 4884 *cam++ = *ba++; bj++; 4885 } 4886 } 4887 } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 4888 ierr = PetscLogEventEnd(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4889 PetscFunctionReturn(0); 4890 } 4891 4892 /*@C 4893 MatMPIAIJGetLocalMatCondensed - Creates a SeqAIJ matrix from an MATMPIAIJ matrix by taking all its local rows and NON-ZERO columns 4894 4895 Not Collective 4896 4897 Input Parameters: 4898 + A - the matrix 4899 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4900 - row, col - index sets of rows and columns to extract (or NULL) 4901 4902 Output Parameter: 4903 . A_loc - the local sequential matrix generated 4904 4905 Level: developer 4906 4907 .seealso: MatGetOwnershipRange(), MatMPIAIJGetLocalMat() 4908 4909 @*/ 4910 PetscErrorCode MatMPIAIJGetLocalMatCondensed(Mat A,MatReuse scall,IS *row,IS *col,Mat *A_loc) 4911 { 4912 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4913 PetscErrorCode ierr; 4914 PetscInt i,start,end,ncols,nzA,nzB,*cmap,imark,*idx; 4915 IS isrowa,iscola; 4916 Mat *aloc; 4917 PetscBool match; 4918 4919 PetscFunctionBegin; 4920 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&match);CHKERRQ(ierr); 4921 if (!match) SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_SUP,"Requires MATMPIAIJ matrix as input"); 4922 ierr = PetscLogEventBegin(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4923 if (!row) { 4924 start = A->rmap->rstart; end = A->rmap->rend; 4925 ierr = ISCreateStride(PETSC_COMM_SELF,end-start,start,1,&isrowa);CHKERRQ(ierr); 4926 } else { 4927 isrowa = *row; 4928 } 4929 if (!col) { 4930 start = A->cmap->rstart; 4931 cmap = a->garray; 4932 nzA = a->A->cmap->n; 4933 nzB = a->B->cmap->n; 4934 ierr = PetscMalloc1(nzA+nzB, &idx);CHKERRQ(ierr); 4935 ncols = 0; 4936 for (i=0; i<nzB; i++) { 4937 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4938 else break; 4939 } 4940 imark = i; 4941 for (i=0; i<nzA; i++) idx[ncols++] = start + i; 4942 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; 4943 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&iscola);CHKERRQ(ierr); 4944 } else { 4945 iscola = *col; 4946 } 4947 if (scall != MAT_INITIAL_MATRIX) { 4948 ierr = PetscMalloc1(1,&aloc);CHKERRQ(ierr); 4949 aloc[0] = *A_loc; 4950 } 4951 ierr = MatCreateSubMatrices(A,1,&isrowa,&iscola,scall,&aloc);CHKERRQ(ierr); 4952 *A_loc = aloc[0]; 4953 ierr = PetscFree(aloc);CHKERRQ(ierr); 4954 if (!row) { 4955 ierr = ISDestroy(&isrowa);CHKERRQ(ierr); 4956 } 4957 if (!col) { 4958 ierr = ISDestroy(&iscola);CHKERRQ(ierr); 4959 } 4960 ierr = PetscLogEventEnd(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4961 PetscFunctionReturn(0); 4962 } 4963 4964 /*@C 4965 MatGetBrowsOfAcols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns of local A 4966 4967 Collective on Mat 4968 4969 Input Parameters: 4970 + A,B - the matrices in mpiaij format 4971 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4972 - rowb, colb - index sets of rows and columns of B to extract (or NULL) 4973 4974 Output Parameter: 4975 + rowb, colb - index sets of rows and columns of B to extract 4976 - B_seq - the sequential matrix generated 4977 4978 Level: developer 4979 4980 @*/ 4981 PetscErrorCode MatGetBrowsOfAcols(Mat A,Mat B,MatReuse scall,IS *rowb,IS *colb,Mat *B_seq) 4982 { 4983 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4984 PetscErrorCode ierr; 4985 PetscInt *idx,i,start,ncols,nzA,nzB,*cmap,imark; 4986 IS isrowb,iscolb; 4987 Mat *bseq=NULL; 4988 4989 PetscFunctionBegin; 4990 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) { 4991 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); 4992 } 4993 ierr = PetscLogEventBegin(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4994 4995 if (scall == MAT_INITIAL_MATRIX) { 4996 start = A->cmap->rstart; 4997 cmap = a->garray; 4998 nzA = a->A->cmap->n; 4999 nzB = a->B->cmap->n; 5000 ierr = PetscMalloc1(nzA+nzB, &idx);CHKERRQ(ierr); 5001 ncols = 0; 5002 for (i=0; i<nzB; i++) { /* row < local row index */ 5003 if (cmap[i] < start) idx[ncols++] = cmap[i]; 5004 else break; 5005 } 5006 imark = i; 5007 for (i=0; i<nzA; i++) idx[ncols++] = start + i; /* local rows */ 5008 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; /* row > local row index */ 5009 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&isrowb);CHKERRQ(ierr); 5010 ierr = ISCreateStride(PETSC_COMM_SELF,B->cmap->N,0,1,&iscolb);CHKERRQ(ierr); 5011 } else { 5012 if (!rowb || !colb) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"IS rowb and colb must be provided for MAT_REUSE_MATRIX"); 5013 isrowb = *rowb; iscolb = *colb; 5014 ierr = PetscMalloc1(1,&bseq);CHKERRQ(ierr); 5015 bseq[0] = *B_seq; 5016 } 5017 ierr = MatCreateSubMatrices(B,1,&isrowb,&iscolb,scall,&bseq);CHKERRQ(ierr); 5018 *B_seq = bseq[0]; 5019 ierr = PetscFree(bseq);CHKERRQ(ierr); 5020 if (!rowb) { 5021 ierr = ISDestroy(&isrowb);CHKERRQ(ierr); 5022 } else { 5023 *rowb = isrowb; 5024 } 5025 if (!colb) { 5026 ierr = ISDestroy(&iscolb);CHKERRQ(ierr); 5027 } else { 5028 *colb = iscolb; 5029 } 5030 ierr = PetscLogEventEnd(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 5031 PetscFunctionReturn(0); 5032 } 5033 5034 /* 5035 MatGetBrowsOfAoCols_MPIAIJ - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns 5036 of the OFF-DIAGONAL portion of local A 5037 5038 Collective on Mat 5039 5040 Input Parameters: 5041 + A,B - the matrices in mpiaij format 5042 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 5043 5044 Output Parameter: 5045 + startsj_s - starting point in B's sending j-arrays, saved for MAT_REUSE (or NULL) 5046 . startsj_r - starting point in B's receiving j-arrays, saved for MAT_REUSE (or NULL) 5047 . bufa_ptr - array for sending matrix values, saved for MAT_REUSE (or NULL) 5048 - B_oth - the sequential matrix generated with size aBn=a->B->cmap->n by B->cmap->N 5049 5050 Level: developer 5051 5052 */ 5053 PetscErrorCode MatGetBrowsOfAoCols_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscInt **startsj_s,PetscInt **startsj_r,MatScalar **bufa_ptr,Mat *B_oth) 5054 { 5055 VecScatter_MPI_General *gen_to,*gen_from; 5056 PetscErrorCode ierr; 5057 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 5058 Mat_SeqAIJ *b_oth; 5059 VecScatter ctx =a->Mvctx; 5060 MPI_Comm comm; 5061 PetscMPIInt *rprocs,*sprocs,tag=((PetscObject)ctx)->tag,rank; 5062 PetscInt *rowlen,*bufj,*bufJ,ncols,aBn=a->B->cmap->n,row,*b_othi,*b_othj; 5063 PetscInt *rvalues,*svalues; 5064 MatScalar *b_otha,*bufa,*bufA; 5065 PetscInt i,j,k,l,ll,nrecvs,nsends,nrows,*srow,*rstarts,*rstartsj = 0,*sstarts,*sstartsj,len; 5066 MPI_Request *rwaits = NULL,*swaits = NULL; 5067 MPI_Status *sstatus,rstatus; 5068 PetscMPIInt jj,size; 5069 PetscInt *cols,sbs,rbs; 5070 PetscScalar *vals; 5071 5072 PetscFunctionBegin; 5073 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 5074 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 5075 5076 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend) { 5077 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); 5078 } 5079 ierr = PetscLogEventBegin(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 5080 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 5081 5082 if (size == 1) { 5083 startsj_s = NULL; 5084 bufa_ptr = NULL; 5085 *B_oth = NULL; 5086 PetscFunctionReturn(0); 5087 } 5088 5089 gen_to = (VecScatter_MPI_General*)ctx->todata; 5090 gen_from = (VecScatter_MPI_General*)ctx->fromdata; 5091 nrecvs = gen_from->n; 5092 nsends = gen_to->n; 5093 5094 ierr = PetscMalloc2(nrecvs,&rwaits,nsends,&swaits);CHKERRQ(ierr); 5095 srow = gen_to->indices; /* local row index to be sent */ 5096 sstarts = gen_to->starts; 5097 sprocs = gen_to->procs; 5098 sstatus = gen_to->sstatus; 5099 sbs = gen_to->bs; 5100 rstarts = gen_from->starts; 5101 rprocs = gen_from->procs; 5102 rbs = gen_from->bs; 5103 5104 if (!startsj_s || !bufa_ptr) scall = MAT_INITIAL_MATRIX; 5105 if (scall == MAT_INITIAL_MATRIX) { 5106 /* i-array */ 5107 /*---------*/ 5108 /* post receives */ 5109 ierr = PetscMalloc1(rbs*(rstarts[nrecvs] - rstarts[0]),&rvalues);CHKERRQ(ierr); 5110 for (i=0; i<nrecvs; i++) { 5111 rowlen = rvalues + rstarts[i]*rbs; 5112 nrows = (rstarts[i+1]-rstarts[i])*rbs; /* num of indices to be received */ 5113 ierr = MPI_Irecv(rowlen,nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5114 } 5115 5116 /* pack the outgoing message */ 5117 ierr = PetscMalloc2(nsends+1,&sstartsj,nrecvs+1,&rstartsj);CHKERRQ(ierr); 5118 5119 sstartsj[0] = 0; 5120 rstartsj[0] = 0; 5121 len = 0; /* total length of j or a array to be sent */ 5122 k = 0; 5123 ierr = PetscMalloc1(sbs*(sstarts[nsends] - sstarts[0]),&svalues);CHKERRQ(ierr); 5124 for (i=0; i<nsends; i++) { 5125 rowlen = svalues + sstarts[i]*sbs; 5126 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5127 for (j=0; j<nrows; j++) { 5128 row = srow[k] + B->rmap->range[rank]; /* global row idx */ 5129 for (l=0; l<sbs; l++) { 5130 ierr = MatGetRow_MPIAIJ(B,row+l,&ncols,NULL,NULL);CHKERRQ(ierr); /* rowlength */ 5131 5132 rowlen[j*sbs+l] = ncols; 5133 5134 len += ncols; 5135 ierr = MatRestoreRow_MPIAIJ(B,row+l,&ncols,NULL,NULL);CHKERRQ(ierr); 5136 } 5137 k++; 5138 } 5139 ierr = MPI_Isend(rowlen,nrows*sbs,MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5140 5141 sstartsj[i+1] = len; /* starting point of (i+1)-th outgoing msg in bufj and bufa */ 5142 } 5143 /* recvs and sends of i-array are completed */ 5144 i = nrecvs; 5145 while (i--) { 5146 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5147 } 5148 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5149 ierr = PetscFree(svalues);CHKERRQ(ierr); 5150 5151 /* allocate buffers for sending j and a arrays */ 5152 ierr = PetscMalloc1(len+1,&bufj);CHKERRQ(ierr); 5153 ierr = PetscMalloc1(len+1,&bufa);CHKERRQ(ierr); 5154 5155 /* create i-array of B_oth */ 5156 ierr = PetscMalloc1(aBn+2,&b_othi);CHKERRQ(ierr); 5157 5158 b_othi[0] = 0; 5159 len = 0; /* total length of j or a array to be received */ 5160 k = 0; 5161 for (i=0; i<nrecvs; i++) { 5162 rowlen = rvalues + rstarts[i]*rbs; 5163 nrows = rbs*(rstarts[i+1]-rstarts[i]); /* num of rows to be received */ 5164 for (j=0; j<nrows; j++) { 5165 b_othi[k+1] = b_othi[k] + rowlen[j]; 5166 ierr = PetscIntSumError(rowlen[j],len,&len);CHKERRQ(ierr); 5167 k++; 5168 } 5169 rstartsj[i+1] = len; /* starting point of (i+1)-th incoming msg in bufj and bufa */ 5170 } 5171 ierr = PetscFree(rvalues);CHKERRQ(ierr); 5172 5173 /* allocate space for j and a arrrays of B_oth */ 5174 ierr = PetscMalloc1(b_othi[aBn]+1,&b_othj);CHKERRQ(ierr); 5175 ierr = PetscMalloc1(b_othi[aBn]+1,&b_otha);CHKERRQ(ierr); 5176 5177 /* j-array */ 5178 /*---------*/ 5179 /* post receives of j-array */ 5180 for (i=0; i<nrecvs; i++) { 5181 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 5182 ierr = MPI_Irecv(b_othj+rstartsj[i],nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5183 } 5184 5185 /* pack the outgoing message j-array */ 5186 k = 0; 5187 for (i=0; i<nsends; i++) { 5188 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5189 bufJ = bufj+sstartsj[i]; 5190 for (j=0; j<nrows; j++) { 5191 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 5192 for (ll=0; ll<sbs; ll++) { 5193 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,&cols,NULL);CHKERRQ(ierr); 5194 for (l=0; l<ncols; l++) { 5195 *bufJ++ = cols[l]; 5196 } 5197 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,&cols,NULL);CHKERRQ(ierr); 5198 } 5199 } 5200 ierr = MPI_Isend(bufj+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5201 } 5202 5203 /* recvs and sends of j-array are completed */ 5204 i = nrecvs; 5205 while (i--) { 5206 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5207 } 5208 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5209 } else if (scall == MAT_REUSE_MATRIX) { 5210 sstartsj = *startsj_s; 5211 rstartsj = *startsj_r; 5212 bufa = *bufa_ptr; 5213 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 5214 b_otha = b_oth->a; 5215 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE, "Matrix P does not posses an object container"); 5216 5217 /* a-array */ 5218 /*---------*/ 5219 /* post receives of a-array */ 5220 for (i=0; i<nrecvs; i++) { 5221 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 5222 ierr = MPI_Irecv(b_otha+rstartsj[i],nrows,MPIU_SCALAR,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5223 } 5224 5225 /* pack the outgoing message a-array */ 5226 k = 0; 5227 for (i=0; i<nsends; i++) { 5228 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5229 bufA = bufa+sstartsj[i]; 5230 for (j=0; j<nrows; j++) { 5231 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 5232 for (ll=0; ll<sbs; ll++) { 5233 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,NULL,&vals);CHKERRQ(ierr); 5234 for (l=0; l<ncols; l++) { 5235 *bufA++ = vals[l]; 5236 } 5237 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,NULL,&vals);CHKERRQ(ierr); 5238 } 5239 } 5240 ierr = MPI_Isend(bufa+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_SCALAR,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5241 } 5242 /* recvs and sends of a-array are completed */ 5243 i = nrecvs; 5244 while (i--) { 5245 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5246 } 5247 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5248 ierr = PetscFree2(rwaits,swaits);CHKERRQ(ierr); 5249 5250 if (scall == MAT_INITIAL_MATRIX) { 5251 /* put together the new matrix */ 5252 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,aBn,B->cmap->N,b_othi,b_othj,b_otha,B_oth);CHKERRQ(ierr); 5253 5254 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 5255 /* Since these are PETSc arrays, change flags to free them as necessary. */ 5256 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 5257 b_oth->free_a = PETSC_TRUE; 5258 b_oth->free_ij = PETSC_TRUE; 5259 b_oth->nonew = 0; 5260 5261 ierr = PetscFree(bufj);CHKERRQ(ierr); 5262 if (!startsj_s || !bufa_ptr) { 5263 ierr = PetscFree2(sstartsj,rstartsj);CHKERRQ(ierr); 5264 ierr = PetscFree(bufa_ptr);CHKERRQ(ierr); 5265 } else { 5266 *startsj_s = sstartsj; 5267 *startsj_r = rstartsj; 5268 *bufa_ptr = bufa; 5269 } 5270 } 5271 ierr = PetscLogEventEnd(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 5272 PetscFunctionReturn(0); 5273 } 5274 5275 /*@C 5276 MatGetCommunicationStructs - Provides access to the communication structures used in matrix-vector multiplication. 5277 5278 Not Collective 5279 5280 Input Parameters: 5281 . A - The matrix in mpiaij format 5282 5283 Output Parameter: 5284 + lvec - The local vector holding off-process values from the argument to a matrix-vector product 5285 . colmap - A map from global column index to local index into lvec 5286 - multScatter - A scatter from the argument of a matrix-vector product to lvec 5287 5288 Level: developer 5289 5290 @*/ 5291 #if defined(PETSC_USE_CTABLE) 5292 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscTable *colmap, VecScatter *multScatter) 5293 #else 5294 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscInt *colmap[], VecScatter *multScatter) 5295 #endif 5296 { 5297 Mat_MPIAIJ *a; 5298 5299 PetscFunctionBegin; 5300 PetscValidHeaderSpecific(A, MAT_CLASSID, 1); 5301 PetscValidPointer(lvec, 2); 5302 PetscValidPointer(colmap, 3); 5303 PetscValidPointer(multScatter, 4); 5304 a = (Mat_MPIAIJ*) A->data; 5305 if (lvec) *lvec = a->lvec; 5306 if (colmap) *colmap = a->colmap; 5307 if (multScatter) *multScatter = a->Mvctx; 5308 PetscFunctionReturn(0); 5309 } 5310 5311 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJCRL(Mat,MatType,MatReuse,Mat*); 5312 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJPERM(Mat,MatType,MatReuse,Mat*); 5313 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPISBAIJ(Mat,MatType,MatReuse,Mat*); 5314 #if defined(PETSC_HAVE_ELEMENTAL) 5315 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_Elemental(Mat,MatType,MatReuse,Mat*); 5316 #endif 5317 #if defined(PETSC_HAVE_HYPRE) 5318 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*); 5319 PETSC_INTERN PetscErrorCode MatMatMatMult_Transpose_AIJ_AIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*); 5320 #endif 5321 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_IS(Mat,MatType,MatReuse,Mat*); 5322 5323 /* 5324 Computes (B'*A')' since computing B*A directly is untenable 5325 5326 n p p 5327 ( ) ( ) ( ) 5328 m ( A ) * n ( B ) = m ( C ) 5329 ( ) ( ) ( ) 5330 5331 */ 5332 PetscErrorCode MatMatMultNumeric_MPIDense_MPIAIJ(Mat A,Mat B,Mat C) 5333 { 5334 PetscErrorCode ierr; 5335 Mat At,Bt,Ct; 5336 5337 PetscFunctionBegin; 5338 ierr = MatTranspose(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 5339 ierr = MatTranspose(B,MAT_INITIAL_MATRIX,&Bt);CHKERRQ(ierr); 5340 ierr = MatMatMult(Bt,At,MAT_INITIAL_MATRIX,1.0,&Ct);CHKERRQ(ierr); 5341 ierr = MatDestroy(&At);CHKERRQ(ierr); 5342 ierr = MatDestroy(&Bt);CHKERRQ(ierr); 5343 ierr = MatTranspose(Ct,MAT_REUSE_MATRIX,&C);CHKERRQ(ierr); 5344 ierr = MatDestroy(&Ct);CHKERRQ(ierr); 5345 PetscFunctionReturn(0); 5346 } 5347 5348 PetscErrorCode MatMatMultSymbolic_MPIDense_MPIAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 5349 { 5350 PetscErrorCode ierr; 5351 PetscInt m=A->rmap->n,n=B->cmap->n; 5352 Mat Cmat; 5353 5354 PetscFunctionBegin; 5355 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); 5356 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 5357 ierr = MatSetSizes(Cmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 5358 ierr = MatSetBlockSizesFromMats(Cmat,A,B);CHKERRQ(ierr); 5359 ierr = MatSetType(Cmat,MATMPIDENSE);CHKERRQ(ierr); 5360 ierr = MatMPIDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 5361 ierr = MatAssemblyBegin(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5362 ierr = MatAssemblyEnd(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5363 5364 Cmat->ops->matmultnumeric = MatMatMultNumeric_MPIDense_MPIAIJ; 5365 5366 *C = Cmat; 5367 PetscFunctionReturn(0); 5368 } 5369 5370 /* ----------------------------------------------------------------*/ 5371 PETSC_INTERN PetscErrorCode MatMatMult_MPIDense_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 5372 { 5373 PetscErrorCode ierr; 5374 5375 PetscFunctionBegin; 5376 if (scall == MAT_INITIAL_MATRIX) { 5377 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 5378 ierr = MatMatMultSymbolic_MPIDense_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 5379 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 5380 } 5381 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 5382 ierr = MatMatMultNumeric_MPIDense_MPIAIJ(A,B,*C);CHKERRQ(ierr); 5383 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 5384 PetscFunctionReturn(0); 5385 } 5386 5387 /*MC 5388 MATMPIAIJ - MATMPIAIJ = "mpiaij" - A matrix type to be used for parallel sparse matrices. 5389 5390 Options Database Keys: 5391 . -mat_type mpiaij - sets the matrix type to "mpiaij" during a call to MatSetFromOptions() 5392 5393 Level: beginner 5394 5395 .seealso: MatCreateAIJ() 5396 M*/ 5397 5398 PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJ(Mat B) 5399 { 5400 Mat_MPIAIJ *b; 5401 PetscErrorCode ierr; 5402 PetscMPIInt size; 5403 5404 PetscFunctionBegin; 5405 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 5406 5407 ierr = PetscNewLog(B,&b);CHKERRQ(ierr); 5408 B->data = (void*)b; 5409 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 5410 B->assembled = PETSC_FALSE; 5411 B->insertmode = NOT_SET_VALUES; 5412 b->size = size; 5413 5414 ierr = MPI_Comm_rank(PetscObjectComm((PetscObject)B),&b->rank);CHKERRQ(ierr); 5415 5416 /* build cache for off array entries formed */ 5417 ierr = MatStashCreate_Private(PetscObjectComm((PetscObject)B),1,&B->stash);CHKERRQ(ierr); 5418 5419 b->donotstash = PETSC_FALSE; 5420 b->colmap = 0; 5421 b->garray = 0; 5422 b->roworiented = PETSC_TRUE; 5423 5424 /* stuff used for matrix vector multiply */ 5425 b->lvec = NULL; 5426 b->Mvctx = NULL; 5427 5428 /* stuff for MatGetRow() */ 5429 b->rowindices = 0; 5430 b->rowvalues = 0; 5431 b->getrowactive = PETSC_FALSE; 5432 5433 /* flexible pointer used in CUSP/CUSPARSE classes */ 5434 b->spptr = NULL; 5435 5436 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetUseScalableIncreaseOverlap_C",MatMPIAIJSetUseScalableIncreaseOverlap_MPIAIJ);CHKERRQ(ierr); 5437 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_MPIAIJ);CHKERRQ(ierr); 5438 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_MPIAIJ);CHKERRQ(ierr); 5439 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_MPIAIJ);CHKERRQ(ierr); 5440 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetPreallocation_C",MatMPIAIJSetPreallocation_MPIAIJ);CHKERRQ(ierr); 5441 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetPreallocationCSR_C",MatMPIAIJSetPreallocationCSR_MPIAIJ);CHKERRQ(ierr); 5442 ierr = PetscObjectComposeFunction((PetscObject)B,"MatDiagonalScaleLocal_C",MatDiagonalScaleLocal_MPIAIJ);CHKERRQ(ierr); 5443 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpiaijperm_C",MatConvert_MPIAIJ_MPIAIJPERM);CHKERRQ(ierr); 5444 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpiaijcrl_C",MatConvert_MPIAIJ_MPIAIJCRL);CHKERRQ(ierr); 5445 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_mpisbaij_C",MatConvert_MPIAIJ_MPISBAIJ);CHKERRQ(ierr); 5446 #if defined(PETSC_HAVE_ELEMENTAL) 5447 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_elemental_C",MatConvert_MPIAIJ_Elemental);CHKERRQ(ierr); 5448 #endif 5449 #if defined(PETSC_HAVE_HYPRE) 5450 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_hypre_C",MatConvert_AIJ_HYPRE);CHKERRQ(ierr); 5451 #endif 5452 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpiaij_is_C",MatConvert_MPIAIJ_IS);CHKERRQ(ierr); 5453 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_mpidense_mpiaij_C",MatMatMult_MPIDense_MPIAIJ);CHKERRQ(ierr); 5454 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_mpidense_mpiaij_C",MatMatMultSymbolic_MPIDense_MPIAIJ);CHKERRQ(ierr); 5455 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_mpidense_mpiaij_C",MatMatMultNumeric_MPIDense_MPIAIJ);CHKERRQ(ierr); 5456 #if defined(PETSC_HAVE_HYPRE) 5457 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMatMult_transpose_mpiaij_mpiaij_C",MatMatMatMult_Transpose_AIJ_AIJ);CHKERRQ(ierr); 5458 #endif 5459 ierr = PetscObjectChangeTypeName((PetscObject)B,MATMPIAIJ);CHKERRQ(ierr); 5460 PetscFunctionReturn(0); 5461 } 5462 5463 /*@C 5464 MatCreateMPIAIJWithSplitArrays - creates a MPI AIJ matrix using arrays that contain the "diagonal" 5465 and "off-diagonal" part of the matrix in CSR format. 5466 5467 Collective on MPI_Comm 5468 5469 Input Parameters: 5470 + comm - MPI communicator 5471 . m - number of local rows (Cannot be PETSC_DECIDE) 5472 . n - This value should be the same as the local size used in creating the 5473 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 5474 calculated if N is given) For square matrices n is almost always m. 5475 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 5476 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 5477 . i - row indices for "diagonal" portion of matrix 5478 . j - column indices 5479 . a - matrix values 5480 . oi - row indices for "off-diagonal" portion of matrix 5481 . oj - column indices 5482 - oa - matrix values 5483 5484 Output Parameter: 5485 . mat - the matrix 5486 5487 Level: advanced 5488 5489 Notes: 5490 The i, j, and a arrays ARE NOT copied by this routine into the internal format used by PETSc. The user 5491 must free the arrays once the matrix has been destroyed and not before. 5492 5493 The i and j indices are 0 based 5494 5495 See MatCreateAIJ() for the definition of "diagonal" and "off-diagonal" portion of the matrix 5496 5497 This sets local rows and cannot be used to set off-processor values. 5498 5499 Use of this routine is discouraged because it is inflexible and cumbersome to use. It is extremely rare that a 5500 legacy application natively assembles into exactly this split format. The code to do so is nontrivial and does 5501 not easily support in-place reassembly. It is recommended to use MatSetValues() (or a variant thereof) because 5502 the resulting assembly is easier to implement, will work with any matrix format, and the user does not have to 5503 keep track of the underlying array. Use MatSetOption(A,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) to disable all 5504 communication if it is known that only local entries will be set. 5505 5506 .keywords: matrix, aij, compressed row, sparse, parallel 5507 5508 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 5509 MATMPIAIJ, MatCreateAIJ(), MatCreateMPIAIJWithArrays() 5510 @*/ 5511 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) 5512 { 5513 PetscErrorCode ierr; 5514 Mat_MPIAIJ *maij; 5515 5516 PetscFunctionBegin; 5517 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 5518 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 5519 if (oi[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"oi (row indices) must start with 0"); 5520 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 5521 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 5522 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 5523 maij = (Mat_MPIAIJ*) (*mat)->data; 5524 5525 (*mat)->preallocated = PETSC_TRUE; 5526 5527 ierr = PetscLayoutSetUp((*mat)->rmap);CHKERRQ(ierr); 5528 ierr = PetscLayoutSetUp((*mat)->cmap);CHKERRQ(ierr); 5529 5530 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,i,j,a,&maij->A);CHKERRQ(ierr); 5531 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,(*mat)->cmap->N,oi,oj,oa,&maij->B);CHKERRQ(ierr); 5532 5533 ierr = MatAssemblyBegin(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5534 ierr = MatAssemblyEnd(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5535 ierr = MatAssemblyBegin(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5536 ierr = MatAssemblyEnd(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5537 5538 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE);CHKERRQ(ierr); 5539 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5540 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5541 ierr = MatSetOption(*mat,MAT_NO_OFF_PROC_ENTRIES,PETSC_FALSE);CHKERRQ(ierr); 5542 ierr = MatSetOption(*mat,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 5543 PetscFunctionReturn(0); 5544 } 5545 5546 /* 5547 Special version for direct calls from Fortran 5548 */ 5549 #include <petsc/private/fortranimpl.h> 5550 5551 /* Change these macros so can be used in void function */ 5552 #undef CHKERRQ 5553 #define CHKERRQ(ierr) CHKERRABORT(PETSC_COMM_WORLD,ierr) 5554 #undef SETERRQ2 5555 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 5556 #undef SETERRQ3 5557 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 5558 #undef SETERRQ 5559 #define SETERRQ(c,ierr,b) CHKERRABORT(c,ierr) 5560 5561 #if defined(PETSC_HAVE_FORTRAN_CAPS) 5562 #define matsetvaluesmpiaij_ MATSETVALUESMPIAIJ 5563 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 5564 #define matsetvaluesmpiaij_ matsetvaluesmpiaij 5565 #else 5566 #endif 5567 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) 5568 { 5569 Mat mat = *mmat; 5570 PetscInt m = *mm, n = *mn; 5571 InsertMode addv = *maddv; 5572 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 5573 PetscScalar value; 5574 PetscErrorCode ierr; 5575 5576 MatCheckPreallocated(mat,1); 5577 if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv; 5578 5579 #if defined(PETSC_USE_DEBUG) 5580 else if (mat->insertmode != addv) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot mix add values and insert values"); 5581 #endif 5582 { 5583 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 5584 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 5585 PetscBool roworiented = aij->roworiented; 5586 5587 /* Some Variables required in the macro */ 5588 Mat A = aij->A; 5589 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 5590 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 5591 MatScalar *aa = a->a; 5592 PetscBool ignorezeroentries = (((a->ignorezeroentries)&&(addv==ADD_VALUES)) ? PETSC_TRUE : PETSC_FALSE); 5593 Mat B = aij->B; 5594 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 5595 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 5596 MatScalar *ba = b->a; 5597 5598 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 5599 PetscInt nonew = a->nonew; 5600 MatScalar *ap1,*ap2; 5601 5602 PetscFunctionBegin; 5603 for (i=0; i<m; i++) { 5604 if (im[i] < 0) continue; 5605 #if defined(PETSC_USE_DEBUG) 5606 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); 5607 #endif 5608 if (im[i] >= rstart && im[i] < rend) { 5609 row = im[i] - rstart; 5610 lastcol1 = -1; 5611 rp1 = aj + ai[row]; 5612 ap1 = aa + ai[row]; 5613 rmax1 = aimax[row]; 5614 nrow1 = ailen[row]; 5615 low1 = 0; 5616 high1 = nrow1; 5617 lastcol2 = -1; 5618 rp2 = bj + bi[row]; 5619 ap2 = ba + bi[row]; 5620 rmax2 = bimax[row]; 5621 nrow2 = bilen[row]; 5622 low2 = 0; 5623 high2 = nrow2; 5624 5625 for (j=0; j<n; j++) { 5626 if (roworiented) value = v[i*n+j]; 5627 else value = v[i+j*m]; 5628 if (in[j] >= cstart && in[j] < cend) { 5629 col = in[j] - cstart; 5630 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 5631 MatSetValues_SeqAIJ_A_Private(row,col,value,addv,im[i],in[j]); 5632 } else if (in[j] < 0) continue; 5633 #if defined(PETSC_USE_DEBUG) 5634 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); 5635 #endif 5636 else { 5637 if (mat->was_assembled) { 5638 if (!aij->colmap) { 5639 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 5640 } 5641 #if defined(PETSC_USE_CTABLE) 5642 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 5643 col--; 5644 #else 5645 col = aij->colmap[in[j]] - 1; 5646 #endif 5647 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES) && row != col) continue; 5648 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 5649 ierr = MatDisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 5650 col = in[j]; 5651 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 5652 B = aij->B; 5653 b = (Mat_SeqAIJ*)B->data; 5654 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; 5655 rp2 = bj + bi[row]; 5656 ap2 = ba + bi[row]; 5657 rmax2 = bimax[row]; 5658 nrow2 = bilen[row]; 5659 low2 = 0; 5660 high2 = nrow2; 5661 bm = aij->B->rmap->n; 5662 ba = b->a; 5663 } 5664 } else col = in[j]; 5665 MatSetValues_SeqAIJ_B_Private(row,col,value,addv,im[i],in[j]); 5666 } 5667 } 5668 } else if (!aij->donotstash) { 5669 if (roworiented) { 5670 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5671 } else { 5672 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5673 } 5674 } 5675 } 5676 } 5677 PetscFunctionReturnVoid(); 5678 } 5679 5680