1 2 #include "../src/mat/impls/aij/mpi/mpiaij.h" /*I "petscmat.h" I*/ 3 #include "petscblaslapack.h" 4 5 #undef __FUNCT__ 6 #define __FUNCT__ "MatFindNonZeroRows_MPIAIJ" 7 PetscErrorCode MatFindNonZeroRows_MPIAIJ(Mat M,IS *keptrows) 8 { 9 PetscErrorCode ierr; 10 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)M->data; 11 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data; 12 Mat_SeqAIJ *b = (Mat_SeqAIJ*)mat->B->data; 13 const PetscInt *ia,*ib; 14 const MatScalar *aa,*bb; 15 PetscInt na,nb,i,j,*rows,cnt=0,n0rows; 16 PetscInt m = M->rmap->n,rstart = M->rmap->rstart; 17 18 PetscFunctionBegin; 19 *keptrows = 0; 20 ia = a->i; 21 ib = b->i; 22 for (i=0; i<m; i++) { 23 na = ia[i+1] - ia[i]; 24 nb = ib[i+1] - ib[i]; 25 if (!na && !nb) { 26 cnt++; 27 goto ok1; 28 } 29 aa = a->a + ia[i]; 30 for (j=0; j<na; j++) { 31 if (aa[j] != 0.0) goto ok1; 32 } 33 bb = b->a + ib[i]; 34 for (j=0; j <nb; j++) { 35 if (bb[j] != 0.0) goto ok1; 36 } 37 cnt++; 38 ok1:; 39 } 40 ierr = MPI_Allreduce(&cnt,&n0rows,1,MPIU_INT,MPI_SUM,((PetscObject)M)->comm);CHKERRQ(ierr); 41 if (!n0rows) PetscFunctionReturn(0); 42 ierr = PetscMalloc((M->rmap->n-cnt)*sizeof(PetscInt),&rows);CHKERRQ(ierr); 43 cnt = 0; 44 for (i=0; i<m; i++) { 45 na = ia[i+1] - ia[i]; 46 nb = ib[i+1] - ib[i]; 47 if (!na && !nb) continue; 48 aa = a->a + ia[i]; 49 for(j=0; j<na;j++) { 50 if (aa[j] != 0.0) { 51 rows[cnt++] = rstart + i; 52 goto ok2; 53 } 54 } 55 bb = b->a + ib[i]; 56 for (j=0; j<nb; j++) { 57 if (bb[j] != 0.0) { 58 rows[cnt++] = rstart + i; 59 goto ok2; 60 } 61 } 62 ok2:; 63 } 64 ierr = ISCreateGeneral(PETSC_COMM_WORLD,cnt,rows,PETSC_OWN_POINTER,keptrows);CHKERRQ(ierr); 65 PetscFunctionReturn(0); 66 } 67 68 #undef __FUNCT__ 69 #define __FUNCT__ "MatDistribute_MPIAIJ" 70 /* 71 Distributes a SeqAIJ matrix across a set of processes. Code stolen from 72 MatLoad_MPIAIJ(). Horrible lack of reuse. Should be a routine for each matrix type. 73 74 Only for square matrices 75 */ 76 PetscErrorCode MatDistribute_MPIAIJ(MPI_Comm comm,Mat gmat,PetscInt m,MatReuse reuse,Mat *inmat) 77 { 78 PetscMPIInt rank,size; 79 PetscInt *rowners,*dlens,*olens,i,rstart,rend,j,jj,nz,*gmataj,cnt,row,*ld; 80 PetscErrorCode ierr; 81 Mat mat; 82 Mat_SeqAIJ *gmata; 83 PetscMPIInt tag; 84 MPI_Status status; 85 PetscBool aij; 86 MatScalar *gmataa,*ao,*ad,*gmataarestore=0; 87 88 PetscFunctionBegin; 89 CHKMEMQ; 90 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 91 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 92 if (!rank) { 93 ierr = PetscTypeCompare((PetscObject)gmat,MATSEQAIJ,&aij);CHKERRQ(ierr); 94 if (!aij) SETERRQ1(((PetscObject)gmat)->comm,PETSC_ERR_SUP,"Currently no support for input matrix of type %s\n",((PetscObject)gmat)->type_name); 95 } 96 if (reuse == MAT_INITIAL_MATRIX) { 97 ierr = MatCreate(comm,&mat);CHKERRQ(ierr); 98 ierr = MatSetSizes(mat,m,m,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 99 ierr = MatSetType(mat,MATAIJ);CHKERRQ(ierr); 100 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&rowners);CHKERRQ(ierr); 101 ierr = PetscMalloc2(m,PetscInt,&dlens,m,PetscInt,&olens);CHKERRQ(ierr); 102 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 103 rowners[0] = 0; 104 for (i=2; i<=size; i++) { 105 rowners[i] += rowners[i-1]; 106 } 107 rstart = rowners[rank]; 108 rend = rowners[rank+1]; 109 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 110 if (!rank) { 111 gmata = (Mat_SeqAIJ*) gmat->data; 112 /* send row lengths to all processors */ 113 for (i=0; i<m; i++) dlens[i] = gmata->ilen[i]; 114 for (i=1; i<size; i++) { 115 ierr = MPI_Send(gmata->ilen + rowners[i],rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 116 } 117 /* determine number diagonal and off-diagonal counts */ 118 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 119 ierr = PetscMalloc(m*sizeof(PetscInt),&ld);CHKERRQ(ierr); 120 ierr = PetscMemzero(ld,m*sizeof(PetscInt));CHKERRQ(ierr); 121 jj = 0; 122 for (i=0; i<m; i++) { 123 for (j=0; j<dlens[i]; j++) { 124 if (gmata->j[jj] < rstart) ld[i]++; 125 if (gmata->j[jj] < rstart || gmata->j[jj] >= rend) olens[i]++; 126 jj++; 127 } 128 } 129 /* send column indices to other processes */ 130 for (i=1; i<size; i++) { 131 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 132 ierr = MPI_Send(&nz,1,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 133 ierr = MPI_Send(gmata->j + gmata->i[rowners[i]],nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 134 } 135 136 /* send numerical values to other processes */ 137 for (i=1; i<size; i++) { 138 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 139 ierr = MPI_Send(gmata->a + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 140 } 141 gmataa = gmata->a; 142 gmataj = gmata->j; 143 144 } else { 145 /* receive row lengths */ 146 ierr = MPI_Recv(dlens,m,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 147 /* receive column indices */ 148 ierr = MPI_Recv(&nz,1,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 149 ierr = PetscMalloc2(nz,PetscScalar,&gmataa,nz,PetscInt,&gmataj);CHKERRQ(ierr); 150 ierr = MPI_Recv(gmataj,nz,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 151 /* determine number diagonal and off-diagonal counts */ 152 ierr = PetscMemzero(olens,m*sizeof(PetscInt));CHKERRQ(ierr); 153 ierr = PetscMalloc(m*sizeof(PetscInt),&ld);CHKERRQ(ierr); 154 ierr = PetscMemzero(ld,m*sizeof(PetscInt));CHKERRQ(ierr); 155 jj = 0; 156 for (i=0; i<m; i++) { 157 for (j=0; j<dlens[i]; j++) { 158 if (gmataj[jj] < rstart) ld[i]++; 159 if (gmataj[jj] < rstart || gmataj[jj] >= rend) olens[i]++; 160 jj++; 161 } 162 } 163 /* receive numerical values */ 164 ierr = PetscMemzero(gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); 165 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 166 } 167 /* set preallocation */ 168 for (i=0; i<m; i++) { 169 dlens[i] -= olens[i]; 170 } 171 ierr = MatSeqAIJSetPreallocation(mat,0,dlens);CHKERRQ(ierr); 172 ierr = MatMPIAIJSetPreallocation(mat,0,dlens,0,olens);CHKERRQ(ierr); 173 174 for (i=0; i<m; i++) { 175 dlens[i] += olens[i]; 176 } 177 cnt = 0; 178 for (i=0; i<m; i++) { 179 row = rstart + i; 180 ierr = MatSetValues(mat,1,&row,dlens[i],gmataj+cnt,gmataa+cnt,INSERT_VALUES);CHKERRQ(ierr); 181 cnt += dlens[i]; 182 } 183 if (rank) { 184 ierr = PetscFree2(gmataa,gmataj);CHKERRQ(ierr); 185 } 186 ierr = PetscFree2(dlens,olens);CHKERRQ(ierr); 187 ierr = PetscFree(rowners);CHKERRQ(ierr); 188 ((Mat_MPIAIJ*)(mat->data))->ld = ld; 189 *inmat = mat; 190 } else { /* column indices are already set; only need to move over numerical values from process 0 */ 191 Mat_SeqAIJ *Ad = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->A->data; 192 Mat_SeqAIJ *Ao = (Mat_SeqAIJ*)((Mat_MPIAIJ*)((*inmat)->data))->B->data; 193 mat = *inmat; 194 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag);CHKERRQ(ierr); 195 if (!rank) { 196 /* send numerical values to other processes */ 197 gmata = (Mat_SeqAIJ*) gmat->data; 198 ierr = MatGetOwnershipRanges(mat,(const PetscInt**)&rowners);CHKERRQ(ierr); 199 gmataa = gmata->a; 200 for (i=1; i<size; i++) { 201 nz = gmata->i[rowners[i+1]]-gmata->i[rowners[i]]; 202 ierr = MPI_Send(gmataa + gmata->i[rowners[i]],nz,MPIU_SCALAR,i,tag,comm);CHKERRQ(ierr); 203 } 204 nz = gmata->i[rowners[1]]-gmata->i[rowners[0]]; 205 } else { 206 /* receive numerical values from process 0*/ 207 nz = Ad->nz + Ao->nz; 208 ierr = PetscMalloc(nz*sizeof(PetscScalar),&gmataa);CHKERRQ(ierr); gmataarestore = gmataa; 209 ierr = MPI_Recv(gmataa,nz,MPIU_SCALAR,0,tag,comm,&status);CHKERRQ(ierr); 210 } 211 /* transfer numerical values into the diagonal A and off diagonal B parts of mat */ 212 ld = ((Mat_MPIAIJ*)(mat->data))->ld; 213 ad = Ad->a; 214 ao = Ao->a; 215 if (mat->rmap->n) { 216 i = 0; 217 nz = ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ao += nz; gmataa += nz; 218 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 219 } 220 for (i=1; i<mat->rmap->n; i++) { 221 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; 222 nz = Ad->i[i+1] - Ad->i[i]; ierr = PetscMemcpy(ad,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ad += nz; gmataa += nz; 223 } 224 i--; 225 if (mat->rmap->n) { 226 nz = Ao->i[i+1] - Ao->i[i] - ld[i]; ierr = PetscMemcpy(ao,gmataa,nz*sizeof(PetscScalar));CHKERRQ(ierr); ao += nz; gmataa += nz; 227 } 228 if (rank) { 229 ierr = PetscFree(gmataarestore);CHKERRQ(ierr); 230 } 231 } 232 ierr = MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 233 ierr = MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 234 CHKMEMQ; 235 PetscFunctionReturn(0); 236 } 237 238 /* 239 Local utility routine that creates a mapping from the global column 240 number to the local number in the off-diagonal part of the local 241 storage of the matrix. When PETSC_USE_CTABLE is used this is scalable at 242 a slightly higher hash table cost; without it it is not scalable (each processor 243 has an order N integer array but is fast to acess. 244 */ 245 #undef __FUNCT__ 246 #define __FUNCT__ "CreateColmap_MPIAIJ_Private" 247 PetscErrorCode CreateColmap_MPIAIJ_Private(Mat mat) 248 { 249 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 250 PetscErrorCode ierr; 251 PetscInt n = aij->B->cmap->n,i; 252 253 PetscFunctionBegin; 254 #if defined (PETSC_USE_CTABLE) 255 ierr = PetscTableCreate(n,&aij->colmap);CHKERRQ(ierr); 256 for (i=0; i<n; i++){ 257 ierr = PetscTableAdd(aij->colmap,aij->garray[i]+1,i+1);CHKERRQ(ierr); 258 } 259 #else 260 ierr = PetscMalloc((mat->cmap->N+1)*sizeof(PetscInt),&aij->colmap);CHKERRQ(ierr); 261 ierr = PetscLogObjectMemory(mat,mat->cmap->N*sizeof(PetscInt));CHKERRQ(ierr); 262 ierr = PetscMemzero(aij->colmap,mat->cmap->N*sizeof(PetscInt));CHKERRQ(ierr); 263 for (i=0; i<n; i++) aij->colmap[aij->garray[i]] = i+1; 264 #endif 265 PetscFunctionReturn(0); 266 } 267 268 #define MatSetValues_SeqAIJ_A_Private(row,col,value,addv) \ 269 { \ 270 if (col <= lastcol1) low1 = 0; else high1 = nrow1; \ 271 lastcol1 = col;\ 272 while (high1-low1 > 5) { \ 273 t = (low1+high1)/2; \ 274 if (rp1[t] > col) high1 = t; \ 275 else low1 = t; \ 276 } \ 277 for (_i=low1; _i<high1; _i++) { \ 278 if (rp1[_i] > col) break; \ 279 if (rp1[_i] == col) { \ 280 if (addv == ADD_VALUES) ap1[_i] += value; \ 281 else ap1[_i] = value; \ 282 goto a_noinsert; \ 283 } \ 284 } \ 285 if (value == 0.0 && ignorezeroentries) {low1 = 0; high1 = nrow1;goto a_noinsert;} \ 286 if (nonew == 1) {low1 = 0; high1 = nrow1; goto a_noinsert;} \ 287 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \ 288 MatSeqXAIJReallocateAIJ(A,am,1,nrow1,row,col,rmax1,aa,ai,aj,rp1,ap1,aimax,nonew,MatScalar); \ 289 N = nrow1++ - 1; a->nz++; high1++; \ 290 /* shift up all the later entries in this row */ \ 291 for (ii=N; ii>=_i; ii--) { \ 292 rp1[ii+1] = rp1[ii]; \ 293 ap1[ii+1] = ap1[ii]; \ 294 } \ 295 rp1[_i] = col; \ 296 ap1[_i] = value; \ 297 a_noinsert: ; \ 298 ailen[row] = nrow1; \ 299 } 300 301 302 #define MatSetValues_SeqAIJ_B_Private(row,col,value,addv) \ 303 { \ 304 if (col <= lastcol2) low2 = 0; else high2 = nrow2; \ 305 lastcol2 = col;\ 306 while (high2-low2 > 5) { \ 307 t = (low2+high2)/2; \ 308 if (rp2[t] > col) high2 = t; \ 309 else low2 = t; \ 310 } \ 311 for (_i=low2; _i<high2; _i++) { \ 312 if (rp2[_i] > col) break; \ 313 if (rp2[_i] == col) { \ 314 if (addv == ADD_VALUES) ap2[_i] += value; \ 315 else ap2[_i] = value; \ 316 goto b_noinsert; \ 317 } \ 318 } \ 319 if (value == 0.0 && ignorezeroentries) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 320 if (nonew == 1) {low2 = 0; high2 = nrow2; goto b_noinsert;} \ 321 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \ 322 MatSeqXAIJReallocateAIJ(B,bm,1,nrow2,row,col,rmax2,ba,bi,bj,rp2,ap2,bimax,nonew,MatScalar); \ 323 N = nrow2++ - 1; b->nz++; high2++; \ 324 /* shift up all the later entries in this row */ \ 325 for (ii=N; ii>=_i; ii--) { \ 326 rp2[ii+1] = rp2[ii]; \ 327 ap2[ii+1] = ap2[ii]; \ 328 } \ 329 rp2[_i] = col; \ 330 ap2[_i] = value; \ 331 b_noinsert: ; \ 332 bilen[row] = nrow2; \ 333 } 334 335 #undef __FUNCT__ 336 #define __FUNCT__ "MatSetValuesRow_MPIAIJ" 337 PetscErrorCode MatSetValuesRow_MPIAIJ(Mat A,PetscInt row,const PetscScalar v[]) 338 { 339 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 340 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data,*b = (Mat_SeqAIJ*)mat->B->data; 341 PetscErrorCode ierr; 342 PetscInt l,*garray = mat->garray,diag; 343 344 PetscFunctionBegin; 345 /* code only works for square matrices A */ 346 347 /* find size of row to the left of the diagonal part */ 348 ierr = MatGetOwnershipRange(A,&diag,0);CHKERRQ(ierr); 349 row = row - diag; 350 for (l=0; l<b->i[row+1]-b->i[row]; l++) { 351 if (garray[b->j[b->i[row]+l]] > diag) break; 352 } 353 ierr = PetscMemcpy(b->a+b->i[row],v,l*sizeof(PetscScalar));CHKERRQ(ierr); 354 355 /* diagonal part */ 356 ierr = PetscMemcpy(a->a+a->i[row],v+l,(a->i[row+1]-a->i[row])*sizeof(PetscScalar));CHKERRQ(ierr); 357 358 /* right of diagonal part */ 359 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); 360 PetscFunctionReturn(0); 361 } 362 363 #undef __FUNCT__ 364 #define __FUNCT__ "MatSetValues_MPIAIJ" 365 PetscErrorCode MatSetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv) 366 { 367 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 368 PetscScalar value; 369 PetscErrorCode ierr; 370 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 371 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 372 PetscBool roworiented = aij->roworiented; 373 374 /* Some Variables required in the macro */ 375 Mat A = aij->A; 376 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 377 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 378 MatScalar *aa = a->a; 379 PetscBool ignorezeroentries = a->ignorezeroentries; 380 Mat B = aij->B; 381 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 382 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 383 MatScalar *ba = b->a; 384 385 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 386 PetscInt nonew = a->nonew; 387 MatScalar *ap1,*ap2; 388 389 PetscFunctionBegin; 390 if (v) PetscValidScalarPointer(v,6); 391 for (i=0; i<m; i++) { 392 if (im[i] < 0) continue; 393 #if defined(PETSC_USE_DEBUG) 394 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); 395 #endif 396 if (im[i] >= rstart && im[i] < rend) { 397 row = im[i] - rstart; 398 lastcol1 = -1; 399 rp1 = aj + ai[row]; 400 ap1 = aa + ai[row]; 401 rmax1 = aimax[row]; 402 nrow1 = ailen[row]; 403 low1 = 0; 404 high1 = nrow1; 405 lastcol2 = -1; 406 rp2 = bj + bi[row]; 407 ap2 = ba + bi[row]; 408 rmax2 = bimax[row]; 409 nrow2 = bilen[row]; 410 low2 = 0; 411 high2 = nrow2; 412 413 for (j=0; j<n; j++) { 414 if (v) {if (roworiented) value = v[i*n+j]; else value = v[i+j*m];} else value = 0.0; 415 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES)) continue; 416 if (in[j] >= cstart && in[j] < cend){ 417 col = in[j] - cstart; 418 MatSetValues_SeqAIJ_A_Private(row,col,value,addv); 419 } else if (in[j] < 0) continue; 420 #if defined(PETSC_USE_DEBUG) 421 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); 422 #endif 423 else { 424 if (mat->was_assembled) { 425 if (!aij->colmap) { 426 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 427 } 428 #if defined (PETSC_USE_CTABLE) 429 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 430 col--; 431 #else 432 col = aij->colmap[in[j]] - 1; 433 #endif 434 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 435 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 436 col = in[j]; 437 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 438 B = aij->B; 439 b = (Mat_SeqAIJ*)B->data; 440 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; ba = b->a; 441 rp2 = bj + bi[row]; 442 ap2 = ba + bi[row]; 443 rmax2 = bimax[row]; 444 nrow2 = bilen[row]; 445 low2 = 0; 446 high2 = nrow2; 447 bm = aij->B->rmap->n; 448 ba = b->a; 449 } 450 } else col = in[j]; 451 MatSetValues_SeqAIJ_B_Private(row,col,value,addv); 452 } 453 } 454 } else { 455 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]); 456 if (!aij->donotstash) { 457 if (roworiented) { 458 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 459 } else { 460 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 461 } 462 } 463 } 464 } 465 PetscFunctionReturn(0); 466 } 467 468 #undef __FUNCT__ 469 #define __FUNCT__ "MatGetValues_MPIAIJ" 470 PetscErrorCode MatGetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[]) 471 { 472 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 473 PetscErrorCode ierr; 474 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 475 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 476 477 PetscFunctionBegin; 478 for (i=0; i<m; i++) { 479 if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]);*/ 480 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); 481 if (idxm[i] >= rstart && idxm[i] < rend) { 482 row = idxm[i] - rstart; 483 for (j=0; j<n; j++) { 484 if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",idxn[j]); */ 485 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); 486 if (idxn[j] >= cstart && idxn[j] < cend){ 487 col = idxn[j] - cstart; 488 ierr = MatGetValues(aij->A,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 489 } else { 490 if (!aij->colmap) { 491 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 492 } 493 #if defined (PETSC_USE_CTABLE) 494 ierr = PetscTableFind(aij->colmap,idxn[j]+1,&col);CHKERRQ(ierr); 495 col --; 496 #else 497 col = aij->colmap[idxn[j]] - 1; 498 #endif 499 if ((col < 0) || (aij->garray[col] != idxn[j])) *(v+i*n+j) = 0.0; 500 else { 501 ierr = MatGetValues(aij->B,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 502 } 503 } 504 } 505 } else { 506 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only local values currently supported"); 507 } 508 } 509 PetscFunctionReturn(0); 510 } 511 512 extern PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat,Vec,Vec); 513 514 #undef __FUNCT__ 515 #define __FUNCT__ "MatAssemblyBegin_MPIAIJ" 516 PetscErrorCode MatAssemblyBegin_MPIAIJ(Mat mat,MatAssemblyType mode) 517 { 518 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 519 PetscErrorCode ierr; 520 PetscInt nstash,reallocs; 521 InsertMode addv; 522 523 PetscFunctionBegin; 524 if (aij->donotstash || mat->nooffprocentries) { 525 PetscFunctionReturn(0); 526 } 527 528 /* make sure all processors are either in INSERTMODE or ADDMODE */ 529 ierr = MPI_Allreduce(&mat->insertmode,&addv,1,MPI_INT,MPI_BOR,((PetscObject)mat)->comm);CHKERRQ(ierr); 530 if (addv == (ADD_VALUES|INSERT_VALUES)) SETERRQ(((PetscObject)mat)->comm,PETSC_ERR_ARG_WRONGSTATE,"Some processors inserted others added"); 531 mat->insertmode = addv; /* in case this processor had no cache */ 532 533 ierr = MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap->range);CHKERRQ(ierr); 534 ierr = MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);CHKERRQ(ierr); 535 ierr = PetscInfo2(aij->A,"Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);CHKERRQ(ierr); 536 PetscFunctionReturn(0); 537 } 538 539 #undef __FUNCT__ 540 #define __FUNCT__ "MatAssemblyEnd_MPIAIJ" 541 PetscErrorCode MatAssemblyEnd_MPIAIJ(Mat mat,MatAssemblyType mode) 542 { 543 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 544 Mat_SeqAIJ *a=(Mat_SeqAIJ *)aij->A->data; 545 PetscErrorCode ierr; 546 PetscMPIInt n; 547 PetscInt i,j,rstart,ncols,flg; 548 PetscInt *row,*col; 549 PetscBool other_disassembled; 550 PetscScalar *val; 551 InsertMode addv = mat->insertmode; 552 553 /* do not use 'b = (Mat_SeqAIJ *)aij->B->data' as B can be reset in disassembly */ 554 PetscFunctionBegin; 555 if (!aij->donotstash && !mat->nooffprocentries) { 556 while (1) { 557 ierr = MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);CHKERRQ(ierr); 558 if (!flg) break; 559 560 for (i=0; i<n;) { 561 /* Now identify the consecutive vals belonging to the same row */ 562 for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; } 563 if (j < n) ncols = j-i; 564 else ncols = n-i; 565 /* Now assemble all these values with a single function call */ 566 ierr = MatSetValues_MPIAIJ(mat,1,row+i,ncols,col+i,val+i,addv);CHKERRQ(ierr); 567 i = j; 568 } 569 } 570 ierr = MatStashScatterEnd_Private(&mat->stash);CHKERRQ(ierr); 571 } 572 ierr = MatAssemblyBegin(aij->A,mode);CHKERRQ(ierr); 573 ierr = MatAssemblyEnd(aij->A,mode);CHKERRQ(ierr); 574 575 /* determine if any processor has disassembled, if so we must 576 also disassemble ourselfs, in order that we may reassemble. */ 577 /* 578 if nonzero structure of submatrix B cannot change then we know that 579 no processor disassembled thus we can skip this stuff 580 */ 581 if (!((Mat_SeqAIJ*)aij->B->data)->nonew) { 582 ierr = MPI_Allreduce(&mat->was_assembled,&other_disassembled,1,MPI_INT,MPI_PROD,((PetscObject)mat)->comm);CHKERRQ(ierr); 583 if (mat->was_assembled && !other_disassembled) { 584 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 585 } 586 } 587 if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) { 588 ierr = MatSetUpMultiply_MPIAIJ(mat);CHKERRQ(ierr); 589 } 590 ierr = MatSetOption(aij->B,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 591 ierr = MatSetOption(aij->B,MAT_CHECK_COMPRESSED_ROW,PETSC_FALSE);CHKERRQ(ierr); 592 ierr = MatAssemblyBegin(aij->B,mode);CHKERRQ(ierr); 593 ierr = MatAssemblyEnd(aij->B,mode);CHKERRQ(ierr); 594 595 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 596 aij->rowvalues = 0; 597 598 /* used by MatAXPY() */ 599 a->xtoy = 0; ((Mat_SeqAIJ *)aij->B->data)->xtoy = 0; /* b->xtoy = 0 */ 600 a->XtoY = 0; ((Mat_SeqAIJ *)aij->B->data)->XtoY = 0; /* b->XtoY = 0 */ 601 602 if (aij->diag) {ierr = VecDestroy(aij->diag);CHKERRQ(ierr);aij->diag = 0;} 603 if (a->inode.size) mat->ops->multdiagonalblock = MatMultDiagonalBlock_MPIAIJ; 604 PetscFunctionReturn(0); 605 } 606 607 #undef __FUNCT__ 608 #define __FUNCT__ "MatZeroEntries_MPIAIJ" 609 PetscErrorCode MatZeroEntries_MPIAIJ(Mat A) 610 { 611 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 612 PetscErrorCode ierr; 613 614 PetscFunctionBegin; 615 ierr = MatZeroEntries(l->A);CHKERRQ(ierr); 616 ierr = MatZeroEntries(l->B);CHKERRQ(ierr); 617 PetscFunctionReturn(0); 618 } 619 620 #undef __FUNCT__ 621 #define __FUNCT__ "MatZeroRows_MPIAIJ" 622 PetscErrorCode MatZeroRows_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 623 { 624 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 625 PetscErrorCode ierr; 626 PetscMPIInt size = l->size,imdex,n,rank = l->rank,tag = ((PetscObject)A)->tag,lastidx = -1; 627 PetscInt i,*owners = A->rmap->range; 628 PetscInt *nprocs,j,idx,nsends,row; 629 PetscInt nmax,*svalues,*starts,*owner,nrecvs; 630 PetscInt *rvalues,count,base,slen,*source; 631 PetscInt *lens,*lrows,*values,rstart=A->rmap->rstart; 632 MPI_Comm comm = ((PetscObject)A)->comm; 633 MPI_Request *send_waits,*recv_waits; 634 MPI_Status recv_status,*send_status; 635 const PetscScalar *xx; 636 PetscScalar *bb; 637 #if defined(PETSC_DEBUG) 638 PetscBool found = PETSC_FALSE; 639 #endif 640 641 PetscFunctionBegin; 642 /* first count number of contributors to each processor */ 643 ierr = PetscMalloc(2*size*sizeof(PetscInt),&nprocs);CHKERRQ(ierr); 644 ierr = PetscMemzero(nprocs,2*size*sizeof(PetscInt));CHKERRQ(ierr); 645 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&owner);CHKERRQ(ierr); /* see note*/ 646 j = 0; 647 for (i=0; i<N; i++) { 648 if (lastidx > (idx = rows[i])) j = 0; 649 lastidx = idx; 650 for (; j<size; j++) { 651 if (idx >= owners[j] && idx < owners[j+1]) { 652 nprocs[2*j]++; 653 nprocs[2*j+1] = 1; 654 owner[i] = j; 655 #if defined(PETSC_DEBUG) 656 found = PETSC_TRUE; 657 #endif 658 break; 659 } 660 } 661 #if defined(PETSC_DEBUG) 662 if (!found) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Index out of range"); 663 found = PETSC_FALSE; 664 #endif 665 } 666 nsends = 0; for (i=0; i<size; i++) { nsends += nprocs[2*i+1];} 667 668 if (A->nooffproczerorows) { 669 if (nsends > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"You called MatSetOption(,MAT_NO_OFF_PROC_ZERO_ROWS,PETSC_TRUE) but set an off process zero row"); 670 nrecvs = nsends; 671 nmax = N; 672 } else { 673 /* inform other processors of number of messages and max length*/ 674 ierr = PetscMaxSum(comm,nprocs,&nmax,&nrecvs);CHKERRQ(ierr); 675 } 676 677 /* post receives: */ 678 ierr = PetscMalloc((nrecvs+1)*(nmax+1)*sizeof(PetscInt),&rvalues);CHKERRQ(ierr); 679 ierr = PetscMalloc((nrecvs+1)*sizeof(MPI_Request),&recv_waits);CHKERRQ(ierr); 680 for (i=0; i<nrecvs; i++) { 681 ierr = MPI_Irecv(rvalues+nmax*i,nmax,MPIU_INT,MPI_ANY_SOURCE,tag,comm,recv_waits+i);CHKERRQ(ierr); 682 } 683 684 /* do sends: 685 1) starts[i] gives the starting index in svalues for stuff going to 686 the ith processor 687 */ 688 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&svalues);CHKERRQ(ierr); 689 ierr = PetscMalloc((nsends+1)*sizeof(MPI_Request),&send_waits);CHKERRQ(ierr); 690 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&starts);CHKERRQ(ierr); 691 starts[0] = 0; 692 for (i=1; i<size; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 693 for (i=0; i<N; i++) { 694 svalues[starts[owner[i]]++] = rows[i]; 695 } 696 697 starts[0] = 0; 698 for (i=1; i<size+1; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 699 count = 0; 700 for (i=0; i<size; i++) { 701 if (nprocs[2*i+1]) { 702 ierr = MPI_Isend(svalues+starts[i],nprocs[2*i],MPIU_INT,i,tag,comm,send_waits+count++);CHKERRQ(ierr); 703 } 704 } 705 ierr = PetscFree(starts);CHKERRQ(ierr); 706 707 base = owners[rank]; 708 709 /* wait on receives */ 710 ierr = PetscMalloc2(nrecvs,PetscInt,&lens,nrecvs,PetscInt,&source);CHKERRQ(ierr); 711 count = nrecvs; slen = 0; 712 while (count) { 713 ierr = MPI_Waitany(nrecvs,recv_waits,&imdex,&recv_status);CHKERRQ(ierr); 714 /* unpack receives into our local space */ 715 ierr = MPI_Get_count(&recv_status,MPIU_INT,&n);CHKERRQ(ierr); 716 source[imdex] = recv_status.MPI_SOURCE; 717 lens[imdex] = n; 718 slen += n; 719 count--; 720 } 721 ierr = PetscFree(recv_waits);CHKERRQ(ierr); 722 723 /* move the data into the send scatter */ 724 ierr = PetscMalloc((slen+1)*sizeof(PetscInt),&lrows);CHKERRQ(ierr); 725 count = 0; 726 for (i=0; i<nrecvs; i++) { 727 values = rvalues + i*nmax; 728 for (j=0; j<lens[i]; j++) { 729 lrows[count++] = values[j] - base; 730 } 731 } 732 ierr = PetscFree(rvalues);CHKERRQ(ierr); 733 ierr = PetscFree2(lens,source);CHKERRQ(ierr); 734 ierr = PetscFree(owner);CHKERRQ(ierr); 735 ierr = PetscFree(nprocs);CHKERRQ(ierr); 736 737 /* fix right hand side if needed */ 738 if (x && b) { 739 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 740 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 741 for (i=0; i<slen; i++) { 742 bb[lrows[i]] = diag*xx[lrows[i]]; 743 } 744 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 745 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 746 } 747 /* 748 Zero the required rows. If the "diagonal block" of the matrix 749 is square and the user wishes to set the diagonal we use separate 750 code so that MatSetValues() is not called for each diagonal allocating 751 new memory, thus calling lots of mallocs and slowing things down. 752 753 */ 754 /* must zero l->B before l->A because the (diag) case below may put values into l->B*/ 755 ierr = MatZeroRows(l->B,slen,lrows,0.0,0,0);CHKERRQ(ierr); 756 if ((diag != 0.0) && (l->A->rmap->N == l->A->cmap->N)) { 757 ierr = MatZeroRows(l->A,slen,lrows,diag,0,0);CHKERRQ(ierr); 758 } else if (diag != 0.0) { 759 ierr = MatZeroRows(l->A,slen,lrows,0.0,0,0);CHKERRQ(ierr); 760 if (((Mat_SeqAIJ*)l->A->data)->nonew) { 761 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatZeroRows() on rectangular matrices cannot be used with the Mat options\n\ 762 MAT_NEW_NONZERO_LOCATIONS,MAT_NEW_NONZERO_LOCATION_ERR,MAT_NEW_NONZERO_ALLOCATION_ERR"); 763 } 764 for (i = 0; i < slen; i++) { 765 row = lrows[i] + rstart; 766 ierr = MatSetValues(A,1,&row,1,&row,&diag,INSERT_VALUES);CHKERRQ(ierr); 767 } 768 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 769 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 770 } else { 771 ierr = MatZeroRows(l->A,slen,lrows,0.0,0,0);CHKERRQ(ierr); 772 } 773 ierr = PetscFree(lrows);CHKERRQ(ierr); 774 775 /* wait on sends */ 776 if (nsends) { 777 ierr = PetscMalloc(nsends*sizeof(MPI_Status),&send_status);CHKERRQ(ierr); 778 ierr = MPI_Waitall(nsends,send_waits,send_status);CHKERRQ(ierr); 779 ierr = PetscFree(send_status);CHKERRQ(ierr); 780 } 781 ierr = PetscFree(send_waits);CHKERRQ(ierr); 782 ierr = PetscFree(svalues);CHKERRQ(ierr); 783 784 PetscFunctionReturn(0); 785 } 786 787 #undef __FUNCT__ 788 #define __FUNCT__ "MatZeroRowsColumns_MPIAIJ" 789 PetscErrorCode MatZeroRowsColumns_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 790 { 791 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 792 PetscErrorCode ierr; 793 PetscMPIInt size = l->size,imdex,n,rank = l->rank,tag = ((PetscObject)A)->tag,lastidx = -1; 794 PetscInt i,*owners = A->rmap->range; 795 PetscInt *nprocs,j,idx,nsends; 796 PetscInt nmax,*svalues,*starts,*owner,nrecvs; 797 PetscInt *rvalues,count,base,slen,*source; 798 PetscInt *lens,*lrows,*values,m; 799 MPI_Comm comm = ((PetscObject)A)->comm; 800 MPI_Request *send_waits,*recv_waits; 801 MPI_Status recv_status,*send_status; 802 const PetscScalar *xx; 803 PetscScalar *bb,*mask; 804 Vec xmask,lmask; 805 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)l->B->data; 806 const PetscInt *aj, *ii,*ridx; 807 PetscScalar *aa; 808 #if defined(PETSC_DEBUG) 809 PetscBool found = PETSC_FALSE; 810 #endif 811 812 PetscFunctionBegin; 813 /* first count number of contributors to each processor */ 814 ierr = PetscMalloc(2*size*sizeof(PetscInt),&nprocs);CHKERRQ(ierr); 815 ierr = PetscMemzero(nprocs,2*size*sizeof(PetscInt));CHKERRQ(ierr); 816 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&owner);CHKERRQ(ierr); /* see note*/ 817 j = 0; 818 for (i=0; i<N; i++) { 819 if (lastidx > (idx = rows[i])) j = 0; 820 lastidx = idx; 821 for (; j<size; j++) { 822 if (idx >= owners[j] && idx < owners[j+1]) { 823 nprocs[2*j]++; 824 nprocs[2*j+1] = 1; 825 owner[i] = j; 826 #if defined(PETSC_DEBUG) 827 found = PETSC_TRUE; 828 #endif 829 break; 830 } 831 } 832 #if defined(PETSC_DEBUG) 833 if (!found) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Index out of range"); 834 found = PETSC_FALSE; 835 #endif 836 } 837 nsends = 0; for (i=0; i<size; i++) { nsends += nprocs[2*i+1];} 838 839 /* inform other processors of number of messages and max length*/ 840 ierr = PetscMaxSum(comm,nprocs,&nmax,&nrecvs);CHKERRQ(ierr); 841 842 /* post receives: */ 843 ierr = PetscMalloc((nrecvs+1)*(nmax+1)*sizeof(PetscInt),&rvalues);CHKERRQ(ierr); 844 ierr = PetscMalloc((nrecvs+1)*sizeof(MPI_Request),&recv_waits);CHKERRQ(ierr); 845 for (i=0; i<nrecvs; i++) { 846 ierr = MPI_Irecv(rvalues+nmax*i,nmax,MPIU_INT,MPI_ANY_SOURCE,tag,comm,recv_waits+i);CHKERRQ(ierr); 847 } 848 849 /* do sends: 850 1) starts[i] gives the starting index in svalues for stuff going to 851 the ith processor 852 */ 853 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&svalues);CHKERRQ(ierr); 854 ierr = PetscMalloc((nsends+1)*sizeof(MPI_Request),&send_waits);CHKERRQ(ierr); 855 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&starts);CHKERRQ(ierr); 856 starts[0] = 0; 857 for (i=1; i<size; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 858 for (i=0; i<N; i++) { 859 svalues[starts[owner[i]]++] = rows[i]; 860 } 861 862 starts[0] = 0; 863 for (i=1; i<size+1; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 864 count = 0; 865 for (i=0; i<size; i++) { 866 if (nprocs[2*i+1]) { 867 ierr = MPI_Isend(svalues+starts[i],nprocs[2*i],MPIU_INT,i,tag,comm,send_waits+count++);CHKERRQ(ierr); 868 } 869 } 870 ierr = PetscFree(starts);CHKERRQ(ierr); 871 872 base = owners[rank]; 873 874 /* wait on receives */ 875 ierr = PetscMalloc2(nrecvs,PetscInt,&lens,nrecvs,PetscInt,&source);CHKERRQ(ierr); 876 count = nrecvs; slen = 0; 877 while (count) { 878 ierr = MPI_Waitany(nrecvs,recv_waits,&imdex,&recv_status);CHKERRQ(ierr); 879 /* unpack receives into our local space */ 880 ierr = MPI_Get_count(&recv_status,MPIU_INT,&n);CHKERRQ(ierr); 881 source[imdex] = recv_status.MPI_SOURCE; 882 lens[imdex] = n; 883 slen += n; 884 count--; 885 } 886 ierr = PetscFree(recv_waits);CHKERRQ(ierr); 887 888 /* move the data into the send scatter */ 889 ierr = PetscMalloc((slen+1)*sizeof(PetscInt),&lrows);CHKERRQ(ierr); 890 count = 0; 891 for (i=0; i<nrecvs; i++) { 892 values = rvalues + i*nmax; 893 for (j=0; j<lens[i]; j++) { 894 lrows[count++] = values[j] - base; 895 } 896 } 897 ierr = PetscFree(rvalues);CHKERRQ(ierr); 898 ierr = PetscFree2(lens,source);CHKERRQ(ierr); 899 ierr = PetscFree(owner);CHKERRQ(ierr); 900 ierr = PetscFree(nprocs);CHKERRQ(ierr); 901 /* lrows are the local rows to be zeroed, slen is the number of local rows */ 902 903 /* zero diagonal part of matrix */ 904 ierr = MatZeroRowsColumns(l->A,slen,lrows,diag,x,b);CHKERRQ(ierr); 905 906 /* handle off diagonal part of matrix */ 907 ierr = MatGetVecs(A,&xmask,PETSC_NULL);CHKERRQ(ierr); 908 ierr = VecDuplicate(l->lvec,&lmask);CHKERRQ(ierr); 909 ierr = VecGetArray(xmask,&bb);CHKERRQ(ierr); 910 for (i=0; i<slen; i++) { 911 bb[lrows[i]] = 1; 912 } 913 ierr = VecRestoreArray(xmask,&bb);CHKERRQ(ierr); 914 ierr = VecScatterBegin(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 915 ierr = VecScatterEnd(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 916 ierr = VecDestroy(xmask);CHKERRQ(ierr); 917 ierr = VecScatterBegin(l->Mvctx,x,l->lvec,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 918 ierr = VecScatterEnd(l->Mvctx,x,l->lvec,ADD_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 919 ierr = VecGetArrayRead(l->lvec,&xx);CHKERRQ(ierr); 920 ierr = VecGetArray(lmask,&mask);CHKERRQ(ierr); 921 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 922 923 /* remove zeroed rows of off diagonal matrix */ 924 ii = aij->i; 925 for (i=0; i<slen; i++) { 926 ierr = PetscMemzero(aij->a + ii[lrows[i]],(ii[lrows[i]+1] - ii[lrows[i]])*sizeof(PetscScalar));CHKERRQ(ierr); 927 } 928 929 /* loop over all elements of off process part of matrix zeroing removed columns*/ 930 if (aij->compressedrow.use){ 931 m = aij->compressedrow.nrows; 932 ii = aij->compressedrow.i; 933 ridx = aij->compressedrow.rindex; 934 for (i=0; i<m; i++){ 935 n = ii[i+1] - ii[i]; 936 aj = aij->j + ii[i]; 937 aa = aij->a + ii[i]; 938 939 for (j=0; j<n; j++) { 940 if (PetscAbsScalar(mask[*aj])) { 941 bb[*ridx] -= *aa*xx[*aj]; 942 *aa = 0.0; 943 } 944 aa++; 945 aj++; 946 } 947 ridx++; 948 } 949 } else { /* do not use compressed row format */ 950 m = l->B->rmap->n; 951 for (i=0; i<m; i++) { 952 n = ii[i+1] - ii[i]; 953 aj = aij->j + ii[i]; 954 aa = aij->a + ii[i]; 955 for (j=0; j<n; j++) { 956 if (PetscAbsScalar(mask[*aj])) { 957 bb[i] -= *aa*xx[*aj]; 958 *aa = 0.0; 959 } 960 aa++; 961 aj++; 962 } 963 } 964 } 965 966 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 967 ierr = VecRestoreArray(lmask,&mask);CHKERRQ(ierr); 968 ierr = VecRestoreArrayRead(l->lvec,&xx);CHKERRQ(ierr); 969 ierr = VecDestroy(lmask);CHKERRQ(ierr); 970 ierr = PetscFree(lrows);CHKERRQ(ierr); 971 972 /* wait on sends */ 973 if (nsends) { 974 ierr = PetscMalloc(nsends*sizeof(MPI_Status),&send_status);CHKERRQ(ierr); 975 ierr = MPI_Waitall(nsends,send_waits,send_status);CHKERRQ(ierr); 976 ierr = PetscFree(send_status);CHKERRQ(ierr); 977 } 978 ierr = PetscFree(send_waits);CHKERRQ(ierr); 979 ierr = PetscFree(svalues);CHKERRQ(ierr); 980 981 PetscFunctionReturn(0); 982 } 983 984 #undef __FUNCT__ 985 #define __FUNCT__ "MatMult_MPIAIJ" 986 PetscErrorCode MatMult_MPIAIJ(Mat A,Vec xx,Vec yy) 987 { 988 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 989 PetscErrorCode ierr; 990 PetscInt nt; 991 992 PetscFunctionBegin; 993 ierr = VecGetLocalSize(xx,&nt);CHKERRQ(ierr); 994 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); 995 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 996 ierr = (*a->A->ops->mult)(a->A,xx,yy);CHKERRQ(ierr); 997 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 998 ierr = (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);CHKERRQ(ierr); 999 PetscFunctionReturn(0); 1000 } 1001 1002 #undef __FUNCT__ 1003 #define __FUNCT__ "MatMultDiagonalBlock_MPIAIJ" 1004 PetscErrorCode MatMultDiagonalBlock_MPIAIJ(Mat A,Vec bb,Vec xx) 1005 { 1006 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1007 PetscErrorCode ierr; 1008 1009 PetscFunctionBegin; 1010 ierr = MatMultDiagonalBlock(a->A,bb,xx);CHKERRQ(ierr); 1011 PetscFunctionReturn(0); 1012 } 1013 1014 #undef __FUNCT__ 1015 #define __FUNCT__ "MatMultAdd_MPIAIJ" 1016 PetscErrorCode MatMultAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1017 { 1018 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1019 PetscErrorCode ierr; 1020 1021 PetscFunctionBegin; 1022 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1023 ierr = (*a->A->ops->multadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 1024 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1025 ierr = (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);CHKERRQ(ierr); 1026 PetscFunctionReturn(0); 1027 } 1028 1029 #undef __FUNCT__ 1030 #define __FUNCT__ "MatMultTranspose_MPIAIJ" 1031 PetscErrorCode MatMultTranspose_MPIAIJ(Mat A,Vec xx,Vec yy) 1032 { 1033 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1034 PetscErrorCode ierr; 1035 PetscBool merged; 1036 1037 PetscFunctionBegin; 1038 ierr = VecScatterGetMerged(a->Mvctx,&merged);CHKERRQ(ierr); 1039 /* do nondiagonal part */ 1040 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 1041 if (!merged) { 1042 /* send it on its way */ 1043 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1044 /* do local part */ 1045 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 1046 /* receive remote parts: note this assumes the values are not actually */ 1047 /* added in yy until the next line, */ 1048 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1049 } else { 1050 /* do local part */ 1051 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 1052 /* send it on its way */ 1053 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1054 /* values actually were received in the Begin() but we need to call this nop */ 1055 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1056 } 1057 PetscFunctionReturn(0); 1058 } 1059 1060 EXTERN_C_BEGIN 1061 #undef __FUNCT__ 1062 #define __FUNCT__ "MatIsTranspose_MPIAIJ" 1063 PetscErrorCode MatIsTranspose_MPIAIJ(Mat Amat,Mat Bmat,PetscReal tol,PetscBool *f) 1064 { 1065 MPI_Comm comm; 1066 Mat_MPIAIJ *Aij = (Mat_MPIAIJ *) Amat->data, *Bij; 1067 Mat Adia = Aij->A, Bdia, Aoff,Boff,*Aoffs,*Boffs; 1068 IS Me,Notme; 1069 PetscErrorCode ierr; 1070 PetscInt M,N,first,last,*notme,i; 1071 PetscMPIInt size; 1072 1073 PetscFunctionBegin; 1074 1075 /* Easy test: symmetric diagonal block */ 1076 Bij = (Mat_MPIAIJ *) Bmat->data; Bdia = Bij->A; 1077 ierr = MatIsTranspose(Adia,Bdia,tol,f);CHKERRQ(ierr); 1078 if (!*f) PetscFunctionReturn(0); 1079 ierr = PetscObjectGetComm((PetscObject)Amat,&comm);CHKERRQ(ierr); 1080 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1081 if (size == 1) PetscFunctionReturn(0); 1082 1083 /* Hard test: off-diagonal block. This takes a MatGetSubMatrix. */ 1084 ierr = MatGetSize(Amat,&M,&N);CHKERRQ(ierr); 1085 ierr = MatGetOwnershipRange(Amat,&first,&last);CHKERRQ(ierr); 1086 ierr = PetscMalloc((N-last+first)*sizeof(PetscInt),¬me);CHKERRQ(ierr); 1087 for (i=0; i<first; i++) notme[i] = i; 1088 for (i=last; i<M; i++) notme[i-last+first] = i; 1089 ierr = ISCreateGeneral(MPI_COMM_SELF,N-last+first,notme,PETSC_COPY_VALUES,&Notme);CHKERRQ(ierr); 1090 ierr = ISCreateStride(MPI_COMM_SELF,last-first,first,1,&Me);CHKERRQ(ierr); 1091 ierr = MatGetSubMatrices(Amat,1,&Me,&Notme,MAT_INITIAL_MATRIX,&Aoffs);CHKERRQ(ierr); 1092 Aoff = Aoffs[0]; 1093 ierr = MatGetSubMatrices(Bmat,1,&Notme,&Me,MAT_INITIAL_MATRIX,&Boffs);CHKERRQ(ierr); 1094 Boff = Boffs[0]; 1095 ierr = MatIsTranspose(Aoff,Boff,tol,f);CHKERRQ(ierr); 1096 ierr = MatDestroyMatrices(1,&Aoffs);CHKERRQ(ierr); 1097 ierr = MatDestroyMatrices(1,&Boffs);CHKERRQ(ierr); 1098 ierr = ISDestroy(Me);CHKERRQ(ierr); 1099 ierr = ISDestroy(Notme);CHKERRQ(ierr); 1100 1101 PetscFunctionReturn(0); 1102 } 1103 EXTERN_C_END 1104 1105 #undef __FUNCT__ 1106 #define __FUNCT__ "MatMultTransposeAdd_MPIAIJ" 1107 PetscErrorCode MatMultTransposeAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1108 { 1109 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1110 PetscErrorCode ierr; 1111 1112 PetscFunctionBegin; 1113 /* do nondiagonal part */ 1114 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 1115 /* send it on its way */ 1116 ierr = VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1117 /* do local part */ 1118 ierr = (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 1119 /* receive remote parts */ 1120 ierr = VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 1121 PetscFunctionReturn(0); 1122 } 1123 1124 /* 1125 This only works correctly for square matrices where the subblock A->A is the 1126 diagonal block 1127 */ 1128 #undef __FUNCT__ 1129 #define __FUNCT__ "MatGetDiagonal_MPIAIJ" 1130 PetscErrorCode MatGetDiagonal_MPIAIJ(Mat A,Vec v) 1131 { 1132 PetscErrorCode ierr; 1133 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1134 1135 PetscFunctionBegin; 1136 if (A->rmap->N != A->cmap->N) SETERRQ(((PetscObject)A)->comm,PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); 1137 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"); 1138 ierr = MatGetDiagonal(a->A,v);CHKERRQ(ierr); 1139 PetscFunctionReturn(0); 1140 } 1141 1142 #undef __FUNCT__ 1143 #define __FUNCT__ "MatScale_MPIAIJ" 1144 PetscErrorCode MatScale_MPIAIJ(Mat A,PetscScalar aa) 1145 { 1146 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1147 PetscErrorCode ierr; 1148 1149 PetscFunctionBegin; 1150 ierr = MatScale(a->A,aa);CHKERRQ(ierr); 1151 ierr = MatScale(a->B,aa);CHKERRQ(ierr); 1152 PetscFunctionReturn(0); 1153 } 1154 1155 #undef __FUNCT__ 1156 #define __FUNCT__ "MatDestroy_MPIAIJ" 1157 PetscErrorCode MatDestroy_MPIAIJ(Mat mat) 1158 { 1159 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1160 PetscErrorCode ierr; 1161 1162 PetscFunctionBegin; 1163 #if defined(PETSC_USE_LOG) 1164 PetscLogObjectState((PetscObject)mat,"Rows=%D, Cols=%D",mat->rmap->N,mat->cmap->N); 1165 #endif 1166 ierr = MatStashDestroy_Private(&mat->stash);CHKERRQ(ierr); 1167 if (aij->diag) {ierr = VecDestroy(aij->diag);CHKERRQ(ierr);} 1168 if (aij->A){ierr = MatDestroy(aij->A);CHKERRQ(ierr);} 1169 if (aij->B){ierr = MatDestroy(aij->B);CHKERRQ(ierr);} 1170 #if defined (PETSC_USE_CTABLE) 1171 if (aij->colmap) {ierr = PetscTableDestroy(aij->colmap);CHKERRQ(ierr);} 1172 #else 1173 ierr = PetscFree(aij->colmap);CHKERRQ(ierr); 1174 #endif 1175 ierr = PetscFree(aij->garray);CHKERRQ(ierr); 1176 if (aij->lvec) {ierr = VecDestroy(aij->lvec);CHKERRQ(ierr);} 1177 if (aij->Mvctx) {ierr = VecScatterDestroy(aij->Mvctx);CHKERRQ(ierr);} 1178 ierr = PetscFree2(aij->rowvalues,aij->rowindices);CHKERRQ(ierr); 1179 ierr = PetscFree(aij->ld);CHKERRQ(ierr); 1180 ierr = PetscFree(aij);CHKERRQ(ierr); 1181 1182 ierr = PetscObjectChangeTypeName((PetscObject)mat,0);CHKERRQ(ierr); 1183 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C","",PETSC_NULL);CHKERRQ(ierr); 1184 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C","",PETSC_NULL);CHKERRQ(ierr); 1185 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatGetDiagonalBlock_C","",PETSC_NULL);CHKERRQ(ierr); 1186 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatIsTranspose_C","",PETSC_NULL);CHKERRQ(ierr); 1187 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocation_C","",PETSC_NULL);CHKERRQ(ierr); 1188 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocationCSR_C","",PETSC_NULL);CHKERRQ(ierr); 1189 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatDiagonalScaleLocal_C","",PETSC_NULL);CHKERRQ(ierr); 1190 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpiaij_mpisbaij_C","",PETSC_NULL);CHKERRQ(ierr); 1191 PetscFunctionReturn(0); 1192 } 1193 1194 #undef __FUNCT__ 1195 #define __FUNCT__ "MatView_MPIAIJ_Binary" 1196 PetscErrorCode MatView_MPIAIJ_Binary(Mat mat,PetscViewer viewer) 1197 { 1198 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1199 Mat_SeqAIJ* A = (Mat_SeqAIJ*)aij->A->data; 1200 Mat_SeqAIJ* B = (Mat_SeqAIJ*)aij->B->data; 1201 PetscErrorCode ierr; 1202 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 1203 int fd; 1204 PetscInt nz,header[4],*row_lengths,*range=0,rlen,i; 1205 PetscInt nzmax,*column_indices,j,k,col,*garray = aij->garray,cnt,cstart = mat->cmap->rstart,rnz; 1206 PetscScalar *column_values; 1207 PetscInt message_count,flowcontrolcount; 1208 1209 PetscFunctionBegin; 1210 ierr = MPI_Comm_rank(((PetscObject)mat)->comm,&rank);CHKERRQ(ierr); 1211 ierr = MPI_Comm_size(((PetscObject)mat)->comm,&size);CHKERRQ(ierr); 1212 nz = A->nz + B->nz; 1213 if (!rank) { 1214 header[0] = MAT_FILE_CLASSID; 1215 header[1] = mat->rmap->N; 1216 header[2] = mat->cmap->N; 1217 ierr = MPI_Reduce(&nz,&header[3],1,MPIU_INT,MPI_SUM,0,((PetscObject)mat)->comm);CHKERRQ(ierr); 1218 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 1219 ierr = PetscBinaryWrite(fd,header,4,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1220 /* get largest number of rows any processor has */ 1221 rlen = mat->rmap->n; 1222 range = mat->rmap->range; 1223 for (i=1; i<size; i++) { 1224 rlen = PetscMax(rlen,range[i+1] - range[i]); 1225 } 1226 } else { 1227 ierr = MPI_Reduce(&nz,0,1,MPIU_INT,MPI_SUM,0,((PetscObject)mat)->comm);CHKERRQ(ierr); 1228 rlen = mat->rmap->n; 1229 } 1230 1231 /* load up the local row counts */ 1232 ierr = PetscMalloc((rlen+1)*sizeof(PetscInt),&row_lengths);CHKERRQ(ierr); 1233 for (i=0; i<mat->rmap->n; i++) { 1234 row_lengths[i] = A->i[i+1] - A->i[i] + B->i[i+1] - B->i[i]; 1235 } 1236 1237 /* store the row lengths to the file */ 1238 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1239 if (!rank) { 1240 MPI_Status status; 1241 ierr = PetscBinaryWrite(fd,row_lengths,mat->rmap->n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1242 for (i=1; i<size; i++) { 1243 ierr = PetscViewerFlowControlStepMaster(viewer,i,message_count,flowcontrolcount);CHKERRQ(ierr); 1244 rlen = range[i+1] - range[i]; 1245 ierr = MPI_Recv(row_lengths,rlen,MPIU_INT,i,tag,((PetscObject)mat)->comm,&status);CHKERRQ(ierr); 1246 ierr = PetscBinaryWrite(fd,row_lengths,rlen,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1247 } 1248 ierr = PetscViewerFlowControlEndMaster(viewer,message_count);CHKERRQ(ierr); 1249 } else { 1250 ierr = PetscViewerFlowControlStepWorker(viewer,rank,message_count);CHKERRQ(ierr); 1251 ierr = MPI_Send(row_lengths,mat->rmap->n,MPIU_INT,0,tag,((PetscObject)mat)->comm);CHKERRQ(ierr); 1252 ierr = PetscViewerFlowControlEndWorker(viewer,message_count);CHKERRQ(ierr); 1253 } 1254 ierr = PetscFree(row_lengths);CHKERRQ(ierr); 1255 1256 /* load up the local column indices */ 1257 nzmax = nz; /* )th processor needs space a largest processor needs */ 1258 ierr = MPI_Reduce(&nz,&nzmax,1,MPIU_INT,MPI_MAX,0,((PetscObject)mat)->comm);CHKERRQ(ierr); 1259 ierr = PetscMalloc((nzmax+1)*sizeof(PetscInt),&column_indices);CHKERRQ(ierr); 1260 cnt = 0; 1261 for (i=0; i<mat->rmap->n; i++) { 1262 for (j=B->i[i]; j<B->i[i+1]; j++) { 1263 if ( (col = garray[B->j[j]]) > cstart) break; 1264 column_indices[cnt++] = col; 1265 } 1266 for (k=A->i[i]; k<A->i[i+1]; k++) { 1267 column_indices[cnt++] = A->j[k] + cstart; 1268 } 1269 for (; j<B->i[i+1]; j++) { 1270 column_indices[cnt++] = garray[B->j[j]]; 1271 } 1272 } 1273 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); 1274 1275 /* store the column indices to the file */ 1276 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1277 if (!rank) { 1278 MPI_Status status; 1279 ierr = PetscBinaryWrite(fd,column_indices,nz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1280 for (i=1; i<size; i++) { 1281 ierr = PetscViewerFlowControlStepMaster(viewer,i,message_count,flowcontrolcount);CHKERRQ(ierr); 1282 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,((PetscObject)mat)->comm,&status);CHKERRQ(ierr); 1283 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1284 ierr = MPI_Recv(column_indices,rnz,MPIU_INT,i,tag,((PetscObject)mat)->comm,&status);CHKERRQ(ierr); 1285 ierr = PetscBinaryWrite(fd,column_indices,rnz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 1286 } 1287 ierr = PetscViewerFlowControlEndMaster(viewer,message_count);CHKERRQ(ierr); 1288 } else { 1289 ierr = PetscViewerFlowControlStepWorker(viewer,rank,message_count);CHKERRQ(ierr); 1290 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,((PetscObject)mat)->comm);CHKERRQ(ierr); 1291 ierr = MPI_Send(column_indices,nz,MPIU_INT,0,tag,((PetscObject)mat)->comm);CHKERRQ(ierr); 1292 ierr = PetscViewerFlowControlEndWorker(viewer,message_count);CHKERRQ(ierr); 1293 } 1294 ierr = PetscFree(column_indices);CHKERRQ(ierr); 1295 1296 /* load up the local column values */ 1297 ierr = PetscMalloc((nzmax+1)*sizeof(PetscScalar),&column_values);CHKERRQ(ierr); 1298 cnt = 0; 1299 for (i=0; i<mat->rmap->n; i++) { 1300 for (j=B->i[i]; j<B->i[i+1]; j++) { 1301 if ( garray[B->j[j]] > cstart) break; 1302 column_values[cnt++] = B->a[j]; 1303 } 1304 for (k=A->i[i]; k<A->i[i+1]; k++) { 1305 column_values[cnt++] = A->a[k]; 1306 } 1307 for (; j<B->i[i+1]; j++) { 1308 column_values[cnt++] = B->a[j]; 1309 } 1310 } 1311 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); 1312 1313 /* store the column values to the file */ 1314 ierr = PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);CHKERRQ(ierr); 1315 if (!rank) { 1316 MPI_Status status; 1317 ierr = PetscBinaryWrite(fd,column_values,nz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1318 for (i=1; i<size; i++) { 1319 ierr = PetscViewerFlowControlStepMaster(viewer,i,message_count,flowcontrolcount);CHKERRQ(ierr); 1320 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,((PetscObject)mat)->comm,&status);CHKERRQ(ierr); 1321 if (rnz > nzmax) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 1322 ierr = MPI_Recv(column_values,rnz,MPIU_SCALAR,i,tag,((PetscObject)mat)->comm,&status);CHKERRQ(ierr); 1323 ierr = PetscBinaryWrite(fd,column_values,rnz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 1324 } 1325 ierr = PetscViewerFlowControlEndMaster(viewer,message_count);CHKERRQ(ierr); 1326 } else { 1327 ierr = PetscViewerFlowControlStepWorker(viewer,rank,message_count);CHKERRQ(ierr); 1328 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,((PetscObject)mat)->comm);CHKERRQ(ierr); 1329 ierr = MPI_Send(column_values,nz,MPIU_SCALAR,0,tag,((PetscObject)mat)->comm);CHKERRQ(ierr); 1330 ierr = PetscViewerFlowControlEndWorker(viewer,message_count);CHKERRQ(ierr); 1331 } 1332 ierr = PetscFree(column_values);CHKERRQ(ierr); 1333 PetscFunctionReturn(0); 1334 } 1335 1336 #undef __FUNCT__ 1337 #define __FUNCT__ "MatView_MPIAIJ_ASCIIorDraworSocket" 1338 PetscErrorCode MatView_MPIAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer) 1339 { 1340 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1341 PetscErrorCode ierr; 1342 PetscMPIInt rank = aij->rank,size = aij->size; 1343 PetscBool isdraw,iascii,isbinary; 1344 PetscViewer sviewer; 1345 PetscViewerFormat format; 1346 1347 PetscFunctionBegin; 1348 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1349 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1350 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1351 if (iascii) { 1352 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 1353 if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 1354 MatInfo info; 1355 PetscBool inodes; 1356 1357 ierr = MPI_Comm_rank(((PetscObject)mat)->comm,&rank);CHKERRQ(ierr); 1358 ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr); 1359 ierr = MatInodeGetInodeSizes(aij->A,PETSC_NULL,(PetscInt **)&inodes,PETSC_NULL);CHKERRQ(ierr); 1360 if (!inodes) { 1361 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, not using I-node routines\n", 1362 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1363 } else { 1364 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, using I-node routines\n", 1365 rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 1366 } 1367 ierr = MatGetInfo(aij->A,MAT_LOCAL,&info);CHKERRQ(ierr); 1368 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1369 ierr = MatGetInfo(aij->B,MAT_LOCAL,&info);CHKERRQ(ierr); 1370 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 1371 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 1372 ierr = PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");CHKERRQ(ierr); 1373 ierr = VecScatterView(aij->Mvctx,viewer);CHKERRQ(ierr); 1374 PetscFunctionReturn(0); 1375 } else if (format == PETSC_VIEWER_ASCII_INFO) { 1376 PetscInt inodecount,inodelimit,*inodes; 1377 ierr = MatInodeGetInodeSizes(aij->A,&inodecount,&inodes,&inodelimit);CHKERRQ(ierr); 1378 if (inodes) { 1379 ierr = PetscViewerASCIIPrintf(viewer,"using I-node (on process 0) routines: found %D nodes, limit used is %D\n",inodecount,inodelimit);CHKERRQ(ierr); 1380 } else { 1381 ierr = PetscViewerASCIIPrintf(viewer,"not using I-node (on process 0) routines\n");CHKERRQ(ierr); 1382 } 1383 PetscFunctionReturn(0); 1384 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) { 1385 PetscFunctionReturn(0); 1386 } 1387 } else if (isbinary) { 1388 if (size == 1) { 1389 ierr = PetscObjectSetName((PetscObject)aij->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1390 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 1391 } else { 1392 ierr = MatView_MPIAIJ_Binary(mat,viewer);CHKERRQ(ierr); 1393 } 1394 PetscFunctionReturn(0); 1395 } else if (isdraw) { 1396 PetscDraw draw; 1397 PetscBool isnull; 1398 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 1399 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); if (isnull) PetscFunctionReturn(0); 1400 } 1401 1402 if (size == 1) { 1403 ierr = PetscObjectSetName((PetscObject)aij->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1404 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 1405 } else { 1406 /* assemble the entire matrix onto first processor. */ 1407 Mat A; 1408 Mat_SeqAIJ *Aloc; 1409 PetscInt M = mat->rmap->N,N = mat->cmap->N,m,*ai,*aj,row,*cols,i,*ct; 1410 MatScalar *a; 1411 1412 if (mat->rmap->N > 1024) { 1413 PetscBool flg = PETSC_FALSE; 1414 1415 ierr = PetscOptionsGetBool(((PetscObject) mat)->prefix, "-mat_ascii_output_large", &flg,PETSC_NULL);CHKERRQ(ierr); 1416 if (!flg) { 1417 SETERRQ(((PetscObject)mat)->comm,PETSC_ERR_ARG_OUTOFRANGE,"ASCII matrix output not allowed for matrices with more than 1024 rows, use binary format instead.\nYou can override this restriction using -mat_ascii_output_large."); 1418 } 1419 } 1420 1421 ierr = MatCreate(((PetscObject)mat)->comm,&A);CHKERRQ(ierr); 1422 if (!rank) { 1423 ierr = MatSetSizes(A,M,N,M,N);CHKERRQ(ierr); 1424 } else { 1425 ierr = MatSetSizes(A,0,0,M,N);CHKERRQ(ierr); 1426 } 1427 /* This is just a temporary matrix, so explicitly using MATMPIAIJ is probably best */ 1428 ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr); 1429 ierr = MatMPIAIJSetPreallocation(A,0,PETSC_NULL,0,PETSC_NULL);CHKERRQ(ierr); 1430 ierr = PetscLogObjectParent(mat,A);CHKERRQ(ierr); 1431 1432 /* copy over the A part */ 1433 Aloc = (Mat_SeqAIJ*)aij->A->data; 1434 m = aij->A->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1435 row = mat->rmap->rstart; 1436 for (i=0; i<ai[m]; i++) {aj[i] += mat->cmap->rstart ;} 1437 for (i=0; i<m; i++) { 1438 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],aj,a,INSERT_VALUES);CHKERRQ(ierr); 1439 row++; a += ai[i+1]-ai[i]; aj += ai[i+1]-ai[i]; 1440 } 1441 aj = Aloc->j; 1442 for (i=0; i<ai[m]; i++) {aj[i] -= mat->cmap->rstart;} 1443 1444 /* copy over the B part */ 1445 Aloc = (Mat_SeqAIJ*)aij->B->data; 1446 m = aij->B->rmap->n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 1447 row = mat->rmap->rstart; 1448 ierr = PetscMalloc((ai[m]+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 1449 ct = cols; 1450 for (i=0; i<ai[m]; i++) {cols[i] = aij->garray[aj[i]];} 1451 for (i=0; i<m; i++) { 1452 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],cols,a,INSERT_VALUES);CHKERRQ(ierr); 1453 row++; a += ai[i+1]-ai[i]; cols += ai[i+1]-ai[i]; 1454 } 1455 ierr = PetscFree(ct);CHKERRQ(ierr); 1456 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1457 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1458 /* 1459 Everyone has to call to draw the matrix since the graphics waits are 1460 synchronized across all processors that share the PetscDraw object 1461 */ 1462 ierr = PetscViewerGetSingleton(viewer,&sviewer);CHKERRQ(ierr); 1463 if (!rank) { 1464 ierr = PetscObjectSetName((PetscObject)((Mat_MPIAIJ*)(A->data))->A,((PetscObject)mat)->name);CHKERRQ(ierr); 1465 /* Set the type name to MATMPIAIJ so that the correct type can be printed out by PetscObjectPrintClassNamePrefixType() in MatView_SeqAIJ_ASCII()*/ 1466 PetscStrcpy(((PetscObject)((Mat_MPIAIJ*)(A->data))->A)->type_name,MATMPIAIJ); 1467 ierr = MatView(((Mat_MPIAIJ*)(A->data))->A,sviewer);CHKERRQ(ierr); 1468 } 1469 ierr = PetscViewerRestoreSingleton(viewer,&sviewer);CHKERRQ(ierr); 1470 ierr = MatDestroy(A);CHKERRQ(ierr); 1471 } 1472 PetscFunctionReturn(0); 1473 } 1474 1475 #undef __FUNCT__ 1476 #define __FUNCT__ "MatView_MPIAIJ" 1477 PetscErrorCode MatView_MPIAIJ(Mat mat,PetscViewer viewer) 1478 { 1479 PetscErrorCode ierr; 1480 PetscBool iascii,isdraw,issocket,isbinary; 1481 1482 PetscFunctionBegin; 1483 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 1484 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 1485 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 1486 ierr = PetscTypeCompare((PetscObject)viewer,PETSCVIEWERSOCKET,&issocket);CHKERRQ(ierr); 1487 if (iascii || isdraw || isbinary || issocket) { 1488 ierr = MatView_MPIAIJ_ASCIIorDraworSocket(mat,viewer);CHKERRQ(ierr); 1489 } else { 1490 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"Viewer type %s not supported by MPIAIJ matrices",((PetscObject)viewer)->type_name); 1491 } 1492 PetscFunctionReturn(0); 1493 } 1494 1495 #undef __FUNCT__ 1496 #define __FUNCT__ "MatSOR_MPIAIJ" 1497 PetscErrorCode MatSOR_MPIAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1498 { 1499 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1500 PetscErrorCode ierr; 1501 Vec bb1 = 0; 1502 PetscBool hasop; 1503 1504 PetscFunctionBegin; 1505 if (its > 1 || ~flag & SOR_ZERO_INITIAL_GUESS || flag & SOR_EISENSTAT) { 1506 ierr = VecDuplicate(bb,&bb1);CHKERRQ(ierr); 1507 } 1508 1509 if (flag == SOR_APPLY_UPPER) { 1510 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1511 PetscFunctionReturn(0); 1512 } 1513 1514 if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){ 1515 if (flag & SOR_ZERO_INITIAL_GUESS) { 1516 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1517 its--; 1518 } 1519 1520 while (its--) { 1521 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1522 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1523 1524 /* update rhs: bb1 = bb - B*x */ 1525 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1526 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1527 1528 /* local sweep */ 1529 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1530 } 1531 } else if (flag & SOR_LOCAL_FORWARD_SWEEP){ 1532 if (flag & SOR_ZERO_INITIAL_GUESS) { 1533 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1534 its--; 1535 } 1536 while (its--) { 1537 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1538 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1539 1540 /* update rhs: bb1 = bb - B*x */ 1541 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1542 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1543 1544 /* local sweep */ 1545 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_FORWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1546 } 1547 } else if (flag & SOR_LOCAL_BACKWARD_SWEEP){ 1548 if (flag & SOR_ZERO_INITIAL_GUESS) { 1549 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);CHKERRQ(ierr); 1550 its--; 1551 } 1552 while (its--) { 1553 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1554 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1555 1556 /* update rhs: bb1 = bb - B*x */ 1557 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1558 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1559 1560 /* local sweep */ 1561 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_BACKWARD_SWEEP,fshift,lits,1,xx);CHKERRQ(ierr); 1562 } 1563 } else if (flag & SOR_EISENSTAT) { 1564 Vec xx1; 1565 1566 ierr = VecDuplicate(bb,&xx1);CHKERRQ(ierr); 1567 ierr = (*mat->A->ops->sor)(mat->A,bb,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_BACKWARD_SWEEP),fshift,lits,1,xx);CHKERRQ(ierr); 1568 1569 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1570 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1571 if (!mat->diag) { 1572 ierr = MatGetVecs(matin,&mat->diag,PETSC_NULL);CHKERRQ(ierr); 1573 ierr = MatGetDiagonal(matin,mat->diag);CHKERRQ(ierr); 1574 } 1575 ierr = MatHasOperation(matin,MATOP_MULT_DIAGONAL_BLOCK,&hasop);CHKERRQ(ierr); 1576 if (hasop) { 1577 ierr = MatMultDiagonalBlock(matin,xx,bb1);CHKERRQ(ierr); 1578 } else { 1579 ierr = VecPointwiseMult(bb1,mat->diag,xx);CHKERRQ(ierr); 1580 } 1581 ierr = VecAYPX(bb1,(omega-2.0)/omega,bb);CHKERRQ(ierr); 1582 1583 ierr = MatMultAdd(mat->B,mat->lvec,bb1,bb1);CHKERRQ(ierr); 1584 1585 /* local sweep */ 1586 ierr = (*mat->A->ops->sor)(mat->A,bb1,omega,(MatSORType)(SOR_ZERO_INITIAL_GUESS | SOR_LOCAL_FORWARD_SWEEP),fshift,lits,1,xx1);CHKERRQ(ierr); 1587 ierr = VecAXPY(xx,1.0,xx1);CHKERRQ(ierr); 1588 ierr = VecDestroy(xx1);CHKERRQ(ierr); 1589 } else { 1590 SETERRQ(((PetscObject)mat)->comm,PETSC_ERR_SUP,"Parallel SOR not supported"); 1591 } 1592 1593 if (bb1) {ierr = VecDestroy(bb1);CHKERRQ(ierr);} 1594 PetscFunctionReturn(0); 1595 } 1596 1597 #undef __FUNCT__ 1598 #define __FUNCT__ "MatPermute_MPIAIJ" 1599 PetscErrorCode MatPermute_MPIAIJ(Mat A,IS rowp,IS colp,Mat *B) 1600 { 1601 MPI_Comm comm,pcomm; 1602 PetscInt first,local_size,nrows; 1603 const PetscInt *rows; 1604 PetscMPIInt size; 1605 IS crowp,growp,irowp,lrowp,lcolp,icolp; 1606 PetscErrorCode ierr; 1607 1608 PetscFunctionBegin; 1609 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1610 /* make a collective version of 'rowp' */ 1611 ierr = PetscObjectGetComm((PetscObject)rowp,&pcomm);CHKERRQ(ierr); 1612 if (pcomm==comm) { 1613 crowp = rowp; 1614 } else { 1615 ierr = ISGetSize(rowp,&nrows);CHKERRQ(ierr); 1616 ierr = ISGetIndices(rowp,&rows);CHKERRQ(ierr); 1617 ierr = ISCreateGeneral(comm,nrows,rows,PETSC_COPY_VALUES,&crowp);CHKERRQ(ierr); 1618 ierr = ISRestoreIndices(rowp,&rows);CHKERRQ(ierr); 1619 } 1620 /* collect the global row permutation and invert it */ 1621 ierr = ISAllGather(crowp,&growp);CHKERRQ(ierr); 1622 ierr = ISSetPermutation(growp);CHKERRQ(ierr); 1623 if (pcomm!=comm) { 1624 ierr = ISDestroy(crowp);CHKERRQ(ierr); 1625 } 1626 ierr = ISInvertPermutation(growp,PETSC_DECIDE,&irowp);CHKERRQ(ierr); 1627 /* get the local target indices */ 1628 ierr = MatGetOwnershipRange(A,&first,PETSC_NULL);CHKERRQ(ierr); 1629 ierr = MatGetLocalSize(A,&local_size,PETSC_NULL);CHKERRQ(ierr); 1630 ierr = ISGetIndices(irowp,&rows);CHKERRQ(ierr); 1631 ierr = ISCreateGeneral(MPI_COMM_SELF,local_size,rows+first,PETSC_COPY_VALUES,&lrowp);CHKERRQ(ierr); 1632 ierr = ISRestoreIndices(irowp,&rows);CHKERRQ(ierr); 1633 ierr = ISDestroy(irowp);CHKERRQ(ierr); 1634 /* the column permutation is so much easier; 1635 make a local version of 'colp' and invert it */ 1636 ierr = PetscObjectGetComm((PetscObject)colp,&pcomm);CHKERRQ(ierr); 1637 ierr = MPI_Comm_size(pcomm,&size);CHKERRQ(ierr); 1638 if (size==1) { 1639 lcolp = colp; 1640 } else { 1641 ierr = ISGetSize(colp,&nrows);CHKERRQ(ierr); 1642 ierr = ISGetIndices(colp,&rows);CHKERRQ(ierr); 1643 ierr = ISCreateGeneral(MPI_COMM_SELF,nrows,rows,PETSC_COPY_VALUES,&lcolp);CHKERRQ(ierr); 1644 } 1645 ierr = ISSetPermutation(lcolp);CHKERRQ(ierr); 1646 ierr = ISInvertPermutation(lcolp,PETSC_DECIDE,&icolp);CHKERRQ(ierr); 1647 ierr = ISSetPermutation(icolp);CHKERRQ(ierr); 1648 if (size>1) { 1649 ierr = ISRestoreIndices(colp,&rows);CHKERRQ(ierr); 1650 ierr = ISDestroy(lcolp);CHKERRQ(ierr); 1651 } 1652 /* now we just get the submatrix */ 1653 ierr = MatGetSubMatrix_MPIAIJ_Private(A,lrowp,icolp,local_size,MAT_INITIAL_MATRIX,B);CHKERRQ(ierr); 1654 /* clean up */ 1655 ierr = ISDestroy(lrowp);CHKERRQ(ierr); 1656 ierr = ISDestroy(icolp);CHKERRQ(ierr); 1657 PetscFunctionReturn(0); 1658 } 1659 1660 #undef __FUNCT__ 1661 #define __FUNCT__ "MatGetInfo_MPIAIJ" 1662 PetscErrorCode MatGetInfo_MPIAIJ(Mat matin,MatInfoType flag,MatInfo *info) 1663 { 1664 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1665 Mat A = mat->A,B = mat->B; 1666 PetscErrorCode ierr; 1667 PetscReal isend[5],irecv[5]; 1668 1669 PetscFunctionBegin; 1670 info->block_size = 1.0; 1671 ierr = MatGetInfo(A,MAT_LOCAL,info);CHKERRQ(ierr); 1672 isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded; 1673 isend[3] = info->memory; isend[4] = info->mallocs; 1674 ierr = MatGetInfo(B,MAT_LOCAL,info);CHKERRQ(ierr); 1675 isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded; 1676 isend[3] += info->memory; isend[4] += info->mallocs; 1677 if (flag == MAT_LOCAL) { 1678 info->nz_used = isend[0]; 1679 info->nz_allocated = isend[1]; 1680 info->nz_unneeded = isend[2]; 1681 info->memory = isend[3]; 1682 info->mallocs = isend[4]; 1683 } else if (flag == MAT_GLOBAL_MAX) { 1684 ierr = MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_MAX,((PetscObject)matin)->comm);CHKERRQ(ierr); 1685 info->nz_used = irecv[0]; 1686 info->nz_allocated = irecv[1]; 1687 info->nz_unneeded = irecv[2]; 1688 info->memory = irecv[3]; 1689 info->mallocs = irecv[4]; 1690 } else if (flag == MAT_GLOBAL_SUM) { 1691 ierr = MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_SUM,((PetscObject)matin)->comm);CHKERRQ(ierr); 1692 info->nz_used = irecv[0]; 1693 info->nz_allocated = irecv[1]; 1694 info->nz_unneeded = irecv[2]; 1695 info->memory = irecv[3]; 1696 info->mallocs = irecv[4]; 1697 } 1698 info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */ 1699 info->fill_ratio_needed = 0; 1700 info->factor_mallocs = 0; 1701 1702 PetscFunctionReturn(0); 1703 } 1704 1705 #undef __FUNCT__ 1706 #define __FUNCT__ "MatSetOption_MPIAIJ" 1707 PetscErrorCode MatSetOption_MPIAIJ(Mat A,MatOption op,PetscBool flg) 1708 { 1709 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1710 PetscErrorCode ierr; 1711 1712 PetscFunctionBegin; 1713 switch (op) { 1714 case MAT_NEW_NONZERO_LOCATIONS: 1715 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1716 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1717 case MAT_KEEP_NONZERO_PATTERN: 1718 case MAT_NEW_NONZERO_LOCATION_ERR: 1719 case MAT_USE_INODES: 1720 case MAT_IGNORE_ZERO_ENTRIES: 1721 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1722 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1723 break; 1724 case MAT_ROW_ORIENTED: 1725 a->roworiented = flg; 1726 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1727 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1728 break; 1729 case MAT_NEW_DIAGONALS: 1730 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1731 break; 1732 case MAT_IGNORE_OFF_PROC_ENTRIES: 1733 a->donotstash = PETSC_TRUE; 1734 break; 1735 case MAT_SPD: 1736 A->spd_set = PETSC_TRUE; 1737 A->spd = flg; 1738 if (flg) { 1739 A->symmetric = PETSC_TRUE; 1740 A->structurally_symmetric = PETSC_TRUE; 1741 A->symmetric_set = PETSC_TRUE; 1742 A->structurally_symmetric_set = PETSC_TRUE; 1743 } 1744 break; 1745 case MAT_SYMMETRIC: 1746 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1747 break; 1748 case MAT_STRUCTURALLY_SYMMETRIC: 1749 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1750 break; 1751 case MAT_HERMITIAN: 1752 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1753 break; 1754 case MAT_SYMMETRY_ETERNAL: 1755 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1756 break; 1757 default: 1758 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1759 } 1760 PetscFunctionReturn(0); 1761 } 1762 1763 #undef __FUNCT__ 1764 #define __FUNCT__ "MatGetRow_MPIAIJ" 1765 PetscErrorCode MatGetRow_MPIAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1766 { 1767 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1768 PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p; 1769 PetscErrorCode ierr; 1770 PetscInt i,*cworkA,*cworkB,**pcA,**pcB,cstart = matin->cmap->rstart; 1771 PetscInt nztot,nzA,nzB,lrow,rstart = matin->rmap->rstart,rend = matin->rmap->rend; 1772 PetscInt *cmap,*idx_p; 1773 1774 PetscFunctionBegin; 1775 if (mat->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Already active"); 1776 mat->getrowactive = PETSC_TRUE; 1777 1778 if (!mat->rowvalues && (idx || v)) { 1779 /* 1780 allocate enough space to hold information from the longest row. 1781 */ 1782 Mat_SeqAIJ *Aa = (Mat_SeqAIJ*)mat->A->data,*Ba = (Mat_SeqAIJ*)mat->B->data; 1783 PetscInt max = 1,tmp; 1784 for (i=0; i<matin->rmap->n; i++) { 1785 tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; 1786 if (max < tmp) { max = tmp; } 1787 } 1788 ierr = PetscMalloc2(max,PetscScalar,&mat->rowvalues,max,PetscInt,&mat->rowindices);CHKERRQ(ierr); 1789 } 1790 1791 if (row < rstart || row >= rend) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Only local rows"); 1792 lrow = row - rstart; 1793 1794 pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB; 1795 if (!v) {pvA = 0; pvB = 0;} 1796 if (!idx) {pcA = 0; if (!v) pcB = 0;} 1797 ierr = (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1798 ierr = (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1799 nztot = nzA + nzB; 1800 1801 cmap = mat->garray; 1802 if (v || idx) { 1803 if (nztot) { 1804 /* Sort by increasing column numbers, assuming A and B already sorted */ 1805 PetscInt imark = -1; 1806 if (v) { 1807 *v = v_p = mat->rowvalues; 1808 for (i=0; i<nzB; i++) { 1809 if (cmap[cworkB[i]] < cstart) v_p[i] = vworkB[i]; 1810 else break; 1811 } 1812 imark = i; 1813 for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i]; 1814 for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i]; 1815 } 1816 if (idx) { 1817 *idx = idx_p = mat->rowindices; 1818 if (imark > -1) { 1819 for (i=0; i<imark; i++) { 1820 idx_p[i] = cmap[cworkB[i]]; 1821 } 1822 } else { 1823 for (i=0; i<nzB; i++) { 1824 if (cmap[cworkB[i]] < cstart) idx_p[i] = cmap[cworkB[i]]; 1825 else break; 1826 } 1827 imark = i; 1828 } 1829 for (i=0; i<nzA; i++) idx_p[imark+i] = cstart + cworkA[i]; 1830 for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]]; 1831 } 1832 } else { 1833 if (idx) *idx = 0; 1834 if (v) *v = 0; 1835 } 1836 } 1837 *nz = nztot; 1838 ierr = (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1839 ierr = (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1840 PetscFunctionReturn(0); 1841 } 1842 1843 #undef __FUNCT__ 1844 #define __FUNCT__ "MatRestoreRow_MPIAIJ" 1845 PetscErrorCode MatRestoreRow_MPIAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1846 { 1847 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1848 1849 PetscFunctionBegin; 1850 if (!aij->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"MatGetRow() must be called first"); 1851 aij->getrowactive = PETSC_FALSE; 1852 PetscFunctionReturn(0); 1853 } 1854 1855 #undef __FUNCT__ 1856 #define __FUNCT__ "MatNorm_MPIAIJ" 1857 PetscErrorCode MatNorm_MPIAIJ(Mat mat,NormType type,PetscReal *norm) 1858 { 1859 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1860 Mat_SeqAIJ *amat = (Mat_SeqAIJ*)aij->A->data,*bmat = (Mat_SeqAIJ*)aij->B->data; 1861 PetscErrorCode ierr; 1862 PetscInt i,j,cstart = mat->cmap->rstart; 1863 PetscReal sum = 0.0; 1864 MatScalar *v; 1865 1866 PetscFunctionBegin; 1867 if (aij->size == 1) { 1868 ierr = MatNorm(aij->A,type,norm);CHKERRQ(ierr); 1869 } else { 1870 if (type == NORM_FROBENIUS) { 1871 v = amat->a; 1872 for (i=0; i<amat->nz; i++) { 1873 #if defined(PETSC_USE_COMPLEX) 1874 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1875 #else 1876 sum += (*v)*(*v); v++; 1877 #endif 1878 } 1879 v = bmat->a; 1880 for (i=0; i<bmat->nz; i++) { 1881 #if defined(PETSC_USE_COMPLEX) 1882 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1883 #else 1884 sum += (*v)*(*v); v++; 1885 #endif 1886 } 1887 ierr = MPI_Allreduce(&sum,norm,1,MPIU_REAL,MPI_SUM,((PetscObject)mat)->comm);CHKERRQ(ierr); 1888 *norm = sqrt(*norm); 1889 } else if (type == NORM_1) { /* max column norm */ 1890 PetscReal *tmp,*tmp2; 1891 PetscInt *jj,*garray = aij->garray; 1892 ierr = PetscMalloc((mat->cmap->N+1)*sizeof(PetscReal),&tmp);CHKERRQ(ierr); 1893 ierr = PetscMalloc((mat->cmap->N+1)*sizeof(PetscReal),&tmp2);CHKERRQ(ierr); 1894 ierr = PetscMemzero(tmp,mat->cmap->N*sizeof(PetscReal));CHKERRQ(ierr); 1895 *norm = 0.0; 1896 v = amat->a; jj = amat->j; 1897 for (j=0; j<amat->nz; j++) { 1898 tmp[cstart + *jj++ ] += PetscAbsScalar(*v); v++; 1899 } 1900 v = bmat->a; jj = bmat->j; 1901 for (j=0; j<bmat->nz; j++) { 1902 tmp[garray[*jj++]] += PetscAbsScalar(*v); v++; 1903 } 1904 ierr = MPI_Allreduce(tmp,tmp2,mat->cmap->N,MPIU_REAL,MPI_SUM,((PetscObject)mat)->comm);CHKERRQ(ierr); 1905 for (j=0; j<mat->cmap->N; j++) { 1906 if (tmp2[j] > *norm) *norm = tmp2[j]; 1907 } 1908 ierr = PetscFree(tmp);CHKERRQ(ierr); 1909 ierr = PetscFree(tmp2);CHKERRQ(ierr); 1910 } else if (type == NORM_INFINITY) { /* max row norm */ 1911 PetscReal ntemp = 0.0; 1912 for (j=0; j<aij->A->rmap->n; j++) { 1913 v = amat->a + amat->i[j]; 1914 sum = 0.0; 1915 for (i=0; i<amat->i[j+1]-amat->i[j]; i++) { 1916 sum += PetscAbsScalar(*v); v++; 1917 } 1918 v = bmat->a + bmat->i[j]; 1919 for (i=0; i<bmat->i[j+1]-bmat->i[j]; i++) { 1920 sum += PetscAbsScalar(*v); v++; 1921 } 1922 if (sum > ntemp) ntemp = sum; 1923 } 1924 ierr = MPI_Allreduce(&ntemp,norm,1,MPIU_REAL,MPI_MAX,((PetscObject)mat)->comm);CHKERRQ(ierr); 1925 } else { 1926 SETERRQ(((PetscObject)mat)->comm,PETSC_ERR_SUP,"No support for two norm"); 1927 } 1928 } 1929 PetscFunctionReturn(0); 1930 } 1931 1932 #undef __FUNCT__ 1933 #define __FUNCT__ "MatTranspose_MPIAIJ" 1934 PetscErrorCode MatTranspose_MPIAIJ(Mat A,MatReuse reuse,Mat *matout) 1935 { 1936 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1937 Mat_SeqAIJ *Aloc=(Mat_SeqAIJ*)a->A->data,*Bloc=(Mat_SeqAIJ*)a->B->data; 1938 PetscErrorCode ierr; 1939 PetscInt M = A->rmap->N,N = A->cmap->N,ma,na,mb,*ai,*aj,*bi,*bj,row,*cols,*cols_tmp,i,*d_nnz; 1940 PetscInt cstart=A->cmap->rstart,ncol; 1941 Mat B; 1942 MatScalar *array; 1943 1944 PetscFunctionBegin; 1945 if (reuse == MAT_REUSE_MATRIX && A == *matout && M != N) SETERRQ(((PetscObject)A)->comm,PETSC_ERR_ARG_SIZ,"Square matrix only for in-place"); 1946 1947 ma = A->rmap->n; na = A->cmap->n; mb = a->B->rmap->n; 1948 ai = Aloc->i; aj = Aloc->j; 1949 bi = Bloc->i; bj = Bloc->j; 1950 if (reuse == MAT_INITIAL_MATRIX || *matout == A) { 1951 /* compute d_nnz for preallocation; o_nnz is approximated by d_nnz to avoid communication */ 1952 ierr = PetscMalloc((1+na)*sizeof(PetscInt),&d_nnz);CHKERRQ(ierr); 1953 ierr = PetscMemzero(d_nnz,(1+na)*sizeof(PetscInt));CHKERRQ(ierr); 1954 for (i=0; i<ai[ma]; i++){ 1955 d_nnz[aj[i]] ++; 1956 aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1957 } 1958 1959 ierr = MatCreate(((PetscObject)A)->comm,&B);CHKERRQ(ierr); 1960 ierr = MatSetSizes(B,A->cmap->n,A->rmap->n,N,M);CHKERRQ(ierr); 1961 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 1962 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,d_nnz);CHKERRQ(ierr); 1963 ierr = PetscFree(d_nnz);CHKERRQ(ierr); 1964 } else { 1965 B = *matout; 1966 } 1967 1968 /* copy over the A part */ 1969 array = Aloc->a; 1970 row = A->rmap->rstart; 1971 for (i=0; i<ma; i++) { 1972 ncol = ai[i+1]-ai[i]; 1973 ierr = MatSetValues(B,ncol,aj,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1974 row++; array += ncol; aj += ncol; 1975 } 1976 aj = Aloc->j; 1977 for (i=0; i<ai[ma]; i++) aj[i] -= cstart; /* resume local col index */ 1978 1979 /* copy over the B part */ 1980 ierr = PetscMalloc(bi[mb]*sizeof(PetscInt),&cols);CHKERRQ(ierr); 1981 ierr = PetscMemzero(cols,bi[mb]*sizeof(PetscInt));CHKERRQ(ierr); 1982 array = Bloc->a; 1983 row = A->rmap->rstart; 1984 for (i=0; i<bi[mb]; i++) {cols[i] = a->garray[bj[i]];} 1985 cols_tmp = cols; 1986 for (i=0; i<mb; i++) { 1987 ncol = bi[i+1]-bi[i]; 1988 ierr = MatSetValues(B,ncol,cols_tmp,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1989 row++; array += ncol; cols_tmp += ncol; 1990 } 1991 ierr = PetscFree(cols);CHKERRQ(ierr); 1992 1993 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1994 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1995 if (reuse == MAT_INITIAL_MATRIX || *matout != A) { 1996 *matout = B; 1997 } else { 1998 ierr = MatHeaderMerge(A,B);CHKERRQ(ierr); 1999 } 2000 PetscFunctionReturn(0); 2001 } 2002 2003 #undef __FUNCT__ 2004 #define __FUNCT__ "MatDiagonalScale_MPIAIJ" 2005 PetscErrorCode MatDiagonalScale_MPIAIJ(Mat mat,Vec ll,Vec rr) 2006 { 2007 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2008 Mat a = aij->A,b = aij->B; 2009 PetscErrorCode ierr; 2010 PetscInt s1,s2,s3; 2011 2012 PetscFunctionBegin; 2013 ierr = MatGetLocalSize(mat,&s2,&s3);CHKERRQ(ierr); 2014 if (rr) { 2015 ierr = VecGetLocalSize(rr,&s1);CHKERRQ(ierr); 2016 if (s1!=s3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"right vector non-conforming local size"); 2017 /* Overlap communication with computation. */ 2018 ierr = VecScatterBegin(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2019 } 2020 if (ll) { 2021 ierr = VecGetLocalSize(ll,&s1);CHKERRQ(ierr); 2022 if (s1!=s2) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"left vector non-conforming local size"); 2023 ierr = (*b->ops->diagonalscale)(b,ll,0);CHKERRQ(ierr); 2024 } 2025 /* scale the diagonal block */ 2026 ierr = (*a->ops->diagonalscale)(a,ll,rr);CHKERRQ(ierr); 2027 2028 if (rr) { 2029 /* Do a scatter end and then right scale the off-diagonal block */ 2030 ierr = VecScatterEnd(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2031 ierr = (*b->ops->diagonalscale)(b,0,aij->lvec);CHKERRQ(ierr); 2032 } 2033 2034 PetscFunctionReturn(0); 2035 } 2036 2037 #undef __FUNCT__ 2038 #define __FUNCT__ "MatSetBlockSize_MPIAIJ" 2039 PetscErrorCode MatSetBlockSize_MPIAIJ(Mat A,PetscInt bs) 2040 { 2041 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2042 PetscErrorCode ierr; 2043 2044 PetscFunctionBegin; 2045 ierr = MatSetBlockSize(a->A,bs);CHKERRQ(ierr); 2046 ierr = MatSetBlockSize(a->B,bs);CHKERRQ(ierr); 2047 ierr = PetscLayoutSetBlockSize(A->rmap,bs);CHKERRQ(ierr); 2048 ierr = PetscLayoutSetBlockSize(A->cmap,bs);CHKERRQ(ierr); 2049 PetscFunctionReturn(0); 2050 } 2051 #undef __FUNCT__ 2052 #define __FUNCT__ "MatSetUnfactored_MPIAIJ" 2053 PetscErrorCode MatSetUnfactored_MPIAIJ(Mat A) 2054 { 2055 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2056 PetscErrorCode ierr; 2057 2058 PetscFunctionBegin; 2059 ierr = MatSetUnfactored(a->A);CHKERRQ(ierr); 2060 PetscFunctionReturn(0); 2061 } 2062 2063 #undef __FUNCT__ 2064 #define __FUNCT__ "MatEqual_MPIAIJ" 2065 PetscErrorCode MatEqual_MPIAIJ(Mat A,Mat B,PetscBool *flag) 2066 { 2067 Mat_MPIAIJ *matB = (Mat_MPIAIJ*)B->data,*matA = (Mat_MPIAIJ*)A->data; 2068 Mat a,b,c,d; 2069 PetscBool flg; 2070 PetscErrorCode ierr; 2071 2072 PetscFunctionBegin; 2073 a = matA->A; b = matA->B; 2074 c = matB->A; d = matB->B; 2075 2076 ierr = MatEqual(a,c,&flg);CHKERRQ(ierr); 2077 if (flg) { 2078 ierr = MatEqual(b,d,&flg);CHKERRQ(ierr); 2079 } 2080 ierr = MPI_Allreduce(&flg,flag,1,MPI_INT,MPI_LAND,((PetscObject)A)->comm);CHKERRQ(ierr); 2081 PetscFunctionReturn(0); 2082 } 2083 2084 #undef __FUNCT__ 2085 #define __FUNCT__ "MatCopy_MPIAIJ" 2086 PetscErrorCode MatCopy_MPIAIJ(Mat A,Mat B,MatStructure str) 2087 { 2088 PetscErrorCode ierr; 2089 Mat_MPIAIJ *a = (Mat_MPIAIJ *)A->data; 2090 Mat_MPIAIJ *b = (Mat_MPIAIJ *)B->data; 2091 2092 PetscFunctionBegin; 2093 /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */ 2094 if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) { 2095 /* because of the column compression in the off-processor part of the matrix a->B, 2096 the number of columns in a->B and b->B may be different, hence we cannot call 2097 the MatCopy() directly on the two parts. If need be, we can provide a more 2098 efficient copy than the MatCopy_Basic() by first uncompressing the a->B matrices 2099 then copying the submatrices */ 2100 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2101 } else { 2102 ierr = MatCopy(a->A,b->A,str);CHKERRQ(ierr); 2103 ierr = MatCopy(a->B,b->B,str);CHKERRQ(ierr); 2104 } 2105 PetscFunctionReturn(0); 2106 } 2107 2108 #undef __FUNCT__ 2109 #define __FUNCT__ "MatSetUpPreallocation_MPIAIJ" 2110 PetscErrorCode MatSetUpPreallocation_MPIAIJ(Mat A) 2111 { 2112 PetscErrorCode ierr; 2113 2114 PetscFunctionBegin; 2115 ierr = MatMPIAIJSetPreallocation(A,PETSC_DEFAULT,0,PETSC_DEFAULT,0);CHKERRQ(ierr); 2116 PetscFunctionReturn(0); 2117 } 2118 2119 #undef __FUNCT__ 2120 #define __FUNCT__ "MatAXPY_MPIAIJ" 2121 PetscErrorCode MatAXPY_MPIAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 2122 { 2123 PetscErrorCode ierr; 2124 PetscInt i; 2125 Mat_MPIAIJ *xx = (Mat_MPIAIJ *)X->data,*yy = (Mat_MPIAIJ *)Y->data; 2126 PetscBLASInt bnz,one=1; 2127 Mat_SeqAIJ *x,*y; 2128 2129 PetscFunctionBegin; 2130 if (str == SAME_NONZERO_PATTERN) { 2131 PetscScalar alpha = a; 2132 x = (Mat_SeqAIJ *)xx->A->data; 2133 y = (Mat_SeqAIJ *)yy->A->data; 2134 bnz = PetscBLASIntCast(x->nz); 2135 BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one); 2136 x = (Mat_SeqAIJ *)xx->B->data; 2137 y = (Mat_SeqAIJ *)yy->B->data; 2138 bnz = PetscBLASIntCast(x->nz); 2139 BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one); 2140 } else if (str == SUBSET_NONZERO_PATTERN) { 2141 ierr = MatAXPY_SeqAIJ(yy->A,a,xx->A,str);CHKERRQ(ierr); 2142 2143 x = (Mat_SeqAIJ *)xx->B->data; 2144 y = (Mat_SeqAIJ *)yy->B->data; 2145 if (y->xtoy && y->XtoY != xx->B) { 2146 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 2147 ierr = MatDestroy(y->XtoY);CHKERRQ(ierr); 2148 } 2149 if (!y->xtoy) { /* get xtoy */ 2150 ierr = MatAXPYGetxtoy_Private(xx->B->rmap->n,x->i,x->j,xx->garray,y->i,y->j,yy->garray,&y->xtoy);CHKERRQ(ierr); 2151 y->XtoY = xx->B; 2152 ierr = PetscObjectReference((PetscObject)xx->B);CHKERRQ(ierr); 2153 } 2154 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 2155 } else { 2156 Mat B; 2157 PetscInt *nnz_d,*nnz_o; 2158 ierr = PetscMalloc(yy->A->rmap->N*sizeof(PetscInt),&nnz_d);CHKERRQ(ierr); 2159 ierr = PetscMalloc(yy->B->rmap->N*sizeof(PetscInt),&nnz_o);CHKERRQ(ierr); 2160 ierr = MatCreate(((PetscObject)Y)->comm,&B);CHKERRQ(ierr); 2161 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 2162 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 2163 ierr = MatSetType(B,MATMPIAIJ);CHKERRQ(ierr); 2164 ierr = MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); 2165 ierr = MatAXPYGetPreallocation_SeqAIJ(yy->B,xx->B,nnz_o);CHKERRQ(ierr); 2166 ierr = MatMPIAIJSetPreallocation(B,PETSC_NULL,nnz_d,PETSC_NULL,nnz_o);CHKERRQ(ierr); 2167 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 2168 ierr = MatHeaderReplace(Y,B); 2169 ierr = PetscFree(nnz_d);CHKERRQ(ierr); 2170 ierr = PetscFree(nnz_o);CHKERRQ(ierr); 2171 } 2172 PetscFunctionReturn(0); 2173 } 2174 2175 extern PetscErrorCode MatConjugate_SeqAIJ(Mat); 2176 2177 #undef __FUNCT__ 2178 #define __FUNCT__ "MatConjugate_MPIAIJ" 2179 PetscErrorCode MatConjugate_MPIAIJ(Mat mat) 2180 { 2181 #if defined(PETSC_USE_COMPLEX) 2182 PetscErrorCode ierr; 2183 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 2184 2185 PetscFunctionBegin; 2186 ierr = MatConjugate_SeqAIJ(aij->A);CHKERRQ(ierr); 2187 ierr = MatConjugate_SeqAIJ(aij->B);CHKERRQ(ierr); 2188 #else 2189 PetscFunctionBegin; 2190 #endif 2191 PetscFunctionReturn(0); 2192 } 2193 2194 #undef __FUNCT__ 2195 #define __FUNCT__ "MatRealPart_MPIAIJ" 2196 PetscErrorCode MatRealPart_MPIAIJ(Mat A) 2197 { 2198 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2199 PetscErrorCode ierr; 2200 2201 PetscFunctionBegin; 2202 ierr = MatRealPart(a->A);CHKERRQ(ierr); 2203 ierr = MatRealPart(a->B);CHKERRQ(ierr); 2204 PetscFunctionReturn(0); 2205 } 2206 2207 #undef __FUNCT__ 2208 #define __FUNCT__ "MatImaginaryPart_MPIAIJ" 2209 PetscErrorCode MatImaginaryPart_MPIAIJ(Mat A) 2210 { 2211 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2212 PetscErrorCode ierr; 2213 2214 PetscFunctionBegin; 2215 ierr = MatImaginaryPart(a->A);CHKERRQ(ierr); 2216 ierr = MatImaginaryPart(a->B);CHKERRQ(ierr); 2217 PetscFunctionReturn(0); 2218 } 2219 2220 #ifdef PETSC_HAVE_PBGL 2221 2222 #include <boost/parallel/mpi/bsp_process_group.hpp> 2223 #include <boost/graph/distributed/ilu_default_graph.hpp> 2224 #include <boost/graph/distributed/ilu_0_block.hpp> 2225 #include <boost/graph/distributed/ilu_preconditioner.hpp> 2226 #include <boost/graph/distributed/petsc/interface.hpp> 2227 #include <boost/multi_array.hpp> 2228 #include <boost/parallel/distributed_property_map->hpp> 2229 2230 #undef __FUNCT__ 2231 #define __FUNCT__ "MatILUFactorSymbolic_MPIAIJ" 2232 /* 2233 This uses the parallel ILU factorization of Peter Gottschling <pgottsch@osl.iu.edu> 2234 */ 2235 PetscErrorCode MatILUFactorSymbolic_MPIAIJ(Mat fact,Mat A, IS isrow, IS iscol, const MatFactorInfo *info) 2236 { 2237 namespace petsc = boost::distributed::petsc; 2238 2239 namespace graph_dist = boost::graph::distributed; 2240 using boost::graph::distributed::ilu_default::process_group_type; 2241 using boost::graph::ilu_permuted; 2242 2243 PetscBool row_identity, col_identity; 2244 PetscContainer c; 2245 PetscInt m, n, M, N; 2246 PetscErrorCode ierr; 2247 2248 PetscFunctionBegin; 2249 if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels = 0 supported for parallel ilu"); 2250 ierr = ISIdentity(isrow, &row_identity);CHKERRQ(ierr); 2251 ierr = ISIdentity(iscol, &col_identity);CHKERRQ(ierr); 2252 if (!row_identity || !col_identity) { 2253 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Row and column permutations must be identity for parallel ILU"); 2254 } 2255 2256 process_group_type pg; 2257 typedef graph_dist::ilu_default::ilu_level_graph_type lgraph_type; 2258 lgraph_type* lgraph_p = new lgraph_type(petsc::num_global_vertices(A), pg, petsc::matrix_distribution(A, pg)); 2259 lgraph_type& level_graph = *lgraph_p; 2260 graph_dist::ilu_default::graph_type& graph(level_graph.graph); 2261 2262 petsc::read_matrix(A, graph, get(boost::edge_weight, graph)); 2263 ilu_permuted(level_graph); 2264 2265 /* put together the new matrix */ 2266 ierr = MatCreate(((PetscObject)A)->comm, fact);CHKERRQ(ierr); 2267 ierr = MatGetLocalSize(A, &m, &n);CHKERRQ(ierr); 2268 ierr = MatGetSize(A, &M, &N);CHKERRQ(ierr); 2269 ierr = MatSetSizes(fact, m, n, M, N);CHKERRQ(ierr); 2270 ierr = MatSetType(fact, ((PetscObject)A)->type_name);CHKERRQ(ierr); 2271 ierr = MatAssemblyBegin(fact, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2272 ierr = MatAssemblyEnd(fact, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2273 2274 ierr = PetscContainerCreate(((PetscObject)A)->comm, &c); 2275 ierr = PetscContainerSetPointer(c, lgraph_p); 2276 ierr = PetscObjectCompose((PetscObject) (fact), "graph", (PetscObject) c); 2277 PetscFunctionReturn(0); 2278 } 2279 2280 #undef __FUNCT__ 2281 #define __FUNCT__ "MatLUFactorNumeric_MPIAIJ" 2282 PetscErrorCode MatLUFactorNumeric_MPIAIJ(Mat B,Mat A, const MatFactorInfo *info) 2283 { 2284 PetscFunctionBegin; 2285 PetscFunctionReturn(0); 2286 } 2287 2288 #undef __FUNCT__ 2289 #define __FUNCT__ "MatSolve_MPIAIJ" 2290 /* 2291 This uses the parallel ILU factorization of Peter Gottschling <pgottsch@osl.iu.edu> 2292 */ 2293 PetscErrorCode MatSolve_MPIAIJ(Mat A, Vec b, Vec x) 2294 { 2295 namespace graph_dist = boost::graph::distributed; 2296 2297 typedef graph_dist::ilu_default::ilu_level_graph_type lgraph_type; 2298 lgraph_type* lgraph_p; 2299 PetscContainer c; 2300 PetscErrorCode ierr; 2301 2302 PetscFunctionBegin; 2303 ierr = PetscObjectQuery((PetscObject) A, "graph", (PetscObject *) &c);CHKERRQ(ierr); 2304 ierr = PetscContainerGetPointer(c, (void **) &lgraph_p);CHKERRQ(ierr); 2305 ierr = VecCopy(b, x);CHKERRQ(ierr); 2306 2307 PetscScalar* array_x; 2308 ierr = VecGetArray(x, &array_x);CHKERRQ(ierr); 2309 PetscInt sx; 2310 ierr = VecGetSize(x, &sx);CHKERRQ(ierr); 2311 2312 PetscScalar* array_b; 2313 ierr = VecGetArray(b, &array_b);CHKERRQ(ierr); 2314 PetscInt sb; 2315 ierr = VecGetSize(b, &sb);CHKERRQ(ierr); 2316 2317 lgraph_type& level_graph = *lgraph_p; 2318 graph_dist::ilu_default::graph_type& graph(level_graph.graph); 2319 2320 typedef boost::multi_array_ref<PetscScalar, 1> array_ref_type; 2321 array_ref_type ref_b(array_b, boost::extents[num_vertices(graph)]), 2322 ref_x(array_x, boost::extents[num_vertices(graph)]); 2323 2324 typedef boost::iterator_property_map<array_ref_type::iterator, 2325 boost::property_map<graph_dist::ilu_default::graph_type, boost::vertex_index_t>::type> gvector_type; 2326 gvector_type vector_b(ref_b.begin(), get(boost::vertex_index, graph)), 2327 vector_x(ref_x.begin(), get(boost::vertex_index, graph)); 2328 2329 ilu_set_solve(*lgraph_p, vector_b, vector_x); 2330 2331 PetscFunctionReturn(0); 2332 } 2333 #endif 2334 2335 typedef struct { /* used by MatGetRedundantMatrix() for reusing matredundant */ 2336 PetscInt nzlocal,nsends,nrecvs; 2337 PetscMPIInt *send_rank,*recv_rank; 2338 PetscInt *sbuf_nz,*rbuf_nz,*sbuf_j,**rbuf_j; 2339 PetscScalar *sbuf_a,**rbuf_a; 2340 PetscErrorCode (*MatDestroy)(Mat); 2341 } Mat_Redundant; 2342 2343 #undef __FUNCT__ 2344 #define __FUNCT__ "PetscContainerDestroy_MatRedundant" 2345 PetscErrorCode PetscContainerDestroy_MatRedundant(void *ptr) 2346 { 2347 PetscErrorCode ierr; 2348 Mat_Redundant *redund=(Mat_Redundant*)ptr; 2349 PetscInt i; 2350 2351 PetscFunctionBegin; 2352 ierr = PetscFree2(redund->send_rank,redund->recv_rank);CHKERRQ(ierr); 2353 ierr = PetscFree(redund->sbuf_j);CHKERRQ(ierr); 2354 ierr = PetscFree(redund->sbuf_a);CHKERRQ(ierr); 2355 for (i=0; i<redund->nrecvs; i++){ 2356 ierr = PetscFree(redund->rbuf_j[i]);CHKERRQ(ierr); 2357 ierr = PetscFree(redund->rbuf_a[i]);CHKERRQ(ierr); 2358 } 2359 ierr = PetscFree4(redund->sbuf_nz,redund->rbuf_nz,redund->rbuf_j,redund->rbuf_a);CHKERRQ(ierr); 2360 ierr = PetscFree(redund);CHKERRQ(ierr); 2361 PetscFunctionReturn(0); 2362 } 2363 2364 #undef __FUNCT__ 2365 #define __FUNCT__ "MatDestroy_MatRedundant" 2366 PetscErrorCode MatDestroy_MatRedundant(Mat A) 2367 { 2368 PetscErrorCode ierr; 2369 PetscContainer container; 2370 Mat_Redundant *redund=PETSC_NULL; 2371 2372 PetscFunctionBegin; 2373 ierr = PetscObjectQuery((PetscObject)A,"Mat_Redundant",(PetscObject *)&container);CHKERRQ(ierr); 2374 if (container) { 2375 ierr = PetscContainerGetPointer(container,(void **)&redund);CHKERRQ(ierr); 2376 } else { 2377 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 2378 } 2379 A->ops->destroy = redund->MatDestroy; 2380 ierr = PetscObjectCompose((PetscObject)A,"Mat_Redundant",0);CHKERRQ(ierr); 2381 ierr = (*A->ops->destroy)(A);CHKERRQ(ierr); 2382 ierr = PetscContainerDestroy(container);CHKERRQ(ierr); 2383 PetscFunctionReturn(0); 2384 } 2385 2386 #undef __FUNCT__ 2387 #define __FUNCT__ "MatGetRedundantMatrix_MPIAIJ" 2388 PetscErrorCode MatGetRedundantMatrix_MPIAIJ(Mat mat,PetscInt nsubcomm,MPI_Comm subcomm,PetscInt mlocal_sub,MatReuse reuse,Mat *matredundant) 2389 { 2390 PetscMPIInt rank,size; 2391 MPI_Comm comm=((PetscObject)mat)->comm; 2392 PetscErrorCode ierr; 2393 PetscInt nsends=0,nrecvs=0,i,rownz_max=0; 2394 PetscMPIInt *send_rank=PETSC_NULL,*recv_rank=PETSC_NULL; 2395 PetscInt *rowrange=mat->rmap->range; 2396 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 2397 Mat A=aij->A,B=aij->B,C=*matredundant; 2398 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data; 2399 PetscScalar *sbuf_a; 2400 PetscInt nzlocal=a->nz+b->nz; 2401 PetscInt j,cstart=mat->cmap->rstart,cend=mat->cmap->rend,row,nzA,nzB,ncols,*cworkA,*cworkB; 2402 PetscInt rstart=mat->rmap->rstart,rend=mat->rmap->rend,*bmap=aij->garray,M,N; 2403 PetscInt *cols,ctmp,lwrite,*rptr,l,*sbuf_j; 2404 MatScalar *aworkA,*aworkB; 2405 PetscScalar *vals; 2406 PetscMPIInt tag1,tag2,tag3,imdex; 2407 MPI_Request *s_waits1=PETSC_NULL,*s_waits2=PETSC_NULL,*s_waits3=PETSC_NULL, 2408 *r_waits1=PETSC_NULL,*r_waits2=PETSC_NULL,*r_waits3=PETSC_NULL; 2409 MPI_Status recv_status,*send_status; 2410 PetscInt *sbuf_nz=PETSC_NULL,*rbuf_nz=PETSC_NULL,count; 2411 PetscInt **rbuf_j=PETSC_NULL; 2412 PetscScalar **rbuf_a=PETSC_NULL; 2413 Mat_Redundant *redund=PETSC_NULL; 2414 PetscContainer container; 2415 2416 PetscFunctionBegin; 2417 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 2418 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 2419 2420 if (reuse == MAT_REUSE_MATRIX) { 2421 ierr = MatGetSize(C,&M,&N);CHKERRQ(ierr); 2422 if (M != N || M != mat->rmap->N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong global size"); 2423 ierr = MatGetLocalSize(C,&M,&N);CHKERRQ(ierr); 2424 if (M != N || M != mlocal_sub) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong local size"); 2425 ierr = PetscObjectQuery((PetscObject)C,"Mat_Redundant",(PetscObject *)&container);CHKERRQ(ierr); 2426 if (container) { 2427 ierr = PetscContainerGetPointer(container,(void **)&redund);CHKERRQ(ierr); 2428 } else { 2429 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Container does not exit"); 2430 } 2431 if (nzlocal != redund->nzlocal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong nzlocal"); 2432 2433 nsends = redund->nsends; 2434 nrecvs = redund->nrecvs; 2435 send_rank = redund->send_rank; 2436 recv_rank = redund->recv_rank; 2437 sbuf_nz = redund->sbuf_nz; 2438 rbuf_nz = redund->rbuf_nz; 2439 sbuf_j = redund->sbuf_j; 2440 sbuf_a = redund->sbuf_a; 2441 rbuf_j = redund->rbuf_j; 2442 rbuf_a = redund->rbuf_a; 2443 } 2444 2445 if (reuse == MAT_INITIAL_MATRIX){ 2446 PetscMPIInt subrank,subsize; 2447 PetscInt nleftover,np_subcomm; 2448 /* get the destination processors' id send_rank, nsends and nrecvs */ 2449 ierr = MPI_Comm_rank(subcomm,&subrank);CHKERRQ(ierr); 2450 ierr = MPI_Comm_size(subcomm,&subsize);CHKERRQ(ierr); 2451 ierr = PetscMalloc2(size,PetscMPIInt,&send_rank,size,PetscMPIInt,&recv_rank); 2452 np_subcomm = size/nsubcomm; 2453 nleftover = size - nsubcomm*np_subcomm; 2454 nsends = 0; nrecvs = 0; 2455 for (i=0; i<size; i++){ /* i=rank*/ 2456 if (subrank == i/nsubcomm && rank != i){ /* my_subrank == other's subrank */ 2457 send_rank[nsends] = i; nsends++; 2458 recv_rank[nrecvs++] = i; 2459 } 2460 } 2461 if (rank >= size - nleftover){/* this proc is a leftover processor */ 2462 i = size-nleftover-1; 2463 j = 0; 2464 while (j < nsubcomm - nleftover){ 2465 send_rank[nsends++] = i; 2466 i--; j++; 2467 } 2468 } 2469 2470 if (nleftover && subsize == size/nsubcomm && subrank==subsize-1){ /* this proc recvs from leftover processors */ 2471 for (i=0; i<nleftover; i++){ 2472 recv_rank[nrecvs++] = size-nleftover+i; 2473 } 2474 } 2475 2476 /* allocate sbuf_j, sbuf_a */ 2477 i = nzlocal + rowrange[rank+1] - rowrange[rank] + 2; 2478 ierr = PetscMalloc(i*sizeof(PetscInt),&sbuf_j);CHKERRQ(ierr); 2479 ierr = PetscMalloc((nzlocal+1)*sizeof(PetscScalar),&sbuf_a);CHKERRQ(ierr); 2480 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 2481 2482 /* copy mat's local entries into the buffers */ 2483 if (reuse == MAT_INITIAL_MATRIX){ 2484 rownz_max = 0; 2485 rptr = sbuf_j; 2486 cols = sbuf_j + rend-rstart + 1; 2487 vals = sbuf_a; 2488 rptr[0] = 0; 2489 for (i=0; i<rend-rstart; i++){ 2490 row = i + rstart; 2491 nzA = a->i[i+1] - a->i[i]; nzB = b->i[i+1] - b->i[i]; 2492 ncols = nzA + nzB; 2493 cworkA = a->j + a->i[i]; cworkB = b->j + b->i[i]; 2494 aworkA = a->a + a->i[i]; aworkB = b->a + b->i[i]; 2495 /* load the column indices for this row into cols */ 2496 lwrite = 0; 2497 for (l=0; l<nzB; l++) { 2498 if ((ctmp = bmap[cworkB[l]]) < cstart){ 2499 vals[lwrite] = aworkB[l]; 2500 cols[lwrite++] = ctmp; 2501 } 2502 } 2503 for (l=0; l<nzA; l++){ 2504 vals[lwrite] = aworkA[l]; 2505 cols[lwrite++] = cstart + cworkA[l]; 2506 } 2507 for (l=0; l<nzB; l++) { 2508 if ((ctmp = bmap[cworkB[l]]) >= cend){ 2509 vals[lwrite] = aworkB[l]; 2510 cols[lwrite++] = ctmp; 2511 } 2512 } 2513 vals += ncols; 2514 cols += ncols; 2515 rptr[i+1] = rptr[i] + ncols; 2516 if (rownz_max < ncols) rownz_max = ncols; 2517 } 2518 if (rptr[rend-rstart] != a->nz + b->nz) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_PLIB, "rptr[%d] %d != %d + %d",rend-rstart,rptr[rend-rstart+1],a->nz,b->nz); 2519 } else { /* only copy matrix values into sbuf_a */ 2520 rptr = sbuf_j; 2521 vals = sbuf_a; 2522 rptr[0] = 0; 2523 for (i=0; i<rend-rstart; i++){ 2524 row = i + rstart; 2525 nzA = a->i[i+1] - a->i[i]; nzB = b->i[i+1] - b->i[i]; 2526 ncols = nzA + nzB; 2527 cworkA = a->j + a->i[i]; cworkB = b->j + b->i[i]; 2528 aworkA = a->a + a->i[i]; aworkB = b->a + b->i[i]; 2529 lwrite = 0; 2530 for (l=0; l<nzB; l++) { 2531 if ((ctmp = bmap[cworkB[l]]) < cstart) vals[lwrite++] = aworkB[l]; 2532 } 2533 for (l=0; l<nzA; l++) vals[lwrite++] = aworkA[l]; 2534 for (l=0; l<nzB; l++) { 2535 if ((ctmp = bmap[cworkB[l]]) >= cend) vals[lwrite++] = aworkB[l]; 2536 } 2537 vals += ncols; 2538 rptr[i+1] = rptr[i] + ncols; 2539 } 2540 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 2541 2542 /* send nzlocal to others, and recv other's nzlocal */ 2543 /*--------------------------------------------------*/ 2544 if (reuse == MAT_INITIAL_MATRIX){ 2545 ierr = PetscMalloc2(3*(nsends + nrecvs)+1,MPI_Request,&s_waits3,nsends+1,MPI_Status,&send_status);CHKERRQ(ierr); 2546 s_waits2 = s_waits3 + nsends; 2547 s_waits1 = s_waits2 + nsends; 2548 r_waits1 = s_waits1 + nsends; 2549 r_waits2 = r_waits1 + nrecvs; 2550 r_waits3 = r_waits2 + nrecvs; 2551 } else { 2552 ierr = PetscMalloc2(nsends + nrecvs +1,MPI_Request,&s_waits3,nsends+1,MPI_Status,&send_status);CHKERRQ(ierr); 2553 r_waits3 = s_waits3 + nsends; 2554 } 2555 2556 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag3);CHKERRQ(ierr); 2557 if (reuse == MAT_INITIAL_MATRIX){ 2558 /* get new tags to keep the communication clean */ 2559 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag1);CHKERRQ(ierr); 2560 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag2);CHKERRQ(ierr); 2561 ierr = PetscMalloc4(nsends,PetscInt,&sbuf_nz,nrecvs,PetscInt,&rbuf_nz,nrecvs,PetscInt*,&rbuf_j,nrecvs,PetscScalar*,&rbuf_a);CHKERRQ(ierr); 2562 2563 /* post receives of other's nzlocal */ 2564 for (i=0; i<nrecvs; i++){ 2565 ierr = MPI_Irecv(rbuf_nz+i,1,MPIU_INT,MPI_ANY_SOURCE,tag1,comm,r_waits1+i);CHKERRQ(ierr); 2566 } 2567 /* send nzlocal to others */ 2568 for (i=0; i<nsends; i++){ 2569 sbuf_nz[i] = nzlocal; 2570 ierr = MPI_Isend(sbuf_nz+i,1,MPIU_INT,send_rank[i],tag1,comm,s_waits1+i);CHKERRQ(ierr); 2571 } 2572 /* wait on receives of nzlocal; allocate space for rbuf_j, rbuf_a */ 2573 count = nrecvs; 2574 while (count) { 2575 ierr = MPI_Waitany(nrecvs,r_waits1,&imdex,&recv_status);CHKERRQ(ierr); 2576 recv_rank[imdex] = recv_status.MPI_SOURCE; 2577 /* allocate rbuf_a and rbuf_j; then post receives of rbuf_j */ 2578 ierr = PetscMalloc((rbuf_nz[imdex]+1)*sizeof(PetscScalar),&rbuf_a[imdex]);CHKERRQ(ierr); 2579 2580 i = rowrange[recv_status.MPI_SOURCE+1] - rowrange[recv_status.MPI_SOURCE]; /* number of expected mat->i */ 2581 rbuf_nz[imdex] += i + 2; 2582 ierr = PetscMalloc(rbuf_nz[imdex]*sizeof(PetscInt),&rbuf_j[imdex]);CHKERRQ(ierr); 2583 ierr = MPI_Irecv(rbuf_j[imdex],rbuf_nz[imdex],MPIU_INT,recv_status.MPI_SOURCE,tag2,comm,r_waits2+imdex);CHKERRQ(ierr); 2584 count--; 2585 } 2586 /* wait on sends of nzlocal */ 2587 if (nsends) {ierr = MPI_Waitall(nsends,s_waits1,send_status);CHKERRQ(ierr);} 2588 /* send mat->i,j to others, and recv from other's */ 2589 /*------------------------------------------------*/ 2590 for (i=0; i<nsends; i++){ 2591 j = nzlocal + rowrange[rank+1] - rowrange[rank] + 1; 2592 ierr = MPI_Isend(sbuf_j,j,MPIU_INT,send_rank[i],tag2,comm,s_waits2+i);CHKERRQ(ierr); 2593 } 2594 /* wait on receives of mat->i,j */ 2595 /*------------------------------*/ 2596 count = nrecvs; 2597 while (count) { 2598 ierr = MPI_Waitany(nrecvs,r_waits2,&imdex,&recv_status);CHKERRQ(ierr); 2599 if (recv_rank[imdex] != recv_status.MPI_SOURCE) SETERRQ2(PETSC_COMM_SELF,1, "recv_rank %d != MPI_SOURCE %d",recv_rank[imdex],recv_status.MPI_SOURCE); 2600 count--; 2601 } 2602 /* wait on sends of mat->i,j */ 2603 /*---------------------------*/ 2604 if (nsends) { 2605 ierr = MPI_Waitall(nsends,s_waits2,send_status);CHKERRQ(ierr); 2606 } 2607 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 2608 2609 /* post receives, send and receive mat->a */ 2610 /*----------------------------------------*/ 2611 for (imdex=0; imdex<nrecvs; imdex++) { 2612 ierr = MPI_Irecv(rbuf_a[imdex],rbuf_nz[imdex],MPIU_SCALAR,recv_rank[imdex],tag3,comm,r_waits3+imdex);CHKERRQ(ierr); 2613 } 2614 for (i=0; i<nsends; i++){ 2615 ierr = MPI_Isend(sbuf_a,nzlocal,MPIU_SCALAR,send_rank[i],tag3,comm,s_waits3+i);CHKERRQ(ierr); 2616 } 2617 count = nrecvs; 2618 while (count) { 2619 ierr = MPI_Waitany(nrecvs,r_waits3,&imdex,&recv_status);CHKERRQ(ierr); 2620 if (recv_rank[imdex] != recv_status.MPI_SOURCE) SETERRQ2(PETSC_COMM_SELF,1, "recv_rank %d != MPI_SOURCE %d",recv_rank[imdex],recv_status.MPI_SOURCE); 2621 count--; 2622 } 2623 if (nsends) { 2624 ierr = MPI_Waitall(nsends,s_waits3,send_status);CHKERRQ(ierr); 2625 } 2626 2627 ierr = PetscFree2(s_waits3,send_status);CHKERRQ(ierr); 2628 2629 /* create redundant matrix */ 2630 /*-------------------------*/ 2631 if (reuse == MAT_INITIAL_MATRIX){ 2632 /* compute rownz_max for preallocation */ 2633 for (imdex=0; imdex<nrecvs; imdex++){ 2634 j = rowrange[recv_rank[imdex]+1] - rowrange[recv_rank[imdex]]; 2635 rptr = rbuf_j[imdex]; 2636 for (i=0; i<j; i++){ 2637 ncols = rptr[i+1] - rptr[i]; 2638 if (rownz_max < ncols) rownz_max = ncols; 2639 } 2640 } 2641 2642 ierr = MatCreate(subcomm,&C);CHKERRQ(ierr); 2643 ierr = MatSetSizes(C,mlocal_sub,mlocal_sub,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr); 2644 ierr = MatSetFromOptions(C);CHKERRQ(ierr); 2645 ierr = MatSeqAIJSetPreallocation(C,rownz_max,PETSC_NULL);CHKERRQ(ierr); 2646 ierr = MatMPIAIJSetPreallocation(C,rownz_max,PETSC_NULL,rownz_max,PETSC_NULL);CHKERRQ(ierr); 2647 } else { 2648 C = *matredundant; 2649 } 2650 2651 /* insert local matrix entries */ 2652 rptr = sbuf_j; 2653 cols = sbuf_j + rend-rstart + 1; 2654 vals = sbuf_a; 2655 for (i=0; i<rend-rstart; i++){ 2656 row = i + rstart; 2657 ncols = rptr[i+1] - rptr[i]; 2658 ierr = MatSetValues(C,1,&row,ncols,cols,vals,INSERT_VALUES);CHKERRQ(ierr); 2659 vals += ncols; 2660 cols += ncols; 2661 } 2662 /* insert received matrix entries */ 2663 for (imdex=0; imdex<nrecvs; imdex++){ 2664 rstart = rowrange[recv_rank[imdex]]; 2665 rend = rowrange[recv_rank[imdex]+1]; 2666 rptr = rbuf_j[imdex]; 2667 cols = rbuf_j[imdex] + rend-rstart + 1; 2668 vals = rbuf_a[imdex]; 2669 for (i=0; i<rend-rstart; i++){ 2670 row = i + rstart; 2671 ncols = rptr[i+1] - rptr[i]; 2672 ierr = MatSetValues(C,1,&row,ncols,cols,vals,INSERT_VALUES);CHKERRQ(ierr); 2673 vals += ncols; 2674 cols += ncols; 2675 } 2676 } 2677 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2678 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2679 ierr = MatGetSize(C,&M,&N);CHKERRQ(ierr); 2680 if (M != mat->rmap->N || N != mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"redundant mat size %d != input mat size %d",M,mat->rmap->N); 2681 if (reuse == MAT_INITIAL_MATRIX){ 2682 PetscContainer container; 2683 *matredundant = C; 2684 /* create a supporting struct and attach it to C for reuse */ 2685 ierr = PetscNewLog(C,Mat_Redundant,&redund);CHKERRQ(ierr); 2686 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 2687 ierr = PetscContainerSetPointer(container,redund);CHKERRQ(ierr); 2688 ierr = PetscObjectCompose((PetscObject)C,"Mat_Redundant",(PetscObject)container);CHKERRQ(ierr); 2689 ierr = PetscContainerSetUserDestroy(container,PetscContainerDestroy_MatRedundant);CHKERRQ(ierr); 2690 2691 redund->nzlocal = nzlocal; 2692 redund->nsends = nsends; 2693 redund->nrecvs = nrecvs; 2694 redund->send_rank = send_rank; 2695 redund->recv_rank = recv_rank; 2696 redund->sbuf_nz = sbuf_nz; 2697 redund->rbuf_nz = rbuf_nz; 2698 redund->sbuf_j = sbuf_j; 2699 redund->sbuf_a = sbuf_a; 2700 redund->rbuf_j = rbuf_j; 2701 redund->rbuf_a = rbuf_a; 2702 2703 redund->MatDestroy = C->ops->destroy; 2704 C->ops->destroy = MatDestroy_MatRedundant; 2705 } 2706 PetscFunctionReturn(0); 2707 } 2708 2709 #undef __FUNCT__ 2710 #define __FUNCT__ "MatGetRowMaxAbs_MPIAIJ" 2711 PetscErrorCode MatGetRowMaxAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2712 { 2713 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2714 PetscErrorCode ierr; 2715 PetscInt i,*idxb = 0; 2716 PetscScalar *va,*vb; 2717 Vec vtmp; 2718 2719 PetscFunctionBegin; 2720 ierr = MatGetRowMaxAbs(a->A,v,idx);CHKERRQ(ierr); 2721 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2722 if (idx) { 2723 for (i=0; i<A->rmap->n; i++) { 2724 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2725 } 2726 } 2727 2728 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2729 if (idx) { 2730 ierr = PetscMalloc(A->rmap->n*sizeof(PetscInt),&idxb);CHKERRQ(ierr); 2731 } 2732 ierr = MatGetRowMaxAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2733 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2734 2735 for (i=0; i<A->rmap->n; i++){ 2736 if (PetscAbsScalar(va[i]) < PetscAbsScalar(vb[i])) { 2737 va[i] = vb[i]; 2738 if (idx) idx[i] = a->garray[idxb[i]]; 2739 } 2740 } 2741 2742 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2743 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2744 ierr = PetscFree(idxb);CHKERRQ(ierr); 2745 ierr = VecDestroy(vtmp);CHKERRQ(ierr); 2746 PetscFunctionReturn(0); 2747 } 2748 2749 #undef __FUNCT__ 2750 #define __FUNCT__ "MatGetRowMinAbs_MPIAIJ" 2751 PetscErrorCode MatGetRowMinAbs_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2752 { 2753 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 2754 PetscErrorCode ierr; 2755 PetscInt i,*idxb = 0; 2756 PetscScalar *va,*vb; 2757 Vec vtmp; 2758 2759 PetscFunctionBegin; 2760 ierr = MatGetRowMinAbs(a->A,v,idx);CHKERRQ(ierr); 2761 ierr = VecGetArray(v,&va);CHKERRQ(ierr); 2762 if (idx) { 2763 for (i=0; i<A->cmap->n; i++) { 2764 if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart; 2765 } 2766 } 2767 2768 ierr = VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);CHKERRQ(ierr); 2769 if (idx) { 2770 ierr = PetscMalloc(A->rmap->n*sizeof(PetscInt),&idxb);CHKERRQ(ierr); 2771 } 2772 ierr = MatGetRowMinAbs(a->B,vtmp,idxb);CHKERRQ(ierr); 2773 ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr); 2774 2775 for (i=0; i<A->rmap->n; i++){ 2776 if (PetscAbsScalar(va[i]) > PetscAbsScalar(vb[i])) { 2777 va[i] = vb[i]; 2778 if (idx) idx[i] = a->garray[idxb[i]]; 2779 } 2780 } 2781 2782 ierr = VecRestoreArray(v,&va);CHKERRQ(ierr); 2783 ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr); 2784 ierr = PetscFree(idxb);CHKERRQ(ierr); 2785 ierr = VecDestroy(vtmp);CHKERRQ(ierr); 2786 PetscFunctionReturn(0); 2787 } 2788 2789 #undef __FUNCT__ 2790 #define __FUNCT__ "MatGetRowMin_MPIAIJ" 2791 PetscErrorCode MatGetRowMin_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2792 { 2793 Mat_MPIAIJ *mat = (Mat_MPIAIJ *) A->data; 2794 PetscInt n = A->rmap->n; 2795 PetscInt cstart = A->cmap->rstart; 2796 PetscInt *cmap = mat->garray; 2797 PetscInt *diagIdx, *offdiagIdx; 2798 Vec diagV, offdiagV; 2799 PetscScalar *a, *diagA, *offdiagA; 2800 PetscInt r; 2801 PetscErrorCode ierr; 2802 2803 PetscFunctionBegin; 2804 ierr = PetscMalloc2(n,PetscInt,&diagIdx,n,PetscInt,&offdiagIdx);CHKERRQ(ierr); 2805 ierr = VecCreateSeq(((PetscObject)A)->comm, n, &diagV);CHKERRQ(ierr); 2806 ierr = VecCreateSeq(((PetscObject)A)->comm, n, &offdiagV);CHKERRQ(ierr); 2807 ierr = MatGetRowMin(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2808 ierr = MatGetRowMin(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2809 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2810 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2811 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2812 for(r = 0; r < n; ++r) { 2813 if (PetscAbsScalar(diagA[r]) <= PetscAbsScalar(offdiagA[r])) { 2814 a[r] = diagA[r]; 2815 idx[r] = cstart + diagIdx[r]; 2816 } else { 2817 a[r] = offdiagA[r]; 2818 idx[r] = cmap[offdiagIdx[r]]; 2819 } 2820 } 2821 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2822 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2823 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2824 ierr = VecDestroy(diagV);CHKERRQ(ierr); 2825 ierr = VecDestroy(offdiagV);CHKERRQ(ierr); 2826 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2827 PetscFunctionReturn(0); 2828 } 2829 2830 #undef __FUNCT__ 2831 #define __FUNCT__ "MatGetRowMax_MPIAIJ" 2832 PetscErrorCode MatGetRowMax_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2833 { 2834 Mat_MPIAIJ *mat = (Mat_MPIAIJ *) A->data; 2835 PetscInt n = A->rmap->n; 2836 PetscInt cstart = A->cmap->rstart; 2837 PetscInt *cmap = mat->garray; 2838 PetscInt *diagIdx, *offdiagIdx; 2839 Vec diagV, offdiagV; 2840 PetscScalar *a, *diagA, *offdiagA; 2841 PetscInt r; 2842 PetscErrorCode ierr; 2843 2844 PetscFunctionBegin; 2845 ierr = PetscMalloc2(n,PetscInt,&diagIdx,n,PetscInt,&offdiagIdx);CHKERRQ(ierr); 2846 ierr = VecCreateSeq(((PetscObject)A)->comm, n, &diagV);CHKERRQ(ierr); 2847 ierr = VecCreateSeq(((PetscObject)A)->comm, n, &offdiagV);CHKERRQ(ierr); 2848 ierr = MatGetRowMax(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2849 ierr = MatGetRowMax(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2850 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2851 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2852 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2853 for(r = 0; r < n; ++r) { 2854 if (PetscAbsScalar(diagA[r]) >= PetscAbsScalar(offdiagA[r])) { 2855 a[r] = diagA[r]; 2856 idx[r] = cstart + diagIdx[r]; 2857 } else { 2858 a[r] = offdiagA[r]; 2859 idx[r] = cmap[offdiagIdx[r]]; 2860 } 2861 } 2862 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2863 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2864 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2865 ierr = VecDestroy(diagV);CHKERRQ(ierr); 2866 ierr = VecDestroy(offdiagV);CHKERRQ(ierr); 2867 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2868 PetscFunctionReturn(0); 2869 } 2870 2871 #undef __FUNCT__ 2872 #define __FUNCT__ "MatGetSeqNonzerostructure_MPIAIJ" 2873 PetscErrorCode MatGetSeqNonzerostructure_MPIAIJ(Mat mat,Mat *newmat) 2874 { 2875 PetscErrorCode ierr; 2876 Mat *dummy; 2877 2878 PetscFunctionBegin; 2879 ierr = MatGetSubMatrix_MPIAIJ_All(mat,MAT_DO_NOT_GET_VALUES,MAT_INITIAL_MATRIX,&dummy);CHKERRQ(ierr); 2880 *newmat = *dummy; 2881 ierr = PetscFree(dummy);CHKERRQ(ierr); 2882 PetscFunctionReturn(0); 2883 } 2884 2885 extern PetscErrorCode MatFDColoringApply_AIJ(Mat,MatFDColoring,Vec,MatStructure*,void*); 2886 /* -------------------------------------------------------------------*/ 2887 static struct _MatOps MatOps_Values = {MatSetValues_MPIAIJ, 2888 MatGetRow_MPIAIJ, 2889 MatRestoreRow_MPIAIJ, 2890 MatMult_MPIAIJ, 2891 /* 4*/ MatMultAdd_MPIAIJ, 2892 MatMultTranspose_MPIAIJ, 2893 MatMultTransposeAdd_MPIAIJ, 2894 #ifdef PETSC_HAVE_PBGL 2895 MatSolve_MPIAIJ, 2896 #else 2897 0, 2898 #endif 2899 0, 2900 0, 2901 /*10*/ 0, 2902 0, 2903 0, 2904 MatSOR_MPIAIJ, 2905 MatTranspose_MPIAIJ, 2906 /*15*/ MatGetInfo_MPIAIJ, 2907 MatEqual_MPIAIJ, 2908 MatGetDiagonal_MPIAIJ, 2909 MatDiagonalScale_MPIAIJ, 2910 MatNorm_MPIAIJ, 2911 /*20*/ MatAssemblyBegin_MPIAIJ, 2912 MatAssemblyEnd_MPIAIJ, 2913 MatSetOption_MPIAIJ, 2914 MatZeroEntries_MPIAIJ, 2915 /*24*/ MatZeroRows_MPIAIJ, 2916 0, 2917 #ifdef PETSC_HAVE_PBGL 2918 0, 2919 #else 2920 0, 2921 #endif 2922 0, 2923 0, 2924 /*29*/ MatSetUpPreallocation_MPIAIJ, 2925 #ifdef PETSC_HAVE_PBGL 2926 0, 2927 #else 2928 0, 2929 #endif 2930 0, 2931 0, 2932 0, 2933 /*34*/ MatDuplicate_MPIAIJ, 2934 0, 2935 0, 2936 0, 2937 0, 2938 /*39*/ MatAXPY_MPIAIJ, 2939 MatGetSubMatrices_MPIAIJ, 2940 MatIncreaseOverlap_MPIAIJ, 2941 MatGetValues_MPIAIJ, 2942 MatCopy_MPIAIJ, 2943 /*44*/ MatGetRowMax_MPIAIJ, 2944 MatScale_MPIAIJ, 2945 0, 2946 0, 2947 MatZeroRowsColumns_MPIAIJ, 2948 /*49*/ MatSetBlockSize_MPIAIJ, 2949 0, 2950 0, 2951 0, 2952 0, 2953 /*54*/ MatFDColoringCreate_MPIAIJ, 2954 0, 2955 MatSetUnfactored_MPIAIJ, 2956 MatPermute_MPIAIJ, 2957 0, 2958 /*59*/ MatGetSubMatrix_MPIAIJ, 2959 MatDestroy_MPIAIJ, 2960 MatView_MPIAIJ, 2961 0, 2962 0, 2963 /*64*/ 0, 2964 0, 2965 0, 2966 0, 2967 0, 2968 /*69*/ MatGetRowMaxAbs_MPIAIJ, 2969 MatGetRowMinAbs_MPIAIJ, 2970 0, 2971 MatSetColoring_MPIAIJ, 2972 #if defined(PETSC_HAVE_ADIC) 2973 MatSetValuesAdic_MPIAIJ, 2974 #else 2975 0, 2976 #endif 2977 MatSetValuesAdifor_MPIAIJ, 2978 /*75*/ MatFDColoringApply_AIJ, 2979 0, 2980 0, 2981 0, 2982 0, 2983 /*80*/ 0, 2984 0, 2985 0, 2986 /*83*/ MatLoad_MPIAIJ, 2987 0, 2988 0, 2989 0, 2990 0, 2991 0, 2992 /*89*/ MatMatMult_MPIAIJ_MPIAIJ, 2993 MatMatMultSymbolic_MPIAIJ_MPIAIJ, 2994 MatMatMultNumeric_MPIAIJ_MPIAIJ, 2995 MatPtAP_Basic, 2996 MatPtAPSymbolic_MPIAIJ, 2997 /*94*/ MatPtAPNumeric_MPIAIJ, 2998 0, 2999 0, 3000 0, 3001 0, 3002 /*99*/ 0, 3003 MatPtAPSymbolic_MPIAIJ_MPIAIJ, 3004 MatPtAPNumeric_MPIAIJ_MPIAIJ, 3005 MatConjugate_MPIAIJ, 3006 0, 3007 /*104*/MatSetValuesRow_MPIAIJ, 3008 MatRealPart_MPIAIJ, 3009 MatImaginaryPart_MPIAIJ, 3010 0, 3011 0, 3012 /*109*/0, 3013 MatGetRedundantMatrix_MPIAIJ, 3014 MatGetRowMin_MPIAIJ, 3015 0, 3016 0, 3017 /*114*/MatGetSeqNonzerostructure_MPIAIJ, 3018 0, 3019 0, 3020 0, 3021 0, 3022 /*119*/0, 3023 0, 3024 0, 3025 0, 3026 MatGetMultiProcBlock_MPIAIJ, 3027 /*124*/MatFindNonZeroRows_MPIAIJ, 3028 0, 3029 0, 3030 0, 3031 MatGetSubMatricesParallel_MPIAIJ 3032 }; 3033 3034 /* ----------------------------------------------------------------------------------------*/ 3035 3036 EXTERN_C_BEGIN 3037 #undef __FUNCT__ 3038 #define __FUNCT__ "MatStoreValues_MPIAIJ" 3039 PetscErrorCode MatStoreValues_MPIAIJ(Mat mat) 3040 { 3041 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 3042 PetscErrorCode ierr; 3043 3044 PetscFunctionBegin; 3045 ierr = MatStoreValues(aij->A);CHKERRQ(ierr); 3046 ierr = MatStoreValues(aij->B);CHKERRQ(ierr); 3047 PetscFunctionReturn(0); 3048 } 3049 EXTERN_C_END 3050 3051 EXTERN_C_BEGIN 3052 #undef __FUNCT__ 3053 #define __FUNCT__ "MatRetrieveValues_MPIAIJ" 3054 PetscErrorCode MatRetrieveValues_MPIAIJ(Mat mat) 3055 { 3056 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 3057 PetscErrorCode ierr; 3058 3059 PetscFunctionBegin; 3060 ierr = MatRetrieveValues(aij->A);CHKERRQ(ierr); 3061 ierr = MatRetrieveValues(aij->B);CHKERRQ(ierr); 3062 PetscFunctionReturn(0); 3063 } 3064 EXTERN_C_END 3065 3066 EXTERN_C_BEGIN 3067 #undef __FUNCT__ 3068 #define __FUNCT__ "MatMPIAIJSetPreallocation_MPIAIJ" 3069 PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJ(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 3070 { 3071 Mat_MPIAIJ *b; 3072 PetscErrorCode ierr; 3073 PetscInt i; 3074 3075 PetscFunctionBegin; 3076 if (d_nz == PETSC_DEFAULT || d_nz == PETSC_DECIDE) d_nz = 5; 3077 if (o_nz == PETSC_DEFAULT || o_nz == PETSC_DECIDE) o_nz = 2; 3078 if (d_nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"d_nz cannot be less than 0: value %D",d_nz); 3079 if (o_nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"o_nz cannot be less than 0: value %D",o_nz); 3080 3081 ierr = PetscLayoutSetBlockSize(B->rmap,1);CHKERRQ(ierr); 3082 ierr = PetscLayoutSetBlockSize(B->cmap,1);CHKERRQ(ierr); 3083 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3084 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3085 if (d_nnz) { 3086 for (i=0; i<B->rmap->n; i++) { 3087 if (d_nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"d_nnz cannot be less than 0: local row %D value %D",i,d_nnz[i]); 3088 } 3089 } 3090 if (o_nnz) { 3091 for (i=0; i<B->rmap->n; i++) { 3092 if (o_nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"o_nnz cannot be less than 0: local row %D value %D",i,o_nnz[i]); 3093 } 3094 } 3095 b = (Mat_MPIAIJ*)B->data; 3096 3097 if (!B->preallocated) { 3098 /* Explicitly create 2 MATSEQAIJ matrices. */ 3099 ierr = MatCreate(PETSC_COMM_SELF,&b->A);CHKERRQ(ierr); 3100 ierr = MatSetSizes(b->A,B->rmap->n,B->cmap->n,B->rmap->n,B->cmap->n);CHKERRQ(ierr); 3101 ierr = MatSetType(b->A,MATSEQAIJ);CHKERRQ(ierr); 3102 ierr = PetscLogObjectParent(B,b->A);CHKERRQ(ierr); 3103 ierr = MatCreate(PETSC_COMM_SELF,&b->B);CHKERRQ(ierr); 3104 ierr = MatSetSizes(b->B,B->rmap->n,B->cmap->N,B->rmap->n,B->cmap->N);CHKERRQ(ierr); 3105 ierr = MatSetType(b->B,MATSEQAIJ);CHKERRQ(ierr); 3106 ierr = PetscLogObjectParent(B,b->B);CHKERRQ(ierr); 3107 } 3108 3109 ierr = MatSeqAIJSetPreallocation(b->A,d_nz,d_nnz);CHKERRQ(ierr); 3110 ierr = MatSeqAIJSetPreallocation(b->B,o_nz,o_nnz);CHKERRQ(ierr); 3111 B->preallocated = PETSC_TRUE; 3112 PetscFunctionReturn(0); 3113 } 3114 EXTERN_C_END 3115 3116 #undef __FUNCT__ 3117 #define __FUNCT__ "MatDuplicate_MPIAIJ" 3118 PetscErrorCode MatDuplicate_MPIAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat) 3119 { 3120 Mat mat; 3121 Mat_MPIAIJ *a,*oldmat = (Mat_MPIAIJ*)matin->data; 3122 PetscErrorCode ierr; 3123 3124 PetscFunctionBegin; 3125 *newmat = 0; 3126 ierr = MatCreate(((PetscObject)matin)->comm,&mat);CHKERRQ(ierr); 3127 ierr = MatSetSizes(mat,matin->rmap->n,matin->cmap->n,matin->rmap->N,matin->cmap->N);CHKERRQ(ierr); 3128 ierr = MatSetType(mat,((PetscObject)matin)->type_name);CHKERRQ(ierr); 3129 ierr = PetscMemcpy(mat->ops,matin->ops,sizeof(struct _MatOps));CHKERRQ(ierr); 3130 a = (Mat_MPIAIJ*)mat->data; 3131 3132 mat->factortype = matin->factortype; 3133 mat->rmap->bs = matin->rmap->bs; 3134 mat->assembled = PETSC_TRUE; 3135 mat->insertmode = NOT_SET_VALUES; 3136 mat->preallocated = PETSC_TRUE; 3137 3138 a->size = oldmat->size; 3139 a->rank = oldmat->rank; 3140 a->donotstash = oldmat->donotstash; 3141 a->roworiented = oldmat->roworiented; 3142 a->rowindices = 0; 3143 a->rowvalues = 0; 3144 a->getrowactive = PETSC_FALSE; 3145 3146 ierr = PetscLayoutCopy(matin->rmap,&mat->rmap);CHKERRQ(ierr); 3147 ierr = PetscLayoutCopy(matin->cmap,&mat->cmap);CHKERRQ(ierr); 3148 3149 if (oldmat->colmap) { 3150 #if defined (PETSC_USE_CTABLE) 3151 ierr = PetscTableCreateCopy(oldmat->colmap,&a->colmap);CHKERRQ(ierr); 3152 #else 3153 ierr = PetscMalloc((mat->cmap->N)*sizeof(PetscInt),&a->colmap);CHKERRQ(ierr); 3154 ierr = PetscLogObjectMemory(mat,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 3155 ierr = PetscMemcpy(a->colmap,oldmat->colmap,(mat->cmap->N)*sizeof(PetscInt));CHKERRQ(ierr); 3156 #endif 3157 } else a->colmap = 0; 3158 if (oldmat->garray) { 3159 PetscInt len; 3160 len = oldmat->B->cmap->n; 3161 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&a->garray);CHKERRQ(ierr); 3162 ierr = PetscLogObjectMemory(mat,len*sizeof(PetscInt));CHKERRQ(ierr); 3163 if (len) { ierr = PetscMemcpy(a->garray,oldmat->garray,len*sizeof(PetscInt));CHKERRQ(ierr); } 3164 } else a->garray = 0; 3165 3166 ierr = VecDuplicate(oldmat->lvec,&a->lvec);CHKERRQ(ierr); 3167 ierr = PetscLogObjectParent(mat,a->lvec);CHKERRQ(ierr); 3168 ierr = VecScatterCopy(oldmat->Mvctx,&a->Mvctx);CHKERRQ(ierr); 3169 ierr = PetscLogObjectParent(mat,a->Mvctx);CHKERRQ(ierr); 3170 ierr = MatDuplicate(oldmat->A,cpvalues,&a->A);CHKERRQ(ierr); 3171 ierr = PetscLogObjectParent(mat,a->A);CHKERRQ(ierr); 3172 ierr = MatDuplicate(oldmat->B,cpvalues,&a->B);CHKERRQ(ierr); 3173 ierr = PetscLogObjectParent(mat,a->B);CHKERRQ(ierr); 3174 ierr = PetscFListDuplicate(((PetscObject)matin)->qlist,&((PetscObject)mat)->qlist);CHKERRQ(ierr); 3175 *newmat = mat; 3176 PetscFunctionReturn(0); 3177 } 3178 3179 /* 3180 Allows sending/receiving larger messages then 2 gigabytes in a single call 3181 */ 3182 static int MPILong_Send(void *mess,PetscInt cnt, MPI_Datatype type,int to, int tag, MPI_Comm comm) 3183 { 3184 int ierr; 3185 static PetscInt CHUNKSIZE = 250000000; /* 250,000,000 */ 3186 PetscInt i,numchunks; 3187 PetscMPIInt icnt; 3188 3189 numchunks = cnt/CHUNKSIZE + 1; 3190 for (i=0; i<numchunks; i++) { 3191 icnt = PetscMPIIntCast((i < numchunks-1) ? CHUNKSIZE : cnt - (numchunks-1)*CHUNKSIZE); 3192 ierr = MPI_Send(mess,icnt,type,to,tag,comm);if (ierr) return ierr; 3193 if (type == MPIU_INT) { 3194 mess = (void*) (((PetscInt*)mess) + CHUNKSIZE); 3195 } else if (type == MPIU_SCALAR) { 3196 mess = (void*) (((PetscScalar*)mess) + CHUNKSIZE); 3197 } else SETERRQ(comm,PETSC_ERR_SUP,"No support for this datatype"); 3198 } 3199 return 0; 3200 } 3201 static int MPILong_Recv(void *mess,PetscInt cnt, MPI_Datatype type,int from, int tag, MPI_Comm comm) 3202 { 3203 int ierr; 3204 static PetscInt CHUNKSIZE = 250000000; /* 250,000,000 */ 3205 MPI_Status status; 3206 PetscInt i,numchunks; 3207 PetscMPIInt icnt; 3208 3209 numchunks = cnt/CHUNKSIZE + 1; 3210 for (i=0; i<numchunks; i++) { 3211 icnt = PetscMPIIntCast((i < numchunks-1) ? CHUNKSIZE : cnt - (numchunks-1)*CHUNKSIZE); 3212 ierr = MPI_Recv(mess,icnt,type,from,tag,comm,&status);if (ierr) return ierr; 3213 if (type == MPIU_INT) { 3214 mess = (void*) (((PetscInt*)mess) + CHUNKSIZE); 3215 } else if (type == MPIU_SCALAR) { 3216 mess = (void*) (((PetscScalar*)mess) + CHUNKSIZE); 3217 } else SETERRQ(comm,PETSC_ERR_SUP,"No support for this datatype"); 3218 } 3219 return 0; 3220 } 3221 3222 #undef __FUNCT__ 3223 #define __FUNCT__ "MatLoad_MPIAIJ" 3224 PetscErrorCode MatLoad_MPIAIJ(Mat newMat, PetscViewer viewer) 3225 { 3226 PetscScalar *vals,*svals; 3227 MPI_Comm comm = ((PetscObject)viewer)->comm; 3228 PetscErrorCode ierr; 3229 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 3230 PetscInt i,nz,j,rstart,rend,mmax,maxnz = 0,grows,gcols; 3231 PetscInt header[4],*rowlengths = 0,M,N,m,*cols; 3232 PetscInt *ourlens = PETSC_NULL,*procsnz = PETSC_NULL,*offlens = PETSC_NULL,jj,*mycols,*smycols; 3233 PetscInt cend,cstart,n,*rowners,sizesset=1; 3234 int fd; 3235 3236 PetscFunctionBegin; 3237 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3238 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3239 if (!rank) { 3240 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 3241 ierr = PetscBinaryRead(fd,(char *)header,4,PETSC_INT);CHKERRQ(ierr); 3242 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object"); 3243 } 3244 3245 if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) sizesset = 0; 3246 3247 ierr = MPI_Bcast(header+1,3,MPIU_INT,0,comm);CHKERRQ(ierr); 3248 M = header[1]; N = header[2]; 3249 /* If global rows/cols are set to PETSC_DECIDE, set it to the sizes given in the file */ 3250 if (sizesset && newMat->rmap->N < 0) newMat->rmap->N = M; 3251 if (sizesset && newMat->cmap->N < 0) newMat->cmap->N = N; 3252 3253 /* If global sizes are set, check if they are consistent with that given in the file */ 3254 if (sizesset) { 3255 ierr = MatGetSize(newMat,&grows,&gcols);CHKERRQ(ierr); 3256 } 3257 if (sizesset && newMat->rmap->N != grows) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Inconsistent # of rows:Matrix in file has (%d) and input matrix has (%d)",M,grows); 3258 if (sizesset && newMat->cmap->N != gcols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Inconsistent # of cols:Matrix in file has (%d) and input matrix has (%d)",N,gcols); 3259 3260 /* determine ownership of all rows */ 3261 if (newMat->rmap->n < 0 ) m = M/size + ((M % size) > rank); /* PETSC_DECIDE */ 3262 else m = newMat->rmap->n; /* Set by user */ 3263 3264 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&rowners);CHKERRQ(ierr); 3265 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 3266 3267 /* First process needs enough room for process with most rows */ 3268 if (!rank) { 3269 mmax = rowners[1]; 3270 for (i=2; i<size; i++) { 3271 mmax = PetscMax(mmax,rowners[i]); 3272 } 3273 } else mmax = m; 3274 3275 rowners[0] = 0; 3276 for (i=2; i<=size; i++) { 3277 rowners[i] += rowners[i-1]; 3278 } 3279 rstart = rowners[rank]; 3280 rend = rowners[rank+1]; 3281 3282 /* distribute row lengths to all processors */ 3283 ierr = PetscMalloc2(mmax,PetscInt,&ourlens,mmax,PetscInt,&offlens);CHKERRQ(ierr); 3284 if (!rank) { 3285 ierr = PetscBinaryRead(fd,ourlens,m,PETSC_INT);CHKERRQ(ierr); 3286 ierr = PetscMalloc(m*sizeof(PetscInt),&rowlengths);CHKERRQ(ierr); 3287 ierr = PetscMalloc(size*sizeof(PetscInt),&procsnz);CHKERRQ(ierr); 3288 ierr = PetscMemzero(procsnz,size*sizeof(PetscInt));CHKERRQ(ierr); 3289 for (j=0; j<m; j++) { 3290 procsnz[0] += ourlens[j]; 3291 } 3292 for (i=1; i<size; i++) { 3293 ierr = PetscBinaryRead(fd,rowlengths,rowners[i+1]-rowners[i],PETSC_INT);CHKERRQ(ierr); 3294 /* calculate the number of nonzeros on each processor */ 3295 for (j=0; j<rowners[i+1]-rowners[i]; j++) { 3296 procsnz[i] += rowlengths[j]; 3297 } 3298 ierr = MPILong_Send(rowlengths,rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 3299 } 3300 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 3301 } else { 3302 ierr = MPILong_Recv(ourlens,m,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 3303 } 3304 3305 if (!rank) { 3306 /* determine max buffer needed and allocate it */ 3307 maxnz = 0; 3308 for (i=0; i<size; i++) { 3309 maxnz = PetscMax(maxnz,procsnz[i]); 3310 } 3311 ierr = PetscMalloc(maxnz*sizeof(PetscInt),&cols);CHKERRQ(ierr); 3312 3313 /* read in my part of the matrix column indices */ 3314 nz = procsnz[0]; 3315 ierr = PetscMalloc(nz*sizeof(PetscInt),&mycols);CHKERRQ(ierr); 3316 ierr = PetscBinaryRead(fd,mycols,nz,PETSC_INT);CHKERRQ(ierr); 3317 3318 /* read in every one elses and ship off */ 3319 for (i=1; i<size; i++) { 3320 nz = procsnz[i]; 3321 ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr); 3322 ierr = MPILong_Send(cols,nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 3323 } 3324 ierr = PetscFree(cols);CHKERRQ(ierr); 3325 } else { 3326 /* determine buffer space needed for message */ 3327 nz = 0; 3328 for (i=0; i<m; i++) { 3329 nz += ourlens[i]; 3330 } 3331 ierr = PetscMalloc(nz*sizeof(PetscInt),&mycols);CHKERRQ(ierr); 3332 3333 /* receive message of column indices*/ 3334 ierr = MPILong_Recv(mycols,nz,MPIU_INT,0,tag,comm);CHKERRQ(ierr); 3335 } 3336 3337 /* determine column ownership if matrix is not square */ 3338 if (N != M) { 3339 if (newMat->cmap->n < 0) n = N/size + ((N % size) > rank); 3340 else n = newMat->cmap->n; 3341 ierr = MPI_Scan(&n,&cend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3342 cstart = cend - n; 3343 } else { 3344 cstart = rstart; 3345 cend = rend; 3346 n = cend - cstart; 3347 } 3348 3349 /* loop over local rows, determining number of off diagonal entries */ 3350 ierr = PetscMemzero(offlens,m*sizeof(PetscInt));CHKERRQ(ierr); 3351 jj = 0; 3352 for (i=0; i<m; i++) { 3353 for (j=0; j<ourlens[i]; j++) { 3354 if (mycols[jj] < cstart || mycols[jj] >= cend) offlens[i]++; 3355 jj++; 3356 } 3357 } 3358 3359 for (i=0; i<m; i++) { 3360 ourlens[i] -= offlens[i]; 3361 } 3362 if (!sizesset) { 3363 ierr = MatSetSizes(newMat,m,n,M,N);CHKERRQ(ierr); 3364 } 3365 ierr = MatMPIAIJSetPreallocation(newMat,0,ourlens,0,offlens);CHKERRQ(ierr); 3366 3367 for (i=0; i<m; i++) { 3368 ourlens[i] += offlens[i]; 3369 } 3370 3371 if (!rank) { 3372 ierr = PetscMalloc((maxnz+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 3373 3374 /* read in my part of the matrix numerical values */ 3375 nz = procsnz[0]; 3376 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 3377 3378 /* insert into matrix */ 3379 jj = rstart; 3380 smycols = mycols; 3381 svals = vals; 3382 for (i=0; i<m; i++) { 3383 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 3384 smycols += ourlens[i]; 3385 svals += ourlens[i]; 3386 jj++; 3387 } 3388 3389 /* read in other processors and ship out */ 3390 for (i=1; i<size; i++) { 3391 nz = procsnz[i]; 3392 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 3393 ierr = MPILong_Send(vals,nz,MPIU_SCALAR,i,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 3394 } 3395 ierr = PetscFree(procsnz);CHKERRQ(ierr); 3396 } else { 3397 /* receive numeric values */ 3398 ierr = PetscMalloc((nz+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 3399 3400 /* receive message of values*/ 3401 ierr = MPILong_Recv(vals,nz,MPIU_SCALAR,0,((PetscObject)newMat)->tag,comm);CHKERRQ(ierr); 3402 3403 /* insert into matrix */ 3404 jj = rstart; 3405 smycols = mycols; 3406 svals = vals; 3407 for (i=0; i<m; i++) { 3408 ierr = MatSetValues_MPIAIJ(newMat,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 3409 smycols += ourlens[i]; 3410 svals += ourlens[i]; 3411 jj++; 3412 } 3413 } 3414 ierr = PetscFree2(ourlens,offlens);CHKERRQ(ierr); 3415 ierr = PetscFree(vals);CHKERRQ(ierr); 3416 ierr = PetscFree(mycols);CHKERRQ(ierr); 3417 ierr = PetscFree(rowners);CHKERRQ(ierr); 3418 3419 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3420 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3421 PetscFunctionReturn(0); 3422 } 3423 3424 #undef __FUNCT__ 3425 #define __FUNCT__ "MatGetSubMatrix_MPIAIJ" 3426 PetscErrorCode MatGetSubMatrix_MPIAIJ(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *newmat) 3427 { 3428 PetscErrorCode ierr; 3429 IS iscol_local; 3430 PetscInt csize; 3431 3432 PetscFunctionBegin; 3433 ierr = ISGetLocalSize(iscol,&csize);CHKERRQ(ierr); 3434 if (call == MAT_REUSE_MATRIX) { 3435 ierr = PetscObjectQuery((PetscObject)*newmat,"ISAllGather",(PetscObject*)&iscol_local);CHKERRQ(ierr); 3436 if (!iscol_local) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3437 } else { 3438 ierr = ISAllGather(iscol,&iscol_local);CHKERRQ(ierr); 3439 } 3440 ierr = MatGetSubMatrix_MPIAIJ_Private(mat,isrow,iscol_local,csize,call,newmat);CHKERRQ(ierr); 3441 if (call == MAT_INITIAL_MATRIX) { 3442 ierr = PetscObjectCompose((PetscObject)*newmat,"ISAllGather",(PetscObject)iscol_local);CHKERRQ(ierr); 3443 ierr = ISDestroy(iscol_local);CHKERRQ(ierr); 3444 } 3445 PetscFunctionReturn(0); 3446 } 3447 3448 #undef __FUNCT__ 3449 #define __FUNCT__ "MatGetSubMatrix_MPIAIJ_Private" 3450 /* 3451 Not great since it makes two copies of the submatrix, first an SeqAIJ 3452 in local and then by concatenating the local matrices the end result. 3453 Writing it directly would be much like MatGetSubMatrices_MPIAIJ() 3454 3455 Note: This requires a sequential iscol with all indices. 3456 */ 3457 PetscErrorCode MatGetSubMatrix_MPIAIJ_Private(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat) 3458 { 3459 PetscErrorCode ierr; 3460 PetscMPIInt rank,size; 3461 PetscInt i,m,n,rstart,row,rend,nz,*cwork,j; 3462 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 3463 Mat *local,M,Mreuse; 3464 MatScalar *vwork,*aa; 3465 MPI_Comm comm = ((PetscObject)mat)->comm; 3466 Mat_SeqAIJ *aij; 3467 3468 3469 PetscFunctionBegin; 3470 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3471 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3472 3473 if (call == MAT_REUSE_MATRIX) { 3474 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject *)&Mreuse);CHKERRQ(ierr); 3475 if (!Mreuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 3476 local = &Mreuse; 3477 ierr = MatGetSubMatrices(mat,1,&isrow,&iscol,MAT_REUSE_MATRIX,&local);CHKERRQ(ierr); 3478 } else { 3479 ierr = MatGetSubMatrices(mat,1,&isrow,&iscol,MAT_INITIAL_MATRIX,&local);CHKERRQ(ierr); 3480 Mreuse = *local; 3481 ierr = PetscFree(local);CHKERRQ(ierr); 3482 } 3483 3484 /* 3485 m - number of local rows 3486 n - number of columns (same on all processors) 3487 rstart - first row in new global matrix generated 3488 */ 3489 ierr = MatGetSize(Mreuse,&m,&n);CHKERRQ(ierr); 3490 if (call == MAT_INITIAL_MATRIX) { 3491 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3492 ii = aij->i; 3493 jj = aij->j; 3494 3495 /* 3496 Determine the number of non-zeros in the diagonal and off-diagonal 3497 portions of the matrix in order to do correct preallocation 3498 */ 3499 3500 /* first get start and end of "diagonal" columns */ 3501 if (csize == PETSC_DECIDE) { 3502 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 3503 if (mglobal == n) { /* square matrix */ 3504 nlocal = m; 3505 } else { 3506 nlocal = n/size + ((n % size) > rank); 3507 } 3508 } else { 3509 nlocal = csize; 3510 } 3511 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3512 rstart = rend - nlocal; 3513 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); 3514 3515 /* next, compute all the lengths */ 3516 ierr = PetscMalloc((2*m+1)*sizeof(PetscInt),&dlens);CHKERRQ(ierr); 3517 olens = dlens + m; 3518 for (i=0; i<m; i++) { 3519 jend = ii[i+1] - ii[i]; 3520 olen = 0; 3521 dlen = 0; 3522 for (j=0; j<jend; j++) { 3523 if (*jj < rstart || *jj >= rend) olen++; 3524 else dlen++; 3525 jj++; 3526 } 3527 olens[i] = olen; 3528 dlens[i] = dlen; 3529 } 3530 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 3531 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,n);CHKERRQ(ierr); 3532 ierr = MatSetType(M,((PetscObject)mat)->type_name);CHKERRQ(ierr); 3533 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 3534 ierr = PetscFree(dlens);CHKERRQ(ierr); 3535 } else { 3536 PetscInt ml,nl; 3537 3538 M = *newmat; 3539 ierr = MatGetLocalSize(M,&ml,&nl);CHKERRQ(ierr); 3540 if (ml != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 3541 ierr = MatZeroEntries(M);CHKERRQ(ierr); 3542 /* 3543 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 3544 rather than the slower MatSetValues(). 3545 */ 3546 M->was_assembled = PETSC_TRUE; 3547 M->assembled = PETSC_FALSE; 3548 } 3549 ierr = MatGetOwnershipRange(M,&rstart,&rend);CHKERRQ(ierr); 3550 aij = (Mat_SeqAIJ*)(Mreuse)->data; 3551 ii = aij->i; 3552 jj = aij->j; 3553 aa = aij->a; 3554 for (i=0; i<m; i++) { 3555 row = rstart + i; 3556 nz = ii[i+1] - ii[i]; 3557 cwork = jj; jj += nz; 3558 vwork = aa; aa += nz; 3559 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,cwork,vwork,INSERT_VALUES);CHKERRQ(ierr); 3560 } 3561 3562 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3563 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3564 *newmat = M; 3565 3566 /* save submatrix used in processor for next request */ 3567 if (call == MAT_INITIAL_MATRIX) { 3568 ierr = PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);CHKERRQ(ierr); 3569 ierr = PetscObjectDereference((PetscObject)Mreuse);CHKERRQ(ierr); 3570 } 3571 3572 PetscFunctionReturn(0); 3573 } 3574 3575 EXTERN_C_BEGIN 3576 #undef __FUNCT__ 3577 #define __FUNCT__ "MatMPIAIJSetPreallocationCSR_MPIAIJ" 3578 PetscErrorCode MatMPIAIJSetPreallocationCSR_MPIAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3579 { 3580 PetscInt m,cstart, cend,j,nnz,i,d; 3581 PetscInt *d_nnz,*o_nnz,nnz_max = 0,rstart,ii; 3582 const PetscInt *JJ; 3583 PetscScalar *values; 3584 PetscErrorCode ierr; 3585 3586 PetscFunctionBegin; 3587 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Ii[0] must be 0 it is %D",Ii[0]); 3588 3589 ierr = PetscLayoutSetBlockSize(B->rmap,1);CHKERRQ(ierr); 3590 ierr = PetscLayoutSetBlockSize(B->cmap,1);CHKERRQ(ierr); 3591 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3592 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3593 m = B->rmap->n; 3594 cstart = B->cmap->rstart; 3595 cend = B->cmap->rend; 3596 rstart = B->rmap->rstart; 3597 3598 ierr = PetscMalloc2(m,PetscInt,&d_nnz,m,PetscInt,&o_nnz);CHKERRQ(ierr); 3599 3600 #if defined(PETSC_USE_DEBUGGING) 3601 for (i=0; i<m; i++) { 3602 nnz = Ii[i+1]- Ii[i]; 3603 JJ = J + Ii[i]; 3604 if (nnz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Local row %D has a negative %D number of columns",i,nnz); 3605 if (nnz && (JJ[0] < 0)) SETERRRQ1(PETSC_ERR_ARG_WRONGSTATE,"Row %D starts with negative column index",i,j); 3606 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); 3607 } 3608 #endif 3609 3610 for (i=0; i<m; i++) { 3611 nnz = Ii[i+1]- Ii[i]; 3612 JJ = J + Ii[i]; 3613 nnz_max = PetscMax(nnz_max,nnz); 3614 d = 0; 3615 for (j=0; j<nnz; j++) { 3616 if (cstart <= JJ[j] && JJ[j] < cend) d++; 3617 } 3618 d_nnz[i] = d; 3619 o_nnz[i] = nnz - d; 3620 } 3621 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 3622 ierr = PetscFree2(d_nnz,o_nnz);CHKERRQ(ierr); 3623 3624 if (v) values = (PetscScalar*)v; 3625 else { 3626 ierr = PetscMalloc((nnz_max+1)*sizeof(PetscScalar),&values);CHKERRQ(ierr); 3627 ierr = PetscMemzero(values,nnz_max*sizeof(PetscScalar));CHKERRQ(ierr); 3628 } 3629 3630 for (i=0; i<m; i++) { 3631 ii = i + rstart; 3632 nnz = Ii[i+1]- Ii[i]; 3633 ierr = MatSetValues_MPIAIJ(B,1,&ii,nnz,J+Ii[i],values+(v ? Ii[i] : 0),INSERT_VALUES);CHKERRQ(ierr); 3634 } 3635 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3636 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3637 3638 if (!v) { 3639 ierr = PetscFree(values);CHKERRQ(ierr); 3640 } 3641 PetscFunctionReturn(0); 3642 } 3643 EXTERN_C_END 3644 3645 #undef __FUNCT__ 3646 #define __FUNCT__ "MatMPIAIJSetPreallocationCSR" 3647 /*@ 3648 MatMPIAIJSetPreallocationCSR - Allocates memory for a sparse parallel matrix in AIJ format 3649 (the default parallel PETSc format). 3650 3651 Collective on MPI_Comm 3652 3653 Input Parameters: 3654 + B - the matrix 3655 . i - the indices into j for the start of each local row (starts with zero) 3656 . j - the column indices for each local row (starts with zero) 3657 - v - optional values in the matrix 3658 3659 Level: developer 3660 3661 Notes: 3662 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3663 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3664 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3665 3666 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3667 3668 The format which is used for the sparse matrix input, is equivalent to a 3669 row-major ordering.. i.e for the following matrix, the input data expected is 3670 as shown: 3671 3672 1 0 0 3673 2 0 3 P0 3674 ------- 3675 4 5 6 P1 3676 3677 Process0 [P0]: rows_owned=[0,1] 3678 i = {0,1,3} [size = nrow+1 = 2+1] 3679 j = {0,0,2} [size = nz = 6] 3680 v = {1,2,3} [size = nz = 6] 3681 3682 Process1 [P1]: rows_owned=[2] 3683 i = {0,3} [size = nrow+1 = 1+1] 3684 j = {0,1,2} [size = nz = 6] 3685 v = {4,5,6} [size = nz = 6] 3686 3687 .keywords: matrix, aij, compressed row, sparse, parallel 3688 3689 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatCreateMPIAIJ(), MPIAIJ, 3690 MatCreateSeqAIJWithArrays(), MatCreateMPIAIJWithSplitArrays() 3691 @*/ 3692 PetscErrorCode MatMPIAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[], const PetscScalar v[]) 3693 { 3694 PetscErrorCode ierr; 3695 3696 PetscFunctionBegin; 3697 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3698 PetscFunctionReturn(0); 3699 } 3700 3701 #undef __FUNCT__ 3702 #define __FUNCT__ "MatMPIAIJSetPreallocation" 3703 /*@C 3704 MatMPIAIJSetPreallocation - Preallocates memory for a sparse parallel matrix in AIJ format 3705 (the default parallel PETSc format). For good matrix assembly performance 3706 the user should preallocate the matrix storage by setting the parameters 3707 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3708 performance can be increased by more than a factor of 50. 3709 3710 Collective on MPI_Comm 3711 3712 Input Parameters: 3713 + A - the matrix 3714 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3715 (same value is used for all local rows) 3716 . d_nnz - array containing the number of nonzeros in the various rows of the 3717 DIAGONAL portion of the local submatrix (possibly different for each row) 3718 or PETSC_NULL, if d_nz is used to specify the nonzero structure. 3719 The size of this array is equal to the number of local rows, i.e 'm'. 3720 You must leave room for the diagonal entry even if it is zero. 3721 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3722 submatrix (same value is used for all local rows). 3723 - o_nnz - array containing the number of nonzeros in the various rows of the 3724 OFF-DIAGONAL portion of the local submatrix (possibly different for 3725 each row) or PETSC_NULL, if o_nz is used to specify the nonzero 3726 structure. The size of this array is equal to the number 3727 of local rows, i.e 'm'. 3728 3729 If the *_nnz parameter is given then the *_nz parameter is ignored 3730 3731 The AIJ format (also called the Yale sparse matrix format or 3732 compressed row storage (CSR)), is fully compatible with standard Fortran 77 3733 storage. The stored row and column indices begin with zero. 3734 See the <A href="../../docs/manual.pdf#nameddest=ch_mat">Mat chapter of the users manual</A> for details. 3735 3736 The parallel matrix is partitioned such that the first m0 rows belong to 3737 process 0, the next m1 rows belong to process 1, the next m2 rows belong 3738 to process 2 etc.. where m0,m1,m2... are the input parameter 'm'. 3739 3740 The DIAGONAL portion of the local submatrix of a processor can be defined 3741 as the submatrix which is obtained by extraction the part corresponding to 3742 the rows r1-r2 and columns c1-c2 of the global matrix, where r1 is the 3743 first row that belongs to the processor, r2 is the last row belonging to 3744 the this processor, and c1-c2 is range of indices of the local part of a 3745 vector suitable for applying the matrix to. This is an mxn matrix. In the 3746 common case of a square matrix, the row and column ranges are the same and 3747 the DIAGONAL part is also square. The remaining portion of the local 3748 submatrix (mxN) constitute the OFF-DIAGONAL portion. 3749 3750 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3751 3752 You can call MatGetInfo() to get information on how effective the preallocation was; 3753 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3754 You can also run with the option -info and look for messages with the string 3755 malloc in them to see if additional memory allocation was needed. 3756 3757 Example usage: 3758 3759 Consider the following 8x8 matrix with 34 non-zero values, that is 3760 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 3761 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 3762 as follows: 3763 3764 .vb 3765 1 2 0 | 0 3 0 | 0 4 3766 Proc0 0 5 6 | 7 0 0 | 8 0 3767 9 0 10 | 11 0 0 | 12 0 3768 ------------------------------------- 3769 13 0 14 | 15 16 17 | 0 0 3770 Proc1 0 18 0 | 19 20 21 | 0 0 3771 0 0 0 | 22 23 0 | 24 0 3772 ------------------------------------- 3773 Proc2 25 26 27 | 0 0 28 | 29 0 3774 30 0 0 | 31 32 33 | 0 34 3775 .ve 3776 3777 This can be represented as a collection of submatrices as: 3778 3779 .vb 3780 A B C 3781 D E F 3782 G H I 3783 .ve 3784 3785 Where the submatrices A,B,C are owned by proc0, D,E,F are 3786 owned by proc1, G,H,I are owned by proc2. 3787 3788 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3789 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3790 The 'M','N' parameters are 8,8, and have the same values on all procs. 3791 3792 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 3793 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 3794 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 3795 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 3796 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 3797 matrix, ans [DF] as another SeqAIJ matrix. 3798 3799 When d_nz, o_nz parameters are specified, d_nz storage elements are 3800 allocated for every row of the local diagonal submatrix, and o_nz 3801 storage locations are allocated for every row of the OFF-DIAGONAL submat. 3802 One way to choose d_nz and o_nz is to use the max nonzerors per local 3803 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 3804 In this case, the values of d_nz,o_nz are: 3805 .vb 3806 proc0 : dnz = 2, o_nz = 2 3807 proc1 : dnz = 3, o_nz = 2 3808 proc2 : dnz = 1, o_nz = 4 3809 .ve 3810 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3811 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3812 for proc3. i.e we are using 12+15+10=37 storage locations to store 3813 34 values. 3814 3815 When d_nnz, o_nnz parameters are specified, the storage is specified 3816 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3817 In the above case the values for d_nnz,o_nnz are: 3818 .vb 3819 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3820 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3821 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3822 .ve 3823 Here the space allocated is sum of all the above values i.e 34, and 3824 hence pre-allocation is perfect. 3825 3826 Level: intermediate 3827 3828 .keywords: matrix, aij, compressed row, sparse, parallel 3829 3830 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateMPIAIJ(), MatMPIAIJSetPreallocationCSR(), 3831 MPIAIJ, MatGetInfo() 3832 @*/ 3833 PetscErrorCode MatMPIAIJSetPreallocation(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 3834 { 3835 PetscErrorCode ierr; 3836 3837 PetscFunctionBegin; 3838 ierr = PetscTryMethod(B,"MatMPIAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[]),(B,d_nz,d_nnz,o_nz,o_nnz));CHKERRQ(ierr); 3839 PetscFunctionReturn(0); 3840 } 3841 3842 #undef __FUNCT__ 3843 #define __FUNCT__ "MatCreateMPIAIJWithArrays" 3844 /*@ 3845 MatCreateMPIAIJWithArrays - creates a MPI AIJ matrix using arrays that contain in standard 3846 CSR format the local rows. 3847 3848 Collective on MPI_Comm 3849 3850 Input Parameters: 3851 + comm - MPI communicator 3852 . m - number of local rows (Cannot be PETSC_DECIDE) 3853 . n - This value should be the same as the local size used in creating the 3854 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3855 calculated if N is given) For square matrices n is almost always m. 3856 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3857 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3858 . i - row indices 3859 . j - column indices 3860 - a - matrix values 3861 3862 Output Parameter: 3863 . mat - the matrix 3864 3865 Level: intermediate 3866 3867 Notes: 3868 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3869 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3870 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3871 3872 The i and j indices are 0 based, and i indices are indices corresponding to the local j array. 3873 3874 The format which is used for the sparse matrix input, is equivalent to a 3875 row-major ordering.. i.e for the following matrix, the input data expected is 3876 as shown: 3877 3878 1 0 0 3879 2 0 3 P0 3880 ------- 3881 4 5 6 P1 3882 3883 Process0 [P0]: rows_owned=[0,1] 3884 i = {0,1,3} [size = nrow+1 = 2+1] 3885 j = {0,0,2} [size = nz = 6] 3886 v = {1,2,3} [size = nz = 6] 3887 3888 Process1 [P1]: rows_owned=[2] 3889 i = {0,3} [size = nrow+1 = 1+1] 3890 j = {0,1,2} [size = nz = 6] 3891 v = {4,5,6} [size = nz = 6] 3892 3893 .keywords: matrix, aij, compressed row, sparse, parallel 3894 3895 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 3896 MPIAIJ, MatCreateMPIAIJ(), MatCreateMPIAIJWithSplitArrays() 3897 @*/ 3898 PetscErrorCode MatCreateMPIAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,const PetscInt i[],const PetscInt j[],const PetscScalar a[],Mat *mat) 3899 { 3900 PetscErrorCode ierr; 3901 3902 PetscFunctionBegin; 3903 if (i[0]) { 3904 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 3905 } 3906 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 3907 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 3908 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 3909 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 3910 ierr = MatMPIAIJSetPreallocationCSR(*mat,i,j,a);CHKERRQ(ierr); 3911 PetscFunctionReturn(0); 3912 } 3913 3914 #undef __FUNCT__ 3915 #define __FUNCT__ "MatCreateMPIAIJ" 3916 /*@C 3917 MatCreateMPIAIJ - Creates a sparse parallel matrix in AIJ format 3918 (the default parallel PETSc format). For good matrix assembly performance 3919 the user should preallocate the matrix storage by setting the parameters 3920 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3921 performance can be increased by more than a factor of 50. 3922 3923 Collective on MPI_Comm 3924 3925 Input Parameters: 3926 + comm - MPI communicator 3927 . m - number of local rows (or PETSC_DECIDE to have calculated if M is given) 3928 This value should be the same as the local size used in creating the 3929 y vector for the matrix-vector product y = Ax. 3930 . n - This value should be the same as the local size used in creating the 3931 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3932 calculated if N is given) For square matrices n is almost always m. 3933 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3934 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3935 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3936 (same value is used for all local rows) 3937 . d_nnz - array containing the number of nonzeros in the various rows of the 3938 DIAGONAL portion of the local submatrix (possibly different for each row) 3939 or PETSC_NULL, if d_nz is used to specify the nonzero structure. 3940 The size of this array is equal to the number of local rows, i.e 'm'. 3941 You must leave room for the diagonal entry even if it is zero. 3942 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3943 submatrix (same value is used for all local rows). 3944 - o_nnz - array containing the number of nonzeros in the various rows of the 3945 OFF-DIAGONAL portion of the local submatrix (possibly different for 3946 each row) or PETSC_NULL, if o_nz is used to specify the nonzero 3947 structure. The size of this array is equal to the number 3948 of local rows, i.e 'm'. 3949 3950 Output Parameter: 3951 . A - the matrix 3952 3953 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3954 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3955 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3956 3957 Notes: 3958 If the *_nnz parameter is given then the *_nz parameter is ignored 3959 3960 m,n,M,N parameters specify the size of the matrix, and its partitioning across 3961 processors, while d_nz,d_nnz,o_nz,o_nnz parameters specify the approximate 3962 storage requirements for this matrix. 3963 3964 If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one 3965 processor than it must be used on all processors that share the object for 3966 that argument. 3967 3968 The user MUST specify either the local or global matrix dimensions 3969 (possibly both). 3970 3971 The parallel matrix is partitioned across processors such that the 3972 first m0 rows belong to process 0, the next m1 rows belong to 3973 process 1, the next m2 rows belong to process 2 etc.. where 3974 m0,m1,m2,.. are the input parameter 'm'. i.e each processor stores 3975 values corresponding to [m x N] submatrix. 3976 3977 The columns are logically partitioned with the n0 columns belonging 3978 to 0th partition, the next n1 columns belonging to the next 3979 partition etc.. where n0,n1,n2... are the the input parameter 'n'. 3980 3981 The DIAGONAL portion of the local submatrix on any given processor 3982 is the submatrix corresponding to the rows and columns m,n 3983 corresponding to the given processor. i.e diagonal matrix on 3984 process 0 is [m0 x n0], diagonal matrix on process 1 is [m1 x n1] 3985 etc. The remaining portion of the local submatrix [m x (N-n)] 3986 constitute the OFF-DIAGONAL portion. The example below better 3987 illustrates this concept. 3988 3989 For a square global matrix we define each processor's diagonal portion 3990 to be its local rows and the corresponding columns (a square submatrix); 3991 each processor's off-diagonal portion encompasses the remainder of the 3992 local matrix (a rectangular submatrix). 3993 3994 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3995 3996 When calling this routine with a single process communicator, a matrix of 3997 type SEQAIJ is returned. If a matrix of type MPIAIJ is desired for this 3998 type of communicator, use the construction mechanism: 3999 MatCreate(...,&A); MatSetType(A,MATMPIAIJ); MatSetSizes(A, m,n,M,N); MatMPIAIJSetPreallocation(A,...); 4000 4001 By default, this format uses inodes (identical nodes) when possible. 4002 We search for consecutive rows with the same nonzero structure, thereby 4003 reusing matrix information to achieve increased efficiency. 4004 4005 Options Database Keys: 4006 + -mat_no_inode - Do not use inodes 4007 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 4008 - -mat_aij_oneindex - Internally use indexing starting at 1 4009 rather than 0. Note that when calling MatSetValues(), 4010 the user still MUST index entries starting at 0! 4011 4012 4013 Example usage: 4014 4015 Consider the following 8x8 matrix with 34 non-zero values, that is 4016 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 4017 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 4018 as follows: 4019 4020 .vb 4021 1 2 0 | 0 3 0 | 0 4 4022 Proc0 0 5 6 | 7 0 0 | 8 0 4023 9 0 10 | 11 0 0 | 12 0 4024 ------------------------------------- 4025 13 0 14 | 15 16 17 | 0 0 4026 Proc1 0 18 0 | 19 20 21 | 0 0 4027 0 0 0 | 22 23 0 | 24 0 4028 ------------------------------------- 4029 Proc2 25 26 27 | 0 0 28 | 29 0 4030 30 0 0 | 31 32 33 | 0 34 4031 .ve 4032 4033 This can be represented as a collection of submatrices as: 4034 4035 .vb 4036 A B C 4037 D E F 4038 G H I 4039 .ve 4040 4041 Where the submatrices A,B,C are owned by proc0, D,E,F are 4042 owned by proc1, G,H,I are owned by proc2. 4043 4044 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 4045 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 4046 The 'M','N' parameters are 8,8, and have the same values on all procs. 4047 4048 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 4049 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 4050 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 4051 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 4052 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 4053 matrix, ans [DF] as another SeqAIJ matrix. 4054 4055 When d_nz, o_nz parameters are specified, d_nz storage elements are 4056 allocated for every row of the local diagonal submatrix, and o_nz 4057 storage locations are allocated for every row of the OFF-DIAGONAL submat. 4058 One way to choose d_nz and o_nz is to use the max nonzerors per local 4059 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 4060 In this case, the values of d_nz,o_nz are: 4061 .vb 4062 proc0 : dnz = 2, o_nz = 2 4063 proc1 : dnz = 3, o_nz = 2 4064 proc2 : dnz = 1, o_nz = 4 4065 .ve 4066 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 4067 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 4068 for proc3. i.e we are using 12+15+10=37 storage locations to store 4069 34 values. 4070 4071 When d_nnz, o_nnz parameters are specified, the storage is specified 4072 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 4073 In the above case the values for d_nnz,o_nnz are: 4074 .vb 4075 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 4076 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 4077 proc2: d_nnz = [1,1] and o_nnz = [4,4] 4078 .ve 4079 Here the space allocated is sum of all the above values i.e 34, and 4080 hence pre-allocation is perfect. 4081 4082 Level: intermediate 4083 4084 .keywords: matrix, aij, compressed row, sparse, parallel 4085 4086 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 4087 MPIAIJ, MatCreateMPIAIJWithArrays() 4088 @*/ 4089 PetscErrorCode MatCreateMPIAIJ(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) 4090 { 4091 PetscErrorCode ierr; 4092 PetscMPIInt size; 4093 4094 PetscFunctionBegin; 4095 ierr = MatCreate(comm,A);CHKERRQ(ierr); 4096 ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr); 4097 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4098 if (size > 1) { 4099 ierr = MatSetType(*A,MATMPIAIJ);CHKERRQ(ierr); 4100 ierr = MatMPIAIJSetPreallocation(*A,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr); 4101 } else { 4102 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 4103 ierr = MatSeqAIJSetPreallocation(*A,d_nz,d_nnz);CHKERRQ(ierr); 4104 } 4105 PetscFunctionReturn(0); 4106 } 4107 4108 #undef __FUNCT__ 4109 #define __FUNCT__ "MatMPIAIJGetSeqAIJ" 4110 PetscErrorCode MatMPIAIJGetSeqAIJ(Mat A,Mat *Ad,Mat *Ao,PetscInt *colmap[]) 4111 { 4112 Mat_MPIAIJ *a = (Mat_MPIAIJ *)A->data; 4113 4114 PetscFunctionBegin; 4115 *Ad = a->A; 4116 *Ao = a->B; 4117 *colmap = a->garray; 4118 PetscFunctionReturn(0); 4119 } 4120 4121 #undef __FUNCT__ 4122 #define __FUNCT__ "MatSetColoring_MPIAIJ" 4123 PetscErrorCode MatSetColoring_MPIAIJ(Mat A,ISColoring coloring) 4124 { 4125 PetscErrorCode ierr; 4126 PetscInt i; 4127 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 4128 4129 PetscFunctionBegin; 4130 if (coloring->ctype == IS_COLORING_GLOBAL) { 4131 ISColoringValue *allcolors,*colors; 4132 ISColoring ocoloring; 4133 4134 /* set coloring for diagonal portion */ 4135 ierr = MatSetColoring_SeqAIJ(a->A,coloring);CHKERRQ(ierr); 4136 4137 /* set coloring for off-diagonal portion */ 4138 ierr = ISAllGatherColors(((PetscObject)A)->comm,coloring->n,coloring->colors,PETSC_NULL,&allcolors);CHKERRQ(ierr); 4139 ierr = PetscMalloc((a->B->cmap->n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4140 for (i=0; i<a->B->cmap->n; i++) { 4141 colors[i] = allcolors[a->garray[i]]; 4142 } 4143 ierr = PetscFree(allcolors);CHKERRQ(ierr); 4144 ierr = ISColoringCreate(MPI_COMM_SELF,coloring->n,a->B->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4145 ierr = MatSetColoring_SeqAIJ(a->B,ocoloring);CHKERRQ(ierr); 4146 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 4147 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 4148 ISColoringValue *colors; 4149 PetscInt *larray; 4150 ISColoring ocoloring; 4151 4152 /* set coloring for diagonal portion */ 4153 ierr = PetscMalloc((a->A->cmap->n+1)*sizeof(PetscInt),&larray);CHKERRQ(ierr); 4154 for (i=0; i<a->A->cmap->n; i++) { 4155 larray[i] = i + A->cmap->rstart; 4156 } 4157 ierr = ISGlobalToLocalMappingApply(A->cmapping,IS_GTOLM_MASK,a->A->cmap->n,larray,PETSC_NULL,larray);CHKERRQ(ierr); 4158 ierr = PetscMalloc((a->A->cmap->n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4159 for (i=0; i<a->A->cmap->n; i++) { 4160 colors[i] = coloring->colors[larray[i]]; 4161 } 4162 ierr = PetscFree(larray);CHKERRQ(ierr); 4163 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,a->A->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4164 ierr = MatSetColoring_SeqAIJ(a->A,ocoloring);CHKERRQ(ierr); 4165 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 4166 4167 /* set coloring for off-diagonal portion */ 4168 ierr = PetscMalloc((a->B->cmap->n+1)*sizeof(PetscInt),&larray);CHKERRQ(ierr); 4169 ierr = ISGlobalToLocalMappingApply(A->cmapping,IS_GTOLM_MASK,a->B->cmap->n,a->garray,PETSC_NULL,larray);CHKERRQ(ierr); 4170 ierr = PetscMalloc((a->B->cmap->n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4171 for (i=0; i<a->B->cmap->n; i++) { 4172 colors[i] = coloring->colors[larray[i]]; 4173 } 4174 ierr = PetscFree(larray);CHKERRQ(ierr); 4175 ierr = ISColoringCreate(MPI_COMM_SELF,coloring->n,a->B->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4176 ierr = MatSetColoring_SeqAIJ(a->B,ocoloring);CHKERRQ(ierr); 4177 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 4178 } else { 4179 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support ISColoringType %d",(int)coloring->ctype); 4180 } 4181 4182 PetscFunctionReturn(0); 4183 } 4184 4185 #if defined(PETSC_HAVE_ADIC) 4186 #undef __FUNCT__ 4187 #define __FUNCT__ "MatSetValuesAdic_MPIAIJ" 4188 PetscErrorCode MatSetValuesAdic_MPIAIJ(Mat A,void *advalues) 4189 { 4190 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 4191 PetscErrorCode ierr; 4192 4193 PetscFunctionBegin; 4194 ierr = MatSetValuesAdic_SeqAIJ(a->A,advalues);CHKERRQ(ierr); 4195 ierr = MatSetValuesAdic_SeqAIJ(a->B,advalues);CHKERRQ(ierr); 4196 PetscFunctionReturn(0); 4197 } 4198 #endif 4199 4200 #undef __FUNCT__ 4201 #define __FUNCT__ "MatSetValuesAdifor_MPIAIJ" 4202 PetscErrorCode MatSetValuesAdifor_MPIAIJ(Mat A,PetscInt nl,void *advalues) 4203 { 4204 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 4205 PetscErrorCode ierr; 4206 4207 PetscFunctionBegin; 4208 ierr = MatSetValuesAdifor_SeqAIJ(a->A,nl,advalues);CHKERRQ(ierr); 4209 ierr = MatSetValuesAdifor_SeqAIJ(a->B,nl,advalues);CHKERRQ(ierr); 4210 PetscFunctionReturn(0); 4211 } 4212 4213 #undef __FUNCT__ 4214 #define __FUNCT__ "MatMerge" 4215 /*@ 4216 MatMerge - Creates a single large PETSc matrix by concatinating sequential 4217 matrices from each processor 4218 4219 Collective on MPI_Comm 4220 4221 Input Parameters: 4222 + comm - the communicators the parallel matrix will live on 4223 . inmat - the input sequential matrices 4224 . n - number of local columns (or PETSC_DECIDE) 4225 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4226 4227 Output Parameter: 4228 . outmat - the parallel matrix generated 4229 4230 Level: advanced 4231 4232 Notes: The number of columns of the matrix in EACH processor MUST be the same. 4233 4234 @*/ 4235 PetscErrorCode MatMerge(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat) 4236 { 4237 PetscErrorCode ierr; 4238 PetscInt m,N,i,rstart,nnz,Ii,*dnz,*onz; 4239 PetscInt *indx; 4240 PetscScalar *values; 4241 4242 PetscFunctionBegin; 4243 ierr = MatGetSize(inmat,&m,&N);CHKERRQ(ierr); 4244 if (scall == MAT_INITIAL_MATRIX){ 4245 /* count nonzeros in each row, for diagonal and off diagonal portion of matrix */ 4246 if (n == PETSC_DECIDE){ 4247 ierr = PetscSplitOwnership(comm,&n,&N);CHKERRQ(ierr); 4248 } 4249 ierr = MPI_Scan(&m, &rstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 4250 rstart -= m; 4251 4252 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 4253 for (i=0;i<m;i++) { 4254 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,PETSC_NULL);CHKERRQ(ierr); 4255 ierr = MatPreallocateSet(i+rstart,nnz,indx,dnz,onz);CHKERRQ(ierr); 4256 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,PETSC_NULL);CHKERRQ(ierr); 4257 } 4258 /* This routine will ONLY return MPIAIJ type matrix */ 4259 ierr = MatCreate(comm,outmat);CHKERRQ(ierr); 4260 ierr = MatSetSizes(*outmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 4261 ierr = MatSetType(*outmat,MATMPIAIJ);CHKERRQ(ierr); 4262 ierr = MatMPIAIJSetPreallocation(*outmat,0,dnz,0,onz);CHKERRQ(ierr); 4263 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 4264 4265 } else if (scall == MAT_REUSE_MATRIX){ 4266 ierr = MatGetOwnershipRange(*outmat,&rstart,PETSC_NULL);CHKERRQ(ierr); 4267 } else { 4268 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 4269 } 4270 4271 for (i=0;i<m;i++) { 4272 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 4273 Ii = i + rstart; 4274 ierr = MatSetValues(*outmat,1,&Ii,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 4275 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 4276 } 4277 ierr = MatDestroy(inmat);CHKERRQ(ierr); 4278 ierr = MatAssemblyBegin(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4279 ierr = MatAssemblyEnd(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4280 4281 PetscFunctionReturn(0); 4282 } 4283 4284 #undef __FUNCT__ 4285 #define __FUNCT__ "MatFileSplit" 4286 PetscErrorCode MatFileSplit(Mat A,char *outfile) 4287 { 4288 PetscErrorCode ierr; 4289 PetscMPIInt rank; 4290 PetscInt m,N,i,rstart,nnz; 4291 size_t len; 4292 const PetscInt *indx; 4293 PetscViewer out; 4294 char *name; 4295 Mat B; 4296 const PetscScalar *values; 4297 4298 PetscFunctionBegin; 4299 ierr = MatGetLocalSize(A,&m,0);CHKERRQ(ierr); 4300 ierr = MatGetSize(A,0,&N);CHKERRQ(ierr); 4301 /* Should this be the type of the diagonal block of A? */ 4302 ierr = MatCreate(PETSC_COMM_SELF,&B);CHKERRQ(ierr); 4303 ierr = MatSetSizes(B,m,N,m,N);CHKERRQ(ierr); 4304 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 4305 ierr = MatSeqAIJSetPreallocation(B,0,PETSC_NULL);CHKERRQ(ierr); 4306 ierr = MatGetOwnershipRange(A,&rstart,0);CHKERRQ(ierr); 4307 for (i=0;i<m;i++) { 4308 ierr = MatGetRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 4309 ierr = MatSetValues(B,1,&i,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 4310 ierr = MatRestoreRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 4311 } 4312 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4313 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4314 4315 ierr = MPI_Comm_rank(((PetscObject)A)->comm,&rank);CHKERRQ(ierr); 4316 ierr = PetscStrlen(outfile,&len);CHKERRQ(ierr); 4317 ierr = PetscMalloc((len+5)*sizeof(char),&name);CHKERRQ(ierr); 4318 sprintf(name,"%s.%d",outfile,rank); 4319 ierr = PetscViewerBinaryOpen(PETSC_COMM_SELF,name,FILE_MODE_APPEND,&out);CHKERRQ(ierr); 4320 ierr = PetscFree(name); 4321 ierr = MatView(B,out);CHKERRQ(ierr); 4322 ierr = PetscViewerDestroy(out);CHKERRQ(ierr); 4323 ierr = MatDestroy(B);CHKERRQ(ierr); 4324 PetscFunctionReturn(0); 4325 } 4326 4327 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 4328 #undef __FUNCT__ 4329 #define __FUNCT__ "MatDestroy_MPIAIJ_SeqsToMPI" 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 4354 ierr = PetscContainerDestroy(container);CHKERRQ(ierr); 4355 ierr = PetscObjectCompose((PetscObject)A,"MatMergeSeqsToMPI",0);CHKERRQ(ierr); 4356 } 4357 ierr = PetscFree(merge);CHKERRQ(ierr); 4358 4359 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 4360 PetscFunctionReturn(0); 4361 } 4362 4363 #include "../src/mat/utils/freespace.h" 4364 #include "petscbt.h" 4365 4366 #undef __FUNCT__ 4367 #define __FUNCT__ "MatMerge_SeqsToMPINumeric" 4368 /*@C 4369 MatMerge_SeqsToMPI - Creates a MPIAIJ matrix by adding sequential 4370 matrices from each processor 4371 4372 Collective on MPI_Comm 4373 4374 Input Parameters: 4375 + comm - the communicators the parallel matrix will live on 4376 . seqmat - the input sequential matrices 4377 . m - number of local rows (or PETSC_DECIDE) 4378 . n - number of local columns (or PETSC_DECIDE) 4379 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4380 4381 Output Parameter: 4382 . mpimat - the parallel matrix generated 4383 4384 Level: advanced 4385 4386 Notes: 4387 The dimensions of the sequential matrix in each processor MUST be the same. 4388 The input seqmat is included into the container "Mat_Merge_SeqsToMPI", and will be 4389 destroyed when mpimat is destroyed. Call PetscObjectQuery() to access seqmat. 4390 @*/ 4391 PetscErrorCode MatMerge_SeqsToMPINumeric(Mat seqmat,Mat mpimat) 4392 { 4393 PetscErrorCode ierr; 4394 MPI_Comm comm=((PetscObject)mpimat)->comm; 4395 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 4396 PetscMPIInt size,rank,taga,*len_s; 4397 PetscInt N=mpimat->cmap->N,i,j,*owners,*ai=a->i,*aj=a->j; 4398 PetscInt proc,m; 4399 PetscInt **buf_ri,**buf_rj; 4400 PetscInt k,anzi,*bj_i,*bi,*bj,arow,bnzi,nextaj; 4401 PetscInt nrows,**buf_ri_k,**nextrow,**nextai; 4402 MPI_Request *s_waits,*r_waits; 4403 MPI_Status *status; 4404 MatScalar *aa=a->a; 4405 MatScalar **abuf_r,*ba_i; 4406 Mat_Merge_SeqsToMPI *merge; 4407 PetscContainer container; 4408 4409 PetscFunctionBegin; 4410 ierr = PetscLogEventBegin(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4411 4412 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4413 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4414 4415 ierr = PetscObjectQuery((PetscObject)mpimat,"MatMergeSeqsToMPI",(PetscObject *)&container);CHKERRQ(ierr); 4416 if (container) { 4417 ierr = PetscContainerGetPointer(container,(void **)&merge);CHKERRQ(ierr); 4418 } 4419 bi = merge->bi; 4420 bj = merge->bj; 4421 buf_ri = merge->buf_ri; 4422 buf_rj = merge->buf_rj; 4423 4424 ierr = PetscMalloc(size*sizeof(MPI_Status),&status);CHKERRQ(ierr); 4425 owners = merge->rowmap->range; 4426 len_s = merge->len_s; 4427 4428 /* send and recv matrix values */ 4429 /*-----------------------------*/ 4430 ierr = PetscObjectGetNewTag((PetscObject)mpimat,&taga);CHKERRQ(ierr); 4431 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 4432 4433 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&s_waits);CHKERRQ(ierr); 4434 for (proc=0,k=0; proc<size; proc++){ 4435 if (!len_s[proc]) continue; 4436 i = owners[proc]; 4437 ierr = MPI_Isend(aa+ai[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 4438 k++; 4439 } 4440 4441 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 4442 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 4443 ierr = PetscFree(status);CHKERRQ(ierr); 4444 4445 ierr = PetscFree(s_waits);CHKERRQ(ierr); 4446 ierr = PetscFree(r_waits);CHKERRQ(ierr); 4447 4448 /* insert mat values of mpimat */ 4449 /*----------------------------*/ 4450 ierr = PetscMalloc(N*sizeof(PetscScalar),&ba_i);CHKERRQ(ierr); 4451 ierr = PetscMalloc3(merge->nrecv,PetscInt*,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextai);CHKERRQ(ierr); 4452 4453 for (k=0; k<merge->nrecv; k++){ 4454 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4455 nrows = *(buf_ri_k[k]); 4456 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 4457 nextai[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 4458 } 4459 4460 /* set values of ba */ 4461 m = merge->rowmap->n; 4462 for (i=0; i<m; i++) { 4463 arow = owners[rank] + i; 4464 bj_i = bj+bi[i]; /* col indices of the i-th row of mpimat */ 4465 bnzi = bi[i+1] - bi[i]; 4466 ierr = PetscMemzero(ba_i,bnzi*sizeof(PetscScalar));CHKERRQ(ierr); 4467 4468 /* add local non-zero vals of this proc's seqmat into ba */ 4469 anzi = ai[arow+1] - ai[arow]; 4470 aj = a->j + ai[arow]; 4471 aa = a->a + ai[arow]; 4472 nextaj = 0; 4473 for (j=0; nextaj<anzi; j++){ 4474 if (*(bj_i + j) == aj[nextaj]){ /* bcol == acol */ 4475 ba_i[j] += aa[nextaj++]; 4476 } 4477 } 4478 4479 /* add received vals into ba */ 4480 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 4481 /* i-th row */ 4482 if (i == *nextrow[k]) { 4483 anzi = *(nextai[k]+1) - *nextai[k]; 4484 aj = buf_rj[k] + *(nextai[k]); 4485 aa = abuf_r[k] + *(nextai[k]); 4486 nextaj = 0; 4487 for (j=0; nextaj<anzi; j++){ 4488 if (*(bj_i + j) == aj[nextaj]){ /* bcol == acol */ 4489 ba_i[j] += aa[nextaj++]; 4490 } 4491 } 4492 nextrow[k]++; nextai[k]++; 4493 } 4494 } 4495 ierr = MatSetValues(mpimat,1,&arow,bnzi,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 4496 } 4497 ierr = MatAssemblyBegin(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4498 ierr = MatAssemblyEnd(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4499 4500 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 4501 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 4502 ierr = PetscFree(ba_i);CHKERRQ(ierr); 4503 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4504 ierr = PetscLogEventEnd(MAT_Seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 4505 PetscFunctionReturn(0); 4506 } 4507 4508 #undef __FUNCT__ 4509 #define __FUNCT__ "MatMerge_SeqsToMPISymbolic" 4510 PetscErrorCode MatMerge_SeqsToMPISymbolic(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,Mat *mpimat) 4511 { 4512 PetscErrorCode ierr; 4513 Mat B_mpi; 4514 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 4515 PetscMPIInt size,rank,tagi,tagj,*len_s,*len_si,*len_ri; 4516 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 4517 PetscInt M=seqmat->rmap->n,N=seqmat->cmap->n,i,*owners,*ai=a->i,*aj=a->j; 4518 PetscInt len,proc,*dnz,*onz; 4519 PetscInt k,anzi,*bi,*bj,*lnk,nlnk,arow,bnzi,nspacedouble=0; 4520 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextai; 4521 MPI_Request *si_waits,*sj_waits,*ri_waits,*rj_waits; 4522 MPI_Status *status; 4523 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 4524 PetscBT lnkbt; 4525 Mat_Merge_SeqsToMPI *merge; 4526 PetscContainer container; 4527 4528 PetscFunctionBegin; 4529 ierr = PetscLogEventBegin(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4530 4531 /* make sure it is a PETSc comm */ 4532 ierr = PetscCommDuplicate(comm,&comm,PETSC_NULL);CHKERRQ(ierr); 4533 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4534 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4535 4536 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 4537 ierr = PetscMalloc(size*sizeof(MPI_Status),&status);CHKERRQ(ierr); 4538 4539 /* determine row ownership */ 4540 /*---------------------------------------------------------*/ 4541 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 4542 ierr = PetscLayoutSetLocalSize(merge->rowmap,m);CHKERRQ(ierr); 4543 ierr = PetscLayoutSetSize(merge->rowmap,M);CHKERRQ(ierr); 4544 ierr = PetscLayoutSetBlockSize(merge->rowmap,1);CHKERRQ(ierr); 4545 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 4546 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 4547 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 4548 4549 m = merge->rowmap->n; 4550 M = merge->rowmap->N; 4551 owners = merge->rowmap->range; 4552 4553 /* determine the number of messages to send, their lengths */ 4554 /*---------------------------------------------------------*/ 4555 len_s = merge->len_s; 4556 4557 len = 0; /* length of buf_si[] */ 4558 merge->nsend = 0; 4559 for (proc=0; proc<size; proc++){ 4560 len_si[proc] = 0; 4561 if (proc == rank){ 4562 len_s[proc] = 0; 4563 } else { 4564 len_si[proc] = owners[proc+1] - owners[proc] + 1; 4565 len_s[proc] = ai[owners[proc+1]] - ai[owners[proc]]; /* num of rows to be sent to [proc] */ 4566 } 4567 if (len_s[proc]) { 4568 merge->nsend++; 4569 nrows = 0; 4570 for (i=owners[proc]; i<owners[proc+1]; i++){ 4571 if (ai[i+1] > ai[i]) nrows++; 4572 } 4573 len_si[proc] = 2*(nrows+1); 4574 len += len_si[proc]; 4575 } 4576 } 4577 4578 /* determine the number and length of messages to receive for ij-structure */ 4579 /*-------------------------------------------------------------------------*/ 4580 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 4581 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 4582 4583 /* post the Irecv of j-structure */ 4584 /*-------------------------------*/ 4585 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 4586 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rj_waits);CHKERRQ(ierr); 4587 4588 /* post the Isend of j-structure */ 4589 /*--------------------------------*/ 4590 ierr = PetscMalloc2(merge->nsend,MPI_Request,&si_waits,merge->nsend,MPI_Request,&sj_waits);CHKERRQ(ierr); 4591 4592 for (proc=0, k=0; proc<size; proc++){ 4593 if (!len_s[proc]) continue; 4594 i = owners[proc]; 4595 ierr = MPI_Isend(aj+ai[i],len_s[proc],MPIU_INT,proc,tagj,comm,sj_waits+k);CHKERRQ(ierr); 4596 k++; 4597 } 4598 4599 /* receives and sends of j-structure are complete */ 4600 /*------------------------------------------------*/ 4601 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,rj_waits,status);CHKERRQ(ierr);} 4602 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,sj_waits,status);CHKERRQ(ierr);} 4603 4604 /* send and recv i-structure */ 4605 /*---------------------------*/ 4606 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 4607 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&ri_waits);CHKERRQ(ierr); 4608 4609 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 4610 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 4611 for (proc=0,k=0; proc<size; proc++){ 4612 if (!len_s[proc]) continue; 4613 /* form outgoing message for i-structure: 4614 buf_si[0]: nrows to be sent 4615 [1:nrows]: row index (global) 4616 [nrows+1:2*nrows+1]: i-structure index 4617 */ 4618 /*-------------------------------------------*/ 4619 nrows = len_si[proc]/2 - 1; 4620 buf_si_i = buf_si + nrows+1; 4621 buf_si[0] = nrows; 4622 buf_si_i[0] = 0; 4623 nrows = 0; 4624 for (i=owners[proc]; i<owners[proc+1]; i++){ 4625 anzi = ai[i+1] - ai[i]; 4626 if (anzi) { 4627 buf_si_i[nrows+1] = buf_si_i[nrows] + anzi; /* i-structure */ 4628 buf_si[nrows+1] = i-owners[proc]; /* local row index */ 4629 nrows++; 4630 } 4631 } 4632 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,si_waits+k);CHKERRQ(ierr); 4633 k++; 4634 buf_si += len_si[proc]; 4635 } 4636 4637 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,ri_waits,status);CHKERRQ(ierr);} 4638 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,si_waits,status);CHKERRQ(ierr);} 4639 4640 ierr = PetscInfo2(seqmat,"nsend: %D, nrecv: %D\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 4641 for (i=0; i<merge->nrecv; i++){ 4642 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); 4643 } 4644 4645 ierr = PetscFree(len_si);CHKERRQ(ierr); 4646 ierr = PetscFree(len_ri);CHKERRQ(ierr); 4647 ierr = PetscFree(rj_waits);CHKERRQ(ierr); 4648 ierr = PetscFree2(si_waits,sj_waits);CHKERRQ(ierr); 4649 ierr = PetscFree(ri_waits);CHKERRQ(ierr); 4650 ierr = PetscFree(buf_s);CHKERRQ(ierr); 4651 ierr = PetscFree(status);CHKERRQ(ierr); 4652 4653 /* compute a local seq matrix in each processor */ 4654 /*----------------------------------------------*/ 4655 /* allocate bi array and free space for accumulating nonzero column info */ 4656 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 4657 bi[0] = 0; 4658 4659 /* create and initialize a linked list */ 4660 nlnk = N+1; 4661 ierr = PetscLLCreate(N,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4662 4663 /* initial FreeSpace size is 2*(num of local nnz(seqmat)) */ 4664 len = 0; 4665 len = ai[owners[rank+1]] - ai[owners[rank]]; 4666 ierr = PetscFreeSpaceGet((PetscInt)(2*len+1),&free_space);CHKERRQ(ierr); 4667 current_space = free_space; 4668 4669 /* determine symbolic info for each local row */ 4670 ierr = PetscMalloc3(merge->nrecv,PetscInt*,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextai);CHKERRQ(ierr); 4671 4672 for (k=0; k<merge->nrecv; k++){ 4673 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 4674 nrows = *buf_ri_k[k]; 4675 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 4676 nextai[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 4677 } 4678 4679 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 4680 len = 0; 4681 for (i=0;i<m;i++) { 4682 bnzi = 0; 4683 /* add local non-zero cols of this proc's seqmat into lnk */ 4684 arow = owners[rank] + i; 4685 anzi = ai[arow+1] - ai[arow]; 4686 aj = a->j + ai[arow]; 4687 ierr = PetscLLAdd(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4688 bnzi += nlnk; 4689 /* add received col data into lnk */ 4690 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 4691 if (i == *nextrow[k]) { /* i-th row */ 4692 anzi = *(nextai[k]+1) - *nextai[k]; 4693 aj = buf_rj[k] + *nextai[k]; 4694 ierr = PetscLLAdd(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 4695 bnzi += nlnk; 4696 nextrow[k]++; nextai[k]++; 4697 } 4698 } 4699 if (len < bnzi) len = bnzi; /* =max(bnzi) */ 4700 4701 /* if free space is not available, make more free space */ 4702 if (current_space->local_remaining<bnzi) { 4703 ierr = PetscFreeSpaceGet(bnzi+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 4704 nspacedouble++; 4705 } 4706 /* copy data into free space, then initialize lnk */ 4707 ierr = PetscLLClean(N,N,bnzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 4708 ierr = MatPreallocateSet(i+owners[rank],bnzi,current_space->array,dnz,onz);CHKERRQ(ierr); 4709 4710 current_space->array += bnzi; 4711 current_space->local_used += bnzi; 4712 current_space->local_remaining -= bnzi; 4713 4714 bi[i+1] = bi[i] + bnzi; 4715 } 4716 4717 ierr = PetscFree3(buf_ri_k,nextrow,nextai);CHKERRQ(ierr); 4718 4719 ierr = PetscMalloc((bi[m]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 4720 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 4721 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 4722 4723 /* create symbolic parallel matrix B_mpi */ 4724 /*---------------------------------------*/ 4725 ierr = MatCreate(comm,&B_mpi);CHKERRQ(ierr); 4726 if (n==PETSC_DECIDE) { 4727 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,N);CHKERRQ(ierr); 4728 } else { 4729 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 4730 } 4731 ierr = MatSetType(B_mpi,MATMPIAIJ);CHKERRQ(ierr); 4732 ierr = MatMPIAIJSetPreallocation(B_mpi,0,dnz,0,onz);CHKERRQ(ierr); 4733 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 4734 4735 /* B_mpi is not ready for use - assembly will be done by MatMerge_SeqsToMPINumeric() */ 4736 B_mpi->assembled = PETSC_FALSE; 4737 B_mpi->ops->destroy = MatDestroy_MPIAIJ_SeqsToMPI; 4738 merge->bi = bi; 4739 merge->bj = bj; 4740 merge->buf_ri = buf_ri; 4741 merge->buf_rj = buf_rj; 4742 merge->coi = PETSC_NULL; 4743 merge->coj = PETSC_NULL; 4744 merge->owners_co = PETSC_NULL; 4745 4746 /* attach the supporting struct to B_mpi for reuse */ 4747 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 4748 ierr = PetscContainerSetPointer(container,merge);CHKERRQ(ierr); 4749 ierr = PetscObjectCompose((PetscObject)B_mpi,"MatMergeSeqsToMPI",(PetscObject)container);CHKERRQ(ierr); 4750 *mpimat = B_mpi; 4751 4752 ierr = PetscCommDestroy(&comm);CHKERRQ(ierr); 4753 ierr = PetscLogEventEnd(MAT_Seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 4754 PetscFunctionReturn(0); 4755 } 4756 4757 #undef __FUNCT__ 4758 #define __FUNCT__ "MatMerge_SeqsToMPI" 4759 PetscErrorCode MatMerge_SeqsToMPI(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,MatReuse scall,Mat *mpimat) 4760 { 4761 PetscErrorCode ierr; 4762 4763 PetscFunctionBegin; 4764 ierr = PetscLogEventBegin(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4765 if (scall == MAT_INITIAL_MATRIX){ 4766 ierr = MatMerge_SeqsToMPISymbolic(comm,seqmat,m,n,mpimat);CHKERRQ(ierr); 4767 } 4768 ierr = MatMerge_SeqsToMPINumeric(seqmat,*mpimat);CHKERRQ(ierr); 4769 ierr = PetscLogEventEnd(MAT_Seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 4770 PetscFunctionReturn(0); 4771 } 4772 4773 #undef __FUNCT__ 4774 #define __FUNCT__ "MatGetLocalMat" 4775 /*@ 4776 MatGetLocalMat - Creates a SeqAIJ matrix by taking all its local rows 4777 4778 Not Collective 4779 4780 Input Parameters: 4781 + A - the matrix 4782 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4783 4784 Output Parameter: 4785 . A_loc - the local sequential matrix generated 4786 4787 Level: developer 4788 4789 @*/ 4790 PetscErrorCode MatGetLocalMat(Mat A,MatReuse scall,Mat *A_loc) 4791 { 4792 PetscErrorCode ierr; 4793 Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data; 4794 Mat_SeqAIJ *mat,*a=(Mat_SeqAIJ*)(mpimat->A)->data,*b=(Mat_SeqAIJ*)(mpimat->B)->data; 4795 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j,*cmap=mpimat->garray; 4796 MatScalar *aa=a->a,*ba=b->a,*cam; 4797 PetscScalar *ca; 4798 PetscInt am=A->rmap->n,i,j,k,cstart=A->cmap->rstart; 4799 PetscInt *ci,*cj,col,ncols_d,ncols_o,jo; 4800 4801 PetscFunctionBegin; 4802 ierr = PetscLogEventBegin(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4803 if (scall == MAT_INITIAL_MATRIX){ 4804 ierr = PetscMalloc((1+am)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 4805 ci[0] = 0; 4806 for (i=0; i<am; i++){ 4807 ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]); 4808 } 4809 ierr = PetscMalloc((1+ci[am])*sizeof(PetscInt),&cj);CHKERRQ(ierr); 4810 ierr = PetscMalloc((1+ci[am])*sizeof(PetscScalar),&ca);CHKERRQ(ierr); 4811 k = 0; 4812 for (i=0; i<am; i++) { 4813 ncols_o = bi[i+1] - bi[i]; 4814 ncols_d = ai[i+1] - ai[i]; 4815 /* off-diagonal portion of A */ 4816 for (jo=0; jo<ncols_o; jo++) { 4817 col = cmap[*bj]; 4818 if (col >= cstart) break; 4819 cj[k] = col; bj++; 4820 ca[k++] = *ba++; 4821 } 4822 /* diagonal portion of A */ 4823 for (j=0; j<ncols_d; j++) { 4824 cj[k] = cstart + *aj++; 4825 ca[k++] = *aa++; 4826 } 4827 /* off-diagonal portion of A */ 4828 for (j=jo; j<ncols_o; j++) { 4829 cj[k] = cmap[*bj++]; 4830 ca[k++] = *ba++; 4831 } 4832 } 4833 /* put together the new matrix */ 4834 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap->N,ci,cj,ca,A_loc);CHKERRQ(ierr); 4835 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4836 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4837 mat = (Mat_SeqAIJ*)(*A_loc)->data; 4838 mat->free_a = PETSC_TRUE; 4839 mat->free_ij = PETSC_TRUE; 4840 mat->nonew = 0; 4841 } else if (scall == MAT_REUSE_MATRIX){ 4842 mat=(Mat_SeqAIJ*)(*A_loc)->data; 4843 ci = mat->i; cj = mat->j; cam = mat->a; 4844 for (i=0; i<am; i++) { 4845 /* off-diagonal portion of A */ 4846 ncols_o = bi[i+1] - bi[i]; 4847 for (jo=0; jo<ncols_o; jo++) { 4848 col = cmap[*bj]; 4849 if (col >= cstart) break; 4850 *cam++ = *ba++; bj++; 4851 } 4852 /* diagonal portion of A */ 4853 ncols_d = ai[i+1] - ai[i]; 4854 for (j=0; j<ncols_d; j++) *cam++ = *aa++; 4855 /* off-diagonal portion of A */ 4856 for (j=jo; j<ncols_o; j++) { 4857 *cam++ = *ba++; bj++; 4858 } 4859 } 4860 } else { 4861 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 4862 } 4863 4864 ierr = PetscLogEventEnd(MAT_Getlocalmat,A,0,0,0);CHKERRQ(ierr); 4865 PetscFunctionReturn(0); 4866 } 4867 4868 #undef __FUNCT__ 4869 #define __FUNCT__ "MatGetLocalMatCondensed" 4870 /*@C 4871 MatGetLocalMatCondensed - Creates a SeqAIJ matrix by taking all its local rows and NON-ZERO columns 4872 4873 Not Collective 4874 4875 Input Parameters: 4876 + A - the matrix 4877 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4878 - row, col - index sets of rows and columns to extract (or PETSC_NULL) 4879 4880 Output Parameter: 4881 . A_loc - the local sequential matrix generated 4882 4883 Level: developer 4884 4885 @*/ 4886 PetscErrorCode MatGetLocalMatCondensed(Mat A,MatReuse scall,IS *row,IS *col,Mat *A_loc) 4887 { 4888 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4889 PetscErrorCode ierr; 4890 PetscInt i,start,end,ncols,nzA,nzB,*cmap,imark,*idx; 4891 IS isrowa,iscola; 4892 Mat *aloc; 4893 4894 PetscFunctionBegin; 4895 ierr = PetscLogEventBegin(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4896 if (!row){ 4897 start = A->rmap->rstart; end = A->rmap->rend; 4898 ierr = ISCreateStride(PETSC_COMM_SELF,end-start,start,1,&isrowa);CHKERRQ(ierr); 4899 } else { 4900 isrowa = *row; 4901 } 4902 if (!col){ 4903 start = A->cmap->rstart; 4904 cmap = a->garray; 4905 nzA = a->A->cmap->n; 4906 nzB = a->B->cmap->n; 4907 ierr = PetscMalloc((nzA+nzB)*sizeof(PetscInt), &idx);CHKERRQ(ierr); 4908 ncols = 0; 4909 for (i=0; i<nzB; i++) { 4910 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4911 else break; 4912 } 4913 imark = i; 4914 for (i=0; i<nzA; i++) idx[ncols++] = start + i; 4915 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; 4916 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&iscola);CHKERRQ(ierr); 4917 } else { 4918 iscola = *col; 4919 } 4920 if (scall != MAT_INITIAL_MATRIX){ 4921 ierr = PetscMalloc(sizeof(Mat),&aloc);CHKERRQ(ierr); 4922 aloc[0] = *A_loc; 4923 } 4924 ierr = MatGetSubMatrices(A,1,&isrowa,&iscola,scall,&aloc);CHKERRQ(ierr); 4925 *A_loc = aloc[0]; 4926 ierr = PetscFree(aloc);CHKERRQ(ierr); 4927 if (!row){ 4928 ierr = ISDestroy(isrowa);CHKERRQ(ierr); 4929 } 4930 if (!col){ 4931 ierr = ISDestroy(iscola);CHKERRQ(ierr); 4932 } 4933 ierr = PetscLogEventEnd(MAT_Getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4934 PetscFunctionReturn(0); 4935 } 4936 4937 #undef __FUNCT__ 4938 #define __FUNCT__ "MatGetBrowsOfAcols" 4939 /*@C 4940 MatGetBrowsOfAcols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns of local A 4941 4942 Collective on Mat 4943 4944 Input Parameters: 4945 + A,B - the matrices in mpiaij format 4946 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4947 - rowb, colb - index sets of rows and columns of B to extract (or PETSC_NULL) 4948 4949 Output Parameter: 4950 + rowb, colb - index sets of rows and columns of B to extract 4951 . brstart - row index of B_seq from which next B->rmap->n rows are taken from B's local rows 4952 - B_seq - the sequential matrix generated 4953 4954 Level: developer 4955 4956 @*/ 4957 PetscErrorCode MatGetBrowsOfAcols(Mat A,Mat B,MatReuse scall,IS *rowb,IS *colb,PetscInt *brstart,Mat *B_seq) 4958 { 4959 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4960 PetscErrorCode ierr; 4961 PetscInt *idx,i,start,ncols,nzA,nzB,*cmap,imark; 4962 IS isrowb,iscolb; 4963 Mat *bseq; 4964 4965 PetscFunctionBegin; 4966 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend){ 4967 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); 4968 } 4969 ierr = PetscLogEventBegin(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4970 4971 if (scall == MAT_INITIAL_MATRIX){ 4972 start = A->cmap->rstart; 4973 cmap = a->garray; 4974 nzA = a->A->cmap->n; 4975 nzB = a->B->cmap->n; 4976 ierr = PetscMalloc((nzA+nzB)*sizeof(PetscInt), &idx);CHKERRQ(ierr); 4977 ncols = 0; 4978 for (i=0; i<nzB; i++) { /* row < local row index */ 4979 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4980 else break; 4981 } 4982 imark = i; 4983 for (i=0; i<nzA; i++) idx[ncols++] = start + i; /* local rows */ 4984 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; /* row > local row index */ 4985 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,PETSC_OWN_POINTER,&isrowb);CHKERRQ(ierr); 4986 *brstart = imark; 4987 ierr = ISCreateStride(PETSC_COMM_SELF,B->cmap->N,0,1,&iscolb);CHKERRQ(ierr); 4988 } else { 4989 if (!rowb || !colb) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"IS rowb and colb must be provided for MAT_REUSE_MATRIX"); 4990 isrowb = *rowb; iscolb = *colb; 4991 ierr = PetscMalloc(sizeof(Mat),&bseq);CHKERRQ(ierr); 4992 bseq[0] = *B_seq; 4993 } 4994 ierr = MatGetSubMatrices(B,1,&isrowb,&iscolb,scall,&bseq);CHKERRQ(ierr); 4995 *B_seq = bseq[0]; 4996 ierr = PetscFree(bseq);CHKERRQ(ierr); 4997 if (!rowb){ 4998 ierr = ISDestroy(isrowb);CHKERRQ(ierr); 4999 } else { 5000 *rowb = isrowb; 5001 } 5002 if (!colb){ 5003 ierr = ISDestroy(iscolb);CHKERRQ(ierr); 5004 } else { 5005 *colb = iscolb; 5006 } 5007 ierr = PetscLogEventEnd(MAT_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 5008 PetscFunctionReturn(0); 5009 } 5010 5011 #undef __FUNCT__ 5012 #define __FUNCT__ "MatGetBrowsOfAoCols" 5013 /*@C 5014 MatGetBrowsOfAoCols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns 5015 of the OFF-DIAGONAL portion of local A 5016 5017 Collective on Mat 5018 5019 Input Parameters: 5020 + A,B - the matrices in mpiaij format 5021 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 5022 . startsj - starting point in B's sending and receiving j-arrays, saved for MAT_REUSE (or PETSC_NULL) 5023 . startsj_r - similar to startsj for receives 5024 - bufa_ptr - array for sending matrix values, saved for MAT_REUSE (or PETSC_NULL) 5025 5026 Output Parameter: 5027 + B_oth - the sequential matrix generated 5028 5029 Level: developer 5030 5031 @*/ 5032 PetscErrorCode MatGetBrowsOfAoCols(Mat A,Mat B,MatReuse scall,PetscInt **startsj,PetscInt **startsj_r,MatScalar **bufa_ptr,Mat *B_oth) 5033 { 5034 VecScatter_MPI_General *gen_to,*gen_from; 5035 PetscErrorCode ierr; 5036 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 5037 Mat_SeqAIJ *b_oth; 5038 VecScatter ctx=a->Mvctx; 5039 MPI_Comm comm=((PetscObject)ctx)->comm; 5040 PetscMPIInt *rprocs,*sprocs,tag=((PetscObject)ctx)->tag,rank; 5041 PetscInt *rowlen,*bufj,*bufJ,ncols,aBn=a->B->cmap->n,row,*b_othi,*b_othj; 5042 PetscScalar *rvalues,*svalues; 5043 MatScalar *b_otha,*bufa,*bufA; 5044 PetscInt i,j,k,l,ll,nrecvs,nsends,nrows,*srow,*rstarts,*rstartsj = 0,*sstarts,*sstartsj,len; 5045 MPI_Request *rwaits = PETSC_NULL,*swaits = PETSC_NULL; 5046 MPI_Status *sstatus,rstatus; 5047 PetscMPIInt jj; 5048 PetscInt *cols,sbs,rbs; 5049 PetscScalar *vals; 5050 5051 PetscFunctionBegin; 5052 if (A->cmap->rstart != B->rmap->rstart || A->cmap->rend != B->rmap->rend){ 5053 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); 5054 } 5055 ierr = PetscLogEventBegin(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 5056 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 5057 5058 gen_to = (VecScatter_MPI_General*)ctx->todata; 5059 gen_from = (VecScatter_MPI_General*)ctx->fromdata; 5060 rvalues = gen_from->values; /* holds the length of receiving row */ 5061 svalues = gen_to->values; /* holds the length of sending row */ 5062 nrecvs = gen_from->n; 5063 nsends = gen_to->n; 5064 5065 ierr = PetscMalloc2(nrecvs,MPI_Request,&rwaits,nsends,MPI_Request,&swaits);CHKERRQ(ierr); 5066 srow = gen_to->indices; /* local row index to be sent */ 5067 sstarts = gen_to->starts; 5068 sprocs = gen_to->procs; 5069 sstatus = gen_to->sstatus; 5070 sbs = gen_to->bs; 5071 rstarts = gen_from->starts; 5072 rprocs = gen_from->procs; 5073 rbs = gen_from->bs; 5074 5075 if (!startsj || !bufa_ptr) scall = MAT_INITIAL_MATRIX; 5076 if (scall == MAT_INITIAL_MATRIX){ 5077 /* i-array */ 5078 /*---------*/ 5079 /* post receives */ 5080 for (i=0; i<nrecvs; i++){ 5081 rowlen = (PetscInt*)rvalues + rstarts[i]*rbs; 5082 nrows = (rstarts[i+1]-rstarts[i])*rbs; /* num of indices to be received */ 5083 ierr = MPI_Irecv(rowlen,nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5084 } 5085 5086 /* pack the outgoing message */ 5087 ierr = PetscMalloc2(nsends+1,PetscInt,&sstartsj,nrecvs+1,PetscInt,&rstartsj);CHKERRQ(ierr); 5088 sstartsj[0] = 0; rstartsj[0] = 0; 5089 len = 0; /* total length of j or a array to be sent */ 5090 k = 0; 5091 for (i=0; i<nsends; i++){ 5092 rowlen = (PetscInt*)svalues + sstarts[i]*sbs; 5093 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5094 for (j=0; j<nrows; j++) { 5095 row = srow[k] + B->rmap->range[rank]; /* global row idx */ 5096 for (l=0; l<sbs; l++){ 5097 ierr = MatGetRow_MPIAIJ(B,row+l,&ncols,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr); /* rowlength */ 5098 rowlen[j*sbs+l] = ncols; 5099 len += ncols; 5100 ierr = MatRestoreRow_MPIAIJ(B,row+l,&ncols,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr); 5101 } 5102 k++; 5103 } 5104 ierr = MPI_Isend(rowlen,nrows*sbs,MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5105 sstartsj[i+1] = len; /* starting point of (i+1)-th outgoing msg in bufj and bufa */ 5106 } 5107 /* recvs and sends of i-array are completed */ 5108 i = nrecvs; 5109 while (i--) { 5110 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5111 } 5112 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5113 5114 /* allocate buffers for sending j and a arrays */ 5115 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&bufj);CHKERRQ(ierr); 5116 ierr = PetscMalloc((len+1)*sizeof(PetscScalar),&bufa);CHKERRQ(ierr); 5117 5118 /* create i-array of B_oth */ 5119 ierr = PetscMalloc((aBn+2)*sizeof(PetscInt),&b_othi);CHKERRQ(ierr); 5120 b_othi[0] = 0; 5121 len = 0; /* total length of j or a array to be received */ 5122 k = 0; 5123 for (i=0; i<nrecvs; i++){ 5124 rowlen = (PetscInt*)rvalues + rstarts[i]*rbs; 5125 nrows = rbs*(rstarts[i+1]-rstarts[i]); /* num of rows to be recieved */ 5126 for (j=0; j<nrows; j++) { 5127 b_othi[k+1] = b_othi[k] + rowlen[j]; 5128 len += rowlen[j]; k++; 5129 } 5130 rstartsj[i+1] = len; /* starting point of (i+1)-th incoming msg in bufj and bufa */ 5131 } 5132 5133 /* allocate space for j and a arrrays of B_oth */ 5134 ierr = PetscMalloc((b_othi[aBn]+1)*sizeof(PetscInt),&b_othj);CHKERRQ(ierr); 5135 ierr = PetscMalloc((b_othi[aBn]+1)*sizeof(MatScalar),&b_otha);CHKERRQ(ierr); 5136 5137 /* j-array */ 5138 /*---------*/ 5139 /* post receives of j-array */ 5140 for (i=0; i<nrecvs; i++){ 5141 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 5142 ierr = MPI_Irecv(b_othj+rstartsj[i],nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5143 } 5144 5145 /* pack the outgoing message j-array */ 5146 k = 0; 5147 for (i=0; i<nsends; i++){ 5148 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5149 bufJ = bufj+sstartsj[i]; 5150 for (j=0; j<nrows; j++) { 5151 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 5152 for (ll=0; ll<sbs; ll++){ 5153 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,&cols,PETSC_NULL);CHKERRQ(ierr); 5154 for (l=0; l<ncols; l++){ 5155 *bufJ++ = cols[l]; 5156 } 5157 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,&cols,PETSC_NULL);CHKERRQ(ierr); 5158 } 5159 } 5160 ierr = MPI_Isend(bufj+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5161 } 5162 5163 /* recvs and sends of j-array are completed */ 5164 i = nrecvs; 5165 while (i--) { 5166 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5167 } 5168 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5169 } else if (scall == MAT_REUSE_MATRIX){ 5170 sstartsj = *startsj; 5171 rstartsj = *startsj_r; 5172 bufa = *bufa_ptr; 5173 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 5174 b_otha = b_oth->a; 5175 } else { 5176 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE, "Matrix P does not posses an object container"); 5177 } 5178 5179 /* a-array */ 5180 /*---------*/ 5181 /* post receives of a-array */ 5182 for (i=0; i<nrecvs; i++){ 5183 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 5184 ierr = MPI_Irecv(b_otha+rstartsj[i],nrows,MPIU_SCALAR,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 5185 } 5186 5187 /* pack the outgoing message a-array */ 5188 k = 0; 5189 for (i=0; i<nsends; i++){ 5190 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 5191 bufA = bufa+sstartsj[i]; 5192 for (j=0; j<nrows; j++) { 5193 row = srow[k++] + B->rmap->range[rank]; /* global row idx */ 5194 for (ll=0; ll<sbs; ll++){ 5195 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,PETSC_NULL,&vals);CHKERRQ(ierr); 5196 for (l=0; l<ncols; l++){ 5197 *bufA++ = vals[l]; 5198 } 5199 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,PETSC_NULL,&vals);CHKERRQ(ierr); 5200 } 5201 } 5202 ierr = MPI_Isend(bufa+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_SCALAR,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 5203 } 5204 /* recvs and sends of a-array are completed */ 5205 i = nrecvs; 5206 while (i--) { 5207 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 5208 } 5209 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 5210 ierr = PetscFree2(rwaits,swaits);CHKERRQ(ierr); 5211 5212 if (scall == MAT_INITIAL_MATRIX){ 5213 /* put together the new matrix */ 5214 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,aBn,B->cmap->N,b_othi,b_othj,b_otha,B_oth);CHKERRQ(ierr); 5215 5216 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 5217 /* Since these are PETSc arrays, change flags to free them as necessary. */ 5218 b_oth = (Mat_SeqAIJ *)(*B_oth)->data; 5219 b_oth->free_a = PETSC_TRUE; 5220 b_oth->free_ij = PETSC_TRUE; 5221 b_oth->nonew = 0; 5222 5223 ierr = PetscFree(bufj);CHKERRQ(ierr); 5224 if (!startsj || !bufa_ptr){ 5225 ierr = PetscFree2(sstartsj,rstartsj);CHKERRQ(ierr); 5226 ierr = PetscFree(bufa_ptr);CHKERRQ(ierr); 5227 } else { 5228 *startsj = sstartsj; 5229 *startsj_r = rstartsj; 5230 *bufa_ptr = bufa; 5231 } 5232 } 5233 ierr = PetscLogEventEnd(MAT_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 5234 PetscFunctionReturn(0); 5235 } 5236 5237 #undef __FUNCT__ 5238 #define __FUNCT__ "MatGetCommunicationStructs" 5239 /*@C 5240 MatGetCommunicationStructs - Provides access to the communication structures used in matrix-vector multiplication. 5241 5242 Not Collective 5243 5244 Input Parameters: 5245 . A - The matrix in mpiaij format 5246 5247 Output Parameter: 5248 + lvec - The local vector holding off-process values from the argument to a matrix-vector product 5249 . colmap - A map from global column index to local index into lvec 5250 - multScatter - A scatter from the argument of a matrix-vector product to lvec 5251 5252 Level: developer 5253 5254 @*/ 5255 #if defined (PETSC_USE_CTABLE) 5256 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscTable *colmap, VecScatter *multScatter) 5257 #else 5258 PetscErrorCode MatGetCommunicationStructs(Mat A, Vec *lvec, PetscInt *colmap[], VecScatter *multScatter) 5259 #endif 5260 { 5261 Mat_MPIAIJ *a; 5262 5263 PetscFunctionBegin; 5264 PetscValidHeaderSpecific(A, MAT_CLASSID, 1); 5265 PetscValidPointer(lvec, 2); 5266 PetscValidPointer(colmap, 3); 5267 PetscValidPointer(multScatter, 4); 5268 a = (Mat_MPIAIJ *) A->data; 5269 if (lvec) *lvec = a->lvec; 5270 if (colmap) *colmap = a->colmap; 5271 if (multScatter) *multScatter = a->Mvctx; 5272 PetscFunctionReturn(0); 5273 } 5274 5275 EXTERN_C_BEGIN 5276 extern PetscErrorCode MatConvert_MPIAIJ_MPIAIJCRL(Mat,const MatType,MatReuse,Mat*); 5277 extern PetscErrorCode MatConvert_MPIAIJ_MPIAIJPERM(Mat,const MatType,MatReuse,Mat*); 5278 extern PetscErrorCode MatConvert_MPIAIJ_MPISBAIJ(Mat,const MatType,MatReuse,Mat*); 5279 EXTERN_C_END 5280 5281 #undef __FUNCT__ 5282 #define __FUNCT__ "MatMatMultNumeric_MPIDense_MPIAIJ" 5283 /* 5284 Computes (B'*A')' since computing B*A directly is untenable 5285 5286 n p p 5287 ( ) ( ) ( ) 5288 m ( A ) * n ( B ) = m ( C ) 5289 ( ) ( ) ( ) 5290 5291 */ 5292 PetscErrorCode MatMatMultNumeric_MPIDense_MPIAIJ(Mat A,Mat B,Mat C) 5293 { 5294 PetscErrorCode ierr; 5295 Mat At,Bt,Ct; 5296 5297 PetscFunctionBegin; 5298 ierr = MatTranspose(A,MAT_INITIAL_MATRIX,&At);CHKERRQ(ierr); 5299 ierr = MatTranspose(B,MAT_INITIAL_MATRIX,&Bt);CHKERRQ(ierr); 5300 ierr = MatMatMult(Bt,At,MAT_INITIAL_MATRIX,1.0,&Ct);CHKERRQ(ierr); 5301 ierr = MatDestroy(At);CHKERRQ(ierr); 5302 ierr = MatDestroy(Bt);CHKERRQ(ierr); 5303 ierr = MatTranspose(Ct,MAT_REUSE_MATRIX,&C);CHKERRQ(ierr); 5304 ierr = MatDestroy(Ct);CHKERRQ(ierr); 5305 PetscFunctionReturn(0); 5306 } 5307 5308 #undef __FUNCT__ 5309 #define __FUNCT__ "MatMatMultSymbolic_MPIDense_MPIAIJ" 5310 PetscErrorCode MatMatMultSymbolic_MPIDense_MPIAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 5311 { 5312 PetscErrorCode ierr; 5313 PetscInt m=A->rmap->n,n=B->cmap->n; 5314 Mat Cmat; 5315 5316 PetscFunctionBegin; 5317 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); 5318 ierr = MatCreate(((PetscObject)A)->comm,&Cmat);CHKERRQ(ierr); 5319 ierr = MatSetSizes(Cmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 5320 ierr = MatSetType(Cmat,MATMPIDENSE);CHKERRQ(ierr); 5321 ierr = MatMPIDenseSetPreallocation(Cmat,PETSC_NULL);CHKERRQ(ierr); 5322 ierr = MatAssemblyBegin(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5323 ierr = MatAssemblyEnd(Cmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5324 *C = Cmat; 5325 PetscFunctionReturn(0); 5326 } 5327 5328 /* ----------------------------------------------------------------*/ 5329 #undef __FUNCT__ 5330 #define __FUNCT__ "MatMatMult_MPIDense_MPIAIJ" 5331 PetscErrorCode MatMatMult_MPIDense_MPIAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 5332 { 5333 PetscErrorCode ierr; 5334 5335 PetscFunctionBegin; 5336 if (scall == MAT_INITIAL_MATRIX){ 5337 ierr = MatMatMultSymbolic_MPIDense_MPIAIJ(A,B,fill,C);CHKERRQ(ierr); 5338 } 5339 ierr = MatMatMultNumeric_MPIDense_MPIAIJ(A,B,*C);CHKERRQ(ierr); 5340 PetscFunctionReturn(0); 5341 } 5342 5343 EXTERN_C_BEGIN 5344 #if defined(PETSC_HAVE_MUMPS) 5345 extern PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*); 5346 #endif 5347 #if defined(PETSC_HAVE_PASTIX) 5348 extern PetscErrorCode MatGetFactor_mpiaij_pastix(Mat,MatFactorType,Mat*); 5349 #endif 5350 #if defined(PETSC_HAVE_SUPERLU_DIST) 5351 extern PetscErrorCode MatGetFactor_mpiaij_superlu_dist(Mat,MatFactorType,Mat*); 5352 #endif 5353 #if defined(PETSC_HAVE_SPOOLES) 5354 extern PetscErrorCode MatGetFactor_mpiaij_spooles(Mat,MatFactorType,Mat*); 5355 #endif 5356 EXTERN_C_END 5357 5358 /*MC 5359 MATMPIAIJ - MATMPIAIJ = "mpiaij" - A matrix type to be used for parallel sparse matrices. 5360 5361 Options Database Keys: 5362 . -mat_type mpiaij - sets the matrix type to "mpiaij" during a call to MatSetFromOptions() 5363 5364 Level: beginner 5365 5366 .seealso: MatCreateMPIAIJ() 5367 M*/ 5368 5369 EXTERN_C_BEGIN 5370 #undef __FUNCT__ 5371 #define __FUNCT__ "MatCreate_MPIAIJ" 5372 PetscErrorCode MatCreate_MPIAIJ(Mat B) 5373 { 5374 Mat_MPIAIJ *b; 5375 PetscErrorCode ierr; 5376 PetscMPIInt size; 5377 5378 PetscFunctionBegin; 5379 ierr = MPI_Comm_size(((PetscObject)B)->comm,&size);CHKERRQ(ierr); 5380 5381 ierr = PetscNewLog(B,Mat_MPIAIJ,&b);CHKERRQ(ierr); 5382 B->data = (void*)b; 5383 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 5384 B->rmap->bs = 1; 5385 B->assembled = PETSC_FALSE; 5386 5387 B->insertmode = NOT_SET_VALUES; 5388 b->size = size; 5389 ierr = MPI_Comm_rank(((PetscObject)B)->comm,&b->rank);CHKERRQ(ierr); 5390 5391 /* build cache for off array entries formed */ 5392 ierr = MatStashCreate_Private(((PetscObject)B)->comm,1,&B->stash);CHKERRQ(ierr); 5393 b->donotstash = PETSC_FALSE; 5394 b->colmap = 0; 5395 b->garray = 0; 5396 b->roworiented = PETSC_TRUE; 5397 5398 /* stuff used for matrix vector multiply */ 5399 b->lvec = PETSC_NULL; 5400 b->Mvctx = PETSC_NULL; 5401 5402 /* stuff for MatGetRow() */ 5403 b->rowindices = 0; 5404 b->rowvalues = 0; 5405 b->getrowactive = PETSC_FALSE; 5406 5407 #if defined(PETSC_HAVE_SPOOLES) 5408 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetFactor_spooles_C", 5409 "MatGetFactor_mpiaij_spooles", 5410 MatGetFactor_mpiaij_spooles);CHKERRQ(ierr); 5411 #endif 5412 #if defined(PETSC_HAVE_MUMPS) 5413 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetFactor_mumps_C", 5414 "MatGetFactor_aij_mumps", 5415 MatGetFactor_aij_mumps);CHKERRQ(ierr); 5416 #endif 5417 #if defined(PETSC_HAVE_PASTIX) 5418 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetFactor_pastix_C", 5419 "MatGetFactor_mpiaij_pastix", 5420 MatGetFactor_mpiaij_pastix);CHKERRQ(ierr); 5421 #endif 5422 #if defined(PETSC_HAVE_SUPERLU_DIST) 5423 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetFactor_superlu_dist_C", 5424 "MatGetFactor_mpiaij_superlu_dist", 5425 MatGetFactor_mpiaij_superlu_dist);CHKERRQ(ierr); 5426 #endif 5427 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C", 5428 "MatStoreValues_MPIAIJ", 5429 MatStoreValues_MPIAIJ);CHKERRQ(ierr); 5430 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C", 5431 "MatRetrieveValues_MPIAIJ", 5432 MatRetrieveValues_MPIAIJ);CHKERRQ(ierr); 5433 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetDiagonalBlock_C", 5434 "MatGetDiagonalBlock_MPIAIJ", 5435 MatGetDiagonalBlock_MPIAIJ);CHKERRQ(ierr); 5436 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatIsTranspose_C", 5437 "MatIsTranspose_MPIAIJ", 5438 MatIsTranspose_MPIAIJ);CHKERRQ(ierr); 5439 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPIAIJSetPreallocation_C", 5440 "MatMPIAIJSetPreallocation_MPIAIJ", 5441 MatMPIAIJSetPreallocation_MPIAIJ);CHKERRQ(ierr); 5442 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPIAIJSetPreallocationCSR_C", 5443 "MatMPIAIJSetPreallocationCSR_MPIAIJ", 5444 MatMPIAIJSetPreallocationCSR_MPIAIJ);CHKERRQ(ierr); 5445 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatDiagonalScaleLocal_C", 5446 "MatDiagonalScaleLocal_MPIAIJ", 5447 MatDiagonalScaleLocal_MPIAIJ);CHKERRQ(ierr); 5448 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_mpiaij_mpiaijperm_C", 5449 "MatConvert_MPIAIJ_MPIAIJPERM", 5450 MatConvert_MPIAIJ_MPIAIJPERM);CHKERRQ(ierr); 5451 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_mpiaij_mpiaijcrl_C", 5452 "MatConvert_MPIAIJ_MPIAIJCRL", 5453 MatConvert_MPIAIJ_MPIAIJCRL);CHKERRQ(ierr); 5454 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_mpiaij_mpisbaij_C", 5455 "MatConvert_MPIAIJ_MPISBAIJ", 5456 MatConvert_MPIAIJ_MPISBAIJ);CHKERRQ(ierr); 5457 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMult_mpidense_mpiaij_C", 5458 "MatMatMult_MPIDense_MPIAIJ", 5459 MatMatMult_MPIDense_MPIAIJ);CHKERRQ(ierr); 5460 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMultSymbolic_mpidense_mpiaij_C", 5461 "MatMatMultSymbolic_MPIDense_MPIAIJ", 5462 MatMatMultSymbolic_MPIDense_MPIAIJ);CHKERRQ(ierr); 5463 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMultNumeric_mpidense_mpiaij_C", 5464 "MatMatMultNumeric_MPIDense_MPIAIJ", 5465 MatMatMultNumeric_MPIDense_MPIAIJ);CHKERRQ(ierr); 5466 ierr = PetscObjectChangeTypeName((PetscObject)B,MATMPIAIJ);CHKERRQ(ierr); 5467 PetscFunctionReturn(0); 5468 } 5469 EXTERN_C_END 5470 5471 #undef __FUNCT__ 5472 #define __FUNCT__ "MatCreateMPIAIJWithSplitArrays" 5473 /*@ 5474 MatCreateMPIAIJWithSplitArrays - creates a MPI AIJ matrix using arrays that contain the "diagonal" 5475 and "off-diagonal" part of the matrix in CSR format. 5476 5477 Collective on MPI_Comm 5478 5479 Input Parameters: 5480 + comm - MPI communicator 5481 . m - number of local rows (Cannot be PETSC_DECIDE) 5482 . n - This value should be the same as the local size used in creating the 5483 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 5484 calculated if N is given) For square matrices n is almost always m. 5485 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 5486 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 5487 . i - row indices for "diagonal" portion of matrix 5488 . j - column indices 5489 . a - matrix values 5490 . oi - row indices for "off-diagonal" portion of matrix 5491 . oj - column indices 5492 - oa - matrix values 5493 5494 Output Parameter: 5495 . mat - the matrix 5496 5497 Level: advanced 5498 5499 Notes: 5500 The i, j, and a arrays ARE NOT copied by this routine into the internal format used by PETSc. The user 5501 must free the arrays once the matrix has been destroyed and not before. 5502 5503 The i and j indices are 0 based 5504 5505 See MatCreateMPIAIJ() for the definition of "diagonal" and "off-diagonal" portion of the matrix 5506 5507 This sets local rows and cannot be used to set off-processor values. 5508 5509 You cannot later use MatSetValues() to change values in this matrix. 5510 5511 .keywords: matrix, aij, compressed row, sparse, parallel 5512 5513 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 5514 MPIAIJ, MatCreateMPIAIJ(), MatCreateMPIAIJWithArrays() 5515 @*/ 5516 PetscErrorCode MatCreateMPIAIJWithSplitArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt i[],PetscInt j[],PetscScalar a[], 5517 PetscInt oi[], PetscInt oj[],PetscScalar oa[],Mat *mat) 5518 { 5519 PetscErrorCode ierr; 5520 Mat_MPIAIJ *maij; 5521 5522 PetscFunctionBegin; 5523 if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 5524 if (i[0]) { 5525 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 5526 } 5527 if (oi[0]) { 5528 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"oi (row indices) must start with 0"); 5529 } 5530 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 5531 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 5532 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 5533 maij = (Mat_MPIAIJ*) (*mat)->data; 5534 maij->donotstash = PETSC_TRUE; 5535 (*mat)->preallocated = PETSC_TRUE; 5536 5537 ierr = PetscLayoutSetBlockSize((*mat)->rmap,1);CHKERRQ(ierr); 5538 ierr = PetscLayoutSetBlockSize((*mat)->cmap,1);CHKERRQ(ierr); 5539 ierr = PetscLayoutSetUp((*mat)->rmap);CHKERRQ(ierr); 5540 ierr = PetscLayoutSetUp((*mat)->cmap);CHKERRQ(ierr); 5541 5542 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,i,j,a,&maij->A);CHKERRQ(ierr); 5543 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,(*mat)->cmap->N,oi,oj,oa,&maij->B);CHKERRQ(ierr); 5544 5545 ierr = MatAssemblyBegin(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5546 ierr = MatAssemblyEnd(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5547 ierr = MatAssemblyBegin(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5548 ierr = MatAssemblyEnd(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5549 5550 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5551 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 5552 PetscFunctionReturn(0); 5553 } 5554 5555 /* 5556 Special version for direct calls from Fortran 5557 */ 5558 #include "private/fortranimpl.h" 5559 5560 #if defined(PETSC_HAVE_FORTRAN_CAPS) 5561 #define matsetvaluesmpiaij_ MATSETVALUESMPIAIJ 5562 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 5563 #define matsetvaluesmpiaij_ matsetvaluesmpiaij 5564 #endif 5565 5566 /* Change these macros so can be used in void function */ 5567 #undef CHKERRQ 5568 #define CHKERRQ(ierr) CHKERRABORT(PETSC_COMM_WORLD,ierr) 5569 #undef SETERRQ2 5570 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 5571 #undef SETERRQ 5572 #define SETERRQ(c,ierr,b) CHKERRABORT(c,ierr) 5573 5574 EXTERN_C_BEGIN 5575 #undef __FUNCT__ 5576 #define __FUNCT__ "matsetvaluesmpiaij_" 5577 void PETSC_STDCALL matsetvaluesmpiaij_(Mat *mmat,PetscInt *mm,const PetscInt im[],PetscInt *mn,const PetscInt in[],const PetscScalar v[],InsertMode *maddv,PetscErrorCode *_ierr) 5578 { 5579 Mat mat = *mmat; 5580 PetscInt m = *mm, n = *mn; 5581 InsertMode addv = *maddv; 5582 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 5583 PetscScalar value; 5584 PetscErrorCode ierr; 5585 5586 ierr = MatPreallocated(mat);CHKERRQ(ierr); 5587 if (mat->insertmode == NOT_SET_VALUES) { 5588 mat->insertmode = addv; 5589 } 5590 #if defined(PETSC_USE_DEBUG) 5591 else if (mat->insertmode != addv) { 5592 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot mix add values and insert values"); 5593 } 5594 #endif 5595 { 5596 PetscInt i,j,rstart = mat->rmap->rstart,rend = mat->rmap->rend; 5597 PetscInt cstart = mat->cmap->rstart,cend = mat->cmap->rend,row,col; 5598 PetscBool roworiented = aij->roworiented; 5599 5600 /* Some Variables required in the macro */ 5601 Mat A = aij->A; 5602 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 5603 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 5604 MatScalar *aa = a->a; 5605 PetscBool ignorezeroentries = (((a->ignorezeroentries)&&(addv==ADD_VALUES))?PETSC_TRUE:PETSC_FALSE); 5606 Mat B = aij->B; 5607 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 5608 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap->n,am = aij->A->rmap->n; 5609 MatScalar *ba = b->a; 5610 5611 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 5612 PetscInt nonew = a->nonew; 5613 MatScalar *ap1,*ap2; 5614 5615 PetscFunctionBegin; 5616 for (i=0; i<m; i++) { 5617 if (im[i] < 0) continue; 5618 #if defined(PETSC_USE_DEBUG) 5619 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); 5620 #endif 5621 if (im[i] >= rstart && im[i] < rend) { 5622 row = im[i] - rstart; 5623 lastcol1 = -1; 5624 rp1 = aj + ai[row]; 5625 ap1 = aa + ai[row]; 5626 rmax1 = aimax[row]; 5627 nrow1 = ailen[row]; 5628 low1 = 0; 5629 high1 = nrow1; 5630 lastcol2 = -1; 5631 rp2 = bj + bi[row]; 5632 ap2 = ba + bi[row]; 5633 rmax2 = bimax[row]; 5634 nrow2 = bilen[row]; 5635 low2 = 0; 5636 high2 = nrow2; 5637 5638 for (j=0; j<n; j++) { 5639 if (roworiented) value = v[i*n+j]; else value = v[i+j*m]; 5640 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES)) continue; 5641 if (in[j] >= cstart && in[j] < cend){ 5642 col = in[j] - cstart; 5643 MatSetValues_SeqAIJ_A_Private(row,col,value,addv); 5644 } else if (in[j] < 0) continue; 5645 #if defined(PETSC_USE_DEBUG) 5646 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); 5647 #endif 5648 else { 5649 if (mat->was_assembled) { 5650 if (!aij->colmap) { 5651 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 5652 } 5653 #if defined (PETSC_USE_CTABLE) 5654 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 5655 col--; 5656 #else 5657 col = aij->colmap[in[j]] - 1; 5658 #endif 5659 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 5660 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 5661 col = in[j]; 5662 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 5663 B = aij->B; 5664 b = (Mat_SeqAIJ*)B->data; 5665 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; 5666 rp2 = bj + bi[row]; 5667 ap2 = ba + bi[row]; 5668 rmax2 = bimax[row]; 5669 nrow2 = bilen[row]; 5670 low2 = 0; 5671 high2 = nrow2; 5672 bm = aij->B->rmap->n; 5673 ba = b->a; 5674 } 5675 } else col = in[j]; 5676 MatSetValues_SeqAIJ_B_Private(row,col,value,addv); 5677 } 5678 } 5679 } else { 5680 if (!aij->donotstash) { 5681 if (roworiented) { 5682 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5683 } else { 5684 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,(PetscBool)(ignorezeroentries && (addv == ADD_VALUES)));CHKERRQ(ierr); 5685 } 5686 } 5687 } 5688 }} 5689 PetscFunctionReturnVoid(); 5690 } 5691 EXTERN_C_END 5692 5693