1 #define PETSCMAT_DLL 2 3 #include "src/mat/impls/aij/mpi/mpiaij.h" /*I "petscmat.h" I*/ 4 #include "src/inline/spops.h" 5 6 /* 7 Local utility routine that creates a mapping from the global column 8 number to the local number in the off-diagonal part of the local 9 storage of the matrix. When PETSC_USE_CTABLE is used this is scalable at 10 a slightly higher hash table cost; without it it is not scalable (each processor 11 has an order N integer array but is fast to acess. 12 */ 13 #undef __FUNCT__ 14 #define __FUNCT__ "CreateColmap_MPIAIJ_Private" 15 PetscErrorCode CreateColmap_MPIAIJ_Private(Mat mat) 16 { 17 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 18 PetscErrorCode ierr; 19 PetscInt n = aij->B->cmap.n,i; 20 21 PetscFunctionBegin; 22 #if defined (PETSC_USE_CTABLE) 23 ierr = PetscTableCreate(n,&aij->colmap);CHKERRQ(ierr); 24 for (i=0; i<n; i++){ 25 ierr = PetscTableAdd(aij->colmap,aij->garray[i]+1,i+1);CHKERRQ(ierr); 26 } 27 #else 28 ierr = PetscMalloc((mat->cmap.N+1)*sizeof(PetscInt),&aij->colmap);CHKERRQ(ierr); 29 ierr = PetscLogObjectMemory(mat,mat->cmap.N*sizeof(PetscInt));CHKERRQ(ierr); 30 ierr = PetscMemzero(aij->colmap,mat->cmap.N*sizeof(PetscInt));CHKERRQ(ierr); 31 for (i=0; i<n; i++) aij->colmap[aij->garray[i]] = i+1; 32 #endif 33 PetscFunctionReturn(0); 34 } 35 36 37 #define CHUNKSIZE 15 38 #define MatSetValues_SeqAIJ_A_Private(row,col,value,addv) \ 39 { \ 40 if (col <= lastcol1) low1 = 0; else high1 = nrow1; \ 41 lastcol1 = col;\ 42 while (high1-low1 > 5) { \ 43 t = (low1+high1)/2; \ 44 if (rp1[t] > col) high1 = t; \ 45 else low1 = t; \ 46 } \ 47 for (_i=low1; _i<high1; _i++) { \ 48 if (rp1[_i] > col) break; \ 49 if (rp1[_i] == col) { \ 50 if (addv == ADD_VALUES) ap1[_i] += value; \ 51 else ap1[_i] = value; \ 52 goto a_noinsert; \ 53 } \ 54 } \ 55 if (value == 0.0 && ignorezeroentries) goto a_noinsert; \ 56 if (nonew == 1) goto a_noinsert; \ 57 if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \ 58 MatSeqXAIJReallocateAIJ(A,am,1,nrow1,row,col,rmax1,aa,ai,aj,rp1,ap1,aimax,nonew,MatScalar); \ 59 N = nrow1++ - 1; a->nz++; high1++; \ 60 /* shift up all the later entries in this row */ \ 61 for (ii=N; ii>=_i; ii--) { \ 62 rp1[ii+1] = rp1[ii]; \ 63 ap1[ii+1] = ap1[ii]; \ 64 } \ 65 rp1[_i] = col; \ 66 ap1[_i] = value; \ 67 a_noinsert: ; \ 68 ailen[row] = nrow1; \ 69 } 70 71 72 #define MatSetValues_SeqAIJ_B_Private(row,col,value,addv) \ 73 { \ 74 if (col <= lastcol2) low2 = 0; else high2 = nrow2; \ 75 lastcol2 = col;\ 76 while (high2-low2 > 5) { \ 77 t = (low2+high2)/2; \ 78 if (rp2[t] > col) high2 = t; \ 79 else low2 = t; \ 80 } \ 81 for (_i=low2; _i<high2; _i++) { \ 82 if (rp2[_i] > col) break; \ 83 if (rp2[_i] == col) { \ 84 if (addv == ADD_VALUES) ap2[_i] += value; \ 85 else ap2[_i] = value; \ 86 goto b_noinsert; \ 87 } \ 88 } \ 89 if (value == 0.0 && ignorezeroentries) goto b_noinsert; \ 90 if (nonew == 1) goto b_noinsert; \ 91 if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \ 92 MatSeqXAIJReallocateAIJ(B,bm,1,nrow2,row,col,rmax2,ba,bi,bj,rp2,ap2,bimax,nonew,MatScalar); \ 93 N = nrow2++ - 1; b->nz++; high2++;\ 94 /* shift up all the later entries in this row */ \ 95 for (ii=N; ii>=_i; ii--) { \ 96 rp2[ii+1] = rp2[ii]; \ 97 ap2[ii+1] = ap2[ii]; \ 98 } \ 99 rp2[_i] = col; \ 100 ap2[_i] = value; \ 101 b_noinsert: ; \ 102 bilen[row] = nrow2; \ 103 } 104 105 #undef __FUNCT__ 106 #define __FUNCT__ "MatSetValuesRow_MPIAIJ" 107 PetscErrorCode MatSetValuesRow_MPIAIJ(Mat A,PetscInt row,const PetscScalar v[]) 108 { 109 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 110 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->A->data,*b = (Mat_SeqAIJ*)mat->B->data; 111 PetscErrorCode ierr; 112 PetscInt l,*garray = mat->garray,diag; 113 114 PetscFunctionBegin; 115 /* code only works for square matrices A */ 116 117 /* find size of row to the left of the diagonal part */ 118 ierr = MatGetOwnershipRange(A,&diag,0);CHKERRQ(ierr); 119 row = row - diag; 120 for (l=0; l<b->i[row+1]-b->i[row]; l++) { 121 if (garray[b->j[b->i[row]+l]] > diag) break; 122 } 123 ierr = PetscMemcpy(b->a+b->i[row],v,l*sizeof(PetscScalar));CHKERRQ(ierr); 124 125 /* diagonal part */ 126 ierr = PetscMemcpy(a->a+a->i[row],v+l,(a->i[row+1]-a->i[row])*sizeof(PetscScalar));CHKERRQ(ierr); 127 128 /* right of diagonal part */ 129 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); 130 PetscFunctionReturn(0); 131 } 132 133 #undef __FUNCT__ 134 #define __FUNCT__ "MatSetValues_MPIAIJ" 135 PetscErrorCode MatSetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv) 136 { 137 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 138 PetscScalar value; 139 PetscErrorCode ierr; 140 PetscInt i,j,rstart = mat->rmap.rstart,rend = mat->rmap.rend; 141 PetscInt cstart = mat->cmap.rstart,cend = mat->cmap.rend,row,col; 142 PetscTruth roworiented = aij->roworiented; 143 144 /* Some Variables required in the macro */ 145 Mat A = aij->A; 146 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 147 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 148 PetscScalar *aa = a->a; 149 PetscTruth ignorezeroentries = a->ignorezeroentries; 150 Mat B = aij->B; 151 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 152 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap.n,am = aij->A->rmap.n; 153 PetscScalar *ba = b->a; 154 155 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 156 PetscInt nonew = a->nonew; 157 PetscScalar *ap1,*ap2; 158 159 PetscFunctionBegin; 160 for (i=0; i<m; i++) { 161 if (im[i] < 0) continue; 162 #if defined(PETSC_USE_DEBUG) 163 if (im[i] >= mat->rmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap.N-1); 164 #endif 165 if (im[i] >= rstart && im[i] < rend) { 166 row = im[i] - rstart; 167 lastcol1 = -1; 168 rp1 = aj + ai[row]; 169 ap1 = aa + ai[row]; 170 rmax1 = aimax[row]; 171 nrow1 = ailen[row]; 172 low1 = 0; 173 high1 = nrow1; 174 lastcol2 = -1; 175 rp2 = bj + bi[row]; 176 ap2 = ba + bi[row]; 177 rmax2 = bimax[row]; 178 nrow2 = bilen[row]; 179 low2 = 0; 180 high2 = nrow2; 181 182 for (j=0; j<n; j++) { 183 if (roworiented) value = v[i*n+j]; else value = v[i+j*m]; 184 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES)) continue; 185 if (in[j] >= cstart && in[j] < cend){ 186 col = in[j] - cstart; 187 MatSetValues_SeqAIJ_A_Private(row,col,value,addv); 188 } else if (in[j] < 0) continue; 189 #if defined(PETSC_USE_DEBUG) 190 else if (in[j] >= mat->cmap.N) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap.N-1);} 191 #endif 192 else { 193 if (mat->was_assembled) { 194 if (!aij->colmap) { 195 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 196 } 197 #if defined (PETSC_USE_CTABLE) 198 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 199 col--; 200 #else 201 col = aij->colmap[in[j]] - 1; 202 #endif 203 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 204 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 205 col = in[j]; 206 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 207 B = aij->B; 208 b = (Mat_SeqAIJ*)B->data; 209 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; 210 rp2 = bj + bi[row]; 211 ap2 = ba + bi[row]; 212 rmax2 = bimax[row]; 213 nrow2 = bilen[row]; 214 low2 = 0; 215 high2 = nrow2; 216 bm = aij->B->rmap.n; 217 ba = b->a; 218 } 219 } else col = in[j]; 220 MatSetValues_SeqAIJ_B_Private(row,col,value,addv); 221 } 222 } 223 } else { 224 if (!aij->donotstash) { 225 if (roworiented) { 226 if (ignorezeroentries && v[i*n] == 0.0) continue; 227 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n);CHKERRQ(ierr); 228 } else { 229 if (ignorezeroentries && v[i] == 0.0) continue; 230 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m);CHKERRQ(ierr); 231 } 232 } 233 } 234 } 235 PetscFunctionReturn(0); 236 } 237 238 #undef __FUNCT__ 239 #define __FUNCT__ "MatGetValues_MPIAIJ" 240 PetscErrorCode MatGetValues_MPIAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[]) 241 { 242 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 243 PetscErrorCode ierr; 244 PetscInt i,j,rstart = mat->rmap.rstart,rend = mat->rmap.rend; 245 PetscInt cstart = mat->cmap.rstart,cend = mat->cmap.rend,row,col; 246 247 PetscFunctionBegin; 248 for (i=0; i<m; i++) { 249 if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]);*/ 250 if (idxm[i] >= mat->rmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",idxm[i],mat->rmap.N-1); 251 if (idxm[i] >= rstart && idxm[i] < rend) { 252 row = idxm[i] - rstart; 253 for (j=0; j<n; j++) { 254 if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",idxn[j]); */ 255 if (idxn[j] >= mat->cmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",idxn[j],mat->cmap.N-1); 256 if (idxn[j] >= cstart && idxn[j] < cend){ 257 col = idxn[j] - cstart; 258 ierr = MatGetValues(aij->A,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 259 } else { 260 if (!aij->colmap) { 261 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 262 } 263 #if defined (PETSC_USE_CTABLE) 264 ierr = PetscTableFind(aij->colmap,idxn[j]+1,&col);CHKERRQ(ierr); 265 col --; 266 #else 267 col = aij->colmap[idxn[j]] - 1; 268 #endif 269 if ((col < 0) || (aij->garray[col] != idxn[j])) *(v+i*n+j) = 0.0; 270 else { 271 ierr = MatGetValues(aij->B,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr); 272 } 273 } 274 } 275 } else { 276 SETERRQ(PETSC_ERR_SUP,"Only local values currently supported"); 277 } 278 } 279 PetscFunctionReturn(0); 280 } 281 282 #undef __FUNCT__ 283 #define __FUNCT__ "MatAssemblyBegin_MPIAIJ" 284 PetscErrorCode MatAssemblyBegin_MPIAIJ(Mat mat,MatAssemblyType mode) 285 { 286 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 287 PetscErrorCode ierr; 288 PetscInt nstash,reallocs; 289 InsertMode addv; 290 291 PetscFunctionBegin; 292 if (aij->donotstash) { 293 PetscFunctionReturn(0); 294 } 295 296 /* make sure all processors are either in INSERTMODE or ADDMODE */ 297 ierr = MPI_Allreduce(&mat->insertmode,&addv,1,MPI_INT,MPI_BOR,mat->comm);CHKERRQ(ierr); 298 if (addv == (ADD_VALUES|INSERT_VALUES)) { 299 SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Some processors inserted others added"); 300 } 301 mat->insertmode = addv; /* in case this processor had no cache */ 302 303 ierr = MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap.range);CHKERRQ(ierr); 304 ierr = MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);CHKERRQ(ierr); 305 ierr = PetscInfo2(aij->A,"Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);CHKERRQ(ierr); 306 PetscFunctionReturn(0); 307 } 308 309 #undef __FUNCT__ 310 #define __FUNCT__ "MatAssemblyEnd_MPIAIJ" 311 PetscErrorCode MatAssemblyEnd_MPIAIJ(Mat mat,MatAssemblyType mode) 312 { 313 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 314 Mat_SeqAIJ *a=(Mat_SeqAIJ *)aij->A->data; 315 PetscErrorCode ierr; 316 PetscMPIInt n; 317 PetscInt i,j,rstart,ncols,flg; 318 PetscInt *row,*col,other_disassembled; 319 PetscScalar *val; 320 InsertMode addv = mat->insertmode; 321 322 /* do not use 'b = (Mat_SeqAIJ *)aij->B->data' as B can be reset in disassembly */ 323 PetscFunctionBegin; 324 if (!aij->donotstash) { 325 while (1) { 326 ierr = MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);CHKERRQ(ierr); 327 if (!flg) break; 328 329 for (i=0; i<n;) { 330 /* Now identify the consecutive vals belonging to the same row */ 331 for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; } 332 if (j < n) ncols = j-i; 333 else ncols = n-i; 334 /* Now assemble all these values with a single function call */ 335 ierr = MatSetValues_MPIAIJ(mat,1,row+i,ncols,col+i,val+i,addv);CHKERRQ(ierr); 336 i = j; 337 } 338 } 339 ierr = MatStashScatterEnd_Private(&mat->stash);CHKERRQ(ierr); 340 } 341 a->compressedrow.use = PETSC_FALSE; 342 ierr = MatAssemblyBegin(aij->A,mode);CHKERRQ(ierr); 343 ierr = MatAssemblyEnd(aij->A,mode);CHKERRQ(ierr); 344 345 /* determine if any processor has disassembled, if so we must 346 also disassemble ourselfs, in order that we may reassemble. */ 347 /* 348 if nonzero structure of submatrix B cannot change then we know that 349 no processor disassembled thus we can skip this stuff 350 */ 351 if (!((Mat_SeqAIJ*)aij->B->data)->nonew) { 352 ierr = MPI_Allreduce(&mat->was_assembled,&other_disassembled,1,MPI_INT,MPI_PROD,mat->comm);CHKERRQ(ierr); 353 if (mat->was_assembled && !other_disassembled) { 354 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 355 } 356 } 357 if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) { 358 ierr = MatSetUpMultiply_MPIAIJ(mat);CHKERRQ(ierr); 359 } 360 ierr = MatSetOption(aij->B,MAT_USE_INODES,PETSC_FALSE);CHKERRQ(ierr); 361 ((Mat_SeqAIJ *)aij->B->data)->compressedrow.use = PETSC_TRUE; /* b->compressedrow.use */ 362 ierr = MatAssemblyBegin(aij->B,mode);CHKERRQ(ierr); 363 ierr = MatAssemblyEnd(aij->B,mode);CHKERRQ(ierr); 364 365 ierr = PetscFree(aij->rowvalues);CHKERRQ(ierr); 366 aij->rowvalues = 0; 367 368 /* used by MatAXPY() */ 369 a->xtoy = 0; ((Mat_SeqAIJ *)aij->B->data)->xtoy = 0; /* b->xtoy = 0 */ 370 a->XtoY = 0; ((Mat_SeqAIJ *)aij->B->data)->XtoY = 0; /* b->XtoY = 0 */ 371 372 PetscFunctionReturn(0); 373 } 374 375 #undef __FUNCT__ 376 #define __FUNCT__ "MatZeroEntries_MPIAIJ" 377 PetscErrorCode MatZeroEntries_MPIAIJ(Mat A) 378 { 379 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 380 PetscErrorCode ierr; 381 382 PetscFunctionBegin; 383 ierr = MatZeroEntries(l->A);CHKERRQ(ierr); 384 ierr = MatZeroEntries(l->B);CHKERRQ(ierr); 385 PetscFunctionReturn(0); 386 } 387 388 #undef __FUNCT__ 389 #define __FUNCT__ "MatZeroRows_MPIAIJ" 390 PetscErrorCode MatZeroRows_MPIAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag) 391 { 392 Mat_MPIAIJ *l = (Mat_MPIAIJ*)A->data; 393 PetscErrorCode ierr; 394 PetscMPIInt size = l->size,imdex,n,rank = l->rank,tag = A->tag,lastidx = -1; 395 PetscInt i,*owners = A->rmap.range; 396 PetscInt *nprocs,j,idx,nsends,row; 397 PetscInt nmax,*svalues,*starts,*owner,nrecvs; 398 PetscInt *rvalues,count,base,slen,*source; 399 PetscInt *lens,*lrows,*values,rstart=A->rmap.rstart; 400 MPI_Comm comm = A->comm; 401 MPI_Request *send_waits,*recv_waits; 402 MPI_Status recv_status,*send_status; 403 #if defined(PETSC_DEBUG) 404 PetscTruth found = PETSC_FALSE; 405 #endif 406 407 PetscFunctionBegin; 408 /* first count number of contributors to each processor */ 409 ierr = PetscMalloc(2*size*sizeof(PetscInt),&nprocs);CHKERRQ(ierr); 410 ierr = PetscMemzero(nprocs,2*size*sizeof(PetscInt));CHKERRQ(ierr); 411 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&owner);CHKERRQ(ierr); /* see note*/ 412 j = 0; 413 for (i=0; i<N; i++) { 414 if (lastidx > (idx = rows[i])) j = 0; 415 lastidx = idx; 416 for (; j<size; j++) { 417 if (idx >= owners[j] && idx < owners[j+1]) { 418 nprocs[2*j]++; 419 nprocs[2*j+1] = 1; 420 owner[i] = j; 421 #if defined(PETSC_DEBUG) 422 found = PETSC_TRUE; 423 #endif 424 break; 425 } 426 } 427 #if defined(PETSC_DEBUG) 428 if (!found) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Index out of range"); 429 found = PETSC_FALSE; 430 #endif 431 } 432 nsends = 0; for (i=0; i<size; i++) { nsends += nprocs[2*i+1];} 433 434 /* inform other processors of number of messages and max length*/ 435 ierr = PetscMaxSum(comm,nprocs,&nmax,&nrecvs);CHKERRQ(ierr); 436 437 /* post receives: */ 438 ierr = PetscMalloc((nrecvs+1)*(nmax+1)*sizeof(PetscInt),&rvalues);CHKERRQ(ierr); 439 ierr = PetscMalloc((nrecvs+1)*sizeof(MPI_Request),&recv_waits);CHKERRQ(ierr); 440 for (i=0; i<nrecvs; i++) { 441 ierr = MPI_Irecv(rvalues+nmax*i,nmax,MPIU_INT,MPI_ANY_SOURCE,tag,comm,recv_waits+i);CHKERRQ(ierr); 442 } 443 444 /* do sends: 445 1) starts[i] gives the starting index in svalues for stuff going to 446 the ith processor 447 */ 448 ierr = PetscMalloc((N+1)*sizeof(PetscInt),&svalues);CHKERRQ(ierr); 449 ierr = PetscMalloc((nsends+1)*sizeof(MPI_Request),&send_waits);CHKERRQ(ierr); 450 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&starts);CHKERRQ(ierr); 451 starts[0] = 0; 452 for (i=1; i<size; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 453 for (i=0; i<N; i++) { 454 svalues[starts[owner[i]]++] = rows[i]; 455 } 456 457 starts[0] = 0; 458 for (i=1; i<size+1; i++) { starts[i] = starts[i-1] + nprocs[2*i-2];} 459 count = 0; 460 for (i=0; i<size; i++) { 461 if (nprocs[2*i+1]) { 462 ierr = MPI_Isend(svalues+starts[i],nprocs[2*i],MPIU_INT,i,tag,comm,send_waits+count++);CHKERRQ(ierr); 463 } 464 } 465 ierr = PetscFree(starts);CHKERRQ(ierr); 466 467 base = owners[rank]; 468 469 /* wait on receives */ 470 ierr = PetscMalloc(2*(nrecvs+1)*sizeof(PetscInt),&lens);CHKERRQ(ierr); 471 source = lens + nrecvs; 472 count = nrecvs; slen = 0; 473 while (count) { 474 ierr = MPI_Waitany(nrecvs,recv_waits,&imdex,&recv_status);CHKERRQ(ierr); 475 /* unpack receives into our local space */ 476 ierr = MPI_Get_count(&recv_status,MPIU_INT,&n);CHKERRQ(ierr); 477 source[imdex] = recv_status.MPI_SOURCE; 478 lens[imdex] = n; 479 slen += n; 480 count--; 481 } 482 ierr = PetscFree(recv_waits);CHKERRQ(ierr); 483 484 /* move the data into the send scatter */ 485 ierr = PetscMalloc((slen+1)*sizeof(PetscInt),&lrows);CHKERRQ(ierr); 486 count = 0; 487 for (i=0; i<nrecvs; i++) { 488 values = rvalues + i*nmax; 489 for (j=0; j<lens[i]; j++) { 490 lrows[count++] = values[j] - base; 491 } 492 } 493 ierr = PetscFree(rvalues);CHKERRQ(ierr); 494 ierr = PetscFree(lens);CHKERRQ(ierr); 495 ierr = PetscFree(owner);CHKERRQ(ierr); 496 ierr = PetscFree(nprocs);CHKERRQ(ierr); 497 498 /* actually zap the local rows */ 499 /* 500 Zero the required rows. If the "diagonal block" of the matrix 501 is square and the user wishes to set the diagonal we use separate 502 code so that MatSetValues() is not called for each diagonal allocating 503 new memory, thus calling lots of mallocs and slowing things down. 504 505 Contributed by: Matthew Knepley 506 */ 507 /* must zero l->B before l->A because the (diag) case below may put values into l->B*/ 508 ierr = MatZeroRows(l->B,slen,lrows,0.0);CHKERRQ(ierr); 509 if ((diag != 0.0) && (l->A->rmap.N == l->A->cmap.N)) { 510 ierr = MatZeroRows(l->A,slen,lrows,diag);CHKERRQ(ierr); 511 } else if (diag != 0.0) { 512 ierr = MatZeroRows(l->A,slen,lrows,0.0);CHKERRQ(ierr); 513 if (((Mat_SeqAIJ*)l->A->data)->nonew) { 514 SETERRQ(PETSC_ERR_SUP,"MatZeroRows() on rectangular matrices cannot be used with the Mat options\n\ 515 MAT_NEW_NONZERO_LOCATIONS,MAT_NEW_NONZERO_LOCATION_ERR,MAT_NEW_NONZERO_ALLOCATION_ERR"); 516 } 517 for (i = 0; i < slen; i++) { 518 row = lrows[i] + rstart; 519 ierr = MatSetValues(A,1,&row,1,&row,&diag,INSERT_VALUES);CHKERRQ(ierr); 520 } 521 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 522 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 523 } else { 524 ierr = MatZeroRows(l->A,slen,lrows,0.0);CHKERRQ(ierr); 525 } 526 ierr = PetscFree(lrows);CHKERRQ(ierr); 527 528 /* wait on sends */ 529 if (nsends) { 530 ierr = PetscMalloc(nsends*sizeof(MPI_Status),&send_status);CHKERRQ(ierr); 531 ierr = MPI_Waitall(nsends,send_waits,send_status);CHKERRQ(ierr); 532 ierr = PetscFree(send_status);CHKERRQ(ierr); 533 } 534 ierr = PetscFree(send_waits);CHKERRQ(ierr); 535 ierr = PetscFree(svalues);CHKERRQ(ierr); 536 537 PetscFunctionReturn(0); 538 } 539 540 #undef __FUNCT__ 541 #define __FUNCT__ "MatMult_MPIAIJ" 542 PetscErrorCode MatMult_MPIAIJ(Mat A,Vec xx,Vec yy) 543 { 544 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 545 PetscErrorCode ierr; 546 PetscInt nt; 547 548 PetscFunctionBegin; 549 ierr = VecGetLocalSize(xx,&nt);CHKERRQ(ierr); 550 if (nt != A->cmap.n) { 551 SETERRQ2(PETSC_ERR_ARG_SIZ,"Incompatible partition of A (%D) and xx (%D)",A->cmap.n,nt); 552 } 553 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 554 ierr = (*a->A->ops->mult)(a->A,xx,yy);CHKERRQ(ierr); 555 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 556 ierr = (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);CHKERRQ(ierr); 557 PetscFunctionReturn(0); 558 } 559 560 #undef __FUNCT__ 561 #define __FUNCT__ "MatMultAdd_MPIAIJ" 562 PetscErrorCode MatMultAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 563 { 564 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 565 PetscErrorCode ierr; 566 567 PetscFunctionBegin; 568 ierr = VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 569 ierr = (*a->A->ops->multadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 570 ierr = VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 571 ierr = (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);CHKERRQ(ierr); 572 PetscFunctionReturn(0); 573 } 574 575 #undef __FUNCT__ 576 #define __FUNCT__ "MatMultTranspose_MPIAIJ" 577 PetscErrorCode MatMultTranspose_MPIAIJ(Mat A,Vec xx,Vec yy) 578 { 579 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 580 PetscErrorCode ierr; 581 PetscTruth merged; 582 583 PetscFunctionBegin; 584 ierr = VecScatterGetMerged(a->Mvctx,&merged);CHKERRQ(ierr); 585 /* do nondiagonal part */ 586 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 587 if (!merged) { 588 /* send it on its way */ 589 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 590 /* do local part */ 591 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 592 /* receive remote parts: note this assumes the values are not actually */ 593 /* added in yy until the next line, */ 594 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 595 } else { 596 /* do local part */ 597 ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr); 598 /* send it on its way */ 599 ierr = VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 600 /* values actually were received in the Begin() but we need to call this nop */ 601 ierr = VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 602 } 603 PetscFunctionReturn(0); 604 } 605 606 EXTERN_C_BEGIN 607 #undef __FUNCT__ 608 #define __FUNCT__ "MatIsTranspose_MPIAIJ" 609 PetscErrorCode PETSCMAT_DLLEXPORT MatIsTranspose_MPIAIJ(Mat Amat,Mat Bmat,PetscReal tol,PetscTruth *f) 610 { 611 MPI_Comm comm; 612 Mat_MPIAIJ *Aij = (Mat_MPIAIJ *) Amat->data, *Bij; 613 Mat Adia = Aij->A, Bdia, Aoff,Boff,*Aoffs,*Boffs; 614 IS Me,Notme; 615 PetscErrorCode ierr; 616 PetscInt M,N,first,last,*notme,i; 617 PetscMPIInt size; 618 619 PetscFunctionBegin; 620 621 /* Easy test: symmetric diagonal block */ 622 Bij = (Mat_MPIAIJ *) Bmat->data; Bdia = Bij->A; 623 ierr = MatIsTranspose(Adia,Bdia,tol,f);CHKERRQ(ierr); 624 if (!*f) PetscFunctionReturn(0); 625 ierr = PetscObjectGetComm((PetscObject)Amat,&comm);CHKERRQ(ierr); 626 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 627 if (size == 1) PetscFunctionReturn(0); 628 629 /* Hard test: off-diagonal block. This takes a MatGetSubMatrix. */ 630 ierr = MatGetSize(Amat,&M,&N);CHKERRQ(ierr); 631 ierr = MatGetOwnershipRange(Amat,&first,&last);CHKERRQ(ierr); 632 ierr = PetscMalloc((N-last+first)*sizeof(PetscInt),¬me);CHKERRQ(ierr); 633 for (i=0; i<first; i++) notme[i] = i; 634 for (i=last; i<M; i++) notme[i-last+first] = i; 635 ierr = ISCreateGeneral(MPI_COMM_SELF,N-last+first,notme,&Notme);CHKERRQ(ierr); 636 ierr = ISCreateStride(MPI_COMM_SELF,last-first,first,1,&Me);CHKERRQ(ierr); 637 ierr = MatGetSubMatrices(Amat,1,&Me,&Notme,MAT_INITIAL_MATRIX,&Aoffs);CHKERRQ(ierr); 638 Aoff = Aoffs[0]; 639 ierr = MatGetSubMatrices(Bmat,1,&Notme,&Me,MAT_INITIAL_MATRIX,&Boffs);CHKERRQ(ierr); 640 Boff = Boffs[0]; 641 ierr = MatIsTranspose(Aoff,Boff,tol,f);CHKERRQ(ierr); 642 ierr = MatDestroyMatrices(1,&Aoffs);CHKERRQ(ierr); 643 ierr = MatDestroyMatrices(1,&Boffs);CHKERRQ(ierr); 644 ierr = ISDestroy(Me);CHKERRQ(ierr); 645 ierr = ISDestroy(Notme);CHKERRQ(ierr); 646 647 PetscFunctionReturn(0); 648 } 649 EXTERN_C_END 650 651 #undef __FUNCT__ 652 #define __FUNCT__ "MatMultTransposeAdd_MPIAIJ" 653 PetscErrorCode MatMultTransposeAdd_MPIAIJ(Mat A,Vec xx,Vec yy,Vec zz) 654 { 655 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 656 PetscErrorCode ierr; 657 658 PetscFunctionBegin; 659 /* do nondiagonal part */ 660 ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr); 661 /* send it on its way */ 662 ierr = VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 663 /* do local part */ 664 ierr = (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);CHKERRQ(ierr); 665 /* receive remote parts */ 666 ierr = VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 667 PetscFunctionReturn(0); 668 } 669 670 /* 671 This only works correctly for square matrices where the subblock A->A is the 672 diagonal block 673 */ 674 #undef __FUNCT__ 675 #define __FUNCT__ "MatGetDiagonal_MPIAIJ" 676 PetscErrorCode MatGetDiagonal_MPIAIJ(Mat A,Vec v) 677 { 678 PetscErrorCode ierr; 679 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 680 681 PetscFunctionBegin; 682 if (A->rmap.N != A->cmap.N) SETERRQ(PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); 683 if (A->rmap.rstart != A->cmap.rstart || A->rmap.rend != A->cmap.rend) { 684 SETERRQ(PETSC_ERR_ARG_SIZ,"row partition must equal col partition"); 685 } 686 ierr = MatGetDiagonal(a->A,v);CHKERRQ(ierr); 687 PetscFunctionReturn(0); 688 } 689 690 #undef __FUNCT__ 691 #define __FUNCT__ "MatScale_MPIAIJ" 692 PetscErrorCode MatScale_MPIAIJ(Mat A,PetscScalar aa) 693 { 694 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 695 PetscErrorCode ierr; 696 697 PetscFunctionBegin; 698 ierr = MatScale(a->A,aa);CHKERRQ(ierr); 699 ierr = MatScale(a->B,aa);CHKERRQ(ierr); 700 PetscFunctionReturn(0); 701 } 702 703 #undef __FUNCT__ 704 #define __FUNCT__ "MatDestroy_MPIAIJ" 705 PetscErrorCode MatDestroy_MPIAIJ(Mat mat) 706 { 707 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 708 PetscErrorCode ierr; 709 710 PetscFunctionBegin; 711 #if defined(PETSC_USE_LOG) 712 PetscLogObjectState((PetscObject)mat,"Rows=%D, Cols=%D",mat->rmap.N,mat->cmap.N); 713 #endif 714 ierr = MatStashDestroy_Private(&mat->stash);CHKERRQ(ierr); 715 ierr = MatDestroy(aij->A);CHKERRQ(ierr); 716 ierr = MatDestroy(aij->B);CHKERRQ(ierr); 717 #if defined (PETSC_USE_CTABLE) 718 if (aij->colmap) {ierr = PetscTableDestroy(aij->colmap);CHKERRQ(ierr);} 719 #else 720 ierr = PetscFree(aij->colmap);CHKERRQ(ierr); 721 #endif 722 ierr = PetscFree(aij->garray);CHKERRQ(ierr); 723 if (aij->lvec) {ierr = VecDestroy(aij->lvec);CHKERRQ(ierr);} 724 if (aij->Mvctx) {ierr = VecScatterDestroy(aij->Mvctx);CHKERRQ(ierr);} 725 ierr = PetscFree(aij->rowvalues);CHKERRQ(ierr); 726 ierr = PetscFree(aij->ld);CHKERRQ(ierr); 727 ierr = PetscFree(aij);CHKERRQ(ierr); 728 729 ierr = PetscObjectChangeTypeName((PetscObject)mat,0);CHKERRQ(ierr); 730 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C","",PETSC_NULL);CHKERRQ(ierr); 731 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C","",PETSC_NULL);CHKERRQ(ierr); 732 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatGetDiagonalBlock_C","",PETSC_NULL);CHKERRQ(ierr); 733 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatIsTranspose_C","",PETSC_NULL);CHKERRQ(ierr); 734 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocation_C","",PETSC_NULL);CHKERRQ(ierr); 735 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatMPIAIJSetPreallocationCSR_C","",PETSC_NULL);CHKERRQ(ierr); 736 ierr = PetscObjectComposeFunction((PetscObject)mat,"MatDiagonalScaleLocal_C","",PETSC_NULL);CHKERRQ(ierr); 737 PetscFunctionReturn(0); 738 } 739 740 #undef __FUNCT__ 741 #define __FUNCT__ "MatView_MPIAIJ_Binary" 742 PetscErrorCode MatView_MPIAIJ_Binary(Mat mat,PetscViewer viewer) 743 { 744 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 745 Mat_SeqAIJ* A = (Mat_SeqAIJ*)aij->A->data; 746 Mat_SeqAIJ* B = (Mat_SeqAIJ*)aij->B->data; 747 PetscErrorCode ierr; 748 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag; 749 int fd; 750 PetscInt nz,header[4],*row_lengths,*range=0,rlen,i; 751 PetscInt nzmax,*column_indices,j,k,col,*garray = aij->garray,cnt,cstart = mat->cmap.rstart,rnz; 752 PetscScalar *column_values; 753 754 PetscFunctionBegin; 755 ierr = MPI_Comm_rank(mat->comm,&rank);CHKERRQ(ierr); 756 ierr = MPI_Comm_size(mat->comm,&size);CHKERRQ(ierr); 757 nz = A->nz + B->nz; 758 if (!rank) { 759 header[0] = MAT_FILE_COOKIE; 760 header[1] = mat->rmap.N; 761 header[2] = mat->cmap.N; 762 ierr = MPI_Reduce(&nz,&header[3],1,MPIU_INT,MPI_SUM,0,mat->comm);CHKERRQ(ierr); 763 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 764 ierr = PetscBinaryWrite(fd,header,4,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 765 /* get largest number of rows any processor has */ 766 rlen = mat->rmap.n; 767 range = mat->rmap.range; 768 for (i=1; i<size; i++) { 769 rlen = PetscMax(rlen,range[i+1] - range[i]); 770 } 771 } else { 772 ierr = MPI_Reduce(&nz,0,1,MPIU_INT,MPI_SUM,0,mat->comm);CHKERRQ(ierr); 773 rlen = mat->rmap.n; 774 } 775 776 /* load up the local row counts */ 777 ierr = PetscMalloc((rlen+1)*sizeof(PetscInt),&row_lengths);CHKERRQ(ierr); 778 for (i=0; i<mat->rmap.n; i++) { 779 row_lengths[i] = A->i[i+1] - A->i[i] + B->i[i+1] - B->i[i]; 780 } 781 782 /* store the row lengths to the file */ 783 if (!rank) { 784 MPI_Status status; 785 ierr = PetscBinaryWrite(fd,row_lengths,mat->rmap.n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 786 for (i=1; i<size; i++) { 787 rlen = range[i+1] - range[i]; 788 ierr = MPI_Recv(row_lengths,rlen,MPIU_INT,i,tag,mat->comm,&status);CHKERRQ(ierr); 789 ierr = PetscBinaryWrite(fd,row_lengths,rlen,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 790 } 791 } else { 792 ierr = MPI_Send(row_lengths,mat->rmap.n,MPIU_INT,0,tag,mat->comm);CHKERRQ(ierr); 793 } 794 ierr = PetscFree(row_lengths);CHKERRQ(ierr); 795 796 /* load up the local column indices */ 797 nzmax = nz; /* )th processor needs space a largest processor needs */ 798 ierr = MPI_Reduce(&nz,&nzmax,1,MPIU_INT,MPI_MAX,0,mat->comm);CHKERRQ(ierr); 799 ierr = PetscMalloc((nzmax+1)*sizeof(PetscInt),&column_indices);CHKERRQ(ierr); 800 cnt = 0; 801 for (i=0; i<mat->rmap.n; i++) { 802 for (j=B->i[i]; j<B->i[i+1]; j++) { 803 if ( (col = garray[B->j[j]]) > cstart) break; 804 column_indices[cnt++] = col; 805 } 806 for (k=A->i[i]; k<A->i[i+1]; k++) { 807 column_indices[cnt++] = A->j[k] + cstart; 808 } 809 for (; j<B->i[i+1]; j++) { 810 column_indices[cnt++] = garray[B->j[j]]; 811 } 812 } 813 if (cnt != A->nz + B->nz) SETERRQ2(PETSC_ERR_LIB,"Internal PETSc error: cnt = %D nz = %D",cnt,A->nz+B->nz); 814 815 /* store the column indices to the file */ 816 if (!rank) { 817 MPI_Status status; 818 ierr = PetscBinaryWrite(fd,column_indices,nz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 819 for (i=1; i<size; i++) { 820 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,mat->comm,&status);CHKERRQ(ierr); 821 if (rnz > nzmax) SETERRQ2(PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 822 ierr = MPI_Recv(column_indices,rnz,MPIU_INT,i,tag,mat->comm,&status);CHKERRQ(ierr); 823 ierr = PetscBinaryWrite(fd,column_indices,rnz,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 824 } 825 } else { 826 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,mat->comm);CHKERRQ(ierr); 827 ierr = MPI_Send(column_indices,nz,MPIU_INT,0,tag,mat->comm);CHKERRQ(ierr); 828 } 829 ierr = PetscFree(column_indices);CHKERRQ(ierr); 830 831 /* load up the local column values */ 832 ierr = PetscMalloc((nzmax+1)*sizeof(PetscScalar),&column_values);CHKERRQ(ierr); 833 cnt = 0; 834 for (i=0; i<mat->rmap.n; i++) { 835 for (j=B->i[i]; j<B->i[i+1]; j++) { 836 if ( garray[B->j[j]] > cstart) break; 837 column_values[cnt++] = B->a[j]; 838 } 839 for (k=A->i[i]; k<A->i[i+1]; k++) { 840 column_values[cnt++] = A->a[k]; 841 } 842 for (; j<B->i[i+1]; j++) { 843 column_values[cnt++] = B->a[j]; 844 } 845 } 846 if (cnt != A->nz + B->nz) SETERRQ2(PETSC_ERR_PLIB,"Internal PETSc error: cnt = %D nz = %D",cnt,A->nz+B->nz); 847 848 /* store the column values to the file */ 849 if (!rank) { 850 MPI_Status status; 851 ierr = PetscBinaryWrite(fd,column_values,nz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 852 for (i=1; i<size; i++) { 853 ierr = MPI_Recv(&rnz,1,MPIU_INT,i,tag,mat->comm,&status);CHKERRQ(ierr); 854 if (rnz > nzmax) SETERRQ2(PETSC_ERR_LIB,"Internal PETSc error: nz = %D nzmax = %D",nz,nzmax); 855 ierr = MPI_Recv(column_values,rnz,MPIU_SCALAR,i,tag,mat->comm,&status);CHKERRQ(ierr); 856 ierr = PetscBinaryWrite(fd,column_values,rnz,PETSC_SCALAR,PETSC_TRUE);CHKERRQ(ierr); 857 } 858 } else { 859 ierr = MPI_Send(&nz,1,MPIU_INT,0,tag,mat->comm);CHKERRQ(ierr); 860 ierr = MPI_Send(column_values,nz,MPIU_SCALAR,0,tag,mat->comm);CHKERRQ(ierr); 861 } 862 ierr = PetscFree(column_values);CHKERRQ(ierr); 863 PetscFunctionReturn(0); 864 } 865 866 #undef __FUNCT__ 867 #define __FUNCT__ "MatView_MPIAIJ_ASCIIorDraworSocket" 868 PetscErrorCode MatView_MPIAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer) 869 { 870 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 871 PetscErrorCode ierr; 872 PetscMPIInt rank = aij->rank,size = aij->size; 873 PetscTruth isdraw,iascii,isbinary; 874 PetscViewer sviewer; 875 PetscViewerFormat format; 876 877 PetscFunctionBegin; 878 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);CHKERRQ(ierr); 879 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);CHKERRQ(ierr); 880 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);CHKERRQ(ierr); 881 if (iascii) { 882 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 883 if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) { 884 MatInfo info; 885 PetscTruth inodes; 886 887 ierr = MPI_Comm_rank(mat->comm,&rank);CHKERRQ(ierr); 888 ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr); 889 ierr = MatInodeGetInodeSizes(aij->A,PETSC_NULL,(PetscInt **)&inodes,PETSC_NULL);CHKERRQ(ierr); 890 if (!inodes) { 891 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, not using I-node routines\n", 892 rank,mat->rmap.n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 893 } else { 894 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D mem %D, using I-node routines\n", 895 rank,mat->rmap.n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,(PetscInt)info.memory);CHKERRQ(ierr); 896 } 897 ierr = MatGetInfo(aij->A,MAT_LOCAL,&info);CHKERRQ(ierr); 898 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 899 ierr = MatGetInfo(aij->B,MAT_LOCAL,&info);CHKERRQ(ierr); 900 ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);CHKERRQ(ierr); 901 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 902 ierr = PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");CHKERRQ(ierr); 903 ierr = VecScatterView(aij->Mvctx,viewer);CHKERRQ(ierr); 904 PetscFunctionReturn(0); 905 } else if (format == PETSC_VIEWER_ASCII_INFO) { 906 PetscInt inodecount,inodelimit,*inodes; 907 ierr = MatInodeGetInodeSizes(aij->A,&inodecount,&inodes,&inodelimit);CHKERRQ(ierr); 908 if (inodes) { 909 ierr = PetscViewerASCIIPrintf(viewer,"using I-node (on process 0) routines: found %D nodes, limit used is %D\n",inodecount,inodelimit);CHKERRQ(ierr); 910 } else { 911 ierr = PetscViewerASCIIPrintf(viewer,"not using I-node (on process 0) routines\n");CHKERRQ(ierr); 912 } 913 PetscFunctionReturn(0); 914 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) { 915 PetscFunctionReturn(0); 916 } 917 } else if (isbinary) { 918 if (size == 1) { 919 ierr = PetscObjectSetName((PetscObject)aij->A,mat->name);CHKERRQ(ierr); 920 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 921 } else { 922 ierr = MatView_MPIAIJ_Binary(mat,viewer);CHKERRQ(ierr); 923 } 924 PetscFunctionReturn(0); 925 } else if (isdraw) { 926 PetscDraw draw; 927 PetscTruth isnull; 928 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 929 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); if (isnull) PetscFunctionReturn(0); 930 } 931 932 if (size == 1) { 933 ierr = PetscObjectSetName((PetscObject)aij->A,mat->name);CHKERRQ(ierr); 934 ierr = MatView(aij->A,viewer);CHKERRQ(ierr); 935 } else { 936 /* assemble the entire matrix onto first processor. */ 937 Mat A; 938 Mat_SeqAIJ *Aloc; 939 PetscInt M = mat->rmap.N,N = mat->cmap.N,m,*ai,*aj,row,*cols,i,*ct; 940 PetscScalar *a; 941 942 ierr = MatCreate(mat->comm,&A);CHKERRQ(ierr); 943 if (!rank) { 944 ierr = MatSetSizes(A,M,N,M,N);CHKERRQ(ierr); 945 } else { 946 ierr = MatSetSizes(A,0,0,M,N);CHKERRQ(ierr); 947 } 948 /* This is just a temporary matrix, so explicitly using MATMPIAIJ is probably best */ 949 ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr); 950 ierr = MatMPIAIJSetPreallocation(A,0,PETSC_NULL,0,PETSC_NULL);CHKERRQ(ierr); 951 ierr = PetscLogObjectParent(mat,A);CHKERRQ(ierr); 952 953 /* copy over the A part */ 954 Aloc = (Mat_SeqAIJ*)aij->A->data; 955 m = aij->A->rmap.n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 956 row = mat->rmap.rstart; 957 for (i=0; i<ai[m]; i++) {aj[i] += mat->cmap.rstart ;} 958 for (i=0; i<m; i++) { 959 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],aj,a,INSERT_VALUES);CHKERRQ(ierr); 960 row++; a += ai[i+1]-ai[i]; aj += ai[i+1]-ai[i]; 961 } 962 aj = Aloc->j; 963 for (i=0; i<ai[m]; i++) {aj[i] -= mat->cmap.rstart;} 964 965 /* copy over the B part */ 966 Aloc = (Mat_SeqAIJ*)aij->B->data; 967 m = aij->B->rmap.n; ai = Aloc->i; aj = Aloc->j; a = Aloc->a; 968 row = mat->rmap.rstart; 969 ierr = PetscMalloc((ai[m]+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 970 ct = cols; 971 for (i=0; i<ai[m]; i++) {cols[i] = aij->garray[aj[i]];} 972 for (i=0; i<m; i++) { 973 ierr = MatSetValues(A,1,&row,ai[i+1]-ai[i],cols,a,INSERT_VALUES);CHKERRQ(ierr); 974 row++; a += ai[i+1]-ai[i]; cols += ai[i+1]-ai[i]; 975 } 976 ierr = PetscFree(ct);CHKERRQ(ierr); 977 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 978 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 979 /* 980 Everyone has to call to draw the matrix since the graphics waits are 981 synchronized across all processors that share the PetscDraw object 982 */ 983 ierr = PetscViewerGetSingleton(viewer,&sviewer);CHKERRQ(ierr); 984 if (!rank) { 985 ierr = PetscObjectSetName((PetscObject)((Mat_MPIAIJ*)(A->data))->A,mat->name);CHKERRQ(ierr); 986 ierr = MatView(((Mat_MPIAIJ*)(A->data))->A,sviewer);CHKERRQ(ierr); 987 } 988 ierr = PetscViewerRestoreSingleton(viewer,&sviewer);CHKERRQ(ierr); 989 ierr = MatDestroy(A);CHKERRQ(ierr); 990 } 991 PetscFunctionReturn(0); 992 } 993 994 #undef __FUNCT__ 995 #define __FUNCT__ "MatView_MPIAIJ" 996 PetscErrorCode MatView_MPIAIJ(Mat mat,PetscViewer viewer) 997 { 998 PetscErrorCode ierr; 999 PetscTruth iascii,isdraw,issocket,isbinary; 1000 1001 PetscFunctionBegin; 1002 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);CHKERRQ(ierr); 1003 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);CHKERRQ(ierr); 1004 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);CHKERRQ(ierr); 1005 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_SOCKET,&issocket);CHKERRQ(ierr); 1006 if (iascii || isdraw || isbinary || issocket) { 1007 ierr = MatView_MPIAIJ_ASCIIorDraworSocket(mat,viewer);CHKERRQ(ierr); 1008 } else { 1009 SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported by MPIAIJ matrices",((PetscObject)viewer)->type_name); 1010 } 1011 PetscFunctionReturn(0); 1012 } 1013 1014 #undef __FUNCT__ 1015 #define __FUNCT__ "MatRelax_MPIAIJ" 1016 PetscErrorCode MatRelax_MPIAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1017 { 1018 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1019 PetscErrorCode ierr; 1020 Vec bb1; 1021 1022 PetscFunctionBegin; 1023 ierr = VecDuplicate(bb,&bb1);CHKERRQ(ierr); 1024 1025 if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){ 1026 if (flag & SOR_ZERO_INITIAL_GUESS) { 1027 ierr = (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,lits,xx);CHKERRQ(ierr); 1028 its--; 1029 } 1030 1031 while (its--) { 1032 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1033 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1034 1035 /* update rhs: bb1 = bb - B*x */ 1036 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1037 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1038 1039 /* local sweep */ 1040 ierr = (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,lits,xx);CHKERRQ(ierr); 1041 } 1042 } else if (flag & SOR_LOCAL_FORWARD_SWEEP){ 1043 if (flag & SOR_ZERO_INITIAL_GUESS) { 1044 ierr = (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,PETSC_NULL,xx);CHKERRQ(ierr); 1045 its--; 1046 } 1047 while (its--) { 1048 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1049 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1050 1051 /* update rhs: bb1 = bb - B*x */ 1052 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1053 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1054 1055 /* local sweep */ 1056 ierr = (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_FORWARD_SWEEP,fshift,lits,PETSC_NULL,xx);CHKERRQ(ierr); 1057 } 1058 } else if (flag & SOR_LOCAL_BACKWARD_SWEEP){ 1059 if (flag & SOR_ZERO_INITIAL_GUESS) { 1060 ierr = (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,PETSC_NULL,xx);CHKERRQ(ierr); 1061 its--; 1062 } 1063 while (its--) { 1064 ierr = VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1065 ierr = VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1066 1067 /* update rhs: bb1 = bb - B*x */ 1068 ierr = VecScale(mat->lvec,-1.0);CHKERRQ(ierr); 1069 ierr = (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);CHKERRQ(ierr); 1070 1071 /* local sweep */ 1072 ierr = (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_BACKWARD_SWEEP,fshift,lits,PETSC_NULL,xx);CHKERRQ(ierr); 1073 } 1074 } else { 1075 SETERRQ(PETSC_ERR_SUP,"Parallel SOR not supported"); 1076 } 1077 1078 ierr = VecDestroy(bb1);CHKERRQ(ierr); 1079 PetscFunctionReturn(0); 1080 } 1081 1082 #undef __FUNCT__ 1083 #define __FUNCT__ "MatPermute_MPIAIJ" 1084 PetscErrorCode MatPermute_MPIAIJ(Mat A,IS rowp,IS colp,Mat *B) 1085 { 1086 MPI_Comm comm,pcomm; 1087 PetscInt first,local_size,nrows,*rows; 1088 int ntids; 1089 IS crowp,growp,irowp,lrowp,lcolp,icolp; 1090 PetscErrorCode ierr; 1091 1092 PetscFunctionBegin; 1093 ierr = PetscObjectGetComm((PetscObject)A,&comm); CHKERRQ(ierr); 1094 /* make a collective version of 'rowp' */ 1095 ierr = PetscObjectGetComm((PetscObject)rowp,&pcomm); CHKERRQ(ierr); 1096 if (pcomm==comm) { 1097 crowp = rowp; 1098 } else { 1099 ierr = ISGetSize(rowp,&nrows); CHKERRQ(ierr); 1100 ierr = ISGetIndices(rowp,&rows); CHKERRQ(ierr); 1101 ierr = ISCreateGeneral(comm,nrows,rows,&crowp); CHKERRQ(ierr); 1102 ierr = ISRestoreIndices(rowp,&rows); CHKERRQ(ierr); 1103 } 1104 /* collect the global row permutation and invert it */ 1105 ierr = ISAllGather(crowp,&growp); CHKERRQ(ierr); 1106 ierr = ISSetPermutation(growp); CHKERRQ(ierr); 1107 if (pcomm!=comm) { 1108 ierr = ISDestroy(crowp); CHKERRQ(ierr); 1109 } 1110 ierr = ISInvertPermutation(growp,PETSC_DECIDE,&irowp);CHKERRQ(ierr); 1111 /* get the local target indices */ 1112 ierr = MatGetOwnershipRange(A,&first,PETSC_NULL); CHKERRQ(ierr); 1113 ierr = MatGetLocalSize(A,&local_size,PETSC_NULL); CHKERRQ(ierr); 1114 ierr = ISGetIndices(irowp,&rows); CHKERRQ(ierr); 1115 ierr = ISCreateGeneral(MPI_COMM_SELF,local_size,rows+first,&lrowp); CHKERRQ(ierr); 1116 ierr = ISRestoreIndices(irowp,&rows); CHKERRQ(ierr); 1117 ierr = ISDestroy(irowp); CHKERRQ(ierr); 1118 /* the column permutation is so much easier; 1119 make a local version of 'colp' and invert it */ 1120 ierr = PetscObjectGetComm((PetscObject)colp,&pcomm); CHKERRQ(ierr); 1121 ierr = MPI_Comm_size(pcomm,&ntids); CHKERRQ(ierr); 1122 if (ntids==1) { 1123 lcolp = colp; 1124 } else { 1125 ierr = ISGetSize(colp,&nrows); CHKERRQ(ierr); 1126 ierr = ISGetIndices(colp,&rows); CHKERRQ(ierr); 1127 ierr = ISCreateGeneral(MPI_COMM_SELF,nrows,rows,&lcolp); CHKERRQ(ierr); 1128 } 1129 ierr = ISInvertPermutation(lcolp,PETSC_DECIDE,&icolp); CHKERRQ(ierr); 1130 ierr = ISSetPermutation(lcolp); CHKERRQ(ierr); 1131 if (ntids>1) { 1132 ierr = ISRestoreIndices(colp,&rows); CHKERRQ(ierr); 1133 ierr = ISDestroy(lcolp); CHKERRQ(ierr); 1134 } 1135 /* now we just get the submatrix */ 1136 ierr = MatGetSubMatrix(A,lrowp,icolp,local_size,MAT_INITIAL_MATRIX,B); CHKERRQ(ierr); 1137 /* clean up */ 1138 ierr = ISDestroy(lrowp); CHKERRQ(ierr); 1139 ierr = ISDestroy(icolp); CHKERRQ(ierr); 1140 PetscFunctionReturn(0); 1141 } 1142 1143 #undef __FUNCT__ 1144 #define __FUNCT__ "MatGetInfo_MPIAIJ" 1145 PetscErrorCode MatGetInfo_MPIAIJ(Mat matin,MatInfoType flag,MatInfo *info) 1146 { 1147 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1148 Mat A = mat->A,B = mat->B; 1149 PetscErrorCode ierr; 1150 PetscReal isend[5],irecv[5]; 1151 1152 PetscFunctionBegin; 1153 info->block_size = 1.0; 1154 ierr = MatGetInfo(A,MAT_LOCAL,info);CHKERRQ(ierr); 1155 isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded; 1156 isend[3] = info->memory; isend[4] = info->mallocs; 1157 ierr = MatGetInfo(B,MAT_LOCAL,info);CHKERRQ(ierr); 1158 isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded; 1159 isend[3] += info->memory; isend[4] += info->mallocs; 1160 if (flag == MAT_LOCAL) { 1161 info->nz_used = isend[0]; 1162 info->nz_allocated = isend[1]; 1163 info->nz_unneeded = isend[2]; 1164 info->memory = isend[3]; 1165 info->mallocs = isend[4]; 1166 } else if (flag == MAT_GLOBAL_MAX) { 1167 ierr = MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_MAX,matin->comm);CHKERRQ(ierr); 1168 info->nz_used = irecv[0]; 1169 info->nz_allocated = irecv[1]; 1170 info->nz_unneeded = irecv[2]; 1171 info->memory = irecv[3]; 1172 info->mallocs = irecv[4]; 1173 } else if (flag == MAT_GLOBAL_SUM) { 1174 ierr = MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_SUM,matin->comm);CHKERRQ(ierr); 1175 info->nz_used = irecv[0]; 1176 info->nz_allocated = irecv[1]; 1177 info->nz_unneeded = irecv[2]; 1178 info->memory = irecv[3]; 1179 info->mallocs = irecv[4]; 1180 } 1181 info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */ 1182 info->fill_ratio_needed = 0; 1183 info->factor_mallocs = 0; 1184 info->rows_global = (double)matin->rmap.N; 1185 info->columns_global = (double)matin->cmap.N; 1186 info->rows_local = (double)matin->rmap.n; 1187 info->columns_local = (double)matin->cmap.N; 1188 1189 PetscFunctionReturn(0); 1190 } 1191 1192 #undef __FUNCT__ 1193 #define __FUNCT__ "MatSetOption_MPIAIJ" 1194 PetscErrorCode MatSetOption_MPIAIJ(Mat A,MatOption op,PetscTruth flg) 1195 { 1196 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1197 PetscErrorCode ierr; 1198 1199 PetscFunctionBegin; 1200 switch (op) { 1201 case MAT_NEW_NONZERO_LOCATIONS: 1202 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1203 case MAT_KEEP_ZEROED_ROWS: 1204 case MAT_NEW_NONZERO_LOCATION_ERR: 1205 case MAT_USE_INODES: 1206 case MAT_IGNORE_ZERO_ENTRIES: 1207 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1208 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1209 break; 1210 case MAT_ROW_ORIENTED: 1211 a->roworiented = flg; 1212 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1213 ierr = MatSetOption(a->B,op,flg);CHKERRQ(ierr); 1214 break; 1215 case MAT_NEW_DIAGONALS: 1216 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1217 break; 1218 case MAT_IGNORE_OFF_PROC_ENTRIES: 1219 a->donotstash = PETSC_TRUE; 1220 break; 1221 case MAT_SYMMETRIC: 1222 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1223 break; 1224 case MAT_STRUCTURALLY_SYMMETRIC: 1225 case MAT_HERMITIAN: 1226 case MAT_SYMMETRY_ETERNAL: 1227 ierr = MatSetOption(a->A,op,flg);CHKERRQ(ierr); 1228 break; 1229 default: 1230 SETERRQ1(PETSC_ERR_SUP,"unknown option %d",op); 1231 } 1232 PetscFunctionReturn(0); 1233 } 1234 1235 #undef __FUNCT__ 1236 #define __FUNCT__ "MatGetRow_MPIAIJ" 1237 PetscErrorCode MatGetRow_MPIAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1238 { 1239 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)matin->data; 1240 PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p; 1241 PetscErrorCode ierr; 1242 PetscInt i,*cworkA,*cworkB,**pcA,**pcB,cstart = matin->cmap.rstart; 1243 PetscInt nztot,nzA,nzB,lrow,rstart = matin->rmap.rstart,rend = matin->rmap.rend; 1244 PetscInt *cmap,*idx_p; 1245 1246 PetscFunctionBegin; 1247 if (mat->getrowactive) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Already active"); 1248 mat->getrowactive = PETSC_TRUE; 1249 1250 if (!mat->rowvalues && (idx || v)) { 1251 /* 1252 allocate enough space to hold information from the longest row. 1253 */ 1254 Mat_SeqAIJ *Aa = (Mat_SeqAIJ*)mat->A->data,*Ba = (Mat_SeqAIJ*)mat->B->data; 1255 PetscInt max = 1,tmp; 1256 for (i=0; i<matin->rmap.n; i++) { 1257 tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; 1258 if (max < tmp) { max = tmp; } 1259 } 1260 ierr = PetscMalloc(max*(sizeof(PetscInt)+sizeof(PetscScalar)),&mat->rowvalues);CHKERRQ(ierr); 1261 mat->rowindices = (PetscInt*)(mat->rowvalues + max); 1262 } 1263 1264 if (row < rstart || row >= rend) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Only local rows") 1265 lrow = row - rstart; 1266 1267 pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB; 1268 if (!v) {pvA = 0; pvB = 0;} 1269 if (!idx) {pcA = 0; if (!v) pcB = 0;} 1270 ierr = (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1271 ierr = (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1272 nztot = nzA + nzB; 1273 1274 cmap = mat->garray; 1275 if (v || idx) { 1276 if (nztot) { 1277 /* Sort by increasing column numbers, assuming A and B already sorted */ 1278 PetscInt imark = -1; 1279 if (v) { 1280 *v = v_p = mat->rowvalues; 1281 for (i=0; i<nzB; i++) { 1282 if (cmap[cworkB[i]] < cstart) v_p[i] = vworkB[i]; 1283 else break; 1284 } 1285 imark = i; 1286 for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i]; 1287 for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i]; 1288 } 1289 if (idx) { 1290 *idx = idx_p = mat->rowindices; 1291 if (imark > -1) { 1292 for (i=0; i<imark; i++) { 1293 idx_p[i] = cmap[cworkB[i]]; 1294 } 1295 } else { 1296 for (i=0; i<nzB; i++) { 1297 if (cmap[cworkB[i]] < cstart) idx_p[i] = cmap[cworkB[i]]; 1298 else break; 1299 } 1300 imark = i; 1301 } 1302 for (i=0; i<nzA; i++) idx_p[imark+i] = cstart + cworkA[i]; 1303 for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]]; 1304 } 1305 } else { 1306 if (idx) *idx = 0; 1307 if (v) *v = 0; 1308 } 1309 } 1310 *nz = nztot; 1311 ierr = (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr); 1312 ierr = (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr); 1313 PetscFunctionReturn(0); 1314 } 1315 1316 #undef __FUNCT__ 1317 #define __FUNCT__ "MatRestoreRow_MPIAIJ" 1318 PetscErrorCode MatRestoreRow_MPIAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1319 { 1320 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1321 1322 PetscFunctionBegin; 1323 if (!aij->getrowactive) { 1324 SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"MatGetRow() must be called first"); 1325 } 1326 aij->getrowactive = PETSC_FALSE; 1327 PetscFunctionReturn(0); 1328 } 1329 1330 #undef __FUNCT__ 1331 #define __FUNCT__ "MatNorm_MPIAIJ" 1332 PetscErrorCode MatNorm_MPIAIJ(Mat mat,NormType type,PetscReal *norm) 1333 { 1334 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1335 Mat_SeqAIJ *amat = (Mat_SeqAIJ*)aij->A->data,*bmat = (Mat_SeqAIJ*)aij->B->data; 1336 PetscErrorCode ierr; 1337 PetscInt i,j,cstart = mat->cmap.rstart; 1338 PetscReal sum = 0.0; 1339 PetscScalar *v; 1340 1341 PetscFunctionBegin; 1342 if (aij->size == 1) { 1343 ierr = MatNorm(aij->A,type,norm);CHKERRQ(ierr); 1344 } else { 1345 if (type == NORM_FROBENIUS) { 1346 v = amat->a; 1347 for (i=0; i<amat->nz; i++) { 1348 #if defined(PETSC_USE_COMPLEX) 1349 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1350 #else 1351 sum += (*v)*(*v); v++; 1352 #endif 1353 } 1354 v = bmat->a; 1355 for (i=0; i<bmat->nz; i++) { 1356 #if defined(PETSC_USE_COMPLEX) 1357 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1358 #else 1359 sum += (*v)*(*v); v++; 1360 #endif 1361 } 1362 ierr = MPI_Allreduce(&sum,norm,1,MPIU_REAL,MPI_SUM,mat->comm);CHKERRQ(ierr); 1363 *norm = sqrt(*norm); 1364 } else if (type == NORM_1) { /* max column norm */ 1365 PetscReal *tmp,*tmp2; 1366 PetscInt *jj,*garray = aij->garray; 1367 ierr = PetscMalloc((mat->cmap.N+1)*sizeof(PetscReal),&tmp);CHKERRQ(ierr); 1368 ierr = PetscMalloc((mat->cmap.N+1)*sizeof(PetscReal),&tmp2);CHKERRQ(ierr); 1369 ierr = PetscMemzero(tmp,mat->cmap.N*sizeof(PetscReal));CHKERRQ(ierr); 1370 *norm = 0.0; 1371 v = amat->a; jj = amat->j; 1372 for (j=0; j<amat->nz; j++) { 1373 tmp[cstart + *jj++ ] += PetscAbsScalar(*v); v++; 1374 } 1375 v = bmat->a; jj = bmat->j; 1376 for (j=0; j<bmat->nz; j++) { 1377 tmp[garray[*jj++]] += PetscAbsScalar(*v); v++; 1378 } 1379 ierr = MPI_Allreduce(tmp,tmp2,mat->cmap.N,MPIU_REAL,MPI_SUM,mat->comm);CHKERRQ(ierr); 1380 for (j=0; j<mat->cmap.N; j++) { 1381 if (tmp2[j] > *norm) *norm = tmp2[j]; 1382 } 1383 ierr = PetscFree(tmp);CHKERRQ(ierr); 1384 ierr = PetscFree(tmp2);CHKERRQ(ierr); 1385 } else if (type == NORM_INFINITY) { /* max row norm */ 1386 PetscReal ntemp = 0.0; 1387 for (j=0; j<aij->A->rmap.n; j++) { 1388 v = amat->a + amat->i[j]; 1389 sum = 0.0; 1390 for (i=0; i<amat->i[j+1]-amat->i[j]; i++) { 1391 sum += PetscAbsScalar(*v); v++; 1392 } 1393 v = bmat->a + bmat->i[j]; 1394 for (i=0; i<bmat->i[j+1]-bmat->i[j]; i++) { 1395 sum += PetscAbsScalar(*v); v++; 1396 } 1397 if (sum > ntemp) ntemp = sum; 1398 } 1399 ierr = MPI_Allreduce(&ntemp,norm,1,MPIU_REAL,MPI_MAX,mat->comm);CHKERRQ(ierr); 1400 } else { 1401 SETERRQ(PETSC_ERR_SUP,"No support for two norm"); 1402 } 1403 } 1404 PetscFunctionReturn(0); 1405 } 1406 1407 #undef __FUNCT__ 1408 #define __FUNCT__ "MatTranspose_MPIAIJ" 1409 PetscErrorCode MatTranspose_MPIAIJ(Mat A,Mat *matout) 1410 { 1411 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1412 Mat_SeqAIJ *Aloc=(Mat_SeqAIJ*)a->A->data,*Bloc=(Mat_SeqAIJ*)a->B->data; 1413 PetscErrorCode ierr; 1414 PetscInt M = A->rmap.N,N = A->cmap.N,ma,na,mb,*ai,*aj,*bi,*bj,row,*cols,i,*d_nnz; 1415 PetscInt cstart=A->cmap.rstart,ncol; 1416 Mat B; 1417 PetscScalar *array; 1418 1419 PetscFunctionBegin; 1420 if (!matout && M != N) SETERRQ(PETSC_ERR_ARG_SIZ,"Square matrix only for in-place"); 1421 1422 /* compute d_nnz for preallocation; o_nnz is approximated by d_nnz to avoid communication */ 1423 ma = A->rmap.n; na = A->cmap.n; mb = a->B->rmap.n; 1424 ai = Aloc->i; aj = Aloc->j; 1425 bi = Bloc->i; bj = Bloc->j; 1426 ierr = PetscMalloc((1+na+bi[mb])*sizeof(PetscInt),&d_nnz);CHKERRQ(ierr); 1427 cols = d_nnz + na + 1; /* work space to be used by B part */ 1428 ierr = PetscMemzero(d_nnz,(1+na)*sizeof(PetscInt));CHKERRQ(ierr); 1429 for (i=0; i<ai[ma]; i++){ 1430 d_nnz[aj[i]] ++; 1431 aj[i] += cstart; /* global col index to be used by MatSetValues() */ 1432 } 1433 1434 ierr = MatCreate(A->comm,&B);CHKERRQ(ierr); 1435 ierr = MatSetSizes(B,A->cmap.n,A->rmap.n,N,M);CHKERRQ(ierr); 1436 ierr = MatSetType(B,A->type_name);CHKERRQ(ierr); 1437 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,d_nnz);CHKERRQ(ierr); 1438 1439 /* copy over the A part */ 1440 array = Aloc->a; 1441 row = A->rmap.rstart; 1442 for (i=0; i<ma; i++) { 1443 ncol = ai[i+1]-ai[i]; 1444 ierr = MatSetValues(B,ncol,aj,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1445 row++; array += ncol; aj += ncol; 1446 } 1447 aj = Aloc->j; 1448 for (i=0; i<ai[ma]; i++) aj[i] -= cstart; /* resume local col index */ 1449 1450 /* copy over the B part */ 1451 array = Bloc->a; 1452 row = A->rmap.rstart; 1453 for (i=0; i<bi[mb]; i++) {cols[i] = a->garray[bj[i]];} 1454 for (i=0; i<mb; i++) { 1455 ncol = bi[i+1]-bi[i]; 1456 ierr = MatSetValues(B,ncol,cols,1,&row,array,INSERT_VALUES);CHKERRQ(ierr); 1457 row++; array += ncol; cols += ncol; 1458 } 1459 ierr = PetscFree(d_nnz);CHKERRQ(ierr); 1460 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1461 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1462 if (matout) { 1463 *matout = B; 1464 } else { 1465 ierr = MatHeaderCopy(A,B);CHKERRQ(ierr); 1466 } 1467 PetscFunctionReturn(0); 1468 } 1469 1470 #undef __FUNCT__ 1471 #define __FUNCT__ "MatDiagonalScale_MPIAIJ" 1472 PetscErrorCode MatDiagonalScale_MPIAIJ(Mat mat,Vec ll,Vec rr) 1473 { 1474 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1475 Mat a = aij->A,b = aij->B; 1476 PetscErrorCode ierr; 1477 PetscInt s1,s2,s3; 1478 1479 PetscFunctionBegin; 1480 ierr = MatGetLocalSize(mat,&s2,&s3);CHKERRQ(ierr); 1481 if (rr) { 1482 ierr = VecGetLocalSize(rr,&s1);CHKERRQ(ierr); 1483 if (s1!=s3) SETERRQ(PETSC_ERR_ARG_SIZ,"right vector non-conforming local size"); 1484 /* Overlap communication with computation. */ 1485 ierr = VecScatterBegin(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1486 } 1487 if (ll) { 1488 ierr = VecGetLocalSize(ll,&s1);CHKERRQ(ierr); 1489 if (s1!=s2) SETERRQ(PETSC_ERR_ARG_SIZ,"left vector non-conforming local size"); 1490 ierr = (*b->ops->diagonalscale)(b,ll,0);CHKERRQ(ierr); 1491 } 1492 /* scale the diagonal block */ 1493 ierr = (*a->ops->diagonalscale)(a,ll,rr);CHKERRQ(ierr); 1494 1495 if (rr) { 1496 /* Do a scatter end and then right scale the off-diagonal block */ 1497 ierr = VecScatterEnd(aij->Mvctx,rr,aij->lvec,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 1498 ierr = (*b->ops->diagonalscale)(b,0,aij->lvec);CHKERRQ(ierr); 1499 } 1500 1501 PetscFunctionReturn(0); 1502 } 1503 1504 #undef __FUNCT__ 1505 #define __FUNCT__ "MatSetBlockSize_MPIAIJ" 1506 PetscErrorCode MatSetBlockSize_MPIAIJ(Mat A,PetscInt bs) 1507 { 1508 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1509 PetscErrorCode ierr; 1510 1511 PetscFunctionBegin; 1512 ierr = MatSetBlockSize(a->A,bs);CHKERRQ(ierr); 1513 ierr = MatSetBlockSize(a->B,bs);CHKERRQ(ierr); 1514 PetscFunctionReturn(0); 1515 } 1516 #undef __FUNCT__ 1517 #define __FUNCT__ "MatSetUnfactored_MPIAIJ" 1518 PetscErrorCode MatSetUnfactored_MPIAIJ(Mat A) 1519 { 1520 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1521 PetscErrorCode ierr; 1522 1523 PetscFunctionBegin; 1524 ierr = MatSetUnfactored(a->A);CHKERRQ(ierr); 1525 PetscFunctionReturn(0); 1526 } 1527 1528 #undef __FUNCT__ 1529 #define __FUNCT__ "MatEqual_MPIAIJ" 1530 PetscErrorCode MatEqual_MPIAIJ(Mat A,Mat B,PetscTruth *flag) 1531 { 1532 Mat_MPIAIJ *matB = (Mat_MPIAIJ*)B->data,*matA = (Mat_MPIAIJ*)A->data; 1533 Mat a,b,c,d; 1534 PetscTruth flg; 1535 PetscErrorCode ierr; 1536 1537 PetscFunctionBegin; 1538 a = matA->A; b = matA->B; 1539 c = matB->A; d = matB->B; 1540 1541 ierr = MatEqual(a,c,&flg);CHKERRQ(ierr); 1542 if (flg) { 1543 ierr = MatEqual(b,d,&flg);CHKERRQ(ierr); 1544 } 1545 ierr = MPI_Allreduce(&flg,flag,1,MPI_INT,MPI_LAND,A->comm);CHKERRQ(ierr); 1546 PetscFunctionReturn(0); 1547 } 1548 1549 #undef __FUNCT__ 1550 #define __FUNCT__ "MatCopy_MPIAIJ" 1551 PetscErrorCode MatCopy_MPIAIJ(Mat A,Mat B,MatStructure str) 1552 { 1553 PetscErrorCode ierr; 1554 Mat_MPIAIJ *a = (Mat_MPIAIJ *)A->data; 1555 Mat_MPIAIJ *b = (Mat_MPIAIJ *)B->data; 1556 1557 PetscFunctionBegin; 1558 /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */ 1559 if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) { 1560 /* because of the column compression in the off-processor part of the matrix a->B, 1561 the number of columns in a->B and b->B may be different, hence we cannot call 1562 the MatCopy() directly on the two parts. If need be, we can provide a more 1563 efficient copy than the MatCopy_Basic() by first uncompressing the a->B matrices 1564 then copying the submatrices */ 1565 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 1566 } else { 1567 ierr = MatCopy(a->A,b->A,str);CHKERRQ(ierr); 1568 ierr = MatCopy(a->B,b->B,str);CHKERRQ(ierr); 1569 } 1570 PetscFunctionReturn(0); 1571 } 1572 1573 #undef __FUNCT__ 1574 #define __FUNCT__ "MatSetUpPreallocation_MPIAIJ" 1575 PetscErrorCode MatSetUpPreallocation_MPIAIJ(Mat A) 1576 { 1577 PetscErrorCode ierr; 1578 1579 PetscFunctionBegin; 1580 ierr = MatMPIAIJSetPreallocation(A,PETSC_DEFAULT,0,PETSC_DEFAULT,0);CHKERRQ(ierr); 1581 PetscFunctionReturn(0); 1582 } 1583 1584 #include "petscblaslapack.h" 1585 #undef __FUNCT__ 1586 #define __FUNCT__ "MatAXPY_MPIAIJ" 1587 PetscErrorCode MatAXPY_MPIAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 1588 { 1589 PetscErrorCode ierr; 1590 PetscInt i; 1591 Mat_MPIAIJ *xx = (Mat_MPIAIJ *)X->data,*yy = (Mat_MPIAIJ *)Y->data; 1592 PetscBLASInt bnz,one=1; 1593 Mat_SeqAIJ *x,*y; 1594 1595 PetscFunctionBegin; 1596 if (str == SAME_NONZERO_PATTERN) { 1597 PetscScalar alpha = a; 1598 x = (Mat_SeqAIJ *)xx->A->data; 1599 y = (Mat_SeqAIJ *)yy->A->data; 1600 bnz = (PetscBLASInt)x->nz; 1601 BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one); 1602 x = (Mat_SeqAIJ *)xx->B->data; 1603 y = (Mat_SeqAIJ *)yy->B->data; 1604 bnz = (PetscBLASInt)x->nz; 1605 BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one); 1606 } else if (str == SUBSET_NONZERO_PATTERN) { 1607 ierr = MatAXPY_SeqAIJ(yy->A,a,xx->A,str);CHKERRQ(ierr); 1608 1609 x = (Mat_SeqAIJ *)xx->B->data; 1610 y = (Mat_SeqAIJ *)yy->B->data; 1611 if (y->xtoy && y->XtoY != xx->B) { 1612 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 1613 ierr = MatDestroy(y->XtoY);CHKERRQ(ierr); 1614 } 1615 if (!y->xtoy) { /* get xtoy */ 1616 ierr = MatAXPYGetxtoy_Private(xx->B->rmap.n,x->i,x->j,xx->garray,y->i,y->j,yy->garray,&y->xtoy);CHKERRQ(ierr); 1617 y->XtoY = xx->B; 1618 ierr = PetscObjectReference((PetscObject)xx->B);CHKERRQ(ierr); 1619 } 1620 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 1621 } else { 1622 ierr = MatAXPY_Basic(Y,a,X,str);CHKERRQ(ierr); 1623 } 1624 PetscFunctionReturn(0); 1625 } 1626 1627 EXTERN PetscErrorCode PETSCMAT_DLLEXPORT MatConjugate_SeqAIJ(Mat); 1628 1629 #undef __FUNCT__ 1630 #define __FUNCT__ "MatConjugate_MPIAIJ" 1631 PetscErrorCode PETSCMAT_DLLEXPORT MatConjugate_MPIAIJ(Mat mat) 1632 { 1633 #if defined(PETSC_USE_COMPLEX) 1634 PetscErrorCode ierr; 1635 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 1636 1637 PetscFunctionBegin; 1638 ierr = MatConjugate_SeqAIJ(aij->A);CHKERRQ(ierr); 1639 ierr = MatConjugate_SeqAIJ(aij->B);CHKERRQ(ierr); 1640 #else 1641 PetscFunctionBegin; 1642 #endif 1643 PetscFunctionReturn(0); 1644 } 1645 1646 #undef __FUNCT__ 1647 #define __FUNCT__ "MatRealPart_MPIAIJ" 1648 PetscErrorCode MatRealPart_MPIAIJ(Mat A) 1649 { 1650 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1651 PetscErrorCode ierr; 1652 1653 PetscFunctionBegin; 1654 ierr = MatRealPart(a->A);CHKERRQ(ierr); 1655 ierr = MatRealPart(a->B);CHKERRQ(ierr); 1656 PetscFunctionReturn(0); 1657 } 1658 1659 #undef __FUNCT__ 1660 #define __FUNCT__ "MatImaginaryPart_MPIAIJ" 1661 PetscErrorCode MatImaginaryPart_MPIAIJ(Mat A) 1662 { 1663 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 1664 PetscErrorCode ierr; 1665 1666 PetscFunctionBegin; 1667 ierr = MatImaginaryPart(a->A);CHKERRQ(ierr); 1668 ierr = MatImaginaryPart(a->B);CHKERRQ(ierr); 1669 PetscFunctionReturn(0); 1670 } 1671 1672 #ifdef PETSC_HAVE_PBGL 1673 1674 #include <boost/parallel/mpi/bsp_process_group.hpp> 1675 #include <boost/graph/distributed/ilu_default_graph.hpp> 1676 #include <boost/graph/distributed/ilu_0_block.hpp> 1677 #include <boost/graph/distributed/ilu_preconditioner.hpp> 1678 #include <boost/graph/distributed/petsc/interface.hpp> 1679 #include <boost/multi_array.hpp> 1680 #include <boost/parallel/distributed_property_map.hpp> 1681 1682 #undef __FUNCT__ 1683 #define __FUNCT__ "MatILUFactorSymbolic_MPIAIJ" 1684 /* 1685 This uses the parallel ILU factorization of Peter Gottschling <pgottsch@osl.iu.edu> 1686 */ 1687 PetscErrorCode MatILUFactorSymbolic_MPIAIJ(Mat A, IS isrow, IS iscol, MatFactorInfo *info, Mat *fact) 1688 { 1689 namespace petsc = boost::distributed::petsc; 1690 1691 namespace graph_dist = boost::graph::distributed; 1692 using boost::graph::distributed::ilu_default::process_group_type; 1693 using boost::graph::ilu_permuted; 1694 1695 PetscTruth row_identity, col_identity; 1696 PetscContainer c; 1697 PetscInt m, n, M, N; 1698 PetscErrorCode ierr; 1699 1700 PetscFunctionBegin; 1701 if (info->levels != 0) SETERRQ(PETSC_ERR_SUP,"Only levels = 0 supported for parallel ilu"); 1702 ierr = ISIdentity(isrow, &row_identity);CHKERRQ(ierr); 1703 ierr = ISIdentity(iscol, &col_identity);CHKERRQ(ierr); 1704 if (!row_identity || !col_identity) { 1705 SETERRQ(PETSC_ERR_ARG_WRONG,"Row and column permutations must be identity for parallel ILU"); 1706 } 1707 1708 process_group_type pg; 1709 typedef graph_dist::ilu_default::ilu_level_graph_type lgraph_type; 1710 lgraph_type* lgraph_p = new lgraph_type(petsc::num_global_vertices(A), pg, petsc::matrix_distribution(A, pg)); 1711 lgraph_type& level_graph = *lgraph_p; 1712 graph_dist::ilu_default::graph_type& graph(level_graph.graph); 1713 1714 petsc::read_matrix(A, graph, get(boost::edge_weight, graph)); 1715 ilu_permuted(level_graph); 1716 1717 /* put together the new matrix */ 1718 ierr = MatCreate(A->comm, fact);CHKERRQ(ierr); 1719 ierr = MatGetLocalSize(A, &m, &n);CHKERRQ(ierr); 1720 ierr = MatGetSize(A, &M, &N);CHKERRQ(ierr); 1721 ierr = MatSetSizes(*fact, m, n, M, N);CHKERRQ(ierr); 1722 ierr = MatSetType(*fact, A->type_name);CHKERRQ(ierr); 1723 ierr = MatAssemblyBegin(*fact, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1724 ierr = MatAssemblyEnd(*fact, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1725 (*fact)->factor = FACTOR_LU; 1726 1727 ierr = PetscContainerCreate(A->comm, &c); 1728 ierr = PetscContainerSetPointer(c, lgraph_p); 1729 ierr = PetscObjectCompose((PetscObject) (*fact), "graph", (PetscObject) c); 1730 PetscFunctionReturn(0); 1731 } 1732 1733 #undef __FUNCT__ 1734 #define __FUNCT__ "MatLUFactorNumeric_MPIAIJ" 1735 PetscErrorCode MatLUFactorNumeric_MPIAIJ(Mat A, MatFactorInfo *info, Mat *B) 1736 { 1737 PetscFunctionBegin; 1738 PetscFunctionReturn(0); 1739 } 1740 1741 #undef __FUNCT__ 1742 #define __FUNCT__ "MatSolve_MPIAIJ" 1743 /* 1744 This uses the parallel ILU factorization of Peter Gottschling <pgottsch@osl.iu.edu> 1745 */ 1746 PetscErrorCode MatSolve_MPIAIJ(Mat A, Vec b, Vec x) 1747 { 1748 namespace graph_dist = boost::graph::distributed; 1749 1750 typedef graph_dist::ilu_default::ilu_level_graph_type lgraph_type; 1751 lgraph_type* lgraph_p; 1752 PetscContainer c; 1753 PetscErrorCode ierr; 1754 1755 PetscFunctionBegin; 1756 ierr = PetscObjectQuery((PetscObject) A, "graph", (PetscObject *) &c);CHKERRQ(ierr); 1757 ierr = PetscContainerGetPointer(c, (void **) &lgraph_p);CHKERRQ(ierr); 1758 ierr = VecCopy(b, x); CHKERRQ(ierr); 1759 1760 PetscScalar* array_x; 1761 ierr = VecGetArray(x, &array_x);CHKERRQ(ierr); 1762 PetscInt sx; 1763 ierr = VecGetSize(x, &sx);CHKERRQ(ierr); 1764 1765 PetscScalar* array_b; 1766 ierr = VecGetArray(b, &array_b);CHKERRQ(ierr); 1767 PetscInt sb; 1768 ierr = VecGetSize(b, &sb);CHKERRQ(ierr); 1769 1770 lgraph_type& level_graph = *lgraph_p; 1771 graph_dist::ilu_default::graph_type& graph(level_graph.graph); 1772 1773 typedef boost::multi_array_ref<PetscScalar, 1> array_ref_type; 1774 array_ref_type ref_b(array_b, boost::extents[num_vertices(graph)]), 1775 ref_x(array_x, boost::extents[num_vertices(graph)]); 1776 1777 typedef boost::iterator_property_map<array_ref_type::iterator, 1778 boost::property_map<graph_dist::ilu_default::graph_type, boost::vertex_index_t>::type> gvector_type; 1779 gvector_type vector_b(ref_b.begin(), get(boost::vertex_index, graph)), 1780 vector_x(ref_x.begin(), get(boost::vertex_index, graph)); 1781 1782 ilu_set_solve(*lgraph_p, vector_b, vector_x); 1783 1784 PetscFunctionReturn(0); 1785 } 1786 #endif 1787 1788 typedef struct { /* used by MatGetRedundantMatrix() for reusing matredundant */ 1789 PetscInt nzlocal,nsends,nrecvs; 1790 PetscMPIInt *send_rank; 1791 PetscInt *sbuf_nz,*sbuf_j,**rbuf_j; 1792 PetscScalar *sbuf_a,**rbuf_a; 1793 PetscErrorCode (*MatDestroy)(Mat); 1794 } Mat_Redundant; 1795 1796 #undef __FUNCT__ 1797 #define __FUNCT__ "PetscContainerDestroy_MatRedundant" 1798 PetscErrorCode PetscContainerDestroy_MatRedundant(void *ptr) 1799 { 1800 PetscErrorCode ierr; 1801 Mat_Redundant *redund=(Mat_Redundant*)ptr; 1802 PetscInt i; 1803 1804 PetscFunctionBegin; 1805 ierr = PetscFree(redund->send_rank);CHKERRQ(ierr); 1806 ierr = PetscFree(redund->sbuf_j);CHKERRQ(ierr); 1807 ierr = PetscFree(redund->sbuf_a);CHKERRQ(ierr); 1808 for (i=0; i<redund->nrecvs; i++){ 1809 ierr = PetscFree(redund->rbuf_j[i]);CHKERRQ(ierr); 1810 ierr = PetscFree(redund->rbuf_a[i]);CHKERRQ(ierr); 1811 } 1812 ierr = PetscFree3(redund->sbuf_nz,redund->rbuf_j,redund->rbuf_a);CHKERRQ(ierr); 1813 ierr = PetscFree(redund);CHKERRQ(ierr); 1814 PetscFunctionReturn(0); 1815 } 1816 1817 #undef __FUNCT__ 1818 #define __FUNCT__ "MatDestroy_MatRedundant" 1819 PetscErrorCode MatDestroy_MatRedundant(Mat A) 1820 { 1821 PetscErrorCode ierr; 1822 PetscContainer container; 1823 Mat_Redundant *redund=PETSC_NULL; 1824 1825 PetscFunctionBegin; 1826 ierr = PetscObjectQuery((PetscObject)A,"Mat_Redundant",(PetscObject *)&container);CHKERRQ(ierr); 1827 if (container) { 1828 ierr = PetscContainerGetPointer(container,(void **)&redund);CHKERRQ(ierr); 1829 } else { 1830 SETERRQ(PETSC_ERR_PLIB,"Container does not exit"); 1831 } 1832 A->ops->destroy = redund->MatDestroy; 1833 ierr = PetscObjectCompose((PetscObject)A,"Mat_Redundant",0);CHKERRQ(ierr); 1834 ierr = (*A->ops->destroy)(A);CHKERRQ(ierr); 1835 ierr = PetscContainerDestroy(container);CHKERRQ(ierr); 1836 PetscFunctionReturn(0); 1837 } 1838 1839 #undef __FUNCT__ 1840 #define __FUNCT__ "MatGetRedundantMatrix_MPIAIJ" 1841 PetscErrorCode MatGetRedundantMatrix_MPIAIJ(Mat mat,PetscInt nsubcomm,MPI_Comm subcomm,PetscInt mlocal_sub,MatReuse reuse,Mat *matredundant) 1842 { 1843 PetscMPIInt rank,size; 1844 MPI_Comm comm=mat->comm; 1845 PetscErrorCode ierr; 1846 PetscInt nsends=0,nrecvs=0,i,rownz_max=0; 1847 PetscMPIInt *send_rank=PETSC_NULL,*recv_rank=PETSC_NULL; 1848 PetscInt *rowrange=mat->rmap.range; 1849 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 1850 Mat A=aij->A,B=aij->B,C=*matredundant; 1851 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)B->data; 1852 PetscScalar *sbuf_a; 1853 PetscInt nzlocal=a->nz+b->nz; 1854 PetscInt j,cstart=mat->cmap.rstart,cend=mat->cmap.rend,row,nzA,nzB,ncols,*cworkA,*cworkB; 1855 PetscInt rstart=mat->rmap.rstart,rend=mat->rmap.rend,*bmap=aij->garray,M,N; 1856 PetscInt *cols,ctmp,lwrite,*rptr,l,*sbuf_j; 1857 PetscScalar *vals,*aworkA,*aworkB; 1858 PetscMPIInt tag1,tag2,tag3,imdex; 1859 MPI_Request *s_waits1=PETSC_NULL,*s_waits2=PETSC_NULL,*s_waits3=PETSC_NULL, 1860 *r_waits1=PETSC_NULL,*r_waits2=PETSC_NULL,*r_waits3=PETSC_NULL; 1861 MPI_Status recv_status,*send_status; 1862 PetscInt *sbuf_nz=PETSC_NULL,*rbuf_nz=PETSC_NULL,count; 1863 PetscInt **rbuf_j=PETSC_NULL; 1864 PetscScalar **rbuf_a=PETSC_NULL; 1865 Mat_Redundant *redund=PETSC_NULL; 1866 PetscContainer container; 1867 1868 PetscFunctionBegin; 1869 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1870 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1871 1872 if (reuse == MAT_REUSE_MATRIX) { 1873 ierr = MatGetSize(C,&M,&N);CHKERRQ(ierr); 1874 if (M != N || M != mat->rmap.N) SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong global size"); 1875 ierr = MatGetLocalSize(C,&M,&N);CHKERRQ(ierr); 1876 if (M != N || M != mlocal_sub) SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong local size"); 1877 ierr = PetscObjectQuery((PetscObject)C,"Mat_Redundant",(PetscObject *)&container);CHKERRQ(ierr); 1878 if (container) { 1879 ierr = PetscContainerGetPointer(container,(void **)&redund);CHKERRQ(ierr); 1880 } else { 1881 SETERRQ(PETSC_ERR_PLIB,"Container does not exit"); 1882 } 1883 if (nzlocal != redund->nzlocal) SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. Wrong nzlocal"); 1884 1885 nsends = redund->nsends; 1886 nrecvs = redund->nrecvs; 1887 send_rank = redund->send_rank; recv_rank = send_rank + size; 1888 sbuf_nz = redund->sbuf_nz; rbuf_nz = sbuf_nz + nsends; 1889 sbuf_j = redund->sbuf_j; 1890 sbuf_a = redund->sbuf_a; 1891 rbuf_j = redund->rbuf_j; 1892 rbuf_a = redund->rbuf_a; 1893 } 1894 1895 if (reuse == MAT_INITIAL_MATRIX){ 1896 PetscMPIInt subrank,subsize; 1897 PetscInt nleftover,np_subcomm; 1898 /* get the destination processors' id send_rank, nsends and nrecvs */ 1899 ierr = MPI_Comm_rank(subcomm,&subrank);CHKERRQ(ierr); 1900 ierr = MPI_Comm_size(subcomm,&subsize);CHKERRQ(ierr); 1901 ierr = PetscMalloc((2*size+1)*sizeof(PetscMPIInt),&send_rank); 1902 recv_rank = send_rank + size; 1903 np_subcomm = size/nsubcomm; 1904 nleftover = size - nsubcomm*np_subcomm; 1905 nsends = 0; nrecvs = 0; 1906 for (i=0; i<size; i++){ /* i=rank*/ 1907 if (subrank == i/nsubcomm && rank != i){ /* my_subrank == other's subrank */ 1908 send_rank[nsends] = i; nsends++; 1909 recv_rank[nrecvs++] = i; 1910 } 1911 } 1912 if (rank >= size - nleftover){/* this proc is a leftover processor */ 1913 i = size-nleftover-1; 1914 j = 0; 1915 while (j < nsubcomm - nleftover){ 1916 send_rank[nsends++] = i; 1917 i--; j++; 1918 } 1919 } 1920 1921 if (nleftover && subsize == size/nsubcomm && subrank==subsize-1){ /* this proc recvs from leftover processors */ 1922 for (i=0; i<nleftover; i++){ 1923 recv_rank[nrecvs++] = size-nleftover+i; 1924 } 1925 } 1926 1927 /* allocate sbuf_j, sbuf_a */ 1928 i = nzlocal + rowrange[rank+1] - rowrange[rank] + 2; 1929 ierr = PetscMalloc(i*sizeof(PetscInt),&sbuf_j);CHKERRQ(ierr); 1930 ierr = PetscMalloc((nzlocal+1)*sizeof(PetscScalar),&sbuf_a);CHKERRQ(ierr); 1931 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 1932 1933 /* copy mat's local entries into the buffers */ 1934 if (reuse == MAT_INITIAL_MATRIX){ 1935 rownz_max = 0; 1936 rptr = sbuf_j; 1937 cols = sbuf_j + rend-rstart + 1; 1938 vals = sbuf_a; 1939 rptr[0] = 0; 1940 for (i=0; i<rend-rstart; i++){ 1941 row = i + rstart; 1942 nzA = a->i[i+1] - a->i[i]; nzB = b->i[i+1] - b->i[i]; 1943 ncols = nzA + nzB; 1944 cworkA = a->j + a->i[i]; cworkB = b->j + b->i[i]; 1945 aworkA = a->a + a->i[i]; aworkB = b->a + b->i[i]; 1946 /* load the column indices for this row into cols */ 1947 lwrite = 0; 1948 for (l=0; l<nzB; l++) { 1949 if ((ctmp = bmap[cworkB[l]]) < cstart){ 1950 vals[lwrite] = aworkB[l]; 1951 cols[lwrite++] = ctmp; 1952 } 1953 } 1954 for (l=0; l<nzA; l++){ 1955 vals[lwrite] = aworkA[l]; 1956 cols[lwrite++] = cstart + cworkA[l]; 1957 } 1958 for (l=0; l<nzB; l++) { 1959 if ((ctmp = bmap[cworkB[l]]) >= cend){ 1960 vals[lwrite] = aworkB[l]; 1961 cols[lwrite++] = ctmp; 1962 } 1963 } 1964 vals += ncols; 1965 cols += ncols; 1966 rptr[i+1] = rptr[i] + ncols; 1967 if (rownz_max < ncols) rownz_max = ncols; 1968 } 1969 if (rptr[rend-rstart] != a->nz + b->nz) SETERRQ4(1, "rptr[%d] %d != %d + %d",rend-rstart,rptr[rend-rstart+1],a->nz,b->nz); 1970 } else { /* only copy matrix values into sbuf_a */ 1971 rptr = sbuf_j; 1972 vals = sbuf_a; 1973 rptr[0] = 0; 1974 for (i=0; i<rend-rstart; i++){ 1975 row = i + rstart; 1976 nzA = a->i[i+1] - a->i[i]; nzB = b->i[i+1] - b->i[i]; 1977 ncols = nzA + nzB; 1978 cworkA = a->j + a->i[i]; cworkB = b->j + b->i[i]; 1979 aworkA = a->a + a->i[i]; aworkB = b->a + b->i[i]; 1980 lwrite = 0; 1981 for (l=0; l<nzB; l++) { 1982 if ((ctmp = bmap[cworkB[l]]) < cstart) vals[lwrite++] = aworkB[l]; 1983 } 1984 for (l=0; l<nzA; l++) vals[lwrite++] = aworkA[l]; 1985 for (l=0; l<nzB; l++) { 1986 if ((ctmp = bmap[cworkB[l]]) >= cend) vals[lwrite++] = aworkB[l]; 1987 } 1988 vals += ncols; 1989 rptr[i+1] = rptr[i] + ncols; 1990 } 1991 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 1992 1993 /* send nzlocal to others, and recv other's nzlocal */ 1994 /*--------------------------------------------------*/ 1995 if (reuse == MAT_INITIAL_MATRIX){ 1996 ierr = PetscMalloc2(3*(nsends + nrecvs)+1,MPI_Request,&s_waits3,nsends+1,MPI_Status,&send_status);CHKERRQ(ierr); 1997 s_waits2 = s_waits3 + nsends; 1998 s_waits1 = s_waits2 + nsends; 1999 r_waits1 = s_waits1 + nsends; 2000 r_waits2 = r_waits1 + nrecvs; 2001 r_waits3 = r_waits2 + nrecvs; 2002 } else { 2003 ierr = PetscMalloc2(nsends + nrecvs +1,MPI_Request,&s_waits3,nsends+1,MPI_Status,&send_status);CHKERRQ(ierr); 2004 r_waits3 = s_waits3 + nsends; 2005 } 2006 2007 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag3);CHKERRQ(ierr); 2008 if (reuse == MAT_INITIAL_MATRIX){ 2009 /* get new tags to keep the communication clean */ 2010 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag1);CHKERRQ(ierr); 2011 ierr = PetscObjectGetNewTag((PetscObject)mat,&tag2);CHKERRQ(ierr); 2012 ierr = PetscMalloc3(nsends+nrecvs+1,PetscInt,&sbuf_nz,nrecvs,PetscInt*,&rbuf_j,nrecvs,PetscScalar*,&rbuf_a);CHKERRQ(ierr); 2013 rbuf_nz = sbuf_nz + nsends; 2014 2015 /* post receives of other's nzlocal */ 2016 for (i=0; i<nrecvs; i++){ 2017 ierr = MPI_Irecv(rbuf_nz+i,1,MPIU_INT,MPI_ANY_SOURCE,tag1,comm,r_waits1+i);CHKERRQ(ierr); 2018 } 2019 /* send nzlocal to others */ 2020 for (i=0; i<nsends; i++){ 2021 sbuf_nz[i] = nzlocal; 2022 ierr = MPI_Isend(sbuf_nz+i,1,MPIU_INT,send_rank[i],tag1,comm,s_waits1+i);CHKERRQ(ierr); 2023 } 2024 /* wait on receives of nzlocal; allocate space for rbuf_j, rbuf_a */ 2025 count = nrecvs; 2026 while (count) { 2027 ierr = MPI_Waitany(nrecvs,r_waits1,&imdex,&recv_status);CHKERRQ(ierr); 2028 recv_rank[imdex] = recv_status.MPI_SOURCE; 2029 /* allocate rbuf_a and rbuf_j; then post receives of rbuf_j */ 2030 ierr = PetscMalloc((rbuf_nz[imdex]+1)*sizeof(PetscScalar),&rbuf_a[imdex]);CHKERRQ(ierr); 2031 2032 i = rowrange[recv_status.MPI_SOURCE+1] - rowrange[recv_status.MPI_SOURCE]; /* number of expected mat->i */ 2033 rbuf_nz[imdex] += i + 2; 2034 ierr = PetscMalloc(rbuf_nz[imdex]*sizeof(PetscInt),&rbuf_j[imdex]);CHKERRQ(ierr); 2035 ierr = MPI_Irecv(rbuf_j[imdex],rbuf_nz[imdex],MPIU_INT,recv_status.MPI_SOURCE,tag2,comm,r_waits2+imdex);CHKERRQ(ierr); 2036 count--; 2037 } 2038 /* wait on sends of nzlocal */ 2039 if (nsends) {ierr = MPI_Waitall(nsends,s_waits1,send_status);CHKERRQ(ierr);} 2040 /* send mat->i,j to others, and recv from other's */ 2041 /*------------------------------------------------*/ 2042 for (i=0; i<nsends; i++){ 2043 j = nzlocal + rowrange[rank+1] - rowrange[rank] + 1; 2044 ierr = MPI_Isend(sbuf_j,j,MPIU_INT,send_rank[i],tag2,comm,s_waits2+i);CHKERRQ(ierr); 2045 } 2046 /* wait on receives of mat->i,j */ 2047 /*------------------------------*/ 2048 count = nrecvs; 2049 while (count) { 2050 ierr = MPI_Waitany(nrecvs,r_waits2,&imdex,&recv_status);CHKERRQ(ierr); 2051 if (recv_rank[imdex] != recv_status.MPI_SOURCE) SETERRQ2(1, "recv_rank %d != MPI_SOURCE %d",recv_rank[imdex],recv_status.MPI_SOURCE); 2052 count--; 2053 } 2054 /* wait on sends of mat->i,j */ 2055 /*---------------------------*/ 2056 if (nsends) { 2057 ierr = MPI_Waitall(nsends,s_waits2,send_status);CHKERRQ(ierr); 2058 } 2059 } /* endof if (reuse == MAT_INITIAL_MATRIX) */ 2060 2061 /* post receives, send and receive mat->a */ 2062 /*----------------------------------------*/ 2063 for (imdex=0; imdex<nrecvs; imdex++) { 2064 ierr = MPI_Irecv(rbuf_a[imdex],rbuf_nz[imdex],MPIU_SCALAR,recv_rank[imdex],tag3,comm,r_waits3+imdex);CHKERRQ(ierr); 2065 } 2066 for (i=0; i<nsends; i++){ 2067 ierr = MPI_Isend(sbuf_a,nzlocal,MPIU_SCALAR,send_rank[i],tag3,comm,s_waits3+i);CHKERRQ(ierr); 2068 } 2069 count = nrecvs; 2070 while (count) { 2071 ierr = MPI_Waitany(nrecvs,r_waits3,&imdex,&recv_status);CHKERRQ(ierr); 2072 if (recv_rank[imdex] != recv_status.MPI_SOURCE) SETERRQ2(1, "recv_rank %d != MPI_SOURCE %d",recv_rank[imdex],recv_status.MPI_SOURCE); 2073 count--; 2074 } 2075 if (nsends) { 2076 ierr = MPI_Waitall(nsends,s_waits3,send_status);CHKERRQ(ierr); 2077 } 2078 2079 ierr = PetscFree2(s_waits3,send_status);CHKERRQ(ierr); 2080 2081 /* create redundant matrix */ 2082 /*-------------------------*/ 2083 if (reuse == MAT_INITIAL_MATRIX){ 2084 /* compute rownz_max for preallocation */ 2085 for (imdex=0; imdex<nrecvs; imdex++){ 2086 j = rowrange[recv_rank[imdex]+1] - rowrange[recv_rank[imdex]]; 2087 rptr = rbuf_j[imdex]; 2088 for (i=0; i<j; i++){ 2089 ncols = rptr[i+1] - rptr[i]; 2090 if (rownz_max < ncols) rownz_max = ncols; 2091 } 2092 } 2093 2094 ierr = MatCreate(subcomm,&C);CHKERRQ(ierr); 2095 ierr = MatSetSizes(C,mlocal_sub,mlocal_sub,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr); 2096 ierr = MatSetFromOptions(C);CHKERRQ(ierr); 2097 ierr = MatSeqAIJSetPreallocation(C,rownz_max,PETSC_NULL);CHKERRQ(ierr); 2098 ierr = MatMPIAIJSetPreallocation(C,rownz_max,PETSC_NULL,rownz_max,PETSC_NULL);CHKERRQ(ierr); 2099 } else { 2100 C = *matredundant; 2101 } 2102 2103 /* insert local matrix entries */ 2104 rptr = sbuf_j; 2105 cols = sbuf_j + rend-rstart + 1; 2106 vals = sbuf_a; 2107 for (i=0; i<rend-rstart; i++){ 2108 row = i + rstart; 2109 ncols = rptr[i+1] - rptr[i]; 2110 ierr = MatSetValues(C,1,&row,ncols,cols,vals,INSERT_VALUES);CHKERRQ(ierr); 2111 vals += ncols; 2112 cols += ncols; 2113 } 2114 /* insert received matrix entries */ 2115 for (imdex=0; imdex<nrecvs; imdex++){ 2116 rstart = rowrange[recv_rank[imdex]]; 2117 rend = rowrange[recv_rank[imdex]+1]; 2118 rptr = rbuf_j[imdex]; 2119 cols = rbuf_j[imdex] + rend-rstart + 1; 2120 vals = rbuf_a[imdex]; 2121 for (i=0; i<rend-rstart; i++){ 2122 row = i + rstart; 2123 ncols = rptr[i+1] - rptr[i]; 2124 ierr = MatSetValues(C,1,&row,ncols,cols,vals,INSERT_VALUES);CHKERRQ(ierr); 2125 vals += ncols; 2126 cols += ncols; 2127 } 2128 } 2129 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2130 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2131 ierr = MatGetSize(C,&M,&N);CHKERRQ(ierr); 2132 if (M != mat->rmap.N || N != mat->cmap.N) SETERRQ2(PETSC_ERR_ARG_INCOMP,"redundant mat size %d != input mat size %d",M,mat->rmap.N); 2133 if (reuse == MAT_INITIAL_MATRIX){ 2134 PetscContainer container; 2135 *matredundant = C; 2136 /* create a supporting struct and attach it to C for reuse */ 2137 ierr = PetscNewLog(C,Mat_Redundant,&redund);CHKERRQ(ierr); 2138 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 2139 ierr = PetscContainerSetPointer(container,redund);CHKERRQ(ierr); 2140 ierr = PetscObjectCompose((PetscObject)C,"Mat_Redundant",(PetscObject)container);CHKERRQ(ierr); 2141 ierr = PetscContainerSetUserDestroy(container,PetscContainerDestroy_MatRedundant);CHKERRQ(ierr); 2142 2143 redund->nzlocal = nzlocal; 2144 redund->nsends = nsends; 2145 redund->nrecvs = nrecvs; 2146 redund->send_rank = send_rank; 2147 redund->sbuf_nz = sbuf_nz; 2148 redund->sbuf_j = sbuf_j; 2149 redund->sbuf_a = sbuf_a; 2150 redund->rbuf_j = rbuf_j; 2151 redund->rbuf_a = rbuf_a; 2152 2153 redund->MatDestroy = C->ops->destroy; 2154 C->ops->destroy = MatDestroy_MatRedundant; 2155 } 2156 PetscFunctionReturn(0); 2157 } 2158 2159 #undef __FUNCT__ 2160 #define __FUNCT__ "MatGetRowMin_MPIAIJ" 2161 PetscErrorCode MatGetRowMin_MPIAIJ(Mat A, Vec v, PetscInt idx[]) 2162 { 2163 Mat_MPIAIJ *mat = (Mat_MPIAIJ *) A->data; 2164 PetscInt n = A->rmap.n; 2165 PetscInt cstart = A->cmap.rstart; 2166 PetscInt *cmap = mat->garray; 2167 PetscInt *diagIdx, *offdiagIdx; 2168 Vec diagV, offdiagV; 2169 PetscScalar *a, *diagA, *offdiagA; 2170 PetscInt r; 2171 PetscErrorCode ierr; 2172 2173 PetscFunctionBegin; 2174 ierr = PetscMalloc2(n,PetscInt,&diagIdx,n,PetscInt,&offdiagIdx);CHKERRQ(ierr); 2175 ierr = VecCreateSeq(A->comm, n, &diagV);CHKERRQ(ierr); 2176 ierr = VecCreateSeq(A->comm, n, &offdiagV);CHKERRQ(ierr); 2177 ierr = MatGetRowMin(mat->A, diagV, diagIdx);CHKERRQ(ierr); 2178 ierr = MatGetRowMin(mat->B, offdiagV, offdiagIdx);CHKERRQ(ierr); 2179 ierr = VecGetArray(v, &a);CHKERRQ(ierr); 2180 ierr = VecGetArray(diagV, &diagA);CHKERRQ(ierr); 2181 ierr = VecGetArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2182 for(r = 0; r < n; ++r) { 2183 if (PetscAbsScalar(diagA[r]) <= PetscAbsScalar(offdiagA[r])) { 2184 a[r] = diagA[r]; 2185 idx[r] = cstart + diagIdx[r]; 2186 } else { 2187 a[r] = offdiagA[r]; 2188 idx[r] = cmap[offdiagIdx[r]]; 2189 } 2190 } 2191 ierr = VecRestoreArray(v, &a);CHKERRQ(ierr); 2192 ierr = VecRestoreArray(diagV, &diagA);CHKERRQ(ierr); 2193 ierr = VecRestoreArray(offdiagV, &offdiagA);CHKERRQ(ierr); 2194 ierr = VecDestroy(diagV);CHKERRQ(ierr); 2195 ierr = VecDestroy(offdiagV);CHKERRQ(ierr); 2196 ierr = PetscFree2(diagIdx, offdiagIdx);CHKERRQ(ierr); 2197 PetscFunctionReturn(0); 2198 } 2199 2200 /* -------------------------------------------------------------------*/ 2201 static struct _MatOps MatOps_Values = {MatSetValues_MPIAIJ, 2202 MatGetRow_MPIAIJ, 2203 MatRestoreRow_MPIAIJ, 2204 MatMult_MPIAIJ, 2205 /* 4*/ MatMultAdd_MPIAIJ, 2206 MatMultTranspose_MPIAIJ, 2207 MatMultTransposeAdd_MPIAIJ, 2208 #ifdef PETSC_HAVE_PBGL 2209 MatSolve_MPIAIJ, 2210 #else 2211 0, 2212 #endif 2213 0, 2214 0, 2215 /*10*/ 0, 2216 0, 2217 0, 2218 MatRelax_MPIAIJ, 2219 MatTranspose_MPIAIJ, 2220 /*15*/ MatGetInfo_MPIAIJ, 2221 MatEqual_MPIAIJ, 2222 MatGetDiagonal_MPIAIJ, 2223 MatDiagonalScale_MPIAIJ, 2224 MatNorm_MPIAIJ, 2225 /*20*/ MatAssemblyBegin_MPIAIJ, 2226 MatAssemblyEnd_MPIAIJ, 2227 0, 2228 MatSetOption_MPIAIJ, 2229 MatZeroEntries_MPIAIJ, 2230 /*25*/ MatZeroRows_MPIAIJ, 2231 0, 2232 #ifdef PETSC_HAVE_PBGL 2233 MatLUFactorNumeric_MPIAIJ, 2234 #else 2235 0, 2236 #endif 2237 0, 2238 0, 2239 /*30*/ MatSetUpPreallocation_MPIAIJ, 2240 #ifdef PETSC_HAVE_PBGL 2241 MatILUFactorSymbolic_MPIAIJ, 2242 #else 2243 0, 2244 #endif 2245 0, 2246 0, 2247 0, 2248 /*35*/ MatDuplicate_MPIAIJ, 2249 0, 2250 0, 2251 0, 2252 0, 2253 /*40*/ MatAXPY_MPIAIJ, 2254 MatGetSubMatrices_MPIAIJ, 2255 MatIncreaseOverlap_MPIAIJ, 2256 MatGetValues_MPIAIJ, 2257 MatCopy_MPIAIJ, 2258 /*45*/ 0, 2259 MatScale_MPIAIJ, 2260 0, 2261 0, 2262 0, 2263 /*50*/ MatSetBlockSize_MPIAIJ, 2264 0, 2265 0, 2266 0, 2267 0, 2268 /*55*/ MatFDColoringCreate_MPIAIJ, 2269 0, 2270 MatSetUnfactored_MPIAIJ, 2271 MatPermute_MPIAIJ, 2272 0, 2273 /*60*/ MatGetSubMatrix_MPIAIJ, 2274 MatDestroy_MPIAIJ, 2275 MatView_MPIAIJ, 2276 0, 2277 0, 2278 /*65*/ 0, 2279 0, 2280 0, 2281 0, 2282 0, 2283 /*70*/ 0, 2284 0, 2285 MatSetColoring_MPIAIJ, 2286 #if defined(PETSC_HAVE_ADIC) 2287 MatSetValuesAdic_MPIAIJ, 2288 #else 2289 0, 2290 #endif 2291 MatSetValuesAdifor_MPIAIJ, 2292 /*75*/ 0, 2293 0, 2294 0, 2295 0, 2296 0, 2297 /*80*/ 0, 2298 0, 2299 0, 2300 0, 2301 /*84*/ MatLoad_MPIAIJ, 2302 0, 2303 0, 2304 0, 2305 0, 2306 0, 2307 /*90*/ MatMatMult_MPIAIJ_MPIAIJ, 2308 MatMatMultSymbolic_MPIAIJ_MPIAIJ, 2309 MatMatMultNumeric_MPIAIJ_MPIAIJ, 2310 MatPtAP_Basic, 2311 MatPtAPSymbolic_MPIAIJ, 2312 /*95*/ MatPtAPNumeric_MPIAIJ, 2313 0, 2314 0, 2315 0, 2316 0, 2317 /*100*/0, 2318 MatPtAPSymbolic_MPIAIJ_MPIAIJ, 2319 MatPtAPNumeric_MPIAIJ_MPIAIJ, 2320 MatConjugate_MPIAIJ, 2321 0, 2322 /*105*/MatSetValuesRow_MPIAIJ, 2323 MatRealPart_MPIAIJ, 2324 MatImaginaryPart_MPIAIJ, 2325 0, 2326 0, 2327 /*110*/0, 2328 MatGetRedundantMatrix_MPIAIJ, 2329 MatGetRowMin_MPIAIJ}; 2330 2331 /* ----------------------------------------------------------------------------------------*/ 2332 2333 EXTERN_C_BEGIN 2334 #undef __FUNCT__ 2335 #define __FUNCT__ "MatStoreValues_MPIAIJ" 2336 PetscErrorCode PETSCMAT_DLLEXPORT MatStoreValues_MPIAIJ(Mat mat) 2337 { 2338 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 2339 PetscErrorCode ierr; 2340 2341 PetscFunctionBegin; 2342 ierr = MatStoreValues(aij->A);CHKERRQ(ierr); 2343 ierr = MatStoreValues(aij->B);CHKERRQ(ierr); 2344 PetscFunctionReturn(0); 2345 } 2346 EXTERN_C_END 2347 2348 EXTERN_C_BEGIN 2349 #undef __FUNCT__ 2350 #define __FUNCT__ "MatRetrieveValues_MPIAIJ" 2351 PetscErrorCode PETSCMAT_DLLEXPORT MatRetrieveValues_MPIAIJ(Mat mat) 2352 { 2353 Mat_MPIAIJ *aij = (Mat_MPIAIJ *)mat->data; 2354 PetscErrorCode ierr; 2355 2356 PetscFunctionBegin; 2357 ierr = MatRetrieveValues(aij->A);CHKERRQ(ierr); 2358 ierr = MatRetrieveValues(aij->B);CHKERRQ(ierr); 2359 PetscFunctionReturn(0); 2360 } 2361 EXTERN_C_END 2362 2363 #include "petscpc.h" 2364 EXTERN_C_BEGIN 2365 #undef __FUNCT__ 2366 #define __FUNCT__ "MatMPIAIJSetPreallocation_MPIAIJ" 2367 PetscErrorCode PETSCMAT_DLLEXPORT MatMPIAIJSetPreallocation_MPIAIJ(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 2368 { 2369 Mat_MPIAIJ *b; 2370 PetscErrorCode ierr; 2371 PetscInt i; 2372 2373 PetscFunctionBegin; 2374 B->preallocated = PETSC_TRUE; 2375 if (d_nz == PETSC_DEFAULT || d_nz == PETSC_DECIDE) d_nz = 5; 2376 if (o_nz == PETSC_DEFAULT || o_nz == PETSC_DECIDE) o_nz = 2; 2377 if (d_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"d_nz cannot be less than 0: value %D",d_nz); 2378 if (o_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"o_nz cannot be less than 0: value %D",o_nz); 2379 2380 B->rmap.bs = B->cmap.bs = 1; 2381 ierr = PetscMapSetUp(&B->rmap);CHKERRQ(ierr); 2382 ierr = PetscMapSetUp(&B->cmap);CHKERRQ(ierr); 2383 if (d_nnz) { 2384 for (i=0; i<B->rmap.n; i++) { 2385 if (d_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"d_nnz cannot be less than 0: local row %D value %D",i,d_nnz[i]); 2386 } 2387 } 2388 if (o_nnz) { 2389 for (i=0; i<B->rmap.n; i++) { 2390 if (o_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"o_nnz cannot be less than 0: local row %D value %D",i,o_nnz[i]); 2391 } 2392 } 2393 b = (Mat_MPIAIJ*)B->data; 2394 2395 /* Explicitly create 2 MATSEQAIJ matrices. */ 2396 ierr = MatCreate(PETSC_COMM_SELF,&b->A);CHKERRQ(ierr); 2397 ierr = MatSetSizes(b->A,B->rmap.n,B->cmap.n,B->rmap.n,B->cmap.n);CHKERRQ(ierr); 2398 ierr = MatSetType(b->A,MATSEQAIJ);CHKERRQ(ierr); 2399 ierr = PetscLogObjectParent(B,b->A);CHKERRQ(ierr); 2400 ierr = MatCreate(PETSC_COMM_SELF,&b->B);CHKERRQ(ierr); 2401 ierr = MatSetSizes(b->B,B->rmap.n,B->cmap.N,B->rmap.n,B->cmap.N);CHKERRQ(ierr); 2402 ierr = MatSetType(b->B,MATSEQAIJ);CHKERRQ(ierr); 2403 ierr = PetscLogObjectParent(B,b->B);CHKERRQ(ierr); 2404 2405 ierr = MatSeqAIJSetPreallocation(b->A,d_nz,d_nnz);CHKERRQ(ierr); 2406 ierr = MatSeqAIJSetPreallocation(b->B,o_nz,o_nnz);CHKERRQ(ierr); 2407 2408 PetscFunctionReturn(0); 2409 } 2410 EXTERN_C_END 2411 2412 #undef __FUNCT__ 2413 #define __FUNCT__ "MatDuplicate_MPIAIJ" 2414 PetscErrorCode MatDuplicate_MPIAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat) 2415 { 2416 Mat mat; 2417 Mat_MPIAIJ *a,*oldmat = (Mat_MPIAIJ*)matin->data; 2418 PetscErrorCode ierr; 2419 2420 PetscFunctionBegin; 2421 *newmat = 0; 2422 ierr = MatCreate(matin->comm,&mat);CHKERRQ(ierr); 2423 ierr = MatSetSizes(mat,matin->rmap.n,matin->cmap.n,matin->rmap.N,matin->cmap.N);CHKERRQ(ierr); 2424 ierr = MatSetType(mat,matin->type_name);CHKERRQ(ierr); 2425 ierr = PetscMemcpy(mat->ops,matin->ops,sizeof(struct _MatOps));CHKERRQ(ierr); 2426 a = (Mat_MPIAIJ*)mat->data; 2427 2428 mat->factor = matin->factor; 2429 mat->rmap.bs = matin->rmap.bs; 2430 mat->assembled = PETSC_TRUE; 2431 mat->insertmode = NOT_SET_VALUES; 2432 mat->preallocated = PETSC_TRUE; 2433 2434 a->size = oldmat->size; 2435 a->rank = oldmat->rank; 2436 a->donotstash = oldmat->donotstash; 2437 a->roworiented = oldmat->roworiented; 2438 a->rowindices = 0; 2439 a->rowvalues = 0; 2440 a->getrowactive = PETSC_FALSE; 2441 2442 ierr = PetscMapCopy(mat->comm,&matin->rmap,&mat->rmap);CHKERRQ(ierr); 2443 ierr = PetscMapCopy(mat->comm,&matin->cmap,&mat->cmap);CHKERRQ(ierr); 2444 2445 ierr = MatStashCreate_Private(matin->comm,1,&mat->stash);CHKERRQ(ierr); 2446 if (oldmat->colmap) { 2447 #if defined (PETSC_USE_CTABLE) 2448 ierr = PetscTableCreateCopy(oldmat->colmap,&a->colmap);CHKERRQ(ierr); 2449 #else 2450 ierr = PetscMalloc((mat->cmap.N)*sizeof(PetscInt),&a->colmap);CHKERRQ(ierr); 2451 ierr = PetscLogObjectMemory(mat,(mat->cmap.N)*sizeof(PetscInt));CHKERRQ(ierr); 2452 ierr = PetscMemcpy(a->colmap,oldmat->colmap,(mat->cmap.N)*sizeof(PetscInt));CHKERRQ(ierr); 2453 #endif 2454 } else a->colmap = 0; 2455 if (oldmat->garray) { 2456 PetscInt len; 2457 len = oldmat->B->cmap.n; 2458 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&a->garray);CHKERRQ(ierr); 2459 ierr = PetscLogObjectMemory(mat,len*sizeof(PetscInt));CHKERRQ(ierr); 2460 if (len) { ierr = PetscMemcpy(a->garray,oldmat->garray,len*sizeof(PetscInt));CHKERRQ(ierr); } 2461 } else a->garray = 0; 2462 2463 ierr = VecDuplicate(oldmat->lvec,&a->lvec);CHKERRQ(ierr); 2464 ierr = PetscLogObjectParent(mat,a->lvec);CHKERRQ(ierr); 2465 ierr = VecScatterCopy(oldmat->Mvctx,&a->Mvctx);CHKERRQ(ierr); 2466 ierr = PetscLogObjectParent(mat,a->Mvctx);CHKERRQ(ierr); 2467 ierr = MatDuplicate(oldmat->A,cpvalues,&a->A);CHKERRQ(ierr); 2468 ierr = PetscLogObjectParent(mat,a->A);CHKERRQ(ierr); 2469 ierr = MatDuplicate(oldmat->B,cpvalues,&a->B);CHKERRQ(ierr); 2470 ierr = PetscLogObjectParent(mat,a->B);CHKERRQ(ierr); 2471 ierr = PetscFListDuplicate(matin->qlist,&mat->qlist);CHKERRQ(ierr); 2472 *newmat = mat; 2473 PetscFunctionReturn(0); 2474 } 2475 2476 #include "petscsys.h" 2477 2478 #undef __FUNCT__ 2479 #define __FUNCT__ "MatLoad_MPIAIJ" 2480 PetscErrorCode MatLoad_MPIAIJ(PetscViewer viewer, MatType type,Mat *newmat) 2481 { 2482 Mat A; 2483 PetscScalar *vals,*svals; 2484 MPI_Comm comm = ((PetscObject)viewer)->comm; 2485 MPI_Status status; 2486 PetscErrorCode ierr; 2487 PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag,maxnz; 2488 PetscInt i,nz,j,rstart,rend,mmax; 2489 PetscInt header[4],*rowlengths = 0,M,N,m,*cols; 2490 PetscInt *ourlens = PETSC_NULL,*procsnz = PETSC_NULL,*offlens = PETSC_NULL,jj,*mycols,*smycols; 2491 PetscInt cend,cstart,n,*rowners; 2492 int fd; 2493 2494 PetscFunctionBegin; 2495 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 2496 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 2497 if (!rank) { 2498 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 2499 ierr = PetscBinaryRead(fd,(char *)header,4,PETSC_INT);CHKERRQ(ierr); 2500 if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object"); 2501 } 2502 2503 ierr = MPI_Bcast(header+1,3,MPIU_INT,0,comm);CHKERRQ(ierr); 2504 M = header[1]; N = header[2]; 2505 /* determine ownership of all rows */ 2506 m = M/size + ((M % size) > rank); 2507 ierr = PetscMalloc((size+1)*sizeof(PetscInt),&rowners);CHKERRQ(ierr); 2508 ierr = MPI_Allgather(&m,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);CHKERRQ(ierr); 2509 2510 /* First process needs enough room for process with most rows */ 2511 if (!rank) { 2512 mmax = rowners[1]; 2513 for (i=2; i<size; i++) { 2514 mmax = PetscMax(mmax,rowners[i]); 2515 } 2516 } else mmax = m; 2517 2518 rowners[0] = 0; 2519 for (i=2; i<=size; i++) { 2520 rowners[i] += rowners[i-1]; 2521 } 2522 rstart = rowners[rank]; 2523 rend = rowners[rank+1]; 2524 2525 /* distribute row lengths to all processors */ 2526 ierr = PetscMalloc2(mmax,PetscInt,&ourlens,mmax,PetscInt,&offlens);CHKERRQ(ierr); 2527 if (!rank) { 2528 ierr = PetscBinaryRead(fd,ourlens,m,PETSC_INT);CHKERRQ(ierr); 2529 ierr = PetscMalloc(m*sizeof(PetscInt),&rowlengths);CHKERRQ(ierr); 2530 ierr = PetscMalloc(size*sizeof(PetscInt),&procsnz);CHKERRQ(ierr); 2531 ierr = PetscMemzero(procsnz,size*sizeof(PetscInt));CHKERRQ(ierr); 2532 for (j=0; j<m; j++) { 2533 procsnz[0] += ourlens[j]; 2534 } 2535 for (i=1; i<size; i++) { 2536 ierr = PetscBinaryRead(fd,rowlengths,rowners[i+1]-rowners[i],PETSC_INT);CHKERRQ(ierr); 2537 /* calculate the number of nonzeros on each processor */ 2538 for (j=0; j<rowners[i+1]-rowners[i]; j++) { 2539 procsnz[i] += rowlengths[j]; 2540 } 2541 ierr = MPI_Send(rowlengths,rowners[i+1]-rowners[i],MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2542 } 2543 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 2544 } else { 2545 ierr = MPI_Recv(ourlens,m,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 2546 } 2547 2548 if (!rank) { 2549 /* determine max buffer needed and allocate it */ 2550 maxnz = 0; 2551 for (i=0; i<size; i++) { 2552 maxnz = PetscMax(maxnz,procsnz[i]); 2553 } 2554 ierr = PetscMalloc(maxnz*sizeof(PetscInt),&cols);CHKERRQ(ierr); 2555 2556 /* read in my part of the matrix column indices */ 2557 nz = procsnz[0]; 2558 ierr = PetscMalloc(nz*sizeof(PetscInt),&mycols);CHKERRQ(ierr); 2559 ierr = PetscBinaryRead(fd,mycols,nz,PETSC_INT);CHKERRQ(ierr); 2560 2561 /* read in every one elses and ship off */ 2562 for (i=1; i<size; i++) { 2563 nz = procsnz[i]; 2564 ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr); 2565 ierr = MPI_Send(cols,nz,MPIU_INT,i,tag,comm);CHKERRQ(ierr); 2566 } 2567 ierr = PetscFree(cols);CHKERRQ(ierr); 2568 } else { 2569 /* determine buffer space needed for message */ 2570 nz = 0; 2571 for (i=0; i<m; i++) { 2572 nz += ourlens[i]; 2573 } 2574 ierr = PetscMalloc(nz*sizeof(PetscInt),&mycols);CHKERRQ(ierr); 2575 2576 /* receive message of column indices*/ 2577 ierr = MPI_Recv(mycols,nz,MPIU_INT,0,tag,comm,&status);CHKERRQ(ierr); 2578 ierr = MPI_Get_count(&status,MPIU_INT,&maxnz);CHKERRQ(ierr); 2579 if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file"); 2580 } 2581 2582 /* determine column ownership if matrix is not square */ 2583 if (N != M) { 2584 n = N/size + ((N % size) > rank); 2585 ierr = MPI_Scan(&n,&cend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 2586 cstart = cend - n; 2587 } else { 2588 cstart = rstart; 2589 cend = rend; 2590 n = cend - cstart; 2591 } 2592 2593 /* loop over local rows, determining number of off diagonal entries */ 2594 ierr = PetscMemzero(offlens,m*sizeof(PetscInt));CHKERRQ(ierr); 2595 jj = 0; 2596 for (i=0; i<m; i++) { 2597 for (j=0; j<ourlens[i]; j++) { 2598 if (mycols[jj] < cstart || mycols[jj] >= cend) offlens[i]++; 2599 jj++; 2600 } 2601 } 2602 2603 /* create our matrix */ 2604 for (i=0; i<m; i++) { 2605 ourlens[i] -= offlens[i]; 2606 } 2607 ierr = MatCreate(comm,&A);CHKERRQ(ierr); 2608 ierr = MatSetSizes(A,m,n,M,N);CHKERRQ(ierr); 2609 ierr = MatSetType(A,type);CHKERRQ(ierr); 2610 ierr = MatMPIAIJSetPreallocation(A,0,ourlens,0,offlens);CHKERRQ(ierr); 2611 2612 for (i=0; i<m; i++) { 2613 ourlens[i] += offlens[i]; 2614 } 2615 2616 if (!rank) { 2617 ierr = PetscMalloc((maxnz+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 2618 2619 /* read in my part of the matrix numerical values */ 2620 nz = procsnz[0]; 2621 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2622 2623 /* insert into matrix */ 2624 jj = rstart; 2625 smycols = mycols; 2626 svals = vals; 2627 for (i=0; i<m; i++) { 2628 ierr = MatSetValues_MPIAIJ(A,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2629 smycols += ourlens[i]; 2630 svals += ourlens[i]; 2631 jj++; 2632 } 2633 2634 /* read in other processors and ship out */ 2635 for (i=1; i<size; i++) { 2636 nz = procsnz[i]; 2637 ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr); 2638 ierr = MPI_Send(vals,nz,MPIU_SCALAR,i,A->tag,comm);CHKERRQ(ierr); 2639 } 2640 ierr = PetscFree(procsnz);CHKERRQ(ierr); 2641 } else { 2642 /* receive numeric values */ 2643 ierr = PetscMalloc((nz+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 2644 2645 /* receive message of values*/ 2646 ierr = MPI_Recv(vals,nz,MPIU_SCALAR,0,A->tag,comm,&status);CHKERRQ(ierr); 2647 ierr = MPI_Get_count(&status,MPIU_SCALAR,&maxnz);CHKERRQ(ierr); 2648 if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file"); 2649 2650 /* insert into matrix */ 2651 jj = rstart; 2652 smycols = mycols; 2653 svals = vals; 2654 for (i=0; i<m; i++) { 2655 ierr = MatSetValues_MPIAIJ(A,1,&jj,ourlens[i],smycols,svals,INSERT_VALUES);CHKERRQ(ierr); 2656 smycols += ourlens[i]; 2657 svals += ourlens[i]; 2658 jj++; 2659 } 2660 } 2661 ierr = PetscFree2(ourlens,offlens);CHKERRQ(ierr); 2662 ierr = PetscFree(vals);CHKERRQ(ierr); 2663 ierr = PetscFree(mycols);CHKERRQ(ierr); 2664 ierr = PetscFree(rowners);CHKERRQ(ierr); 2665 2666 ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2667 ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2668 *newmat = A; 2669 PetscFunctionReturn(0); 2670 } 2671 2672 #undef __FUNCT__ 2673 #define __FUNCT__ "MatGetSubMatrix_MPIAIJ" 2674 /* 2675 Not great since it makes two copies of the submatrix, first an SeqAIJ 2676 in local and then by concatenating the local matrices the end result. 2677 Writing it directly would be much like MatGetSubMatrices_MPIAIJ() 2678 */ 2679 PetscErrorCode MatGetSubMatrix_MPIAIJ(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat) 2680 { 2681 PetscErrorCode ierr; 2682 PetscMPIInt rank,size; 2683 PetscInt i,m,n,rstart,row,rend,nz,*cwork,j; 2684 PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal; 2685 Mat *local,M,Mreuse; 2686 PetscScalar *vwork,*aa; 2687 MPI_Comm comm = mat->comm; 2688 Mat_SeqAIJ *aij; 2689 2690 2691 PetscFunctionBegin; 2692 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 2693 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 2694 2695 if (call == MAT_REUSE_MATRIX) { 2696 ierr = PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject *)&Mreuse);CHKERRQ(ierr); 2697 if (!Mreuse) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse"); 2698 local = &Mreuse; 2699 ierr = MatGetSubMatrices(mat,1,&isrow,&iscol,MAT_REUSE_MATRIX,&local);CHKERRQ(ierr); 2700 } else { 2701 ierr = MatGetSubMatrices(mat,1,&isrow,&iscol,MAT_INITIAL_MATRIX,&local);CHKERRQ(ierr); 2702 Mreuse = *local; 2703 ierr = PetscFree(local);CHKERRQ(ierr); 2704 } 2705 2706 /* 2707 m - number of local rows 2708 n - number of columns (same on all processors) 2709 rstart - first row in new global matrix generated 2710 */ 2711 ierr = MatGetSize(Mreuse,&m,&n);CHKERRQ(ierr); 2712 if (call == MAT_INITIAL_MATRIX) { 2713 aij = (Mat_SeqAIJ*)(Mreuse)->data; 2714 ii = aij->i; 2715 jj = aij->j; 2716 2717 /* 2718 Determine the number of non-zeros in the diagonal and off-diagonal 2719 portions of the matrix in order to do correct preallocation 2720 */ 2721 2722 /* first get start and end of "diagonal" columns */ 2723 if (csize == PETSC_DECIDE) { 2724 ierr = ISGetSize(isrow,&mglobal);CHKERRQ(ierr); 2725 if (mglobal == n) { /* square matrix */ 2726 nlocal = m; 2727 } else { 2728 nlocal = n/size + ((n % size) > rank); 2729 } 2730 } else { 2731 nlocal = csize; 2732 } 2733 ierr = MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 2734 rstart = rend - nlocal; 2735 if (rank == size - 1 && rend != n) { 2736 SETERRQ2(PETSC_ERR_ARG_SIZ,"Local column sizes %D do not add up to total number of columns %D",rend,n); 2737 } 2738 2739 /* next, compute all the lengths */ 2740 ierr = PetscMalloc((2*m+1)*sizeof(PetscInt),&dlens);CHKERRQ(ierr); 2741 olens = dlens + m; 2742 for (i=0; i<m; i++) { 2743 jend = ii[i+1] - ii[i]; 2744 olen = 0; 2745 dlen = 0; 2746 for (j=0; j<jend; j++) { 2747 if (*jj < rstart || *jj >= rend) olen++; 2748 else dlen++; 2749 jj++; 2750 } 2751 olens[i] = olen; 2752 dlens[i] = dlen; 2753 } 2754 ierr = MatCreate(comm,&M);CHKERRQ(ierr); 2755 ierr = MatSetSizes(M,m,nlocal,PETSC_DECIDE,n);CHKERRQ(ierr); 2756 ierr = MatSetType(M,mat->type_name);CHKERRQ(ierr); 2757 ierr = MatMPIAIJSetPreallocation(M,0,dlens,0,olens);CHKERRQ(ierr); 2758 ierr = PetscFree(dlens);CHKERRQ(ierr); 2759 } else { 2760 PetscInt ml,nl; 2761 2762 M = *newmat; 2763 ierr = MatGetLocalSize(M,&ml,&nl);CHKERRQ(ierr); 2764 if (ml != m) SETERRQ(PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request"); 2765 ierr = MatZeroEntries(M);CHKERRQ(ierr); 2766 /* 2767 The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly, 2768 rather than the slower MatSetValues(). 2769 */ 2770 M->was_assembled = PETSC_TRUE; 2771 M->assembled = PETSC_FALSE; 2772 } 2773 ierr = MatGetOwnershipRange(M,&rstart,&rend);CHKERRQ(ierr); 2774 aij = (Mat_SeqAIJ*)(Mreuse)->data; 2775 ii = aij->i; 2776 jj = aij->j; 2777 aa = aij->a; 2778 for (i=0; i<m; i++) { 2779 row = rstart + i; 2780 nz = ii[i+1] - ii[i]; 2781 cwork = jj; jj += nz; 2782 vwork = aa; aa += nz; 2783 ierr = MatSetValues_MPIAIJ(M,1,&row,nz,cwork,vwork,INSERT_VALUES);CHKERRQ(ierr); 2784 } 2785 2786 ierr = MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2787 ierr = MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2788 *newmat = M; 2789 2790 /* save submatrix used in processor for next request */ 2791 if (call == MAT_INITIAL_MATRIX) { 2792 ierr = PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);CHKERRQ(ierr); 2793 ierr = PetscObjectDereference((PetscObject)Mreuse);CHKERRQ(ierr); 2794 } 2795 2796 PetscFunctionReturn(0); 2797 } 2798 2799 EXTERN_C_BEGIN 2800 #undef __FUNCT__ 2801 #define __FUNCT__ "MatMPIAIJSetPreallocationCSR_MPIAIJ" 2802 PetscErrorCode PETSCMAT_DLLEXPORT MatMPIAIJSetPreallocationCSR_MPIAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 2803 { 2804 PetscInt m,cstart, cend,j,nnz,i,d; 2805 PetscInt *d_nnz,*o_nnz,nnz_max = 0,rstart,ii; 2806 const PetscInt *JJ; 2807 PetscScalar *values; 2808 PetscErrorCode ierr; 2809 2810 PetscFunctionBegin; 2811 if (Ii[0]) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Ii[0] must be 0 it is %D",Ii[0]); 2812 2813 B->rmap.bs = B->cmap.bs = 1; 2814 ierr = PetscMapSetUp(&B->rmap);CHKERRQ(ierr); 2815 ierr = PetscMapSetUp(&B->cmap);CHKERRQ(ierr); 2816 m = B->rmap.n; 2817 cstart = B->cmap.rstart; 2818 cend = B->cmap.rend; 2819 rstart = B->rmap.rstart; 2820 2821 ierr = PetscMalloc((2*m+1)*sizeof(PetscInt),&d_nnz);CHKERRQ(ierr); 2822 o_nnz = d_nnz + m; 2823 2824 for (i=0; i<m; i++) { 2825 nnz = Ii[i+1]- Ii[i]; 2826 JJ = J + Ii[i]; 2827 nnz_max = PetscMax(nnz_max,nnz); 2828 if (nnz < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Local row %D has a negative %D number of columns",i,nnz); 2829 for (j=0; j<nnz; j++) { 2830 if (*JJ >= cstart) break; 2831 JJ++; 2832 } 2833 d = 0; 2834 for (; j<nnz; j++) { 2835 if (*JJ++ >= cend) break; 2836 d++; 2837 } 2838 d_nnz[i] = d; 2839 o_nnz[i] = nnz - d; 2840 } 2841 ierr = MatMPIAIJSetPreallocation(B,0,d_nnz,0,o_nnz);CHKERRQ(ierr); 2842 ierr = PetscFree(d_nnz);CHKERRQ(ierr); 2843 2844 if (v) values = (PetscScalar*)v; 2845 else { 2846 ierr = PetscMalloc((nnz_max+1)*sizeof(PetscScalar),&values);CHKERRQ(ierr); 2847 ierr = PetscMemzero(values,nnz_max*sizeof(PetscScalar));CHKERRQ(ierr); 2848 } 2849 2850 for (i=0; i<m; i++) { 2851 ii = i + rstart; 2852 nnz = Ii[i+1]- Ii[i]; 2853 ierr = MatSetValues_MPIAIJ(B,1,&ii,nnz,J+Ii[i],values+(v ? Ii[i] : 0),INSERT_VALUES);CHKERRQ(ierr); 2854 } 2855 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2856 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2857 2858 if (!v) { 2859 ierr = PetscFree(values);CHKERRQ(ierr); 2860 } 2861 PetscFunctionReturn(0); 2862 } 2863 EXTERN_C_END 2864 2865 #undef __FUNCT__ 2866 #define __FUNCT__ "MatMPIAIJSetPreallocationCSR" 2867 /*@ 2868 MatMPIAIJSetPreallocationCSR - Allocates memory for a sparse parallel matrix in AIJ format 2869 (the default parallel PETSc format). 2870 2871 Collective on MPI_Comm 2872 2873 Input Parameters: 2874 + B - the matrix 2875 . i - the indices into j for the start of each local row (starts with zero) 2876 . j - the column indices for each local row (starts with zero) these must be sorted for each row 2877 - v - optional values in the matrix 2878 2879 Level: developer 2880 2881 Notes: this actually copies the values from i[], j[], and a[] to put them into PETSc's internal 2882 storage format. Thus changing the values in a[] after this call will not effect the matrix values. 2883 2884 .keywords: matrix, aij, compressed row, sparse, parallel 2885 2886 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatCreateMPIAIJ(), MPIAIJ, 2887 MatCreateSeqAIJWithArrays(), MatCreateMPIAIJWithSplitArrays() 2888 @*/ 2889 PetscErrorCode PETSCMAT_DLLEXPORT MatMPIAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[], const PetscScalar v[]) 2890 { 2891 PetscErrorCode ierr,(*f)(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]); 2892 2893 PetscFunctionBegin; 2894 ierr = PetscObjectQueryFunction((PetscObject)B,"MatMPIAIJSetPreallocationCSR_C",(void (**)(void))&f);CHKERRQ(ierr); 2895 if (f) { 2896 ierr = (*f)(B,i,j,v);CHKERRQ(ierr); 2897 } 2898 PetscFunctionReturn(0); 2899 } 2900 2901 #undef __FUNCT__ 2902 #define __FUNCT__ "MatMPIAIJSetPreallocation" 2903 /*@C 2904 MatMPIAIJSetPreallocation - Preallocates memory for a sparse parallel matrix in AIJ format 2905 (the default parallel PETSc format). For good matrix assembly performance 2906 the user should preallocate the matrix storage by setting the parameters 2907 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 2908 performance can be increased by more than a factor of 50. 2909 2910 Collective on MPI_Comm 2911 2912 Input Parameters: 2913 + A - the matrix 2914 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 2915 (same value is used for all local rows) 2916 . d_nnz - array containing the number of nonzeros in the various rows of the 2917 DIAGONAL portion of the local submatrix (possibly different for each row) 2918 or PETSC_NULL, if d_nz is used to specify the nonzero structure. 2919 The size of this array is equal to the number of local rows, i.e 'm'. 2920 You must leave room for the diagonal entry even if it is zero. 2921 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 2922 submatrix (same value is used for all local rows). 2923 - o_nnz - array containing the number of nonzeros in the various rows of the 2924 OFF-DIAGONAL portion of the local submatrix (possibly different for 2925 each row) or PETSC_NULL, if o_nz is used to specify the nonzero 2926 structure. The size of this array is equal to the number 2927 of local rows, i.e 'm'. 2928 2929 If the *_nnz parameter is given then the *_nz parameter is ignored 2930 2931 The AIJ format (also called the Yale sparse matrix format or 2932 compressed row storage (CSR)), is fully compatible with standard Fortran 77 2933 storage. The stored row and column indices begin with zero. See the users manual for details. 2934 2935 The parallel matrix is partitioned such that the first m0 rows belong to 2936 process 0, the next m1 rows belong to process 1, the next m2 rows belong 2937 to process 2 etc.. where m0,m1,m2... are the input parameter 'm'. 2938 2939 The DIAGONAL portion of the local submatrix of a processor can be defined 2940 as the submatrix which is obtained by extraction the part corresponding 2941 to the rows r1-r2 and columns r1-r2 of the global matrix, where r1 is the 2942 first row that belongs to the processor, and r2 is the last row belonging 2943 to the this processor. This is a square mxm matrix. The remaining portion 2944 of the local submatrix (mxN) constitute the OFF-DIAGONAL portion. 2945 2946 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 2947 2948 Example usage: 2949 2950 Consider the following 8x8 matrix with 34 non-zero values, that is 2951 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 2952 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 2953 as follows: 2954 2955 .vb 2956 1 2 0 | 0 3 0 | 0 4 2957 Proc0 0 5 6 | 7 0 0 | 8 0 2958 9 0 10 | 11 0 0 | 12 0 2959 ------------------------------------- 2960 13 0 14 | 15 16 17 | 0 0 2961 Proc1 0 18 0 | 19 20 21 | 0 0 2962 0 0 0 | 22 23 0 | 24 0 2963 ------------------------------------- 2964 Proc2 25 26 27 | 0 0 28 | 29 0 2965 30 0 0 | 31 32 33 | 0 34 2966 .ve 2967 2968 This can be represented as a collection of submatrices as: 2969 2970 .vb 2971 A B C 2972 D E F 2973 G H I 2974 .ve 2975 2976 Where the submatrices A,B,C are owned by proc0, D,E,F are 2977 owned by proc1, G,H,I are owned by proc2. 2978 2979 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 2980 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 2981 The 'M','N' parameters are 8,8, and have the same values on all procs. 2982 2983 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 2984 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 2985 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 2986 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 2987 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 2988 matrix, ans [DF] as another SeqAIJ matrix. 2989 2990 When d_nz, o_nz parameters are specified, d_nz storage elements are 2991 allocated for every row of the local diagonal submatrix, and o_nz 2992 storage locations are allocated for every row of the OFF-DIAGONAL submat. 2993 One way to choose d_nz and o_nz is to use the max nonzerors per local 2994 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 2995 In this case, the values of d_nz,o_nz are: 2996 .vb 2997 proc0 : dnz = 2, o_nz = 2 2998 proc1 : dnz = 3, o_nz = 2 2999 proc2 : dnz = 1, o_nz = 4 3000 .ve 3001 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3002 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3003 for proc3. i.e we are using 12+15+10=37 storage locations to store 3004 34 values. 3005 3006 When d_nnz, o_nnz parameters are specified, the storage is specified 3007 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3008 In the above case the values for d_nnz,o_nnz are: 3009 .vb 3010 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3011 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3012 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3013 .ve 3014 Here the space allocated is sum of all the above values i.e 34, and 3015 hence pre-allocation is perfect. 3016 3017 Level: intermediate 3018 3019 .keywords: matrix, aij, compressed row, sparse, parallel 3020 3021 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateMPIAIJ(), MatMPIAIJSetPreallocationCSR(), 3022 MPIAIJ 3023 @*/ 3024 PetscErrorCode PETSCMAT_DLLEXPORT MatMPIAIJSetPreallocation(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[]) 3025 { 3026 PetscErrorCode ierr,(*f)(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[]); 3027 3028 PetscFunctionBegin; 3029 ierr = PetscObjectQueryFunction((PetscObject)B,"MatMPIAIJSetPreallocation_C",(void (**)(void))&f);CHKERRQ(ierr); 3030 if (f) { 3031 ierr = (*f)(B,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr); 3032 } 3033 PetscFunctionReturn(0); 3034 } 3035 3036 #undef __FUNCT__ 3037 #define __FUNCT__ "MatCreateMPIAIJWithArrays" 3038 /*@ 3039 MatCreateMPIAIJWithArrays - creates a MPI AIJ matrix using arrays that contain in standard 3040 CSR format the local rows. 3041 3042 Collective on MPI_Comm 3043 3044 Input Parameters: 3045 + comm - MPI communicator 3046 . m - number of local rows (Cannot be PETSC_DECIDE) 3047 . n - This value should be the same as the local size used in creating the 3048 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3049 calculated if N is given) For square matrices n is almost always m. 3050 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3051 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3052 . i - row indices 3053 . j - column indices 3054 - a - matrix values 3055 3056 Output Parameter: 3057 . mat - the matrix 3058 3059 Level: intermediate 3060 3061 Notes: 3062 The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc; 3063 thus you CANNOT change the matrix entries by changing the values of a[] after you have 3064 called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays. 3065 3066 The i and j indices are 0 based 3067 3068 .keywords: matrix, aij, compressed row, sparse, parallel 3069 3070 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 3071 MPIAIJ, MatCreateMPIAIJ(), MatCreateMPIAIJWithSplitArrays() 3072 @*/ 3073 PetscErrorCode PETSCMAT_DLLEXPORT MatCreateMPIAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,const PetscInt i[],const PetscInt j[],const PetscScalar a[],Mat *mat) 3074 { 3075 PetscErrorCode ierr; 3076 3077 PetscFunctionBegin; 3078 if (i[0]) { 3079 SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 3080 } 3081 if (m < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 3082 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 3083 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 3084 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 3085 ierr = MatMPIAIJSetPreallocationCSR(*mat,i,j,a);CHKERRQ(ierr); 3086 PetscFunctionReturn(0); 3087 } 3088 3089 #undef __FUNCT__ 3090 #define __FUNCT__ "MatCreateMPIAIJ" 3091 /*@C 3092 MatCreateMPIAIJ - Creates a sparse parallel matrix in AIJ format 3093 (the default parallel PETSc format). For good matrix assembly performance 3094 the user should preallocate the matrix storage by setting the parameters 3095 d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately, 3096 performance can be increased by more than a factor of 50. 3097 3098 Collective on MPI_Comm 3099 3100 Input Parameters: 3101 + comm - MPI communicator 3102 . m - number of local rows (or PETSC_DECIDE to have calculated if M is given) 3103 This value should be the same as the local size used in creating the 3104 y vector for the matrix-vector product y = Ax. 3105 . n - This value should be the same as the local size used in creating the 3106 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 3107 calculated if N is given) For square matrices n is almost always m. 3108 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 3109 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 3110 . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix 3111 (same value is used for all local rows) 3112 . d_nnz - array containing the number of nonzeros in the various rows of the 3113 DIAGONAL portion of the local submatrix (possibly different for each row) 3114 or PETSC_NULL, if d_nz is used to specify the nonzero structure. 3115 The size of this array is equal to the number of local rows, i.e 'm'. 3116 You must leave room for the diagonal entry even if it is zero. 3117 . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local 3118 submatrix (same value is used for all local rows). 3119 - o_nnz - array containing the number of nonzeros in the various rows of the 3120 OFF-DIAGONAL portion of the local submatrix (possibly different for 3121 each row) or PETSC_NULL, if o_nz is used to specify the nonzero 3122 structure. The size of this array is equal to the number 3123 of local rows, i.e 'm'. 3124 3125 Output Parameter: 3126 . A - the matrix 3127 3128 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3129 MatXXXXSetPreallocation() paradgm instead of this routine directly. This is definitely 3130 true if you plan to use the external direct solvers such as SuperLU, MUMPS or Spooles. 3131 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3132 3133 Notes: 3134 If the *_nnz parameter is given then the *_nz parameter is ignored 3135 3136 m,n,M,N parameters specify the size of the matrix, and its partitioning across 3137 processors, while d_nz,d_nnz,o_nz,o_nnz parameters specify the approximate 3138 storage requirements for this matrix. 3139 3140 If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one 3141 processor than it must be used on all processors that share the object for 3142 that argument. 3143 3144 The user MUST specify either the local or global matrix dimensions 3145 (possibly both). 3146 3147 The parallel matrix is partitioned across processors such that the 3148 first m0 rows belong to process 0, the next m1 rows belong to 3149 process 1, the next m2 rows belong to process 2 etc.. where 3150 m0,m1,m2,.. are the input parameter 'm'. i.e each processor stores 3151 values corresponding to [m x N] submatrix. 3152 3153 The columns are logically partitioned with the n0 columns belonging 3154 to 0th partition, the next n1 columns belonging to the next 3155 partition etc.. where n0,n1,n2... are the the input parameter 'n'. 3156 3157 The DIAGONAL portion of the local submatrix on any given processor 3158 is the submatrix corresponding to the rows and columns m,n 3159 corresponding to the given processor. i.e diagonal matrix on 3160 process 0 is [m0 x n0], diagonal matrix on process 1 is [m1 x n1] 3161 etc. The remaining portion of the local submatrix [m x (N-n)] 3162 constitute the OFF-DIAGONAL portion. The example below better 3163 illustrates this concept. 3164 3165 For a square global matrix we define each processor's diagonal portion 3166 to be its local rows and the corresponding columns (a square submatrix); 3167 each processor's off-diagonal portion encompasses the remainder of the 3168 local matrix (a rectangular submatrix). 3169 3170 If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored. 3171 3172 When calling this routine with a single process communicator, a matrix of 3173 type SEQAIJ is returned. If a matrix of type MPIAIJ is desired for this 3174 type of communicator, use the construction mechanism: 3175 MatCreate(...,&A); MatSetType(A,MPIAIJ); MatMPIAIJSetPreallocation(A,...); 3176 3177 By default, this format uses inodes (identical nodes) when possible. 3178 We search for consecutive rows with the same nonzero structure, thereby 3179 reusing matrix information to achieve increased efficiency. 3180 3181 Options Database Keys: 3182 + -mat_no_inode - Do not use inodes 3183 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3184 - -mat_aij_oneindex - Internally use indexing starting at 1 3185 rather than 0. Note that when calling MatSetValues(), 3186 the user still MUST index entries starting at 0! 3187 3188 3189 Example usage: 3190 3191 Consider the following 8x8 matrix with 34 non-zero values, that is 3192 assembled across 3 processors. Lets assume that proc0 owns 3 rows, 3193 proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown 3194 as follows: 3195 3196 .vb 3197 1 2 0 | 0 3 0 | 0 4 3198 Proc0 0 5 6 | 7 0 0 | 8 0 3199 9 0 10 | 11 0 0 | 12 0 3200 ------------------------------------- 3201 13 0 14 | 15 16 17 | 0 0 3202 Proc1 0 18 0 | 19 20 21 | 0 0 3203 0 0 0 | 22 23 0 | 24 0 3204 ------------------------------------- 3205 Proc2 25 26 27 | 0 0 28 | 29 0 3206 30 0 0 | 31 32 33 | 0 34 3207 .ve 3208 3209 This can be represented as a collection of submatrices as: 3210 3211 .vb 3212 A B C 3213 D E F 3214 G H I 3215 .ve 3216 3217 Where the submatrices A,B,C are owned by proc0, D,E,F are 3218 owned by proc1, G,H,I are owned by proc2. 3219 3220 The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3221 The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively. 3222 The 'M','N' parameters are 8,8, and have the same values on all procs. 3223 3224 The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are 3225 submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices 3226 corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively. 3227 Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL 3228 part as SeqAIJ matrices. for eg: proc1 will store [E] as a SeqAIJ 3229 matrix, ans [DF] as another SeqAIJ matrix. 3230 3231 When d_nz, o_nz parameters are specified, d_nz storage elements are 3232 allocated for every row of the local diagonal submatrix, and o_nz 3233 storage locations are allocated for every row of the OFF-DIAGONAL submat. 3234 One way to choose d_nz and o_nz is to use the max nonzerors per local 3235 rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices. 3236 In this case, the values of d_nz,o_nz are: 3237 .vb 3238 proc0 : dnz = 2, o_nz = 2 3239 proc1 : dnz = 3, o_nz = 2 3240 proc2 : dnz = 1, o_nz = 4 3241 .ve 3242 We are allocating m*(d_nz+o_nz) storage locations for every proc. This 3243 translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10 3244 for proc3. i.e we are using 12+15+10=37 storage locations to store 3245 34 values. 3246 3247 When d_nnz, o_nnz parameters are specified, the storage is specified 3248 for every row, coresponding to both DIAGONAL and OFF-DIAGONAL submatrices. 3249 In the above case the values for d_nnz,o_nnz are: 3250 .vb 3251 proc0: d_nnz = [2,2,2] and o_nnz = [2,2,2] 3252 proc1: d_nnz = [3,3,2] and o_nnz = [2,1,1] 3253 proc2: d_nnz = [1,1] and o_nnz = [4,4] 3254 .ve 3255 Here the space allocated is sum of all the above values i.e 34, and 3256 hence pre-allocation is perfect. 3257 3258 Level: intermediate 3259 3260 .keywords: matrix, aij, compressed row, sparse, parallel 3261 3262 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 3263 MPIAIJ, MatCreateMPIAIJWithArrays() 3264 @*/ 3265 PetscErrorCode PETSCMAT_DLLEXPORT 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) 3266 { 3267 PetscErrorCode ierr; 3268 PetscMPIInt size; 3269 3270 PetscFunctionBegin; 3271 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3272 ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr); 3273 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3274 if (size > 1) { 3275 ierr = MatSetType(*A,MATMPIAIJ);CHKERRQ(ierr); 3276 ierr = MatMPIAIJSetPreallocation(*A,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr); 3277 } else { 3278 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3279 ierr = MatSeqAIJSetPreallocation(*A,d_nz,d_nnz);CHKERRQ(ierr); 3280 } 3281 PetscFunctionReturn(0); 3282 } 3283 3284 #undef __FUNCT__ 3285 #define __FUNCT__ "MatMPIAIJGetSeqAIJ" 3286 PetscErrorCode PETSCMAT_DLLEXPORT MatMPIAIJGetSeqAIJ(Mat A,Mat *Ad,Mat *Ao,PetscInt *colmap[]) 3287 { 3288 Mat_MPIAIJ *a = (Mat_MPIAIJ *)A->data; 3289 3290 PetscFunctionBegin; 3291 *Ad = a->A; 3292 *Ao = a->B; 3293 *colmap = a->garray; 3294 PetscFunctionReturn(0); 3295 } 3296 3297 #undef __FUNCT__ 3298 #define __FUNCT__ "MatSetColoring_MPIAIJ" 3299 PetscErrorCode MatSetColoring_MPIAIJ(Mat A,ISColoring coloring) 3300 { 3301 PetscErrorCode ierr; 3302 PetscInt i; 3303 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 3304 3305 PetscFunctionBegin; 3306 if (coloring->ctype == IS_COLORING_GLOBAL) { 3307 ISColoringValue *allcolors,*colors; 3308 ISColoring ocoloring; 3309 3310 /* set coloring for diagonal portion */ 3311 ierr = MatSetColoring_SeqAIJ(a->A,coloring);CHKERRQ(ierr); 3312 3313 /* set coloring for off-diagonal portion */ 3314 ierr = ISAllGatherColors(A->comm,coloring->n,coloring->colors,PETSC_NULL,&allcolors);CHKERRQ(ierr); 3315 ierr = PetscMalloc((a->B->cmap.n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 3316 for (i=0; i<a->B->cmap.n; i++) { 3317 colors[i] = allcolors[a->garray[i]]; 3318 } 3319 ierr = PetscFree(allcolors);CHKERRQ(ierr); 3320 ierr = ISColoringCreate(MPI_COMM_SELF,coloring->n,a->B->cmap.n,colors,&ocoloring);CHKERRQ(ierr); 3321 ierr = MatSetColoring_SeqAIJ(a->B,ocoloring);CHKERRQ(ierr); 3322 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 3323 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 3324 ISColoringValue *colors; 3325 PetscInt *larray; 3326 ISColoring ocoloring; 3327 3328 /* set coloring for diagonal portion */ 3329 ierr = PetscMalloc((a->A->cmap.n+1)*sizeof(PetscInt),&larray);CHKERRQ(ierr); 3330 for (i=0; i<a->A->cmap.n; i++) { 3331 larray[i] = i + A->cmap.rstart; 3332 } 3333 ierr = ISGlobalToLocalMappingApply(A->mapping,IS_GTOLM_MASK,a->A->cmap.n,larray,PETSC_NULL,larray);CHKERRQ(ierr); 3334 ierr = PetscMalloc((a->A->cmap.n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 3335 for (i=0; i<a->A->cmap.n; i++) { 3336 colors[i] = coloring->colors[larray[i]]; 3337 } 3338 ierr = PetscFree(larray);CHKERRQ(ierr); 3339 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,a->A->cmap.n,colors,&ocoloring);CHKERRQ(ierr); 3340 ierr = MatSetColoring_SeqAIJ(a->A,ocoloring);CHKERRQ(ierr); 3341 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 3342 3343 /* set coloring for off-diagonal portion */ 3344 ierr = PetscMalloc((a->B->cmap.n+1)*sizeof(PetscInt),&larray);CHKERRQ(ierr); 3345 ierr = ISGlobalToLocalMappingApply(A->mapping,IS_GTOLM_MASK,a->B->cmap.n,a->garray,PETSC_NULL,larray);CHKERRQ(ierr); 3346 ierr = PetscMalloc((a->B->cmap.n+1)*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 3347 for (i=0; i<a->B->cmap.n; i++) { 3348 colors[i] = coloring->colors[larray[i]]; 3349 } 3350 ierr = PetscFree(larray);CHKERRQ(ierr); 3351 ierr = ISColoringCreate(MPI_COMM_SELF,coloring->n,a->B->cmap.n,colors,&ocoloring);CHKERRQ(ierr); 3352 ierr = MatSetColoring_SeqAIJ(a->B,ocoloring);CHKERRQ(ierr); 3353 ierr = ISColoringDestroy(ocoloring);CHKERRQ(ierr); 3354 } else { 3355 SETERRQ1(PETSC_ERR_SUP,"No support ISColoringType %d",(int)coloring->ctype); 3356 } 3357 3358 PetscFunctionReturn(0); 3359 } 3360 3361 #if defined(PETSC_HAVE_ADIC) 3362 #undef __FUNCT__ 3363 #define __FUNCT__ "MatSetValuesAdic_MPIAIJ" 3364 PetscErrorCode MatSetValuesAdic_MPIAIJ(Mat A,void *advalues) 3365 { 3366 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 3367 PetscErrorCode ierr; 3368 3369 PetscFunctionBegin; 3370 ierr = MatSetValuesAdic_SeqAIJ(a->A,advalues);CHKERRQ(ierr); 3371 ierr = MatSetValuesAdic_SeqAIJ(a->B,advalues);CHKERRQ(ierr); 3372 PetscFunctionReturn(0); 3373 } 3374 #endif 3375 3376 #undef __FUNCT__ 3377 #define __FUNCT__ "MatSetValuesAdifor_MPIAIJ" 3378 PetscErrorCode MatSetValuesAdifor_MPIAIJ(Mat A,PetscInt nl,void *advalues) 3379 { 3380 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 3381 PetscErrorCode ierr; 3382 3383 PetscFunctionBegin; 3384 ierr = MatSetValuesAdifor_SeqAIJ(a->A,nl,advalues);CHKERRQ(ierr); 3385 ierr = MatSetValuesAdifor_SeqAIJ(a->B,nl,advalues);CHKERRQ(ierr); 3386 PetscFunctionReturn(0); 3387 } 3388 3389 #undef __FUNCT__ 3390 #define __FUNCT__ "MatMerge" 3391 /*@ 3392 MatMerge - Creates a single large PETSc matrix by concatinating sequential 3393 matrices from each processor 3394 3395 Collective on MPI_Comm 3396 3397 Input Parameters: 3398 + comm - the communicators the parallel matrix will live on 3399 . inmat - the input sequential matrices 3400 . n - number of local columns (or PETSC_DECIDE) 3401 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 3402 3403 Output Parameter: 3404 . outmat - the parallel matrix generated 3405 3406 Level: advanced 3407 3408 Notes: The number of columns of the matrix in EACH processor MUST be the same. 3409 3410 @*/ 3411 PetscErrorCode PETSCMAT_DLLEXPORT MatMerge(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat) 3412 { 3413 PetscErrorCode ierr; 3414 PetscInt m,N,i,rstart,nnz,Ii,*dnz,*onz; 3415 PetscInt *indx; 3416 PetscScalar *values; 3417 3418 PetscFunctionBegin; 3419 ierr = MatGetSize(inmat,&m,&N);CHKERRQ(ierr); 3420 if (scall == MAT_INITIAL_MATRIX){ 3421 /* count nonzeros in each row, for diagonal and off diagonal portion of matrix */ 3422 if (n == PETSC_DECIDE){ 3423 ierr = PetscSplitOwnership(comm,&n,&N);CHKERRQ(ierr); 3424 } 3425 ierr = MPI_Scan(&m, &rstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr); 3426 rstart -= m; 3427 3428 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 3429 for (i=0;i<m;i++) { 3430 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,PETSC_NULL);CHKERRQ(ierr); 3431 ierr = MatPreallocateSet(i+rstart,nnz,indx,dnz,onz);CHKERRQ(ierr); 3432 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,PETSC_NULL);CHKERRQ(ierr); 3433 } 3434 /* This routine will ONLY return MPIAIJ type matrix */ 3435 ierr = MatCreate(comm,outmat);CHKERRQ(ierr); 3436 ierr = MatSetSizes(*outmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 3437 ierr = MatSetType(*outmat,MATMPIAIJ);CHKERRQ(ierr); 3438 ierr = MatMPIAIJSetPreallocation(*outmat,0,dnz,0,onz);CHKERRQ(ierr); 3439 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 3440 3441 } else if (scall == MAT_REUSE_MATRIX){ 3442 ierr = MatGetOwnershipRange(*outmat,&rstart,PETSC_NULL);CHKERRQ(ierr); 3443 } else { 3444 SETERRQ1(PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 3445 } 3446 3447 for (i=0;i<m;i++) { 3448 ierr = MatGetRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 3449 Ii = i + rstart; 3450 ierr = MatSetValues(*outmat,1,&Ii,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 3451 ierr = MatRestoreRow_SeqAIJ(inmat,i,&nnz,&indx,&values);CHKERRQ(ierr); 3452 } 3453 ierr = MatDestroy(inmat);CHKERRQ(ierr); 3454 ierr = MatAssemblyBegin(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3455 ierr = MatAssemblyEnd(*outmat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3456 3457 PetscFunctionReturn(0); 3458 } 3459 3460 #undef __FUNCT__ 3461 #define __FUNCT__ "MatFileSplit" 3462 PetscErrorCode MatFileSplit(Mat A,char *outfile) 3463 { 3464 PetscErrorCode ierr; 3465 PetscMPIInt rank; 3466 PetscInt m,N,i,rstart,nnz; 3467 size_t len; 3468 const PetscInt *indx; 3469 PetscViewer out; 3470 char *name; 3471 Mat B; 3472 const PetscScalar *values; 3473 3474 PetscFunctionBegin; 3475 ierr = MatGetLocalSize(A,&m,0);CHKERRQ(ierr); 3476 ierr = MatGetSize(A,0,&N);CHKERRQ(ierr); 3477 /* Should this be the type of the diagonal block of A? */ 3478 ierr = MatCreate(PETSC_COMM_SELF,&B);CHKERRQ(ierr); 3479 ierr = MatSetSizes(B,m,N,m,N);CHKERRQ(ierr); 3480 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 3481 ierr = MatSeqAIJSetPreallocation(B,0,PETSC_NULL);CHKERRQ(ierr); 3482 ierr = MatGetOwnershipRange(A,&rstart,0);CHKERRQ(ierr); 3483 for (i=0;i<m;i++) { 3484 ierr = MatGetRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 3485 ierr = MatSetValues(B,1,&i,nnz,indx,values,INSERT_VALUES);CHKERRQ(ierr); 3486 ierr = MatRestoreRow(A,i+rstart,&nnz,&indx,&values);CHKERRQ(ierr); 3487 } 3488 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3489 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3490 3491 ierr = MPI_Comm_rank(A->comm,&rank);CHKERRQ(ierr); 3492 ierr = PetscStrlen(outfile,&len);CHKERRQ(ierr); 3493 ierr = PetscMalloc((len+5)*sizeof(char),&name);CHKERRQ(ierr); 3494 sprintf(name,"%s.%d",outfile,rank); 3495 ierr = PetscViewerBinaryOpen(PETSC_COMM_SELF,name,FILE_MODE_APPEND,&out);CHKERRQ(ierr); 3496 ierr = PetscFree(name); 3497 ierr = MatView(B,out);CHKERRQ(ierr); 3498 ierr = PetscViewerDestroy(out);CHKERRQ(ierr); 3499 ierr = MatDestroy(B);CHKERRQ(ierr); 3500 PetscFunctionReturn(0); 3501 } 3502 3503 EXTERN PetscErrorCode MatDestroy_MPIAIJ(Mat); 3504 #undef __FUNCT__ 3505 #define __FUNCT__ "MatDestroy_MPIAIJ_SeqsToMPI" 3506 PetscErrorCode PETSCMAT_DLLEXPORT MatDestroy_MPIAIJ_SeqsToMPI(Mat A) 3507 { 3508 PetscErrorCode ierr; 3509 Mat_Merge_SeqsToMPI *merge; 3510 PetscContainer container; 3511 3512 PetscFunctionBegin; 3513 ierr = PetscObjectQuery((PetscObject)A,"MatMergeSeqsToMPI",(PetscObject *)&container);CHKERRQ(ierr); 3514 if (container) { 3515 ierr = PetscContainerGetPointer(container,(void **)&merge);CHKERRQ(ierr); 3516 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 3517 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 3518 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 3519 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 3520 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 3521 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 3522 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 3523 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 3524 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 3525 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 3526 ierr = PetscFree(merge->rowmap.range);CHKERRQ(ierr); 3527 3528 ierr = PetscContainerDestroy(container);CHKERRQ(ierr); 3529 ierr = PetscObjectCompose((PetscObject)A,"MatMergeSeqsToMPI",0);CHKERRQ(ierr); 3530 } 3531 ierr = PetscFree(merge);CHKERRQ(ierr); 3532 3533 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 3534 PetscFunctionReturn(0); 3535 } 3536 3537 #include "src/mat/utils/freespace.h" 3538 #include "petscbt.h" 3539 static PetscEvent logkey_seqstompinum = 0; 3540 #undef __FUNCT__ 3541 #define __FUNCT__ "MatMerge_SeqsToMPINumeric" 3542 /*@C 3543 MatMerge_SeqsToMPI - Creates a MPIAIJ matrix by adding sequential 3544 matrices from each processor 3545 3546 Collective on MPI_Comm 3547 3548 Input Parameters: 3549 + comm - the communicators the parallel matrix will live on 3550 . seqmat - the input sequential matrices 3551 . m - number of local rows (or PETSC_DECIDE) 3552 . n - number of local columns (or PETSC_DECIDE) 3553 - scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 3554 3555 Output Parameter: 3556 . mpimat - the parallel matrix generated 3557 3558 Level: advanced 3559 3560 Notes: 3561 The dimensions of the sequential matrix in each processor MUST be the same. 3562 The input seqmat is included into the container "Mat_Merge_SeqsToMPI", and will be 3563 destroyed when mpimat is destroyed. Call PetscObjectQuery() to access seqmat. 3564 @*/ 3565 PetscErrorCode PETSCMAT_DLLEXPORT MatMerge_SeqsToMPINumeric(Mat seqmat,Mat mpimat) 3566 { 3567 PetscErrorCode ierr; 3568 MPI_Comm comm=mpimat->comm; 3569 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 3570 PetscMPIInt size,rank,taga,*len_s; 3571 PetscInt N=mpimat->cmap.N,i,j,*owners,*ai=a->i,*aj=a->j; 3572 PetscInt proc,m; 3573 PetscInt **buf_ri,**buf_rj; 3574 PetscInt k,anzi,*bj_i,*bi,*bj,arow,bnzi,nextaj; 3575 PetscInt nrows,**buf_ri_k,**nextrow,**nextai; 3576 MPI_Request *s_waits,*r_waits; 3577 MPI_Status *status; 3578 MatScalar *aa=a->a,**abuf_r,*ba_i; 3579 Mat_Merge_SeqsToMPI *merge; 3580 PetscContainer container; 3581 3582 PetscFunctionBegin; 3583 if (!logkey_seqstompinum) { 3584 ierr = PetscLogEventRegister(&logkey_seqstompinum,"MatMerge_SeqsToMPINumeric",MAT_COOKIE); 3585 } 3586 ierr = PetscLogEventBegin(logkey_seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 3587 3588 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3589 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3590 3591 ierr = PetscObjectQuery((PetscObject)mpimat,"MatMergeSeqsToMPI",(PetscObject *)&container);CHKERRQ(ierr); 3592 if (container) { 3593 ierr = PetscContainerGetPointer(container,(void **)&merge);CHKERRQ(ierr); 3594 } 3595 bi = merge->bi; 3596 bj = merge->bj; 3597 buf_ri = merge->buf_ri; 3598 buf_rj = merge->buf_rj; 3599 3600 ierr = PetscMalloc(size*sizeof(MPI_Status),&status);CHKERRQ(ierr); 3601 owners = merge->rowmap.range; 3602 len_s = merge->len_s; 3603 3604 /* send and recv matrix values */ 3605 /*-----------------------------*/ 3606 ierr = PetscObjectGetNewTag((PetscObject)mpimat,&taga);CHKERRQ(ierr); 3607 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 3608 3609 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&s_waits);CHKERRQ(ierr); 3610 for (proc=0,k=0; proc<size; proc++){ 3611 if (!len_s[proc]) continue; 3612 i = owners[proc]; 3613 ierr = MPI_Isend(aa+ai[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 3614 k++; 3615 } 3616 3617 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 3618 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 3619 ierr = PetscFree(status);CHKERRQ(ierr); 3620 3621 ierr = PetscFree(s_waits);CHKERRQ(ierr); 3622 ierr = PetscFree(r_waits);CHKERRQ(ierr); 3623 3624 /* insert mat values of mpimat */ 3625 /*----------------------------*/ 3626 ierr = PetscMalloc(N*sizeof(MatScalar),&ba_i);CHKERRQ(ierr); 3627 ierr = PetscMalloc((3*merge->nrecv+1)*sizeof(PetscInt**),&buf_ri_k);CHKERRQ(ierr); 3628 nextrow = buf_ri_k + merge->nrecv; 3629 nextai = nextrow + merge->nrecv; 3630 3631 for (k=0; k<merge->nrecv; k++){ 3632 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 3633 nrows = *(buf_ri_k[k]); 3634 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 3635 nextai[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 3636 } 3637 3638 /* set values of ba */ 3639 m = merge->rowmap.n; 3640 for (i=0; i<m; i++) { 3641 arow = owners[rank] + i; 3642 bj_i = bj+bi[i]; /* col indices of the i-th row of mpimat */ 3643 bnzi = bi[i+1] - bi[i]; 3644 ierr = PetscMemzero(ba_i,bnzi*sizeof(MatScalar));CHKERRQ(ierr); 3645 3646 /* add local non-zero vals of this proc's seqmat into ba */ 3647 anzi = ai[arow+1] - ai[arow]; 3648 aj = a->j + ai[arow]; 3649 aa = a->a + ai[arow]; 3650 nextaj = 0; 3651 for (j=0; nextaj<anzi; j++){ 3652 if (*(bj_i + j) == aj[nextaj]){ /* bcol == acol */ 3653 ba_i[j] += aa[nextaj++]; 3654 } 3655 } 3656 3657 /* add received vals into ba */ 3658 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 3659 /* i-th row */ 3660 if (i == *nextrow[k]) { 3661 anzi = *(nextai[k]+1) - *nextai[k]; 3662 aj = buf_rj[k] + *(nextai[k]); 3663 aa = abuf_r[k] + *(nextai[k]); 3664 nextaj = 0; 3665 for (j=0; nextaj<anzi; j++){ 3666 if (*(bj_i + j) == aj[nextaj]){ /* bcol == acol */ 3667 ba_i[j] += aa[nextaj++]; 3668 } 3669 } 3670 nextrow[k]++; nextai[k]++; 3671 } 3672 } 3673 ierr = MatSetValues(mpimat,1,&arow,bnzi,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 3674 } 3675 ierr = MatAssemblyBegin(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3676 ierr = MatAssemblyEnd(mpimat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3677 3678 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 3679 ierr = PetscFree(ba_i);CHKERRQ(ierr); 3680 ierr = PetscFree(buf_ri_k);CHKERRQ(ierr); 3681 ierr = PetscLogEventEnd(logkey_seqstompinum,seqmat,0,0,0);CHKERRQ(ierr); 3682 PetscFunctionReturn(0); 3683 } 3684 3685 static PetscEvent logkey_seqstompisym = 0; 3686 #undef __FUNCT__ 3687 #define __FUNCT__ "MatMerge_SeqsToMPISymbolic" 3688 PetscErrorCode PETSCMAT_DLLEXPORT MatMerge_SeqsToMPISymbolic(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,Mat *mpimat) 3689 { 3690 PetscErrorCode ierr; 3691 Mat B_mpi; 3692 Mat_SeqAIJ *a=(Mat_SeqAIJ*)seqmat->data; 3693 PetscMPIInt size,rank,tagi,tagj,*len_s,*len_si,*len_ri; 3694 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 3695 PetscInt M=seqmat->rmap.n,N=seqmat->cmap.n,i,*owners,*ai=a->i,*aj=a->j; 3696 PetscInt len,proc,*dnz,*onz; 3697 PetscInt k,anzi,*bi,*bj,*lnk,nlnk,arow,bnzi,nspacedouble=0; 3698 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextai; 3699 MPI_Request *si_waits,*sj_waits,*ri_waits,*rj_waits; 3700 MPI_Status *status; 3701 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 3702 PetscBT lnkbt; 3703 Mat_Merge_SeqsToMPI *merge; 3704 PetscContainer container; 3705 3706 PetscFunctionBegin; 3707 if (!logkey_seqstompisym) { 3708 ierr = PetscLogEventRegister(&logkey_seqstompisym,"MatMerge_SeqsToMPISymbolic",MAT_COOKIE); 3709 } 3710 ierr = PetscLogEventBegin(logkey_seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 3711 3712 /* make sure it is a PETSc comm */ 3713 ierr = PetscCommDuplicate(comm,&comm,PETSC_NULL);CHKERRQ(ierr); 3714 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 3715 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 3716 3717 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 3718 ierr = PetscMalloc(size*sizeof(MPI_Status),&status);CHKERRQ(ierr); 3719 3720 /* determine row ownership */ 3721 /*---------------------------------------------------------*/ 3722 ierr = PetscMapInitialize(comm,&merge->rowmap);CHKERRQ(ierr); 3723 merge->rowmap.n = m; 3724 merge->rowmap.N = M; 3725 merge->rowmap.bs = 1; 3726 ierr = PetscMapSetUp(&merge->rowmap);CHKERRQ(ierr); 3727 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 3728 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 3729 3730 m = merge->rowmap.n; 3731 M = merge->rowmap.N; 3732 owners = merge->rowmap.range; 3733 3734 /* determine the number of messages to send, their lengths */ 3735 /*---------------------------------------------------------*/ 3736 len_s = merge->len_s; 3737 3738 len = 0; /* length of buf_si[] */ 3739 merge->nsend = 0; 3740 for (proc=0; proc<size; proc++){ 3741 len_si[proc] = 0; 3742 if (proc == rank){ 3743 len_s[proc] = 0; 3744 } else { 3745 len_si[proc] = owners[proc+1] - owners[proc] + 1; 3746 len_s[proc] = ai[owners[proc+1]] - ai[owners[proc]]; /* num of rows to be sent to [proc] */ 3747 } 3748 if (len_s[proc]) { 3749 merge->nsend++; 3750 nrows = 0; 3751 for (i=owners[proc]; i<owners[proc+1]; i++){ 3752 if (ai[i+1] > ai[i]) nrows++; 3753 } 3754 len_si[proc] = 2*(nrows+1); 3755 len += len_si[proc]; 3756 } 3757 } 3758 3759 /* determine the number and length of messages to receive for ij-structure */ 3760 /*-------------------------------------------------------------------------*/ 3761 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 3762 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 3763 3764 /* post the Irecv of j-structure */ 3765 /*-------------------------------*/ 3766 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 3767 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rj_waits);CHKERRQ(ierr); 3768 3769 /* post the Isend of j-structure */ 3770 /*--------------------------------*/ 3771 ierr = PetscMalloc((2*merge->nsend+1)*sizeof(MPI_Request),&si_waits);CHKERRQ(ierr); 3772 sj_waits = si_waits + merge->nsend; 3773 3774 for (proc=0, k=0; proc<size; proc++){ 3775 if (!len_s[proc]) continue; 3776 i = owners[proc]; 3777 ierr = MPI_Isend(aj+ai[i],len_s[proc],MPIU_INT,proc,tagj,comm,sj_waits+k);CHKERRQ(ierr); 3778 k++; 3779 } 3780 3781 /* receives and sends of j-structure are complete */ 3782 /*------------------------------------------------*/ 3783 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,rj_waits,status);CHKERRQ(ierr);} 3784 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,sj_waits,status);CHKERRQ(ierr);} 3785 3786 /* send and recv i-structure */ 3787 /*---------------------------*/ 3788 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 3789 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&ri_waits);CHKERRQ(ierr); 3790 3791 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 3792 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 3793 for (proc=0,k=0; proc<size; proc++){ 3794 if (!len_s[proc]) continue; 3795 /* form outgoing message for i-structure: 3796 buf_si[0]: nrows to be sent 3797 [1:nrows]: row index (global) 3798 [nrows+1:2*nrows+1]: i-structure index 3799 */ 3800 /*-------------------------------------------*/ 3801 nrows = len_si[proc]/2 - 1; 3802 buf_si_i = buf_si + nrows+1; 3803 buf_si[0] = nrows; 3804 buf_si_i[0] = 0; 3805 nrows = 0; 3806 for (i=owners[proc]; i<owners[proc+1]; i++){ 3807 anzi = ai[i+1] - ai[i]; 3808 if (anzi) { 3809 buf_si_i[nrows+1] = buf_si_i[nrows] + anzi; /* i-structure */ 3810 buf_si[nrows+1] = i-owners[proc]; /* local row index */ 3811 nrows++; 3812 } 3813 } 3814 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,si_waits+k);CHKERRQ(ierr); 3815 k++; 3816 buf_si += len_si[proc]; 3817 } 3818 3819 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,ri_waits,status);CHKERRQ(ierr);} 3820 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,si_waits,status);CHKERRQ(ierr);} 3821 3822 ierr = PetscInfo2(seqmat,"nsend: %D, nrecv: %D\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 3823 for (i=0; i<merge->nrecv; i++){ 3824 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); 3825 } 3826 3827 ierr = PetscFree(len_si);CHKERRQ(ierr); 3828 ierr = PetscFree(len_ri);CHKERRQ(ierr); 3829 ierr = PetscFree(rj_waits);CHKERRQ(ierr); 3830 ierr = PetscFree(si_waits);CHKERRQ(ierr); 3831 ierr = PetscFree(ri_waits);CHKERRQ(ierr); 3832 ierr = PetscFree(buf_s);CHKERRQ(ierr); 3833 ierr = PetscFree(status);CHKERRQ(ierr); 3834 3835 /* compute a local seq matrix in each processor */ 3836 /*----------------------------------------------*/ 3837 /* allocate bi array and free space for accumulating nonzero column info */ 3838 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 3839 bi[0] = 0; 3840 3841 /* create and initialize a linked list */ 3842 nlnk = N+1; 3843 ierr = PetscLLCreate(N,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 3844 3845 /* initial FreeSpace size is 2*(num of local nnz(seqmat)) */ 3846 len = 0; 3847 len = ai[owners[rank+1]] - ai[owners[rank]]; 3848 ierr = PetscFreeSpaceGet((PetscInt)(2*len+1),&free_space);CHKERRQ(ierr); 3849 current_space = free_space; 3850 3851 /* determine symbolic info for each local row */ 3852 ierr = PetscMalloc((3*merge->nrecv+1)*sizeof(PetscInt**),&buf_ri_k);CHKERRQ(ierr); 3853 nextrow = buf_ri_k + merge->nrecv; 3854 nextai = nextrow + merge->nrecv; 3855 for (k=0; k<merge->nrecv; k++){ 3856 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 3857 nrows = *buf_ri_k[k]; 3858 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 3859 nextai[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 3860 } 3861 3862 ierr = MatPreallocateInitialize(comm,m,n,dnz,onz);CHKERRQ(ierr); 3863 len = 0; 3864 for (i=0;i<m;i++) { 3865 bnzi = 0; 3866 /* add local non-zero cols of this proc's seqmat into lnk */ 3867 arow = owners[rank] + i; 3868 anzi = ai[arow+1] - ai[arow]; 3869 aj = a->j + ai[arow]; 3870 ierr = PetscLLAdd(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 3871 bnzi += nlnk; 3872 /* add received col data into lnk */ 3873 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 3874 if (i == *nextrow[k]) { /* i-th row */ 3875 anzi = *(nextai[k]+1) - *nextai[k]; 3876 aj = buf_rj[k] + *nextai[k]; 3877 ierr = PetscLLAdd(anzi,aj,N,nlnk,lnk,lnkbt);CHKERRQ(ierr); 3878 bnzi += nlnk; 3879 nextrow[k]++; nextai[k]++; 3880 } 3881 } 3882 if (len < bnzi) len = bnzi; /* =max(bnzi) */ 3883 3884 /* if free space is not available, make more free space */ 3885 if (current_space->local_remaining<bnzi) { 3886 ierr = PetscFreeSpaceGet(current_space->total_array_size,¤t_space);CHKERRQ(ierr); 3887 nspacedouble++; 3888 } 3889 /* copy data into free space, then initialize lnk */ 3890 ierr = PetscLLClean(N,N,bnzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 3891 ierr = MatPreallocateSet(i+owners[rank],bnzi,current_space->array,dnz,onz);CHKERRQ(ierr); 3892 3893 current_space->array += bnzi; 3894 current_space->local_used += bnzi; 3895 current_space->local_remaining -= bnzi; 3896 3897 bi[i+1] = bi[i] + bnzi; 3898 } 3899 3900 ierr = PetscFree(buf_ri_k);CHKERRQ(ierr); 3901 3902 ierr = PetscMalloc((bi[m]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 3903 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 3904 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 3905 3906 /* create symbolic parallel matrix B_mpi */ 3907 /*---------------------------------------*/ 3908 ierr = MatCreate(comm,&B_mpi);CHKERRQ(ierr); 3909 if (n==PETSC_DECIDE) { 3910 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,N);CHKERRQ(ierr); 3911 } else { 3912 ierr = MatSetSizes(B_mpi,m,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 3913 } 3914 ierr = MatSetType(B_mpi,MATMPIAIJ);CHKERRQ(ierr); 3915 ierr = MatMPIAIJSetPreallocation(B_mpi,0,dnz,0,onz);CHKERRQ(ierr); 3916 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 3917 3918 /* B_mpi is not ready for use - assembly will be done by MatMerge_SeqsToMPINumeric() */ 3919 B_mpi->assembled = PETSC_FALSE; 3920 B_mpi->ops->destroy = MatDestroy_MPIAIJ_SeqsToMPI; 3921 merge->bi = bi; 3922 merge->bj = bj; 3923 merge->buf_ri = buf_ri; 3924 merge->buf_rj = buf_rj; 3925 merge->coi = PETSC_NULL; 3926 merge->coj = PETSC_NULL; 3927 merge->owners_co = PETSC_NULL; 3928 3929 /* attach the supporting struct to B_mpi for reuse */ 3930 ierr = PetscContainerCreate(PETSC_COMM_SELF,&container);CHKERRQ(ierr); 3931 ierr = PetscContainerSetPointer(container,merge);CHKERRQ(ierr); 3932 ierr = PetscObjectCompose((PetscObject)B_mpi,"MatMergeSeqsToMPI",(PetscObject)container);CHKERRQ(ierr); 3933 *mpimat = B_mpi; 3934 3935 ierr = PetscCommDestroy(&comm);CHKERRQ(ierr); 3936 ierr = PetscLogEventEnd(logkey_seqstompisym,seqmat,0,0,0);CHKERRQ(ierr); 3937 PetscFunctionReturn(0); 3938 } 3939 3940 static PetscEvent logkey_seqstompi = 0; 3941 #undef __FUNCT__ 3942 #define __FUNCT__ "MatMerge_SeqsToMPI" 3943 PetscErrorCode PETSCMAT_DLLEXPORT MatMerge_SeqsToMPI(MPI_Comm comm,Mat seqmat,PetscInt m,PetscInt n,MatReuse scall,Mat *mpimat) 3944 { 3945 PetscErrorCode ierr; 3946 3947 PetscFunctionBegin; 3948 if (!logkey_seqstompi) { 3949 ierr = PetscLogEventRegister(&logkey_seqstompi,"MatMerge_SeqsToMPI",MAT_COOKIE); 3950 } 3951 ierr = PetscLogEventBegin(logkey_seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 3952 if (scall == MAT_INITIAL_MATRIX){ 3953 ierr = MatMerge_SeqsToMPISymbolic(comm,seqmat,m,n,mpimat);CHKERRQ(ierr); 3954 } 3955 ierr = MatMerge_SeqsToMPINumeric(seqmat,*mpimat);CHKERRQ(ierr); 3956 ierr = PetscLogEventEnd(logkey_seqstompi,seqmat,0,0,0);CHKERRQ(ierr); 3957 PetscFunctionReturn(0); 3958 } 3959 static PetscEvent logkey_getlocalmat = 0; 3960 #undef __FUNCT__ 3961 #define __FUNCT__ "MatGetLocalMat" 3962 /*@ 3963 MatGetLocalMat - Creates a SeqAIJ matrix by taking all its local rows 3964 3965 Not Collective 3966 3967 Input Parameters: 3968 + A - the matrix 3969 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 3970 3971 Output Parameter: 3972 . A_loc - the local sequential matrix generated 3973 3974 Level: developer 3975 3976 @*/ 3977 PetscErrorCode PETSCMAT_DLLEXPORT MatGetLocalMat(Mat A,MatReuse scall,Mat *A_loc) 3978 { 3979 PetscErrorCode ierr; 3980 Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data; 3981 Mat_SeqAIJ *mat,*a=(Mat_SeqAIJ*)(mpimat->A)->data,*b=(Mat_SeqAIJ*)(mpimat->B)->data; 3982 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j,*cmap=mpimat->garray; 3983 PetscScalar *aa=a->a,*ba=b->a,*ca; 3984 PetscInt am=A->rmap.n,i,j,k,cstart=A->cmap.rstart; 3985 PetscInt *ci,*cj,col,ncols_d,ncols_o,jo; 3986 3987 PetscFunctionBegin; 3988 if (!logkey_getlocalmat) { 3989 ierr = PetscLogEventRegister(&logkey_getlocalmat,"MatGetLocalMat",MAT_COOKIE); 3990 } 3991 ierr = PetscLogEventBegin(logkey_getlocalmat,A,0,0,0);CHKERRQ(ierr); 3992 if (scall == MAT_INITIAL_MATRIX){ 3993 ierr = PetscMalloc((1+am)*sizeof(PetscInt),&ci);CHKERRQ(ierr); 3994 ci[0] = 0; 3995 for (i=0; i<am; i++){ 3996 ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]); 3997 } 3998 ierr = PetscMalloc((1+ci[am])*sizeof(PetscInt),&cj);CHKERRQ(ierr); 3999 ierr = PetscMalloc((1+ci[am])*sizeof(PetscScalar),&ca);CHKERRQ(ierr); 4000 k = 0; 4001 for (i=0; i<am; i++) { 4002 ncols_o = bi[i+1] - bi[i]; 4003 ncols_d = ai[i+1] - ai[i]; 4004 /* off-diagonal portion of A */ 4005 for (jo=0; jo<ncols_o; jo++) { 4006 col = cmap[*bj]; 4007 if (col >= cstart) break; 4008 cj[k] = col; bj++; 4009 ca[k++] = *ba++; 4010 } 4011 /* diagonal portion of A */ 4012 for (j=0; j<ncols_d; j++) { 4013 cj[k] = cstart + *aj++; 4014 ca[k++] = *aa++; 4015 } 4016 /* off-diagonal portion of A */ 4017 for (j=jo; j<ncols_o; j++) { 4018 cj[k] = cmap[*bj++]; 4019 ca[k++] = *ba++; 4020 } 4021 } 4022 /* put together the new matrix */ 4023 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,A->cmap.N,ci,cj,ca,A_loc);CHKERRQ(ierr); 4024 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4025 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4026 mat = (Mat_SeqAIJ*)(*A_loc)->data; 4027 mat->free_a = PETSC_TRUE; 4028 mat->free_ij = PETSC_TRUE; 4029 mat->nonew = 0; 4030 } else if (scall == MAT_REUSE_MATRIX){ 4031 mat=(Mat_SeqAIJ*)(*A_loc)->data; 4032 ci = mat->i; cj = mat->j; ca = mat->a; 4033 for (i=0; i<am; i++) { 4034 /* off-diagonal portion of A */ 4035 ncols_o = bi[i+1] - bi[i]; 4036 for (jo=0; jo<ncols_o; jo++) { 4037 col = cmap[*bj]; 4038 if (col >= cstart) break; 4039 *ca++ = *ba++; bj++; 4040 } 4041 /* diagonal portion of A */ 4042 ncols_d = ai[i+1] - ai[i]; 4043 for (j=0; j<ncols_d; j++) *ca++ = *aa++; 4044 /* off-diagonal portion of A */ 4045 for (j=jo; j<ncols_o; j++) { 4046 *ca++ = *ba++; bj++; 4047 } 4048 } 4049 } else { 4050 SETERRQ1(PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall); 4051 } 4052 4053 ierr = PetscLogEventEnd(logkey_getlocalmat,A,0,0,0);CHKERRQ(ierr); 4054 PetscFunctionReturn(0); 4055 } 4056 4057 static PetscEvent logkey_getlocalmatcondensed = 0; 4058 #undef __FUNCT__ 4059 #define __FUNCT__ "MatGetLocalMatCondensed" 4060 /*@C 4061 MatGetLocalMatCondensed - Creates a SeqAIJ matrix by taking all its local rows and NON-ZERO columns 4062 4063 Not Collective 4064 4065 Input Parameters: 4066 + A - the matrix 4067 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4068 - row, col - index sets of rows and columns to extract (or PETSC_NULL) 4069 4070 Output Parameter: 4071 . A_loc - the local sequential matrix generated 4072 4073 Level: developer 4074 4075 @*/ 4076 PetscErrorCode PETSCMAT_DLLEXPORT MatGetLocalMatCondensed(Mat A,MatReuse scall,IS *row,IS *col,Mat *A_loc) 4077 { 4078 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4079 PetscErrorCode ierr; 4080 PetscInt i,start,end,ncols,nzA,nzB,*cmap,imark,*idx; 4081 IS isrowa,iscola; 4082 Mat *aloc; 4083 4084 PetscFunctionBegin; 4085 if (!logkey_getlocalmatcondensed) { 4086 ierr = PetscLogEventRegister(&logkey_getlocalmatcondensed,"MatGetLocalMatCondensed",MAT_COOKIE); 4087 } 4088 ierr = PetscLogEventBegin(logkey_getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4089 if (!row){ 4090 start = A->rmap.rstart; end = A->rmap.rend; 4091 ierr = ISCreateStride(PETSC_COMM_SELF,end-start,start,1,&isrowa);CHKERRQ(ierr); 4092 } else { 4093 isrowa = *row; 4094 } 4095 if (!col){ 4096 start = A->cmap.rstart; 4097 cmap = a->garray; 4098 nzA = a->A->cmap.n; 4099 nzB = a->B->cmap.n; 4100 ierr = PetscMalloc((nzA+nzB)*sizeof(PetscInt), &idx);CHKERRQ(ierr); 4101 ncols = 0; 4102 for (i=0; i<nzB; i++) { 4103 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4104 else break; 4105 } 4106 imark = i; 4107 for (i=0; i<nzA; i++) idx[ncols++] = start + i; 4108 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; 4109 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,&iscola);CHKERRQ(ierr); 4110 ierr = PetscFree(idx);CHKERRQ(ierr); 4111 } else { 4112 iscola = *col; 4113 } 4114 if (scall != MAT_INITIAL_MATRIX){ 4115 ierr = PetscMalloc(sizeof(Mat),&aloc);CHKERRQ(ierr); 4116 aloc[0] = *A_loc; 4117 } 4118 ierr = MatGetSubMatrices(A,1,&isrowa,&iscola,scall,&aloc);CHKERRQ(ierr); 4119 *A_loc = aloc[0]; 4120 ierr = PetscFree(aloc);CHKERRQ(ierr); 4121 if (!row){ 4122 ierr = ISDestroy(isrowa);CHKERRQ(ierr); 4123 } 4124 if (!col){ 4125 ierr = ISDestroy(iscola);CHKERRQ(ierr); 4126 } 4127 ierr = PetscLogEventEnd(logkey_getlocalmatcondensed,A,0,0,0);CHKERRQ(ierr); 4128 PetscFunctionReturn(0); 4129 } 4130 4131 static PetscEvent logkey_GetBrowsOfAcols = 0; 4132 #undef __FUNCT__ 4133 #define __FUNCT__ "MatGetBrowsOfAcols" 4134 /*@C 4135 MatGetBrowsOfAcols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns of local A 4136 4137 Collective on Mat 4138 4139 Input Parameters: 4140 + A,B - the matrices in mpiaij format 4141 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4142 - rowb, colb - index sets of rows and columns of B to extract (or PETSC_NULL) 4143 4144 Output Parameter: 4145 + rowb, colb - index sets of rows and columns of B to extract 4146 . brstart - row index of B_seq from which next B->rmap.n rows are taken from B's local rows 4147 - B_seq - the sequential matrix generated 4148 4149 Level: developer 4150 4151 @*/ 4152 PetscErrorCode PETSCMAT_DLLEXPORT MatGetBrowsOfAcols(Mat A,Mat B,MatReuse scall,IS *rowb,IS *colb,PetscInt *brstart,Mat *B_seq) 4153 { 4154 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4155 PetscErrorCode ierr; 4156 PetscInt *idx,i,start,ncols,nzA,nzB,*cmap,imark; 4157 IS isrowb,iscolb; 4158 Mat *bseq; 4159 4160 PetscFunctionBegin; 4161 if (A->cmap.rstart != B->rmap.rstart || A->cmap.rend != B->rmap.rend){ 4162 SETERRQ4(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); 4163 } 4164 if (!logkey_GetBrowsOfAcols) { 4165 ierr = PetscLogEventRegister(&logkey_GetBrowsOfAcols,"MatGetBrowsOfAcols",MAT_COOKIE); 4166 } 4167 ierr = PetscLogEventBegin(logkey_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4168 4169 if (scall == MAT_INITIAL_MATRIX){ 4170 start = A->cmap.rstart; 4171 cmap = a->garray; 4172 nzA = a->A->cmap.n; 4173 nzB = a->B->cmap.n; 4174 ierr = PetscMalloc((nzA+nzB)*sizeof(PetscInt), &idx);CHKERRQ(ierr); 4175 ncols = 0; 4176 for (i=0; i<nzB; i++) { /* row < local row index */ 4177 if (cmap[i] < start) idx[ncols++] = cmap[i]; 4178 else break; 4179 } 4180 imark = i; 4181 for (i=0; i<nzA; i++) idx[ncols++] = start + i; /* local rows */ 4182 for (i=imark; i<nzB; i++) idx[ncols++] = cmap[i]; /* row > local row index */ 4183 ierr = ISCreateGeneral(PETSC_COMM_SELF,ncols,idx,&isrowb);CHKERRQ(ierr); 4184 ierr = PetscFree(idx);CHKERRQ(ierr); 4185 *brstart = imark; 4186 ierr = ISCreateStride(PETSC_COMM_SELF,B->cmap.N,0,1,&iscolb);CHKERRQ(ierr); 4187 } else { 4188 if (!rowb || !colb) SETERRQ(PETSC_ERR_SUP,"IS rowb and colb must be provided for MAT_REUSE_MATRIX"); 4189 isrowb = *rowb; iscolb = *colb; 4190 ierr = PetscMalloc(sizeof(Mat),&bseq);CHKERRQ(ierr); 4191 bseq[0] = *B_seq; 4192 } 4193 ierr = MatGetSubMatrices(B,1,&isrowb,&iscolb,scall,&bseq);CHKERRQ(ierr); 4194 *B_seq = bseq[0]; 4195 ierr = PetscFree(bseq);CHKERRQ(ierr); 4196 if (!rowb){ 4197 ierr = ISDestroy(isrowb);CHKERRQ(ierr); 4198 } else { 4199 *rowb = isrowb; 4200 } 4201 if (!colb){ 4202 ierr = ISDestroy(iscolb);CHKERRQ(ierr); 4203 } else { 4204 *colb = iscolb; 4205 } 4206 ierr = PetscLogEventEnd(logkey_GetBrowsOfAcols,A,B,0,0);CHKERRQ(ierr); 4207 PetscFunctionReturn(0); 4208 } 4209 4210 static PetscEvent logkey_GetBrowsOfAocols = 0; 4211 #undef __FUNCT__ 4212 #define __FUNCT__ "MatGetBrowsOfAoCols" 4213 /*@C 4214 MatGetBrowsOfAoCols - Creates a SeqAIJ matrix by taking rows of B that equal to nonzero columns 4215 of the OFF-DIAGONAL portion of local A 4216 4217 Collective on Mat 4218 4219 Input Parameters: 4220 + A,B - the matrices in mpiaij format 4221 . scall - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4222 . startsj - starting point in B's sending and receiving j-arrays, saved for MAT_REUSE (or PETSC_NULL) 4223 - bufa_ptr - array for sending matrix values, saved for MAT_REUSE (or PETSC_NULL) 4224 4225 Output Parameter: 4226 + B_oth - the sequential matrix generated 4227 4228 Level: developer 4229 4230 @*/ 4231 PetscErrorCode PETSCMAT_DLLEXPORT MatGetBrowsOfAoCols(Mat A,Mat B,MatReuse scall,PetscInt **startsj,PetscScalar **bufa_ptr,Mat *B_oth) 4232 { 4233 VecScatter_MPI_General *gen_to,*gen_from; 4234 PetscErrorCode ierr; 4235 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 4236 Mat_SeqAIJ *b_oth; 4237 VecScatter ctx=a->Mvctx; 4238 MPI_Comm comm=ctx->comm; 4239 PetscMPIInt *rprocs,*sprocs,tag=ctx->tag,rank; 4240 PetscInt *rowlen,*bufj,*bufJ,ncols,aBn=a->B->cmap.n,row,*b_othi,*b_othj; 4241 PetscScalar *rvalues,*svalues,*b_otha,*bufa,*bufA; 4242 PetscInt i,j,k,l,ll,nrecvs,nsends,nrows,*srow,*rstarts,*rstartsj = 0,*sstarts,*sstartsj,len; 4243 MPI_Request *rwaits = PETSC_NULL,*swaits = PETSC_NULL; 4244 MPI_Status *sstatus,rstatus; 4245 PetscMPIInt jj; 4246 PetscInt *cols,sbs,rbs; 4247 PetscScalar *vals; 4248 4249 PetscFunctionBegin; 4250 if (A->cmap.rstart != B->rmap.rstart || A->cmap.rend != B->rmap.rend){ 4251 SETERRQ4(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); 4252 } 4253 if (!logkey_GetBrowsOfAocols) { 4254 ierr = PetscLogEventRegister(&logkey_GetBrowsOfAocols,"MatGetBrAoCol",MAT_COOKIE); 4255 } 4256 ierr = PetscLogEventBegin(logkey_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 4257 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 4258 4259 gen_to = (VecScatter_MPI_General*)ctx->todata; 4260 gen_from = (VecScatter_MPI_General*)ctx->fromdata; 4261 rvalues = gen_from->values; /* holds the length of receiving row */ 4262 svalues = gen_to->values; /* holds the length of sending row */ 4263 nrecvs = gen_from->n; 4264 nsends = gen_to->n; 4265 4266 ierr = PetscMalloc2(nrecvs,MPI_Request,&rwaits,nsends,MPI_Request,&swaits);CHKERRQ(ierr); 4267 srow = gen_to->indices; /* local row index to be sent */ 4268 sstarts = gen_to->starts; 4269 sprocs = gen_to->procs; 4270 sstatus = gen_to->sstatus; 4271 sbs = gen_to->bs; 4272 rstarts = gen_from->starts; 4273 rprocs = gen_from->procs; 4274 rbs = gen_from->bs; 4275 4276 if (!startsj || !bufa_ptr) scall = MAT_INITIAL_MATRIX; 4277 if (scall == MAT_INITIAL_MATRIX){ 4278 /* i-array */ 4279 /*---------*/ 4280 /* post receives */ 4281 for (i=0; i<nrecvs; i++){ 4282 rowlen = (PetscInt*)rvalues + rstarts[i]*rbs; 4283 nrows = (rstarts[i+1]-rstarts[i])*rbs; /* num of indices to be received */ 4284 ierr = MPI_Irecv(rowlen,nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4285 } 4286 4287 /* pack the outgoing message */ 4288 ierr = PetscMalloc((nsends+nrecvs+3)*sizeof(PetscInt),&sstartsj);CHKERRQ(ierr); 4289 rstartsj = sstartsj + nsends +1; 4290 sstartsj[0] = 0; rstartsj[0] = 0; 4291 len = 0; /* total length of j or a array to be sent */ 4292 k = 0; 4293 for (i=0; i<nsends; i++){ 4294 rowlen = (PetscInt*)svalues + sstarts[i]*sbs; 4295 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4296 for (j=0; j<nrows; j++) { 4297 row = srow[k] + B->rmap.range[rank]; /* global row idx */ 4298 for (l=0; l<sbs; l++){ 4299 ierr = MatGetRow_MPIAIJ(B,row+l,&ncols,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr); /* rowlength */ 4300 rowlen[j*sbs+l] = ncols; 4301 len += ncols; 4302 ierr = MatRestoreRow_MPIAIJ(B,row+l,&ncols,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr); 4303 } 4304 k++; 4305 } 4306 ierr = MPI_Isend(rowlen,nrows*sbs,MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4307 sstartsj[i+1] = len; /* starting point of (i+1)-th outgoing msg in bufj and bufa */ 4308 } 4309 /* recvs and sends of i-array are completed */ 4310 i = nrecvs; 4311 while (i--) { 4312 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4313 } 4314 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4315 4316 /* allocate buffers for sending j and a arrays */ 4317 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&bufj);CHKERRQ(ierr); 4318 ierr = PetscMalloc((len+1)*sizeof(PetscScalar),&bufa);CHKERRQ(ierr); 4319 4320 /* create i-array of B_oth */ 4321 ierr = PetscMalloc((aBn+2)*sizeof(PetscInt),&b_othi);CHKERRQ(ierr); 4322 b_othi[0] = 0; 4323 len = 0; /* total length of j or a array to be received */ 4324 k = 0; 4325 for (i=0; i<nrecvs; i++){ 4326 rowlen = (PetscInt*)rvalues + rstarts[i]*rbs; 4327 nrows = rbs*(rstarts[i+1]-rstarts[i]); /* num of rows to be recieved */ 4328 for (j=0; j<nrows; j++) { 4329 b_othi[k+1] = b_othi[k] + rowlen[j]; 4330 len += rowlen[j]; k++; 4331 } 4332 rstartsj[i+1] = len; /* starting point of (i+1)-th incoming msg in bufj and bufa */ 4333 } 4334 4335 /* allocate space for j and a arrrays of B_oth */ 4336 ierr = PetscMalloc((b_othi[aBn]+1)*sizeof(PetscInt),&b_othj);CHKERRQ(ierr); 4337 ierr = PetscMalloc((b_othi[aBn]+1)*sizeof(PetscScalar),&b_otha);CHKERRQ(ierr); 4338 4339 /* j-array */ 4340 /*---------*/ 4341 /* post receives of j-array */ 4342 for (i=0; i<nrecvs; i++){ 4343 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 4344 ierr = MPI_Irecv(b_othj+rstartsj[i],nrows,MPIU_INT,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4345 } 4346 4347 /* pack the outgoing message j-array */ 4348 k = 0; 4349 for (i=0; i<nsends; i++){ 4350 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4351 bufJ = bufj+sstartsj[i]; 4352 for (j=0; j<nrows; j++) { 4353 row = srow[k++] + B->rmap.range[rank]; /* global row idx */ 4354 for (ll=0; ll<sbs; ll++){ 4355 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,&cols,PETSC_NULL);CHKERRQ(ierr); 4356 for (l=0; l<ncols; l++){ 4357 *bufJ++ = cols[l]; 4358 } 4359 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,&cols,PETSC_NULL);CHKERRQ(ierr); 4360 } 4361 } 4362 ierr = MPI_Isend(bufj+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_INT,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4363 } 4364 4365 /* recvs and sends of j-array are completed */ 4366 i = nrecvs; 4367 while (i--) { 4368 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4369 } 4370 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4371 } else if (scall == MAT_REUSE_MATRIX){ 4372 sstartsj = *startsj; 4373 rstartsj = sstartsj + nsends +1; 4374 bufa = *bufa_ptr; 4375 b_oth = (Mat_SeqAIJ*)(*B_oth)->data; 4376 b_otha = b_oth->a; 4377 } else { 4378 SETERRQ(PETSC_ERR_ARG_WRONGSTATE, "Matrix P does not posses an object container"); 4379 } 4380 4381 /* a-array */ 4382 /*---------*/ 4383 /* post receives of a-array */ 4384 for (i=0; i<nrecvs; i++){ 4385 nrows = rstartsj[i+1]-rstartsj[i]; /* length of the msg received */ 4386 ierr = MPI_Irecv(b_otha+rstartsj[i],nrows,MPIU_SCALAR,rprocs[i],tag,comm,rwaits+i);CHKERRQ(ierr); 4387 } 4388 4389 /* pack the outgoing message a-array */ 4390 k = 0; 4391 for (i=0; i<nsends; i++){ 4392 nrows = sstarts[i+1]-sstarts[i]; /* num of block rows */ 4393 bufA = bufa+sstartsj[i]; 4394 for (j=0; j<nrows; j++) { 4395 row = srow[k++] + B->rmap.range[rank]; /* global row idx */ 4396 for (ll=0; ll<sbs; ll++){ 4397 ierr = MatGetRow_MPIAIJ(B,row+ll,&ncols,PETSC_NULL,&vals);CHKERRQ(ierr); 4398 for (l=0; l<ncols; l++){ 4399 *bufA++ = vals[l]; 4400 } 4401 ierr = MatRestoreRow_MPIAIJ(B,row+ll,&ncols,PETSC_NULL,&vals);CHKERRQ(ierr); 4402 } 4403 } 4404 ierr = MPI_Isend(bufa+sstartsj[i],sstartsj[i+1]-sstartsj[i],MPIU_SCALAR,sprocs[i],tag,comm,swaits+i);CHKERRQ(ierr); 4405 } 4406 /* recvs and sends of a-array are completed */ 4407 i = nrecvs; 4408 while (i--) { 4409 ierr = MPI_Waitany(nrecvs,rwaits,&jj,&rstatus);CHKERRQ(ierr); 4410 } 4411 if (nsends) {ierr = MPI_Waitall(nsends,swaits,sstatus);CHKERRQ(ierr);} 4412 ierr = PetscFree2(rwaits,swaits);CHKERRQ(ierr); 4413 4414 if (scall == MAT_INITIAL_MATRIX){ 4415 /* put together the new matrix */ 4416 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,aBn,B->cmap.N,b_othi,b_othj,b_otha,B_oth);CHKERRQ(ierr); 4417 4418 /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */ 4419 /* Since these are PETSc arrays, change flags to free them as necessary. */ 4420 b_oth = (Mat_SeqAIJ *)(*B_oth)->data; 4421 b_oth->free_a = PETSC_TRUE; 4422 b_oth->free_ij = PETSC_TRUE; 4423 b_oth->nonew = 0; 4424 4425 ierr = PetscFree(bufj);CHKERRQ(ierr); 4426 if (!startsj || !bufa_ptr){ 4427 ierr = PetscFree(sstartsj);CHKERRQ(ierr); 4428 ierr = PetscFree(bufa_ptr);CHKERRQ(ierr); 4429 } else { 4430 *startsj = sstartsj; 4431 *bufa_ptr = bufa; 4432 } 4433 } 4434 ierr = PetscLogEventEnd(logkey_GetBrowsOfAocols,A,B,0,0);CHKERRQ(ierr); 4435 PetscFunctionReturn(0); 4436 } 4437 4438 #undef __FUNCT__ 4439 #define __FUNCT__ "MatGetCommunicationStructs" 4440 /*@C 4441 MatGetCommunicationStructs - Provides access to the communication structures used in matrix-vector multiplication. 4442 4443 Not Collective 4444 4445 Input Parameters: 4446 . A - The matrix in mpiaij format 4447 4448 Output Parameter: 4449 + lvec - The local vector holding off-process values from the argument to a matrix-vector product 4450 . colmap - A map from global column index to local index into lvec 4451 - multScatter - A scatter from the argument of a matrix-vector product to lvec 4452 4453 Level: developer 4454 4455 @*/ 4456 #if defined (PETSC_USE_CTABLE) 4457 PetscErrorCode PETSCMAT_DLLEXPORT MatGetCommunicationStructs(Mat A, Vec *lvec, PetscTable *colmap, VecScatter *multScatter) 4458 #else 4459 PetscErrorCode PETSCMAT_DLLEXPORT MatGetCommunicationStructs(Mat A, Vec *lvec, PetscInt *colmap[], VecScatter *multScatter) 4460 #endif 4461 { 4462 Mat_MPIAIJ *a; 4463 4464 PetscFunctionBegin; 4465 PetscValidHeaderSpecific(A, MAT_COOKIE, 1); 4466 PetscValidPointer(lvec, 2) 4467 PetscValidPointer(colmap, 3) 4468 PetscValidPointer(multScatter, 4) 4469 a = (Mat_MPIAIJ *) A->data; 4470 if (lvec) *lvec = a->lvec; 4471 if (colmap) *colmap = a->colmap; 4472 if (multScatter) *multScatter = a->Mvctx; 4473 PetscFunctionReturn(0); 4474 } 4475 4476 EXTERN_C_BEGIN 4477 extern PetscErrorCode PETSCMAT_DLLEXPORT MatConvert_MPIAIJ_MPICRL(Mat,MatType,MatReuse,Mat*); 4478 extern PetscErrorCode PETSCMAT_DLLEXPORT MatConvert_MPIAIJ_MPICSRPERM(Mat,MatType,MatReuse,Mat*); 4479 EXTERN_C_END 4480 4481 /*MC 4482 MATMPIAIJ - MATMPIAIJ = "mpiaij" - A matrix type to be used for parallel sparse matrices. 4483 4484 Options Database Keys: 4485 . -mat_type mpiaij - sets the matrix type to "mpiaij" during a call to MatSetFromOptions() 4486 4487 Level: beginner 4488 4489 .seealso: MatCreateMPIAIJ() 4490 M*/ 4491 4492 EXTERN_C_BEGIN 4493 #undef __FUNCT__ 4494 #define __FUNCT__ "MatCreate_MPIAIJ" 4495 PetscErrorCode PETSCMAT_DLLEXPORT MatCreate_MPIAIJ(Mat B) 4496 { 4497 Mat_MPIAIJ *b; 4498 PetscErrorCode ierr; 4499 PetscMPIInt size; 4500 4501 PetscFunctionBegin; 4502 ierr = MPI_Comm_size(B->comm,&size);CHKERRQ(ierr); 4503 4504 ierr = PetscNewLog(B,Mat_MPIAIJ,&b);CHKERRQ(ierr); 4505 B->data = (void*)b; 4506 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 4507 B->factor = 0; 4508 B->rmap.bs = 1; 4509 B->assembled = PETSC_FALSE; 4510 B->mapping = 0; 4511 4512 B->insertmode = NOT_SET_VALUES; 4513 b->size = size; 4514 ierr = MPI_Comm_rank(B->comm,&b->rank);CHKERRQ(ierr); 4515 4516 /* build cache for off array entries formed */ 4517 ierr = MatStashCreate_Private(B->comm,1,&B->stash);CHKERRQ(ierr); 4518 b->donotstash = PETSC_FALSE; 4519 b->colmap = 0; 4520 b->garray = 0; 4521 b->roworiented = PETSC_TRUE; 4522 4523 /* stuff used for matrix vector multiply */ 4524 b->lvec = PETSC_NULL; 4525 b->Mvctx = PETSC_NULL; 4526 4527 /* stuff for MatGetRow() */ 4528 b->rowindices = 0; 4529 b->rowvalues = 0; 4530 b->getrowactive = PETSC_FALSE; 4531 4532 4533 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C", 4534 "MatStoreValues_MPIAIJ", 4535 MatStoreValues_MPIAIJ);CHKERRQ(ierr); 4536 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C", 4537 "MatRetrieveValues_MPIAIJ", 4538 MatRetrieveValues_MPIAIJ);CHKERRQ(ierr); 4539 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetDiagonalBlock_C", 4540 "MatGetDiagonalBlock_MPIAIJ", 4541 MatGetDiagonalBlock_MPIAIJ);CHKERRQ(ierr); 4542 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatIsTranspose_C", 4543 "MatIsTranspose_MPIAIJ", 4544 MatIsTranspose_MPIAIJ);CHKERRQ(ierr); 4545 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPIAIJSetPreallocation_C", 4546 "MatMPIAIJSetPreallocation_MPIAIJ", 4547 MatMPIAIJSetPreallocation_MPIAIJ);CHKERRQ(ierr); 4548 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPIAIJSetPreallocationCSR_C", 4549 "MatMPIAIJSetPreallocationCSR_MPIAIJ", 4550 MatMPIAIJSetPreallocationCSR_MPIAIJ);CHKERRQ(ierr); 4551 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatDiagonalScaleLocal_C", 4552 "MatDiagonalScaleLocal_MPIAIJ", 4553 MatDiagonalScaleLocal_MPIAIJ);CHKERRQ(ierr); 4554 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_mpiaij_mpicsrperm_C", 4555 "MatConvert_MPIAIJ_MPICSRPERM", 4556 MatConvert_MPIAIJ_MPICSRPERM);CHKERRQ(ierr); 4557 ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_mpiaij_mpicrl_C", 4558 "MatConvert_MPIAIJ_MPICRL", 4559 MatConvert_MPIAIJ_MPICRL);CHKERRQ(ierr); 4560 ierr = PetscObjectChangeTypeName((PetscObject)B,MATMPIAIJ);CHKERRQ(ierr); 4561 PetscFunctionReturn(0); 4562 } 4563 EXTERN_C_END 4564 4565 #undef __FUNCT__ 4566 #define __FUNCT__ "MatCreateMPIAIJWithSplitArrays" 4567 /*@ 4568 MatCreateMPIAIJWithSplitArrays - creates a MPI AIJ matrix using arrays that contain the "diagonal" 4569 and "off-diagonal" part of the matrix in CSR format. 4570 4571 Collective on MPI_Comm 4572 4573 Input Parameters: 4574 + comm - MPI communicator 4575 . m - number of local rows (Cannot be PETSC_DECIDE) 4576 . n - This value should be the same as the local size used in creating the 4577 x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have 4578 calculated if N is given) For square matrices n is almost always m. 4579 . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given) 4580 . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given) 4581 . i - row indices for "diagonal" portion of matrix 4582 . j - column indices 4583 . a - matrix values 4584 . oi - row indices for "off-diagonal" portion of matrix 4585 . oj - column indices 4586 - oa - matrix values 4587 4588 Output Parameter: 4589 . mat - the matrix 4590 4591 Level: advanced 4592 4593 Notes: 4594 The i, j, and a arrays ARE NOT copied by this routine into the internal format used by PETSc. 4595 4596 The i and j indices are 0 based 4597 4598 See MatCreateMPIAIJ() for the definition of "diagonal" and "off-diagonal" portion of the matrix 4599 4600 4601 .keywords: matrix, aij, compressed row, sparse, parallel 4602 4603 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(), 4604 MPIAIJ, MatCreateMPIAIJ(), MatCreateMPIAIJWithArrays() 4605 @*/ 4606 PetscErrorCode PETSCMAT_DLLEXPORT MatCreateMPIAIJWithSplitArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt i[],PetscInt j[],PetscScalar a[], 4607 PetscInt oi[], PetscInt oj[],PetscScalar oa[],Mat *mat) 4608 { 4609 PetscErrorCode ierr; 4610 Mat_MPIAIJ *maij; 4611 4612 PetscFunctionBegin; 4613 if (m < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative"); 4614 if (i[0]) { 4615 SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4616 } 4617 if (oi[0]) { 4618 SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"oi (row indices) must start with 0"); 4619 } 4620 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4621 ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr); 4622 ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr); 4623 maij = (Mat_MPIAIJ*) (*mat)->data; 4624 maij->donotstash = PETSC_TRUE; 4625 (*mat)->preallocated = PETSC_TRUE; 4626 4627 (*mat)->rmap.bs = (*mat)->cmap.bs = 1; 4628 ierr = PetscMapSetUp(&(*mat)->rmap);CHKERRQ(ierr); 4629 ierr = PetscMapSetUp(&(*mat)->cmap);CHKERRQ(ierr); 4630 4631 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,i,j,a,&maij->A);CHKERRQ(ierr); 4632 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,(*mat)->cmap.N,oi,oj,oa,&maij->B);CHKERRQ(ierr); 4633 4634 ierr = MatAssemblyBegin(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4635 ierr = MatAssemblyEnd(maij->A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4636 ierr = MatAssemblyBegin(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4637 ierr = MatAssemblyEnd(maij->B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4638 4639 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4640 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4641 PetscFunctionReturn(0); 4642 } 4643 4644 /* 4645 Special version for direct calls from Fortran 4646 */ 4647 #if defined(PETSC_HAVE_FORTRAN_CAPS) 4648 #define matsetvaluesmpiaij_ MATSETVALUESMPIAIJ 4649 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 4650 #define matsetvaluesmpiaij_ matsetvaluesmpiaij 4651 #endif 4652 4653 /* Change these macros so can be used in void function */ 4654 #undef CHKERRQ 4655 #define CHKERRQ(ierr) CHKERRABORT(mat->comm,ierr) 4656 #undef SETERRQ2 4657 #define SETERRQ2(ierr,b,c,d) CHKERRABORT(mat->comm,ierr) 4658 #undef SETERRQ 4659 #define SETERRQ(ierr,b) CHKERRABORT(mat->comm,ierr) 4660 4661 EXTERN_C_BEGIN 4662 #undef __FUNCT__ 4663 #define __FUNCT__ "matsetvaluesmpiaij_" 4664 void PETSC_STDCALL matsetvaluesmpiaij_(Mat *mmat,PetscInt *mm,const PetscInt im[],PetscInt *mn,const PetscInt in[],const PetscScalar v[],InsertMode *maddv,PetscErrorCode *_ierr) 4665 { 4666 Mat mat = *mmat; 4667 PetscInt m = *mm, n = *mn; 4668 InsertMode addv = *maddv; 4669 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 4670 PetscScalar value; 4671 PetscErrorCode ierr; 4672 4673 MatPreallocated(mat); 4674 if (mat->insertmode == NOT_SET_VALUES) { 4675 mat->insertmode = addv; 4676 } 4677 #if defined(PETSC_USE_DEBUG) 4678 else if (mat->insertmode != addv) { 4679 SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Cannot mix add values and insert values"); 4680 } 4681 #endif 4682 { 4683 PetscInt i,j,rstart = mat->rmap.rstart,rend = mat->rmap.rend; 4684 PetscInt cstart = mat->cmap.rstart,cend = mat->cmap.rend,row,col; 4685 PetscTruth roworiented = aij->roworiented; 4686 4687 /* Some Variables required in the macro */ 4688 Mat A = aij->A; 4689 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4690 PetscInt *aimax = a->imax,*ai = a->i,*ailen = a->ilen,*aj = a->j; 4691 PetscScalar *aa = a->a; 4692 PetscTruth ignorezeroentries = (((a->ignorezeroentries)&&(addv==ADD_VALUES))?PETSC_TRUE:PETSC_FALSE); 4693 Mat B = aij->B; 4694 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 4695 PetscInt *bimax = b->imax,*bi = b->i,*bilen = b->ilen,*bj = b->j,bm = aij->B->rmap.n,am = aij->A->rmap.n; 4696 PetscScalar *ba = b->a; 4697 4698 PetscInt *rp1,*rp2,ii,nrow1,nrow2,_i,rmax1,rmax2,N,low1,high1,low2,high2,t,lastcol1,lastcol2; 4699 PetscInt nonew = a->nonew; 4700 PetscScalar *ap1,*ap2; 4701 4702 PetscFunctionBegin; 4703 for (i=0; i<m; i++) { 4704 if (im[i] < 0) continue; 4705 #if defined(PETSC_USE_DEBUG) 4706 if (im[i] >= mat->rmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap.N-1); 4707 #endif 4708 if (im[i] >= rstart && im[i] < rend) { 4709 row = im[i] - rstart; 4710 lastcol1 = -1; 4711 rp1 = aj + ai[row]; 4712 ap1 = aa + ai[row]; 4713 rmax1 = aimax[row]; 4714 nrow1 = ailen[row]; 4715 low1 = 0; 4716 high1 = nrow1; 4717 lastcol2 = -1; 4718 rp2 = bj + bi[row]; 4719 ap2 = ba + bi[row]; 4720 rmax2 = bimax[row]; 4721 nrow2 = bilen[row]; 4722 low2 = 0; 4723 high2 = nrow2; 4724 4725 for (j=0; j<n; j++) { 4726 if (roworiented) value = v[i*n+j]; else value = v[i+j*m]; 4727 if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES)) continue; 4728 if (in[j] >= cstart && in[j] < cend){ 4729 col = in[j] - cstart; 4730 MatSetValues_SeqAIJ_A_Private(row,col,value,addv); 4731 } else if (in[j] < 0) continue; 4732 #if defined(PETSC_USE_DEBUG) 4733 else if (in[j] >= mat->cmap.N) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap.N-1);} 4734 #endif 4735 else { 4736 if (mat->was_assembled) { 4737 if (!aij->colmap) { 4738 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 4739 } 4740 #if defined (PETSC_USE_CTABLE) 4741 ierr = PetscTableFind(aij->colmap,in[j]+1,&col);CHKERRQ(ierr); 4742 col--; 4743 #else 4744 col = aij->colmap[in[j]] - 1; 4745 #endif 4746 if (col < 0 && !((Mat_SeqAIJ*)(aij->A->data))->nonew) { 4747 ierr = DisAssemble_MPIAIJ(mat);CHKERRQ(ierr); 4748 col = in[j]; 4749 /* Reinitialize the variables required by MatSetValues_SeqAIJ_B_Private() */ 4750 B = aij->B; 4751 b = (Mat_SeqAIJ*)B->data; 4752 bimax = b->imax; bi = b->i; bilen = b->ilen; bj = b->j; 4753 rp2 = bj + bi[row]; 4754 ap2 = ba + bi[row]; 4755 rmax2 = bimax[row]; 4756 nrow2 = bilen[row]; 4757 low2 = 0; 4758 high2 = nrow2; 4759 bm = aij->B->rmap.n; 4760 ba = b->a; 4761 } 4762 } else col = in[j]; 4763 MatSetValues_SeqAIJ_B_Private(row,col,value,addv); 4764 } 4765 } 4766 } else { 4767 if (!aij->donotstash) { 4768 if (roworiented) { 4769 if (ignorezeroentries && v[i*n] == 0.0) continue; 4770 ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n);CHKERRQ(ierr); 4771 } else { 4772 if (ignorezeroentries && v[i] == 0.0) continue; 4773 ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m);CHKERRQ(ierr); 4774 } 4775 } 4776 } 4777 }} 4778 PetscFunctionReturnVoid(); 4779 } 4780 EXTERN_C_END 4781 4782