1 2 #include <../src/mat/impls/aij/mpi/mpiaij.h> 3 4 extern PetscErrorCode MatCreateColmap_MPIAIJ_Private(Mat); 5 6 #undef __FUNCT__ 7 #define __FUNCT__ "MatFDColoringApply_MPIAIJ" 8 PetscErrorCode MatFDColoringApply_MPIAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 9 { 10 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 11 PetscErrorCode ierr; 12 PetscInt k,cstart,cend,l,row,col,colb,nz; 13 PetscScalar dx=0.0,*y,*xx,*w3_array; 14 PetscScalar *vscale_array; 15 PetscReal epsilon=coloring->error_rel,umin=coloring->umin,unorm; 16 Vec w1=coloring->w1,w2=coloring->w2,w3,vscale=coloring->vscale; 17 void *fctx=coloring->fctx; 18 PetscBool flg=PETSC_FALSE; 19 PetscInt ctype=coloring->ctype,nxloc; 20 Mat_MPIAIJ *aij=(Mat_MPIAIJ*)J->data; 21 MatEntry *Jentry=coloring->matentry; 22 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 23 24 PetscFunctionBegin; 25 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 26 ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 27 if (flg) { 28 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 29 } else { 30 PetscBool assembled; 31 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 32 if (assembled) { 33 ierr = MatZeroEntries(J);CHKERRQ(ierr); 34 } 35 } 36 37 /* create vscale for storing dx */ 38 if (!vscale) { 39 if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] == 'd') { 40 ierr = VecCreateGhost(PetscObjectComm((PetscObject)J),J->cmap->n,PETSC_DETERMINE,aij->B->cmap->n,aij->garray,&vscale);CHKERRQ(ierr); 41 } else if (ctype == IS_COLORING_GHOSTED) { 42 ierr = VecDuplicate(x1,&vscale);CHKERRQ(ierr); 43 } 44 coloring->vscale = vscale; 45 } 46 47 /* (1) Set w1 = F(x1) */ 48 if (!coloring->fset) { 49 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 50 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 51 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 52 } else { 53 coloring->fset = PETSC_FALSE; 54 } 55 56 /* (2) Compute vscale = 1./dx - the local scale factors, including ghost points */ 57 if (coloring->htype[0] == 'w') { 58 /* vscale = dx is a constant scalar */ 59 ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr); 60 dx = 1.0/((1.0 + unorm)*epsilon); 61 } else { 62 ierr = VecGetLocalSize(x1,&nxloc);CHKERRQ(ierr); 63 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr); 64 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 65 for (col=0; col<nxloc; col++) { 66 dx = xx[col]; 67 if (PetscAbsScalar(dx) < umin) { 68 if (PetscRealPart(dx) >= 0.0) dx = umin; 69 else if (PetscRealPart(dx) < 0.0 ) dx = -umin; 70 } 71 dx *= epsilon; 72 vscale_array[col] = 1.0/dx; 73 } 74 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 75 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 76 } 77 if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] != 'w') { 78 ierr = VecGhostUpdateBegin(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 79 ierr = VecGhostUpdateEnd(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 80 } 81 82 /* (3) Loop over each color */ 83 if (!coloring->w3) { 84 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 85 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 86 } 87 w3 = coloring->w3; 88 89 ierr = VecGetOwnershipRange(x1,&cstart,&cend);CHKERRQ(ierr); /* used by ghosted vscale */ 90 if (vscale) { 91 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 92 } 93 nz = 0; 94 for (k=0; k<ncolors; k++) { 95 coloring->currentcolor = k; 96 97 /* 98 (3-1) Loop over each column associated with color 99 adding the perturbation to the vector w3 = x1 + dx. 100 */ 101 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 102 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 103 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array - cstart; /* shift pointer so global index can be used */ 104 for (l=0; l<ncolumns[k]; l++) { 105 col = coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */ 106 if (coloring->htype[0] == 'w') { 107 w3_array[col] += 1.0/dx; 108 continue; 109 } 110 /* convert global col to local colb */ 111 if (col >= cstart && col < cend) { 112 colb = col - cstart; 113 } else { 114 #if defined(PETSC_USE_CTABLE) 115 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 116 colb--; 117 #else 118 colb = aij->colmap[col] - 1; /* local column index */ 119 #endif 120 colb += cend - cstart; 121 } 122 w3_array[col] += 1.0/vscale_array[colb]; 123 } 124 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array + cstart; 125 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 126 127 /* 128 (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 129 w2 = F(x1 + dx) - F(x1) 130 */ 131 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 132 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 133 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 134 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 135 136 /* 137 (3-3) Loop over rows of vector, putting results into Jacobian matrix 138 */ 139 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 140 for (l=0; l<nrows[k]; l++) { 141 row = Jentry[nz].row; /* local row index */ 142 if (coloring->htype[0] == 'w') { 143 *(Jentry[nz].valaddr) = y[row]*dx; 144 } else { 145 *(Jentry[nz].valaddr) = y[row]*vscale_array[Jentry[nz].col]; 146 } 147 nz++; 148 } 149 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 150 } 151 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 152 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 153 if (vscale) { 154 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 155 } 156 157 coloring->currentcolor = -1; 158 PetscFunctionReturn(0); 159 } 160 161 #undef __FUNCT__ 162 #define __FUNCT__ "MatFDColoringCreate_MPIAIJ_new" 163 PetscErrorCode MatFDColoringCreate_MPIAIJ_new(Mat mat,ISColoring iscoloring,MatFDColoring c) 164 { 165 Mat_MPIAIJ *aij=(Mat_MPIAIJ*)mat->data; 166 PetscErrorCode ierr; 167 PetscMPIInt size,*ncolsonproc,*disp,nn; 168 PetscInt i,n,nrows,j,k,m,ncols,col; 169 const PetscInt *is,*A_ci,*A_cj,*B_ci,*B_cj,*row = NULL,*ltog=NULL; 170 PetscInt nis=iscoloring->n,nctot,*cols; 171 PetscInt *rowhit,cstart,cend,colb; 172 IS *isa; 173 ISLocalToGlobalMapping map=mat->cmap->mapping; 174 PetscInt ctype=c->ctype; 175 Mat A=aij->A,B=aij->B; 176 Mat_SeqAIJ *spA=(Mat_SeqAIJ*)A->data,*spB=(Mat_SeqAIJ*)B->data; 177 PetscScalar *A_val=spA->a,*B_val=spB->a; 178 PetscInt spidx; 179 PetscInt *spidxA,*spidxB,nz; 180 PetscScalar **valaddrhit; 181 MatEntry *Jentry; 182 183 PetscFunctionBegin; 184 if (ctype == IS_COLORING_GHOSTED) { 185 if(!map) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping"); 186 ierr = ISLocalToGlobalMappingGetIndices(map,<og);CHKERRQ(ierr); 187 } 188 189 m = mat->rmap->n; 190 cstart = mat->cmap->rstart; 191 cend = mat->cmap->rend; 192 c->M = mat->rmap->N; /* set the global rows and columns and local rows */ 193 c->N = mat->cmap->N; 194 c->m = m; 195 c->rstart = mat->rmap->rstart; 196 197 c->ncolors = nis; 198 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 199 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr); 200 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 201 ierr = PetscLogObjectMemory((PetscObject)c,3*nis*sizeof(PetscInt));CHKERRQ(ierr); 202 203 nz = spA->nz + spB->nz; /* total nonzero entries of mat */ 204 ierr = PetscMalloc(nz*sizeof(MatEntry),&Jentry);CHKERRQ(ierr); 205 ierr = PetscLogObjectMemory((PetscObject)c,nz*sizeof(MatEntry));CHKERRQ(ierr); 206 c->matentry = Jentry; 207 208 /* Allow access to data structures of local part of matrix 209 - creates aij->colmap which maps global column number to local number in part B */ 210 if (!aij->colmap) { 211 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 212 } 213 ierr = MatGetColumnIJ_SeqAIJ_Color(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 214 ierr = MatGetColumnIJ_SeqAIJ_Color(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 215 216 ierr = PetscMalloc2(m+1,PetscInt,&rowhit,m+1,PetscScalar*,&valaddrhit);CHKERRQ(ierr); 217 nz = 0; 218 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 219 for (i=0; i<nis; i++) { /* for each local color */ 220 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 221 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 222 223 c->ncolumns[i] = n; /* local number of columns of this color on this process */ 224 if (n) { 225 ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr); 226 ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr); 227 ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr); 228 } else { 229 c->columns[i] = 0; 230 } 231 232 if (ctype == IS_COLORING_GLOBAL) { 233 /* Determine nctot, the total (parallel) number of columns of this color */ 234 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr); 235 ierr = PetscMalloc2(size,PetscMPIInt,&ncolsonproc,size,PetscMPIInt,&disp);CHKERRQ(ierr); 236 237 /* ncolsonproc[j]: local ncolumns on proc[j] of this color */ 238 ierr = PetscMPIIntCast(n,&nn);CHKERRQ(ierr); 239 ierr = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 240 nctot = 0; for (j=0; j<size; j++) nctot += ncolsonproc[j]; 241 if (!nctot) { 242 ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr); 243 } 244 245 disp[0] = 0; 246 for (j=1; j<size; j++) { 247 disp[j] = disp[j-1] + ncolsonproc[j-1]; 248 } 249 250 /* Get cols, the complete list of columns for this color on each process */ 251 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 252 ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 253 ierr = PetscFree2(ncolsonproc,disp);CHKERRQ(ierr); 254 } else if (ctype == IS_COLORING_GHOSTED) { 255 /* Determine local number of columns of this color on this process, including ghost points */ 256 nctot = n; 257 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 258 ierr = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr); 259 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not provided for this MatFDColoring type"); 260 261 /* Mark all rows affect by these columns */ 262 ierr = PetscMemzero(rowhit,m*sizeof(PetscInt));CHKERRQ(ierr); 263 for (j=0; j<nctot; j++) { /* loop over columns*/ 264 if (ctype == IS_COLORING_GHOSTED) { 265 col = ltog[cols[j]]; 266 } else { 267 col = cols[j]; 268 } 269 if (col >= cstart && col < cend) { /* column is in diagonal block of matrix A */ 270 row = A_cj + A_ci[col-cstart]; 271 nrows = A_ci[col-cstart+1] - A_ci[col-cstart]; 272 /* loop over columns of A marking them in rowhit */ 273 for (k=0; k<nrows; k++) { 274 /* set valaddrhit for part A */ 275 spidx = spidxA[A_ci[col-cstart] + k]; 276 valaddrhit[*row] = &A_val[spidx]; 277 rowhit[*row++] = col - cstart + 1; /* local column index */ 278 } 279 } else { /* column is in off-diagonal block of matrix B */ 280 #if defined(PETSC_USE_CTABLE) 281 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 282 colb--; 283 #else 284 colb = aij->colmap[col] - 1; /* local column index */ 285 #endif 286 if (colb == -1) { 287 nrows = 0; 288 } else { 289 row = B_cj + B_ci[colb]; 290 nrows = B_ci[colb+1] - B_ci[colb]; 291 } 292 /* loop over columns of B marking them in rowhit */ 293 for (k=0; k<nrows; k++) { 294 /* set valaddrhit for part B */ 295 spidx = spidxB[B_ci[colb] + k]; 296 valaddrhit[*row] = &B_val[spidx]; 297 rowhit[*row++] = colb + 1 + cend - cstart; /* local column index */ 298 } 299 } 300 } 301 302 /* count the number of hits */ 303 nrows = 0; 304 for (j=0; j<m; j++) { 305 if (rowhit[j]) nrows++; 306 } 307 c->nrows[i] = nrows; 308 309 for (j=0; j<m; j++) { 310 if (rowhit[j]) { 311 Jentry[nz].row = j; /* local row index */ 312 Jentry[nz].col = rowhit[j] - 1; /* local column index */ 313 Jentry[nz].valaddr = valaddrhit[j]; /* address of mat value for this entry */ 314 nz++; 315 } 316 } 317 ierr = PetscFree(cols);CHKERRQ(ierr); 318 } 319 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 320 if (nz != spA->nz + spB->nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"nz %d != mat->nz %d\n",nz,spA->nz+spB->nz); 321 322 ierr = PetscFree2(rowhit,valaddrhit);CHKERRQ(ierr); 323 ierr = MatRestoreColumnIJ_SeqAIJ_Color(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 324 ierr = MatRestoreColumnIJ_SeqAIJ_Color(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 325 if (ctype == IS_COLORING_GHOSTED) { 326 ierr = ISLocalToGlobalMappingRestoreIndices(map,<og);CHKERRQ(ierr); 327 } 328 329 mat->ops->fdcoloringapply = MatFDColoringApply_MPIAIJ; 330 PetscFunctionReturn(0); 331 } 332 333 /*------------------------------------------------------*/ 334 #undef __FUNCT__ 335 #define __FUNCT__ "MatFDColoringCreate_MPIAIJ" 336 PetscErrorCode MatFDColoringCreate_MPIAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c) 337 { 338 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 339 PetscErrorCode ierr; 340 PetscMPIInt size,*ncolsonproc,*disp,nn; 341 PetscInt i,n,nrows,j,k,m,ncols,col; 342 const PetscInt *is,*A_ci,*A_cj,*B_ci,*B_cj,*rows = 0,*ltog; 343 PetscInt nis = iscoloring->n,nctot,*cols; 344 PetscInt *rowhit,M,cstart,cend,colb; 345 PetscInt *columnsforrow,l; 346 IS *isa; 347 PetscBool done,flg; 348 ISLocalToGlobalMapping map = mat->cmap->mapping; 349 PetscInt ctype=c->ctype; 350 PetscBool new_impl=PETSC_FALSE; 351 352 PetscFunctionBegin; 353 ierr = PetscOptionsName("-new","using new impls","",&new_impl);CHKERRQ(ierr); 354 if (new_impl){ 355 ierr = MatFDColoringCreate_MPIAIJ_new(mat,iscoloring,c);CHKERRQ(ierr); 356 PetscFunctionReturn(0); 357 } 358 if (ctype == IS_COLORING_GHOSTED && !map) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping"); 359 360 if (map) {ierr = ISLocalToGlobalMappingGetIndices(map,<og);CHKERRQ(ierr);} 361 else ltog = NULL; 362 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 363 364 M = mat->rmap->n; 365 cstart = mat->cmap->rstart; 366 cend = mat->cmap->rend; 367 c->M = mat->rmap->N; /* set the global rows and columns and local rows */ 368 c->N = mat->cmap->N; 369 c->m = mat->rmap->n; 370 c->rstart = mat->rmap->rstart; 371 372 c->ncolors = nis; 373 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 374 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr); 375 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 376 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->rows);CHKERRQ(ierr); 377 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columnsforrow);CHKERRQ(ierr); 378 ierr = PetscLogObjectMemory((PetscObject)c,5*nis*sizeof(PetscInt));CHKERRQ(ierr); 379 380 /* Allow access to data structures of local part of matrix */ 381 if (!aij->colmap) { 382 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 383 } 384 ierr = MatGetColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr); 385 ierr = MatGetColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr); 386 387 ierr = PetscMalloc((M+1)*sizeof(PetscInt),&rowhit);CHKERRQ(ierr); 388 ierr = PetscMalloc((M+1)*sizeof(PetscInt),&columnsforrow);CHKERRQ(ierr); 389 390 for (i=0; i<nis; i++) { 391 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 392 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 393 394 c->ncolumns[i] = n; /* local number of columns of this color on this process */ 395 if (n) { 396 ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr); 397 ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr); 398 ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr); 399 } else { 400 c->columns[i] = 0; 401 } 402 403 if (ctype == IS_COLORING_GLOBAL) { 404 /* Determine the total (parallel) number of columns of this color */ 405 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr); 406 ierr = PetscMalloc2(size,PetscMPIInt,&ncolsonproc,size,PetscMPIInt,&disp);CHKERRQ(ierr); 407 408 ierr = PetscMPIIntCast(n,&nn);CHKERRQ(ierr); 409 ierr = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 410 nctot = 0; for (j=0; j<size; j++) nctot += ncolsonproc[j]; 411 if (!nctot) { 412 ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr); 413 } 414 415 disp[0] = 0; 416 for (j=1; j<size; j++) { 417 disp[j] = disp[j-1] + ncolsonproc[j-1]; 418 } 419 420 /* Get complete list of columns for color on each processor */ 421 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 422 ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 423 ierr = PetscFree2(ncolsonproc,disp);CHKERRQ(ierr); 424 } else if (ctype == IS_COLORING_GHOSTED) { 425 /* Determine local number of columns of this color on this process, including ghost points */ 426 nctot = n; 427 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 428 ierr = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr); 429 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not provided for this MatFDColoring type"); 430 431 /* 432 Mark all rows affect by these columns 433 */ 434 /* Temporary option to allow for debugging/testing */ 435 flg = PETSC_FALSE; 436 ierr = PetscOptionsGetBool(NULL,"-matfdcoloring_slow",&flg,NULL);CHKERRQ(ierr); 437 if (!flg) { /*-----------------------------------------------------------------------------*/ 438 /* crude, fast version */ 439 ierr = PetscMemzero(rowhit,M*sizeof(PetscInt));CHKERRQ(ierr); 440 /* loop over columns*/ 441 for (j=0; j<nctot; j++) { 442 if (ctype == IS_COLORING_GHOSTED) { 443 col = ltog[cols[j]]; 444 } else { 445 col = cols[j]; 446 } 447 if (col >= cstart && col < cend) { 448 /* column is in diagonal block of matrix */ 449 rows = A_cj + A_ci[col-cstart]; 450 m = A_ci[col-cstart+1] - A_ci[col-cstart]; 451 } else { 452 #if defined(PETSC_USE_CTABLE) 453 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 454 colb--; 455 #else 456 colb = aij->colmap[col] - 1; 457 #endif 458 if (colb == -1) { 459 m = 0; 460 } else { 461 rows = B_cj + B_ci[colb]; 462 m = B_ci[colb+1] - B_ci[colb]; 463 } 464 } 465 /* loop over columns marking them in rowhit */ 466 for (k=0; k<m; k++) { 467 rowhit[*rows++] = col + 1; 468 } 469 } 470 471 /* count the number of hits */ 472 nrows = 0; 473 for (j=0; j<M; j++) { 474 if (rowhit[j]) nrows++; 475 } 476 c->nrows[i] = nrows; 477 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr); 478 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr); 479 ierr = PetscLogObjectMemory((PetscObject)c,2*(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr); 480 nrows = 0; 481 for (j=0; j<M; j++) { 482 if (rowhit[j]) { 483 c->rows[i][nrows] = j; /* local row index */ 484 c->columnsforrow[i][nrows] = rowhit[j] - 1; /* global column index */ 485 nrows++; 486 } 487 } 488 } else { /*-------------------------------------------------------------------------------*/ 489 /* slow version, using rowhit as a linked list */ 490 PetscInt currentcol,fm,mfm; 491 rowhit[M] = M; 492 nrows = 0; 493 /* loop over columns*/ 494 for (j=0; j<nctot; j++) { 495 if (ctype == IS_COLORING_GHOSTED) { 496 col = ltog[cols[j]]; 497 } else { 498 col = cols[j]; 499 } 500 if (col >= cstart && col < cend) { 501 /* column is in diagonal block of matrix */ 502 rows = A_cj + A_ci[col-cstart]; 503 m = A_ci[col-cstart+1] - A_ci[col-cstart]; 504 } else { 505 #if defined(PETSC_USE_CTABLE) 506 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 507 colb--; 508 #else 509 colb = aij->colmap[col] - 1; 510 #endif 511 if (colb == -1) { 512 m = 0; 513 } else { 514 rows = B_cj + B_ci[colb]; 515 m = B_ci[colb+1] - B_ci[colb]; 516 } 517 } 518 519 /* loop over columns marking them in rowhit */ 520 fm = M; /* fm points to first entry in linked list */ 521 for (k=0; k<m; k++) { 522 currentcol = *rows++; 523 /* is it already in the list? */ 524 do { 525 mfm = fm; 526 fm = rowhit[fm]; 527 } while (fm < currentcol); 528 /* not in list so add it */ 529 if (fm != currentcol) { 530 nrows++; 531 columnsforrow[currentcol] = col; 532 /* next three lines insert new entry into linked list */ 533 rowhit[mfm] = currentcol; 534 rowhit[currentcol] = fm; 535 fm = currentcol; 536 /* fm points to present position in list since we know the columns are sorted */ 537 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Invalid coloring of matrix detected"); 538 } 539 } 540 c->nrows[i] = nrows; 541 542 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr); 543 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr); 544 ierr = PetscLogObjectMemory((PetscObject)c,(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr); 545 /* now store the linked list of rows into c->rows[i] */ 546 nrows = 0; 547 fm = rowhit[M]; 548 do { 549 c->rows[i][nrows] = fm; 550 c->columnsforrow[i][nrows++] = columnsforrow[fm]; 551 fm = rowhit[fm]; 552 } while (fm < M); 553 } /* ---------------------------------------------------------------------------------------*/ 554 ierr = PetscFree(cols);CHKERRQ(ierr); 555 } 556 557 /* Optimize by adding the vscale, and scaleforrow[][] fields */ 558 /* 559 vscale will contain the "diagonal" on processor scalings followed by the off processor 560 */ 561 if (ctype == IS_COLORING_GLOBAL) { 562 ierr = VecCreateGhost(PetscObjectComm((PetscObject)mat),aij->A->rmap->n,PETSC_DETERMINE,aij->B->cmap->n,aij->garray,&c->vscale);CHKERRQ(ierr); 563 ierr = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr); 564 for (k=0; k<c->ncolors; k++) { 565 ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr); 566 for (l=0; l<c->nrows[k]; l++) { 567 col = c->columnsforrow[k][l]; 568 if (col >= cstart && col < cend) { 569 /* column is in diagonal block of matrix */ 570 colb = col - cstart; 571 } else { 572 /* column is in "off-processor" part */ 573 #if defined(PETSC_USE_CTABLE) 574 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 575 colb--; 576 #else 577 colb = aij->colmap[col] - 1; 578 #endif 579 colb += cend - cstart; 580 } 581 c->vscaleforrow[k][l] = colb; 582 } 583 } 584 } else if (ctype == IS_COLORING_GHOSTED) { 585 /* Get gtol mapping */ 586 PetscInt N = mat->cmap->N,nlocal,*gtol; 587 ierr = PetscMalloc((N+1)*sizeof(PetscInt),>ol);CHKERRQ(ierr); 588 for (i=0; i<N; i++) gtol[i] = -1; 589 ierr = ISLocalToGlobalMappingGetSize(map,&nlocal);CHKERRQ(ierr); 590 for (i=0; i<nlocal; i++) gtol[ltog[i]] = i; 591 592 c->vscale = 0; /* will be created in MatFDColoringApply() */ 593 ierr = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr); 594 for (k=0; k<c->ncolors; k++) { 595 ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr); 596 for (l=0; l<c->nrows[k]; l++) { 597 col = c->columnsforrow[k][l]; /* global column index */ 598 c->vscaleforrow[k][l] = gtol[col]; /* local column index */ 599 } 600 } 601 ierr = PetscFree(gtol);CHKERRQ(ierr); 602 } 603 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 604 605 ierr = PetscFree(rowhit);CHKERRQ(ierr); 606 ierr = PetscFree(columnsforrow);CHKERRQ(ierr); 607 ierr = MatRestoreColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr); 608 ierr = MatRestoreColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr); 609 if (map) {ierr = ISLocalToGlobalMappingRestoreIndices(map,<og);CHKERRQ(ierr);} 610 PetscFunctionReturn(0); 611 } 612 613 614 615 616 617 618