1 2 #include <../src/mat/impls/aij/mpi/mpiaij.h> 3 #include <../src/mat/impls/baij/mpi/mpibaij.h> 4 5 #undef __FUNCT__ 6 #define __FUNCT__ "MatFDColoringApply_BAIJ" 7 PetscErrorCode MatFDColoringApply_BAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 8 { 9 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 10 PetscErrorCode ierr; 11 PetscInt k,cstart,cend,l,row,col,nz,spidx,i,j; 12 PetscScalar dx=0.0,*xx,*w3_array,*dy_i,*dy=coloring->dy; 13 PetscScalar *vscale_array; 14 PetscReal epsilon=coloring->error_rel,umin=coloring->umin,unorm; 15 Vec w1=coloring->w1,w2=coloring->w2,w3,vscale=coloring->vscale; 16 void *fctx=coloring->fctx; 17 PetscInt ctype=coloring->ctype,nxloc,nrows_k; 18 PetscScalar *valaddr; 19 MatEntry *Jentry=coloring->matentry; 20 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 21 PetscInt bs=J->rmap->bs; 22 23 PetscFunctionBegin; 24 /* (1) Set w1 = F(x1) */ 25 if (!coloring->fset) { 26 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 27 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 28 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 29 } else { 30 coloring->fset = PETSC_FALSE; 31 } 32 33 /* (2) Compute vscale = 1./dx - the local scale factors, including ghost points */ 34 ierr = VecGetLocalSize(x1,&nxloc);CHKERRQ(ierr); 35 if (coloring->htype[0] == 'w') { 36 /* vscale = dx is a constant scalar */ 37 ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr); 38 dx = 1.0/(PetscSqrtReal(1.0 + unorm)*epsilon); 39 } else { 40 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr); 41 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 42 for (col=0; col<nxloc; col++) { 43 dx = xx[col]; 44 if (PetscAbsScalar(dx) < umin) { 45 if (PetscRealPart(dx) >= 0.0) dx = umin; 46 else if (PetscRealPart(dx) < 0.0 ) dx = -umin; 47 } 48 dx *= epsilon; 49 vscale_array[col] = 1.0/dx; 50 } 51 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 52 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 53 } 54 if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] == 'd') { 55 ierr = VecGhostUpdateBegin(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 56 ierr = VecGhostUpdateEnd(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 57 } 58 59 /* (3) Loop over each color */ 60 if (!coloring->w3) { 61 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 62 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 63 } 64 w3 = coloring->w3; 65 66 ierr = VecGetOwnershipRange(x1,&cstart,&cend);CHKERRQ(ierr); /* used by ghosted vscale */ 67 if (vscale) { 68 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 69 } 70 nz = 0; 71 for (k=0; k<ncolors; k++) { 72 coloring->currentcolor = k; 73 74 /* 75 (3-1) Loop over each column associated with color 76 adding the perturbation to the vector w3 = x1 + dx. 77 */ 78 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 79 dy_i = dy; 80 for (i=0; i<bs; i++) { /* Loop over a block of columns */ 81 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 82 if (ctype == IS_COLORING_GLOBAL) w3_array -= cstart; /* shift pointer so global index can be used */ 83 if (coloring->htype[0] == 'w') { 84 for (l=0; l<ncolumns[k]; l++) { 85 col = i + bs*coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */ 86 w3_array[col] += 1.0/dx; 87 if (i) w3_array[col-1] -= 1.0/dx; /* resume original w3[col-1] */ 88 } 89 } else { /* htype == 'ds' */ 90 vscale_array -= cstart; /* shift pointer so global index can be used */ 91 for (l=0; l<ncolumns[k]; l++) { 92 col = i + bs*coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */ 93 w3_array[col] += 1.0/vscale_array[col]; 94 if (i) w3_array[col-1] -= 1.0/vscale_array[col-1]; /* resume original w3[col-1] */ 95 } 96 vscale_array += cstart; 97 } 98 if (ctype == IS_COLORING_GLOBAL) w3_array += cstart; 99 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 100 101 /* 102 (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 103 w2 = F(x1 + dx) - F(x1) 104 */ 105 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 106 ierr = VecPlaceArray(w2,dy_i);CHKERRQ(ierr); /* place w2 to the array dy_i */ 107 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 108 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 109 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 110 ierr = VecResetArray(w2);CHKERRQ(ierr); 111 dy_i += nxloc; /* points to dy+i*nxloc */ 112 } 113 114 /* 115 (3-3) Loop over rows of vector, putting results into Jacobian matrix 116 */ 117 nrows_k = nrows[k]; 118 if (coloring->htype[0] == 'w') { 119 for (l=0; l<nrows_k; l++) { 120 row = bs*Jentry[nz].row; /* local row index */ 121 valaddr = Jentry[nz++].valaddr; 122 spidx = 0; 123 dy_i = dy; 124 for (i=0; i<bs; i++) { /* column of the block */ 125 for (j=0; j<bs; j++) { /* row of the block */ 126 valaddr[spidx++] = dy_i[row+j]*dx; 127 } 128 dy_i += nxloc; /* points to dy+i*nxloc */ 129 } 130 } 131 } else { /* htype == 'ds' */ 132 for (l=0; l<nrows_k; l++) { 133 row = bs*Jentry[nz].row; /* local row index */ 134 col = bs*Jentry[nz].col; /* local column index */ 135 valaddr = Jentry[nz++].valaddr; 136 spidx = 0; 137 dy_i = dy; 138 for (i=0; i<bs; i++) { /* column of the block */ 139 for (j=0; j<bs; j++) { /* row of the block */ 140 valaddr[spidx++] = dy_i[row+j]*vscale_array[col+i]; 141 } 142 dy_i += nxloc; /* points to dy+i*nxloc */ 143 } 144 } 145 } 146 } 147 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 148 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 149 if (vscale) { 150 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 151 } 152 153 coloring->currentcolor = -1; 154 PetscFunctionReturn(0); 155 } 156 157 #undef __FUNCT__ 158 #define __FUNCT__ "MatFDColoringApply_AIJ" 159 PetscErrorCode MatFDColoringApply_AIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 160 { 161 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 162 PetscErrorCode ierr; 163 PetscInt k,cstart,cend,l,row,col,nz; 164 PetscScalar dx=0.0,*y,*xx,*w3_array; 165 PetscScalar *vscale_array; 166 PetscReal epsilon=coloring->error_rel,umin=coloring->umin,unorm; 167 Vec w1=coloring->w1,w2=coloring->w2,w3,vscale=coloring->vscale; 168 void *fctx=coloring->fctx; 169 PetscInt ctype=coloring->ctype,nxloc,nrows_k; 170 MatEntry *Jentry=coloring->matentry; 171 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 172 173 PetscFunctionBegin; 174 /* (1) Set w1 = F(x1) */ 175 if (!coloring->fset) { 176 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 177 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 178 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 179 } else { 180 coloring->fset = PETSC_FALSE; 181 } 182 183 /* (2) Compute vscale = 1./dx - the local scale factors, including ghost points */ 184 if (coloring->htype[0] == 'w') { 185 /* vscale = 1./dx is a constant scalar */ 186 ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr); 187 dx = 1.0/(PetscSqrtReal(1.0 + unorm)*epsilon); 188 } else { 189 ierr = VecGetLocalSize(x1,&nxloc);CHKERRQ(ierr); 190 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr); 191 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 192 for (col=0; col<nxloc; col++) { 193 dx = xx[col]; 194 if (PetscAbsScalar(dx) < umin) { 195 if (PetscRealPart(dx) >= 0.0) dx = umin; 196 else if (PetscRealPart(dx) < 0.0 ) dx = -umin; 197 } 198 dx *= epsilon; 199 vscale_array[col] = 1.0/dx; 200 } 201 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 202 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 203 } 204 if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] == 'd') { 205 ierr = VecGhostUpdateBegin(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 206 ierr = VecGhostUpdateEnd(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 207 } 208 209 /* (3) Loop over each color */ 210 if (!coloring->w3) { 211 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 212 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 213 } 214 w3 = coloring->w3; 215 216 ierr = VecGetOwnershipRange(x1,&cstart,&cend);CHKERRQ(ierr); /* used by ghosted vscale */ 217 if (vscale) { 218 ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr); 219 } 220 nz = 0; 221 222 if (coloring->bcols > 1) { /* use blocked insertion of Jentry */ 223 PetscInt i,m=J->rmap->n,nbcols,bcols=coloring->bcols; 224 PetscScalar *dy=coloring->dy,*dy_k; 225 226 nbcols = 0; 227 for (k=0; k<ncolors; k+=bcols) { 228 coloring->currentcolor = k; 229 230 /* 231 (3-1) Loop over each column associated with color 232 adding the perturbation to the vector w3 = x1 + dx. 233 */ 234 235 dy_k = dy; 236 if (k + bcols > ncolors) bcols = ncolors - k; 237 for (i=0; i<bcols; i++) { 238 239 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 240 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 241 if (ctype == IS_COLORING_GLOBAL) w3_array -= cstart; /* shift pointer so global index can be used */ 242 if (coloring->htype[0] == 'w') { 243 for (l=0; l<ncolumns[k+i]; l++) { 244 col = coloring->columns[k+i][l]; /* local column (in global index!) of the matrix we are probing for */ 245 w3_array[col] += 1.0/dx; 246 } 247 } else { /* htype == 'ds' */ 248 vscale_array -= cstart; /* shift pointer so global index can be used */ 249 for (l=0; l<ncolumns[k+i]; l++) { 250 col = coloring->columns[k+i][l]; /* local column (in global index!) of the matrix we are probing for */ 251 w3_array[col] += 1.0/vscale_array[col]; 252 } 253 vscale_array += cstart; 254 } 255 if (ctype == IS_COLORING_GLOBAL) w3_array += cstart; 256 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 257 258 /* 259 (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 260 w2 = F(x1 + dx) - F(x1) 261 */ 262 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 263 ierr = VecPlaceArray(w2,dy_k);CHKERRQ(ierr); /* place w2 to the array dy_i */ 264 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 265 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 266 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 267 ierr = VecResetArray(w2);CHKERRQ(ierr); 268 dy_k += m; /* points to dy+i*nxloc */ 269 } 270 271 /* 272 (3-3) Loop over block rows of vector, putting results into Jacobian matrix 273 */ 274 nrows_k = nrows[nbcols++]; 275 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 276 277 if (coloring->htype[0] == 'w') { 278 for (l=0; l<nrows_k; l++) { 279 row = Jentry[nz].row; /* local row index */ 280 *(Jentry[nz++].valaddr) = dy[row]*dx; 281 } 282 } else { /* htype == 'ds' */ 283 for (l=0; l<nrows_k; l++) { 284 row = Jentry[nz].row; /* local row index */ 285 *(Jentry[nz].valaddr) = dy[row]*vscale_array[Jentry[nz].col]; 286 nz++; 287 } 288 } 289 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 290 } 291 } else { /* bcols == 1 */ 292 for (k=0; k<ncolors; k++) { 293 coloring->currentcolor = k; 294 295 /* 296 (3-1) Loop over each column associated with color 297 adding the perturbation to the vector w3 = x1 + dx. 298 */ 299 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 300 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 301 if (ctype == IS_COLORING_GLOBAL) w3_array -= cstart; /* shift pointer so global index can be used */ 302 if (coloring->htype[0] == 'w') { 303 for (l=0; l<ncolumns[k]; l++) { 304 col = coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */ 305 w3_array[col] += 1.0/dx; 306 } 307 } else { /* htype == 'ds' */ 308 vscale_array -= cstart; /* shift pointer so global index can be used */ 309 for (l=0; l<ncolumns[k]; l++) { 310 col = coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */ 311 w3_array[col] += 1.0/vscale_array[col]; 312 } 313 vscale_array += cstart; 314 } 315 if (ctype == IS_COLORING_GLOBAL) w3_array += cstart; 316 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 317 318 /* 319 (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 320 w2 = F(x1 + dx) - F(x1) 321 */ 322 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 323 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 324 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 325 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 326 327 /* 328 (3-3) Loop over rows of vector, putting results into Jacobian matrix 329 */ 330 nrows_k = nrows[k]; 331 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 332 if (coloring->htype[0] == 'w') { 333 for (l=0; l<nrows_k; l++) { 334 row = Jentry[nz].row; /* local row index */ 335 *(Jentry[nz++].valaddr) = y[row]*dx; 336 } 337 } else { /* htype == 'ds' */ 338 for (l=0; l<nrows_k; l++) { 339 row = Jentry[nz].row; /* local row index */ 340 *(Jentry[nz].valaddr) = y[row]*vscale_array[Jentry[nz].col]; 341 nz++; 342 } 343 } 344 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 345 } 346 } 347 348 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 349 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 350 if (vscale) { 351 ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr); 352 } 353 coloring->currentcolor = -1; 354 PetscFunctionReturn(0); 355 } 356 357 #undef __FUNCT__ 358 #define __FUNCT__ "MatFDColoringSetUp_MPIXAIJ" 359 PetscErrorCode MatFDColoringSetUp_MPIXAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c) 360 { 361 PetscErrorCode ierr; 362 PetscMPIInt size,*ncolsonproc,*disp,nn; 363 PetscInt i,n,nrows,nrows_i,j,k,m,ncols,col,*rowhit,cstart,cend,colb; 364 const PetscInt *is,*A_ci,*A_cj,*B_ci,*B_cj,*row=NULL,*ltog=NULL; 365 PetscInt nis=iscoloring->n,nctot,*cols; 366 IS *isa; 367 ISLocalToGlobalMapping map=mat->cmap->mapping; 368 PetscInt ctype=c->ctype,*spidxA,*spidxB,nz,bs,bs2,spidx; 369 Mat A,B; 370 PetscScalar *A_val,*B_val,**valaddrhit; 371 MatEntry *Jentry; 372 PetscBool isBAIJ; 373 PetscInt bcols=c->bcols; 374 #if defined(PETSC_USE_CTABLE) 375 PetscTable colmap=NULL; 376 #else 377 PetscInt *colmap=NULL; /* local col number of off-diag col */ 378 #endif 379 380 PetscFunctionBegin; 381 if (ctype == IS_COLORING_GHOSTED) { 382 if (!map) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping"); 383 ierr = ISLocalToGlobalMappingGetIndices(map,<og);CHKERRQ(ierr); 384 } 385 386 ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr); 387 ierr = PetscObjectTypeCompare((PetscObject)mat,MATMPIBAIJ,&isBAIJ);CHKERRQ(ierr); 388 if (isBAIJ) { 389 Mat_MPIBAIJ *baij=(Mat_MPIBAIJ*)mat->data; 390 Mat_SeqBAIJ *spA,*spB; 391 A = baij->A; spA = (Mat_SeqBAIJ*)A->data; A_val = spA->a; 392 B = baij->B; spB = (Mat_SeqBAIJ*)B->data; B_val = spB->a; 393 nz = spA->nz + spB->nz; /* total nonzero entries of mat */ 394 if (!baij->colmap) { 395 ierr = MatCreateColmap_MPIBAIJ_Private(mat);CHKERRQ(ierr); 396 colmap = baij->colmap; 397 } 398 ierr = MatGetColumnIJ_SeqBAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 399 ierr = MatGetColumnIJ_SeqBAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 400 401 if (ctype == IS_COLORING_GLOBAL && c->htype[0] == 'd') { /* create vscale for storing dx */ 402 PetscInt *garray; 403 ierr = PetscMalloc(B->cmap->n*sizeof(PetscInt),&garray);CHKERRQ(ierr); 404 for (i=0; i<baij->B->cmap->n/bs; i++) { 405 for (j=0; j<bs; j++) { 406 garray[i*bs+j] = bs*baij->garray[i]+j; 407 } 408 } 409 ierr = VecCreateGhost(PetscObjectComm((PetscObject)mat),mat->cmap->n,PETSC_DETERMINE,B->cmap->n,garray,&c->vscale);CHKERRQ(ierr); 410 ierr = PetscFree(garray);CHKERRQ(ierr); 411 } 412 } else { 413 Mat_MPIAIJ *aij=(Mat_MPIAIJ*)mat->data; 414 Mat_SeqAIJ *spA,*spB; 415 A = aij->A; spA = (Mat_SeqAIJ*)A->data; A_val = spA->a; 416 B = aij->B; spB = (Mat_SeqAIJ*)B->data; B_val = spB->a; 417 nz = spA->nz + spB->nz; /* total nonzero entries of mat */ 418 if (!aij->colmap) { 419 /* Allow access to data structures of local part of matrix 420 - creates aij->colmap which maps global column number to local number in part B */ 421 ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 422 colmap = aij->colmap; 423 } 424 ierr = MatGetColumnIJ_SeqAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 425 ierr = MatGetColumnIJ_SeqAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 426 427 bs = 1; /* only bs=1 is supported for non MPIBAIJ matrix */ 428 429 if (ctype == IS_COLORING_GLOBAL && c->htype[0] == 'd') { /* create vscale for storing dx */ 430 ierr = VecCreateGhost(PetscObjectComm((PetscObject)mat),mat->cmap->n,PETSC_DETERMINE,B->cmap->n,aij->garray,&c->vscale);CHKERRQ(ierr); 431 } 432 } 433 434 m = mat->rmap->n/bs; 435 cstart = mat->cmap->rstart/bs; 436 cend = mat->cmap->rend/bs; 437 438 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 439 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr); 440 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 441 ierr = PetscLogObjectMemory((PetscObject)c,3*nis*sizeof(PetscInt));CHKERRQ(ierr); 442 443 ierr = PetscMalloc(nz*sizeof(MatEntry),&Jentry);CHKERRQ(ierr); 444 ierr = PetscLogObjectMemory((PetscObject)c,nz*sizeof(MatEntry));CHKERRQ(ierr); 445 c->matentry = Jentry; 446 447 ierr = PetscMalloc2(m+1,PetscInt,&rowhit,m+1,PetscScalar*,&valaddrhit);CHKERRQ(ierr); 448 nz = 0; 449 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 450 for (i=0; i<nis; i++) { /* for each local color */ 451 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 452 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 453 454 c->ncolumns[i] = n; /* local number of columns of this color on this process */ 455 if (n) { 456 ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr); 457 ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr); 458 ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr); 459 } else { 460 c->columns[i] = 0; 461 } 462 463 if (ctype == IS_COLORING_GLOBAL) { 464 /* Determine nctot, the total (parallel) number of columns of this color */ 465 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr); 466 ierr = PetscMalloc2(size,PetscMPIInt,&ncolsonproc,size,PetscMPIInt,&disp);CHKERRQ(ierr); 467 468 /* ncolsonproc[j]: local ncolumns on proc[j] of this color */ 469 ierr = PetscMPIIntCast(n,&nn);CHKERRQ(ierr); 470 ierr = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 471 nctot = 0; for (j=0; j<size; j++) nctot += ncolsonproc[j]; 472 if (!nctot) { 473 ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr); 474 } 475 476 disp[0] = 0; 477 for (j=1; j<size; j++) { 478 disp[j] = disp[j-1] + ncolsonproc[j-1]; 479 } 480 481 /* Get cols, the complete list of columns for this color on each process */ 482 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 483 ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr); 484 ierr = PetscFree2(ncolsonproc,disp);CHKERRQ(ierr); 485 } else if (ctype == IS_COLORING_GHOSTED) { 486 /* Determine local number of columns of this color on this process, including ghost points */ 487 nctot = n; 488 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 489 ierr = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr); 490 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not provided for this MatFDColoring type"); 491 492 /* Mark all rows affect by these columns */ 493 ierr = PetscMemzero(rowhit,m*sizeof(PetscInt));CHKERRQ(ierr); 494 bs2 = bs*bs; 495 nrows_i = 0; 496 for (j=0; j<nctot; j++) { /* loop over columns*/ 497 if (ctype == IS_COLORING_GHOSTED) { 498 col = ltog[cols[j]]; 499 } else { 500 col = cols[j]; 501 } 502 if (col >= cstart && col < cend) { /* column is in A, diagonal block of mat */ 503 row = A_cj + A_ci[col-cstart]; 504 nrows = A_ci[col-cstart+1] - A_ci[col-cstart]; 505 nrows_i += nrows; 506 /* loop over columns of A marking them in rowhit */ 507 for (k=0; k<nrows; k++) { 508 /* set valaddrhit for part A */ 509 spidx = bs2*spidxA[A_ci[col-cstart] + k]; 510 valaddrhit[*row] = &A_val[spidx]; 511 rowhit[*row++] = col - cstart + 1; /* local column index */ 512 } 513 } else { /* column is in B, off-diagonal block of mat */ 514 #if defined(PETSC_USE_CTABLE) 515 ierr = PetscTableFind(colmap,col+1,&colb);CHKERRQ(ierr); 516 colb--; 517 #else 518 colb = colmap[col] - 1; /* local column index */ 519 #endif 520 if (colb == -1) { 521 nrows = 0; 522 } else { 523 colb = colb/bs; 524 row = B_cj + B_ci[colb]; 525 nrows = B_ci[colb+1] - B_ci[colb]; 526 } 527 nrows_i += nrows; 528 /* loop over columns of B marking them in rowhit */ 529 for (k=0; k<nrows; k++) { 530 /* set valaddrhit for part B */ 531 spidx = bs2*spidxB[B_ci[colb] + k]; 532 valaddrhit[*row] = &B_val[spidx]; 533 rowhit[*row++] = colb + 1 + cend - cstart; /* local column index */ 534 } 535 } 536 } 537 c->nrows[i] = nrows_i; 538 539 for (j=0; j<m; j++) { 540 if (rowhit[j]) { 541 Jentry[nz].row = j; /* local row index */ 542 Jentry[nz].col = rowhit[j] - 1; /* local column index */ 543 Jentry[nz].valaddr = valaddrhit[j]; /* address of mat value for this entry */ 544 nz++; 545 } 546 } 547 ierr = PetscFree(cols);CHKERRQ(ierr); 548 } 549 550 if (bcols > 1) { /* reorder Jentry for faster MatFDColoringApply() */ 551 ierr = MatFDColoringSetUpBlocked_AIJ_Private(mat,c,nz);CHKERRQ(ierr); 552 } 553 554 if (isBAIJ) { 555 ierr = MatRestoreColumnIJ_SeqBAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 556 ierr = MatRestoreColumnIJ_SeqBAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 557 ierr = PetscMalloc(bs*mat->rmap->n*sizeof(PetscScalar),&c->dy);CHKERRQ(ierr); 558 } else { 559 ierr = MatRestoreColumnIJ_SeqAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr); 560 ierr = MatRestoreColumnIJ_SeqAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr); 561 } 562 563 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 564 ierr = PetscFree2(rowhit,valaddrhit);CHKERRQ(ierr); 565 566 if (ctype == IS_COLORING_GHOSTED) { 567 ierr = ISLocalToGlobalMappingRestoreIndices(map,<og);CHKERRQ(ierr); 568 } 569 #if defined(PETSC_USE_INFO) 570 ierr = PetscInfo3(c,"ncolors %D, brows %D and bcols %D are used.\n",c->ncolors,c->brows,c->bcols);CHKERRQ(ierr); 571 #endif 572 PetscFunctionReturn(0); 573 } 574 575 #undef __FUNCT__ 576 #define __FUNCT__ "MatFDColoringCreate_MPIXAIJ" 577 PetscErrorCode MatFDColoringCreate_MPIXAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c) 578 { 579 PetscErrorCode ierr; 580 PetscInt bs,nis=iscoloring->n,m=mat->rmap->n; 581 PetscBool isBAIJ; 582 583 PetscFunctionBegin; 584 /* set default brows and bcols for speedup inserting the dense matrix into sparse Jacobian; 585 bcols is chosen s.t. dy-array takes 50% of memory space as mat */ 586 ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr); 587 ierr = PetscObjectTypeCompare((PetscObject)mat,MATMPIBAIJ,&isBAIJ);CHKERRQ(ierr); 588 if (isBAIJ) { 589 c->brows = m; 590 c->bcols = 1; 591 } else { /* mpiaij matrix */ 592 /* bcols is chosen s.t. dy-array takes 50% of local memory space as mat */ 593 Mat_MPIAIJ *aij=(Mat_MPIAIJ*)mat->data; 594 Mat_SeqAIJ *spA,*spB; 595 Mat A,B; 596 PetscInt nz,brows,bcols; 597 PetscReal mem; 598 599 bs = 1; /* only bs=1 is supported for MPIAIJ matrix */ 600 601 A = aij->A; spA = (Mat_SeqAIJ*)A->data; 602 B = aij->B; spB = (Mat_SeqAIJ*)B->data; 603 nz = spA->nz + spB->nz; /* total local nonzero entries of mat */ 604 mem = nz*(sizeof(PetscScalar) + sizeof(PetscInt)) + 3*m*sizeof(PetscInt); 605 bcols = (PetscInt)(0.5*mem /(m*sizeof(PetscScalar))); 606 brows = 1000/bcols; 607 if (bcols > nis) bcols = nis; 608 if (brows == 0 || brows > m) brows = m; 609 c->brows = brows; 610 c->bcols = bcols; 611 } 612 613 c->M = mat->rmap->N/bs; /* set the global rows and columns and local rows */ 614 c->N = mat->cmap->N/bs; 615 c->m = mat->rmap->n/bs; 616 c->rstart = mat->rmap->rstart/bs; 617 c->ncolors = nis; 618 PetscFunctionReturn(0); 619 } 620