1 #define PETSCMAT_DLL 2 3 #include "../src/mat/impls/aij/mpi/mpiaij.h" 4 5 EXTERN PetscErrorCode CreateColmap_MPIAIJ_Private(Mat); 6 7 #undef __FUNCT__ 8 #define __FUNCT__ "MatFDColoringCreate_MPIAIJ" 9 PetscErrorCode MatFDColoringCreate_MPIAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c) 10 { 11 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)mat->data; 12 PetscErrorCode ierr; 13 PetscMPIInt size,*ncolsonproc,*disp,nn; 14 PetscInt i,n,nrows,j,k,m,*rows = 0,*A_ci,*A_cj,ncols,col; 15 const PetscInt *is; 16 PetscInt nis = iscoloring->n,nctot,*cols,*B_ci,*B_cj; 17 PetscInt *rowhit,M,cstart,cend,colb; 18 PetscInt *columnsforrow,l; 19 IS *isa; 20 PetscTruth done,flg; 21 ISLocalToGlobalMapping map = mat->mapping; 22 PetscInt *ltog = (map ? map->indices : (PetscInt*) PETSC_NULL) ,ctype=c->ctype; 23 24 PetscFunctionBegin; 25 if (!mat->assembled) { 26 SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Matrix must be assembled first; MatAssemblyBegin/End();"); 27 } 28 if (ctype == IS_COLORING_GHOSTED && !map) SETERRQ(PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping"); 29 30 ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr); 31 32 M = mat->rmap->n; 33 cstart = mat->cmap->rstart; 34 cend = mat->cmap->rend; 35 c->M = mat->rmap->N; /* set the global rows and columns and local rows */ 36 c->N = mat->cmap->N; 37 c->m = mat->rmap->n; 38 c->rstart = mat->rmap->rstart; 39 40 c->ncolors = nis; 41 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr); 42 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr); 43 ierr = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr); 44 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->rows);CHKERRQ(ierr); 45 ierr = PetscMalloc(nis*sizeof(PetscInt*),&c->columnsforrow);CHKERRQ(ierr); 46 ierr = PetscLogObjectMemory(c,5*nis*sizeof(PetscInt));CHKERRQ(ierr); 47 48 /* Allow access to data structures of local part of matrix */ 49 if (!aij->colmap) { 50 ierr = CreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr); 51 } 52 ierr = MatGetColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr); 53 ierr = MatGetColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr); 54 55 ierr = PetscMalloc((M+1)*sizeof(PetscInt),&rowhit);CHKERRQ(ierr); 56 ierr = PetscMalloc((M+1)*sizeof(PetscInt),&columnsforrow);CHKERRQ(ierr); 57 58 for (i=0; i<nis; i++) { 59 ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr); 60 ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr); 61 c->ncolumns[i] = n; 62 if (n) { 63 ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr); 64 ierr = PetscLogObjectMemory(c,n*sizeof(PetscInt));CHKERRQ(ierr); 65 ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr); 66 } else { 67 c->columns[i] = 0; 68 } 69 70 if (ctype == IS_COLORING_GLOBAL){ 71 /* Determine the total (parallel) number of columns of this color */ 72 ierr = MPI_Comm_size(((PetscObject)mat)->comm,&size);CHKERRQ(ierr); 73 ierr = PetscMalloc(2*size*sizeof(PetscInt*),&ncolsonproc);CHKERRQ(ierr); 74 disp = ncolsonproc + size; 75 76 nn = PetscMPIIntCast(n); 77 ierr = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,((PetscObject)mat)->comm);CHKERRQ(ierr); 78 nctot = 0; for (j=0; j<size; j++) {nctot += ncolsonproc[j];} 79 if (!nctot) { 80 ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr); 81 } 82 83 disp[0] = 0; 84 for (j=1; j<size; j++) { 85 disp[j] = disp[j-1] + ncolsonproc[j-1]; 86 } 87 88 /* Get complete list of columns for color on each processor */ 89 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 90 ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,((PetscObject)mat)->comm);CHKERRQ(ierr); 91 ierr = PetscFree(ncolsonproc);CHKERRQ(ierr); 92 } else if (ctype == IS_COLORING_GHOSTED){ 93 /* Determine local number of columns of this color on this process, including ghost points */ 94 nctot = n; 95 ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr); 96 ierr = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr); 97 } else { 98 SETERRQ(PETSC_ERR_SUP,"Not provided for this MatFDColoring type"); 99 } 100 101 /* 102 Mark all rows affect by these columns 103 */ 104 /* Temporary option to allow for debugging/testing */ 105 flg = PETSC_FALSE; 106 ierr = PetscOptionsGetTruth(PETSC_NULL,"-matfdcoloring_slow",&flg,PETSC_NULL);CHKERRQ(ierr); 107 if (!flg) {/*-----------------------------------------------------------------------------*/ 108 /* crude, fast version */ 109 ierr = PetscMemzero(rowhit,M*sizeof(PetscInt));CHKERRQ(ierr); 110 /* loop over columns*/ 111 for (j=0; j<nctot; j++) { 112 if (ctype == IS_COLORING_GHOSTED) { 113 col = ltog[cols[j]]; 114 } else { 115 col = cols[j]; 116 } 117 if (col >= cstart && col < cend) { 118 /* column is in diagonal block of matrix */ 119 rows = A_cj + A_ci[col-cstart]; 120 m = A_ci[col-cstart+1] - A_ci[col-cstart]; 121 } else { 122 #if defined (PETSC_USE_CTABLE) 123 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr) 124 colb --; 125 #else 126 colb = aij->colmap[col] - 1; 127 #endif 128 if (colb == -1) { 129 m = 0; 130 } else { 131 rows = B_cj + B_ci[colb]; 132 m = B_ci[colb+1] - B_ci[colb]; 133 } 134 } 135 /* loop over columns marking them in rowhit */ 136 for (k=0; k<m; k++) { 137 rowhit[*rows++] = col + 1; 138 } 139 } 140 141 /* count the number of hits */ 142 nrows = 0; 143 for (j=0; j<M; j++) { 144 if (rowhit[j]) nrows++; 145 } 146 c->nrows[i] = nrows; 147 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr); 148 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr); 149 ierr = PetscLogObjectMemory(c,2*(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr); 150 nrows = 0; 151 for (j=0; j<M; j++) { 152 if (rowhit[j]) { 153 c->rows[i][nrows] = j; 154 c->columnsforrow[i][nrows] = rowhit[j] - 1; 155 nrows++; 156 } 157 } 158 } else {/*-------------------------------------------------------------------------------*/ 159 /* slow version, using rowhit as a linked list */ 160 PetscInt currentcol,fm,mfm; 161 rowhit[M] = M; 162 nrows = 0; 163 /* loop over columns*/ 164 for (j=0; j<nctot; j++) { 165 if (ctype == IS_COLORING_GHOSTED) { 166 col = ltog[cols[j]]; 167 } else { 168 col = cols[j]; 169 } 170 if (col >= cstart && col < cend) { 171 /* column is in diagonal block of matrix */ 172 rows = A_cj + A_ci[col-cstart]; 173 m = A_ci[col-cstart+1] - A_ci[col-cstart]; 174 } else { 175 #if defined (PETSC_USE_CTABLE) 176 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 177 colb --; 178 #else 179 colb = aij->colmap[col] - 1; 180 #endif 181 if (colb == -1) { 182 m = 0; 183 } else { 184 rows = B_cj + B_ci[colb]; 185 m = B_ci[colb+1] - B_ci[colb]; 186 } 187 } 188 189 /* loop over columns marking them in rowhit */ 190 fm = M; /* fm points to first entry in linked list */ 191 for (k=0; k<m; k++) { 192 currentcol = *rows++; 193 /* is it already in the list? */ 194 do { 195 mfm = fm; 196 fm = rowhit[fm]; 197 } while (fm < currentcol); 198 /* not in list so add it */ 199 if (fm != currentcol) { 200 nrows++; 201 columnsforrow[currentcol] = col; 202 /* next three lines insert new entry into linked list */ 203 rowhit[mfm] = currentcol; 204 rowhit[currentcol] = fm; 205 fm = currentcol; 206 /* fm points to present position in list since we know the columns are sorted */ 207 } else { 208 SETERRQ(PETSC_ERR_PLIB,"Invalid coloring of matrix detected"); 209 } 210 } 211 } 212 c->nrows[i] = nrows; 213 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr); 214 ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr); 215 ierr = PetscLogObjectMemory(c,(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr); 216 /* now store the linked list of rows into c->rows[i] */ 217 nrows = 0; 218 fm = rowhit[M]; 219 do { 220 c->rows[i][nrows] = fm; 221 c->columnsforrow[i][nrows++] = columnsforrow[fm]; 222 fm = rowhit[fm]; 223 } while (fm < M); 224 } /* ---------------------------------------------------------------------------------------*/ 225 ierr = PetscFree(cols);CHKERRQ(ierr); 226 } 227 228 /* Optimize by adding the vscale, and scaleforrow[][] fields */ 229 /* 230 vscale will contain the "diagonal" on processor scalings followed by the off processor 231 */ 232 if (ctype == IS_COLORING_GLOBAL) { 233 ierr = VecCreateGhost(((PetscObject)mat)->comm,aij->A->rmap->n,PETSC_DETERMINE,aij->B->cmap->n,aij->garray,&c->vscale);CHKERRQ(ierr); 234 ierr = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr); 235 for (k=0; k<c->ncolors; k++) { 236 ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr); 237 for (l=0; l<c->nrows[k]; l++) { 238 col = c->columnsforrow[k][l]; 239 if (col >= cstart && col < cend) { 240 /* column is in diagonal block of matrix */ 241 colb = col - cstart; 242 } else { 243 /* column is in "off-processor" part */ 244 #if defined (PETSC_USE_CTABLE) 245 ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr); 246 colb --; 247 #else 248 colb = aij->colmap[col] - 1; 249 #endif 250 colb += cend - cstart; 251 } 252 c->vscaleforrow[k][l] = colb; 253 } 254 } 255 } else if (ctype == IS_COLORING_GHOSTED) { 256 /* Get gtol mapping */ 257 PetscInt N = mat->cmap->N, *gtol; 258 ierr = PetscMalloc((N+1)*sizeof(PetscInt),>ol);CHKERRQ(ierr); 259 for (i=0; i<N; i++) gtol[i] = -1; 260 for (i=0; i<map->n; i++) gtol[ltog[i]] = i; 261 262 c->vscale = 0; /* will be created in MatFDColoringApply() */ 263 ierr = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr); 264 for (k=0; k<c->ncolors; k++) { 265 ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr); 266 for (l=0; l<c->nrows[k]; l++) { 267 col = c->columnsforrow[k][l]; /* global column index */ 268 c->vscaleforrow[k][l] = gtol[col]; /* local column index */ 269 } 270 } 271 ierr = PetscFree(gtol);CHKERRQ(ierr); 272 } 273 ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr); 274 275 ierr = PetscFree(rowhit);CHKERRQ(ierr); 276 ierr = PetscFree(columnsforrow);CHKERRQ(ierr); 277 ierr = MatRestoreColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr); 278 ierr = MatRestoreColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr); 279 PetscFunctionReturn(0); 280 } 281 282 283 284 285 286 287