1 2 /* 3 Defines a matrix-vector product for the MATMPIAIJCRL matrix class. 4 This class is derived from the MATMPIAIJ class and retains the 5 compressed row storage (aka Yale sparse matrix format) but augments 6 it with a column oriented storage that is more efficient for 7 matrix vector products on Vector machines. 8 9 CRL stands for constant row length (that is the same number of columns 10 is kept (padded with zeros) for each row of the sparse matrix. 11 12 See src/mat/impls/aij/seq/crl/crl.c for the sequential version 13 */ 14 15 #include <../src/mat/impls/aij/mpi/mpiaij.h> 16 #include <../src/mat/impls/aij/seq/crl/crl.h> 17 18 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 19 20 PetscErrorCode MatDestroy_MPIAIJCRL(Mat A) 21 { 22 PetscErrorCode ierr; 23 Mat_AIJCRL *aijcrl = (Mat_AIJCRL*) A->spptr; 24 25 /* Free everything in the Mat_AIJCRL data structure. */ 26 if (aijcrl) { 27 ierr = PetscFree2(aijcrl->acols,aijcrl->icols);CHKERRQ(ierr); 28 ierr = VecDestroy(&aijcrl->fwork);CHKERRQ(ierr); 29 ierr = VecDestroy(&aijcrl->xwork);CHKERRQ(ierr); 30 ierr = PetscFree(aijcrl->array);CHKERRQ(ierr); 31 } 32 ierr = PetscFree(A->spptr);CHKERRQ(ierr); 33 34 ierr = PetscObjectChangeTypeName((PetscObject)A, MATMPIAIJ);CHKERRQ(ierr); 35 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 36 PetscFunctionReturn(0); 37 } 38 39 PetscErrorCode MatMPIAIJCRL_create_aijcrl(Mat A) 40 { 41 Mat_MPIAIJ *a = (Mat_MPIAIJ*)(A)->data; 42 Mat_SeqAIJ *Aij = (Mat_SeqAIJ*)(a->A->data), *Bij = (Mat_SeqAIJ*)(a->B->data); 43 Mat_AIJCRL *aijcrl = (Mat_AIJCRL*) A->spptr; 44 PetscInt m = A->rmap->n; /* Number of rows in the matrix. */ 45 PetscInt nd = a->A->cmap->n; /* number of columns in diagonal portion */ 46 PetscInt *aj = Aij->j,*bj = Bij->j; /* From the CSR representation; points to the beginning of each row. */ 47 PetscInt i, j,rmax = 0,*icols, *ailen = Aij->ilen, *bilen = Bij->ilen; 48 PetscScalar *aa = Aij->a,*ba = Bij->a,*acols,*array; 49 PetscErrorCode ierr; 50 51 PetscFunctionBegin; 52 /* determine the row with the most columns */ 53 for (i=0; i<m; i++) { 54 rmax = PetscMax(rmax,ailen[i]+bilen[i]); 55 } 56 aijcrl->nz = Aij->nz+Bij->nz; 57 aijcrl->m = A->rmap->n; 58 aijcrl->rmax = rmax; 59 60 ierr = PetscFree2(aijcrl->acols,aijcrl->icols);CHKERRQ(ierr); 61 ierr = PetscMalloc2(rmax*m,&aijcrl->acols,rmax*m,&aijcrl->icols);CHKERRQ(ierr); 62 acols = aijcrl->acols; 63 icols = aijcrl->icols; 64 for (i=0; i<m; i++) { 65 for (j=0; j<ailen[i]; j++) { 66 acols[j*m+i] = *aa++; 67 icols[j*m+i] = *aj++; 68 } 69 for (; j<ailen[i]+bilen[i]; j++) { 70 acols[j*m+i] = *ba++; 71 icols[j*m+i] = nd + *bj++; 72 } 73 for (; j<rmax; j++) { /* empty column entries */ 74 acols[j*m+i] = 0.0; 75 icols[j*m+i] = (j) ? icols[(j-1)*m+i] : 0; /* handle case where row is EMPTY */ 76 } 77 } 78 ierr = PetscInfo1(A,"Percentage of 0's introduced for vectorized multiply %g\n",1.0-((double)(aijcrl->nz))/((double)(rmax*m)));CHKERRQ(ierr); 79 80 ierr = PetscFree(aijcrl->array);CHKERRQ(ierr); 81 ierr = PetscMalloc1(a->B->cmap->n+nd,&array);CHKERRQ(ierr); 82 /* xwork array is actually B->n+nd long, but we define xwork this length so can copy into it */ 83 ierr = VecDestroy(&aijcrl->xwork);CHKERRQ(ierr); 84 ierr = VecCreateMPIWithArray(PetscObjectComm((PetscObject)A),1,nd,PETSC_DECIDE,array,&aijcrl->xwork);CHKERRQ(ierr); 85 ierr = VecDestroy(&aijcrl->fwork);CHKERRQ(ierr); 86 ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,a->B->cmap->n,array+nd,&aijcrl->fwork);CHKERRQ(ierr); 87 88 aijcrl->array = array; 89 aijcrl->xscat = a->Mvctx; 90 PetscFunctionReturn(0); 91 } 92 93 extern PetscErrorCode MatAssemblyEnd_MPIAIJ(Mat,MatAssemblyType); 94 95 PetscErrorCode MatAssemblyEnd_MPIAIJCRL(Mat A, MatAssemblyType mode) 96 { 97 PetscErrorCode ierr; 98 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 99 Mat_SeqAIJ *Aij = (Mat_SeqAIJ*)(a->A->data), *Bij = (Mat_SeqAIJ*)(a->A->data); 100 101 PetscFunctionBegin; 102 Aij->inode.use = PETSC_FALSE; 103 Bij->inode.use = PETSC_FALSE; 104 105 ierr = MatAssemblyEnd_MPIAIJ(A,mode);CHKERRQ(ierr); 106 if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(0); 107 108 /* Now calculate the permutation and grouping information. */ 109 ierr = MatMPIAIJCRL_create_aijcrl(A);CHKERRQ(ierr); 110 PetscFunctionReturn(0); 111 } 112 113 extern PetscErrorCode MatMult_AIJCRL(Mat,Vec,Vec); 114 extern PetscErrorCode MatDuplicate_AIJCRL(Mat,MatDuplicateOption,Mat*); 115 116 /* MatConvert_MPIAIJ_MPIAIJCRL converts a MPIAIJ matrix into a 117 * MPIAIJCRL matrix. This routine is called by the MatCreate_MPIAIJCRL() 118 * routine, but can also be used to convert an assembled MPIAIJ matrix 119 * into a MPIAIJCRL one. */ 120 121 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJCRL(Mat A,MatType type,MatReuse reuse,Mat *newmat) 122 { 123 PetscErrorCode ierr; 124 Mat B = *newmat; 125 Mat_AIJCRL *aijcrl; 126 127 PetscFunctionBegin; 128 if (reuse == MAT_INITIAL_MATRIX) { 129 ierr = MatDuplicate(A,MAT_COPY_VALUES,&B);CHKERRQ(ierr); 130 } 131 132 ierr = PetscNewLog(B,&aijcrl);CHKERRQ(ierr); 133 B->spptr = (void*) aijcrl; 134 135 /* Set function pointers for methods that we inherit from AIJ but override. */ 136 B->ops->duplicate = MatDuplicate_AIJCRL; 137 B->ops->assemblyend = MatAssemblyEnd_MPIAIJCRL; 138 B->ops->destroy = MatDestroy_MPIAIJCRL; 139 B->ops->mult = MatMult_AIJCRL; 140 141 /* If A has already been assembled, compute the permutation. */ 142 if (A->assembled) { 143 ierr = MatMPIAIJCRL_create_aijcrl(B);CHKERRQ(ierr); 144 } 145 ierr = PetscObjectChangeTypeName((PetscObject)B,MATMPIAIJCRL);CHKERRQ(ierr); 146 *newmat = B; 147 PetscFunctionReturn(0); 148 } 149 150 /*@C 151 MatCreateMPIAIJCRL - Creates a sparse matrix of type MPIAIJCRL. 152 This type inherits from AIJ, but stores some additional 153 information that is used to allow better vectorization of 154 the matrix-vector product. At the cost of increased storage, the AIJ formatted 155 matrix can be copied to a format in which pieces of the matrix are 156 stored in ELLPACK format, allowing the vectorized matrix multiply 157 routine to use stride-1 memory accesses. As with the AIJ type, it is 158 important to preallocate matrix storage in order to get good assembly 159 performance. 160 161 Collective on MPI_Comm 162 163 Input Parameters: 164 + comm - MPI communicator, set to PETSC_COMM_SELF 165 . m - number of rows 166 . n - number of columns 167 . nz - number of nonzeros per row (same for all rows) 168 - nnz - array containing the number of nonzeros in the various rows 169 (possibly different for each row) or NULL 170 171 Output Parameter: 172 . A - the matrix 173 174 Notes: 175 If nnz is given then nz is ignored 176 177 Level: intermediate 178 179 .keywords: matrix, cray, sparse, parallel 180 181 .seealso: MatCreate(), MatCreateMPIAIJPERM(), MatSetValues() 182 @*/ 183 PetscErrorCode MatCreateMPIAIJCRL(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],PetscInt onz,const PetscInt onnz[],Mat *A) 184 { 185 PetscErrorCode ierr; 186 187 PetscFunctionBegin; 188 ierr = MatCreate(comm,A);CHKERRQ(ierr); 189 ierr = MatSetSizes(*A,m,n,m,n);CHKERRQ(ierr); 190 ierr = MatSetType(*A,MATMPIAIJCRL);CHKERRQ(ierr); 191 ierr = MatMPIAIJSetPreallocation_MPIAIJ(*A,nz,(PetscInt*)nnz,onz,(PetscInt*)onnz);CHKERRQ(ierr); 192 PetscFunctionReturn(0); 193 } 194 195 PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJCRL(Mat A) 196 { 197 PetscErrorCode ierr; 198 199 PetscFunctionBegin; 200 ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr); 201 ierr = MatConvert_MPIAIJ_MPIAIJCRL(A,MATMPIAIJCRL,MAT_INPLACE_MATRIX,&A);CHKERRQ(ierr); 202 PetscFunctionReturn(0); 203 } 204 205