1 2 #include <petsc/private/matimpl.h> /*I "petscmat.h" I*/ 3 4 typedef struct { 5 Mat A; /* sparse matrix */ 6 Mat U,V; /* dense tall-skinny matrices */ 7 Vec c; /* sequential vector containing the diagonal of C */ 8 Vec work1,work2; /* sequential vectors that hold partial products */ 9 PetscMPIInt nwork; /* length of work vectors */ 10 Vec xl,yl; /* auxiliary sequential vectors for matmult operation */ 11 } Mat_LRC; 12 13 14 #undef __FUNCT__ 15 #define __FUNCT__ "MatMult_LRC" 16 PetscErrorCode MatMult_LRC(Mat N,Vec x,Vec y) 17 { 18 Mat_LRC *Na = (Mat_LRC*)N->data; 19 PetscErrorCode ierr; 20 PetscScalar *w1,*w2; 21 const PetscScalar *a; 22 23 PetscFunctionBegin; 24 ierr = VecGetArrayRead(x,&a);CHKERRQ(ierr); 25 ierr = VecPlaceArray(Na->xl,a);CHKERRQ(ierr); 26 ierr = VecGetLocalVector(y,Na->yl);CHKERRQ(ierr); 27 28 /* multiply the local part of V with the local part of x */ 29 #if defined(PETSC_USE_COMPLEX) 30 ierr = MatMultHermitianTranspose(Na->V,Na->xl,Na->work1);CHKERRQ(ierr); 31 #else 32 ierr = MatMultTranspose(Na->V,Na->xl,Na->work1);CHKERRQ(ierr); 33 #endif 34 35 /* form the sum of all the local multiplies: this is work2 = V'*x = 36 sum_{all processors} work1 */ 37 ierr = VecGetArray(Na->work1,&w1);CHKERRQ(ierr); 38 ierr = VecGetArray(Na->work2,&w2);CHKERRQ(ierr); 39 ierr = MPIU_Allreduce(w1,w2,Na->nwork,MPIU_SCALAR,MPIU_SUM,PetscObjectComm((PetscObject)N));CHKERRQ(ierr); 40 ierr = VecRestoreArray(Na->work1,&w1);CHKERRQ(ierr); 41 ierr = VecRestoreArray(Na->work2,&w2);CHKERRQ(ierr); 42 43 if (Na->c) { /* work2 = C*work2 */ 44 ierr = VecPointwiseMult(Na->work2,Na->c,Na->work2);CHKERRQ(ierr); 45 } 46 47 if (Na->A) { 48 /* form y = A*x */ 49 ierr = MatMult(Na->A,x,y);CHKERRQ(ierr); 50 /* multiply-add y = y + U*work2 */ 51 ierr = MatMultAdd(Na->U,Na->work2,Na->yl,Na->yl);CHKERRQ(ierr); 52 } else { 53 /* multiply y = U*work2 */ 54 ierr = MatMult(Na->U,Na->work2,Na->yl);CHKERRQ(ierr); 55 } 56 57 ierr = VecRestoreArrayRead(x,&a);CHKERRQ(ierr); 58 ierr = VecResetArray(Na->xl);CHKERRQ(ierr); 59 ierr = VecRestoreLocalVector(y,Na->yl);CHKERRQ(ierr); 60 PetscFunctionReturn(0); 61 } 62 63 #undef __FUNCT__ 64 #define __FUNCT__ "MatDestroy_LRC" 65 PetscErrorCode MatDestroy_LRC(Mat N) 66 { 67 Mat_LRC *Na = (Mat_LRC*)N->data; 68 PetscErrorCode ierr; 69 70 PetscFunctionBegin; 71 ierr = MatDestroy(&Na->A);CHKERRQ(ierr); 72 ierr = MatDestroy(&Na->U);CHKERRQ(ierr); 73 ierr = MatDestroy(&Na->V);CHKERRQ(ierr); 74 ierr = VecDestroy(&Na->c);CHKERRQ(ierr); 75 ierr = VecDestroy(&Na->work1);CHKERRQ(ierr); 76 ierr = VecDestroy(&Na->work2);CHKERRQ(ierr); 77 ierr = VecDestroy(&Na->xl);CHKERRQ(ierr); 78 ierr = VecDestroy(&Na->yl);CHKERRQ(ierr); 79 ierr = PetscFree(N->data);CHKERRQ(ierr); 80 PetscFunctionReturn(0); 81 } 82 83 #undef __FUNCT__ 84 #define __FUNCT__ "MatCreateLRC" 85 /*@ 86 MatCreateLRC - Creates a new matrix object that behaves like A + U*C*V' 87 88 Collective on Mat 89 90 Input Parameters: 91 + A - the (sparse) matrix (can be NULL) 92 . U, V - two dense rectangular (tall and skinny) matrices 93 - c - a sequential vector containing the diagonal of C (can be NULL) 94 95 Output Parameter: 96 . N - the matrix that represents A + U*C*V' 97 98 Notes: 99 The matrix A + U*C*V' is not formed! Rather the new matrix 100 object performs the matrix-vector product by first multiplying by 101 A and then adding the other term. 102 103 C is a diagonal matrix (represented as a vector) of order k, 104 where k is the number of columns of both U and V. 105 106 If A is NULL then the new object behaves like a low-rank matrix U*C*V'. 107 108 Use V=U (or V=NULL) for a symmetric low-rank correction, A + U*C*U'. 109 110 If c is NULL then the low-rank correction is just U*V'. 111 112 Level: intermediate 113 @*/ 114 PetscErrorCode MatCreateLRC(Mat A,Mat U,Vec c,Mat V,Mat *N) 115 { 116 PetscErrorCode ierr; 117 PetscBool match; 118 PetscInt m,n,k,m1,n1,k1; 119 Mat_LRC *Na; 120 121 PetscFunctionBegin; 122 if (A) PetscValidHeaderSpecific(A,MAT_CLASSID,1); 123 PetscValidHeaderSpecific(U,MAT_CLASSID,2); 124 if (c) PetscValidHeaderSpecific(c,VEC_CLASSID,3); 125 if (V) PetscValidHeaderSpecific(V,MAT_CLASSID,4); 126 else V=U; 127 if (A) PetscCheckSameComm(A,1,U,2); 128 PetscCheckSameComm(U,2,V,4); 129 130 ierr = PetscObjectTypeCompareAny((PetscObject)U,&match,MATSEQDENSE,MATMPIDENSE,"");CHKERRQ(ierr); 131 if (!match) SETERRQ(PetscObjectComm((PetscObject)U),PETSC_ERR_SUP,"Matrix U must be of type dense"); 132 if (V) { 133 ierr = PetscObjectTypeCompareAny((PetscObject)V,&match,MATSEQDENSE,MATMPIDENSE,"");CHKERRQ(ierr); 134 if (!match) SETERRQ(PetscObjectComm((PetscObject)V),PETSC_ERR_SUP,"Matrix V must be of type dense"); 135 } 136 137 ierr = MatGetSize(U,NULL,&k);CHKERRQ(ierr); 138 ierr = MatGetSize(V,NULL,&k1);CHKERRQ(ierr); 139 if (k!=k1) SETERRQ(PetscObjectComm((PetscObject)U),PETSC_ERR_ARG_INCOMP,"U and V have different number of columns"); 140 ierr = MatGetLocalSize(U,&m,NULL);CHKERRQ(ierr); 141 ierr = MatGetLocalSize(V,&n,NULL);CHKERRQ(ierr); 142 if (A) { 143 ierr = MatGetLocalSize(A,&m1,&n1);CHKERRQ(ierr); 144 if (m!=m1) SETERRQ(PetscObjectComm((PetscObject)U),PETSC_ERR_ARG_INCOMP,"Local dimensions of U and A do not match"); 145 if (n!=n1) SETERRQ(PetscObjectComm((PetscObject)V),PETSC_ERR_ARG_INCOMP,"Local dimensions of V and A do not match"); 146 } 147 if (c) { 148 ierr = VecGetSize(c,&k1);CHKERRQ(ierr); 149 if (k!=k1) SETERRQ(PetscObjectComm((PetscObject)c),PETSC_ERR_ARG_INCOMP,"The length of c does not match the number of columns of U and V"); 150 ierr = VecGetLocalSize(c,&k1);CHKERRQ(ierr); 151 if (k!=k1) SETERRQ(PetscObjectComm((PetscObject)c),PETSC_ERR_ARG_INCOMP,"c must be a sequential vector"); 152 } 153 154 ierr = MatCreate(PetscObjectComm((PetscObject)U),N);CHKERRQ(ierr); 155 ierr = MatSetSizes(*N,m,n,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr); 156 ierr = PetscObjectChangeTypeName((PetscObject)*N,MATLRC);CHKERRQ(ierr); 157 158 ierr = PetscNewLog(*N,&Na);CHKERRQ(ierr); 159 (*N)->data = (void*)Na; 160 Na->A = A; 161 Na->c = c; 162 163 ierr = MatDenseGetLocalMatrix(U,&Na->U);CHKERRQ(ierr); 164 ierr = MatDenseGetLocalMatrix(V,&Na->V);CHKERRQ(ierr); 165 if (A) { ierr = PetscObjectReference((PetscObject)A);CHKERRQ(ierr); } 166 ierr = PetscObjectReference((PetscObject)Na->U);CHKERRQ(ierr); 167 ierr = PetscObjectReference((PetscObject)Na->V);CHKERRQ(ierr); 168 if (c) { ierr = PetscObjectReference((PetscObject)c);CHKERRQ(ierr); } 169 170 ierr = VecCreateSeq(PETSC_COMM_SELF,U->cmap->N,&Na->work1);CHKERRQ(ierr); 171 ierr = VecDuplicate(Na->work1,&Na->work2);CHKERRQ(ierr); 172 ierr = PetscMPIIntCast(U->cmap->N,&Na->nwork);CHKERRQ(ierr); 173 174 ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,V->rmap->n,NULL,&Na->xl);CHKERRQ(ierr); 175 ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,U->rmap->n,NULL,&Na->yl);CHKERRQ(ierr); 176 177 (*N)->ops->destroy = MatDestroy_LRC; 178 (*N)->ops->mult = MatMult_LRC; 179 (*N)->assembled = PETSC_TRUE; 180 (*N)->preallocated = PETSC_TRUE; 181 (*N)->cmap->N = V->rmap->N; 182 (*N)->rmap->N = U->rmap->N; 183 (*N)->cmap->n = V->rmap->n; 184 (*N)->rmap->n = U->rmap->n; 185 PetscFunctionReturn(0); 186 } 187 188