Actual source code: mpiaijviennacl.cxx

  1: #define PETSC_SKIP_IMMINTRIN_H_CUDAWORKAROUND 1

  3: #include <petscconf.h>
  4: #include <../src/mat/impls/aij/mpi/mpiaij.h>
  5: #include <../src/mat/impls/aij/seq/seqviennacl/viennaclmatimpl.h>

  7: static PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJViennaCL(Mat B, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[])
  8: {
  9:   Mat_MPIAIJ *b = (Mat_MPIAIJ *)B->data;

 11:   PetscFunctionBegin;
 12:   PetscCall(PetscLayoutSetUp(B->rmap));
 13:   PetscCall(PetscLayoutSetUp(B->cmap));
 14:   if (!B->preallocated) {
 15:     /* Explicitly create the two MATSEQAIJVIENNACL matrices. */
 16:     PetscCall(MatCreate(PETSC_COMM_SELF, &b->A));
 17:     PetscCall(MatSetSizes(b->A, B->rmap->n, B->cmap->n, B->rmap->n, B->cmap->n));
 18:     PetscCall(MatSetType(b->A, MATSEQAIJVIENNACL));
 19:     PetscCall(MatCreate(PETSC_COMM_SELF, &b->B));
 20:     PetscCall(MatSetSizes(b->B, B->rmap->n, B->cmap->N, B->rmap->n, B->cmap->N));
 21:     PetscCall(MatSetType(b->B, MATSEQAIJVIENNACL));
 22:   }
 23:   PetscCall(MatSeqAIJSetPreallocation(b->A, d_nz, d_nnz));
 24:   PetscCall(MatSeqAIJSetPreallocation(b->B, o_nz, o_nnz));
 25:   B->preallocated = PETSC_TRUE;
 26:   PetscFunctionReturn(PETSC_SUCCESS);
 27: }

 29: static PetscErrorCode MatAssemblyEnd_MPIAIJViennaCL(Mat A, MatAssemblyType mode)
 30: {
 31:   Mat_MPIAIJ *b = (Mat_MPIAIJ *)A->data;
 32:   PetscBool   v;

 34:   PetscFunctionBegin;
 35:   PetscCall(MatAssemblyEnd_MPIAIJ(A, mode));
 36:   PetscCall(PetscObjectTypeCompare((PetscObject)b->lvec, VECSEQVIENNACL, &v));
 37:   if (!v) {
 38:     PetscInt m;
 39:     PetscCall(VecGetSize(b->lvec, &m));
 40:     PetscCall(VecDestroy(&b->lvec));
 41:     PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, m, &b->lvec));
 42:   }
 43:   PetscFunctionReturn(PETSC_SUCCESS);
 44: }

 46: PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJViennaCL(Mat A)
 47: {
 48:   PetscFunctionBegin;
 49:   PetscCall(MatCreate_MPIAIJ(A));
 50:   A->boundtocpu = PETSC_FALSE;
 51:   PetscCall(PetscFree(A->defaultvectype));
 52:   PetscCall(PetscStrallocpy(VECVIENNACL, &A->defaultvectype));
 53:   PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatMPIAIJSetPreallocation_C", MatMPIAIJSetPreallocation_MPIAIJViennaCL));
 54:   A->ops->assemblyend = MatAssemblyEnd_MPIAIJViennaCL;
 55:   PetscCall(PetscObjectChangeTypeName((PetscObject)A, MATMPIAIJVIENNACL));
 56:   PetscFunctionReturn(PETSC_SUCCESS);
 57: }

 59: /*@C
 60:   MatCreateAIJViennaCL - Creates a sparse matrix in `MATAIJ` (compressed row) format
 61:   (the default parallel PETSc format).  This matrix will ultimately be pushed down
 62:   to GPUs and use the ViennaCL library for calculations.

 64:   Collective

 66:   Input Parameters:
 67: + comm  - MPI communicator, set to `PETSC_COMM_SELF`
 68: . m     - number of rows, or `PETSC_DECIDE` if `M` is provided
 69: . n     - number of columns, or `PETSC_DECIDE` if `N` is provided
 70: . M     - number of global rows in the matrix, or `PETSC_DETERMINE` if `m` is provided
 71: . N     - number of global columns in the matrix, or `PETSC_DETERMINE` if `n` is provided
 72: . d_nz  - number of nonzeros per row in DIAGONAL portion of local submatrix
 73:            (same value is used for all local rows)
 74: . d_nnz - array containing the number of nonzeros in the various rows of the
 75:            DIAGONAL portion of the local submatrix (possibly different for each row)
 76:            or `NULL`, if `d_nz` is used to specify the nonzero structure.
 77:            The size of this array is equal to the number of local rows, i.e `m`.
 78:            For matrices you plan to factor you must leave room for the diagonal entry and
 79:            put in the entry even if it is zero.
 80: . o_nz  - number of nonzeros per row in the OFF-DIAGONAL portion of local
 81:            submatrix (same value is used for all local rows).
 82: - o_nnz - array containing the number of nonzeros in the various rows of the
 83:            OFF-DIAGONAL portion of the local submatrix (possibly different for
 84:            each row) or `NULL`, if `o_nz` is used to specify the nonzero
 85:            structure. The size of this array is equal to the number
 86:            of local rows, i.e `m`.

 88:   Output Parameter:
 89: . A - the matrix

 91:   Level: intermediate

 93:   Notes:
 94:   It is recommended that one use the `MatCreate()`, `MatSetType()` and/or `MatSetFromOptions()`,
 95:   MatXXXXSetPreallocation() paradigm instead of this routine directly.
 96:   [MatXXXXSetPreallocation() is, for example, `MatSeqAIJSetPreallocation()`]

 98:   The AIJ format, also called
 99:   compressed row storage), is fully compatible with standard Fortran
100:   storage.  That is, the stored row and column indices can begin at
101:   either one (as in Fortran) or zero.

103: .seealso: `Mat`, `MatCreate()`, `MatCreateAIJ()`, `MatCreateAIJCUSPARSE()`, `MatSetValues()`, `MatSeqAIJSetColumnIndices()`, `MatCreateSeqAIJWithArrays()`,
104:           `MATMPIAIJVIENNACL`, `MATAIJVIENNACL`
105: @*/
106: PetscErrorCode MatCreateAIJViennaCL(MPI_Comm comm, PetscInt m, PetscInt n, PetscInt M, PetscInt N, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[], Mat *A)
107: {
108:   PetscMPIInt size;

110:   PetscFunctionBegin;
111:   PetscCall(MatCreate(comm, A));
112:   PetscCall(MatSetSizes(*A, m, n, M, N));
113:   PetscCallMPI(MPI_Comm_size(comm, &size));
114:   if (size > 1) {
115:     PetscCall(MatSetType(*A, MATMPIAIJVIENNACL));
116:     PetscCall(MatMPIAIJSetPreallocation(*A, d_nz, d_nnz, o_nz, o_nnz));
117:   } else {
118:     PetscCall(MatSetType(*A, MATSEQAIJVIENNACL));
119:     PetscCall(MatSeqAIJSetPreallocation(*A, d_nz, d_nnz));
120:   }
121:   PetscFunctionReturn(PETSC_SUCCESS);
122: }

124: /*MC
125:    MATAIJVIENNACL - MATMPIAIJVIENNACL= "aijviennacl" = "mpiaijviennacl" - A matrix type to be used for sparse matrices.

127:    A matrix type (CSR format) whose data resides on GPUs.
128:    All matrix calculations are performed using the ViennaCL library.

130:    This matrix type is identical to `MATSEQAIJVIENNACL` when constructed with a single process communicator,
131:    and `MATMPIAIJVIENNACL` otherwise.  As a result, for single process communicators,
132:    `MatSeqAIJSetPreallocation()` is supported, and similarly `MatMPIAIJSetPreallocation()` is supported
133:    for communicators controlling multiple processes.  It is recommended that you call both of
134:    the above preallocation routines for simplicity.

136:    Options Database Keys:
137: .  -mat_type mpiaijviennacl - sets the matrix type to `MATAIJVIENNACL` during a call to `MatSetFromOptions()`

139:   Level: beginner

141: .seealso: `Mat`, `MatType`, `MatCreateAIJViennaCL()`, `MATSEQAIJVIENNACL`, `MatCreateSeqAIJVIENNACL()`
142: M*/