Actual source code: mpiaijperm.c


  2: #include <../src/mat/impls/aij/mpi/mpiaij.h>
  3: /*@C
  4:    MatCreateMPIAIJPERM - Creates a sparse parallel matrix whose local
  5:    portions are stored as `MATSEQAIJPERM` matrices (a matrix class that inherits
  6:    from SEQAIJ but includes some optimizations to allow more effective
  7:    vectorization).

  9:       Collective

 11:    Input Parameters:
 12: +  comm - MPI communicator
 13: .  m - number of local rows (or `PETSC_DECIDE` to have calculated if `M` is given)
 14:            This value should be the same as the local size used in creating the
 15:            y vector for the matrix-vector product y = Ax.
 16: .  n - This value should be the same as the local size used in creating the
 17:        x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have
 18:        calculated if `N` is given) For square matrices `n` is almost always `m`.
 19: .  M - number of global rows (or `PETSC_DETERMINE` to have calculated if `m` is given)
 20: .  N - number of global columns (or `PETSC_DETERMINE` to have calculated if `n` is given)
 21: .  d_nz  - number of nonzeros per row in DIAGONAL portion of local submatrix
 22:            (same value is used for all local rows)
 23: .  d_nnz - array containing the number of nonzeros in the various rows of the
 24:            DIAGONAL portion of the local submatrix (possibly different for each row)
 25:            or `NULL`, if `d_nz` is used to specify the nonzero structure.
 26:            The size of this array is equal to the number of local rows, i.e `m`.
 27:            For matrices you plan to factor you must leave room for the diagonal entry and
 28:            put in the entry even if it is zero.
 29: .  o_nz  - number of nonzeros per row in the OFF-DIAGONAL portion of local
 30:            submatrix (same value is used for all local rows).
 31: -  o_nnz - array containing the number of nonzeros in the various rows of the
 32:            OFF-DIAGONAL portion of the local submatrix (possibly different for
 33:            each row) or `NULL`, if `o_nz` is used to specify the nonzero
 34:            structure. The size of this array is equal to the number
 35:            of local rows, i.e `m`.

 37:    Output Parameter:
 38: .  A - the matrix

 40:    Options Database Keys:
 41: +  -mat_no_inode  - Do not use inodes
 42: -  -mat_inode_limit <limit> - Sets inode limit (max limit=5)

 44:    Level: intermediate

 46:    Notes:
 47:    If the *_nnz parameter is given then the *_nz parameter is ignored

 49:    `m`,`n`,`M`,`N` parameters specify the size of the matrix, and its partitioning across
 50:    processors, while `d_nz`,`d_nnz`,`o_nz`,`o_nnz` parameters specify the approximate
 51:    storage requirements for this matrix.

 53:    If `PETSC_DECIDE` or  `PETSC_DETERMINE` is used for a particular argument on one
 54:    processor than it must be used on all processors that share the object for
 55:    that argument.

 57:    The user MUST specify either the local or global matrix dimensions
 58:    (possibly both).

 60:    The parallel matrix is partitioned such that the first m0 rows belong to
 61:    process 0, the next m1 rows belong to process 1, the next m2 rows belong
 62:    to process 2 etc.. where m0,m1,m2... are the input parameter `m`.

 64:    The DIAGONAL portion of the local submatrix of a processor can be defined
 65:    as the submatrix which is obtained by extraction the part corresponding
 66:    to the rows r1-r2 and columns r1-r2 of the global matrix, where r1 is the
 67:    first row that belongs to the processor, and r2 is the last row belonging
 68:    to the this processor. This is a square mxm matrix. The remaining portion
 69:    of the local submatrix (mxN) constitute the OFF-DIAGONAL portion.

 71:    If `o_nnz`, `d_nnz` are specified, then `o_nz`, and `d_nz` are ignored.

 73:    When calling this routine with a single process communicator, a matrix of
 74:    type `MATSEQAIJPERM` is returned.  If a matrix of type `MATMPIAIJPERM` is desired
 75:    for this type of communicator, use the construction mechanism
 76: .vb
 77:    MatCreate(...,&A);
 78:    MatSetType(A,MPIAIJ);
 79:    MatMPIAIJSetPreallocation(A,...);
 80: .ve

 82:    By default, this format uses inodes (identical nodes) when possible.
 83:    We search for consecutive rows with the same nonzero structure, thereby
 84:    reusing matrix information to achieve increased efficiency.

 86: .seealso: [](ch_matrices), `Mat`, [Sparse Matrix Creation](sec_matsparse), `MATMPIAIJPERM`, `MatCreate()`, `MatCreateSeqAIJPERM()`, `MatSetValues()`
 87: @*/
 88: PetscErrorCode MatCreateMPIAIJPERM(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)
 89: {
 90:   PetscMPIInt size;

 92:   PetscFunctionBegin;
 93:   PetscCall(MatCreate(comm, A));
 94:   PetscCall(MatSetSizes(*A, m, n, M, N));
 95:   PetscCallMPI(MPI_Comm_size(comm, &size));
 96:   if (size > 1) {
 97:     PetscCall(MatSetType(*A, MATMPIAIJPERM));
 98:     PetscCall(MatMPIAIJSetPreallocation(*A, d_nz, d_nnz, o_nz, o_nnz));
 99:   } else {
100:     PetscCall(MatSetType(*A, MATSEQAIJPERM));
101:     PetscCall(MatSeqAIJSetPreallocation(*A, d_nz, d_nnz));
102:   }
103:   PetscFunctionReturn(PETSC_SUCCESS);
104: }

106: PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJPERM(Mat B, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[])
107: {
108:   Mat_MPIAIJ *b = (Mat_MPIAIJ *)B->data;

110:   PetscFunctionBegin;
111:   PetscCall(MatMPIAIJSetPreallocation_MPIAIJ(B, d_nz, d_nnz, o_nz, o_nnz));
112:   PetscCall(MatConvert_SeqAIJ_SeqAIJPERM(b->A, MATSEQAIJPERM, MAT_INPLACE_MATRIX, &b->A));
113:   PetscCall(MatConvert_SeqAIJ_SeqAIJPERM(b->B, MATSEQAIJPERM, MAT_INPLACE_MATRIX, &b->B));
114:   PetscFunctionReturn(PETSC_SUCCESS);
115: }

117: PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJPERM(Mat A, MatType type, MatReuse reuse, Mat *newmat)
118: {
119:   Mat B = *newmat;

121:   PetscFunctionBegin;
122:   if (reuse == MAT_INITIAL_MATRIX) PetscCall(MatDuplicate(A, MAT_COPY_VALUES, &B));

124:   PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATMPIAIJPERM));
125:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatMPIAIJSetPreallocation_C", MatMPIAIJSetPreallocation_MPIAIJPERM));
126:   *newmat = B;
127:   PetscFunctionReturn(PETSC_SUCCESS);
128: }

130: PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJPERM(Mat A)
131: {
132:   PetscFunctionBegin;
133:   PetscCall(MatSetType(A, MATMPIAIJ));
134:   PetscCall(MatConvert_MPIAIJ_MPIAIJPERM(A, MATMPIAIJPERM, MAT_INPLACE_MATRIX, &A));
135:   PetscFunctionReturn(PETSC_SUCCESS);
136: }

138: /*MC
139:    MATAIJPERM - "AIJPERM" - A matrix type to be used for sparse matrices.

141:    This matrix type is identical to `MATSEQAIJPERM` when constructed with a single process communicator,
142:    and `MATMPIAIJPERM` otherwise.  As a result, for single process communicators,
143:   `MatSeqAIJSetPreallocation()` is supported, and similarly `MatMPIAIJSetPreallocation()` is supported
144:   for communicators controlling multiple processes.  It is recommended that you call both of
145:   the above preallocation routines for simplicity.

147:    Options Database Key:
148: . -mat_type aijperm - sets the matrix type to `MATAIJPERM`

150:   Level: beginner

152: .seealso: [](ch_matrices), `Mat`, `MatCreateMPIAIJPERM()`, `MATSEQAIJPERM`, `MATMPIAIJPERM`, `MATSEQAIJ`, `MATMPIAIJ`, `MATSEQAIJMKL`, `MATMPIAIJMKL`, `MATSEQAIJSELL`, `MATMPIAIJSELL`
153: M*/