xref: /petsc/src/mat/impls/aij/mpi/aijmkl/mpiaijmkl.c (revision bcee047adeeb73090d7e36cc71e39fc287cdbb97)
1 #include <../src/mat/impls/aij/mpi/mpiaij.h>
2 /*@C
3    MatCreateMPIAIJMKL - Creates a sparse parallel matrix whose local
4    portions are stored as `MATSEQAIJMKL` matrices (a matrix class that inherits
5    from `MATSEQAIJ` but uses some operations provided by Intel MKL).
6 
7       Collective
8 
9    Input Parameters:
10 +  comm - MPI communicator
11 .  m - number of local rows (or `PETSC_DECIDE` to have calculated if `M` is given)
12            This value should be the same as the local size used in creating the
13            y vector for the matrix-vector product y = Ax.
14 .  n - This value should be the same as the local size used in creating the
15        x vector for the matrix-vector product y = Ax. (or `PETSC_DECIDE` to have
16        calculated if N is given) For square matrices n is almost always `m`.
17 .  M - number of global rows (or `PETSC_DETERMINE` to have calculated if `m` is given)
18 .  N - number of global columns (or `PETSC_DETERMINE` to have calculated if `n` is given)
19 .  d_nz  - number of nonzeros per row in DIAGONAL portion of local submatrix
20            (same value is used for all local rows)
21 .  d_nnz - array containing the number of nonzeros in the various rows of the
22            DIAGONAL portion of the local submatrix (possibly different for each row)
23            or `NULL`, if `d_nz` is used to specify the nonzero structure.
24            The size of this array is equal to the number of local rows, i.e `m`.
25            For matrices you plan to factor you must leave room for the diagonal entry and
26            put in the entry even if it is zero.
27 .  o_nz  - number of nonzeros per row in the OFF-DIAGONAL portion of local
28            submatrix (same value is used for all local rows).
29 -  o_nnz - array containing the number of nonzeros in the various rows of the
30            OFF-DIAGONAL portion of the local submatrix (possibly different for
31            each row) or `NULL`, if `o_nz` is used to specify the nonzero
32            structure. The size of this array is equal to the number
33            of local rows, i.e `m`.
34 
35    Output Parameter:
36 .  A - the matrix
37 
38    Options Database Key:
39 .  -mat_aijmkl_no_spmv2 - disables use of the SpMV2 inspector-executor routines
40 
41    Level: intermediate
42 
43    Notes:
44    If the *_nnz parameter is given then the *_nz parameter is ignored
45 
46    `m`,`n`,`M`,`N` parameters specify the size of the matrix, and its partitioning across
47    processors, while `d_nz`,`d_nnz`,`o_nz`,`o_nnz` parameters specify the approximate
48    storage requirements for this matrix.
49 
50    If `PETSC_DECIDE` or `PETSC_DETERMINE` is used for a particular argument on one
51    processor than it must be used on all processors that share the object for
52    that argument.
53 
54    The user MUST specify either the local or global matrix dimensions
55    (possibly both).
56 
57    The parallel matrix is partitioned such that the first m0 rows belong to
58    process 0, the next m1 rows belong to process 1, the next m2 rows belong
59    to process 2 etc.. where m0,m1,m2... are the input parameter `m`.
60 
61    The DIAGONAL portion of the local submatrix of a processor can be defined
62    as the submatrix which is obtained by extraction the part corresponding
63    to the rows r1-r2 and columns r1-r2 of the global matrix, where r1 is the
64    first row that belongs to the processor, and r2 is the last row belonging
65    to the this processor. This is a square mxm matrix. The remaining portion
66    of the local submatrix (mxN) constitute the OFF-DIAGONAL portion.
67 
68    If `o_nnz`, `d_nnz` are specified, then `o_nz`, and `d_nz` are ignored.
69 
70    When calling this routine with a single process communicator, a matrix of
71    type `MATSEQAIJMKL` is returned.  If a matrix of type `MATMPIAIJMKL` is desired
72    for this type of communicator, use the construction mechanism
73 .vb
74   MatCreate(...,&A);
75   MatSetType(A,MPIAIJMKL);
76   MatMPIAIJSetPreallocation(A,...);
77 .ve
78 
79 .seealso: [](ch_matrices), `Mat`, [Sparse Matrix Creation](sec_matsparse), `MATMPIAIJMKL`, `MatCreate()`, `MatCreateSeqAIJMKL()`, `MatSetValues()`
80 @*/
81 PetscErrorCode MatCreateMPIAIJMKL(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)
82 {
83   PetscMPIInt size;
84 
85   PetscFunctionBegin;
86   PetscCall(MatCreate(comm, A));
87   PetscCall(MatSetSizes(*A, m, n, M, N));
88   PetscCallMPI(MPI_Comm_size(comm, &size));
89   if (size > 1) {
90     PetscCall(MatSetType(*A, MATMPIAIJMKL));
91     PetscCall(MatMPIAIJSetPreallocation(*A, d_nz, d_nnz, o_nz, o_nnz));
92   } else {
93     PetscCall(MatSetType(*A, MATSEQAIJMKL));
94     PetscCall(MatSeqAIJSetPreallocation(*A, d_nz, d_nnz));
95   }
96   PetscFunctionReturn(PETSC_SUCCESS);
97 }
98 
99 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJMKL(Mat, MatType, MatReuse, Mat *);
100 
101 PetscErrorCode MatMPIAIJSetPreallocation_MPIAIJMKL(Mat B, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[])
102 {
103   Mat_MPIAIJ *b = (Mat_MPIAIJ *)B->data;
104 
105   PetscFunctionBegin;
106   PetscCall(MatMPIAIJSetPreallocation_MPIAIJ(B, d_nz, d_nnz, o_nz, o_nnz));
107   PetscCall(MatConvert_SeqAIJ_SeqAIJMKL(b->A, MATSEQAIJMKL, MAT_INPLACE_MATRIX, &b->A));
108   PetscCall(MatConvert_SeqAIJ_SeqAIJMKL(b->B, MATSEQAIJMKL, MAT_INPLACE_MATRIX, &b->B));
109   PetscFunctionReturn(PETSC_SUCCESS);
110 }
111 
112 PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJMKL(Mat A, MatType type, MatReuse reuse, Mat *newmat)
113 {
114   Mat B = *newmat;
115 
116   PetscFunctionBegin;
117   if (reuse == MAT_INITIAL_MATRIX) PetscCall(MatDuplicate(A, MAT_COPY_VALUES, &B));
118   PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATMPIAIJMKL));
119   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatMPIAIJSetPreallocation_C", MatMPIAIJSetPreallocation_MPIAIJMKL));
120   *newmat = B;
121   PetscFunctionReturn(PETSC_SUCCESS);
122 }
123 
124 PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJMKL(Mat A)
125 {
126   PetscFunctionBegin;
127   PetscCall(MatSetType(A, MATMPIAIJ));
128   PetscCall(MatConvert_MPIAIJ_MPIAIJMKL(A, MATMPIAIJMKL, MAT_INPLACE_MATRIX, &A));
129   PetscFunctionReturn(PETSC_SUCCESS);
130 }
131 
132 /*MC
133    MATAIJMKL - MATAIJMKL = "AIJMKL" - A matrix type to be used for sparse matrices.
134 
135    This matrix type is identical to `MATSEQAIJMKL` when constructed with a single process communicator,
136    and `MATMPIAIJMKL` otherwise.  As a result, for single process communicators,
137   MatSeqAIJSetPreallocation() is supported, and similarly `MatMPIAIJSetPreallocation()` is supported
138   for communicators controlling multiple processes.  It is recommended that you call both of
139   the above preallocation routines for simplicity.
140 
141    Options Database Key:
142 . -mat_type aijmkl - sets the matrix type to `MATAIJMKL` during a call to `MatSetFromOptions()`
143 
144   Level: beginner
145 
146 .seealso: [](ch_matrices), `Mat`, `MATMPIAIJMKL`, `MATSEQAIJMKL`, `MatCreateMPIAIJMKL()`, `MATSEQAIJMKL`, `MATMPIAIJMKL`, `MATSEQAIJSELL`, `MATMPIAIJSELL`, `MATSEQAIJPERM`, `MATMPIAIJPERM`
147 M*/
148