xref: /petsc/src/mat/impls/baij/mpi/baijmkl/mpibaijmkl.c (revision bcee047adeeb73090d7e36cc71e39fc287cdbb97)
1 #include <../src/mat/impls/baij/mpi/mpibaij.h>
2 
3 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat, MatType, MatReuse, Mat *);
4 
5 static PetscErrorCode MatMPIBAIJSetPreallocation_MPIBAIJMKL(Mat B, PetscInt bs, PetscInt d_nz, const PetscInt *d_nnz, PetscInt o_nz, const PetscInt *o_nnz)
6 {
7   Mat_MPIBAIJ *b = (Mat_MPIBAIJ *)B->data;
8 
9   PetscFunctionBegin;
10   PetscCall(MatMPIBAIJSetPreallocation_MPIBAIJ(B, bs, d_nz, d_nnz, o_nz, o_nnz));
11   PetscCall(MatConvert_SeqBAIJ_SeqBAIJMKL(b->A, MATSEQBAIJMKL, MAT_INPLACE_MATRIX, &b->A));
12   PetscCall(MatConvert_SeqBAIJ_SeqBAIJMKL(b->B, MATSEQBAIJMKL, MAT_INPLACE_MATRIX, &b->B));
13   PetscFunctionReturn(PETSC_SUCCESS);
14 }
15 
16 static PetscErrorCode MatConvert_MPIBAIJ_MPIBAIJMKL(Mat A, MatType type, MatReuse reuse, Mat *newmat)
17 {
18   Mat B = *newmat;
19 
20   PetscFunctionBegin;
21   if (reuse == MAT_INITIAL_MATRIX) PetscCall(MatDuplicate(A, MAT_COPY_VALUES, &B));
22 
23   PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATMPIBAIJMKL));
24   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatMPIBAIJSetPreallocation_C", MatMPIBAIJSetPreallocation_MPIBAIJMKL));
25   *newmat = B;
26   PetscFunctionReturn(PETSC_SUCCESS);
27 }
28 
29 /*@C
30    MatCreateBAIJMKL - Creates a sparse parallel matrix in `MATBAIJMKL` format (block compressed row).
31 
32    Collective
33 
34    Input Parameters:
35 +  comm - MPI communicator
36 .  bs   - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row
37           blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()`
38 .  m - number of local rows (or `PETSC_DECIDE` to have calculated if `M` is given)
39            This value should be the same as the local size used in creating the
40            y vector for the matrix-vector product y = Ax.
41 .  n - number of local columns (or `PETSC_DECIDE` to have calculated if `N` is given)
42            This value should be the same as the local size used in creating the
43            x vector for the matrix-vector product y = Ax.
44 .  M - number of global rows (or `PETSC_DETERMINE` to have calculated if `m` is given)
45 .  N - number of global columns (or `PETSC_DETERMINE` to have calculated if `n` is given)
46 .  d_nz  - number of nonzero blocks per block row in diagonal portion of local
47            submatrix  (same for all local rows)
48 .  d_nnz - array containing the number of nonzero blocks in the various block rows
49            of the in diagonal portion of the local (possibly different for each block
50            row) or `NULL`.  If you plan to factor the matrix you must leave room for the diagonal entry
51            and set it even if it is zero.
52 .  o_nz  - number of nonzero blocks per block row in the off-diagonal portion of local
53            submatrix (same for all local rows).
54 -  o_nnz - array containing the number of nonzero blocks in the various block rows of the
55            off-diagonal portion of the local submatrix (possibly different for
56            each block row) or `NULL`.
57 
58    Output Parameter:
59 .  A - the matrix
60 
61    Options Database Keys:
62 +   -mat_block_size - size of the blocks to use
63 -   -mat_use_hash_table <fact> - set hash table factor
64 
65      Level: intermediate
66 
67    Notes:
68    It is recommended that one use the `MatCreate()`, `MatSetType()` and/or `MatSetFromOptions()`,
69    MatXXXXSetPreallocation() paradigm instead of this routine directly.
70    [MatXXXXSetPreallocation() is, for example, `MatSeqAIJSetPreallocation()`]
71 
72    This type inherits from `MATBAIJ` and is largely identical, but uses sparse BLAS
73    routines from Intel MKL whenever possible.
74    `MatMult()`, `MatMultAdd()`, `MatMultTranspose()`, and `MatMultTransposeAdd()`
75    operations are currently supported.
76    If the installed version of MKL supports the "SpMV2" sparse
77    inspector-executor routines, then those are used by default.
78    Default PETSc kernels are used otherwise.
79    For good matrix assembly performance the user should preallocate the matrix
80    storage by setting the parameters `d_nz` (or `d_nnz`) and `o_nz` (or `o_nnz`).
81    By setting these parameters accurately, performance can be increased by more
82    than a factor of 50.
83 
84    If the *_nnz parameter is given then the *_nz parameter is ignored
85 
86    A nonzero block is any block that as 1 or more nonzeros in it
87 
88    The user MUST specify either the local or global matrix dimensions
89    (possibly both).
90 
91    If `PETSC_DECIDE` or  `PETSC_DETERMINE` is used for a particular argument on one processor
92    than it must be used on all processors that share the object for that argument.
93 
94    Storage Information:
95    For a square global matrix we define each processor's diagonal portion
96    to be its local rows and the corresponding columns (a square submatrix);
97    each processor's off-diagonal portion encompasses the remainder of the
98    local matrix (a rectangular submatrix).
99 
100    The user can specify preallocated storage for the diagonal part of
101    the local submatrix with either `d_nz` or `d_nnz` (not both).  Set
102    `d_nz` = `PETSC_DEFAULT` and `d_nnz` = `NULL` for PETSc to control dynamic
103    memory allocation.  Likewise, specify preallocated storage for the
104    off-diagonal part of the local submatrix with `o_nz` or `o_nnz` (not both).
105 
106    Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
107    the figure below we depict these three local rows and all columns (0-11).
108 
109 .vb
110            0 1 2 3 4 5 6 7 8 9 10 11
111           --------------------------
112    row 3  |o o o d d d o o o o  o  o
113    row 4  |o o o d d d o o o o  o  o
114    row 5  |o o o d d d o o o o  o  o
115           --------------------------
116 .ve
117 
118    Thus, any entries in the d locations are stored in the d (diagonal)
119    submatrix, and any entries in the o locations are stored in the
120    o (off-diagonal) submatrix.  Note that the d and the o submatrices are
121    stored simply in the `MATSEQBAIJMKL` format for compressed row storage.
122 
123    Now `d_nz` should indicate the number of block nonzeros per row in the d matrix,
124    and `o_nz` should indicate the number of block nonzeros per row in the o matrix.
125    In general, for PDE problems in which most nonzeros are near the diagonal,
126    one expects `d_nz` >> `o_nz`.
127 
128 .seealso: [](ch_matrices), `Mat`, `MATBAIJMKL`, `MATBAIJ`, `MatCreate()`, `MatCreateSeqBAIJMKL()`, `MatSetValues()`, `MatCreateBAIJMKL()`, `MatMPIBAIJSetPreallocation()`, `MatMPIBAIJSetPreallocationCSR()`
129 @*/
130 
131 PetscErrorCode MatCreateBAIJMKL(MPI_Comm comm, PetscInt bs, PetscInt m, PetscInt n, PetscInt M, PetscInt N, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[], Mat *A)
132 {
133   PetscMPIInt size;
134 
135   PetscFunctionBegin;
136   PetscCall(MatCreate(comm, A));
137   PetscCall(MatSetSizes(*A, m, n, M, N));
138   PetscCallMPI(MPI_Comm_size(comm, &size));
139   if (size > 1) {
140     PetscCall(MatSetType(*A, MATMPIBAIJMKL));
141     PetscCall(MatMPIBAIJSetPreallocation(*A, bs, d_nz, d_nnz, o_nz, o_nnz));
142   } else {
143     PetscCall(MatSetType(*A, MATSEQBAIJMKL));
144     PetscCall(MatSeqBAIJSetPreallocation(*A, bs, d_nz, d_nnz));
145   }
146   PetscFunctionReturn(PETSC_SUCCESS);
147 }
148 
149 PETSC_EXTERN PetscErrorCode MatCreate_MPIBAIJMKL(Mat A)
150 {
151   PetscFunctionBegin;
152   PetscCall(MatSetType(A, MATMPIBAIJ));
153   PetscCall(MatConvert_MPIBAIJ_MPIBAIJMKL(A, MATMPIBAIJMKL, MAT_INPLACE_MATRIX, &A));
154   PetscFunctionReturn(PETSC_SUCCESS);
155 }
156 
157 /*MC
158    MATBAIJMKL - MATBAIJMKL = "BAIJMKL" - A matrix type to be used for sparse matrices.
159 
160    This matrix type is identical to `MATSEQBAIJMKL` when constructed with a single process communicator,
161    and `MATMPIBAIJMKL` otherwise.  As a result, for single process communicators,
162   `MatSeqBAIJSetPreallocation()` is supported, and similarly `MatMPIBAIJSetPreallocation()` is supported
163   for communicators controlling multiple processes.  It is recommended that you call both of
164   the above preallocation routines for simplicity.
165 
166    Options Database Key:
167 . -mat_type baijmkl - sets the matrix type to `MATBAIJMKL` during a call to `MatSetFromOptions()`
168 
169   Level: beginner
170 
171 .seealso: [](ch_matrices), `Mat`, `MatCreateBAIJMKL()`, `MATSEQBAIJMKL`, `MATMPIBAIJMKL`
172 M*/
173