Actual source code: mpibaijmkl.c

  1: #include <../src/mat/impls/baij/mpi/mpibaij.h>

  3: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat, MatType, MatReuse, Mat *);

  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;

  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: }

 16: static PetscErrorCode MatConvert_MPIBAIJ_MPIBAIJMKL(Mat A, MatType type, MatReuse reuse, Mat *newmat)
 17: {
 18:   Mat B = *newmat;

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

 23:   PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATMPIBAIJMKL));
 24:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatMPIBAIJSetPreallocation_C", MatMPIBAIJSetPreallocation_MPIBAIJMKL));
 25:   *newmat = B;
 26:   PetscFunctionReturn(PETSC_SUCCESS);
 27: }

 29: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
 30: /*@C
 31:   MatCreateBAIJMKL - Creates a sparse parallel matrix in `MATBAIJMKL` format (block compressed row).

 33:   Collective

 35:   Input Parameters:
 36: + comm  - MPI communicator
 37: . bs    - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row
 38:           blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()`
 39: . m     - number of local rows (or `PETSC_DECIDE` to have calculated if `M` is given)
 40:            This value should be the same as the local size used in creating the
 41:            y vector for the matrix-vector product y = Ax.
 42: . n     - number of local columns (or `PETSC_DECIDE` to have calculated if `N` is given)
 43:            This value should be the same as the local size used in creating the
 44:            x vector for the matrix-vector product y = Ax.
 45: . M     - number of global rows (or `PETSC_DETERMINE` to have calculated if `m` is given)
 46: . N     - number of global columns (or `PETSC_DETERMINE` to have calculated if `n` is given)
 47: . d_nz  - number of nonzero blocks per block row in diagonal portion of local
 48:            submatrix  (same for all local rows)
 49: . d_nnz - array containing the number of nonzero blocks in the various block rows
 50:            of the in diagonal portion of the local (possibly different for each block
 51:            row) or `NULL`.  If you plan to factor the matrix you must leave room for the diagonal entry
 52:            and set it even if it is zero.
 53: . o_nz  - number of nonzero blocks per block row in the off-diagonal portion of local
 54:            submatrix (same for all local rows).
 55: - o_nnz - array containing the number of nonzero blocks in the various block rows of the
 56:            off-diagonal portion of the local submatrix (possibly different for
 57:            each block row) or `NULL`.

 59:   Output Parameter:
 60: . A - the matrix

 62:   Options Database Keys:
 63: + -mat_block_size            - size of the blocks to use
 64: - -mat_use_hash_table <fact> - set hash table factor

 66:   Level: intermediate

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

 73:   This type inherits from `MATBAIJ` and is largely identical, but uses sparse BLAS
 74:   routines from Intel MKL whenever possible.
 75:   `MatMult()`, `MatMultAdd()`, `MatMultTranspose()`, and `MatMultTransposeAdd()`
 76:   operations are currently supported.
 77:   If the installed version of MKL supports the "SpMV2" sparse
 78:   inspector-executor routines, then those are used by default.
 79:   Default PETSc kernels are used otherwise.
 80:   For good matrix assembly performance the user should preallocate the matrix
 81:   storage by setting the parameters `d_nz` (or `d_nnz`) and `o_nz` (or `o_nnz`).
 82:   By setting these parameters accurately, performance can be increased by more
 83:   than a factor of 50.

 85:   If the *_nnz parameter is given then the *_nz parameter is ignored

 87:   A nonzero block is any block that as 1 or more nonzeros in it

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

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

 95:   Storage Information:
 96:   For a square global matrix we define each processor's diagonal portion
 97:   to be its local rows and the corresponding columns (a square submatrix);
 98:   each processor's off-diagonal portion encompasses the remainder of the
 99:   local matrix (a rectangular submatrix).

101:   The user can specify preallocated storage for the diagonal part of
102:   the local submatrix with either `d_nz` or `d_nnz` (not both).  Set
103:   `d_nz` = `PETSC_DEFAULT` and `d_nnz` = `NULL` for PETSc to control dynamic
104:   memory allocation.  Likewise, specify preallocated storage for the
105:   off-diagonal part of the local submatrix with `o_nz` or `o_nnz` (not both).

107:   Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
108:   the figure below we depict these three local rows and all columns (0-11).

110: .vb
111:            0 1 2 3 4 5 6 7 8 9 10 11
112:           --------------------------
113:    row 3  |o o o d d d o o o o  o  o
114:    row 4  |o o o d d d o o o o  o  o
115:    row 5  |o o o d d d o o o o  o  o
116:           --------------------------
117: .ve

119:   Thus, any entries in the d locations are stored in the d (diagonal)
120:   submatrix, and any entries in the o locations are stored in the
121:   o (off-diagonal) submatrix.  Note that the d and the o submatrices are
122:   stored simply in the `MATSEQBAIJMKL` format for compressed row storage.

124:   Now `d_nz` should indicate the number of block nonzeros per row in the d matrix,
125:   and `o_nz` should indicate the number of block nonzeros per row in the o matrix.
126:   In general, for PDE problems in which most nonzeros are near the diagonal,
127:   one expects `d_nz` >> `o_nz`.

129: .seealso: [](ch_matrices), `Mat`, `MATBAIJMKL`, `MATBAIJ`, `MatCreate()`, `MatCreateSeqBAIJMKL()`, `MatSetValues()`, `MatMPIBAIJSetPreallocation()`, `MatMPIBAIJSetPreallocationCSR()`
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;

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: }

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: }

157: /*MC
158:    MATBAIJMKL - MATBAIJMKL = "BAIJMKL" - A matrix type to be used for sparse matrices.

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.

166:    Options Database Key:
167: . -mat_type baijmkl - sets the matrix type to `MATBAIJMKL` during a call to `MatSetFromOptions()`

169:   Level: beginner

171: .seealso: [](ch_matrices), `Mat`, `MatCreateBAIJMKL()`, `MATSEQBAIJMKL`, `MATMPIBAIJMKL`
172: M*/