xref: /petsc/src/mat/impls/aij/mpi/aijperm/mpiaijperm.c (revision d083f849a86f1f43e18d534ee43954e2786cb29a)
1938d9b04SRichard Tran Mills 
2938d9b04SRichard Tran Mills #include <../src/mat/impls/aij/mpi/mpiaij.h>
3938d9b04SRichard Tran Mills /*@C
4938d9b04SRichard Tran Mills    MatCreateMPIAIJPERM - Creates a sparse parallel matrix whose local
5938d9b04SRichard Tran Mills    portions are stored as SEQAIJPERM matrices (a matrix class that inherits
6938d9b04SRichard Tran Mills    from SEQAIJ but includes some optimizations to allow more effective
7938d9b04SRichard Tran Mills    vectorization).  The same guidelines that apply to MPIAIJ matrices for
8938d9b04SRichard Tran Mills    preallocating the matrix storage apply here as well.
9938d9b04SRichard Tran Mills 
10*d083f849SBarry Smith       Collective
11938d9b04SRichard Tran Mills 
12938d9b04SRichard Tran Mills    Input Parameters:
13938d9b04SRichard Tran Mills +  comm - MPI communicator
14938d9b04SRichard Tran Mills .  m - number of local rows (or PETSC_DECIDE to have calculated if M is given)
15938d9b04SRichard Tran Mills            This value should be the same as the local size used in creating the
16938d9b04SRichard Tran Mills            y vector for the matrix-vector product y = Ax.
17938d9b04SRichard Tran Mills .  n - This value should be the same as the local size used in creating the
18938d9b04SRichard Tran Mills        x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have
19938d9b04SRichard Tran Mills        calculated if N is given) For square matrices n is almost always m.
20938d9b04SRichard Tran Mills .  M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
21938d9b04SRichard Tran Mills .  N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
22938d9b04SRichard Tran Mills .  d_nz  - number of nonzeros per row in DIAGONAL portion of local submatrix
23938d9b04SRichard Tran Mills            (same value is used for all local rows)
24938d9b04SRichard Tran Mills .  d_nnz - array containing the number of nonzeros in the various rows of the
25938d9b04SRichard Tran Mills            DIAGONAL portion of the local submatrix (possibly different for each row)
26938d9b04SRichard Tran Mills            or NULL, if d_nz is used to specify the nonzero structure.
27938d9b04SRichard Tran Mills            The size of this array is equal to the number of local rows, i.e 'm'.
28938d9b04SRichard Tran Mills            For matrices you plan to factor you must leave room for the diagonal entry and
29938d9b04SRichard Tran Mills            put in the entry even if it is zero.
30938d9b04SRichard Tran Mills .  o_nz  - number of nonzeros per row in the OFF-DIAGONAL portion of local
31938d9b04SRichard Tran Mills            submatrix (same value is used for all local rows).
32938d9b04SRichard Tran Mills -  o_nnz - array containing the number of nonzeros in the various rows of the
33938d9b04SRichard Tran Mills            OFF-DIAGONAL portion of the local submatrix (possibly different for
34938d9b04SRichard Tran Mills            each row) or NULL, if o_nz is used to specify the nonzero
35938d9b04SRichard Tran Mills            structure. The size of this array is equal to the number
36938d9b04SRichard Tran Mills            of local rows, i.e 'm'.
37938d9b04SRichard Tran Mills 
38938d9b04SRichard Tran Mills    Output Parameter:
39938d9b04SRichard Tran Mills .  A - the matrix
40938d9b04SRichard Tran Mills 
41938d9b04SRichard Tran Mills    Notes:
42938d9b04SRichard Tran Mills    If the *_nnz parameter is given then the *_nz parameter is ignored
43938d9b04SRichard Tran Mills 
44938d9b04SRichard Tran Mills    m,n,M,N parameters specify the size of the matrix, and its partitioning across
45938d9b04SRichard Tran Mills    processors, while d_nz,d_nnz,o_nz,o_nnz parameters specify the approximate
46938d9b04SRichard Tran Mills    storage requirements for this matrix.
47938d9b04SRichard Tran Mills 
48938d9b04SRichard Tran Mills    If PETSC_DECIDE or  PETSC_DETERMINE is used for a particular argument on one
49938d9b04SRichard Tran Mills    processor than it must be used on all processors that share the object for
50938d9b04SRichard Tran Mills    that argument.
51938d9b04SRichard Tran Mills 
52938d9b04SRichard Tran Mills    The user MUST specify either the local or global matrix dimensions
53938d9b04SRichard Tran Mills    (possibly both).
54938d9b04SRichard Tran Mills 
55938d9b04SRichard Tran Mills    The parallel matrix is partitioned such that the first m0 rows belong to
56938d9b04SRichard Tran Mills    process 0, the next m1 rows belong to process 1, the next m2 rows belong
57938d9b04SRichard Tran Mills    to process 2 etc.. where m0,m1,m2... are the input parameter 'm'.
58938d9b04SRichard Tran Mills 
59938d9b04SRichard Tran Mills    The DIAGONAL portion of the local submatrix of a processor can be defined
60938d9b04SRichard Tran Mills    as the submatrix which is obtained by extraction the part corresponding
61938d9b04SRichard Tran Mills    to the rows r1-r2 and columns r1-r2 of the global matrix, where r1 is the
62938d9b04SRichard Tran Mills    first row that belongs to the processor, and r2 is the last row belonging
63938d9b04SRichard Tran Mills    to the this processor. This is a square mxm matrix. The remaining portion
64938d9b04SRichard Tran Mills    of the local submatrix (mxN) constitute the OFF-DIAGONAL portion.
65938d9b04SRichard Tran Mills 
66938d9b04SRichard Tran Mills    If o_nnz, d_nnz are specified, then o_nz, and d_nz are ignored.
67938d9b04SRichard Tran Mills 
68938d9b04SRichard Tran Mills    When calling this routine with a single process communicator, a matrix of
69938d9b04SRichard Tran Mills    type SEQAIJPERM is returned.  If a matrix of type MPIAIJPERM is desired
70938d9b04SRichard Tran Mills    for this type of communicator, use the construction mechanism:
71938d9b04SRichard Tran Mills      MatCreate(...,&A); MatSetType(A,MPIAIJ); MatMPIAIJSetPreallocation(A,...);
72938d9b04SRichard Tran Mills 
73938d9b04SRichard Tran Mills    By default, this format uses inodes (identical nodes) when possible.
74938d9b04SRichard Tran Mills    We search for consecutive rows with the same nonzero structure, thereby
75938d9b04SRichard Tran Mills    reusing matrix information to achieve increased efficiency.
76938d9b04SRichard Tran Mills 
77938d9b04SRichard Tran Mills    Options Database Keys:
78938d9b04SRichard Tran Mills +  -mat_no_inode  - Do not use inodes
79938d9b04SRichard Tran Mills -  -mat_inode_limit <limit> - Sets inode limit (max limit=5)
80938d9b04SRichard Tran Mills 
81938d9b04SRichard Tran Mills    Level: intermediate
82938d9b04SRichard Tran Mills 
83938d9b04SRichard Tran Mills .seealso: MatCreate(), MatCreateSeqAIJPERM(), MatSetValues()
84938d9b04SRichard Tran Mills @*/
85938d9b04SRichard Tran Mills 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)
86938d9b04SRichard Tran Mills {
87938d9b04SRichard Tran Mills   PetscErrorCode ierr;
88938d9b04SRichard Tran Mills   PetscMPIInt    size;
89938d9b04SRichard Tran Mills 
90938d9b04SRichard Tran Mills   PetscFunctionBegin;
91938d9b04SRichard Tran Mills   ierr = MatCreate(comm,A);CHKERRQ(ierr);
92938d9b04SRichard Tran Mills   ierr = MatSetSizes(*A,m,n,M,N);CHKERRQ(ierr);
93938d9b04SRichard Tran Mills   ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
94938d9b04SRichard Tran Mills   if (size > 1) {
95938d9b04SRichard Tran Mills     ierr = MatSetType(*A,MATMPIAIJPERM);CHKERRQ(ierr);
96938d9b04SRichard Tran Mills     ierr = MatMPIAIJSetPreallocation(*A,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr);
97938d9b04SRichard Tran Mills   } else {
98938d9b04SRichard Tran Mills     ierr = MatSetType(*A,MATSEQAIJPERM);CHKERRQ(ierr);
99938d9b04SRichard Tran Mills     ierr = MatSeqAIJSetPreallocation(*A,d_nz,d_nnz);CHKERRQ(ierr);
100938d9b04SRichard Tran Mills   }
101938d9b04SRichard Tran Mills   PetscFunctionReturn(0);
102938d9b04SRichard Tran Mills }
103938d9b04SRichard Tran Mills 
104938d9b04SRichard Tran Mills PetscErrorCode  MatMPIAIJSetPreallocation_MPIAIJPERM(Mat B,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[])
105938d9b04SRichard Tran Mills {
106938d9b04SRichard Tran Mills   Mat_MPIAIJ     *b = (Mat_MPIAIJ*)B->data;
107938d9b04SRichard Tran Mills   PetscErrorCode ierr;
108938d9b04SRichard Tran Mills 
109938d9b04SRichard Tran Mills   PetscFunctionBegin;
110938d9b04SRichard Tran Mills   ierr = MatMPIAIJSetPreallocation_MPIAIJ(B,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr);
111938d9b04SRichard Tran Mills   ierr = MatConvert_SeqAIJ_SeqAIJPERM(b->A, MATSEQAIJPERM, MAT_INPLACE_MATRIX, &b->A);CHKERRQ(ierr);
112938d9b04SRichard Tran Mills   ierr = MatConvert_SeqAIJ_SeqAIJPERM(b->B, MATSEQAIJPERM, MAT_INPLACE_MATRIX, &b->B);CHKERRQ(ierr);
113938d9b04SRichard Tran Mills   PetscFunctionReturn(0);
114938d9b04SRichard Tran Mills }
115938d9b04SRichard Tran Mills 
116938d9b04SRichard Tran Mills PETSC_INTERN PetscErrorCode MatConvert_MPIAIJ_MPIAIJPERM(Mat A,MatType type,MatReuse reuse,Mat *newmat)
117938d9b04SRichard Tran Mills {
118938d9b04SRichard Tran Mills   PetscErrorCode ierr;
119938d9b04SRichard Tran Mills   Mat            B = *newmat;
120938d9b04SRichard Tran Mills 
121938d9b04SRichard Tran Mills   PetscFunctionBegin;
122938d9b04SRichard Tran Mills   if (reuse == MAT_INITIAL_MATRIX) {
123938d9b04SRichard Tran Mills     ierr = MatDuplicate(A,MAT_COPY_VALUES,&B);CHKERRQ(ierr);
124938d9b04SRichard Tran Mills   }
125938d9b04SRichard Tran Mills 
126938d9b04SRichard Tran Mills   ierr = PetscObjectChangeTypeName((PetscObject) B, MATMPIAIJPERM);CHKERRQ(ierr);
127938d9b04SRichard Tran Mills   ierr = PetscObjectComposeFunction((PetscObject)B,"MatMPIAIJSetPreallocation_C",MatMPIAIJSetPreallocation_MPIAIJPERM);CHKERRQ(ierr);
128938d9b04SRichard Tran Mills   *newmat = B;
129938d9b04SRichard Tran Mills   PetscFunctionReturn(0);
130938d9b04SRichard Tran Mills }
131938d9b04SRichard Tran Mills 
132938d9b04SRichard Tran Mills PETSC_EXTERN PetscErrorCode MatCreate_MPIAIJPERM(Mat A)
133938d9b04SRichard Tran Mills {
134938d9b04SRichard Tran Mills   PetscErrorCode ierr;
135938d9b04SRichard Tran Mills 
136938d9b04SRichard Tran Mills   PetscFunctionBegin;
137938d9b04SRichard Tran Mills   ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr);
138938d9b04SRichard Tran Mills   ierr = MatConvert_MPIAIJ_MPIAIJPERM(A,MATMPIAIJPERM,MAT_INPLACE_MATRIX,&A);CHKERRQ(ierr);
139938d9b04SRichard Tran Mills   PetscFunctionReturn(0);
140938d9b04SRichard Tran Mills }
141938d9b04SRichard Tran Mills 
142938d9b04SRichard Tran Mills /*MC
143938d9b04SRichard Tran Mills    MATAIJPERM - MATAIJPERM = "AIJPERM" - A matrix type to be used for sparse matrices.
144938d9b04SRichard Tran Mills 
145938d9b04SRichard Tran Mills    This matrix type is identical to MATSEQAIJPERM when constructed with a single process communicator,
146938d9b04SRichard Tran Mills    and MATMPIAIJPERM otherwise.  As a result, for single process communicators,
147938d9b04SRichard Tran Mills   MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported
148938d9b04SRichard Tran Mills   for communicators controlling multiple processes.  It is recommended that you call both of
149938d9b04SRichard Tran Mills   the above preallocation routines for simplicity.
150938d9b04SRichard Tran Mills 
151938d9b04SRichard Tran Mills    Options Database Keys:
152938d9b04SRichard Tran Mills . -mat_type aijperm - sets the matrix type to "AIJPERM" during a call to MatSetFromOptions()
153938d9b04SRichard Tran Mills 
154938d9b04SRichard Tran Mills   Level: beginner
155938d9b04SRichard Tran Mills 
156938d9b04SRichard Tran Mills .seealso: MatCreateMPIAIJPERM(), MATSEQAIJPERM, MATMPIAIJPERM
157938d9b04SRichard Tran Mills M*/
158938d9b04SRichard Tran Mills 
159