1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
|
\newpage
%===============================================================================
\subsection{GraphBLAS import/export: using copy semantics} %====================
%===============================================================================
\label{GrB_import_export}
The v2.0 C API includes import/export methods for matrices (not vectors) using
a different strategy as compared to the \verb'GxB_Container' methods. The
\verb'GxB_Container' methods are based on {\em move semantics}, in which
ownership of arrays is passed between SuiteSparse:GraphBLAS and the user
application. This allows the \verb'GxB_Container' methods to work in $O(1)$
time, and require no additional memory, but it requires that GraphBLAS and the
user application agree on which memory manager to use. This is done via
\verb'GxB_init'. This allows GraphBLAS to \verb'malloc' an array that can be
later \verb'free'd by the user application, and visa versa.
The \verb'GrB' import/export methods take a different approach. The data
is always copied in and out between the opaque GraphBLAS matrix and the
user arrays. This takes $\Omega(e)$ time, if the matrix has $e$ entries,
and requires more memory. It has the advantage that it does not require
GraphBLAS and the user application to agree on what memory manager to use,
since no ownership of allocated arrays is changed.
The format for \verb'GrB_Matrix_import' and \verb'GrB_Matrix_export' is
controlled by the following enum:
{\footnotesize
\begin{verbatim}
typedef enum
{
GrB_CSR_FORMAT = 0, // CSR format (equiv to GxB_SPARSE with GrB_ROWMAJOR)
GrB_CSC_FORMAT = 1, // CSC format (equiv to GxB_SPARSE with GrB_COLMAJOR)
GrB_COO_FORMAT = 2 // triplet format (like input to GrB*build)
}
GrB_Format ; \end{verbatim}}
\newpage
%-------------------------------------------------------------------------------
\subsubsection{{\sf GrB\_Matrix\_import:} import a matrix}
%-------------------------------------------------------------------------------
\label{GrB_matrix_import}
\begin{mdframed}[userdefinedwidth=6in]
{\footnotesize
\begin{verbatim}
GrB_Info GrB_Matrix_import // import a matrix
(
GrB_Matrix *A, // handle of matrix to create
GrB_Type type, // type of matrix to create
GrB_Index nrows, // number of rows of the matrix
GrB_Index ncols, // number of columns of the matrix
const GrB_Index *Ap, // pointers for CSR, CSC, column indices for COO
const GrB_Index *Ai, // row indices for CSR, CSC
const <type> *Ax, // values
GrB_Index Ap_len, // number of entries in Ap (not # of bytes)
GrB_Index Ai_len, // number of entries in Ai (not # of bytes)
GrB_Index Ax_len, // number of entries in Ax (not # of bytes)
int format // import format (GrB_Format)
) ;
\end{verbatim}
} \end{mdframed}
The \verb'GrB_Matrix_import' method copies from user-provided arrays into an
opaque \verb'GrB_Matrix' and \verb'GrB_Matrix_export' copies data out, from an
opaque \verb'GrB_Matrix' into user-provided arrays.
The suffix \verb'TYPE' in the prototype above is one of \verb'BOOL',
\verb'INT8', \verb'INT16', etc, for built-n types, or \verb'UDT' for
user-defined types. The type of the \verb'Ax' array must match this type. No
typecasting is performed.
Unlike the \verb'GxB_Container' methods, memory is not handed off between the
user application and GraphBLAS. The three arrays \verb'Ap', \verb'Ai'. and
\verb'Ax' are not modified, and are still owned by the user application when
the method finishes.
\verb'GrB_Matrix_import' does not support the creation of matrices with 32-bit
integer indices.
The matrix can be imported in one of three different formats:
\begin{packed_itemize}
\item \verb'GrB_CSR_FORMAT': % CSR format (equiv to GxB_SPARSE with GrB_ROWMAJOR)
Compressed-row format. \verb'Ap' is an array of size \verb'nrows+1'.
The arrays \verb'Ai' and \verb'Ax' are of size \verb'nvals = Ap [nrows]',
and \verb'Ap[0]' must be zero.
The column indices of entries in the \verb'i'th row appear in
\verb'Ai[Ap[i]...Ap[i+1]-1]', and the values of those entries appear in
the same locations in \verb'Ax'.
The column indices need not be in any particular order.
See Section~\ref{format_sparse_by_row} for details of the sparse-by-row (CSR) format.
\item \verb'GrB_CSC_FORMAT': % CSC format (equiv to GxB_SPARSE with GrB_COLMAJOR)
Compressed-column format. \verb'Ap' is an array of size \verb'ncols+1'.
The arrays \verb'Ai' and \verb'Ax' are of size \verb'nvals = Ap [ncols]',
and \verb'Ap[0]' must be zero.
The row indices of entries in the \verb'j'th column appear in
\verb'Ai[Ap[j]...Ap[j+1]-1]', and the values of those entries appear in
the same locations in \verb'Ax'.
The row indices need not be in any particular order.
See Section~\ref{format_sparse_by_col} for details of the sparse-by-column (CSC) format.
\item \verb'GrB_COO_FORMAT': % triplet format (like input to GrB*build)
Coordinate format. This is the same format as the \verb'I', \verb'J',
\verb'X' inputs to \verb'GrB_Matrix_build'. The three arrays
\verb'Ap', \verb'Ai', and \verb'Ax' have the same size. The \verb'k'th
tuple has row index \verb'Ai[k]', column index \verb'Ap[k]', and value
\verb'Ax[k]'. The tuples can appear any order, but no duplicates are
permitted.
% \item \verb'GrB_DENSE_ROW_FORMAT': % FullR format (GxB_FULL with GrB_ROWMAJOR)
% Dense matrix format, held by row. Only the \verb'Ax' array is used, of
% size \verb'nrows*ncols'.
% It holds the matrix in dense format, in row major order.
%
% \item \verb'GrB_DENSE_COL_FORMAT': % FullC format (GxB_FULL with GrB_ROWMAJOR)
% Dense matrix format, held by column. Only the \verb'Ax' array is used, of
% size \verb'nrows*ncols'.
% It holds the matrix in dense format, in column major order.
\end{packed_itemize}
%-------------------------------------------------------------------------------
\subsubsection{{\sf GrB\_Matrix\_export:} export a matrix}
%-------------------------------------------------------------------------------
\label{GrB_matrix_export}
\begin{mdframed}[userdefinedwidth=6in]
{\footnotesize
\begin{verbatim}
GrB_Info GrB_Matrix_export // export a matrix
(
GrB_Index *Ap, // pointers for CSR, CSC, column indices for COO
GrB_Index *Ai, // col indices for CSR/COO, row indices for CSC
<type> *Ax, // values (must match the type of A_input)
GrB_Index *Ap_len, // number of entries in Ap (not # of bytes)
GrB_Index *Ai_len, // number of entries in Ai (not # of bytes)
GrB_Index *Ax_len, // number of entries in Ax (not # of bytes)
int format, // export format (GrB_Format)
GrB_Matrix A // matrix to export
) ;
\end{verbatim}
} \end{mdframed}
\verb'GrB_Matrix_export' copies the contents of a matrix into three
user-provided arrays, using any one of the three different formats
described in Section~\ref{GrB_matrix_import}. The size of the arrays must be
at least as large as the lengths returned by \verb'GrB_Matrix_exportSize'. The
matrix \verb'A' is not modified.
On input, the size of the three arrays \verb'Ap', \verb'Ai', and \verb'Ax' is
given by \verb'Ap_len', \verb'Ai_len', and \verb'Ax_len', respectively. These
values are in terms of the number of entries in these arrays, not the number of
bytes. On output, these three value are adjusted to report the number of
entries written to the three arrays.
The suffix \verb'TYPE' in the prototype above is one of \verb'BOOL',
\verb'INT8', \verb'INT16', etc, for built-n types, or \verb'UDT' for
user-defined types. The type of the \verb'Ax' array must match this type. No
typecasting is performed.
\verb'GrB_Matrix_export' always exports the indices and offsets of the matrix
using 64-bit integer indices, even if they are held internally using 32-bit
integers.
% The \verb'GrB_DENSE_ROW_FORMAT' and \verb'GrB_DENSE_COL_FORMAT' formats can
% only be used if all entries are present in the matrix. That is,
% \verb'GrB_Matrix_nvals (&nvals,A)' must return \verb'nvals' equal to
% \verb'nrows*ncols'.
\newpage
%-------------------------------------------------------------------------------
\subsubsection{{\sf GrB\_Matrix\_exportSize:} determine size of export}
%-------------------------------------------------------------------------------
\label{export_size}
\begin{mdframed}[userdefinedwidth=6in]
{\footnotesize
\begin{verbatim}
GrB_Info GrB_Matrix_exportSize // determine sizes of user arrays for export
(
GrB_Index *Ap_len, // # of entries required for Ap (not # of bytes)
GrB_Index *Ai_len, // # of entries required for Ai (not # of bytes)
GrB_Index *Ax_len, // # of entries required for Ax (not # of bytes)
int format, // export format (GrB_Format)
GrB_Matrix A // matrix to export
) ;
\end{verbatim}
} \end{mdframed}
Returns the required sizes of the arrays \verb'Ap', \verb'Ai', and \verb'Ax'
for exporting a matrix using \verb'GrB_Matrix_export', using the same
\verb'format'.
%-------------------------------------------------------------------------------
\subsubsection{{\sf GrB\_Matrix\_exportHint:} determine best export format}
%-------------------------------------------------------------------------------
\label{export_hint}
\begin{mdframed}[userdefinedwidth=6in]
{\footnotesize
\begin{verbatim}
GrB_Info GrB_Matrix_exportHint // suggest the best export format
(
int *format, // export format (GrB_Format)
GrB_Matrix A // matrix to export
) ;
\end{verbatim}
} \end{mdframed}
This method suggests the most efficient format for the export of a given
matrix. For SuiteSparse:GraphBLAS, the hint depends on the current
format of the \verb'GrB_Matrix':
\begin{packed_itemize}
\item \verb'GxB_SPARSE', \verb'GrB_ROWMAJOR': export as \verb'GrB_CSR_FORMAT'
\item \verb'GxB_SPARSE', \verb'GrB_COLMAJOR': export as \verb'GrB_CSC_FORMAT'
\item \verb'GxB_HYPERSPARSE': export as \verb'GrB_COO_FORMAT'
\item \verb'GxB_BITMAP', \verb'GrB_ROWMAJOR': export as \verb'GrB_CSR_FORMAT'
\item \verb'GxB_BITMAP', \verb'GrB_COLMAJOR': export as \verb'GrB_CSC_FORMAT'
%\item \verb'GxB_FULL', \verb'GrB_ROWMAJOR': export as \verb'GrB_DENSE_ROW_FORMAT'
%\item \verb'GxB_FULL', \verb'GrB_COLMAJOR': export as \verb'GrB_DENSE_COL_FORMAT'
\item \verb'GxB_FULL', \verb'GrB_ROWMAJOR': export as \verb'GrB_CSR_FORMAT'
\item \verb'GxB_FULL', \verb'GrB_COLMAJOR': export as \verb'GrB_CSC_FORMAT'
\end{packed_itemize}
|
