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
|
//------------------------------------------------------------------------------
// GB_AxB_saxpy5: compute C+=A*B where A is bitmap/full and B is sparse/hyper
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// GB_AxB_saxpy5 computes C+=A*B where C is full, A is bitmap/full, and B is
// sparse/hypersparse. No mask is present, C_replace is false, and the accum
// matches the monoid.
// The type of A and C must match the multiply op. B can be typecasted
// if using the JIT.
// See also GB_AxB_saxpy4, which computes C+=A*B but with the sparsity formats
// of A and B reversed.
// Typically, if B is a large sparse matrix, the number of columns of A and C
// will be large, so A and C will typically be short-and-fat dense matrices.
// As a result, only a coarse-grain method is used, where no atomics are
// needed.
// The ANY monoid is not supported, since its use as accum would be unusual.
// TODO: if the monoid is ANY, quick return GrB_SUCCESS and done_in_place
// true, also for saxpy4. No work is needed and C doesn't change.
// FUTURE: expand use of AVX to more semirings
//------------------------------------------------------------------------------
#include "mxm/GB_mxm.h"
#include "jitifyer/GB_stringify.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "FactoryKernels/GB_AxB__include2.h"
#endif
#define GB_FREE_WORKSPACE \
{ \
GB_WERK_POP (B_slice, int64_t) ; \
}
#define GB_FREE_ALL \
{ \
GB_FREE_WORKSPACE ; \
GB_phybix_free (C) ; \
}
//------------------------------------------------------------------------------
// GB_AxB_saxpy5: compute C+=A*B in-place
//------------------------------------------------------------------------------
GrB_Info GB_AxB_saxpy5 // C += A*B
(
GrB_Matrix C, // users input/output matrix
const GrB_Matrix A, // input matrix A
const GrB_Matrix B, // input matrix B
const GrB_Semiring semiring, // semiring that defines C=A*B and accum
const bool flipxy, // if true, do z=fmult(b,a) vs fmult(a,b)
bool *done_in_place, // if true, saxpy5 has computed the result
GB_Werk Werk
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
GrB_Info info ;
GB_WERK_DECLARE (B_slice, int64_t) ;
// C type must match the ztype of the monoid
ASSERT_MATRIX_OK (C, "C for saxpy5 C+=A*B", GB0) ;
ASSERT (GB_IS_FULL (C)) ;
ASSERT (!GB_PENDING (C)) ;
ASSERT (!GB_JUMBLED (C)) ;
ASSERT (!GB_ZOMBIES (C)) ;
ASSERT (C->type == semiring->add->op->ztype) ;
// A cannot be typecasted
ASSERT_MATRIX_OK (A, "A for saxpy5 C+=A*B", GB0) ;
ASSERT (GB_IS_BITMAP (A) || GB_IS_FULL (A)) ;
ASSERT (!GB_PENDING (A)) ;
ASSERT (!GB_JUMBLED (A)) ;
ASSERT (!GB_ZOMBIES (A)) ;
ASSERT (A->type == (flipxy ? semiring->multiply->ytype :
semiring->multiply->xtype)) ;
// B can have any type for the JIT kernel
ASSERT_MATRIX_OK (B, "B for saxpy5 C+=A*B", GB0) ;
ASSERT (GB_IS_SPARSE (B) || GB_IS_HYPERSPARSE (B)) ;
ASSERT (!GB_PENDING (B)) ;
ASSERT (GB_JUMBLED_OK (B)) ;
ASSERT (!GB_ZOMBIES (B)) ;
ASSERT_SEMIRING_OK (semiring, "semiring for saxpy5 C+=A*B", GB0) ;
ASSERT (A->vdim == B->vlen) ;
//--------------------------------------------------------------------------
// get the semiring operators
//--------------------------------------------------------------------------
GrB_BinaryOp mult = semiring->multiply ;
ASSERT (mult->ztype == semiring->add->op->ztype) ;
bool A_is_pattern, B_is_pattern ;
GB_binop_pattern (&A_is_pattern, &B_is_pattern, flipxy, mult->opcode) ;
GB_Opcode mult_binop_code, add_binop_code ;
GB_Type_code xcode, ycode, zcode ;
bool builtin_semiring = GB_AxB_semiring_builtin (A, A_is_pattern, B,
B_is_pattern, semiring, flipxy, &mult_binop_code, &add_binop_code,
&xcode, &ycode, &zcode) ;
if (add_binop_code == GB_ANY_binop_code)
{
// The semiring cannot use the ANY monoid.
// The semiring must be builtin, or use the JIT (no generic method).
GBURBLE ("(punt) ") ;
return (GrB_NO_VALUE) ;
}
GBURBLE ("(saxpy5: %s += %s*%s) ",
GB_sparsity_char_matrix (C),
GB_sparsity_char_matrix (A),
GB_sparsity_char_matrix (B)) ;
//--------------------------------------------------------------------------
// ensure C is non-iso
//--------------------------------------------------------------------------
GB_OK (GB_convert_any_to_non_iso (C, true)) ;
//--------------------------------------------------------------------------
// determine the # of threads to use and the parallel tasks
//--------------------------------------------------------------------------
int64_t anz = GB_nnz_held (A) ;
int64_t bnz = GB_nnz_held (B) ;
int64_t bnvec = B->nvec ;
int nthreads_max = GB_Context_nthreads_max ( ) ;
double chunk = GB_Context_chunk ( ) ;
int nthreads = GB_nthreads (anz + bnz, chunk, nthreads_max) ;
int ntasks = (nthreads == 1) ? 1 : 4 * nthreads ;
ntasks = GB_IMIN (ntasks, bnvec) ;
GB_WERK_PUSH (B_slice, ntasks + 1, int64_t) ;
if (B_slice == NULL)
{
// out of memory
GB_FREE_WORKSPACE ;
return (GrB_OUT_OF_MEMORY) ;
}
GB_p_slice (B_slice, B->p, B->p_is_32, bnvec, ntasks, false) ;
//--------------------------------------------------------------------------
// via the factory kernel
//--------------------------------------------------------------------------
info = GrB_NO_VALUE ;
#ifndef GBCOMPACT
GB_IF_FACTORY_KERNELS_ENABLED
{
//----------------------------------------------------------------------
// define the worker for the switch factory
//----------------------------------------------------------------------
#define GB_Asaxpy5B(add,mult,xname) \
GB (_Asaxpy5B_ ## add ## mult ## xname)
#define GB_AxB_WORKER(add,mult,xname) \
{ \
info = GB_Asaxpy5B (add,mult,xname) (C, A, B, \
ntasks, nthreads, B_slice) ; \
} \
break ;
//----------------------------------------------------------------------
// launch the switch factory
//----------------------------------------------------------------------
// disable the ANY monoid
#define GB_NO_ANY_MONOID
if (builtin_semiring)
{
#include "mxm/factory/GB_AxB_factory.c"
}
}
#endif
//--------------------------------------------------------------------------
// via the JIT or PreJIT kernel
//--------------------------------------------------------------------------
if (info == GrB_NO_VALUE)
{
info = GB_AxB_saxpy5_jit (C, A, B, semiring, flipxy, ntasks, nthreads,
B_slice) ;
}
//--------------------------------------------------------------------------
// free workspace and return result
//--------------------------------------------------------------------------
GB_FREE_WORKSPACE ;
if (info == GrB_NO_VALUE)
{
// saxpy5 doesn't handle this case; punt to saxpy3, bitmap saxpy, etc
GBURBLE ("(punt) ") ;
}
else if (info == GrB_SUCCESS)
{
ASSERT_MATRIX_OK (C, "saxpy5: output", GB0) ;
(*done_in_place) = true ;
}
else
{
// out of memory, or other error
GB_FREE_ALL ;
}
return (info) ;
}
|
