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
|
//------------------------------------------------------------------------------
// GB_dense_subassign_23: C += B where C is dense and B is sparse or dense
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// C and B must have the same vector dimension and vector length.
// FUTURE::: the transposed case, C+=B' could easily be done.
// The parallelism used is identical to GB_AxB_colscale.
// The type of C must match the type of x and z for the accum function, since
// C(i,j) = accum (C(i,j), B(i,j)) is handled. The generic case here can
// typecast B(i,j) but not C(i,j). The case for typecasting of C is handled by
// Method 04.
// The caller passes in the second matrix as A, but it is called B here to
// match its use as the 2nd input to the binary accum operator. C and B can
// have any sparsity structure, but C must be dense.
#include "GB_dense.h"
#include "GB_binop.h"
#ifndef GBCUDA_DEV
#include "GB_binop__include.h"
#endif
#include "GB_unused.h"
#define GB_FREE_ALL \
{ \
GB_WERK_POP (B_ek_slicing, int64_t) ; \
}
GrB_Info GB_dense_subassign_23 // C += B; C is dense, B is sparse or dense
(
GrB_Matrix C, // input/output matrix
const GrB_Matrix B, // input matrix
const GrB_BinaryOp accum, // operator to apply
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (!GB_aliased (C, B)) ; // NO ALIAS of C==A (A is called B here)
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
GrB_Info info ;
ASSERT_MATRIX_OK (C, "C for C+=B", GB0) ;
ASSERT (!GB_PENDING (C)) ;
ASSERT (!GB_JUMBLED (C)) ;
ASSERT (!GB_ZOMBIES (C)) ;
ASSERT (GB_is_dense (C)) ;
ASSERT_MATRIX_OK (B, "B for C+=B", GB0) ;
ASSERT (!GB_PENDING (B)) ;
ASSERT (GB_JUMBLED_OK (B)) ;
ASSERT (!GB_ZOMBIES (B)) ;
ASSERT_BINARYOP_OK (accum, "accum for C+=B", GB0) ;
ASSERT (!GB_OP_IS_POSITIONAL (accum)) ;
ASSERT (B->vlen == C->vlen) ;
ASSERT (B->vdim == C->vdim) ;
GB_ENSURE_FULL (C) ; // convert C to full, if sparsity control allows it
//--------------------------------------------------------------------------
// get the operator
//--------------------------------------------------------------------------
if (accum->opcode == GB_FIRST_binop_code || C->iso)
{
// nothing to do
return (GrB_SUCCESS) ;
}
// C = accum (C,B) will be computed
ASSERT (C->type == accum->ztype) ;
ASSERT (C->type == accum->xtype) ;
ASSERT (GB_Type_compatible (B->type, accum->ytype)) ;
//--------------------------------------------------------------------------
// determine the number of threads to use
//--------------------------------------------------------------------------
GB_GET_NTHREADS_MAX (nthreads_max, chunk, Context) ;
//--------------------------------------------------------------------------
// slice the entries for each task
//--------------------------------------------------------------------------
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
int B_ntasks, B_nthreads ;
if (GB_IS_BITMAP (B) || GB_as_if_full (B))
{
// C is dense and B is bitmap or as-if-full
GBURBLE ("(Z bitmap/as-if-full) ") ;
int64_t bnvec = B->nvec ;
int64_t bnz = GB_nnz_held (B) ;
B_nthreads = GB_nthreads (bnz + bnvec, chunk, nthreads_max) ;
B_ntasks = 0 ; // unused
ASSERT (B_ek_slicing == NULL) ;
}
else
{
// create tasks to compute over the matrix B
GB_SLICE_MATRIX (B, 32, chunk) ;
ASSERT (B_ek_slicing != NULL) ;
}
//--------------------------------------------------------------------------
// C += B, sparse accum into dense, with built-in binary operators
//--------------------------------------------------------------------------
bool done = false ;
#ifndef GBCUDA_DEV
//----------------------------------------------------------------------
// define the worker for the switch factory
//----------------------------------------------------------------------
#define GB_Cdense_accumB(accum,xname) \
GB (_Cdense_accumB_ ## accum ## xname)
#define GB_BINOP_WORKER(accum,xname) \
{ \
info = GB_Cdense_accumB(accum,xname) (C, B, \
B_ek_slicing, B_ntasks, B_nthreads) ; \
done = (info != GrB_NO_VALUE) ; \
} \
break ;
//----------------------------------------------------------------------
// launch the switch factory
//----------------------------------------------------------------------
GB_Opcode opcode ;
GB_Type_code xcode, ycode, zcode ;
if (GB_binop_builtin (C->type, false, B->type, false, // C = C + B
accum, false, &opcode, &xcode, &ycode, &zcode))
{
// accumulate sparse matrix into dense matrix with built-in operator
#include "GB_binop_factory.c"
}
#endif
//--------------------------------------------------------------------------
// C += B, sparse accum into dense, with typecasting or user-defined op
//--------------------------------------------------------------------------
if (!done)
{
//----------------------------------------------------------------------
// get operators, functions, workspace, contents of B and C
//----------------------------------------------------------------------
GB_BURBLE_MATRIX (B, "(generic C+=B) ") ;
GxB_binary_function fadd = accum->binop_function ;
size_t csize = C->type->size ;
size_t bsize = B->type->size ;
size_t ysize = accum->ytype->size ;
GB_cast_function cast_B_to_Y ;
// B is typecasted to y
cast_B_to_Y = GB_cast_factory (accum->ytype->code, B->type->code) ;
//----------------------------------------------------------------------
// C += B via function pointers, and typecasting
//----------------------------------------------------------------------
// bij = B(i,j), located in Bx [pB]. Note that GB_GETB is used,
// since B appears as the 2nd input to z = fadd (x,y)
#define GB_GETB(bij,Bx,pB,B_iso) \
GB_void bij [GB_VLA(ysize)] ; \
cast_B_to_Y (bij, Bx +(B_iso ? 0:(pB)*bsize), bsize)
// address of Cx [p]
#define GB_CX(p) Cx +((p)*csize)
#define GB_BTYPE GB_void
#define GB_CTYPE GB_void
// no vectorization
#define GB_PRAGMA_SIMD_VECTORIZE ;
#define GB_BINOP(z,x,y,i,j) fadd (z,x,y)
#include "GB_dense_subassign_23_template.c"
}
//--------------------------------------------------------------------------
// free workspace and return result
//--------------------------------------------------------------------------
GB_FREE_ALL ;
ASSERT_MATRIX_OK (C, "C+=B output", GB0) ;
return (GrB_SUCCESS) ;
}
|
