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
|
//------------------------------------------------------------------------------
// GB_subassign_23_template: C += A where C is full
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Method 23: C += A, where C is full
// M: NULL
// Mask_comp: false
// Mask_struct: ignored
// C_replace: false
// accum: present
// A: matrix
// S: none
// The type of C must match the type of x and z for the accum function, since
// C(i,j) = accum (C(i,j), A(i,j)) is handled. The generic case here can
// typecast A(i,j) but not C(i,j). The case for typecasting of C is handled by
// Method 04.
// C and A can have any sparsity structure, but C must be as-if-full.
#include "include/GB_unused.h"
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_WERK_POP (A_ek_slicing, int64_t) ; \
}
{
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
const bool A_is_bitmap = GB_IS_BITMAP (A) ;
const bool A_is_full = GB_IS_FULL (A) ;
const bool A_iso = A->iso ;
//--------------------------------------------------------------------------
// slice the A matrix
//--------------------------------------------------------------------------
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
int A_ntasks, A_nthreads ;
GB_A_NHELD (anz) ; // int64_t anz = GB_nnz_held (A) ;
double work = anz + A->nvec ;
if (GB_A_IS_BITMAP || GB_A_IS_FULL)
{
// C is full and A is bitmap or full: A_ek_slicing is not created.
A_nthreads = GB_nthreads (work, chunk, nthreads_max) ;
A_ntasks = 0 ; // unused
ASSERT (A_ek_slicing == NULL) ;
}
else
{
// create tasks to compute over the matrix A
GB_SLICE_MATRIX_WORK (A, 32, work, anz) ;
ASSERT (A_ek_slicing != NULL) ;
}
//--------------------------------------------------------------------------
// get C and A
//--------------------------------------------------------------------------
ASSERT (!C->iso) ;
const GB_A_TYPE *restrict Ax = (GB_A_TYPE *) A->x ;
GB_C_TYPE *restrict Cx = (GB_C_TYPE *) C->x ;
ASSERT (GB_IS_FULL (C)) ;
GB_C_NHELD (cnz) ; // const int64_t cnz = GB_nnz_held (C) ;
GB_DECLAREY (ywork) ;
if (GB_A_ISO)
{
// get the iso value of A and typecast it to Y
// ywork = (ytype) Ax [0]
GB_COPY_aij_to_ywork (ywork, Ax, 0, true) ;
}
if (GB_A_IS_BITMAP)
{
//----------------------------------------------------------------------
// C += A when C is full and A is bitmap
//----------------------------------------------------------------------
const int8_t *restrict Ab = A->b ;
int64_t p ;
#pragma omp parallel for num_threads(A_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
if (!Ab [p]) continue ;
// Cx [p] += (ytype) Ax [p], with typecasting
GB_ACCUMULATE_aij (Cx, p, Ax, p, GB_A_ISO, ywork, false) ;
}
}
else if (GB_A_IS_FULL)
{
//----------------------------------------------------------------------
// C += A when both C and A are ffull
//----------------------------------------------------------------------
int64_t p ;
#pragma omp parallel for num_threads(A_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
// Cx [p] += (ytype) Ax [p], with typecasting
GB_ACCUMULATE_aij (Cx, p, Ax, p, GB_A_ISO, ywork, false) ;
}
}
else
{
//----------------------------------------------------------------------
// C += A when C is full and A is sparse
//----------------------------------------------------------------------
ASSERT (GB_JUMBLED_OK (A)) ;
GB_Ap_DECLARE (Ap, const) ; GB_Ap_PTR (Ap, A) ;
GB_Ah_DECLARE (Ah, const) ; GB_Ah_PTR (Ah, A) ;
GB_Ai_DECLARE (Ai, const) ; GB_Ai_PTR (Ai, A) ;
const int64_t avlen = A->vlen ;
const int64_t Cvlen = C->vlen ;
bool A_jumbled = A->jumbled ;
const int64_t *restrict kfirst_Aslice = A_ek_slicing ;
const int64_t *restrict klast_Aslice = kfirst_Aslice + A_ntasks ;
const int64_t *restrict pstart_Aslice = klast_Aslice + A_ntasks ;
int taskid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (taskid = 0 ; taskid < A_ntasks ; taskid++)
{
// if kfirst > klast then taskid does no work at all
int64_t kfirst = kfirst_Aslice [taskid] ;
int64_t klast = klast_Aslice [taskid] ;
//------------------------------------------------------------------
// C(:,kfirst:klast) += A(:,kfirst:klast)
//------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//--------------------------------------------------------------
// find the part of A(:,k) and C(:,k) for this task
//--------------------------------------------------------------
int64_t j = GBh_A (Ah, k) ;
int64_t pA_start = GB_IGET (Ap, k) ;
int64_t pA_end = GB_IGET (Ap, k+1) ;
GB_GET_PA (my_pA_start, my_pA_end, taskid, k,
kfirst, klast, pstart_Aslice, pA_start, pA_end) ;
bool ajdense = ((pA_end - pA_start) == Cvlen) ;
// pC points to the start of C(:,j)
int64_t pC = j * Cvlen ;
//--------------------------------------------------------------
// C(:,j) += A(:,j)
//--------------------------------------------------------------
if (ajdense && !A_jumbled)
{
//----------------------------------------------------------
// A(:,j) is dense
//----------------------------------------------------------
GB_PRAGMA_SIMD_VECTORIZE
for (int64_t pA = my_pA_start ; pA < my_pA_end ; pA++)
{
int64_t i = pA - pA_start ;
int64_t p = pC + i ;
// Cx [p] += (ytype) Ax [pA], with typecasting
GB_ACCUMULATE_aij (Cx, p, Ax, pA, GB_A_ISO, ywork,
false) ;
}
}
else
{
//----------------------------------------------------------
// A(:,j) is sparse
//----------------------------------------------------------
GB_PRAGMA_SIMD_VECTORIZE
for (int64_t pA = my_pA_start ; pA < my_pA_end ; pA++)
{
int64_t i = GB_IGET (Ai, pA) ;
int64_t p = pC + i ;
// Cx [p] += (ytype) Ax [pA], with typecasting
GB_ACCUMULATE_aij (Cx, p, Ax, pA, GB_A_ISO, ywork,
false) ;
}
}
}
}
}
GB_FREE_ALL ;
}
#undef GB_FREE_ALL
#define GB_FREE_ALL ;
|
