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
|
//------------------------------------------------------------------------------
// GB_subassign_25_template: C<M> = A where C is empty and A is dense
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Method 25: C(:,:)<M,s> = A ; C is empty, M structural, A bitmap/as-if-full
// M: present
// Mask_comp: false
// Mask_struct: true
// C_replace: effectively false (not relevant since C is empty)
// accum: NULL
// A: matrix
// S: none
// C and M are sparse or hypersparse. A can have any sparsity structure, even
// bitmap, but it must either be bitmap, or as-if-full. M may be jumbled. If
// so, C is constructed as jumbled. C is reconstructed with the same structure
// as M and can have any sparsity structure on input. The only constraint on C
// is nnz(C) is zero on input.
// C is iso if A is iso
// C<M> = A where C starts as empty, M is structural, and A is dense. The
// pattern of C is an exact copy of M. A is full, dense, or bitmap.
// M is sparse or hypersparse, and C is constructed with the same pattern as M.
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_WERK_POP (M_ek_slicing, int64_t) ; \
}
{
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
const bool A_is_bitmap = GB_IS_BITMAP (A) ;
const bool A_iso = A->iso ;
ASSERT (GB_IS_FULL (A) || GB_A_IS_BITMAP) ;
//--------------------------------------------------------------------------
// Parallel: slice M into equal-sized chunks
//--------------------------------------------------------------------------
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
int M_nthreads, M_ntasks ;
GB_M_NHELD (M_nnz_held) ;
GB_SLICE_MATRIX_WORK (M, 8, M_nnz_held + M->nvec, M_nnz_held) ;
//--------------------------------------------------------------------------
// get C, M, and A
//--------------------------------------------------------------------------
GB_Ci_DECLARE (Ci, ) ; GB_Ci_PTR (Ci, C) ;
ASSERT (GB_IS_SPARSE (M) || GB_IS_HYPERSPARSE (M)) ;
ASSERT (GB_JUMBLED_OK (M)) ;
GB_Mp_DECLARE (Mp, const) ; GB_Mp_PTR (Mp, M) ;
GB_Mh_DECLARE (Mh, const) ; GB_Mh_PTR (Mh, M) ;
GB_Mi_DECLARE (Mi, const) ; GB_Mi_PTR (Mi, M) ;
const int64_t Mvlen = M->vlen ;
const int8_t *restrict Ab = A->b ;
const int64_t avlen = A->vlen ;
bool C_iso = C->iso ;
ASSERT (C->iso == A->iso) ;
#ifdef GB_ISO_ASSIGN
ASSERT (C->iso) ;
#else
ASSERT (!C->iso) ;
const GB_A_TYPE *restrict Ax = (GB_A_TYPE *) A->x ;
GB_C_TYPE *restrict Cx = (GB_C_TYPE *) C->x ;
GB_DECLAREC (cwork) ;
if (GB_A_ISO)
{
// get the iso value of A and typecast to C->type
// cwork = (ctype) Ax [0]
// This is no longer used. If A is iso, so is C, and in that case,
// GB_ISO_ASSIGN is true and cwork is not used here.
ASSERT (GB_DEAD_CODE) ;
GB_COPY_aij_to_cwork (cwork, Ax, 0, true) ;
}
#endif
//--------------------------------------------------------------------------
// C<M> = A
//--------------------------------------------------------------------------
if (GB_A_IS_BITMAP)
{
//----------------------------------------------------------------------
// A is bitmap, so zombies can be created in C
//----------------------------------------------------------------------
int64_t nzombies = 0 ;
int tid ;
#pragma omp parallel for num_threads(M_nthreads) schedule(dynamic,1) \
reduction(+:nzombies)
for (tid = 0 ; tid < M_ntasks ; tid++)
{
// if kfirst > klast then task tid does no work at all
int64_t kfirst = kfirst_Mslice [tid] ;
int64_t klast = klast_Mslice [tid] ;
int64_t task_nzombies = 0 ;
//------------------------------------------------------------------
// C<M(:,kfirst:klast)> = A(:,kfirst:klast)
//------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//--------------------------------------------------------------
// find the part of M(:,k) to be operated on by this task
//--------------------------------------------------------------
int64_t j = GBh_M (Mh, k) ;
GB_GET_PA (pM_start, pM_end, tid, k, kfirst, klast,
pstart_Mslice, GB_IGET (Mp, k), GB_IGET (Mp, k+1)) ;
//--------------------------------------------------------------
// C<M(:,j)> = A(:,j)
//--------------------------------------------------------------
// M is hypersparse or sparse. C is the same as M.
// pA points to the start of A(:,j) since A is dense
int64_t pA = j * avlen ;
for (int64_t pM = pM_start ; pM < pM_end ; pM++)
{
int64_t i = GB_IGET (Mi, pM) ;
int64_t p = pA + i ;
if (Ab [p])
{
// C(i,j) = A(i,j)
#ifndef GB_ISO_ASSIGN
GB_COPY_aij_to_C (Cx, pM, Ax, p, GB_A_ISO, cwork,
GB_C_ISO) ;
#endif
}
else
{
// C(i,j) becomes a zombie
task_nzombies++ ;
i = GB_ZOMBIE (i) ;
GB_ISET (Ci, pM, i) ; // Ci [pM] = i
}
}
}
nzombies += task_nzombies ;
}
C->nzombies = nzombies ;
}
else
{
//----------------------------------------------------------------------
// A is full, so no zombies will appear in C
//----------------------------------------------------------------------
#ifndef GB_ISO_ASSIGN
{
int tid ;
#pragma omp parallel for num_threads(M_nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < M_ntasks ; tid++)
{
// if kfirst > klast then task tid does no work at all
int64_t kfirst = kfirst_Mslice [tid] ;
int64_t klast = klast_Mslice [tid] ;
//--------------------------------------------------------------
// C<M(:,kfirst:klast)> = A(:,kfirst:klast)
//--------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//----------------------------------------------------------
// find the part of M(:,k) to be operated on by this task
//----------------------------------------------------------
int64_t j = GBh_M (Mh, k) ;
GB_GET_PA (pM_start, pM_end, tid, k, kfirst, klast,
pstart_Mslice, GB_IGET (Mp, k), GB_IGET (Mp, k+1)) ;
//----------------------------------------------------------
// C<M(:,j)> = A(:,j)
//----------------------------------------------------------
// M is hypersparse or sparse. C is the same as M.
// pA points to the start of A(:,j) since A is dense
int64_t pA = j * avlen ;
GB_PRAGMA_SIMD_VECTORIZE
for (int64_t pM = pM_start ; pM < pM_end ; pM++)
{
// C(i,j) = A(i,j)
int64_t p = pA + GB_IGET (Mi, pM) ;
GB_COPY_aij_to_C (Cx, pM, Ax, p,
GB_A_ISO, cwork, GB_C_ISO) ;
}
}
}
}
#endif
}
GB_FREE_ALL ;
}
#undef GB_ISO_ASSIGN
#undef GB_FREE_ALL
#define GB_FREE_ALL ;
|
