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
|
//------------------------------------------------------------------------------
// GB_add_bitmap_M_sparse: C<!M>=A+B, C bitmap, M sparse/hyper and comp.
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// C is bitmap.
// M is sparse/hyper and complemented.
// A and B can have any format, except at least one is bitmap/full.
{
//--------------------------------------------------------------------------
// C is bitmap, M is sparse or hyper and complemented
//--------------------------------------------------------------------------
// ------------------------------------------
// C <!M> = A + B
// ------------------------------------------
// bitmap sparse sparse bitmap
// bitmap sparse sparse full
// bitmap sparse bitmap sparse
// bitmap sparse bitmap bitmap
// bitmap sparse bitmap full
// bitmap sparse full sparse
// bitmap sparse full bitmap
// bitmap sparse full full
// M is sparse and complemented. If M is sparse and not complemented, then
// C is constructed as sparse, not bitmap.
ASSERT (Mask_comp) ;
// C(i,j) = A(i,j) + B(i,j) can only be computed where M(i,j) is not
// present in the sparse pattern of M, and where it is present but equal to
// zero.
//--------------------------------------------------------------------------
// scatter M into the C bitmap
//--------------------------------------------------------------------------
const int64_t *kfirst_Mslice = M_ek_slicing ;
const int64_t *klast_Mslice = M_ek_slicing + M_ntasks ;
const int64_t *pstart_Mslice = M_ek_slicing + M_ntasks*2 ;
#pragma omp parallel for num_threads(M_nthreads) schedule(dynamic,1)
for (taskid = 0 ; taskid < M_ntasks ; taskid++)
{
int64_t kfirst = kfirst_Mslice [taskid] ;
int64_t klast = klast_Mslice [taskid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
// find the part of M(:,k) for this task
int64_t j = GBh_M (Mh, k) ;
GB_GET_PA (pM_start, pM_end, taskid, k, kfirst, klast,
pstart_Mslice, GB_IGET (Mp, k), GB_IGET (Mp, k+1)) ;
int64_t pC_start = j * vlen ;
// traverse over M(:,j), the kth vector of M
for (int64_t pM = pM_start ; pM < pM_end ; pM++)
{
// mark C(i,j) if M(i,j) is true
bool mij = GB_MCAST (Mx, pM, msize) ;
if (mij)
{
int64_t i = GB_IGET (Mi, pM) ;
int64_t p = pC_start + i ;
Cb [p] = 2 ;
}
}
}
}
// C(i,j) has been marked, in Cb, with the value 2 where M(i,j)=1.
// These positions will not be computed in C(i,j). C(i,j) can only
// be modified where Cb [p] is zero.
//--------------------------------------------------------------------------
// compute C<!M>=A+B using the mask scattered in C
//--------------------------------------------------------------------------
#ifdef GB_JIT_KERNEL
#if (GB_A_IS_BITMAP || GB_A_IS_FULL) && (GB_B_IS_BITMAP || GB_B_IS_FULL)
{
// A and B are both bitmap/full
#include "template/GB_add_bitmap_M_sparse_24.c"
#define M_cleared true
}
#elif (GB_A_IS_BITMAP || GB_A_IS_FULL)
{
// A is bitmap/full, B is sparse/hyper
#include "template/GB_add_bitmap_M_sparse_25.c"
#define M_cleared false
}
#else
{
// A is sparse/hyper, B is bitmap/full
#include "template/GB_add_bitmap_M_sparse_26.c"
#define M_cleared false
}
#endif
#else
bool M_cleared = false ;
if ((A_is_bitmap || A_is_full) && (B_is_bitmap || B_is_full))
{
// A and B are both bitmap/full
#include "template/GB_add_bitmap_M_sparse_24.c"
M_cleared = true ; // M has also been cleared from C
}
else if (A_is_bitmap || A_is_full)
{
// A is bitmap/full, B is sparse/hyper
#include "template/GB_add_bitmap_M_sparse_25.c"
}
else
{
// A is sparse/hyper, B is bitmap/full
#include "template/GB_add_bitmap_M_sparse_26.c"
}
#endif
//-------------------------------------------------------------------------
// clear M from C
//-------------------------------------------------------------------------
if (!M_cleared)
{
// This step is required if either A or B are sparse/hyper (if
// one is sparse/hyper, the other must be bitmap). It requires
// an extra pass over the mask M, so this might be slower than
// postponing the application of the mask, and doing it later.
#pragma omp parallel for num_threads(M_nthreads) schedule(dynamic,1)
for (taskid = 0 ; taskid < M_ntasks ; taskid++)
{
int64_t kfirst = kfirst_Mslice [taskid] ;
int64_t klast = klast_Mslice [taskid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
// find the part of M(:,k) for this task
int64_t j = GBh_M (Mh, k) ;
GB_GET_PA (pM_start, pM_end, taskid, k, kfirst, klast,
pstart_Mslice, GB_IGET (Mp, k), GB_IGET (Mp, k+1)) ;
int64_t pC_start = j * vlen ;
// traverse over M(:,j), the kth vector of M
for (int64_t pM = pM_start ; pM < pM_end ; pM++)
{
// mark C(i,j) if M(i,j) is true
bool mij = GB_MCAST (Mx, pM, msize) ;
if (mij)
{
int64_t i = GB_IGET (Mi, pM) ;
int64_t p = pC_start + i ;
Cb [p] = 0 ;
}
}
}
}
}
}
|
