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
|
//------------------------------------------------------------------------------
// GB_emult_02_template: C = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// C is sparse, with the same sparsity structure as A. No mask is present, or
// M is bitmap/full. A is sparse/hyper, and B is bitmap/full. This method
// also handles the case when the original input A is bitmap/full and B is
// sparse/hyper, by computing B.*A with the operator flipped.
{
//--------------------------------------------------------------------------
// get A, B, and C
//--------------------------------------------------------------------------
const int64_t *restrict Ap = A->p ;
const int64_t *restrict Ah = A->h ;
const int64_t *restrict Ai = A->i ;
const int64_t vlen = A->vlen ;
const int8_t *restrict Bb = B->b ;
const int64_t *restrict kfirst_Aslice = A_ek_slicing ;
const int64_t *restrict klast_Aslice = A_ek_slicing + A_ntasks ;
const int64_t *restrict pstart_Aslice = A_ek_slicing + A_ntasks * 2 ;
const bool A_iso = A->iso ;
const bool B_iso = B->iso ;
#ifdef GB_ISO_EMULT
ASSERT (C->iso) ;
#else
ASSERT (!C->iso) ;
ASSERT (!(A_iso && B_iso)) ; // one of A or B can be iso, but not both
#if GB_FLIPPED
const GB_BTYPE *restrict Ax = (GB_BTYPE *) A->x ;
const GB_ATYPE *restrict Bx = (GB_ATYPE *) B->x ;
#else
const GB_ATYPE *restrict Ax = (GB_ATYPE *) A->x ;
const GB_BTYPE *restrict Bx = (GB_BTYPE *) B->x ;
#endif
GB_CTYPE *restrict Cx = (GB_CTYPE *) C->x ;
#endif
const int64_t *restrict Cp = C->p ;
int64_t *restrict Ci = C->i ;
//--------------------------------------------------------------------------
// C=A.*B or C<#M>=A.*B
//--------------------------------------------------------------------------
if (M == NULL)
{
//----------------------------------------------------------------------
// C = A.*B
//----------------------------------------------------------------------
if (GB_IS_BITMAP (B))
{
//------------------------------------------------------------------
// Method2(a): C=A.*B where A is sparse/hyper and B is bitmap
//------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < A_ntasks ; tid++)
{
int64_t kfirst = kfirst_Aslice [tid] ;
int64_t klast = klast_Aslice [tid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
int64_t j = GBH (Ah, k) ;
int64_t pB_start = j * vlen ;
int64_t pA, pA_end, pC ;
GB_get_pA_and_pC (&pA, &pA_end, &pC, tid, k, kfirst, klast,
pstart_Aslice, Cp_kfirst, Cp, vlen, Ap, vlen) ;
for ( ; pA < pA_end ; pA++)
{
int64_t i = Ai [pA] ;
int64_t pB = pB_start + i ;
if (!Bb [pB]) continue ;
// C (i,j) = A (i,j) .* B (i,j)
Ci [pC] = i ;
#ifndef GB_ISO_EMULT
GB_GETA (aij, Ax, pA, A_iso) ;
GB_GETB (bij, Bx, pB, B_iso) ;
#if GB_FLIPPED
GB_BINOP (GB_CX (pC), bij, aij, i, j) ;
#else
GB_BINOP (GB_CX (pC), aij, bij, i, j) ;
#endif
#endif
pC++ ;
}
}
}
}
else
{
//------------------------------------------------------------------
// Method2(b): C=A.*B where A is sparse/hyper and B is full
//------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < A_ntasks ; tid++)
{
int64_t kfirst = kfirst_Aslice [tid] ;
int64_t klast = klast_Aslice [tid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
int64_t j = GBH (Ah, k) ;
int64_t pB_start = j * vlen ;
int64_t pA, pA_end ;
GB_get_pA (&pA, &pA_end, tid, k, kfirst, klast,
pstart_Aslice, Ap, vlen) ;
for ( ; pA < pA_end ; pA++)
{
// C (i,j) = A (i,j) .* B (i,j)
int64_t i = Ai [pA] ;
int64_t pB = pB_start + i ;
// Ci [pA] = i ; already defined
#ifndef GB_ISO_EMULT
GB_GETA (aij, Ax, pA, A_iso) ;
GB_GETB (bij, Bx, pB, B_iso) ;
#if GB_FLIPPED
GB_BINOP (GB_CX (pA), bij, aij, i, j) ;
#else
GB_BINOP (GB_CX (pA), aij, bij, i, j) ;
#endif
#endif
}
}
}
}
}
else
{
//----------------------------------------------------------------------
// Method2(c): C<#M>=A.*B, A is sparse/hyper, M and B are bitmap/full
//----------------------------------------------------------------------
const int8_t *restrict Mb = M->b ;
const GB_void *restrict Mx = (Mask_struct) ? NULL : ((GB_void *) M->x) ;
const size_t msize = M->type->size ;
int tid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < A_ntasks ; tid++)
{
int64_t kfirst = kfirst_Aslice [tid] ;
int64_t klast = klast_Aslice [tid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
int64_t j = GBH (Ah, k) ;
int64_t pB_start = j * vlen ;
int64_t pA, pA_end, pC ;
GB_get_pA_and_pC (&pA, &pA_end, &pC, tid, k, kfirst, klast,
pstart_Aslice, Cp_kfirst, Cp, vlen, Ap, vlen) ;
for ( ; pA < pA_end ; pA++)
{
int64_t i = Ai [pA] ;
int64_t pB = pB_start + i ;
if (!GBB (Bb, pB)) continue ;
bool mij = GBB (Mb, pB) && GB_mcast (Mx, pB, msize) ;
mij = mij ^ Mask_comp ;
if (!mij) continue ;
// C (i,j) = A (i,j) .* B (i,j)
Ci [pC] = i ;
#ifndef GB_ISO_EMULT
GB_GETA (aij, Ax, pA, A_iso) ;
GB_GETB (bij, Bx, pB, B_iso) ;
#if GB_FLIPPED
GB_BINOP (GB_CX (pC), bij, aij, i, j) ;
#else
GB_BINOP (GB_CX (pC), aij, bij, i, j) ;
#endif
#endif
pC++ ;
}
}
}
}
}
|
