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
|
//------------------------------------------------------------------------------
// GB_select_phase2: C=select(A,thunk)
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
{
//--------------------------------------------------------------------------
// get A
//--------------------------------------------------------------------------
const int64_t *restrict Ap = A->p ;
const int64_t *restrict Ah = A->h ;
const int64_t *restrict Ai = A->i ;
// if A is iso and the op is user-defined, Ax [0] is passed to the user
// selectop
const GB_ATYPE *restrict Ax = (GB_ATYPE *) A->x ;
size_t asize = A->type->size ;
int64_t avlen = A->vlen ;
int64_t avdim = A->vdim ;
// if A is bitmap, the bitmap selector is always used instead
ASSERT (!GB_IS_BITMAP (A)) ;
#ifndef GB_DIAG_SELECTOR
// if A is full, all opcodes except DIAG use the bitmap selector instead
ASSERT (!GB_IS_FULL (A)) ;
#endif
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 ;
//--------------------------------------------------------------------------
// C = select (A)
//--------------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < A_ntasks ; tid++)
{
// if kfirst > klast then task tid does no work at all
int64_t kfirst = kfirst_Aslice [tid] ;
int64_t klast = klast_Aslice [tid] ;
//----------------------------------------------------------------------
// selection from vectors kfirst to klast
//----------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//------------------------------------------------------------------
// find the part of A(:,k) to be operated on by this task
//------------------------------------------------------------------
int64_t pA_start, pA_end, pC ;
GB_get_pA_and_pC (&pA_start, &pA_end, &pC, tid, k, kfirst, klast,
pstart_Aslice, Cp_kfirst, Cp, avlen, Ap, avlen) ;
//------------------------------------------------------------------
// compact Ai and Ax [pA_start ... pA_end-1] into Ci and Cx
//------------------------------------------------------------------
#if defined ( GB_ENTRY_SELECTOR )
int64_t j = GBH (Ah, k) ;
for (int64_t pA = pA_start ; pA < pA_end ; pA++)
{
// A is never full; that case is now handled by the
// bitmap selector instead.
ASSERT (Ai != NULL) ;
int64_t i = Ai [pA] ;
GB_TEST_VALUE_OF_ENTRY (keep, pA) ;
if (keep)
{
ASSERT (pC >= Cp [k] && pC < Cp [k+1]) ;
Ci [pC] = i ;
// Cx [pC] = Ax [pA] ;
GB_SELECT_ENTRY (Cx, pC, Ax, pA) ;
pC++ ;
}
}
#elif defined ( GB_TRIL_SELECTOR ) || \
defined ( GB_ROWGT_SELECTOR )
// keep Zp [k] to pA_end-1
int64_t p = GB_IMAX (Zp [k], pA_start) ;
int64_t mynz = pA_end - p ;
if (mynz > 0)
{
// A and C are both sparse or hypersparse
ASSERT (pA_start <= p && p + mynz <= pA_end) ;
ASSERT (pC >= Cp [k] && pC + mynz <= Cp [k+1]) ;
ASSERT (Ai != NULL) ;
memcpy (Ci +pC, Ai +p, mynz*sizeof (int64_t)) ;
#if !GB_ISO_SELECT
memcpy (Cx +pC*asize, Ax +p*asize, mynz*asize) ;
#endif
}
#elif defined ( GB_TRIU_SELECTOR ) || \
defined ( GB_ROWLE_SELECTOR )
// keep pA_start to Zp[k]-1
int64_t p = GB_IMIN (Zp [k], pA_end) ;
int64_t mynz = p - pA_start ;
if (mynz > 0)
{
// A and C are both sparse or hypersparse
ASSERT (pC >= Cp [k] && pC + mynz <= Cp [k+1]) ;
ASSERT (Ai != NULL) ;
memcpy (Ci +pC, Ai +pA_start, mynz*sizeof (int64_t)) ;
#if !GB_ISO_SELECT
memcpy (Cx +pC*asize, Ax +pA_start*asize, mynz*asize) ;
#endif
}
#elif defined ( GB_DIAG_SELECTOR )
// task that owns the diagonal entry does this work
// A can be sparse or full, but not bitmap
int64_t p = Zp [k] ;
if (pA_start <= p && p < pA_end)
{
ASSERT (pC >= Cp [k] && pC + 1 <= Cp [k+1]) ;
Ci [pC] = GBI (Ai, p, avlen) ;
#if !GB_ISO_SELECT
memcpy (Cx +pC*asize, Ax +p*asize, asize) ;
#endif
}
#elif defined ( GB_OFFDIAG_SELECTOR ) || \
defined ( GB_ROWINDEX_SELECTOR )
// keep pA_start to Zp[k]-1
int64_t p = GB_IMIN (Zp [k], pA_end) ;
int64_t mynz = p - pA_start ;
if (mynz > 0)
{
// A and C are both sparse or hypersparse
ASSERT (pC >= Cp [k] && pC + mynz <= Cp [k+1]) ;
ASSERT (Ai != NULL) ;
memcpy (Ci +pC, Ai +pA_start, mynz*sizeof (int64_t)) ;
#if !GB_ISO_SELECT
memcpy (Cx +pC*asize, Ax +pA_start*asize, mynz*asize) ;
#endif
pC += mynz ;
}
// keep Zp[k]+1 to pA_end-1
p = GB_IMAX (Zp [k]+1, pA_start) ;
mynz = pA_end - p ;
if (mynz > 0)
{
// A and C are both sparse or hypersparse
ASSERT (pA_start <= p && p < pA_end) ;
ASSERT (pC >= Cp [k] && pC + mynz <= Cp [k+1]) ;
ASSERT (Ai != NULL) ;
memcpy (Ci +pC, Ai +p, mynz*sizeof (int64_t)) ;
#if !GB_ISO_SELECT
memcpy (Cx +pC*asize, Ax +p*asize, mynz*asize) ;
#endif
}
#endif
}
}
}
|
