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
|
//------------------------------------------------------------------------------
// GB_bitmap_assign_A_template: traverse over A for bitmap assignment into C
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// This template traverses over all the entries of the matrix A and operates on
// the corresponding entry in C(i,j), using the GB_AIJ_WORK macro. A can be
// hypersparse, sparse, bitmap, or full. It is not a scalar. The matrix
// C must be bitmap or full.
//
// The workspace must already be declared as follows:
//
// GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
// int A_ntasks = 0, A_nthreads = 0 ;
//
// The workspace is allocated and tasks are computed, if not already done.
// It is not freed, so it can be used for subsequent uses of this template.
// To free the workspace, the method that uses this template must do:
//
// GB_WERK_POP (A_ek_slicing, int64_t) ;
{
//--------------------------------------------------------------------------
// slice the matrix A
//--------------------------------------------------------------------------
if (A_ek_slicing == NULL)
{
GB_A_NHELD (A_nnz_held) ;
GB_SLICE_MATRIX_WORK (A, 8, A_nnz_held + A->nvec, A_nnz_held) ;
}
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 ;
//--------------------------------------------------------------------------
// traverse the entries of the matrix A
//--------------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1) \
reduction(+:cnvals)
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] ;
int64_t task_cnvals = 0 ;
//----------------------------------------------------------------------
// traverse over A (:,kfirst:klast)
//----------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//------------------------------------------------------------------
// find the part of A(:,k) for this task
//------------------------------------------------------------------
int64_t jA = GBh_A (Ah, k) ;
GB_GET_PA (pA_start, pA_end, tid, k, kfirst, klast, pstart_Aslice,
GBp_A (Ap, k, nI), GBp_A (Ap, k+1, nI)) ;
//------------------------------------------------------------------
// traverse over A(:,jA), the kth vector of A
//------------------------------------------------------------------
int64_t jC = GB_IJLIST (J, jA, GB_J_KIND, Jcolon) ;
int64_t pC0 = jC * Cvlen ; // first entry in C(:,jC)
for (int64_t pA = pA_start ; pA < pA_end ; pA++)
{
if (!GBb_A (Ab, pA)) continue ;
int64_t iA = GBi_A (Ai, pA, nI) ;
int64_t iC = GB_IJLIST (I, iA, GB_I_KIND, Icolon) ;
int64_t pC = iC + pC0 ;
// operate on C(iC,jC) at pC, and A(iA,jA) at pA. The mask
// can be accessed at pC if M is bitmap or full. A has any
// sparsity format so only A(iA,jA) can be accessed at pA.
// To access a full matrix M for the subassign case, use
// the position (iA + jA*nI).
GB_AIJ_WORK (pC, pA) ;
}
}
#ifndef GB_NO_CNVALS
cnvals += task_cnvals ;
#endif
}
}
|
