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
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@generated from sparse/control/magma_zparic_kernels.cpp, normal z -> s, Wed Jan 22 14:42:32 2025
@author Hartwig Anzt
*/
#include "magmasparse_internal.h"
#ifdef _OPENMP
#include <omp.h>
#endif
#define SWAP(a, b) { val_swap = a; a = b; b = val_swap; }
/***************************************************************************//**
Purpose
-------
This function does one asynchronous ParILU sweep (symmetric case).
Input and output array is identical.
Arguments
---------
@param[in]
A magma_s_matrix
System matrix in COO.
@param[in]
L magma_s_matrix*
Current approximation for the lower triangular factor
The format is sorted CSR.
@param[in]
queue magma_queue_t
Queue to execute in.
@ingroup magmasparse_saux
*******************************************************************************/
extern "C" magma_int_t
magma_sparic_sweep(
magma_s_matrix A,
magma_s_matrix *L,
magma_queue_t queue )
{
magma_int_t info = 0;
int i, j;
float zero = MAGMA_S_MAKE(0.0, 0.0);
int il, iu, jl, ju;
#pragma omp parallel for
for (int k=0; k < A.nnz; k++) {
i = A.rowidx[k];
j = A.col[k];
float s, sp;
s = A.val[k];
sp = zero;
il = L->row[i];
iu = L->row[j];
while (il < L->row[i+1] && iu < L->row[j+1])
{
sp = zero;
jl = L->col[il];
ju = L->col[iu];
// avoid branching
sp = ( jl == ju ) ? L->val[il] * L->val[iu] : sp;
s = ( jl == ju ) ? s-sp : s;
il = ( jl <= ju ) ? il+1 : il;
iu = ( jl >= ju ) ? iu+1 : iu;
}
// undo the last operation (it must be the last)
s += sp;
if ( i > j ) // modify l entry
L->val[il-1] = s / L->val[L->row[j+1]-1];
else { // modify u entry
L->val[iu-1] = MAGMA_S_MAKE( sqrt( fabs( MAGMA_S_REAL(s) )), 0.0 );
}
}
return info;
}
/***************************************************************************//**
Purpose
-------
This function does one synchronized ParILU sweep (symmetric case).
Input and output are different arrays.
Arguments
---------
@param[in]
A magma_s_matrix
System matrix in COO.
@param[in]
L magma_s_matrix*
Current approximation for the lower triangular factor
The format is sorted CSR.
@param[in]
queue magma_queue_t
Queue to execute in.
@ingroup magmasparse_saux
*******************************************************************************/
extern "C" magma_int_t
magma_sparic_sweep_sync(
magma_s_matrix A,
magma_s_matrix *L,
magma_queue_t queue )
{
magma_int_t info = 0;
int i, j;
int il, iu, jl, ju;
float zero = MAGMA_S_MAKE(0.0, 0.0);
float *L_new_val = NULL, *val_swap = NULL;
CHECK( magma_smalloc_cpu( &L_new_val, L->nnz ));
#pragma omp parallel for
for (int k=0; k < A.nnz; k++) {
i = A.rowidx[k];
j = A.col[k];
float s, sp;
s = A.val[k];
sp = zero;
il = L->row[i];
iu = L->row[j];
while (il < L->row[i+1] && iu < L->row[j+1])
{
sp = zero;
jl = L->col[il];
ju = L->col[iu];
// avoid branching
sp = ( jl == ju ) ? L->val[il] * L->val[iu] : sp;
s = ( jl == ju ) ? s-sp : s;
il = ( jl <= ju ) ? il+1 : il;
iu = ( jl >= ju ) ? iu+1 : iu;
}
// undo the last operation (it must be the last)
s += sp;
if ( i > j ) // modify l entry
L_new_val[il-1] = s / L->val[L->row[j+1]-1];
else { // modify u entry
L_new_val[iu-1] = MAGMA_S_MAKE( sqrt( fabs( MAGMA_S_REAL(s) )), 0.0 );
}
}
// swap old and new values
SWAP( L_new_val, L->val );
cleanup:
magma_free_cpu( L_new_val );
return info;
}
|
