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
|
//==============================================================================
//=== sfmult_AN_XT_YT ==========================================================
//==============================================================================
// SFMULT, Copyright (c) 2009, Timothy A Davis. All Rights Reserved.
// SPDX-License-Identifier: BSD-3-clause
// y = (A*x')' A is m-by-n, x is k-by-n, y is k-by-m
// compare with sfmult_AN_XT_YN for kernel usage
// compare with sfmult_AN_XN_YT for outer loop structure but different kernels
#include "sfmult.h"
mxArray *sfmult_AN_XT_YT // y = (A*x')'
(
const mxArray *A,
const mxArray *X,
int ac, // if true: conj(A) if false: A. ignored if A real
int xc, // if true: conj(x) if false: x. ignored if x real
int yc // if true: conj(y) if false: y. ignored if y real
)
{
mxArray *Y ;
double *Ax, *Az, *Xx, *Xz, *Yx, *Yz, *Wx, *Wz ;
Int *Ap, *Ai ;
Int m, n, k, k1, i ;
int Acomplex, Xcomplex, Ycomplex ;
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
m = mxGetM (A) ;
n = mxGetN (A) ;
k = mxGetM (X) ;
if (n != mxGetN (X)) sfmult_invalid ( ) ;
Acomplex = mxIsComplex (A) ;
Xcomplex = mxIsComplex (X) ;
Ap = mxGetJc (A) ;
Ai = mxGetIr (A) ;
Ax = mxGetPr (A) ;
Az = mxGetPi (A) ;
Xx = mxGetPr (X) ;
Xz = mxGetPi (X) ;
//--------------------------------------------------------------------------
// allocate result
//--------------------------------------------------------------------------
Ycomplex = Acomplex || Xcomplex ;
Y = sfmult_yalloc (k, m, Ycomplex) ;
Yx = mxGetPr (Y) ;
Yz = mxGetPi (Y) ;
//--------------------------------------------------------------------------
// special cases
//--------------------------------------------------------------------------
if (k == 0 || m == 0 || n == 0 || Ap [n] == 0)
{
// Y = 0
return (sfmult_yzero (Y)) ;
}
if (k == 1)
{
// Y = A*X
sfmult_AN_x_1 (Yx, Yz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc) ;
return (Y) ;
}
if (k == 2)
{
// Y = (A * X')'
sfmult_AN_XT_YT_2 (Yx, Yz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k);
return (Y) ;
}
if (k == 4)
{
// Y = (A * X')'
sfmult_AN_XT_YT_4 (Yx, Yz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k);
return (Y) ;
}
if (k > 12 && Ap [n] < 8 * MAX (m,n)) // (TO DO) check overflow
{
// Y = (A * X')' when A is moderately sparse and X is large
mxArray *C ;
double *Cx, *Cz ;
Int *Cp, *Ci ;
// C = A' ;
C = ssmult_transpose (A, 0) ;
// (TO DO) if C is NULL, skip this and try using (A*X')' below
// Y = (C' * X')' when A is moderately sparse and X is large
Cp = mxGetJc (C) ;
Ci = mxGetIr (C) ;
Cx = mxGetPr (C) ;
Cz = mxGetPi (C) ;
sfmult_xA (Yx, Yz, Cp, Ci, Cx, Cz, n, m, Xx, Xz, ac, xc, yc, k) ;
mxDestroyArray (C) ;
return (Y) ;
}
//--------------------------------------------------------------------------
// allocate workspace
//--------------------------------------------------------------------------
sfmult_walloc (4, m, &Wx, &Wz) ;
// (TO DO) if walloc fails, use a workspace-free technique.
// This may require new sparse-times-vector kernels (both x and y with
// non-unit strides)
//--------------------------------------------------------------------------
// Y = (A*X')', in blocks of up to 4 columns of X, using sfmult_anxtyt
//--------------------------------------------------------------------------
k1 = k % 4 ;
if (k1 == 1)
{
// W = A * X(1,:)'
sfmult_AN_xk_1 (Wx, Wz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k) ;
// Y (1,:) = W'
for (i = 0 ; i < m ; i++)
{
Yx [k*i] = Wx [i] ;
}
Yx += 1 ;
Yz += 1 ;
Xx += 1 ;
Xz += 1 ;
}
else if (k1 == 2)
{
// W = (A * X(1:2,:)')'
sfmult_AN_XT_YT_2 (Wx, Wz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k);
// Y (1:2,:) = W
for (i = 0 ; i < m ; i++)
{
Yx [k*i ] = Wx [2*i ] ;
Yx [k*i+1] = Wx [2*i+1] ;
}
Yx += 2 ;
Yz += 2 ;
Xx += 2 ;
Xz += 2 ;
}
else if (k1 == 3)
{
// W = (A * X(1:3,:)')'
sfmult_AN_XT_YT_3 (Wx, Wz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k);
// Y (1:3,:) = W
for (i = 0 ; i < m ; i++)
{
Yx [k*i ] = Wx [4*i ] ;
Yx [k*i+1] = Wx [4*i+1] ;
Yx [k*i+2] = Wx [4*i+2] ;
}
Yx += 3 ;
Yz += 3 ;
Xx += 3 ;
Xz += 3 ;
}
for ( ; k1 < k ; k1 += 4)
{
// W = (A * X(1:4,:)')'
sfmult_AN_XT_YT_4 (Wx, Wz, Ap, Ai, Ax, Az, m, n, Xx, Xz, ac, xc, yc, k);
// Y (k1+(1:4),:) = W
for (i = 0 ; i < m ; i++)
{
Yx [k*i ] = Wx [4*i ] ;
Yx [k*i+1] = Wx [4*i+1] ;
Yx [k*i+2] = Wx [4*i+2] ;
Yx [k*i+3] = Wx [4*i+3] ;
}
Yx += 4 ;
Yz += 4 ;
Xx += 4 ;
Xz += 4 ;
}
//--------------------------------------------------------------------------
// free workspace and return result
//--------------------------------------------------------------------------
mxFree (Wx) ;
return (Y) ;
}
|
