Mercurial > octave-nkf
annotate libcruft/blas-xtra/zconv2.f @ 14596:ad1d7a53b1ca stable
perform convolution properly independently of matrix orientation (bug #34893)
* cconv2.f (cconv2i), csconv2.f (csconv2i), dconv2.f (dconv2i),
sconv2.f (sconv2i), zconv2.f (zconv2i), zdconv2.f (zdconv2i): Fix
iteration over the a matrix, so it's properly reversed.
* conv2.cc: Uncomment test to check for this, and add a new xtest.
author | Sunghyun Cho <sodomau@gmail.com> |
---|---|
date | Tue, 17 Apr 2012 12:47:34 +0100 |
parents | 72c96de7a403 |
children |
rev | line source |
---|---|
14138
72c96de7a403
maint: update copyright notices for 2012
John W. Eaton <jwe@octave.org>
parents:
11523
diff
changeset
|
1 c Copyright (C) 2010-2012 VZLU Prague, a.s., Czech Republic |
10388
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
2 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
3 c Author: Jaroslav Hajek <highegg@gmail.com> |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
4 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
5 c This file is part of Octave. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
6 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
7 c Octave is free software; you can redistribute it and/or modify |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
8 c it under the terms of the GNU General Public License as published by |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
9 c the Free Software Foundation; either version 3 of the License, or |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
10 c (at your option) any later version. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
11 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
12 c This program is distributed in the hope that it will be useful, |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
13 c but WITHOUT ANY WARRANTY; without even the implied warranty of |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
14 c MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
15 c GNU General Public License for more details. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
16 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
17 c You should have received a copy of the GNU General Public License |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
18 c along with this software; see the file COPYING. If not, see |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
19 c <http://www.gnu.org/licenses/>. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
20 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
21 subroutine zconv2o(ma,na,a,mb,nb,b,c) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
22 c purpose: a 2-dimensional outer additive convolution. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
23 c equivalent to the following: |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
24 c for i = 1:ma |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
25 c for j = 1:na |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
26 c c(i:i+mb-1,j:j+mb-1) += a(i,j)*b |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
27 c endfor |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
28 c endfor |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
29 c arguments: |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
30 c ma,na (in) dimensions of a |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
31 c a (in) 1st matrix |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
32 c mb,nb (in) dimensions of b |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
33 c b (in) 2nd matrix |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
34 c c (inout) accumulator matrix, size (ma+mb-1, na+nb-1) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
35 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
36 integer ma,na,mb,nb |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
37 double complex a(ma,na),b(mb,nb) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
38 double complex c(ma+mb-1,na+nb-1) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
39 integer i,j,k |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
40 external zaxpy |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
41 do k = 1,na |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
42 do j = 1,nb |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
43 do i = 1,mb |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
44 call zaxpy(ma,b(i,j),a(1,k),1,c(i,j+k-1),1) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
45 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
46 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
47 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
48 end subroutine |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
49 |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
50 subroutine zconv2i(ma,na,a,mb,nb,b,c) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
51 c purpose: a 2-dimensional inner additive convolution. |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
52 c equivalent to the following: |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
53 c for i = 1:ma-mb+1 |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
54 c for j = 1:na-nb+1 |
10946
1094868ca10d
fix bugs in inner convolution
Jaroslav Hajek <highegg@gmail.com>
parents:
10388
diff
changeset
|
55 c c(i,j) = sum (sum (a(i+mb-1:-1:i,j+nb-1:-1:j) .* b)) |
10388
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
56 c endfor |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
57 c endfor |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
58 c arguments: |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
59 c ma,na (in) dimensions of a |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
60 c a (in) 1st matrix |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
61 c mb,nb (in) dimensions of b |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
62 c b (in) 2nd matrix |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
63 c c (inout) accumulator matrix, size (ma+mb-1, na+nb-1) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
64 c |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
65 integer ma,na,mb,nb |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
66 double complex a(ma,na),b(mb,nb) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
67 double complex c(ma-mb+1,na-nb+1) |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
68 integer i,j,k |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
69 external zaxpy |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
70 do k = 1,na-nb+1 |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
71 do j = 1,nb |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
72 do i = 1,mb |
14596
ad1d7a53b1ca
perform convolution properly independently of matrix orientation (bug #34893)
Sunghyun Cho <sodomau@gmail.com>
parents:
14138
diff
changeset
|
73 call zaxpy(ma-mb+1,b(i,j),a(mb+1-i,k+nb-j),1,c(1,k),1) |
10388
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
74 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
75 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
76 end do |
5af0b4bb384d
rewrite convn optimizations based on xAXPY
Jaroslav Hajek <highegg@gmail.com>
parents:
diff
changeset
|
77 end subroutine |