Mercurial > hg > octave-lyh

annotate liboctave/floatQRP.cc @ 10396:a0b51ac0f88a

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
optimize accumdim with summation
author Jaroslav Hajek <highegg@gmail.com>
date Fri, 05 Mar 2010 12:31:30 +0100
parents 12884915a8e4
children 4d1fc073fbb7
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
1 /*
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
2
8920
eb63fbe60fab update copyright notices
John W. Eaton <jwe@octave.org>
parents: 8811
diff changeset
3 Copyright (C) 1994, 1995, 1996, 1997, 2002, 2003, 2004, 2005, 2007,
eb63fbe60fab update copyright notices
John W. Eaton <jwe@octave.org>
parents: 8811
diff changeset
4 2008, 2009 John W. Eaton
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
5
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
6 This file is part of Octave.
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
7
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
8 Octave is free software; you can redistribute it and/or modify it
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
9 under the terms of the GNU General Public License as published by the
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
10 Free Software Foundation; either version 3 of the License, or (at your
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
11 option) any later version.
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
12
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
13 Octave is distributed in the hope that it will be useful, but WITHOUT
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
16 for more details.
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
17
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
18 You should have received a copy of the GNU General Public License
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
19 along with Octave; see the file COPYING. If not, see
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
20 <http://www.gnu.org/licenses/>.
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
21
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
22 */
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
23
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
24 #ifdef HAVE_CONFIG_H
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
25 #include <config.h>
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
26 #endif
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
27
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
28 #include <cassert>
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
29
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
30 #include "floatQRP.h"
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
31 #include "f77-fcn.h"
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
32 #include "lo-error.h"
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
33 #include "oct-locbuf.h"
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
34
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
35 extern "C"
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
36 {
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
37 F77_RET_T
8368
c72c1c9bccdc call blocked permuted qr factorization routines from LAPACK
Jaroslav Hajek <highegg@gmail.com>
parents: 8367
diff changeset
38 F77_FUNC (sgeqp3, SGEQP3) (const octave_idx_type&, const octave_idx_type&, float*,
10314
07ebe522dac2 untabify liboctave C++ sources
John W. Eaton <jwe@octave.org>
parents: 10158
diff changeset
39 const octave_idx_type&, octave_idx_type*, float*, float*,
8368
c72c1c9bccdc call blocked permuted qr factorization routines from LAPACK
Jaroslav Hajek <highegg@gmail.com>
parents: 8367
diff changeset
40 const octave_idx_type&, octave_idx_type&);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
41 }
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
42
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
43 // It would be best to share some of this code with QR class...
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
44
9713
7918eb15040c refactor the QR classes onto a templated base
Jaroslav Hajek <highegg@gmail.com>
parents: 8920
diff changeset
45 FloatQRP::FloatQRP (const FloatMatrix& a, qr_type_t qr_type)
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
46 : FloatQR (), p ()
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
47 {
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
48 init (a, qr_type);
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
49 }
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
50
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
51 void
9713
7918eb15040c refactor the QR classes onto a templated base
Jaroslav Hajek <highegg@gmail.com>
parents: 8920
diff changeset
52 FloatQRP::init (const FloatMatrix& a, qr_type_t qr_type)
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
53 {
9713
7918eb15040c refactor the QR classes onto a templated base
Jaroslav Hajek <highegg@gmail.com>
parents: 8920
diff changeset
54 assert (qr_type != qr_type_raw);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
55
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
56 octave_idx_type m = a.rows ();
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
57 octave_idx_type n = a.cols ();
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
58
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
59 octave_idx_type min_mn = m < n ? m : n;
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
60 OCTAVE_LOCAL_BUFFER (float, tau, min_mn);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
61
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
62 octave_idx_type info = 0;
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
63
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
64 FloatMatrix afact = a;
9713
7918eb15040c refactor the QR classes onto a templated base
Jaroslav Hajek <highegg@gmail.com>
parents: 8920
diff changeset
65 if (m > n && qr_type == qr_type_std)
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
66 afact.resize (m, m);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
67
10350
12884915a8e4 merge MArray classes & improve Array interface
Jaroslav Hajek <highegg@gmail.com>
parents: 10314
diff changeset
68 MArray<octave_idx_type> jpvt (n, 1, 0);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
69
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
70 if (m > 0)
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
71 {
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
72 // workspace query.
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
73 float rlwork;
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
74 F77_XFCN (sgeqp3, SGEQP3, (m, n, afact.fortran_vec (), m, jpvt.fortran_vec (),
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
75 tau, &rlwork, -1, info));
8368
c72c1c9bccdc call blocked permuted qr factorization routines from LAPACK
Jaroslav Hajek <highegg@gmail.com>
parents: 8367
diff changeset
76
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
77 // allocate buffer and do the job.
8811
20dfb885f877 int -> octave_idx fixes
John W. Eaton <jwe@octave.org>
parents: 8597
diff changeset
78 octave_idx_type lwork = rlwork;
20dfb885f877 int -> octave_idx fixes
John W. Eaton <jwe@octave.org>
parents: 8597
diff changeset
79 lwork = std::max (lwork, static_cast<octave_idx_type> (1));
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
80 OCTAVE_LOCAL_BUFFER (float, work, lwork);
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
81 F77_XFCN (sgeqp3, SGEQP3, (m, n, afact.fortran_vec (), m, jpvt.fortran_vec (),
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
82 tau, work, lwork, info));
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
83 }
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
84 else
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
85 for (octave_idx_type i = 0; i < n; i++) jpvt(i) = i+1;
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
86
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
87 // Form Permutation matrix (if economy is requested, return the
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
88 // indices only!)
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
89
8811
20dfb885f877 int -> octave_idx fixes
John W. Eaton <jwe@octave.org>
parents: 8597
diff changeset
90 jpvt -= static_cast<octave_idx_type> (1);
8367
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
91 p = PermMatrix (jpvt, true);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
92
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
93
8597
c86718093c1b improve & fix QR classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8375
diff changeset
94 form (n, afact, tau, qr_type);
7792
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
95 }
39c1026191e9 add missing files from single-precision merge
John W. Eaton <jwe@octave.org>
parents:
diff changeset
96
8367
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
97 FloatColumnVector
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
98 FloatQRP::Pvec (void) const
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
99 {
8375
e3c9102431a9 fix design problems of diag & perm matrix classes
Jaroslav Hajek <highegg@gmail.com>
parents: 8368
diff changeset
100 Array<float> pa (p.pvec ());
8367
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
101 FloatColumnVector pv (MArray<float> (pa) + 1.0f);
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
102 return pv;
445d27d79f4e support permutation matrix objects
Jaroslav Hajek <highegg@gmail.com>
parents: 7792
diff changeset
103 }