nmatrix 0.0.7 → 0.0.8
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/History.txt +11 -0
- data/Manifest.txt +1 -2
- data/README.rdoc +2 -2
- data/ext/nmatrix/extconf.rb +4 -3
- data/ext/nmatrix/math.cpp +2 -0
- data/ext/nmatrix/math/getri.h +1 -0
- data/ext/nmatrix/math/inc.h +41 -0
- data/ext/nmatrix/math/math.h +0 -1
- data/ext/nmatrix/nmatrix.cpp +1 -0
- data/lib/nmatrix/version.rb +1 -1
- data/nmatrix.gemspec +1 -1
- data/spec/lapack_spec.rb +2 -3
- metadata +7 -7
- data/ext/nmatrix/math/lapack.h +0 -879
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA1:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 29f34a19d63da6254ccd6557dc500391a6c766d2
|
|
4
|
+
data.tar.gz: e70dcc15d4d154d60dfdc12ffc28eaf05705b37a
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: 96ce3ba8f13ac7ee5f75a8e27cc3e878dc7d12aa98691c00082ceb342c86f9123e43a1e6593c87142c10bd119f571ad8ff4fb77f48b555b6dc4d86290de24778
|
|
7
|
+
data.tar.gz: 232b743d7b6272e4abcd15a82298d3179fa2f222dc16d754bc5f898bff98aab1c0d0a28d387527dce14b4e500d4d167e61d5636c0570303ffae2238f1b6dd825
|
data/History.txt
CHANGED
|
@@ -378,3 +378,14 @@
|
|
|
378
378
|
* Fixed conversion from Ruby object to Complex and Rational
|
|
379
379
|
|
|
380
380
|
* Fixed memory leak in slicing
|
|
381
|
+
|
|
382
|
+
=== 0.0.8 / 2013-08-23
|
|
383
|
+
|
|
384
|
+
* 2 bug fixes:
|
|
385
|
+
|
|
386
|
+
* Fixed Ubuntu compilation bug caused by math header file
|
|
387
|
+
refactoring
|
|
388
|
+
|
|
389
|
+
* Fixed pry version error which manifests on some systems but not
|
|
390
|
+
others
|
|
391
|
+
|
data/Manifest.txt
CHANGED
|
@@ -54,7 +54,6 @@ ext/nmatrix/storage/yale.h
|
|
|
54
54
|
ext/nmatrix/util/sl_list.cpp
|
|
55
55
|
ext/nmatrix/util/sl_list.h
|
|
56
56
|
ext/nmatrix/util/util.h
|
|
57
|
-
ext/nmatrix/util/lapack.h
|
|
58
57
|
ext/nmatrix/math.cpp
|
|
59
58
|
ext/nmatrix/math/asum.h
|
|
60
59
|
ext/nmatrix/math/geev.h
|
|
@@ -68,7 +67,7 @@ ext/nmatrix/math/getrf.h
|
|
|
68
67
|
ext/nmatrix/math/getri.h
|
|
69
68
|
ext/nmatrix/math/getrs.h
|
|
70
69
|
ext/nmatrix/math/idamax.h
|
|
71
|
-
ext/nmatrix/math/
|
|
70
|
+
ext/nmatrix/math/inc.h
|
|
72
71
|
ext/nmatrix/math/laswp.h
|
|
73
72
|
ext/nmatrix/math/long_dtype.h
|
|
74
73
|
ext/nmatrix/math/math.h
|
data/README.rdoc
CHANGED
|
@@ -37,7 +37,7 @@ If you want to obtain the latest (development) code, you should generally do:
|
|
|
37
37
|
bundle install
|
|
38
38
|
bundle exec rake compile
|
|
39
39
|
bundle exec rake repackage
|
|
40
|
-
gem install pkg/nmatrix-0.0.
|
|
40
|
+
gem install pkg/nmatrix-0.0.8.gem
|
|
41
41
|
|
|
42
42
|
Detailed instructions are available for {Mac}[https://github.com/SciRuby/nmatrix/wiki/Installation#mac-os-x] and {Linux}[https://github.com/SciRuby/nmatrix/wiki/Installation#linux].
|
|
43
43
|
|
|
@@ -79,7 +79,7 @@ Read the instructions in +CONTRIBUTING.md+ if you want to help NMatrix.
|
|
|
79
79
|
|
|
80
80
|
== Features
|
|
81
81
|
|
|
82
|
-
The following features exist in the current version of NMatrix (0.0.
|
|
82
|
+
The following features exist in the current version of NMatrix (0.0.8):
|
|
83
83
|
|
|
84
84
|
* Matrix and vector storage containers: dense, yale, list (more to come)
|
|
85
85
|
* Data types: byte (uint8), int8, int16, int32, int64, float32, float64, complex64, complex128, rational64, rational128,
|
data/ext/nmatrix/extconf.rb
CHANGED
|
@@ -147,15 +147,16 @@ end
|
|
|
147
147
|
have_header("clapack.h")
|
|
148
148
|
have_header("cblas.h")
|
|
149
149
|
|
|
150
|
-
have_func("clapack_dgetrf", "clapack.h")
|
|
150
|
+
have_func("clapack_dgetrf", ["cblas.h", "clapack.h"])
|
|
151
|
+
have_func("clapack_dgetri", ["cblas.h", "clapack.h"])
|
|
151
152
|
have_func("dgesvd_", "clapack.h")
|
|
152
153
|
|
|
154
|
+
have_func("cblas_dgemm", "cblas.h")
|
|
155
|
+
|
|
153
156
|
|
|
154
157
|
#find_library("cblas", "cblas_dgemm")
|
|
155
158
|
#find_library("atlas", "ATL_dgemmNN")
|
|
156
159
|
|
|
157
|
-
have_func("cblas_dgemm", "cblas.h")
|
|
158
|
-
|
|
159
160
|
# Order matters here: ATLAS has to go after LAPACK: http://mail.scipy.org/pipermail/scipy-user/2007-January/010717.html
|
|
160
161
|
$libs += " -llapack -lcblas -latlas "
|
|
161
162
|
|
data/ext/nmatrix/math.cpp
CHANGED
data/ext/nmatrix/math/getri.h
CHANGED
|
@@ -0,0 +1,41 @@
|
|
|
1
|
+
/////////////////////////////////////////////////////////////////////
|
|
2
|
+
// = NMatrix
|
|
3
|
+
//
|
|
4
|
+
// A linear algebra library for scientific computation in Ruby.
|
|
5
|
+
// NMatrix is part of SciRuby.
|
|
6
|
+
//
|
|
7
|
+
// NMatrix was originally inspired by and derived from NArray, by
|
|
8
|
+
// Masahiro Tanaka: http://narray.rubyforge.org
|
|
9
|
+
//
|
|
10
|
+
// == Copyright Information
|
|
11
|
+
//
|
|
12
|
+
// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
|
|
13
|
+
// NMatrix is Copyright (c) 2013, Ruby Science Foundation
|
|
14
|
+
//
|
|
15
|
+
// Please see LICENSE.txt for additional copyright notices.
|
|
16
|
+
//
|
|
17
|
+
// == Contributing
|
|
18
|
+
//
|
|
19
|
+
// By contributing source code to SciRuby, you agree to be bound by
|
|
20
|
+
// our Contributor Agreement:
|
|
21
|
+
//
|
|
22
|
+
// * https://github.com/SciRuby/sciruby/wiki/Contributor-Agreement
|
|
23
|
+
//
|
|
24
|
+
// == inc.h
|
|
25
|
+
//
|
|
26
|
+
// Includes needed for LAPACK, CLAPACK, and CBLAS functions.
|
|
27
|
+
//
|
|
28
|
+
|
|
29
|
+
#ifndef INC_H
|
|
30
|
+
# define INC_H
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
extern "C" { // These need to be in an extern "C" block or you'll get all kinds of undefined symbol errors.
|
|
34
|
+
#include <cblas.h>
|
|
35
|
+
|
|
36
|
+
#ifdef HAVE_CLAPACK_H
|
|
37
|
+
#include <clapack.h>
|
|
38
|
+
#endif
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
#endif // INC_H
|
data/ext/nmatrix/math/math.h
CHANGED
data/ext/nmatrix/nmatrix.cpp
CHANGED
data/lib/nmatrix/version.rb
CHANGED
data/nmatrix.gemspec
CHANGED
|
@@ -54,7 +54,7 @@ EOF
|
|
|
54
54
|
gem.add_development_dependency 'rake', '~>0.9'
|
|
55
55
|
gem.add_development_dependency 'bundler'
|
|
56
56
|
gem.add_development_dependency 'rspec', '~>2.9.0'
|
|
57
|
-
gem.add_development_dependency 'pry'
|
|
57
|
+
gem.add_development_dependency 'pry'
|
|
58
58
|
gem.add_development_dependency 'guard-rspec', '~>0.7.0'
|
|
59
59
|
gem.add_development_dependency 'rake-compiler', '~>0.8.1'
|
|
60
60
|
end
|
data/spec/lapack_spec.rb
CHANGED
|
@@ -1,4 +1,3 @@
|
|
|
1
|
-
require 'pry'
|
|
2
1
|
# = NMatrix
|
|
3
2
|
#
|
|
4
3
|
# A linear algebra library for scientific computation in Ruby.
|
|
@@ -126,7 +125,7 @@ describe NMatrix::LAPACK do
|
|
|
126
125
|
end
|
|
127
126
|
end
|
|
128
127
|
|
|
129
|
-
it "exposes gesdd" do
|
|
128
|
+
it "exposes lapack gesdd" do
|
|
130
129
|
if [:float32, :float64].include? dtype
|
|
131
130
|
a = NMatrix.new([5,6], %w|8.79 9.93 9.83 5.45 3.16
|
|
132
131
|
6.11 6.91 5.04 -0.27 7.98
|
|
@@ -187,7 +186,7 @@ describe NMatrix::LAPACK do
|
|
|
187
186
|
end
|
|
188
187
|
|
|
189
188
|
|
|
190
|
-
it "exposes gesvd" do
|
|
189
|
+
it "exposes lapack gesvd" do
|
|
191
190
|
# http://software.intel.com/sites/products/documentation/doclib/mkl_sa/11/mkl_lapack_examples/dgesvd_ex.c.htm
|
|
192
191
|
if [:float32, :float64].include? dtype
|
|
193
192
|
a = NMatrix.new([5,6], %w|8.79 9.93 9.83 5.45 3.16
|
metadata
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
|
2
2
|
name: nmatrix
|
|
3
3
|
version: !ruby/object:Gem::Version
|
|
4
|
-
version: 0.0.
|
|
4
|
+
version: 0.0.8
|
|
5
5
|
platform: ruby
|
|
6
6
|
authors:
|
|
7
7
|
- John Woods
|
|
@@ -10,7 +10,7 @@ authors:
|
|
|
10
10
|
autorequire:
|
|
11
11
|
bindir: bin
|
|
12
12
|
cert_chain: []
|
|
13
|
-
date: 2013-08-
|
|
13
|
+
date: 2013-08-23 00:00:00.000000000 Z
|
|
14
14
|
dependencies:
|
|
15
15
|
- !ruby/object:Gem::Dependency
|
|
16
16
|
name: rdoc
|
|
@@ -72,16 +72,16 @@ dependencies:
|
|
|
72
72
|
name: pry
|
|
73
73
|
requirement: !ruby/object:Gem::Requirement
|
|
74
74
|
requirements:
|
|
75
|
-
- -
|
|
75
|
+
- - '>='
|
|
76
76
|
- !ruby/object:Gem::Version
|
|
77
|
-
version: 0
|
|
77
|
+
version: '0'
|
|
78
78
|
type: :development
|
|
79
79
|
prerelease: false
|
|
80
80
|
version_requirements: !ruby/object:Gem::Requirement
|
|
81
81
|
requirements:
|
|
82
|
-
- -
|
|
82
|
+
- - '>='
|
|
83
83
|
- !ruby/object:Gem::Version
|
|
84
|
-
version: 0
|
|
84
|
+
version: '0'
|
|
85
85
|
- !ruby/object:Gem::Dependency
|
|
86
86
|
name: guard-rspec
|
|
87
87
|
requirement: !ruby/object:Gem::Requirement
|
|
@@ -149,7 +149,7 @@ files:
|
|
|
149
149
|
- ext/nmatrix/math/getri.h
|
|
150
150
|
- ext/nmatrix/math/getrs.h
|
|
151
151
|
- ext/nmatrix/math/idamax.h
|
|
152
|
-
- ext/nmatrix/math/
|
|
152
|
+
- ext/nmatrix/math/inc.h
|
|
153
153
|
- ext/nmatrix/math/laswp.h
|
|
154
154
|
- ext/nmatrix/math/long_dtype.h
|
|
155
155
|
- ext/nmatrix/math/math.h
|
data/ext/nmatrix/math/lapack.h
DELETED
|
@@ -1,879 +0,0 @@
|
|
|
1
|
-
/////////////////////////////////////////////////////////////////////
|
|
2
|
-
// = NMatrix
|
|
3
|
-
//
|
|
4
|
-
// A linear algebra library for scientific computation in Ruby.
|
|
5
|
-
// NMatrix is part of SciRuby.
|
|
6
|
-
//
|
|
7
|
-
// NMatrix was originally inspired by and derived from NArray, by
|
|
8
|
-
// Masahiro Tanaka: http://narray.rubyforge.org
|
|
9
|
-
//
|
|
10
|
-
// == Copyright Information
|
|
11
|
-
//
|
|
12
|
-
// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
|
|
13
|
-
// NMatrix is Copyright (c) 2013, Ruby Science Foundation
|
|
14
|
-
//
|
|
15
|
-
// Please see LICENSE.txt for additional copyright notices.
|
|
16
|
-
//
|
|
17
|
-
// == Contributing
|
|
18
|
-
//
|
|
19
|
-
// By contributing source code to SciRuby, you agree to be bound by
|
|
20
|
-
// our Contributor Agreement:
|
|
21
|
-
//
|
|
22
|
-
// * https://github.com/SciRuby/sciruby/wiki/Contributor-Agreement
|
|
23
|
-
//
|
|
24
|
-
// == lapack.h
|
|
25
|
-
//
|
|
26
|
-
// Templated versions of LAPACK functions, in C++.
|
|
27
|
-
|
|
28
|
-
#ifndef LAPACK_H
|
|
29
|
-
#define LAPACK_H
|
|
30
|
-
|
|
31
|
-
#include <cmath> // std::round
|
|
32
|
-
|
|
33
|
-
#include "math.h"
|
|
34
|
-
|
|
35
|
-
namespace nm { namespace math { namespace lapack {
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
/* -- LAPACK auxiliary routine (version 3.2) -- */
|
|
40
|
-
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
|
|
41
|
-
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
|
|
42
|
-
/* November 2006 */
|
|
43
|
-
|
|
44
|
-
/* .. Scalar Arguments .. */
|
|
45
|
-
|
|
46
|
-
/* Purpose */
|
|
47
|
-
/* ======= */
|
|
48
|
-
|
|
49
|
-
/* This program sets problem and machine dependent parameters */
|
|
50
|
-
/* useful for xHSEQR and its subroutines. It is called whenever */
|
|
51
|
-
/* ILAENV is called with 12 <= ISPEC <= 16 */
|
|
52
|
-
|
|
53
|
-
/* Arguments */
|
|
54
|
-
/* ========= */
|
|
55
|
-
|
|
56
|
-
/* ISPEC (input) int scalar */
|
|
57
|
-
/* ISPEC specifies which tunable parameter IPARMQ should */
|
|
58
|
-
/* return. */
|
|
59
|
-
|
|
60
|
-
/* ISPEC=12: (INMIN) Matrices of order nmin or less */
|
|
61
|
-
/* are sent directly to xLAHQR, the implicit */
|
|
62
|
-
/* double shift QR algorithm. NMIN must be */
|
|
63
|
-
/* at least 11. */
|
|
64
|
-
|
|
65
|
-
/* ISPEC=13: (INWIN) Size of the deflation window. */
|
|
66
|
-
/* This is best set greater than or equal to */
|
|
67
|
-
/* the number of simultaneous shifts NS. */
|
|
68
|
-
/* Larger matrices benefit from larger deflation */
|
|
69
|
-
/* windows. */
|
|
70
|
-
|
|
71
|
-
/* ISPEC=14: (INIBL) Determines when to stop nibbling and */
|
|
72
|
-
/* invest in an (expensive) multi-shift QR sweep. */
|
|
73
|
-
/* If the aggressive early deflation subroutine */
|
|
74
|
-
/* finds LD converged eigenvalues from an order */
|
|
75
|
-
/* NW deflation window and LD.GT.(NW*NIBBLE)/100, */
|
|
76
|
-
/* then the next QR sweep is skipped and early */
|
|
77
|
-
/* deflation is applied immediately to the */
|
|
78
|
-
/* remaining active diagonal block. Setting */
|
|
79
|
-
/* IPARMQ(ISPEC=14) = 0 causes TTQRE to skip a */
|
|
80
|
-
/* multi-shift QR sweep whenever early deflation */
|
|
81
|
-
/* finds a converged eigenvalue. Setting */
|
|
82
|
-
/* IPARMQ(ISPEC=14) greater than or equal to 100 */
|
|
83
|
-
/* prevents TTQRE from skipping a multi-shift */
|
|
84
|
-
/* QR sweep. */
|
|
85
|
-
|
|
86
|
-
/* ISPEC=15: (NSHFTS) The number of simultaneous shifts in */
|
|
87
|
-
/* a multi-shift QR iteration. */
|
|
88
|
-
|
|
89
|
-
/* ISPEC=16: (IACC22) IPARMQ is set to 0, 1 or 2 with the */
|
|
90
|
-
/* following meanings. */
|
|
91
|
-
/* 0: During the multi-shift QR sweep, */
|
|
92
|
-
/* xLAQR5 does not accumulate reflections and */
|
|
93
|
-
/* does not use matrix-matrix multiply to */
|
|
94
|
-
/* update the far-from-diagonal matrix */
|
|
95
|
-
/* entries. */
|
|
96
|
-
/* 1: During the multi-shift QR sweep, */
|
|
97
|
-
/* xLAQR5 and/or xLAQRaccumulates reflections and uses */
|
|
98
|
-
/* matrix-matrix multiply to update the */
|
|
99
|
-
/* far-from-diagonal matrix entries. */
|
|
100
|
-
/* 2: During the multi-shift QR sweep. */
|
|
101
|
-
/* xLAQR5 accumulates reflections and takes */
|
|
102
|
-
/* advantage of 2-by-2 block structure during */
|
|
103
|
-
/* matrix-matrix multiplies. */
|
|
104
|
-
/* (If xTRMM is slower than xGEMM, then */
|
|
105
|
-
/* IPARMQ(ISPEC=16)=1 may be more efficient than */
|
|
106
|
-
/* IPARMQ(ISPEC=16)=2 despite the greater level of */
|
|
107
|
-
/* arithmetic work implied by the latter choice.) */
|
|
108
|
-
|
|
109
|
-
/* NAME (input) character string */
|
|
110
|
-
/* Name of the calling subroutine */
|
|
111
|
-
|
|
112
|
-
/* OPTS (input) character string */
|
|
113
|
-
/* This is a concatenation of the string arguments to */
|
|
114
|
-
/* TTQRE. */
|
|
115
|
-
|
|
116
|
-
/* N (input) int scalar */
|
|
117
|
-
/* N is the order of the Hessenberg matrix H. */
|
|
118
|
-
|
|
119
|
-
/* ILO (input) INTEGER */
|
|
120
|
-
/* IHI (input) INTEGER */
|
|
121
|
-
/* It is assumed that H is already upper triangular */
|
|
122
|
-
/* in rows and columns 1:ILO-1 and IHI+1:N. */
|
|
123
|
-
|
|
124
|
-
/* LWORK (input) int scalar */
|
|
125
|
-
/* The amount of workspace available. */
|
|
126
|
-
|
|
127
|
-
/* Further Details */
|
|
128
|
-
/* =============== */
|
|
129
|
-
|
|
130
|
-
/* Little is known about how best to choose these parameters. */
|
|
131
|
-
/* It is possible to use different values of the parameters */
|
|
132
|
-
/* for each of CHSEQR, DHSEQR, SHSEQR and ZHSEQR. */
|
|
133
|
-
|
|
134
|
-
/* It is probably best to choose different parameters for */
|
|
135
|
-
/* different matrices and different parameters at different */
|
|
136
|
-
/* times during the iteration, but this has not been */
|
|
137
|
-
/* implemented --- yet. */
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
/* The best choices of most of the parameters depend */
|
|
141
|
-
/* in an ill-understood way on the relative execution */
|
|
142
|
-
/* rate of xLAQR3 and xLAQR5 and on the nature of each */
|
|
143
|
-
/* particular eigenvalue problem. Experiment may be the */
|
|
144
|
-
/* only practical way to determine which choices are most */
|
|
145
|
-
/* effective. */
|
|
146
|
-
|
|
147
|
-
/* Following is a list of default values supplied by IPARMQ. */
|
|
148
|
-
/* These defaults may be adjusted in order to attain better */
|
|
149
|
-
/* performance in any particular computational environment. */
|
|
150
|
-
|
|
151
|
-
/* IPARMQ(ISPEC=12) The xLAHQR vs xLAQR0 crossover point. */
|
|
152
|
-
/* Default: 75. (Must be at least 11.) */
|
|
153
|
-
|
|
154
|
-
/* IPARMQ(ISPEC=13) Recommended deflation window size. */
|
|
155
|
-
/* This depends on ILO, IHI and NS, the */
|
|
156
|
-
/* number of simultaneous shifts returned */
|
|
157
|
-
/* by IPARMQ(ISPEC=15). The default for */
|
|
158
|
-
/* (IHI-ILO+1).LE.500 is NS. The default */
|
|
159
|
-
/* for (IHI-ILO+1).GT.500 is 3*NS/2. */
|
|
160
|
-
|
|
161
|
-
/* IPARMQ(ISPEC=14) Nibble crossover point. Default: 14. */
|
|
162
|
-
|
|
163
|
-
/* IPARMQ(ISPEC=15) Number of simultaneous shifts, NS. */
|
|
164
|
-
/* a multi-shift QR iteration. */
|
|
165
|
-
|
|
166
|
-
/* If IHI-ILO+1 is ... */
|
|
167
|
-
|
|
168
|
-
/* greater than ...but less ... the */
|
|
169
|
-
/* or equal to ... than default is */
|
|
170
|
-
|
|
171
|
-
/* 0 30 NS = 2+ */
|
|
172
|
-
/* 30 60 NS = 4+ */
|
|
173
|
-
/* 60 150 NS = 10 */
|
|
174
|
-
/* 150 590 NS = ** */
|
|
175
|
-
/* 590 3000 NS = 64 */
|
|
176
|
-
/* 3000 6000 NS = 128 */
|
|
177
|
-
/* 6000 infinity NS = 256 */
|
|
178
|
-
|
|
179
|
-
/* (+) By default matrices of this order are */
|
|
180
|
-
/* passed to the implicit double shift routine */
|
|
181
|
-
/* xLAHQR. See IPARMQ(ISPEC=12) above. These */
|
|
182
|
-
/* values of NS are used only in case of a rare */
|
|
183
|
-
/* xLAHQR failure. */
|
|
184
|
-
|
|
185
|
-
/* (**) The asterisks (**) indicate an ad-hoc */
|
|
186
|
-
/* function increasing from 10 to 64. */
|
|
187
|
-
|
|
188
|
-
/* IPARMQ(ISPEC=16) Select structured matrix multiply. */
|
|
189
|
-
/* (See ISPEC=16 above for details.) */
|
|
190
|
-
/* Default: 3. */
|
|
191
|
-
|
|
192
|
-
/* ================================================================ */
|
|
193
|
-
inline int iparmq(int ispec, int ilo, int ihi) {
|
|
194
|
-
|
|
195
|
-
const int INMIN = 12, INWIN = 13, INIBL = 14, ISHFTS = 15, IACC22 = 16;
|
|
196
|
-
const int NMIN = 75, K22MIN = 14, KACMIN = 14, NIBBLE = 14, KNWSWP = 500;
|
|
197
|
-
|
|
198
|
-
int ns = 2, nh = ihi - ilo + 1;
|
|
199
|
-
|
|
200
|
-
if (ispec == ISHFTS || ispec == INWIN|| ispec == IACC22) {
|
|
201
|
-
|
|
202
|
-
/* ==== Set the number of simultaneous shifts ==== */
|
|
203
|
-
if (nh >= 30) ns = 4;
|
|
204
|
-
if (nh >= 60) ns = 10;
|
|
205
|
-
if (nh >= 150) ns = std::max(10, (int)(nh / std::round(std::log((float) (ihi - ilo + 1)) / log(2.f)))); /* Computing MAX */
|
|
206
|
-
if (nh >= 590) ns = 64;
|
|
207
|
-
if (nh >= 3000) ns = 128;
|
|
208
|
-
if (nh >= 6000) ns = 256;
|
|
209
|
-
ns = std::max(2,ns - ns % 2); /* Computing MAX */
|
|
210
|
-
}
|
|
211
|
-
|
|
212
|
-
if (ispec == INMIN) {
|
|
213
|
-
/* ===== Matrices of order smaller than NMIN get sent */
|
|
214
|
-
/* . to xLAHQR, the classic double shift algorithm. */
|
|
215
|
-
/* . This must be at least 11. ==== */
|
|
216
|
-
return NMIN;
|
|
217
|
-
|
|
218
|
-
} else if (ispec == INIBL) {
|
|
219
|
-
|
|
220
|
-
/* ==== INIBL: skip a multi-shift qr iteration and */
|
|
221
|
-
/* . whenever aggressive early deflation finds */
|
|
222
|
-
/* . at least (NIBBLE*(window size)/100) deflations. ==== */
|
|
223
|
-
|
|
224
|
-
return NIBBLE;
|
|
225
|
-
|
|
226
|
-
} else if (ispec == ISHFTS) {
|
|
227
|
-
|
|
228
|
-
/* ==== NSHFTS: The number of simultaneous shifts ===== */
|
|
229
|
-
return ns;
|
|
230
|
-
|
|
231
|
-
} else if (ispec == INWIN) {
|
|
232
|
-
|
|
233
|
-
/* ==== NW: deflation window size. ==== */
|
|
234
|
-
|
|
235
|
-
if (nh <= KNWSWP) return ns;
|
|
236
|
-
else return ns * 3 / 2;
|
|
237
|
-
|
|
238
|
-
} else if (ispec == 16) {
|
|
239
|
-
|
|
240
|
-
/* ==== IACC22: Whether to accumulate reflections */
|
|
241
|
-
/* . before updating the far-from-diagonal elements */
|
|
242
|
-
/* . and whether to use 2-by-2 block structure while */
|
|
243
|
-
/* . doing it. A small amount of work could be saved */
|
|
244
|
-
/* . by making this choice dependent also upon the */
|
|
245
|
-
/* . NH=IHI-ILO+1. */
|
|
246
|
-
|
|
247
|
-
if (ns >= KACMIN) return 1;
|
|
248
|
-
if (ns >= K22MIN) return 2;
|
|
249
|
-
|
|
250
|
-
}
|
|
251
|
-
|
|
252
|
-
return -1;
|
|
253
|
-
} /* iparmq_ */
|
|
254
|
-
|
|
255
|
-
|
|
256
|
-
|
|
257
|
-
|
|
258
|
-
/* Purpose */
|
|
259
|
-
/* ======= */
|
|
260
|
-
|
|
261
|
-
/* DGER performs the rank 1 operation */
|
|
262
|
-
|
|
263
|
-
/* A := alpha*x*y**T + A, */
|
|
264
|
-
|
|
265
|
-
/* where alpha is a scalar, x is an m element vector, y is an n element */
|
|
266
|
-
/* vector and A is an m by n matrix. */
|
|
267
|
-
|
|
268
|
-
/* Arguments */
|
|
269
|
-
/* ========== */
|
|
270
|
-
|
|
271
|
-
/* M - INTEGER. */
|
|
272
|
-
/* On entry, M specifies the number of rows of the matrix A. */
|
|
273
|
-
/* M must be at least zero. */
|
|
274
|
-
/* Unchanged on exit. */
|
|
275
|
-
|
|
276
|
-
/* N - INTEGER. */
|
|
277
|
-
/* On entry, N specifies the number of columns of the matrix A. */
|
|
278
|
-
/* N must be at least zero. */
|
|
279
|
-
/* Unchanged on exit. */
|
|
280
|
-
|
|
281
|
-
/* ALPHA - DOUBLE PRECISION. */
|
|
282
|
-
/* On entry, ALPHA specifies the scalar alpha. */
|
|
283
|
-
/* Unchanged on exit. */
|
|
284
|
-
|
|
285
|
-
/* X - DOUBLE PRECISION array of dimension at least */
|
|
286
|
-
/* ( 1 + ( m - 1 )*abs( INCX ) ). */
|
|
287
|
-
/* Before entry, the incremented array X must contain the m */
|
|
288
|
-
/* element vector x. */
|
|
289
|
-
/* Unchanged on exit. */
|
|
290
|
-
|
|
291
|
-
/* INCX - INTEGER. */
|
|
292
|
-
/* On entry, INCX specifies the increment for the elements of */
|
|
293
|
-
/* X. INCX must not be zero. */
|
|
294
|
-
/* Unchanged on exit. */
|
|
295
|
-
|
|
296
|
-
/* Y - DOUBLE PRECISION array of dimension at least */
|
|
297
|
-
/* ( 1 + ( n - 1 )*abs( INCY ) ). */
|
|
298
|
-
/* Before entry, the incremented array Y must contain the n */
|
|
299
|
-
/* element vector y. */
|
|
300
|
-
/* Unchanged on exit. */
|
|
301
|
-
|
|
302
|
-
/* INCY - INTEGER. */
|
|
303
|
-
/* On entry, INCY specifies the increment for the elements of */
|
|
304
|
-
/* Y. INCY must not be zero. */
|
|
305
|
-
/* Unchanged on exit. */
|
|
306
|
-
|
|
307
|
-
/* A - DOUBLE PRECISION array of DIMENSION ( LDA, n ). */
|
|
308
|
-
/* Before entry, the leading m by n part of the array A must */
|
|
309
|
-
/* contain the matrix of coefficients. On exit, A is */
|
|
310
|
-
/* overwritten by the updated matrix. */
|
|
311
|
-
|
|
312
|
-
/* LDA - INTEGER. */
|
|
313
|
-
/* On entry, LDA specifies the first dimension of A as declared */
|
|
314
|
-
/* in the calling (sub) program. LDA must be at least */
|
|
315
|
-
/* max( 1, m ). */
|
|
316
|
-
/* Unchanged on exit. */
|
|
317
|
-
|
|
318
|
-
/* Further Details */
|
|
319
|
-
/* =============== */
|
|
320
|
-
|
|
321
|
-
/* Level 2 Blas routine. */
|
|
322
|
-
|
|
323
|
-
/* -- Written on 22-October-1986. */
|
|
324
|
-
/* Jack Dongarra, Argonne National Lab. */
|
|
325
|
-
/* Jeremy Du Croz, Nag Central Office. */
|
|
326
|
-
/* Sven Hammarling, Nag Central Office. */
|
|
327
|
-
/* Richard Hanson, Sandia National Labs. */
|
|
328
|
-
|
|
329
|
-
/* ===================================================================== */
|
|
330
|
-
|
|
331
|
-
|
|
332
|
-
|
|
333
|
-
|
|
334
|
-
/* Purpose */
|
|
335
|
-
/* ======= */
|
|
336
|
-
|
|
337
|
-
/* interchanges two vectors. */
|
|
338
|
-
/* uses unrolled loops for increments equal one. */
|
|
339
|
-
|
|
340
|
-
/* Further Details */
|
|
341
|
-
/* =============== */
|
|
342
|
-
|
|
343
|
-
/* jack dongarra, linpack, 3/11/78. */
|
|
344
|
-
/* modified 12/3/93, array(1) declarations changed to array(*) */
|
|
345
|
-
|
|
346
|
-
/* ===================================================================== */
|
|
347
|
-
// Formerly dswap
|
|
348
|
-
|
|
349
|
-
|
|
350
|
-
|
|
351
|
-
|
|
352
|
-
/* -- LAPACK auxiliary routine (version 3.3.1) -- */
|
|
353
|
-
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
|
|
354
|
-
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
|
|
355
|
-
/* -- April 2011 -- */
|
|
356
|
-
|
|
357
|
-
/* .. Scalar Arguments .. */
|
|
358
|
-
/* .. */
|
|
359
|
-
|
|
360
|
-
/* Purpose */
|
|
361
|
-
/* ======= */
|
|
362
|
-
|
|
363
|
-
/* IEEECK is called from the ILAENV to verify that Infinity and */
|
|
364
|
-
/* possibly NaN arithmetic is safe (i.e. will not trap). */
|
|
365
|
-
// FIXME: Can we use std::numeric_limits::traps for this?
|
|
366
|
-
|
|
367
|
-
/* Arguments */
|
|
368
|
-
/* ========= */
|
|
369
|
-
|
|
370
|
-
/* ISPEC (input) INTEGER */
|
|
371
|
-
/* Specifies whether to test just for inifinity arithmetic */
|
|
372
|
-
/* or whether to test for infinity and NaN arithmetic. */
|
|
373
|
-
/* = 0: Verify infinity arithmetic only. */
|
|
374
|
-
/* = 1: Verify infinity and NaN arithmetic. */
|
|
375
|
-
|
|
376
|
-
/* ZERO (input) REAL */
|
|
377
|
-
/* Must contain the value 0.0 */
|
|
378
|
-
/* This is passed to prevent the compiler from optimizing */
|
|
379
|
-
/* away this code. */
|
|
380
|
-
|
|
381
|
-
/* ONE (input) REAL */
|
|
382
|
-
/* Must contain the value 1.0 */
|
|
383
|
-
/* This is passed to prevent the compiler from optimizing */
|
|
384
|
-
/* away this code. */
|
|
385
|
-
|
|
386
|
-
/* RETURN VALUE: INTEGER */
|
|
387
|
-
/* = 0: Arithmetic failed to produce the correct answers */
|
|
388
|
-
/* = 1: Arithmetic produced the correct answers */
|
|
389
|
-
|
|
390
|
-
/* ===================================================================== */
|
|
391
|
-
|
|
392
|
-
/*
|
|
393
|
-
* Note from John: This seems totally unnecessary in modern C++.
|
|
394
|
-
* FIXME: Remove this after testing that on modern systems this always returns 1.
|
|
395
|
-
*/
|
|
396
|
-
|
|
397
|
-
inline int ieeeck(bool ispec) {
|
|
398
|
-
|
|
399
|
-
float posinf = 1.0 / 0.0;
|
|
400
|
-
if (posinf <= 1.0) return 0;
|
|
401
|
-
|
|
402
|
-
float neginf = -1.0 / 0.0;
|
|
403
|
-
if (neginf >= 0.0) return 0;
|
|
404
|
-
|
|
405
|
-
float negzro = 1.0 / (neginf + 1.0);
|
|
406
|
-
if (negzro != 0.0) return 0;
|
|
407
|
-
|
|
408
|
-
neginf = 1.0 / negzro;
|
|
409
|
-
if (neginf >= 0.0) return 0;
|
|
410
|
-
|
|
411
|
-
float newzro = negzro + 0.0;
|
|
412
|
-
if (newzro != 0.0) return 0;
|
|
413
|
-
|
|
414
|
-
posinf = 1.0 / newzro;
|
|
415
|
-
if (posinf <= 1.0) return 0;
|
|
416
|
-
|
|
417
|
-
neginf *= posinf;
|
|
418
|
-
if (neginf >= 0.0) return 0;
|
|
419
|
-
|
|
420
|
-
posinf *= posinf;
|
|
421
|
-
if (posinf <= 1.0) return 0;
|
|
422
|
-
|
|
423
|
-
|
|
424
|
-
/* Return if we were only asked to check infinity arithmetic */
|
|
425
|
-
|
|
426
|
-
if (!ispec) return 1;
|
|
427
|
-
|
|
428
|
-
float nan1 = posinf + neginf;
|
|
429
|
-
float nan2 = posinf / neginf;
|
|
430
|
-
float nan3 = posinf / posinf;
|
|
431
|
-
float nan4 = posinf * 0.0;
|
|
432
|
-
float nan5 = neginf * negzro;
|
|
433
|
-
float nan6 = nan5 * 0.0;
|
|
434
|
-
|
|
435
|
-
if (nan1 == nan1) return 0;
|
|
436
|
-
if (nan2 == nan2) return 0;
|
|
437
|
-
if (nan3 == nan3) return 0;
|
|
438
|
-
if (nan4 == nan4) return 0;
|
|
439
|
-
if (nan5 == nan5) return 0;
|
|
440
|
-
if (nan6 == nan6) return 0;
|
|
441
|
-
|
|
442
|
-
return 1;
|
|
443
|
-
} /* ieeeck_ */
|
|
444
|
-
|
|
445
|
-
|
|
446
|
-
|
|
447
|
-
|
|
448
|
-
inline int ilaenv_block_size(int n2, int n4, const std::string& c2, const std::string& c3, const std::string& c4, bool sname, bool cname) {
|
|
449
|
-
if (c2 == "GE") { //(s_cmp(c2, "GE", (size_t)2, (size_t)2) == 0) {
|
|
450
|
-
if (c3 == "TRF") { //if (s_cmp(c3, "TRF", (size_t)3, (size_t)3) == 0) {
|
|
451
|
-
if (sname) return 64;
|
|
452
|
-
else return 64;
|
|
453
|
-
} else if (c3 == "QRF" || c3 == "RQF" || c3 == "LQF" || c3 == "QLF") { //(s_cmp(c3, "QRF", (size_t)3, (size_t)3) == 0 || s_cmp(c3, "RQF", (size_t)3, (size_t)3) == 0 || s_cmp(c3, "LQF", (size_t) 3, (size_t)3) == 0 || s_cmp(c3, "QLF", (size_t)3, (size_t)3) == 0) {
|
|
454
|
-
if (sname) return 32;
|
|
455
|
-
else return 32;
|
|
456
|
-
} else if (c3 == "HRD") {
|
|
457
|
-
if (sname) return 32;
|
|
458
|
-
else return 32;
|
|
459
|
-
} else if (c3 == "BRD") {
|
|
460
|
-
if (sname) return 32;
|
|
461
|
-
else return 32;
|
|
462
|
-
} else if (c3 == "TRI") {
|
|
463
|
-
if (sname) return 64;
|
|
464
|
-
else return 64;
|
|
465
|
-
}
|
|
466
|
-
} else if (c2 == "PO") {
|
|
467
|
-
if (c3 == "TRF") {
|
|
468
|
-
if (sname) return 64;
|
|
469
|
-
else return 64;
|
|
470
|
-
}
|
|
471
|
-
} else if (c2 == "SY") {
|
|
472
|
-
if (c3 == "TRF") {
|
|
473
|
-
if (sname) return 64;
|
|
474
|
-
else return 64;
|
|
475
|
-
} else if (sname && c3 == "TRD") {
|
|
476
|
-
return 32;
|
|
477
|
-
} else if (sname && c3 == "GST") {
|
|
478
|
-
return 64;
|
|
479
|
-
}
|
|
480
|
-
} else if (cname && c2 == "HE") {
|
|
481
|
-
if (c3 == "TRF") return 64;
|
|
482
|
-
else if (c3 == "TRD") return 32;
|
|
483
|
-
else if (c3 == "GST") return 64;
|
|
484
|
-
} else if (sname && c2 == "OR") {
|
|
485
|
-
if (c3.at(0) == 'G') {
|
|
486
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") return 32;
|
|
487
|
-
} else if (c3.at(0) == 'M') {
|
|
488
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") return 32;
|
|
489
|
-
}
|
|
490
|
-
} else if (cname && c2 == "UN") {
|
|
491
|
-
if (c3.at(0) == 'G') {
|
|
492
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") return 32;
|
|
493
|
-
} else if (c3.at(0) == 'M') {
|
|
494
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") return 32;
|
|
495
|
-
}
|
|
496
|
-
} else if (c2 == "GB") {
|
|
497
|
-
if (c3 == "TRF") {
|
|
498
|
-
if (sname) {
|
|
499
|
-
if (n4 <= 64) return 1;
|
|
500
|
-
else return 32;
|
|
501
|
-
} else {
|
|
502
|
-
if (n4 <= 64) return 1;
|
|
503
|
-
else return 32;
|
|
504
|
-
}
|
|
505
|
-
}
|
|
506
|
-
} else if (c2 == "PB") {
|
|
507
|
-
if (c3 == "TRF") {
|
|
508
|
-
if (sname) {
|
|
509
|
-
if (n2 <= 64) return 1;
|
|
510
|
-
else return 32;
|
|
511
|
-
} else {
|
|
512
|
-
if (n2 <= 64) return 1;
|
|
513
|
-
else return 32;
|
|
514
|
-
}
|
|
515
|
-
}
|
|
516
|
-
} else if (c2 == "TR") {
|
|
517
|
-
if (c3 == "TRI") {
|
|
518
|
-
if (sname) return 64;
|
|
519
|
-
else return 64;
|
|
520
|
-
}
|
|
521
|
-
} else if (c2 == "LA") {
|
|
522
|
-
if (c3 == "UUM") {
|
|
523
|
-
if (sname) return 64;
|
|
524
|
-
else return 64;
|
|
525
|
-
}
|
|
526
|
-
} else if (sname && c2 == "ST") {
|
|
527
|
-
if (c3 == "EBZ") return 1;
|
|
528
|
-
}
|
|
529
|
-
return 1;
|
|
530
|
-
}
|
|
531
|
-
|
|
532
|
-
|
|
533
|
-
inline int ilaenv_min_block_size(const std::string& c2, const std::string& c3, const std::string& c4, bool sname, bool cname) {
|
|
534
|
-
|
|
535
|
-
if (c2 == "GE") {
|
|
536
|
-
if (c3 == "QRF" || c3 == "RQF" || c3 == "LQF" || c3 == "QLF") {
|
|
537
|
-
if (sname) {
|
|
538
|
-
return 2;
|
|
539
|
-
} else {
|
|
540
|
-
return 2;
|
|
541
|
-
}
|
|
542
|
-
} else if (c3 == "HRD") {
|
|
543
|
-
if (sname) {
|
|
544
|
-
return 2;
|
|
545
|
-
} else {
|
|
546
|
-
return 2;
|
|
547
|
-
}
|
|
548
|
-
} else if (c3 == "BRD") {
|
|
549
|
-
if (sname) {
|
|
550
|
-
return 2;
|
|
551
|
-
} else {
|
|
552
|
-
return 2;
|
|
553
|
-
}
|
|
554
|
-
} else if (c3 == "TRI") {
|
|
555
|
-
if (sname) {
|
|
556
|
-
return 2;
|
|
557
|
-
} else {
|
|
558
|
-
return 2;
|
|
559
|
-
}
|
|
560
|
-
}
|
|
561
|
-
} else if (c2 == "SY") {
|
|
562
|
-
if (c3 == "TRF") {
|
|
563
|
-
if (sname) {
|
|
564
|
-
return 8;
|
|
565
|
-
} else {
|
|
566
|
-
return 8;
|
|
567
|
-
}
|
|
568
|
-
} else if (sname && c3 == "TRD") {
|
|
569
|
-
return 2;
|
|
570
|
-
}
|
|
571
|
-
} else if (cname && c2 == "HE") {
|
|
572
|
-
if (c3 == "TRD") {
|
|
573
|
-
return 2;
|
|
574
|
-
}
|
|
575
|
-
} else if (sname && c2 == "OR") {
|
|
576
|
-
if (c3.at(0) == 'G') {
|
|
577
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
578
|
-
return 2;
|
|
579
|
-
}
|
|
580
|
-
} else if (c3.at(0) == 'M') {
|
|
581
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
582
|
-
return 2;
|
|
583
|
-
}
|
|
584
|
-
}
|
|
585
|
-
} else if (cname && c2 == "UN") {
|
|
586
|
-
if (c3.at(0) == 'G') {
|
|
587
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
588
|
-
return 2;
|
|
589
|
-
}
|
|
590
|
-
} else if (c3.at(0) == 'M') {
|
|
591
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
592
|
-
return 2;
|
|
593
|
-
}
|
|
594
|
-
}
|
|
595
|
-
}
|
|
596
|
-
return 2;
|
|
597
|
-
}
|
|
598
|
-
|
|
599
|
-
|
|
600
|
-
inline int ilaenv_crossover_point(const std::string& c2, const std::string& c3, const std::string& c4, bool sname, bool cname) {
|
|
601
|
-
if (c2 == "GE") {
|
|
602
|
-
if (c3 == "QRF" || c3 == "RQF" || c3 == "LQF" || c3 == "QLF") {
|
|
603
|
-
if (sname) {
|
|
604
|
-
return 128;
|
|
605
|
-
} else {
|
|
606
|
-
return 128;
|
|
607
|
-
}
|
|
608
|
-
} else if (c3 == "HRD") {
|
|
609
|
-
if (sname) {
|
|
610
|
-
return 128;
|
|
611
|
-
} else {
|
|
612
|
-
return 128;
|
|
613
|
-
}
|
|
614
|
-
} else if (c3 == "BRD") {
|
|
615
|
-
if (sname) {
|
|
616
|
-
return 128;
|
|
617
|
-
} else {
|
|
618
|
-
return 128;
|
|
619
|
-
}
|
|
620
|
-
}
|
|
621
|
-
} else if (c2 == "SY") {
|
|
622
|
-
if (sname && c3 == "TRD") {
|
|
623
|
-
return 32;
|
|
624
|
-
}
|
|
625
|
-
} else if (cname && c2 == "HE") {
|
|
626
|
-
if (c3 == "TRD") {
|
|
627
|
-
return 32;
|
|
628
|
-
}
|
|
629
|
-
} else if (sname && c2 == "OR") {
|
|
630
|
-
if (c3.at(0) == 'G') {
|
|
631
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
632
|
-
return 128;
|
|
633
|
-
}
|
|
634
|
-
}
|
|
635
|
-
} else if (cname && c2 == "UN") {
|
|
636
|
-
if (c3.at(0) == 'G') {
|
|
637
|
-
if (c4 == "QR" || c4 == "RQ" || c4 == "LQ" || c4 == "QL" || c4 == "HR" || c4 == "TR" || c4 == "BR") {
|
|
638
|
-
return 128;
|
|
639
|
-
}
|
|
640
|
-
}
|
|
641
|
-
}
|
|
642
|
-
return 0;
|
|
643
|
-
}
|
|
644
|
-
|
|
645
|
-
|
|
646
|
-
/* -- LAPACK auxiliary routine (version 3.2.1) -- */
|
|
647
|
-
|
|
648
|
-
/* -- April 2009 -- */
|
|
649
|
-
|
|
650
|
-
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
|
|
651
|
-
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
|
|
652
|
-
|
|
653
|
-
/* .. Scalar Arguments .. */
|
|
654
|
-
/* .. */
|
|
655
|
-
|
|
656
|
-
/* Purpose */
|
|
657
|
-
/* ======= */
|
|
658
|
-
|
|
659
|
-
/* ILAENV is called from the LAPACK routines to choose problem-dependent */
|
|
660
|
-
/* parameters for the local environment. See ISPEC for a description of */
|
|
661
|
-
/* the parameters. */
|
|
662
|
-
|
|
663
|
-
/* ILAENV returns an INTEGER */
|
|
664
|
-
/* if ILAENV >= 0: ILAENV returns the value of the parameter specified by ISPEC */
|
|
665
|
-
/* if ILAENV < 0: if ILAENV = -k, the k-th argument had an illegal value. */
|
|
666
|
-
|
|
667
|
-
/* This version provides a set of parameters which should give good, */
|
|
668
|
-
/* but not optimal, performance on many of the currently available */
|
|
669
|
-
/* computers. Users are encouraged to modify this subroutine to set */
|
|
670
|
-
/* the tuning parameters for their particular machine using the option */
|
|
671
|
-
/* and problem size information in the arguments. */
|
|
672
|
-
|
|
673
|
-
/* This routine will not function correctly if it is converted to all */
|
|
674
|
-
/* lower case. Converting it to all upper case is allowed. */
|
|
675
|
-
|
|
676
|
-
/* Arguments */
|
|
677
|
-
/* ========= */
|
|
678
|
-
|
|
679
|
-
/* ISPEC (input) INTEGER */
|
|
680
|
-
/* Specifies the parameter to be returned as the value of */
|
|
681
|
-
/* ILAENV. */
|
|
682
|
-
/* = 1: the optimal blocksize; if this value is 1, an unblocked */
|
|
683
|
-
/* algorithm will give the best performance. */
|
|
684
|
-
/* = 2: the minimum block size for which the block routine */
|
|
685
|
-
/* should be used; if the usable block size is less than */
|
|
686
|
-
/* this value, an unblocked routine should be used. */
|
|
687
|
-
/* = 3: the crossover point (in a block routine, for N less */
|
|
688
|
-
/* than this value, an unblocked routine should be used) */
|
|
689
|
-
/* = 4: the number of shifts, used in the nonsymmetric */
|
|
690
|
-
/* eigenvalue routines (DEPRECATED) */
|
|
691
|
-
/* = 5: the minimum column dimension for blocking to be used; */
|
|
692
|
-
/* rectangular blocks must have dimension at least k by m, */
|
|
693
|
-
/* where k is given by ILAENV(2,...) and m by ILAENV(5,...) */
|
|
694
|
-
/* = 6: the crossover point for the SVD (when reducing an m by n */
|
|
695
|
-
/* matrix to bidiagonal form, if max(m,n)/min(m,n) exceeds */
|
|
696
|
-
/* this value, a QR factorization is used first to reduce */
|
|
697
|
-
/* the matrix to a triangular form.) */
|
|
698
|
-
/* = 7: the number of processors */
|
|
699
|
-
/* = 8: the crossover point for the multishift QR method */
|
|
700
|
-
/* for nonsymmetric eigenvalue problems (DEPRECATED) */
|
|
701
|
-
/* = 9: maximum size of the subproblems at the bottom of the */
|
|
702
|
-
/* computation tree in the divide-and-conquer algorithm */
|
|
703
|
-
/* (used by xGELSD and xGESDD) */
|
|
704
|
-
/* =10: ieee NaN arithmetic can be trusted not to trap */
|
|
705
|
-
/* =11: infinity arithmetic can be trusted not to trap */
|
|
706
|
-
/* 12 <= ISPEC <= 16: */
|
|
707
|
-
/* xHSEQR or one of its subroutines, */
|
|
708
|
-
/* see IPARMQ for detailed explanation */
|
|
709
|
-
|
|
710
|
-
/* NAME (input) CHARACTER*(*) */
|
|
711
|
-
/* The name of the calling subroutine, in either upper case or */
|
|
712
|
-
/* lower case. */
|
|
713
|
-
|
|
714
|
-
/* OPTS (input) CHARACTER*(*) */
|
|
715
|
-
/* The character options to the subroutine NAME, concatenated */
|
|
716
|
-
/* into a single character string. For example, UPLO = 'U', */
|
|
717
|
-
/* TRANS = 'T', and DIAG = 'N' for a triangular routine would */
|
|
718
|
-
/* be specified as OPTS = 'UTN'. */
|
|
719
|
-
|
|
720
|
-
/* N1 (input) INTEGER */
|
|
721
|
-
/* N2 (input) INTEGER */
|
|
722
|
-
/* N3 (input) INTEGER */
|
|
723
|
-
/* N4 (input) INTEGER */
|
|
724
|
-
/* Problem dimensions for the subroutine NAME; these may not all */
|
|
725
|
-
/* be required. */
|
|
726
|
-
|
|
727
|
-
/* Further Details */
|
|
728
|
-
/* =============== */
|
|
729
|
-
|
|
730
|
-
/* The following conventions have been used when calling ILAENV from the */
|
|
731
|
-
/* LAPACK routines: */
|
|
732
|
-
/* 1) OPTS is a concatenation of all of the character options to */
|
|
733
|
-
/* subroutine NAME, in the same order that they appear in the */
|
|
734
|
-
/* argument list for NAME, even if they are not used in determining */
|
|
735
|
-
/* the value of the parameter specified by ISPEC. */
|
|
736
|
-
/* 2) The problem dimensions N1, N2, N3, N4 are specified in the order */
|
|
737
|
-
/* that they appear in the argument list for NAME. N1 is used */
|
|
738
|
-
/* first, N2 second, and so on, and unused problem dimensions are */
|
|
739
|
-
/* passed a value of -1. */
|
|
740
|
-
/* 3) The parameter value returned by ILAENV is checked for validity in */
|
|
741
|
-
/* the calling subroutine. For example, ILAENV is used to retrieve */
|
|
742
|
-
/* the optimal blocksize for STRTRI as follows: */
|
|
743
|
-
|
|
744
|
-
/* NB = ILAENV( 1, 'STRTRI', UPLO // DIAG, N, -1, -1, -1 ) */
|
|
745
|
-
/* IF( NB.LE.1 ) NB = MAX( 1, N ) */
|
|
746
|
-
|
|
747
|
-
/* ===================================================================== */
|
|
748
|
-
inline int ilaenv(int ispec, const std::string& name, int n1, int n2, int n3, int n4) {
|
|
749
|
-
|
|
750
|
-
if (ispec < 1 || ispec > 3) {
|
|
751
|
-
switch (ispec) {
|
|
752
|
-
case 4: return 6; /* ISPEC = 4: number of shifts (used by xHSEQR) */
|
|
753
|
-
case 5: return 2; /* ISPEC = 5: minimum column dimension (not used) */
|
|
754
|
-
case 6: /* ISPEC = 6: crossover point for SVD (used by xGELSS and xGESVD) */
|
|
755
|
-
return (int) ((float)std::min(n1, n2) * 1.6f);
|
|
756
|
-
case 7: return 1; /* ISPEC = 7: number of processors (not used) */
|
|
757
|
-
case 8: return 50; /* ISPEC = 8: crossover point for multishift (used by xHSEQR) */
|
|
758
|
-
case 9: return 25; /* ISPEC = 9: maximum size of the subproblems at the bottom of the */
|
|
759
|
-
/* computation tree in the divide-and-conquer algorithm */
|
|
760
|
-
/* (used by xGELSD and xGESDD) */
|
|
761
|
-
case 10: /* ISPEC = 10: ieee NaN arithmetic can be trusted not to trap */
|
|
762
|
-
return ieeeck(1);
|
|
763
|
-
|
|
764
|
-
case 11: /* ISPEC = 11: infinity arithmetic can be trusted not to trap */
|
|
765
|
-
return ieeeck(0);
|
|
766
|
-
|
|
767
|
-
default:
|
|
768
|
-
if (ispec >= 12 && ispec <= 16) { /* 12 <= ISPEC <= 16: xHSEQR or one of its subroutines. */
|
|
769
|
-
return iparmq(ispec, n2, n3);
|
|
770
|
-
} else {
|
|
771
|
-
return -1; /* Invalid value for ISPEC */
|
|
772
|
-
}
|
|
773
|
-
}
|
|
774
|
-
}
|
|
775
|
-
|
|
776
|
-
|
|
777
|
-
/* Convert NAME to upper case if the first character is lower case. */
|
|
778
|
-
|
|
779
|
-
std::string subnam(name);
|
|
780
|
-
std::transform(subnam.begin(), subnam.end(), subnam.begin(), ::toupper);
|
|
781
|
-
std::string c1(subnam);
|
|
782
|
-
|
|
783
|
-
bool sname = c1.at(0) == 'S' || c1.at(0) == 'D',
|
|
784
|
-
cname = c1.at(0) == 'C' || c1.at(0) == 'Z';
|
|
785
|
-
|
|
786
|
-
if (! (cname || sname)) return 1;
|
|
787
|
-
|
|
788
|
-
std::string c2(subnam.substr(1, 2)),
|
|
789
|
-
c3(subnam.substr(3, 3)),
|
|
790
|
-
c4(c3.substr(1, 2));
|
|
791
|
-
|
|
792
|
-
if (ispec == 2) return ilaenv_min_block_size(c2, c3, c4, sname, cname);
|
|
793
|
-
if (ispec == 3) return ilaenv_crossover_point(c2, c3, c4, sname, cname);
|
|
794
|
-
return ilaenv_block_size(n2, n4, c2, c3, c4, sname, cname);
|
|
795
|
-
|
|
796
|
-
} /* ilaenv_ */
|
|
797
|
-
|
|
798
|
-
|
|
799
|
-
|
|
800
|
-
template <typename DType>
|
|
801
|
-
inline int lsame(char const *ca, char const *cb, int ca_len, int cb_len)
|
|
802
|
-
{
|
|
803
|
-
int ret_val;
|
|
804
|
-
|
|
805
|
-
int static inta, intb, zcode;
|
|
806
|
-
/* -- LAPACK auxiliary routine (version 3.2) -- */
|
|
807
|
-
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
|
|
808
|
-
/* November 2006 */
|
|
809
|
-
|
|
810
|
-
/* .. Scalar Arguments .. */
|
|
811
|
-
/* .. */
|
|
812
|
-
|
|
813
|
-
/* Purpose */
|
|
814
|
-
/* ======= */
|
|
815
|
-
|
|
816
|
-
/* LSAME returns .TRUE. if CA is the same letter as CB regardless of */
|
|
817
|
-
/* case. */
|
|
818
|
-
|
|
819
|
-
/* Arguments */
|
|
820
|
-
/* ========= */
|
|
821
|
-
|
|
822
|
-
/* CA (input) CHARACTER*1 */
|
|
823
|
-
/* CB (input) CHARACTER*1 */
|
|
824
|
-
/* CA and CB specify the single characters to be compared. */
|
|
825
|
-
|
|
826
|
-
/* ===================================================================== */
|
|
827
|
-
inta = *(unsigned char *)ca;
|
|
828
|
-
intb = *(unsigned char *)cb;
|
|
829
|
-
ret_val = inta == intb;
|
|
830
|
-
if (ret_val) {
|
|
831
|
-
return ret_val;
|
|
832
|
-
}
|
|
833
|
-
|
|
834
|
-
zcode = (int)"Z";
|
|
835
|
-
|
|
836
|
-
if (zcode == 90 || zcode == 122 ) {
|
|
837
|
-
/* ASCII is assumed - ZCODE is the ASCII code of either lower or */
|
|
838
|
-
/* upper case 'Z'. */
|
|
839
|
-
if (inta >=97 && inta <= 122) {
|
|
840
|
-
inta += -32;
|
|
841
|
-
}
|
|
842
|
-
if (intb >= 97 && intb <= 122) {
|
|
843
|
-
intb += -32;
|
|
844
|
-
}
|
|
845
|
-
} else if (zcode == 233 || zcode == 169) {
|
|
846
|
-
/* EBCDIC is assumed - ZCODE is the EBCDIC code of either lower or */
|
|
847
|
-
/* upper case 'Z'. */
|
|
848
|
-
if (inta >= 129 && inta <= 137 || inta >= 145 && inta <= 153 || inta >= 162 && inta <= 169) {
|
|
849
|
-
inta += 64;
|
|
850
|
-
}
|
|
851
|
-
if (intb >= 129 && intb <= 137 || intb >= 145 && intb <= 153 || intb >= 162 && intb <= 169) {
|
|
852
|
-
intb += 64;
|
|
853
|
-
}
|
|
854
|
-
} else if (zcode = 218 || zcode == 250) {
|
|
855
|
-
/* ASCII is assumed, on Prime machines - ZCODE is the ASCII code */
|
|
856
|
-
/* plus 128 of either lower or upper case 'Z'. */
|
|
857
|
-
if (inta >= 225 && inta <= 250) {
|
|
858
|
-
inta += -32;
|
|
859
|
-
}
|
|
860
|
-
if (intb >= 225 && intb <= 250) {
|
|
861
|
-
intb += -32;
|
|
862
|
-
}
|
|
863
|
-
}
|
|
864
|
-
ret_val = inta == intb;
|
|
865
|
-
|
|
866
|
-
return ret_val;
|
|
867
|
-
}
|
|
868
|
-
|
|
869
|
-
|
|
870
|
-
|
|
871
|
-
|
|
872
|
-
} // end namespace lapack
|
|
873
|
-
|
|
874
|
-
|
|
875
|
-
|
|
876
|
-
|
|
877
|
-
}}
|
|
878
|
-
|
|
879
|
-
#endif
|