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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
|
// Copyright (c) 2017-2023, University of Tennessee. All rights reserved.
// SPDX-License-Identifier: BSD-3-Clause
// This program is free software: you can redistribute it and/or modify it under
// the terms of the BSD 3-Clause license. See the accompanying LICENSE file.
#include "test.hh"
#include "lapack.hh"
#include "lapack/flops.hh"
#include "print_matrix.hh"
#include "error.hh"
#include "lapacke_wrappers.hh"
#include <vector>
#if LAPACK_VERSION >= 30700 // >= 3.7.0
// -----------------------------------------------------------------------------
// tplqt is blocked version
// tplqt2 is non-blocked version
// todo: merge test_tplqt2.cc and test_tplqt.cc
template< typename scalar_t >
void test_tplqt_work( Params& params, bool run )
{
using real_t = blas::real_type< scalar_t >;
// get & mark input values
int64_t m = params.dim.m();
int64_t n = params.dim.n();
int64_t L = params.l();
int64_t mb = params.nb(); // rename nb => mb
int64_t align = params.align();
int64_t verbose = params.verbose();
real_t eps = std::numeric_limits< real_t >::epsilon();
real_t tol = params.tol() * eps;
// mark non-standard output values
params.ref_time();
params.ref_gflops();
params.gflops();
params.msg();
if (! run)
return;
// skip invalid sizes
if (blas::min(m, n) < L || m < mb || mb < 1) {
params.msg() = "skipping: requires min(m, n) >= L >= 0 and m >= mb >= 1";
return;
}
// ---------- setup
int64_t lda = roundup( blas::max( 1, m ), align );
int64_t ldb = roundup( blas::max( 1, m ), align );
int64_t ldt = roundup( mb, align );
size_t size_A = (size_t) lda * m; // m-by-m
size_t size_B = (size_t) ldb * n; // m-by-n
size_t size_T = (size_t) ldt * m; // mb-by-m
std::vector< scalar_t > A_tst( size_A );
std::vector< scalar_t > A_ref( size_A );
std::vector< scalar_t > B_tst( size_B );
std::vector< scalar_t > B_ref( size_B );
std::vector< scalar_t > T_tst( size_T );
std::vector< scalar_t > T_ref( size_T );
int64_t idist = 1;
int64_t iseed[4] = { 0, 1, 2, 3 };
lapack::larnv( idist, iseed, A_tst.size(), &A_tst[0] );
lapack::larnv( idist, iseed, B_tst.size(), &B_tst[0] );
A_ref = A_tst;
B_ref = B_tst;
if (verbose >= 1) {
printf( "\n"
"A m-by-m=%5lld, lda=%5lld triangle\n"
"B m=%5lld, n=%5lld, ldb=%5lld, rows m-L=%5lld rect, L=%5lld trapezoid\n",
llong( m ), llong( lda ),
llong( m ), llong( n ), llong( ldb ), llong( (m-L) ), llong( L ) );
}
if (verbose >= 2) {
printf( "A = " ); print_matrix( m, m, &A_tst[0], lda );
printf( "B1 = " ); print_matrix( m, n-L, &B_tst[0], ldb );
printf( "B2 = " ); print_matrix( m, L, &B_tst[(n-L)*ldb], ldb );
printf( "\n" );
}
// ---------- run test
testsweeper::flush_cache( params.cache() );
double time = testsweeper::get_wtime();
int64_t info_tst = lapack::tplqt(
m, n, L, mb, &A_tst[0], lda, &B_tst[0], ldb, &T_tst[0], ldt );
time = testsweeper::get_wtime() - time;
if (info_tst != 0) {
fprintf( stderr, "lapack::tplqt returned error %lld\n", llong( info_tst ) );
}
params.time() = time;
double gflop = lapack::Gflop< scalar_t >::gelqf( m, n ); // under-estimate
params.gflops() = gflop / time;
if (verbose >= 2) {
printf( "Ahat = " ); print_matrix( m, m, &A_tst[0], lda );
printf( "Bhat1 = " ); print_matrix( m, n-L, &B_tst[0], ldb );
printf( "Bhat2 = " ); print_matrix( m, L, &B_tst[(n-L)*ldb], ldb );
printf( "T = " ); print_matrix( mb, m, &T_tst[0], ldt );
printf( "\n" );
}
if (params.check() == 'y') {
// ---------- check error
// Relative backwards error = ||A - LQ|| / (m * ||A||).
// todo: Orthogonality check = ||I - Q^H Q|| / m.
// LA = lower( A ); LB = zeros( m, n );
std::vector< scalar_t > LA( size_A, 0.0 );
std::vector< scalar_t > LB( size_B, 0.0 );
lapack::lacpy( lapack::MatrixType::Lower, m, m, &A_tst[0], lda, &LA[0], lda );
if (verbose >= 2) {
printf( "LA = " ); print_matrix( m, m, &LA[0], lda );
printf( "LB1 = " ); print_matrix( m, n-L, &LB[0], ldb );
printf( "LB2 = " ); print_matrix( m, L, &LB[(n-L)*ldb], ldb );
printf( "\n" );
}
// zero out upper triangle of A (above col 1) and B (above col n-L+1).
lapack::laset( lapack::MatrixType::Upper, m-1, m-1, 0.0, 0.0, &A_ref[1*lda], lda );
lapack::laset( lapack::MatrixType::Upper, L-1, L-1, 0.0, 0.0, &B_ref[(n-L+1)*ldb], ldb );
if (verbose >= 2) {
printf( "Aref = " ); print_matrix( m, m, &A_ref[0], lda );
printf( "Bref1 = " ); print_matrix( m, n-L, &B_ref[0], ldb );
printf( "Bref2 = " ); print_matrix( m, L, &B_ref[(n-L)*ldb], ldb );
printf( "\n" );
}
// [ LA LB ] = [ LA LB ] * Q
info_tst = lapack::tpmlqt(
lapack::Side::Right, lapack::Op::NoTrans, m, n, m, L, mb,
&B_tst[0], ldb, &T_tst[0], ldt, &LA[0], lda, &LB[0], ldb );
assert( info_tst == 0 );
real_t Anorm = std::max(
lapack::lange( lapack::Norm::One, m, m, &A_ref[0], lda ),
lapack::lange( lapack::Norm::One, m, n, &B_ref[0], ldb ) );
// [ A B ] - ([ LA LB ] * Q)
blas::axpy( size_A, -1.0, &LA[0], 1, &A_ref[0], 1 );
blas::axpy( size_B, -1.0, &LB[0], 1, &B_ref[0], 1 );
real_t error = std::max(
lapack::lange( lapack::Norm::One, m, m, &A_ref[0], lda ),
lapack::lange( lapack::Norm::One, m, n, &B_ref[0], ldb ) );
error /= ((m + n)*Anorm);
params.error() = error;
params.okay() = (error < tol);
if (verbose >= 2) {
printf( "A_QR0 = " ); print_matrix( m, m, &A_ref[0], lda );
printf( "A_QR1 = " ); print_matrix( m, n-L, &B_ref[0], ldb );
printf( "A_QR2 = " ); print_matrix( m, L, &B_ref[m-L], ldb );
printf( "\n" );
}
}
if (params.ref() == 'y') {
// ---------- run reference
// Reset A, B in case check wiped them out.
lapack::larnv( idist, iseed, A_ref.size(), &A_ref[0] );
lapack::larnv( idist, iseed, B_ref.size(), &B_ref[0] );
testsweeper::flush_cache( params.cache() );
time = testsweeper::get_wtime();
int64_t info_ref = LAPACKE_tplqt(
m, n, L, mb, &A_ref[0], lda, &B_ref[0], ldb, &T_ref[0], ldt );
time = testsweeper::get_wtime() - time;
if (info_ref != 0) {
fprintf( stderr, "LAPACKE_tplqt returned error %lld\n", llong( info_ref ) );
}
params.ref_time() = time;
params.ref_gflops() = gflop / time;
}
}
// -----------------------------------------------------------------------------
void test_tplqt( Params& params, bool run )
{
switch (params.datatype()) {
case testsweeper::DataType::Single:
test_tplqt_work< float >( params, run );
break;
case testsweeper::DataType::Double:
test_tplqt_work< double >( params, run );
break;
case testsweeper::DataType::SingleComplex:
test_tplqt_work< std::complex<float> >( params, run );
break;
case testsweeper::DataType::DoubleComplex:
test_tplqt_work< std::complex<double> >( params, run );
break;
default:
throw std::runtime_error( "unknown datatype" );
break;
}
}
#else
// -----------------------------------------------------------------------------
void test_tplqt( Params& params, bool run )
{
fprintf( stderr, "tplqt requires LAPACK >= 3.7.0\n\n" );
exit(0);
}
#endif // LAPACK >= 3.7.0
|
