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
220
221
222
|
// 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/device.hh"
#include "lapack/flops.hh"
#include "print_matrix.hh"
#include "error.hh"
#include "lapacke_wrappers.hh"
#include <vector>
// -----------------------------------------------------------------------------
template< typename scalar_t >
void test_geqrf_device_work( Params& params, bool run )
{
using lapack::device_info_int;
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 device = params.device();
int64_t align = params.align();
int64_t verbose = params.verbose();
params.matrix.mark();
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.ortho();
if (! run)
return;
if (blas::get_device_count() == 0) {
params.msg() = "skipping: no GPU devices or no GPU support";
return;
}
// ---------- setup
int64_t lda = roundup( blas::max( 1, m ), align );
size_t size_A = (size_t) lda * n;
size_t size_tau = (size_t) blas::min( m, n );
int64_t minmn = blas::min( m, n );
std::vector< scalar_t > A_tst( size_A );
std::vector< scalar_t > A_ref( size_A );
std::vector< scalar_t > tau_tst( size_tau );
std::vector< scalar_t > tau_ref( size_tau );
lapack::generate_matrix( params.matrix, m, n, &A_tst[0], lda );
A_ref = A_tst;
// Allocate and copy to GPU.
lapack::Queue queue( device );
scalar_t* dA_tst = blas::device_malloc< scalar_t >( size_A, queue );
scalar_t* d_tau = blas::device_malloc< scalar_t >( size_tau, queue );
device_info_int* d_info = blas::device_malloc< device_info_int >( 1, queue );
blas::device_copy_matrix( m, n, A_tst.data(), lda, dA_tst, lda, queue );
if (verbose >= 1) {
printf( "\n"
"A m=%5lld, n=%5lld, lda=%5lld\n",
llong( m ), llong( n ), llong( lda ) );
}
if (verbose >= 2) {
printf( "A = " ); print_matrix( m, n, &A_tst[0], lda );
}
// Allocate workspace
size_t d_size, h_size;
lapack::geqrf_work_size_bytes( m, n, dA_tst, lda, &d_size, &h_size, queue );
char* d_work = blas::device_malloc< char >( d_size, queue );
std::vector<char> h_work_vector( h_size );
char* h_work = h_work_vector.data();
// test error exits
if (params.error_exit() == 'y') {
assert_throw( lapack::geqrf( -1, n, dA_tst, lda, d_tau, d_work, d_size, h_work, h_size, d_info, queue ), lapack::Error );
assert_throw( lapack::geqrf( m, -1, dA_tst, lda, d_tau, d_work, d_size, h_work, h_size, d_info, queue ), lapack::Error );
assert_throw( lapack::geqrf( m, n, dA_tst, m-1, d_tau, d_work, d_size, h_work, h_size, d_info, queue ), lapack::Error );
}
// ---------- run test
testsweeper::flush_cache( params.cache() );
queue.sync();
double time = testsweeper::get_wtime();
lapack::geqrf( m, n, dA_tst, lda, d_tau,
d_work, d_size, h_work, h_size, d_info, queue );
queue.sync();
time = testsweeper::get_wtime() - time;
params.time() = time;
double gflop = lapack::Gflop< scalar_t >::geqrf( m, n );
params.gflops() = gflop / time;
// Copy result back to CPU.
device_info_int info_tst;
blas::device_copy_matrix( m, n, dA_tst, lda, A_tst.data(), lda, queue );
blas::device_memcpy( &info_tst, d_info, 1, queue );
blas::device_memcpy( &tau_tst[0], d_tau, size_tau, queue );
queue.sync();
if (info_tst != 0) {
fprintf( stderr, "lapack::geqrf returned error %lld\n", llong( info_tst ) );
}
// Cleanup GPU memory.
blas::device_free( dA_tst, queue );
blas::device_free( d_tau, queue );
blas::device_free( d_info, queue );
blas::device_free( d_work, queue );
if (verbose >= 2) {
printf( "A_factor = " ); print_matrix( m, n, &A_tst[0], lda );
printf( "tau = " ); print_matrix( 1, minmn, &tau_tst[0], 1 );
}
if (params.check() == 'y') {
// ---------- check error
// comparing to ref. solution doesn't work
// Following lapack/TESTING/LIN/zqrt01.f but using smaller Q and R
int64_t ldq = m;
std::vector< scalar_t > Q( m * minmn ); // m by k
int64_t ldr = minmn;
std::vector< scalar_t > R( minmn * n ); // k by n
// Copy details of Q
real_t rogue = -10000000000; // -1D+10
lapack::laset( lapack::MatrixType::General, m, minmn, rogue, rogue, &Q[0], ldq );
lapack::lacpy( lapack::MatrixType::Lower, m, minmn, &A_tst[0], lda, &Q[0], ldq );
// Generate the m-by-m matrix Q
int64_t info_ungqr = lapack::ungqr( m, minmn, minmn, &Q[0], ldq, &tau_tst[0] );
if (info_ungqr != 0) {
fprintf( stderr, "lapack::ungqr returned error %lld\n", llong( info_ungqr ) );
}
// Copy R
lapack::laset( lapack::MatrixType::Lower, minmn, n, 0.0, 0.0, &R[0], ldr );
lapack::lacpy( lapack::MatrixType::Upper, minmn, n, &A_tst[0], lda, &R[0], ldr );
// Compute R - Q'*A
blas::gemm( blas::Layout::ColMajor,
blas::Op::ConjTrans, blas::Op::NoTrans, minmn, n, m,
-1.0, &Q[0], ldq, &A_ref[0], lda, 1.0, &R[0], ldr );
// Compute norm( R - Q'*A ) / ( M * norm(A) * EPS )
real_t Anorm = lapack::lange( lapack::Norm::One, m, n, &A_ref[0], lda );
real_t resid1 = lapack::lange( lapack::Norm::One, minmn, n, &R[0], ldr );
real_t error1 = 0;
if (Anorm > 0)
error1 = resid1 / ( n * Anorm );
// Compute I - Q'*Q
lapack::laset( lapack::MatrixType::Upper, minmn, minmn, 0.0, 1.0, &R[0], ldr );
blas::herk( blas::Layout::ColMajor, blas::Uplo::Upper, blas::Op::ConjTrans,
minmn, m, -1.0, &Q[0], ldq, 1.0, &R[0], ldr );
// Compute norm( I - Q'*Q ) / ( M * EPS ) .
real_t resid2 = lapack::lanhe( lapack::Norm::One, lapack::Uplo::Upper, minmn, &R[0], ldr );
real_t error2 = ( resid2 / n );
params.error() = error1;
params.ortho() = error2;
params.okay() = (error1 < tol) && (error2 < tol);
}
if (params.ref() == 'y') {
// ---------- run reference
testsweeper::flush_cache( params.cache() );
time = testsweeper::get_wtime();
int64_t info_ref = LAPACKE_geqrf( m, n, &A_ref[0], lda, &tau_ref[0] );
time = testsweeper::get_wtime() - time;
if (info_ref != 0) {
fprintf( stderr, "LAPACKE_geqrf returned error %lld\n", llong( info_ref ) );
}
params.ref_time() = time;
params.ref_gflops() = gflop / time;
if (verbose >= 2) {
printf( "Aref_factor = " ); print_matrix( m, n, &A_ref[0], lda );
}
}
}
// -----------------------------------------------------------------------------
void test_geqrf_device( Params& params, bool run )
{
switch (params.datatype()) {
case testsweeper::DataType::Single:
test_geqrf_device_work< float >( params, run );
break;
case testsweeper::DataType::Double:
test_geqrf_device_work< double >( params, run );
break;
case testsweeper::DataType::SingleComplex:
test_geqrf_device_work< std::complex<float> >( params, run );
break;
case testsweeper::DataType::DoubleComplex:
test_geqrf_device_work< std::complex<double> >( params, run );
break;
default:
throw std::runtime_error( "unknown datatype" );
break;
}
}
|
