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
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@generated from testing/testing_zpotrf_gpu.cpp, normal z -> c, Wed Jan 22 14:40:26 2025
*/
// includes, system
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <math.h>
// includes, project
#include "flops.h"
#include "magma_v2.h"
#include "magma_lapack.h"
#include "testings.h"
/* ////////////////////////////////////////////////////////////////////////////
-- Testing cpotrf
*/
int main( int argc, char** argv)
{
TESTING_CHECK( magma_init() );
magma_print_environment();
// constants
const magmaFloatComplex c_neg_one = MAGMA_C_NEG_ONE;
const magma_int_t ione = 1;
// locals
real_Double_t gflops, gpu_perf=0, gpu_time=0, cpu_perf=0, cpu_time=0;
magmaFloatComplex *h_A, *h_R;
magmaFloatComplex_ptr d_A;
magma_int_t N, n2, lda, ldda, info;
float Anorm, error, work[1], *sigma;
int status = 0;
magma_opts opts;
opts.matrix = "rand_dominant"; // default
opts.parse_opts( argc, argv );
opts.lapack |= opts.check; // check (-c) implies lapack (-l)
float tol = opts.tolerance * lapackf77_slamch("E");
// for expert API testing
magma_device_t cdev;
magma_queue_t queues[2];
magma_event_t events[2];
magma_getdevice( &cdev );
magma_queue_create( cdev, &queues[0] );
magma_queue_create( cdev, &queues[1] );
magma_event_create(&events[0]);
magma_event_create(&events[1]);
printf("%% uplo = %s\n", lapack_uplo_const(opts.uplo) );
printf("%% N CPU Gflop/s (sec) GPU Gflop/s (sec) ||R_magma - R_lapack||_F / ||R_lapack||_F\n");
printf("%%=======================================================\n");
for( int itest = 0; itest < opts.ntest; ++itest ) {
for( int iter = 0; iter < opts.niter; ++iter ) {
N = opts.nsize[itest];
lda = max(1, N);
n2 = lda*N;
ldda = max(1, magma_roundup( N, opts.align )); // multiple of 32 by default
gflops = FLOPS_CPOTRF( N ) / 1e9;
TESTING_CHECK( magma_cmalloc_cpu( &h_A, n2 ));
TESTING_CHECK( magma_smalloc_cpu( &sigma, N ));
TESTING_CHECK( magma_cmalloc_pinned( &h_R, n2 ));
TESTING_CHECK( magma_cmalloc( &d_A, ldda*N ));
/* Initialize the matrix */
magma_generate_matrix( opts, N, N, h_A, lda, sigma );
lapackf77_clacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda );
magma_csetmatrix( N, N, h_A, lda, d_A, ldda, opts.queue );
/* ====================================================================
Performs operation using MAGMA
=================================================================== */
if(opts.version == 1){
gpu_time = magma_wtime();
magma_cpotrf_gpu( opts.uplo, N, d_A, ldda, &info );
gpu_time = magma_wtime() - gpu_time;
}
else if(opts.version == 2){
gpu_time = magma_wtime();
magma_cpotrf_native(opts.uplo, N, d_A, ldda, &info );
gpu_time = magma_wtime() - gpu_time;
}
else if(opts.version == 3 || opts.version == 4) {
// expert interface
magma_mode_t mode = (opts.version == 3) ? MagmaHybrid : MagmaNative;
magma_int_t nb = magma_get_cpotrf_nb( N );
magma_int_t recnb = 128;
// query workspace
void *hwork = NULL, *dwork=NULL;
magma_int_t lhwork[1] = {-1}, ldwork[1] = {-1};
magma_cpotrf_expert_gpu_work(
opts.uplo, N, NULL, ldda, &info,
mode, nb, recnb,
NULL, lhwork, NULL, ldwork,
events, queues );
// alloc workspace
if( lhwork[0] > 0 ) {
magma_malloc_pinned( (void**)&hwork, lhwork[0] );
}
if( ldwork[0] > 0 ) {
magma_malloc( (void**)&dwork, ldwork[0] );
}
// time actual call only
gpu_time = magma_wtime();
magma_cpotrf_expert_gpu_work(
opts.uplo, N, d_A, ldda, &info,
mode, nb, recnb,
hwork, lhwork, dwork, ldwork,
events, queues );
magma_queue_sync( queues[0] );
magma_queue_sync( queues[1] );
gpu_time = magma_wtime() - gpu_time;
// free workspace
if( hwork != NULL) {
magma_free_pinned( hwork );
}
if( dwork != NULL ) {
magma_free( dwork );
}
}
gpu_perf = gflops / gpu_time;
if (info != 0) {
printf("magma_cpotrf_gpu returned error %lld: %s.\n",
(long long) info, magma_strerror( info ));
}
if ( opts.lapack ) {
/* =====================================================================
Performs operation using LAPACK
=================================================================== */
cpu_time = magma_wtime();
lapackf77_cpotrf( lapack_uplo_const(opts.uplo), &N, h_A, &lda, &info );
cpu_time = magma_wtime() - cpu_time;
cpu_perf = gflops / cpu_time;
if (info != 0) {
printf("lapackf77_cpotrf returned error %lld: %s.\n",
(long long) info, magma_strerror( info ));
}
/* =====================================================================
Check the result compared to LAPACK
=================================================================== */
magma_cgetmatrix( N, N, d_A, ldda, h_R, lda, opts.queue );
blasf77_caxpy(&n2, &c_neg_one, h_A, &ione, h_R, &ione);
#ifndef MAGMA_HAVE_HIP
Anorm = lapackf77_clange("f", &N, &N, h_A, &lda, work);
error = lapackf77_clange("f", &N, &N, h_R, &lda, work) / Anorm;
#else
// TODO: use clange when the herk/syrk implementations are standardized.
// For HIP, the current herk/syrk routines overwrite the entire diagonal
// blocks of the matrix, so using clange causes the error check to fail
Anorm = safe_lapackf77_clanhe( "f", lapack_uplo_const(opts.uplo), &N, h_A, &lda, work );
error = safe_lapackf77_clanhe( "f", lapack_uplo_const(opts.uplo), &N, h_R, &lda, work ) / Anorm;
#endif
if (N == 0)
error = 0.0;
printf("%5lld %7.2f (%7.2f) %7.2f (%7.2f) %8.2e %s\n",
(long long) N, cpu_perf, cpu_time, gpu_perf, gpu_time,
error, (error < tol ? "ok" : "failed") );
status += ! (error < tol);
}
else {
printf("%5lld --- ( --- ) %7.2f (%7.2f) --- \n",
(long long) N, gpu_perf, gpu_time );
}
magma_free_cpu( h_A );
magma_free_cpu( sigma );
magma_free_pinned( h_R );
magma_free( d_A );
fflush( stdout );
}
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
magma_event_destroy( events[0] );
magma_event_destroy( events[1] );
magma_queue_destroy( queues[0] );
magma_queue_destroy( queues[1] );
opts.cleanup();
TESTING_CHECK( magma_finalize() );
return status;
}
|
