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
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@author Azzam Haidar
@author Ahmad Abdelfattah
@generated from testing/testing_zpotrf_vbatched.cpp, normal z -> d, Wed Jan 22 14:40:47 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"
#if defined(_OPENMP)
#include <omp.h>
#include "../control/magma_threadsetting.h" // internal header
#endif
/* ////////////////////////////////////////////////////////////////////////////
-- Testing dpotrf_vbatched
*/
int main( int argc, char** argv)
{
TESTING_CHECK( magma_init() );
magma_print_environment();
real_Double_t gflops, gpu_perf, gpu_time, cpu_perf, cpu_time;
double *h_A, *h_R;
double *d_A;
magma_int_t N, total_size_cpu, total_size_dev, info;
double c_neg_one = MAGMA_D_NEG_ONE;
magma_int_t ione = 1;
magma_int_t ISEED[4] = {0,0,0,1};
double work[1], Anorm, error, magma_error;
int status = 0;
double **h_A_array=NULL, **d_A_array = NULL;
magma_int_t *dinfo_magma;
magma_int_t *hinfo_magma;
magma_int_t max_N, batchCount;
magma_int_t *h_N, *d_N;
magma_int_t *h_lda, *h_ldda, *d_ldda;
double *h_A_tmp, *h_R_tmp, *d_A_tmp;
magma_opts opts( MagmaOptsBatched );
opts.parse_opts( argc, argv);
opts.lapack |= opts.check; // check (-c) implies lapack (-l)
batchCount = opts.batchcount;
double tol = opts.tolerance * lapackf77_dlamch("E");
TESTING_CHECK( magma_imalloc_cpu(&h_N, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&h_ldda, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&hinfo_magma, batchCount) );
TESTING_CHECK( magma_imalloc(&d_N, batchCount+1) );
TESTING_CHECK( magma_imalloc(&d_ldda, batchCount+1) );
TESTING_CHECK( magma_imalloc(&dinfo_magma, batchCount) );
TESTING_CHECK( magma_malloc_cpu((void**)&h_A_array, batchCount * sizeof(double*)) );
TESTING_CHECK( magma_malloc((void**)&d_A_array, batchCount * sizeof(double*)) );
h_lda = h_N;
printf("%% max\n");
printf("%% BatchCount N CPU Gflop/s (ms) MAGMA Gflop/s (ms) ||R_magma - R_lapack||_F / ||R_lapack||_F\n");
printf("%%=====================================================================================================\n");
for( int i = 0; i < opts.ntest; ++i ) {
for( int iter = 0; iter < opts.niter; ++iter ) {
srand(1000); // guarantee reproducible sizes
N = opts.nsize[i];
total_size_cpu = total_size_dev = 0;
gflops = 0;
max_N = 0;
for(int k = 0; k < batchCount; k++){
h_N[k] = 1 + (rand() % N);
max_N = max( max_N, h_N[k] );
h_ldda[k] = magma_roundup( h_N[k], opts.align ); // multiple of 32 by default
total_size_cpu += h_N[k] * h_lda[k];
total_size_dev += h_N[k] * h_ldda[k];
gflops += FLOPS_DPOTRF( h_N[k] ) / 1e9;
}
TESTING_CHECK( magma_dmalloc_cpu(&h_A, total_size_cpu) );
TESTING_CHECK( magma_dmalloc_pinned(&h_R, total_size_cpu) );
TESTING_CHECK( magma_dmalloc(&d_A, total_size_dev) );
/* Initialize the matrix */
lapackf77_dlarnv( &ione, ISEED, &total_size_cpu, h_A );
h_A_tmp = h_A;
for (int s=0; s < batchCount; s++){
magma_dmake_hpd( h_N[s], h_A_tmp, h_lda[s]); // need modification
h_A_tmp += h_N[s] * h_lda[s];
}
h_A_tmp = h_A;
h_R_tmp = h_R;
d_A_tmp = d_A;
for(int s = 0; s < batchCount; s++){
lapackf77_dlacpy( MagmaFullStr, &h_N[s], &h_N[s], h_A_tmp, &h_lda[s], h_R_tmp, &h_lda[s] );
magma_dsetmatrix( h_N[s], h_N[s], h_A_tmp, h_lda[s], d_A_tmp, h_ldda[s], opts.queue );
h_A_tmp += h_N[s] * h_lda[s];
h_R_tmp += h_N[s] * h_lda[s];
d_A_tmp += h_N[s] * h_ldda[s];
}
h_A_array[0] = d_A;
for(int s = 1; s < batchCount; s++){
h_A_array[s] = h_A_array[s-1] + h_N[s-1] * h_ldda[s-1];
}
magma_setvector(batchCount, sizeof(double*), h_A_array, 1, d_A_array, 1, opts.queue);
magma_setvector(batchCount, sizeof(magma_int_t), h_N, 1, d_N, 1, opts.queue);
magma_setvector(batchCount, sizeof(magma_int_t), h_ldda, 1, d_ldda, 1, opts.queue);
/* ====================================================================
Performs operation using MAGMA
=================================================================== */
memset(hinfo_magma, 0, batchCount * sizeof(magma_int_t));
magma_setvector(batchCount, sizeof(magma_int_t), hinfo_magma, 1, dinfo_magma, 1, opts.queue);
gpu_time = magma_sync_wtime( opts.queue );
info = magma_dpotrf_vbatched_max_nocheck(
opts.uplo, d_N,
d_A_array, d_ldda,
dinfo_magma, batchCount, max_N,
opts.queue);
gpu_time = magma_sync_wtime( opts.queue ) - gpu_time;
gpu_perf = gflops / gpu_time;
magma_getvector( batchCount, sizeof(magma_int_t), dinfo_magma, 1, hinfo_magma, 1, opts.queue);
for (int s=0; s < batchCount; s++){
if (hinfo_magma[s] != 0 ) {
printf("magma_dpotrf_vbatched matrix %d returned internal error %d\n", s, (int)hinfo_magma[s] );
status = -1;
goto cleanup;
}
}
if (info != 0){
status = -1;
goto cleanup;
}
/* =====================================================================
Performs operation using LAPACK
=================================================================== */
if ( opts.lapack ){
h_A_array[0] = h_A;
for(int s = 1; s < batchCount; s++){
h_A_array[s] = h_A_array[s-1] + h_N[s-1] * h_lda[s-1];
}
cpu_time = magma_wtime();
//#define BATCHED_DISABLE_PARCPU
#if !defined (BATCHED_DISABLE_PARCPU) && defined(_OPENMP)
magma_int_t nthreads = magma_get_lapack_numthreads();
magma_set_lapack_numthreads(1);
magma_set_omp_numthreads(nthreads);
#pragma omp parallel for schedule(dynamic)
#endif
for (magma_int_t s=0; s < batchCount; s++){
lapackf77_dpotrf( lapack_uplo_const(opts.uplo), &h_N[s], h_A_array[s], &h_lda[s], &info );
if (info != 0)
printf("lapackf77_dpotrf matrix %d returned err %d: %s.\n", (int) s, (int) info, magma_strerror( info ));
}
#if !defined (BATCHED_DISABLE_PARCPU) && defined(_OPENMP)
magma_set_lapack_numthreads(nthreads);
#endif
cpu_time = magma_wtime() - cpu_time;
cpu_perf = gflops / cpu_time;
/* =====================================================================
Check the result compared to LAPACK
=================================================================== */
h_R_tmp = h_R;
d_A_tmp = d_A;
for(int s = 0; s < batchCount; s++){
magma_dgetmatrix(h_N[s], h_N[s], d_A_tmp, h_ldda[s], h_R_tmp, h_lda[s], opts.queue);
h_R_tmp += h_N[s] * h_lda[s];
d_A_tmp += h_N[s] * h_ldda[s];
}
magma_error = 0.0;
h_A_tmp = h_A;
h_R_tmp = h_R;
for (int s=0; s < batchCount; s++)
{
magma_int_t Asize = h_lda[s] * h_N[s];
Anorm = lapackf77_dlansy("f", lapack_uplo_const(opts.uplo), &h_N[s], h_A_tmp, &h_lda[s], work);
blasf77_daxpy(&Asize, &c_neg_one, h_A_tmp, &ione, h_R_tmp, &ione);
error = lapackf77_dlansy("f", lapack_uplo_const(opts.uplo), &h_N[s], h_R_tmp, &h_lda[s], work) / Anorm;
magma_error = magma_max_nan( magma_error, error );
h_A_tmp += h_N[s] * h_lda[s];
h_R_tmp += h_N[s] * h_lda[s];
}
bool okay = (magma_error < tol);
status += ! okay;
printf(" %10lld %5lld %7.2f (%7.2f) %7.2f (%7.2f) %8.2e %s\n",
(long long)batchCount, (long long)max_N,
cpu_perf, cpu_time*1000.,
gpu_perf, gpu_time*1000.,
magma_error, (magma_error < tol ? "ok" : "failed"));
}
else {
printf(" %10lld %5lld --- ( --- ) %7.2f (%7.2f) ---\n",
(long long)batchCount, (long long)max_N,
gpu_perf, gpu_time*1000. );
}
cleanup:
magma_free_pinned( h_R );
magma_free_cpu( h_A );
magma_free( d_A );
if (status == -1)
break;
}
if (status == -1)
break;
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
magma_free( d_N );
magma_free( d_ldda );
magma_free( d_A_array );
magma_free( dinfo_magma );
magma_free_cpu( h_N );
magma_free_cpu( h_ldda );
magma_free_cpu( h_A_array );
magma_free_cpu( hinfo_magma );
opts.cleanup();
TESTING_CHECK( magma_finalize() );
return status;
}
|
