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
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@precisions normal z -> c d s
*/
// 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 zpotrf_mgpu
*/
int main( int argc, char** argv )
{
TESTING_CHECK( magma_init() );
magma_print_environment();
real_Double_t gflops, gpu_perf, gpu_time, cpu_perf=0, cpu_time=0;
double Anorm, error, work[1];
magmaDoubleComplex c_neg_one = MAGMA_Z_NEG_ONE;
magmaDoubleComplex *h_A, *h_R;
magmaDoubleComplex_ptr d_lA[ MagmaMaxGPUs ];
magma_int_t N, n2, lda, ldda, max_size, ngpu;
magma_int_t info, nb;
magma_int_t ione = 1;
int status = 0;
magma_opts opts;
opts.matrix = "rand_dominant"; // default
opts.parse_opts( argc, argv );
opts.ngpu = abs( opts.ngpu ); // always uses multi-GPU code
opts.lapack |= opts.check; // check (-c) implies lapack (-l)
double tol = opts.tolerance * lapackf77_dlamch("E");
magma_queue_t queues[ MagmaMaxAccelerators ] = { NULL };
for( int dev=0; dev < opts.ngpu; ++dev ) {
magma_queue_create( dev, &queues[dev] );
}
printf("%% ngpu = %lld, uplo = %s\n", (long long) opts.ngpu, lapack_uplo_const(opts.uplo) );
printf("%% N CPU Gflop/s (sec) GPU Gflop/s (sec) ||R||_F / ||A||_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 = N;
n2 = lda*N;
nb = magma_get_zpotrf_nb( N );
gflops = FLOPS_ZPOTRF( N ) / 1e9;
// ngpu must be at least the number of blocks
ngpu = min( opts.ngpu, magma_ceildiv(N,nb) );
if ( ngpu < opts.ngpu ) {
printf( " * too many GPUs for the matrix size, using %lld GPUs\n", (long long) ngpu );
}
// Allocate host memory for the matrix
TESTING_CHECK( magma_zmalloc_cpu( &h_A, n2 ));
TESTING_CHECK( magma_zmalloc_pinned( &h_R, n2 ));
// Allocate device memory
// matrix is distributed by block-rows or block-columns
// this is maximum size that any GPU stores;
// size is rounded up to full blocks in both rows and columns
max_size = (1+N/(nb*ngpu))*nb * magma_roundup( N, nb );
for( int dev=0; dev < ngpu; dev++ ) {
magma_setdevice( dev );
TESTING_CHECK( magma_zmalloc( &d_lA[dev], max_size ));
}
/* Initialize the matrix */
magma_generate_matrix( opts, N, N, h_A, lda );
lapackf77_zlacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda );
/* =====================================================================
Performs operation using LAPACK
=================================================================== */
if ( opts.lapack ) {
cpu_time = magma_wtime();
lapackf77_zpotrf( 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_zpotrf returned error %lld: %s.\n",
(long long) info, magma_strerror( info ));
}
}
/* ====================================================================
Performs operation using MAGMA
=================================================================== */
if ( opts.uplo == MagmaUpper ) {
ldda = magma_roundup( N, nb );
magma_zsetmatrix_1D_col_bcyclic( ngpu, N, N, nb, h_R, lda, d_lA, ldda, queues );
}
else {
ldda = (1+N/(nb*ngpu))*nb;
magma_zsetmatrix_1D_row_bcyclic( ngpu, N, N, nb, h_R, lda, d_lA, ldda, queues );
}
gpu_time = magma_wtime();
magma_zpotrf_mgpu( ngpu, opts.uplo, N, d_lA, ldda, &info );
gpu_time = magma_wtime() - gpu_time;
gpu_perf = gflops / gpu_time;
if (info != 0) {
printf("magma_zpotrf_mgpu returned error %lld: %s.\n",
(long long) info, magma_strerror( info ));
}
if ( opts.uplo == MagmaUpper ) {
magma_zgetmatrix_1D_col_bcyclic( ngpu, N, N, nb, d_lA, ldda, h_R, lda, queues );
}
else {
magma_zgetmatrix_1D_row_bcyclic( ngpu, N, N, nb, d_lA, ldda, h_R, lda, queues );
}
/* =====================================================================
Check the result compared to LAPACK
=================================================================== */
if ( opts.lapack ) {
blasf77_zaxpy( &n2, &c_neg_one, h_A, &ione, h_R, &ione );
Anorm = lapackf77_zlange("f", &N, &N, h_A, &lda, work );
error = lapackf77_zlange("f", &N, &N, h_R, &lda, work ) / Anorm;
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_pinned( h_R );
for( int dev=0; dev < ngpu; dev++ ) {
magma_setdevice( dev );
magma_free( d_lA[dev] );
}
fflush( stdout );
}
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
for( int dev=0; dev < opts.ngpu; ++dev ) {
magma_queue_destroy( queues[dev] );
}
opts.cleanup();
TESTING_CHECK( magma_finalize() );
return status;
}
|
