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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@generated from testing/testing_zher2k_vbatched.cpp, normal z -> s, Wed Jan 22 14:40:47 2025
@author Mark Gates
@author Azzam Haidar
@author Tingxing Dong
@author Ahmad Abdelfattah
*/
// 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>
#endif
#include "../control/magma_threadsetting.h" // internal header
#define REAL
/* ////////////////////////////////////////////////////////////////////////////
-- Testing ssyr2k_vbatched
*/
int main( int argc, char** argv)
{
TESTING_CHECK( magma_init() );
magma_print_environment();
real_Double_t gflops, magma_perf, magma_time, cpu_perf, cpu_time;
float error, magma_error, normalize, work[1];
magma_int_t N, K;
magma_int_t *An, *Ak, *Bn, *Bk;
magma_int_t total_size_A_cpu = 0, total_size_B_cpu = 0, total_size_C_cpu = 0;
magma_int_t total_size_A_dev = 0, total_size_B_dev = 0, total_size_C_dev = 0;
magma_int_t ione = 1;
magma_int_t ISEED[4] = {0,0,0,1};
int status = 0;
magma_int_t batchCount;
magma_int_t max_N, max_K;
float *h_A, *h_B, *h_C, *h_Cmagma;
float *d_A, *d_B, *d_C;
float **h_A_array = NULL;
float **h_B_array = NULL;
float **h_C_array = NULL;
float **d_A_array = NULL;
float **d_B_array = NULL;
float **d_C_array = NULL;
float *h_A_tmp, *h_B_tmp, *h_C_tmp;
magma_int_t *h_lda, *h_ldda, *d_ldda;
magma_int_t *h_ldb, *h_lddb, *d_lddb;
magma_int_t *h_ldc, *h_lddc, *d_lddc;
magma_int_t *h_N, *h_K; // hold the sizes on cpu
magma_int_t *d_N, *d_K; // hold the sizes on gpu
float c_neg_one = MAGMA_S_NEG_ONE;
float alpha = MAGMA_S_MAKE( 0.29, -0.86 );
float beta = MAGMA_D_MAKE( -0.48, 0.38 );
magma_opts opts( MagmaOptsBatched );
opts.parse_opts( argc, argv );
opts.lapack |= opts.check;
batchCount = opts.batchcount;
// allocate space for the sizes/leading dim.
TESTING_CHECK( magma_imalloc_cpu(&h_N, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&h_K, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&h_ldda, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&h_lddb, batchCount) );
TESTING_CHECK( magma_imalloc_cpu(&h_lddc, batchCount) );
TESTING_CHECK( magma_imalloc(&d_N, batchCount+1) );
TESTING_CHECK( magma_imalloc(&d_K, batchCount+1) );
TESTING_CHECK( magma_imalloc(&d_ldda, batchCount+1) );
TESTING_CHECK( magma_imalloc(&d_lddb, batchCount+1) );
TESTING_CHECK( magma_imalloc(&d_lddc, batchCount+1) );
float *Anorm, *Bnorm, *Cnorm;
TESTING_CHECK( magma_smalloc_cpu( &Anorm, batchCount ));
TESTING_CHECK( magma_smalloc_cpu( &Bnorm, batchCount ));
TESTING_CHECK( magma_smalloc_cpu( &Cnorm, batchCount ));
TESTING_CHECK( magma_malloc_cpu((void**)&h_A_array, batchCount*sizeof(float*)) );
TESTING_CHECK( magma_malloc_cpu((void**)&h_B_array, batchCount*sizeof(float*)) );
TESTING_CHECK( magma_malloc_cpu((void**)&h_C_array, batchCount*sizeof(float*)) );
TESTING_CHECK( magma_malloc((void**)&d_A_array, batchCount*sizeof(float*)) );
TESTING_CHECK( magma_malloc((void**)&d_B_array, batchCount*sizeof(float*)) );
TESTING_CHECK( magma_malloc((void**)&d_C_array, batchCount*sizeof(float*)) );
// See testing_sgemm about tolerance.
float eps = lapackf77_slamch("E");
float tol = 3*eps;
#ifdef COMPLEX
if (opts.transA == MagmaTrans) {
opts.transA = MagmaConjTrans;
printf("%% WARNING: transA = MagmaTrans changed to MagmaConjTrans\n");
}
#endif
printf("%% If running lapack (option --lapack), MAGMA error is computed\n"
"%% relative to CPU BLAS result.\n\n"
"%% uplo = %s, trans = %s\n",
lapack_uplo_const(opts.uplo),
lapack_trans_const(opts.transA));
printf("%% max max\n");
printf("%% BatchCount N K MAGMA Gflop/s (ms) CPU Gflop/s (ms) MAGMA error\n");
printf("%%=============================================================================\n");
for( int itest = 0; itest < opts.ntest; ++itest ) {
for( int iter = 0; iter < opts.niter; ++iter ) {
N = opts.nsize[itest];
K = opts.ksize[itest];
if ( opts.transA == MagmaNoTrans ) {
h_lda = An = h_N;
h_ldb = Bn = h_N;
Ak = h_K;
Bk = h_K;
}
else {
h_lda = An = h_K;
h_ldb = Bn = h_K;
Ak = h_N;
Bk = h_N;
}
h_ldc = h_N;
// guarantee reproducible sizes
srand(1000);
gflops = 0;
max_N = max_K = 0;
total_size_A_cpu = total_size_A_dev = 0;
total_size_B_cpu = total_size_B_dev = 0;
total_size_C_cpu = total_size_C_dev = 0;
for (int i = 0; i < batchCount; i++) {
h_N[i] = 1 + (rand() % N);
h_K[i] = 1 + (rand() % K);
max_N = max( max_N, h_N[i] );
max_K = max( max_K, h_K[i] );
h_ldda[i] = magma_roundup( h_lda[i], opts.align ); // multiple of 32 by default
h_lddb[i] = magma_roundup( h_ldb[i], opts.align ); // multiple of 32 by default
h_lddc[i] = magma_roundup( h_ldc[i], opts.align ); // multiple of 32 by default
total_size_A_cpu += Ak[i] * h_lda[i];
total_size_A_dev += Ak[i] * h_ldda[i];
total_size_B_cpu += Bk[i] * h_ldb[i];
total_size_B_dev += Bk[i] * h_lddb[i];
total_size_C_cpu += h_N[i] * h_ldc[i];
total_size_C_dev += h_N[i] * h_lddc[i];
gflops += FLOPS_SSYR2K( h_N[i], h_K[i] ) / 1e9;
}
TESTING_CHECK( magma_smalloc_cpu(&h_A, total_size_A_cpu) );
TESTING_CHECK( magma_smalloc_cpu(&h_B, total_size_B_cpu) );
TESTING_CHECK( magma_smalloc_cpu(&h_C, total_size_C_cpu) );
TESTING_CHECK( magma_smalloc_cpu(&h_Cmagma, total_size_C_cpu) );
TESTING_CHECK( magma_smalloc(&d_A, total_size_A_dev) );
TESTING_CHECK( magma_smalloc(&d_B, total_size_B_dev) );
TESTING_CHECK( magma_smalloc(&d_C, total_size_C_dev) );
/* Initialize the matrices */
lapackf77_slarnv( &ione, ISEED, &total_size_A_cpu, h_A );
lapackf77_slarnv( &ione, ISEED, &total_size_B_cpu, h_B );
lapackf77_slarnv( &ione, ISEED, &total_size_C_cpu, h_C );
// Compute norms for error
h_A_tmp = h_A;
h_B_tmp = h_B;
h_C_tmp = h_C;
for (int s = 0; s < batchCount; ++s) {
Anorm[s] = lapackf77_slange( "F", &An[s], &Ak[s], h_A_tmp, &h_lda[s], work );
Bnorm[s] = lapackf77_slange( "F", &Bn[s], &Bk[s], h_B_tmp, &h_ldb[s], work );
Cnorm[s] = safe_lapackf77_slansy( "F", lapack_uplo_const(opts.uplo), &h_N[s], h_C_tmp, &h_ldc[s], work );
h_A_tmp += Ak[s] * h_lda[s];
h_B_tmp += Bk[s] * h_ldb[s];
h_C_tmp += h_N[s] * h_ldc[s];
}
/* =====================================================================
Performs operation using MAGMABLAS
=================================================================== */
magma_setvector(batchCount, sizeof(magma_int_t), h_N, 1, d_N, 1, opts.queue );
magma_setvector(batchCount, sizeof(magma_int_t), h_K, 1, d_K, 1, opts.queue );
magma_setvector(batchCount, sizeof(magma_int_t), h_ldda, 1, d_ldda, 1, opts.queue );
magma_setvector(batchCount, sizeof(magma_int_t), h_lddb, 1, d_lddb, 1, opts.queue );
magma_setvector(batchCount, sizeof(magma_int_t), h_lddc, 1, d_lddc, 1, opts.queue );
h_A_array[0] = d_A;
h_B_array[0] = d_B;
h_C_array[0] = d_C;
for (int i = 1; i < batchCount; i++) {
h_A_array[i] = h_A_array[i-1] + Ak[i-1] * h_ldda[i-1];
h_B_array[i] = h_B_array[i-1] + Bk[i-1] * h_lddb[i-1];
h_C_array[i] = h_C_array[i-1] + h_N[i-1] * h_lddc[i-1];
}
magma_setvector(batchCount, sizeof(float*), h_A_array, 1, d_A_array, 1, opts.queue );
magma_setvector(batchCount, sizeof(float*), h_B_array, 1, d_B_array, 1, opts.queue );
magma_setvector(batchCount, sizeof(float*), h_C_array, 1, d_C_array, 1, opts.queue );
h_A_tmp = h_A;
h_B_tmp = h_B;
h_C_tmp = h_C;
for (int i = 0; i < batchCount; i++) {
magma_ssetmatrix( An[i], Ak[i], h_A_tmp, h_lda[i], h_A_array[i], h_ldda[i], opts.queue );
magma_ssetmatrix( Bn[i], Bk[i], h_B_tmp, h_ldb[i], h_B_array[i], h_lddb[i], opts.queue );
magma_ssetmatrix( h_N[i], h_N[i], h_C_tmp, h_ldc[i], h_C_array[i], h_lddc[i], opts.queue );
h_A_tmp += Ak[i] * h_lda[i];
h_B_tmp += Bk[i] * h_ldb[i];
h_C_tmp += h_N[i] * h_ldc[i];
}
magma_time = magma_sync_wtime( opts.queue );
#if 0
magmablas_ssyr2k_vbatched_max_nocheck(opts.uplo, opts.transA, d_N, d_K,
alpha, d_A_array, d_ldda,
d_B_array, d_lddb,
beta, d_C_array, d_lddc,
batchCount, max_N, max_K, opts.queue );
#else
magmablas_ssyr2k_vbatched(opts.uplo, opts.transA, d_N, d_K,
alpha, d_A_array, d_ldda,
d_B_array, d_lddb,
beta, d_C_array, d_lddc,
batchCount, opts.queue );
#endif
magma_time = magma_sync_wtime( opts.queue ) - magma_time;
magma_perf = gflops / magma_time;
h_C_tmp = h_Cmagma;
for (int i = 0; i < batchCount; i++) {
magma_sgetmatrix( h_N[i], h_N[i], h_C_array[i], h_lddc[i], h_C_tmp, h_ldc[i], opts.queue );
h_C_tmp += h_N[i] * h_ldc[i];
}
/* =====================================================================
Performs operation using CPU BLAS
=================================================================== */
if ( opts.lapack ) {
// displace pointers for the cpu, reuse h_A_array, h_B_array, h_C_array
h_A_array[0] = h_A;
h_B_array[0] = h_B;
h_C_array[0] = h_C;
for (int i = 1; i < batchCount; i++) {
h_A_array[i] = h_A_array[i-1] + Ak[i-1] * h_lda[i-1];
h_B_array[i] = h_B_array[i-1] + Bk[i-1] * h_ldb[i-1];
h_C_array[i] = h_C_array[i-1] + h_N[i-1] * h_ldc[i-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++)
{
blasf77_ssyr2k( lapack_uplo_const(opts.uplo),
lapack_trans_const(opts.transA),
&h_N[s], &h_K[s],
&alpha, h_A_array[s], &h_lda[s],
h_B_array[s], &h_ldb[s],
&beta, h_C_array[s], &h_ldc[s] );
}
#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
=================================================================== */
if ( opts.lapack ) {
// compute error compared lapack
// error = |dC - C| / (2*gamma_{k+2}|A||B| + gamma_2|Cin|)
magma_error = 0;
h_C_tmp = h_C;
float* h_Cmagma_tmp = h_Cmagma;
for (int s=0; s < batchCount; s++) {
normalize = 2*sqrt(float(h_K[s]+2))*Anorm[s]*Bnorm[s] + 2*Cnorm[s];
if (normalize == 0)
normalize = 1;
magma_int_t Csize = h_ldc[s] * h_N[s];
blasf77_saxpy( &Csize, &c_neg_one, h_C_tmp, &ione, h_Cmagma_tmp, &ione );
error = safe_lapackf77_slansy( "F", lapack_uplo_const(opts.uplo), &h_N[s], h_Cmagma_tmp, &h_ldc[s], work )
/ normalize;
magma_error = magma_max_nan( error, magma_error );
h_C_tmp += h_N[s] * h_ldc[s];
h_Cmagma_tmp += h_N[s] * h_ldc[s];
}
bool okay = (magma_error < tol);
status += ! okay;
printf(" %10lld %5lld %5lld %7.2f (%7.2f) %7.2f (%7.2f) %8.2e %s\n",
(long long) batchCount, (long long) max_N, (long long) max_K,
magma_perf, 1000.*magma_time,
cpu_perf, 1000.*cpu_time,
magma_error, (okay ? "ok" : "failed"));
}
else {
printf(" %10lld %5lld %5lld %7.2f (%7.2f) --- ( --- ) ---\n",
(long long) batchCount, (long long) max_N, (long long) max_K,
magma_perf, 1000.*magma_time);
}
magma_free_cpu( h_A );
magma_free_cpu( h_B );
magma_free_cpu( h_C );
magma_free_cpu( h_Cmagma );
magma_free( d_A );
magma_free( d_B );
magma_free( d_C );
fflush( stdout);
}
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
// free resources
magma_free_cpu( h_N );
magma_free_cpu( h_K );
magma_free_cpu( h_ldda );
magma_free_cpu( h_lddb );
magma_free_cpu( h_lddc );
magma_free_cpu( Anorm );
magma_free_cpu( Bnorm );
magma_free_cpu( Cnorm );
magma_free_cpu( h_A_array );
magma_free_cpu( h_B_array );
magma_free_cpu( h_C_array );
magma_free( d_N );
magma_free( d_K );
magma_free( d_ldda );
magma_free( d_lddb );
magma_free( d_lddc );
magma_free( d_A_array );
magma_free( d_B_array );
magma_free( d_C_array );
opts.cleanup();
TESTING_CHECK( magma_finalize() );
return status;
}
|
