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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@author Raffaele Solca
@author Azzam Haidar
@author Mark Gates
@generated from testing/testing_zheevdx_2stage.cpp, normal z -> d, Wed Jan 22 14:40:38 2025
*/
// includes, system
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <math.h>
// includes, project
#include "magma_v2.h"
#include "magma_lapack.h"
#include "testings.h"
#include "../control/magma_threadsetting.h" // internal header
#define REAL
static magma_int_t check_orthogonality(magma_int_t M, magma_int_t N, double *Q, magma_int_t LDQ, double eps);
static magma_int_t check_reduction(magma_uplo_t uplo, magma_int_t N, magma_int_t bw, double *A, double *D, magma_int_t LDA, double *Q, double eps );
static magma_int_t check_solution(magma_int_t N, magma_int_t Nfound, double *E1, double *E2, double tolulp);
/* ////////////////////////////////////////////////////////////////////////////
-- Testing dsygvdx
*/
int main( int argc, char** argv)
{
TESTING_CHECK( magma_init() );
magma_print_environment();
real_Double_t gpu_time;
double *h_A, *h_R, *h_work, unused[1];;
#ifdef COMPLEX
double *rwork;
magma_int_t lrwork;
#endif
double *w1, *w2, result[4]={0, 0, 0, 0}, runused[1];
magma_int_t *iwork;
magma_int_t N, Nfound, n2, info, lda, lwork, liwork;
magma_int_t info_ortho = 0;
magma_int_t info_solution = 0;
magma_int_t info_reduction = 0;
magma_int_t ione = 1;
magma_int_t izero = 0;
int status = 0;
magma_opts opts;
opts.parse_opts( argc, argv );
double tol = opts.tolerance * lapackf77_dlamch("E");
//double tolulp = opts.tolerance * lapackf77_dlamch("P");
// pass ngpu = -1 to test multi-GPU code using 1 gpu
magma_int_t abs_ngpu = abs( opts.ngpu );
printf("%% jobz = %s, uplo = %s, ngpu %lld\n",
lapack_vec_const(opts.jobz), lapack_uplo_const(opts.uplo),
(long long) abs_ngpu);
printf("%% N M GPU Time (sec) ||I-Q^H Q||/N ||A-QDQ^H||/(||A||N) |D-D_magma|/(|D| * N)\n");
printf("%%=========================================================================================\n");
magma_int_t threads = magma_get_parallel_numthreads();
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;
magma_range_t range;
magma_int_t il, iu;
double vl, vu;
opts.get_range( N, &range, &vl, &vu, &il, &iu );
magma_dsyevdx_getworksize(N, threads, (opts.jobz == MagmaVec),
&lwork,
#ifdef COMPLEX
&lrwork,
#endif
&liwork);
if( opts.jobz == MagmaNoVec ) {
// For LAPACK test using dsyevx.
#ifdef COMPLEX
lrwork = 7*N;
#endif
liwork = 5*N;
}
/* Allocate host memory for the matrix */
TESTING_CHECK( magma_dmalloc_cpu( &h_A, n2 ));
TESTING_CHECK( magma_dmalloc_cpu( &w1, N ));
TESTING_CHECK( magma_dmalloc_cpu( &w2, N ));
TESTING_CHECK( magma_imalloc_cpu( &iwork, liwork ));
TESTING_CHECK( magma_dmalloc_pinned( &h_R, n2 ));
TESTING_CHECK( magma_dmalloc_pinned( &h_work, lwork ));
#ifdef COMPLEX
TESTING_CHECK( magma_dmalloc_pinned( &rwork, lrwork ));
#endif
/* Initialize the matrix */
magma_generate_matrix( opts, N, N, h_A, lda );
if (opts.warmup) {
// ==================================================================
// Warmup using MAGMA
// ==================================================================
lapackf77_dlacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda );
if (opts.ngpu == 1) {
//printf("calling dsyevdx_2stage 1 GPU\n");
magma_dsyevdx_2stage( opts.jobz, range, opts.uplo, N,
h_R, lda,
vl, vu, il, iu,
&Nfound, w1,
h_work, lwork,
#ifdef COMPLEX
rwork, lrwork,
#endif
iwork, liwork,
&info );
} else {
//printf("calling dsyevdx_2stage_m %lld GPU\n", (long long) opts.ngpu);
magma_dsyevdx_2stage_m( abs_ngpu, opts.jobz, range, opts.uplo, N,
h_R, lda,
vl, vu, il, iu,
&Nfound, w1,
h_work, lwork,
#ifdef COMPLEX
rwork, lrwork,
#endif
iwork, liwork,
&info );
}
}
// ===================================================================
// Performs operation using MAGMA
// ===================================================================
lapackf77_dlacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda );
gpu_time = magma_wtime();
if (opts.ngpu == 1) {
//printf("calling dsyevdx_2stage 1 GPU\n");
magma_dsyevdx_2stage( opts.jobz, range, opts.uplo, N,
h_R, lda,
vl, vu, il, iu,
&Nfound, w1,
h_work, lwork,
#ifdef COMPLEX
rwork, lrwork,
#endif
iwork, liwork,
&info );
} else {
//printf("calling dsyevdx_2stage_m %lld GPU\n", (long long) opts.ngpu);
magma_dsyevdx_2stage_m( abs_ngpu, opts.jobz, range, opts.uplo, N,
h_R, lda,
vl, vu, il, iu,
&Nfound, w1,
h_work, lwork,
#ifdef COMPLEX
rwork, lrwork,
#endif
iwork, liwork,
&info );
}
gpu_time = magma_wtime() - gpu_time;
if (info != 0) {
printf("magma_dsyevdx_2stage returned error %lld: %s.\n",
(long long) info, magma_strerror( info ));
}
printf("%5lld %5lld %7.2f ",
(long long) N, (long long) Nfound, gpu_time );
if ( opts.check ) {
info_solution = 0;
info_ortho = 0;
info_reduction = 0;
//double eps = lapackf77_dlamch("E")*lapackf77_dlamch("B");
double eps = lapackf77_dlamch("E");
double abstol = 2 * lapackf77_dlamch("S");
double* h_Z;
magma_int_t* ifail;
TESTING_CHECK( magma_dmalloc_cpu( &h_Z, N*lda ));
TESTING_CHECK( magma_imalloc_cpu( &ifail, N ));
/* Check the orthogonality, reduction and the eigen solutions */
if (opts.jobz == MagmaVec) {
// Disable the following old tests and use LAPACK routines.
//info_ortho = check_orthogonality(N, N, h_R, lda, eps);
//info_reduction = check_reduction(opts.uplo, N, 1, h_A, w1, lda, h_R, eps);
/* =====================================================================
Check the results following the LAPACK's [zcds]drvst routine.
A is factored as A = U S U^H and the following 3 tests computed:
(1) | A - U S U^H | / ( |A| N ) if all eigenvectors were computed
| U^H A U - S | / ( |A| Nfound ) otherwise
(2) | I - U^H U | / ( N )
(3) | S(with U) - S(w/o U) | / | S | // currently disabled, but compares to LAPACK
=================================================================== */
double* work;
TESTING_CHECK( magma_dmalloc_cpu( &work, 2*N*N ));
// e is unused since kband=0; tau is unused since itype=1
if( Nfound == N ) {
lapackf77_dsyt21( &ione, lapack_uplo_const(opts.uplo), &N, &izero,
h_A, &lda,
w1, runused,
h_R, &lda,
h_R, &lda,
unused, work,
#ifdef COMPLEX
rwork,
#endif
&result[0] );
} else {
lapackf77_dsyt22( &ione, lapack_uplo_const(opts.uplo), &N, &Nfound, &izero,
h_A, &lda,
w1, runused,
h_R, &lda,
h_R, &lda,
unused, work,
#ifdef COMPLEX
rwork,
#endif
&result[0] );
}
result[0] *= eps;
result[1] *= eps;
info_reduction = result[0] >= tol;
info_ortho = result[1] >= tol;
printf(" %8.2e %8.2e", result[0], result[1] );
magma_free_cpu( work ); work=NULL;
} else {
printf(" --- --- ");
}
lapackf77_dsyevx( "N",
lapack_range_const(range),
lapack_uplo_const(opts.uplo),
&N, h_A, &lda,
&vl, &vu, &il, &iu, &abstol,
&Nfound, w2,
h_Z, &lda,
h_work, &lwork,
#ifdef COMPLEX
rwork,
#endif
iwork,
ifail,
&info );
info_solution = check_solution(N, Nfound, w2, w1, eps);
magma_free_cpu(h_Z);
magma_free_cpu(ifail);
bool okay = (info_solution == 0) && (info_ortho == 0) && (info_reduction == 0);
status += ! okay;
printf(" %s", (okay ? "ok" : "failed"));
}
printf("\n");
magma_free_cpu( h_A );
magma_free_cpu( w1 );
magma_free_cpu( w2 );
magma_free_cpu( iwork );
magma_free_pinned( h_R );
magma_free_pinned( h_work );
#ifdef COMPLEX
magma_free_pinned( rwork );
#endif
fflush( stdout );
}
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
opts.cleanup();
TESTING_CHECK( magma_finalize() );
return status;
}
/*-------------------------------------------------------------------
* Check the orthogonality of Q
*/
static magma_int_t check_orthogonality(magma_int_t M, magma_int_t N, double *Q, magma_int_t LDQ, double eps)
{
double d_one = 1.0;
double d_neg_one = -1.0;
double c_zero = MAGMA_D_ZERO;
double c_one = MAGMA_D_ONE;
double normQ, result;
magma_int_t info_ortho;
magma_int_t minMN = min(M, N);
double *work;
TESTING_CHECK( magma_dmalloc_cpu( &work, minMN ));
/* Build the idendity matrix */
double *Id;
TESTING_CHECK( magma_dmalloc_cpu( &Id, minMN*minMN ));
lapackf77_dlaset("A", &minMN, &minMN, &c_zero, &c_one, Id, &minMN);
/* Perform Id - Q^H Q */
if (M >= N)
blasf77_dsyrk("U", "C", &N, &M, &d_one, Q, &LDQ, &d_neg_one, Id, &N);
else
blasf77_dsyrk("U", "N", &M, &N, &d_one, Q, &LDQ, &d_neg_one, Id, &M);
normQ = safe_lapackf77_dlansy("I", "U", &minMN, Id, &minMN, work);
result = normQ / (minMN * eps);
printf( " %8.2e", normQ / minMN );
// TODO: use opts.tolerance instead of hard coding 60
if (std::isnan(result) || std::isinf(result) || (result > 60.0)) {
info_ortho = 1;
}
else {
info_ortho = 0;
}
magma_free_cpu( work );
magma_free_cpu( Id );
return info_ortho;
}
/*------------------------------------------------------------
* Check the reduction
*/
static magma_int_t check_reduction(magma_uplo_t uplo, magma_int_t N, magma_int_t bw, double *A, double *D, magma_int_t LDA, double *Q, double eps )
{
double c_one = MAGMA_D_ONE;
double c_neg_one = MAGMA_D_NEG_ONE;
double *TEMP, *Residual;
double *work;
double Anorm, Rnorm, result;
magma_int_t info_reduction;
magma_int_t i;
magma_int_t ione=1;
TESTING_CHECK( magma_dmalloc_cpu( &TEMP, N*N ));
TESTING_CHECK( magma_dmalloc_cpu( &Residual, N*N ));
TESTING_CHECK( magma_dmalloc_cpu( &work, N ));
/* Compute TEMP = Q * LAMBDA */
lapackf77_dlacpy("A", &N, &N, Q, &LDA, TEMP, &N);
for (i = 0; i < N; i++) {
blasf77_dscal(&N, &D[i], &(TEMP[i*N]), &ione);
}
/* Compute Residual = A - Q * LAMBDA * Q^H */
/* A is symmetric but both upper and lower
* are assumed valable here for checking
* otherwise it need to be symetrized before
* checking.
*/
lapackf77_dlacpy("A", &N, &N, A, &LDA, Residual, &N);
blasf77_dgemm("N", "C", &N, &N, &N, &c_neg_one, TEMP, &N, Q, &LDA, &c_one, Residual, &N);
// since A has been generated by larnv and we did not symmetrize,
// so only the uplo portion of A should be equal to Q*LAMBDA*Q^H
// for that Rnorm use dlansy instead of dlange
Rnorm = safe_lapackf77_dlansy("1", lapack_uplo_const(uplo), &N, Residual, &N, work);
Anorm = safe_lapackf77_dlansy("1", lapack_uplo_const(uplo), &N, A, &LDA, work);
result = Rnorm / ( Anorm * N * eps);
printf(" %8.2e", Rnorm / ( Anorm * N));
// TODO: use opts.tolerance instead of hard coding 60
if (std::isnan(result) || std::isinf(result) || (result > 60.0)) {
info_reduction = 1;
}
else {
info_reduction = 0;
}
magma_free_cpu( TEMP );
magma_free_cpu( Residual );
magma_free_cpu( work );
return info_reduction;
}
/*------------------------------------------------------------
* Check the eigenvalues
*/
static magma_int_t check_solution(magma_int_t N, magma_int_t Nfound, double *E1, double *E2, double tolulp)
{
magma_int_t info_solution, i;
double unfl = lapackf77_dlamch("Safe minimum");
double resid;
double maxtmp;
double maxdif = fabs( fabs(E1[0]) - fabs(E2[0]) );
double maxeig = max( fabs(E1[0]), fabs(E2[0]) );
for (i = 1; i < Nfound; i++) {
resid = fabs(fabs(E1[i])-fabs(E2[i]));
maxtmp = max(fabs(E1[i]), fabs(E2[i]));
/* Update */
maxeig = max(maxtmp, maxeig);
maxdif = max(resid, maxdif );
}
maxtmp = maxdif / max(unfl, max(maxeig, maxdif));
printf(" %8.2e", maxdif / (max(maxeig, maxdif)) );
if (std::isnan(maxtmp) || std::isinf(maxtmp) || (maxtmp >= tolulp * N)) {
info_solution = 1;
}
else {
info_solution = 0;
}
return info_solution;
}
|