1/* 2 * (c) 2002 Fabrice Bellard 3 * 4 * This file is part of Libav. 5 * 6 * Libav is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * Libav is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with Libav; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 */ 20 21/** 22 * @file 23 * FFT and MDCT tests. 24 */ 25 26#include "libavutil/mathematics.h" 27#include "libavutil/lfg.h" 28#include "libavutil/log.h" 29#include "fft.h" 30#if CONFIG_FFT_FLOAT 31#include "dct.h" 32#include "rdft.h" 33#endif 34#include <math.h> 35#include <unistd.h> 36#include <sys/time.h> 37#include <stdlib.h> 38#include <string.h> 39 40/* reference fft */ 41 42#define MUL16(a,b) ((a) * (b)) 43 44#define CMAC(pre, pim, are, aim, bre, bim) \ 45{\ 46 pre += (MUL16(are, bre) - MUL16(aim, bim));\ 47 pim += (MUL16(are, bim) + MUL16(bre, aim));\ 48} 49 50#if CONFIG_FFT_FLOAT 51# define RANGE 1.0 52# define REF_SCALE(x, bits) (x) 53# define FMT "%10.6f" 54#else 55# define RANGE 16384 56# define REF_SCALE(x, bits) ((x) / (1<<(bits))) 57# define FMT "%6d" 58#endif 59 60struct { 61 float re, im; 62} *exptab; 63 64static void fft_ref_init(int nbits, int inverse) 65{ 66 int n, i; 67 double c1, s1, alpha; 68 69 n = 1 << nbits; 70 exptab = av_malloc((n / 2) * sizeof(*exptab)); 71 72 for (i = 0; i < (n/2); i++) { 73 alpha = 2 * M_PI * (float)i / (float)n; 74 c1 = cos(alpha); 75 s1 = sin(alpha); 76 if (!inverse) 77 s1 = -s1; 78 exptab[i].re = c1; 79 exptab[i].im = s1; 80 } 81} 82 83static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits) 84{ 85 int n, i, j, k, n2; 86 double tmp_re, tmp_im, s, c; 87 FFTComplex *q; 88 89 n = 1 << nbits; 90 n2 = n >> 1; 91 for (i = 0; i < n; i++) { 92 tmp_re = 0; 93 tmp_im = 0; 94 q = tab; 95 for (j = 0; j < n; j++) { 96 k = (i * j) & (n - 1); 97 if (k >= n2) { 98 c = -exptab[k - n2].re; 99 s = -exptab[k - n2].im; 100 } else { 101 c = exptab[k].re; 102 s = exptab[k].im; 103 } 104 CMAC(tmp_re, tmp_im, c, s, q->re, q->im); 105 q++; 106 } 107 tabr[i].re = REF_SCALE(tmp_re, nbits); 108 tabr[i].im = REF_SCALE(tmp_im, nbits); 109 } 110} 111 112static void imdct_ref(FFTSample *out, FFTSample *in, int nbits) 113{ 114 int n = 1<<nbits; 115 int k, i, a; 116 double sum, f; 117 118 for (i = 0; i < n; i++) { 119 sum = 0; 120 for (k = 0; k < n/2; k++) { 121 a = (2 * i + 1 + (n / 2)) * (2 * k + 1); 122 f = cos(M_PI * a / (double)(2 * n)); 123 sum += f * in[k]; 124 } 125 out[i] = REF_SCALE(-sum, nbits - 2); 126 } 127} 128 129/* NOTE: no normalisation by 1 / N is done */ 130static void mdct_ref(FFTSample *output, FFTSample *input, int nbits) 131{ 132 int n = 1<<nbits; 133 int k, i; 134 double a, s; 135 136 /* do it by hand */ 137 for (k = 0; k < n/2; k++) { 138 s = 0; 139 for (i = 0; i < n; i++) { 140 a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n)); 141 s += input[i] * cos(a); 142 } 143 output[k] = REF_SCALE(s, nbits - 1); 144 } 145} 146 147#if CONFIG_FFT_FLOAT 148static void idct_ref(float *output, float *input, int nbits) 149{ 150 int n = 1<<nbits; 151 int k, i; 152 double a, s; 153 154 /* do it by hand */ 155 for (i = 0; i < n; i++) { 156 s = 0.5 * input[0]; 157 for (k = 1; k < n; k++) { 158 a = M_PI*k*(i+0.5) / n; 159 s += input[k] * cos(a); 160 } 161 output[i] = 2 * s / n; 162 } 163} 164static void dct_ref(float *output, float *input, int nbits) 165{ 166 int n = 1<<nbits; 167 int k, i; 168 double a, s; 169 170 /* do it by hand */ 171 for (k = 0; k < n; k++) { 172 s = 0; 173 for (i = 0; i < n; i++) { 174 a = M_PI*k*(i+0.5) / n; 175 s += input[i] * cos(a); 176 } 177 output[k] = s; 178 } 179} 180#endif 181 182 183static FFTSample frandom(AVLFG *prng) 184{ 185 return (int16_t)av_lfg_get(prng) / 32768.0 * RANGE; 186} 187 188static int64_t gettime(void) 189{ 190 struct timeval tv; 191 gettimeofday(&tv,NULL); 192 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec; 193} 194 195static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale) 196{ 197 int i; 198 double max= 0; 199 double error= 0; 200 int err = 0; 201 202 for (i = 0; i < n; i++) { 203 double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE; 204 if (e >= 1e-3) { 205 av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n", 206 i, tab1[i], tab2[i]); 207 err = 1; 208 } 209 error+= e*e; 210 if(e>max) max= e; 211 } 212 av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error)/n); 213 return err; 214} 215 216 217static void help(void) 218{ 219 av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n" 220 "-h print this help\n" 221 "-s speed test\n" 222 "-m (I)MDCT test\n" 223 "-d (I)DCT test\n" 224 "-r (I)RDFT test\n" 225 "-i inverse transform test\n" 226 "-n b set the transform size to 2^b\n" 227 "-f x set scale factor for output data of (I)MDCT to x\n" 228 ); 229} 230 231enum tf_transform { 232 TRANSFORM_FFT, 233 TRANSFORM_MDCT, 234 TRANSFORM_RDFT, 235 TRANSFORM_DCT, 236}; 237 238int main(int argc, char **argv) 239{ 240 FFTComplex *tab, *tab1, *tab_ref; 241 FFTSample *tab2; 242 int it, i, c; 243 int do_speed = 0; 244 int err = 1; 245 enum tf_transform transform = TRANSFORM_FFT; 246 int do_inverse = 0; 247 FFTContext s1, *s = &s1; 248 FFTContext m1, *m = &m1; 249#if CONFIG_FFT_FLOAT 250 RDFTContext r1, *r = &r1; 251 DCTContext d1, *d = &d1; 252 int fft_size_2; 253#endif 254 int fft_nbits, fft_size; 255 double scale = 1.0; 256 AVLFG prng; 257 av_lfg_init(&prng, 1); 258 259 fft_nbits = 9; 260 for(;;) { 261 c = getopt(argc, argv, "hsimrdn:f:"); 262 if (c == -1) 263 break; 264 switch(c) { 265 case 'h': 266 help(); 267 return 1; 268 case 's': 269 do_speed = 1; 270 break; 271 case 'i': 272 do_inverse = 1; 273 break; 274 case 'm': 275 transform = TRANSFORM_MDCT; 276 break; 277 case 'r': 278 transform = TRANSFORM_RDFT; 279 break; 280 case 'd': 281 transform = TRANSFORM_DCT; 282 break; 283 case 'n': 284 fft_nbits = atoi(optarg); 285 break; 286 case 'f': 287 scale = atof(optarg); 288 break; 289 } 290 } 291 292 fft_size = 1 << fft_nbits; 293 tab = av_malloc(fft_size * sizeof(FFTComplex)); 294 tab1 = av_malloc(fft_size * sizeof(FFTComplex)); 295 tab_ref = av_malloc(fft_size * sizeof(FFTComplex)); 296 tab2 = av_malloc(fft_size * sizeof(FFTSample)); 297 298 switch (transform) { 299 case TRANSFORM_MDCT: 300 av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale); 301 if (do_inverse) 302 av_log(NULL, AV_LOG_INFO,"IMDCT"); 303 else 304 av_log(NULL, AV_LOG_INFO,"MDCT"); 305 ff_mdct_init(m, fft_nbits, do_inverse, scale); 306 break; 307 case TRANSFORM_FFT: 308 if (do_inverse) 309 av_log(NULL, AV_LOG_INFO,"IFFT"); 310 else 311 av_log(NULL, AV_LOG_INFO,"FFT"); 312 ff_fft_init(s, fft_nbits, do_inverse); 313 fft_ref_init(fft_nbits, do_inverse); 314 break; 315#if CONFIG_FFT_FLOAT 316 case TRANSFORM_RDFT: 317 if (do_inverse) 318 av_log(NULL, AV_LOG_INFO,"IDFT_C2R"); 319 else 320 av_log(NULL, AV_LOG_INFO,"DFT_R2C"); 321 ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C); 322 fft_ref_init(fft_nbits, do_inverse); 323 break; 324 case TRANSFORM_DCT: 325 if (do_inverse) 326 av_log(NULL, AV_LOG_INFO,"DCT_III"); 327 else 328 av_log(NULL, AV_LOG_INFO,"DCT_II"); 329 ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II); 330 break; 331#endif 332 default: 333 av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n"); 334 return 1; 335 } 336 av_log(NULL, AV_LOG_INFO," %d test\n", fft_size); 337 338 /* generate random data */ 339 340 for (i = 0; i < fft_size; i++) { 341 tab1[i].re = frandom(&prng); 342 tab1[i].im = frandom(&prng); 343 } 344 345 /* checking result */ 346 av_log(NULL, AV_LOG_INFO,"Checking...\n"); 347 348 switch (transform) { 349 case TRANSFORM_MDCT: 350 if (do_inverse) { 351 imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits); 352 m->imdct_calc(m, tab2, (FFTSample *)tab1); 353 err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale); 354 } else { 355 mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits); 356 357 m->mdct_calc(m, tab2, (FFTSample *)tab1); 358 359 err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale); 360 } 361 break; 362 case TRANSFORM_FFT: 363 memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); 364 s->fft_permute(s, tab); 365 s->fft_calc(s, tab); 366 367 fft_ref(tab_ref, tab1, fft_nbits); 368 err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0); 369 break; 370#if CONFIG_FFT_FLOAT 371 case TRANSFORM_RDFT: 372 fft_size_2 = fft_size >> 1; 373 if (do_inverse) { 374 tab1[ 0].im = 0; 375 tab1[fft_size_2].im = 0; 376 for (i = 1; i < fft_size_2; i++) { 377 tab1[fft_size_2+i].re = tab1[fft_size_2-i].re; 378 tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im; 379 } 380 381 memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); 382 tab2[1] = tab1[fft_size_2].re; 383 384 r->rdft_calc(r, tab2); 385 fft_ref(tab_ref, tab1, fft_nbits); 386 for (i = 0; i < fft_size; i++) { 387 tab[i].re = tab2[i]; 388 tab[i].im = 0; 389 } 390 err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5); 391 } else { 392 for (i = 0; i < fft_size; i++) { 393 tab2[i] = tab1[i].re; 394 tab1[i].im = 0; 395 } 396 r->rdft_calc(r, tab2); 397 fft_ref(tab_ref, tab1, fft_nbits); 398 tab_ref[0].im = tab_ref[fft_size_2].re; 399 err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0); 400 } 401 break; 402 case TRANSFORM_DCT: 403 memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); 404 d->dct_calc(d, tab); 405 if (do_inverse) { 406 idct_ref(tab_ref, tab1, fft_nbits); 407 } else { 408 dct_ref(tab_ref, tab1, fft_nbits); 409 } 410 err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0); 411 break; 412#endif 413 } 414 415 /* do a speed test */ 416 417 if (do_speed) { 418 int64_t time_start, duration; 419 int nb_its; 420 421 av_log(NULL, AV_LOG_INFO,"Speed test...\n"); 422 /* we measure during about 1 seconds */ 423 nb_its = 1; 424 for(;;) { 425 time_start = gettime(); 426 for (it = 0; it < nb_its; it++) { 427 switch (transform) { 428 case TRANSFORM_MDCT: 429 if (do_inverse) { 430 m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1); 431 } else { 432 m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1); 433 } 434 break; 435 case TRANSFORM_FFT: 436 memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); 437 s->fft_calc(s, tab); 438 break; 439#if CONFIG_FFT_FLOAT 440 case TRANSFORM_RDFT: 441 memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); 442 r->rdft_calc(r, tab2); 443 break; 444 case TRANSFORM_DCT: 445 memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); 446 d->dct_calc(d, tab2); 447 break; 448#endif 449 } 450 } 451 duration = gettime() - time_start; 452 if (duration >= 1000000) 453 break; 454 nb_its *= 2; 455 } 456 av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n", 457 (double)duration / nb_its, 458 (double)duration / 1000000.0, 459 nb_its); 460 } 461 462 switch (transform) { 463 case TRANSFORM_MDCT: 464 ff_mdct_end(m); 465 break; 466 case TRANSFORM_FFT: 467 ff_fft_end(s); 468 break; 469#if CONFIG_FFT_FLOAT 470 case TRANSFORM_RDFT: 471 ff_rdft_end(r); 472 break; 473 case TRANSFORM_DCT: 474 ff_dct_end(d); 475 break; 476#endif 477 } 478 479 av_free(tab); 480 av_free(tab1); 481 av_free(tab2); 482 av_free(tab_ref); 483 av_free(exptab); 484 485 return err; 486} 487