| 1 | /* |
| 2 | * (c) 2002 Fabrice Bellard |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg 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 | * FFmpeg 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 FFmpeg; 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 "config.h" |
| 27 | |
| 28 | #include <math.h> |
| 29 | #if HAVE_UNISTD_H |
| 30 | #include <unistd.h> |
| 31 | #endif |
| 32 | #include <stdio.h> |
| 33 | #include <stdlib.h> |
| 34 | #include <string.h> |
| 35 | |
| 36 | #include "libavutil/cpu.h" |
| 37 | #include "libavutil/lfg.h" |
| 38 | #include "libavutil/log.h" |
| 39 | #include "libavutil/mathematics.h" |
| 40 | #include "libavutil/time.h" |
| 41 | |
| 42 | #include "fft.h" |
| 43 | #if FFT_FLOAT |
| 44 | #include "dct.h" |
| 45 | #include "rdft.h" |
| 46 | #endif |
| 47 | |
| 48 | /* reference fft */ |
| 49 | |
| 50 | #define MUL16(a, b) ((a) * (b)) |
| 51 | |
| 52 | #define CMAC(pre, pim, are, aim, bre, bim) \ |
| 53 | { \ |
| 54 | pre += (MUL16(are, bre) - MUL16(aim, bim)); \ |
| 55 | pim += (MUL16(are, bim) + MUL16(bre, aim)); \ |
| 56 | } |
| 57 | |
| 58 | #if FFT_FLOAT |
| 59 | #define RANGE 1.0 |
| 60 | #define REF_SCALE(x, bits) (x) |
| 61 | #define FMT "%10.6f" |
| 62 | #elif FFT_FIXED_32 |
| 63 | #define RANGE 8388608 |
| 64 | #define REF_SCALE(x, bits) (x) |
| 65 | #define FMT "%6d" |
| 66 | #else |
| 67 | #define RANGE 16384 |
| 68 | #define REF_SCALE(x, bits) ((x) / (1 << (bits))) |
| 69 | #define FMT "%6d" |
| 70 | #endif |
| 71 | |
| 72 | static struct { |
| 73 | float re, im; |
| 74 | } *exptab; |
| 75 | |
| 76 | static int fft_ref_init(int nbits, int inverse) |
| 77 | { |
| 78 | int i, n = 1 << nbits; |
| 79 | |
| 80 | exptab = av_malloc_array((n / 2), sizeof(*exptab)); |
| 81 | if (!exptab) |
| 82 | return AVERROR(ENOMEM); |
| 83 | |
| 84 | for (i = 0; i < (n / 2); i++) { |
| 85 | double alpha = 2 * M_PI * (float) i / (float) n; |
| 86 | double c1 = cos(alpha), s1 = sin(alpha); |
| 87 | if (!inverse) |
| 88 | s1 = -s1; |
| 89 | exptab[i].re = c1; |
| 90 | exptab[i].im = s1; |
| 91 | } |
| 92 | return 0; |
| 93 | } |
| 94 | |
| 95 | static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits) |
| 96 | { |
| 97 | int i, j; |
| 98 | int n = 1 << nbits; |
| 99 | int n2 = n >> 1; |
| 100 | |
| 101 | for (i = 0; i < n; i++) { |
| 102 | double tmp_re = 0, tmp_im = 0; |
| 103 | FFTComplex *q = tab; |
| 104 | for (j = 0; j < n; j++) { |
| 105 | double s, c; |
| 106 | int k = (i * j) & (n - 1); |
| 107 | if (k >= n2) { |
| 108 | c = -exptab[k - n2].re; |
| 109 | s = -exptab[k - n2].im; |
| 110 | } else { |
| 111 | c = exptab[k].re; |
| 112 | s = exptab[k].im; |
| 113 | } |
| 114 | CMAC(tmp_re, tmp_im, c, s, q->re, q->im); |
| 115 | q++; |
| 116 | } |
| 117 | tabr[i].re = REF_SCALE(tmp_re, nbits); |
| 118 | tabr[i].im = REF_SCALE(tmp_im, nbits); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | #if CONFIG_MDCT |
| 123 | static void imdct_ref(FFTSample *out, FFTSample *in, int nbits) |
| 124 | { |
| 125 | int i, k, n = 1 << nbits; |
| 126 | |
| 127 | for (i = 0; i < n; i++) { |
| 128 | double sum = 0; |
| 129 | for (k = 0; k < n / 2; k++) { |
| 130 | int a = (2 * i + 1 + (n / 2)) * (2 * k + 1); |
| 131 | double f = cos(M_PI * a / (double) (2 * n)); |
| 132 | sum += f * in[k]; |
| 133 | } |
| 134 | out[i] = REF_SCALE(-sum, nbits - 2); |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | /* NOTE: no normalisation by 1 / N is done */ |
| 139 | static void mdct_ref(FFTSample *output, FFTSample *input, int nbits) |
| 140 | { |
| 141 | int i, k, n = 1 << nbits; |
| 142 | |
| 143 | /* do it by hand */ |
| 144 | for (k = 0; k < n / 2; k++) { |
| 145 | double s = 0; |
| 146 | for (i = 0; i < n; i++) { |
| 147 | double a = (2 * M_PI * (2 * i + 1 + n / 2) * (2 * k + 1) / (4 * n)); |
| 148 | s += input[i] * cos(a); |
| 149 | } |
| 150 | output[k] = REF_SCALE(s, nbits - 1); |
| 151 | } |
| 152 | } |
| 153 | #endif /* CONFIG_MDCT */ |
| 154 | |
| 155 | #if FFT_FLOAT |
| 156 | #if CONFIG_DCT |
| 157 | static void idct_ref(FFTSample *output, FFTSample *input, int nbits) |
| 158 | { |
| 159 | int i, k, n = 1 << nbits; |
| 160 | |
| 161 | /* do it by hand */ |
| 162 | for (i = 0; i < n; i++) { |
| 163 | double s = 0.5 * input[0]; |
| 164 | for (k = 1; k < n; k++) { |
| 165 | double a = M_PI * k * (i + 0.5) / n; |
| 166 | s += input[k] * cos(a); |
| 167 | } |
| 168 | output[i] = 2 * s / n; |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | static void dct_ref(FFTSample *output, FFTSample *input, int nbits) |
| 173 | { |
| 174 | int i, k, n = 1 << nbits; |
| 175 | |
| 176 | /* do it by hand */ |
| 177 | for (k = 0; k < n; k++) { |
| 178 | double s = 0; |
| 179 | for (i = 0; i < n; i++) { |
| 180 | double a = M_PI * k * (i + 0.5) / n; |
| 181 | s += input[i] * cos(a); |
| 182 | } |
| 183 | output[k] = s; |
| 184 | } |
| 185 | } |
| 186 | #endif /* CONFIG_DCT */ |
| 187 | #endif /* FFT_FLOAT */ |
| 188 | |
| 189 | static FFTSample frandom(AVLFG *prng) |
| 190 | { |
| 191 | return (int16_t) av_lfg_get(prng) / 32768.0 * RANGE; |
| 192 | } |
| 193 | |
| 194 | static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale) |
| 195 | { |
| 196 | int i, err = 0; |
| 197 | double error = 0, max = 0; |
| 198 | |
| 199 | for (i = 0; i < n; i++) { |
| 200 | double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE; |
| 201 | if (e >= 1e-3) { |
| 202 | av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n", |
| 203 | i, tab1[i], tab2[i]); |
| 204 | err = 1; |
| 205 | } |
| 206 | error += e * e; |
| 207 | if (e > max) |
| 208 | max = e; |
| 209 | } |
| 210 | av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error / n)); |
| 211 | return err; |
| 212 | } |
| 213 | |
| 214 | static void help(void) |
| 215 | { |
| 216 | av_log(NULL, AV_LOG_INFO, |
| 217 | "usage: fft-test [-h] [-s] [-i] [-n b]\n" |
| 218 | "-h print this help\n" |
| 219 | "-s speed test\n" |
| 220 | "-m (I)MDCT test\n" |
| 221 | "-d (I)DCT test\n" |
| 222 | "-r (I)RDFT test\n" |
| 223 | "-i inverse transform test\n" |
| 224 | "-n b set the transform size to 2^b\n" |
| 225 | "-f x set scale factor for output data of (I)MDCT to x\n"); |
| 226 | } |
| 227 | |
| 228 | enum tf_transform { |
| 229 | TRANSFORM_FFT, |
| 230 | TRANSFORM_MDCT, |
| 231 | TRANSFORM_RDFT, |
| 232 | TRANSFORM_DCT, |
| 233 | }; |
| 234 | |
| 235 | #if !HAVE_GETOPT |
| 236 | #include "compat/getopt.c" |
| 237 | #endif |
| 238 | |
| 239 | int main(int argc, char **argv) |
| 240 | { |
| 241 | FFTComplex *tab, *tab1, *tab_ref; |
| 242 | FFTSample *tab2; |
| 243 | enum tf_transform transform = TRANSFORM_FFT; |
| 244 | FFTContext m, s; |
| 245 | #if FFT_FLOAT |
| 246 | RDFTContext r; |
| 247 | DCTContext d; |
| 248 | #endif /* FFT_FLOAT */ |
| 249 | int it, i, err = 1; |
| 250 | int do_speed = 0, do_inverse = 0; |
| 251 | int fft_nbits = 9, fft_size; |
| 252 | double scale = 1.0; |
| 253 | AVLFG prng; |
| 254 | |
| 255 | av_lfg_init(&prng, 1); |
| 256 | |
| 257 | for (;;) { |
| 258 | int c = getopt(argc, argv, "hsimrdn:f:c:"); |
| 259 | if (c == -1) |
| 260 | break; |
| 261 | switch (c) { |
| 262 | case 'h': |
| 263 | help(); |
| 264 | return 1; |
| 265 | case 's': |
| 266 | do_speed = 1; |
| 267 | break; |
| 268 | case 'i': |
| 269 | do_inverse = 1; |
| 270 | break; |
| 271 | case 'm': |
| 272 | transform = TRANSFORM_MDCT; |
| 273 | break; |
| 274 | case 'r': |
| 275 | transform = TRANSFORM_RDFT; |
| 276 | break; |
| 277 | case 'd': |
| 278 | transform = TRANSFORM_DCT; |
| 279 | break; |
| 280 | case 'n': |
| 281 | fft_nbits = atoi(optarg); |
| 282 | break; |
| 283 | case 'f': |
| 284 | scale = atof(optarg); |
| 285 | break; |
| 286 | case 'c': |
| 287 | { |
| 288 | int cpuflags = av_get_cpu_flags(); |
| 289 | |
| 290 | if (av_parse_cpu_caps(&cpuflags, optarg) < 0) |
| 291 | return 1; |
| 292 | |
| 293 | av_force_cpu_flags(cpuflags); |
| 294 | break; |
| 295 | } |
| 296 | } |
| 297 | } |
| 298 | |
| 299 | fft_size = 1 << fft_nbits; |
| 300 | tab = av_malloc_array(fft_size, sizeof(FFTComplex)); |
| 301 | tab1 = av_malloc_array(fft_size, sizeof(FFTComplex)); |
| 302 | tab_ref = av_malloc_array(fft_size, sizeof(FFTComplex)); |
| 303 | tab2 = av_malloc_array(fft_size, sizeof(FFTSample)); |
| 304 | |
| 305 | if (!(tab && tab1 && tab_ref && tab2)) |
| 306 | goto cleanup; |
| 307 | |
| 308 | switch (transform) { |
| 309 | #if CONFIG_MDCT |
| 310 | case TRANSFORM_MDCT: |
| 311 | av_log(NULL, AV_LOG_INFO, "Scale factor is set to %f\n", scale); |
| 312 | if (do_inverse) |
| 313 | av_log(NULL, AV_LOG_INFO, "IMDCT"); |
| 314 | else |
| 315 | av_log(NULL, AV_LOG_INFO, "MDCT"); |
| 316 | ff_mdct_init(&m, fft_nbits, do_inverse, scale); |
| 317 | break; |
| 318 | #endif /* CONFIG_MDCT */ |
| 319 | case TRANSFORM_FFT: |
| 320 | if (do_inverse) |
| 321 | av_log(NULL, AV_LOG_INFO, "IFFT"); |
| 322 | else |
| 323 | av_log(NULL, AV_LOG_INFO, "FFT"); |
| 324 | ff_fft_init(&s, fft_nbits, do_inverse); |
| 325 | if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0) |
| 326 | goto cleanup; |
| 327 | break; |
| 328 | #if FFT_FLOAT |
| 329 | # if CONFIG_RDFT |
| 330 | case TRANSFORM_RDFT: |
| 331 | if (do_inverse) |
| 332 | av_log(NULL, AV_LOG_INFO, "IDFT_C2R"); |
| 333 | else |
| 334 | av_log(NULL, AV_LOG_INFO, "DFT_R2C"); |
| 335 | ff_rdft_init(&r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C); |
| 336 | if ((err = fft_ref_init(fft_nbits, do_inverse)) < 0) |
| 337 | goto cleanup; |
| 338 | break; |
| 339 | # endif /* CONFIG_RDFT */ |
| 340 | # if CONFIG_DCT |
| 341 | case TRANSFORM_DCT: |
| 342 | if (do_inverse) |
| 343 | av_log(NULL, AV_LOG_INFO, "DCT_III"); |
| 344 | else |
| 345 | av_log(NULL, AV_LOG_INFO, "DCT_II"); |
| 346 | ff_dct_init(&d, fft_nbits, do_inverse ? DCT_III : DCT_II); |
| 347 | break; |
| 348 | # endif /* CONFIG_DCT */ |
| 349 | #endif /* FFT_FLOAT */ |
| 350 | default: |
| 351 | av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n"); |
| 352 | goto cleanup; |
| 353 | } |
| 354 | av_log(NULL, AV_LOG_INFO, " %d test\n", fft_size); |
| 355 | |
| 356 | /* generate random data */ |
| 357 | |
| 358 | for (i = 0; i < fft_size; i++) { |
| 359 | tab1[i].re = frandom(&prng); |
| 360 | tab1[i].im = frandom(&prng); |
| 361 | } |
| 362 | |
| 363 | /* checking result */ |
| 364 | av_log(NULL, AV_LOG_INFO, "Checking...\n"); |
| 365 | |
| 366 | switch (transform) { |
| 367 | #if CONFIG_MDCT |
| 368 | case TRANSFORM_MDCT: |
| 369 | if (do_inverse) { |
| 370 | imdct_ref(&tab_ref->re, &tab1->re, fft_nbits); |
| 371 | m.imdct_calc(&m, tab2, &tab1->re); |
| 372 | err = check_diff(&tab_ref->re, tab2, fft_size, scale); |
| 373 | } else { |
| 374 | mdct_ref(&tab_ref->re, &tab1->re, fft_nbits); |
| 375 | m.mdct_calc(&m, tab2, &tab1->re); |
| 376 | err = check_diff(&tab_ref->re, tab2, fft_size / 2, scale); |
| 377 | } |
| 378 | break; |
| 379 | #endif /* CONFIG_MDCT */ |
| 380 | case TRANSFORM_FFT: |
| 381 | memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); |
| 382 | s.fft_permute(&s, tab); |
| 383 | s.fft_calc(&s, tab); |
| 384 | |
| 385 | fft_ref(tab_ref, tab1, fft_nbits); |
| 386 | err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 1.0); |
| 387 | break; |
| 388 | #if FFT_FLOAT |
| 389 | #if CONFIG_RDFT |
| 390 | case TRANSFORM_RDFT: |
| 391 | { |
| 392 | int fft_size_2 = fft_size >> 1; |
| 393 | if (do_inverse) { |
| 394 | tab1[0].im = 0; |
| 395 | tab1[fft_size_2].im = 0; |
| 396 | for (i = 1; i < fft_size_2; i++) { |
| 397 | tab1[fft_size_2 + i].re = tab1[fft_size_2 - i].re; |
| 398 | tab1[fft_size_2 + i].im = -tab1[fft_size_2 - i].im; |
| 399 | } |
| 400 | |
| 401 | memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); |
| 402 | tab2[1] = tab1[fft_size_2].re; |
| 403 | |
| 404 | r.rdft_calc(&r, tab2); |
| 405 | fft_ref(tab_ref, tab1, fft_nbits); |
| 406 | for (i = 0; i < fft_size; i++) { |
| 407 | tab[i].re = tab2[i]; |
| 408 | tab[i].im = 0; |
| 409 | } |
| 410 | err = check_diff(&tab_ref->re, &tab->re, fft_size * 2, 0.5); |
| 411 | } else { |
| 412 | for (i = 0; i < fft_size; i++) { |
| 413 | tab2[i] = tab1[i].re; |
| 414 | tab1[i].im = 0; |
| 415 | } |
| 416 | r.rdft_calc(&r, tab2); |
| 417 | fft_ref(tab_ref, tab1, fft_nbits); |
| 418 | tab_ref[0].im = tab_ref[fft_size_2].re; |
| 419 | err = check_diff(&tab_ref->re, tab2, fft_size, 1.0); |
| 420 | } |
| 421 | break; |
| 422 | } |
| 423 | #endif /* CONFIG_RDFT */ |
| 424 | #if CONFIG_DCT |
| 425 | case TRANSFORM_DCT: |
| 426 | memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); |
| 427 | d.dct_calc(&d, &tab->re); |
| 428 | if (do_inverse) |
| 429 | idct_ref(&tab_ref->re, &tab1->re, fft_nbits); |
| 430 | else |
| 431 | dct_ref(&tab_ref->re, &tab1->re, fft_nbits); |
| 432 | err = check_diff(&tab_ref->re, &tab->re, fft_size, 1.0); |
| 433 | break; |
| 434 | #endif /* CONFIG_DCT */ |
| 435 | #endif /* FFT_FLOAT */ |
| 436 | } |
| 437 | |
| 438 | /* do a speed test */ |
| 439 | |
| 440 | if (do_speed) { |
| 441 | int64_t time_start, duration; |
| 442 | int nb_its; |
| 443 | |
| 444 | av_log(NULL, AV_LOG_INFO, "Speed test...\n"); |
| 445 | /* we measure during about 1 seconds */ |
| 446 | nb_its = 1; |
| 447 | for (;;) { |
| 448 | time_start = av_gettime_relative(); |
| 449 | for (it = 0; it < nb_its; it++) { |
| 450 | switch (transform) { |
| 451 | case TRANSFORM_MDCT: |
| 452 | if (do_inverse) |
| 453 | m.imdct_calc(&m, &tab->re, &tab1->re); |
| 454 | else |
| 455 | m.mdct_calc(&m, &tab->re, &tab1->re); |
| 456 | break; |
| 457 | case TRANSFORM_FFT: |
| 458 | memcpy(tab, tab1, fft_size * sizeof(FFTComplex)); |
| 459 | s.fft_calc(&s, tab); |
| 460 | break; |
| 461 | #if FFT_FLOAT |
| 462 | case TRANSFORM_RDFT: |
| 463 | memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); |
| 464 | r.rdft_calc(&r, tab2); |
| 465 | break; |
| 466 | case TRANSFORM_DCT: |
| 467 | memcpy(tab2, tab1, fft_size * sizeof(FFTSample)); |
| 468 | d.dct_calc(&d, tab2); |
| 469 | break; |
| 470 | #endif /* FFT_FLOAT */ |
| 471 | } |
| 472 | } |
| 473 | duration = av_gettime_relative() - time_start; |
| 474 | if (duration >= 1000000) |
| 475 | break; |
| 476 | nb_its *= 2; |
| 477 | } |
| 478 | av_log(NULL, AV_LOG_INFO, |
| 479 | "time: %0.1f us/transform [total time=%0.2f s its=%d]\n", |
| 480 | (double) duration / nb_its, |
| 481 | (double) duration / 1000000.0, |
| 482 | nb_its); |
| 483 | } |
| 484 | |
| 485 | switch (transform) { |
| 486 | #if CONFIG_MDCT |
| 487 | case TRANSFORM_MDCT: |
| 488 | ff_mdct_end(&m); |
| 489 | break; |
| 490 | #endif /* CONFIG_MDCT */ |
| 491 | case TRANSFORM_FFT: |
| 492 | ff_fft_end(&s); |
| 493 | break; |
| 494 | #if FFT_FLOAT |
| 495 | # if CONFIG_RDFT |
| 496 | case TRANSFORM_RDFT: |
| 497 | ff_rdft_end(&r); |
| 498 | break; |
| 499 | # endif /* CONFIG_RDFT */ |
| 500 | # if CONFIG_DCT |
| 501 | case TRANSFORM_DCT: |
| 502 | ff_dct_end(&d); |
| 503 | break; |
| 504 | # endif /* CONFIG_DCT */ |
| 505 | #endif /* FFT_FLOAT */ |
| 506 | } |
| 507 | |
| 508 | cleanup: |
| 509 | av_free(tab); |
| 510 | av_free(tab1); |
| 511 | av_free(tab2); |
| 512 | av_free(tab_ref); |
| 513 | av_free(exptab); |
| 514 | |
| 515 | if (err) |
| 516 | printf("Error: %d.\n", err); |
| 517 | |
| 518 | return !!err; |
| 519 | } |