2 * (c) 2002 Fabrice Bellard
4 * This file is part of FFmpeg.
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.
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.
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
26 #include "libavutil/mathematics.h"
27 #include "libavutil/lfg.h"
28 #include "libavutil/log.h"
42 #define MUL16(a,b) ((a) * (b))
44 #define CMAC(pre, pim, are, aim, bre, bim) \
46 pre += (MUL16(are, bre) - MUL16(aim, bim));\
47 pim += (MUL16(are, bim) + MUL16(bre, aim));\
52 static void fft_ref_init(int nbits, int inverse)
58 exptab = av_malloc((n / 2) * sizeof(FFTComplex));
60 for (i = 0; i < (n/2); i++) {
61 alpha = 2 * M_PI * (float)i / (float)n;
71 static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
74 double tmp_re, tmp_im, s, c;
79 for (i = 0; i < n; i++) {
83 for (j = 0; j < n; j++) {
84 k = (i * j) & (n - 1);
86 c = -exptab[k - n2].re;
87 s = -exptab[k - n2].im;
92 CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
100 static void imdct_ref(float *out, float *in, int nbits)
106 for (i = 0; i < n; i++) {
108 for (k = 0; k < n/2; k++) {
109 a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
110 f = cos(M_PI * a / (double)(2 * n));
117 /* NOTE: no normalisation by 1 / N is done */
118 static void mdct_ref(float *output, float *input, int nbits)
125 for (k = 0; k < n/2; k++) {
127 for (i = 0; i < n; i++) {
128 a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
129 s += input[i] * cos(a);
135 static void idct_ref(float *output, float *input, int nbits)
142 for (i = 0; i < n; i++) {
144 for (k = 1; k < n; k++) {
145 a = M_PI*k*(i+0.5) / n;
146 s += input[k] * cos(a);
148 output[i] = 2 * s / n;
151 static void dct_ref(float *output, float *input, int nbits)
158 for (k = 0; k < n; k++) {
160 for (i = 0; i < n; i++) {
161 a = M_PI*k*(i+0.5) / n;
162 s += input[i] * cos(a);
169 static float frandom(AVLFG *prng)
171 return (int16_t)av_lfg_get(prng) / 32768.0;
174 static int64_t gettime(void)
177 gettimeofday(&tv,NULL);
178 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
181 static int check_diff(float *tab1, float *tab2, int n, double scale)
188 for (i = 0; i < n; i++) {
189 double e= fabsf(tab1[i] - (tab2[i] / scale));
191 av_log(NULL, AV_LOG_ERROR, "ERROR %5d: %10.6f %10.6f\n",
192 i, tab1[i], tab2[i]);
198 av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error)/n);
203 static void help(void)
205 av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
206 "-h print this help\n"
211 "-i inverse transform test\n"
212 "-n b set the transform size to 2^b\n"
213 "-f x set scale factor for output data of (I)MDCT to x\n"
225 int main(int argc, char **argv)
227 FFTComplex *tab, *tab1, *tab_ref;
232 enum tf_transform transform = TRANSFORM_FFT;
234 FFTContext s1, *s = &s1;
235 FFTContext m1, *m = &m1;
236 RDFTContext r1, *r = &r1;
237 DCTContext d1, *d = &d1;
238 int fft_nbits, fft_size, fft_size_2;
241 av_lfg_init(&prng, 1);
245 c = getopt(argc, argv, "hsimrdn:f:");
259 transform = TRANSFORM_MDCT;
262 transform = TRANSFORM_RDFT;
265 transform = TRANSFORM_DCT;
268 fft_nbits = atoi(optarg);
271 scale = atof(optarg);
276 fft_size = 1 << fft_nbits;
277 fft_size_2 = fft_size >> 1;
278 tab = av_malloc(fft_size * sizeof(FFTComplex));
279 tab1 = av_malloc(fft_size * sizeof(FFTComplex));
280 tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
281 tab2 = av_malloc(fft_size * sizeof(FFTSample));
285 av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale);
287 av_log(NULL, AV_LOG_INFO,"IMDCT");
289 av_log(NULL, AV_LOG_INFO,"MDCT");
290 ff_mdct_init(m, fft_nbits, do_inverse, scale);
294 av_log(NULL, AV_LOG_INFO,"IFFT");
296 av_log(NULL, AV_LOG_INFO,"FFT");
297 ff_fft_init(s, fft_nbits, do_inverse);
298 fft_ref_init(fft_nbits, do_inverse);
302 av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
304 av_log(NULL, AV_LOG_INFO,"DFT_R2C");
305 ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
306 fft_ref_init(fft_nbits, do_inverse);
310 av_log(NULL, AV_LOG_INFO,"DCT_III");
312 av_log(NULL, AV_LOG_INFO,"DCT_II");
313 ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
316 av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
318 /* generate random data */
320 for (i = 0; i < fft_size; i++) {
321 tab1[i].re = frandom(&prng);
322 tab1[i].im = frandom(&prng);
325 /* checking result */
326 av_log(NULL, AV_LOG_INFO,"Checking...\n");
331 imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
332 m->imdct_calc(m, tab2, (float *)tab1);
333 err = check_diff((float *)tab_ref, tab2, fft_size, scale);
335 mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
337 m->mdct_calc(m, tab2, (float *)tab1);
339 err = check_diff((float *)tab_ref, tab2, fft_size / 2, scale);
343 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
344 s->fft_permute(s, tab);
347 fft_ref(tab_ref, tab1, fft_nbits);
348 err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 1.0);
353 tab1[fft_size_2].im = 0;
354 for (i = 1; i < fft_size_2; i++) {
355 tab1[fft_size_2+i].re = tab1[fft_size_2-i].re;
356 tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im;
359 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
360 tab2[1] = tab1[fft_size_2].re;
362 r->rdft_calc(r, tab2);
363 fft_ref(tab_ref, tab1, fft_nbits);
364 for (i = 0; i < fft_size; i++) {
368 err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
370 for (i = 0; i < fft_size; i++) {
371 tab2[i] = tab1[i].re;
374 r->rdft_calc(r, tab2);
375 fft_ref(tab_ref, tab1, fft_nbits);
376 tab_ref[0].im = tab_ref[fft_size_2].re;
377 err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
381 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
384 idct_ref(tab_ref, tab1, fft_nbits);
386 dct_ref(tab_ref, tab1, fft_nbits);
388 err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
392 /* do a speed test */
395 int64_t time_start, duration;
398 av_log(NULL, AV_LOG_INFO,"Speed test...\n");
399 /* we measure during about 1 seconds */
402 time_start = gettime();
403 for (it = 0; it < nb_its; it++) {
407 m->imdct_calc(m, (float *)tab, (float *)tab1);
409 m->mdct_calc(m, (float *)tab, (float *)tab1);
413 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
417 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
418 r->rdft_calc(r, tab2);
421 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
422 d->dct_calc(d, tab2);
426 duration = gettime() - time_start;
427 if (duration >= 1000000)
431 av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
432 (double)duration / nb_its,
433 (double)duration / 1000000.0,