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"
44 #define MUL16(a,b) ((a) * (b))
46 #define CMAC(pre, pim, are, aim, bre, bim) \
48 pre += (MUL16(are, bre) - MUL16(aim, bim));\
49 pim += (MUL16(are, bim) + MUL16(bre, aim));\
54 # define REF_SCALE(x, bits) (x)
58 # define REF_SCALE(x, bits) ((x) / (1<<(bits)))
66 static void fft_ref_init(int nbits, int inverse)
72 exptab = av_malloc((n / 2) * sizeof(*exptab));
74 for (i = 0; i < (n/2); i++) {
75 alpha = 2 * M_PI * (float)i / (float)n;
85 static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
88 double tmp_re, tmp_im, s, c;
93 for (i = 0; i < n; i++) {
97 for (j = 0; j < n; j++) {
98 k = (i * j) & (n - 1);
100 c = -exptab[k - n2].re;
101 s = -exptab[k - n2].im;
106 CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
109 tabr[i].re = REF_SCALE(tmp_re, nbits);
110 tabr[i].im = REF_SCALE(tmp_im, nbits);
114 static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
120 for (i = 0; i < n; i++) {
122 for (k = 0; k < n/2; k++) {
123 a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
124 f = cos(M_PI * a / (double)(2 * n));
127 out[i] = REF_SCALE(-sum, nbits - 2);
131 /* NOTE: no normalisation by 1 / N is done */
132 static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
139 for (k = 0; k < n/2; k++) {
141 for (i = 0; i < n; i++) {
142 a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
143 s += input[i] * cos(a);
145 output[k] = REF_SCALE(s, nbits - 1);
150 static void idct_ref(float *output, float *input, int nbits)
157 for (i = 0; i < n; i++) {
159 for (k = 1; k < n; k++) {
160 a = M_PI*k*(i+0.5) / n;
161 s += input[k] * cos(a);
163 output[i] = 2 * s / n;
166 static void dct_ref(float *output, float *input, int nbits)
173 for (k = 0; k < n; k++) {
175 for (i = 0; i < n; i++) {
176 a = M_PI*k*(i+0.5) / n;
177 s += input[i] * cos(a);
185 static FFTSample frandom(AVLFG *prng)
187 return (int16_t)av_lfg_get(prng) / 32768.0 * RANGE;
190 static int64_t gettime(void)
193 gettimeofday(&tv,NULL);
194 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
197 static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
204 for (i = 0; i < n; i++) {
205 double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE;
207 av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
208 i, tab1[i], tab2[i]);
214 av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error)/n);
219 static void help(void)
221 av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
222 "-h print this help\n"
227 "-i inverse transform test\n"
228 "-n b set the transform size to 2^b\n"
229 "-f x set scale factor for output data of (I)MDCT to x\n"
241 int main(int argc, char **argv)
243 FFTComplex *tab, *tab1, *tab_ref;
248 enum tf_transform transform = TRANSFORM_FFT;
250 FFTContext s1, *s = &s1;
251 FFTContext m1, *m = &m1;
253 RDFTContext r1, *r = &r1;
254 DCTContext d1, *d = &d1;
256 int fft_nbits, fft_size, fft_size_2;
259 av_lfg_init(&prng, 1);
263 c = getopt(argc, argv, "hsimrdn:f:");
277 transform = TRANSFORM_MDCT;
280 transform = TRANSFORM_RDFT;
283 transform = TRANSFORM_DCT;
286 fft_nbits = atoi(optarg);
289 scale = atof(optarg);
294 fft_size = 1 << fft_nbits;
295 fft_size_2 = fft_size >> 1;
296 tab = av_malloc(fft_size * sizeof(FFTComplex));
297 tab1 = av_malloc(fft_size * sizeof(FFTComplex));
298 tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
299 tab2 = av_malloc(fft_size * sizeof(FFTSample));
303 av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale);
305 av_log(NULL, AV_LOG_INFO,"IMDCT");
307 av_log(NULL, AV_LOG_INFO,"MDCT");
308 ff_mdct_init(m, fft_nbits, do_inverse, scale);
312 av_log(NULL, AV_LOG_INFO,"IFFT");
314 av_log(NULL, AV_LOG_INFO,"FFT");
315 ff_fft_init(s, fft_nbits, do_inverse);
316 fft_ref_init(fft_nbits, do_inverse);
321 av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
323 av_log(NULL, AV_LOG_INFO,"DFT_R2C");
324 ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
325 fft_ref_init(fft_nbits, do_inverse);
329 av_log(NULL, AV_LOG_INFO,"DCT_III");
331 av_log(NULL, AV_LOG_INFO,"DCT_II");
332 ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
336 av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
339 av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
341 /* generate random data */
343 for (i = 0; i < fft_size; i++) {
344 tab1[i].re = frandom(&prng);
345 tab1[i].im = frandom(&prng);
348 /* checking result */
349 av_log(NULL, AV_LOG_INFO,"Checking...\n");
354 imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
355 m->imdct_calc(m, tab2, (FFTSample *)tab1);
356 err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale);
358 mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
360 m->mdct_calc(m, tab2, (FFTSample *)tab1);
362 err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale);
366 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
367 s->fft_permute(s, tab);
370 fft_ref(tab_ref, tab1, fft_nbits);
371 err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0);
377 tab1[fft_size_2].im = 0;
378 for (i = 1; i < fft_size_2; i++) {
379 tab1[fft_size_2+i].re = tab1[fft_size_2-i].re;
380 tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im;
383 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
384 tab2[1] = tab1[fft_size_2].re;
386 r->rdft_calc(r, tab2);
387 fft_ref(tab_ref, tab1, fft_nbits);
388 for (i = 0; i < fft_size; i++) {
392 err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
394 for (i = 0; i < fft_size; i++) {
395 tab2[i] = tab1[i].re;
398 r->rdft_calc(r, tab2);
399 fft_ref(tab_ref, tab1, fft_nbits);
400 tab_ref[0].im = tab_ref[fft_size_2].re;
401 err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
405 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
408 idct_ref(tab_ref, tab1, fft_nbits);
410 dct_ref(tab_ref, tab1, fft_nbits);
412 err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
417 /* do a speed test */
420 int64_t time_start, duration;
423 av_log(NULL, AV_LOG_INFO,"Speed test...\n");
424 /* we measure during about 1 seconds */
427 time_start = gettime();
428 for (it = 0; it < nb_its; it++) {
432 m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
434 m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
438 memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
443 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
444 r->rdft_calc(r, tab2);
447 memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
448 d->dct_calc(d, tab2);
453 duration = gettime() - time_start;
454 if (duration >= 1000000)
458 av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
459 (double)duration / nb_its,
460 (double)duration / 1000000.0,