FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
fft-test.c
Go to the documentation of this file.
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 "libavutil/cpu.h"
27 #include "libavutil/mathematics.h"
28 #include "libavutil/lfg.h"
29 #include "libavutil/log.h"
30 #include "libavutil/time.h"
31 #include "fft.h"
32 #if CONFIG_FFT_FLOAT
33 #include "dct.h"
34 #include "rdft.h"
35 #endif
36 #include <math.h>
37 #if HAVE_UNISTD_H
38 #include <unistd.h>
39 #endif
40 #include <stdlib.h>
41 #include <string.h>
42 
43 /* reference fft */
44 
45 #define MUL16(a,b) ((a) * (b))
46 
47 #define CMAC(pre, pim, are, aim, bre, bim) \
48 {\
49  pre += (MUL16(are, bre) - MUL16(aim, bim));\
50  pim += (MUL16(are, bim) + MUL16(bre, aim));\
51 }
52 
53 #if CONFIG_FFT_FLOAT
54 # define RANGE 1.0
55 # define REF_SCALE(x, bits) (x)
56 # define FMT "%10.6f"
57 #else
58 # define RANGE 16384
59 # define REF_SCALE(x, bits) ((x) / (1<<(bits)))
60 # define FMT "%6d"
61 #endif
62 
63 struct {
64  float re, im;
65 } *exptab;
66 
67 static void fft_ref_init(int nbits, int inverse)
68 {
69  int n, i;
70  double c1, s1, alpha;
71 
72  n = 1 << nbits;
73  exptab = av_malloc((n / 2) * sizeof(*exptab));
74 
75  for (i = 0; i < (n/2); i++) {
76  alpha = 2 * M_PI * (float)i / (float)n;
77  c1 = cos(alpha);
78  s1 = sin(alpha);
79  if (!inverse)
80  s1 = -s1;
81  exptab[i].re = c1;
82  exptab[i].im = s1;
83  }
84 }
85 
86 static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
87 {
88  int n, i, j, k, n2;
89  double tmp_re, tmp_im, s, c;
90  FFTComplex *q;
91 
92  n = 1 << nbits;
93  n2 = n >> 1;
94  for (i = 0; i < n; i++) {
95  tmp_re = 0;
96  tmp_im = 0;
97  q = tab;
98  for (j = 0; j < n; j++) {
99  k = (i * j) & (n - 1);
100  if (k >= n2) {
101  c = -exptab[k - n2].re;
102  s = -exptab[k - n2].im;
103  } else {
104  c = exptab[k].re;
105  s = exptab[k].im;
106  }
107  CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
108  q++;
109  }
110  tabr[i].re = REF_SCALE(tmp_re, nbits);
111  tabr[i].im = REF_SCALE(tmp_im, nbits);
112  }
113 }
114 
115 static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
116 {
117  int n = 1<<nbits;
118  int k, i, a;
119  double sum, f;
120 
121  for (i = 0; i < n; i++) {
122  sum = 0;
123  for (k = 0; k < n/2; k++) {
124  a = (2 * i + 1 + (n / 2)) * (2 * k + 1);
125  f = cos(M_PI * a / (double)(2 * n));
126  sum += f * in[k];
127  }
128  out[i] = REF_SCALE(-sum, nbits - 2);
129  }
130 }
131 
132 /* NOTE: no normalisation by 1 / N is done */
133 static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
134 {
135  int n = 1<<nbits;
136  int k, i;
137  double a, s;
138 
139  /* do it by hand */
140  for (k = 0; k < n/2; k++) {
141  s = 0;
142  for (i = 0; i < n; i++) {
143  a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
144  s += input[i] * cos(a);
145  }
146  output[k] = REF_SCALE(s, nbits - 1);
147  }
148 }
149 
150 #if CONFIG_FFT_FLOAT
151 static void idct_ref(FFTSample *output, FFTSample *input, int nbits)
152 {
153  int n = 1<<nbits;
154  int k, i;
155  double a, s;
156 
157  /* do it by hand */
158  for (i = 0; i < n; i++) {
159  s = 0.5 * input[0];
160  for (k = 1; k < n; k++) {
161  a = M_PI*k*(i+0.5) / n;
162  s += input[k] * cos(a);
163  }
164  output[i] = 2 * s / n;
165  }
166 }
167 static void dct_ref(FFTSample *output, FFTSample *input, int nbits)
168 {
169  int n = 1<<nbits;
170  int k, i;
171  double a, s;
172 
173  /* do it by hand */
174  for (k = 0; k < n; k++) {
175  s = 0;
176  for (i = 0; i < n; i++) {
177  a = M_PI*k*(i+0.5) / n;
178  s += input[i] * cos(a);
179  }
180  output[k] = s;
181  }
182 }
183 #endif
184 
185 
186 static FFTSample frandom(AVLFG *prng)
187 {
188  return (int16_t)av_lfg_get(prng) / 32768.0 * RANGE;
189 }
190 
191 static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
192 {
193  int i;
194  double max= 0;
195  double error= 0;
196  int err = 0;
197 
198  for (i = 0; i < n; i++) {
199  double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE;
200  if (e >= 1e-3) {
201  av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
202  i, tab1[i], tab2[i]);
203  err = 1;
204  }
205  error+= e*e;
206  if(e>max) max= e;
207  }
208  av_log(NULL, AV_LOG_INFO, "max:%f e:%g\n", max, sqrt(error/n));
209  return err;
210 }
211 
212 
213 static void help(void)
214 {
215  av_log(NULL, AV_LOG_INFO,"usage: fft-test [-h] [-s] [-i] [-n b]\n"
216  "-h print this help\n"
217  "-s speed test\n"
218  "-m (I)MDCT test\n"
219  "-d (I)DCT test\n"
220  "-r (I)RDFT test\n"
221  "-i inverse transform test\n"
222  "-n b set the transform size to 2^b\n"
223  "-f x set scale factor for output data of (I)MDCT to x\n"
224  );
225 }
226 
232 };
233 
234 #if !HAVE_GETOPT
235 #include "compat/getopt.c"
236 #endif
237 
238 int main(int argc, char **argv)
239 {
240  FFTComplex *tab, *tab1, *tab_ref;
241  FFTSample *tab2;
242  int it, i, c;
243  int cpuflags;
244  int do_speed = 0;
245  int err = 1;
246  enum tf_transform transform = TRANSFORM_FFT;
247  int do_inverse = 0;
248  FFTContext s1, *s = &s1;
249  FFTContext m1, *m = &m1;
250 #if CONFIG_FFT_FLOAT
251  RDFTContext r1, *r = &r1;
252  DCTContext d1, *d = &d1;
253  int fft_size_2;
254 #endif
255  int fft_nbits, fft_size;
256  double scale = 1.0;
257  AVLFG prng;
258  av_lfg_init(&prng, 1);
259 
260  fft_nbits = 9;
261  for(;;) {
262  c = getopt(argc, argv, "hsimrdn:f:c:");
263  if (c == -1)
264  break;
265  switch(c) {
266  case 'h':
267  help();
268  return 1;
269  case 's':
270  do_speed = 1;
271  break;
272  case 'i':
273  do_inverse = 1;
274  break;
275  case 'm':
276  transform = TRANSFORM_MDCT;
277  break;
278  case 'r':
279  transform = TRANSFORM_RDFT;
280  break;
281  case 'd':
282  transform = TRANSFORM_DCT;
283  break;
284  case 'n':
285  fft_nbits = atoi(optarg);
286  break;
287  case 'f':
288  scale = atof(optarg);
289  break;
290  case 'c':
291  cpuflags = av_get_cpu_flags();
292 
293  if (av_parse_cpu_caps(&cpuflags, optarg) < 0)
294  return 1;
295 
296  av_force_cpu_flags(cpuflags);
297  break;
298  }
299  }
300 
301  fft_size = 1 << fft_nbits;
302  tab = av_malloc(fft_size * sizeof(FFTComplex));
303  tab1 = av_malloc(fft_size * sizeof(FFTComplex));
304  tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
305  tab2 = av_malloc(fft_size * sizeof(FFTSample));
306 
307  switch (transform) {
308  case TRANSFORM_MDCT:
309  av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale);
310  if (do_inverse)
311  av_log(NULL, AV_LOG_INFO,"IMDCT");
312  else
313  av_log(NULL, AV_LOG_INFO,"MDCT");
314  ff_mdct_init(m, fft_nbits, do_inverse, scale);
315  break;
316  case TRANSFORM_FFT:
317  if (do_inverse)
318  av_log(NULL, AV_LOG_INFO,"IFFT");
319  else
320  av_log(NULL, AV_LOG_INFO,"FFT");
321  ff_fft_init(s, fft_nbits, do_inverse);
322  fft_ref_init(fft_nbits, do_inverse);
323  break;
324 #if CONFIG_FFT_FLOAT
325  case TRANSFORM_RDFT:
326  if (do_inverse)
327  av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
328  else
329  av_log(NULL, AV_LOG_INFO,"DFT_R2C");
330  ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
331  fft_ref_init(fft_nbits, do_inverse);
332  break;
333  case TRANSFORM_DCT:
334  if (do_inverse)
335  av_log(NULL, AV_LOG_INFO,"DCT_III");
336  else
337  av_log(NULL, AV_LOG_INFO,"DCT_II");
338  ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
339  break;
340 #endif
341  default:
342  av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
343  return 1;
344  }
345  av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
346 
347  /* generate random data */
348 
349  for (i = 0; i < fft_size; i++) {
350  tab1[i].re = frandom(&prng);
351  tab1[i].im = frandom(&prng);
352  }
353 
354  /* checking result */
355  av_log(NULL, AV_LOG_INFO,"Checking...\n");
356 
357  switch (transform) {
358  case TRANSFORM_MDCT:
359  if (do_inverse) {
360  imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
361  m->imdct_calc(m, tab2, (FFTSample *)tab1);
362  err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale);
363  } else {
364  mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
365 
366  m->mdct_calc(m, tab2, (FFTSample *)tab1);
367 
368  err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale);
369  }
370  break;
371  case TRANSFORM_FFT:
372  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
373  s->fft_permute(s, tab);
374  s->fft_calc(s, tab);
375 
376  fft_ref(tab_ref, tab1, fft_nbits);
377  err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0);
378  break;
379 #if CONFIG_FFT_FLOAT
380  case TRANSFORM_RDFT:
381  fft_size_2 = fft_size >> 1;
382  if (do_inverse) {
383  tab1[ 0].im = 0;
384  tab1[fft_size_2].im = 0;
385  for (i = 1; i < fft_size_2; i++) {
386  tab1[fft_size_2+i].re = tab1[fft_size_2-i].re;
387  tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im;
388  }
389 
390  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
391  tab2[1] = tab1[fft_size_2].re;
392 
393  r->rdft_calc(r, tab2);
394  fft_ref(tab_ref, tab1, fft_nbits);
395  for (i = 0; i < fft_size; i++) {
396  tab[i].re = tab2[i];
397  tab[i].im = 0;
398  }
399  err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
400  } else {
401  for (i = 0; i < fft_size; i++) {
402  tab2[i] = tab1[i].re;
403  tab1[i].im = 0;
404  }
405  r->rdft_calc(r, tab2);
406  fft_ref(tab_ref, tab1, fft_nbits);
407  tab_ref[0].im = tab_ref[fft_size_2].re;
408  err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
409  }
410  break;
411  case TRANSFORM_DCT:
412  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
413  d->dct_calc(d, (FFTSample *)tab);
414  if (do_inverse) {
415  idct_ref((FFTSample*)tab_ref, (FFTSample *)tab1, fft_nbits);
416  } else {
417  dct_ref((FFTSample*)tab_ref, (FFTSample *)tab1, fft_nbits);
418  }
419  err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
420  break;
421 #endif
422  }
423 
424  /* do a speed test */
425 
426  if (do_speed) {
427  int64_t time_start, duration;
428  int nb_its;
429 
430  av_log(NULL, AV_LOG_INFO,"Speed test...\n");
431  /* we measure during about 1 seconds */
432  nb_its = 1;
433  for(;;) {
434  time_start = av_gettime();
435  for (it = 0; it < nb_its; it++) {
436  switch (transform) {
437  case TRANSFORM_MDCT:
438  if (do_inverse) {
439  m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
440  } else {
441  m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
442  }
443  break;
444  case TRANSFORM_FFT:
445  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
446  s->fft_calc(s, tab);
447  break;
448 #if CONFIG_FFT_FLOAT
449  case TRANSFORM_RDFT:
450  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
451  r->rdft_calc(r, tab2);
452  break;
453  case TRANSFORM_DCT:
454  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
455  d->dct_calc(d, tab2);
456  break;
457 #endif
458  }
459  }
460  duration = av_gettime() - time_start;
461  if (duration >= 1000000)
462  break;
463  nb_its *= 2;
464  }
465  av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
466  (double)duration / nb_its,
467  (double)duration / 1000000.0,
468  nb_its);
469  }
470 
471  switch (transform) {
472  case TRANSFORM_MDCT:
473  ff_mdct_end(m);
474  break;
475  case TRANSFORM_FFT:
476  ff_fft_end(s);
477  break;
478 #if CONFIG_FFT_FLOAT
479  case TRANSFORM_RDFT:
480  ff_rdft_end(r);
481  break;
482  case TRANSFORM_DCT:
483  ff_dct_end(d);
484  break;
485 #endif
486  }
487 
488  av_free(tab);
489  av_free(tab1);
490  av_free(tab2);
491  av_free(tab_ref);
492  av_free(exptab);
493 
494  return err;
495 }