fft-test.c
Go to the documentation of this file.
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 
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(float *output, float *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(float *output, float *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_parse_cpu_flags(optarg);
292  if (cpuflags < 0)
293  return 1;
294  av_set_cpu_flags_mask(cpuflags);
295  break;
296  }
297  }
298 
299  fft_size = 1 << fft_nbits;
300  tab = av_malloc(fft_size * sizeof(FFTComplex));
301  tab1 = av_malloc(fft_size * sizeof(FFTComplex));
302  tab_ref = av_malloc(fft_size * sizeof(FFTComplex));
303  tab2 = av_malloc(fft_size * sizeof(FFTSample));
304 
305  switch (transform) {
306  case TRANSFORM_MDCT:
307  av_log(NULL, AV_LOG_INFO,"Scale factor is set to %f\n", scale);
308  if (do_inverse)
309  av_log(NULL, AV_LOG_INFO,"IMDCT");
310  else
311  av_log(NULL, AV_LOG_INFO,"MDCT");
312  ff_mdct_init(m, fft_nbits, do_inverse, scale);
313  break;
314  case TRANSFORM_FFT:
315  if (do_inverse)
316  av_log(NULL, AV_LOG_INFO,"IFFT");
317  else
318  av_log(NULL, AV_LOG_INFO,"FFT");
319  ff_fft_init(s, fft_nbits, do_inverse);
320  fft_ref_init(fft_nbits, do_inverse);
321  break;
322 #if CONFIG_FFT_FLOAT
323  case TRANSFORM_RDFT:
324  if (do_inverse)
325  av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
326  else
327  av_log(NULL, AV_LOG_INFO,"DFT_R2C");
328  ff_rdft_init(r, fft_nbits, do_inverse ? IDFT_C2R : DFT_R2C);
329  fft_ref_init(fft_nbits, do_inverse);
330  break;
331  case TRANSFORM_DCT:
332  if (do_inverse)
333  av_log(NULL, AV_LOG_INFO,"DCT_III");
334  else
335  av_log(NULL, AV_LOG_INFO,"DCT_II");
336  ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
337  break;
338 #endif
339  default:
340  av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
341  return 1;
342  }
343  av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
344 
345  /* generate random data */
346 
347  for (i = 0; i < fft_size; i++) {
348  tab1[i].re = frandom(&prng);
349  tab1[i].im = frandom(&prng);
350  }
351 
352  /* checking result */
353  av_log(NULL, AV_LOG_INFO,"Checking...\n");
354 
355  switch (transform) {
356  case TRANSFORM_MDCT:
357  if (do_inverse) {
358  imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
359  m->imdct_calc(m, tab2, (FFTSample *)tab1);
360  err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale);
361  } else {
362  mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
363 
364  m->mdct_calc(m, tab2, (FFTSample *)tab1);
365 
366  err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale);
367  }
368  break;
369  case TRANSFORM_FFT:
370  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
371  s->fft_permute(s, tab);
372  s->fft_calc(s, tab);
373 
374  fft_ref(tab_ref, tab1, fft_nbits);
375  err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0);
376  break;
377 #if CONFIG_FFT_FLOAT
378  case TRANSFORM_RDFT:
379  fft_size_2 = fft_size >> 1;
380  if (do_inverse) {
381  tab1[ 0].im = 0;
382  tab1[fft_size_2].im = 0;
383  for (i = 1; i < fft_size_2; i++) {
384  tab1[fft_size_2+i].re = tab1[fft_size_2-i].re;
385  tab1[fft_size_2+i].im = -tab1[fft_size_2-i].im;
386  }
387 
388  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
389  tab2[1] = tab1[fft_size_2].re;
390 
391  r->rdft_calc(r, tab2);
392  fft_ref(tab_ref, tab1, fft_nbits);
393  for (i = 0; i < fft_size; i++) {
394  tab[i].re = tab2[i];
395  tab[i].im = 0;
396  }
397  err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 0.5);
398  } else {
399  for (i = 0; i < fft_size; i++) {
400  tab2[i] = tab1[i].re;
401  tab1[i].im = 0;
402  }
403  r->rdft_calc(r, tab2);
404  fft_ref(tab_ref, tab1, fft_nbits);
405  tab_ref[0].im = tab_ref[fft_size_2].re;
406  err = check_diff((float *)tab_ref, (float *)tab2, fft_size, 1.0);
407  }
408  break;
409  case TRANSFORM_DCT:
410  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
411  d->dct_calc(d, tab);
412  if (do_inverse) {
413  idct_ref(tab_ref, tab1, fft_nbits);
414  } else {
415  dct_ref(tab_ref, tab1, fft_nbits);
416  }
417  err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
418  break;
419 #endif
420  }
421 
422  /* do a speed test */
423 
424  if (do_speed) {
425  int64_t time_start, duration;
426  int nb_its;
427 
428  av_log(NULL, AV_LOG_INFO,"Speed test...\n");
429  /* we measure during about 1 seconds */
430  nb_its = 1;
431  for(;;) {
432  time_start = av_gettime();
433  for (it = 0; it < nb_its; it++) {
434  switch (transform) {
435  case TRANSFORM_MDCT:
436  if (do_inverse) {
437  m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
438  } else {
439  m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
440  }
441  break;
442  case TRANSFORM_FFT:
443  memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
444  s->fft_calc(s, tab);
445  break;
446 #if CONFIG_FFT_FLOAT
447  case TRANSFORM_RDFT:
448  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
449  r->rdft_calc(r, tab2);
450  break;
451  case TRANSFORM_DCT:
452  memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
453  d->dct_calc(d, tab2);
454  break;
455 #endif
456  }
457  }
458  duration = av_gettime() - time_start;
459  if (duration >= 1000000)
460  break;
461  nb_its *= 2;
462  }
463  av_log(NULL, AV_LOG_INFO,"time: %0.1f us/transform [total time=%0.2f s its=%d]\n",
464  (double)duration / nb_its,
465  (double)duration / 1000000.0,
466  nb_its);
467  }
468 
469  switch (transform) {
470  case TRANSFORM_MDCT:
471  ff_mdct_end(m);
472  break;
473  case TRANSFORM_FFT:
474  ff_fft_end(s);
475  break;
476 #if CONFIG_FFT_FLOAT
477  case TRANSFORM_RDFT:
478  ff_rdft_end(r);
479  break;
480  case TRANSFORM_DCT:
481  ff_dct_end(d);
482  break;
483 #endif
484  }
485 
486  av_free(tab);
487  av_free(tab1);
488  av_free(tab2);
489  av_free(tab_ref);
490  av_free(exptab);
491 
492  return err;
493 }
Definition: lfg.h:25
av_cold void ff_rdft_end(RDFTContext *s)
Definition: rdft.c:130
void * av_malloc(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
Definition: mem.c:61
void(* dct_calc)(struct DCTContext *s, FFTSample *data)
Definition: dct.h:35
void(* mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input)
Definition: fft.h:83
Definition: avfft.h:95
void av_set_cpu_flags_mask(int mask)
Set a mask on flags returned by av_get_cpu_flags().
Definition: cpu.c:42
void(* fft_permute)(struct FFTContext *s, FFTComplex *z)
Do the permutation needed BEFORE calling fft_calc().
Definition: fft.h:75
FFTSample re
Definition: avfft.h:38
static int64_t duration
Definition: avplay.c:249
static void idct_ref(float *output, float *input, int nbits)
Definition: fft-test.c:151
tf_transform
Definition: fft-test.c:227
static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
Definition: fft-test.c:86
static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
Definition: fft-test.c:115
int av_parse_cpu_flags(const char *s)
Parse CPU flags from a string.
Definition: cpu.c:48
static void help(void)
Definition: fft-test.c:213
#define r
Definition: input.c:51
void av_free(void *ptr)
Free a memory block which has been allocated with av_malloc(z)() or av_realloc(). ...
Definition: mem.c:139
struct @17 * exptab
static FFTSample frandom(AVLFG *prng)
Definition: fft-test.c:186
void av_log(void *avcl, int level, const char *fmt,...)
Definition: log.c:146
Definition: avfft.h:73
#define ff_mdct_init
Definition: fft.h:146
float FFTSample
Definition: avfft.h:35
void(* rdft_calc)(struct RDFTContext *s, FFTSample *z)
Definition: rdft.h:60
void(* imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input)
Definition: fft.h:81
Definition: fft.h:62
static void fft_ref_init(int nbits, int inverse)
Definition: fft-test.c:67
#define CMAC(pre, pim, are, aim, bre, bim)
Definition: fft-test.c:47
#define ff_fft_init
Definition: fft.h:126
Definition: dct.h:29
Definition: avfft.h:72
int64_t av_gettime(void)
Get the current time in microseconds.
Definition: time.c:37
#define REF_SCALE(x, bits)
Definition: fft-test.c:55
NULL
Definition: eval.c:52
float im
Definition: fft-test.c:64
static void dct_ref(float *output, float *input, int nbits)
Definition: fft-test.c:167
static int getopt(int argc, char *argv[], char *opts)
Definition: getopt.c:43
static unsigned int av_lfg_get(AVLFG *c)
Get the next random unsigned 32-bit number using an ALFG.
Definition: lfg.h:38
const int16_t * tab1
Definition: mace.c:144
av_cold int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType inverse)
Set up DCT.
Definition: dct.c:177
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:30
#define s1
Definition: regdef.h:38
#define FMT
Definition: fft-test.c:56
static const uint16_t scale[4]
#define RANGE
Definition: fft-test.c:54
int main(int argc, char **argv)
Definition: fft-test.c:238
FFTSample im
Definition: avfft.h:38
#define ff_mdct_end
Definition: fft.h:147
#define ff_fft_end
Definition: fft.h:127
void(* fft_calc)(struct FFTContext *s, FFTComplex *z)
Do a complex FFT with the parameters defined in ff_fft_init().
Definition: fft.h:80
Definition: avfft.h:94
static char * optarg
Definition: getopt.c:39
static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
Definition: fft-test.c:191
float re
Definition: fft-test.c:64
av_cold void ff_dct_end(DCTContext *s)
Definition: dct.c:218
static const struct twinvq_data tab
static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
Definition: fft-test.c:133
static uint32_t inverse(uint32_t v)
find multiplicative inverse modulo 2 ^ 32
Definition: asfcrypt.c:35
av_cold int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans)
Set up a real FFT.
Definition: rdft.c:99
const int16_t * tab2
Definition: mace.c:144
#define c1
Definition: idct_sh4.c:27