FFmpeg  4.4
dct.c
Go to the documentation of this file.
1 /*
2  * (c) 2001 Fabrice Bellard
3  * 2007 Marc Hoffman <marc.hoffman@analog.com>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * DCT test (c) 2001 Fabrice Bellard
25  * Started from sample code by Juan J. Sierralta P.
26  */
27 
28 #include "config.h"
29 #include <stdlib.h>
30 #include <stdio.h>
31 #include <string.h>
32 #if HAVE_UNISTD_H
33 #include <unistd.h>
34 #endif
35 #include <math.h>
36 
37 #include "libavutil/cpu.h"
38 #include "libavutil/common.h"
39 #include "libavutil/internal.h"
40 #include "libavutil/lfg.h"
41 #include "libavutil/mem_internal.h"
42 #include "libavutil/time.h"
43 
44 #include "libavcodec/dct.h"
45 #include "libavcodec/idctdsp.h"
46 #include "libavcodec/simple_idct.h"
47 #include "libavcodec/xvididct.h"
48 #include "libavcodec/aandcttab.h"
49 #include "libavcodec/faandct.h"
50 #include "libavcodec/faanidct.h"
51 #include "libavcodec/dctref.h"
52 
53 struct algo {
54  const char *name;
55  void (*func)(int16_t *block);
57  int cpu_flag;
58  int nonspec;
59 };
60 
61 static const struct algo fdct_tab[] = {
62  { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
63  { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
64  { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
65 #if CONFIG_FAANDCT
66  { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
67 #endif /* CONFIG_FAANDCT */
68 };
69 
70 static void ff_prores_idct_wrap(int16_t *dst){
71  LOCAL_ALIGNED(16, int16_t, qmat, [64]);
72  int i;
73 
74  for(i=0; i<64; i++){
75  qmat[i]=4;
76  }
77  ff_prores_idct_10(dst, qmat);
78  for(i=0; i<64; i++) {
79  dst[i] -= 512;
80  }
81 }
82 
83 static const struct algo idct_tab[] = {
84  { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
88  { "SIMPLE-C12", ff_simple_idct_int16_12bit, FF_IDCT_PERM_NONE, 0, 1 },
89  { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 },
90 #if CONFIG_FAANIDCT
91  { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
92 #endif /* CONFIG_FAANIDCT */
93 #if CONFIG_MPEG4_DECODER
94  { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
95 #endif /* CONFIG_MPEG4_DECODER */
96 };
97 
98 #if ARCH_AARCH64
99 #include "aarch64/dct.c"
100 #elif ARCH_ARM
101 #include "arm/dct.c"
102 #elif ARCH_PPC
103 #include "ppc/dct.c"
104 #elif ARCH_X86
105 #include "x86/dct.c"
106 #else
107 static const struct algo fdct_tab_arch[] = { { 0 } };
108 static const struct algo idct_tab_arch[] = { { 0 } };
109 #endif
110 
111 #define AANSCALE_BITS 12
112 
113 #define NB_ITS 20000
114 #define NB_ITS_SPEED 50000
115 
116 DECLARE_ALIGNED(16, static int16_t, block)[64];
117 DECLARE_ALIGNED(8, static int16_t, block1)[64];
118 
119 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
120 {
121  int i, j;
122 
123  memset(block, 0, 64 * sizeof(*block));
124 
125  switch (test) {
126  case 0:
127  for (i = 0; i < 64; i++)
128  block[i] = (av_lfg_get(prng) % (2*vals)) -vals;
129  if (is_idct) {
131  for (i = 0; i < 64; i++)
132  block[i] >>= 3;
133  }
134  break;
135  case 1:
136  j = av_lfg_get(prng) % 10 + 1;
137  for (i = 0; i < j; i++) {
138  int idx = av_lfg_get(prng) % 64;
139  block[idx] = av_lfg_get(prng) % (2*vals) -vals;
140  }
141  break;
142  case 2:
143  block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);
144  block[63] = (block[0] & 1) ^ 1;
145  break;
146  }
147 }
148 
149 static void permute(int16_t dst[64], const int16_t src[64],
151 {
152  int i;
153 
154 #if ARCH_X86
155  if (permute_x86(dst, src, perm_type))
156  return;
157 #endif
158 
159  switch (perm_type) {
161  for (i = 0; i < 64; i++)
162  dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
163  break;
165  for (i = 0; i < 64; i++)
166  dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
167  break;
169  for (i = 0; i < 64; i++)
170  dst[(i>>3) | ((i<<3)&0x38)] = src[i];
171  break;
172  default:
173  for (i = 0; i < 64; i++)
174  dst[i] = src[i];
175  break;
176  }
177 }
178 
179 static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
180 {
181  void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
182  int it, i, scale;
183  int err_inf, v;
184  int64_t err2, ti, ti1, it1, err_sum = 0;
185  int64_t sysErr[64], sysErrMax = 0;
186  int64_t err2_matrix[64], err2_max = 0;
187  int maxout = 0;
188  int blockSumErrMax = 0, blockSumErr;
189  AVLFG prng;
190  const int vals=1<<bits;
191  double omse, ome;
192  int spec_err;
193 
194  av_lfg_init(&prng, 1);
195 
196  err_inf = 0;
197  err2 = 0;
198  for (i = 0; i < 64; i++)
199  err2_matrix[i] = sysErr[i] = 0;
200  for (it = 0; it < NB_ITS; it++) {
201  init_block(block1, test, is_idct, &prng, vals);
202  permute(block, block1, dct->perm_type);
203 
204  dct->func(block);
205  emms_c();
206 
207  if (!strcmp(dct->name, "IJG-AAN-INT")) {
208  for (i = 0; i < 64; i++) {
209  scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
210  block[i] = (block[i] * scale) >> AANSCALE_BITS;
211  }
212  }
213 
214  ref(block1);
215  if (!strcmp(dct->name, "PR-SSE2"))
216  for (i = 0; i < 64; i++)
217  block1[i] = av_clip(block1[i], 4-512, 1019-512);
218 
219  blockSumErr = 0;
220  for (i = 0; i < 64; i++) {
221  int err = block[i] - block1[i];
222  err_sum += err;
223  v = abs(err);
224  if (v > err_inf)
225  err_inf = v;
226  err2_matrix[i] += v * v;
227  err2 += v * v;
228  sysErr[i] += block[i] - block1[i];
229  blockSumErr += v;
230  if (abs(block[i]) > maxout)
231  maxout = abs(block[i]);
232  }
233  if (blockSumErrMax < blockSumErr)
234  blockSumErrMax = blockSumErr;
235  }
236  for (i = 0; i < 64; i++) {
237  sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
238  err2_max = FFMAX(err2_max , FFABS(err2_matrix[i]));
239  }
240 
241  for (i = 0; i < 64; i++) {
242  if (i % 8 == 0)
243  printf("\n");
244  printf("%7d ", (int) sysErr[i]);
245  }
246  printf("\n");
247 
248  omse = (double) err2 / NB_ITS / 64;
249  ome = (double) err_sum / NB_ITS / 64;
250 
251  spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
252  if (test < 2)
253  spec_err = is_idct && ((double) err2_max / NB_ITS > 0.06 || (double) sysErrMax / NB_ITS > 0.015);
254 
255  printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
256  is_idct ? "IDCT" : "DCT", dct->name, err_inf,
257  omse, ome, (double) sysErrMax / NB_ITS,
258  maxout, blockSumErrMax);
259 
260  if (spec_err && !dct->nonspec) {
261  printf("Failed!\n");
262  return 1;
263  }
264 
265  if (!speed)
266  return 0;
267 
268  /* speed test */
269 
270  init_block(block, test, is_idct, &prng, vals);
271  permute(block1, block, dct->perm_type);
272 
273  ti = av_gettime_relative();
274  it1 = 0;
275  do {
276  for (it = 0; it < NB_ITS_SPEED; it++) {
277  memcpy(block, block1, sizeof(block));
278  dct->func(block);
279  }
280  emms_c();
281  it1 += NB_ITS_SPEED;
282  ti1 = av_gettime_relative() - ti;
283  } while (ti1 < 1000000);
284 
285  printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
286  (double) it1 * 1000.0 / (double) ti1);
287 
288  return 0;
289 }
290 
293 
294 static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
295 {
296  static int init;
297  static double c8[8][8];
298  static double c4[4][4];
299  double block1[64], block2[64], block3[64];
300  double s, sum, v;
301  int i, j, k;
302 
303  if (!init) {
304  init = 1;
305 
306  for (i = 0; i < 8; i++) {
307  sum = 0;
308  for (j = 0; j < 8; j++) {
309  s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
310  c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
311  sum += c8[i][j] * c8[i][j];
312  }
313  }
314 
315  for (i = 0; i < 4; i++) {
316  sum = 0;
317  for (j = 0; j < 4; j++) {
318  s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
319  c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
320  sum += c4[i][j] * c4[i][j];
321  }
322  }
323  }
324 
325  /* butterfly */
326  s = 0.5 * sqrt(2.0);
327  for (i = 0; i < 4; i++) {
328  for (j = 0; j < 8; j++) {
329  block1[8 * (2 * i) + j] =
330  (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
331  block1[8 * (2 * i + 1) + j] =
332  (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
333  }
334  }
335 
336  /* idct8 on lines */
337  for (i = 0; i < 8; i++) {
338  for (j = 0; j < 8; j++) {
339  sum = 0;
340  for (k = 0; k < 8; k++)
341  sum += c8[k][j] * block1[8 * i + k];
342  block2[8 * i + j] = sum;
343  }
344  }
345 
346  /* idct4 */
347  for (i = 0; i < 8; i++) {
348  for (j = 0; j < 4; j++) {
349  /* top */
350  sum = 0;
351  for (k = 0; k < 4; k++)
352  sum += c4[k][j] * block2[8 * (2 * k) + i];
353  block3[8 * (2 * j) + i] = sum;
354 
355  /* bottom */
356  sum = 0;
357  for (k = 0; k < 4; k++)
358  sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
359  block3[8 * (2 * j + 1) + i] = sum;
360  }
361  }
362 
363  /* clamp and store the result */
364  for (i = 0; i < 8; i++) {
365  for (j = 0; j < 8; j++) {
366  v = block3[8 * i + j];
367  if (v < 0) v = 0;
368  else if (v > 255) v = 255;
369  dest[i * linesize + j] = (int) rint(v);
370  }
371  }
372 }
373 
374 static void idct248_error(const char *name,
375  void (*idct248_put)(uint8_t *dest,
376  ptrdiff_t line_size,
377  int16_t *block),
378  int speed)
379 {
380  int it, i, it1, ti, ti1, err_max, v;
381  AVLFG prng;
382 
383  av_lfg_init(&prng, 1);
384 
385  /* just one test to see if code is correct (precision is less
386  important here) */
387  err_max = 0;
388  for (it = 0; it < NB_ITS; it++) {
389  /* XXX: use forward transform to generate values */
390  for (i = 0; i < 64; i++)
391  block1[i] = av_lfg_get(&prng) % 256 - 128;
392  block1[0] += 1024;
393 
394  for (i = 0; i < 64; i++)
395  block[i] = block1[i];
397 
398  for (i = 0; i < 64; i++)
399  block[i] = block1[i];
400  idct248_put(img_dest, 8, block);
401 
402  for (i = 0; i < 64; i++) {
403  v = abs((int) img_dest[i] - (int) img_dest1[i]);
404  if (v == 255)
405  printf("%d %d\n", img_dest[i], img_dest1[i]);
406  if (v > err_max)
407  err_max = v;
408  }
409 #if 0
410  printf("ref=\n");
411  for(i=0;i<8;i++) {
412  int j;
413  for(j=0;j<8;j++) {
414  printf(" %3d", img_dest1[i*8+j]);
415  }
416  printf("\n");
417  }
418 
419  printf("out=\n");
420  for(i=0;i<8;i++) {
421  int j;
422  for(j=0;j<8;j++) {
423  printf(" %3d", img_dest[i*8+j]);
424  }
425  printf("\n");
426  }
427 #endif
428  }
429  printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
430 
431  if (!speed)
432  return;
433 
434  ti = av_gettime_relative();
435  it1 = 0;
436  do {
437  for (it = 0; it < NB_ITS_SPEED; it++) {
438  for (i = 0; i < 64; i++)
439  block[i] = block1[i];
440  idct248_put(img_dest, 8, block);
441  }
442  emms_c();
443  it1 += NB_ITS_SPEED;
444  ti1 = av_gettime_relative() - ti;
445  } while (ti1 < 1000000);
446 
447  printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
448  (double) it1 * 1000.0 / (double) ti1);
449 }
450 
451 static void help(void)
452 {
453  printf("dct-test [-i] [<test-number>] [<bits>]\n"
454  "test-number 0 -> test with random matrixes\n"
455  " 1 -> test with random sparse matrixes\n"
456  " 2 -> do 3. test from MPEG-4 std\n"
457  "bits Number of time domain bits to use, 8 is default\n"
458  "-i test IDCT implementations\n"
459  "-4 test IDCT248 implementations\n"
460  "-t speed test\n");
461 }
462 
463 #if !HAVE_GETOPT
464 #include "compat/getopt.c"
465 #endif
466 
467 int main(int argc, char **argv)
468 {
469  int test_idct = 0, test_248_dct = 0;
470  int c, i;
471  int test = 1;
472  int speed = 0;
473  int err = 0;
474  int bits=8;
475 
476  ff_ref_dct_init();
477 
478  for (;;) {
479  c = getopt(argc, argv, "ih4t");
480  if (c == -1)
481  break;
482  switch (c) {
483  case 'i':
484  test_idct = 1;
485  break;
486  case '4':
487  test_248_dct = 1;
488  break;
489  case 't':
490  speed = 1;
491  break;
492  default:
493  case 'h':
494  help();
495  return 0;
496  }
497  }
498 
499  if (optind < argc)
500  test = atoi(argv[optind]);
501  if(optind+1 < argc) bits= atoi(argv[optind+1]);
502 
503  printf("ffmpeg DCT/IDCT test\n");
504 
505  if (test_248_dct) {
506  idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
507  } else {
508  const int cpu_flags = av_get_cpu_flags();
509  if (test_idct) {
510  for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
511  err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);
512 
513  for (i = 0; idct_tab_arch[i].name; i++)
514  if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
515  err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);
516  }
517 #if CONFIG_FDCTDSP
518  else {
519  for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
520  err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);
521 
522  for (i = 0; fdct_tab_arch[i].name; i++)
523  if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
524  err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);
525  }
526 #endif /* CONFIG_FDCTDSP */
527  }
528 
529  if (err)
530  printf("Error: %d.\n", err);
531 
532  return !!err;
533 }
const uint16_t ff_aanscales[64]
Definition: aandcttab.c:26
AAN (Arai, Agui and Nakajima) (I)DCT tables.
static void dct(AudioRNNContext *s, float *out, const float *in)
Definition: af_arnndn.c:1024
uint8_t
static av_cold int init(AVCodecContext *avctx)
Definition: avrndec.c:31
#define s(width, name)
Definition: cbs_vp9.c:257
int cpu_flag
Definition: checkasm.c:282
common internal and external API header
#define av_clip
Definition: common.h:122
#define FFMAX(a, b)
Definition: common.h:103
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:72
static atomic_int cpu_flags
Definition: cpu.c:50
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:95
#define abs(x)
Definition: cuda_runtime.h:35
static __device__ float fabs(float a)
Definition: cuda_runtime.h:182
void ff_j_rev_dct(int16_t *data)
void ff_fdct_ifast(int16_t *data)
Definition: jfdctfst.c:208
void ff_jpeg_fdct_islow_8(int16_t *data)
av_cold void ff_ref_dct_init(void)
Initialize the double precision discrete cosine transform functions fdct & idct.
Definition: dctref.c:41
void ff_ref_idct(short *block)
Transform 8x8 block of data with a double precision inverse DCT This is a reference implementation.
Definition: dctref.c:95
void ff_ref_fdct(short *block)
Transform 8x8 block of data with a double precision forward DCT This is a reference implementation.
Definition: dctref.c:59
void ff_faandct(int16_t *data)
Definition: faandct.c:114
Floating point AAN DCT.
void ff_faanidct(int16_t block[64])
Definition: faanidct.c:127
int
static int getopt(int argc, char *argv[], char *opts)
Definition: getopt.c:41
static int optind
Definition: getopt.c:37
#define DECLARE_ALIGNED(n, t, v)
Declare a variable that is aligned in memory.
Definition: mem.h:117
idct_permutation_type
Definition: idctdsp.h:37
@ FF_IDCT_PERM_LIBMPEG2
Definition: idctdsp.h:39
@ FF_IDCT_PERM_NONE
Definition: idctdsp.h:38
@ FF_IDCT_PERM_PARTTRANS
Definition: idctdsp.h:42
@ FF_IDCT_PERM_TRANSPOSE
Definition: idctdsp.h:41
int i
Definition: input.c:407
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:32
static unsigned int av_lfg_get(AVLFG *c)
Get the next random unsigned 32-bit number using an ALFG.
Definition: lfg.h:53
common internal API header
#define emms_c()
Definition: internal.h:54
#define M_PI
Definition: mathematics.h:52
#define LOCAL_ALIGNED(a, t, v,...)
Definition: mem_internal.h:113
static void test(const char *pattern, const char *host)
Definition: noproxy.c:23
const char * name
Definition: qsvenc.c:46
typedef void(RENAME(mix_any_func_type))
void ff_prores_idct_10(int16_t *block, const int16_t *qmat)
Special version of ff_simple_idct_int16_10bit() which does dequantization and scales by a factor of 2...
Definition: simple_idct.c:240
void ff_simple_idct248_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
Definition: simple_idct.c:106
simple idct header.
void ff_simple_idct_int16_8bit(int16_t *block)
void ff_simple_idct_int16_12bit(int16_t *block)
void ff_simple_idct_int16_10bit(int16_t *block)
#define FF_ARRAY_ELEMS(a)
Context structure for the Lagged Fibonacci PRNG.
Definition: lfg.h:33
Definition: dct.c:53
enum idct_permutation_type perm_type
Definition: dct.c:56
int cpu_flag
Definition: dct.c:57
int nonspec
Definition: dct.c:58
void(* func)(int16_t *block)
Definition: dct.c:55
const char * name
Definition: dct.c:54
#define rint
Definition: tablegen.h:41
#define src
Definition: vp8dsp.c:255
static void idct248_error(const char *name, void(*idct248_put)(uint8_t *dest, ptrdiff_t line_size, int16_t *block), int speed)
Definition: dct.c:374
static const struct algo fdct_tab[]
Definition: dct.c:61
static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
Definition: dct.c:294
#define AANSCALE_BITS
Definition: dct.c:111
static uint8_t img_dest1[64]
Definition: dct.c:292
static int16_t block[64]
Definition: dct.c:116
int main(int argc, char **argv)
Definition: dct.c:467
static const struct algo fdct_tab_arch[]
Definition: dct.c:107
static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals)
Definition: dct.c:119
#define NB_ITS
Definition: dct.c:113
#define NB_ITS_SPEED
Definition: dct.c:114
static void permute(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
Definition: dct.c:149
static void ff_prores_idct_wrap(int16_t *dst)
Definition: dct.c:70
static const struct algo idct_tab[]
Definition: dct.c:83
static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits)
Definition: dct.c:179
static void help(void)
Definition: dct.c:451
static const struct algo idct_tab_arch[]
Definition: dct.c:108
static int16_t block1[64]
Definition: dct.c:117
static uint8_t img_dest[64]
Definition: dct.c:291
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
static int permute_x86(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
Definition: dct.c:120
int64_t av_gettime_relative(void)
Get the current time in microseconds since some unspecified starting point.
Definition: time.c:56
uint8_t bits
Definition: vp3data.h:141
static double c[64]
void ff_xvid_idct(int16_t *const in)
Definition: xvididct.c:291