Libav
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 Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
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/time.h"
42 
43 #include "libavcodec/aandcttab.h"
44 #include "libavcodec/dct.h"
45 #include "libavcodec/dctref.h"
46 #include "libavcodec/faandct.h"
47 #include "libavcodec/faanidct.h"
48 #include "libavcodec/idctdsp.h"
49 #include "libavcodec/simple_idct.h"
50 #include "libavcodec/xvididct.h"
51 
52 struct algo {
53  const char *name;
54  void (*func)(int16_t *block);
56  int cpu_flag;
57  int nonspec;
58 };
59 
60 static const struct algo fdct_tab[] = {
61  { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
62  { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
63  { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
64 #if CONFIG_FAANDCT
65  { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
66 #endif /* CONFIG_FAANDCT */
67 };
68 
69 static const struct algo idct_tab[] = {
70  { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
72  { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
73 #if CONFIG_FAANIDCT
74  { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
75 #endif /* CONFIG_FAANIDCT */
76 #if CONFIG_MPEG4_DECODER
77  { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
78 #endif /* CONFIG_MPEG4_DECODER */
79 };
80 
81 #if ARCH_ARM
82 #include "arm/dct.c"
83 #elif ARCH_PPC
84 #include "ppc/dct.c"
85 #elif ARCH_X86
86 #include "x86/dct.c"
87 #else
88 static const struct algo fdct_tab_arch[] = { { 0 } };
89 static const struct algo idct_tab_arch[] = { { 0 } };
90 #endif
91 
92 #define AANSCALE_BITS 12
93 
94 #define NB_ITS 20000
95 #define NB_ITS_SPEED 50000
96 
97 DECLARE_ALIGNED(16, static int16_t, block)[64];
98 DECLARE_ALIGNED(8, static int16_t, block1)[64];
99 
100 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
101 {
102  int i, j;
103 
104  memset(block, 0, 64 * sizeof(*block));
105 
106  switch (test) {
107  case 0:
108  for (i = 0; i < 64; i++)
109  block[i] = (av_lfg_get(prng) % 512) - 256;
110  if (is_idct) {
111  ff_ref_fdct(block);
112  for (i = 0; i < 64; i++)
113  block[i] >>= 3;
114  }
115  break;
116  case 1:
117  j = av_lfg_get(prng) % 10 + 1;
118  for (i = 0; i < j; i++)
119  block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;
120  break;
121  case 2:
122  block[ 0] = av_lfg_get(prng) % 4096 - 2048;
123  block[63] = (block[0] & 1) ^ 1;
124  break;
125  }
126 }
127 
128 static void permute(int16_t dst[64], const int16_t src[64],
130 {
131  int i;
132 
133 #if ARCH_X86
134  if (permute_x86(dst, src, perm_type))
135  return;
136 #endif
137 
138  switch (perm_type) {
140  for (i = 0; i < 64; i++)
141  dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
142  break;
144  for (i = 0; i < 64; i++)
145  dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
146  break;
147  default:
148  for (i = 0; i < 64; i++)
149  dst[i] = src[i];
150  break;
151  }
152 }
153 
154 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
155 {
156  void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
157  int it, i, scale;
158  int err_inf, v;
159  int64_t err2, ti, ti1, it1, err_sum = 0;
160  int64_t sysErr[64], sysErrMax = 0;
161  int maxout = 0;
162  int blockSumErrMax = 0, blockSumErr;
163  AVLFG prng;
164  double omse, ome;
165  int spec_err;
166 
167  av_lfg_init(&prng, 1);
168 
169  err_inf = 0;
170  err2 = 0;
171  for (i = 0; i < 64; i++)
172  sysErr[i] = 0;
173  for (it = 0; it < NB_ITS; it++) {
174  init_block(block1, test, is_idct, &prng);
175  permute(block, block1, dct->perm_type);
176 
177  dct->func(block);
178  emms_c();
179 
180  if (!strcmp(dct->name, "IJG-AAN-INT")) {
181  for (i = 0; i < 64; i++) {
182  scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
183  block[i] = (block[i] * scale) >> AANSCALE_BITS;
184  }
185  }
186 
187  ref(block1);
188 
189  blockSumErr = 0;
190  for (i = 0; i < 64; i++) {
191  int err = block[i] - block1[i];
192  err_sum += err;
193  v = abs(err);
194  if (v > err_inf)
195  err_inf = v;
196  err2 += v * v;
197  sysErr[i] += block[i] - block1[i];
198  blockSumErr += v;
199  if (abs(block[i]) > maxout)
200  maxout = abs(block[i]);
201  }
202  if (blockSumErrMax < blockSumErr)
203  blockSumErrMax = blockSumErr;
204  }
205  for (i = 0; i < 64; i++)
206  sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
207 
208  for (i = 0; i < 64; i++) {
209  if (i % 8 == 0)
210  printf("\n");
211  printf("%7d ", (int) sysErr[i]);
212  }
213  printf("\n");
214 
215  omse = (double) err2 / NB_ITS / 64;
216  ome = (double) err_sum / NB_ITS / 64;
217 
218  spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
219 
220  printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
221  is_idct ? "IDCT" : "DCT", dct->name, err_inf,
222  omse, ome, (double) sysErrMax / NB_ITS,
223  maxout, blockSumErrMax);
224 
225  if (spec_err && !dct->nonspec)
226  return 1;
227 
228  if (!speed)
229  return 0;
230 
231  /* speed test */
232  init_block(block, test, is_idct, &prng);
233  permute(block1, block, dct->perm_type);
234 
235  ti = av_gettime_relative();
236  it1 = 0;
237  do {
238  for (it = 0; it < NB_ITS_SPEED; it++) {
239  memcpy(block, block1, sizeof(block));
240  dct->func(block);
241  }
242  it1 += NB_ITS_SPEED;
243  ti1 = av_gettime_relative() - ti;
244  } while (ti1 < 1000000);
245  emms_c();
246 
247  printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
248  (double) it1 * 1000.0 / (double) ti1);
249 
250  return 0;
251 }
252 
255 
256 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
257 {
258  static int init;
259  static double c8[8][8];
260  static double c4[4][4];
261  double block1[64], block2[64], block3[64];
262  double s, sum, v;
263  int i, j, k;
264 
265  if (!init) {
266  init = 1;
267 
268  for (i = 0; i < 8; i++) {
269  sum = 0;
270  for (j = 0; j < 8; j++) {
271  s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
272  c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
273  sum += c8[i][j] * c8[i][j];
274  }
275  }
276 
277  for (i = 0; i < 4; i++) {
278  sum = 0;
279  for (j = 0; j < 4; j++) {
280  s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
281  c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
282  sum += c4[i][j] * c4[i][j];
283  }
284  }
285  }
286 
287  /* butterfly */
288  s = 0.5 * sqrt(2.0);
289  for (i = 0; i < 4; i++) {
290  for (j = 0; j < 8; j++) {
291  block1[8 * (2 * i) + j] =
292  (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
293  block1[8 * (2 * i + 1) + j] =
294  (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
295  }
296  }
297 
298  /* idct8 on lines */
299  for (i = 0; i < 8; i++) {
300  for (j = 0; j < 8; j++) {
301  sum = 0;
302  for (k = 0; k < 8; k++)
303  sum += c8[k][j] * block1[8 * i + k];
304  block2[8 * i + j] = sum;
305  }
306  }
307 
308  /* idct4 */
309  for (i = 0; i < 8; i++) {
310  for (j = 0; j < 4; j++) {
311  /* top */
312  sum = 0;
313  for (k = 0; k < 4; k++)
314  sum += c4[k][j] * block2[8 * (2 * k) + i];
315  block3[8 * (2 * j) + i] = sum;
316 
317  /* bottom */
318  sum = 0;
319  for (k = 0; k < 4; k++)
320  sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
321  block3[8 * (2 * j + 1) + i] = sum;
322  }
323  }
324 
325  /* clamp and store the result */
326  for (i = 0; i < 8; i++) {
327  for (j = 0; j < 8; j++) {
328  v = block3[8 * i + j];
329  if (v < 0) v = 0;
330  else if (v > 255) v = 255;
331  dest[i * linesize + j] = (int) rint(v);
332  }
333  }
334 }
335 
336 static void idct248_error(const char *name,
337  void (*idct248_put)(uint8_t *dest, int line_size,
338  int16_t *block),
339  int speed)
340 {
341  int it, i, it1, ti, ti1, err_max, v;
342  AVLFG prng;
343 
344  av_lfg_init(&prng, 1);
345 
346  /* just one test to see if code is correct (precision is less
347  important here) */
348  err_max = 0;
349  for (it = 0; it < NB_ITS; it++) {
350  /* XXX: use forward transform to generate values */
351  for (i = 0; i < 64; i++)
352  block1[i] = av_lfg_get(&prng) % 256 - 128;
353  block1[0] += 1024;
354 
355  for (i = 0; i < 64; i++)
356  block[i] = block1[i];
357  idct248_ref(img_dest1, 8, block);
358 
359  for (i = 0; i < 64; i++)
360  block[i] = block1[i];
361  idct248_put(img_dest, 8, block);
362 
363  for (i = 0; i < 64; i++) {
364  v = abs((int) img_dest[i] - (int) img_dest1[i]);
365  if (v == 255)
366  printf("%d %d\n", img_dest[i], img_dest1[i]);
367  if (v > err_max)
368  err_max = v;
369  }
370  }
371  printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
372 
373  if (!speed)
374  return;
375 
376  ti = av_gettime_relative();
377  it1 = 0;
378  do {
379  for (it = 0; it < NB_ITS_SPEED; it++) {
380  for (i = 0; i < 64; i++)
381  block[i] = block1[i];
382  idct248_put(img_dest, 8, block);
383  }
384  it1 += NB_ITS_SPEED;
385  ti1 = av_gettime_relative() - ti;
386  } while (ti1 < 1000000);
387  emms_c();
388 
389  printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
390  (double) it1 * 1000.0 / (double) ti1);
391 }
392 
393 static void help(void)
394 {
395  printf("dct-test [-i] [<test-number>]\n"
396  "test-number 0 -> test with random matrixes\n"
397  " 1 -> test with random sparse matrixes\n"
398  " 2 -> do 3. test from MPEG-4 std\n"
399  "-i test IDCT implementations\n"
400  "-4 test IDCT248 implementations\n"
401  "-t speed test\n");
402 }
403 
404 #if !HAVE_GETOPT
405 #include "compat/getopt.c"
406 #endif
407 
408 int main(int argc, char **argv)
409 {
410  int test_idct = 0, test_248_dct = 0;
411  int c, i;
412  int test = 1;
413  int speed = 0;
414  int err = 0;
415 
416  ff_ref_dct_init();
417 
418  for (;;) {
419  c = getopt(argc, argv, "ih4t");
420  if (c == -1)
421  break;
422  switch (c) {
423  case 'i':
424  test_idct = 1;
425  break;
426  case '4':
427  test_248_dct = 1;
428  break;
429  case 't':
430  speed = 1;
431  break;
432  default:
433  case 'h':
434  help();
435  return 0;
436  }
437  }
438 
439  if (optind < argc)
440  test = atoi(argv[optind]);
441 
442  printf("Libav DCT/IDCT test\n");
443 
444  if (test_248_dct) {
445  idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
446  } else {
447  const int cpu_flags = av_get_cpu_flags();
448  if (test_idct) {
449  for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
450  err |= dct_error(&idct_tab[i], test, test_idct, speed);
451 
452  for (i = 0; idct_tab_arch[i].name; i++)
453  if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
454  err |= dct_error(&idct_tab_arch[i], test, test_idct, speed);
455  }
456 #if CONFIG_FDCTDSP
457  else {
458  for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
459  err |= dct_error(&fdct_tab[i], test, test_idct, speed);
460 
461  for (i = 0; fdct_tab_arch[i].name; i++)
462  if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
463  err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed);
464  }
465 #endif /* CONFIG_FDCTDSP */
466  }
467 
468  if (err)
469  printf("Error: %d.\n", err);
470 
471  return !!err;
472 }
static const struct algo idct_tab_arch[]
Definition: dct.c:89
Definition: lfg.h:25
idct_permutation_type
Definition: idctdsp.h:35
#define AANSCALE_BITS
Definition: dct.c:92
static double rint(double x)
Definition: libm.h:130
void ff_fdct_ifast(int16_t *data)
Definition: jfdctfst.c:208
static int optind
Definition: getopt.c:37
const char * name
Definition: dct.c:53
static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
Definition: dct.c:100
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:58
#define FF_ARRAY_ELEMS(a)
av_cold void ff_ref_dct_init(void)
Initialize the double precision discrete cosine transform functions fdct & idct.
Definition: dctref.c:41
static int16_t block[64]
Definition: dct.c:97
int nonspec
Definition: dct.c:57
uint8_t
void ff_faanidct(int16_t block[64])
Definition: faanidct.c:125
static int permute_x86(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
Definition: dct.c:69
#define emms_c()
Definition: internal.h:48
int cpu_flag
Definition: dct.c:56
void ff_simple_idct248_put(uint8_t *dest, int line_size, int16_t *block)
Definition: simple_idct.c:88
const uint16_t ff_aanscales[64]
Definition: aandcttab.c:26
static void idct248_error(const char *name, void(*idct248_put)(uint8_t *dest, int line_size, int16_t *block), int speed)
Definition: dct.c:336
static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
Definition: dct.c:154
static void help(void)
Definition: dct.c:393
#define src
Definition: vp8dsp.c:254
static uint8_t img_dest1[64]
Definition: dct.c:254
#define FFMAX(a, b)
Definition: common.h:64
void(* func)(int16_t *block)
Definition: dct.c:54
common internal API header
Definition: dct.c:52
int main(int argc, char **argv)
Definition: dct.c:408
static void permute(int16_t dst[64], const int16_t src[64], enum idct_permutation_type perm_type)
Definition: dct.c:128
static const struct algo idct_tab[]
Definition: dct.c:69
#define FFABS(a)
Definition: common.h:61
static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
Definition: dct.c:256
void ff_jpeg_fdct_islow_8(int16_t *data)
void ff_j_rev_dct(int16_t *data)
void ff_faandct(int16_t *data)
Definition: faandct.c:115
static void test(const char *pattern, const char *host)
Definition: noproxy.c:23
static void(WINAPI *cond_broadcast)(pthread_cond_t *cond)
static int getopt(int argc, char *argv[], char *opts)
Definition: getopt.c:41
static unsigned int av_lfg_get(AVLFG *c)
Get the next random unsigned 32-bit number using an ALFG.
Definition: lfg.h:38
static const struct algo fdct_tab_arch[]
Definition: dct.c:88
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
av_cold void av_lfg_init(AVLFG *c, unsigned int seed)
Definition: lfg.c:30
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:47
void ff_xvid_idct(int16_t *const in)
Definition: xvididct.c:291
int64_t av_gettime_relative(void)
Get the current time in microseconds since some unspecified starting point.
Definition: time.c:57
static uint8_t img_dest[64]
Definition: dct.c:253
AAN (Arai Agui Nakajima) (I)DCT tables.
common internal and external API header
#define NB_ITS_SPEED
Definition: dct.c:95
enum idct_permutation_type perm_type
Definition: dct.c:55
static av_cold int init(AVCodecParserContext *s)
Definition: h264_parser.c:582
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
simple idct header.
static int16_t block1[64]
Definition: dct.c:98
Floating point AAN DCT
static const struct algo fdct_tab[]
Definition: dct.c:60
#define NB_ITS
Definition: dct.c:94
void ff_simple_idct_8(int16_t *block)