falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
40234 Commits
32 Branches
871MB
Tree: e0a2e60c0a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e0a2e60c0a'
${ noResults }
FFmpeg/libavcodec/dct-test.c

510 lines
14KB
Raw Blame History 

  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. 

  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/lfg.h"

  40. #include "libavutil/time.h"

  41. 

  42. #include "dct.h"

  43. #include "idctdsp.h"

  44. #include "simple_idct.h"

  45. #include "aandcttab.h"

  46. #include "faandct.h"

  47. #include "faanidct.h"

  48. #include "x86/idct_xvid.h"

  49. #include "dctref.h"

  50. 

  51. // ALTIVEC

  52. void ff_fdct_altivec(int16_t *block);

  53. 

  54. // ARM

  55. void ff_j_rev_dct_arm(int16_t *data);

  56. void ff_simple_idct_arm(int16_t *data);

  57. void ff_simple_idct_armv5te(int16_t *data);

  58. void ff_simple_idct_armv6(int16_t *data);

  59. void ff_simple_idct_neon(int16_t *data);

  60. 

  61. struct algo {

  62.     const char *name;

  63.     void (*func)(int16_t *block);

  64.     enum idct_permutation_type perm_type;

  65.     int cpu_flag;

  66.     int nonspec;

  67. };

  68. 

  69. static const struct algo fdct_tab[] = {

  70.     { "REF-DBL",     ff_ref_fdct,          FF_IDCT_PERM_NONE },

  71.     { "FAAN",        ff_faandct,           FF_IDCT_PERM_NONE },

  72.     { "IJG-AAN-INT", ff_fdct_ifast,        FF_IDCT_PERM_NONE },

  73.     { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },

  74. 

  75. #if HAVE_MMX_INLINE

  76.     { "MMX",         ff_fdct_mmx,          FF_IDCT_PERM_NONE, AV_CPU_FLAG_MMX },

  77. #endif

  78. #if HAVE_MMXEXT_INLINE

  79.     { "MMXEXT",      ff_fdct_mmxext,       FF_IDCT_PERM_NONE, AV_CPU_FLAG_MMXEXT },

  80. #endif

  81. #if HAVE_SSE2_INLINE

  82.     { "SSE2",        ff_fdct_sse2,         FF_IDCT_PERM_NONE, AV_CPU_FLAG_SSE2 },

  83. #endif

  84. 

  85. #if HAVE_ALTIVEC

  86.     { "altivecfdct", ff_fdct_altivec,      FF_IDCT_PERM_NONE, AV_CPU_FLAG_ALTIVEC },

  87. #endif

  88. 

  89.     { 0 }

  90. };

  91. 

  92. static const struct algo idct_tab[] = {

  93.     { "FAANI",       ff_faanidct,          FF_IDCT_PERM_NONE },

  94.     { "REF-DBL",     ff_ref_idct,          FF_IDCT_PERM_NONE },

  95.     { "INT",         ff_j_rev_dct,         FF_IDCT_PERM_LIBMPEG2 },

  96.     { "SIMPLE-C",    ff_simple_idct_8,     FF_IDCT_PERM_NONE },

  97. 

  98. #if HAVE_MMX_INLINE

  99.     { "SIMPLE-MMX",     ff_simple_idct_mmx,     FF_IDCT_PERM_SIMPLE,    AV_CPU_FLAG_MMX },

  100.     { "XVID-MMX",       ff_idct_xvid_mmx,       FF_IDCT_PERM_NONE,      AV_CPU_FLAG_MMX,    1 },

  101. #endif

  102. #if HAVE_MMXEXT_INLINE

  103.     { "XVID-MMXEXT",    ff_idct_xvid_mmxext,    FF_IDCT_PERM_NONE,      AV_CPU_FLAG_MMXEXT, 1 },

  104. #endif

  105. #if HAVE_SSE2_INLINE

  106.     { "XVID-SSE2",      ff_idct_xvid_sse2,      FF_IDCT_PERM_SSE2,      AV_CPU_FLAG_SSE2,   1 },

  107. #endif

  108. 

  109. #if ARCH_ARM

  110.     { "SIMPLE-ARM",     ff_simple_idct_arm,     FF_IDCT_PERM_NONE },

  111.     { "INT-ARM",        ff_j_rev_dct_arm,       FF_IDCT_PERM_LIBMPEG2 },

  112. #endif

  113. #if HAVE_ARMV5TE

  114.     { "SIMPLE-ARMV5TE", ff_simple_idct_armv5te, FF_IDCT_PERM_NONE,      AV_CPU_FLAG_ARMV5TE },

  115. #endif

  116. #if HAVE_ARMV6

  117.     { "SIMPLE-ARMV6",   ff_simple_idct_armv6,   FF_IDCT_PERM_LIBMPEG2,  AV_CPU_FLAG_ARMV6 },

  118. #endif

  119. #if HAVE_NEON && ARCH_ARM

  120.     { "SIMPLE-NEON",    ff_simple_idct_neon,    FF_IDCT_PERM_PARTTRANS, AV_CPU_FLAG_NEON },

  121. #endif

  122. 

  123.     { 0 }

  124. };

  125. 

  126. #define AANSCALE_BITS 12

  127. 

  128. #define NB_ITS 20000

  129. #define NB_ITS_SPEED 50000

  130. 

  131. static short idct_simple_mmx_perm[64] = {

  132.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  133.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  134.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  135.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  136.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  137.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  138.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  139.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  140. };

  141. 

  142. static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };

  143. 

  144. DECLARE_ALIGNED(16, static int16_t, block)[64];

  145. DECLARE_ALIGNED(8,  static int16_t, block1)[64];

  146. 

  147. static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)

  148. {

  149.     int i, j;

  150. 

  151.     memset(block, 0, 64 * sizeof(*block));

  152. 

  153.     switch (test) {

  154.     case 0:

  155.         for (i = 0; i < 64; i++)

  156.             block[i] = (av_lfg_get(prng) % 512) - 256;

  157.         if (is_idct) {

  158.             ff_ref_fdct(block);

  159.             for (i = 0; i < 64; i++)

  160.                 block[i] >>= 3;

  161.         }

  162.         break;

  163.     case 1:

  164.         j = av_lfg_get(prng) % 10 + 1;

  165.         for (i = 0; i < j; i++)

  166.             block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;

  167.         break;

  168.     case 2:

  169.         block[ 0] = av_lfg_get(prng) % 4096 - 2048;

  170.         block[63] = (block[0] & 1) ^ 1;

  171.         break;

  172.     }

  173. }

  174. 

  175. static void permute(int16_t dst[64], const int16_t src[64],

  176.                     enum idct_permutation_type perm_type)

  177. {

  178.     int i;

  179. 

  180.     switch (perm_type) {

  181.     case FF_IDCT_PERM_LIBMPEG2:

  182.         for (i = 0; i < 64; i++)

  183.             dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];

  184.         break;

  185.     case FF_IDCT_PERM_SIMPLE:

  186.         for (i = 0; i < 64; i++)

  187.             dst[idct_simple_mmx_perm[i]] = src[i];

  188.         break;

  189.     case FF_IDCT_PERM_SSE2:

  190.         for (i = 0; i < 64; i++)

  191.             dst[(i & 0x38) | idct_sse2_row_perm[i & 7]] = src[i];

  192.         break;

  193.     case FF_IDCT_PERM_PARTTRANS:

  194.         for (i = 0; i < 64; i++)

  195.             dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];

  196.         break;

  197.     default:

  198.         for (i = 0; i < 64; i++)

  199.             dst[i] = src[i];

  200.         break;

  201.     }

  202. }

  203. 

  204. static int dct_error(const struct algo *dct, int test, int is_idct, int speed)

  205. {

  206.     void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;

  207.     int it, i, scale;

  208.     int err_inf, v;

  209.     int64_t err2, ti, ti1, it1, err_sum = 0;

  210.     int64_t sysErr[64], sysErrMax = 0;

  211.     int maxout = 0;

  212.     int blockSumErrMax = 0, blockSumErr;

  213.     AVLFG prng;

  214.     double omse, ome;

  215.     int spec_err;

  216. 

  217.     av_lfg_init(&prng, 1);

  218. 

  219.     err_inf = 0;

  220.     err2 = 0;

  221.     for (i = 0; i < 64; i++)

  222.         sysErr[i] = 0;

  223.     for (it = 0; it < NB_ITS; it++) {

  224.         init_block(block1, test, is_idct, &prng);

  225.         permute(block, block1, dct->perm_type);

  226. 

  227.         dct->func(block);

  228.         emms_c();

  229. 

  230.         if (!strcmp(dct->name, "IJG-AAN-INT")) {

  231.             for (i = 0; i < 64; i++) {

  232.                 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];

  233.                 block[i] = (block[i] * scale) >> AANSCALE_BITS;

  234.             }

  235.         }

  236. 

  237.         ref(block1);

  238. 

  239.         blockSumErr = 0;

  240.         for (i = 0; i < 64; i++) {

  241.             int err = block[i] - block1[i];

  242.             err_sum += err;

  243.             v = abs(err);

  244.             if (v > err_inf)

  245.                 err_inf = v;

  246.             err2 += v * v;

  247.             sysErr[i] += block[i] - block1[i];

  248.             blockSumErr += v;

  249.             if (abs(block[i]) > maxout)

  250.                 maxout = abs(block[i]);

  251.         }

  252.         if (blockSumErrMax < blockSumErr)

  253.             blockSumErrMax = blockSumErr;

  254.     }

  255.     for (i = 0; i < 64; i++)

  256.         sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));

  257. 

  258.     for (i = 0; i < 64; i++) {

  259.         if (i % 8 == 0)

  260.             printf("\n");

  261.         printf("%7d ", (int) sysErr[i]);

  262.     }

  263.     printf("\n");

  264. 

  265.     omse = (double) err2 / NB_ITS / 64;

  266.     ome  = (double) err_sum / NB_ITS / 64;

  267. 

  268.     spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);

  269. 

  270.     printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",

  271.            is_idct ? "IDCT" : "DCT", dct->name, err_inf,

  272.            omse, ome, (double) sysErrMax / NB_ITS,

  273.            maxout, blockSumErrMax);

  274. 

  275.     if (spec_err && !dct->nonspec)

  276.         return 1;

  277. 

  278.     if (!speed)

  279.         return 0;

  280. 

  281.     /* speed test */

  282.     init_block(block, test, is_idct, &prng);

  283.     permute(block1, block, dct->perm_type);

  284. 

  285.     ti = av_gettime();

  286.     it1 = 0;

  287.     do {

  288.         for (it = 0; it < NB_ITS_SPEED; it++) {

  289.             memcpy(block, block1, sizeof(block));

  290.             dct->func(block);

  291.         }

  292.         it1 += NB_ITS_SPEED;

  293.         ti1 = av_gettime() - ti;

  294.     } while (ti1 < 1000000);

  295.     emms_c();

  296. 

  297.     printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,

  298.            (double) it1 * 1000.0 / (double) ti1);

  299. 

  300.     return 0;

  301. }

  302. 

  303. DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];

  304. DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];

  305. 

  306. static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)

  307. {

  308.     static int init;

  309.     static double c8[8][8];

  310.     static double c4[4][4];

  311.     double block1[64], block2[64], block3[64];

  312.     double s, sum, v;

  313.     int i, j, k;

  314. 

  315.     if (!init) {

  316.         init = 1;

  317. 

  318.         for (i = 0; i < 8; i++) {

  319.             sum = 0;

  320.             for (j = 0; j < 8; j++) {

  321.                 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);

  322.                 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);

  323.                 sum += c8[i][j] * c8[i][j];

  324.             }

  325.         }

  326. 

  327.         for (i = 0; i < 4; i++) {

  328.             sum = 0;

  329.             for (j = 0; j < 4; j++) {

  330.                 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);

  331.                 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);

  332.                 sum += c4[i][j] * c4[i][j];

  333.             }

  334.         }

  335.     }

  336. 

  337.     /* butterfly */

  338.     s = 0.5 * sqrt(2.0);

  339.     for (i = 0; i < 4; i++) {

  340.         for (j = 0; j < 8; j++) {

  341.             block1[8 * (2 * i) + j] =

  342.                 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;

  343.             block1[8 * (2 * i + 1) + j] =

  344.                 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;

  345.         }

  346.     }

  347. 

  348.     /* idct8 on lines */

  349.     for (i = 0; i < 8; i++) {

  350.         for (j = 0; j < 8; j++) {

  351.             sum = 0;

  352.             for (k = 0; k < 8; k++)

  353.                 sum += c8[k][j] * block1[8 * i + k];

  354.             block2[8 * i + j] = sum;

  355.         }

  356.     }

  357. 

  358.     /* idct4 */

  359.     for (i = 0; i < 8; i++) {

  360.         for (j = 0; j < 4; j++) {

  361.             /* top */

  362.             sum = 0;

  363.             for (k = 0; k < 4; k++)

  364.                 sum += c4[k][j] * block2[8 * (2 * k) + i];

  365.             block3[8 * (2 * j) + i] = sum;

  366. 

  367.             /* bottom */

  368.             sum = 0;

  369.             for (k = 0; k < 4; k++)

  370.                 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];

  371.             block3[8 * (2 * j + 1) + i] = sum;

  372.         }

  373.     }

  374. 

  375.     /* clamp and store the result */

  376.     for (i = 0; i < 8; i++) {

  377.         for (j = 0; j < 8; j++) {

  378.             v = block3[8 * i + j];

  379.             if      (v < 0)   v = 0;

  380.             else if (v > 255) v = 255;

  381.             dest[i * linesize + j] = (int) rint(v);

  382.         }

  383.     }

  384. }

  385. 

  386. static void idct248_error(const char *name,

  387.                           void (*idct248_put)(uint8_t *dest, int line_size,

  388.                                               int16_t *block),

  389.                           int speed)

  390. {

  391.     int it, i, it1, ti, ti1, err_max, v;

  392.     AVLFG prng;

  393. 

  394.     av_lfg_init(&prng, 1);

  395. 

  396.     /* just one test to see if code is correct (precision is less

  397.        important here) */

  398.     err_max = 0;

  399.     for (it = 0; it < NB_ITS; it++) {

  400.         /* XXX: use forward transform to generate values */

  401.         for (i = 0; i < 64; i++)

  402.             block1[i] = av_lfg_get(&prng) % 256 - 128;

  403.         block1[0] += 1024;

  404. 

  405.         for (i = 0; i < 64; i++)

  406.             block[i] = block1[i];

  407.         idct248_ref(img_dest1, 8, block);

  408. 

  409.         for (i = 0; i < 64; i++)

  410.             block[i] = block1[i];

  411.         idct248_put(img_dest, 8, block);

  412. 

  413.         for (i = 0; i < 64; i++) {

  414.             v = abs((int) img_dest[i] - (int) img_dest1[i]);

  415.             if (v == 255)

  416.                 printf("%d %d\n", img_dest[i], img_dest1[i]);

  417.             if (v > err_max)

  418.                 err_max = v;

  419.         }

  420.     }

  421.     printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);

  422. 

  423.     if (!speed)

  424.         return;

  425. 

  426.     ti = av_gettime();

  427.     it1 = 0;

  428.     do {

  429.         for (it = 0; it < NB_ITS_SPEED; it++) {

  430.             for (i = 0; i < 64; i++)

  431.                 block[i] = block1[i];

  432.             idct248_put(img_dest, 8, block);

  433.         }

  434.         it1 += NB_ITS_SPEED;

  435.         ti1 = av_gettime() - ti;

  436.     } while (ti1 < 1000000);

  437.     emms_c();

  438. 

  439.     printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,

  440.            (double) it1 * 1000.0 / (double) ti1);

  441. }

  442. 

  443. static void help(void)

  444. {

  445.     printf("dct-test [-i] [<test-number>]\n"

  446.            "test-number 0 -> test with random matrixes\n"

  447.            "            1 -> test with random sparse matrixes\n"

  448.            "            2 -> do 3. test from mpeg4 std\n"

  449.            "-i          test IDCT implementations\n"

  450.            "-4          test IDCT248 implementations\n"

  451.            "-t          speed test\n");

  452. }

  453. 

  454. #if !HAVE_GETOPT

  455. #include "compat/getopt.c"

  456. #endif

  457. 

  458. int main(int argc, char **argv)

  459. {

  460.     int test_idct = 0, test_248_dct = 0;

  461.     int c, i;

  462.     int test = 1;

  463.     int speed = 0;

  464.     int err = 0;

  465. 

  466.     ff_ref_dct_init();

  467. 

  468.     for (;;) {

  469.         c = getopt(argc, argv, "ih4t");

  470.         if (c == -1)

  471.             break;

  472.         switch (c) {

  473.         case 'i':

  474.             test_idct = 1;

  475.             break;

  476.         case '4':

  477.             test_248_dct = 1;

  478.             break;

  479.         case 't':

  480.             speed = 1;

  481.             break;

  482.         default:

  483.         case 'h':

  484.             help();

  485.             return 0;

  486.         }

  487.     }

  488. 

  489.     if (optind < argc)

  490.         test = atoi(argv[optind]);

  491. 

  492.     printf("Libav DCT/IDCT test\n");

  493. 

  494.     if (test_248_dct) {

  495.         idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);

  496.     } else {

  497.         const int cpu_flags = av_get_cpu_flags();

  498.         const struct algo *algos = test_idct ? idct_tab : fdct_tab;

  499.         for (i = 0; algos[i].name; i++)

  500.             if (!(~cpu_flags & algos[i].cpu_flag)) {

  501.                 err |= dct_error(&algos[i], test, test_idct, speed);

  502.             }

  503.     }

  504. 

  505.     if (err)

  506.         printf("Error: %d.\n", err);

  507. 

  508.     return !!err;

  509. }