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.
39979 Commits
32 Branches
649MB
Tree: 593886b5dd
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '593886b5dd'
${ noResults }
FFmpeg/libavcodec/dct-test.c

531 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 "simple_idct.h"

  44. #include "aandcttab.h"

  45. #include "faandct.h"

  46. #include "faanidct.h"

  47. #include "x86/idct_xvid.h"

  48. #include "dctref.h"

  49. 

  50. // BFIN

  51. void ff_bfin_idct(int16_t *block);

  52. void ff_bfin_fdct(int16_t *block);

  53. 

  54. // ALTIVEC

  55. void ff_fdct_altivec(int16_t *block);

  56. 

  57. // ARM

  58. void ff_j_rev_dct_arm(int16_t *data);

  59. void ff_simple_idct_arm(int16_t *data);

  60. void ff_simple_idct_armv5te(int16_t *data);

  61. void ff_simple_idct_armv6(int16_t *data);

  62. void ff_simple_idct_neon(int16_t *data);

  63. 

  64. struct algo {

  65.     const char *name;

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

  67.     enum formattag { NO_PERM, MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM,

  68.                      SSE2_PERM, PARTTRANS_PERM } format;

  69.     int mm_support;

  70.     int nonspec;

  71. };

  72. 

  73. static int cpu_flags;

  74. 

  75. static const struct algo fdct_tab[] = {

  76.     { "REF-DBL",        ff_ref_fdct,           NO_PERM    },

  77.     { "FAAN",           ff_faandct,            NO_PERM    },

  78.     { "IJG-AAN-INT",    ff_fdct_ifast,         SCALE_PERM },

  79.     { "IJG-LLM-INT",    ff_jpeg_fdct_islow_8,  NO_PERM    },

  80. 

  81. #if HAVE_MMX_INLINE

  82.     { "MMX",            ff_fdct_mmx,           NO_PERM,   AV_CPU_FLAG_MMX     },

  83. #endif

  84. #if HAVE_MMXEXT_INLINE

  85.     { "MMXEXT",         ff_fdct_mmxext,        NO_PERM,   AV_CPU_FLAG_MMXEXT  },

  86. #endif

  87. #if HAVE_SSE2_INLINE

  88.     { "SSE2",           ff_fdct_sse2,          NO_PERM,   AV_CPU_FLAG_SSE2    },

  89. #endif

  90. 

  91. #if HAVE_ALTIVEC

  92.     { "altivecfdct",    ff_fdct_altivec,       NO_PERM,   AV_CPU_FLAG_ALTIVEC },

  93. #endif

  94. 

  95. #if ARCH_BFIN

  96.     { "BFINfdct",       ff_bfin_fdct,          NO_PERM  },

  97. #endif

  98. 

  99.     { 0 }

  100. };

  101. 

  102. static const struct algo idct_tab[] = {

  103.     { "FAANI",          ff_faanidct,           NO_PERM  },

  104.     { "REF-DBL",        ff_ref_idct,           NO_PERM  },

  105.     { "INT",            ff_j_rev_dct,          MMX_PERM },

  106.     { "SIMPLE-C",       ff_simple_idct_8,      NO_PERM  },

  107. 

  108. #if HAVE_MMX_INLINE

  109.     { "SIMPLE-MMX",     ff_simple_idct_mmx,  MMX_SIMPLE_PERM, AV_CPU_FLAG_MMX },

  110.     { "XVID-MMX",       ff_idct_xvid_mmx,      NO_PERM,   AV_CPU_FLAG_MMX,  1 },

  111. #endif

  112. #if HAVE_MMXEXT_INLINE

  113.     { "XVID-MMXEXT",    ff_idct_xvid_mmxext,   NO_PERM,   AV_CPU_FLAG_MMXEXT, 1 },

  114. #endif

  115. #if HAVE_SSE2_INLINE

  116.     { "XVID-SSE2",      ff_idct_xvid_sse2,     SSE2_PERM, AV_CPU_FLAG_SSE2, 1 },

  117. #endif

  118. 

  119. #if ARCH_BFIN

  120.     { "BFINidct",       ff_bfin_idct,          NO_PERM  },

  121. #endif

  122. 

  123. #if ARCH_ARM

  124.     { "SIMPLE-ARM",     ff_simple_idct_arm,    NO_PERM  },

  125.     { "INT-ARM",        ff_j_rev_dct_arm,      MMX_PERM },

  126. #endif

  127. #if HAVE_ARMV5TE

  128.     { "SIMPLE-ARMV5TE", ff_simple_idct_armv5te,NO_PERM,   AV_CPU_FLAG_ARMV5TE },

  129. #endif

  130. #if HAVE_ARMV6

  131.     { "SIMPLE-ARMV6",   ff_simple_idct_armv6,  MMX_PERM,  AV_CPU_FLAG_ARMV6   },

  132. #endif

  133. #if HAVE_NEON && ARCH_ARM

  134.     { "SIMPLE-NEON",    ff_simple_idct_neon, PARTTRANS_PERM, AV_CPU_FLAG_NEON },

  135. #endif

  136. 

  137.     { 0 }

  138. };

  139. 

  140. #define AANSCALE_BITS 12

  141. 

  142. #define NB_ITS 20000

  143. #define NB_ITS_SPEED 50000

  144. 

  145. static short idct_mmx_perm[64];

  146. 

  147. static short idct_simple_mmx_perm[64] = {

  148.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  149.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  150.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  151.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  152.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  153.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  154.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  155.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  156. };

  157. 

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

  159. 

  160. static void idct_mmx_init(void)

  161. {

  162.     int i;

  163. 

  164.     /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */

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

  166.         idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);

  167.     }

  168. }

  169. 

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

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

  172. 

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

  174. {

  175.     int i, j;

  176. 

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

  178. 

  179.     switch (test) {

  180.     case 0:

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

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

  183.         if (is_idct) {

  184.             ff_ref_fdct(block);

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

  186.                 block[i] >>= 3;

  187.         }

  188.         break;

  189.     case 1:

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

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

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

  193.         break;

  194.     case 2:

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

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

  197.         break;

  198.     }

  199. }

  200. 

  201. static void permute(int16_t dst[64], const int16_t src[64], int perm)

  202. {

  203.     int i;

  204. 

  205.     if (perm == MMX_PERM) {

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

  207.             dst[idct_mmx_perm[i]] = src[i];

  208.     } else if (perm == MMX_SIMPLE_PERM) {

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

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

  211.     } else if (perm == SSE2_PERM) {

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

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

  214.     } else if (perm == PARTTRANS_PERM) {

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

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

  217.     } else {

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

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

  220.     }

  221. }

  222. 

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

  224. {

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

  226.     int it, i, scale;

  227.     int err_inf, v;

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

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

  230.     int maxout = 0;

  231.     int blockSumErrMax = 0, blockSumErr;

  232.     AVLFG prng;

  233.     double omse, ome;

  234.     int spec_err;

  235. 

  236.     av_lfg_init(&prng, 1);

  237. 

  238.     err_inf = 0;

  239.     err2 = 0;

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

  241.         sysErr[i] = 0;

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

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

  244.         permute(block, block1, dct->format);

  245. 

  246.         dct->func(block);

  247.         emms_c();

  248. 

  249.         if (dct->format == SCALE_PERM) {

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

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

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

  253.             }

  254.         }

  255. 

  256.         ref(block1);

  257. 

  258.         blockSumErr = 0;

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

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

  261.             err_sum += err;

  262.             v = abs(err);

  263.             if (v > err_inf)

  264.                 err_inf = v;

  265.             err2 += v * v;

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

  267.             blockSumErr += v;

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

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

  270.         }

  271.         if (blockSumErrMax < blockSumErr)

  272.             blockSumErrMax = blockSumErr;

  273.     }

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

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

  276. 

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

  278.         if (i % 8 == 0)

  279.             printf("\n");

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

  281.     }

  282.     printf("\n");

  283. 

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

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

  286. 

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

  288. 

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

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

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

  292.            maxout, blockSumErrMax);

  293. 

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

  295.         return 1;

  296. 

  297.     if (!speed)

  298.         return 0;

  299. 

  300.     /* speed test */

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

  302.     permute(block1, block, dct->format);

  303. 

  304.     ti = av_gettime();

  305.     it1 = 0;

  306.     do {

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

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

  309.             dct->func(block);

  310.         }

  311.         it1 += NB_ITS_SPEED;

  312.         ti1 = av_gettime() - ti;

  313.     } while (ti1 < 1000000);

  314.     emms_c();

  315. 

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

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

  318. 

  319.     return 0;

  320. }

  321. 

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

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

  324. 

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

  326. {

  327.     static int init;

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

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

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

  331.     double s, sum, v;

  332.     int i, j, k;

  333. 

  334.     if (!init) {

  335.         init = 1;

  336. 

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

  338.             sum = 0;

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

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

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

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

  343.             }

  344.         }

  345. 

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

  347.             sum = 0;

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

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

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

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

  352.             }

  353.         }

  354.     }

  355. 

  356.     /* butterfly */

  357.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  364.         }

  365.     }

  366. 

  367.     /* idct8 on lines */

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

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

  370.             sum = 0;

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

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

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

  374.         }

  375.     }

  376. 

  377.     /* idct4 */

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

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

  380.             /* top */

  381.             sum = 0;

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

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

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

  385. 

  386.             /* bottom */

  387.             sum = 0;

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

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

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

  391.         }

  392.     }

  393. 

  394.     /* clamp and store the result */

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

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

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

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

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

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

  401.         }

  402.     }

  403. }

  404. 

  405. static void idct248_error(const char *name,

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

  407.                                               int16_t *block),

  408.                           int speed)

  409. {

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

  411.     AVLFG prng;

  412. 

  413.     av_lfg_init(&prng, 1);

  414. 

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

  416.        important here) */

  417.     err_max = 0;

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

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

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

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

  422.         block1[0] += 1024;

  423. 

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

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

  426.         idct248_ref(img_dest1, 8, block);

  427. 

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

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

  430.         idct248_put(img_dest, 8, block);

  431. 

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

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

  434.             if (v == 255)

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

  436.             if (v > err_max)

  437.                 err_max = v;

  438.         }

  439.     }

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

  441. 

  442.     if (!speed)

  443.         return;

  444. 

  445.     ti = av_gettime();

  446.     it1 = 0;

  447.     do {

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

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

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

  451.             idct248_put(img_dest, 8, block);

  452.         }

  453.         it1 += NB_ITS_SPEED;

  454.         ti1 = av_gettime() - ti;

  455.     } while (ti1 < 1000000);

  456.     emms_c();

  457. 

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

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

  460. }

  461. 

  462. static void help(void)

  463. {

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

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

  466.            "            1 -> test with random sparse matrixes\n"

  467.            "            2 -> do 3. test from mpeg4 std\n"

  468.            "-i          test IDCT implementations\n"

  469.            "-4          test IDCT248 implementations\n"

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

  471. }

  472. 

  473. #if !HAVE_GETOPT

  474. #include "compat/getopt.c"

  475. #endif

  476. 

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

  478. {

  479.     int test_idct = 0, test_248_dct = 0;

  480.     int c, i;

  481.     int test = 1;

  482.     int speed = 0;

  483.     int err = 0;

  484. 

  485.     cpu_flags = av_get_cpu_flags();

  486. 

  487.     ff_ref_dct_init();

  488.     idct_mmx_init();

  489. 

  490.     for (;;) {

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

  492.         if (c == -1)

  493.             break;

  494.         switch (c) {

  495.         case 'i':

  496.             test_idct = 1;

  497.             break;

  498.         case '4':

  499.             test_248_dct = 1;

  500.             break;

  501.         case 't':

  502.             speed = 1;

  503.             break;

  504.         default:

  505.         case 'h':

  506.             help();

  507.             return 0;

  508.         }

  509.     }

  510. 

  511.     if (optind < argc)

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

  513. 

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

  515. 

  516.     if (test_248_dct) {

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

  518.     } else {

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

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

  521.             if (!(~cpu_flags & algos[i].mm_support)) {

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

  523.             }

  524.     }

  525. 

  526.     if (err)

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

  528. 

  529.     return !!err;

  530. }