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.
40228 Commits
32 Branches
411MB
Tree: c0de9159a7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c0de9159a7'
${ noResults }
FFmpeg/libavcodec/dct-test.c

516 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. // ALTIVEC

  51. void ff_fdct_altivec(int16_t *block);

  52. 

  53. // ARM

  54. void ff_j_rev_dct_arm(int16_t *data);

  55. void ff_simple_idct_arm(int16_t *data);

  56. void ff_simple_idct_armv5te(int16_t *data);

  57. void ff_simple_idct_armv6(int16_t *data);

  58. void ff_simple_idct_neon(int16_t *data);

  59. 

  60. struct algo {

  61.     const char *name;

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

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

  64.                      SSE2_PERM, PARTTRANS_PERM } format;

  65.     int cpu_flag;

  66.     int nonspec;

  67. };

  68. 

  69. static const struct algo fdct_tab[] = {

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

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

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

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

  74. 

  75. #if HAVE_MMX_INLINE

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

  77. #endif

  78. #if HAVE_MMXEXT_INLINE

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

  80. #endif

  81. #if HAVE_SSE2_INLINE

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

  83. #endif

  84. 

  85. #if HAVE_ALTIVEC

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

  87. #endif

  88. 

  89.     { 0 }

  90. };

  91. 

  92. static const struct algo idct_tab[] = {

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

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

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

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

  97. 

  98. #if HAVE_MMX_INLINE

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

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

  101. #endif

  102. #if HAVE_MMXEXT_INLINE

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

  104. #endif

  105. #if HAVE_SSE2_INLINE

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

  107. #endif

  108. 

  109. #if ARCH_ARM

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

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

  112. #endif

  113. #if HAVE_ARMV5TE

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

  115. #endif

  116. #if HAVE_ARMV6

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

  118. #endif

  119. #if HAVE_NEON && ARCH_ARM

  120.     { "SIMPLE-NEON",    ff_simple_idct_neon, PARTTRANS_PERM, 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_mmx_perm[64];

  132. 

  133. static short idct_simple_mmx_perm[64] = {

  134.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  135.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  136.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  137.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  138.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  139.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  140.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  141.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  142. };

  143. 

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

  145. 

  146. static void idct_mmx_init(void)

  147. {

  148.     int i;

  149. 

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

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

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

  153.     }

  154. }

  155. 

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

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

  158. 

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

  160. {

  161.     int i, j;

  162. 

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

  164. 

  165.     switch (test) {

  166.     case 0:

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

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

  169.         if (is_idct) {

  170.             ff_ref_fdct(block);

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

  172.                 block[i] >>= 3;

  173.         }

  174.         break;

  175.     case 1:

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

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

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

  179.         break;

  180.     case 2:

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

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

  183.         break;

  184.     }

  185. }

  186. 

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

  188. {

  189.     int i;

  190. 

  191.     if (perm == MMX_PERM) {

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

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

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

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

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

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

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

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

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

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

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

  203.     } else {

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

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

  206.     }

  207. }

  208. 

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

  210. {

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

  212.     int it, i, scale;

  213.     int err_inf, v;

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

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

  216.     int maxout = 0;

  217.     int blockSumErrMax = 0, blockSumErr;

  218.     AVLFG prng;

  219.     double omse, ome;

  220.     int spec_err;

  221. 

  222.     av_lfg_init(&prng, 1);

  223. 

  224.     err_inf = 0;

  225.     err2 = 0;

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

  227.         sysErr[i] = 0;

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

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

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

  231. 

  232.         dct->func(block);

  233.         emms_c();

  234. 

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

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

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

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

  239.             }

  240.         }

  241. 

  242.         ref(block1);

  243. 

  244.         blockSumErr = 0;

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

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

  247.             err_sum += err;

  248.             v = abs(err);

  249.             if (v > err_inf)

  250.                 err_inf = v;

  251.             err2 += v * v;

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

  253.             blockSumErr += v;

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

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

  256.         }

  257.         if (blockSumErrMax < blockSumErr)

  258.             blockSumErrMax = blockSumErr;

  259.     }

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

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

  262. 

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

  264.         if (i % 8 == 0)

  265.             printf("\n");

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

  267.     }

  268.     printf("\n");

  269. 

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

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

  272. 

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

  274. 

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

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

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

  278.            maxout, blockSumErrMax);

  279. 

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

  281.         return 1;

  282. 

  283.     if (!speed)

  284.         return 0;

  285. 

  286.     /* speed test */

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

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

  289. 

  290.     ti = av_gettime();

  291.     it1 = 0;

  292.     do {

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

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

  295.             dct->func(block);

  296.         }

  297.         it1 += NB_ITS_SPEED;

  298.         ti1 = av_gettime() - ti;

  299.     } while (ti1 < 1000000);

  300.     emms_c();

  301. 

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

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

  304. 

  305.     return 0;

  306. }

  307. 

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

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

  310. 

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

  312. {

  313.     static int init;

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

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

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

  317.     double s, sum, v;

  318.     int i, j, k;

  319. 

  320.     if (!init) {

  321.         init = 1;

  322. 

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

  324.             sum = 0;

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

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

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

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

  329.             }

  330.         }

  331. 

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

  333.             sum = 0;

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

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

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

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

  338.             }

  339.         }

  340.     }

  341. 

  342.     /* butterfly */

  343.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  350.         }

  351.     }

  352. 

  353.     /* idct8 on lines */

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

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

  356.             sum = 0;

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

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

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

  360.         }

  361.     }

  362. 

  363.     /* idct4 */

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

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

  366.             /* top */

  367.             sum = 0;

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

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

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

  371. 

  372.             /* bottom */

  373.             sum = 0;

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

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

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

  377.         }

  378.     }

  379. 

  380.     /* clamp and store the result */

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

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

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

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

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

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

  387.         }

  388.     }

  389. }

  390. 

  391. static void idct248_error(const char *name,

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

  393.                                               int16_t *block),

  394.                           int speed)

  395. {

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

  397.     AVLFG prng;

  398. 

  399.     av_lfg_init(&prng, 1);

  400. 

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

  402.        important here) */

  403.     err_max = 0;

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

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

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

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

  408.         block1[0] += 1024;

  409. 

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

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

  412.         idct248_ref(img_dest1, 8, block);

  413. 

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

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

  416.         idct248_put(img_dest, 8, block);

  417. 

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

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

  420.             if (v == 255)

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

  422.             if (v > err_max)

  423.                 err_max = v;

  424.         }

  425.     }

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

  427. 

  428.     if (!speed)

  429.         return;

  430. 

  431.     ti = av_gettime();

  432.     it1 = 0;

  433.     do {

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

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

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

  437.             idct248_put(img_dest, 8, block);

  438.         }

  439.         it1 += NB_ITS_SPEED;

  440.         ti1 = av_gettime() - ti;

  441.     } while (ti1 < 1000000);

  442.     emms_c();

  443. 

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

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

  446. }

  447. 

  448. static void help(void)

  449. {

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

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

  452.            "            1 -> test with random sparse matrixes\n"

  453.            "            2 -> do 3. test from mpeg4 std\n"

  454.            "-i          test IDCT implementations\n"

  455.            "-4          test IDCT248 implementations\n"

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

  457. }

  458. 

  459. #if !HAVE_GETOPT

  460. #include "compat/getopt.c"

  461. #endif

  462. 

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

  464. {

  465.     int test_idct = 0, test_248_dct = 0;

  466.     int c, i;

  467.     int test = 1;

  468.     int speed = 0;

  469.     int err = 0;

  470. 

  471.     ff_ref_dct_init();

  472.     idct_mmx_init();

  473. 

  474.     for (;;) {

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

  476.         if (c == -1)

  477.             break;

  478.         switch (c) {

  479.         case 'i':

  480.             test_idct = 1;

  481.             break;

  482.         case '4':

  483.             test_248_dct = 1;

  484.             break;

  485.         case 't':

  486.             speed = 1;

  487.             break;

  488.         default:

  489.         case 'h':

  490.             help();

  491.             return 0;

  492.         }

  493.     }

  494. 

  495.     if (optind < argc)

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

  497. 

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

  499. 

  500.     if (test_248_dct) {

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

  502.     } else {

  503.         const int cpu_flags = av_get_cpu_flags();

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

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

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

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

  508.             }

  509.     }

  510. 

  511.     if (err)

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

  513. 

  514.     return !!err;

  515. }