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.
86972 Commits
32 Branches
369MB
Tree: 86cbffdc4d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '86cbffdc4d'
${ noResults }
FFmpeg/libavcodec/tests/dct.c

529 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 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/time.h"

  42. 

  43. #include "libavcodec/dct.h"

  44. #include "libavcodec/idctdsp.h"

  45. #include "libavcodec/simple_idct.h"

  46. #include "libavcodec/xvididct.h"

  47. #include "libavcodec/aandcttab.h"

  48. #include "libavcodec/faandct.h"

  49. #include "libavcodec/faanidct.h"

  50. #include "libavcodec/dctref.h"

  51. 

  52. struct algo {

  53.     const char *name;

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

  55.     enum idct_permutation_type perm_type;

  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 void ff_prores_idct_wrap(int16_t *dst){

  70.     LOCAL_ALIGNED(16, int16_t, qmat, [64]);

  71.     int i;

  72. 

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

  74.         qmat[i]=4;

  75.     }

  76.     ff_prores_idct(dst, qmat);

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

  78.          dst[i] -= 512;

  79.     }

  80. }

  81. 

  82. static const struct algo idct_tab[] = {

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

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

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

  86.     { "SIMPLE-C10",  ff_simple_idct_10,    FF_IDCT_PERM_NONE },

  87.     { "SIMPLE-C12",  ff_simple_idct_12,    FF_IDCT_PERM_NONE, 0, 1 },

  88.     { "PR-C",        ff_prores_idct_wrap,  FF_IDCT_PERM_NONE, 0, 1 },

  89. #if CONFIG_FAANIDCT

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

  91. #endif /* CONFIG_FAANIDCT */

  92. #if CONFIG_MPEG4_DECODER

  93.     { "XVID",        ff_xvid_idct,         FF_IDCT_PERM_NONE, 0, 1 },

  94. #endif /* CONFIG_MPEG4_DECODER */

  95. };

  96. 

  97. #if ARCH_AARCH64

  98. #include "aarch64/dct.c"

  99. #elif ARCH_ARM

  100. #include "arm/dct.c"

  101. #elif ARCH_PPC

  102. #include "ppc/dct.c"

  103. #elif ARCH_X86

  104. #include "x86/dct.c"

  105. #else

  106. static const struct algo fdct_tab_arch[] = { { 0 } };

  107. static const struct algo idct_tab_arch[] = { { 0 } };

  108. #endif

  109. 

  110. #define AANSCALE_BITS 12

  111. 

  112. #define NB_ITS 20000

  113. #define NB_ITS_SPEED 50000

  114. 

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

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

  117. 

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

  119. {

  120.     int i, j;

  121. 

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

  123. 

  124.     switch (test) {

  125.     case 0:

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

  127.             block[i] = (av_lfg_get(prng) % (2*vals)) -vals;

  128.         if (is_idct) {

  129.             ff_ref_fdct(block);

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

  131.                 block[i] >>= 3;

  132.         }

  133.         break;

  134.     case 1:

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

  136.         for (i = 0; i < j; i++) {

  137.             int idx = av_lfg_get(prng) % 64;

  138.             block[idx] = av_lfg_get(prng) % (2*vals) -vals;

  139.         }

  140.         break;

  141.     case 2:

  142.         block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals);

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

  144.         break;

  145.     }

  146. }

  147. 

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

  149.                     enum idct_permutation_type perm_type)

  150. {

  151.     int i;

  152. 

  153. #if ARCH_X86

  154.     if (permute_x86(dst, src, perm_type))

  155.         return;

  156. #endif

  157. 

  158.     switch (perm_type) {

  159.     case FF_IDCT_PERM_LIBMPEG2:

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

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

  162.         break;

  163.     case FF_IDCT_PERM_PARTTRANS:

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

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

  166.         break;

  167.     case FF_IDCT_PERM_TRANSPOSE:

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

  169.             dst[(i>>3) | ((i<<3)&0x38)] = src[i];

  170.         break;

  171.     default:

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

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

  174.         break;

  175.     }

  176. }

  177. 

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

  179. {

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

  181.     int it, i, scale;

  182.     int err_inf, v;

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

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

  185.     int maxout = 0;

  186.     int blockSumErrMax = 0, blockSumErr;

  187.     AVLFG prng;

  188.     const int vals=1<<bits;

  189.     double omse, ome;

  190.     int spec_err;

  191. 

  192.     av_lfg_init(&prng, 1);

  193. 

  194.     err_inf = 0;

  195.     err2 = 0;

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

  197.         sysErr[i] = 0;

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

  199.         init_block(block1, test, is_idct, &prng, vals);

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

  201. 

  202.         dct->func(block);

  203.         emms_c();

  204. 

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

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

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

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

  209.             }

  210.         }

  211. 

  212.         ref(block1);

  213.         if (!strcmp(dct->name, "PR-SSE2"))

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

  215.                 block1[i] = av_clip(block1[i], 4-512, 1019-512);

  216. 

  217.         blockSumErr = 0;

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

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

  220.             err_sum += err;

  221.             v = abs(err);

  222.             if (v > err_inf)

  223.                 err_inf = v;

  224.             err2 += v * v;

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

  226.             blockSumErr += v;

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

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

  229.         }

  230.         if (blockSumErrMax < blockSumErr)

  231.             blockSumErrMax = blockSumErr;

  232.     }

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

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

  235. 

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

  237.         if (i % 8 == 0)

  238.             printf("\n");

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

  240.     }

  241.     printf("\n");

  242. 

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

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

  245. 

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

  247.     if (test < 2)

  248.         spec_err = is_idct && ((double) sysErrMax / NB_ITS > 0.015);

  249. 

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

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

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

  253.            maxout, blockSumErrMax);

  254. 

  255.     if (spec_err && !dct->nonspec) {

  256.         printf("Failed!\n");

  257.         return 1;

  258.     }

  259. 

  260.     if (!speed)

  261.         return 0;

  262. 

  263.     /* speed test */

  264. 

  265.     init_block(block, test, is_idct, &prng, vals);

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

  267. 

  268.     ti = av_gettime_relative();

  269.     it1 = 0;

  270.     do {

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

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

  273.             dct->func(block);

  274.         }

  275.         emms_c();

  276.         it1 += NB_ITS_SPEED;

  277.         ti1 = av_gettime_relative() - ti;

  278.     } while (ti1 < 1000000);

  279. 

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

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

  282. 

  283.     return 0;

  284. }

  285. 

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

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

  288. 

  289. static void idct248_ref(uint8_t *dest, ptrdiff_t linesize, int16_t *block)

  290. {

  291.     static int init;

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

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

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

  295.     double s, sum, v;

  296.     int i, j, k;

  297. 

  298.     if (!init) {

  299.         init = 1;

  300. 

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

  302.             sum = 0;

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

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

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

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

  307.             }

  308.         }

  309. 

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

  311.             sum = 0;

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

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

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

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

  316.             }

  317.         }

  318.     }

  319. 

  320.     /* butterfly */

  321.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  328.         }

  329.     }

  330. 

  331.     /* idct8 on lines */

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

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

  334.             sum = 0;

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

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

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

  338.         }

  339.     }

  340. 

  341.     /* idct4 */

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

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

  344.             /* top */

  345.             sum = 0;

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

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

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

  349. 

  350.             /* bottom */

  351.             sum = 0;

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

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

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

  355.         }

  356.     }

  357. 

  358.     /* clamp and store the result */

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

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

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

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

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

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

  365.         }

  366.     }

  367. }

  368. 

  369. static void idct248_error(const char *name,

  370.                           void (*idct248_put)(uint8_t *dest,

  371.                                               ptrdiff_t line_size,

  372.                                               int16_t *block),

  373.                           int speed)

  374. {

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

  376.     AVLFG prng;

  377. 

  378.     av_lfg_init(&prng, 1);

  379. 

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

  381.        important here) */

  382.     err_max = 0;

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

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

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

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

  387.         block1[0] += 1024;

  388. 

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

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

  391.         idct248_ref(img_dest1, 8, block);

  392. 

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

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

  395.         idct248_put(img_dest, 8, block);

  396. 

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

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

  399.             if (v == 255)

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

  401.             if (v > err_max)

  402.                 err_max = v;

  403.         }

  404. #if 0

  405.         printf("ref=\n");

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

  407.             int j;

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

  409.                 printf(" %3d", img_dest1[i*8+j]);

  410.             }

  411.             printf("\n");

  412.         }

  413. 

  414.         printf("out=\n");

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

  416.             int j;

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

  418.                 printf(" %3d", img_dest[i*8+j]);

  419.             }

  420.             printf("\n");

  421.         }

  422. #endif

  423.     }

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

  425. 

  426.     if (!speed)

  427.         return;

  428. 

  429.     ti = av_gettime_relative();

  430.     it1 = 0;

  431.     do {

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

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

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

  435.             idct248_put(img_dest, 8, block);

  436.         }

  437.         emms_c();

  438.         it1 += NB_ITS_SPEED;

  439.         ti1 = av_gettime_relative() - ti;

  440.     } while (ti1 < 1000000);

  441. 

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

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

  444. }

  445. 

  446. static void help(void)

  447. {

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

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

  450.            "            1 -> test with random sparse matrixes\n"

  451.            "            2 -> do 3. test from MPEG-4 std\n"

  452.            "bits        Number of time domain bits to use, 8 is default\n"

  453.            "-i          test IDCT implementations\n"

  454.            "-4          test IDCT248 implementations\n"

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

  456. }

  457. 

  458. #if !HAVE_GETOPT

  459. #include "compat/getopt.c"

  460. #endif

  461. 

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

  463. {

  464.     int test_idct = 0, test_248_dct = 0;

  465.     int c, i;

  466.     int test = 1;

  467.     int speed = 0;

  468.     int err = 0;

  469.     int bits=8;

  470. 

  471.     ff_ref_dct_init();

  472. 

  473.     for (;;) {

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

  475.         if (c == -1)

  476.             break;

  477.         switch (c) {

  478.         case 'i':

  479.             test_idct = 1;

  480.             break;

  481.         case '4':

  482.             test_248_dct = 1;

  483.             break;

  484.         case 't':

  485.             speed = 1;

  486.             break;

  487.         default:

  488.         case 'h':

  489.             help();

  490.             return 0;

  491.         }

  492.     }

  493. 

  494.     if (optind < argc)

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

  496.     if(optind+1 < argc) bits= atoi(argv[optind+1]);

  497. 

  498.     printf("ffmpeg 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.         if (test_idct) {

  505.             for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)

  506.                 err |= dct_error(&idct_tab[i], test, test_idct, speed, bits);

  507. 

  508.             for (i = 0; idct_tab_arch[i].name; i++)

  509.                 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))

  510.                     err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits);

  511.         }

  512. #if CONFIG_FDCTDSP

  513.         else {

  514.             for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)

  515.                 err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits);

  516. 

  517.             for (i = 0; fdct_tab_arch[i].name; i++)

  518.                 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))

  519.                     err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits);

  520.         }

  521. #endif /* CONFIG_FDCTDSP */

  522.     }

  523. 

  524.     if (err)

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

  526. 

  527.     return !!err;

  528. }