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.
64841 Commits
32 Branches
687MB
Tree: d986c414de
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd986c414de'
${ noResults }
FFmpeg/libavcodec/dct-test.c

508 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/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 "dctref.h"

  49. 

  50. struct algo {

  51.     const char *name;

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

  53.     enum idct_permutation_type perm_type;

  54.     int cpu_flag;

  55.     int nonspec;

  56. };

  57. 

  58. static const struct algo fdct_tab[4] = {

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

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

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

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

  63. };

  64. 

  65. static void ff_prores_idct_wrap(int16_t *dst){

  66.     DECLARE_ALIGNED(16, static int16_t, qmat)[64];

  67.     int i;

  68. 

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

  70.         qmat[i]=4;

  71.     }

  72.     ff_prores_idct(dst, qmat);

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

  74.          dst[i] -= 512;

  75.     }

  76. }

  77. 

  78. static const struct algo idct_tab[5] = {

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

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

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

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

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

  84. };

  85. 

  86. #if ARCH_ARM

  87. #include "arm/dct-test.c"

  88. #elif ARCH_PPC

  89. #include "ppc/dct-test.c"

  90. #elif ARCH_X86

  91. #include "x86/dct-test.c"

  92. #else

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

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

  95. #endif

  96. 

  97. #define AANSCALE_BITS 12

  98. 

  99. #define NB_ITS 20000

  100. #define NB_ITS_SPEED 50000

  101. 

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

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

  104. 

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

  106. {

  107.     int i, j;

  108. 

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

  110. 

  111.     switch (test) {

  112.     case 0:

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

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

  115.         if (is_idct) {

  116.             ff_ref_fdct(block);

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

  118.                 block[i] >>= 3;

  119.         }

  120.         break;

  121.     case 1:

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

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

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

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

  126.         }

  127.         break;

  128.     case 2:

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

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

  131.         break;

  132.     }

  133. }

  134. 

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

  136.                     enum idct_permutation_type perm_type)

  137. {

  138.     int i;

  139. 

  140. #if ARCH_X86

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

  142.         return;

  143. #endif

  144. 

  145.     switch (perm_type) {

  146.     case FF_IDCT_PERM_LIBMPEG2:

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

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

  149.         break;

  150.     case FF_IDCT_PERM_PARTTRANS:

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

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

  153.         break;

  154.     case FF_IDCT_PERM_TRANSPOSE:

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

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

  157.         break;

  158.     default:

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

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

  161.         break;

  162.     }

  163. }

  164. 

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

  166. {

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

  168.     int it, i, scale;

  169.     int err_inf, v;

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

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

  172.     int maxout = 0;

  173.     int blockSumErrMax = 0, blockSumErr;

  174.     AVLFG prng;

  175.     const int vals=1<<bits;

  176.     double omse, ome;

  177.     int spec_err;

  178. 

  179.     av_lfg_init(&prng, 1);

  180. 

  181.     err_inf = 0;

  182.     err2 = 0;

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

  184.         sysErr[i] = 0;

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

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

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

  188. 

  189.         dct->func(block);

  190.         emms_c();

  191. 

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

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

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

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

  196.             }

  197.         }

  198. 

  199.         ref(block1);

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

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

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

  203. 

  204.         blockSumErr = 0;

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

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

  207.             err_sum += err;

  208.             v = abs(err);

  209.             if (v > err_inf)

  210.                 err_inf = v;

  211.             err2 += v * v;

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

  213.             blockSumErr += v;

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

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

  216.         }

  217.         if (blockSumErrMax < blockSumErr)

  218.             blockSumErrMax = blockSumErr;

  219.     }

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

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

  222. 

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

  224.         if (i % 8 == 0)

  225.             printf("\n");

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

  227.     }

  228.     printf("\n");

  229. 

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

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

  232. 

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

  234. 

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

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

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

  238.            maxout, blockSumErrMax);

  239. 

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

  241.         return 1;

  242. 

  243.     if (!speed)

  244.         return 0;

  245. 

  246.     /* speed test */

  247. 

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

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

  250. 

  251.     ti = av_gettime_relative();

  252.     it1 = 0;

  253.     do {

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

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

  256.             dct->func(block);

  257.         }

  258.         emms_c();

  259.         it1 += NB_ITS_SPEED;

  260.         ti1 = av_gettime_relative() - ti;

  261.     } while (ti1 < 1000000);

  262. 

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

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

  265. 

  266.     return 0;

  267. }

  268. 

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

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

  271. 

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

  273. {

  274.     static int init;

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

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

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

  278.     double s, sum, v;

  279.     int i, j, k;

  280. 

  281.     if (!init) {

  282.         init = 1;

  283. 

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

  285.             sum = 0;

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

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

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

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

  290.             }

  291.         }

  292. 

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

  294.             sum = 0;

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

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

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

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

  299.             }

  300.         }

  301.     }

  302. 

  303.     /* butterfly */

  304.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  311.         }

  312.     }

  313. 

  314.     /* idct8 on lines */

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

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

  317.             sum = 0;

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

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

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

  321.         }

  322.     }

  323. 

  324.     /* idct4 */

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

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

  327.             /* top */

  328.             sum = 0;

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

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

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

  332. 

  333.             /* bottom */

  334.             sum = 0;

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

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

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

  338.         }

  339.     }

  340. 

  341.     /* clamp and store the result */

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

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

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

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

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

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

  348.         }

  349.     }

  350. }

  351. 

  352. static void idct248_error(const char *name,

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

  354.                                               int16_t *block),

  355.                           int speed)

  356. {

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

  358.     AVLFG prng;

  359. 

  360.     av_lfg_init(&prng, 1);

  361. 

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

  363.        important here) */

  364.     err_max = 0;

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

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

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

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

  369.         block1[0] += 1024;

  370. 

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

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

  373.         idct248_ref(img_dest1, 8, block);

  374. 

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

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

  377.         idct248_put(img_dest, 8, block);

  378. 

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

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

  381.             if (v == 255)

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

  383.             if (v > err_max)

  384.                 err_max = v;

  385.         }

  386. #if 0

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

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

  389.             int j;

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

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

  392.             }

  393.             printf("\n");

  394.         }

  395. 

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

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

  398.             int j;

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

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

  401.             }

  402.             printf("\n");

  403.         }

  404. #endif

  405.     }

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

  407. 

  408.     if (!speed)

  409.         return;

  410. 

  411.     ti = av_gettime_relative();

  412.     it1 = 0;

  413.     do {

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

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

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

  417.             idct248_put(img_dest, 8, block);

  418.         }

  419.         emms_c();

  420.         it1 += NB_ITS_SPEED;

  421.         ti1 = av_gettime_relative() - ti;

  422.     } while (ti1 < 1000000);

  423. 

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

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

  426. }

  427. 

  428. static void help(void)

  429. {

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

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

  432.            "            1 -> test with random sparse matrixes\n"

  433.            "            2 -> do 3. test from mpeg4 std\n"

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

  435.            "-i          test IDCT implementations\n"

  436.            "-4          test IDCT248 implementations\n"

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

  438. }

  439. 

  440. #if !HAVE_GETOPT

  441. #include "compat/getopt.c"

  442. #endif

  443. 

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

  445. {

  446.     int test_idct = 0, test_248_dct = 0;

  447.     int c, i;

  448.     int test = 1;

  449.     int speed = 0;

  450.     int err = 0;

  451.     int bits=8;

  452. 

  453.     ff_ref_dct_init();

  454. 

  455.     for (;;) {

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

  457.         if (c == -1)

  458.             break;

  459.         switch (c) {

  460.         case 'i':

  461.             test_idct = 1;

  462.             break;

  463.         case '4':

  464.             test_248_dct = 1;

  465.             break;

  466.         case 't':

  467.             speed = 1;

  468.             break;

  469.         default:

  470.         case 'h':

  471.             help();

  472.             return 0;

  473.         }

  474.     }

  475. 

  476.     if (optind < argc)

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

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

  479. 

  480.     printf("ffmpeg DCT/IDCT test\n");

  481. 

  482.     if (test_248_dct) {

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

  484.     } else {

  485.         const int cpu_flags = av_get_cpu_flags();

  486.         if (test_idct) {

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

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

  489. 

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

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

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

  493.         } else {

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

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

  496. 

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

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

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

  500.         }

  501.     }

  502. 

  503.     if (err)

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

  505. 

  506.     return !!err;

  507. }