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.
38308 Commits
32 Branches
776MB
Tree: ce1e8045e0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ce1e8045e0'
${ noResults }
FFmpeg/libavcodec/dct-test.c

536 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. #undef printf

  51. 

  52. // BFIN

  53. void ff_bfin_idct(int16_t *block);

  54. void ff_bfin_fdct(int16_t *block);

  55. 

  56. // ALTIVEC

  57. void ff_fdct_altivec(int16_t *block);

  58. 

  59. // ARM

  60. void ff_j_rev_dct_arm(int16_t *data);

  61. void ff_simple_idct_arm(int16_t *data);

  62. void ff_simple_idct_armv5te(int16_t *data);

  63. void ff_simple_idct_armv6(int16_t *data);

  64. void ff_simple_idct_neon(int16_t *data);

  65. 

  66. void ff_simple_idct_axp(int16_t *data);

  67. 

  68. struct algo {

  69.     const char *name;

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

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

  72.                      SSE2_PERM, PARTTRANS_PERM } format;

  73.     int mm_support;

  74.     int nonspec;

  75. };

  76. 

  77. static int cpu_flags;

  78. 

  79. static const struct algo fdct_tab[] = {

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

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

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

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

  84. 

  85. #if HAVE_MMX_INLINE

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

  87. #endif

  88. #if HAVE_MMXEXT_INLINE

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

  90. #endif

  91. #if HAVE_SSE2_INLINE

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

  93. #endif

  94. 

  95. #if HAVE_ALTIVEC

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

  97. #endif

  98. 

  99. #if ARCH_BFIN

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

  101. #endif

  102. 

  103.     { 0 }

  104. };

  105. 

  106. static const struct algo idct_tab[] = {

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

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

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

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

  111. 

  112. #if HAVE_MMX_INLINE

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

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

  115. #endif

  116. #if HAVE_MMXEXT_INLINE

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

  118. #endif

  119. #if HAVE_SSE2_INLINE

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

  121. #endif

  122. 

  123. #if ARCH_BFIN

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

  125. #endif

  126. 

  127. #if ARCH_ARM

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

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

  130. #endif

  131. #if HAVE_ARMV5TE

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

  133. #endif

  134. #if HAVE_ARMV6

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

  136. #endif

  137. #if HAVE_NEON

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

  139. #endif

  140. 

  141. #if ARCH_ALPHA

  142.     { "SIMPLE-ALPHA",   ff_simple_idct_axp,    NO_PERM },

  143. #endif

  144. 

  145.     { 0 }

  146. };

  147. 

  148. #define AANSCALE_BITS 12

  149. 

  150. #define NB_ITS 20000

  151. #define NB_ITS_SPEED 50000

  152. 

  153. static short idct_mmx_perm[64];

  154. 

  155. static short idct_simple_mmx_perm[64] = {

  156.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  157.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  158.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  159.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  160.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  161.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  162.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  163.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  164. };

  165. 

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

  167. 

  168. static void idct_mmx_init(void)

  169. {

  170.     int i;

  171. 

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

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

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

  175.     }

  176. }

  177. 

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

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

  180. 

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

  182. {

  183.     int i, j;

  184. 

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

  186. 

  187.     switch (test) {

  188.     case 0:

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

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

  191.         if (is_idct) {

  192.             ff_ref_fdct(block);

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

  194.                 block[i] >>= 3;

  195.         }

  196.         break;

  197.     case 1:

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

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

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

  201.         break;

  202.     case 2:

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

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

  205.         break;

  206.     }

  207. }

  208. 

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

  210. {

  211.     int i;

  212. 

  213.     if (perm == MMX_PERM) {

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

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

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

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

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

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

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

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

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

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

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

  225.     } else {

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

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

  228.     }

  229. }

  230. 

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

  232. {

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

  234.     int it, i, scale;

  235.     int err_inf, v;

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

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

  238.     int maxout = 0;

  239.     int blockSumErrMax = 0, blockSumErr;

  240.     AVLFG prng;

  241.     double omse, ome;

  242.     int spec_err;

  243. 

  244.     av_lfg_init(&prng, 1);

  245. 

  246.     err_inf = 0;

  247.     err2 = 0;

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

  249.         sysErr[i] = 0;

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

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

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

  253. 

  254.         dct->func(block);

  255.         emms_c();

  256. 

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

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

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

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

  261.             }

  262.         }

  263. 

  264.         ref(block1);

  265. 

  266.         blockSumErr = 0;

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

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

  269.             err_sum += err;

  270.             v = abs(err);

  271.             if (v > err_inf)

  272.                 err_inf = v;

  273.             err2 += v * v;

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

  275.             blockSumErr += v;

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

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

  278.         }

  279.         if (blockSumErrMax < blockSumErr)

  280.             blockSumErrMax = blockSumErr;

  281.     }

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

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

  284. 

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

  286.         if (i % 8 == 0)

  287.             printf("\n");

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

  289.     }

  290.     printf("\n");

  291. 

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

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

  294. 

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

  296. 

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

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

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

  300.            maxout, blockSumErrMax);

  301. 

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

  303.         return 1;

  304. 

  305.     if (!speed)

  306.         return 0;

  307. 

  308.     /* speed test */

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

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

  311. 

  312.     ti = av_gettime();

  313.     it1 = 0;

  314.     do {

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

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

  317.             dct->func(block);

  318.         }

  319.         it1 += NB_ITS_SPEED;

  320.         ti1 = av_gettime() - ti;

  321.     } while (ti1 < 1000000);

  322.     emms_c();

  323. 

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

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

  326. 

  327.     return 0;

  328. }

  329. 

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

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

  332. 

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

  334. {

  335.     static int init;

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

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

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

  339.     double s, sum, v;

  340.     int i, j, k;

  341. 

  342.     if (!init) {

  343.         init = 1;

  344. 

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

  346.             sum = 0;

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

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

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

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

  351.             }

  352.         }

  353. 

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

  355.             sum = 0;

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

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

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

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

  360.             }

  361.         }

  362.     }

  363. 

  364.     /* butterfly */

  365.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  372.         }

  373.     }

  374. 

  375.     /* idct8 on lines */

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

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

  378.             sum = 0;

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

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

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

  382.         }

  383.     }

  384. 

  385.     /* idct4 */

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

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

  388.             /* top */

  389.             sum = 0;

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

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

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

  393. 

  394.             /* bottom */

  395.             sum = 0;

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

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

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

  399.         }

  400.     }

  401. 

  402.     /* clamp and store the result */

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

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

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

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

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

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

  409.         }

  410.     }

  411. }

  412. 

  413. static void idct248_error(const char *name,

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

  415.                                               int16_t *block),

  416.                           int speed)

  417. {

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

  419.     AVLFG prng;

  420. 

  421.     av_lfg_init(&prng, 1);

  422. 

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

  424.        important here) */

  425.     err_max = 0;

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

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

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

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

  430.         block1[0] += 1024;

  431. 

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

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

  434.         idct248_ref(img_dest1, 8, block);

  435. 

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

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

  438.         idct248_put(img_dest, 8, block);

  439. 

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

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

  442.             if (v == 255)

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

  444.             if (v > err_max)

  445.                 err_max = v;

  446.         }

  447.     }

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

  449. 

  450.     if (!speed)

  451.         return;

  452. 

  453.     ti = av_gettime();

  454.     it1 = 0;

  455.     do {

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

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

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

  459.             idct248_put(img_dest, 8, block);

  460.         }

  461.         it1 += NB_ITS_SPEED;

  462.         ti1 = av_gettime() - ti;

  463.     } while (ti1 < 1000000);

  464.     emms_c();

  465. 

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

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

  468. }

  469. 

  470. static void help(void)

  471. {

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

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

  474.            "            1 -> test with random sparse matrixes\n"

  475.            "            2 -> do 3. test from mpeg4 std\n"

  476.            "-i          test IDCT implementations\n"

  477.            "-4          test IDCT248 implementations\n"

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

  479. }

  480. 

  481. #if !HAVE_GETOPT

  482. #include "compat/getopt.c"

  483. #endif

  484. 

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

  486. {

  487.     int test_idct = 0, test_248_dct = 0;

  488.     int c, i;

  489.     int test = 1;

  490.     int speed = 0;

  491.     int err = 0;

  492. 

  493.     cpu_flags = av_get_cpu_flags();

  494. 

  495.     ff_ref_dct_init();

  496.     idct_mmx_init();

  497. 

  498.     for (;;) {

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

  500.         if (c == -1)

  501.             break;

  502.         switch (c) {

  503.         case 'i':

  504.             test_idct = 1;

  505.             break;

  506.         case '4':

  507.             test_248_dct = 1;

  508.             break;

  509.         case 't':

  510.             speed = 1;

  511.             break;

  512.         default:

  513.         case 'h':

  514.             help();

  515.             return 0;

  516.         }

  517.     }

  518. 

  519.     if (optind < argc)

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

  521. 

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

  523. 

  524.     if (test_248_dct) {

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

  526.     } else {

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

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

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

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

  531.             }

  532.     }

  533. 

  534.     return err;

  535. }