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.
33576 Commits
32 Branches
940MB
Tree: db465be45d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'db465be45d'
${ noResults }
FFmpeg/libavcodec/dct-test.c

543 lines
15KB
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 <stdlib.h>

  29. #include <stdio.h>

  30. #include <string.h>

  31. #include <sys/time.h>

  32. #include <unistd.h>

  33. #include <math.h>

  34. 

  35. #include "libavutil/cpu.h"

  36. #include "libavutil/common.h"

  37. #include "libavutil/lfg.h"

  38. 

  39. #include "simple_idct.h"

  40. #include "aandcttab.h"

  41. #include "faandct.h"

  42. #include "faanidct.h"

  43. #include "x86/idct_xvid.h"

  44. #include "dctref.h"

  45. 

  46. #undef printf

  47. 

  48. void ff_mmx_idct(DCTELEM *data);

  49. void ff_mmxext_idct(DCTELEM *data);

  50. 

  51. // BFIN

  52. void ff_bfin_idct(DCTELEM *block);

  53. void ff_bfin_fdct(DCTELEM *block);

  54. 

  55. // ALTIVEC

  56. void ff_fdct_altivec(DCTELEM *block);

  57. //void ff_idct_altivec(DCTELEM *block);?? no routine

  58. 

  59. // ARM

  60. void ff_j_rev_dct_arm(DCTELEM *data);

  61. void ff_simple_idct_arm(DCTELEM *data);

  62. void ff_simple_idct_armv5te(DCTELEM *data);

  63. void ff_simple_idct_armv6(DCTELEM *data);

  64. void ff_simple_idct_neon(DCTELEM *data);

  65. 

  66. void ff_simple_idct_axp(DCTELEM *data);

  67. 

  68. struct algo {

  69.     const char *name;

  70.     void (*func)(DCTELEM *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

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

  87.     { "MMX2",           ff_fdct_mmx2,          NO_PERM,   AV_CPU_FLAG_MMX2    },

  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

  109. #if CONFIG_GPL

  110.     { "LIBMPEG2-MMX",   ff_mmx_idct,           MMX_PERM,  AV_CPU_FLAG_MMX,  1 },

  111.     { "LIBMPEG2-MMX2",  ff_mmxext_idct,        MMX_PERM,  AV_CPU_FLAG_MMX2, 1 },

  112. #endif

  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.     { "XVID-MMX2",      ff_idct_xvid_mmx2,     NO_PERM,   AV_CPU_FLAG_MMX2, 1 },

  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  },

  129. #endif

  130. #if HAVE_ARMV6

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

  132. #endif

  133. #if HAVE_NEON

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

  135. #endif

  136. 

  137. #if ARCH_ALPHA

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

  139. #endif

  140. 

  141.     { 0 }

  142. };

  143. 

  144. #define AANSCALE_BITS 12

  145. 

  146. static int64_t gettime(void)

  147. {

  148.     struct timeval tv;

  149.     gettimeofday(&tv, NULL);

  150.     return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;

  151. }

  152. 

  153. #define NB_ITS 20000

  154. #define NB_ITS_SPEED 50000

  155. 

  156. static short idct_mmx_perm[64];

  157. 

  158. static short idct_simple_mmx_perm[64] = {

  159.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  160.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  161.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  162.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  163.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  164.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  165.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  166.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  167. };

  168. 

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

  170. 

  171. static void idct_mmx_init(void)

  172. {

  173.     int i;

  174. 

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

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

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

  178.     }

  179. }

  180. 

  181. DECLARE_ALIGNED(16, static DCTELEM, block)[64];

  182. DECLARE_ALIGNED(8,  static DCTELEM, block1)[64];

  183. 

  184. static inline void mmx_emms(void)

  185. {

  186. #if HAVE_MMX

  187.     if (cpu_flags & AV_CPU_FLAG_MMX)

  188.         __asm__ volatile ("emms\n\t");

  189. #endif

  190. }

  191. 

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

  193. {

  194.     int i, j;

  195. 

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

  197. 

  198.     switch (test) {

  199.     case 0:

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

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

  202.         if (is_idct) {

  203.             ff_ref_fdct(block);

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

  205.                 block[i] >>= 3;

  206.         }

  207.         break;

  208.     case 1:

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

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

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

  212.         break;

  213.     case 2:

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

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

  216.         break;

  217.     }

  218. }

  219. 

  220. static void permute(DCTELEM dst[64], const DCTELEM src[64], int perm)

  221. {

  222.     int i;

  223. 

  224.     if (perm == MMX_PERM) {

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

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

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

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

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

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

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

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

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

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

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

  236.     } else {

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

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

  239.     }

  240. }

  241. 

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

  243. {

  244.     void (*ref)(DCTELEM *block) = is_idct ? ff_ref_idct : ff_ref_fdct;

  245.     int it, i, scale;

  246.     int err_inf, v;

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

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

  249.     int maxout = 0;

  250.     int blockSumErrMax = 0, blockSumErr;

  251.     AVLFG prng;

  252.     double omse, ome;

  253.     int spec_err;

  254. 

  255.     av_lfg_init(&prng, 1);

  256. 

  257.     err_inf = 0;

  258.     err2 = 0;

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

  260.         sysErr[i] = 0;

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

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

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

  264. 

  265.         dct->func(block);

  266.         mmx_emms();

  267. 

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

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

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

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

  272.             }

  273.         }

  274. 

  275.         ref(block1);

  276. 

  277.         blockSumErr = 0;

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

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

  280.             err_sum += err;

  281.             v = abs(err);

  282.             if (v > err_inf)

  283.                 err_inf = v;

  284.             err2 += v * v;

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

  286.             blockSumErr += v;

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

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

  289.         }

  290.         if (blockSumErrMax < blockSumErr)

  291.             blockSumErrMax = blockSumErr;

  292.     }

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

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

  295. 

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

  297.         if (i % 8 == 0)

  298.             printf("\n");

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

  300.     }

  301.     printf("\n");

  302. 

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

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

  305. 

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

  307. 

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

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

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

  311.            maxout, blockSumErrMax);

  312. 

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

  314.         return 1;

  315. 

  316.     if (!speed)

  317.         return 0;

  318. 

  319.     /* speed test */

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

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

  322. 

  323.     ti = gettime();

  324.     it1 = 0;

  325.     do {

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

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

  328.             dct->func(block);

  329.         }

  330.         it1 += NB_ITS_SPEED;

  331.         ti1 = gettime() - ti;

  332.     } while (ti1 < 1000000);

  333.     mmx_emms();

  334. 

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

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

  337. 

  338.     return 0;

  339. }

  340. 

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

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

  343. 

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

  345. {

  346.     static int init;

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

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

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

  350.     double s, sum, v;

  351.     int i, j, k;

  352. 

  353.     if (!init) {

  354.         init = 1;

  355. 

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

  357.             sum = 0;

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

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

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

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

  362.             }

  363.         }

  364. 

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

  366.             sum = 0;

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

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

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

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

  371.             }

  372.         }

  373.     }

  374. 

  375.     /* butterfly */

  376.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  383.         }

  384.     }

  385. 

  386.     /* idct8 on lines */

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

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

  389.             sum = 0;

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

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

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

  393.         }

  394.     }

  395. 

  396.     /* idct4 */

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

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

  399.             /* top */

  400.             sum = 0;

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

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

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

  404. 

  405.             /* bottom */

  406.             sum = 0;

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

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

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

  410.         }

  411.     }

  412. 

  413.     /* clamp and store the result */

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

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

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

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

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

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

  420.         }

  421.     }

  422. }

  423. 

  424. static void idct248_error(const char *name,

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

  426.                                               int16_t *block),

  427.                           int speed)

  428. {

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

  430.     AVLFG prng;

  431. 

  432.     av_lfg_init(&prng, 1);

  433. 

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

  435.        important here) */

  436.     err_max = 0;

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

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

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

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

  441.         block1[0] += 1024;

  442. 

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

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

  445.         idct248_ref(img_dest1, 8, block);

  446. 

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

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

  449.         idct248_put(img_dest, 8, block);

  450. 

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

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

  453.             if (v == 255)

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

  455.             if (v > err_max)

  456.                 err_max = v;

  457.         }

  458.     }

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

  460. 

  461.     if (!speed)

  462.         return;

  463. 

  464.     ti = gettime();

  465.     it1 = 0;

  466.     do {

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

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

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

  470.             idct248_put(img_dest, 8, block);

  471.         }

  472.         it1 += NB_ITS_SPEED;

  473.         ti1 = gettime() - ti;

  474.     } while (ti1 < 1000000);

  475.     mmx_emms();

  476. 

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

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

  479. }

  480. 

  481. static void help(void)

  482. {

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

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

  485.            "            1 -> test with random sparse matrixes\n"

  486.            "            2 -> do 3. test from mpeg4 std\n"

  487.            "-i          test IDCT implementations\n"

  488.            "-4          test IDCT248 implementations\n"

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

  490. }

  491. 

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

  493. {

  494.     int test_idct = 0, test_248_dct = 0;

  495.     int c, i;

  496.     int test = 1;

  497.     int speed = 0;

  498.     int err = 0;

  499. 

  500.     cpu_flags = av_get_cpu_flags();

  501. 

  502.     ff_ref_dct_init();

  503.     idct_mmx_init();

  504. 

  505.     for (;;) {

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

  507.         if (c == -1)

  508.             break;

  509.         switch (c) {

  510.         case 'i':

  511.             test_idct = 1;

  512.             break;

  513.         case '4':

  514.             test_248_dct = 1;

  515.             break;

  516.         case 't':

  517.             speed = 1;

  518.             break;

  519.         default:

  520.         case 'h':

  521.             help();

  522.             return 0;

  523.         }

  524.     }

  525. 

  526.     if (optind < argc)

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

  528. 

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

  530. 

  531.     if (test_248_dct) {

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

  533.     } else {

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

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

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

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

  538.             }

  539.     }

  540. 

  541.     return err;

  542. }