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.
34105 Commits
32 Branches
283MB
Tree: 667fb97a65
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '667fb97a65'
${ noResults }
FFmpeg/libavcodec/dct-test.c

535 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 "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 "simple_idct.h"

  43. #include "aandcttab.h"

  44. #include "faandct.h"

  45. #include "faanidct.h"

  46. #include "x86/idct_xvid.h"

  47. #include "dctref.h"

  48. 

  49. #undef printf

  50. 

  51. void ff_mmx_idct(DCTELEM *data);

  52. void ff_mmxext_idct(DCTELEM *data);

  53. 

  54. // BFIN

  55. void ff_bfin_idct(DCTELEM *block);

  56. void ff_bfin_fdct(DCTELEM *block);

  57. 

  58. // ALTIVEC

  59. void ff_fdct_altivec(DCTELEM *block);

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

  61. 

  62. // ARM

  63. void ff_j_rev_dct_arm(DCTELEM *data);

  64. void ff_simple_idct_arm(DCTELEM *data);

  65. void ff_simple_idct_armv5te(DCTELEM *data);

  66. void ff_simple_idct_armv6(DCTELEM *data);

  67. void ff_simple_idct_neon(DCTELEM *data);

  68. 

  69. void ff_simple_idct_axp(DCTELEM *data);

  70. 

  71. struct algo {

  72.     const char *name;

  73.     void (*func)(DCTELEM *block);

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

  75.                      SSE2_PERM, PARTTRANS_PERM } format;

  76.     int mm_support;

  77.     int nonspec;

  78. };

  79. 

  80. static int cpu_flags;

  81. 

  82. static const struct algo fdct_tab[] = {

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

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

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

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

  87. 

  88. #if HAVE_MMX

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

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

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

  92. #endif

  93. 

  94. #if HAVE_ALTIVEC

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

  96. #endif

  97. 

  98. #if ARCH_BFIN

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

  100. #endif

  101. 

  102.     { 0 }

  103. };

  104. 

  105. static const struct algo idct_tab[] = {

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

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

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

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

  110. 

  111. #if HAVE_MMX

  112. #if CONFIG_GPL

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

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

  115. #endif

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

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

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

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

  120. #endif

  121. 

  122. #if ARCH_BFIN

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

  124. #endif

  125. 

  126. #if ARCH_ARM

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

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

  129. #endif

  130. #if HAVE_ARMV5TE

  131.     { "SIMPLE-ARMV5TE", ff_simple_idct_armv5te,NO_PERM  },

  132. #endif

  133. #if HAVE_ARMV6

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

  135. #endif

  136. #if HAVE_NEON

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

  138. #endif

  139. 

  140. #if ARCH_ALPHA

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

  142. #endif

  143. 

  144.     { 0 }

  145. };

  146. 

  147. #define AANSCALE_BITS 12

  148. 

  149. #define NB_ITS 20000

  150. #define NB_ITS_SPEED 50000

  151. 

  152. static short idct_mmx_perm[64];

  153. 

  154. static short idct_simple_mmx_perm[64] = {

  155.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  156.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  157.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  158.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  159.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  160.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  161.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  162.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  163. };

  164. 

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

  166. 

  167. static void idct_mmx_init(void)

  168. {

  169.     int i;

  170. 

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

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

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

  174.     }

  175. }

  176. 

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

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

  179. 

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

  181. {

  182.     int i, j;

  183. 

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

  185. 

  186.     switch (test) {

  187.     case 0:

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

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

  190.         if (is_idct) {

  191.             ff_ref_fdct(block);

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

  193.                 block[i] >>= 3;

  194.         }

  195.         break;

  196.     case 1:

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

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

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

  200.         break;

  201.     case 2:

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

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

  204.         break;

  205.     }

  206. }

  207. 

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

  209. {

  210.     int i;

  211. 

  212.     if (perm == MMX_PERM) {

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

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

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

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

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

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

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

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

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

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

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

  224.     } else {

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

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

  227.     }

  228. }

  229. 

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

  231. {

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

  233.     int it, i, scale;

  234.     int err_inf, v;

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

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

  237.     int maxout = 0;

  238.     int blockSumErrMax = 0, blockSumErr;

  239.     AVLFG prng;

  240.     double omse, ome;

  241.     int spec_err;

  242. 

  243.     av_lfg_init(&prng, 1);

  244. 

  245.     err_inf = 0;

  246.     err2 = 0;

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

  248.         sysErr[i] = 0;

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

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

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

  252. 

  253.         dct->func(block);

  254.         emms_c();

  255. 

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

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

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

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

  260.             }

  261.         }

  262. 

  263.         ref(block1);

  264. 

  265.         blockSumErr = 0;

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

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

  268.             err_sum += err;

  269.             v = abs(err);

  270.             if (v > err_inf)

  271.                 err_inf = v;

  272.             err2 += v * v;

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

  274.             blockSumErr += v;

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

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

  277.         }

  278.         if (blockSumErrMax < blockSumErr)

  279.             blockSumErrMax = blockSumErr;

  280.     }

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

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

  283. 

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

  285.         if (i % 8 == 0)

  286.             printf("\n");

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

  288.     }

  289.     printf("\n");

  290. 

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

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

  293. 

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

  295. 

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

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

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

  299.            maxout, blockSumErrMax);

  300. 

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

  302.         return 1;

  303. 

  304.     if (!speed)

  305.         return 0;

  306. 

  307.     /* speed test */

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

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

  310. 

  311.     ti = av_gettime();

  312.     it1 = 0;

  313.     do {

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

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

  316.             dct->func(block);

  317.         }

  318.         it1 += NB_ITS_SPEED;

  319.         ti1 = av_gettime() - ti;

  320.     } while (ti1 < 1000000);

  321.     emms_c();

  322. 

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

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

  325. 

  326.     return 0;

  327. }

  328. 

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

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

  331. 

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

  333. {

  334.     static int init;

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

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

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

  338.     double s, sum, v;

  339.     int i, j, k;

  340. 

  341.     if (!init) {

  342.         init = 1;

  343. 

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

  345.             sum = 0;

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

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

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

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

  350.             }

  351.         }

  352. 

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

  354.             sum = 0;

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

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

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

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

  359.             }

  360.         }

  361.     }

  362. 

  363.     /* butterfly */

  364.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  371.         }

  372.     }

  373. 

  374.     /* idct8 on lines */

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

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

  377.             sum = 0;

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

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

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

  381.         }

  382.     }

  383. 

  384.     /* idct4 */

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

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

  387.             /* top */

  388.             sum = 0;

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

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

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

  392. 

  393.             /* bottom */

  394.             sum = 0;

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

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

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

  398.         }

  399.     }

  400. 

  401.     /* clamp and store the result */

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

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

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

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

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

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

  408.         }

  409.     }

  410. }

  411. 

  412. static void idct248_error(const char *name,

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

  414.                                               int16_t *block),

  415.                           int speed)

  416. {

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

  418.     AVLFG prng;

  419. 

  420.     av_lfg_init(&prng, 1);

  421. 

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

  423.        important here) */

  424.     err_max = 0;

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

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

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

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

  429.         block1[0] += 1024;

  430. 

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

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

  433.         idct248_ref(img_dest1, 8, block);

  434. 

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

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

  437.         idct248_put(img_dest, 8, block);

  438. 

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

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

  441.             if (v == 255)

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

  443.             if (v > err_max)

  444.                 err_max = v;

  445.         }

  446.     }

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

  448. 

  449.     if (!speed)

  450.         return;

  451. 

  452.     ti = av_gettime();

  453.     it1 = 0;

  454.     do {

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

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

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

  458.             idct248_put(img_dest, 8, block);

  459.         }

  460.         it1 += NB_ITS_SPEED;

  461.         ti1 = av_gettime() - ti;

  462.     } while (ti1 < 1000000);

  463.     emms_c();

  464. 

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

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

  467. }

  468. 

  469. static void help(void)

  470. {

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

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

  473.            "            1 -> test with random sparse matrixes\n"

  474.            "            2 -> do 3. test from mpeg4 std\n"

  475.            "-i          test IDCT implementations\n"

  476.            "-4          test IDCT248 implementations\n"

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

  478. }

  479. 

  480. #if !HAVE_GETOPT

  481. #include "compat/getopt.c"

  482. #endif

  483. 

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

  485. {

  486.     int test_idct = 0, test_248_dct = 0;

  487.     int c, i;

  488.     int test = 1;

  489.     int speed = 0;

  490.     int err = 0;

  491. 

  492.     cpu_flags = av_get_cpu_flags();

  493. 

  494.     ff_ref_dct_init();

  495.     idct_mmx_init();

  496. 

  497.     for (;;) {

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

  499.         if (c == -1)

  500.             break;

  501.         switch (c) {

  502.         case 'i':

  503.             test_idct = 1;

  504.             break;

  505.         case '4':

  506.             test_248_dct = 1;

  507.             break;

  508.         case 't':

  509.             speed = 1;

  510.             break;

  511.         default:

  512.         case 'h':

  513.             help();

  514.             return 0;

  515.         }

  516.     }

  517. 

  518.     if (optind < argc)

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

  520. 

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

  522. 

  523.     if (test_248_dct) {

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

  525.     } else {

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

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

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

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

  530.             }

  531.     }

  532. 

  533.     return err;

  534. }