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.
49674 Commits
32 Branches
649MB
Tree: b5884db437
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b5884db437'
${ noResults }
FFmpeg/libavcodec/dct-test.c

584 lines
16KB
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 "dsputil.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. void ff_mmx_idct(int16_t *data);

  53. void ff_mmxext_idct(int16_t *data);

  54. 

  55. // BFIN

  56. void ff_bfin_idct(int16_t *block);

  57. void ff_bfin_fdct(int16_t *block);

  58. 

  59. // ALTIVEC

  60. void ff_fdct_altivec(int16_t *block);

  61. 

  62. // ARM

  63. void ff_j_rev_dct_arm(int16_t *data);

  64. void ff_simple_idct_arm(int16_t *data);

  65. void ff_simple_idct_armv5te(int16_t *data);

  66. void ff_simple_idct_armv6(int16_t *data);

  67. void ff_simple_idct_neon(int16_t *data);

  68. 

  69. void ff_simple_idct_axp(int16_t *data);

  70. 

  71. struct algo {

  72.     const char *name;

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

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

  75.                      SSE2_PERM, PARTTRANS_PERM, TRANSPOSE_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_INLINE

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

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

  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. #if ARCH_X86_64 && HAVE_MMX && HAVE_YASM

  106. void ff_prores_idct_put_10_sse2(uint16_t *dst, int linesize,

  107.                                 int16_t *block, int16_t *qmat);

  108. 

  109. static void ff_prores_idct_put_10_sse2_wrap(int16_t *dst){

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

  111.     DECLARE_ALIGNED(16, static int16_t, tmp)[64];

  112.     int i;

  113. 

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

  115.         qmat[i]=4;

  116.         tmp[i]= dst[i];

  117.     }

  118.     ff_prores_idct_put_10_sse2(dst, 16, tmp, qmat);

  119. }

  120. #endif

  121. 

  122. static const struct algo idct_tab[] = {

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

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

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

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

  127. 

  128. #if HAVE_MMX_INLINE

  129. #if CONFIG_GPL

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

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

  132. #endif

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

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

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

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

  137. #if ARCH_X86_64 && HAVE_YASM

  138.     { "PR-SSE2",        ff_prores_idct_put_10_sse2_wrap,     TRANSPOSE_PERM, AV_CPU_FLAG_SSE2, 1 },

  139. #endif

  140. #endif

  141. 

  142. #if ARCH_BFIN

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

  144. #endif

  145. 

  146. #if ARCH_ARM

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

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

  149. #endif

  150. #if HAVE_ARMV5TE

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

  152. #endif

  153. #if HAVE_ARMV6

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

  155. #endif

  156. #if HAVE_NEON

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

  158. #endif

  159. 

  160. #if ARCH_ALPHA

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

  162. #endif

  163. 

  164.     { 0 }

  165. };

  166. 

  167. #define AANSCALE_BITS 12

  168. 

  169. #define NB_ITS 20000

  170. #define NB_ITS_SPEED 50000

  171. 

  172. static short idct_mmx_perm[64];

  173. 

  174. static short idct_simple_mmx_perm[64] = {

  175.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  176.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  177.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  178.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  179.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  180.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  181.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  182.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  183. };

  184. 

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

  186. 

  187. static void idct_mmx_init(void)

  188. {

  189.     int i;

  190. 

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

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

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

  194.     }

  195. }

  196. 

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

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

  199. 

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

  201. {

  202.     int i, j;

  203. 

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

  205. 

  206.     switch (test) {

  207.     case 0:

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

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

  210.         if (is_idct) {

  211.             ff_ref_fdct(block);

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

  213.                 block[i] >>= 3;

  214.         }

  215.         break;

  216.     case 1:

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

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

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

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

  221.         }

  222.         break;

  223.     case 2:

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

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

  226.         break;

  227.     }

  228. }

  229. 

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

  231. {

  232.     int i;

  233. 

  234.     if (perm == MMX_PERM) {

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

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

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

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

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

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

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

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

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

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

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

  246.     } else if (perm == TRANSPOSE_PERM) {

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

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

  249.     } else {

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

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

  252.     }

  253. }

  254. 

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

  256. {

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

  258.     int it, i, scale;

  259.     int err_inf, v;

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

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

  262.     int maxout = 0;

  263.     int blockSumErrMax = 0, blockSumErr;

  264.     AVLFG prng;

  265.     const int vals=1<<bits;

  266.     double omse, ome;

  267.     int spec_err;

  268. 

  269.     av_lfg_init(&prng, 1);

  270. 

  271.     err_inf = 0;

  272.     err2 = 0;

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

  274.         sysErr[i] = 0;

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

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

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

  278. 

  279.         dct->func(block);

  280.         emms_c();

  281. 

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

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

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

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

  286.             }

  287.         }

  288. 

  289.         ref(block1);

  290. 

  291.         blockSumErr = 0;

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

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

  294.             err_sum += err;

  295.             v = abs(err);

  296.             if (v > err_inf)

  297.                 err_inf = v;

  298.             err2 += v * v;

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

  300.             blockSumErr += v;

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

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

  303.         }

  304.         if (blockSumErrMax < blockSumErr)

  305.             blockSumErrMax = blockSumErr;

  306.     }

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

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

  309. 

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

  311.         if (i % 8 == 0)

  312.             printf("\n");

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

  314.     }

  315.     printf("\n");

  316. 

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

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

  319. 

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

  321. 

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

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

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

  325.            maxout, blockSumErrMax);

  326. 

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

  328.         return 1;

  329. 

  330.     if (!speed)

  331.         return 0;

  332. 

  333.     /* speed test */

  334. 

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

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

  337. 

  338.     ti = av_gettime();

  339.     it1 = 0;

  340.     do {

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

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

  343.             dct->func(block);

  344.         }

  345.         emms_c();

  346.         it1 += NB_ITS_SPEED;

  347.         ti1 = av_gettime() - ti;

  348.     } while (ti1 < 1000000);

  349. 

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

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

  352. 

  353.     return 0;

  354. }

  355. 

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

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

  358. 

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

  360. {

  361.     static int init;

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

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

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

  365.     double s, sum, v;

  366.     int i, j, k;

  367. 

  368.     if (!init) {

  369.         init = 1;

  370. 

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

  372.             sum = 0;

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

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

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

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

  377.             }

  378.         }

  379. 

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

  381.             sum = 0;

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

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

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

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

  386.             }

  387.         }

  388.     }

  389. 

  390.     /* butterfly */

  391.     s = 0.5 * sqrt(2.0);

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

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

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

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

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

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

  398.         }

  399.     }

  400. 

  401.     /* idct8 on lines */

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

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

  404.             sum = 0;

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

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

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

  408.         }

  409.     }

  410. 

  411.     /* idct4 */

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

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

  414.             /* top */

  415.             sum = 0;

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

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

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

  419. 

  420.             /* bottom */

  421.             sum = 0;

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

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

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

  425.         }

  426.     }

  427. 

  428.     /* clamp and store the result */

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

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

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

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

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

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

  435.         }

  436.     }

  437. }

  438. 

  439. static void idct248_error(const char *name,

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

  441.                                               int16_t *block),

  442.                           int speed)

  443. {

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

  445.     AVLFG prng;

  446. 

  447.     av_lfg_init(&prng, 1);

  448. 

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

  450.        important here) */

  451.     err_max = 0;

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

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

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

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

  456.         block1[0] += 1024;

  457. 

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

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

  460.         idct248_ref(img_dest1, 8, block);

  461. 

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

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

  464.         idct248_put(img_dest, 8, block);

  465. 

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

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

  468.             if (v == 255)

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

  470.             if (v > err_max)

  471.                 err_max = v;

  472.         }

  473. #if 0

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

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

  476.             int j;

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

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

  479.             }

  480.             printf("\n");

  481.         }

  482. 

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

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

  485.             int j;

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

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

  488.             }

  489.             printf("\n");

  490.         }

  491. #endif

  492.     }

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

  494. 

  495.     if (!speed)

  496.         return;

  497. 

  498.     ti = av_gettime();

  499.     it1 = 0;

  500.     do {

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

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

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

  504.             idct248_put(img_dest, 8, block);

  505.         }

  506.         emms_c();

  507.         it1 += NB_ITS_SPEED;

  508.         ti1 = av_gettime() - ti;

  509.     } while (ti1 < 1000000);

  510. 

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

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

  513. }

  514. 

  515. static void help(void)

  516. {

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

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

  519.            "            1 -> test with random sparse matrixes\n"

  520.            "            2 -> do 3. test from mpeg4 std\n"

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

  522.            "-i          test IDCT implementations\n"

  523.            "-4          test IDCT248 implementations\n"

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

  525. }

  526. 

  527. #if !HAVE_GETOPT

  528. #include "compat/getopt.c"

  529. #endif

  530. 

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

  532. {

  533.     int test_idct = 0, test_248_dct = 0;

  534.     int c, i;

  535.     int test = 1;

  536.     int speed = 0;

  537.     int err = 0;

  538.     int bits=8;

  539. 

  540.     cpu_flags = av_get_cpu_flags();

  541. 

  542.     ff_ref_dct_init();

  543.     idct_mmx_init();

  544. 

  545.     for (;;) {

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

  547.         if (c == -1)

  548.             break;

  549.         switch (c) {

  550.         case 'i':

  551.             test_idct = 1;

  552.             break;

  553.         case '4':

  554.             test_248_dct = 1;

  555.             break;

  556.         case 't':

  557.             speed = 1;

  558.             break;

  559.         default:

  560.         case 'h':

  561.             help();

  562.             return 0;

  563.         }

  564.     }

  565. 

  566.     if (optind < argc)

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

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

  569. 

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

  571. 

  572.     if (test_248_dct) {

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

  574.     } else {

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

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

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

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

  579.             }

  580.     }

  581. 

  582.     return err;

  583. }