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.
29807 Commits
32 Branches
444MB
Tree: 59eb12faff
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '59eb12faff'
${ noResults }
FFmpeg/libavcodec/dct-test.c

601 lines
17KB
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 <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. void odivx_idct_c(short *block);

  52. 

  53. // BFIN

  54. void ff_bfin_idct(DCTELEM *block);

  55. void ff_bfin_fdct(DCTELEM *block);

  56. 

  57. // ALTIVEC

  58. void fdct_altivec(DCTELEM *block);

  59. //void idct_altivec(DCTELEM *block);?? no routine

  60. 

  61. // ARM

  62. void ff_j_rev_dct_arm(DCTELEM *data);

  63. void ff_simple_idct_arm(DCTELEM *data);

  64. void ff_simple_idct_armv5te(DCTELEM *data);

  65. void ff_simple_idct_armv6(DCTELEM *data);

  66. void ff_simple_idct_neon(DCTELEM *data);

  67. 

  68. void ff_simple_idct_axp(DCTELEM *data);

  69. 

  70. struct algo {

  71.   const char *name;

  72.   enum { FDCT, IDCT } is_idct;

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

  74.   void (* ref)  (DCTELEM *block);

  75.   enum formattag { NO_PERM,MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM, SSE2_PERM, PARTTRANS_PERM } format;

  76.   int  mm_support;

  77. };

  78. 

  79. #ifndef FAAN_POSTSCALE

  80. #define FAAN_SCALE SCALE_PERM

  81. #else

  82. #define FAAN_SCALE NO_PERM

  83. #endif

  84. 

  85. static int cpu_flags;

  86. 

  87. struct algo algos[] = {

  88.   {"REF-DBL",         0, ff_ref_fdct,        ff_ref_fdct, NO_PERM},

  89.   {"FAAN",            0, ff_faandct,         ff_ref_fdct, FAAN_SCALE},

  90.   {"FAANI",           1, ff_faanidct,        ff_ref_idct, NO_PERM},

  91.   {"IJG-AAN-INT",     0, fdct_ifast,         ff_ref_fdct, SCALE_PERM},

  92.   {"IJG-LLM-INT",     0, ff_jpeg_fdct_islow, ff_ref_fdct, NO_PERM},

  93.   {"REF-DBL",         1, ff_ref_idct,        ff_ref_idct, NO_PERM},

  94.   {"INT",             1, j_rev_dct,          ff_ref_idct, MMX_PERM},

  95.   {"SIMPLE-C",        1, ff_simple_idct,     ff_ref_idct, NO_PERM},

  96. 

  97. #if HAVE_MMX

  98.   {"MMX",             0, ff_fdct_mmx,        ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX},

  99. #if HAVE_MMX2

  100.   {"MMX2",            0, ff_fdct_mmx2,       ff_ref_fdct, NO_PERM, AV_CPU_FLAG_MMX2},

  101.   {"SSE2",            0, ff_fdct_sse2,       ff_ref_fdct, NO_PERM, AV_CPU_FLAG_SSE2},

  102. #endif

  103. 

  104. #if CONFIG_GPL

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

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

  107. #endif

  108.   {"SIMPLE-MMX",      1, ff_simple_idct_mmx, ff_ref_idct, MMX_SIMPLE_PERM, AV_CPU_FLAG_MMX},

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

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

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

  112. #endif

  113. 

  114. #if HAVE_ALTIVEC

  115.   {"altivecfdct",     0, fdct_altivec,       ff_ref_fdct, NO_PERM, AV_CPU_FLAG_ALTIVEC},

  116. #endif

  117. 

  118. #if ARCH_BFIN

  119.   {"BFINfdct",        0, ff_bfin_fdct,       ff_ref_fdct, NO_PERM},

  120.   {"BFINidct",        1, ff_bfin_idct,       ff_ref_idct, NO_PERM},

  121. #endif

  122. 

  123. #if ARCH_ARM

  124.   {"SIMPLE-ARM",      1, ff_simple_idct_arm, ff_ref_idct, NO_PERM },

  125.   {"INT-ARM",         1, ff_j_rev_dct_arm,   ff_ref_idct, MMX_PERM },

  126. #if HAVE_ARMV5TE

  127.   {"SIMPLE-ARMV5TE",  1, ff_simple_idct_armv5te, ff_ref_idct, NO_PERM },

  128. #endif

  129. #if HAVE_ARMV6

  130.   {"SIMPLE-ARMV6",    1, ff_simple_idct_armv6, ff_ref_idct, MMX_PERM },

  131. #endif

  132. #if HAVE_NEON

  133.   {"SIMPLE-NEON",     1, ff_simple_idct_neon, ff_ref_idct, PARTTRANS_PERM },

  134. #endif

  135. #endif /* ARCH_ARM */

  136. 

  137. #if ARCH_ALPHA

  138.   {"SIMPLE-ALPHA",    1, ff_simple_idct_axp,  ff_ref_idct, NO_PERM },

  139. #endif

  140. 

  141.   { 0 }

  142. };

  143. 

  144. #define AANSCALE_BITS 12

  145. 

  146. uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];

  147. 

  148. static int64_t gettime(void)

  149. {

  150.     struct timeval tv;

  151.     gettimeofday(&tv,NULL);

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

  153. }

  154. 

  155. #define NB_ITS 20000

  156. #define NB_ITS_SPEED 50000

  157. 

  158. static short idct_mmx_perm[64];

  159. 

  160. static short idct_simple_mmx_perm[64]={

  161.         0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  162.         0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  163.         0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  164.         0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  165.         0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  166.         0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  167.         0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  168.         0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  169. };

  170. 

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

  172. 

  173. static void idct_mmx_init(void)

  174. {

  175.     int i;

  176. 

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

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

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

  180. //        idct_simple_mmx_perm[i] = simple_block_permute_op(i);

  181.     }

  182. }

  183. 

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

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

  186. DECLARE_ALIGNED(8, static DCTELEM, block_org)[64];

  187. 

  188. static inline void mmx_emms(void)

  189. {

  190. #if HAVE_MMX

  191.     if (cpu_flags & AV_CPU_FLAG_MMX)

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

  193. #endif

  194. }

  195. 

  196. static void dct_error(const char *name, int is_idct,

  197.                void (*fdct_func)(DCTELEM *block),

  198.                void (*fdct_ref)(DCTELEM *block), int form, int test, const int bits)

  199. {

  200.     int it, i, scale;

  201.     int err_inf, v;

  202.     int64_t err2, ti, ti1, it1;

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

  204.     int maxout=0;

  205.     int blockSumErrMax=0, blockSumErr;

  206.     AVLFG prng;

  207.     const int vals=1<<bits;

  208. 

  209.     av_lfg_init(&prng, 1);

  210. 

  211.     err_inf = 0;

  212.     err2 = 0;

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

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

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

  216.             block1[i] = 0;

  217.         switch(test){

  218.         case 0:

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

  220.                 block1[i] = (av_lfg_get(&prng) % (2*vals)) -vals;

  221.             if (is_idct){

  222.                 ff_ref_fdct(block1);

  223. 

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

  225.                     block1[i]>>=3;

  226.             }

  227.         break;

  228.         case 1:{

  229.             int num = av_lfg_get(&prng) % 10 + 1;

  230.             for(i=0;i<num;i++)

  231.                 block1[av_lfg_get(&prng) % 64] = av_lfg_get(&prng) % (2*vals) -vals;

  232.         }break;

  233.         case 2:

  234.             block1[0] = av_lfg_get(&prng) % (16*vals) - (8*vals);

  235.             block1[63]= (block1[0]&1)^1;

  236.         break;

  237.         }

  238. 

  239. #if 0 // simulate mismatch control

  240. { int sum=0;

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

  242.            sum+=block1[i];

  243. 

  244.         if((sum&1)==0) block1[63]^=1;

  245. }

  246. #endif

  247. 

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

  249.             block_org[i]= block1[i];

  250. 

  251.         if (form == MMX_PERM) {

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

  253.                 block[idct_mmx_perm[i]] = block1[i];

  254.             } else if (form == MMX_SIMPLE_PERM) {

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

  256.                 block[idct_simple_mmx_perm[i]] = block1[i];

  257. 

  258.         } else if (form == SSE2_PERM) {

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

  260.                 block[(i&0x38) | idct_sse2_row_perm[i&7]] = block1[i];

  261.         } else if (form == PARTTRANS_PERM) {

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

  263.                 block[(i&0x24) | ((i&3)<<3) | ((i>>3)&3)] = block1[i];

  264.         } else {

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

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

  267.         }

  268. #if 0 // simulate mismatch control for tested IDCT but not the ref

  269. { int sum=0;

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

  271.            sum+=block[i];

  272. 

  273.         if((sum&1)==0) block[63]^=1;

  274. }

  275. #endif

  276. 

  277.         fdct_func(block);

  278.         mmx_emms();

  279. 

  280.         if (form == SCALE_PERM) {

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

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

  283.                 block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS;

  284.             }

  285.         }

  286. 

  287.         fdct_ref(block1);

  288. 

  289.         blockSumErr=0;

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

  291.             v = abs(block[i] - block1[i]);

  292.             if (v > err_inf)

  293.                 err_inf = v;

  294.             err2 += v * v;

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

  296.             blockSumErr += v;

  297.             if( abs(block[i])>maxout) maxout=abs(block[i]);

  298.         }

  299.         if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr;

  300. #if 0 // print different matrix pairs

  301.         if(blockSumErr){

  302.             printf("\n");

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

  304.                 if((i&7)==0) printf("\n");

  305.                 printf("%4d ", block_org[i]);

  306.             }

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

  308.                 if((i&7)==0) printf("\n");

  309.                 printf("%4d ", block[i] - block1[i]);

  310.             }

  311.         }

  312. #endif

  313.     }

  314.     for(i=0; i<64; i++) sysErrMax= FFMAX(sysErrMax, FFABS(sysErr[i]));

  315. 

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

  317.         if(i%8==0) printf("\n");

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

  319.     }

  320.     printf("\n");

  321. 

  322.     printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",

  323.            is_idct ? "IDCT" : "DCT",

  324.            name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax);

  325. 

  326.     /* speed test */

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

  328.         block1[i] = 0;

  329.     switch(test){

  330.     case 0:

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

  332.             block1[i] = av_lfg_get(&prng) % (2*vals) -vals;

  333.         if (is_idct){

  334.             ff_ref_fdct(block1);

  335. 

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

  337.                 block1[i]>>=3;

  338.         }

  339.     break;

  340.     case 1:{

  341.     case 2:

  342.         block1[0] = av_lfg_get(&prng) % (2*vals) -vals;

  343.         block1[1] = av_lfg_get(&prng) % (2*vals) -vals;

  344.         block1[2] = av_lfg_get(&prng) % (2*vals) -vals;

  345.         block1[3] = av_lfg_get(&prng) % (2*vals) -vals;

  346.     }break;

  347.     }

  348. 

  349.     if (form == MMX_PERM) {

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

  351.             block[idct_mmx_perm[i]] = block1[i];

  352.     } else if(form == MMX_SIMPLE_PERM) {

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

  354.             block[idct_simple_mmx_perm[i]] = block1[i];

  355.     } else {

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

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

  358.     }

  359. 

  360.     ti = gettime();

  361.     it1 = 0;

  362.     do {

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

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

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

  366. //            memcpy(block, block1, sizeof(DCTELEM) * 64);

  367. // do not memcpy especially not fastmemcpy because it does movntq !!!

  368.             fdct_func(block);

  369.         }

  370.         it1 += NB_ITS_SPEED;

  371.         ti1 = gettime() - ti;

  372.     } while (ti1 < 1000000);

  373.     mmx_emms();

  374. 

  375.     printf("%s %s: %0.1f kdct/s\n",

  376.            is_idct ? "IDCT" : "DCT",

  377.            name, (double)it1 * 1000.0 / (double)ti1);

  378. }

  379. 

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

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

  382. 

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

  384. {

  385.     static int init;

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

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

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

  389.     double s, sum, v;

  390.     int i, j, k;

  391. 

  392.     if (!init) {

  393.         init = 1;

  394. 

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

  396.             sum = 0;

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

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

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

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

  401.             }

  402.         }

  403. 

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

  405.             sum = 0;

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

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

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

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

  410.             }

  411.         }

  412.     }

  413. 

  414.     /* butterfly */

  415.     s = 0.5 * sqrt(2.0);

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

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

  418.             block1[8*(2*i)+j] = (block[8*(2*i)+j] + block[8*(2*i+1)+j]) * s;

  419.             block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s;

  420.         }

  421.     }

  422. 

  423.     /* idct8 on lines */

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

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

  426.             sum = 0;

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

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

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

  430.         }

  431.     }

  432. 

  433.     /* idct4 */

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

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

  436.             /* top */

  437.             sum = 0;

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

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

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

  441. 

  442.             /* bottom */

  443.             sum = 0;

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

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

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

  447.         }

  448.     }

  449. 

  450.     /* clamp and store the result */

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

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

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

  454.             if (v < 0)

  455.                 v = 0;

  456.             else if (v > 255)

  457.                 v = 255;

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

  459.         }

  460.     }

  461. }

  462. 

  463. static void idct248_error(const char *name,

  464.                     void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block))

  465. {

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

  467. 

  468.     AVLFG prng;

  469. 

  470.     av_lfg_init(&prng, 1);

  471. 

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

  473.        important here) */

  474.     err_max = 0;

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

  476. 

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

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

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

  480.         block1[0] += 1024;

  481. 

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

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

  484.         idct248_ref(img_dest1, 8, block);

  485. 

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

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

  488.         idct248_put(img_dest, 8, block);

  489. 

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

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

  492.             if (v == 255)

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

  494.             if (v > err_max)

  495.                 err_max = v;

  496.         }

  497. #if 0

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

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

  500.             int j;

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

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

  503.             }

  504.             printf("\n");

  505.         }

  506. 

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

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

  509.             int j;

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

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

  512.             }

  513.             printf("\n");

  514.         }

  515. #endif

  516.     }

  517.     printf("%s %s: err_inf=%d\n",

  518.            1 ? "IDCT248" : "DCT248",

  519.            name, err_max);

  520. 

  521.     ti = gettime();

  522.     it1 = 0;

  523.     do {

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

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

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

  527. //            memcpy(block, block1, sizeof(DCTELEM) * 64);

  528. // do not memcpy especially not fastmemcpy because it does movntq !!!

  529.             idct248_put(img_dest, 8, block);

  530.         }

  531.         it1 += NB_ITS_SPEED;

  532.         ti1 = gettime() - ti;

  533.     } while (ti1 < 1000000);

  534.     mmx_emms();

  535. 

  536.     printf("%s %s: %0.1f kdct/s\n",

  537.            1 ? "IDCT248" : "DCT248",

  538.            name, (double)it1 * 1000.0 / (double)ti1);

  539. }

  540. 

  541. static void help(void)

  542. {

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

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

  545.            "            1 -> test with random sparse matrixes\n"

  546.            "            2 -> do 3. test from mpeg4 std\n"

  547.            "-i          test IDCT implementations\n"

  548.            "-4          test IDCT248 implementations\n");

  549. }

  550. 

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

  552. {

  553.     int test_idct = 0, test_248_dct = 0;

  554.     int c,i;

  555.     int test=1;

  556.     int bits=8;

  557.     cpu_flags = av_get_cpu_flags();

  558. 

  559.     ff_ref_dct_init();

  560.     idct_mmx_init();

  561. 

  562.     for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;

  563.     for(i=0;i<MAX_NEG_CROP;i++) {

  564.         cropTbl[i] = 0;

  565.         cropTbl[i + MAX_NEG_CROP + 256] = 255;

  566.     }

  567. 

  568.     for(;;) {

  569.         c = getopt(argc, argv, "ih4");

  570.         if (c == -1)

  571.             break;

  572.         switch(c) {

  573.         case 'i':

  574.             test_idct = 1;

  575.             break;

  576.         case '4':

  577.             test_248_dct = 1;

  578.             break;

  579.         default :

  580.         case 'h':

  581.             help();

  582.             return 0;

  583.         }

  584.     }

  585. 

  586.     if(optind <argc) test= atoi(argv[optind]);

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

  588. 

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

  590. 

  591.     if (test_248_dct) {

  592.         idct248_error("SIMPLE-C", ff_simple_idct248_put);

  593.     } else {

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

  595.         if (algos[i].is_idct == test_idct && !(~cpu_flags & algos[i].mm_support)) {

  596.           dct_error (algos[i].name, algos[i].is_idct, algos[i].func, algos[i].ref, algos[i].format, test, bits);

  597.         }

  598.     }

  599.     return 0;

  600. }