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.
64266 Commits
32 Branches
313MB
Tree: 09a7a4704e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '09a7a4704e'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1492 lines
49KB
Raw Blame History 

  1. /*

  2.  * DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

  8.  * FFmpeg is free software; you can redistribute it and/or

  9.  * modify it under the terms of the GNU Lesser General Public

  10.  * License as published by the Free Software Foundation; either

  11.  * version 2.1 of the License, or (at your option) any later version.

  12.  *

  13.  * FFmpeg is distributed in the hope that it will be useful,

  14.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  15.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  16.  * Lesser General Public License for more details.

  17.  *

  18.  * You should have received a copy of the GNU Lesser General Public

  19.  * License along with FFmpeg; if not, write to the Free Software

  20.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  21.  */

  22. 

  23. /**

  24.  * @file

  25.  * DSP utils

  26.  */

  27. 

  28. #include "libavutil/attributes.h"

  29. #include "libavutil/imgutils.h"

  30. #include "libavutil/internal.h"

  31. #include "avcodec.h"

  32. #include "copy_block.h"

  33. #include "dct.h"

  34. #include "dsputil.h"

  35. #include "simple_idct.h"

  36. #include "faandct.h"

  37. #include "faanidct.h"

  38. #include "imgconvert.h"

  39. #include "mathops.h"

  40. #include "mpegvideo.h"

  41. #include "config.h"

  42. 

  43. uint32_t ff_square_tab[512] = { 0, };

  44. 

  45. #define BIT_DEPTH 16

  46. #include "dsputilenc_template.c"

  47. #undef BIT_DEPTH

  48. 

  49. #define BIT_DEPTH 8

  50. #include "dsputilenc_template.c"

  51. 

  52. av_cold void ff_init_scantable(uint8_t *permutation, ScanTable *st,

  53.                                const uint8_t *src_scantable)

  54. {

  55.     int i, end;

  56. 

  57.     st->scantable = src_scantable;

  58. 

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

  60.         int j = src_scantable[i];

  61.         st->permutated[i] = permutation[j];

  62.     }

  63. 

  64.     end = -1;

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

  66.         int j = st->permutated[i];

  67.         if (j > end)

  68.             end = j;

  69.         st->raster_end[i] = end;

  70.     }

  71. }

  72. 

  73. av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,

  74.                                            int idct_permutation_type)

  75. {

  76.     int i;

  77. 

  78.     if (ARCH_X86)

  79.         if (ff_init_scantable_permutation_x86(idct_permutation,

  80.                                               idct_permutation_type))

  81.             return;

  82. 

  83.     switch (idct_permutation_type) {

  84.     case FF_NO_IDCT_PERM:

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

  86.             idct_permutation[i] = i;

  87.         break;

  88.     case FF_LIBMPEG2_IDCT_PERM:

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

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

  91.         break;

  92.     case FF_TRANSPOSE_IDCT_PERM:

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

  94.             idct_permutation[i] = ((i & 7) << 3) | (i >> 3);

  95.         break;

  96.     case FF_PARTTRANS_IDCT_PERM:

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

  98.             idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3);

  99.         break;

  100.     default:

  101.         av_log(NULL, AV_LOG_ERROR,

  102.                "Internal error, IDCT permutation not set\n");

  103.     }

  104. }

  105. 

  106. static int pix_sum_c(uint8_t *pix, int line_size)

  107. {

  108.     int s = 0, i, j;

  109. 

  110.     for (i = 0; i < 16; i++) {

  111.         for (j = 0; j < 16; j += 8) {

  112.             s   += pix[0];

  113.             s   += pix[1];

  114.             s   += pix[2];

  115.             s   += pix[3];

  116.             s   += pix[4];

  117.             s   += pix[5];

  118.             s   += pix[6];

  119.             s   += pix[7];

  120.             pix += 8;

  121.         }

  122.         pix += line_size - 16;

  123.     }

  124.     return s;

  125. }

  126. 

  127. static int pix_norm1_c(uint8_t *pix, int line_size)

  128. {

  129.     int s = 0, i, j;

  130.     uint32_t *sq = ff_square_tab + 256;

  131. 

  132.     for (i = 0; i < 16; i++) {

  133.         for (j = 0; j < 16; j += 8) {

  134. #if 0

  135.             s += sq[pix[0]];

  136.             s += sq[pix[1]];

  137.             s += sq[pix[2]];

  138.             s += sq[pix[3]];

  139.             s += sq[pix[4]];

  140.             s += sq[pix[5]];

  141.             s += sq[pix[6]];

  142.             s += sq[pix[7]];

  143. #else

  144. #if HAVE_FAST_64BIT

  145.             register uint64_t x = *(uint64_t *) pix;

  146.             s += sq[x         & 0xff];

  147.             s += sq[(x >>  8) & 0xff];

  148.             s += sq[(x >> 16) & 0xff];

  149.             s += sq[(x >> 24) & 0xff];

  150.             s += sq[(x >> 32) & 0xff];

  151.             s += sq[(x >> 40) & 0xff];

  152.             s += sq[(x >> 48) & 0xff];

  153.             s += sq[(x >> 56) & 0xff];

  154. #else

  155.             register uint32_t x = *(uint32_t *) pix;

  156.             s += sq[x         & 0xff];

  157.             s += sq[(x >>  8) & 0xff];

  158.             s += sq[(x >> 16) & 0xff];

  159.             s += sq[(x >> 24) & 0xff];

  160.             x  = *(uint32_t *) (pix + 4);

  161.             s += sq[x         & 0xff];

  162.             s += sq[(x >>  8) & 0xff];

  163.             s += sq[(x >> 16) & 0xff];

  164.             s += sq[(x >> 24) & 0xff];

  165. #endif

  166. #endif

  167.             pix += 8;

  168.         }

  169.         pix += line_size - 16;

  170.     }

  171.     return s;

  172. }

  173. 

  174. static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  175.                   int line_size, int h)

  176. {

  177.     int s = 0, i;

  178.     uint32_t *sq = ff_square_tab + 256;

  179. 

  180.     for (i = 0; i < h; i++) {

  181.         s    += sq[pix1[0] - pix2[0]];

  182.         s    += sq[pix1[1] - pix2[1]];

  183.         s    += sq[pix1[2] - pix2[2]];

  184.         s    += sq[pix1[3] - pix2[3]];

  185.         pix1 += line_size;

  186.         pix2 += line_size;

  187.     }

  188.     return s;

  189. }

  190. 

  191. static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  192.                   int line_size, int h)

  193. {

  194.     int s = 0, i;

  195.     uint32_t *sq = ff_square_tab + 256;

  196. 

  197.     for (i = 0; i < h; i++) {

  198.         s    += sq[pix1[0] - pix2[0]];

  199.         s    += sq[pix1[1] - pix2[1]];

  200.         s    += sq[pix1[2] - pix2[2]];

  201.         s    += sq[pix1[3] - pix2[3]];

  202.         s    += sq[pix1[4] - pix2[4]];

  203.         s    += sq[pix1[5] - pix2[5]];

  204.         s    += sq[pix1[6] - pix2[6]];

  205.         s    += sq[pix1[7] - pix2[7]];

  206.         pix1 += line_size;

  207.         pix2 += line_size;

  208.     }

  209.     return s;

  210. }

  211. 

  212. static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  213.                    int line_size, int h)

  214. {

  215.     int s = 0, i;

  216.     uint32_t *sq = ff_square_tab + 256;

  217. 

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

  219.         s += sq[pix1[0]  - pix2[0]];

  220.         s += sq[pix1[1]  - pix2[1]];

  221.         s += sq[pix1[2]  - pix2[2]];

  222.         s += sq[pix1[3]  - pix2[3]];

  223.         s += sq[pix1[4]  - pix2[4]];

  224.         s += sq[pix1[5]  - pix2[5]];

  225.         s += sq[pix1[6]  - pix2[6]];

  226.         s += sq[pix1[7]  - pix2[7]];

  227.         s += sq[pix1[8]  - pix2[8]];

  228.         s += sq[pix1[9]  - pix2[9]];

  229.         s += sq[pix1[10] - pix2[10]];

  230.         s += sq[pix1[11] - pix2[11]];

  231.         s += sq[pix1[12] - pix2[12]];

  232.         s += sq[pix1[13] - pix2[13]];

  233.         s += sq[pix1[14] - pix2[14]];

  234.         s += sq[pix1[15] - pix2[15]];

  235. 

  236.         pix1 += line_size;

  237.         pix2 += line_size;

  238.     }

  239.     return s;

  240. }

  241. 

  242. static void diff_pixels_c(int16_t *av_restrict block, const uint8_t *s1,

  243.                           const uint8_t *s2, int stride)

  244. {

  245.     int i;

  246. 

  247.     /* read the pixels */

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

  249.         block[0] = s1[0] - s2[0];

  250.         block[1] = s1[1] - s2[1];

  251.         block[2] = s1[2] - s2[2];

  252.         block[3] = s1[3] - s2[3];

  253.         block[4] = s1[4] - s2[4];

  254.         block[5] = s1[5] - s2[5];

  255.         block[6] = s1[6] - s2[6];

  256.         block[7] = s1[7] - s2[7];

  257.         s1      += stride;

  258.         s2      += stride;

  259.         block   += 8;

  260.     }

  261. }

  262. 

  263. static void put_pixels_clamped_c(const int16_t *block, uint8_t *av_restrict pixels,

  264.                                  int line_size)

  265. {

  266.     int i;

  267. 

  268.     /* read the pixels */

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

  270.         pixels[0] = av_clip_uint8(block[0]);

  271.         pixels[1] = av_clip_uint8(block[1]);

  272.         pixels[2] = av_clip_uint8(block[2]);

  273.         pixels[3] = av_clip_uint8(block[3]);

  274.         pixels[4] = av_clip_uint8(block[4]);

  275.         pixels[5] = av_clip_uint8(block[5]);

  276.         pixels[6] = av_clip_uint8(block[6]);

  277.         pixels[7] = av_clip_uint8(block[7]);

  278. 

  279.         pixels += line_size;

  280.         block  += 8;

  281.     }

  282. }

  283. 

  284. static void put_pixels_clamped4_c(const int16_t *block, uint8_t *av_restrict pixels,

  285.                                  int line_size)

  286. {

  287.     int i;

  288. 

  289.     /* read the pixels */

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

  291.         pixels[0] = av_clip_uint8(block[0]);

  292.         pixels[1] = av_clip_uint8(block[1]);

  293.         pixels[2] = av_clip_uint8(block[2]);

  294.         pixels[3] = av_clip_uint8(block[3]);

  295. 

  296.         pixels += line_size;

  297.         block += 8;

  298.     }

  299. }

  300. 

  301. static void put_pixels_clamped2_c(const int16_t *block, uint8_t *av_restrict pixels,

  302.                                  int line_size)

  303. {

  304.     int i;

  305. 

  306.     /* read the pixels */

  307.     for(i=0;i<2;i++) {

  308.         pixels[0] = av_clip_uint8(block[0]);

  309.         pixels[1] = av_clip_uint8(block[1]);

  310. 

  311.         pixels += line_size;

  312.         block += 8;

  313.     }

  314. }

  315. 

  316. static void put_signed_pixels_clamped_c(const int16_t *block,

  317.                                         uint8_t *av_restrict pixels,

  318.                                         int line_size)

  319. {

  320.     int i, j;

  321. 

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

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

  324.             if (*block < -128)

  325.                 *pixels = 0;

  326.             else if (*block > 127)

  327.                 *pixels = 255;

  328.             else

  329.                 *pixels = (uint8_t) (*block + 128);

  330.             block++;

  331.             pixels++;

  332.         }

  333.         pixels += (line_size - 8);

  334.     }

  335. }

  336. 

  337. static void add_pixels_clamped_c(const int16_t *block, uint8_t *av_restrict pixels,

  338.                                  int line_size)

  339. {

  340.     int i;

  341. 

  342.     /* read the pixels */

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

  344.         pixels[0] = av_clip_uint8(pixels[0] + block[0]);

  345.         pixels[1] = av_clip_uint8(pixels[1] + block[1]);

  346.         pixels[2] = av_clip_uint8(pixels[2] + block[2]);

  347.         pixels[3] = av_clip_uint8(pixels[3] + block[3]);

  348.         pixels[4] = av_clip_uint8(pixels[4] + block[4]);

  349.         pixels[5] = av_clip_uint8(pixels[5] + block[5]);

  350.         pixels[6] = av_clip_uint8(pixels[6] + block[6]);

  351.         pixels[7] = av_clip_uint8(pixels[7] + block[7]);

  352.         pixels   += line_size;

  353.         block    += 8;

  354.     }

  355. }

  356. 

  357. static void add_pixels_clamped4_c(const int16_t *block, uint8_t *av_restrict pixels,

  358.                           int line_size)

  359. {

  360.     int i;

  361. 

  362.     /* read the pixels */

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

  364.         pixels[0] = av_clip_uint8(pixels[0] + block[0]);

  365.         pixels[1] = av_clip_uint8(pixels[1] + block[1]);

  366.         pixels[2] = av_clip_uint8(pixels[2] + block[2]);

  367.         pixels[3] = av_clip_uint8(pixels[3] + block[3]);

  368.         pixels += line_size;

  369.         block += 8;

  370.     }

  371. }

  372. 

  373. static void add_pixels_clamped2_c(const int16_t *block, uint8_t *av_restrict pixels,

  374.                           int line_size)

  375. {

  376.     int i;

  377. 

  378.     /* read the pixels */

  379.     for(i=0;i<2;i++) {

  380.         pixels[0] = av_clip_uint8(pixels[0] + block[0]);

  381.         pixels[1] = av_clip_uint8(pixels[1] + block[1]);

  382.         pixels += line_size;

  383.         block += 8;

  384.     }

  385. }

  386. 

  387. static int sum_abs_dctelem_c(int16_t *block)

  388. {

  389.     int sum = 0, i;

  390. 

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

  392.         sum += FFABS(block[i]);

  393.     return sum;

  394. }

  395. 

  396. #define avg2(a, b) ((a + b + 1) >> 1)

  397. #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)

  398. 

  399. static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  400.                               int line_size, int h)

  401. {

  402.     int s = 0, i;

  403. 

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

  405.         s    += abs(pix1[0]  - pix2[0]);

  406.         s    += abs(pix1[1]  - pix2[1]);

  407.         s    += abs(pix1[2]  - pix2[2]);

  408.         s    += abs(pix1[3]  - pix2[3]);

  409.         s    += abs(pix1[4]  - pix2[4]);

  410.         s    += abs(pix1[5]  - pix2[5]);

  411.         s    += abs(pix1[6]  - pix2[6]);

  412.         s    += abs(pix1[7]  - pix2[7]);

  413.         s    += abs(pix1[8]  - pix2[8]);

  414.         s    += abs(pix1[9]  - pix2[9]);

  415.         s    += abs(pix1[10] - pix2[10]);

  416.         s    += abs(pix1[11] - pix2[11]);

  417.         s    += abs(pix1[12] - pix2[12]);

  418.         s    += abs(pix1[13] - pix2[13]);

  419.         s    += abs(pix1[14] - pix2[14]);

  420.         s    += abs(pix1[15] - pix2[15]);

  421.         pix1 += line_size;

  422.         pix2 += line_size;

  423.     }

  424.     return s;

  425. }

  426. 

  427. static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  428.                           int line_size, int h)

  429. {

  430.     int s = 0, i;

  431. 

  432.     for (i = 0; i < h; i++) {

  433.         s    += abs(pix1[0]  - avg2(pix2[0],  pix2[1]));

  434.         s    += abs(pix1[1]  - avg2(pix2[1],  pix2[2]));

  435.         s    += abs(pix1[2]  - avg2(pix2[2],  pix2[3]));

  436.         s    += abs(pix1[3]  - avg2(pix2[3],  pix2[4]));

  437.         s    += abs(pix1[4]  - avg2(pix2[4],  pix2[5]));

  438.         s    += abs(pix1[5]  - avg2(pix2[5],  pix2[6]));

  439.         s    += abs(pix1[6]  - avg2(pix2[6],  pix2[7]));

  440.         s    += abs(pix1[7]  - avg2(pix2[7],  pix2[8]));

  441.         s    += abs(pix1[8]  - avg2(pix2[8],  pix2[9]));

  442.         s    += abs(pix1[9]  - avg2(pix2[9],  pix2[10]));

  443.         s    += abs(pix1[10] - avg2(pix2[10], pix2[11]));

  444.         s    += abs(pix1[11] - avg2(pix2[11], pix2[12]));

  445.         s    += abs(pix1[12] - avg2(pix2[12], pix2[13]));

  446.         s    += abs(pix1[13] - avg2(pix2[13], pix2[14]));

  447.         s    += abs(pix1[14] - avg2(pix2[14], pix2[15]));

  448.         s    += abs(pix1[15] - avg2(pix2[15], pix2[16]));

  449.         pix1 += line_size;

  450.         pix2 += line_size;

  451.     }

  452.     return s;

  453. }

  454. 

  455. static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  456.                           int line_size, int h)

  457. {

  458.     int s = 0, i;

  459.     uint8_t *pix3 = pix2 + line_size;

  460. 

  461.     for (i = 0; i < h; i++) {

  462.         s    += abs(pix1[0]  - avg2(pix2[0],  pix3[0]));

  463.         s    += abs(pix1[1]  - avg2(pix2[1],  pix3[1]));

  464.         s    += abs(pix1[2]  - avg2(pix2[2],  pix3[2]));

  465.         s    += abs(pix1[3]  - avg2(pix2[3],  pix3[3]));

  466.         s    += abs(pix1[4]  - avg2(pix2[4],  pix3[4]));

  467.         s    += abs(pix1[5]  - avg2(pix2[5],  pix3[5]));

  468.         s    += abs(pix1[6]  - avg2(pix2[6],  pix3[6]));

  469.         s    += abs(pix1[7]  - avg2(pix2[7],  pix3[7]));

  470.         s    += abs(pix1[8]  - avg2(pix2[8],  pix3[8]));

  471.         s    += abs(pix1[9]  - avg2(pix2[9],  pix3[9]));

  472.         s    += abs(pix1[10] - avg2(pix2[10], pix3[10]));

  473.         s    += abs(pix1[11] - avg2(pix2[11], pix3[11]));

  474.         s    += abs(pix1[12] - avg2(pix2[12], pix3[12]));

  475.         s    += abs(pix1[13] - avg2(pix2[13], pix3[13]));

  476.         s    += abs(pix1[14] - avg2(pix2[14], pix3[14]));

  477.         s    += abs(pix1[15] - avg2(pix2[15], pix3[15]));

  478.         pix1 += line_size;

  479.         pix2 += line_size;

  480.         pix3 += line_size;

  481.     }

  482.     return s;

  483. }

  484. 

  485. static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  486.                            int line_size, int h)

  487. {

  488.     int s = 0, i;

  489.     uint8_t *pix3 = pix2 + line_size;

  490. 

  491.     for (i = 0; i < h; i++) {

  492.         s    += abs(pix1[0]  - avg4(pix2[0],  pix2[1],  pix3[0],  pix3[1]));

  493.         s    += abs(pix1[1]  - avg4(pix2[1],  pix2[2],  pix3[1],  pix3[2]));

  494.         s    += abs(pix1[2]  - avg4(pix2[2],  pix2[3],  pix3[2],  pix3[3]));

  495.         s    += abs(pix1[3]  - avg4(pix2[3],  pix2[4],  pix3[3],  pix3[4]));

  496.         s    += abs(pix1[4]  - avg4(pix2[4],  pix2[5],  pix3[4],  pix3[5]));

  497.         s    += abs(pix1[5]  - avg4(pix2[5],  pix2[6],  pix3[5],  pix3[6]));

  498.         s    += abs(pix1[6]  - avg4(pix2[6],  pix2[7],  pix3[6],  pix3[7]));

  499.         s    += abs(pix1[7]  - avg4(pix2[7],  pix2[8],  pix3[7],  pix3[8]));

  500.         s    += abs(pix1[8]  - avg4(pix2[8],  pix2[9],  pix3[8],  pix3[9]));

  501.         s    += abs(pix1[9]  - avg4(pix2[9],  pix2[10], pix3[9],  pix3[10]));

  502.         s    += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));

  503.         s    += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));

  504.         s    += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));

  505.         s    += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));

  506.         s    += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));

  507.         s    += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));

  508.         pix1 += line_size;

  509.         pix2 += line_size;

  510.         pix3 += line_size;

  511.     }

  512.     return s;

  513. }

  514. 

  515. static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  516.                              int line_size, int h)

  517. {

  518.     int s = 0, i;

  519. 

  520.     for (i = 0; i < h; i++) {

  521.         s    += abs(pix1[0] - pix2[0]);

  522.         s    += abs(pix1[1] - pix2[1]);

  523.         s    += abs(pix1[2] - pix2[2]);

  524.         s    += abs(pix1[3] - pix2[3]);

  525.         s    += abs(pix1[4] - pix2[4]);

  526.         s    += abs(pix1[5] - pix2[5]);

  527.         s    += abs(pix1[6] - pix2[6]);

  528.         s    += abs(pix1[7] - pix2[7]);

  529.         pix1 += line_size;

  530.         pix2 += line_size;

  531.     }

  532.     return s;

  533. }

  534. 

  535. static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  536.                          int line_size, int h)

  537. {

  538.     int s = 0, i;

  539. 

  540.     for (i = 0; i < h; i++) {

  541.         s    += abs(pix1[0] - avg2(pix2[0], pix2[1]));

  542.         s    += abs(pix1[1] - avg2(pix2[1], pix2[2]));

  543.         s    += abs(pix1[2] - avg2(pix2[2], pix2[3]));

  544.         s    += abs(pix1[3] - avg2(pix2[3], pix2[4]));

  545.         s    += abs(pix1[4] - avg2(pix2[4], pix2[5]));

  546.         s    += abs(pix1[5] - avg2(pix2[5], pix2[6]));

  547.         s    += abs(pix1[6] - avg2(pix2[6], pix2[7]));

  548.         s    += abs(pix1[7] - avg2(pix2[7], pix2[8]));

  549.         pix1 += line_size;

  550.         pix2 += line_size;

  551.     }

  552.     return s;

  553. }

  554. 

  555. static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  556.                          int line_size, int h)

  557. {

  558.     int s = 0, i;

  559.     uint8_t *pix3 = pix2 + line_size;

  560. 

  561.     for (i = 0; i < h; i++) {

  562.         s    += abs(pix1[0] - avg2(pix2[0], pix3[0]));

  563.         s    += abs(pix1[1] - avg2(pix2[1], pix3[1]));

  564.         s    += abs(pix1[2] - avg2(pix2[2], pix3[2]));

  565.         s    += abs(pix1[3] - avg2(pix2[3], pix3[3]));

  566.         s    += abs(pix1[4] - avg2(pix2[4], pix3[4]));

  567.         s    += abs(pix1[5] - avg2(pix2[5], pix3[5]));

  568.         s    += abs(pix1[6] - avg2(pix2[6], pix3[6]));

  569.         s    += abs(pix1[7] - avg2(pix2[7], pix3[7]));

  570.         pix1 += line_size;

  571.         pix2 += line_size;

  572.         pix3 += line_size;

  573.     }

  574.     return s;

  575. }

  576. 

  577. static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  578.                           int line_size, int h)

  579. {

  580.     int s = 0, i;

  581.     uint8_t *pix3 = pix2 + line_size;

  582. 

  583.     for (i = 0; i < h; i++) {

  584.         s    += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));

  585.         s    += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));

  586.         s    += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));

  587.         s    += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));

  588.         s    += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));

  589.         s    += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));

  590.         s    += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));

  591.         s    += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));

  592.         pix1 += line_size;

  593.         pix2 += line_size;

  594.         pix3 += line_size;

  595.     }

  596.     return s;

  597. }

  598. 

  599. static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  600. {

  601.     int score1 = 0, score2 = 0, x, y;

  602. 

  603.     for (y = 0; y < h; y++) {

  604.         for (x = 0; x < 16; x++)

  605.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  606.         if (y + 1 < h) {

  607.             for (x = 0; x < 15; x++)

  608.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  609.                                 s1[x + 1] + s1[x + stride + 1]) -

  610.                           FFABS(s2[x]     - s2[x + stride] -

  611.                                 s2[x + 1] + s2[x + stride + 1]);

  612.         }

  613.         s1 += stride;

  614.         s2 += stride;

  615.     }

  616. 

  617.     if (c)

  618.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  619.     else

  620.         return score1 + FFABS(score2) * 8;

  621. }

  622. 

  623. static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  624. {

  625.     int score1 = 0, score2 = 0, x, y;

  626. 

  627.     for (y = 0; y < h; y++) {

  628.         for (x = 0; x < 8; x++)

  629.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  630.         if (y + 1 < h) {

  631.             for (x = 0; x < 7; x++)

  632.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  633.                                 s1[x + 1] + s1[x + stride + 1]) -

  634.                           FFABS(s2[x]     - s2[x + stride] -

  635.                                 s2[x + 1] + s2[x + stride + 1]);

  636.         }

  637.         s1 += stride;

  638.         s2 += stride;

  639.     }

  640. 

  641.     if (c)

  642.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  643.     else

  644.         return score1 + FFABS(score2) * 8;

  645. }

  646. 

  647. static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],

  648.                           int16_t basis[64], int scale)

  649. {

  650.     int i;

  651.     unsigned int sum = 0;

  652. 

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

  654.         int b = rem[i] + ((basis[i] * scale +

  655.                            (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  656.                           (BASIS_SHIFT - RECON_SHIFT));

  657.         int w = weight[i];

  658.         b >>= RECON_SHIFT;

  659.         av_assert2(-512 < b && b < 512);

  660. 

  661.         sum += (w * b) * (w * b) >> 4;

  662.     }

  663.     return sum >> 2;

  664. }

  665. 

  666. static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)

  667. {

  668.     int i;

  669. 

  670.     for (i = 0; i < 8 * 8; i++)

  671.         rem[i] += (basis[i] * scale +

  672.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  673.                   (BASIS_SHIFT - RECON_SHIFT);

  674. }

  675. 

  676. static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,

  677.                     int stride, int h)

  678. {

  679.     return 0;

  680. }

  681. 

  682. void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)

  683. {

  684.     int i;

  685. 

  686.     memset(cmp, 0, sizeof(void *) * 6);

  687. 

  688.     for (i = 0; i < 6; i++) {

  689.         switch (type & 0xFF) {

  690.         case FF_CMP_SAD:

  691.             cmp[i] = c->sad[i];

  692.             break;

  693.         case FF_CMP_SATD:

  694.             cmp[i] = c->hadamard8_diff[i];

  695.             break;

  696.         case FF_CMP_SSE:

  697.             cmp[i] = c->sse[i];

  698.             break;

  699.         case FF_CMP_DCT:

  700.             cmp[i] = c->dct_sad[i];

  701.             break;

  702.         case FF_CMP_DCT264:

  703.             cmp[i] = c->dct264_sad[i];

  704.             break;

  705.         case FF_CMP_DCTMAX:

  706.             cmp[i] = c->dct_max[i];

  707.             break;

  708.         case FF_CMP_PSNR:

  709.             cmp[i] = c->quant_psnr[i];

  710.             break;

  711.         case FF_CMP_BIT:

  712.             cmp[i] = c->bit[i];

  713.             break;

  714.         case FF_CMP_RD:

  715.             cmp[i] = c->rd[i];

  716.             break;

  717.         case FF_CMP_VSAD:

  718.             cmp[i] = c->vsad[i];

  719.             break;

  720.         case FF_CMP_VSSE:

  721.             cmp[i] = c->vsse[i];

  722.             break;

  723.         case FF_CMP_ZERO:

  724.             cmp[i] = zero_cmp;

  725.             break;

  726.         case FF_CMP_NSSE:

  727.             cmp[i] = c->nsse[i];

  728.             break;

  729. #if CONFIG_DWT

  730.         case FF_CMP_W53:

  731.             cmp[i]= c->w53[i];

  732.             break;

  733.         case FF_CMP_W97:

  734.             cmp[i]= c->w97[i];

  735.             break;

  736. #endif

  737.         default:

  738.             av_log(NULL, AV_LOG_ERROR,

  739.                    "internal error in cmp function selection\n");

  740.         }

  741.     }

  742. }

  743. 

  744. #define BUTTERFLY2(o1, o2, i1, i2)              \

  745.     o1 = (i1) + (i2);                           \

  746.     o2 = (i1) - (i2);

  747. 

  748. #define BUTTERFLY1(x, y)                        \

  749.     {                                           \

  750.         int a, b;                               \

  751.         a = x;                                  \

  752.         b = y;                                  \

  753.         x = a + b;                              \

  754.         y = a - b;                              \

  755.     }

  756. 

  757. #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))

  758. 

  759. static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,

  760.                                uint8_t *src, int stride, int h)

  761. {

  762.     int i, temp[64], sum = 0;

  763. 

  764.     av_assert2(h == 8);

  765. 

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

  767.         // FIXME: try pointer walks

  768.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  769.                    src[stride * i + 0] - dst[stride * i + 0],

  770.                    src[stride * i + 1] - dst[stride * i + 1]);

  771.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  772.                    src[stride * i + 2] - dst[stride * i + 2],

  773.                    src[stride * i + 3] - dst[stride * i + 3]);

  774.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  775.                    src[stride * i + 4] - dst[stride * i + 4],

  776.                    src[stride * i + 5] - dst[stride * i + 5]);

  777.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  778.                    src[stride * i + 6] - dst[stride * i + 6],

  779.                    src[stride * i + 7] - dst[stride * i + 7]);

  780. 

  781.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  782.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  783.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  784.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  785. 

  786.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  787.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  788.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  789.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  790.     }

  791. 

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

  793.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  794.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  795.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  796.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  797. 

  798.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  799.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  800.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  801.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  802. 

  803.         sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +

  804.                BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +

  805.                BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +

  806.                BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  807.     }

  808.     return sum;

  809. }

  810. 

  811. static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,

  812.                                 uint8_t *dummy, int stride, int h)

  813. {

  814.     int i, temp[64], sum = 0;

  815. 

  816.     av_assert2(h == 8);

  817. 

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

  819.         // FIXME: try pointer walks

  820.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  821.                    src[stride * i + 0], src[stride * i + 1]);

  822.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  823.                    src[stride * i + 2], src[stride * i + 3]);

  824.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  825.                    src[stride * i + 4], src[stride * i + 5]);

  826.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  827.                    src[stride * i + 6], src[stride * i + 7]);

  828. 

  829.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  830.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  831.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  832.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  833. 

  834.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  835.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  836.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  837.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  838.     }

  839. 

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

  841.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  842.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  843.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  844.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  845. 

  846.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  847.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  848.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  849.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  850. 

  851.         sum +=

  852.             BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])

  853.             + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])

  854.             + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])

  855.             + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  856.     }

  857. 

  858.     sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean

  859. 

  860.     return sum;

  861. }

  862. 

  863. static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  864.                         uint8_t *src2, int stride, int h)

  865. {

  866.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  867. 

  868.     av_assert2(h == 8);

  869. 

  870.     s->dsp.diff_pixels(temp, src1, src2, stride);

  871.     s->dsp.fdct(temp);

  872.     return s->dsp.sum_abs_dctelem(temp);

  873. }

  874. 

  875. #if CONFIG_GPL

  876. #define DCT8_1D                                         \

  877.     {                                                   \

  878.         const int s07 = SRC(0) + SRC(7);                \

  879.         const int s16 = SRC(1) + SRC(6);                \

  880.         const int s25 = SRC(2) + SRC(5);                \

  881.         const int s34 = SRC(3) + SRC(4);                \

  882.         const int a0  = s07 + s34;                      \

  883.         const int a1  = s16 + s25;                      \

  884.         const int a2  = s07 - s34;                      \

  885.         const int a3  = s16 - s25;                      \

  886.         const int d07 = SRC(0) - SRC(7);                \

  887.         const int d16 = SRC(1) - SRC(6);                \

  888.         const int d25 = SRC(2) - SRC(5);                \

  889.         const int d34 = SRC(3) - SRC(4);                \

  890.         const int a4  = d16 + d25 + (d07 + (d07 >> 1)); \

  891.         const int a5  = d07 - d34 - (d25 + (d25 >> 1)); \

  892.         const int a6  = d07 + d34 - (d16 + (d16 >> 1)); \

  893.         const int a7  = d16 - d25 + (d34 + (d34 >> 1)); \

  894.         DST(0, a0 + a1);                                \

  895.         DST(1, a4 + (a7 >> 2));                         \

  896.         DST(2, a2 + (a3 >> 1));                         \

  897.         DST(3, a5 + (a6 >> 2));                         \

  898.         DST(4, a0 - a1);                                \

  899.         DST(5, a6 - (a5 >> 2));                         \

  900.         DST(6, (a2 >> 1) - a3);                         \

  901.         DST(7, (a4 >> 2) - a7);                         \

  902.     }

  903. 

  904. static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  905.                            uint8_t *src2, int stride, int h)

  906. {

  907.     int16_t dct[8][8];

  908.     int i, sum = 0;

  909. 

  910.     s->dsp.diff_pixels(dct[0], src1, src2, stride);

  911. 

  912. #define SRC(x) dct[i][x]

  913. #define DST(x, v) dct[i][x] = v

  914.     for (i = 0; i < 8; i++)

  915.         DCT8_1D

  916. #undef SRC

  917. #undef DST

  918. 

  919. #define SRC(x) dct[x][i]

  920. #define DST(x, v) sum += FFABS(v)

  921.         for (i = 0; i < 8; i++)

  922.             DCT8_1D

  923. #undef SRC

  924. #undef DST

  925.             return sum;

  926. }

  927. #endif

  928. 

  929. static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,

  930.                         uint8_t *src2, int stride, int h)

  931. {

  932.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  933.     int sum = 0, i;

  934. 

  935.     av_assert2(h == 8);

  936. 

  937.     s->dsp.diff_pixels(temp, src1, src2, stride);

  938.     s->dsp.fdct(temp);

  939. 

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

  941.         sum = FFMAX(sum, FFABS(temp[i]));

  942. 

  943.     return sum;

  944. }

  945. 

  946. static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,

  947.                            uint8_t *src2, int stride, int h)

  948. {

  949.     LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);

  950.     int16_t *const bak = temp + 64;

  951.     int sum = 0, i;

  952. 

  953.     av_assert2(h == 8);

  954.     s->mb_intra = 0;

  955. 

  956.     s->dsp.diff_pixels(temp, src1, src2, stride);

  957. 

  958.     memcpy(bak, temp, 64 * sizeof(int16_t));

  959. 

  960.     s->block_last_index[0 /* FIXME */] =

  961.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  962.     s->dct_unquantize_inter(s, temp, 0, s->qscale);

  963.     ff_simple_idct_8(temp); // FIXME

  964. 

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

  966.         sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);

  967. 

  968.     return sum;

  969. }

  970. 

  971. static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  972.                    int stride, int h)

  973. {

  974.     const uint8_t *scantable = s->intra_scantable.permutated;

  975.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  976.     LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);

  977.     LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);

  978.     int i, last, run, bits, level, distortion, start_i;

  979.     const int esc_length = s->ac_esc_length;

  980.     uint8_t *length, *last_length;

  981. 

  982.     av_assert2(h == 8);

  983. 

  984.     copy_block8(lsrc1, src1, 8, stride, 8);

  985.     copy_block8(lsrc2, src2, 8, stride, 8);

  986. 

  987.     s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);

  988. 

  989.     s->block_last_index[0 /* FIXME */] =

  990.     last                               =

  991.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  992. 

  993.     bits = 0;

  994. 

  995.     if (s->mb_intra) {

  996.         start_i     = 1;

  997.         length      = s->intra_ac_vlc_length;

  998.         last_length = s->intra_ac_vlc_last_length;

  999.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1000.     } else {

  1001.         start_i     = 0;

  1002.         length      = s->inter_ac_vlc_length;

  1003.         last_length = s->inter_ac_vlc_last_length;

  1004.     }

  1005. 

  1006.     if (last >= start_i) {

  1007.         run = 0;

  1008.         for (i = start_i; i < last; i++) {

  1009.             int j = scantable[i];

  1010.             level = temp[j];

  1011. 

  1012.             if (level) {

  1013.                 level += 64;

  1014.                 if ((level & (~127)) == 0)

  1015.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1016.                 else

  1017.                     bits += esc_length;

  1018.                 run = 0;

  1019.             } else

  1020.                 run++;

  1021.         }

  1022.         i = scantable[last];

  1023. 

  1024.         level = temp[i] + 64;

  1025. 

  1026.         av_assert2(level - 64);

  1027. 

  1028.         if ((level & (~127)) == 0) {

  1029.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1030.         } else

  1031.             bits += esc_length;

  1032.     }

  1033. 

  1034.     if (last >= 0) {

  1035.         if (s->mb_intra)

  1036.             s->dct_unquantize_intra(s, temp, 0, s->qscale);

  1037.         else

  1038.             s->dct_unquantize_inter(s, temp, 0, s->qscale);

  1039.     }

  1040. 

  1041.     s->dsp.idct_add(lsrc2, 8, temp);

  1042. 

  1043.     distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);

  1044. 

  1045.     return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);

  1046. }

  1047. 

  1048. static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  1049.                     int stride, int h)

  1050. {

  1051.     const uint8_t *scantable = s->intra_scantable.permutated;

  1052.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1053.     int i, last, run, bits, level, start_i;

  1054.     const int esc_length = s->ac_esc_length;

  1055.     uint8_t *length, *last_length;

  1056. 

  1057.     av_assert2(h == 8);

  1058. 

  1059.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1060. 

  1061.     s->block_last_index[0 /* FIXME */] =

  1062.     last                               =

  1063.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1064. 

  1065.     bits = 0;

  1066. 

  1067.     if (s->mb_intra) {

  1068.         start_i     = 1;

  1069.         length      = s->intra_ac_vlc_length;

  1070.         last_length = s->intra_ac_vlc_last_length;

  1071.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1072.     } else {

  1073.         start_i     = 0;

  1074.         length      = s->inter_ac_vlc_length;

  1075.         last_length = s->inter_ac_vlc_last_length;

  1076.     }

  1077. 

  1078.     if (last >= start_i) {

  1079.         run = 0;

  1080.         for (i = start_i; i < last; i++) {

  1081.             int j = scantable[i];

  1082.             level = temp[j];

  1083. 

  1084.             if (level) {

  1085.                 level += 64;

  1086.                 if ((level & (~127)) == 0)

  1087.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1088.                 else

  1089.                     bits += esc_length;

  1090.                 run = 0;

  1091.             } else

  1092.                 run++;

  1093.         }

  1094.         i = scantable[last];

  1095. 

  1096.         level = temp[i] + 64;

  1097. 

  1098.         av_assert2(level - 64);

  1099. 

  1100.         if ((level & (~127)) == 0)

  1101.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1102.         else

  1103.             bits += esc_length;

  1104.     }

  1105. 

  1106.     return bits;

  1107. }

  1108. 

  1109. #define VSAD_INTRA(size)                                                \

  1110. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  1111.                                     uint8_t *s, uint8_t *dummy,         \

  1112.                                     int stride, int h)                  \

  1113. {                                                                       \

  1114.     int score = 0, x, y;                                                \

  1115.                                                                         \

  1116.     for (y = 1; y < h; y++) {                                           \

  1117.         for (x = 0; x < size; x += 4) {                                 \

  1118.             score += FFABS(s[x]     - s[x + stride])     +              \

  1119.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  1120.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  1121.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  1122.         }                                                               \

  1123.         s += stride;                                                    \

  1124.     }                                                                   \

  1125.                                                                         \

  1126.     return score;                                                       \

  1127. }

  1128. VSAD_INTRA(8)

  1129. VSAD_INTRA(16)

  1130. 

  1131. #define VSAD(size)                                                             \

  1132. static int vsad ## size ## _c(MpegEncContext *c,                               \

  1133.                               uint8_t *s1, uint8_t *s2,                        \

  1134.                               int stride, int h)                               \

  1135. {                                                                              \

  1136.     int score = 0, x, y;                                                       \

  1137.                                                                                \

  1138.     for (y = 1; y < h; y++) {                                                  \

  1139.         for (x = 0; x < size; x++)                                             \

  1140.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);   \

  1141.         s1 += stride;                                                          \

  1142.         s2 += stride;                                                          \

  1143.     }                                                                          \

  1144.                                                                                \

  1145.     return score;                                                              \

  1146. }

  1147. VSAD(8)

  1148. VSAD(16)

  1149. 

  1150. #define SQ(a) ((a) * (a))

  1151. #define VSSE_INTRA(size)                                                \

  1152. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  1153.                                     uint8_t *s, uint8_t *dummy,         \

  1154.                                     int stride, int h)                  \

  1155. {                                                                       \

  1156.     int score = 0, x, y;                                                \

  1157.                                                                         \

  1158.     for (y = 1; y < h; y++) {                                           \

  1159.         for (x = 0; x < size; x += 4) {                                 \

  1160.             score += SQ(s[x]     - s[x + stride]) +                     \

  1161.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  1162.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  1163.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  1164.         }                                                               \

  1165.         s += stride;                                                    \

  1166.     }                                                                   \

  1167.                                                                         \

  1168.     return score;                                                       \

  1169. }

  1170. VSSE_INTRA(8)

  1171. VSSE_INTRA(16)

  1172. 

  1173. #define VSSE(size)                                                             \

  1174. static int vsse ## size ## _c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,     \

  1175.                     int stride, int h)                                         \

  1176. {                                                                              \

  1177.     int score = 0, x, y;                                                       \

  1178.                                                                                \

  1179.     for (y = 1; y < h; y++) {                                                  \

  1180.         for (x = 0; x < size; x++)                                             \

  1181.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);      \

  1182.         s1 += stride;                                                          \

  1183.         s2 += stride;                                                          \

  1184.     }                                                                          \

  1185.                                                                                \

  1186.     return score;                                                              \

  1187. }

  1188. VSSE(8)

  1189. VSSE(16)

  1190. 

  1191. #define WRAPPER8_16_SQ(name8, name16)                                   \

  1192. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  1193.                   int stride, int h)                                    \

  1194. {                                                                       \

  1195.     int score = 0;                                                      \

  1196.                                                                         \

  1197.     score += name8(s, dst, src, stride, 8);                             \

  1198.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1199.     if (h == 16) {                                                      \

  1200.         dst   += 8 * stride;                                            \

  1201.         src   += 8 * stride;                                            \

  1202.         score += name8(s, dst, src, stride, 8);                         \

  1203.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1204.     }                                                                   \

  1205.     return score;                                                       \

  1206. }

  1207. 

  1208. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1209. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1210. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1211. #if CONFIG_GPL

  1212. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1213. #endif

  1214. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1215. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1216. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1217. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1218. 

  1219. static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)

  1220. {

  1221.     ff_j_rev_dct(block);

  1222.     put_pixels_clamped_c(block, dest, line_size);

  1223. }

  1224. 

  1225. static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)

  1226. {

  1227.     ff_j_rev_dct(block);

  1228.     add_pixels_clamped_c(block, dest, line_size);

  1229. }

  1230. 

  1231. static void ff_jref_idct4_put(uint8_t *dest, int line_size, int16_t *block)

  1232. {

  1233.     ff_j_rev_dct4 (block);

  1234.     put_pixels_clamped4_c(block, dest, line_size);

  1235. }

  1236. static void ff_jref_idct4_add(uint8_t *dest, int line_size, int16_t *block)

  1237. {

  1238.     ff_j_rev_dct4 (block);

  1239.     add_pixels_clamped4_c(block, dest, line_size);

  1240. }

  1241. 

  1242. static void ff_jref_idct2_put(uint8_t *dest, int line_size, int16_t *block)

  1243. {

  1244.     ff_j_rev_dct2 (block);

  1245.     put_pixels_clamped2_c(block, dest, line_size);

  1246. }

  1247. static void ff_jref_idct2_add(uint8_t *dest, int line_size, int16_t *block)

  1248. {

  1249.     ff_j_rev_dct2 (block);

  1250.     add_pixels_clamped2_c(block, dest, line_size);

  1251. }

  1252. 

  1253. static void ff_jref_idct1_put(uint8_t *dest, int line_size, int16_t *block)

  1254. {

  1255.     dest[0] = av_clip_uint8((block[0] + 4)>>3);

  1256. }

  1257. static void ff_jref_idct1_add(uint8_t *dest, int line_size, int16_t *block)

  1258. {

  1259.     dest[0] = av_clip_uint8(dest[0] + ((block[0] + 4)>>3));

  1260. }

  1261. 

  1262. /* draw the edges of width 'w' of an image of size width, height */

  1263. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1264. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1265.                            int w, int h, int sides)

  1266. {

  1267.     uint8_t *ptr = buf, *last_line;

  1268.     int i;

  1269. 

  1270.     /* left and right */

  1271.     for (i = 0; i < height; i++) {

  1272.         memset(ptr - w, ptr[0], w);

  1273.         memset(ptr + width, ptr[width - 1], w);

  1274.         ptr += wrap;

  1275.     }

  1276. 

  1277.     /* top and bottom + corners */

  1278.     buf -= w;

  1279.     last_line = buf + (height - 1) * wrap;

  1280.     if (sides & EDGE_TOP)

  1281.         for (i = 0; i < h; i++)

  1282.             // top

  1283.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1284.     if (sides & EDGE_BOTTOM)

  1285.         for (i = 0; i < h; i++)

  1286.             // bottom

  1287.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1288. }

  1289. 

  1290. /* init static data */

  1291. av_cold void ff_dsputil_static_init(void)

  1292. {

  1293.     int i;

  1294. 

  1295.     for (i = 0; i < 512; i++)

  1296.         ff_square_tab[i] = (i - 256) * (i - 256);

  1297. }

  1298. 

  1299. int ff_check_alignment(void)

  1300. {

  1301.     static int did_fail = 0;

  1302.     LOCAL_ALIGNED_16(int, aligned, [4]);

  1303. 

  1304.     if ((intptr_t)aligned & 15) {

  1305.         if (!did_fail) {

  1306. #if HAVE_MMX || HAVE_ALTIVEC

  1307.             av_log(NULL, AV_LOG_ERROR,

  1308.                 "Compiler did not align stack variables. Libavcodec has been miscompiled\n"

  1309.                 "and may be very slow or crash. This is not a bug in libavcodec,\n"

  1310.                 "but in the compiler. You may try recompiling using gcc >= 4.2.\n"

  1311.                 "Do not report crashes to FFmpeg developers.\n");

  1312. #endif

  1313.             did_fail=1;

  1314.         }

  1315.         return -1;

  1316.     }

  1317.     return 0;

  1318. }

  1319. 

  1320. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1321. {

  1322.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1323. 

  1324.     ff_check_alignment();

  1325. 

  1326. #if CONFIG_ENCODERS

  1327.     if (avctx->bits_per_raw_sample == 10) {

  1328.         c->fdct    = ff_jpeg_fdct_islow_10;

  1329.         c->fdct248 = ff_fdct248_islow_10;

  1330.     } else {

  1331.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1332.             c->fdct    = ff_fdct_ifast;

  1333.             c->fdct248 = ff_fdct_ifast248;

  1334.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1335.             c->fdct    = ff_faandct;

  1336.             c->fdct248 = ff_faandct248;

  1337.         } else {

  1338.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1339.             c->fdct248 = ff_fdct248_islow_8;

  1340.         }

  1341.     }

  1342. #endif /* CONFIG_ENCODERS */

  1343. 

  1344.     if (avctx->lowres==1) {

  1345.         c->idct_put              = ff_jref_idct4_put;

  1346.         c->idct_add              = ff_jref_idct4_add;

  1347.         c->idct                  = ff_j_rev_dct4;

  1348.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1349.     } else if (avctx->lowres==2) {

  1350.         c->idct_put              =  ff_jref_idct2_put;

  1351.         c->idct_add              =  ff_jref_idct2_add;

  1352.         c->idct                  =  ff_j_rev_dct2;

  1353.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1354.     } else if (avctx->lowres==3) {

  1355.         c->idct_put              =  ff_jref_idct1_put;

  1356.         c->idct_add              =  ff_jref_idct1_add;

  1357.         c->idct                  =  ff_j_rev_dct1;

  1358.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1359.     } else {

  1360.         if (avctx->bits_per_raw_sample == 10) {

  1361.             c->idct_put              = ff_simple_idct_put_10;

  1362.             c->idct_add              = ff_simple_idct_add_10;

  1363.             c->idct                  = ff_simple_idct_10;

  1364.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1365.         } else if (avctx->bits_per_raw_sample == 12) {

  1366.             c->idct_put              = ff_simple_idct_put_12;

  1367.             c->idct_add              = ff_simple_idct_add_12;

  1368.             c->idct                  = ff_simple_idct_12;

  1369.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1370.         } else {

  1371.         if (avctx->idct_algo == FF_IDCT_INT) {

  1372.             c->idct_put              = jref_idct_put;

  1373.             c->idct_add              = jref_idct_add;

  1374.             c->idct                  = ff_j_rev_dct;

  1375.             c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;

  1376.         } else if (avctx->idct_algo == FF_IDCT_FAAN) {

  1377.             c->idct_put              = ff_faanidct_put;

  1378.             c->idct_add              = ff_faanidct_add;

  1379.             c->idct                  = ff_faanidct;

  1380.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1381.         } else { // accurate/default

  1382.             c->idct_put              = ff_simple_idct_put_8;

  1383.             c->idct_add              = ff_simple_idct_add_8;

  1384.             c->idct                  = ff_simple_idct_8;

  1385.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1386.         }

  1387.         }

  1388.     }

  1389. 

  1390.     c->diff_pixels = diff_pixels_c;

  1391. 

  1392.     c->put_pixels_clamped        = put_pixels_clamped_c;

  1393.     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;

  1394.     c->add_pixels_clamped        = add_pixels_clamped_c;

  1395. 

  1396.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1397. 

  1398.     c->pix_sum   = pix_sum_c;

  1399.     c->pix_norm1 = pix_norm1_c;

  1400. 

  1401.     /* TODO [0] 16  [1] 8 */

  1402.     c->pix_abs[0][0] = pix_abs16_c;

  1403.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1404.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1405.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1406.     c->pix_abs[1][0] = pix_abs8_c;

  1407.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1408.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1409.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1410. 

  1411. #define SET_CMP_FUNC(name)                      \

  1412.     c->name[0] = name ## 16_c;                  \

  1413.     c->name[1] = name ## 8x8_c;

  1414. 

  1415.     SET_CMP_FUNC(hadamard8_diff)

  1416.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1417.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1418.     SET_CMP_FUNC(dct_sad)

  1419.     SET_CMP_FUNC(dct_max)

  1420. #if CONFIG_GPL

  1421.     SET_CMP_FUNC(dct264_sad)

  1422. #endif

  1423.     c->sad[0] = pix_abs16_c;

  1424.     c->sad[1] = pix_abs8_c;

  1425.     c->sse[0] = sse16_c;

  1426.     c->sse[1] = sse8_c;

  1427.     c->sse[2] = sse4_c;

  1428.     SET_CMP_FUNC(quant_psnr)

  1429.     SET_CMP_FUNC(rd)

  1430.     SET_CMP_FUNC(bit)

  1431.     c->vsad[0] = vsad16_c;

  1432.     c->vsad[1] = vsad8_c;

  1433.     c->vsad[4] = vsad_intra16_c;

  1434.     c->vsad[5] = vsad_intra8_c;

  1435.     c->vsse[0] = vsse16_c;

  1436.     c->vsse[1] = vsse8_c;

  1437.     c->vsse[4] = vsse_intra16_c;

  1438.     c->vsse[5] = vsse_intra8_c;

  1439.     c->nsse[0] = nsse16_c;

  1440.     c->nsse[1] = nsse8_c;

  1441. #if CONFIG_SNOW_DECODER || CONFIG_SNOW_ENCODER

  1442.     ff_dsputil_init_dwt(c);

  1443. #endif

  1444. 

  1445.     c->try_8x8basis = try_8x8basis_c;

  1446.     c->add_8x8basis = add_8x8basis_c;

  1447. 

  1448.     c->shrink[0] = av_image_copy_plane;

  1449.     c->shrink[1] = ff_shrink22;

  1450.     c->shrink[2] = ff_shrink44;

  1451.     c->shrink[3] = ff_shrink88;

  1452. 

  1453.     c->draw_edges = draw_edges_8_c;

  1454. 

  1455.     switch (avctx->bits_per_raw_sample) {

  1456.     case 9:

  1457.     case 10:

  1458.     case 12:

  1459.     case 14:

  1460.         c->get_pixels = get_pixels_16_c;

  1461.         break;

  1462.     default:

  1463.         if (avctx->bits_per_raw_sample<=8 || avctx->codec_type != AVMEDIA_TYPE_VIDEO) {

  1464.             c->get_pixels = get_pixels_8_c;

  1465.         }

  1466.         break;

  1467.     }

  1468. 

  1469. 

  1470.     if (ARCH_ALPHA)

  1471.         ff_dsputil_init_alpha(c, avctx);

  1472.     if (ARCH_ARM)

  1473.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1474.     if (ARCH_PPC)

  1475.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1476.     if (ARCH_X86)

  1477.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1478. 

  1479.     ff_init_scantable_permutation(c->idct_permutation,

  1480.                                   c->idct_permutation_type);

  1481. }

  1482. 

  1483. av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)

  1484. {

  1485.     ff_dsputil_init(c, avctx);

  1486. }

  1487. 

  1488. av_cold void avpriv_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1489. {

  1490.     ff_dsputil_init(c, avctx);

  1491. }