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.
40139 Commits
32 Branches
5.3GB
Tree: adcb8392c9
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'adcb8392c9'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1316 lines
42KB
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 Libav.

  7.  *

  8.  * Libav 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.  * Libav 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 Libav; 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 "avcodec.h"

  31. #include "copy_block.h"

  32. #include "dct.h"

  33. #include "dsputil.h"

  34. #include "simple_idct.h"

  35. #include "faandct.h"

  36. #include "faanidct.h"

  37. #include "imgconvert.h"

  38. #include "mathops.h"

  39. #include "mpegvideo.h"

  40. #include "config.h"

  41. 

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

  43. 

  44. #define BIT_DEPTH 16

  45. #include "dsputilenc_template.c"

  46. #undef BIT_DEPTH

  47. 

  48. #define BIT_DEPTH 8

  49. #include "dsputilenc_template.c"

  50. 

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

  52.                                const uint8_t *src_scantable)

  53. {

  54.     int i, end;

  55. 

  56.     st->scantable = src_scantable;

  57. 

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

  59.         int j = src_scantable[i];

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

  61.     }

  62. 

  63.     end = -1;

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

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

  66.         if (j > end)

  67.             end = j;

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

  69.     }

  70. }

  71. 

  72. av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,

  73.                                            int idct_permutation_type)

  74. {

  75.     int i;

  76. 

  77.     if (ARCH_X86)

  78.         if (ff_init_scantable_permutation_x86(idct_permutation,

  79.                                               idct_permutation_type))

  80.             return;

  81. 

  82.     switch (idct_permutation_type) {

  83.     case FF_NO_IDCT_PERM:

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

  85.             idct_permutation[i] = i;

  86.         break;

  87.     case FF_LIBMPEG2_IDCT_PERM:

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

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

  90.         break;

  91.     case FF_TRANSPOSE_IDCT_PERM:

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

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

  94.         break;

  95.     case FF_PARTTRANS_IDCT_PERM:

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

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

  98.         break;

  99.     default:

  100.         av_log(NULL, AV_LOG_ERROR,

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

  102.     }

  103. }

  104. 

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

  106. {

  107.     int s = 0, i, j;

  108. 

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

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

  111.             s   += pix[0];

  112.             s   += pix[1];

  113.             s   += pix[2];

  114.             s   += pix[3];

  115.             s   += pix[4];

  116.             s   += pix[5];

  117.             s   += pix[6];

  118.             s   += pix[7];

  119.             pix += 8;

  120.         }

  121.         pix += line_size - 16;

  122.     }

  123.     return s;

  124. }

  125. 

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

  127. {

  128.     int s = 0, i, j;

  129.     uint32_t *sq = ff_square_tab + 256;

  130. 

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

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

  133. #if 0

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

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

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

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

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

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

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

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

  142. #else

  143. #if HAVE_FAST_64BIT

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

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

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

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

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

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

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

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

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

  153. #else

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

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

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

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

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

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

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

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

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

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

  164. #endif

  165. #endif

  166.             pix += 8;

  167.         }

  168.         pix += line_size - 16;

  169.     }

  170.     return s;

  171. }

  172. 

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

  174.                   int line_size, int h)

  175. {

  176.     int s = 0, i;

  177.     uint32_t *sq = ff_square_tab + 256;

  178. 

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

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

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

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

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

  184.         pix1 += line_size;

  185.         pix2 += line_size;

  186.     }

  187.     return s;

  188. }

  189. 

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

  191.                   int line_size, int h)

  192. {

  193.     int s = 0, i;

  194.     uint32_t *sq = ff_square_tab + 256;

  195. 

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

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

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

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

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

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

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

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

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

  205.         pix1 += line_size;

  206.         pix2 += line_size;

  207.     }

  208.     return s;

  209. }

  210. 

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

  212.                    int line_size, int h)

  213. {

  214.     int s = 0, i;

  215.     uint32_t *sq = ff_square_tab + 256;

  216. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  234. 

  235.         pix1 += line_size;

  236.         pix2 += line_size;

  237.     }

  238.     return s;

  239. }

  240. 

  241. static void diff_pixels_c(int16_t *restrict block, const uint8_t *s1,

  242.                           const uint8_t *s2, int stride)

  243. {

  244.     int i;

  245. 

  246.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  256.         s1      += stride;

  257.         s2      += stride;

  258.         block   += 8;

  259.     }

  260. }

  261. 

  262. static void put_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,

  263.                                  int line_size)

  264. {

  265.     int i;

  266. 

  267.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  277. 

  278.         pixels += line_size;

  279.         block  += 8;

  280.     }

  281. }

  282. 

  283. static void put_signed_pixels_clamped_c(const int16_t *block,

  284.                                         uint8_t *restrict pixels,

  285.                                         int line_size)

  286. {

  287.     int i, j;

  288. 

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

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

  291.             if (*block < -128)

  292.                 *pixels = 0;

  293.             else if (*block > 127)

  294.                 *pixels = 255;

  295.             else

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

  297.             block++;

  298.             pixels++;

  299.         }

  300.         pixels += (line_size - 8);

  301.     }

  302. }

  303. 

  304. static void add_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,

  305.                                  int line_size)

  306. {

  307.     int i;

  308. 

  309.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  319.         pixels   += line_size;

  320.         block    += 8;

  321.     }

  322. }

  323. 

  324. static int sum_abs_dctelem_c(int16_t *block)

  325. {

  326.     int sum = 0, i;

  327. 

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

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

  330.     return sum;

  331. }

  332. 

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

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

  335. 

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

  337.                               int line_size, int h)

  338. {

  339.     int s = 0, i;

  340. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  358.         pix1 += line_size;

  359.         pix2 += line_size;

  360.     }

  361.     return s;

  362. }

  363. 

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

  365.                           int line_size, int h)

  366. {

  367.     int s = 0, i;

  368. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  386.         pix1 += line_size;

  387.         pix2 += line_size;

  388.     }

  389.     return s;

  390. }

  391. 

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

  393.                           int line_size, int h)

  394. {

  395.     int s = 0, i;

  396.     uint8_t *pix3 = pix2 + line_size;

  397. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  415.         pix1 += line_size;

  416.         pix2 += line_size;

  417.         pix3 += line_size;

  418.     }

  419.     return s;

  420. }

  421. 

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

  423.                            int line_size, int h)

  424. {

  425.     int s = 0, i;

  426.     uint8_t *pix3 = pix2 + line_size;

  427. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  445.         pix1 += line_size;

  446.         pix2 += line_size;

  447.         pix3 += line_size;

  448.     }

  449.     return s;

  450. }

  451. 

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

  453.                              int line_size, int h)

  454. {

  455.     int s = 0, i;

  456. 

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

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

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

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

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

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

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

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

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

  466.         pix1 += line_size;

  467.         pix2 += line_size;

  468.     }

  469.     return s;

  470. }

  471. 

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

  473.                          int line_size, int h)

  474. {

  475.     int s = 0, i;

  476. 

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

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

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

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

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

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

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

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

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

  486.         pix1 += line_size;

  487.         pix2 += line_size;

  488.     }

  489.     return s;

  490. }

  491. 

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

  493.                          int line_size, int h)

  494. {

  495.     int s = 0, i;

  496.     uint8_t *pix3 = pix2 + line_size;

  497. 

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

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

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

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

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

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

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

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

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

  507.         pix1 += line_size;

  508.         pix2 += line_size;

  509.         pix3 += line_size;

  510.     }

  511.     return s;

  512. }

  513. 

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

  515.                           int line_size, int h)

  516. {

  517.     int s = 0, i;

  518.     uint8_t *pix3 = pix2 + line_size;

  519. 

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

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

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

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

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

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

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

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

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

  529.         pix1 += line_size;

  530.         pix2 += line_size;

  531.         pix3 += line_size;

  532.     }

  533.     return s;

  534. }

  535. 

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

  537. {

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

  539. 

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

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

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

  543.         if (y + 1 < h) {

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

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

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

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

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

  549.         }

  550.         s1 += stride;

  551.         s2 += stride;

  552.     }

  553. 

  554.     if (c)

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

  556.     else

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

  558. }

  559. 

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

  561. {

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

  563. 

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

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

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

  567.         if (y + 1 < h) {

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

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

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

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

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

  573.         }

  574.         s1 += stride;

  575.         s2 += stride;

  576.     }

  577. 

  578.     if (c)

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

  580.     else

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

  582. }

  583. 

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

  585.                           int16_t basis[64], int scale)

  586. {

  587.     int i;

  588.     unsigned int sum = 0;

  589. 

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

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

  592.                            (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  593.                           (BASIS_SHIFT - RECON_SHIFT));

  594.         int w = weight[i];

  595.         b >>= RECON_SHIFT;

  596.         assert(-512 < b && b < 512);

  597. 

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

  599.     }

  600.     return sum >> 2;

  601. }

  602. 

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

  604. {

  605.     int i;

  606. 

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

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

  609.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  610.                   (BASIS_SHIFT - RECON_SHIFT);

  611. }

  612. 

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

  614.                     int stride, int h)

  615. {

  616.     return 0;

  617. }

  618. 

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

  620. {

  621.     int i;

  622. 

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

  624. 

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

  626.         switch (type & 0xFF) {

  627.         case FF_CMP_SAD:

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

  629.             break;

  630.         case FF_CMP_SATD:

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

  632.             break;

  633.         case FF_CMP_SSE:

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

  635.             break;

  636.         case FF_CMP_DCT:

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

  638.             break;

  639.         case FF_CMP_DCT264:

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

  641.             break;

  642.         case FF_CMP_DCTMAX:

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

  644.             break;

  645.         case FF_CMP_PSNR:

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

  647.             break;

  648.         case FF_CMP_BIT:

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

  650.             break;

  651.         case FF_CMP_RD:

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

  653.             break;

  654.         case FF_CMP_VSAD:

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

  656.             break;

  657.         case FF_CMP_VSSE:

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

  659.             break;

  660.         case FF_CMP_ZERO:

  661.             cmp[i] = zero_cmp;

  662.             break;

  663.         case FF_CMP_NSSE:

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

  665.             break;

  666.         default:

  667.             av_log(NULL, AV_LOG_ERROR,

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

  669.         }

  670.     }

  671. }

  672. 

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

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

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

  676. 

  677. #define BUTTERFLY1(x, y)                        \

  678.     {                                           \

  679.         int a, b;                               \

  680.         a = x;                                  \

  681.         b = y;                                  \

  682.         x = a + b;                              \

  683.         y = a - b;                              \

  684.     }

  685. 

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

  687. 

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

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

  690. {

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

  692. 

  693.     assert(h == 8);

  694. 

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

  696.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  709. 

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

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

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

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

  714. 

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

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

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

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

  719.     }

  720. 

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

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

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

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

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

  726. 

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

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

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

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

  731. 

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

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

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

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

  736.     }

  737.     return sum;

  738. }

  739. 

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

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

  742. {

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

  744. 

  745.     assert(h == 8);

  746. 

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

  748.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  757. 

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

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

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

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

  762. 

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

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

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

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

  767.     }

  768. 

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

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

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

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

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

  774. 

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

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

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

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

  779. 

  780.         sum +=

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

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

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

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

  785.     }

  786. 

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

  788. 

  789.     return sum;

  790. }

  791. 

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

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

  794. {

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

  796. 

  797.     assert(h == 8);

  798. 

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

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

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

  802. }

  803. 

  804. #if CONFIG_GPL

  805. #define DCT8_1D                                         \

  806.     {                                                   \

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

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

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

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

  811.         const int a0  = s07 + s34;                      \

  812.         const int a1  = s16 + s25;                      \

  813.         const int a2  = s07 - s34;                      \

  814.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  831.     }

  832. 

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

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

  835. {

  836.     int16_t dct[8][8];

  837.     int i, sum = 0;

  838. 

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

  840. 

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

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

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

  844.         DCT8_1D

  845. #undef SRC

  846. #undef DST

  847. 

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

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

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

  851.             DCT8_1D

  852. #undef SRC

  853. #undef DST

  854.             return sum;

  855. }

  856. #endif

  857. 

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

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

  860. {

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

  862.     int sum = 0, i;

  863. 

  864.     assert(h == 8);

  865. 

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

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

  868. 

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

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

  871. 

  872.     return sum;

  873. }

  874. 

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

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

  877. {

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

  879.     int16_t *const bak = temp + 64;

  880.     int sum = 0, i;

  881. 

  882.     assert(h == 8);

  883.     s->mb_intra = 0;

  884. 

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

  886. 

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

  888. 

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

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

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

  892.     ff_simple_idct_8(temp); // FIXME

  893. 

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

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

  896. 

  897.     return sum;

  898. }

  899. 

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

  901.                    int stride, int h)

  902. {

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

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

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

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

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

  908.     const int esc_length = s->ac_esc_length;

  909.     uint8_t *length, *last_length;

  910. 

  911.     assert(h == 8);

  912. 

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

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

  915. 

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

  917. 

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

  919.     last                               =

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

  921. 

  922.     bits = 0;

  923. 

  924.     if (s->mb_intra) {

  925.         start_i     = 1;

  926.         length      = s->intra_ac_vlc_length;

  927.         last_length = s->intra_ac_vlc_last_length;

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

  929.     } else {

  930.         start_i     = 0;

  931.         length      = s->inter_ac_vlc_length;

  932.         last_length = s->inter_ac_vlc_last_length;

  933.     }

  934. 

  935.     if (last >= start_i) {

  936.         run = 0;

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

  938.             int j = scantable[i];

  939.             level = temp[j];

  940. 

  941.             if (level) {

  942.                 level += 64;

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

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

  945.                 else

  946.                     bits += esc_length;

  947.                 run = 0;

  948.             } else

  949.                 run++;

  950.         }

  951.         i = scantable[last];

  952. 

  953.         level = temp[i] + 64;

  954. 

  955.         assert(level - 64);

  956. 

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

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

  959.         } else

  960.             bits += esc_length;

  961.     }

  962. 

  963.     if (last >= 0) {

  964.         if (s->mb_intra)

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

  966.         else

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

  968.     }

  969. 

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

  971. 

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

  973. 

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

  975. }

  976. 

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

  978.                     int stride, int h)

  979. {

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

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

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

  983.     const int esc_length = s->ac_esc_length;

  984.     uint8_t *length, *last_length;

  985. 

  986.     assert(h == 8);

  987. 

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

  989. 

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

  991.     last                               =

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

  993. 

  994.     bits = 0;

  995. 

  996.     if (s->mb_intra) {

  997.         start_i     = 1;

  998.         length      = s->intra_ac_vlc_length;

  999.         last_length = s->intra_ac_vlc_last_length;

  1000.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1001.     } else {

  1002.         start_i     = 0;

  1003.         length      = s->inter_ac_vlc_length;

  1004.         last_length = s->inter_ac_vlc_last_length;

  1005.     }

  1006. 

  1007.     if (last >= start_i) {

  1008.         run = 0;

  1009.         for (i = start_i; i < last; i++) {

  1010.             int j = scantable[i];

  1011.             level = temp[j];

  1012. 

  1013.             if (level) {

  1014.                 level += 64;

  1015.                 if ((level & (~127)) == 0)

  1016.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1017.                 else

  1018.                     bits += esc_length;

  1019.                 run = 0;

  1020.             } else

  1021.                 run++;

  1022.         }

  1023.         i = scantable[last];

  1024. 

  1025.         level = temp[i] + 64;

  1026. 

  1027.         assert(level - 64);

  1028. 

  1029.         if ((level & (~127)) == 0)

  1030.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1031.         else

  1032.             bits += esc_length;

  1033.     }

  1034. 

  1035.     return bits;

  1036. }

  1037. 

  1038. #define VSAD_INTRA(size)                                                \

  1039. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  1040.                                     uint8_t *s, uint8_t *dummy,         \

  1041.                                     int stride, int h)                  \

  1042. {                                                                       \

  1043.     int score = 0, x, y;                                                \

  1044.                                                                         \

  1045.     for (y = 1; y < h; y++) {                                           \

  1046.         for (x = 0; x < size; x += 4) {                                 \

  1047.             score += FFABS(s[x]     - s[x + stride])     +              \

  1048.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  1049.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  1050.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  1051.         }                                                               \

  1052.         s += stride;                                                    \

  1053.     }                                                                   \

  1054.                                                                         \

  1055.     return score;                                                       \

  1056. }

  1057. VSAD_INTRA(8)

  1058. VSAD_INTRA(16)

  1059. 

  1060. static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1061.                     int stride, int h)

  1062. {

  1063.     int score = 0, x, y;

  1064. 

  1065.     for (y = 1; y < h; y++) {

  1066.         for (x = 0; x < 16; x++)

  1067.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1068.         s1 += stride;

  1069.         s2 += stride;

  1070.     }

  1071. 

  1072.     return score;

  1073. }

  1074. 

  1075. #define SQ(a) ((a) * (a))

  1076. #define VSSE_INTRA(size)                                                \

  1077. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  1078.                                     uint8_t *s, uint8_t *dummy,         \

  1079.                                     int stride, int h)                  \

  1080. {                                                                       \

  1081.     int score = 0, x, y;                                                \

  1082.                                                                         \

  1083.     for (y = 1; y < h; y++) {                                           \

  1084.         for (x = 0; x < size; x += 4) {                                 \

  1085.             score += SQ(s[x]     - s[x + stride]) +                     \

  1086.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  1087.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  1088.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  1089.         }                                                               \

  1090.         s += stride;                                                    \

  1091.     }                                                                   \

  1092.                                                                         \

  1093.     return score;                                                       \

  1094. }

  1095. VSSE_INTRA(8)

  1096. VSSE_INTRA(16)

  1097. 

  1098. static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1099.                     int stride, int h)

  1100. {

  1101.     int score = 0, x, y;

  1102. 

  1103.     for (y = 1; y < h; y++) {

  1104.         for (x = 0; x < 16; x++)

  1105.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1106.         s1 += stride;

  1107.         s2 += stride;

  1108.     }

  1109. 

  1110.     return score;

  1111. }

  1112. 

  1113. #define WRAPPER8_16_SQ(name8, name16)                                   \

  1114. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  1115.                   int stride, int h)                                    \

  1116. {                                                                       \

  1117.     int score = 0;                                                      \

  1118.                                                                         \

  1119.     score += name8(s, dst, src, stride, 8);                             \

  1120.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1121.     if (h == 16) {                                                      \

  1122.         dst   += 8 * stride;                                            \

  1123.         src   += 8 * stride;                                            \

  1124.         score += name8(s, dst, src, stride, 8);                         \

  1125.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1126.     }                                                                   \

  1127.     return score;                                                       \

  1128. }

  1129. 

  1130. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1131. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1132. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1133. #if CONFIG_GPL

  1134. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1135. #endif

  1136. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1137. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1138. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1139. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1140. 

  1141. static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)

  1142. {

  1143.     ff_j_rev_dct(block);

  1144.     put_pixels_clamped_c(block, dest, line_size);

  1145. }

  1146. 

  1147. static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)

  1148. {

  1149.     ff_j_rev_dct(block);

  1150.     add_pixels_clamped_c(block, dest, line_size);

  1151. }

  1152. 

  1153. /* draw the edges of width 'w' of an image of size width, height */

  1154. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1155. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1156.                            int w, int h, int sides)

  1157. {

  1158.     uint8_t *ptr = buf, *last_line;

  1159.     int i;

  1160. 

  1161.     /* left and right */

  1162.     for (i = 0; i < height; i++) {

  1163.         memset(ptr - w, ptr[0], w);

  1164.         memset(ptr + width, ptr[width - 1], w);

  1165.         ptr += wrap;

  1166.     }

  1167. 

  1168.     /* top and bottom + corners */

  1169.     buf -= w;

  1170.     last_line = buf + (height - 1) * wrap;

  1171.     if (sides & EDGE_TOP)

  1172.         for (i = 0; i < h; i++)

  1173.             // top

  1174.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1175.     if (sides & EDGE_BOTTOM)

  1176.         for (i = 0; i < h; i++)

  1177.             // bottom

  1178.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1179. }

  1180. 

  1181. /* init static data */

  1182. av_cold void ff_dsputil_static_init(void)

  1183. {

  1184.     int i;

  1185. 

  1186.     for (i = 0; i < 512; i++)

  1187.         ff_square_tab[i] = (i - 256) * (i - 256);

  1188. }

  1189. 

  1190. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1191. {

  1192.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1193. 

  1194. #if CONFIG_ENCODERS

  1195.     if (avctx->bits_per_raw_sample == 10) {

  1196.         c->fdct    = ff_jpeg_fdct_islow_10;

  1197.         c->fdct248 = ff_fdct248_islow_10;

  1198.     } else {

  1199.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1200.             c->fdct    = ff_fdct_ifast;

  1201.             c->fdct248 = ff_fdct_ifast248;

  1202.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1203.             c->fdct    = ff_faandct;

  1204.             c->fdct248 = ff_faandct248;

  1205.         } else {

  1206.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1207.             c->fdct248 = ff_fdct248_islow_8;

  1208.         }

  1209.     }

  1210. #endif /* CONFIG_ENCODERS */

  1211. 

  1212.     if (avctx->bits_per_raw_sample == 10) {

  1213.         c->idct_put              = ff_simple_idct_put_10;

  1214.         c->idct_add              = ff_simple_idct_add_10;

  1215.         c->idct                  = ff_simple_idct_10;

  1216.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1217.     } else {

  1218.         if (avctx->idct_algo == FF_IDCT_INT) {

  1219.             c->idct_put              = jref_idct_put;

  1220.             c->idct_add              = jref_idct_add;

  1221.             c->idct                  = ff_j_rev_dct;

  1222.             c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;

  1223.         } else if (avctx->idct_algo == FF_IDCT_FAAN) {

  1224.             c->idct_put              = ff_faanidct_put;

  1225.             c->idct_add              = ff_faanidct_add;

  1226.             c->idct                  = ff_faanidct;

  1227.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1228.         } else { // accurate/default

  1229.             c->idct_put              = ff_simple_idct_put_8;

  1230.             c->idct_add              = ff_simple_idct_add_8;

  1231.             c->idct                  = ff_simple_idct_8;

  1232.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1233.         }

  1234.     }

  1235. 

  1236.     c->diff_pixels = diff_pixels_c;

  1237. 

  1238.     c->put_pixels_clamped        = put_pixels_clamped_c;

  1239.     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;

  1240.     c->add_pixels_clamped        = add_pixels_clamped_c;

  1241. 

  1242.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1243. 

  1244.     c->pix_sum   = pix_sum_c;

  1245.     c->pix_norm1 = pix_norm1_c;

  1246. 

  1247.     /* TODO [0] 16  [1] 8 */

  1248.     c->pix_abs[0][0] = pix_abs16_c;

  1249.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1250.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1251.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1252.     c->pix_abs[1][0] = pix_abs8_c;

  1253.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1254.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1255.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1256. 

  1257. #define SET_CMP_FUNC(name)                      \

  1258.     c->name[0] = name ## 16_c;                  \

  1259.     c->name[1] = name ## 8x8_c;

  1260. 

  1261.     SET_CMP_FUNC(hadamard8_diff)

  1262.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1263.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1264.     SET_CMP_FUNC(dct_sad)

  1265.     SET_CMP_FUNC(dct_max)

  1266. #if CONFIG_GPL

  1267.     SET_CMP_FUNC(dct264_sad)

  1268. #endif

  1269.     c->sad[0] = pix_abs16_c;

  1270.     c->sad[1] = pix_abs8_c;

  1271.     c->sse[0] = sse16_c;

  1272.     c->sse[1] = sse8_c;

  1273.     c->sse[2] = sse4_c;

  1274.     SET_CMP_FUNC(quant_psnr)

  1275.     SET_CMP_FUNC(rd)

  1276.     SET_CMP_FUNC(bit)

  1277.     c->vsad[0] = vsad16_c;

  1278.     c->vsad[4] = vsad_intra16_c;

  1279.     c->vsad[5] = vsad_intra8_c;

  1280.     c->vsse[0] = vsse16_c;

  1281.     c->vsse[4] = vsse_intra16_c;

  1282.     c->vsse[5] = vsse_intra8_c;

  1283.     c->nsse[0] = nsse16_c;

  1284.     c->nsse[1] = nsse8_c;

  1285. 

  1286.     c->try_8x8basis = try_8x8basis_c;

  1287.     c->add_8x8basis = add_8x8basis_c;

  1288. 

  1289.     c->shrink[0] = av_image_copy_plane;

  1290.     c->shrink[1] = ff_shrink22;

  1291.     c->shrink[2] = ff_shrink44;

  1292.     c->shrink[3] = ff_shrink88;

  1293. 

  1294.     c->draw_edges = draw_edges_8_c;

  1295. 

  1296.     switch (avctx->bits_per_raw_sample) {

  1297.     case 9:

  1298.     case 10:

  1299.         c->get_pixels = get_pixels_16_c;

  1300.         break;

  1301.     default:

  1302.         c->get_pixels = get_pixels_8_c;

  1303.         break;

  1304.     }

  1305. 

  1306.     if (ARCH_ARM)

  1307.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1308.     if (ARCH_PPC)

  1309.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1310.     if (ARCH_X86)

  1311.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1312. 

  1313.     ff_init_scantable_permutation(c->idct_permutation,

  1314.                                   c->idct_permutation_type);

  1315. }