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.
40127 Commits
32 Branches
788MB
Tree: 182d3f8221
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '182d3f8221'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1333 lines
43KB
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. /* Input permutation for the simple_idct_mmx */

  52. static const uint8_t simple_mmx_permutation[64] = {

  53.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  54.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  55.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  56.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  57.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  58.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  59.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  60.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  61. };

  62. 

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

  64. 

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

  66.                                const uint8_t *src_scantable)

  67. {

  68.     int i, end;

  69. 

  70.     st->scantable = src_scantable;

  71. 

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

  73.         int j = src_scantable[i];

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

  75.     }

  76. 

  77.     end = -1;

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

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

  80.         if (j > end)

  81.             end = j;

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

  83.     }

  84. }

  85. 

  86. av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,

  87.                                            int idct_permutation_type)

  88. {

  89.     int i;

  90. 

  91.     switch (idct_permutation_type) {

  92.     case FF_NO_IDCT_PERM:

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

  94.             idct_permutation[i] = i;

  95.         break;

  96.     case FF_LIBMPEG2_IDCT_PERM:

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

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

  99.         break;

  100.     case FF_SIMPLE_IDCT_PERM:

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

  102.             idct_permutation[i] = simple_mmx_permutation[i];

  103.         break;

  104.     case FF_TRANSPOSE_IDCT_PERM:

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

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

  107.         break;

  108.     case FF_PARTTRANS_IDCT_PERM:

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

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

  111.         break;

  112.     case FF_SSE2_IDCT_PERM:

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

  114.             idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];

  115.         break;

  116.     default:

  117.         av_log(NULL, AV_LOG_ERROR,

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

  119.     }

  120. }

  121. 

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

  123. {

  124.     int s = 0, i, j;

  125. 

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

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

  128.             s   += pix[0];

  129.             s   += pix[1];

  130.             s   += pix[2];

  131.             s   += pix[3];

  132.             s   += pix[4];

  133.             s   += pix[5];

  134.             s   += pix[6];

  135.             s   += pix[7];

  136.             pix += 8;

  137.         }

  138.         pix += line_size - 16;

  139.     }

  140.     return s;

  141. }

  142. 

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

  144. {

  145.     int s = 0, i, j;

  146.     uint32_t *sq = ff_square_tab + 256;

  147. 

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

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

  150. #if 0

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

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

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

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

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

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

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

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

  159. #else

  160. #if HAVE_FAST_64BIT

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

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

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

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

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

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

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

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

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

  170. #else

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

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

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

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

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

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

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

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

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

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

  181. #endif

  182. #endif

  183.             pix += 8;

  184.         }

  185.         pix += line_size - 16;

  186.     }

  187.     return s;

  188. }

  189. 

  190. static int sse4_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.         pix1 += line_size;

  202.         pix2 += line_size;

  203.     }

  204.     return s;

  205. }

  206. 

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

  208.                   int line_size, int h)

  209. {

  210.     int s = 0, i;

  211.     uint32_t *sq = ff_square_tab + 256;

  212. 

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

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

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

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

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

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

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

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

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

  222.         pix1 += line_size;

  223.         pix2 += line_size;

  224.     }

  225.     return s;

  226. }

  227. 

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

  229.                    int line_size, int h)

  230. {

  231.     int s = 0, i;

  232.     uint32_t *sq = ff_square_tab + 256;

  233. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  251. 

  252.         pix1 += line_size;

  253.         pix2 += line_size;

  254.     }

  255.     return s;

  256. }

  257. 

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

  259.                           const uint8_t *s2, int stride)

  260. {

  261.     int i;

  262. 

  263.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  273.         s1      += stride;

  274.         s2      += stride;

  275.         block   += 8;

  276.     }

  277. }

  278. 

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

  280.                                  int line_size)

  281. {

  282.     int i;

  283. 

  284.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  294. 

  295.         pixels += line_size;

  296.         block  += 8;

  297.     }

  298. }

  299. 

  300. static void put_signed_pixels_clamped_c(const int16_t *block,

  301.                                         uint8_t *restrict pixels,

  302.                                         int line_size)

  303. {

  304.     int i, j;

  305. 

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

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

  308.             if (*block < -128)

  309.                 *pixels = 0;

  310.             else if (*block > 127)

  311.                 *pixels = 255;

  312.             else

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

  314.             block++;

  315.             pixels++;

  316.         }

  317.         pixels += (line_size - 8);

  318.     }

  319. }

  320. 

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

  322.                                  int line_size)

  323. {

  324.     int i;

  325. 

  326.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  336.         pixels   += line_size;

  337.         block    += 8;

  338.     }

  339. }

  340. 

  341. static int sum_abs_dctelem_c(int16_t *block)

  342. {

  343.     int sum = 0, i;

  344. 

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

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

  347.     return sum;

  348. }

  349. 

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

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

  352. 

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

  354.                               int line_size, int h)

  355. {

  356.     int s = 0, i;

  357. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  375.         pix1 += line_size;

  376.         pix2 += line_size;

  377.     }

  378.     return s;

  379. }

  380. 

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

  382.                           int line_size, int h)

  383. {

  384.     int s = 0, i;

  385. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  403.         pix1 += line_size;

  404.         pix2 += line_size;

  405.     }

  406.     return s;

  407. }

  408. 

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

  410.                           int line_size, int h)

  411. {

  412.     int s = 0, i;

  413.     uint8_t *pix3 = pix2 + line_size;

  414. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  432.         pix1 += line_size;

  433.         pix2 += line_size;

  434.         pix3 += line_size;

  435.     }

  436.     return s;

  437. }

  438. 

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

  440.                            int line_size, int h)

  441. {

  442.     int s = 0, i;

  443.     uint8_t *pix3 = pix2 + line_size;

  444. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  462.         pix1 += line_size;

  463.         pix2 += line_size;

  464.         pix3 += line_size;

  465.     }

  466.     return s;

  467. }

  468. 

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

  470.                              int line_size, int h)

  471. {

  472.     int s = 0, i;

  473. 

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

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

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

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

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

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

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

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

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

  483.         pix1 += line_size;

  484.         pix2 += line_size;

  485.     }

  486.     return s;

  487. }

  488. 

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

  490.                          int line_size, int h)

  491. {

  492.     int s = 0, i;

  493. 

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

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

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

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

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

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

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

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

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

  503.         pix1 += line_size;

  504.         pix2 += line_size;

  505.     }

  506.     return s;

  507. }

  508. 

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

  510.                          int line_size, int h)

  511. {

  512.     int s = 0, i;

  513.     uint8_t *pix3 = pix2 + line_size;

  514. 

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

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

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

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

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

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

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

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

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

  524.         pix1 += line_size;

  525.         pix2 += line_size;

  526.         pix3 += line_size;

  527.     }

  528.     return s;

  529. }

  530. 

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

  532.                           int line_size, int h)

  533. {

  534.     int s = 0, i;

  535.     uint8_t *pix3 = pix2 + line_size;

  536. 

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

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

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

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

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

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

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

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

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

  546.         pix1 += line_size;

  547.         pix2 += line_size;

  548.         pix3 += line_size;

  549.     }

  550.     return s;

  551. }

  552. 

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

  554. {

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

  556. 

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

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

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

  560.         if (y + 1 < h) {

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

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

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

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

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

  566.         }

  567.         s1 += stride;

  568.         s2 += stride;

  569.     }

  570. 

  571.     if (c)

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

  573.     else

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

  575. }

  576. 

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

  578. {

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

  580. 

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

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

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

  584.         if (y + 1 < h) {

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

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

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

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

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

  590.         }

  591.         s1 += stride;

  592.         s2 += stride;

  593.     }

  594. 

  595.     if (c)

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

  597.     else

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

  599. }

  600. 

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

  602.                           int16_t basis[64], int scale)

  603. {

  604.     int i;

  605.     unsigned int sum = 0;

  606. 

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

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

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

  610.                           (BASIS_SHIFT - RECON_SHIFT));

  611.         int w = weight[i];

  612.         b >>= RECON_SHIFT;

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

  614. 

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

  616.     }

  617.     return sum >> 2;

  618. }

  619. 

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

  621. {

  622.     int i;

  623. 

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

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

  626.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  627.                   (BASIS_SHIFT - RECON_SHIFT);

  628. }

  629. 

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

  631.                     int stride, int h)

  632. {

  633.     return 0;

  634. }

  635. 

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

  637. {

  638.     int i;

  639. 

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

  641. 

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

  643.         switch (type & 0xFF) {

  644.         case FF_CMP_SAD:

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

  646.             break;

  647.         case FF_CMP_SATD:

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

  649.             break;

  650.         case FF_CMP_SSE:

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

  652.             break;

  653.         case FF_CMP_DCT:

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

  655.             break;

  656.         case FF_CMP_DCT264:

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

  658.             break;

  659.         case FF_CMP_DCTMAX:

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

  661.             break;

  662.         case FF_CMP_PSNR:

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

  664.             break;

  665.         case FF_CMP_BIT:

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

  667.             break;

  668.         case FF_CMP_RD:

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

  670.             break;

  671.         case FF_CMP_VSAD:

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

  673.             break;

  674.         case FF_CMP_VSSE:

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

  676.             break;

  677.         case FF_CMP_ZERO:

  678.             cmp[i] = zero_cmp;

  679.             break;

  680.         case FF_CMP_NSSE:

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

  682.             break;

  683.         default:

  684.             av_log(NULL, AV_LOG_ERROR,

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

  686.         }

  687.     }

  688. }

  689. 

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

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

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

  693. 

  694. #define BUTTERFLY1(x, y)                        \

  695.     {                                           \

  696.         int a, b;                               \

  697.         a = x;                                  \

  698.         b = y;                                  \

  699.         x = a + b;                              \

  700.         y = a - b;                              \

  701.     }

  702. 

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

  704. 

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

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

  707. {

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

  709. 

  710.     assert(h == 8);

  711. 

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

  713.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  726. 

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

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

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

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

  731. 

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

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

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

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

  736.     }

  737. 

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

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

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

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

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

  743. 

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

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

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

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

  748. 

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

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

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

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

  753.     }

  754.     return sum;

  755. }

  756. 

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

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

  759. {

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

  761. 

  762.     assert(h == 8);

  763. 

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

  765.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  774. 

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

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

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

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

  779. 

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

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

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

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

  784.     }

  785. 

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

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

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

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

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

  791. 

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

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

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

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

  796. 

  797.         sum +=

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

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

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

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

  802.     }

  803. 

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

  805. 

  806.     return sum;

  807. }

  808. 

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

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

  811. {

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

  813. 

  814.     assert(h == 8);

  815. 

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

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

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

  819. }

  820. 

  821. #if CONFIG_GPL

  822. #define DCT8_1D                                         \

  823.     {                                                   \

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

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

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

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

  828.         const int a0  = s07 + s34;                      \

  829.         const int a1  = s16 + s25;                      \

  830.         const int a2  = s07 - s34;                      \

  831.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  848.     }

  849. 

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

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

  852. {

  853.     int16_t dct[8][8];

  854.     int i, sum = 0;

  855. 

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

  857. 

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

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

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

  861.         DCT8_1D

  862. #undef SRC

  863. #undef DST

  864. 

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

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

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

  868.             DCT8_1D

  869. #undef SRC

  870. #undef DST

  871.             return sum;

  872. }

  873. #endif

  874. 

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

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

  877. {

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

  879.     int sum = 0, i;

  880. 

  881.     assert(h == 8);

  882. 

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

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

  885. 

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

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

  888. 

  889.     return sum;

  890. }

  891. 

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

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

  894. {

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

  896.     int16_t *const bak = temp + 64;

  897.     int sum = 0, i;

  898. 

  899.     assert(h == 8);

  900.     s->mb_intra = 0;

  901. 

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

  903. 

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

  905. 

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

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

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

  909.     ff_simple_idct_8(temp); // FIXME

  910. 

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

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

  913. 

  914.     return sum;

  915. }

  916. 

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

  918.                    int stride, int h)

  919. {

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

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

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

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

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

  925.     const int esc_length = s->ac_esc_length;

  926.     uint8_t *length, *last_length;

  927. 

  928.     assert(h == 8);

  929. 

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

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

  932. 

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

  934. 

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

  936.     last                               =

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

  938. 

  939.     bits = 0;

  940. 

  941.     if (s->mb_intra) {

  942.         start_i     = 1;

  943.         length      = s->intra_ac_vlc_length;

  944.         last_length = s->intra_ac_vlc_last_length;

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

  946.     } else {

  947.         start_i     = 0;

  948.         length      = s->inter_ac_vlc_length;

  949.         last_length = s->inter_ac_vlc_last_length;

  950.     }

  951. 

  952.     if (last >= start_i) {

  953.         run = 0;

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

  955.             int j = scantable[i];

  956.             level = temp[j];

  957. 

  958.             if (level) {

  959.                 level += 64;

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

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

  962.                 else

  963.                     bits += esc_length;

  964.                 run = 0;

  965.             } else

  966.                 run++;

  967.         }

  968.         i = scantable[last];

  969. 

  970.         level = temp[i] + 64;

  971. 

  972.         assert(level - 64);

  973. 

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

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

  976.         } else

  977.             bits += esc_length;

  978.     }

  979. 

  980.     if (last >= 0) {

  981.         if (s->mb_intra)

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

  983.         else

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

  985.     }

  986. 

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

  988. 

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

  990. 

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

  992. }

  993. 

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

  995.                     int stride, int h)

  996. {

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

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

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

  1000.     const int esc_length = s->ac_esc_length;

  1001.     uint8_t *length, *last_length;

  1002. 

  1003.     assert(h == 8);

  1004. 

  1005.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1006. 

  1007.     s->block_last_index[0 /* FIXME */] =

  1008.     last                               =

  1009.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1010. 

  1011.     bits = 0;

  1012. 

  1013.     if (s->mb_intra) {

  1014.         start_i     = 1;

  1015.         length      = s->intra_ac_vlc_length;

  1016.         last_length = s->intra_ac_vlc_last_length;

  1017.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1018.     } else {

  1019.         start_i     = 0;

  1020.         length      = s->inter_ac_vlc_length;

  1021.         last_length = s->inter_ac_vlc_last_length;

  1022.     }

  1023. 

  1024.     if (last >= start_i) {

  1025.         run = 0;

  1026.         for (i = start_i; i < last; i++) {

  1027.             int j = scantable[i];

  1028.             level = temp[j];

  1029. 

  1030.             if (level) {

  1031.                 level += 64;

  1032.                 if ((level & (~127)) == 0)

  1033.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1034.                 else

  1035.                     bits += esc_length;

  1036.                 run = 0;

  1037.             } else

  1038.                 run++;

  1039.         }

  1040.         i = scantable[last];

  1041. 

  1042.         level = temp[i] + 64;

  1043. 

  1044.         assert(level - 64);

  1045. 

  1046.         if ((level & (~127)) == 0)

  1047.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1048.         else

  1049.             bits += esc_length;

  1050.     }

  1051. 

  1052.     return bits;

  1053. }

  1054. 

  1055. #define VSAD_INTRA(size)                                                \

  1056. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  1057.                                     uint8_t *s, uint8_t *dummy,         \

  1058.                                     int stride, int h)                  \

  1059. {                                                                       \

  1060.     int score = 0, x, y;                                                \

  1061.                                                                         \

  1062.     for (y = 1; y < h; y++) {                                           \

  1063.         for (x = 0; x < size; x += 4) {                                 \

  1064.             score += FFABS(s[x]     - s[x + stride])     +              \

  1065.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  1066.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  1067.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  1068.         }                                                               \

  1069.         s += stride;                                                    \

  1070.     }                                                                   \

  1071.                                                                         \

  1072.     return score;                                                       \

  1073. }

  1074. VSAD_INTRA(8)

  1075. VSAD_INTRA(16)

  1076. 

  1077. static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1078.                     int stride, int h)

  1079. {

  1080.     int score = 0, x, y;

  1081. 

  1082.     for (y = 1; y < h; y++) {

  1083.         for (x = 0; x < 16; x++)

  1084.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1085.         s1 += stride;

  1086.         s2 += stride;

  1087.     }

  1088. 

  1089.     return score;

  1090. }

  1091. 

  1092. #define SQ(a) ((a) * (a))

  1093. #define VSSE_INTRA(size)                                                \

  1094. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  1095.                                     uint8_t *s, uint8_t *dummy,         \

  1096.                                     int stride, int h)                  \

  1097. {                                                                       \

  1098.     int score = 0, x, y;                                                \

  1099.                                                                         \

  1100.     for (y = 1; y < h; y++) {                                           \

  1101.         for (x = 0; x < size; x += 4) {                                 \

  1102.             score += SQ(s[x]     - s[x + stride]) +                     \

  1103.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  1104.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  1105.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  1106.         }                                                               \

  1107.         s += stride;                                                    \

  1108.     }                                                                   \

  1109.                                                                         \

  1110.     return score;                                                       \

  1111. }

  1112. VSSE_INTRA(8)

  1113. VSSE_INTRA(16)

  1114. 

  1115. static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1116.                     int stride, int h)

  1117. {

  1118.     int score = 0, x, y;

  1119. 

  1120.     for (y = 1; y < h; y++) {

  1121.         for (x = 0; x < 16; x++)

  1122.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1123.         s1 += stride;

  1124.         s2 += stride;

  1125.     }

  1126. 

  1127.     return score;

  1128. }

  1129. 

  1130. #define WRAPPER8_16_SQ(name8, name16)                                   \

  1131. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  1132.                   int stride, int h)                                    \

  1133. {                                                                       \

  1134.     int score = 0;                                                      \

  1135.                                                                         \

  1136.     score += name8(s, dst, src, stride, 8);                             \

  1137.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1138.     if (h == 16) {                                                      \

  1139.         dst   += 8 * stride;                                            \

  1140.         src   += 8 * stride;                                            \

  1141.         score += name8(s, dst, src, stride, 8);                         \

  1142.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1143.     }                                                                   \

  1144.     return score;                                                       \

  1145. }

  1146. 

  1147. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1148. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1149. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1150. #if CONFIG_GPL

  1151. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1152. #endif

  1153. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1154. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1155. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1156. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1157. 

  1158. static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)

  1159. {

  1160.     ff_j_rev_dct(block);

  1161.     put_pixels_clamped_c(block, dest, line_size);

  1162. }

  1163. 

  1164. static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)

  1165. {

  1166.     ff_j_rev_dct(block);

  1167.     add_pixels_clamped_c(block, dest, line_size);

  1168. }

  1169. 

  1170. /* draw the edges of width 'w' of an image of size width, height */

  1171. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1172. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1173.                            int w, int h, int sides)

  1174. {

  1175.     uint8_t *ptr = buf, *last_line;

  1176.     int i;

  1177. 

  1178.     /* left and right */

  1179.     for (i = 0; i < height; i++) {

  1180.         memset(ptr - w, ptr[0], w);

  1181.         memset(ptr + width, ptr[width - 1], w);

  1182.         ptr += wrap;

  1183.     }

  1184. 

  1185.     /* top and bottom + corners */

  1186.     buf -= w;

  1187.     last_line = buf + (height - 1) * wrap;

  1188.     if (sides & EDGE_TOP)

  1189.         for (i = 0; i < h; i++)

  1190.             // top

  1191.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1192.     if (sides & EDGE_BOTTOM)

  1193.         for (i = 0; i < h; i++)

  1194.             // bottom

  1195.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1196. }

  1197. 

  1198. /* init static data */

  1199. av_cold void ff_dsputil_static_init(void)

  1200. {

  1201.     int i;

  1202. 

  1203.     for (i = 0; i < 512; i++)

  1204.         ff_square_tab[i] = (i - 256) * (i - 256);

  1205. }

  1206. 

  1207. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1208. {

  1209.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1210. 

  1211. #if CONFIG_ENCODERS

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

  1213.         c->fdct    = ff_jpeg_fdct_islow_10;

  1214.         c->fdct248 = ff_fdct248_islow_10;

  1215.     } else {

  1216.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1217.             c->fdct    = ff_fdct_ifast;

  1218.             c->fdct248 = ff_fdct_ifast248;

  1219.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1220.             c->fdct    = ff_faandct;

  1221.             c->fdct248 = ff_faandct248;

  1222.         } else {

  1223.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1224.             c->fdct248 = ff_fdct248_islow_8;

  1225.         }

  1226.     }

  1227. #endif /* CONFIG_ENCODERS */

  1228. 

  1229.     if (avctx->bits_per_raw_sample == 10) {

  1230.         c->idct_put              = ff_simple_idct_put_10;

  1231.         c->idct_add              = ff_simple_idct_add_10;

  1232.         c->idct                  = ff_simple_idct_10;

  1233.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1234.     } else {

  1235.         if (avctx->idct_algo == FF_IDCT_INT) {

  1236.             c->idct_put              = jref_idct_put;

  1237.             c->idct_add              = jref_idct_add;

  1238.             c->idct                  = ff_j_rev_dct;

  1239.             c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;

  1240.         } else if (avctx->idct_algo == FF_IDCT_FAAN) {

  1241.             c->idct_put              = ff_faanidct_put;

  1242.             c->idct_add              = ff_faanidct_add;

  1243.             c->idct                  = ff_faanidct;

  1244.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1245.         } else { // accurate/default

  1246.             c->idct_put              = ff_simple_idct_put_8;

  1247.             c->idct_add              = ff_simple_idct_add_8;

  1248.             c->idct                  = ff_simple_idct_8;

  1249.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1250.         }

  1251.     }

  1252. 

  1253.     c->diff_pixels = diff_pixels_c;

  1254. 

  1255.     c->put_pixels_clamped        = put_pixels_clamped_c;

  1256.     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;

  1257.     c->add_pixels_clamped        = add_pixels_clamped_c;

  1258. 

  1259.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1260. 

  1261.     c->pix_sum   = pix_sum_c;

  1262.     c->pix_norm1 = pix_norm1_c;

  1263. 

  1264.     /* TODO [0] 16  [1] 8 */

  1265.     c->pix_abs[0][0] = pix_abs16_c;

  1266.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1267.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1268.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1269.     c->pix_abs[1][0] = pix_abs8_c;

  1270.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1271.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1272.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1273. 

  1274. #define SET_CMP_FUNC(name)                      \

  1275.     c->name[0] = name ## 16_c;                  \

  1276.     c->name[1] = name ## 8x8_c;

  1277. 

  1278.     SET_CMP_FUNC(hadamard8_diff)

  1279.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1280.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1281.     SET_CMP_FUNC(dct_sad)

  1282.     SET_CMP_FUNC(dct_max)

  1283. #if CONFIG_GPL

  1284.     SET_CMP_FUNC(dct264_sad)

  1285. #endif

  1286.     c->sad[0] = pix_abs16_c;

  1287.     c->sad[1] = pix_abs8_c;

  1288.     c->sse[0] = sse16_c;

  1289.     c->sse[1] = sse8_c;

  1290.     c->sse[2] = sse4_c;

  1291.     SET_CMP_FUNC(quant_psnr)

  1292.     SET_CMP_FUNC(rd)

  1293.     SET_CMP_FUNC(bit)

  1294.     c->vsad[0] = vsad16_c;

  1295.     c->vsad[4] = vsad_intra16_c;

  1296.     c->vsad[5] = vsad_intra8_c;

  1297.     c->vsse[0] = vsse16_c;

  1298.     c->vsse[4] = vsse_intra16_c;

  1299.     c->vsse[5] = vsse_intra8_c;

  1300.     c->nsse[0] = nsse16_c;

  1301.     c->nsse[1] = nsse8_c;

  1302. 

  1303.     c->try_8x8basis = try_8x8basis_c;

  1304.     c->add_8x8basis = add_8x8basis_c;

  1305. 

  1306.     c->shrink[0] = av_image_copy_plane;

  1307.     c->shrink[1] = ff_shrink22;

  1308.     c->shrink[2] = ff_shrink44;

  1309.     c->shrink[3] = ff_shrink88;

  1310. 

  1311.     c->draw_edges = draw_edges_8_c;

  1312. 

  1313.     switch (avctx->bits_per_raw_sample) {

  1314.     case 9:

  1315.     case 10:

  1316.         c->get_pixels = get_pixels_16_c;

  1317.         break;

  1318.     default:

  1319.         c->get_pixels = get_pixels_8_c;

  1320.         break;

  1321.     }

  1322. 

  1323.     if (ARCH_ARM)

  1324.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1325.     if (ARCH_PPC)

  1326.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1327.     if (ARCH_X86)

  1328.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1329. 

  1330.     ff_init_scantable_permutation(c->idct_permutation,

  1331.                                   c->idct_permutation_type);

  1332. }