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.
71971 Commits
32 Branches
293MB
Tree: b28d5c49f7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b28d5c49f7'
${ noResults }
FFmpeg/libavcodec/vp9dsp_template.c

2518 lines
86KB
Raw Blame History 

  1. /*

  2.  * VP9 compatible video decoder

  3.  *

  4.  * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>

  5.  * Copyright (C) 2013 Clément Bœsch <u pkh me>

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

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

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

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

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

  13.  *

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

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

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

  17.  * Lesser General Public License for more details.

  18.  *

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

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

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

  22.  */

  23. 

  24. #include "libavutil/common.h"

  25. #include "bit_depth_template.c"

  26. #include "vp9dsp.h"

  27. 

  28. // FIXME see whether we can merge parts of this (perhaps at least 4x4 and 8x8)

  29. // back with h264pred.[ch]

  30. 

  31. static void vert_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  32.                        const uint8_t *left, const uint8_t *_top)

  33. {

  34.     pixel *dst = (pixel *) _dst;

  35.     const pixel *top = (const pixel *) _top;

  36.     pixel4 p4 = AV_RN4PA(top);

  37. 

  38.     stride /= sizeof(pixel);

  39.     AV_WN4PA(dst + stride * 0, p4);

  40.     AV_WN4PA(dst + stride * 1, p4);

  41.     AV_WN4PA(dst + stride * 2, p4);

  42.     AV_WN4PA(dst + stride * 3, p4);

  43. }

  44. 

  45. static void vert_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  46.                        const uint8_t *left, const uint8_t *_top)

  47. {

  48.     pixel *dst = (pixel *) _dst;

  49.     const pixel *top = (const pixel *) _top;

  50.     pixel4 p4a = AV_RN4PA(top + 0);

  51.     pixel4 p4b = AV_RN4PA(top + 4);

  52.     int y;

  53. 

  54.     stride /= sizeof(pixel);

  55.     for (y = 0; y < 8; y++) {

  56.         AV_WN4PA(dst + 0, p4a);

  57.         AV_WN4PA(dst + 4, p4b);

  58.         dst += stride;

  59.     }

  60. }

  61. 

  62. static void vert_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  63.                          const uint8_t *left, const uint8_t *_top)

  64. {

  65.     pixel *dst = (pixel *) _dst;

  66.     const pixel *top = (const pixel *) _top;

  67.     pixel4 p4a = AV_RN4PA(top +  0);

  68.     pixel4 p4b = AV_RN4PA(top +  4);

  69.     pixel4 p4c = AV_RN4PA(top +  8);

  70.     pixel4 p4d = AV_RN4PA(top + 12);

  71.     int y;

  72. 

  73.     stride /= sizeof(pixel);

  74.     for (y = 0; y < 16; y++) {

  75.         AV_WN4PA(dst +  0, p4a);

  76.         AV_WN4PA(dst +  4, p4b);

  77.         AV_WN4PA(dst +  8, p4c);

  78.         AV_WN4PA(dst + 12, p4d);

  79.         dst += stride;

  80.     }

  81. }

  82. 

  83. static void vert_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  84.                          const uint8_t *left, const uint8_t *_top)

  85. {

  86.     pixel *dst = (pixel *) _dst;

  87.     const pixel *top = (const pixel *) _top;

  88.     pixel4 p4a = AV_RN4PA(top +  0);

  89.     pixel4 p4b = AV_RN4PA(top +  4);

  90.     pixel4 p4c = AV_RN4PA(top +  8);

  91.     pixel4 p4d = AV_RN4PA(top + 12);

  92.     pixel4 p4e = AV_RN4PA(top + 16);

  93.     pixel4 p4f = AV_RN4PA(top + 20);

  94.     pixel4 p4g = AV_RN4PA(top + 24);

  95.     pixel4 p4h = AV_RN4PA(top + 28);

  96.     int y;

  97. 

  98.     stride /= sizeof(pixel);

  99.     for (y = 0; y < 32; y++) {

  100.         AV_WN4PA(dst +  0, p4a);

  101.         AV_WN4PA(dst +  4, p4b);

  102.         AV_WN4PA(dst +  8, p4c);

  103.         AV_WN4PA(dst + 12, p4d);

  104.         AV_WN4PA(dst + 16, p4e);

  105.         AV_WN4PA(dst + 20, p4f);

  106.         AV_WN4PA(dst + 24, p4g);

  107.         AV_WN4PA(dst + 28, p4h);

  108.         dst += stride;

  109.     }

  110. }

  111. 

  112. static void hor_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  113.                       const uint8_t *_left, const uint8_t *top)

  114. {

  115.     pixel *dst = (pixel *) _dst;

  116.     const pixel *left = (const pixel *) _left;

  117. 

  118.     stride /= sizeof(pixel);

  119.     AV_WN4PA(dst + stride * 0, PIXEL_SPLAT_X4(left[3]));

  120.     AV_WN4PA(dst + stride * 1, PIXEL_SPLAT_X4(left[2]));

  121.     AV_WN4PA(dst + stride * 2, PIXEL_SPLAT_X4(left[1]));

  122.     AV_WN4PA(dst + stride * 3, PIXEL_SPLAT_X4(left[0]));

  123. }

  124. 

  125. static void hor_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  126.                       const uint8_t *_left, const uint8_t *top)

  127. {

  128.     pixel *dst = (pixel *) _dst;

  129.     const pixel *left = (const pixel *) _left;

  130.     int y;

  131. 

  132.     stride /= sizeof(pixel);

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

  134.         pixel4 p4 = PIXEL_SPLAT_X4(left[7 - y]);

  135. 

  136.         AV_WN4PA(dst + 0, p4);

  137.         AV_WN4PA(dst + 4, p4);

  138.         dst += stride;

  139.     }

  140. }

  141. 

  142. static void hor_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  143.                         const uint8_t *_left, const uint8_t *top)

  144. {

  145.     pixel *dst = (pixel *) _dst;

  146.     const pixel *left = (const pixel *) _left;

  147.     int y;

  148. 

  149.     stride /= sizeof(pixel);

  150.     for (y = 0; y < 16; y++) {

  151.         pixel4 p4 = PIXEL_SPLAT_X4(left[15 - y]);

  152. 

  153.         AV_WN4PA(dst +  0, p4);

  154.         AV_WN4PA(dst +  4, p4);

  155.         AV_WN4PA(dst +  8, p4);

  156.         AV_WN4PA(dst + 12, p4);

  157.         dst += stride;

  158.     }

  159. }

  160. 

  161. static void hor_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  162.                         const uint8_t *_left, const uint8_t *top)

  163. {

  164.     pixel *dst = (pixel *) _dst;

  165.     const pixel *left = (const pixel *) _left;

  166.     int y;

  167. 

  168.     stride /= sizeof(pixel);

  169.     for (y = 0; y < 32; y++) {

  170.         pixel4 p4 = PIXEL_SPLAT_X4(left[31 - y]);

  171. 

  172.         AV_WN4PA(dst +  0, p4);

  173.         AV_WN4PA(dst +  4, p4);

  174.         AV_WN4PA(dst +  8, p4);

  175.         AV_WN4PA(dst + 12, p4);

  176.         AV_WN4PA(dst + 16, p4);

  177.         AV_WN4PA(dst + 20, p4);

  178.         AV_WN4PA(dst + 24, p4);

  179.         AV_WN4PA(dst + 28, p4);

  180.         dst += stride;

  181.     }

  182. }

  183. 

  184. static void tm_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  185.                      const uint8_t *_left, const uint8_t *_top)

  186. {

  187.     pixel *dst = (pixel *) _dst;

  188.     const pixel *left = (const pixel *) _left;

  189.     const pixel *top = (const pixel *) _top;

  190.     int y, tl = top[-1];

  191. 

  192.     stride /= sizeof(pixel);

  193.     for (y = 0; y < 4; y++) {

  194.         int l_m_tl = left[3 - y] - tl;

  195. 

  196.         dst[0] = av_clip_pixel(top[0] + l_m_tl);

  197.         dst[1] = av_clip_pixel(top[1] + l_m_tl);

  198.         dst[2] = av_clip_pixel(top[2] + l_m_tl);

  199.         dst[3] = av_clip_pixel(top[3] + l_m_tl);

  200.         dst += stride;

  201.     }

  202. }

  203. 

  204. static void tm_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  205.                      const uint8_t *_left, const uint8_t *_top)

  206. {

  207.     pixel *dst = (pixel *) _dst;

  208.     const pixel *left = (const pixel *) _left;

  209.     const pixel *top = (const pixel *) _top;

  210.     int y, tl = top[-1];

  211. 

  212.     stride /= sizeof(pixel);

  213.     for (y = 0; y < 8; y++) {

  214.         int l_m_tl = left[7 - y] - tl;

  215. 

  216.         dst[0] = av_clip_pixel(top[0] + l_m_tl);

  217.         dst[1] = av_clip_pixel(top[1] + l_m_tl);

  218.         dst[2] = av_clip_pixel(top[2] + l_m_tl);

  219.         dst[3] = av_clip_pixel(top[3] + l_m_tl);

  220.         dst[4] = av_clip_pixel(top[4] + l_m_tl);

  221.         dst[5] = av_clip_pixel(top[5] + l_m_tl);

  222.         dst[6] = av_clip_pixel(top[6] + l_m_tl);

  223.         dst[7] = av_clip_pixel(top[7] + l_m_tl);

  224.         dst += stride;

  225.     }

  226. }

  227. 

  228. static void tm_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  229.                        const uint8_t *_left, const uint8_t *_top)

  230. {

  231.     pixel *dst = (pixel *) _dst;

  232.     const pixel *left = (const pixel *) _left;

  233.     const pixel *top = (const pixel *) _top;

  234.     int y, tl = top[-1];

  235. 

  236.     stride /= sizeof(pixel);

  237.     for (y = 0; y < 16; y++) {

  238.         int l_m_tl = left[15 - y] - tl;

  239. 

  240.         dst[ 0] = av_clip_pixel(top[ 0] + l_m_tl);

  241.         dst[ 1] = av_clip_pixel(top[ 1] + l_m_tl);

  242.         dst[ 2] = av_clip_pixel(top[ 2] + l_m_tl);

  243.         dst[ 3] = av_clip_pixel(top[ 3] + l_m_tl);

  244.         dst[ 4] = av_clip_pixel(top[ 4] + l_m_tl);

  245.         dst[ 5] = av_clip_pixel(top[ 5] + l_m_tl);

  246.         dst[ 6] = av_clip_pixel(top[ 6] + l_m_tl);

  247.         dst[ 7] = av_clip_pixel(top[ 7] + l_m_tl);

  248.         dst[ 8] = av_clip_pixel(top[ 8] + l_m_tl);

  249.         dst[ 9] = av_clip_pixel(top[ 9] + l_m_tl);

  250.         dst[10] = av_clip_pixel(top[10] + l_m_tl);

  251.         dst[11] = av_clip_pixel(top[11] + l_m_tl);

  252.         dst[12] = av_clip_pixel(top[12] + l_m_tl);

  253.         dst[13] = av_clip_pixel(top[13] + l_m_tl);

  254.         dst[14] = av_clip_pixel(top[14] + l_m_tl);

  255.         dst[15] = av_clip_pixel(top[15] + l_m_tl);

  256.         dst += stride;

  257.     }

  258. }

  259. 

  260. static void tm_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  261.                        const uint8_t *_left, const uint8_t *_top)

  262. {

  263.     pixel *dst = (pixel *) _dst;

  264.     const pixel *left = (const pixel *) _left;

  265.     const pixel *top = (const pixel *) _top;

  266.     int y, tl = top[-1];

  267. 

  268.     stride /= sizeof(pixel);

  269.     for (y = 0; y < 32; y++) {

  270.         int l_m_tl = left[31 - y] - tl;

  271. 

  272.         dst[ 0] = av_clip_pixel(top[ 0] + l_m_tl);

  273.         dst[ 1] = av_clip_pixel(top[ 1] + l_m_tl);

  274.         dst[ 2] = av_clip_pixel(top[ 2] + l_m_tl);

  275.         dst[ 3] = av_clip_pixel(top[ 3] + l_m_tl);

  276.         dst[ 4] = av_clip_pixel(top[ 4] + l_m_tl);

  277.         dst[ 5] = av_clip_pixel(top[ 5] + l_m_tl);

  278.         dst[ 6] = av_clip_pixel(top[ 6] + l_m_tl);

  279.         dst[ 7] = av_clip_pixel(top[ 7] + l_m_tl);

  280.         dst[ 8] = av_clip_pixel(top[ 8] + l_m_tl);

  281.         dst[ 9] = av_clip_pixel(top[ 9] + l_m_tl);

  282.         dst[10] = av_clip_pixel(top[10] + l_m_tl);

  283.         dst[11] = av_clip_pixel(top[11] + l_m_tl);

  284.         dst[12] = av_clip_pixel(top[12] + l_m_tl);

  285.         dst[13] = av_clip_pixel(top[13] + l_m_tl);

  286.         dst[14] = av_clip_pixel(top[14] + l_m_tl);

  287.         dst[15] = av_clip_pixel(top[15] + l_m_tl);

  288.         dst[16] = av_clip_pixel(top[16] + l_m_tl);

  289.         dst[17] = av_clip_pixel(top[17] + l_m_tl);

  290.         dst[18] = av_clip_pixel(top[18] + l_m_tl);

  291.         dst[19] = av_clip_pixel(top[19] + l_m_tl);

  292.         dst[20] = av_clip_pixel(top[20] + l_m_tl);

  293.         dst[21] = av_clip_pixel(top[21] + l_m_tl);

  294.         dst[22] = av_clip_pixel(top[22] + l_m_tl);

  295.         dst[23] = av_clip_pixel(top[23] + l_m_tl);

  296.         dst[24] = av_clip_pixel(top[24] + l_m_tl);

  297.         dst[25] = av_clip_pixel(top[25] + l_m_tl);

  298.         dst[26] = av_clip_pixel(top[26] + l_m_tl);

  299.         dst[27] = av_clip_pixel(top[27] + l_m_tl);

  300.         dst[28] = av_clip_pixel(top[28] + l_m_tl);

  301.         dst[29] = av_clip_pixel(top[29] + l_m_tl);

  302.         dst[30] = av_clip_pixel(top[30] + l_m_tl);

  303.         dst[31] = av_clip_pixel(top[31] + l_m_tl);

  304.         dst += stride;

  305.     }

  306. }

  307. 

  308. static void dc_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  309.                      const uint8_t *_left, const uint8_t *_top)

  310. {

  311.     pixel *dst = (pixel *) _dst;

  312.     const pixel *left = (const pixel *) _left;

  313.     const pixel *top = (const pixel *) _top;

  314.     pixel4 dc = PIXEL_SPLAT_X4((left[0] + left[1] + left[2] + left[3] +

  315.                                 top[0] + top[1] + top[2] + top[3] + 4) >> 3);

  316. 

  317.     stride /= sizeof(pixel);

  318.     AV_WN4PA(dst + stride * 0, dc);

  319.     AV_WN4PA(dst + stride * 1, dc);

  320.     AV_WN4PA(dst + stride * 2, dc);

  321.     AV_WN4PA(dst + stride * 3, dc);

  322. }

  323. 

  324. static void dc_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  325.                      const uint8_t *_left, const uint8_t *_top)

  326. {

  327.     pixel *dst = (pixel *) _dst;

  328.     const pixel *left = (const pixel *) _left;

  329.     const pixel *top = (const pixel *) _top;

  330.     pixel4 dc = PIXEL_SPLAT_X4

  331.         ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +

  332.           left[6] + left[7] + top[0] + top[1] + top[2] + top[3] +

  333.           top[4] + top[5] + top[6] + top[7] + 8) >> 4);

  334.     int y;

  335. 

  336.     stride /= sizeof(pixel);

  337.     for (y = 0; y < 8; y++) {

  338.         AV_WN4PA(dst + 0, dc);

  339.         AV_WN4PA(dst + 4, dc);

  340.         dst += stride;

  341.     }

  342. }

  343. 

  344. static void dc_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  345.                        const uint8_t *_left, const uint8_t *_top)

  346. {

  347.     pixel *dst = (pixel *) _dst;

  348.     const pixel *left = (const pixel *) _left;

  349.     const pixel *top = (const pixel *) _top;

  350.     pixel4 dc = PIXEL_SPLAT_X4

  351.         ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] + left[6] +

  352.           left[7] + left[8] + left[9] + left[10] + left[11] + left[12] +

  353.           left[13] + left[14] + left[15] + top[0] + top[1] + top[2] + top[3] +

  354.           top[4] + top[5] + top[6] + top[7] + top[8] + top[9] + top[10] +

  355.           top[11] + top[12] + top[13] + top[14] + top[15] + 16) >> 5);

  356.     int y;

  357. 

  358.     stride /= sizeof(pixel);

  359.     for (y = 0; y < 16; y++) {

  360.         AV_WN4PA(dst +  0, dc);

  361.         AV_WN4PA(dst +  4, dc);

  362.         AV_WN4PA(dst +  8, dc);

  363.         AV_WN4PA(dst + 12, dc);

  364.         dst += stride;

  365.     }

  366. }

  367. 

  368. static void dc_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  369.                        const uint8_t *_left, const uint8_t *_top)

  370. {

  371.     pixel *dst = (pixel *) _dst;

  372.     const pixel *left = (const pixel *) _left;

  373.     const pixel *top = (const pixel *) _top;

  374.     pixel4 dc = PIXEL_SPLAT_X4

  375.         ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] + left[6] +

  376.           left[7] + left[8] + left[9] + left[10] + left[11] + left[12] +

  377.           left[13] + left[14] + left[15] + left[16] + left[17] + left[18] +

  378.           left[19] + left[20] + left[21] + left[22] + left[23] + left[24] +

  379.           left[25] + left[26] + left[27] + left[28] + left[29] + left[30] +

  380.           left[31] + top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +

  381.           top[6] + top[7] + top[8] + top[9] + top[10] + top[11] + top[12] +

  382.           top[13] + top[14] + top[15] + top[16] + top[17] + top[18] + top[19] +

  383.           top[20] + top[21] + top[22] + top[23] + top[24] + top[25] + top[26] +

  384.           top[27] + top[28] + top[29] + top[30] + top[31] + 32) >> 6);

  385.     int y;

  386. 

  387.     stride /= sizeof(pixel);

  388.     for (y = 0; y < 32; y++) {

  389.         AV_WN4PA(dst +  0, dc);

  390.         AV_WN4PA(dst +  4, dc);

  391.         AV_WN4PA(dst +  8, dc);

  392.         AV_WN4PA(dst + 12, dc);

  393.         AV_WN4PA(dst + 16, dc);

  394.         AV_WN4PA(dst + 20, dc);

  395.         AV_WN4PA(dst + 24, dc);

  396.         AV_WN4PA(dst + 28, dc);

  397.         dst += stride;

  398.     }

  399. }

  400. 

  401. static void dc_left_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  402.                           const uint8_t *_left, const uint8_t *top)

  403. {

  404.     pixel *dst = (pixel *) _dst;

  405.     const pixel *left = (const pixel *) _left;

  406.     pixel4 dc = PIXEL_SPLAT_X4((left[0] + left[1] + left[2] + left[3] + 2) >> 2);

  407. 

  408.     stride /= sizeof(pixel);

  409.     AV_WN4PA(dst + stride * 0, dc);

  410.     AV_WN4PA(dst + stride * 1, dc);

  411.     AV_WN4PA(dst + stride * 2, dc);

  412.     AV_WN4PA(dst + stride * 3, dc);

  413. }

  414. 

  415. static void dc_left_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  416.                           const uint8_t *_left, const uint8_t *top)

  417. {

  418.     pixel *dst = (pixel *) _dst;

  419.     const pixel *left = (const pixel *) _left;

  420.     pixel4 dc = PIXEL_SPLAT_X4

  421.         ((left[0] + left[1] + left[2] + left[3] +

  422.           left[4] + left[5] + left[6] + left[7] + 4) >> 3);

  423.     int y;

  424. 

  425.     stride /= sizeof(pixel);

  426.     for (y = 0; y < 8; y++) {

  427.         AV_WN4PA(dst + 0, dc);

  428.         AV_WN4PA(dst + 4, dc);

  429.         dst += stride;

  430.     }

  431. }

  432. 

  433. static void dc_left_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  434.                             const uint8_t *_left, const uint8_t *top)

  435. {

  436.     pixel *dst = (pixel *) _dst;

  437.     const pixel *left = (const pixel *) _left;

  438.     pixel4 dc = PIXEL_SPLAT_X4

  439.         ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +

  440.           left[6] + left[7] + left[8] + left[9] + left[10] + left[11] +

  441.           left[12] + left[13] + left[14] + left[15] + 8) >> 4);

  442.     int y;

  443. 

  444.     stride /= sizeof(pixel);

  445.     for (y = 0; y < 16; y++) {

  446.         AV_WN4PA(dst +  0, dc);

  447.         AV_WN4PA(dst +  4, dc);

  448.         AV_WN4PA(dst +  8, dc);

  449.         AV_WN4PA(dst + 12, dc);

  450.         dst += stride;

  451.     }

  452. }

  453. 

  454. static void dc_left_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  455.                             const uint8_t *_left, const uint8_t *top)

  456. {

  457.     pixel *dst = (pixel *) _dst;

  458.     const pixel *left = (const pixel *) _left;

  459.     pixel4 dc = PIXEL_SPLAT_X4

  460.         ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +

  461.           left[6] + left[7] + left[8] + left[9] + left[10] + left[11] +

  462.           left[12] + left[13] + left[14] + left[15] + left[16] + left[17] +

  463.           left[18] + left[19] + left[20] + left[21] + left[22] + left[23] +

  464.           left[24] + left[25] + left[26] + left[27] + left[28] + left[29] +

  465.           left[30] + left[31] + 16) >> 5);

  466.     int y;

  467. 

  468.     stride /= sizeof(pixel);

  469.     for (y = 0; y < 32; y++) {

  470.         AV_WN4PA(dst +  0, dc);

  471.         AV_WN4PA(dst +  4, dc);

  472.         AV_WN4PA(dst +  8, dc);

  473.         AV_WN4PA(dst + 12, dc);

  474.         AV_WN4PA(dst + 16, dc);

  475.         AV_WN4PA(dst + 20, dc);

  476.         AV_WN4PA(dst + 24, dc);

  477.         AV_WN4PA(dst + 28, dc);

  478.         dst += stride;

  479.     }

  480. }

  481. 

  482. static void dc_top_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  483.                          const uint8_t *left, const uint8_t *_top)

  484. {

  485.     pixel *dst = (pixel *) _dst;

  486.     const pixel *top = (const pixel *) _top;

  487.     pixel4 dc = PIXEL_SPLAT_X4((top[0] + top[1] + top[2] + top[3] + 2) >> 2);

  488. 

  489.     stride /= sizeof(pixel);

  490.     AV_WN4PA(dst + stride * 0, dc);

  491.     AV_WN4PA(dst + stride * 1, dc);

  492.     AV_WN4PA(dst + stride * 2, dc);

  493.     AV_WN4PA(dst + stride * 3, dc);

  494. }

  495. 

  496. static void dc_top_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  497.                          const uint8_t *left, const uint8_t *_top)

  498. {

  499.     pixel *dst = (pixel *) _dst;

  500.     const pixel *top = (const pixel *) _top;

  501.     pixel4 dc = PIXEL_SPLAT_X4

  502.         ((top[0] + top[1] + top[2] + top[3] +

  503.           top[4] + top[5] + top[6] + top[7] + 4) >> 3);

  504.     int y;

  505. 

  506.     stride /= sizeof(pixel);

  507.     for (y = 0; y < 8; y++) {

  508.         AV_WN4PA(dst + 0, dc);

  509.         AV_WN4PA(dst + 4, dc);

  510.         dst += stride;

  511.     }

  512. }

  513. 

  514. static void dc_top_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  515.                            const uint8_t *left, const uint8_t *_top)

  516. {

  517.     pixel *dst = (pixel *) _dst;

  518.     const pixel *top = (const pixel *) _top;

  519.     pixel4 dc = PIXEL_SPLAT_X4

  520.         ((top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +

  521.           top[6] + top[7] + top[8] + top[9] + top[10] + top[11] +

  522.           top[12] + top[13] + top[14] + top[15] + 8) >> 4);

  523.     int y;

  524. 

  525.     stride /= sizeof(pixel);

  526.     for (y = 0; y < 16; y++) {

  527.         AV_WN4PA(dst +  0, dc);

  528.         AV_WN4PA(dst +  4, dc);

  529.         AV_WN4PA(dst +  8, dc);

  530.         AV_WN4PA(dst + 12, dc);

  531.         dst += stride;

  532.     }

  533. }

  534. 

  535. static void dc_top_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  536.                            const uint8_t *left, const uint8_t *_top)

  537. {

  538.     pixel *dst = (pixel *) _dst;

  539.     const pixel *top = (const pixel *) _top;

  540.     pixel4 dc = PIXEL_SPLAT_X4

  541.         ((top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +

  542.           top[6] + top[7] + top[8] + top[9] + top[10] + top[11] +

  543.           top[12] + top[13] + top[14] + top[15] + top[16] + top[17] +

  544.           top[18] + top[19] + top[20] + top[21] + top[22] + top[23] +

  545.           top[24] + top[25] + top[26] + top[27] + top[28] + top[29] +

  546.           top[30] + top[31] + 16) >> 5);

  547.     int y;

  548. 

  549.     stride /= sizeof(pixel);

  550.     for (y = 0; y < 32; y++) {

  551.         AV_WN4PA(dst +  0, dc);

  552.         AV_WN4PA(dst +  4, dc);

  553.         AV_WN4PA(dst +  8, dc);

  554.         AV_WN4PA(dst + 12, dc);

  555.         AV_WN4PA(dst + 16, dc);

  556.         AV_WN4PA(dst + 20, dc);

  557.         AV_WN4PA(dst + 24, dc);

  558.         AV_WN4PA(dst + 28, dc);

  559.         dst += stride;

  560.     }

  561. }

  562. 

  563. static void dc_128_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  564.                          const uint8_t *left, const uint8_t *top)

  565. {

  566.     pixel *dst = (pixel *) _dst;

  567.     pixel4 val = PIXEL_SPLAT_X4(128 << (BIT_DEPTH - 8));

  568. 

  569.     stride /= sizeof(pixel);

  570.     AV_WN4PA(dst + stride * 0, val);

  571.     AV_WN4PA(dst + stride * 1, val);

  572.     AV_WN4PA(dst + stride * 2, val);

  573.     AV_WN4PA(dst + stride * 3, val);

  574. }

  575. 

  576. static void dc_128_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  577.                          const uint8_t *left, const uint8_t *top)

  578. {

  579.     pixel *dst = (pixel *) _dst;

  580.     pixel4 val = PIXEL_SPLAT_X4(128 << (BIT_DEPTH - 8));

  581.     int y;

  582. 

  583.     stride /= sizeof(pixel);

  584.     for (y = 0; y < 8; y++) {

  585.         AV_WN4PA(dst + 0, val);

  586.         AV_WN4PA(dst + 4, val);

  587.         dst += stride;

  588.     }

  589. }

  590. 

  591. static void dc_128_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  592.                            const uint8_t *left, const uint8_t *top)

  593. {

  594.     pixel *dst = (pixel *) _dst;

  595.     pixel4 val = PIXEL_SPLAT_X4(128 << (BIT_DEPTH - 8));

  596.     int y;

  597. 

  598.     stride /= sizeof(pixel);

  599.     for (y = 0; y < 16; y++) {

  600.         AV_WN4PA(dst +  0, val);

  601.         AV_WN4PA(dst +  4, val);

  602.         AV_WN4PA(dst +  8, val);

  603.         AV_WN4PA(dst + 12, val);

  604.         dst += stride;

  605.     }

  606. }

  607. 

  608. static void dc_128_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  609.                            const uint8_t *left, const uint8_t *top)

  610. {

  611.     pixel *dst = (pixel *) _dst;

  612.     pixel4 val = PIXEL_SPLAT_X4(128 << (BIT_DEPTH - 8));

  613.     int y;

  614. 

  615.     stride /= sizeof(pixel);

  616.     for (y = 0; y < 32; y++) {

  617.         AV_WN4PA(dst +  0, val);

  618.         AV_WN4PA(dst +  4, val);

  619.         AV_WN4PA(dst +  8, val);

  620.         AV_WN4PA(dst + 12, val);

  621.         AV_WN4PA(dst + 16, val);

  622.         AV_WN4PA(dst + 20, val);

  623.         AV_WN4PA(dst + 24, val);

  624.         AV_WN4PA(dst + 28, val);

  625.         dst += stride;

  626.     }

  627. }

  628. 

  629. static void dc_127_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  630.                          const uint8_t *left, const uint8_t *top)

  631. {

  632.     pixel *dst = (pixel *) _dst;

  633.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) - 1);

  634. 

  635.     stride /= sizeof(pixel);

  636.     AV_WN4PA(dst + stride * 0, val);

  637.     AV_WN4PA(dst + stride * 1, val);

  638.     AV_WN4PA(dst + stride * 2, val);

  639.     AV_WN4PA(dst + stride * 3, val);}

  640. 

  641. static void dc_127_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  642.                          const uint8_t *left, const uint8_t *top)

  643. {

  644.     pixel *dst = (pixel *) _dst;

  645.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) - 1);

  646.     int y;

  647. 

  648.     stride /= sizeof(pixel);

  649.     for (y = 0; y < 8; y++) {

  650.         AV_WN4PA(dst + 0, val);

  651.         AV_WN4PA(dst + 4, val);

  652.         dst += stride;

  653.     }

  654. }

  655. 

  656. static void dc_127_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  657.                            const uint8_t *left, const uint8_t *top)

  658. {

  659.     pixel *dst = (pixel *) _dst;

  660.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) - 1);

  661.     int y;

  662. 

  663.     stride /= sizeof(pixel);

  664.     for (y = 0; y < 16; y++) {

  665.         AV_WN4PA(dst +  0, val);

  666.         AV_WN4PA(dst +  4, val);

  667.         AV_WN4PA(dst +  8, val);

  668.         AV_WN4PA(dst + 12, val);

  669.         dst += stride;

  670.     }

  671. }

  672. 

  673. static void dc_127_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  674.                            const uint8_t *left, const uint8_t *top)

  675. {

  676.     pixel *dst = (pixel *) _dst;

  677.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) - 1);

  678.     int y;

  679. 

  680.     stride /= sizeof(pixel);

  681.     for (y = 0; y < 32; y++) {

  682.         AV_WN4PA(dst +  0, val);

  683.         AV_WN4PA(dst +  4, val);

  684.         AV_WN4PA(dst +  8, val);

  685.         AV_WN4PA(dst + 12, val);

  686.         AV_WN4PA(dst + 16, val);

  687.         AV_WN4PA(dst + 20, val);

  688.         AV_WN4PA(dst + 24, val);

  689.         AV_WN4PA(dst + 28, val);

  690.         dst += stride;

  691.     }

  692. }

  693. 

  694. static void dc_129_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  695.                          const uint8_t *left, const uint8_t *top)

  696. {

  697.     pixel *dst = (pixel *) _dst;

  698.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) + 1);

  699. 

  700.     stride /= sizeof(pixel);

  701.     AV_WN4PA(dst + stride * 0, val);

  702.     AV_WN4PA(dst + stride * 1, val);

  703.     AV_WN4PA(dst + stride * 2, val);

  704.     AV_WN4PA(dst + stride * 3, val);

  705. }

  706. 

  707. static void dc_129_8x8_c(uint8_t *_dst, ptrdiff_t stride,

  708.                          const uint8_t *left, const uint8_t *top)

  709. {

  710.     pixel *dst = (pixel *) _dst;

  711.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) + 1);

  712.     int y;

  713. 

  714.     stride /= sizeof(pixel);

  715.     for (y = 0; y < 8; y++) {

  716.         AV_WN4PA(dst + 0, val);

  717.         AV_WN4PA(dst + 4, val);

  718.         dst += stride;

  719.     }

  720. }

  721. 

  722. static void dc_129_16x16_c(uint8_t *_dst, ptrdiff_t stride,

  723.                            const uint8_t *left, const uint8_t *top)

  724. {

  725.     pixel *dst = (pixel *) _dst;

  726.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) + 1);

  727.     int y;

  728. 

  729.     stride /= sizeof(pixel);

  730.     for (y = 0; y < 16; y++) {

  731.         AV_WN4PA(dst +  0, val);

  732.         AV_WN4PA(dst +  4, val);

  733.         AV_WN4PA(dst +  8, val);

  734.         AV_WN4PA(dst + 12, val);

  735.         dst += stride;

  736.     }

  737. }

  738. 

  739. static void dc_129_32x32_c(uint8_t *_dst, ptrdiff_t stride,

  740.                            const uint8_t *left, const uint8_t *top)

  741. {

  742.     pixel *dst = (pixel *) _dst;

  743.     pixel4 val = PIXEL_SPLAT_X4((128 << (BIT_DEPTH - 8)) + 1);

  744.     int y;

  745. 

  746.     stride /= sizeof(pixel);

  747.     for (y = 0; y < 32; y++) {

  748.         AV_WN4PA(dst +  0, val);

  749.         AV_WN4PA(dst +  4, val);

  750.         AV_WN4PA(dst +  8, val);

  751.         AV_WN4PA(dst + 12, val);

  752.         AV_WN4PA(dst + 16, val);

  753.         AV_WN4PA(dst + 20, val);

  754.         AV_WN4PA(dst + 24, val);

  755.         AV_WN4PA(dst + 28, val);

  756.         dst += stride;

  757.     }

  758. }

  759. 

  760. #if BIT_DEPTH == 8

  761. #define memset_bpc memset

  762. #else

  763. static inline void memset_bpc(uint16_t *dst, int val, int len) {

  764.     int n;

  765.     for (n = 0; n < len; n++) {

  766.         dst[n] = val;

  767.     }

  768. }

  769. #endif

  770. 

  771. #define DST(x, y) dst[(x) + (y) * stride]

  772. 

  773. static void diag_downleft_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  774.                                 const uint8_t *left, const uint8_t *_top)

  775. {

  776.     pixel *dst = (pixel *) _dst;

  777.     const pixel *top = (const pixel *) _top;

  778.     int a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],

  779.         a4 = top[4], a5 = top[5], a6 = top[6], a7 = top[7];

  780. 

  781.     stride /= sizeof(pixel);

  782.     DST(0,0) = (a0 + a1 * 2 + a2 + 2) >> 2;

  783.     DST(1,0) = DST(0,1) = (a1 + a2 * 2 + a3 + 2) >> 2;

  784.     DST(2,0) = DST(1,1) = DST(0,2) = (a2 + a3 * 2 + a4 + 2) >> 2;

  785.     DST(3,0) = DST(2,1) = DST(1,2) = DST(0,3) = (a3 + a4 * 2 + a5 + 2) >> 2;

  786.     DST(3,1) = DST(2,2) = DST(1,3) = (a4 + a5 * 2 + a6 + 2) >> 2;

  787.     DST(3,2) = DST(2,3) = (a5 + a6 * 2 + a7 + 2) >> 2;

  788.     DST(3,3) = a7;  // note: this is different from vp8 and such

  789. }

  790. 

  791. #define def_diag_downleft(size) \

  792. static void diag_downleft_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  793.                                               const uint8_t *left, const uint8_t *_top) \

  794. { \

  795.     pixel *dst = (pixel *) _dst; \

  796.     const pixel *top = (const pixel *) _top; \

  797.     int i, j; \

  798.     pixel v[size - 1]; \

  799. \

  800.     stride /= sizeof(pixel); \

  801.     for (i = 0; i < size - 2; i++) \

  802.         v[i] = (top[i] + top[i + 1] * 2 + top[i + 2] + 2) >> 2; \

  803.     v[size - 2] = (top[size - 2] + top[size - 1] * 3 + 2) >> 2; \

  804. \

  805.     for (j = 0; j < size; j++) { \

  806.         memcpy(dst + j*stride, v + j, (size - 1 - j) * sizeof(pixel)); \

  807.         memset_bpc(dst + j*stride + size - 1 - j, top[size - 1], j + 1); \

  808.     } \

  809. }

  810. 

  811. def_diag_downleft(8)

  812. def_diag_downleft(16)

  813. def_diag_downleft(32)

  814. 

  815. static void diag_downright_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  816.                                  const uint8_t *_left, const uint8_t *_top)

  817. {

  818.     pixel *dst = (pixel *) _dst;

  819.     const pixel *top = (const pixel *) _top;

  820.     const pixel *left = (const pixel *) _left;

  821.     int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],

  822.         l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0];

  823. 

  824.     stride /= sizeof(pixel);

  825.     DST(0,3) = (l1 + l2 * 2 + l3 + 2) >> 2;

  826.     DST(0,2) = DST(1,3) = (l0 + l1 * 2 + l2 + 2) >> 2;

  827.     DST(0,1) = DST(1,2) = DST(2,3) = (tl + l0 * 2 + l1 + 2) >> 2;

  828.     DST(0,0) = DST(1,1) = DST(2,2) = DST(3,3) = (l0 + tl * 2 + a0 + 2) >> 2;

  829.     DST(1,0) = DST(2,1) = DST(3,2) = (tl + a0 * 2 + a1 + 2) >> 2;

  830.     DST(2,0) = DST(3,1) = (a0 + a1 * 2 + a2 + 2) >> 2;

  831.     DST(3,0) = (a1 + a2 * 2 + a3 + 2) >> 2;

  832. }

  833. 

  834. #define def_diag_downright(size) \

  835. static void diag_downright_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  836.                                                const uint8_t *_left, const uint8_t *_top) \

  837. { \

  838.     pixel *dst = (pixel *) _dst; \

  839.     const pixel *top = (const pixel *) _top; \

  840.     const pixel *left = (const pixel *) _left; \

  841.     int i, j; \

  842.     pixel v[size + size - 1]; \

  843. \

  844.     stride /= sizeof(pixel); \

  845.     for (i = 0; i < size - 2; i++) { \

  846.         v[i           ] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \

  847.         v[size + 1 + i] = (top[i]  + top[i + 1]  * 2 + top[i + 2]  + 2) >> 2; \

  848.     } \

  849.     v[size - 2] = (left[size - 2] + left[size - 1] * 2 + top[-1] + 2) >> 2; \

  850.     v[size - 1] = (left[size - 1] + top[-1] * 2 + top[ 0] + 2) >> 2; \

  851.     v[size    ] = (top[-1] + top[0]  * 2 + top[ 1] + 2) >> 2; \

  852. \

  853.     for (j = 0; j < size; j++) \

  854.         memcpy(dst + j*stride, v + size - 1 - j, size * sizeof(pixel)); \

  855. }

  856. 

  857. def_diag_downright(8)

  858. def_diag_downright(16)

  859. def_diag_downright(32)

  860. 

  861. static void vert_right_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  862.                              const uint8_t *_left, const uint8_t *_top)

  863. {

  864.     pixel *dst = (pixel *) _dst;

  865.     const pixel *top = (const pixel *) _top;

  866.     const pixel *left = (const pixel *) _left;

  867.     int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],

  868.         l0 = left[3], l1 = left[2], l2 = left[1];

  869. 

  870.     stride /= sizeof(pixel);

  871.     DST(0,3) = (l0 + l1 * 2 + l2 + 2) >> 2;

  872.     DST(0,2) = (tl + l0 * 2 + l1 + 2) >> 2;

  873.     DST(0,0) = DST(1,2) = (tl + a0 + 1) >> 1;

  874.     DST(0,1) = DST(1,3) = (l0 + tl * 2 + a0 + 2) >> 2;

  875.     DST(1,0) = DST(2,2) = (a0 + a1 + 1) >> 1;

  876.     DST(1,1) = DST(2,3) = (tl + a0 * 2 + a1 + 2) >> 2;

  877.     DST(2,0) = DST(3,2) = (a1 + a2 + 1) >> 1;

  878.     DST(2,1) = DST(3,3) = (a0 + a1 * 2 + a2 + 2) >> 2;

  879.     DST(3,0) = (a2 + a3 + 1) >> 1;

  880.     DST(3,1) = (a1 + a2 * 2 + a3 + 2) >> 2;

  881. }

  882. 

  883. #define def_vert_right(size) \

  884. static void vert_right_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  885.                                            const uint8_t *_left, const uint8_t *_top) \

  886. { \

  887.     pixel *dst = (pixel *) _dst; \

  888.     const pixel *top = (const pixel *) _top; \

  889.     const pixel *left = (const pixel *) _left; \

  890.     int i, j; \

  891.     pixel ve[size + size/2 - 1], vo[size + size/2 - 1]; \

  892. \

  893.     stride /= sizeof(pixel); \

  894.     for (i = 0; i < size/2 - 2; i++) { \

  895.         vo[i] = (left[i*2 + 3] + left[i*2 + 2] * 2 + left[i*2 + 1] + 2) >> 2; \

  896.         ve[i] = (left[i*2 + 4] + left[i*2 + 3] * 2 + left[i*2 + 2] + 2) >> 2; \

  897.     } \

  898.     vo[size/2 - 2] = (left[size - 1] + left[size - 2] * 2 + left[size - 3] + 2) >> 2; \

  899.     ve[size/2 - 2] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \

  900. \

  901.     ve[size/2 - 1] = (top[-1] + top[0] + 1) >> 1; \

  902.     vo[size/2 - 1] = (left[size - 1] + top[-1] * 2 + top[0] + 2) >> 2; \

  903.     for (i = 0; i < size - 1; i++) { \

  904.         ve[size/2 + i] = (top[i] + top[i + 1] + 1) >> 1; \

  905.         vo[size/2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \

  906.     } \

  907. \

  908.     for (j = 0; j < size / 2; j++) { \

  909.         memcpy(dst +  j*2     *stride, ve + size/2 - 1 - j, size * sizeof(pixel)); \

  910.         memcpy(dst + (j*2 + 1)*stride, vo + size/2 - 1 - j, size * sizeof(pixel)); \

  911.     } \

  912. }

  913. 

  914. def_vert_right(8)

  915. def_vert_right(16)

  916. def_vert_right(32)

  917. 

  918. static void hor_down_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  919.                            const uint8_t *_left, const uint8_t *_top)

  920. {

  921.     pixel *dst = (pixel *) _dst;

  922.     const pixel *top = (const pixel *) _top;

  923.     const pixel *left = (const pixel *) _left;

  924.     int l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0],

  925.         tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2];

  926. 

  927.     stride /= sizeof(pixel);

  928.     DST(2,0) = (tl + a0 * 2 + a1 + 2) >> 2;

  929.     DST(3,0) = (a0 + a1 * 2 + a2 + 2) >> 2;

  930.     DST(0,0) = DST(2,1) = (tl + l0 + 1) >> 1;

  931.     DST(1,0) = DST(3,1) = (a0 + tl * 2 + l0 + 2) >> 2;

  932.     DST(0,1) = DST(2,2) = (l0 + l1 + 1) >> 1;

  933.     DST(1,1) = DST(3,2) = (tl + l0 * 2 + l1 + 2) >> 2;

  934.     DST(0,2) = DST(2,3) = (l1 + l2 + 1) >> 1;

  935.     DST(1,2) = DST(3,3) = (l0 + l1 * 2 + l2 + 2) >> 2;

  936.     DST(0,3) = (l2 + l3 + 1) >> 1;

  937.     DST(1,3) = (l1 + l2 * 2 + l3 + 2) >> 2;

  938. }

  939. 

  940. #define def_hor_down(size) \

  941. static void hor_down_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  942.                                          const uint8_t *_left, const uint8_t *_top) \

  943. { \

  944.     pixel *dst = (pixel *) _dst; \

  945.     const pixel *top = (const pixel *) _top; \

  946.     const pixel *left = (const pixel *) _left; \

  947.     int i, j; \

  948.     pixel v[size * 3 - 2]; \

  949. \

  950.     stride /= sizeof(pixel); \

  951.     for (i = 0; i < size - 2; i++) { \

  952.         v[i*2       ] = (left[i + 1] + left[i + 0] + 1) >> 1; \

  953.         v[i*2    + 1] = (left[i + 2] + left[i + 1] * 2 + left[i + 0] + 2) >> 2; \

  954.         v[size*2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \

  955.     } \

  956.     v[size*2 - 2] = (top[-1] + left[size - 1] + 1) >> 1; \

  957.     v[size*2 - 4] = (left[size - 1] + left[size - 2] + 1) >> 1; \

  958.     v[size*2 - 1] = (top[0]  + top[-1] * 2 + left[size - 1] + 2) >> 2; \

  959.     v[size*2 - 3] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \

  960. \

  961.     for (j = 0; j < size; j++) \

  962.         memcpy(dst + j*stride, v + size*2 - 2 - j*2, size * sizeof(pixel)); \

  963. }

  964. 

  965. def_hor_down(8)

  966. def_hor_down(16)

  967. def_hor_down(32)

  968. 

  969. static void vert_left_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  970.                             const uint8_t *left, const uint8_t *_top)

  971. {

  972.     pixel *dst = (pixel *) _dst;

  973.     const pixel *top = (const pixel *) _top;

  974.     int a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],

  975.         a4 = top[4], a5 = top[5], a6 = top[6];

  976. 

  977.     stride /= sizeof(pixel);

  978.     DST(0,0) = (a0 + a1 + 1) >> 1;

  979.     DST(0,1) = (a0 + a1 * 2 + a2 + 2) >> 2;

  980.     DST(1,0) = DST(0,2) = (a1 + a2 + 1) >> 1;

  981.     DST(1,1) = DST(0,3) = (a1 + a2 * 2 + a3 + 2) >> 2;

  982.     DST(2,0) = DST(1,2) = (a2 + a3 + 1) >> 1;

  983.     DST(2,1) = DST(1,3) = (a2 + a3 * 2 + a4 + 2) >> 2;

  984.     DST(3,0) = DST(2,2) = (a3 + a4 + 1) >> 1;

  985.     DST(3,1) = DST(2,3) = (a3 + a4 * 2 + a5 + 2) >> 2;

  986.     DST(3,2) = (a4 + a5 + 1) >> 1;

  987.     DST(3,3) = (a4 + a5 * 2 + a6 + 2) >> 2;

  988. }

  989. 

  990. #define def_vert_left(size) \

  991. static void vert_left_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  992.                                           const uint8_t *left, const uint8_t *_top) \

  993. { \

  994.     pixel *dst = (pixel *) _dst; \

  995.     const pixel *top = (const pixel *) _top; \

  996.     int i, j; \

  997.     pixel ve[size - 1], vo[size - 1]; \

  998. \

  999.     stride /= sizeof(pixel); \

  1000.     for (i = 0; i < size - 2; i++) { \

  1001.         ve[i] = (top[i] + top[i + 1] + 1) >> 1; \

  1002.         vo[i] = (top[i] + top[i + 1] * 2 + top[i + 2] + 2) >> 2; \

  1003.     } \

  1004.     ve[size - 2] = (top[size - 2] + top[size - 1] + 1) >> 1; \

  1005.     vo[size - 2] = (top[size - 2] + top[size - 1] * 3 + 2) >> 2; \

  1006. \

  1007.     for (j = 0; j < size / 2; j++) { \

  1008.         memcpy(dst +  j*2      * stride, ve + j, (size - j - 1) * sizeof(pixel)); \

  1009.         memset_bpc(dst +  j*2      * stride + size - j - 1, top[size - 1], j + 1); \

  1010.         memcpy(dst + (j*2 + 1) * stride, vo + j, (size - j - 1) * sizeof(pixel)); \

  1011.         memset_bpc(dst + (j*2 + 1) * stride + size - j - 1, top[size - 1], j + 1); \

  1012.     } \

  1013. }

  1014. 

  1015. def_vert_left(8)

  1016. def_vert_left(16)

  1017. def_vert_left(32)

  1018. 

  1019. static void hor_up_4x4_c(uint8_t *_dst, ptrdiff_t stride,

  1020.                          const uint8_t *_left, const uint8_t *top)

  1021. {

  1022.     pixel *dst = (pixel *) _dst;

  1023.     const pixel *left = (const pixel *) _left;

  1024.     int l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3];

  1025. 

  1026.     stride /= sizeof(pixel);

  1027.     DST(0,0) = (l0 + l1 + 1) >> 1;

  1028.     DST(1,0) = (l0 + l1 * 2 + l2 + 2) >> 2;

  1029.     DST(0,1) = DST(2,0) = (l1 + l2 + 1) >> 1;

  1030.     DST(1,1) = DST(3,0) = (l1 + l2 * 2 + l3 + 2) >> 2;

  1031.     DST(0,2) = DST(2,1) = (l2 + l3 + 1) >> 1;

  1032.     DST(1,2) = DST(3,1) = (l2 + l3 * 3 + 2) >> 2;

  1033.     DST(0,3) = DST(1,3) = DST(2,2) = DST(2,3) = DST(3,2) = DST(3,3) = l3;

  1034. }

  1035. 

  1036. #define def_hor_up(size) \

  1037. static void hor_up_##size##x##size##_c(uint8_t *_dst, ptrdiff_t stride, \

  1038.                                        const uint8_t *_left, const uint8_t *top) \

  1039. { \

  1040.     pixel *dst = (pixel *) _dst; \

  1041.     const pixel *left = (const pixel *) _left; \

  1042.     int i, j; \

  1043.     pixel v[size*2 - 2]; \

  1044. \

  1045.     stride /= sizeof(pixel); \

  1046.     for (i = 0; i < size - 2; i++) { \

  1047.         v[i*2    ] = (left[i] + left[i + 1] + 1) >> 1; \

  1048.         v[i*2 + 1] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \

  1049.     } \

  1050.     v[size*2 - 4] = (left[size - 2] + left[size - 1] + 1) >> 1; \

  1051.     v[size*2 - 3] = (left[size - 2] + left[size - 1] * 3 + 2) >> 2; \

  1052. \

  1053.     for (j = 0; j < size / 2; j++) \

  1054.         memcpy(dst + j*stride, v + j*2, size * sizeof(pixel)); \

  1055.     for (j = size / 2; j < size; j++) { \

  1056.         memcpy(dst + j*stride, v + j*2, (size*2 - 2 - j*2) * sizeof(pixel)); \

  1057.         memset_bpc(dst + j*stride + size*2 - 2 - j*2, left[size - 1], \

  1058.                    2 + j*2 - size); \

  1059.     } \

  1060. }

  1061. 

  1062. def_hor_up(8)

  1063. def_hor_up(16)

  1064. def_hor_up(32)

  1065. 

  1066. #undef DST

  1067. 

  1068. static av_cold void vp9dsp_intrapred_init(VP9DSPContext *dsp)

  1069. {

  1070. #define init_intra_pred(tx, sz) \

  1071.     dsp->intra_pred[tx][VERT_PRED]            = vert_##sz##_c; \

  1072.     dsp->intra_pred[tx][HOR_PRED]             = hor_##sz##_c; \

  1073.     dsp->intra_pred[tx][DC_PRED]              = dc_##sz##_c; \

  1074.     dsp->intra_pred[tx][DIAG_DOWN_LEFT_PRED]  = diag_downleft_##sz##_c; \

  1075.     dsp->intra_pred[tx][DIAG_DOWN_RIGHT_PRED] = diag_downright_##sz##_c; \

  1076.     dsp->intra_pred[tx][VERT_RIGHT_PRED]      = vert_right_##sz##_c; \

  1077.     dsp->intra_pred[tx][HOR_DOWN_PRED]        = hor_down_##sz##_c; \

  1078.     dsp->intra_pred[tx][VERT_LEFT_PRED]       = vert_left_##sz##_c; \

  1079.     dsp->intra_pred[tx][HOR_UP_PRED]          = hor_up_##sz##_c; \

  1080.     dsp->intra_pred[tx][TM_VP8_PRED]          = tm_##sz##_c; \

  1081.     dsp->intra_pred[tx][LEFT_DC_PRED]         = dc_left_##sz##_c; \

  1082.     dsp->intra_pred[tx][TOP_DC_PRED]          = dc_top_##sz##_c; \

  1083.     dsp->intra_pred[tx][DC_128_PRED]          = dc_128_##sz##_c; \

  1084.     dsp->intra_pred[tx][DC_127_PRED]          = dc_127_##sz##_c; \

  1085.     dsp->intra_pred[tx][DC_129_PRED]          = dc_129_##sz##_c

  1086. 

  1087.     init_intra_pred(TX_4X4,   4x4);

  1088.     init_intra_pred(TX_8X8,   8x8);

  1089.     init_intra_pred(TX_16X16, 16x16);

  1090.     init_intra_pred(TX_32X32, 32x32);

  1091. 

  1092. #undef init_intra_pred

  1093. }

  1094. 

  1095. #define itxfm_wrapper(type_a, type_b, sz, bits, has_dconly) \

  1096. static void type_a##_##type_b##_##sz##x##sz##_add_c(uint8_t *_dst, \

  1097.                                                     ptrdiff_t stride, \

  1098.                                                     int16_t *_block, int eob) \

  1099. { \

  1100.     int i, j; \

  1101.     pixel *dst = (pixel *) _dst; \

  1102.     dctcoef *block = (dctcoef *) _block, tmp[sz * sz], out[sz]; \

  1103. \

  1104.     stride /= sizeof(pixel); \

  1105.     if (has_dconly && eob == 1) { \

  1106.         const int t  = (((block[0] * 11585 + (1 << 13)) >> 14) \

  1107.                                    * 11585 + (1 << 13)) >> 14; \

  1108.         block[0] = 0; \

  1109.         for (i = 0; i < sz; i++) { \

  1110.             for (j = 0; j < sz; j++) \

  1111.                 dst[j * stride] = av_clip_pixel(dst[j * stride] + \

  1112.                                                 (bits ? \

  1113.                                                  (t + (1 << (bits - 1))) >> bits : \

  1114.                                                  t)); \

  1115.             dst++; \

  1116.         } \

  1117.         return; \

  1118.     } \

  1119. \

  1120.     for (i = 0; i < sz; i++) \

  1121.         type_a##sz##_1d(block + i, sz, tmp + i * sz, 0); \

  1122.     memset(block, 0, sz * sz * sizeof(*block)); \

  1123.     for (i = 0; i < sz; i++) { \

  1124.         type_b##sz##_1d(tmp + i, sz, out, 1); \

  1125.         for (j = 0; j < sz; j++) \

  1126.             dst[j * stride] = av_clip_pixel(dst[j * stride] + \

  1127.                                             (bits ? \

  1128.                                              (out[j] + (1 << (bits - 1))) >> bits : \

  1129.                                              out[j])); \

  1130.         dst++; \

  1131.     } \

  1132. }

  1133. 

  1134. #define itxfm_wrap(sz, bits) \

  1135. itxfm_wrapper(idct,  idct,  sz, bits, 1) \

  1136. itxfm_wrapper(iadst, idct,  sz, bits, 0) \

  1137. itxfm_wrapper(idct,  iadst, sz, bits, 0) \

  1138. itxfm_wrapper(iadst, iadst, sz, bits, 0)

  1139. 

  1140. #define IN(x) ((dctint) in[(x) * stride])

  1141. 

  1142. static av_always_inline void idct4_1d(const dctcoef *in, ptrdiff_t stride,

  1143.                                       dctcoef *out, int pass)

  1144. {

  1145.     dctint t0, t1, t2, t3;

  1146. 

  1147.     t0 = ((IN(0) + IN(2)) * 11585 + (1 << 13)) >> 14;

  1148.     t1 = ((IN(0) - IN(2)) * 11585 + (1 << 13)) >> 14;

  1149.     t2 = (IN(1) *  6270 - IN(3) * 15137 + (1 << 13)) >> 14;

  1150.     t3 = (IN(1) * 15137 + IN(3) *  6270 + (1 << 13)) >> 14;

  1151. 

  1152.     out[0] = t0 + t3;

  1153.     out[1] = t1 + t2;

  1154.     out[2] = t1 - t2;

  1155.     out[3] = t0 - t3;

  1156. }

  1157. 

  1158. static av_always_inline void iadst4_1d(const dctcoef *in, ptrdiff_t stride,

  1159.                                        dctcoef *out, int pass)

  1160. {

  1161.     int t0, t1, t2, t3;

  1162. 

  1163.     t0 =  5283 * IN(0) + 15212 * IN(2) +  9929 * IN(3);

  1164.     t1 =  9929 * IN(0) -  5283 * IN(2) - 15212 * IN(3);

  1165.     t2 = 13377 * (IN(0) - IN(2) + IN(3));

  1166.     t3 = 13377 * IN(1);

  1167. 

  1168.     out[0] = (t0 + t3      + (1 << 13)) >> 14;

  1169.     out[1] = (t1 + t3      + (1 << 13)) >> 14;

  1170.     out[2] = (t2           + (1 << 13)) >> 14;

  1171.     out[3] = (t0 + t1 - t3 + (1 << 13)) >> 14;

  1172. }

  1173. 

  1174. itxfm_wrap(4, 4)

  1175. 

  1176. static av_always_inline void idct8_1d(const dctcoef *in, ptrdiff_t stride,

  1177.                                       dctcoef *out, int pass)

  1178. {

  1179.     dctint t0, t0a, t1, t1a, t2, t2a, t3, t3a, t4, t4a, t5, t5a, t6, t6a, t7, t7a;

  1180. 

  1181.     t0a = ((IN(0) + IN(4)) * 11585 + (1 << 13)) >> 14;

  1182.     t1a = ((IN(0) - IN(4)) * 11585 + (1 << 13)) >> 14;

  1183.     t2a = (IN(2) *  6270 - IN(6) * 15137 + (1 << 13)) >> 14;

  1184.     t3a = (IN(2) * 15137 + IN(6) *  6270 + (1 << 13)) >> 14;

  1185.     t4a = (IN(1) *  3196 - IN(7) * 16069 + (1 << 13)) >> 14;

  1186.     t5a = (IN(5) * 13623 - IN(3) *  9102 + (1 << 13)) >> 14;

  1187.     t6a = (IN(5) *  9102 + IN(3) * 13623 + (1 << 13)) >> 14;

  1188.     t7a = (IN(1) * 16069 + IN(7) *  3196 + (1 << 13)) >> 14;

  1189. 

  1190.     t0  = t0a + t3a;

  1191.     t1  = t1a + t2a;

  1192.     t2  = t1a - t2a;

  1193.     t3  = t0a - t3a;

  1194.     t4  = t4a + t5a;

  1195.     t5a = t4a - t5a;

  1196.     t7  = t7a + t6a;

  1197.     t6a = t7a - t6a;

  1198. 

  1199.     t5  = ((t6a - t5a) * 11585 + (1 << 13)) >> 14;

  1200.     t6  = ((t6a + t5a) * 11585 + (1 << 13)) >> 14;

  1201. 

  1202.     out[0] = t0 + t7;

  1203.     out[1] = t1 + t6;

  1204.     out[2] = t2 + t5;

  1205.     out[3] = t3 + t4;

  1206.     out[4] = t3 - t4;

  1207.     out[5] = t2 - t5;

  1208.     out[6] = t1 - t6;

  1209.     out[7] = t0 - t7;

  1210. }

  1211. 

  1212. static av_always_inline void iadst8_1d(const dctcoef *in, ptrdiff_t stride,

  1213.                                        dctcoef *out, int pass)

  1214. {

  1215.     dctint t0, t0a, t1, t1a, t2, t2a, t3, t3a, t4, t4a, t5, t5a, t6, t6a, t7, t7a;

  1216. 

  1217.     t0a = 16305 * IN(7) +  1606 * IN(0);

  1218.     t1a =  1606 * IN(7) - 16305 * IN(0);

  1219.     t2a = 14449 * IN(5) +  7723 * IN(2);

  1220.     t3a =  7723 * IN(5) - 14449 * IN(2);

  1221.     t4a = 10394 * IN(3) + 12665 * IN(4);

  1222.     t5a = 12665 * IN(3) - 10394 * IN(4);

  1223.     t6a =  4756 * IN(1) + 15679 * IN(6);

  1224.     t7a = 15679 * IN(1) -  4756 * IN(6);

  1225. 

  1226.     t0 = (t0a + t4a + (1 << 13)) >> 14;

  1227.     t1 = (t1a + t5a + (1 << 13)) >> 14;

  1228.     t2 = (t2a + t6a + (1 << 13)) >> 14;

  1229.     t3 = (t3a + t7a + (1 << 13)) >> 14;

  1230.     t4 = (t0a - t4a + (1 << 13)) >> 14;

  1231.     t5 = (t1a - t5a + (1 << 13)) >> 14;

  1232.     t6 = (t2a - t6a + (1 << 13)) >> 14;

  1233.     t7 = (t3a - t7a + (1 << 13)) >> 14;

  1234. 

  1235.     t4a = 15137 * t4 +  6270 * t5;

  1236.     t5a =  6270 * t4 - 15137 * t5;

  1237.     t6a = 15137 * t7 -  6270 * t6;

  1238.     t7a =  6270 * t7 + 15137 * t6;

  1239. 

  1240.     out[0] =   t0 + t2;

  1241.     out[7] = -(t1 + t3);

  1242.     t2     =   t0 - t2;

  1243.     t3     =   t1 - t3;

  1244. 

  1245.     out[1] = -((t4a + t6a + (1 << 13)) >> 14);

  1246.     out[6] =   (t5a + t7a + (1 << 13)) >> 14;

  1247.     t6     =   (t4a - t6a + (1 << 13)) >> 14;

  1248.     t7     =   (t5a - t7a + (1 << 13)) >> 14;

  1249. 

  1250.     out[3] = -(((t2 + t3) * 11585 + (1 << 13)) >> 14);

  1251.     out[4] =   ((t2 - t3) * 11585 + (1 << 13)) >> 14;

  1252.     out[2] =   ((t6 + t7) * 11585 + (1 << 13)) >> 14;

  1253.     out[5] = -(((t6 - t7) * 11585 + (1 << 13)) >> 14);

  1254. }

  1255. 

  1256. itxfm_wrap(8, 5)

  1257. 

  1258. static av_always_inline void idct16_1d(const dctcoef *in, ptrdiff_t stride,

  1259.                                        dctcoef *out, int pass)

  1260. {

  1261.     dctint t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12, t13, t14, t15;

  1262.     dctint t0a, t1a, t2a, t3a, t4a, t5a, t6a, t7a;

  1263.     dctint t8a, t9a, t10a, t11a, t12a, t13a, t14a, t15a;

  1264. 

  1265.     t0a  = ((IN(0) + IN(8)) * 11585 + (1 << 13)) >> 14;

  1266.     t1a  = ((IN(0) - IN(8)) * 11585 + (1 << 13)) >> 14;

  1267.     t2a  = (IN(4)  *  6270 - IN(12) * 15137 + (1 << 13)) >> 14;

  1268.     t3a  = (IN(4)  * 15137 + IN(12) *  6270 + (1 << 13)) >> 14;

  1269.     t4a  = (IN(2)  *  3196 - IN(14) * 16069 + (1 << 13)) >> 14;

  1270.     t7a  = (IN(2)  * 16069 + IN(14) *  3196 + (1 << 13)) >> 14;

  1271.     t5a  = (IN(10) * 13623 - IN(6)  *  9102 + (1 << 13)) >> 14;

  1272.     t6a  = (IN(10) *  9102 + IN(6)  * 13623 + (1 << 13)) >> 14;

  1273.     t8a  = (IN(1)  *  1606 - IN(15) * 16305 + (1 << 13)) >> 14;

  1274.     t15a = (IN(1)  * 16305 + IN(15) *  1606 + (1 << 13)) >> 14;

  1275.     t9a  = (IN(9)  * 12665 - IN(7)  * 10394 + (1 << 13)) >> 14;

  1276.     t14a = (IN(9)  * 10394 + IN(7)  * 12665 + (1 << 13)) >> 14;

  1277.     t10a = (IN(5)  *  7723 - IN(11) * 14449 + (1 << 13)) >> 14;

  1278.     t13a = (IN(5)  * 14449 + IN(11) *  7723 + (1 << 13)) >> 14;

  1279.     t11a = (IN(13) * 15679 - IN(3)  *  4756 + (1 << 13)) >> 14;

  1280.     t12a = (IN(13) *  4756 + IN(3)  * 15679 + (1 << 13)) >> 14;

  1281. 

  1282.     t0  = t0a  + t3a;

  1283.     t1  = t1a  + t2a;

  1284.     t2  = t1a  - t2a;

  1285.     t3  = t0a  - t3a;

  1286.     t4  = t4a  + t5a;

  1287.     t5  = t4a  - t5a;

  1288.     t6  = t7a  - t6a;

  1289.     t7  = t7a  + t6a;

  1290.     t8  = t8a  + t9a;

  1291.     t9  = t8a  - t9a;

  1292.     t10 = t11a - t10a;

  1293.     t11 = t11a + t10a;

  1294.     t12 = t12a + t13a;

  1295.     t13 = t12a - t13a;

  1296.     t14 = t15a - t14a;

  1297.     t15 = t15a + t14a;

  1298. 

  1299.     t5a  = ((t6 - t5) * 11585 + (1 << 13)) >> 14;

  1300.     t6a  = ((t6 + t5) * 11585 + (1 << 13)) >> 14;

  1301.     t9a  = (  t14 *  6270 - t9  * 15137  + (1 << 13)) >> 14;

  1302.     t14a = (  t14 * 15137 + t9  *  6270  + (1 << 13)) >> 14;

  1303.     t10a = (-(t13 * 15137 + t10 *  6270) + (1 << 13)) >> 14;

  1304.     t13a = (  t13 *  6270 - t10 * 15137  + (1 << 13)) >> 14;

  1305. 

  1306.     t0a  = t0   + t7;

  1307.     t1a  = t1   + t6a;

  1308.     t2a  = t2   + t5a;

  1309.     t3a  = t3   + t4;

  1310.     t4   = t3   - t4;

  1311.     t5   = t2   - t5a;

  1312.     t6   = t1   - t6a;

  1313.     t7   = t0   - t7;

  1314.     t8a  = t8   + t11;

  1315.     t9   = t9a  + t10a;

  1316.     t10  = t9a  - t10a;

  1317.     t11a = t8   - t11;

  1318.     t12a = t15  - t12;

  1319.     t13  = t14a - t13a;

  1320.     t14  = t14a + t13a;

  1321.     t15a = t15  + t12;

  1322. 

  1323.     t10a = ((t13  - t10)  * 11585 + (1 << 13)) >> 14;

  1324.     t13a = ((t13  + t10)  * 11585 + (1 << 13)) >> 14;

  1325.     t11  = ((t12a - t11a) * 11585 + (1 << 13)) >> 14;

  1326.     t12  = ((t12a + t11a) * 11585 + (1 << 13)) >> 14;

  1327. 

  1328.     out[ 0] = t0a + t15a;

  1329.     out[ 1] = t1a + t14;

  1330.     out[ 2] = t2a + t13a;

  1331.     out[ 3] = t3a + t12;

  1332.     out[ 4] = t4  + t11;

  1333.     out[ 5] = t5  + t10a;

  1334.     out[ 6] = t6  + t9;

  1335.     out[ 7] = t7  + t8a;

  1336.     out[ 8] = t7  - t8a;

  1337.     out[ 9] = t6  - t9;

  1338.     out[10] = t5  - t10a;

  1339.     out[11] = t4  - t11;

  1340.     out[12] = t3a - t12;

  1341.     out[13] = t2a - t13a;

  1342.     out[14] = t1a - t14;

  1343.     out[15] = t0a - t15a;

  1344. }

  1345. 

  1346. static av_always_inline void iadst16_1d(const dctcoef *in, ptrdiff_t stride,

  1347.                                         dctcoef *out, int pass)

  1348. {

  1349.     dctint t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12, t13, t14, t15;

  1350.     dctint t0a, t1a, t2a, t3a, t4a, t5a, t6a, t7a;

  1351.     dctint t8a, t9a, t10a, t11a, t12a, t13a, t14a, t15a;

  1352. 

  1353.     t0  = IN(15) * 16364 + IN(0)  *   804;

  1354.     t1  = IN(15) *   804 - IN(0)  * 16364;

  1355.     t2  = IN(13) * 15893 + IN(2)  *  3981;

  1356.     t3  = IN(13) *  3981 - IN(2)  * 15893;

  1357.     t4  = IN(11) * 14811 + IN(4)  *  7005;

  1358.     t5  = IN(11) *  7005 - IN(4)  * 14811;

  1359.     t6  = IN(9)  * 13160 + IN(6)  *  9760;

  1360.     t7  = IN(9)  *  9760 - IN(6)  * 13160;

  1361.     t8  = IN(7)  * 11003 + IN(8)  * 12140;

  1362.     t9  = IN(7)  * 12140 - IN(8)  * 11003;

  1363.     t10 = IN(5)  *  8423 + IN(10) * 14053;

  1364.     t11 = IN(5)  * 14053 - IN(10) *  8423;

  1365.     t12 = IN(3)  *  5520 + IN(12) * 15426;

  1366.     t13 = IN(3)  * 15426 - IN(12) *  5520;

  1367.     t14 = IN(1)  *  2404 + IN(14) * 16207;

  1368.     t15 = IN(1)  * 16207 - IN(14) *  2404;

  1369. 

  1370.     t0a  = (t0 + t8  + (1 << 13)) >> 14;

  1371.     t1a  = (t1 + t9  + (1 << 13)) >> 14;

  1372.     t2a  = (t2 + t10 + (1 << 13)) >> 14;

  1373.     t3a  = (t3 + t11 + (1 << 13)) >> 14;

  1374.     t4a  = (t4 + t12 + (1 << 13)) >> 14;

  1375.     t5a  = (t5 + t13 + (1 << 13)) >> 14;

  1376.     t6a  = (t6 + t14 + (1 << 13)) >> 14;

  1377.     t7a  = (t7 + t15 + (1 << 13)) >> 14;

  1378.     t8a  = (t0 - t8  + (1 << 13)) >> 14;

  1379.     t9a  = (t1 - t9  + (1 << 13)) >> 14;

  1380.     t10a = (t2 - t10 + (1 << 13)) >> 14;

  1381.     t11a = (t3 - t11 + (1 << 13)) >> 14;

  1382.     t12a = (t4 - t12 + (1 << 13)) >> 14;

  1383.     t13a = (t5 - t13 + (1 << 13)) >> 14;

  1384.     t14a = (t6 - t14 + (1 << 13)) >> 14;

  1385.     t15a = (t7 - t15 + (1 << 13)) >> 14;

  1386. 

  1387.     t8   = t8a  * 16069 + t9a  *  3196;

  1388.     t9   = t8a  *  3196 - t9a  * 16069;

  1389.     t10  = t10a *  9102 + t11a * 13623;

  1390.     t11  = t10a * 13623 - t11a *  9102;

  1391.     t12  = t13a * 16069 - t12a *  3196;

  1392.     t13  = t13a *  3196 + t12a * 16069;

  1393.     t14  = t15a *  9102 - t14a * 13623;

  1394.     t15  = t15a * 13623 + t14a *  9102;

  1395. 

  1396.     t0   = t0a + t4a;

  1397.     t1   = t1a + t5a;

  1398.     t2   = t2a + t6a;

  1399.     t3   = t3a + t7a;

  1400.     t4   = t0a - t4a;

  1401.     t5   = t1a - t5a;

  1402.     t6   = t2a - t6a;

  1403.     t7   = t3a - t7a;

  1404.     t8a  = (t8  + t12 + (1 << 13)) >> 14;

  1405.     t9a  = (t9  + t13 + (1 << 13)) >> 14;

  1406.     t10a = (t10 + t14 + (1 << 13)) >> 14;

  1407.     t11a = (t11 + t15 + (1 << 13)) >> 14;

  1408.     t12a = (t8  - t12 + (1 << 13)) >> 14;

  1409.     t13a = (t9  - t13 + (1 << 13)) >> 14;

  1410.     t14a = (t10 - t14 + (1 << 13)) >> 14;

  1411.     t15a = (t11 - t15 + (1 << 13)) >> 14;

  1412. 

  1413.     t4a  = t4 * 15137 + t5 *  6270;

  1414.     t5a  = t4 *  6270 - t5 * 15137;

  1415.     t6a  = t7 * 15137 - t6 *  6270;

  1416.     t7a  = t7 *  6270 + t6 * 15137;

  1417.     t12  = t12a * 15137 + t13a *  6270;

  1418.     t13  = t12a *  6270 - t13a * 15137;

  1419.     t14  = t15a * 15137 - t14a *  6270;

  1420.     t15  = t15a *  6270 + t14a * 15137;

  1421. 

  1422.     out[ 0] =   t0 + t2;

  1423.     out[15] = -(t1 + t3);

  1424.     t2a     =   t0 - t2;

  1425.     t3a     =   t1 - t3;

  1426.     out[ 3] = -((t4a + t6a + (1 << 13)) >> 14);

  1427.     out[12] =   (t5a + t7a + (1 << 13)) >> 14;

  1428.     t6      =   (t4a - t6a + (1 << 13)) >> 14;

  1429.     t7      =   (t5a - t7a + (1 << 13)) >> 14;

  1430.     out[ 1] = -(t8a + t10a);

  1431.     out[14] =   t9a + t11a;

  1432.     t10     =   t8a - t10a;

  1433.     t11     =   t9a - t11a;

  1434.     out[ 2] =   (t12 + t14 + (1 << 13)) >> 14;

  1435.     out[13] = -((t13 + t15 + (1 << 13)) >> 14);

  1436.     t14a    =   (t12 - t14 + (1 << 13)) >> 14;

  1437.     t15a    =   (t13 - t15 + (1 << 13)) >> 14;

  1438. 

  1439.     out[ 7] = ((t2a  + t3a)  * -11585 + (1 << 13)) >> 14;

  1440.     out[ 8] = ((t2a  - t3a)  *  11585 + (1 << 13)) >> 14;

  1441.     out[ 4] = ((t7   + t6)   *  11585 + (1 << 13)) >> 14;

  1442.     out[11] = ((t7   - t6)   *  11585 + (1 << 13)) >> 14;

  1443.     out[ 6] = ((t11  + t10)  *  11585 + (1 << 13)) >> 14;

  1444.     out[ 9] = ((t11  - t10)  *  11585 + (1 << 13)) >> 14;

  1445.     out[ 5] = ((t14a + t15a) * -11585 + (1 << 13)) >> 14;

  1446.     out[10] = ((t14a - t15a) *  11585 + (1 << 13)) >> 14;

  1447. }

  1448. 

  1449. itxfm_wrap(16, 6)

  1450. 

  1451. static av_always_inline void idct32_1d(const dctcoef *in, ptrdiff_t stride,

  1452.                                        dctcoef *out, int pass)

  1453. {

  1454.     dctint t0a  = ((IN(0) + IN(16)) * 11585 + (1 << 13)) >> 14;

  1455.     dctint t1a  = ((IN(0) - IN(16)) * 11585 + (1 << 13)) >> 14;

  1456.     dctint t2a  = (IN( 8) *  6270 - IN(24) * 15137 + (1 << 13)) >> 14;

  1457.     dctint t3a  = (IN( 8) * 15137 + IN(24) *  6270 + (1 << 13)) >> 14;

  1458.     dctint t4a  = (IN( 4) *  3196 - IN(28) * 16069 + (1 << 13)) >> 14;

  1459.     dctint t7a  = (IN( 4) * 16069 + IN(28) *  3196 + (1 << 13)) >> 14;

  1460.     dctint t5a  = (IN(20) * 13623 - IN(12) *  9102 + (1 << 13)) >> 14;

  1461.     dctint t6a  = (IN(20) *  9102 + IN(12) * 13623 + (1 << 13)) >> 14;

  1462.     dctint t8a  = (IN( 2) *  1606 - IN(30) * 16305 + (1 << 13)) >> 14;

  1463.     dctint t15a = (IN( 2) * 16305 + IN(30) *  1606 + (1 << 13)) >> 14;

  1464.     dctint t9a  = (IN(18) * 12665 - IN(14) * 10394 + (1 << 13)) >> 14;

  1465.     dctint t14a = (IN(18) * 10394 + IN(14) * 12665 + (1 << 13)) >> 14;

  1466.     dctint t10a = (IN(10) *  7723 - IN(22) * 14449 + (1 << 13)) >> 14;

  1467.     dctint t13a = (IN(10) * 14449 + IN(22) *  7723 + (1 << 13)) >> 14;

  1468.     dctint t11a = (IN(26) * 15679 - IN( 6) *  4756 + (1 << 13)) >> 14;

  1469.     dctint t12a = (IN(26) *  4756 + IN( 6) * 15679 + (1 << 13)) >> 14;

  1470.     dctint t16a = (IN( 1) *   804 - IN(31) * 16364 + (1 << 13)) >> 14;

  1471.     dctint t31a = (IN( 1) * 16364 + IN(31) *   804 + (1 << 13)) >> 14;

  1472.     dctint t17a = (IN(17) * 12140 - IN(15) * 11003 + (1 << 13)) >> 14;

  1473.     dctint t30a = (IN(17) * 11003 + IN(15) * 12140 + (1 << 13)) >> 14;

  1474.     dctint t18a = (IN( 9) *  7005 - IN(23) * 14811 + (1 << 13)) >> 14;

  1475.     dctint t29a = (IN( 9) * 14811 + IN(23) *  7005 + (1 << 13)) >> 14;

  1476.     dctint t19a = (IN(25) * 15426 - IN( 7) *  5520 + (1 << 13)) >> 14;

  1477.     dctint t28a = (IN(25) *  5520 + IN( 7) * 15426 + (1 << 13)) >> 14;

  1478.     dctint t20a = (IN( 5) *  3981 - IN(27) * 15893 + (1 << 13)) >> 14;

  1479.     dctint t27a = (IN( 5) * 15893 + IN(27) *  3981 + (1 << 13)) >> 14;

  1480.     dctint t21a = (IN(21) * 14053 - IN(11) *  8423 + (1 << 13)) >> 14;

  1481.     dctint t26a = (IN(21) *  8423 + IN(11) * 14053 + (1 << 13)) >> 14;

  1482.     dctint t22a = (IN(13) *  9760 - IN(19) * 13160 + (1 << 13)) >> 14;

  1483.     dctint t25a = (IN(13) * 13160 + IN(19) *  9760 + (1 << 13)) >> 14;

  1484.     dctint t23a = (IN(29) * 16207 - IN( 3) *  2404 + (1 << 13)) >> 14;

  1485.     dctint t24a = (IN(29) *  2404 + IN( 3) * 16207 + (1 << 13)) >> 14;

  1486. 

  1487.     dctint t0  = t0a  + t3a;

  1488.     dctint t1  = t1a  + t2a;

  1489.     dctint t2  = t1a  - t2a;

  1490.     dctint t3  = t0a  - t3a;

  1491.     dctint t4  = t4a  + t5a;

  1492.     dctint t5  = t4a  - t5a;

  1493.     dctint t6  = t7a  - t6a;

  1494.     dctint t7  = t7a  + t6a;

  1495.     dctint t8  = t8a  + t9a;

  1496.     dctint t9  = t8a  - t9a;

  1497.     dctint t10 = t11a - t10a;

  1498.     dctint t11 = t11a + t10a;

  1499.     dctint t12 = t12a + t13a;

  1500.     dctint t13 = t12a - t13a;

  1501.     dctint t14 = t15a - t14a;

  1502.     dctint t15 = t15a + t14a;

  1503.     dctint t16 = t16a + t17a;

  1504.     dctint t17 = t16a - t17a;

  1505.     dctint t18 = t19a - t18a;

  1506.     dctint t19 = t19a + t18a;

  1507.     dctint t20 = t20a + t21a;

  1508.     dctint t21 = t20a - t21a;

  1509.     dctint t22 = t23a - t22a;

  1510.     dctint t23 = t23a + t22a;

  1511.     dctint t24 = t24a + t25a;

  1512.     dctint t25 = t24a - t25a;

  1513.     dctint t26 = t27a - t26a;

  1514.     dctint t27 = t27a + t26a;

  1515.     dctint t28 = t28a + t29a;

  1516.     dctint t29 = t28a - t29a;

  1517.     dctint t30 = t31a - t30a;

  1518.     dctint t31 = t31a + t30a;

  1519. 

  1520.     t5a = ((t6 - t5) * 11585 + (1 << 13)) >> 14;

  1521.     t6a = ((t6 + t5) * 11585 + (1 << 13)) >> 14;

  1522.     t9a  = (  t14 *  6270 - t9  * 15137  + (1 << 13)) >> 14;

  1523.     t14a = (  t14 * 15137 + t9  *  6270  + (1 << 13)) >> 14;

  1524.     t10a = (-(t13 * 15137 + t10 *  6270) + (1 << 13)) >> 14;

  1525.     t13a = (  t13 *  6270 - t10 * 15137  + (1 << 13)) >> 14;

  1526.     t17a = (  t30 *  3196 - t17 * 16069  + (1 << 13)) >> 14;

  1527.     t30a = (  t30 * 16069 + t17 *  3196  + (1 << 13)) >> 14;

  1528.     t18a = (-(t29 * 16069 + t18 *  3196) + (1 << 13)) >> 14;

  1529.     t29a = (  t29 *  3196 - t18 * 16069  + (1 << 13)) >> 14;

  1530.     t21a = (  t26 * 13623 - t21 *  9102  + (1 << 13)) >> 14;

  1531.     t26a = (  t26 *  9102 + t21 * 13623  + (1 << 13)) >> 14;

  1532.     t22a = (-(t25 *  9102 + t22 * 13623) + (1 << 13)) >> 14;

  1533.     t25a = (  t25 * 13623 - t22 *  9102  + (1 << 13)) >> 14;

  1534. 

  1535.     t0a  = t0   + t7;

  1536.     t1a  = t1   + t6a;

  1537.     t2a  = t2   + t5a;

  1538.     t3a  = t3   + t4;

  1539.     t4a  = t3   - t4;

  1540.     t5   = t2   - t5a;

  1541.     t6   = t1   - t6a;

  1542.     t7a  = t0   - t7;

  1543.     t8a  = t8   + t11;

  1544.     t9   = t9a  + t10a;

  1545.     t10  = t9a  - t10a;

  1546.     t11a = t8   - t11;

  1547.     t12a = t15  - t12;

  1548.     t13  = t14a - t13a;

  1549.     t14  = t14a + t13a;

  1550.     t15a = t15  + t12;

  1551.     t16a = t16  + t19;

  1552.     t17  = t17a + t18a;

  1553.     t18  = t17a - t18a;

  1554.     t19a = t16  - t19;

  1555.     t20a = t23  - t20;

  1556.     t21  = t22a - t21a;

  1557.     t22  = t22a + t21a;

  1558.     t23a = t23  + t20;

  1559.     t24a = t24  + t27;

  1560.     t25  = t25a + t26a;

  1561.     t26  = t25a - t26a;

  1562.     t27a = t24  - t27;

  1563.     t28a = t31  - t28;

  1564.     t29  = t30a - t29a;

  1565.     t30  = t30a + t29a;

  1566.     t31a = t31  + t28;

  1567. 

  1568.     t10a = ((t13  - t10)  * 11585 + (1 << 13)) >> 14;

  1569.     t13a = ((t13  + t10)  * 11585 + (1 << 13)) >> 14;

  1570.     t11  = ((t12a - t11a) * 11585 + (1 << 13)) >> 14;

  1571.     t12  = ((t12a + t11a) * 11585 + (1 << 13)) >> 14;

  1572.     t18a = (  t29  *  6270 - t18  * 15137  + (1 << 13)) >> 14;

  1573.     t29a = (  t29  * 15137 + t18  *  6270  + (1 << 13)) >> 14;

  1574.     t19  = (  t28a *  6270 - t19a * 15137  + (1 << 13)) >> 14;

  1575.     t28  = (  t28a * 15137 + t19a *  6270  + (1 << 13)) >> 14;

  1576.     t20  = (-(t27a * 15137 + t20a *  6270) + (1 << 13)) >> 14;

  1577.     t27  = (  t27a *  6270 - t20a * 15137  + (1 << 13)) >> 14;

  1578.     t21a = (-(t26  * 15137 + t21  *  6270) + (1 << 13)) >> 14;

  1579.     t26a = (  t26  *  6270 - t21  * 15137  + (1 << 13)) >> 14;

  1580. 

  1581.     t0   = t0a + t15a;

  1582.     t1   = t1a + t14;

  1583.     t2   = t2a + t13a;

  1584.     t3   = t3a + t12;

  1585.     t4   = t4a + t11;

  1586.     t5a  = t5  + t10a;

  1587.     t6a  = t6  + t9;

  1588.     t7   = t7a + t8a;

  1589.     t8   = t7a - t8a;

  1590.     t9a  = t6  - t9;

  1591.     t10  = t5  - t10a;

  1592.     t11a = t4a - t11;

  1593.     t12a = t3a - t12;

  1594.     t13  = t2a - t13a;

  1595.     t14a = t1a - t14;

  1596.     t15  = t0a - t15a;

  1597.     t16  = t16a + t23a;

  1598.     t17a = t17  + t22;

  1599.     t18  = t18a + t21a;

  1600.     t19a = t19  + t20;

  1601.     t20a = t19  - t20;

  1602.     t21  = t18a - t21a;

  1603.     t22a = t17  - t22;

  1604.     t23  = t16a - t23a;

  1605.     t24  = t31a - t24a;

  1606.     t25a = t30  - t25;

  1607.     t26  = t29a - t26a;

  1608.     t27a = t28  - t27;

  1609.     t28a = t28  + t27;

  1610.     t29  = t29a + t26a;

  1611.     t30a = t30  + t25;

  1612.     t31  = t31a + t24a;

  1613. 

  1614.     t20  = ((t27a - t20a) * 11585 + (1 << 13)) >> 14;

  1615.     t27  = ((t27a + t20a) * 11585 + (1 << 13)) >> 14;

  1616.     t21a = ((t26  - t21 ) * 11585 + (1 << 13)) >> 14;

  1617.     t26a = ((t26  + t21 ) * 11585 + (1 << 13)) >> 14;

  1618.     t22  = ((t25a - t22a) * 11585 + (1 << 13)) >> 14;

  1619.     t25  = ((t25a + t22a) * 11585 + (1 << 13)) >> 14;

  1620.     t23a = ((t24  - t23 ) * 11585 + (1 << 13)) >> 14;

  1621.     t24a = ((t24  + t23 ) * 11585 + (1 << 13)) >> 14;

  1622. 

  1623.     out[ 0] = t0   + t31;

  1624.     out[ 1] = t1   + t30a;

  1625.     out[ 2] = t2   + t29;

  1626.     out[ 3] = t3   + t28a;

  1627.     out[ 4] = t4   + t27;

  1628.     out[ 5] = t5a  + t26a;

  1629.     out[ 6] = t6a  + t25;

  1630.     out[ 7] = t7   + t24a;

  1631.     out[ 8] = t8   + t23a;

  1632.     out[ 9] = t9a  + t22;

  1633.     out[10] = t10  + t21a;

  1634.     out[11] = t11a + t20;

  1635.     out[12] = t12a + t19a;

  1636.     out[13] = t13  + t18;

  1637.     out[14] = t14a + t17a;

  1638.     out[15] = t15  + t16;

  1639.     out[16] = t15  - t16;

  1640.     out[17] = t14a - t17a;

  1641.     out[18] = t13  - t18;

  1642.     out[19] = t12a - t19a;

  1643.     out[20] = t11a - t20;

  1644.     out[21] = t10  - t21a;

  1645.     out[22] = t9a  - t22;

  1646.     out[23] = t8   - t23a;

  1647.     out[24] = t7   - t24a;

  1648.     out[25] = t6a  - t25;

  1649.     out[26] = t5a  - t26a;

  1650.     out[27] = t4   - t27;

  1651.     out[28] = t3   - t28a;

  1652.     out[29] = t2   - t29;

  1653.     out[30] = t1   - t30a;

  1654.     out[31] = t0   - t31;

  1655. }

  1656. 

  1657. itxfm_wrapper(idct, idct, 32, 6, 1)

  1658. 

  1659. static av_always_inline void iwht4_1d(const dctcoef *in, ptrdiff_t stride,

  1660.                                       dctcoef *out, int pass)

  1661. {

  1662.     int t0, t1, t2, t3, t4;

  1663. 

  1664.     if (pass == 0) {

  1665.         t0 = IN(0) >> 2;

  1666.         t1 = IN(3) >> 2;

  1667.         t2 = IN(1) >> 2;

  1668.         t3 = IN(2) >> 2;

  1669.     } else {

  1670.         t0 = IN(0);

  1671.         t1 = IN(3);

  1672.         t2 = IN(1);

  1673.         t3 = IN(2);

  1674.     }

  1675. 

  1676.     t0 += t2;

  1677.     t3 -= t1;

  1678.     t4 = (t0 - t3) >> 1;

  1679.     t1 = t4 - t1;

  1680.     t2 = t4 - t2;

  1681.     t0 -= t1;

  1682.     t3 += t2;

  1683. 

  1684.     out[0] = t0;

  1685.     out[1] = t1;

  1686.     out[2] = t2;

  1687.     out[3] = t3;

  1688. }

  1689. 

  1690. itxfm_wrapper(iwht, iwht, 4, 0, 0)

  1691. 

  1692. #undef IN

  1693. #undef itxfm_wrapper

  1694. #undef itxfm_wrap

  1695. 

  1696. static av_cold void vp9dsp_itxfm_init(VP9DSPContext *dsp)

  1697. {

  1698. #define init_itxfm(tx, sz) \

  1699.     dsp->itxfm_add[tx][DCT_DCT]   = idct_idct_##sz##_add_c; \

  1700.     dsp->itxfm_add[tx][DCT_ADST]  = iadst_idct_##sz##_add_c; \

  1701.     dsp->itxfm_add[tx][ADST_DCT]  = idct_iadst_##sz##_add_c; \

  1702.     dsp->itxfm_add[tx][ADST_ADST] = iadst_iadst_##sz##_add_c

  1703. 

  1704. #define init_idct(tx, nm) \

  1705.     dsp->itxfm_add[tx][DCT_DCT]   = \

  1706.     dsp->itxfm_add[tx][ADST_DCT]  = \

  1707.     dsp->itxfm_add[tx][DCT_ADST]  = \

  1708.     dsp->itxfm_add[tx][ADST_ADST] = nm##_add_c

  1709. 

  1710.     init_itxfm(TX_4X4,   4x4);

  1711.     init_itxfm(TX_8X8,   8x8);

  1712.     init_itxfm(TX_16X16, 16x16);

  1713.     init_idct(TX_32X32,  idct_idct_32x32);

  1714.     init_idct(4 /* lossless */, iwht_iwht_4x4);

  1715. 

  1716. #undef init_itxfm

  1717. #undef init_idct

  1718. }

  1719. 

  1720. static av_always_inline void loop_filter(pixel *dst, int E, int I, int H,

  1721.                                          ptrdiff_t stridea, ptrdiff_t strideb,

  1722.                                          int wd)

  1723. {

  1724.     int i, F = 1 << (BIT_DEPTH - 8);

  1725. 

  1726.     E <<= (BIT_DEPTH - 8);

  1727.     I <<= (BIT_DEPTH - 8);

  1728.     H <<= (BIT_DEPTH - 8);

  1729.     for (i = 0; i < 8; i++, dst += stridea) {

  1730.         int p7, p6, p5, p4;

  1731.         int p3 = dst[strideb * -4], p2 = dst[strideb * -3];

  1732.         int p1 = dst[strideb * -2], p0 = dst[strideb * -1];

  1733.         int q0 = dst[strideb * +0], q1 = dst[strideb * +1];

  1734.         int q2 = dst[strideb * +2], q3 = dst[strideb * +3];

  1735.         int q4, q5, q6, q7;

  1736.         int fm = FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I &&

  1737.                  FFABS(p1 - p0) <= I && FFABS(q1 - q0) <= I &&

  1738.                  FFABS(q2 - q1) <= I && FFABS(q3 - q2) <= I &&

  1739.                  FFABS(p0 - q0) * 2 + (FFABS(p1 - q1) >> 1) <= E;

  1740.         int flat8out, flat8in;

  1741. 

  1742.         if (!fm)

  1743.             continue;

  1744. 

  1745.         if (wd >= 16) {

  1746.             p7 = dst[strideb * -8];

  1747.             p6 = dst[strideb * -7];

  1748.             p5 = dst[strideb * -6];

  1749.             p4 = dst[strideb * -5];

  1750.             q4 = dst[strideb * +4];

  1751.             q5 = dst[strideb * +5];

  1752.             q6 = dst[strideb * +6];

  1753.             q7 = dst[strideb * +7];

  1754. 

  1755.             flat8out = FFABS(p7 - p0) <= F && FFABS(p6 - p0) <= F &&

  1756.                        FFABS(p5 - p0) <= F && FFABS(p4 - p0) <= F &&

  1757.                        FFABS(q4 - q0) <= F && FFABS(q5 - q0) <= F &&

  1758.                        FFABS(q6 - q0) <= F && FFABS(q7 - q0) <= F;

  1759.         }

  1760. 

  1761.         if (wd >= 8)

  1762.             flat8in = FFABS(p3 - p0) <= F && FFABS(p2 - p0) <= F &&

  1763.                       FFABS(p1 - p0) <= F && FFABS(q1 - q0) <= F &&

  1764.                       FFABS(q2 - q0) <= F && FFABS(q3 - q0) <= F;

  1765. 

  1766.         if (wd >= 16 && flat8out && flat8in) {

  1767.             dst[strideb * -7] = (p7 + p7 + p7 + p7 + p7 + p7 + p7 + p6 * 2 +

  1768.                                  p5 + p4 + p3 + p2 + p1 + p0 + q0 + 8) >> 4;

  1769.             dst[strideb * -6] = (p7 + p7 + p7 + p7 + p7 + p7 + p6 + p5 * 2 +

  1770.                                  p4 + p3 + p2 + p1 + p0 + q0 + q1 + 8) >> 4;

  1771.             dst[strideb * -5] = (p7 + p7 + p7 + p7 + p7 + p6 + p5 + p4 * 2 +

  1772.                                  p3 + p2 + p1 + p0 + q0 + q1 + q2 + 8) >> 4;

  1773.             dst[strideb * -4] = (p7 + p7 + p7 + p7 + p6 + p5 + p4 + p3 * 2 +

  1774.                                  p2 + p1 + p0 + q0 + q1 + q2 + q3 + 8) >> 4;

  1775.             dst[strideb * -3] = (p7 + p7 + p7 + p6 + p5 + p4 + p3 + p2 * 2 +

  1776.                                  p1 + p0 + q0 + q1 + q2 + q3 + q4 + 8) >> 4;

  1777.             dst[strideb * -2] = (p7 + p7 + p6 + p5 + p4 + p3 + p2 + p1 * 2 +

  1778.                                  p0 + q0 + q1 + q2 + q3 + q4 + q5 + 8) >> 4;

  1779.             dst[strideb * -1] = (p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 +

  1780.                                  q0 + q1 + q2 + q3 + q4 + q5 + q6 + 8) >> 4;

  1781.             dst[strideb * +0] = (p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 +

  1782.                                  q1 + q2 + q3 + q4 + q5 + q6 + q7 + 8) >> 4;

  1783.             dst[strideb * +1] = (p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 +

  1784.                                  q2 + q3 + q4 + q5 + q6 + q7 + q7 + 8) >> 4;

  1785.             dst[strideb * +2] = (p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 +

  1786.                                  q3 + q4 + q5 + q6 + q7 + q7 + q7 + 8) >> 4;

  1787.             dst[strideb * +3] = (p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 +

  1788.                                  q4 + q5 + q6 + q7 + q7 + q7 + q7 + 8) >> 4;

  1789.             dst[strideb * +4] = (p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 +

  1790.                                  q5 + q6 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;

  1791.             dst[strideb * +5] = (p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 +

  1792.                                  q6 + q7 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;

  1793.             dst[strideb * +6] = (p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 +

  1794.                                  q7 + q7 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;

  1795.         } else if (wd >= 8 && flat8in) {

  1796.             dst[strideb * -3] = (p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0 + 4) >> 3;

  1797.             dst[strideb * -2] = (p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1 + 4) >> 3;

  1798.             dst[strideb * -1] = (p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2 + 4) >> 3;

  1799.             dst[strideb * +0] = (p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3 + 4) >> 3;

  1800.             dst[strideb * +1] = (p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3 + 4) >> 3;

  1801.             dst[strideb * +2] = (p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3 + 4) >> 3;

  1802.         } else {

  1803.             int hev = FFABS(p1 - p0) > H || FFABS(q1 - q0) > H;

  1804. 

  1805.             if (hev) {

  1806.                 int f = av_clip_intp2(p1 - q1, BIT_DEPTH - 1), f1, f2;

  1807.                 f = av_clip_intp2(3 * (q0 - p0) + f, BIT_DEPTH - 1);

  1808. 

  1809.                 f1 = FFMIN(f + 4, (1 << (BIT_DEPTH - 1)) - 1) >> 3;

  1810.                 f2 = FFMIN(f + 3, (1 << (BIT_DEPTH - 1)) - 1) >> 3;

  1811. 

  1812.                 dst[strideb * -1] = av_clip_pixel(p0 + f2);

  1813.                 dst[strideb * +0] = av_clip_pixel(q0 - f1);

  1814.             } else {

  1815.                 int f = av_clip_intp2(3 * (q0 - p0), BIT_DEPTH - 1), f1, f2;

  1816. 

  1817.                 f1 = FFMIN(f + 4, (1 << (BIT_DEPTH - 1)) - 1) >> 3;

  1818.                 f2 = FFMIN(f + 3, (1 << (BIT_DEPTH - 1)) - 1) >> 3;

  1819. 

  1820.                 dst[strideb * -1] = av_clip_pixel(p0 + f2);

  1821.                 dst[strideb * +0] = av_clip_pixel(q0 - f1);

  1822. 

  1823.                 f = (f1 + 1) >> 1;

  1824.                 dst[strideb * -2] = av_clip_pixel(p1 + f);

  1825.                 dst[strideb * +1] = av_clip_pixel(q1 - f);

  1826.             }

  1827.         }

  1828.     }

  1829. }

  1830. 

  1831. #define lf_8_fn(dir, wd, stridea, strideb) \

  1832. static void loop_filter_##dir##_##wd##_8_c(uint8_t *_dst, \

  1833.                                            ptrdiff_t stride, \

  1834.                                            int E, int I, int H) \

  1835. { \

  1836.     pixel *dst = (pixel *) _dst; \

  1837.     stride /= sizeof(pixel); \

  1838.     loop_filter(dst, E, I, H, stridea, strideb, wd); \

  1839. }

  1840. 

  1841. #define lf_8_fns(wd) \

  1842. lf_8_fn(h, wd, stride, 1) \

  1843. lf_8_fn(v, wd, 1, stride)

  1844. 

  1845. lf_8_fns(4)

  1846. lf_8_fns(8)

  1847. lf_8_fns(16)

  1848. 

  1849. #undef lf_8_fn

  1850. #undef lf_8_fns

  1851. 

  1852. #define lf_16_fn(dir, stridea) \

  1853. static void loop_filter_##dir##_16_16_c(uint8_t *dst, \

  1854.                                         ptrdiff_t stride, \

  1855.                                         int E, int I, int H) \

  1856. { \

  1857.     loop_filter_##dir##_16_8_c(dst, stride, E, I, H); \

  1858.     loop_filter_##dir##_16_8_c(dst + 8 * stridea, stride, E, I, H); \

  1859. }

  1860. 

  1861. lf_16_fn(h, stride)

  1862. lf_16_fn(v, sizeof(pixel))

  1863. 

  1864. #undef lf_16_fn

  1865. 

  1866. #define lf_mix_fn(dir, wd1, wd2, stridea) \

  1867. static void loop_filter_##dir##_##wd1##wd2##_16_c(uint8_t *dst, \

  1868.                                                   ptrdiff_t stride, \

  1869.                                                   int E, int I, int H) \

  1870. { \

  1871.     loop_filter_##dir##_##wd1##_8_c(dst, stride, E & 0xff, I & 0xff, H & 0xff); \

  1872.     loop_filter_##dir##_##wd2##_8_c(dst + 8 * stridea, stride, E >> 8, I >> 8, H >> 8); \

  1873. }

  1874. 

  1875. #define lf_mix_fns(wd1, wd2) \

  1876. lf_mix_fn(h, wd1, wd2, stride) \

  1877. lf_mix_fn(v, wd1, wd2, sizeof(pixel))

  1878. 

  1879. lf_mix_fns(4, 4)

  1880. lf_mix_fns(4, 8)

  1881. lf_mix_fns(8, 4)

  1882. lf_mix_fns(8, 8)

  1883. 

  1884. #undef lf_mix_fn

  1885. #undef lf_mix_fns

  1886. 

  1887. static av_cold void vp9dsp_loopfilter_init(VP9DSPContext *dsp)

  1888. {

  1889.     dsp->loop_filter_8[0][0] = loop_filter_h_4_8_c;

  1890.     dsp->loop_filter_8[0][1] = loop_filter_v_4_8_c;

  1891.     dsp->loop_filter_8[1][0] = loop_filter_h_8_8_c;

  1892.     dsp->loop_filter_8[1][1] = loop_filter_v_8_8_c;

  1893.     dsp->loop_filter_8[2][0] = loop_filter_h_16_8_c;

  1894.     dsp->loop_filter_8[2][1] = loop_filter_v_16_8_c;

  1895. 

  1896.     dsp->loop_filter_16[0] = loop_filter_h_16_16_c;

  1897.     dsp->loop_filter_16[1] = loop_filter_v_16_16_c;

  1898. 

  1899.     dsp->loop_filter_mix2[0][0][0] = loop_filter_h_44_16_c;

  1900.     dsp->loop_filter_mix2[0][0][1] = loop_filter_v_44_16_c;

  1901.     dsp->loop_filter_mix2[0][1][0] = loop_filter_h_48_16_c;

  1902.     dsp->loop_filter_mix2[0][1][1] = loop_filter_v_48_16_c;

  1903.     dsp->loop_filter_mix2[1][0][0] = loop_filter_h_84_16_c;

  1904.     dsp->loop_filter_mix2[1][0][1] = loop_filter_v_84_16_c;

  1905.     dsp->loop_filter_mix2[1][1][0] = loop_filter_h_88_16_c;

  1906.     dsp->loop_filter_mix2[1][1][1] = loop_filter_v_88_16_c;

  1907. }

  1908. 

  1909. static av_always_inline void copy_c(uint8_t *dst, ptrdiff_t dst_stride,

  1910.                                     const uint8_t *src, ptrdiff_t src_stride,

  1911.                                     int w, int h)

  1912. {

  1913.     do {

  1914.         memcpy(dst, src, w * sizeof(pixel));

  1915. 

  1916.         dst += dst_stride;

  1917.         src += src_stride;

  1918.     } while (--h);

  1919. }

  1920. 

  1921. static av_always_inline void avg_c(uint8_t *_dst, ptrdiff_t dst_stride,

  1922.                                    const uint8_t *_src, ptrdiff_t src_stride,

  1923.                                    int w, int h)

  1924. {

  1925.     pixel *dst = (pixel *) _dst;

  1926.     const pixel *src = (const pixel *) _src;

  1927. 

  1928.     dst_stride /= sizeof(pixel);

  1929.     src_stride /= sizeof(pixel);

  1930.     do {

  1931.         int x;

  1932. 

  1933.         for (x = 0; x < w; x += 4)

  1934.             AV_WN4PA(&dst[x], rnd_avg_pixel4(AV_RN4PA(&dst[x]), AV_RN4P(&src[x])));

  1935. 

  1936.         dst += dst_stride;

  1937.         src += src_stride;

  1938.     } while (--h);

  1939. }

  1940. 

  1941. #define fpel_fn(type, sz) \

  1942. static void type##sz##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  1943.                          const uint8_t *src, ptrdiff_t src_stride, \

  1944.                          int h, int mx, int my) \

  1945. { \

  1946.     type##_c(dst, dst_stride, src, src_stride, sz, h); \

  1947. }

  1948. 

  1949. #define copy_avg_fn(sz) \

  1950. fpel_fn(copy, sz) \

  1951. fpel_fn(avg,  sz)

  1952. 

  1953. copy_avg_fn(64)

  1954. copy_avg_fn(32)

  1955. copy_avg_fn(16)

  1956. copy_avg_fn(8)

  1957. copy_avg_fn(4)

  1958. 

  1959. #undef fpel_fn

  1960. #undef copy_avg_fn

  1961. 

  1962. static const int16_t vp9_subpel_filters[3][16][8] = {

  1963.     [FILTER_8TAP_REGULAR] = {

  1964.         {  0,  0,   0, 128,   0,   0,  0,  0 },

  1965.         {  0,  1,  -5, 126,   8,  -3,  1,  0 },

  1966.         { -1,  3, -10, 122,  18,  -6,  2,  0 },

  1967.         { -1,  4, -13, 118,  27,  -9,  3, -1 },

  1968.         { -1,  4, -16, 112,  37, -11,  4, -1 },

  1969.         { -1,  5, -18, 105,  48, -14,  4, -1 },

  1970.         { -1,  5, -19,  97,  58, -16,  5, -1 },

  1971.         { -1,  6, -19,  88,  68, -18,  5, -1 },

  1972.         { -1,  6, -19,  78,  78, -19,  6, -1 },

  1973.         { -1,  5, -18,  68,  88, -19,  6, -1 },

  1974.         { -1,  5, -16,  58,  97, -19,  5, -1 },

  1975.         { -1,  4, -14,  48, 105, -18,  5, -1 },

  1976.         { -1,  4, -11,  37, 112, -16,  4, -1 },

  1977.         { -1,  3,  -9,  27, 118, -13,  4, -1 },

  1978.         {  0,  2,  -6,  18, 122, -10,  3, -1 },

  1979.         {  0,  1,  -3,   8, 126,  -5,  1,  0 },

  1980.     }, [FILTER_8TAP_SHARP] = {

  1981.         {  0,  0,   0, 128,   0,   0,  0,  0 },

  1982.         { -1,  3,  -7, 127,   8,  -3,  1,  0 },

  1983.         { -2,  5, -13, 125,  17,  -6,  3, -1 },

  1984.         { -3,  7, -17, 121,  27, -10,  5, -2 },

  1985.         { -4,  9, -20, 115,  37, -13,  6, -2 },

  1986.         { -4, 10, -23, 108,  48, -16,  8, -3 },

  1987.         { -4, 10, -24, 100,  59, -19,  9, -3 },

  1988.         { -4, 11, -24,  90,  70, -21, 10, -4 },

  1989.         { -4, 11, -23,  80,  80, -23, 11, -4 },

  1990.         { -4, 10, -21,  70,  90, -24, 11, -4 },

  1991.         { -3,  9, -19,  59, 100, -24, 10, -4 },

  1992.         { -3,  8, -16,  48, 108, -23, 10, -4 },

  1993.         { -2,  6, -13,  37, 115, -20,  9, -4 },

  1994.         { -2,  5, -10,  27, 121, -17,  7, -3 },

  1995.         { -1,  3,  -6,  17, 125, -13,  5, -2 },

  1996.         {  0,  1,  -3,   8, 127,  -7,  3, -1 },

  1997.     }, [FILTER_8TAP_SMOOTH] = {

  1998.         {  0,  0,   0, 128,   0,   0,  0,  0 },

  1999.         { -3, -1,  32,  64,  38,   1, -3,  0 },

  2000.         { -2, -2,  29,  63,  41,   2, -3,  0 },

  2001.         { -2, -2,  26,  63,  43,   4, -4,  0 },

  2002.         { -2, -3,  24,  62,  46,   5, -4,  0 },

  2003.         { -2, -3,  21,  60,  49,   7, -4,  0 },

  2004.         { -1, -4,  18,  59,  51,   9, -4,  0 },

  2005.         { -1, -4,  16,  57,  53,  12, -4, -1 },

  2006.         { -1, -4,  14,  55,  55,  14, -4, -1 },

  2007.         { -1, -4,  12,  53,  57,  16, -4, -1 },

  2008.         {  0, -4,   9,  51,  59,  18, -4, -1 },

  2009.         {  0, -4,   7,  49,  60,  21, -3, -2 },

  2010.         {  0, -4,   5,  46,  62,  24, -3, -2 },

  2011.         {  0, -4,   4,  43,  63,  26, -2, -2 },

  2012.         {  0, -3,   2,  41,  63,  29, -2, -2 },

  2013.         {  0, -3,   1,  38,  64,  32, -1, -3 },

  2014.     }

  2015. };

  2016. 

  2017. #define FILTER_8TAP(src, x, F, stride) \

  2018.     av_clip_pixel((F[0] * src[x + -3 * stride] + \

  2019.                    F[1] * src[x + -2 * stride] + \

  2020.                    F[2] * src[x + -1 * stride] + \

  2021.                    F[3] * src[x + +0 * stride] + \

  2022.                    F[4] * src[x + +1 * stride] + \

  2023.                    F[5] * src[x + +2 * stride] + \

  2024.                    F[6] * src[x + +3 * stride] + \

  2025.                    F[7] * src[x + +4 * stride] + 64) >> 7)

  2026. 

  2027. static av_always_inline void do_8tap_1d_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2028.                                           const uint8_t *_src, ptrdiff_t src_stride,

  2029.                                           int w, int h, ptrdiff_t ds,

  2030.                                           const int16_t *filter, int avg)

  2031. {

  2032.     pixel *dst = (pixel *) _dst;

  2033.     const pixel *src = (const pixel *) _src;

  2034. 

  2035.     dst_stride /= sizeof(pixel);

  2036.     src_stride /= sizeof(pixel);

  2037.     do {

  2038.         int x;

  2039. 

  2040.         for (x = 0; x < w; x++)

  2041.             if (avg) {

  2042.                 dst[x] = (dst[x] + FILTER_8TAP(src, x, filter, ds) + 1) >> 1;

  2043.             } else {

  2044.                 dst[x] = FILTER_8TAP(src, x, filter, ds);

  2045.             }

  2046. 

  2047.         dst += dst_stride;

  2048.         src += src_stride;

  2049.     } while (--h);

  2050. }

  2051. 

  2052. #define filter_8tap_1d_fn(opn, opa, dir, ds) \

  2053. static av_noinline void opn##_8tap_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2054.                                                 const uint8_t *src, ptrdiff_t src_stride, \

  2055.                                                 int w, int h, const int16_t *filter) \

  2056. { \

  2057.     do_8tap_1d_c(dst, dst_stride, src, src_stride, w, h, ds, filter, opa); \

  2058. }

  2059. 

  2060. filter_8tap_1d_fn(put, 0, v, src_stride / sizeof(pixel))

  2061. filter_8tap_1d_fn(put, 0, h, 1)

  2062. filter_8tap_1d_fn(avg, 1, v, src_stride / sizeof(pixel))

  2063. filter_8tap_1d_fn(avg, 1, h, 1)

  2064. 

  2065. #undef filter_8tap_1d_fn

  2066. 

  2067. static av_always_inline void do_8tap_2d_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2068.                                           const uint8_t *_src, ptrdiff_t src_stride,

  2069.                                           int w, int h, const int16_t *filterx,

  2070.                                           const int16_t *filtery, int avg)

  2071. {

  2072.     int tmp_h = h + 7;

  2073.     pixel tmp[64 * 71], *tmp_ptr = tmp;

  2074.     pixel *dst = (pixel *) _dst;

  2075.     const pixel *src = (const pixel *) _src;

  2076. 

  2077.     dst_stride /= sizeof(pixel);

  2078.     src_stride /= sizeof(pixel);

  2079.     src -= src_stride * 3;

  2080.     do {

  2081.         int x;

  2082. 

  2083.         for (x = 0; x < w; x++)

  2084.             tmp_ptr[x] = FILTER_8TAP(src, x, filterx, 1);

  2085. 

  2086.         tmp_ptr += 64;

  2087.         src += src_stride;

  2088.     } while (--tmp_h);

  2089. 

  2090.     tmp_ptr = tmp + 64 * 3;

  2091.     do {

  2092.         int x;

  2093. 

  2094.         for (x = 0; x < w; x++)

  2095.             if (avg) {

  2096.                 dst[x] = (dst[x] + FILTER_8TAP(tmp_ptr, x, filtery, 64) + 1) >> 1;

  2097.             } else {

  2098.                 dst[x] = FILTER_8TAP(tmp_ptr, x, filtery, 64);

  2099.             }

  2100. 

  2101.         tmp_ptr += 64;

  2102.         dst += dst_stride;

  2103.     } while (--h);

  2104. }

  2105. 

  2106. #define filter_8tap_2d_fn(opn, opa) \

  2107. static av_noinline void opn##_8tap_2d_hv_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2108.                                            const uint8_t *src, ptrdiff_t src_stride, \

  2109.                                            int w, int h, const int16_t *filterx, \

  2110.                                            const int16_t *filtery) \

  2111. { \

  2112.     do_8tap_2d_c(dst, dst_stride, src, src_stride, w, h, filterx, filtery, opa); \

  2113. }

  2114. 

  2115. filter_8tap_2d_fn(put, 0)

  2116. filter_8tap_2d_fn(avg, 1)

  2117. 

  2118. #undef filter_8tap_2d_fn

  2119. 

  2120. #define filter_fn_1d(sz, dir, dir_m, type, type_idx, avg) \

  2121. static void avg##_8tap_##type##_##sz##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2122.                                               const uint8_t *src, ptrdiff_t src_stride, \

  2123.                                               int h, int mx, int my) \

  2124. { \

  2125.     avg##_8tap_1d_##dir##_c(dst, dst_stride, src, src_stride, sz, h, \

  2126.                             vp9_subpel_filters[type_idx][dir_m]); \

  2127. }

  2128. 

  2129. #define filter_fn_2d(sz, type, type_idx, avg) \

  2130. static void avg##_8tap_##type##_##sz##hv_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2131.                                            const uint8_t *src, ptrdiff_t src_stride, \

  2132.                                            int h, int mx, int my) \

  2133. { \

  2134.     avg##_8tap_2d_hv_c(dst, dst_stride, src, src_stride, sz, h, \

  2135.                        vp9_subpel_filters[type_idx][mx], \

  2136.                        vp9_subpel_filters[type_idx][my]); \

  2137. }

  2138. 

  2139. #define FILTER_BILIN(src, x, mxy, stride) \

  2140.     (src[x] + ((mxy * (src[x + stride] - src[x]) + 8) >> 4))

  2141. 

  2142. static av_always_inline void do_bilin_1d_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2143.                                            const uint8_t *_src, ptrdiff_t src_stride,

  2144.                                            int w, int h, ptrdiff_t ds, int mxy, int avg)

  2145. {

  2146.     pixel *dst = (pixel *) _dst;

  2147.     const pixel *src = (const pixel *) _src;

  2148. 

  2149.     dst_stride /= sizeof(pixel);

  2150.     src_stride /= sizeof(pixel);

  2151.     do {

  2152.         int x;

  2153. 

  2154.         for (x = 0; x < w; x++)

  2155.             if (avg) {

  2156.                 dst[x] = (dst[x] + FILTER_BILIN(src, x, mxy, ds) + 1) >> 1;

  2157.             } else {

  2158.                 dst[x] = FILTER_BILIN(src, x, mxy, ds);

  2159.             }

  2160. 

  2161.         dst += dst_stride;

  2162.         src += src_stride;

  2163.     } while (--h);

  2164. }

  2165. 

  2166. #define bilin_1d_fn(opn, opa, dir, ds) \

  2167. static av_noinline void opn##_bilin_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2168.                                                  const uint8_t *src, ptrdiff_t src_stride, \

  2169.                                                  int w, int h, int mxy) \

  2170. { \

  2171.     do_bilin_1d_c(dst, dst_stride, src, src_stride, w, h, ds, mxy, opa); \

  2172. }

  2173. 

  2174. bilin_1d_fn(put, 0, v, src_stride / sizeof(pixel))

  2175. bilin_1d_fn(put, 0, h, 1)

  2176. bilin_1d_fn(avg, 1, v, src_stride / sizeof(pixel))

  2177. bilin_1d_fn(avg, 1, h, 1)

  2178. 

  2179. #undef bilin_1d_fn

  2180. 

  2181. static av_always_inline void do_bilin_2d_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2182.                                            const uint8_t *_src, ptrdiff_t src_stride,

  2183.                                            int w, int h, int mx, int my, int avg)

  2184. {

  2185.     pixel tmp[64 * 65], *tmp_ptr = tmp;

  2186.     int tmp_h = h + 1;

  2187.     pixel *dst = (pixel *) _dst;

  2188.     const pixel *src = (const pixel *) _src;

  2189. 

  2190.     dst_stride /= sizeof(pixel);

  2191.     src_stride /= sizeof(pixel);

  2192.     do {

  2193.         int x;

  2194. 

  2195.         for (x = 0; x < w; x++)

  2196.             tmp_ptr[x] = FILTER_BILIN(src, x, mx, 1);

  2197. 

  2198.         tmp_ptr += 64;

  2199.         src += src_stride;

  2200.     } while (--tmp_h);

  2201. 

  2202.     tmp_ptr = tmp;

  2203.     do {

  2204.         int x;

  2205. 

  2206.         for (x = 0; x < w; x++)

  2207.             if (avg) {

  2208.                 dst[x] = (dst[x] + FILTER_BILIN(tmp_ptr, x, my, 64) + 1) >> 1;

  2209.             } else {

  2210.                 dst[x] = FILTER_BILIN(tmp_ptr, x, my, 64);

  2211.             }

  2212. 

  2213.         tmp_ptr += 64;

  2214.         dst += dst_stride;

  2215.     } while (--h);

  2216. }

  2217. 

  2218. #define bilin_2d_fn(opn, opa) \

  2219. static av_noinline void opn##_bilin_2d_hv_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2220.                                             const uint8_t *src, ptrdiff_t src_stride, \

  2221.                                             int w, int h, int mx, int my) \

  2222. { \

  2223.     do_bilin_2d_c(dst, dst_stride, src, src_stride, w, h, mx, my, opa); \

  2224. }

  2225. 

  2226. bilin_2d_fn(put, 0)

  2227. bilin_2d_fn(avg, 1)

  2228. 

  2229. #undef bilin_2d_fn

  2230. 

  2231. #define bilinf_fn_1d(sz, dir, dir_m, avg) \

  2232. static void avg##_bilin_##sz##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2233.                                       const uint8_t *src, ptrdiff_t src_stride, \

  2234.                                       int h, int mx, int my) \

  2235. { \

  2236.     avg##_bilin_1d_##dir##_c(dst, dst_stride, src, src_stride, sz, h, dir_m); \

  2237. }

  2238. 

  2239. #define bilinf_fn_2d(sz, avg) \

  2240. static void avg##_bilin_##sz##hv_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2241.                                    const uint8_t *src, ptrdiff_t src_stride, \

  2242.                                    int h, int mx, int my) \

  2243. { \

  2244.     avg##_bilin_2d_hv_c(dst, dst_stride, src, src_stride, sz, h, mx, my); \

  2245. }

  2246. 

  2247. #define filter_fn(sz, avg) \

  2248. filter_fn_1d(sz, h, mx, regular, FILTER_8TAP_REGULAR, avg) \

  2249. filter_fn_1d(sz, v, my, regular, FILTER_8TAP_REGULAR, avg) \

  2250. filter_fn_2d(sz,        regular, FILTER_8TAP_REGULAR, avg) \

  2251. filter_fn_1d(sz, h, mx, smooth,  FILTER_8TAP_SMOOTH,  avg) \

  2252. filter_fn_1d(sz, v, my, smooth,  FILTER_8TAP_SMOOTH,  avg) \

  2253. filter_fn_2d(sz,        smooth,  FILTER_8TAP_SMOOTH,  avg) \

  2254. filter_fn_1d(sz, h, mx, sharp,   FILTER_8TAP_SHARP,   avg) \

  2255. filter_fn_1d(sz, v, my, sharp,   FILTER_8TAP_SHARP,   avg) \

  2256. filter_fn_2d(sz,        sharp,   FILTER_8TAP_SHARP,   avg) \

  2257. bilinf_fn_1d(sz, h, mx,                               avg) \

  2258. bilinf_fn_1d(sz, v, my,                               avg) \

  2259. bilinf_fn_2d(sz,                                      avg)

  2260. 

  2261. #define filter_fn_set(avg) \

  2262. filter_fn(64, avg) \

  2263. filter_fn(32, avg) \

  2264. filter_fn(16, avg) \

  2265. filter_fn(8,  avg) \

  2266. filter_fn(4,  avg)

  2267. 

  2268. filter_fn_set(put)

  2269. filter_fn_set(avg)

  2270. 

  2271. #undef filter_fn

  2272. #undef filter_fn_set

  2273. #undef filter_fn_1d

  2274. #undef filter_fn_2d

  2275. #undef bilinf_fn_1d

  2276. #undef bilinf_fn_2d

  2277. 

  2278. static av_cold void vp9dsp_mc_init(VP9DSPContext *dsp)

  2279. {

  2280. #define init_fpel(idx1, idx2, sz, type) \

  2281.     dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = type##sz##_c; \

  2282.     dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = type##sz##_c; \

  2283.     dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][0][0] = type##sz##_c; \

  2284.     dsp->mc[idx1][FILTER_BILINEAR    ][idx2][0][0] = type##sz##_c

  2285. 

  2286. #define init_copy_avg(idx, sz) \

  2287.     init_fpel(idx, 0, sz, copy); \

  2288.     init_fpel(idx, 1, sz, avg)

  2289. 

  2290.     init_copy_avg(0, 64);

  2291.     init_copy_avg(1, 32);

  2292.     init_copy_avg(2, 16);

  2293.     init_copy_avg(3,  8);

  2294.     init_copy_avg(4,  4);

  2295. 

  2296. #undef init_copy_avg

  2297. #undef init_fpel

  2298. 

  2299. #define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type) \

  2300.     dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_c; \

  2301.     dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_c; \

  2302.     dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_c; \

  2303.     dsp->mc[idx1][FILTER_BILINEAR    ][idx2][idxh][idxv] = type##_bilin_##sz##dir##_c

  2304. 

  2305. #define init_subpel2(idx, idxh, idxv, dir, type) \

  2306.     init_subpel1(0, idx, idxh, idxv, 64, dir, type); \

  2307.     init_subpel1(1, idx, idxh, idxv, 32, dir, type); \

  2308.     init_subpel1(2, idx, idxh, idxv, 16, dir, type); \

  2309.     init_subpel1(3, idx, idxh, idxv,  8, dir, type); \

  2310.     init_subpel1(4, idx, idxh, idxv,  4, dir, type)

  2311. 

  2312. #define init_subpel3(idx, type) \

  2313.     init_subpel2(idx, 1, 1, hv, type); \

  2314.     init_subpel2(idx, 0, 1, v, type); \

  2315.     init_subpel2(idx, 1, 0, h, type)

  2316. 

  2317.     init_subpel3(0, put);

  2318.     init_subpel3(1, avg);

  2319. 

  2320. #undef init_subpel1

  2321. #undef init_subpel2

  2322. #undef init_subpel3

  2323. }

  2324. 

  2325. static av_always_inline void do_scaled_8tap_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2326.                                               const uint8_t *_src, ptrdiff_t src_stride,

  2327.                                               int w, int h, int mx, int my,

  2328.                                               int dx, int dy, int avg,

  2329.                                               const int16_t (*filters)[8])

  2330. {

  2331.     int tmp_h = (((h - 1) * dy + my) >> 4) + 8;

  2332.     pixel tmp[64 * 135], *tmp_ptr = tmp;

  2333.     pixel *dst = (pixel *) _dst;

  2334.     const pixel *src = (const pixel *) _src;

  2335. 

  2336.     dst_stride /= sizeof(pixel);

  2337.     src_stride /= sizeof(pixel);

  2338.     src -= src_stride * 3;

  2339.     do {

  2340.         int x;

  2341.         int imx = mx, ioff = 0;

  2342. 

  2343.         for (x = 0; x < w; x++) {

  2344.             tmp_ptr[x] = FILTER_8TAP(src, ioff, filters[imx], 1);

  2345.             imx += dx;

  2346.             ioff += imx >> 4;

  2347.             imx &= 0xf;

  2348.         }

  2349. 

  2350.         tmp_ptr += 64;

  2351.         src += src_stride;

  2352.     } while (--tmp_h);

  2353. 

  2354.     tmp_ptr = tmp + 64 * 3;

  2355.     do {

  2356.         int x;

  2357.         const int16_t *filter = filters[my];

  2358. 

  2359.         for (x = 0; x < w; x++)

  2360.             if (avg) {

  2361.                 dst[x] = (dst[x] + FILTER_8TAP(tmp_ptr, x, filter, 64) + 1) >> 1;

  2362.             } else {

  2363.                 dst[x] = FILTER_8TAP(tmp_ptr, x, filter, 64);

  2364.             }

  2365. 

  2366.         my += dy;

  2367.         tmp_ptr += (my >> 4) * 64;

  2368.         my &= 0xf;

  2369.         dst += dst_stride;

  2370.     } while (--h);

  2371. }

  2372. 

  2373. #define scaled_filter_8tap_fn(opn, opa) \

  2374. static av_noinline void opn##_scaled_8tap_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2375.                                             const uint8_t *src, ptrdiff_t src_stride, \

  2376.                                             int w, int h, int mx, int my, int dx, int dy, \

  2377.                                             const int16_t (*filters)[8]) \

  2378. { \

  2379.     do_scaled_8tap_c(dst, dst_stride, src, src_stride, w, h, mx, my, dx, dy, \

  2380.                      opa, filters); \

  2381. }

  2382. 

  2383. scaled_filter_8tap_fn(put, 0)

  2384. scaled_filter_8tap_fn(avg, 1)

  2385. 

  2386. #undef scaled_filter_8tap_fn

  2387. 

  2388. #undef FILTER_8TAP

  2389. 

  2390. #define scaled_filter_fn(sz, type, type_idx, avg) \

  2391. static void avg##_scaled_##type##_##sz##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2392.                                            const uint8_t *src, ptrdiff_t src_stride, \

  2393.                                            int h, int mx, int my, int dx, int dy) \

  2394. { \

  2395.     avg##_scaled_8tap_c(dst, dst_stride, src, src_stride, sz, h, mx, my, dx, dy, \

  2396.                         vp9_subpel_filters[type_idx]); \

  2397. }

  2398. 

  2399. static av_always_inline void do_scaled_bilin_c(uint8_t *_dst, ptrdiff_t dst_stride,

  2400.                                                const uint8_t *_src, ptrdiff_t src_stride,

  2401.                                                int w, int h, int mx, int my,

  2402.                                                int dx, int dy, int avg)

  2403. {

  2404.     pixel tmp[64 * 129], *tmp_ptr = tmp;

  2405.     int tmp_h = (((h - 1) * dy + my) >> 4) + 2;

  2406.     pixel *dst = (pixel *) _dst;

  2407.     const pixel *src = (const pixel *) _src;

  2408. 

  2409.     dst_stride /= sizeof(pixel);

  2410.     src_stride /= sizeof(pixel);

  2411.     do {

  2412.         int x;

  2413.         int imx = mx, ioff = 0;

  2414. 

  2415.         for (x = 0; x < w; x++) {

  2416.             tmp_ptr[x] = FILTER_BILIN(src, ioff, imx, 1);

  2417.             imx += dx;

  2418.             ioff += imx >> 4;

  2419.             imx &= 0xf;

  2420.         }

  2421. 

  2422.         tmp_ptr += 64;

  2423.         src += src_stride;

  2424.     } while (--tmp_h);

  2425. 

  2426.     tmp_ptr = tmp;

  2427.     do {

  2428.         int x;

  2429. 

  2430.         for (x = 0; x < w; x++)

  2431.             if (avg) {

  2432.                 dst[x] = (dst[x] + FILTER_BILIN(tmp_ptr, x, my, 64) + 1) >> 1;

  2433.             } else {

  2434.                 dst[x] = FILTER_BILIN(tmp_ptr, x, my, 64);

  2435.             }

  2436. 

  2437.         my += dy;

  2438.         tmp_ptr += (my >> 4) * 64;

  2439.         my &= 0xf;

  2440.         dst += dst_stride;

  2441.     } while (--h);

  2442. }

  2443. 

  2444. #define scaled_bilin_fn(opn, opa) \

  2445. static av_noinline void opn##_scaled_bilin_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2446.                                              const uint8_t *src, ptrdiff_t src_stride, \

  2447.                                              int w, int h, int mx, int my, int dx, int dy) \

  2448. { \

  2449.     do_scaled_bilin_c(dst, dst_stride, src, src_stride, w, h, mx, my, dx, dy, opa); \

  2450. }

  2451. 

  2452. scaled_bilin_fn(put, 0)

  2453. scaled_bilin_fn(avg, 1)

  2454. 

  2455. #undef scaled_bilin_fn

  2456. 

  2457. #undef FILTER_BILIN

  2458. 

  2459. #define scaled_bilinf_fn(sz, avg) \

  2460. static void avg##_scaled_bilin_##sz##_c(uint8_t *dst, ptrdiff_t dst_stride, \

  2461.                                         const uint8_t *src, ptrdiff_t src_stride, \

  2462.                                         int h, int mx, int my, int dx, int dy) \

  2463. { \

  2464.     avg##_scaled_bilin_c(dst, dst_stride, src, src_stride, sz, h, mx, my, dx, dy); \

  2465. }

  2466. 

  2467. #define scaled_filter_fns(sz, avg) \

  2468. scaled_filter_fn(sz,        regular, FILTER_8TAP_REGULAR, avg) \

  2469. scaled_filter_fn(sz,        smooth,  FILTER_8TAP_SMOOTH,  avg) \

  2470. scaled_filter_fn(sz,        sharp,   FILTER_8TAP_SHARP,   avg) \

  2471. scaled_bilinf_fn(sz,                                      avg)

  2472. 

  2473. #define scaled_filter_fn_set(avg) \

  2474. scaled_filter_fns(64, avg) \

  2475. scaled_filter_fns(32, avg) \

  2476. scaled_filter_fns(16, avg) \

  2477. scaled_filter_fns(8,  avg) \

  2478. scaled_filter_fns(4,  avg)

  2479. 

  2480. scaled_filter_fn_set(put)

  2481. scaled_filter_fn_set(avg)

  2482. 

  2483. #undef scaled_filter_fns

  2484. #undef scaled_filter_fn_set

  2485. #undef scaled_filter_fn

  2486. #undef scaled_bilinf_fn

  2487. 

  2488. static av_cold void vp9dsp_scaled_mc_init(VP9DSPContext *dsp)

  2489. {

  2490. #define init_scaled(idx1, idx2, sz, type) \

  2491.     dsp->smc[idx1][FILTER_8TAP_SMOOTH ][idx2] = type##_scaled_smooth_##sz##_c; \

  2492.     dsp->smc[idx1][FILTER_8TAP_REGULAR][idx2] = type##_scaled_regular_##sz##_c; \

  2493.     dsp->smc[idx1][FILTER_8TAP_SHARP  ][idx2] = type##_scaled_sharp_##sz##_c; \

  2494.     dsp->smc[idx1][FILTER_BILINEAR    ][idx2] = type##_scaled_bilin_##sz##_c

  2495. 

  2496. #define init_scaled_put_avg(idx, sz) \

  2497.     init_scaled(idx, 0, sz, put); \

  2498.     init_scaled(idx, 1, sz, avg)

  2499. 

  2500.     init_scaled_put_avg(0, 64);

  2501.     init_scaled_put_avg(1, 32);

  2502.     init_scaled_put_avg(2, 16);

  2503.     init_scaled_put_avg(3,  8);

  2504.     init_scaled_put_avg(4,  4);

  2505. 

  2506. #undef init_scaled_put_avg

  2507. #undef init_scaled

  2508. }

  2509. 

  2510. av_cold void FUNC(ff_vp9dsp_init)(VP9DSPContext *dsp)

  2511. {

  2512.     vp9dsp_intrapred_init(dsp);

  2513.     vp9dsp_itxfm_init(dsp);

  2514.     vp9dsp_loopfilter_init(dsp);

  2515.     vp9dsp_mc_init(dsp);

  2516.     vp9dsp_scaled_mc_init(dsp);

  2517. }