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.
63293 Commits
32 Branches
979MB
Tree: 3791741cf9
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3791741cf9'
${ noResults }
FFmpeg/libavcodec/hevcpred_template.c

556 lines
22KB
Raw Blame History 

  1. /*

  2.  * HEVC video decoder

  3.  *

  4.  * Copyright (C) 2012 - 2013 Guillaume Martres

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

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

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

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

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

  12.  *

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

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

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

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

  21.  */

  22. 

  23. #include "libavutil/pixdesc.h"

  24. 

  25. #include "bit_depth_template.c"

  26. #include "hevcpred.h"

  27. 

  28. #define POS(x, y) src[(x) + stride * (y)]

  29. 

  30. static av_always_inline void FUNC(intra_pred)(HEVCContext *s, int x0, int y0,

  31.                                               int log2_size, int c_idx)

  32. {

  33. #define PU(x) \

  34.     ((x) >> s->sps->log2_min_pu_size)

  35. #define MVF(x, y) \

  36.     (s->ref->tab_mvf[(x) + (y) * min_pu_width])

  37. #define MVF_PU(x, y) \

  38.     MVF(PU(x0 + ((x) << hshift)), PU(y0 + ((y) << vshift)))

  39. #define IS_INTRA(x, y) \

  40.     (MVF_PU(x, y).pred_flag == PF_INTRA)

  41. #define MIN_TB_ADDR_ZS(x, y) \

  42.     s->pps->min_tb_addr_zs[(y) * s->sps->min_tb_width + (x)]

  43. 

  44. #define EXTEND(ptr, val, len)         \

  45. do {                                  \

  46.     pixel4 pix = PIXEL_SPLAT_X4(val); \

  47.     for (i = 0; i < (len); i += 4)    \

  48.         AV_WN4P(ptr + i, pix);        \

  49. } while (0)

  50. 

  51. #define EXTEND_RIGHT_CIP(ptr, start, length)                                   \

  52.         for (i = start; i < (start) + (length); i += 4)                        \

  53.             if (!IS_INTRA(i, -1))                                              \

  54.                 AV_WN4P(&ptr[i], a);                                           \

  55.             else                                                               \

  56.                 a = PIXEL_SPLAT_X4(ptr[i+3])

  57. #define EXTEND_LEFT_CIP(ptr, start, length) \

  58.         for (i = start; i > (start) - (length); i--) \

  59.             if (!IS_INTRA(i - 1, -1)) \

  60.                 ptr[i - 1] = ptr[i]

  61. #define EXTEND_UP_CIP(ptr, start, length)                                      \

  62.         for (i = (start); i > (start) - (length); i -= 4)                      \

  63.             if (!IS_INTRA(-1, i - 3))                                          \

  64.                 AV_WN4P(&ptr[i - 3], a);                                       \

  65.             else                                                               \

  66.                 a = PIXEL_SPLAT_X4(ptr[i - 3])

  67. #define EXTEND_DOWN_CIP(ptr, start, length)                                    \

  68.         for (i = start; i < (start) + (length); i += 4)                        \

  69.             if (!IS_INTRA(-1, i))                                              \

  70.                 AV_WN4P(&ptr[i], a);                                           \

  71.             else                                                               \

  72.                 a = PIXEL_SPLAT_X4(ptr[i + 3])

  73. 

  74.     HEVCLocalContext *lc = s->HEVClc;

  75.     int i;

  76.     int hshift = s->sps->hshift[c_idx];

  77.     int vshift = s->sps->vshift[c_idx];

  78.     int size = (1 << log2_size);

  79.     int size_in_luma_h = size << hshift;

  80.     int size_in_tbs_h  = size_in_luma_h >> s->sps->log2_min_tb_size;

  81.     int size_in_luma_v = size << vshift;

  82.     int size_in_tbs_v  = size_in_luma_v >> s->sps->log2_min_tb_size;

  83.     int x = x0 >> hshift;

  84.     int y = y0 >> vshift;

  85.     int x_tb = x0 >> s->sps->log2_min_tb_size;

  86.     int y_tb = y0 >> s->sps->log2_min_tb_size;

  87.     int cur_tb_addr = MIN_TB_ADDR_ZS(x_tb, y_tb);

  88. 

  89.     ptrdiff_t stride = s->frame->linesize[c_idx] / sizeof(pixel);

  90.     pixel *src = (pixel*)s->frame->data[c_idx] + x + y * stride;

  91. 

  92.     int min_pu_width = s->sps->min_pu_width;

  93. 

  94.     enum IntraPredMode mode = c_idx ? lc->pu.intra_pred_mode_c :

  95.                               lc->tu.cur_intra_pred_mode;

  96.     pixel4 a;

  97.     pixel  left_array[2 * MAX_TB_SIZE + 1];

  98.     pixel  filtered_left_array[2 * MAX_TB_SIZE + 1];

  99.     pixel  top_array[2 * MAX_TB_SIZE + 1];

  100.     pixel  filtered_top_array[2 * MAX_TB_SIZE + 1];

  101. 

  102.     pixel  *left          = left_array + 1;

  103.     pixel  *top           = top_array  + 1;

  104.     pixel  *filtered_left = filtered_left_array + 1;

  105.     pixel  *filtered_top  = filtered_top_array  + 1;

  106. 

  107.     int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs_v);

  108.     int cand_left        = lc->na.cand_left;

  109.     int cand_up_left     = lc->na.cand_up_left;

  110.     int cand_up          = lc->na.cand_up;

  111.     int cand_up_right    = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs_h, y_tb - 1);

  112. 

  113.     int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma_v, s->sps->height) -

  114.                            (y0 + size_in_luma_v)) >> vshift;

  115.     int top_right_size   = (FFMIN(x0 + 2 * size_in_luma_h, s->sps->width) -

  116.                            (x0 + size_in_luma_h)) >> hshift;

  117. 

  118.     if (s->pps->constrained_intra_pred_flag == 1) {

  119.         int size_in_luma_pu_v = PU(size_in_luma_v);

  120.         int size_in_luma_pu_h = PU(size_in_luma_h);

  121.         int on_pu_edge_x    = !(x0 & ((1 << s->sps->log2_min_pu_size) - 1));

  122.         int on_pu_edge_y    = !(y0 & ((1 << s->sps->log2_min_pu_size) - 1));

  123.         if (!size_in_luma_pu_h)

  124.             size_in_luma_pu_h++;

  125.         if (cand_bottom_left == 1 && on_pu_edge_x) {

  126.             int x_left_pu   = PU(x0 - 1);

  127.             int y_bottom_pu = PU(y0 + size_in_luma_v);

  128.             int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_bottom_pu);

  129.             cand_bottom_left = 0;

  130.             for (i = 0; i < max; i += 2)

  131.                 cand_bottom_left |= (MVF(x_left_pu, y_bottom_pu + i).pred_flag == PF_INTRA);

  132.         }

  133.         if (cand_left == 1 && on_pu_edge_x) {

  134.             int x_left_pu   = PU(x0 - 1);

  135.             int y_left_pu   = PU(y0);

  136.             int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_left_pu);

  137.             cand_left = 0;

  138.             for (i = 0; i < max; i += 2)

  139.                 cand_left |= (MVF(x_left_pu, y_left_pu + i).pred_flag == PF_INTRA);

  140.         }

  141.         if (cand_up_left == 1) {

  142.             int x_left_pu   = PU(x0 - 1);

  143.             int y_top_pu    = PU(y0 - 1);

  144.             cand_up_left = MVF(x_left_pu, y_top_pu).pred_flag == PF_INTRA;

  145.         }

  146.         if (cand_up == 1 && on_pu_edge_y) {

  147.             int x_top_pu    = PU(x0);

  148.             int y_top_pu    = PU(y0 - 1);

  149.             int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_top_pu);

  150.             cand_up = 0;

  151.             for (i = 0; i < max; i += 2)

  152.                 cand_up |= (MVF(x_top_pu + i, y_top_pu).pred_flag == PF_INTRA);

  153.         }

  154.         if (cand_up_right == 1 && on_pu_edge_y) {

  155.             int y_top_pu    = PU(y0 - 1);

  156.             int x_right_pu  = PU(x0 + size_in_luma_h);

  157.             int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_right_pu);

  158.             cand_up_right = 0;

  159.             for (i = 0; i < max; i += 2)

  160.                 cand_up_right |= (MVF(x_right_pu + i, y_top_pu).pred_flag == PF_INTRA);

  161.         }

  162.         memset(left, 128, 2 * MAX_TB_SIZE*sizeof(pixel));

  163.         memset(top , 128, 2 * MAX_TB_SIZE*sizeof(pixel));

  164.         top[-1] = 128;

  165.     }

  166.     if (cand_up_left) {

  167.         left[-1] = POS(-1, -1);

  168.         top[-1]  = left[-1];

  169.     }

  170.     if (cand_up)

  171.         memcpy(top, src - stride, size * sizeof(pixel));

  172.     if (cand_up_right) {

  173.         memcpy(top + size, src - stride + size, size * sizeof(pixel));

  174.         EXTEND(top + size + top_right_size, POS(size + top_right_size - 1, -1),

  175.                size - top_right_size);

  176.     }

  177.     if (cand_left)

  178.         for (i = 0; i < size; i++)

  179.             left[i] = POS(-1, i);

  180.     if (cand_bottom_left) {

  181.         for (i = size; i < size + bottom_left_size; i++)

  182.             left[i] = POS(-1, i);

  183.         EXTEND(left + size + bottom_left_size, POS(-1, size + bottom_left_size - 1),

  184.                size - bottom_left_size);

  185.     }

  186. 

  187.     if (s->pps->constrained_intra_pred_flag == 1) {

  188.         if (cand_bottom_left || cand_left || cand_up_left || cand_up || cand_up_right) {

  189.             int size_max_x = x0 + ((2 * size) << hshift) < s->sps->width ?

  190.                                     2 * size : (s->sps->width - x0) >> hshift;

  191.             int size_max_y = y0 + ((2 * size) << vshift) < s->sps->height ?

  192.                                     2 * size : (s->sps->height - y0) >> vshift;

  193.             int j = size + (cand_bottom_left? bottom_left_size: 0) -1;

  194.             if (!cand_up_right) {

  195.                 size_max_x = x0 + ((size) << hshift) < s->sps->width ?

  196.                                                     size : (s->sps->width - x0) >> hshift;

  197.             }

  198.             if (!cand_bottom_left) {

  199.                 size_max_y = y0 + (( size) << vshift) < s->sps->height ?

  200.                                                      size : (s->sps->height - y0) >> vshift;

  201.             }

  202.             if (cand_bottom_left || cand_left || cand_up_left) {

  203.                 while (j > -1 && !IS_INTRA(-1, j))

  204.                     j--;

  205.                 if (!IS_INTRA(-1, j)) {

  206.                     j = 0;

  207.                     while (j < size_max_x && !IS_INTRA(j, -1))

  208.                         j++;

  209.                     EXTEND_LEFT_CIP(top, j, j + 1);

  210.                     left[-1] = top[-1];

  211.                     j        = 0;

  212.                 }

  213.             } else {

  214.                 j = 0;

  215.                 while (j < size_max_x && !IS_INTRA(j, -1))

  216.                     j++;

  217.                 if (j > 0)

  218.                     if (x0 > 0) {

  219.                         EXTEND_LEFT_CIP(top, j, j + 1);

  220.                     } else {

  221.                         EXTEND_LEFT_CIP(top, j, j);

  222.                         top[-1] = top[0];

  223.                     }

  224.                 left[-1] = top[-1];

  225.                 j        = 0;

  226.             }

  227.             left[-1] = top[-1];

  228.             if (cand_bottom_left || cand_left) {

  229.                 a = PIXEL_SPLAT_X4(left[-1]);

  230.                 EXTEND_DOWN_CIP(left, 0, size_max_y);

  231.             }

  232.             if (!cand_left)

  233.                 EXTEND(left, left[-1], size);

  234.             if (!cand_bottom_left)

  235.                 EXTEND(left + size, left[size - 1], size);

  236.             if (x0 != 0 && y0 != 0) {

  237.                 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);

  238.                 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);

  239.                 if (!IS_INTRA(-1, - 1))

  240.                     left[-1] = left[0];

  241.             } else if (x0 == 0) {

  242.                 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);

  243.                 EXTEND(left, 0, size_max_y);

  244.             } else {

  245.                 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);

  246.                 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);

  247.             }

  248.             top[-1] = left[-1];

  249.             if (y0 != 0) {

  250.                 a = PIXEL_SPLAT_X4(left[-1]);

  251.                 EXTEND_RIGHT_CIP(top, 0, size_max_x);

  252.             }

  253.         }

  254.     }

  255.     // Infer the unavailable samples

  256.     if (!cand_bottom_left) {

  257.         if (cand_left) {

  258.             EXTEND(left + size, left[size - 1], size);

  259.         } else if (cand_up_left) {

  260.             EXTEND(left, left[-1], 2 * size);

  261.             cand_left = 1;

  262.         } else if (cand_up) {

  263.             left[-1] = top[0];

  264.             EXTEND(left, left[-1], 2 * size);

  265.             cand_up_left = 1;

  266.             cand_left    = 1;

  267.         } else if (cand_up_right) {

  268.             EXTEND(top, top[size], size);

  269.             left[-1] = top[size];

  270.             EXTEND(left, left[-1], 2 * size);

  271.             cand_up      = 1;

  272.             cand_up_left = 1;

  273.             cand_left    = 1;

  274.         } else { // No samples available

  275.             left[-1] = (1 << (BIT_DEPTH - 1));

  276.             EXTEND(top,  left[-1], 2 * size);

  277.             EXTEND(left, left[-1], 2 * size);

  278.         }

  279.     }

  280. 

  281.     if (!cand_left)

  282.         EXTEND(left, left[size], size);

  283.     if (!cand_up_left) {

  284.         left[-1] = left[0];

  285.     }

  286.     if (!cand_up)

  287.         EXTEND(top, left[-1], size);

  288.     if (!cand_up_right)

  289.         EXTEND(top + size, top[size - 1], size);

  290. 

  291.     top[-1] = left[-1];

  292. 

  293.     // Filtering process

  294.     if (c_idx == 0) {

  295.         if (mode != INTRA_DC && size != 4){

  296.             int intra_hor_ver_dist_thresh[] = { 7, 1, 0 };

  297.             int min_dist_vert_hor = FFMIN(FFABS((int)(mode - 26U)),

  298.                                           FFABS((int)(mode - 10U)));

  299.             if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) {

  300.                 int threshold = 1 << (BIT_DEPTH - 5);

  301.                 if (s->sps->sps_strong_intra_smoothing_enable_flag && c_idx == 0 &&

  302.                     log2_size == 5 &&

  303.                     FFABS(top[-1]  + top[63]  - 2 * top[31])  < threshold &&

  304.                     FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) {

  305.                     // We can't just overwrite values in top because it could be

  306.                     // a pointer into src

  307.                     filtered_top[-1] = top[-1];

  308.                     filtered_top[63] = top[63];

  309.                     for (i = 0; i < 63; i++)

  310.                         filtered_top[i] = ((64 - (i + 1)) * top[-1] +

  311.                                            (i + 1)  * top[63] + 32) >> 6;

  312.                     for (i = 0; i < 63; i++)

  313.                         left[i] = ((64 - (i + 1)) * left[-1] +

  314.                                    (i + 1)  * left[63] + 32) >> 6;

  315.                     top = filtered_top;

  316.                 } else {

  317.                     filtered_left[2 * size - 1] = left[2 * size - 1];

  318.                     filtered_top[2 * size - 1]  = top[2 * size - 1];

  319.                     for (i = 2 * size - 2; i >= 0; i--)

  320.                         filtered_left[i] = (left[i + 1] + 2 * left[i] +

  321.                                             left[i - 1] + 2) >> 2;

  322.                     filtered_top[-1]  =

  323.                     filtered_left[-1] = (left[0] + 2 * left[-1] + top[0] + 2) >> 2;

  324.                     for (i = 2 * size - 2; i >= 0; i--)

  325.                         filtered_top[i] = (top[i + 1] + 2 * top[i] +

  326.                                            top[i - 1] + 2) >> 2;

  327.                     left = filtered_left;

  328.                     top  = filtered_top;

  329.                 }

  330.             }

  331.         }

  332.     }

  333. 

  334.     switch (mode) {

  335.     case INTRA_PLANAR:

  336.         s->hpc.pred_planar[log2_size - 2]((uint8_t *)src, (uint8_t *)top,

  337.                                           (uint8_t *)left, stride);

  338.         break;

  339.     case INTRA_DC:

  340.         s->hpc.pred_dc((uint8_t *)src, (uint8_t *)top,

  341.                        (uint8_t *)left, stride, log2_size, c_idx);

  342.         break;

  343.     default:

  344.         s->hpc.pred_angular[log2_size - 2]((uint8_t *)src, (uint8_t *)top,

  345.                                            (uint8_t *)left, stride, c_idx,

  346.                                            mode);

  347.         break;

  348.     }

  349. }

  350. 

  351. #define INTRA_PRED(size)                                                            \

  352. static void FUNC(intra_pred_ ## size)(HEVCContext *s, int x0, int y0, int c_idx)    \

  353. {                                                                                   \

  354.     FUNC(intra_pred)(s, x0, y0, size, c_idx);                                       \

  355. }

  356. 

  357. INTRA_PRED(2)

  358. INTRA_PRED(3)

  359. INTRA_PRED(4)

  360. INTRA_PRED(5)

  361. 

  362. #undef INTRA_PRED

  363. 

  364. static av_always_inline void FUNC(pred_planar)(uint8_t *_src, const uint8_t *_top,

  365.                                   const uint8_t *_left, ptrdiff_t stride,

  366.                                   int trafo_size)

  367. {

  368.     int x, y;

  369.     pixel *src        = (pixel *)_src;

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

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

  372.     int size = 1 << trafo_size;

  373.     for (y = 0; y < size; y++)

  374.         for (x = 0; x < size; x++)

  375.             POS(x, y) = ((size - 1 - x) * left[y] + (x + 1) * top[size]  +

  376.                          (size - 1 - y) * top[x]  + (y + 1) * left[size] + size) >> (trafo_size + 1);

  377. }

  378. 

  379. #define PRED_PLANAR(size)\

  380. static void FUNC(pred_planar_ ## size)(uint8_t *src, const uint8_t *top,        \

  381.                                        const uint8_t *left, ptrdiff_t stride)   \

  382. {                                                                               \

  383.     FUNC(pred_planar)(src, top, left, stride, size + 2);                        \

  384. }

  385. 

  386. PRED_PLANAR(0)

  387. PRED_PLANAR(1)

  388. PRED_PLANAR(2)

  389. PRED_PLANAR(3)

  390. 

  391. #undef PRED_PLANAR

  392. 

  393. static void FUNC(pred_dc)(uint8_t *_src, const uint8_t *_top,

  394.                           const uint8_t *_left,

  395.                           ptrdiff_t stride, int log2_size, int c_idx)

  396. {

  397.     int i, j, x, y;

  398.     int size          = (1 << log2_size);

  399.     pixel *src        = (pixel *)_src;

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

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

  402.     int dc            = size;

  403.     pixel4 a;

  404.     for (i = 0; i < size; i++)

  405.         dc += left[i] + top[i];

  406. 

  407.     dc >>= log2_size + 1;

  408. 

  409.     a = PIXEL_SPLAT_X4(dc);

  410. 

  411.     for (i = 0; i < size; i++)

  412.         for (j = 0; j < size; j+=4)

  413.             AV_WN4P(&POS(j, i), a);

  414. 

  415.     if (c_idx == 0 && size < 32) {

  416.         POS(0, 0) = (left[0] + 2 * dc + top[0] + 2) >> 2;

  417.         for (x = 1; x < size; x++)

  418.             POS(x, 0) = (top[x] + 3 * dc + 2) >> 2;

  419.         for (y = 1; y < size; y++)

  420.             POS(0, y) = (left[y] + 3 * dc + 2) >> 2;

  421.     }

  422. }

  423. 

  424. static av_always_inline void FUNC(pred_angular)(uint8_t *_src,

  425.                                                 const uint8_t *_top,

  426.                                                 const uint8_t *_left,

  427.                                                 ptrdiff_t stride, int c_idx,

  428.                                                 int mode, int size)

  429. {

  430.     int x, y;

  431.     pixel *src        = (pixel *)_src;

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

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

  434. 

  435.     static const int intra_pred_angle[] = {

  436.          32,  26,  21,  17, 13,  9,  5, 2, 0, -2, -5, -9, -13, -17, -21, -26, -32,

  437.         -26, -21, -17, -13, -9, -5, -2, 0, 2,  5,  9, 13,  17,  21,  26,  32

  438.     };

  439.     static const int inv_angle[] = {

  440.         -4096, -1638, -910, -630, -482, -390, -315, -256, -315, -390, -482,

  441.         -630, -910, -1638, -4096

  442.     };

  443. 

  444.     int angle = intra_pred_angle[mode - 2];

  445.     pixel ref_array[3 * MAX_TB_SIZE + 4];

  446.     pixel *ref_tmp = ref_array + size;

  447.     const pixel *ref;

  448.     int last = (size * angle) >> 5;

  449. 

  450.     if (mode >= 18) {

  451.         ref = top - 1;

  452.         if (angle < 0 && last < -1) {

  453.             for (x = 0; x <= size; x += 4)

  454.                 AV_WN4P(&ref_tmp[x], AV_RN4P(&top[x - 1]));

  455.             for (x = last; x <= -1; x++)

  456.                 ref_tmp[x] = left[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];

  457.             ref = ref_tmp;

  458.         }

  459. 

  460.         for (y = 0; y < size; y++) {

  461.             int idx  = ((y + 1) * angle) >> 5;

  462.             int fact = ((y + 1) * angle) & 31;

  463.             if (fact) {

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

  465.                     POS(x    , y) = ((32 - fact) * ref[x + idx + 1] +

  466.                                            fact  * ref[x + idx + 2] + 16) >> 5;

  467.                     POS(x + 1, y) = ((32 - fact) * ref[x + 1 + idx + 1] +

  468.                                            fact  * ref[x + 1 + idx + 2] + 16) >> 5;

  469.                     POS(x + 2, y) = ((32 - fact) * ref[x + 2 + idx + 1] +

  470.                                            fact  * ref[x + 2 + idx + 2] + 16) >> 5;

  471.                     POS(x + 3, y) = ((32 - fact) * ref[x + 3 + idx + 1] +

  472.                                            fact  * ref[x + 3 + idx + 2] + 16) >> 5;

  473.                 }

  474.             } else {

  475.                 for (x = 0; x < size; x += 4)

  476.                     AV_WN4P(&POS(x, y), AV_RN4P(&ref[x + idx + 1]));

  477.             }

  478.         }

  479.         if (mode == 26 && c_idx == 0 && size < 32) {

  480.             for (y = 0; y < size; y++)

  481.                 POS(0, y) = av_clip_pixel(top[0] + ((left[y] - left[-1]) >> 1));

  482.         }

  483.     } else {

  484.         ref = left - 1;

  485.         if (angle < 0 && last < -1) {

  486.             for (x = 0; x <= size; x += 4)

  487.                 AV_WN4P(&ref_tmp[x], AV_RN4P(&left[x - 1]));

  488.             for (x = last; x <= -1; x++)

  489.                 ref_tmp[x] = top[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];

  490.             ref = ref_tmp;

  491.         }

  492. 

  493.         for (x = 0; x < size; x++) {

  494.             int idx  = ((x + 1) * angle) >> 5;

  495.             int fact = ((x + 1) * angle) & 31;

  496.             if (fact) {

  497.                 for (y = 0; y < size; y++) {

  498.                     POS(x, y) = ((32 - fact) * ref[y + idx + 1] +

  499.                                        fact  * ref[y + idx + 2] + 16) >> 5;

  500.                 }

  501.             } else {

  502.                 for (y = 0; y < size; y++)

  503.                     POS(x, y) = ref[y + idx + 1];

  504.             }

  505.         }

  506.         if (mode == 10 && c_idx == 0 && size < 32) {

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

  508.                 POS(x,     0) = av_clip_pixel(left[0] + ((top[x    ] - top[-1]) >> 1));

  509.                 POS(x + 1, 0) = av_clip_pixel(left[0] + ((top[x + 1] - top[-1]) >> 1));

  510.                 POS(x + 2, 0) = av_clip_pixel(left[0] + ((top[x + 2] - top[-1]) >> 1));

  511.                 POS(x + 3, 0) = av_clip_pixel(left[0] + ((top[x + 3] - top[-1]) >> 1));

  512.             }

  513.         }

  514.     }

  515. }

  516. 

  517. static void FUNC(pred_angular_0)(uint8_t *src, const uint8_t *top,

  518.                                  const uint8_t *left,

  519.                                  ptrdiff_t stride, int c_idx, int mode)

  520. {

  521.     FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 2);

  522. }

  523. 

  524. static void FUNC(pred_angular_1)(uint8_t *src, const uint8_t *top,

  525.                                  const uint8_t *left,

  526.                                  ptrdiff_t stride, int c_idx, int mode)

  527. {

  528.     FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 3);

  529. }

  530. 

  531. static void FUNC(pred_angular_2)(uint8_t *src, const uint8_t *top,

  532.                                  const uint8_t *left,

  533.                                  ptrdiff_t stride, int c_idx, int mode)

  534. {

  535.     FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 4);

  536. }

  537. 

  538. static void FUNC(pred_angular_3)(uint8_t *src, const uint8_t *top,

  539.                                  const uint8_t *left,

  540.                                  ptrdiff_t stride, int c_idx, int mode)

  541. {

  542.     FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 5);

  543. }

  544. 

  545. #undef EXTEND_LEFT_CIP

  546. #undef EXTEND_RIGHT_CIP

  547. #undef EXTEND_UP_CIP

  548. #undef EXTEND_DOWN_CIP

  549. #undef IS_INTRA

  550. #undef MVF_PU

  551. #undef MVF

  552. #undef PU

  553. #undef EXTEND

  554. #undef MIN_TB_ADDR_ZS

  555. #undef POS