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.
41838 Commits
32 Branches
1.0GB
Tree: 3058872c29
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3058872c29'
${ noResults }
FFmpeg/libavcodec/cavs.c

827 lines
29KB
Raw Blame History 

  1. /*

  2.  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.

  3.  * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>

  4.  *

  5.  * This file is part of Libav.

  6.  *

  7.  * Libav is free software; you can redistribute it and/or

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

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

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

  11.  *

  12.  * Libav is distributed in the hope that it will be useful,

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

  18.  * License along with Libav; if not, write to the Free Software

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

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder

  25.  * @author Stefan Gehrer <stefan.gehrer@gmx.de>

  26.  */

  27. 

  28. #include "avcodec.h"

  29. #include "get_bits.h"

  30. #include "golomb.h"

  31. #include "h264chroma.h"

  32. #include "idctdsp.h"

  33. #include "internal.h"

  34. #include "mathops.h"

  35. #include "qpeldsp.h"

  36. #include "cavs.h"

  37. 

  38. static const uint8_t alpha_tab[64] = {

  39.      0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  2,  2,  2,  3,  3,

  40.      4,  4,  5,  5,  6,  7,  8,  9, 10, 11, 12, 13, 15, 16, 18, 20,

  41.     22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,

  42.     46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64

  43. };

  44. 

  45. static const uint8_t beta_tab[64] = {

  46.      0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,

  47.      2,  2,  3,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,  5,  6,  6,

  48.      6,  7,  7,  7,  8,  8,  8,  9,  9, 10, 10, 11, 11, 12, 13, 14,

  49.     15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27

  50. };

  51. 

  52. static const uint8_t tc_tab[64] = {

  53.     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,

  54.     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,

  55.     2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,

  56.     5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9

  57. };

  58. 

  59. /** mark block as unavailable, i.e. out of picture

  60.  *  or not yet decoded */

  61. static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL };

  62. 

  63. static const int8_t left_modifier_l[8] = {  0, -1,  6, -1, -1, 7, 6, 7 };

  64. static const int8_t top_modifier_l[8]  = { -1,  1,  5, -1, -1, 5, 7, 7 };

  65. static const int8_t left_modifier_c[7] = {  5, -1,  2, -1,  6, 5, 6 };

  66. static const int8_t top_modifier_c[7]  = {  4,  1, -1, -1,  4, 6, 6 };

  67. 

  68. /*****************************************************************************

  69.  *

  70.  * in-loop deblocking filter

  71.  *

  72.  ****************************************************************************/

  73. 

  74. static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b)

  75. {

  76.     if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))

  77.         return 2;

  78.     if ((abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4))

  79.         return 1;

  80.     if (b) {

  81.         mvP += MV_BWD_OFFS;

  82.         mvQ += MV_BWD_OFFS;

  83.         if ((abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4))

  84.             return 1;

  85.     } else {

  86.         if (mvP->ref != mvQ->ref)

  87.             return 1;

  88.     }

  89.     return 0;

  90. }

  91. 

  92. #define SET_PARAMS                                                \

  93.     alpha = alpha_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];  \

  94.     beta  =  beta_tab[av_clip_uintp2(qp_avg + h->beta_offset,  6)];  \

  95.     tc    =    tc_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];

  96. 

  97. /**

  98.  * in-loop deblocking filter for a single macroblock

  99.  *

  100.  * boundary strength (bs) mapping:

  101.  *

  102.  * --4---5--

  103.  * 0   2   |

  104.  * | 6 | 7 |

  105.  * 1   3   |

  106.  * ---------

  107.  *

  108.  */

  109. void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type)

  110. {

  111.     uint8_t bs[8];

  112.     int qp_avg, alpha, beta, tc;

  113.     int i;

  114. 

  115.     /* save un-deblocked lines */

  116.     h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15];

  117.     h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8];

  118.     h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8];

  119.     memcpy(&h->top_border_y[h->mbx * 16],     h->cy + 15 * h->l_stride, 16);

  120.     memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu +  7 * h->c_stride, 8);

  121.     memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv +  7 * h->c_stride, 8);

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

  123.         h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride);

  124.         h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride);

  125.         h->left_border_u[i + 1]     = *(h->cu + 7  +  i          * h->c_stride);

  126.         h->left_border_v[i + 1]     = *(h->cv + 7  +  i          * h->c_stride);

  127.     }

  128.     if (!h->loop_filter_disable) {

  129.         /* determine bs */

  130.         if (mb_type == I_8X8)

  131.             memset(bs, 2, 8);

  132.         else {

  133.             memset(bs, 0, 8);

  134.             if (ff_cavs_partition_flags[mb_type] & SPLITV) {

  135.                 bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);

  136.                 bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);

  137.             }

  138.             if (ff_cavs_partition_flags[mb_type] & SPLITH) {

  139.                 bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);

  140.                 bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);

  141.             }

  142.             bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);

  143.             bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);

  144.             bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);

  145.             bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);

  146.         }

  147.         if (AV_RN64(bs)) {

  148.             if (h->flags & A_AVAIL) {

  149.                 qp_avg = (h->qp + h->left_qp + 1) >> 1;

  150.                 SET_PARAMS;

  151.                 h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]);

  152.                 h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]);

  153.                 h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]);

  154.             }

  155.             qp_avg = h->qp;

  156.             SET_PARAMS;

  157.             h->cdsp.cavs_filter_lv(h->cy + 8,               h->l_stride, alpha, beta, tc, bs[2], bs[3]);

  158.             h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]);

  159. 

  160.             if (h->flags & B_AVAIL) {

  161.                 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;

  162.                 SET_PARAMS;

  163.                 h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]);

  164.                 h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]);

  165.                 h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]);

  166.             }

  167.         }

  168.     }

  169.     h->left_qp        = h->qp;

  170.     h->top_qp[h->mbx] = h->qp;

  171. }

  172. 

  173. #undef SET_PARAMS

  174. 

  175. /*****************************************************************************

  176.  *

  177.  * spatial intra prediction

  178.  *

  179.  ****************************************************************************/

  180. 

  181. void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top,

  182.                                   uint8_t **left, int block)

  183. {

  184.     int i;

  185. 

  186.     switch (block) {

  187.     case 0:

  188.         *left               = h->left_border_y;

  189.         h->left_border_y[0] = h->left_border_y[1];

  190.         memset(&h->left_border_y[17], h->left_border_y[16], 9);

  191.         memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16);

  192.         top[17] = top[16];

  193.         top[0]  = top[1];

  194.         if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL))

  195.             h->left_border_y[0] = top[0] = h->topleft_border_y;

  196.         break;

  197.     case 1:

  198.         *left = h->intern_border_y;

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

  200.             h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride);

  201.         memset(&h->intern_border_y[9], h->intern_border_y[8], 9);

  202.         h->intern_border_y[0] = h->intern_border_y[1];

  203.         memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8);

  204.         if (h->flags & C_AVAIL)

  205.             memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8);

  206.         else

  207.             memset(&top[9], top[8], 9);

  208.         top[17] = top[16];

  209.         top[0]  = top[1];

  210.         if (h->flags & B_AVAIL)

  211.             h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7];

  212.         break;

  213.     case 2:

  214.         *left = &h->left_border_y[8];

  215.         memcpy(&top[1], h->cy + 7 * h->l_stride, 16);

  216.         top[17] = top[16];

  217.         top[0]  = top[1];

  218.         if (h->flags & A_AVAIL)

  219.             top[0] = h->left_border_y[8];

  220.         break;

  221.     case 3:

  222.         *left = &h->intern_border_y[8];

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

  224.             h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride);

  225.         memset(&h->intern_border_y[17], h->intern_border_y[16], 9);

  226.         memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9);

  227.         memset(&top[9], top[8], 9);

  228.         break;

  229.     }

  230. }

  231. 

  232. void ff_cavs_load_intra_pred_chroma(AVSContext *h)

  233. {

  234.     /* extend borders by one pixel */

  235.     h->left_border_u[9]              = h->left_border_u[8];

  236.     h->left_border_v[9]              = h->left_border_v[8];

  237.     h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8];

  238.     h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8];

  239.     if (h->mbx && h->mby) {

  240.         h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u;

  241.         h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v;

  242.     } else {

  243.         h->left_border_u[0]          = h->left_border_u[1];

  244.         h->left_border_v[0]          = h->left_border_v[1];

  245.         h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1];

  246.         h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1];

  247.     }

  248. }

  249. 

  250. static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  251. {

  252.     int y;

  253.     uint64_t a = AV_RN64(&top[1]);

  254.     for (y = 0; y < 8; y++)

  255.         *((uint64_t *)(d + y * stride)) = a;

  256. }

  257. 

  258. static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  259. {

  260.     int y;

  261.     uint64_t a;

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

  263.         a = left[y + 1] * 0x0101010101010101ULL;

  264.         *((uint64_t *)(d + y * stride)) = a;

  265.     }

  266. }

  267. 

  268. static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  269. {

  270.     int y;

  271.     uint64_t a = 0x8080808080808080ULL;

  272.     for (y = 0; y < 8; y++)

  273.         *((uint64_t *)(d + y * stride)) = a;

  274. }

  275. 

  276. static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  277. {

  278.     int x, y, ia;

  279.     int ih = 0;

  280.     int iv = 0;

  281.     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;

  282. 

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

  284.         ih += (x + 1) *  (top[5 + x] -  top[3 - x]);

  285.         iv += (x + 1) * (left[5 + x] - left[3 - x]);

  286.     }

  287.     ia = (top[8] + left[8]) << 4;

  288.     ih = (17 * ih + 16) >> 5;

  289.     iv = (17 * iv + 16) >> 5;

  290.     for (y = 0; y < 8; y++)

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

  292.             d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5];

  293. }

  294. 

  295. #define LOWPASS(ARRAY, INDEX)                                           \

  296.     ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2)

  297. 

  298. static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  299. {

  300.     int x, y;

  301.     for (y = 0; y < 8; y++)

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

  303.             d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1;

  304. }

  305. 

  306. static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  307. {

  308.     int x, y;

  309.     for (y = 0; y < 8; y++)

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

  311.             d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1;

  312. }

  313. 

  314. static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  315. {

  316.     int x, y;

  317.     for (y = 0; y < 8; y++)

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

  319.             if (x == y)

  320.                 d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2;

  321.             else if (x > y)

  322.                 d[y * stride + x] = LOWPASS(top, x - y);

  323.             else

  324.                 d[y * stride + x] = LOWPASS(left, y - x);

  325. }

  326. 

  327. static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  328. {

  329.     int x, y;

  330.     for (y = 0; y < 8; y++)

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

  332.             d[y * stride + x] = LOWPASS(left, y + 1);

  333. }

  334. 

  335. static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, int stride)

  336. {

  337.     int x, y;

  338.     for (y = 0; y < 8; y++)

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

  340.             d[y * stride + x] = LOWPASS(top, x + 1);

  341. }

  342. 

  343. #undef LOWPASS

  344. 

  345. static inline void modify_pred(const int8_t *mod_table, int *mode)

  346. {

  347.     *mode = mod_table[*mode];

  348.     if (*mode < 0) {

  349.         av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");

  350.         *mode = 0;

  351.     }

  352. }

  353. 

  354. void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv)

  355. {

  356.     /* save pred modes before they get modified */

  357.     h->pred_mode_Y[3]             = h->pred_mode_Y[5];

  358.     h->pred_mode_Y[6]             = h->pred_mode_Y[8];

  359.     h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7];

  360.     h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8];

  361. 

  362.     /* modify pred modes according to availability of neighbour samples */

  363.     if (!(h->flags & A_AVAIL)) {

  364.         modify_pred(left_modifier_l, &h->pred_mode_Y[4]);

  365.         modify_pred(left_modifier_l, &h->pred_mode_Y[7]);

  366.         modify_pred(left_modifier_c, pred_mode_uv);

  367.     }

  368.     if (!(h->flags & B_AVAIL)) {

  369.         modify_pred(top_modifier_l, &h->pred_mode_Y[4]);

  370.         modify_pred(top_modifier_l, &h->pred_mode_Y[5]);

  371.         modify_pred(top_modifier_c, pred_mode_uv);

  372.     }

  373. }

  374. 

  375. /*****************************************************************************

  376.  *

  377.  * motion compensation

  378.  *

  379.  ****************************************************************************/

  380. 

  381. static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,

  382.                                int delta, int list, uint8_t *dest_y,

  383.                                uint8_t *dest_cb, uint8_t *dest_cr,

  384.                                int src_x_offset, int src_y_offset,

  385.                                qpel_mc_func *qpix_op,

  386.                                h264_chroma_mc_func chroma_op, cavs_vector *mv)

  387. {

  388.     const int mx         = mv->x + src_x_offset * 8;

  389.     const int my         = mv->y + src_y_offset * 8;

  390.     const int luma_xy    = (mx & 3) + ((my & 3) << 2);

  391.     uint8_t *src_y       = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride;

  392.     uint8_t *src_cb      = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride;

  393.     uint8_t *src_cr      = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride;

  394.     int extra_width      = 0;

  395.     int extra_height     = extra_width;

  396.     const int full_mx    = mx >> 2;

  397.     const int full_my    = my >> 2;

  398.     const int pic_width  = 16 * h->mb_width;

  399.     const int pic_height = 16 * h->mb_height;

  400.     int emu = 0;

  401. 

  402.     if (!pic->data[0])

  403.         return;

  404.     if (mx & 7)

  405.         extra_width  -= 3;

  406.     if (my & 7)

  407.         extra_height -= 3;

  408. 

  409.     if (full_mx < 0 - extra_width ||

  410.         full_my < 0 - extra_height ||

  411.         full_mx + 16 /* FIXME */ > pic_width + extra_width ||

  412.         full_my + 16 /* FIXME */ > pic_height + extra_height) {

  413.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer,

  414.                                  src_y - 2 - 2 * h->l_stride,

  415.                                  h->l_stride, h->l_stride,

  416.                                  16 + 5, 16 + 5 /* FIXME */,

  417.                                  full_mx - 2, full_my - 2,

  418.                                  pic_width, pic_height);

  419.         src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride;

  420.         emu   = 1;

  421.     }

  422. 

  423.     // FIXME try variable height perhaps?

  424.     qpix_op[luma_xy](dest_y, src_y, h->l_stride);

  425. 

  426.     if (emu) {

  427.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,

  428.                                  h->c_stride, h->c_stride,

  429.                                  9, 9 /* FIXME */,

  430.                                  mx >> 3, my >> 3,

  431.                                  pic_width >> 1, pic_height >> 1);

  432.         src_cb = h->edge_emu_buffer;

  433.     }

  434.     chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7);

  435. 

  436.     if (emu) {

  437.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,

  438.                                  h->c_stride, h->c_stride,

  439.                                  9, 9 /* FIXME */,

  440.                                  mx >> 3, my >> 3,

  441.                                  pic_width >> 1, pic_height >> 1);

  442.         src_cr = h->edge_emu_buffer;

  443.     }

  444.     chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7);

  445. }

  446. 

  447. static inline void mc_part_std(AVSContext *h, int chroma_height, int delta,

  448.                                uint8_t *dest_y,

  449.                                uint8_t *dest_cb,

  450.                                uint8_t *dest_cr,

  451.                                int x_offset, int y_offset,

  452.                                qpel_mc_func *qpix_put,

  453.                                h264_chroma_mc_func chroma_put,

  454.                                qpel_mc_func *qpix_avg,

  455.                                h264_chroma_mc_func chroma_avg,

  456.                                cavs_vector *mv)

  457. {

  458.     qpel_mc_func *qpix_op =  qpix_put;

  459.     h264_chroma_mc_func chroma_op = chroma_put;

  460. 

  461.     dest_y   += x_offset * 2 + y_offset * h->l_stride * 2;

  462.     dest_cb  += x_offset     + y_offset * h->c_stride;

  463.     dest_cr  += x_offset     + y_offset * h->c_stride;

  464.     x_offset += 8 * h->mbx;

  465.     y_offset += 8 * h->mby;

  466. 

  467.     if (mv->ref >= 0) {

  468.         AVFrame *ref = h->DPB[mv->ref].f;

  469.         mc_dir_part(h, ref, chroma_height, delta, 0,

  470.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  471.                     qpix_op, chroma_op, mv);

  472. 

  473.         qpix_op   = qpix_avg;

  474.         chroma_op = chroma_avg;

  475.     }

  476. 

  477.     if ((mv + MV_BWD_OFFS)->ref >= 0) {

  478.         AVFrame *ref = h->DPB[0].f;

  479.         mc_dir_part(h, ref, chroma_height, delta, 1,

  480.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  481.                     qpix_op, chroma_op, mv + MV_BWD_OFFS);

  482.     }

  483. }

  484. 

  485. void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type)

  486. {

  487.     if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16

  488.         mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0,

  489.                     h->cdsp.put_cavs_qpel_pixels_tab[0],

  490.                     h->h264chroma.put_h264_chroma_pixels_tab[0],

  491.                     h->cdsp.avg_cavs_qpel_pixels_tab[0],

  492.                     h->h264chroma.avg_h264_chroma_pixels_tab[0],

  493.                     &h->mv[MV_FWD_X0]);

  494.     } else {

  495.         mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0,

  496.                     h->cdsp.put_cavs_qpel_pixels_tab[1],

  497.                     h->h264chroma.put_h264_chroma_pixels_tab[1],

  498.                     h->cdsp.avg_cavs_qpel_pixels_tab[1],

  499.                     h->h264chroma.avg_h264_chroma_pixels_tab[1],

  500.                     &h->mv[MV_FWD_X0]);

  501.         mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0,

  502.                     h->cdsp.put_cavs_qpel_pixels_tab[1],

  503.                     h->h264chroma.put_h264_chroma_pixels_tab[1],

  504.                     h->cdsp.avg_cavs_qpel_pixels_tab[1],

  505.                     h->h264chroma.avg_h264_chroma_pixels_tab[1],

  506.                     &h->mv[MV_FWD_X1]);

  507.         mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4,

  508.                     h->cdsp.put_cavs_qpel_pixels_tab[1],

  509.                     h->h264chroma.put_h264_chroma_pixels_tab[1],

  510.                     h->cdsp.avg_cavs_qpel_pixels_tab[1],

  511.                     h->h264chroma.avg_h264_chroma_pixels_tab[1],

  512.                     &h->mv[MV_FWD_X2]);

  513.         mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4,

  514.                     h->cdsp.put_cavs_qpel_pixels_tab[1],

  515.                     h->h264chroma.put_h264_chroma_pixels_tab[1],

  516.                     h->cdsp.avg_cavs_qpel_pixels_tab[1],

  517.                     h->h264chroma.avg_h264_chroma_pixels_tab[1],

  518.                     &h->mv[MV_FWD_X3]);

  519.     }

  520. }

  521. 

  522. /*****************************************************************************

  523.  *

  524.  * motion vector prediction

  525.  *

  526.  ****************************************************************************/

  527. 

  528. static inline void scale_mv(AVSContext *h, int *d_x, int *d_y,

  529.                             cavs_vector *src, int distp)

  530. {

  531.     int den = h->scale_den[src->ref];

  532. 

  533.     *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9;

  534.     *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9;

  535. }

  536. 

  537. static inline void mv_pred_median(AVSContext *h,

  538.                                   cavs_vector *mvP,

  539.                                   cavs_vector *mvA,

  540.                                   cavs_vector *mvB,

  541.                                   cavs_vector *mvC)

  542. {

  543.     int ax, ay, bx, by, cx, cy;

  544.     int len_ab, len_bc, len_ca, len_mid;

  545. 

  546.     /* scale candidates according to their temporal span */

  547.     scale_mv(h, &ax, &ay, mvA, mvP->dist);

  548.     scale_mv(h, &bx, &by, mvB, mvP->dist);

  549.     scale_mv(h, &cx, &cy, mvC, mvP->dist);

  550.     /* find the geometrical median of the three candidates */

  551.     len_ab  = abs(ax - bx) + abs(ay - by);

  552.     len_bc  = abs(bx - cx) + abs(by - cy);

  553.     len_ca  = abs(cx - ax) + abs(cy - ay);

  554.     len_mid = mid_pred(len_ab, len_bc, len_ca);

  555.     if (len_mid == len_ab) {

  556.         mvP->x = cx;

  557.         mvP->y = cy;

  558.     } else if (len_mid == len_bc) {

  559.         mvP->x = ax;

  560.         mvP->y = ay;

  561.     } else {

  562.         mvP->x = bx;

  563.         mvP->y = by;

  564.     }

  565. }

  566. 

  567. void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,

  568.                 enum cavs_mv_pred mode, enum cavs_block size, int ref)

  569. {

  570.     cavs_vector *mvP = &h->mv[nP];

  571.     cavs_vector *mvA = &h->mv[nP-1];

  572.     cavs_vector *mvB = &h->mv[nP-4];

  573.     cavs_vector *mvC = &h->mv[nC];

  574.     const cavs_vector *mvP2 = NULL;

  575. 

  576.     mvP->ref  = ref;

  577.     mvP->dist = h->dist[mvP->ref];

  578.     if (mvC->ref == NOT_AVAIL)

  579.         mvC = &h->mv[nP - 5];  // set to top-left (mvD)

  580.     if (mode == MV_PRED_PSKIP &&

  581.         (mvA->ref == NOT_AVAIL ||

  582.          mvB->ref == NOT_AVAIL ||

  583.          (mvA->x | mvA->y | mvA->ref) == 0 ||

  584.          (mvB->x | mvB->y | mvB->ref) == 0)) {

  585.         mvP2 = &un_mv;

  586.     /* if there is only one suitable candidate, take it */

  587.     } else if (mvA->ref >= 0 && mvB->ref < 0  && mvC->ref < 0) {

  588.         mvP2 = mvA;

  589.     } else if (mvA->ref < 0  && mvB->ref >= 0 && mvC->ref < 0) {

  590.         mvP2 = mvB;

  591.     } else if (mvA->ref < 0  && mvB->ref < 0  && mvC->ref >= 0) {

  592.         mvP2 = mvC;

  593.     } else if (mode == MV_PRED_LEFT     && mvA->ref == ref) {

  594.         mvP2 = mvA;

  595.     } else if (mode == MV_PRED_TOP      && mvB->ref == ref) {

  596.         mvP2 = mvB;

  597.     } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) {

  598.         mvP2 = mvC;

  599.     }

  600.     if (mvP2) {

  601.         mvP->x = mvP2->x;

  602.         mvP->y = mvP2->y;

  603.     } else

  604.         mv_pred_median(h, mvP, mvA, mvB, mvC);

  605. 

  606.     if (mode < MV_PRED_PSKIP) {

  607.         mvP->x += get_se_golomb(&h->gb);

  608.         mvP->y += get_se_golomb(&h->gb);

  609.     }

  610.     set_mvs(mvP, size);

  611. }

  612. 

  613. /*****************************************************************************

  614.  *

  615.  * macroblock level

  616.  *

  617.  ****************************************************************************/

  618. 

  619. /**

  620.  * initialise predictors for motion vectors and intra prediction

  621.  */

  622. void ff_cavs_init_mb(AVSContext *h)

  623. {

  624.     int i;

  625. 

  626.     /* copy predictors from top line (MB B and C) into cache */

  627.     for (i = 0; i < 3; i++) {

  628.         h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i];

  629.         h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i];

  630.     }

  631.     h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0];

  632.     h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1];

  633.     /* clear top predictors if MB B is not available */

  634.     if (!(h->flags & B_AVAIL)) {

  635.         h->mv[MV_FWD_B2]  = un_mv;

  636.         h->mv[MV_FWD_B3]  = un_mv;

  637.         h->mv[MV_BWD_B2]  = un_mv;

  638.         h->mv[MV_BWD_B3]  = un_mv;

  639.         h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;

  640.         h->flags         &= ~(C_AVAIL | D_AVAIL);

  641.     } else if (h->mbx) {

  642.         h->flags |= D_AVAIL;

  643.     }

  644.     if (h->mbx == h->mb_width - 1) // MB C not available

  645.         h->flags &= ~C_AVAIL;

  646.     /* clear top-right predictors if MB C is not available */

  647.     if (!(h->flags & C_AVAIL)) {

  648.         h->mv[MV_FWD_C2] = un_mv;

  649.         h->mv[MV_BWD_C2] = un_mv;

  650.     }

  651.     /* clear top-left predictors if MB D is not available */

  652.     if (!(h->flags & D_AVAIL)) {

  653.         h->mv[MV_FWD_D3] = un_mv;

  654.         h->mv[MV_BWD_D3] = un_mv;

  655.     }

  656. }

  657. 

  658. /**

  659.  * save predictors for later macroblocks and increase

  660.  * macroblock address

  661.  * @return 0 if end of frame is reached, 1 otherwise

  662.  */

  663. int ff_cavs_next_mb(AVSContext *h)

  664. {

  665.     int i;

  666. 

  667.     h->flags |= A_AVAIL;

  668.     h->cy    += 16;

  669.     h->cu    += 8;

  670.     h->cv    += 8;

  671.     /* copy mvs as predictors to the left */

  672.     for (i = 0; i <= 20; i += 4)

  673.         h->mv[i] = h->mv[i + 2];

  674.     /* copy bottom mvs from cache to top line */

  675.     h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2];

  676.     h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3];

  677.     h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2];

  678.     h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3];

  679.     /* next MB address */

  680.     h->mbidx++;

  681.     h->mbx++;

  682.     if (h->mbx == h->mb_width) { // New mb line

  683.         h->flags = B_AVAIL | C_AVAIL;

  684.         /* clear left pred_modes */

  685.         h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;

  686.         /* clear left mv predictors */

  687.         for (i = 0; i <= 20; i += 4)

  688.             h->mv[i] = un_mv;

  689.         h->mbx = 0;

  690.         h->mby++;

  691.         /* re-calculate sample pointers */

  692.         h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride;

  693.         h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride;

  694.         h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride;

  695.         if (h->mby == h->mb_height) { // Frame end

  696.             return 0;

  697.         }

  698.     }

  699.     return 1;

  700. }

  701. 

  702. /*****************************************************************************

  703.  *

  704.  * frame level

  705.  *

  706.  ****************************************************************************/

  707. 

  708. void ff_cavs_init_pic(AVSContext *h)

  709. {

  710.     int i;

  711. 

  712.     /* clear some predictors */

  713.     for (i = 0; i <= 20; i += 4)

  714.         h->mv[i] = un_mv;

  715.     h->mv[MV_BWD_X0] = ff_cavs_dir_mv;

  716.     set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);

  717.     h->mv[MV_FWD_X0] = ff_cavs_dir_mv;

  718.     set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);

  719.     h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;

  720.     h->cy             = h->cur.f->data[0];

  721.     h->cu             = h->cur.f->data[1];

  722.     h->cv             = h->cur.f->data[2];

  723.     h->l_stride       = h->cur.f->linesize[0];

  724.     h->c_stride       = h->cur.f->linesize[1];

  725.     h->luma_scan[2]   = 8 * h->l_stride;

  726.     h->luma_scan[3]   = 8 * h->l_stride + 8;

  727.     h->mbx            = h->mby = h->mbidx = 0;

  728.     h->flags          = 0;

  729. }

  730. 

  731. /*****************************************************************************

  732.  *

  733.  * headers and interface

  734.  *

  735.  ****************************************************************************/

  736. 

  737. /**

  738.  * some predictions require data from the top-neighbouring macroblock.

  739.  * this data has to be stored for one complete row of macroblocks

  740.  * and this storage space is allocated here

  741.  */

  742. void ff_cavs_init_top_lines(AVSContext *h)

  743. {

  744.     /* alloc top line of predictors */

  745.     h->top_qp       = av_mallocz(h->mb_width);

  746.     h->top_mv[0]    = av_mallocz((h->mb_width * 2 + 1) * sizeof(cavs_vector));

  747.     h->top_mv[1]    = av_mallocz((h->mb_width * 2 + 1) * sizeof(cavs_vector));

  748.     h->top_pred_Y   = av_mallocz(h->mb_width * 2 * sizeof(*h->top_pred_Y));

  749.     h->top_border_y = av_mallocz((h->mb_width + 1) * 16);

  750.     h->top_border_u = av_mallocz(h->mb_width * 10);

  751.     h->top_border_v = av_mallocz(h->mb_width * 10);

  752. 

  753.     /* alloc space for co-located MVs and types */

  754.     h->col_mv        = av_mallocz(h->mb_width * h->mb_height * 4 *

  755.                                   sizeof(cavs_vector));

  756.     h->col_type_base = av_mallocz(h->mb_width * h->mb_height);

  757.     h->block         = av_mallocz(64 * sizeof(int16_t));

  758. }

  759. 

  760. av_cold int ff_cavs_init(AVCodecContext *avctx)

  761. {

  762.     AVSContext *h = avctx->priv_data;

  763. 

  764.     ff_blockdsp_init(&h->bdsp, avctx);

  765.     ff_h264chroma_init(&h->h264chroma, 8);

  766.     ff_idctdsp_init(&h->idsp, avctx);

  767.     ff_videodsp_init(&h->vdsp, 8);

  768.     ff_cavsdsp_init(&h->cdsp, avctx);

  769.     ff_init_scantable_permutation(h->idsp.idct_permutation,

  770.                                   h->cdsp.idct_perm);

  771.     ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct);

  772. 

  773.     h->avctx       = avctx;

  774.     avctx->pix_fmt = AV_PIX_FMT_YUV420P;

  775. 

  776.     h->cur.f    = av_frame_alloc();

  777.     h->DPB[0].f = av_frame_alloc();

  778.     h->DPB[1].f = av_frame_alloc();

  779.     if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f) {

  780.         ff_cavs_end(avctx);

  781.         return AVERROR(ENOMEM);

  782.     }

  783. 

  784.     h->luma_scan[0]                     = 0;

  785.     h->luma_scan[1]                     = 8;

  786.     h->intra_pred_l[INTRA_L_VERT]       = intra_pred_vert;

  787.     h->intra_pred_l[INTRA_L_HORIZ]      = intra_pred_horiz;

  788.     h->intra_pred_l[INTRA_L_LP]         = intra_pred_lp;

  789.     h->intra_pred_l[INTRA_L_DOWN_LEFT]  = intra_pred_down_left;

  790.     h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;

  791.     h->intra_pred_l[INTRA_L_LP_LEFT]    = intra_pred_lp_left;

  792.     h->intra_pred_l[INTRA_L_LP_TOP]     = intra_pred_lp_top;

  793.     h->intra_pred_l[INTRA_L_DC_128]     = intra_pred_dc_128;

  794.     h->intra_pred_c[INTRA_C_LP]         = intra_pred_lp;

  795.     h->intra_pred_c[INTRA_C_HORIZ]      = intra_pred_horiz;

  796.     h->intra_pred_c[INTRA_C_VERT]       = intra_pred_vert;

  797.     h->intra_pred_c[INTRA_C_PLANE]      = intra_pred_plane;

  798.     h->intra_pred_c[INTRA_C_LP_LEFT]    = intra_pred_lp_left;

  799.     h->intra_pred_c[INTRA_C_LP_TOP]     = intra_pred_lp_top;

  800.     h->intra_pred_c[INTRA_C_DC_128]     = intra_pred_dc_128;

  801.     h->mv[7]                            = un_mv;

  802.     h->mv[19]                           = un_mv;

  803.     return 0;

  804. }

  805. 

  806. av_cold int ff_cavs_end(AVCodecContext *avctx)

  807. {

  808.     AVSContext *h = avctx->priv_data;

  809. 

  810.     av_frame_free(&h->cur.f);

  811.     av_frame_free(&h->DPB[0].f);

  812.     av_frame_free(&h->DPB[1].f);

  813. 

  814.     av_free(h->top_qp);

  815.     av_free(h->top_mv[0]);

  816.     av_free(h->top_mv[1]);

  817.     av_free(h->top_pred_Y);

  818.     av_free(h->top_border_y);

  819.     av_free(h->top_border_u);

  820.     av_free(h->top_border_v);

  821.     av_free(h->col_mv);

  822.     av_free(h->col_type_base);

  823.     av_free(h->block);

  824.     av_freep(&h->edge_emu_buffer);

  825.     return 0;

  826. }