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.
38308 Commits
32 Branches
428MB
Tree: ce1e8045e0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ce1e8045e0'
${ noResults }
FFmpeg/libavcodec/cavs.c

789 lines
28KB
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 "mathops.h"

  33. #include "cavs.h"

  34. 

  35. static const uint8_t alpha_tab[64] = {

  36.    0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  2,  2,  2,  3,  3,

  37.    4,  4,  5,  5,  6,  7,  8,  9, 10, 11, 12, 13, 15, 16, 18, 20,

  38.   22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,

  39.   46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64

  40. };

  41. 

  42. static const uint8_t beta_tab[64] = {

  43.    0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,

  44.    2,  2,  3,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,  5,  6,  6,

  45.    6,  7,  7,  7,  8,  8,  8,  9,  9, 10, 10, 11, 11, 12, 13, 14,

  46.   15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27

  47. };

  48. 

  49. static const uint8_t tc_tab[64] = {

  50.   0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,

  51.   1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,

  52.   2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,

  53.   5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9

  54. };

  55. 

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

  57.     or not yet decoded */

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

  59. 

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

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

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

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

  64. 

  65. /*****************************************************************************

  66.  *

  67.  * in-loop deblocking filter

  68.  *

  69.  ****************************************************************************/

  70. 

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

  72. {

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

  74.         return 2;

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

  76.         return 1;

  77.     if (b) {

  78.         mvP += MV_BWD_OFFS;

  79.         mvQ += MV_BWD_OFFS;

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

  81.             return 1;

  82.     } else {

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

  84.             return 1;

  85.     }

  86.     return 0;

  87. }

  88. 

  89. #define SET_PARAMS                                                \

  90.     alpha = alpha_tab[av_clip(qp_avg + h->alpha_offset, 0, 63)];  \

  91.     beta  =  beta_tab[av_clip(qp_avg + h->beta_offset,  0, 63)];  \

  92.     tc    =    tc_tab[av_clip(qp_avg + h->alpha_offset, 0, 63)];

  93. 

  94. /**

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

  96.  *

  97.  * boundary strength (bs) mapping:

  98.  *

  99.  * --4---5--

  100.  * 0   2   |

  101.  * | 6 | 7 |

  102.  * 1   3   |

  103.  * ---------

  104.  *

  105.  */

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

  107. {

  108.     uint8_t bs[8];

  109.     int qp_avg, alpha, beta, tc;

  110.     int i;

  111. 

  112.     /* save un-deblocked lines */

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

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

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

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

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

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

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

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

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

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

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

  124.     }

  125.     if (!h->loop_filter_disable) {

  126.         /* determine bs */

  127.         if (mb_type == I_8X8)

  128.             memset(bs, 2, 8);

  129.         else{

  130.             memset(bs, 0, 8);

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

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

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

  134.             }

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

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

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

  138.             }

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

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

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

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

  143.         }

  144.         if (AV_RN64(bs)) {

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

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

  147.                 SET_PARAMS;

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

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

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

  151.             }

  152.             qp_avg = h->qp;

  153.             SET_PARAMS;

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

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

  156. 

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

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

  159.                 SET_PARAMS;

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

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

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

  163.             }

  164.         }

  165.     }

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

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

  168. }

  169. 

  170. #undef SET_PARAMS

  171. 

  172. /*****************************************************************************

  173.  *

  174.  * spatial intra prediction

  175.  *

  176.  ****************************************************************************/

  177. 

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

  179.                                   uint8_t **left, int block)

  180. {

  181.     int i;

  182. 

  183.     switch (block) {

  184.     case 0:

  185.         *left               = h->left_border_y;

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

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

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

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

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

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

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

  193.         break;

  194.     case 1:

  195.         *left = h->intern_border_y;

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

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

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

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

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

  201.         if (h->flags & C_AVAIL)

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

  203.         else

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

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

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

  207.         if (h->flags & B_AVAIL)

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

  209.         break;

  210.     case 2:

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

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

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

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

  215.         if (h->flags & A_AVAIL)

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

  217.         break;

  218.     case 3:

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

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

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

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

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

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

  225.         break;

  226.     }

  227. }

  228. 

  229. void ff_cavs_load_intra_pred_chroma(AVSContext *h)

  230. {

  231.     /* extend borders by one pixel */

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

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

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

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

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

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

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

  239.     } else {

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

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

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

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

  244.     }

  245. }

  246. 

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

  248. {

  249.     int y;

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

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

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

  253.     }

  254. }

  255. 

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

  257. {

  258.     int y;

  259.     uint64_t a;

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

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

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

  263.     }

  264. }

  265. 

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

  267. {

  268.     int y;

  269.     uint64_t a = 0x8080808080808080ULL;

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

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

  272. }

  273. 

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

  275. {

  276.     int x, y, ia;

  277.     int ih = 0;

  278.     int iv = 0;

  279.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  280. 

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

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

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

  284.     }

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

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

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

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

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

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

  291. }

  292. 

  293. #define LOWPASS(ARRAY,INDEX)                                            \

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

  295. 

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

  297. {

  298.     int x, y;

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

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

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

  302. }

  303. 

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

  305. {

  306.     int x, y;

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

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

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

  310. }

  311. 

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

  313. {

  314.     int x, y;

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

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

  317.             if (x == y)

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

  319.             else if (x > y)

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

  321.             else

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

  323. }

  324. 

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

  326. {

  327.     int x, y;

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

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

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

  331. }

  332. 

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

  334. {

  335.     int x, y;

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

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

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

  339. }

  340. 

  341. #undef LOWPASS

  342. 

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

  344. {

  345.     *mode = mod_table[*mode];

  346.     if (*mode < 0) {

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

  348.         *mode = 0;

  349.     }

  350. }

  351. 

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

  353. {

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

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

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

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

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

  359. 

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

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

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

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

  364.         modify_pred(left_modifier_c, pred_mode_uv);

  365.     }

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

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

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

  369.         modify_pred(top_modifier_c, pred_mode_uv);

  370.     }

  371. }

  372. 

  373. /*****************************************************************************

  374.  *

  375.  * motion compensation

  376.  *

  377.  ****************************************************************************/

  378. 

  379. static inline void mc_dir_part(AVSContext *h, AVFrame *pic,

  380.                                int chroma_height,int delta,int list,uint8_t *dest_y,

  381.                                uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,

  382.                                int src_y_offset,qpel_mc_func *qpix_op,

  383.                                h264_chroma_mc_func chroma_op,cavs_vector *mv)

  384. {

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

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

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

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

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

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

  391.     int extra_width = 0;

  392.     int extra_height= extra_width;

  393.     int emu=0;

  394.     const int full_mx= mx>>2;

  395.     const int full_my= my>>2;

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

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

  398. 

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

  400.         return;

  401.     if(mx&7) extra_width -= 3;

  402.     if(my&7) extra_height -= 3;

  403. 

  404.     if(   full_mx < 0-extra_width

  405.           || full_my < 0-extra_height

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

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

  408.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,

  409.                             16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

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

  411.         emu=1;

  412.     }

  413. 

  414.     qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?

  415. 

  416.     if(emu){

  417.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->c_stride,

  418.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  419.         src_cb= h->edge_emu_buffer;

  420.     }

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

  422. 

  423.     if(emu){

  424.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->c_stride,

  425.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  426.         src_cr= h->edge_emu_buffer;

  427.     }

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

  429. }

  430. 

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

  432.                                uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,

  433.                                int x_offset, int y_offset,qpel_mc_func *qpix_put,

  434.                                h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,

  435.                                h264_chroma_mc_func chroma_avg, cavs_vector *mv)

  436. {

  437.     qpel_mc_func *qpix_op=  qpix_put;

  438.     h264_chroma_mc_func chroma_op= chroma_put;

  439. 

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

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

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

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

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

  445. 

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

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

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

  449.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  450.                     qpix_op, chroma_op, mv);

  451. 

  452.         qpix_op=  qpix_avg;

  453.         chroma_op= chroma_avg;

  454.     }

  455. 

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

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

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

  459.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  460.                     qpix_op, chroma_op, mv+MV_BWD_OFFS);

  461.     }

  462. }

  463. 

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

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

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

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

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

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

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

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

  472.     }else{

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  497.     }

  498. }

  499. 

  500. /*****************************************************************************

  501.  *

  502.  * motion vector prediction

  503.  *

  504.  ****************************************************************************/

  505. 

  506. static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, cavs_vector *src, int distp) {

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

  508. 

  509.     *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;

  510.     *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;

  511. }

  512. 

  513. static inline void mv_pred_median(AVSContext *h, cavs_vector *mvP,

  514.                         cavs_vector *mvA, cavs_vector *mvB, cavs_vector *mvC) {

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

  516.     int len_ab, len_bc, len_ca, len_mid;

  517. 

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

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

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

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

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

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

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

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

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

  527.     if(len_mid == len_ab) {

  528.         mvP->x = cx;

  529.         mvP->y = cy;

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

  531.         mvP->x = ax;

  532.         mvP->y = ay;

  533.     } else {

  534.         mvP->x = bx;

  535.         mvP->y = by;

  536.     }

  537. }

  538. 

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

  540.                 enum cavs_mv_pred mode, enum cavs_block size, int ref) {

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

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

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

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

  545.     const cavs_vector *mvP2 = NULL;

  546. 

  547.     mvP->ref = ref;

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

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

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

  551.     if((mode == MV_PRED_PSKIP) &&

  552.        ((mvA->ref == NOT_AVAIL) || (mvB->ref == NOT_AVAIL) ||

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

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

  555.         mvP2 = &un_mv;

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

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

  558.         mvP2= mvA;

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

  560.         mvP2= mvB;

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

  562.         mvP2= mvC;

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

  564.         mvP2= mvA;

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

  566.         mvP2= mvB;

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

  568.         mvP2= mvC;

  569.     }

  570.     if(mvP2){

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

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

  573.     }else

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

  575. 

  576.     if(mode < MV_PRED_PSKIP) {

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

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

  579.     }

  580.     set_mvs(mvP,size);

  581. }

  582. 

  583. /*****************************************************************************

  584.  *

  585.  * macroblock level

  586.  *

  587.  ****************************************************************************/

  588. 

  589. /**

  590.  * initialise predictors for motion vectors and intra prediction

  591.  */

  592. void ff_cavs_init_mb(AVSContext *h) {

  593.     int i;

  594. 

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

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

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

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

  599.     }

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

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

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

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

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

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

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

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

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

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

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

  611.         h->flags |= D_AVAIL;

  612.     }

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

  614.         h->flags &= ~C_AVAIL;

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

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

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

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

  619.     }

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

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

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

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

  624.     }

  625. }

  626. 

  627. /**

  628.  * save predictors for later macroblocks and increase

  629.  * macroblock address

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

  631.  */

  632. int ff_cavs_next_mb(AVSContext *h) {

  633.     int i;

  634. 

  635.     h->flags |= A_AVAIL;

  636.     h->cy += 16;

  637.     h->cu += 8;

  638.     h->cv += 8;

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

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

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

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

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

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

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

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

  647.     /* next MB address */

  648.     h->mbidx++;

  649.     h->mbx++;

  650.     if(h->mbx == h->mb_width) { //new mb line

  651.         h->flags = B_AVAIL|C_AVAIL;

  652.         /* clear left pred_modes */

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

  654.         /* clear left mv predictors */

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

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

  657.         h->mbx = 0;

  658.         h->mby++;

  659.         /* re-calculate sample pointers */

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

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

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

  663.         if(h->mby == h->mb_height) { //frame end

  664.             return 0;

  665.         }

  666.     }

  667.     return 1;

  668. }

  669. 

  670. /*****************************************************************************

  671.  *

  672.  * frame level

  673.  *

  674.  ****************************************************************************/

  675. 

  676. void ff_cavs_init_pic(AVSContext *h) {

  677.     int i;

  678. 

  679.     /* clear some predictors */

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  695.     h->flags = 0;

  696. }

  697. 

  698. /*****************************************************************************

  699.  *

  700.  * headers and interface

  701.  *

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

  703. 

  704. /**

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

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

  707.  * and this storage space is allocated here

  708.  */

  709. void ff_cavs_init_top_lines(AVSContext *h) {

  710.     /* alloc top line of predictors */

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

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

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

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

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

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

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

  718. 

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

  720.     h->col_mv       = av_mallocz( h->mb_width*h->mb_height*4*sizeof(cavs_vector));

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

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

  723. }

  724. 

  725. av_cold int ff_cavs_init(AVCodecContext *avctx) {

  726.     AVSContext *h = avctx->priv_data;

  727. 

  728.     ff_dsputil_init(&h->dsp, avctx);

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

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

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

  732.     ff_init_scantable_permutation(h->dsp.idct_permutation,

  733.                                   h->cdsp.idct_perm);

  734.     ff_init_scantable(h->dsp.idct_permutation, &h->scantable, ff_zigzag_direct);

  735. 

  736.     h->avctx = avctx;

  737.     avctx->pix_fmt= AV_PIX_FMT_YUV420P;

  738. 

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

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

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

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

  743.         ff_cavs_end(avctx);

  744.         return AVERROR(ENOMEM);

  745.     }

  746. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  766.     return 0;

  767. }

  768. 

  769. av_cold int ff_cavs_end(AVCodecContext *avctx) {

  770.     AVSContext *h = avctx->priv_data;

  771. 

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

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

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

  775. 

  776.     av_free(h->top_qp);

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

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

  779.     av_free(h->top_pred_Y);

  780.     av_free(h->top_border_y);

  781.     av_free(h->top_border_u);

  782.     av_free(h->top_border_v);

  783.     av_free(h->col_mv);

  784.     av_free(h->col_type_base);

  785.     av_free(h->block);

  786.     av_freep(&h->edge_emu_buffer);

  787.     return 0;

  788. }