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.
35406 Commits
32 Branches
935MB
Tree: c3e15f7b39
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c3e15f7b39'
${ noResults }
FFmpeg/libavcodec/cavs.c

754 lines
27KB
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 "mathops.h"

  32. #include "cavs.h"

  33. 

  34. static const uint8_t alpha_tab[64] = {

  35.    0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  2,  2,  2,  3,  3,

  36.    4,  4,  5,  5,  6,  7,  8,  9, 10, 11, 12, 13, 15, 16, 18, 20,

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

  38.   46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64

  39. };

  40. 

  41. static const uint8_t beta_tab[64] = {

  42.    0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,

  43.    2,  2,  3,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,  5,  6,  6,

  44.    6,  7,  7,  7,  8,  8,  8,  9,  9, 10, 10, 11, 11, 12, 13, 14,

  45.   15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27

  46. };

  47. 

  48. static const uint8_t tc_tab[64] = {

  49.   0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,

  50.   1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,

  51.   2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,

  52.   5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9

  53. };

  54. 

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

  56.     or not yet decoded */

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

  58. 

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

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

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

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

  63. 

  64. /*****************************************************************************

  65.  *

  66.  * in-loop deblocking filter

  67.  *

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

  69. 

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

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

  72.         return 2;

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

  74.         return 1;

  75.     if(b){

  76.         mvP += MV_BWD_OFFS;

  77.         mvQ += MV_BWD_OFFS;

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

  79.             return 1;

  80.     }else{

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

  82.             return 1;

  83.     }

  84.     return 0;

  85. }

  86. 

  87. #define SET_PARAMS                                            \

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

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

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

  91. 

  92. /**

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

  94.  *

  95.  * boundary strength (bs) mapping:

  96.  *

  97.  * --4---5--

  98.  * 0   2   |

  99.  * | 6 | 7 |

  100.  * 1   3   |

  101.  * ---------

  102.  *

  103.  */

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

  105.     uint8_t bs[8];

  106.     int qp_avg, alpha, beta, tc;

  107.     int i;

  108. 

  109.     /* save un-deblocked lines */

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

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

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

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

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

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

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

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

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

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

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

  121.     }

  122.     if(!h->loop_filter_disable) {

  123.         /* determine bs */

  124.         if(mb_type == I_8X8)

  125.             memset(bs,2,8);

  126.         else{

  127.             memset(bs,0,8);

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

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

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

  131.             }

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

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

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

  135.             }

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

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

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

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

  140.         }

  141.         if(AV_RN64(bs)) {

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

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

  144.                 SET_PARAMS;

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

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

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

  148.             }

  149.             qp_avg = h->qp;

  150.             SET_PARAMS;

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

  152.             h->cdsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,

  153.                            bs[6],bs[7]);

  154. 

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

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

  157.                 SET_PARAMS;

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

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

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

  161.             }

  162.         }

  163.     }

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

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

  166. }

  167. 

  168. #undef SET_PARAMS

  169. 

  170. /*****************************************************************************

  171.  *

  172.  * spatial intra prediction

  173.  *

  174.  ****************************************************************************/

  175. 

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

  177.                                         uint8_t **left, int block) {

  178.     int i;

  179. 

  180.     switch(block) {

  181.     case 0:

  182.         *left = h->left_border_y;

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

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

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

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

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

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

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

  190.         break;

  191.     case 1:

  192.         *left = h->intern_border_y;

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

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

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

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

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

  198.         if(h->flags & C_AVAIL)

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

  200.         else

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

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

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

  204.         if(h->flags & B_AVAIL)

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

  206.         break;

  207.     case 2:

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

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

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

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

  212.         if(h->flags & A_AVAIL)

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

  214.         break;

  215.     case 3:

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

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

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

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

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

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

  222.         break;

  223.     }

  224. }

  225. 

  226. void ff_cavs_load_intra_pred_chroma(AVSContext *h) {

  227.     /* extend borders by one pixel */

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

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

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

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

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

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

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

  235.     } else {

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

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

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

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

  240.     }

  241. }

  242. 

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

  244.     int y;

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

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

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

  248.     }

  249. }

  250. 

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

  252.     int y;

  253.     uint64_t a;

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

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

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

  257.     }

  258. }

  259. 

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

  261.     int y;

  262.     uint64_t a = 0x8080808080808080ULL;

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

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

  265. }

  266. 

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

  268.     int x,y,ia;

  269.     int ih = 0;

  270.     int iv = 0;

  271.     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  272. 

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

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

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

  276.     }

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

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

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

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

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

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

  283. }

  284. 

  285. #define LOWPASS(ARRAY,INDEX)                                            \

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

  287. 

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

  289.     int x,y;

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

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

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

  293. }

  294. 

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

  296.     int x,y;

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

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

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

  300. }

  301. 

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

  303.     int x,y;

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

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

  306.             if(x==y)

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

  308.             else if(x>y)

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

  310.             else

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

  312. }

  313. 

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

  315.     int x,y;

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

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

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

  319. }

  320. 

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

  322.     int x,y;

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

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

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

  326. }

  327. 

  328. #undef LOWPASS

  329. 

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

  331. {

  332.     *mode = mod_table[*mode];

  333.     if(*mode < 0) {

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

  335.         *mode = 0;

  336.     }

  337. }

  338. 

  339. void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv) {

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

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

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

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

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

  345. 

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

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

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

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

  350.         modify_pred(left_modifier_c, pred_mode_uv);

  351.     }

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

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

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

  355.         modify_pred(top_modifier_c, pred_mode_uv);

  356.     }

  357. }

  358. 

  359. /*****************************************************************************

  360.  *

  361.  * motion compensation

  362.  *

  363.  ****************************************************************************/

  364. 

  365. static inline void mc_dir_part(AVSContext *h,Picture *pic,

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

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

  368.                                int src_y_offset,qpel_mc_func *qpix_op,

  369.                                h264_chroma_mc_func chroma_op,cavs_vector *mv)

  370. {

  371.     MpegEncContext * const s = &h->s;

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

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

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

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

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

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

  378.     int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;

  379.     int extra_height= extra_width;

  380.     int emu=0;

  381.     const int full_mx= mx>>2;

  382.     const int full_my= my>>2;

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

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

  385. 

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

  387.         return;

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

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

  390. 

  391.     if(   full_mx < 0-extra_width

  392.           || full_my < 0-extra_height

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

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

  395.         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,

  396.                             16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

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

  398.         emu=1;

  399.     }

  400. 

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

  402. 

  403.     if(emu){

  404.         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,

  405.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  406.         src_cb= s->edge_emu_buffer;

  407.     }

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

  409. 

  410.     if(emu){

  411.         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,

  412.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  413.         src_cr= s->edge_emu_buffer;

  414.     }

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

  416. }

  417. 

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

  419.                                uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,

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

  421.                                h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,

  422.                                h264_chroma_mc_func chroma_avg, cavs_vector *mv)

  423. {

  424.     qpel_mc_func *qpix_op=  qpix_put;

  425.     h264_chroma_mc_func chroma_op= chroma_put;

  426. 

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

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

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

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

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

  432. 

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

  434.         Picture *ref= &h->DPB[mv->ref];

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

  436.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  437.                     qpix_op, chroma_op, mv);

  438. 

  439.         qpix_op=  qpix_avg;

  440.         chroma_op= chroma_avg;

  441.     }

  442. 

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

  444.         Picture *ref= &h->DPB[0];

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

  446.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  447.                     qpix_op, chroma_op, mv+MV_BWD_OFFS);

  448.     }

  449. }

  450. 

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

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

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

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

  455.                 h->s.dsp.put_h264_chroma_pixels_tab[0],

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

  457.                 h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);

  458.     }else{

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

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

  461.                 h->s.dsp.put_h264_chroma_pixels_tab[1],

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

  463.                 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);

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

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

  466.                 h->s.dsp.put_h264_chroma_pixels_tab[1],

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

  468.                 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);

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

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

  471.                 h->s.dsp.put_h264_chroma_pixels_tab[1],

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

  473.                 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);

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

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

  476.                 h->s.dsp.put_h264_chroma_pixels_tab[1],

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

  478.                 h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);

  479.     }

  480. }

  481. 

  482. /*****************************************************************************

  483.  *

  484.  * motion vector prediction

  485.  *

  486.  ****************************************************************************/

  487. 

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

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

  490. 

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

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

  493. }

  494. 

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

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

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

  498.     int len_ab, len_bc, len_ca, len_mid;

  499. 

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

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

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

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

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

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

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

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

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

  509.     if(len_mid == len_ab) {

  510.         mvP->x = cx;

  511.         mvP->y = cy;

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

  513.         mvP->x = ax;

  514.         mvP->y = ay;

  515.     } else {

  516.         mvP->x = bx;

  517.         mvP->y = by;

  518.     }

  519. }

  520. 

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

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

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

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

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

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

  527.     const cavs_vector *mvP2 = NULL;

  528. 

  529.     mvP->ref = ref;

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

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

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

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

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

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

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

  537.         mvP2 = &un_mv;

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

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

  540.         mvP2= mvA;

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

  542.         mvP2= mvB;

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

  544.         mvP2= mvC;

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

  546.         mvP2= mvA;

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

  548.         mvP2= mvB;

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

  550.         mvP2= mvC;

  551.     }

  552.     if(mvP2){

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

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

  555.     }else

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

  557. 

  558.     if(mode < MV_PRED_PSKIP) {

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

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

  561.     }

  562.     set_mvs(mvP,size);

  563. }

  564. 

  565. /*****************************************************************************

  566.  *

  567.  * macroblock level

  568.  *

  569.  ****************************************************************************/

  570. 

  571. /**

  572.  * initialise predictors for motion vectors and intra prediction

  573.  */

  574. void ff_cavs_init_mb(AVSContext *h) {

  575.     int i;

  576. 

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

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

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

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

  581.     }

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

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

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

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

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

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

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

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

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

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

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

  593.         h->flags |= D_AVAIL;

  594.     }

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

  596.         h->flags &= ~C_AVAIL;

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

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

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

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

  601.     }

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

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

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

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

  606.     }

  607. }

  608. 

  609. /**

  610.  * save predictors for later macroblocks and increase

  611.  * macroblock address

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

  613.  */

  614. int ff_cavs_next_mb(AVSContext *h) {

  615.     int i;

  616. 

  617.     h->flags |= A_AVAIL;

  618.     h->cy += 16;

  619.     h->cu += 8;

  620.     h->cv += 8;

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

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

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

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

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

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

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

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

  629.     /* next MB address */

  630.     h->mbidx++;

  631.     h->mbx++;

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

  633.         h->flags = B_AVAIL|C_AVAIL;

  634.         /* clear left pred_modes */

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

  636.         /* clear left mv predictors */

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

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

  639.         h->mbx = 0;

  640.         h->mby++;

  641.         /* re-calculate sample pointers */

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

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

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

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

  646.             return 0;

  647.         }

  648.     }

  649.     return 1;

  650. }

  651. 

  652. /*****************************************************************************

  653.  *

  654.  * frame level

  655.  *

  656.  ****************************************************************************/

  657. 

  658. void ff_cavs_init_pic(AVSContext *h) {

  659.     int i;

  660. 

  661.     /* clear some predictors */

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

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

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

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

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

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

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

  669.     h->cy           = h->picture.f.data[0];

  670.     h->cu           = h->picture.f.data[1];

  671.     h->cv           = h->picture.f.data[2];

  672.     h->l_stride     = h->picture.f.linesize[0];

  673.     h->c_stride     = h->picture.f.linesize[1];

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

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

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

  677.     h->flags = 0;

  678. }

  679. 

  680. /*****************************************************************************

  681.  *

  682.  * headers and interface

  683.  *

  684.  ****************************************************************************/

  685. 

  686. /**

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

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

  689.  * and this storage space is allocated here

  690.  */

  691. void ff_cavs_init_top_lines(AVSContext *h) {

  692.     /* alloc top line of predictors */

  693.     h->top_qp       = av_malloc( h->mb_width);

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

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

  696.     h->top_pred_Y   = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));

  697.     h->top_border_y = av_malloc((h->mb_width+1)*16);

  698.     h->top_border_u = av_malloc( h->mb_width * 10);

  699.     h->top_border_v = av_malloc( h->mb_width * 10);

  700. 

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

  702.     h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(cavs_vector));

  703.     h->col_type_base = av_malloc(h->mb_width*h->mb_height);

  704.     h->block        = av_mallocz(64*sizeof(DCTELEM));

  705. }

  706. 

  707. av_cold int ff_cavs_init(AVCodecContext *avctx) {

  708.     AVSContext *h = avctx->priv_data;

  709.     MpegEncContext * const s = &h->s;

  710. 

  711.     ff_MPV_decode_defaults(s);

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

  713.     s->avctx = avctx;

  714. 

  715.     avctx->pix_fmt= AV_PIX_FMT_YUV420P;

  716. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  736.     return 0;

  737. }

  738. 

  739. av_cold int ff_cavs_end(AVCodecContext *avctx) {

  740.     AVSContext *h = avctx->priv_data;

  741. 

  742.     av_free(h->top_qp);

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

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

  745.     av_free(h->top_pred_Y);

  746.     av_free(h->top_border_y);

  747.     av_free(h->top_border_u);

  748.     av_free(h->top_border_v);

  749.     av_free(h->col_mv);

  750.     av_free(h->col_type_base);

  751.     av_free(h->block);

  752.     return 0;

  753. }