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.
13133 Commits
32 Branches
449MB
Tree: c1ac39b3df
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c1ac39b3df'
${ noResults }
FFmpeg/libavcodec/cavs.c

717 lines
25KB
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 FFmpeg.

  6.  *

  7.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; 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 cavs.c

  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 "bitstream.h"

  30. #include "golomb.h"

  31. #include "cavs.h"

  32. #include "cavsdata.h"

  33. 

  34. /*****************************************************************************

  35.  *

  36.  * in-loop deblocking filter

  37.  *

  38.  ****************************************************************************/

  39. 

  40. static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {

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

  42.         return 2;

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

  44.         return 1;

  45.     if(b){

  46.         mvP += MV_BWD_OFFS;

  47.         mvQ += MV_BWD_OFFS;

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

  49.             return 1;

  50.     }else{

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

  52.             return 1;

  53.     }

  54.     return 0;

  55. }

  56. 

  57. #define SET_PARAMS                                            \

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

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

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

  61. 

  62. /**

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

  64.  *

  65.  * boundary strength (bs) mapping:

  66.  *

  67.  * --4---5--

  68.  * 0   2   |

  69.  * | 6 | 7 |

  70.  * 1   3   |

  71.  * ---------

  72.  *

  73.  */

  74. void ff_cavs_filter(AVSContext *h, enum mb_t mb_type) {

  75.     DECLARE_ALIGNED_8(uint8_t, bs[8]);

  76.     int qp_avg, alpha, beta, tc;

  77.     int i;

  78. 

  79.     /* save un-deblocked lines */

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

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

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

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

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

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

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

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

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

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

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

  91.     }

  92.     if(!h->loop_filter_disable) {

  93.         /* determine bs */

  94.         if(mb_type == I_8X8)

  95.             *((uint64_t *)bs) = 0x0202020202020202ULL;

  96.         else{

  97.             *((uint64_t *)bs) = 0;

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

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

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

  101.             }

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

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

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

  105.             }

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

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

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

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

  110.         }

  111.         if( *((uint64_t *)bs) ) {

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

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

  114.                 SET_PARAMS;

  115.                 h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);

  116.                 h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);

  117.                 h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);

  118.             }

  119.             qp_avg = h->qp;

  120.             SET_PARAMS;

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

  122.             h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,

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

  124. 

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

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

  127.                 SET_PARAMS;

  128.                 h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);

  129.                 h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);

  130.                 h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);

  131.             }

  132.         }

  133.     }

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

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

  136. }

  137. 

  138. #undef SET_PARAMS

  139. 

  140. /*****************************************************************************

  141.  *

  142.  * spatial intra prediction

  143.  *

  144.  ****************************************************************************/

  145. 

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

  147.                                         uint8_t **left, int block) {

  148.     int i;

  149. 

  150.     switch(block) {

  151.     case 0:

  152.         *left = h->left_border_y;

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

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

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

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

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

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

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

  160.         break;

  161.     case 1:

  162.         *left = h->intern_border_y;

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

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

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

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

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

  168.         if(h->flags & C_AVAIL)

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

  170.         else

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

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

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

  174.         if(h->flags & B_AVAIL)

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

  176.         break;

  177.     case 2:

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

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

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

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

  182.         if(h->flags & A_AVAIL)

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

  184.         break;

  185.     case 3:

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

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

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

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

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

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

  192.         break;

  193.     }

  194. }

  195. 

  196. void ff_cavs_load_intra_pred_chroma(AVSContext *h) {

  197.     /* extend borders by one pixel */

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

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

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

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

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

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

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

  205.     } else {

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

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

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

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

  210.     }

  211. }

  212. 

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

  214.     int y;

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

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

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

  218.     }

  219. }

  220. 

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

  222.     int y;

  223.     uint64_t a;

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

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

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

  227.     }

  228. }

  229. 

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

  231.     int y;

  232.     uint64_t a = 0x8080808080808080ULL;

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

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

  235. }

  236. 

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

  238.     int x,y,ia;

  239.     int ih = 0;

  240.     int iv = 0;

  241.     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  242. 

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

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

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

  246.     }

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

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

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

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

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

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

  253. }

  254. 

  255. #define LOWPASS(ARRAY,INDEX)                                            \

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

  257. 

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

  259.     int x,y;

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

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

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

  263. }

  264. 

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

  266.     int x,y;

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

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

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

  270. }

  271. 

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

  273.     int x,y;

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

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

  276.             if(x==y)

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

  278.             else if(x>y)

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

  280.             else

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

  282. }

  283. 

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

  285.     int x,y;

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

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

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

  289. }

  290. 

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

  292.     int x,y;

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

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

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

  296. }

  297. 

  298. #undef LOWPASS

  299. 

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

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

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

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

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

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

  306. 

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

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

  309.         modify_pred(ff_left_modifier_l, &h->pred_mode_Y[4] );

  310.         modify_pred(ff_left_modifier_l, &h->pred_mode_Y[7] );

  311.         modify_pred(ff_left_modifier_c, pred_mode_uv );

  312.     }

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

  314.         modify_pred(ff_top_modifier_l, &h->pred_mode_Y[4] );

  315.         modify_pred(ff_top_modifier_l, &h->pred_mode_Y[5] );

  316.         modify_pred(ff_top_modifier_c, pred_mode_uv );

  317.     }

  318. }

  319. 

  320. /*****************************************************************************

  321.  *

  322.  * motion compensation

  323.  *

  324.  ****************************************************************************/

  325. 

  326. static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,

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

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

  329.                         int src_y_offset,qpel_mc_func *qpix_op,

  330.                         h264_chroma_mc_func chroma_op,vector_t *mv){

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

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

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

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

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

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

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

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

  339.     int extra_height= extra_width;

  340.     int emu=0;

  341.     const int full_mx= mx>>2;

  342.     const int full_my= my>>2;

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

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

  345. 

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

  347.         return;

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

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

  350. 

  351.     if(   full_mx < 0-extra_width

  352.           || full_my < 0-extra_height

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

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

  355.         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,

  356.                             16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

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

  358.         emu=1;

  359.     }

  360. 

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

  362.     if(!square){

  363.         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);

  364.     }

  365. 

  366.     if(emu){

  367.         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,

  368.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  369.         src_cb= s->edge_emu_buffer;

  370.     }

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

  372. 

  373.     if(emu){

  374.         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,

  375.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  376.         src_cr= s->edge_emu_buffer;

  377.     }

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

  379. }

  380. 

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

  382.                         uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,

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

  384.                         h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,

  385.                         h264_chroma_mc_func chroma_avg, vector_t *mv){

  386.     qpel_mc_func *qpix_op=  qpix_put;

  387.     h264_chroma_mc_func chroma_op= chroma_put;

  388. 

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

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

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

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

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

  394. 

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

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

  397.         mc_dir_part(h, ref, square, chroma_height, delta, 0,

  398.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  399.                     qpix_op, chroma_op, mv);

  400. 

  401.         qpix_op=  qpix_avg;

  402.         chroma_op= chroma_avg;

  403.     }

  404. 

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

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

  407.         mc_dir_part(h, ref, square, chroma_height, delta, 1,

  408.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  409.                     qpix_op, chroma_op, mv+MV_BWD_OFFS);

  410.     }

  411. }

  412. 

  413. void ff_cavs_inter(AVSContext *h, enum mb_t mb_type) {

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

  415.         mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,

  416.                 h->s.dsp.put_cavs_qpel_pixels_tab[0],

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

  418.                 h->s.dsp.avg_cavs_qpel_pixels_tab[0],

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

  420.     }else{

  421.         mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,

  422.                 h->s.dsp.put_cavs_qpel_pixels_tab[1],

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

  424.                 h->s.dsp.avg_cavs_qpel_pixels_tab[1],

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

  426.         mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,

  427.                 h->s.dsp.put_cavs_qpel_pixels_tab[1],

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

  429.                 h->s.dsp.avg_cavs_qpel_pixels_tab[1],

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

  431.         mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,

  432.                 h->s.dsp.put_cavs_qpel_pixels_tab[1],

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

  434.                 h->s.dsp.avg_cavs_qpel_pixels_tab[1],

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

  436.         mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,

  437.                 h->s.dsp.put_cavs_qpel_pixels_tab[1],

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

  439.                 h->s.dsp.avg_cavs_qpel_pixels_tab[1],

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

  441.     }

  442. }

  443. 

  444. /*****************************************************************************

  445.  *

  446.  * motion vector prediction

  447.  *

  448.  ****************************************************************************/

  449. 

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

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

  452. 

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

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

  455. }

  456. 

  457. static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {

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

  459.     int len_ab, len_bc, len_ca, len_mid;

  460. 

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

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

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

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

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

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

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

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

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

  470.     if(len_mid == len_ab) {

  471.         mvP->x = cx;

  472.         mvP->y = cy;

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

  474.         mvP->x = ax;

  475.         mvP->y = ay;

  476.     } else {

  477.         mvP->x = bx;

  478.         mvP->y = by;

  479.     }

  480. }

  481. 

  482. void ff_cavs_mv(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,

  483.                 enum mv_pred_t mode, enum block_t size, int ref) {

  484.     vector_t *mvP = &h->mv[nP];

  485.     vector_t *mvA = &h->mv[nP-1];

  486.     vector_t *mvB = &h->mv[nP-4];

  487.     vector_t *mvC = &h->mv[nC];

  488.     const vector_t *mvP2 = NULL;

  489. 

  490.     mvP->ref = ref;

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

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

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

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

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

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

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

  498.         mvP2 = &ff_cavs_un_mv;

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

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

  501.         mvP2= mvA;

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

  503.         mvP2= mvB;

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

  505.         mvP2= mvC;

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

  507.         mvP2= mvA;

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

  509.         mvP2= mvB;

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

  511.         mvP2= mvC;

  512.     }

  513.     if(mvP2){

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

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

  516.     }else

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

  518. 

  519.     if(mode < MV_PRED_PSKIP) {

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

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

  522.     }

  523.     set_mvs(mvP,size);

  524. }

  525. 

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

  527.  *

  528.  * macroblock level

  529.  *

  530.  ****************************************************************************/

  531. 

  532. /**

  533.  * initialise predictors for motion vectors and intra prediction

  534.  */

  535. void ff_cavs_init_mb(AVSContext *h) {

  536.     int i;

  537. 

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

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

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

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

  542.     }

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

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

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

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

  547.         h->mv[MV_FWD_B2] = ff_cavs_un_mv;

  548.         h->mv[MV_FWD_B3] = ff_cavs_un_mv;

  549.         h->mv[MV_BWD_B2] = ff_cavs_un_mv;

  550.         h->mv[MV_BWD_B3] = ff_cavs_un_mv;

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

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

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

  554.         h->flags |= D_AVAIL;

  555.     }

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

  557.         h->flags &= ~C_AVAIL;

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

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

  560.         h->mv[MV_FWD_C2] = ff_cavs_un_mv;

  561.         h->mv[MV_BWD_C2] = ff_cavs_un_mv;

  562.     }

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

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

  565.         h->mv[MV_FWD_D3] = ff_cavs_un_mv;

  566.         h->mv[MV_BWD_D3] = ff_cavs_un_mv;

  567.     }

  568.     /* set pointer for co-located macroblock type */

  569.     h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];

  570. }

  571. 

  572. /**

  573.  * save predictors for later macroblocks and increase

  574.  * macroblock address

  575.  * @returns 0 if end of frame is reached, 1 otherwise

  576.  */

  577. int ff_cavs_next_mb(AVSContext *h) {

  578.     int i;

  579. 

  580.     h->flags |= A_AVAIL;

  581.     h->cy += 16;

  582.     h->cu += 8;

  583.     h->cv += 8;

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

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

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

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

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

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

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

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

  592.     /* next MB address */

  593.     h->mbx++;

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

  595.         h->flags = B_AVAIL|C_AVAIL;

  596.         /* clear left pred_modes */

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

  598.         /* clear left mv predictors */

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

  600.             h->mv[i] = ff_cavs_un_mv;

  601.         h->mbx = 0;

  602.         h->mby++;

  603.         /* re-calculate sample pointers */

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

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

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

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

  608.             return 0;

  609.         } else {

  610.             //check_for_slice(h);

  611.         }

  612.     }

  613.     return 1;

  614. }

  615. 

  616. /*****************************************************************************

  617.  *

  618.  * frame level

  619.  *

  620.  ****************************************************************************/

  621. 

  622. void ff_cavs_init_pic(AVSContext *h) {

  623.     int i;

  624. 

  625.     /* clear some predictors */

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

  627.         h->mv[i] = ff_cavs_un_mv;

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

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

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

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

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

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

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

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

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

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

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

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

  640.     h->mbx = h->mby = 0;

  641.     h->flags = 0;

  642. }

  643. 

  644. /*****************************************************************************

  645.  *

  646.  * headers and interface

  647.  *

  648.  ****************************************************************************/

  649. 

  650. /**

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

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

  653.  * and this storage space is allocated here

  654.  */

  655. void ff_cavs_init_top_lines(AVSContext *h) {

  656.     /* alloc top line of predictors */

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

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

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

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

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

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

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

  664. 

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

  666.     h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));

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

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

  669. }

  670. 

  671. av_cold int ff_cavs_init(AVCodecContext *avctx) {

  672.     AVSContext *h = avctx->priv_data;

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

  674. 

  675.     MPV_decode_defaults(s);

  676.     s->avctx = avctx;

  677. 

  678.     avctx->pix_fmt= PIX_FMT_YUV420P;

  679. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  697.     h->mv[ 7] = ff_cavs_un_mv;

  698.     h->mv[19] = ff_cavs_un_mv;

  699.     return 0;

  700. }

  701. 

  702. av_cold int ff_cavs_end(AVCodecContext *avctx) {

  703.     AVSContext *h = avctx->priv_data;

  704. 

  705.     av_free(h->top_qp);

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

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

  708.     av_free(h->top_pred_Y);

  709.     av_free(h->top_border_y);

  710.     av_free(h->top_border_u);

  711.     av_free(h->top_border_v);

  712.     av_free(h->col_mv);

  713.     av_free(h->col_type_base);

  714.     av_free(h->block);

  715.     return 0;

  716. }