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.
9969 Commits
32 Branches
283MB
Tree: b8524fd13f
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b8524fd13f'
${ noResults }
FFmpeg/libavcodec/cavs.c

1229 lines
43KB
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. static void filter_mb(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(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(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. static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {

  147.     int y;

  148.     uint64_t a = unaligned64(&top[1]);

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

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

  151.     }

  152. }

  153. 

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

  155.     int y;

  156.     uint64_t a;

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

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

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

  160.     }

  161. }

  162. 

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

  164.     int y;

  165.     uint64_t a = 0x8080808080808080ULL;

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

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

  168. }

  169. 

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

  171.     int x,y,ia;

  172.     int ih = 0;

  173.     int iv = 0;

  174.     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  175. 

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

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

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

  179.     }

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

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

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

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

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

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

  186. }

  187. 

  188. #define LOWPASS(ARRAY,INDEX)                                            \

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

  190. 

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

  192.     int x,y;

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

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

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

  196. }

  197. 

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

  199.     int x,y;

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

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

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

  203. }

  204. 

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

  206.     int x,y;

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

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

  209.             if(x==y)

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

  211.             else if(x>y)

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

  213.             else

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

  215. }

  216. 

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

  218.     int x,y;

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

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

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

  222. }

  223. 

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

  225.     int x,y;

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

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

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

  229. }

  230. 

  231. #undef LOWPASS

  232. 

  233. /*****************************************************************************

  234.  *

  235.  * motion compensation

  236.  *

  237.  ****************************************************************************/

  238. 

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

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

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

  242.                         int src_y_offset,qpel_mc_func *qpix_op,

  243.                         h264_chroma_mc_func chroma_op,vector_t *mv){

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

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

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

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

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

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

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

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

  252.     int extra_height= extra_width;

  253.     int emu=0;

  254.     const int full_mx= mx>>2;

  255.     const int full_my= my>>2;

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

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

  258. 

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

  260.         return;

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

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

  263. 

  264.     if(   full_mx < 0-extra_width

  265.           || full_my < 0-extra_height

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

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

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

  269.                             16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

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

  271.         emu=1;

  272.     }

  273. 

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

  275.     if(!square){

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

  277.     }

  278. 

  279.     if(emu){

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

  281.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  282.         src_cb= s->edge_emu_buffer;

  283.     }

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

  285. 

  286.     if(emu){

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

  288.                             9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  289.         src_cr= s->edge_emu_buffer;

  290.     }

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

  292. }

  293. 

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

  295.                         uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,

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

  297.                         h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,

  298.                         h264_chroma_mc_func chroma_avg, vector_t *mv){

  299.     qpel_mc_func *qpix_op=  qpix_put;

  300.     h264_chroma_mc_func chroma_op= chroma_put;

  301. 

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

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

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

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

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

  307. 

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

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

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

  311.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  312.                     qpix_op, chroma_op, mv);

  313. 

  314.         qpix_op=  qpix_avg;

  315.         chroma_op= chroma_avg;

  316.     }

  317. 

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

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

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

  321.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  322.                     qpix_op, chroma_op, mv+MV_BWD_OFFS);

  323.     }

  324. }

  325. 

  326. static void inter_pred(AVSContext *h, enum mb_t mb_type) {

  327.     if(partition_flags[mb_type] == 0){ // 16x16

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

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

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

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

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

  333.     }else{

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  354.     }

  355.     /* set intra prediction modes to default values */

  356.     h->pred_mode_Y[3] =  h->pred_mode_Y[6] = INTRA_L_LP;

  357.     h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;

  358. }

  359. 

  360. /*****************************************************************************

  361.  *

  362.  * motion vector prediction

  363.  *

  364.  ****************************************************************************/

  365. 

  366. static inline void store_mvs(AVSContext *h) {

  367.     h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];

  368.     h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];

  369.     h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];

  370.     h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];

  371. }

  372. 

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

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

  375. 

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

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

  378. }

  379. 

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

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

  382.     int len_ab, len_bc, len_ca, len_mid;

  383. 

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

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

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

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

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

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

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

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

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

  393.     if(len_mid == len_ab) {

  394.         mvP->x = cx;

  395.         mvP->y = cy;

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

  397.         mvP->x = ax;

  398.         mvP->y = ay;

  399.     } else {

  400.         mvP->x = bx;

  401.         mvP->y = by;

  402.     }

  403. }

  404. 

  405. static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,

  406.                                   vector_t *col_mv) {

  407.     vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;

  408.     int den = h->direct_den[col_mv->ref];

  409.     int m = col_mv->x >> 31;

  410. 

  411.     pmv_fw->dist = h->dist[1];

  412.     pmv_bw->dist = h->dist[0];

  413.     pmv_fw->ref = 1;

  414.     pmv_bw->ref = 0;

  415.     /* scale the co-located motion vector according to its temporal span */

  416.     pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;

  417.     pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);

  418.     m = col_mv->y >> 31;

  419.     pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;

  420.     pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);

  421. }

  422. 

  423. static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {

  424.     vector_t *dst = src + MV_BWD_OFFS;

  425. 

  426.     /* backward mv is the scaled and negated forward mv */

  427.     dst->x = -((src->x * h->sym_factor + 256) >> 9);

  428.     dst->y = -((src->y * h->sym_factor + 256) >> 9);

  429.     dst->ref = 0;

  430.     dst->dist = h->dist[0];

  431.     set_mvs(dst, size);

  432. }

  433. 

  434. static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,

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

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

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

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

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

  440.     const vector_t *mvP2 = NULL;

  441. 

  442.     mvP->ref = ref;

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

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

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

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

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

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

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

  450.         mvP2 = &ff_cavs_un_mv;

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

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

  453.         mvP2= mvA;

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

  455.         mvP2= mvB;

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

  457.         mvP2= mvC;

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

  459.         mvP2= mvA;

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

  461.         mvP2= mvB;

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

  463.         mvP2= mvC;

  464.     }

  465.     if(mvP2){

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

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

  468.     }else

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

  470. 

  471.     if(mode < MV_PRED_PSKIP) {

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

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

  474.     }

  475.     set_mvs(mvP,size);

  476. }

  477. 

  478. /*****************************************************************************

  479.  *

  480.  * residual data decoding

  481.  *

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

  483. 

  484. /** kth-order exponential golomb code */

  485. static inline int get_ue_code(GetBitContext *gb, int order) {

  486.     if(order) {

  487.         int ret = get_ue_golomb(gb) << order;

  488.         return ret + get_bits(gb,order);

  489.     }

  490.     return get_ue_golomb(gb);

  491. }

  492. 

  493. /**

  494.  * decode coefficients from one 8x8 block, dequantize, inverse transform

  495.  *  and add them to sample block

  496.  * @param r pointer to 2D VLC table

  497.  * @param esc_golomb_order escape codes are k-golomb with this order k

  498.  * @param qp quantizer

  499.  * @param dst location of sample block

  500.  * @param stride line stride in frame buffer

  501.  */

  502. static int decode_residual_block(AVSContext *h, GetBitContext *gb,

  503.                                  const residual_vlc_t *r, int esc_golomb_order,

  504.                                  int qp, uint8_t *dst, int stride) {

  505.     int i,pos = -1;

  506.     int level_code, esc_code, level, run, mask;

  507.     int level_buf[64];

  508.     int run_buf[64];

  509.     int dqm = dequant_mul[qp];

  510.     int dqs = dequant_shift[qp];

  511.     int dqa = 1 << (dqs - 1);

  512.     const uint8_t *scantab = h->scantable.permutated;

  513.     DCTELEM *block = h->block;

  514. 

  515.     for(i=0;i<65;i++) {

  516.         level_code = get_ue_code(gb,r->golomb_order);

  517.         if(level_code >= ESCAPE_CODE) {

  518.             run = ((level_code - ESCAPE_CODE) >> 1) + 1;

  519.             esc_code = get_ue_code(gb,esc_golomb_order);

  520.             level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);

  521.             while(level > r->inc_limit)

  522.                 r++;

  523.             mask = -(level_code & 1);

  524.             level = (level^mask) - mask;

  525.         } else {

  526.             level = r->rltab[level_code][0];

  527.             if(!level) //end of block signal

  528.                 break;

  529.             run   = r->rltab[level_code][1];

  530.             r += r->rltab[level_code][2];

  531.         }

  532.         level_buf[i] = level;

  533.         run_buf[i] = run;

  534.     }

  535.     /* inverse scan and dequantization */

  536.     while(--i >= 0){

  537.         pos += run_buf[i];

  538.         if(pos > 63) {

  539.             av_log(h->s.avctx, AV_LOG_ERROR,

  540.                    "position out of block bounds at pic %d MB(%d,%d)\n",

  541.                    h->picture.poc, h->mbx, h->mby);

  542.             return -1;

  543.         }

  544.         block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;

  545.     }

  546.     h->s.dsp.cavs_idct8_add(dst,block,stride);

  547.     return 0;

  548. }

  549. 

  550. 

  551. static inline void decode_residual_chroma(AVSContext *h) {

  552.     if(h->cbp & (1<<4))

  553.         decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],

  554.                               h->cu,h->c_stride);

  555.     if(h->cbp & (1<<5))

  556.         decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],

  557.                               h->cv,h->c_stride);

  558. }

  559. 

  560. static inline int decode_residual_inter(AVSContext *h) {

  561.     int block;

  562. 

  563.     /* get coded block pattern */

  564.     int cbp= get_ue_golomb(&h->s.gb);

  565.     if(cbp > 63){

  566.         av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");

  567.         return -1;

  568.     }

  569.     h->cbp = cbp_tab[cbp][1];

  570. 

  571.     /* get quantizer */

  572.     if(h->cbp && !h->qp_fixed)

  573.         h->qp = (h->qp + get_se_golomb(&h->s.gb)) & 63;

  574.     for(block=0;block<4;block++)

  575.         if(h->cbp & (1<<block))

  576.             decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,

  577.                                   h->cy + h->luma_scan[block], h->l_stride);

  578.     decode_residual_chroma(h);

  579. 

  580.     return 0;

  581. }

  582. 

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

  584.  *

  585.  * macroblock level

  586.  *

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

  588. 

  589. static int decode_mb_i(AVSContext *h, int cbp_code) {

  590.     GetBitContext *gb = &h->s.gb;

  591.     int block, pred_mode_uv;

  592.     uint8_t top[18];

  593.     uint8_t *left = NULL;

  594.     uint8_t *d;

  595. 

  596.     init_mb(h);

  597. 

  598.     /* get intra prediction modes from stream */

  599.     for(block=0;block<4;block++) {

  600.         int nA,nB,predpred;

  601.         int pos = scan3x3[block];

  602. 

  603.         nA = h->pred_mode_Y[pos-1];

  604.         nB = h->pred_mode_Y[pos-3];

  605.         predpred = FFMIN(nA,nB);

  606.         if(predpred == NOT_AVAIL) // if either is not available

  607.             predpred = INTRA_L_LP;

  608.         if(!get_bits1(gb)){

  609.             int rem_mode= get_bits(gb, 2);

  610.             predpred = rem_mode + (rem_mode >= predpred);

  611.         }

  612.         h->pred_mode_Y[pos] = predpred;

  613.     }

  614.     pred_mode_uv = get_ue_golomb(gb);

  615.     if(pred_mode_uv > 6) {

  616.         av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");

  617.         return -1;

  618.     }

  619. 

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

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

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

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

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

  625. 

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

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

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

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

  630.         modify_pred(left_modifier_c, &pred_mode_uv );

  631.     }

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

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

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

  635.         modify_pred(top_modifier_c, &pred_mode_uv );

  636.     }

  637. 

  638.     /* get coded block pattern */

  639.     if(h->pic_type == FF_I_TYPE)

  640.         cbp_code = get_ue_golomb(gb);

  641.     if(cbp_code > 63){

  642.         av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");

  643.         return -1;

  644.     }

  645.     h->cbp = cbp_tab[cbp_code][0];

  646.     if(h->cbp && !h->qp_fixed)

  647.         h->qp = (h->qp + get_se_golomb(gb)) & 63; //qp_delta

  648. 

  649.     /* luma intra prediction interleaved with residual decode/transform/add */

  650.     for(block=0;block<4;block++) {

  651.         d = h->cy + h->luma_scan[block];

  652.         load_intra_pred_luma(h, top, &left, block);

  653.         h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]

  654.             (d, top, left, h->l_stride);

  655.         if(h->cbp & (1<<block))

  656.             decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);

  657.     }

  658. 

  659.     /* chroma intra prediction */

  660.     /* extend borders by one pixel */

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

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

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

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

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

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

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

  668.     } else {

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

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

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

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

  673.     }

  674.     h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],

  675.                                   h->left_border_u, h->c_stride);

  676.     h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],

  677.                                   h->left_border_v, h->c_stride);

  678. 

  679.     decode_residual_chroma(h);

  680.     filter_mb(h,I_8X8);

  681. 

  682.     /* mark motion vectors as intra */

  683.     h->mv[MV_FWD_X0] = ff_cavs_intra_mv;

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

  685.     h->mv[MV_BWD_X0] = ff_cavs_intra_mv;

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

  687.     if(h->pic_type != FF_B_TYPE)

  688.         *h->col_type = I_8X8;

  689. 

  690.     return 0;

  691. }

  692. 

  693. static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {

  694.     GetBitContext *gb = &h->s.gb;

  695.     int ref[4];

  696. 

  697.     init_mb(h);

  698.     switch(mb_type) {

  699.     case P_SKIP:

  700.         mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);

  701.         break;

  702.     case P_16X16:

  703.         ref[0] = h->ref_flag ? 0 : get_bits1(gb);

  704.         mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_16X16,ref[0]);

  705.         break;

  706.     case P_16X8:

  707.         ref[0] = h->ref_flag ? 0 : get_bits1(gb);

  708.         ref[2] = h->ref_flag ? 0 : get_bits1(gb);

  709.         mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,      BLK_16X8, ref[0]);

  710.         mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT,     BLK_16X8, ref[2]);

  711.         break;

  712.     case P_8X16:

  713.         ref[0] = h->ref_flag ? 0 : get_bits1(gb);

  714.         ref[1] = h->ref_flag ? 0 : get_bits1(gb);

  715.         mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT,     BLK_8X16, ref[0]);

  716.         mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);

  717.         break;

  718.     case P_8X8:

  719.         ref[0] = h->ref_flag ? 0 : get_bits1(gb);

  720.         ref[1] = h->ref_flag ? 0 : get_bits1(gb);

  721.         ref[2] = h->ref_flag ? 0 : get_bits1(gb);

  722.         ref[3] = h->ref_flag ? 0 : get_bits1(gb);

  723.         mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN,   BLK_8X8, ref[0]);

  724.         mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_8X8, ref[1]);

  725.         mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN,   BLK_8X8, ref[2]);

  726.         mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN,   BLK_8X8, ref[3]);

  727.     }

  728.     inter_pred(h, mb_type);

  729.     store_mvs(h);

  730.     if(mb_type != P_SKIP)

  731.         decode_residual_inter(h);

  732.     filter_mb(h,mb_type);

  733.     *h->col_type = mb_type;

  734. }

  735. 

  736. static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {

  737.     int block;

  738.     enum sub_mb_t sub_type[4];

  739.     int flags;

  740. 

  741.     init_mb(h);

  742. 

  743.     /* reset all MVs */

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

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

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

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

  748.     switch(mb_type) {

  749.     case B_SKIP:

  750.     case B_DIRECT:

  751.         if(!(*h->col_type)) {

  752.             /* intra MB at co-location, do in-plane prediction */

  753.             mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);

  754.             mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);

  755.         } else

  756.             /* direct prediction from co-located P MB, block-wise */

  757.             for(block=0;block<4;block++)

  758.                 mv_pred_direct(h,&h->mv[mv_scan[block]],

  759.                             &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);

  760.         break;

  761.     case B_FWD_16X16:

  762.         mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);

  763.         break;

  764.     case B_SYM_16X16:

  765.         mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);

  766.         mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);

  767.         break;

  768.     case B_BWD_16X16:

  769.         mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);

  770.         break;

  771.     case B_8X8:

  772.         for(block=0;block<4;block++)

  773.             sub_type[block] = get_bits(&h->s.gb,2);

  774.         for(block=0;block<4;block++) {

  775.             switch(sub_type[block]) {

  776.             case B_SUB_DIRECT:

  777.                 if(!(*h->col_type)) {

  778.                     /* intra MB at co-location, do in-plane prediction */

  779.                     mv_pred(h, mv_scan[block], mv_scan[block]-3,

  780.                             MV_PRED_BSKIP, BLK_8X8, 1);

  781.                     mv_pred(h, mv_scan[block]+MV_BWD_OFFS,

  782.                             mv_scan[block]-3+MV_BWD_OFFS,

  783.                             MV_PRED_BSKIP, BLK_8X8, 0);

  784.                 } else

  785.                     mv_pred_direct(h,&h->mv[mv_scan[block]],

  786.                                    &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);

  787.                 break;

  788.             case B_SUB_FWD:

  789.                 mv_pred(h, mv_scan[block], mv_scan[block]-3,

  790.                         MV_PRED_MEDIAN, BLK_8X8, 1);

  791.                 break;

  792.             case B_SUB_SYM:

  793.                 mv_pred(h, mv_scan[block], mv_scan[block]-3,

  794.                         MV_PRED_MEDIAN, BLK_8X8, 1);

  795.                 mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);

  796.                 break;

  797.             }

  798.         }

  799.         for(block=0;block<4;block++) {

  800.             if(sub_type[block] == B_SUB_BWD)

  801.                 mv_pred(h, mv_scan[block]+MV_BWD_OFFS,

  802.                         mv_scan[block]+MV_BWD_OFFS-3,

  803.                         MV_PRED_MEDIAN, BLK_8X8, 0);

  804.         }

  805.         break;

  806.     default:

  807.         assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));

  808.         flags = partition_flags[mb_type];

  809.         if(mb_type & 1) { /* 16x8 macroblock types */

  810.             if(flags & FWD0)

  811.                 mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,  BLK_16X8, 1);

  812.             if(flags & SYM0)

  813.                 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);

  814.             if(flags & FWD1)

  815.                 mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);

  816.             if(flags & SYM1)

  817.                 mv_pred_sym(h, &h->mv[MV_FWD_X2], BLK_16X8);

  818.             if(flags & BWD0)

  819.                 mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP,  BLK_16X8, 0);

  820.             if(flags & BWD1)

  821.                 mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);

  822.         } else {          /* 8x16 macroblock types */

  823.             if(flags & FWD0)

  824.                 mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);

  825.             if(flags & SYM0)

  826.                 mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);

  827.             if(flags & FWD1)

  828.                 mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);

  829.             if(flags & SYM1)

  830.                 mv_pred_sym(h, &h->mv[MV_FWD_X1], BLK_8X16);

  831.             if(flags & BWD0)

  832.                 mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);

  833.             if(flags & BWD1)

  834.                 mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);

  835.         }

  836.     }

  837.     inter_pred(h, mb_type);

  838.     if(mb_type != B_SKIP)

  839.         decode_residual_inter(h);

  840.     filter_mb(h,mb_type);

  841. }

  842. 

  843. /*****************************************************************************

  844.  *

  845.  * slice level

  846.  *

  847.  ****************************************************************************/

  848. 

  849. static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {

  850.     if(h->stc > 0xAF)

  851.         av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);

  852.     h->mby = h->stc;

  853.     if((h->mby == 0) && (!h->qp_fixed)){

  854.         h->qp_fixed = get_bits1(gb);

  855.         h->qp = get_bits(gb,6);

  856.     }

  857.     /* inter frame or second slice can have weighting params */

  858.     if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))

  859.         if(get_bits1(gb)) { //slice_weighting_flag

  860.             av_log(h->s.avctx, AV_LOG_ERROR,

  861.                    "weighted prediction not yet supported\n");

  862.         }

  863.     return 0;

  864. }

  865. 

  866. static inline void check_for_slice(AVSContext *h) {

  867.     GetBitContext *gb = &h->s.gb;

  868.     int align;

  869.     align = (-get_bits_count(gb)) & 7;

  870.     if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {

  871.         get_bits_long(gb,24+align);

  872.         h->stc = get_bits(gb,8);

  873.         decode_slice_header(h,gb);

  874.     }

  875. }

  876. 

  877. /*****************************************************************************

  878.  *

  879.  * frame level

  880.  *

  881.  ****************************************************************************/

  882. 

  883. static void init_pic(AVSContext *h) {

  884.     int i;

  885. 

  886.     /* clear some predictors */

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  902.     h->flags = 0;

  903. }

  904. 

  905. static int decode_pic(AVSContext *h) {

  906.     MpegEncContext *s = &h->s;

  907.     int skip_count;

  908.     enum mb_t mb_type;

  909. 

  910.     if (!s->context_initialized) {

  911.         s->avctx->idct_algo = FF_IDCT_CAVS;

  912.         if (MPV_common_init(s) < 0)

  913.             return -1;

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

  915.     }

  916.     get_bits(&s->gb,16);//bbv_dwlay

  917.     if(h->stc == PIC_PB_START_CODE) {

  918.         h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;

  919.         if(h->pic_type > FF_B_TYPE) {

  920.             av_log(s->avctx, AV_LOG_ERROR, "illegal picture type\n");

  921.             return -1;

  922.         }

  923.         /* make sure we have the reference frames we need */

  924.         if(!h->DPB[0].data[0] ||

  925.           (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))

  926.             return -1;

  927.     } else {

  928.         h->pic_type = FF_I_TYPE;

  929.         if(get_bits1(&s->gb))

  930.             get_bits(&s->gb,16);//time_code

  931.     }

  932.     /* release last B frame */

  933.     if(h->picture.data[0])

  934.         s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);

  935. 

  936.     s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);

  937.     init_pic(h);

  938.     h->picture.poc = get_bits(&s->gb,8)*2;

  939. 

  940.     /* get temporal distances and MV scaling factors */

  941.     if(h->pic_type != FF_B_TYPE) {

  942.         h->dist[0] = (h->picture.poc - h->DPB[0].poc  + 512) % 512;

  943.     } else {

  944.         h->dist[0] = (h->DPB[0].poc  - h->picture.poc + 512) % 512;

  945.     }

  946.     h->dist[1] = (h->picture.poc - h->DPB[1].poc  + 512) % 512;

  947.     h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;

  948.     h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;

  949.     if(h->pic_type == FF_B_TYPE) {

  950.         h->sym_factor = h->dist[0]*h->scale_den[1];

  951.     } else {

  952.         h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;

  953.         h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;

  954.     }

  955. 

  956.     if(s->low_delay)

  957.         get_ue_golomb(&s->gb); //bbv_check_times

  958.     h->progressive             = get_bits1(&s->gb);

  959.     if(h->progressive)

  960.         h->pic_structure = 1;

  961.     else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )

  962.         get_bits1(&s->gb);     //advanced_pred_mode_disable

  963.     skip_bits1(&s->gb);        //top_field_first

  964.     skip_bits1(&s->gb);        //repeat_first_field

  965.     h->qp_fixed                = get_bits1(&s->gb);

  966.     h->qp                      = get_bits(&s->gb,6);

  967.     if(h->pic_type == FF_I_TYPE) {

  968.         if(!h->progressive && !h->pic_structure)

  969.             skip_bits1(&s->gb);//what is this?

  970.         skip_bits(&s->gb,4);   //reserved bits

  971.     } else {

  972.         if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))

  973.             h->ref_flag        = get_bits1(&s->gb);

  974.         skip_bits(&s->gb,4);   //reserved bits

  975.         h->skip_mode_flag      = get_bits1(&s->gb);

  976.     }

  977.     h->loop_filter_disable     = get_bits1(&s->gb);

  978.     if(!h->loop_filter_disable && get_bits1(&s->gb)) {

  979.         h->alpha_offset        = get_se_golomb(&s->gb);

  980.         h->beta_offset         = get_se_golomb(&s->gb);

  981.     } else {

  982.         h->alpha_offset = h->beta_offset  = 0;

  983.     }

  984.     check_for_slice(h);

  985.     if(h->pic_type == FF_I_TYPE) {

  986.         do {

  987.             decode_mb_i(h, 0);

  988.         } while(next_mb(h));

  989.     } else if(h->pic_type == FF_P_TYPE) {

  990.         do {

  991.             if(h->skip_mode_flag) {

  992.                 skip_count = get_ue_golomb(&s->gb);

  993.                 while(skip_count--) {

  994.                     decode_mb_p(h,P_SKIP);

  995.                     if(!next_mb(h))

  996.                         goto done;

  997.                 }

  998.                 mb_type = get_ue_golomb(&s->gb) + P_16X16;

  999.             } else

  1000.                 mb_type = get_ue_golomb(&s->gb) + P_SKIP;

  1001.             if(mb_type > P_8X8) {

  1002.                 decode_mb_i(h, mb_type - P_8X8 - 1);

  1003.             } else

  1004.                 decode_mb_p(h,mb_type);

  1005.         } while(next_mb(h));

  1006.     } else { /* FF_B_TYPE */

  1007.         do {

  1008.             if(h->skip_mode_flag) {

  1009.                 skip_count = get_ue_golomb(&s->gb);

  1010.                 while(skip_count--) {

  1011.                     decode_mb_b(h,B_SKIP);

  1012.                     if(!next_mb(h))

  1013.                         goto done;

  1014.                 }

  1015.                 mb_type = get_ue_golomb(&s->gb) + B_DIRECT;

  1016.             } else

  1017.                 mb_type = get_ue_golomb(&s->gb) + B_SKIP;

  1018.             if(mb_type > B_8X8) {

  1019.                 decode_mb_i(h, mb_type - B_8X8 - 1);

  1020.             } else

  1021.                 decode_mb_b(h,mb_type);

  1022.         } while(next_mb(h));

  1023.     }

  1024.  done:

  1025.     if(h->pic_type != FF_B_TYPE) {

  1026.         if(h->DPB[1].data[0])

  1027.             s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);

  1028.         memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));

  1029.         memcpy(&h->DPB[0], &h->picture, sizeof(Picture));

  1030.         memset(&h->picture,0,sizeof(Picture));

  1031.     }

  1032.     return 0;

  1033. }

  1034. 

  1035. /*****************************************************************************

  1036.  *

  1037.  * headers and interface

  1038.  *

  1039.  ****************************************************************************/

  1040. 

  1041. /**

  1042.  * some predictions require data from the top-neighbouring macroblock.

  1043.  * this data has to be stored for one complete row of macroblocks

  1044.  * and this storage space is allocated here

  1045.  */

  1046. static void init_top_lines(AVSContext *h) {

  1047.     /* alloc top line of predictors */

  1048.     h->top_qp       = av_malloc( h->mb_width);

  1049.     h->top_mv[0]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));

  1050.     h->top_mv[1]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));

  1051.     h->top_pred_Y   = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));

  1052.     h->top_border_y = av_malloc((h->mb_width+1)*16);

  1053.     h->top_border_u = av_malloc((h->mb_width)*10);

  1054.     h->top_border_v = av_malloc((h->mb_width)*10);

  1055. 

  1056.     /* alloc space for co-located MVs and types */

  1057.     h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));

  1058.     h->col_type_base = av_malloc(h->mb_width*h->mb_height);

  1059.     h->block        = av_mallocz(64*sizeof(DCTELEM));

  1060. }

  1061. 

  1062. static int decode_seq_header(AVSContext *h) {

  1063.     MpegEncContext *s = &h->s;

  1064.     int frame_rate_code;

  1065. 

  1066.     h->profile =         get_bits(&s->gb,8);

  1067.     h->level =           get_bits(&s->gb,8);

  1068.     skip_bits1(&s->gb); //progressive sequence

  1069.     s->width =           get_bits(&s->gb,14);

  1070.     s->height =          get_bits(&s->gb,14);

  1071.     skip_bits(&s->gb,2); //chroma format

  1072.     skip_bits(&s->gb,3); //sample_precision

  1073.     h->aspect_ratio =    get_bits(&s->gb,4);

  1074.     frame_rate_code =    get_bits(&s->gb,4);

  1075.     skip_bits(&s->gb,18);//bit_rate_lower

  1076.     skip_bits1(&s->gb);  //marker_bit

  1077.     skip_bits(&s->gb,12);//bit_rate_upper

  1078.     s->low_delay =       get_bits1(&s->gb);

  1079.     h->mb_width  = (s->width  + 15) >> 4;

  1080.     h->mb_height = (s->height + 15) >> 4;

  1081.     h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;

  1082.     h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;

  1083.     h->s.avctx->width  = s->width;

  1084.     h->s.avctx->height = s->height;

  1085.     if(!h->top_qp)

  1086.         init_top_lines(h);

  1087.     return 0;

  1088. }

  1089. 

  1090. static void cavs_flush(AVCodecContext * avctx) {

  1091.     AVSContext *h = avctx->priv_data;

  1092.     h->got_keyframe = 0;

  1093. }

  1094. 

  1095. static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,

  1096.                              uint8_t * buf, int buf_size) {

  1097.     AVSContext *h = avctx->priv_data;

  1098.     MpegEncContext *s = &h->s;

  1099.     int input_size;

  1100.     const uint8_t *buf_end;

  1101.     const uint8_t *buf_ptr;

  1102.     AVFrame *picture = data;

  1103.     uint32_t stc;

  1104. 

  1105.     s->avctx = avctx;

  1106. 

  1107.     if (buf_size == 0) {

  1108.         if(!s->low_delay && h->DPB[0].data[0]) {

  1109.             *data_size = sizeof(AVPicture);

  1110.             *picture = *(AVFrame *) &h->DPB[0];

  1111.         }

  1112.         return 0;

  1113.     }

  1114. 

  1115.     buf_ptr = buf;

  1116.     buf_end = buf + buf_size;

  1117.     for(;;) {

  1118.         buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);

  1119.         if(stc & 0xFFFFFE00)

  1120.             return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);

  1121.         input_size = (buf_end - buf_ptr)*8;

  1122.         switch(stc) {

  1123.         case CAVS_START_CODE:

  1124.             init_get_bits(&s->gb, buf_ptr, input_size);

  1125.             decode_seq_header(h);

  1126.             break;

  1127.         case PIC_I_START_CODE:

  1128.             if(!h->got_keyframe) {

  1129.                 if(h->DPB[0].data[0])

  1130.                     avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);

  1131.                 if(h->DPB[1].data[0])

  1132.                     avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);

  1133.                 h->got_keyframe = 1;

  1134.             }

  1135.         case PIC_PB_START_CODE:

  1136.             *data_size = 0;

  1137.             if(!h->got_keyframe)

  1138.                 break;

  1139.             init_get_bits(&s->gb, buf_ptr, input_size);

  1140.             h->stc = stc;

  1141.             if(decode_pic(h))

  1142.                 break;

  1143.             *data_size = sizeof(AVPicture);

  1144.             if(h->pic_type != FF_B_TYPE) {

  1145.                 if(h->DPB[1].data[0]) {

  1146.                     *picture = *(AVFrame *) &h->DPB[1];

  1147.                 } else {

  1148.                     *data_size = 0;

  1149.                 }

  1150.             } else

  1151.                 *picture = *(AVFrame *) &h->picture;

  1152.             break;

  1153.         case EXT_START_CODE:

  1154.             //mpeg_decode_extension(avctx,buf_ptr, input_size);

  1155.             break;

  1156.         case USER_START_CODE:

  1157.             //mpeg_decode_user_data(avctx,buf_ptr, input_size);

  1158.             break;

  1159.         default:

  1160.             if (stc >= SLICE_MIN_START_CODE &&

  1161.                 stc <= SLICE_MAX_START_CODE) {

  1162.                 init_get_bits(&s->gb, buf_ptr, input_size);

  1163.                 decode_slice_header(h, &s->gb);

  1164.             }

  1165.             break;

  1166.         }

  1167.     }

  1168. }

  1169. 

  1170. static int cavs_decode_init(AVCodecContext * avctx) {

  1171.     AVSContext *h = avctx->priv_data;

  1172.     MpegEncContext * const s = &h->s;

  1173. 

  1174.     MPV_decode_defaults(s);

  1175.     s->avctx = avctx;

  1176. 

  1177.     avctx->pix_fmt= PIX_FMT_YUV420P;

  1178. 

  1179.     h->luma_scan[0] = 0;

  1180.     h->luma_scan[1] = 8;

  1181.     h->intra_pred_l[      INTRA_L_VERT] = intra_pred_vert;

  1182.     h->intra_pred_l[     INTRA_L_HORIZ] = intra_pred_horiz;

  1183.     h->intra_pred_l[        INTRA_L_LP] = intra_pred_lp;

  1184.     h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;

  1185.     h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;

  1186.     h->intra_pred_l[   INTRA_L_LP_LEFT] = intra_pred_lp_left;

  1187.     h->intra_pred_l[    INTRA_L_LP_TOP] = intra_pred_lp_top;

  1188.     h->intra_pred_l[    INTRA_L_DC_128] = intra_pred_dc_128;

  1189.     h->intra_pred_c[        INTRA_C_LP] = intra_pred_lp;

  1190.     h->intra_pred_c[     INTRA_C_HORIZ] = intra_pred_horiz;

  1191.     h->intra_pred_c[      INTRA_C_VERT] = intra_pred_vert;

  1192.     h->intra_pred_c[     INTRA_C_PLANE] = intra_pred_plane;

  1193.     h->intra_pred_c[   INTRA_C_LP_LEFT] = intra_pred_lp_left;

  1194.     h->intra_pred_c[    INTRA_C_LP_TOP] = intra_pred_lp_top;

  1195.     h->intra_pred_c[    INTRA_C_DC_128] = intra_pred_dc_128;

  1196.     h->mv[ 7] = ff_cavs_un_mv;

  1197.     h->mv[19] = ff_cavs_un_mv;

  1198.     return 0;

  1199. }

  1200. 

  1201. static int cavs_decode_end(AVCodecContext * avctx) {

  1202.     AVSContext *h = avctx->priv_data;

  1203. 

  1204.     av_free(h->top_qp);

  1205.     av_free(h->top_mv[0]);

  1206.     av_free(h->top_mv[1]);

  1207.     av_free(h->top_pred_Y);

  1208.     av_free(h->top_border_y);

  1209.     av_free(h->top_border_u);

  1210.     av_free(h->top_border_v);

  1211.     av_free(h->col_mv);

  1212.     av_free(h->col_type_base);

  1213.     av_free(h->block);

  1214.     return 0;

  1215. }

  1216. 

  1217. AVCodec cavs_decoder = {

  1218.     "cavs",

  1219.     CODEC_TYPE_VIDEO,

  1220.     CODEC_ID_CAVS,

  1221.     sizeof(AVSContext),

  1222.     cavs_decode_init,

  1223.     NULL,

  1224.     cavs_decode_end,

  1225.     cavs_decode_frame,

  1226.     CODEC_CAP_DR1 | CODEC_CAP_DELAY,

  1227.     .flush= cavs_flush,

  1228. };