Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases 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.
22750 Commits
37 Branches
400 Tags
464 MiB
Tag: Branch: Tree: 83a6c1a69c
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '83a6c1a69c'
${ noResults }
FFmpeg/libavcodec/h264_cavlc.c

1031 lines
37 KiB
Raw Normal View History Unescape

Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
/*
* H.26L/H.264/AVC/JVT/14496-10/... cavlc bitstream decoding
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file libavcodec/h264_cavlc.c
* H.264 / AVC / MPEG4 part10 cavlc bitstream decoding.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
Replace cabac checks in inline functions from h264.h with constants. No benchmark because its just replacing variables with litteral constants (so no risk for slowdown outside gcc silliness) and i need sleep. Originally committed as revision 21237 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
#define CABAC 0
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
#include "internal.h"
#include "avcodec.h"
#include "mpegvideo.h"
#include "h264.h"
#include "h264data.h" // FIXME FIXME FIXME
#include "h264_mvpred.h"
#include "golomb.h"
//#undef NDEBUG
#include <assert.h>
static const uint8_t golomb_to_inter_cbp_gray[16]={
0, 1, 2, 4, 8, 3, 5,10,12,15, 7,11,13,14, 6, 9,
};
static const uint8_t golomb_to_intra4x4_cbp_gray[16]={
15, 0, 7,11,13,14, 3, 5,10,12, 1, 2, 4, 8, 6, 9,
};
static const uint8_t chroma_dc_coeff_token_len[4*5]={
2, 0, 0, 0,
6, 1, 0, 0,
6, 6, 3, 0,
6, 7, 7, 6,
6, 8, 8, 7,
};
static const uint8_t chroma_dc_coeff_token_bits[4*5]={
1, 0, 0, 0,
7, 1, 0, 0,
4, 6, 1, 0,
3, 3, 2, 5,
2, 3, 2, 0,
};
static const uint8_t coeff_token_len[4][4*17]={
{
1, 0, 0, 0,
6, 2, 0, 0, 8, 6, 3, 0, 9, 8, 7, 5, 10, 9, 8, 6,
11,10, 9, 7, 13,11,10, 8, 13,13,11, 9, 13,13,13,10,
14,14,13,11, 14,14,14,13, 15,15,14,14, 15,15,15,14,
16,15,15,15, 16,16,16,15, 16,16,16,16, 16,16,16,16,
},
{
2, 0, 0, 0,
6, 2, 0, 0, 6, 5, 3, 0, 7, 6, 6, 4, 8, 6, 6, 4,
8, 7, 7, 5, 9, 8, 8, 6, 11, 9, 9, 6, 11,11,11, 7,
12,11,11, 9, 12,12,12,11, 12,12,12,11, 13,13,13,12,
13,13,13,13, 13,14,13,13, 14,14,14,13, 14,14,14,14,
},
{
4, 0, 0, 0,
6, 4, 0, 0, 6, 5, 4, 0, 6, 5, 5, 4, 7, 5, 5, 4,
7, 5, 5, 4, 7, 6, 6, 4, 7, 6, 6, 4, 8, 7, 7, 5,
8, 8, 7, 6, 9, 8, 8, 7, 9, 9, 8, 8, 9, 9, 9, 8,
10, 9, 9, 9, 10,10,10,10, 10,10,10,10, 10,10,10,10,
},
{
6, 0, 0, 0,
6, 6, 0, 0, 6, 6, 6, 0, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
}
};
static const uint8_t coeff_token_bits[4][4*17]={
{
1, 0, 0, 0,
5, 1, 0, 0, 7, 4, 1, 0, 7, 6, 5, 3, 7, 6, 5, 3,
7, 6, 5, 4, 15, 6, 5, 4, 11,14, 5, 4, 8,10,13, 4,
15,14, 9, 4, 11,10,13,12, 15,14, 9,12, 11,10,13, 8,
15, 1, 9,12, 11,14,13, 8, 7,10, 9,12, 4, 6, 5, 8,
},
{
3, 0, 0, 0,
11, 2, 0, 0, 7, 7, 3, 0, 7,10, 9, 5, 7, 6, 5, 4,
4, 6, 5, 6, 7, 6, 5, 8, 15, 6, 5, 4, 11,14,13, 4,
15,10, 9, 4, 11,14,13,12, 8,10, 9, 8, 15,14,13,12,
11,10, 9,12, 7,11, 6, 8, 9, 8,10, 1, 7, 6, 5, 4,
},
{
15, 0, 0, 0,
15,14, 0, 0, 11,15,13, 0, 8,12,14,12, 15,10,11,11,
11, 8, 9,10, 9,14,13, 9, 8,10, 9, 8, 15,14,13,13,
11,14,10,12, 15,10,13,12, 11,14, 9,12, 8,10,13, 8,
13, 7, 9,12, 9,12,11,10, 5, 8, 7, 6, 1, 4, 3, 2,
},
{
3, 0, 0, 0,
0, 1, 0, 0, 4, 5, 6, 0, 8, 9,10,11, 12,13,14,15,
16,17,18,19, 20,21,22,23, 24,25,26,27, 28,29,30,31,
32,33,34,35, 36,37,38,39, 40,41,42,43, 44,45,46,47,
48,49,50,51, 52,53,54,55, 56,57,58,59, 60,61,62,63,
}
};
static const uint8_t total_zeros_len[16][16]= {
{1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9},
{3,3,3,3,3,4,4,4,4,5,5,6,6,6,6},
{4,3,3,3,4,4,3,3,4,5,5,6,5,6},
{5,3,4,4,3,3,3,4,3,4,5,5,5},
{4,4,4,3,3,3,3,3,4,5,4,5},
{6,5,3,3,3,3,3,3,4,3,6},
{6,5,3,3,3,2,3,4,3,6},
{6,4,5,3,2,2,3,3,6},
{6,6,4,2,2,3,2,5},
{5,5,3,2,2,2,4},
{4,4,3,3,1,3},
{4,4,2,1,3},
{3,3,1,2},
{2,2,1},
{1,1},
};
static const uint8_t total_zeros_bits[16][16]= {
{1,3,2,3,2,3,2,3,2,3,2,3,2,3,2,1},
{7,6,5,4,3,5,4,3,2,3,2,3,2,1,0},
{5,7,6,5,4,3,4,3,2,3,2,1,1,0},
{3,7,5,4,6,5,4,3,3,2,2,1,0},
{5,4,3,7,6,5,4,3,2,1,1,0},
{1,1,7,6,5,4,3,2,1,1,0},
{1,1,5,4,3,3,2,1,1,0},
{1,1,1,3,3,2,2,1,0},
{1,0,1,3,2,1,1,1},
{1,0,1,3,2,1,1},
{0,1,1,2,1,3},
{0,1,1,1,1},
{0,1,1,1},
{0,1,1},
{0,1},
};
static const uint8_t chroma_dc_total_zeros_len[3][4]= {
{ 1, 2, 3, 3,},
{ 1, 2, 2, 0,},
{ 1, 1, 0, 0,},
};
static const uint8_t chroma_dc_total_zeros_bits[3][4]= {
{ 1, 1, 1, 0,},
{ 1, 1, 0, 0,},
{ 1, 0, 0, 0,},
};
static const uint8_t run_len[7][16]={
{1,1},
{1,2,2},
{2,2,2,2},
{2,2,2,3,3},
{2,2,3,3,3,3},
{2,3,3,3,3,3,3},
{3,3,3,3,3,3,3,4,5,6,7,8,9,10,11},
};
static const uint8_t run_bits[7][16]={
{1,0},
{1,1,0},
{3,2,1,0},
{3,2,1,1,0},
{3,2,3,2,1,0},
{3,0,1,3,2,5,4},
{7,6,5,4,3,2,1,1,1,1,1,1,1,1,1},
};
static VLC coeff_token_vlc[4];
static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
static VLC chroma_dc_coeff_token_vlc;
static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
static const int chroma_dc_coeff_token_vlc_table_size = 256;
static VLC total_zeros_vlc[15];
static VLC_TYPE total_zeros_vlc_tables[15][512][2];
static const int total_zeros_vlc_tables_size = 512;
static VLC chroma_dc_total_zeros_vlc[3];
static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
static const int chroma_dc_total_zeros_vlc_tables_size = 8;
static VLC run_vlc[6];
static VLC_TYPE run_vlc_tables[6][8][2];
static const int run_vlc_tables_size = 8;
static VLC run7_vlc;
static VLC_TYPE run7_vlc_table[96][2];
static const int run7_vlc_table_size = 96;
#define LEVEL_TAB_BITS 8
static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
/**
* gets the predicted number of non-zero coefficients.
* @param n block index
*/
static inline int pred_non_zero_count(H264Context *h, int n){
const int index8= scan8[n];
const int left= h->non_zero_count_cache[index8 - 1];
const int top = h->non_zero_count_cache[index8 - 8];
int i= left + top;
if(i<64) i= (i+1)>>1;
tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
return i&31;
}
static av_cold void init_cavlc_level_tab(void){
int suffix_length, mask;
unsigned int i;
for(suffix_length=0; suffix_length<7; suffix_length++){
for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
int prefix= LEVEL_TAB_BITS - av_log2(2*i);
int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
mask= -(level_code&1);
level_code= (((2+level_code)>>1) ^ mask) - mask;
if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
cavlc_level_tab[suffix_length][i][0]= level_code;
cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
}else if(prefix + 1 <= LEVEL_TAB_BITS){
cavlc_level_tab[suffix_length][i][0]= prefix+100;
cavlc_level_tab[suffix_length][i][1]= prefix + 1;
}else{
cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
}
}
}
}
av_cold void ff_h264_decode_init_vlc(void){
static int done = 0;
if (!done) {
int i;
int offset;
done = 1;
chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
&chroma_dc_coeff_token_len [0], 1, 1,
&chroma_dc_coeff_token_bits[0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
offset = 0;
for(i=0; i<4; i++){
coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
&coeff_token_len [i][0], 1, 1,
&coeff_token_bits[i][0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
offset += coeff_token_vlc_tables_size[i];
}
/*
* This is a one time safety check to make sure that
* the packed static coeff_token_vlc table sizes
* were initialized correctly.
*/
assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
for(i=0; i<3; i++){
chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
init_vlc(&chroma_dc_total_zeros_vlc[i],
CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
&chroma_dc_total_zeros_len [i][0], 1, 1,
&chroma_dc_total_zeros_bits[i][0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
}
for(i=0; i<15; i++){
total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
init_vlc(&total_zeros_vlc[i],
TOTAL_ZEROS_VLC_BITS, 16,
&total_zeros_len [i][0], 1, 1,
&total_zeros_bits[i][0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
}
for(i=0; i<6; i++){
run_vlc[i].table = run_vlc_tables[i];
run_vlc[i].table_allocated = run_vlc_tables_size;
init_vlc(&run_vlc[i],
RUN_VLC_BITS, 7,
&run_len [i][0], 1, 1,
&run_bits[i][0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
}
run7_vlc.table = run7_vlc_table,
run7_vlc.table_allocated = run7_vlc_table_size;
init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
&run_len [6][0], 1, 1,
&run_bits[6][0], 1, 1,
INIT_VLC_USE_NEW_STATIC);
init_cavlc_level_tab();
}
}
/**
*
*/
static inline int get_level_prefix(GetBitContext *gb){
unsigned int buf;
int log;
OPEN_READER(re, gb);
UPDATE_CACHE(re, gb);
buf=GET_CACHE(re, gb);
log= 32 - av_log2(buf);
#ifdef TRACE
print_bin(buf>>(32-log), log);
av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
#endif
LAST_SKIP_BITS(re, gb, log);
CLOSE_READER(re, gb);
return log-1;
}
/**
* decodes a residual block.
* @param n block index
* @param scantable scantable
* @param max_coeff number of coefficients in the block
* @return <0 if an error occurred
*/
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
MpegEncContext * const s = &h->s;
static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
int level[16];
int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
//FIXME put trailing_onex into the context
if(n == CHROMA_DC_BLOCK_INDEX){
coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
total_coeff= coeff_token>>2;
}else{
if(n == LUMA_DC_BLOCK_INDEX){
total_coeff= pred_non_zero_count(h, 0);
coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
total_coeff= coeff_token>>2;
}else{
total_coeff= pred_non_zero_count(h, n);
coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
total_coeff= coeff_token>>2;
h->non_zero_count_cache[ scan8[n] ]= total_coeff;
}
}
//FIXME set last_non_zero?
if(total_coeff==0)
return 0;
if(total_coeff > (unsigned)max_coeff) {
av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
return -1;
}
trailing_ones= coeff_token&3;
tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
assert(total_coeff<=16);
i = show_bits(gb, 3);
skip_bits(gb, trailing_ones);
level[0] = 1-((i&4)>>1);
level[1] = 1-((i&2) );
level[2] = 1-((i&1)<<1);
if(trailing_ones<total_coeff) {
int mask, prefix;
1 cpu cycle faster suffix_length calculation. Originally committed as revision 21425 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
int suffix_length = total_coeff > 10 & trailing_ones < 3;
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
int bitsi= show_bits(gb, LEVEL_TAB_BITS);
int level_code= cavlc_level_tab[suffix_length][bitsi][0];
skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
if(level_code >= 100){
prefix= level_code - 100;
if(prefix == LEVEL_TAB_BITS)
prefix += get_level_prefix(gb);
//first coefficient has suffix_length equal to 0 or 1
if(prefix<14){ //FIXME try to build a large unified VLC table for all this
if(suffix_length)
level_code= (prefix<<1) + get_bits1(gb); //part
else
level_code= prefix; //part
}else if(prefix==14){
if(suffix_length)
level_code= (prefix<<1) + get_bits1(gb); //part
else
level_code= prefix + get_bits(gb, 4); //part
}else{
level_code= 30 + get_bits(gb, prefix-3); //part
if(prefix>=16)
level_code += (1<<(prefix-3))-4096;
}
if(trailing_ones < 3) level_code += 2;
suffix_length = 2;
mask= -(level_code&1);
level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
}else{
Optimize level_code computation, 6cpu cycles speedup. Originally committed as revision 21428 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
level_code += ((level_code>>31)|1) & -(trailing_ones < 3);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
Simplify suffix_length computation, same speed. Originally committed as revision 21430 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
suffix_length = 1 + (level_code + 3U > 6U);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
level[trailing_ones]= level_code;
}
//remaining coefficients have suffix_length > 0
for(i=trailing_ones+1;i<total_coeff;i++) {
static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
int bitsi= show_bits(gb, LEVEL_TAB_BITS);
level_code= cavlc_level_tab[suffix_length][bitsi][0];
skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
if(level_code >= 100){
prefix= level_code - 100;
if(prefix == LEVEL_TAB_BITS){
prefix += get_level_prefix(gb);
}
if(prefix<15){
level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
}else{
level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
if(prefix>=16)
level_code += (1<<(prefix-3))-4096;
}
mask= -(level_code&1);
level_code= (((2+level_code)>>1) ^ mask) - mask;
}
level[i]= level_code;
Optimize suffix_length computation, 1 cpu cycle speedup. Originally committed as revision 21431 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
suffix_length+= suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length];
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}
}
if(total_coeff == max_coeff)
zeros_left=0;
else{
if(n == CHROMA_DC_BLOCK_INDEX)
a[b-1] -> (a-1)[b]. Helps gcc not to add seperate -1 instructions. Originally committed as revision 21432 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
zeros_left= get_vlc2(gb, (chroma_dc_total_zeros_vlc-1)[ total_coeff ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
else
a[b-1] -> (a-1)[b]. Helps gcc not to add seperate -1 instructions. Originally committed as revision 21432 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
zeros_left= get_vlc2(gb, (total_zeros_vlc-1)[ total_coeff ].table, TOTAL_ZEROS_VLC_BITS, 1);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}
coeff_num = zeros_left + total_coeff - 1;
j = scantable[coeff_num];
if(n > 24){
block[j] = level[0];
for(i=1;i<total_coeff;i++) {
if(zeros_left <= 0)
run_before = 0;
else if(zeros_left < 7){
a[b-1] -> (a-1)[b]. Helps gcc not to add seperate -1 instructions. Originally committed as revision 21432 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}else{
run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
}
zeros_left -= run_before;
coeff_num -= 1 + run_before;
j= scantable[ coeff_num ];
block[j]= level[i];
}
}else{
block[j] = (level[0] * qmul[j] + 32)>>6;
for(i=1;i<total_coeff;i++) {
if(zeros_left <= 0)
run_before = 0;
else if(zeros_left < 7){
a[b-1] -> (a-1)[b]. Helps gcc not to add seperate -1 instructions. Originally committed as revision 21432 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}else{
run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
}
zeros_left -= run_before;
coeff_num -= 1 + run_before;
j= scantable[ coeff_num ];
block[j]= (level[i] * qmul[j] + 32)>>6;
}
}
if(zeros_left<0){
av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
return -1;
}
return 0;
}
int ff_h264_decode_mb_cavlc(H264Context *h){
MpegEncContext * const s = &h->s;
int mb_xy;
int partition_count;
unsigned int mb_type, cbp;
int dct8x8_allowed= h->pps.transform_8x8_mode;
mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
cbp = 0; /* avoid warning. FIXME: find a solution without slowing
down the code */
if(h->slice_type_nos != FF_I_TYPE){
if(s->mb_skip_run==-1)
s->mb_skip_run= get_ue_golomb(&s->gb);
if (s->mb_skip_run--) {
if(FRAME_MBAFF && (s->mb_y&1) == 0){
if(s->mb_skip_run==0)
h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
else
predict_field_decoding_flag(h);
}
decode_mb_skip(h);
return 0;
}
}
if(FRAME_MBAFF){
if( (s->mb_y&1) == 0 )
h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
}
h->prev_mb_skipped= 0;
mb_type= get_ue_golomb(&s->gb);
if(h->slice_type_nos == FF_B_TYPE){
if(mb_type < 23){
partition_count= b_mb_type_info[mb_type].partition_count;
mb_type= b_mb_type_info[mb_type].type;
}else{
mb_type -= 23;
goto decode_intra_mb;
}
}else if(h->slice_type_nos == FF_P_TYPE){
if(mb_type < 5){
partition_count= p_mb_type_info[mb_type].partition_count;
mb_type= p_mb_type_info[mb_type].type;
}else{
mb_type -= 5;
goto decode_intra_mb;
}
}else{
assert(h->slice_type_nos == FF_I_TYPE);
if(h->slice_type == FF_SI_TYPE && mb_type)
mb_type--;
decode_intra_mb:
if(mb_type > 25){
av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
return -1;
}
partition_count=0;
cbp= i_mb_type_info[mb_type].cbp;
h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
mb_type= i_mb_type_info[mb_type].type;
}
if(MB_FIELD)
mb_type |= MB_TYPE_INTERLACED;
h->slice_table[ mb_xy ]= h->slice_num;
if(IS_INTRA_PCM(mb_type)){
unsigned int x;
// We assume these blocks are very rare so we do not optimize it.
align_get_bits(&s->gb);
// The pixels are stored in the same order as levels in h->mb array.
for(x=0; x < (CHROMA ? 384 : 256); x++){
((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
}
// In deblocking, the quantizer is 0
s->current_picture.qscale_table[mb_xy]= 0;
// All coeffs are present
Rearchitecturing the stiched up goose part 1 Run loop filter per row instead of per MB, this also should make it much easier to switch to per frame filtering and also doing so in a seperate thread in the future if some volunteer wants to try. Overall decoding speedup of 1.7% (single thread on pentium dual / cathedral sample) This change also allows some optimizations to be tried that would not have been possible before. Originally committed as revision 21270 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
memset(h->non_zero_count[mb_xy], 16, 32);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
s->current_picture.mb_type[mb_xy]= mb_type;
return 0;
}
if(MB_MBAFF){
h->ref_count[0] <<= 1;
h->ref_count[1] <<= 1;
}
Split fill_caches() between filter and decoder. Originally committed as revision 21271 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
fill_decode_caches(h, mb_type);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
//mb_pred
if(IS_INTRA(mb_type)){
int pred_mode;
// init_top_left_availability(h);
if(IS_INTRA4x4(mb_type)){
int i;
int di = 1;
if(dct8x8_allowed && get_bits1(&s->gb)){
mb_type |= MB_TYPE_8x8DCT;
di = 4;
}
// fill_intra4x4_pred_table(h);
for(i=0; i<16; i+=di){
int mode= pred_intra_mode(h, i);
if(!get_bits1(&s->gb)){
const int rem_mode= get_bits(&s->gb, 3);
mode = rem_mode + (rem_mode >= mode);
}
if(di==4)
fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
else
h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
}
ff_h264_write_back_intra_pred_mode(h);
if( ff_h264_check_intra4x4_pred_mode(h) < 0)
return -1;
}else{
h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode(h, h->intra16x16_pred_mode);
if(h->intra16x16_pred_mode < 0)
return -1;
}
if(CHROMA){
pred_mode= ff_h264_check_intra_pred_mode(h, get_ue_golomb_31(&s->gb));
if(pred_mode < 0)
return -1;
h->chroma_pred_mode= pred_mode;
}
}else if(partition_count==4){
int i, j, sub_partition_count[4], list, ref[2][4];
if(h->slice_type_nos == FF_B_TYPE){
for(i=0; i<4; i++){
h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
if(h->sub_mb_type[i] >=13){
av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
return -1;
}
sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
}
Merge multiple IS_* macro uses where possible. Originally committed as revision 21340 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
if( IS_DIRECT(h->sub_mb_type[0]|h->sub_mb_type[1]|h->sub_mb_type[2]|h->sub_mb_type[3])) {
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
ff_h264_pred_direct_motion(h, &mb_type);
h->ref_cache[0][scan8[4]] =
h->ref_cache[1][scan8[4]] =
h->ref_cache[0][scan8[12]] =
h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
}
}else{
assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
for(i=0; i<4; i++){
h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
if(h->sub_mb_type[i] >=4){
av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
return -1;
}
sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
}
}
for(list=0; list<h->list_count; list++){
int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
for(i=0; i<4; i++){
if(IS_DIRECT(h->sub_mb_type[i])) continue;
if(IS_DIR(h->sub_mb_type[i], 0, list)){
unsigned int tmp;
if(ref_count == 1){
tmp= 0;
}else if(ref_count == 2){
tmp= get_bits1(&s->gb)^1;
}else{
tmp= get_ue_golomb_31(&s->gb);
if(tmp>=ref_count){
av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
return -1;
}
}
ref[list][i]= tmp;
}else{
//FIXME
ref[list][i] = -1;
}
}
}
if(dct8x8_allowed)
dct8x8_allowed = get_dct8x8_allowed(h);
for(list=0; list<h->list_count; list++){
for(i=0; i<4; i++){
if(IS_DIRECT(h->sub_mb_type[i])) {
h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
continue;
}
h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
if(IS_DIR(h->sub_mb_type[i], 0, list)){
const int sub_mb_type= h->sub_mb_type[i];
const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
for(j=0; j<sub_partition_count[i]; j++){
int mx, my;
const int index= 4*i + block_width*j;
int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
mx += get_se_golomb(&s->gb);
my += get_se_golomb(&s->gb);
tprintf(s->avctx, "final mv:%d %d\n", mx, my);
if(IS_SUB_8X8(sub_mb_type)){
mv_cache[ 1 ][0]=
mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
mv_cache[ 1 ][1]=
mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
}else if(IS_SUB_8X4(sub_mb_type)){
mv_cache[ 1 ][0]= mx;
mv_cache[ 1 ][1]= my;
}else if(IS_SUB_4X8(sub_mb_type)){
mv_cache[ 8 ][0]= mx;
mv_cache[ 8 ][1]= my;
}
mv_cache[ 0 ][0]= mx;
mv_cache[ 0 ][1]= my;
}
}else{
uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
p[0] = p[1]=
p[8] = p[9]= 0;
}
}
}
}else if(IS_DIRECT(mb_type)){
ff_h264_pred_direct_motion(h, &mb_type);
dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
}else{
int list, mx, my, i;
//FIXME we should set ref_idx_l? to 0 if we use that later ...
if(IS_16X16(mb_type)){
for(list=0; list<h->list_count; list++){
unsigned int val;
if(IS_DIR(mb_type, 0, list)){
if(h->ref_count[list]==1){
val= 0;
}else if(h->ref_count[list]==2){
val= get_bits1(&s->gb)^1;
}else{
val= get_ue_golomb_31(&s->gb);
if(val >= h->ref_count[list]){
av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
return -1;
}
}
}else
val= LIST_NOT_USED&0xFF;
fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
}
for(list=0; list<h->list_count; list++){
unsigned int val;
if(IS_DIR(mb_type, 0, list)){
pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
mx += get_se_golomb(&s->gb);
my += get_se_golomb(&s->gb);
tprintf(s->avctx, "final mv:%d %d\n", mx, my);
val= pack16to32(mx,my);
}else
val=0;
fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, val, 4);
}
}
else if(IS_16X8(mb_type)){
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
unsigned int val;
if(IS_DIR(mb_type, i, list)){
if(h->ref_count[list] == 1){
val= 0;
}else if(h->ref_count[list] == 2){
val= get_bits1(&s->gb)^1;
}else{
val= get_ue_golomb_31(&s->gb);
if(val >= h->ref_count[list]){
av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
return -1;
}
}
}else
val= LIST_NOT_USED&0xFF;
fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
}
}
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
unsigned int val;
if(IS_DIR(mb_type, i, list)){
pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
mx += get_se_golomb(&s->gb);
my += get_se_golomb(&s->gb);
tprintf(s->avctx, "final mv:%d %d\n", mx, my);
val= pack16to32(mx,my);
}else
val=0;
fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
}
}
}else{
assert(IS_8X16(mb_type));
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
unsigned int val;
if(IS_DIR(mb_type, i, list)){ //FIXME optimize
if(h->ref_count[list]==1){
val= 0;
}else if(h->ref_count[list]==2){
val= get_bits1(&s->gb)^1;
}else{
val= get_ue_golomb_31(&s->gb);
if(val >= h->ref_count[list]){
av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
return -1;
}
}
}else
val= LIST_NOT_USED&0xFF;
fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
}
}
for(list=0; list<h->list_count; list++){
for(i=0; i<2; i++){
unsigned int val;
if(IS_DIR(mb_type, i, list)){
pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
mx += get_se_golomb(&s->gb);
my += get_se_golomb(&s->gb);
tprintf(s->avctx, "final mv:%d %d\n", mx, my);
val= pack16to32(mx,my);
}else
val=0;
fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
}
}
}
}
if(IS_INTER(mb_type))
write_back_motion(h, mb_type);
if(!IS_INTRA16x16(mb_type)){
cbp= get_ue_golomb(&s->gb);
if(cbp > 47){
av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, s->mb_x, s->mb_y);
return -1;
}
if(CHROMA){
if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
else cbp= golomb_to_inter_cbp [cbp];
}else{
if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
else cbp= golomb_to_inter_cbp_gray[cbp];
}
}
if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
Branchless setting of MB_TYPE_8x8DCT. Not benchmarked as i failed to find a sample that uses this one. But it should be faster. Originally committed as revision 21435 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
mb_type |= MB_TYPE_8x8DCT*get_bits1(&s->gb);
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}
filter_mb_fast needs cbp_table to be set. Originally committed as revision 21290 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
h->cbp=
h->cbp_table[mb_xy]= cbp;
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
s->current_picture.mb_type[mb_xy]= mb_type;
if(cbp || IS_INTRA16x16(mb_type)){
int i8x8, i4x4, chroma_idx;
int dquant;
GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
const uint8_t *scan, *scan8x8, *dc_scan;
if(IS_INTERLACED(mb_type)){
scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
scan= s->qscale ? h->field_scan : h->field_scan_q0;
dc_scan= luma_dc_field_scan;
}else{
scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
dc_scan= luma_dc_zigzag_scan;
}
dquant= get_se_golomb(&s->gb);
s->qscale += dquant;
Move dquant check into qscale overflow check. This should be faster (couldnt meassue a difference), and its less picky on slightly out of spec dquant. Originally committed as revision 21373 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
if(((unsigned)s->qscale) > 51){
if(s->qscale<0) s->qscale+= 52;
else s->qscale-= 52;
Move dquant check into qscale overflow check. This should be faster (couldnt meassue a difference), and its less picky on slightly out of spec dquant. Originally committed as revision 21373 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
if(((unsigned)s->qscale) > 51){
av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
return -1;
}
Split cavlc out of h264.c. Seems to speed the code up a little... The placement of many generic functions between h264.c and h264.h is still open Currently they are a little randomly placed between them. Originally committed as revision 21178 to svn://svn.ffmpeg.org/ffmpeg/trunk
15 years ago
}
h->chroma_qp[0]= get_chroma_qp(h, 0, s->qscale);
h->chroma_qp[1]= get_chroma_qp(h, 1, s->qscale);
if(IS_INTRA16x16(mb_type)){
if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){
return -1; //FIXME continue if partitioned and other return -1 too
}
assert((cbp&15) == 0 || (cbp&15) == 15);
if(cbp&15){
for(i8x8=0; i8x8<4; i8x8++){
for(i4x4=0; i4x4<4; i4x4++){
const int index= i4x4 + 4*i8x8;
if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
return -1;
}
}
}
}else{
fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
}
}else{
for(i8x8=0; i8x8<4; i8x8++){
if(cbp & (1<<i8x8)){
if(IS_8x8DCT(mb_type)){
DCTELEM *buf = &h->mb[64*i8x8];
uint8_t *nnz;
for(i4x4=0; i4x4<4; i4x4++){
if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,
h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )
return -1;
}
nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
nnz[0] += nnz[1] + nnz[8] + nnz[9];
}else{
for(i4x4=0; i4x4<4; i4x4++){
const int index= i4x4 + 4*i8x8;
if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
return -1;
}
}
}
}else{
uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
}
}
}
if(cbp&0x30){
for(chroma_idx=0; chroma_idx<2; chroma_idx++)
if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){
return -1;
}
}
if(cbp&0x20){
for(chroma_idx=0; chroma_idx<2; chroma_idx++){
const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
for(i4x4=0; i4x4<4; i4x4++){
const int index= 16 + 4*chroma_idx + i4x4;
if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, qmul, 15) < 0){
return -1;
}
}
}
}else{
uint8_t * const nnz= &h->non_zero_count_cache[0];
nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
}
}else{
uint8_t * const nnz= &h->non_zero_count_cache[0];
fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
}
s->current_picture.qscale_table[mb_xy]= s->qscale;
write_back_non_zero_count(h);
if(MB_MBAFF){
h->ref_count[0] >>= 1;
h->ref_count[1] >>= 1;
}
return 0;
}