Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Split filter_mb_dir() out of filter_mb().

Browse Source 
1% overall decoding speed up for cathedral-beta2-400extra-crop-avc.mp4
no speed change for Aladin.mpg
Benchmarks done on Pentium dual

Originally committed as revision 16182 to svn://svn.ffmpeg.org/ffmpeg/trunk
pull/126/head
Michael Niedermayer 16 years ago
parent 6c414bb6d4
commit ad9ca7e720
1 changed files with 211 additions and 203 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 248
      libavcodec/h264.c

248
libavcodec/h264.c
Unescape View File

@ -6241,126 +6241,9 @@ static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
} }
} }
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
MpegEncContext * const s = &h->s;
const int mb_xy= mb_x + mb_y*s->mb_stride;
const int mb_type = s->current_picture.mb_type[mb_xy];
const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
int first_vertical_edge_done = 0;
int dir;
//for sufficiently low qp, filtering wouldn't do anything static void av_always_inline filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int dir) {
//this is a conservative estimate: could also check beta_offset and more accurate chroma_qp MpegEncContext * const s = &h->s;
if(!FRAME_MBAFF){
int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
int qp = s->current_picture.qscale_table[mb_xy];
if(qp <= qp_thresh
&& (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
&& (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){
return;
}
}
// CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
if(!h->pps.cabac && h->pps.transform_8x8_mode){
int top_type, left_type[2];
top_type = s->current_picture.mb_type[h->top_mb_xy] ;
left_type[0] = s->current_picture.mb_type[h->left_mb_xy[0]];
left_type[1] = s->current_picture.mb_type[h->left_mb_xy[1]];
if(IS_8x8DCT(top_type)){
h->non_zero_count_cache[4+8*0]=
h->non_zero_count_cache[5+8*0]= h->cbp_table[h->top_mb_xy] & 4;
h->non_zero_count_cache[6+8*0]=
h->non_zero_count_cache[7+8*0]= h->cbp_table[h->top_mb_xy] & 8;
}
if(IS_8x8DCT(left_type[0])){
h->non_zero_count_cache[3+8*1]=
h->non_zero_count_cache[3+8*2]= h->cbp_table[h->left_mb_xy[0]]&2; //FIXME check MBAFF
}
if(IS_8x8DCT(left_type[1])){
h->non_zero_count_cache[3+8*3]=
h->non_zero_count_cache[3+8*4]= h->cbp_table[h->left_mb_xy[1]]&8; //FIXME check MBAFF
}
if(IS_8x8DCT(mb_type)){
h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp_table[mb_xy] & 1;
h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp_table[mb_xy] & 2;
h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp_table[mb_xy] & 4;
h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp_table[mb_xy] & 8;
}
}
if (FRAME_MBAFF
// left mb is in picture
&& h->slice_table[mb_xy-1] != 0xFFFF
// and current and left pair do not have the same interlaced type
&& (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
// and left mb is in the same slice if deblocking_filter == 2
&& (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
/* First vertical edge is different in MBAFF frames
* There are 8 different bS to compute and 2 different Qp
*/
const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
int16_t bS[8];
int qp[2];
int bqp[2];
int rqp[2];
int mb_qp, mbn0_qp, mbn1_qp;
int i;
first_vertical_edge_done = 1;
if( IS_INTRA(mb_type) )
bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
else {
for( i = 0; i < 8; i++ ) {
int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
bS[i] = 4;
else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
(h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
:
h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2]))
bS[i] = 2;
else
bS[i] = 1;
}
}
mb_qp = s->current_picture.qscale_table[mb_xy];
mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
/* Filter edge */
tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);
{ int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, bqp );
filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, rqp );
}
/* dir : 0 -> vertical edge, 1 -> horizontal edge */
for( dir = 0; dir < 2; dir++ )
{
int edge; int edge;
const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
const int mbm_type = s->current_picture.mb_type[mbm_xy]; const int mbm_type = s->current_picture.mb_type[mbm_xy];
@ -6378,7 +6261,6 @@ static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8
if (first_vertical_edge_done) { if (first_vertical_edge_done) {
start = 1; start = 1;
first_vertical_edge_done = 0;
} }
if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy]) if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
@ -6561,7 +6443,133 @@ static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8
} }
} }
} }
}
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
MpegEncContext * const s = &h->s;
const int mb_xy= mb_x + mb_y*s->mb_stride;
const int mb_type = s->current_picture.mb_type[mb_xy];
const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
int first_vertical_edge_done = 0;
int dir;
//for sufficiently low qp, filtering wouldn't do anything
//this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
if(!FRAME_MBAFF){
int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
int qp = s->current_picture.qscale_table[mb_xy];
if(qp <= qp_thresh
&& (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
&& (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){
return;
}
}
// CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
if(!h->pps.cabac && h->pps.transform_8x8_mode){
int top_type, left_type[2];
top_type = s->current_picture.mb_type[h->top_mb_xy] ;
left_type[0] = s->current_picture.mb_type[h->left_mb_xy[0]];
left_type[1] = s->current_picture.mb_type[h->left_mb_xy[1]];
if(IS_8x8DCT(top_type)){
h->non_zero_count_cache[4+8*0]=
h->non_zero_count_cache[5+8*0]= h->cbp_table[h->top_mb_xy] & 4;
h->non_zero_count_cache[6+8*0]=
h->non_zero_count_cache[7+8*0]= h->cbp_table[h->top_mb_xy] & 8;
}
if(IS_8x8DCT(left_type[0])){
h->non_zero_count_cache[3+8*1]=
h->non_zero_count_cache[3+8*2]= h->cbp_table[h->left_mb_xy[0]]&2; //FIXME check MBAFF
}
if(IS_8x8DCT(left_type[1])){
h->non_zero_count_cache[3+8*3]=
h->non_zero_count_cache[3+8*4]= h->cbp_table[h->left_mb_xy[1]]&8; //FIXME check MBAFF
}
if(IS_8x8DCT(mb_type)){
h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp_table[mb_xy] & 1;
h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp_table[mb_xy] & 2;
h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp_table[mb_xy] & 4;
h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp_table[mb_xy] & 8;
} }
}
if (FRAME_MBAFF
// left mb is in picture
&& h->slice_table[mb_xy-1] != 0xFFFF
// and current and left pair do not have the same interlaced type
&& (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
// and left mb is in the same slice if deblocking_filter == 2
&& (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
/* First vertical edge is different in MBAFF frames
* There are 8 different bS to compute and 2 different Qp
*/
const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
int16_t bS[8];
int qp[2];
int bqp[2];
int rqp[2];
int mb_qp, mbn0_qp, mbn1_qp;
int i;
first_vertical_edge_done = 1;
if( IS_INTRA(mb_type) )
bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
else {
for( i = 0; i < 8; i++ ) {
int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
bS[i] = 4;
else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
(h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
:
h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2]))
bS[i] = 2;
else
bS[i] = 1;
}
}
mb_qp = s->current_picture.qscale_table[mb_xy];
mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
/* Filter edge */
tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);
{ int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, bqp );
filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, rqp );
}
#ifdef CONFIG_SMALL
for( dir = 0; dir < 2; dir++ )
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, dir);
#else
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, 0);
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
#endif
} }
static int decode_slice(struct AVCodecContext *avctx, void *arg){ static int decode_slice(struct AVCodecContext *avctx, void *arg){

Write Preview
Loading…
Cancel
Save