[FFmpeg-cvslog] H.264: add filter_mb_fast support for >8-bit decoding
Jason Garrett-Glaser
git at videolan.org
Tue Jul 12 01:44:23 CEST 2011
ffmpeg | branch: master | Jason Garrett-Glaser <jason at x264.com> | Mon Jul 11 14:46:37 2011 -0700| [b5bbc84fe2ae07d98764361d93d40ee2781467ab] | committer: Jason Garrett-Glaser
H.264: add filter_mb_fast support for >8-bit decoding
Much faster high bit depth deblocking.
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=b5bbc84fe2ae07d98764361d93d40ee2781467ab
---
libavcodec/h264_loopfilter.c | 104 +++++++++++++++++++++++-------------------
libavcodec/x86/h264dsp_mmx.c | 3 +-
2 files changed, 59 insertions(+), 48 deletions(-)
diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c
index f94e13a..377968f 100644
--- a/libavcodec/h264_loopfilter.c
+++ b/libavcodec/h264_loopfilter.c
@@ -207,56 +207,48 @@ static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, const i
}
}
-void ff_h264_filter_mb_fast( 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) {
+static void av_always_inline h264_filter_mb_fast_internal( 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 pixel_shift) {
MpegEncContext * const s = &h->s;
- int mb_xy;
- int mb_type, left_type, top_type;
- int qp, qp0, qp1, qpc, qpc0, qpc1;
int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
int chroma444 = CHROMA444;
- int qp_bd_offset, a, b;
- mb_xy = h->mb_xy;
+ int mb_xy = h->mb_xy;
+ int left_type= h->left_type[LTOP];
+ int top_type= h->top_type;
- if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
- ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
- return;
- }
- assert(!FRAME_MBAFF);
- left_type= h->left_type[LTOP];
- top_type= h->top_type;
-
- mb_type = s->current_picture.f.mb_type[mb_xy];
- qp = s->current_picture.f.qscale_table[mb_xy];
- qp0 = s->current_picture.f.qscale_table[mb_xy - 1];
- qp1 = s->current_picture.f.qscale_table[h->top_mb_xy];
- qpc = get_chroma_qp( h, 0, qp );
- qpc0 = get_chroma_qp( h, 0, qp0 );
- qpc1 = get_chroma_qp( h, 0, qp1 );
+ int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ int a = h->slice_alpha_c0_offset - qp_bd_offset;
+ int b = h->slice_beta_offset - qp_bd_offset;
+
+ int mb_type = s->current_picture.f.mb_type[mb_xy];
+ int qp = s->current_picture.f.qscale_table[mb_xy];
+ int qp0 = s->current_picture.f.qscale_table[mb_xy - 1];
+ int qp1 = s->current_picture.f.qscale_table[h->top_mb_xy];
+ int qpc = get_chroma_qp( h, 0, qp );
+ int qpc0 = get_chroma_qp( h, 0, qp0 );
+ int qpc1 = get_chroma_qp( h, 0, qp1 );
qp0 = (qp + qp0 + 1) >> 1;
qp1 = (qp + qp1 + 1) >> 1;
qpc0 = (qpc + qpc0 + 1) >> 1;
qpc1 = (qpc + qpc1 + 1) >> 1;
- qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
- a = h->slice_alpha_c0_offset - qp_bd_offset;
- b = h->slice_beta_offset - qp_bd_offset;
if( IS_INTRA(mb_type) ) {
static const int16_t bS4[4] = {4,4,4,4};
static const int16_t bS3[4] = {3,3,3,3};
const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
if(left_type)
- filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, a, b, h, 1);
+ filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);
if( IS_8x8DCT(mb_type) ) {
- filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, a, b, h, 0);
+ filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
if(top_type){
filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
}
filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
} else {
- filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, a, b, h, 0);
- filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, a, b, h, 0);
- filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, a, b, h, 0);
+ filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
+ filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
+ filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
if(top_type){
filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
}
@@ -267,12 +259,12 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
if(chroma){
if(chroma444){
if(left_type){
- filter_mb_edgev( &img_cb[4*0], linesize, bS4, qpc0, a, b, h, 1);
- filter_mb_edgev( &img_cr[4*0], linesize, bS4, qpc0, a, b, h, 1);
+ filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
+ filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
}
if( IS_8x8DCT(mb_type) ) {
- filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
if(top_type){
filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
@@ -280,12 +272,12 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
} else {
- filter_mb_edgev( &img_cb[4*1], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cr[4*1], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cb[4*3], linesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgev( &img_cr[4*3], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
if(top_type){
filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
@@ -299,11 +291,11 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
}
}else{
if(left_type){
- filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, a, b, h, 1);
- filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, a, b, h, 1);
+ filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
+ filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
}
- filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, a, b, h, 0);
- filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
+ filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
if(top_type){
filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
@@ -337,14 +329,14 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
#define FILTER(hv,dir,edge,intra)\
if(AV_RN64A(bS[dir][edge])) { \
- filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, a, b, h, intra );\
+ filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qp : qp##dir, a, b, h, intra );\
if(chroma){\
if(chroma444){\
- filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
- filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
+ filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
+ filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
} else if(!(edge&1)) {\
- filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
- filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
+ filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
+ filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
}\
}\
}
@@ -372,6 +364,24 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
}
}
+void ff_h264_filter_mb_fast( 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) {
+ assert(!FRAME_MBAFF);
+ if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
+ ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
+ return;
+ }
+
+#if CONFIG_SMALL
+ h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
+#else
+ if(h->pixel_shift){
+ h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
+ }else{
+ h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
+ }
+#endif
+}
+
static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
int v;
diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
index 1042552..35ec267 100644
--- a/libavcodec/x86/h264dsp_mmx.c
+++ b/libavcodec/x86/h264dsp_mmx.c
@@ -354,10 +354,11 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
{
int mm_flags = av_get_cpu_flags();
- if (bit_depth == 8) {
if (mm_flags & AV_CPU_FLAG_MMX2) {
c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
}
+
+ if (bit_depth == 8) {
#if HAVE_YASM
if (mm_flags & AV_CPU_FLAG_MMX) {
c->h264_idct_dc_add =
More information about the ffmpeg-cvslog
mailing list