Skip to content

F
ffmpeg-mt
Commit 544ab882 authored by mru's avatar mru
Browse files
Options

H264: use alias-safe macros 

This eliminates all aliasing violation warnings in h264 code.
No measurable speed difference with gcc-4.4.3 on i7.

git-svn-id: file:///var/local/repositories/ffmpeg/trunk@21881 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b
parent 977aca53
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 86 additions and 82 deletions
libavcodec/h264.c
View file @ 544ab882
...@@ -156,11 +156,11 @@ const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_l ...@@ -156,11 +156,11 @@ const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_l
# if HAVE_FAST_64BIT # if HAVE_FAST_64BIT
# define RS 7 # define RS 7
for(i=0; i+1<length; i+=9){ for(i=0; i+1<length; i+=9){
if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL)) if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
# else # else
# define RS 3 # define RS 3
for(i=0; i+1<length; i+=5){ for(i=0; i+1<length; i+=5){
if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U)) if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
# endif # endif
continue; continue;
if(i>0 && !src[i]) i--; if(i>0 && !src[i]) i--;
......
libavcodec/h264.h
View file @ 544ab882
This diff is collapsed.
libavcodec/h264_direct.c
View file @ 544ab882
...@@ -183,11 +183,11 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){ ...@@ -183,11 +183,11 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
}else { }else {
assert(match_count==1); assert(match_count==1);
if(left_ref==ref[list]){ if(left_ref==ref[list]){
mv[list]= *(uint32_t*)A; mv[list]= AV_RN32A(A);
}else if(top_ref==ref[list]){ }else if(top_ref==ref[list]){
mv[list]= *(uint32_t*)B; mv[list]= AV_RN32A(B);
}else{ }else{
mv[list]= *(uint32_t*)C; mv[list]= AV_RN32A(C);
} }
} }
}else{ }else{
...@@ -362,9 +362,9 @@ single_col: ...@@ -362,9 +362,9 @@ single_col:
const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
if(ref[0] == 0) if(ref[0] == 0)
*(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0; AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
if(ref[1] == 0) if(ref[1] == 0)
*(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0; AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
m++; m++;
} }
} }
...@@ -571,8 +571,8 @@ single_col: ...@@ -571,8 +571,8 @@ single_col:
int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
mv_l0[0] = (scale * mv_col[0] + 128) >> 8; mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8; mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
*(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
} }
} }
} }
......
libavcodec/h264_loopfilter.c
View file @ 544ab882
...@@ -25,6 +25,7 @@ ...@@ -25,6 +25,7 @@
* @author Michael Niedermayer <michaelni@gmx.at> * @author Michael Niedermayer <michaelni@gmx.at>
*/ */
#include "libavutil/intreadwrite.h"
#include "internal.h" #include "internal.h"
#include "dsputil.h" #include "dsputil.h"
#include "avcodec.h" #include "avcodec.h"
...@@ -368,11 +369,13 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, ...@@ -368,11 +369,13 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
return; return;
} else { } else {
LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]); LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
int edges; int edges;
if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
edges = 4; edges = 4;
bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL; AV_WN64A(bS[0][0], 0x0002000200020002ULL);
AV_WN64A(bS[0][2], 0x0002000200020002ULL);
AV_WN64A(bS[1][0], 0x0002000200020002ULL);
AV_WN64A(bS[1][2], 0x0002000200020002ULL);
} else { } else {
int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0; int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0;
int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0; int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;
...@@ -382,12 +385,12 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, ...@@ -382,12 +385,12 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
} }
if( IS_INTRA(left_type) ) if( IS_INTRA(left_type) )
bSv[0][0] = 0x0004000400040004ULL; AV_WN64A(bS[0][0], 0x0004000400040004ULL);
if( IS_INTRA(h->top_type) ) if( IS_INTRA(h->top_type) )
bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL; AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
#define FILTER(hv,dir,edge)\ #define FILTER(hv,dir,edge)\
if(bSv[dir][edge]) {\ 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, h );\ filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
if(!(edge&1)) {\ if(!(edge&1)) {\
filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
...@@ -477,7 +480,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u ...@@ -477,7 +480,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
DECLARE_ALIGNED_8(int16_t, bS)[4]; DECLARE_ALIGNED_8(int16_t, bS)[4];
int qp; int qp;
if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) { if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
*(uint64_t*)bS= 0x0003000300030003ULL; AV_WN64A(bS, 0x0003000300030003ULL);
} else { } else {
if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){ if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]); bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
...@@ -508,17 +511,17 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u ...@@ -508,17 +511,17 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
int qp; int qp;
if( IS_INTRA(mb_type|mbm_type)) { if( IS_INTRA(mb_type|mbm_type)) {
*(uint64_t*)bS= 0x0003000300030003ULL; AV_WN64A(bS, 0x0003000300030003ULL);
if ( (!IS_INTERLACED(mb_type|mbm_type)) if ( (!IS_INTERLACED(mb_type|mbm_type))
|| ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
) )
*(uint64_t*)bS= 0x0004000400040004ULL; AV_WN64A(bS, 0x0004000400040004ULL);
} else { } else {
int i; int i;
int mv_done; int mv_done;
if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) { if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
*(uint64_t*)bS= 0x0001000100010001ULL; AV_WN64A(bS, 0x0001000100010001ULL);
mv_done = 1; mv_done = 1;
} }
else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
...@@ -588,13 +591,13 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u ...@@ -588,13 +591,13 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
continue; continue;
if( IS_INTRA(mb_type)) { if( IS_INTRA(mb_type)) {
*(uint64_t*)bS= 0x0003000300030003ULL; AV_WN64A(bS, 0x0003000300030003ULL);
} else { } else {
int i; int i;
int mv_done; int mv_done;
if( edge & mask_edge ) { if( edge & mask_edge ) {
*(uint64_t*)bS= 0; AV_ZERO64(bS);
mv_done = 1; mv_done = 1;
} }
else if( mask_par0 ) { else if( mask_par0 ) {
...@@ -674,10 +677,10 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint ...@@ -674,10 +677,10 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint
int i; int i;
first_vertical_edge_done = 1; first_vertical_edge_done = 1;
if( IS_INTRA(mb_type) ) if( IS_INTRA(mb_type) ) {
*(uint64_t*)&bS[0]= AV_WN64A(&bS[0], 0x0004000400040004ULL);
*(uint64_t*)&bS[4]= 0x0004000400040004ULL; AV_WN64A(&bS[4], 0x0004000400040004ULL);
else { } else {
static const uint8_t offset[2][2][8]={ static const uint8_t offset[2][2][8]={
{ {
{7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1}, {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
......
libavcodec/h264_mvpred.h
View file @ 544ab882
...@@ -58,7 +58,7 @@ static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, in ...@@ -58,7 +58,7 @@ static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, in
&& h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
const uint32_t *mb_types = s->current_picture_ptr->mb_type; const uint32_t *mb_types = s->current_picture_ptr->mb_type;
const int16_t *mv; const int16_t *mv;
*(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0; AV_ZERO32(h->mv_cache[list][scan8[0]-2]);
*C = h->mv_cache[list][scan8[0]-2]; *C = h->mv_cache[list][scan8[0]-2];
if(!MB_FIELD if(!MB_FIELD
...@@ -220,8 +220,8 @@ static inline void pred_pskip_motion(H264Context * const h, int * const mx, int ...@@ -220,8 +220,8 @@ static inline void pred_pskip_motion(H264Context * const h, int * const mx, int
tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y); tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
|| !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ]) || !( top_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 8 ]))
|| !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){ || !(left_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 1 ]))){
*mx = *my = 0; *mx = *my = 0;
return; return;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment