Revision b4c2ada5 libavcodec/x86/h264dsp_mmx.c
| libavcodec/x86/h264dsp_mmx.c | ||
|---|---|---|
| 812 | 812 |
// could do a special case for dir==0 && edges==1, but it only reduces the |
| 813 | 813 |
// average filter time by 1.2% |
| 814 | 814 |
for( dir=1; dir>=0; dir-- ) {
|
| 815 |
const int d_idx = dir ? -8 : -1;
|
|
| 815 |
const x86_reg d_idx = dir ? -8 : -1;
|
|
| 816 | 816 |
const int mask_mv = dir ? mask_mv1 : mask_mv0; |
| 817 | 817 |
DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL; |
| 818 | 818 |
int b_idx, edge, l; |
| ... | ... | |
| 825 | 825 |
__asm__ volatile("pxor %%mm0, %%mm0 \n\t":);
|
| 826 | 826 |
for( l = bidir; l >= 0; l-- ) {
|
| 827 | 827 |
__asm__ volatile( |
| 828 |
"movd %0, %%mm1 \n\t"
|
|
| 829 |
"punpckldq %1, %%mm1 \n\t"
|
|
| 828 |
"movd (%0), %%mm1 \n\t"
|
|
| 829 |
"punpckldq (%0,%1), %%mm1 \n\t"
|
|
| 830 | 830 |
"punpckldq %%mm1, %%mm2 \n\t" |
| 831 | 831 |
"pcmpeqb %%mm2, %%mm1 \n\t" |
| 832 | 832 |
"paddb %%mm6, %%mm1 \n\t" |
| 833 | 833 |
"punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn] |
| 834 | 834 |
"por %%mm1, %%mm0 \n\t" |
| 835 | 835 |
|
| 836 |
"movq %2, %%mm1 \n\t"
|
|
| 837 |
"movq 8+1*%2, %%mm2 \n\t"
|
|
| 838 |
"psubw %3, %%mm1 \n\t"
|
|
| 839 |
"psubw 8+1*%3, %%mm2 \n\t"
|
|
| 836 |
"movq (%2), %%mm1 \n\t"
|
|
| 837 |
"movq 8(%2), %%mm2 \n\t"
|
|
| 838 |
"psubw (%2,%1,4), %%mm1 \n\t"
|
|
| 839 |
"psubw 8(%2,%1,4), %%mm2 \n\t"
|
|
| 840 | 840 |
"packsswb %%mm2, %%mm1 \n\t" |
| 841 | 841 |
"paddb %%mm5, %%mm1 \n\t" |
| 842 | 842 |
"pminub %%mm4, %%mm1 \n\t" |
| 843 | 843 |
"pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit |
| 844 | 844 |
"por %%mm1, %%mm0 \n\t" |
| 845 |
::"m"(ref[l][b_idx]), |
|
| 846 |
"m"(ref[l][b_idx+d_idx]), |
|
| 847 |
"m"(mv[l][b_idx][0]), |
|
| 848 |
"m"(mv[l][b_idx+d_idx][0]) |
|
| 845 |
::"r"(ref[l]+b_idx), |
|
| 846 |
"r"(d_idx), |
|
| 847 |
"r"(mv[l]+b_idx) |
|
| 849 | 848 |
); |
| 850 | 849 |
} |
| 851 | 850 |
if(bidir==1){
|
| 852 | 851 |
__asm__ volatile("pxor %%mm3, %%mm3 \n\t":);
|
| 853 | 852 |
for( l = bidir; l >= 0; l-- ) {
|
| 854 | 853 |
__asm__ volatile( |
| 855 |
"movd %0, %%mm1 \n\t"
|
|
| 856 |
"punpckldq %1, %%mm1 \n\t"
|
|
| 854 |
"movd (%0), %%mm1 \n\t"
|
|
| 855 |
"punpckldq (%1), %%mm1 \n\t"
|
|
| 857 | 856 |
"punpckldq %%mm1, %%mm2 \n\t" |
| 858 | 857 |
"pcmpeqb %%mm2, %%mm1 \n\t" |
| 859 | 858 |
"paddb %%mm6, %%mm1 \n\t" |
| 860 | 859 |
"punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn] |
| 861 | 860 |
"por %%mm1, %%mm3 \n\t" |
| 862 | 861 |
|
| 863 |
"movq %2, %%mm1 \n\t"
|
|
| 864 |
"movq 8+1*%2, %%mm2 \n\t"
|
|
| 865 |
"psubw %3, %%mm1 \n\t"
|
|
| 866 |
"psubw 8+1*%3, %%mm2 \n\t"
|
|
| 862 |
"movq (%2), %%mm1 \n\t"
|
|
| 863 |
"movq 8(%2), %%mm2 \n\t"
|
|
| 864 |
"psubw (%3), %%mm1 \n\t"
|
|
| 865 |
"psubw 8(%3), %%mm2 \n\t"
|
|
| 867 | 866 |
"packsswb %%mm2, %%mm1 \n\t" |
| 868 | 867 |
"paddb %%mm5, %%mm1 \n\t" |
| 869 | 868 |
"pminub %%mm4, %%mm1 \n\t" |
| 870 | 869 |
"pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit |
| 871 | 870 |
"por %%mm1, %%mm3 \n\t" |
| 872 |
::"m"(ref[l][b_idx]),
|
|
| 873 |
"m"(ref[1-l][b_idx+d_idx]),
|
|
| 874 |
"m"(mv[l][b_idx][0]),
|
|
| 875 |
"m"(mv[1-l][b_idx+d_idx][0])
|
|
| 871 |
::"r"(ref[l]+b_idx),
|
|
| 872 |
"r"(ref[1-l]+b_idx+d_idx),
|
|
| 873 |
"r"(mv[l][b_idx]),
|
|
| 874 |
"r"(mv[1-l][b_idx+d_idx])
|
|
| 876 | 875 |
); |
| 877 | 876 |
} |
| 878 | 877 |
__asm__ volatile( |
Also available in: Unified diff