diff options
author | Christophe Gisquet <christophe.gisquet@gmail.com> | 2014-04-04 14:21:40 -0300 |
---|---|---|
committer | Michael Niedermayer <michaelni@gmx.at> | 2014-04-04 22:35:45 +0200 |
commit | dfd865e51b890d9be394804bccddf55198f4a251 (patch) | |
tree | 9d634104e21241bd01a7ab1b5f4b69a1a4ae0a33 /libavcodec | |
parent | 3aac5fcfa9d3748659d78ab2a66d0ccce22cfd4f (diff) | |
download | ffmpeg-dfd865e51b890d9be394804bccddf55198f4a251.tar.gz |
x86/synth_filter: remove the main loop when it's not needed
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec')
-rw-r--r-- | libavcodec/x86/dcadsp.asm | 11 |
1 files changed, 11 insertions, 0 deletions
diff --git a/libavcodec/x86/dcadsp.asm b/libavcodec/x86/dcadsp.asm index c5a1e77bc2..10cb2983aa 100644 --- a/libavcodec/x86/dcadsp.asm +++ b/libavcodec/x86/dcadsp.asm @@ -311,9 +311,14 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \ sub r5q, offmp and r5q, -64 shl r5q, 2 +%if ARCH_X86_32 || mmsize < 32 mov OFFQ, r5q %define i r5q mov i, 16 * 4 - (ARCH_X86_64 + 1) * mmsize ; main loop counter +%else +%define i 0 +%define OFFQ r5q +%endif %define buf2 synth_buf2q %if ARCH_X86_32 @@ -332,8 +337,10 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \ %define j r3q mov win, windowm mov ptr1, synth_bufm +%if ARCH_X86_32 || mmsize < 32 add win, i add ptr1, i +%endif %else ; ARCH_X86_64 %define ptr1 r6q %define ptr2 r7q ; must be loaded @@ -349,7 +356,9 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \ mov ptr2, synth_bufmp ; prepare the inner loop counter mov j, OFFQ +%if ARCH_X86_32 || mmsize < 32 sub ptr2, i +%endif .loop1: INNER_LOOP 0 jge .loop1 @@ -394,8 +403,10 @@ cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \ mova [outq + i + 0 * 4 + mmsize], m7 mova [outq + i + 16 * 4 + mmsize], m8 %endif +%if ARCH_X86_32 || mmsize < 32 sub i, (ARCH_X86_64 + 1) * mmsize jge .mainloop +%endif RET %endmacro |