ref: 3db194267e46239bd5ceaa526b4dd433ee14bbb9
parent: ded8ed3f49c022359107fa7f4752b84389716b00
author: Victorien Le Couviour--Tuffet <victorien.lecouviour.tuffet@gmail.com>
date: Wed Feb 13 12:36:19 EST 2019
x86: improve AVX2 cdef_filter macro consistency - consistently use %3 instead of hardcoded value for tmp stride - also correct a comment
--- a/src/x86/cdef.asm
+++ b/src/x86/cdef.asm
@@ -104,8 +104,8 @@
pminsw m12, m10
psignw m11, m5 ; constrain(diff_p0)
psignw m12, m6 ; constrain(diff_p1)
- pmullw m11, %4 ; constrain(diff_p0) * pri_taps
- pmullw m12, %4 ; constrain(diff_p1) * pri_taps
+ pmullw m11, %4 ; constrain(diff_p0) * taps
+ pmullw m12, %4 ; constrain(diff_p1) * taps
paddw m15, m11
paddw m15, m12
%endmacro
@@ -258,15 +258,15 @@
%if %2 == 8
pmovzxbw xm2, [leftq+ 8]
%endif
- movd [px+0*32-4], xm1
- pextrd [px+1*32-4], xm1, 1
- pextrd [px+2*32-4], xm1, 2
- pextrd [px+3*32-4], xm1, 3
+ movd [px+0*%3-4], xm1
+ pextrd [px+1*%3-4], xm1, 1
+ pextrd [px+2*%3-4], xm1, 2
+ pextrd [px+3*%3-4], xm1, 3
%if %2 == 8
- movd [px+4*32-4], xm2
- pextrd [px+5*32-4], xm2, 1
- pextrd [px+6*32-4], xm2, 2
- pextrd [px+7*32-4], xm2, 3
+ movd [px+4*%3-4], xm2
+ pextrd [px+5*%3-4], xm2, 1
+ pextrd [px+6*%3-4], xm2, 2
+ pextrd [px+7*%3-4], xm2, 3
%endif
jmp .left_done
.no_left: