RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

sbr_hf_gen_sse: Optimize code a bit more. 

Core I7 (Sandy Bridge) 135 to 107 cycles
Core i5 (Arrandale) 162 to 142 (Thanks to Christophe Gisquet for testing)

Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer 2012-12-07 23:59:20 +01:00
parent 7f154bd54f
commit 0110108a7c
1 changed files with 18 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
libavcodec/x86/sbrdsp.asm
View File

@ -134,7 +134,6 @@ cglobal sbr_hf_gen, 4,4,8, X_high, X_low, alpha0, alpha1, BW, S, E
mulps m2, bw ; (a1[0] a1[1])*bw*bw = (a0 a1) mulps m2, bw ; (a1[0] a1[1])*bw*bw = (a0 a1)
mova m3, m1 mova m3, m1
mova m4, m2 mova m4, m2
mova m7, [ps_mask]
; Set pointers ; Set pointers
%if ARCH_X86_64 == 0 || WIN64 %if ARCH_X86_64 == 0 || WIN64
@ -154,30 +153,28 @@ cglobal sbr_hf_gen, 4,4,8, X_high, X_low, alpha0, alpha1, BW, S, E
shl start, 3 ; offset from num loops shl start, 3 ; offset from num loops
mova m0, [X_lowq + start] mova m0, [X_lowq + start]
movlhps m1, m1 ; (a2 a3 a2 a3) shufps m3, m3, q1111
movlhps m2, m2 ; (a0 a1 a0 a1) shufps m4, m4, q1111
shufps m3, m3, q0101 ; (a3 a2 a3 a2) xorps m3, [ps_mask]
shufps m4, m4, q0101 ; (a1 a0 a1 a0) shufps m1, m1, q0000
xorps m3, m7 ; (-a3 a2 -a3 a2) shufps m2, m2, q0000
xorps m4, m7 ; (-a1 a0 -a1 a0) xorps m4, [ps_mask]
.loop2: .loop2:
mova m5, m0 movu m7, [X_lowq + start + 8] ; BbCc
mova m6, m0 mova m6, m0
shufps m0, m0, q2200 ; {Xl[-2][0],",Xl[-1][0],"} mova m5, m7
shufps m5, m5, q3311 ; {Xl[-2][1],",Xl[-1][1],"} shufps m0, m0, q2301 ; aAbB
mulps m0, m2 shufps m7, m7, q2301 ; bBcC
mulps m5, m4 mulps m0, m4
mova m7, m6
addps m5, m0
mova m0, [X_lowq + start + 2*2*4]
shufps m6, m0, q0022 ; {Xl[-1][0],",Xl[0][0],"}
shufps m7, m0, q1133 ; {Xl[-1][1],",Xl[1][1],"}
mulps m6, m1
mulps m7, m3 mulps m7, m3
addps m5, m6 mulps m6, m2
mulps m5, m1
addps m7, m0 addps m7, m0
addps m5, m7 mova m0, [X_lowq + start +16] ; CcDd
mova [X_highq + start], m5 addps m7, m0
addps m6, m5
addps m7, m6
mova [X_highq + start], m7
add start, 16 add start, 16
jnz .loop2 jnz .loop2
RET RET