virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

x86: float dsp: unroll SSE versions

Browse Source 
vector_fmul and vector_fmac_scalar are guaranteed that they can process in
batch of 16 elements, but their SSE versions only does 8 at a time.

Therefore, unroll them a bit.
299 to 261c for 256 elements in vector_fmac_scalar on Arrandale/Win64.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Christophe Gisquet 2014-02-14 16:00:51 +00:00 committed by Michael Niedermayer
parent 1db8bc5648
commit 133b34207c
1 changed files with 24 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
libavutil/x86/float_dsp.asm
View File

@ -29,17 +29,21 @@ SECTION .text
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL 0
cglobal vector_fmul, 4,4,2, dst, src0, src1, len
lea lenq, [lend*4 - 2*mmsize]
lea lenq, [lend*4 - 64]
ALIGN 16
.loop:
mova m0, [src0q + lenq]
mova m1, [src0q + lenq + mmsize]
mulps m0, m0, [src1q + lenq]
mulps m1, m1, [src1q + lenq + mmsize]
mova [dstq + lenq], m0
mova [dstq + lenq + mmsize], m1
%assign a 0
%rep 32/mmsize
mova m0, [src0q + lenq + (a+0)*mmsize]
mova m1, [src0q + lenq + (a+1)*mmsize]
mulps m0, m0, [src1q + lenq + (a+0)*mmsize]
mulps m1, m1, [src1q + lenq + (a+1)*mmsize]
mova [dstq + lenq + (a+0)*mmsize], m0
mova [dstq + lenq + (a+1)*mmsize], m1
%assign a a+2
%endrep
sub lenq, 2*mmsize
sub lenq, 64
jge .loop
REP_RET
%endmacro
@ -72,15 +76,19 @@ cglobal vector_fmac_scalar, 4,4,3, dst, src, mul, len
vinsertf128 m0, m0, xmm0, 1
%endif
%endif
lea lenq, [lend*4-2*mmsize]
lea lenq, [lend*4-64]
.loop:
mulps m1, m0, [srcq+lenq ]
mulps m2, m0, [srcq+lenq+mmsize]
addps m1, m1, [dstq+lenq ]
addps m2, m2, [dstq+lenq+mmsize]
mova [dstq+lenq ], m1
mova [dstq+lenq+mmsize], m2
sub lenq, 2*mmsize
%assign a 0
%rep 32/mmsize
mulps m1, m0, [srcq+lenq+(a+0)*mmsize]
mulps m2, m0, [srcq+lenq+(a+1)*mmsize]
addps m1, m1, [dstq+lenq+(a+0)*mmsize]
addps m2, m2, [dstq+lenq+(a+1)*mmsize]
mova [dstq+lenq+(a+0)*mmsize], m1
mova [dstq+lenq+(a+1)*mmsize], m2
%assign a a+2
%endrep
sub lenq, 64
jge .loop
REP_RET
%endmacro