Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
110797 Commits
37 Branches
400 Tags
465 MiB
Tag: Branch: Tree: 6ee57fd2b6
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '6ee57fd2b6'
${ noResults }
FFmpeg/libavcodec/x86/takdsp.asm

117 lines
4.0 KiB
Raw Normal View History Unescape

avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
;******************************************************************************
;* TAK DSP SIMD optimizations
;*
;* Copyright (C) 2015 Paul B Mahol
;*
;* This file is part of FFmpeg.
;*
;* FFmpeg is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* FFmpeg is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with FFmpeg; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION_RODATA
pd_128: times 4 dd 128
SECTION .text
INIT_XMM sse2
cglobal tak_decorrelate_ls, 3, 3, 2, p1, p2, length
shl lengthd, 2
add p1q, lengthq
add p2q, lengthq
neg lengthq
.loop:
mova m0, [p1q+lengthq+mmsize*0]
mova m1, [p1q+lengthq+mmsize*1]
paddd m0, [p2q+lengthq+mmsize*0]
paddd m1, [p2q+lengthq+mmsize*1]
mova [p2q+lengthq+mmsize*0], m0
mova [p2q+lengthq+mmsize*1], m1
add lengthq, mmsize*2
jl .loop
x86: replace explicit REP_RETs with RETs From x86inc: > On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either > a branch or a branch target. So switch to a 2-byte form of ret in that case. > We can automatically detect "follows a branch", but not a branch target. > (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.) x86inc can automatically determine whether to use REP_RET rather than REP in most of these cases, so impact is minimal. Additionally, a few REP_RETs were used unnecessary, despite the return being nowhere near a branch. The only CPUs affected were AMD K10s, made between 2007 and 2011, 16 years ago and 12 years ago, respectively. In the future, everyone involved with x86inc should consider dropping REP_RETs altogether.
2 years ago
RET
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
cglobal tak_decorrelate_sr, 3, 3, 2, p1, p2, length
shl lengthd, 2
add p1q, lengthq
add p2q, lengthq
neg lengthq
.loop:
mova m0, [p2q+lengthq+mmsize*0]
mova m1, [p2q+lengthq+mmsize*1]
psubd m0, [p1q+lengthq+mmsize*0]
psubd m1, [p1q+lengthq+mmsize*1]
mova [p1q+lengthq+mmsize*0], m0
mova [p1q+lengthq+mmsize*1], m1
add lengthq, mmsize*2
jl .loop
x86: replace explicit REP_RETs with RETs From x86inc: > On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either > a branch or a branch target. So switch to a 2-byte form of ret in that case. > We can automatically detect "follows a branch", but not a branch target. > (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.) x86inc can automatically determine whether to use REP_RET rather than REP in most of these cases, so impact is minimal. Additionally, a few REP_RETs were used unnecessary, despite the return being nowhere near a branch. The only CPUs affected were AMD K10s, made between 2007 and 2011, 16 years ago and 12 years ago, respectively. In the future, everyone involved with x86inc should consider dropping REP_RETs altogether.
2 years ago
RET
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
cglobal tak_decorrelate_sm, 3, 3, 6, p1, p2, length
shl lengthd, 2
add p1q, lengthq
add p2q, lengthq
neg lengthq
.loop:
mova m0, [p1q+lengthq]
mova m1, [p2q+lengthq]
mova m3, [p1q+lengthq+mmsize]
mova m4, [p2q+lengthq+mmsize]
mova m2, m1
mova m5, m4
x86/takdsp: use arithmetic shift instructions p1 and p2 are int32_t. Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
9 years ago
psrad m2, 1
psrad m5, 1
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
psubd m0, m2
psubd m3, m5
paddd m1, m0
paddd m4, m3
mova [p1q+lengthq], m0
mova [p2q+lengthq], m1
mova [p1q+lengthq+mmsize], m3
mova [p2q+lengthq+mmsize], m4
add lengthq, mmsize*2
jl .loop
x86: replace explicit REP_RETs with RETs From x86inc: > On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either > a branch or a branch target. So switch to a 2-byte form of ret in that case. > We can automatically detect "follows a branch", but not a branch target. > (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.) x86inc can automatically determine whether to use REP_RET rather than REP in most of these cases, so impact is minimal. Additionally, a few REP_RETs were used unnecessary, despite the return being nowhere near a branch. The only CPUs affected were AMD K10s, made between 2007 and 2011, 16 years ago and 12 years ago, respectively. In the future, everyone involved with x86inc should consider dropping REP_RETs altogether.
2 years ago
RET
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
INIT_XMM sse4
cglobal tak_decorrelate_sf, 3, 3, 5, p1, p2, length, dshift, dfactor
shl lengthd, 2
add p1q, lengthq
add p2q, lengthq
neg lengthq
movd m2, dshiftm
movd m3, dfactorm
pshufd m3, m3, 0
mova m4, [pd_128]
.loop:
mova m0, [p1q+lengthq]
mova m1, [p2q+lengthq]
x86/takdsp: use arithmetic shift instructions p1 and p2 are int32_t. Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
9 years ago
psrad m1, m2
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
pmulld m1, m3
paddd m1, m4
x86/takdsp: use arithmetic shift instructions p1 and p2 are int32_t. Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
9 years ago
psrad m1, 8
avcodec/takdec: add x86 SIMD for rest of decorrelation modes Signed-off-by: Paul B Mahol <onemda@gmail.com>
9 years ago
pslld m1, m2
psubd m1, m0
mova [p1q+lengthq], m1
add lengthq, mmsize
jl .loop
x86: replace explicit REP_RETs with RETs From x86inc: > On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either > a branch or a branch target. So switch to a 2-byte form of ret in that case. > We can automatically detect "follows a branch", but not a branch target. > (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.) x86inc can automatically determine whether to use REP_RET rather than REP in most of these cases, so impact is minimal. Additionally, a few REP_RETs were used unnecessary, despite the return being nowhere near a branch. The only CPUs affected were AMD K10s, made between 2007 and 2011, 16 years ago and 12 years ago, respectively. In the future, everyone involved with x86inc should consider dropping REP_RETs altogether.
2 years ago
RET