[FFmpeg-trac] #8696(avcodec:new): enhancement: more neon accerleration for vc-1
FFmpeg
trac at avcodec.org
Sat May 30 21:00:18 EEST 2020
#8696: enhancement: more neon accerleration for vc-1
---------------------------------+---------------------------------------
Reporter: spotter | Type: enhancement
Status: new | Priority: normal
Component: avcodec | Version: git-master
Keywords: vc-1 | Blocked By:
Blocking: | Reproduced by developer: 0
Analyzed by developer: 0 |
---------------------------------+---------------------------------------
Summary of the bug:
comparing x86 optimizations to arm, it seems there are a number of
optimizations that exist in the x86 space that don't exist for arm.
I've noticed this on my Sony Android TV which seems to be defective in
that even though it has support for VC-1, it's not exporting that support
via mediacodecs.
see: https://github.com/CiNcH83/bravia_atv2/issues/4 (and even their built
in player is buggy with it see
https://github.com/CiNcH83/bravia_atv2/issues/76)
so decided to explore what could be done to optimize vc-1 software
rendering.
I cam across the samsung paper that shows major improvements:
http://www.ncc.org.in/download.php?f=NCC2010/B2_3.pdf (in digging into the
asm hooks that ffmpeg's video decoder has, I'm guessing this paper isn't
new to ffmpeg's developers)
and then dug into ffmpeg where I saw that it seems that there a lot more
x86 hooks that do't exist for arm (example loopfilter). I'm wondering if
this is a conscious decision? (i.e. because of limitations of neon it
can't be done effectively) or an oversight and in truth arm should have an
equivalent set of vector instructions like x86?
--
Ticket URL: <https://trac.ffmpeg.org/ticket/8696>
FFmpeg <https://ffmpeg.org>
FFmpeg issue tracker
More information about the FFmpeg-trac
mailing list