Web lists-archives.com

Re: [MPlayer-dev-eng] [PATCH 1/3] osd_template.c: Add intrinsic based SSE2 support.




On Tue, May 21, 2019 at 09:32:49PM +0200, Reimar Döffinger wrote:
> Intrinsics are not well liked for good reason,
> but it is much faster to implement and compilers
> are no longer completely horrible for x86.
> Speed is about 2x-3x and it matches the C code instead
> of the less accurate MMX code.
> Primary motivation is though that MMX is starting to
> have worse and worse support in CPUs.
> To benchmark, use this command and remove the reading code
> from rawvideo and the memcpy_pic from vf_expand.
> ./mplayer /dev/zero -demuxer rawvideo -rawvideo w=1920:h=1080:format=yuy2 -vo null -osdlevel 3 -benchmark -quiet -frames 50000 -vf expand=osd=1 -subfont-osd-scale 20

I plan on pushing this patch series soonish,
even if the practical benefits right now are likely
rather limited (as impressive as 2x speed might
sound on paper, I doubt these functions are actually
much used).
If anyone is aware of some low-hanging, relatively
easy to optimize code relevant to them I'd be
happy to have a look (also for other architectures).
_______________________________________________
MPlayer-dev-eng mailing list
MPlayer-dev-eng@xxxxxxxxxxxx
https://lists.mplayerhq.hu/mailman/listinfo/mplayer-dev-eng