On 01/12/14 10:03, [email protected] wrote:
Hi,

Currently, seems there are still no much assembly optimization(such
as arm-neon) for the latest codecs, include HEVC and VP9 I just want
to know if libav org have plan to implement more optimization for
them. As far as I know, the decoders(HEVC/VP9) released in the latest
android(lollipop) packages have better performance than libav on ARM
platform.

Hi,

We are still interested in writing SIMD optimizations for our codecs for sure, Libav is an volunteer based organization so interest and time from time to time might not match. In this case there are other good reasons not to optimize early though.

## VP9

I'm waiting for Google to provide a specification to make sure any optimization doesn't clash with the soon-to-be-released specification and/or newer version of libvpx. Last time I discussed about having access to a pre-release was during the Videolan Dev Days in September, I guess it is time to ask again and see if there is a change in the situation, recently they started to provide access to drafting documents for the new features so there might be something regarding the features that should be set in stone brewing.

Optimizing for a moving target is not a really nice way to spend your spare time.

## HEVC

The current C codebase is working but is not feature complete nor is as polished as many would.

During our past Sprints[1] some time had been devoted on cleaning it up and it lead on some unexpected speedups. Once that step is done and we integrate the support for a larger part of the standard I'm sure that some will start working on SIMD optimizations. Probably I'll try to dig some time to write some Altivec for it since I'm itching to play a bit more since the ppcle64 announcement (the fact I broke my ppc64 setup by foolishly updating udev prevented me to do much though).

Is Freescale interested in providing access to hardware to test simd optimization? That (and sponsoring) is always appreciated.

lu

[1] https://wiki.libav.org/Sprint
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to