[FFmpeg-devel] Loongson Optimization Patchs Submitting Question - new branch of mipsel64

Michael Niedermayer michaelni at gmx.at
Fri Mar 20 18:05:23 CET 2015


On Fri, Mar 20, 2015 at 04:06:56PM +0800, 周晓勇 wrote:
> Dear Michael,
> 
> 
>     Hi, ffmpeg and mplayer have achieved huge performance improvement on x86 under your leadership. And MIPS corporation have contributed the optimizations on their chips, which mainly aiming at audio codecs. As we know, ffmpeg and mplayer began to support loongson as a branch of mipsel from about version n0.6.x, but without assembly optimization. In 2009 and 2011, we taped out Loongson-3A and Loongson-3B. Loongson-3 series have the same core with Loongson-2E and Loongson-2F named GS464, which also added multimedia extension instructs but more powerful. 
> 
> 
>     From last year, we began to optimize the H.264 and MPEG1/2/4 codecs based on ffmpeg-0.10.x, and achieved about 70%~80% boost with H.264 decode, about 30% boost with MPEG4 decode and 130% boost with RV40 decode(the xls data attached in the accessory).  The rough test was done on Loongson-3A,

thats great


> and we are very pleased to present some machines to the community for testing and verifying. Just send me your address.

ok, ill send you my address seperately, unless someone else volunteers


>
> 
>     In this year, we will release Loongson Fedora19 based on the community, and I have finished parts of optimization for n2.1.7. By contrast, the decode boost(VC+VO+A) come up with 170% for H264-720P(one is calvc, the other one is cabac), 78% for H264-1080P(two cabac), 59% for RV40, and 70% for MPEG4. The optimization mainly include libavcodec, libavutil, and mplayer's fast_memcpy.
> 
> 
>     Now, for questions.
>     1.Our optimization based on n2.1.7, would you like me to send you the n2.1.7-patches or patches on master?

please submit patches against master


>     2.Same question for mplayer patches based on SVN-r37150.

mplayer patches too should be on the latest checkout not some
release


>     3.Could we use the suffix of "mmi" for the optimized function? For instance below, does ARM have registered "mmi" like Intel does for "MMX"?
>     -----------------------------------------------------------------------
>     av_cold void ff_h264dsp_init_loongson(...){
>         ...
>         c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmi;
>         ...
>     }
>     -----------------------------------------------------------------------

you can use any suffix that is appropriate

Thanks

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

He who knows, does not speak. He who speaks, does not know. -- Lao Tsu
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20150320/300b7cb8/attachment.asc>


More information about the ffmpeg-devel mailing list