FFMPEG based Theora Video Decoder performance??

[ajitsdeshpande@gmail.com]

Hi,

I am in process of porting and optimization of the theora video decoder in the ffmpeg-0.5 package to ARM-Cortex-A8 -Neon processor @ 667 MHz.
I am looking for some target estimate for frames per second the decoder library alone should achieve after full optimization (C level and Neon assembly / Intrinsics) for 720x480 Progressive content for a 2Mbps stream.
I have a Real Video 9 decoder on cortex-A8 which gives around 40 fps for the same stream above.(720x480, 2Mbps)
How can i extrapolate this data based on relative complexities of RV9 and Theora and get a fps estimate for theora decoder Cortex-A8?

I am aware the performance depends upon the cache configuration of the h/w, etc...,but any
Any pointers will help.

Thanks,

-AD

[Dark Shikari]

Porting? Optimization? Huh?

The ffmpeg Theora decoder already supports ARM (there's no "porting" necessary) and has a great deal of assembly on ARM (there's not much "optimization" to be done either).

[hellfred]

Originally Posted by ajitsdeshpande@gmail.com

Hi,

I am in process of porting and optimization of the theora video decoder in the ffmpeg-0.5 package to ARM-Cortex-A8 -Neon processor @ 667 MHz.

The theora decoder underwent many performance improvements (algorithmic improvements!) since the 0.5 release. Do use the current code from FFmpeg subversion repository. For a list of changes see e. g. the verion history of vp3.c.

[hellfred]

Do you know this news?

[Dark Shikari]

Originally Posted by hellfred

Do you know this news?

Yes. ffmpeg is still faster by a significant margin.

[hellfred]

Originally Posted by Dark Shikari

Yes. ffmpeg is still faster by a significant margin.

That link was meant for ajitsdeshpande to point him to another theora ARM decoder codebase.