[vdr] Re: Softdevice + EPIA-M / Sound resampling

[Holger Waechtler <holger@convergence.de>]

Colin Paton wrote:
> Hi,
>
> I suspect that the audio thread may be being starved of CPU access which is
> why I'm getting funny audio.
>
> I did rough benchmarks with 'top' and got the following results on an M10000
> Nehemiah CPU:
>
> Playing back with sound + MPEG decoding but no video display (but call to
> 'Flip') = ~17% CPU
> Playing back with blit added (get video display) = ~78% CPU
That's quite a lot. Such problems typycally occur when you take 
unaccelerated paths through the library or place video surfaces in video 
memory where they would be better places in system memory and vice 
versa. Can you please describe your scenario on the DirectFB mailing 
list and ask there for help optimizing your code paths?

btw: are you still using the default RBG/YUV conversion code? Have you 
tried using the libavcodec img_convert() functions instead?

Are you converting directly into the locked target overlay surface or 
are you explictly blitting?

> Playing back with FB sync but no blit = ~46% CPU
> Playing back with blit + FB sync = ~98% CPU
>
> Without sync and blit audio sounds OK.
Then maybe the buffers are not large enough to hold data in burst cases.

Holger



--
Info:
To unsubscribe send a mail to ecartis@linuxtv.org with "unsubscribe vdr" as subject.