Re: [xine-devel] video_out flaw?

[Michel L. <wa...@zo...>]

On Mon, Apr 29, 2002 at 11:06:53PM +0200, Gildas Bazin wrote:
> If you are interested, I made several comparable experiments with the 
> VideoLAN client which I think uses the same process to decode mpeg2 frames.

Yes, the mpeg2 decoder in videolan is really close to libmpeg2 these
days.

> My last experiment was to try to use AGP memory for the frame buffers and 
> then DMA the buffers into video memory. This should have had the advantages 
> of both worlds: fast r/w memory access and no memcpy required!
> Unfortunately I learnt afterwards that Windows disable all caching on AGP 
> memory to keep a coherent view between the CPU and Graphics card. And 
> indeed the performance is at least as bad as using video memory directly.

Hmmm. Interesting. Is there no way to ask windows to not disable the
caching ? It's really too bad - it would be easy enough to make sure
we write back the cache before we switch the video buffer...

> I'm still waiting for a graphics card that supports DMA transfers from 
> system memory to video memory...

That'd be really cool :)

-- 
Michel "Walken" LESPINASSE
Is this the best that god can do ? Then I'm not impressed.