> Good idea. I submitted a patch to do just that. Works great, except
> it still winds up being ahead by ~1 second about 90% of the time. So
> I just add another second to the latency and calculate using that.
> Perhaps the accuracy of audiobuffer_get_latency() could be improved.

Could we use a smaller DMA length so that we get more DMA0 interrupts?
That would allow us to update the latency more often, rather than once per
block. I think. I'm not looking at the code right now and I have a
sketchy understanding at best anyway

(Note I'm not suggesting we change the blocksize, just that we set program
the DMA length to be MIN( length, SMALL ) where SMALL could be 1024 for
example).

d

_______________________________________________
http://cool.haxx.se/mailman/listinfo/rockbox
Received on Tue Jul 12 14:48:33 2005