[vdr] Re: [PATCH] Reduce CPU usage during playback

[Jon Burgess <mplayer@jburgess.uklinux.net>]

Sascha Volkenandt wrote:
> At Freitag, 13. Juni 2003 19:15 Jon Burgess wrote:
> Remarkable is, that this patch applies at a completely different location 
> (Poller GETS data, but no playFrame is available afterwards) and drops CPU 
This patch (from Sascha) works for me as well, in fact it works better 
than the one I posted. I think they both accomplish more or less the 
same thing - to slow down the iteractions around the dvbplayer::Action 
loop when waiting for the non blocking file reader to provide data for 
the ringbuffer + output device.

I think your patch works better because it inserts the delay when there 
is no data to take from the ringbuffer to the output device, where as my 
patch inserts the delay when there was no data to take from the file 
reader to the ringbuffer. Hence my approach would insert a delay, even 
if there was actually data remaining in the ringbuffer which could be 
fed to the output device. Your approach only waits when there is truly 
nothing that the action loop can do to help provide data to the output 
device.

Both patches catch the real CPU hog situation which occurs when both the 
ringbuffer and the file reader have no data available, but the output 
device was will to accept data (hence the poll() returns immeadiately).

As another data point, I only see the excess CPU load after a playback 
has been running for some 5 - 10 seconds (without the patch), perhaps 
there is some subtle timing or interaction problem that causes the worst 
case to only appear on some systems.

	Jon




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