26 Jan
2011
26 Jan
'11
8:59 p.m.
For some reason, gmail has stopped delivering my LAD messages, so
apologies if this reply isn't working correctly...
I'm attempting to run, in general, a 128 sample buffer (frames per
fragment). There is a lower latency mode though which is 64 samples.
I have two fragments. Maybe if I want to do this I should be using
three at least or make 128 the minimum.
But I'm a little confused at the lack of ins and outs being linked.
For a full duplex loop, I'd have to have the input before I can
process output for example, so I don't see how that works unless I
just ignore input requests from ALSA if I am not yet ready to receive.
I'm defining "ready to receive input" as the time after I run my
processing loop and then get a POLLOUT request. However, when I
ignore POLLINs like this, it seems like ALSA immediately POLLINs again
and I wind up with something like 50,000 polling requests handled in a
second, which drives the CPU usage through the roof.
the in & out are not linked in any way as far as
ALSA is concerned
(even though they may be at the h/w level).
a 64 sample buffer (not period) is very, very small. you need very
tight code and a well configured machine to operate like this. what
period size and/or number of periods are you attempting to use?
> Hey, I've been having a lot of trouble with ALSA and processing input
> in a full duplex loop, and I was wondering if anyone here had any
> insight. I'm using the ALSA poll file descriptor method with the mmap
> routines. Typically, my processing loop is blowing past the poll
> statement with input requests (POLLIN), causing the CPU usage to go
> off the charts as it's basically looping as fast as it can go.
>
> Now, I'm handling the inputs in a way that might not be correct.
> Often, ALSA is saying that the available frames is 1.5 times my buffer
> size (e.g., I set a 64 sample buffer and ALSA offers me 96 samples).
> I only get (begin/commit) input when the buffer size matches or
> exceeds the buffer length I send out to the card (the set buffer
> size), and if it exceeds that, I only pick up (commit) one buffer's
> length of frames. Is this wrong? If there is more input waiting,
> does ALSA immediately go and POLLIN again, or should I expect it to
> not bug me again until I've processed a POLLOUT event (the in and out
> are linked)? I'm not really sure how to handle this..