Subject: Re: [linux-audio-dev] Realtime
From: Paul Barton-Davis (pbd_AT_Op.Net)
Date: Fri Jun 30 2000 - 04:49:09 EEST
>> alas, its nots so simple. there are always buffers on the audio
>> interface which accumulate sample data in both directions. plus, there
>> is the PCI bus burst transfer size (64 bytes). so, lets assume a
>> minimum transfer size of 64 bytes:
>
>The buffering is not configurable?
it is, but not down in this range. you typically get to choose from
between 32 and 8192 samples in the buffer. the PCI burst transfer
size, as i understand it, is a design spec for the bus, and is not
configurable.
>>
>> 16 32 bit samples @ 48kHz: 320usec
>> 32 16 bit samples @ 48kHz: 640usec
>600 usecs is a very long time for PIII700 or an Alpha.
>It should be more than 600,000 instructions.
>There are no interesting things to do in such times?
>Is it interesting to do "live" effects?
> Music in -> do something ->Music out
> |->log
Thats *exactly* the point! And thats one of the many reasons we want
improved latency/scheduling jitter figures for the main kernel.
BTW, we can chew up those 600,000 instructions in no time at all. Just
put a decent reverb algorithm in there with early, middle and late
reflections computed from a physical room model, run it on, oh, about
6 channels, and your 600usec will be gone in an instant. I may be
underestimating the number of channels, but not by much. Most current
dedicated FX processors can rarely do good reverb for more than 4
channels, and very few do more than 2. Thats with kick-ass dedicated
DSP.
--p
This archive was generated by hypermail 2b28 : Fri Jun 30 2000 - 05:38:26 EEST