Subject: Re: [alsa-devel] Re: [linux-audio-dev] laaga, round 2
From: Abramo Bagnara (abramo_AT_alsa-project.org)
Date: Tue May 15 2001 - 17:42:53 EEST
Paul Davis wrote:
>
> Whether we retain busses or not, I propose the following: unlike the
> context switch costs that we've been discussing, the cost of zeroing
> memory goes down more or less linearly with processor speed (helped
> along by RAM speed, which is increasing much more slowly). Because of
> this, we should consider the cost to zero a buffer to be an
> essentially vanishing item in the overall operation of an audio
> system. Because of this, all output memory buffers should be used as
> "busses" - they are, as tom put it, "summing nodes". All operations to
> store data in them should use "+=", and not "=". The host will provide
> buffers, and will ensure that they are zeroed when appropriate.
I think you should have benchmarked this statement before to write it
;-)
If buffer is in cache, zero + sum take more than 3 times the cycles
needed to do assignment only.
On a 333 Mhz machine this means to use an extra 0.1% CPU every 48000
samples processed.
*Now* we may decide if it's relevant or not.
-- Abramo Bagnara mailto:abramo_AT_alsa-project.orgOpera Unica Phone: +39.546.656023 Via Emilia Interna, 140 48014 Castel Bolognese (RA) - Italy
ALSA project http://www.alsa-project.org It sounds good!
This archive was generated by hypermail 2b28 : Tue May 15 2001 - 18:00:26 EEST