25 Aug
2004
25 Aug
'04
12:16 p.m.
Just out of curiousity, what optimizations did he suggest and which ones were a good
idea?
Taybin
-----Original Message-----
From: Erik de Castro Lopo <erikd-lad(a)mega-nerd.com>
Sent: Aug 25, 2004 3:58 AM
To:
The Linux Audio Developers' Mailing List <linux-audio-dev(a)music.columbia.edu>
Subject: Re: [linux-audio-dev] Hello - FYI - intro
On Tue, 24 Aug 2004 19:56:27 -0400
Paul Winkler <pw_lists(a)slinkp.com> wrote:
On Tue, Aug 24, 2004 at 06:35:12PM -0400, John Check
wrote:
Are you sure you wanted to say that ;)
Yes. First rule of optimization: don't optimize something that's
statistically irrelevant.
<vent=11/10>
I wish that attitude was more prevalent in the ricer community.
I've been getting emails from a Gentoo ricer suggesting a whole
bunch of "optimisations" for libsndfile. A small number of these
were valid, but the vast majority were not measurably faster as
measured by the benchmark program shipped with libsndfile.
</vent>
Erik
--
+-----------------------------------------------------------+
Erik de Castro Lopo nospam(a)mega-nerd.com (Yes it's valid)
+-----------------------------------------------------------+
"The X-files is too optimistic. The truth is not out there."
-- Anthony Ord
It's completely dwarfed by compile time.
25 Aug
25 Aug
7:59 p.m.
New subject: [linux-audio-dev] Hello - FYI - intro
On Wed, 25 Aug 2004 10:00:30 -0400 (GMT-04:00)
Taybin Rutkin <taybin(a)earthlink.net> wrote:
Just out of curiousity, what optimizations did he
suggest and
which ones were a good idea?
The main two "optimisations" were measurable in isolation (small
test programs), but not in the benchmark program:
- I was using my own endswap macros. I had benchmarked mine
and they performed well in comparison to one I found in
ffmpeg but not in comparion to the ones in <byteswap.h>
- Inner loops were changed from:
while (count) { count -- ; /* whatever */ }
to :
while (--count >= 0) { /* whatever */ }
Other suggestions (which couldn't be easily tested in isolation
and made no measurable difference in the real benchmark) included:
- Rewrite a loop which contains a read() system call on 16k
of data so that the "long long" which is incremented inside
the loop can be changed to an int.
- Changing a whole bunch of functions to inline.
The above exercise took up a whole bunch of my time for negligible
gain. Personally I wish I had that time back to wok on more useful
and interesting stuff.
Erik
--
+-----------------------------------------------------------+
Erik de Castro Lopo nospam(a)mega-nerd.com (Yes it's valid)
+-----------------------------------------------------------+
"UNIX was not designed to stop you from doing stupid things, because
that would also stop you from doing clever things." -- Doug Gwyn
7428
days inactive
7428
days old
linux-audio-dev@lists.linuxaudio.org
1 comments
2 participants
participants (2)
-
Erik de Castro Lopo -
Taybin Rutkin