The new Judy (probably rev 2.0) is going very slowly. I am having problems debugging it quickly,
while testing the effects of modifications on speed. So, I decided to use the 1.0.5 version of Judy
to try some of the speed fixes. There is one "fix" to "JudyGet.c" that improves the speed about
50%. Some of the others changes require many file changes. I will make that one "fix" and send
you the source to try out, if you are interested. I will incorporate some of the other changes
in rev 1.0.6 when I know how to use
"popc" in Windows and Apple compilers. Perhaps, I
will only use it in the 64bit versions (or until Core 2 duo becomes obsolete).
I only use one SSE instruction in the new Judy so far -- popc (bit counting). I would be very interested
what other you use to get the %20 speed gain.
Thanks for you interest,
Doug Baskins <firstname.lastname@example.org>
From: gwenael chailleu <email@example.com>
To: Doug Baskins <firstname.lastname@example.org>
Sent: Saturday, November 3, 2012 7:42 AM
Subject: Re: better thank twice than not at all !
We open-sourced our gnu m4 extension (using JudyBL built over JudyL) last week and we were wondering how your JudyL recoding was going ?
On another matter, we have dowloaded openMalaria, an open source BOINC project in order to optimize it.
We reached à 20 % speed gain only with a few SSE math functions.
Are you thinking of using SIMD instructions in your new version ?