Re: [tablix-list] Scope for small (5%) speed improvement in Tablix

From: Tomaz Solc <tomaz.solcREMOVE@THISsiol.net>
Date: Sat Aug 07 2004 - 18:48:11 CEST

Hi

I've changed all *map structures to simple arrays as you described and
did some benchmarking. I've made three 10 minute tests with modified and
unmodified version of Tablix and noted the average GPM value.
Command line I used: tablix -t10 -d3 sample.xml

Modified Tablix: 953.9, 960.9, 958.3
average: 958.5

Unmodified Tablix 0.0.8: 973.5, 966.3, 975.5
average: 971.8

It seems that at least on my machine (athlon xp 2500) the unmodified
version is slightly faster.

I'm also experimenting with using mmx optimized memcpy routine from
MPlayer. gprof shows that about 20% of CPU time is spent in
mate_chromo() which is essentialy moving blocks of chromosomes around in
memory. Some optimization here would probably have a large effect on
overall speed of Tablix.

I've created a new branch in CVS repository called EXPERIMENTAL_0_1_0
which contains these modified versions, so you can test them on your
machine if you wish.

Best regards
Tomaz Solc

Received on Sat Aug 07 19:09:12 2004

This archive was generated by hypermail 2.1.8 : Tue Aug 16 2005 - 20:42:18 CEST