Re: [brlcad-devel] A more efficient algorithm for implementing bn_mat_identity and bn_mat_inverse

Fri, 26 Jul 2013 11:08:33 -0700

On Jul 26, 2013, at 12:51 PM, Check Nyah wrote:

I have already sent you the tests; you requested. Just check in your mail

Got 'em!  I've reviewed and revised...  An updated version of the tests is posted and attached for you (or anyone else) to do a local comparison.  You'll need an optimized compile of BRL-CAD (cmake -DCMAKE_BUILD_TYPE=Release) and will need to edit the Makefile to point to the header/lib dir and maybe change -rpath to -Wl,-rpath but it should otherwise be made to work easily.

The cycle counts were not a useful reflection of actual time, so I rewrote the timings to use our performance timer and to isolate the new functions in question.  This gives solid comparison performance numbers that should take about 2 and 5-10 minutes respectively to run the determinant and inverse test (unless you have a crazy-fast/slow machine).

Also, were the numbers stable?  That is, your two results are approximately 65s vs 55s which is about 15% reduction in time.  If you ran the comparison a couple more times with different matrices, is always about 15%?

Yes. Well I can't remember the exact figures. but my implementations especially for the bn_mat_inverse was significantly faster than that which was originally implemented on brlcad.

What I'm seeing is a fair bit of variability from platform to platform, but stable per environment.  So far, I've tested on three wildly different platforms: a Mac Core2, a FreeBSD Xeon, and an IBM Power7 (PPC64).

The new determinant function was actually 8% SLOWER on the Mac, but then 32% faster on FreeBSD and 4% faster on the Power7.

The new inverse function was 11% faster on the Mac, 8% faster on the FreeBSD, and a whooping 50% faster on Power7 (awesome!).

The numbers were relatively stable +-2 percent and predominantly varied on how many iterations were performed.  In the tests, I made it so one can either test for a set amount of time or a fixed number of iterations.

Would love to see what other people's numbers look like.


p.s. Please remember about bottom-posting and deletion

Well this is the test i'm trying already. So how is my performance at this one?

Outstanding.

Cheers!
Sean


Attachment: bnmatinvdet.tar.gz
Description: GNU Zip compressed data


------------------------------------------------------------------------------
See everything from the browser to the database with AppDynamics
Get end-to-end visibility with application monitoring from AppDynamics
Isolate bottlenecks and diagnose root cause in seconds.
Start your free trial of AppDynamics Pro today!
http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk
_______________________________________________
BRL-CAD Developer mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/brlcad-devel

Reply via email to