Re: Compiled 3.8.0 from source on Ubuntu 12.10 slower than 3.6.3 from PPA

[Aielyn]

Mike Miller wrote
> If you can share your code, or some representative code that does
> similar operations and shows the same performance penalty, that might
> help others help you. As a start, I tried your simple one-liner for
> loop on one machine and I got a very slight (less than 5%) improvement
> in 3.8.0 over 3.6.4, even with JIT disabled, both self-compiled and
> using reference BLAS and LAPACK. So with that small sample, I'm not
> seeing a regression.
> 
> You might also try compiling 3.6.4 the same way you are compiling
> 3.8.0 as another point of comparison.

I can't really share my code, and the one-liner is close enough to being
representative to serve as a suitable test - at the very least, that
one-liner shows the same "regression"...

And to that point, I just tried compiling 3.6.4, as you suggested, starting
with the basic, default compile setting. With the one-liner, Octave 3.6.3
(from binary) takes roughly 3 seconds. Octave 3.8.0 takes roughly 4.2-4.3
seconds. Octave 3.6.4 takes about 2.5 seconds (unsure of why I'm getting
better performance - better loop-handling?).

For my code, 3.6.3 takes about 97 seconds for 5 steps. 3.8.0 takes about 149
seconds. 3.6.4 takes about 100 seconds. So it seems that there is some sort
of regression in Octave 3.8.0, or at least in its configuration process.

I suppose that information about my laptop will probably help, here. It's an
Intel(R) Core(TM) i3-3110M CPU @ 2.40GHz (four core), running Xubuntu 12.10,
Linux Kernel is 3.9.4-030904-generic. 

Side note: all times above had the CPU running at low speed, 1.2 GHz. I just
noticed it, and have retested both 3.8.0 and 3.6.4 with my code. The former
took about 69 seconds and the latter about 49 seconds.



--
View this message in context: 
http://octave.1599824.n4.nabble.com/Compiled-3-8-0-from-source-on-Ubuntu-12-10-slower-than-3-6-3-from-PPA-tp4662951p4663017.html
Sent from the Octave - General mailing list archive at Nabble.com.