help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: False alarm (Re: reshape slowdown?)


From: Dmitri A. Sergatskov
Subject: Re: False alarm (Re: reshape slowdown?)
Date: Tue, 22 Feb 2005 19:21:24 -0700
User-agent: Mozilla Thunderbird (X11/20041216)

David Bateman wrote:
Dmitri,


It seems you've fall on the same problem as my colleague Laurent came across, though he identified it as happening with the fft function I believe. In any case I came confirm this is an issue... Though last I heard Laurent was as lost as you.


I know what is the problem now. I do not know how to solve it.
The problem is that when running in SMP mode the task is bouncing between
two CPUs flushing the caches. That can be verified booting in UP mode.
I had this problem some 5 years ago (in the time of kernel 2.2).
Back then the problem was hardware dependent -- I had two SMP system
(one intel 440BX and another 440 LX). The problem was with some
compiled C code which would run significantly slower on LX
while in SMP mode but almost the same speed in UP mode.
One could also see significantly more system time when time
the task (in BX system most time was a user time). I had some
exchange with Alan Cox on the matter, and while he conferred
the problem it was no easy solution. These systems are retired
now...
So on my system in UP mode:

octave:1> a=rand(3000);
octave:2> tic; b=a'; toc
ans = 0.80099

Which is 3 times faster!

I am trying to reproduce the problem with a simple standalone program,
so I can bug RH people with it.
If it is possible to strip out the matrix transpose code from octave,
that would be great, but I tried and could not really make it.

Regards
David


Sincerely,

Dmitri.
--



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------



reply via email to

[Prev in Thread] Current Thread [Next in Thread]