Re: slow LU decomposiotion? for octave 4 for windows

[Dmitri A. Sergatskov]

On Tue, May 31, 2016 at 7:42 AM, Dmitri A. Sergatskov <address@hidden> wrote:

On Tue, May 31, 2016 at 1:41 AM, Tatsuro MATSUOKA <address@hidden> wrote:

Hello

I am now preparing the lecture notes for my class using octave on my University.
I noticed that bench mark for the lu decomposition is slow on octave 4. 

​<...........>​

 

************************************
Octave-4.0.2 mingw (64 bit)

>> lutest
Simple left division
Elapsed time is 0.526351 seconds.
LU decomposition
Elapsed time is 0.112075 seconds.

************************************
octave-4.1.0+ mingw (64bit)
(hg clone on May 28, 2016)

>> lutest
Simple left division
Elapsed time is 2.01935 seconds.
LU decomposition
Elapsed time is 0.125084 seconds.
************************************

The lu decomposition test is very slow on octave 4.
While simple division is fastest on octave 4.0 but is slowest on 4.1.0+.

What happens ?

Does anyone see this on other platforms?

​On Fedora 23 (x86_64) 4.1.0+ (self-compiled) is slightly faster on lutest than 4.0.2​ dist rpm):

LD_PRELOAD=/usr/lib64/libopenblasp.so octave --no-gui
GNU Octave, version 4.0.2

>> a=randn(4000);
>> tic; inv(a)*a ; toc
Elapsed time is 6.08218 seconds.
>> lutest
Simple left division
Elapsed time is 0.242294 seconds.
LU decomposition
Elapsed time is 0.11199 seconds.
>>

LD_PRELOAD=/usr/lib64/libopenblasp.so ./run-octave --no-gui
GNU Octave, version 4.1.0+

>> lutest
Simple left division
Elapsed time is 0.241105 seconds.
LU decomposition
Elapsed time is 0.11608 seconds.
>> a=randn(3000);
>> tic; inv(a)*a; toc
Elapsed time is 1.97684 seconds.

​(wrong array size)​

>> a=randn(4000);
>> tic; inv(a)*a; toc
Elapsed time is 5.71351 seconds.

 

​This is on good old ​i7-2600K CPU @ 3.40GHz

 

Tatsuro MATSUOKA

​Dmitri.
--
​