[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: benchmark octave with atlas and vecLib
From: |
Dmitri A. Sergatskov |
Subject: |
Re: benchmark octave with atlas and vecLib |
Date: |
Wed, 05 Nov 2003 14:53:03 -0700 |
User-agent: |
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030624 Netscape/7.1 |
Gavin Jin wrote:
I just ran Octave2.m, a benchmark test file I obtained from
http://www.sciviews.org/other/benchmark.htm, on the same powerMac G4
dual 1.25GHz
with 2GB of RAM with the following two version and library combinations:
1. octave-atlas I installed with fink
GNU Octave, version 2.1.35 (powerpc-apple-darwin6.3).
...
Creation, transp., deformation of a 1500x1500 matrix (sec): 6.91675
...
<other results deleted>
2. Octave 2.1.50 with vecLib and fftw that I obtained from
http://homepage.mac.com/chibaf/math/macosx/install/
....
Creation, transp., deformation of a 1500x1500 matrix (sec): 2.479
...
<other results deleted>
The dramatic improvement here is probably due to dramatic improvement of rand()
function from octave 2.1.35 to octave 2.1.50:
(From Octave2.m)
<<<<
disp(' I. Matrix calculation')
disp(' ---------------------')
% (1)
cumulate = 0; a = 0; b = 0;
for i = 1:runs
tic;
a = abs(randn(1500, 1500)/10);
b = a';
a = reshape(b, 750, 3000);
b = a';
timing = toc;
cumulate = cumulate + timing;
end;
>>>>
I'm not sure why we're seeing impressive performance with vecLib only on
I.A, I.E, II.C, and II.E, but not other tests.
I would also question quality of Atlas library used for the first test.
Here is results from my Octave2.1.50/Athlon 2GHz/Atlas3.5.3
(which are more in line with your second test):
Octave Benchmark 2
==================
Number of times each test is run__________________________: 3
I. Matrix calculation
---------------------
Creation, transp., deformation of a 1500x1500 matrix (sec): 1.459
800x800 normal distributed random matrix ^1000______ (sec): 0.1039
Sorting of 2,000,000 random values__________________ (sec): 7.866
700x700 cross-product matrix (b = a' * a)___________ (sec): 0.2788
Linear regression over a 600x600 matrix (c = a \ b') (sec): 0.1699
------------------------------------------------------
Trimmed geom. mean (2 extremes eliminated): 0.2805
II. Matrix functions
--------------------
FFT over 800,000 random values______________________ (sec): 0.709
Eigenvalues of a 320x320 random matrix______________ (sec): 1.997
Determinant of a 650x650 random matrix______________ (sec): 0.2093
Cholesky decomposition of a 900x900 matrix__________ (sec): 0.2299
Inverse of a 400x400 random matrix__________________ (sec): 0.119
------------------------------------------------------
Trimmed geom. mean (2 extremes eliminated): 0.3243
III. Programmation
------------------
750,000 Fibonacci numbers calculation (vector calc)_ (sec): 0.3955
Creation of a 2250x2250 Hilbert matrix (matrix calc) (sec): 0.8678
(If I remove rand() from the timing loop for the first test,
the number goes down to 0.4657)
I do not know what this vecLib does, but if this is for ALTIVEC
then octave has hardly any use of it...
Regards,
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
-------------------------------------------------------------