Segmentation violation in test_sparse.m octave built on mingw with GotoBLAS

[Tatsuro MATSUOKA]

Hello

I have been trying to build an octave with fast matrix calculation on windows.

My tryial is ALTLAS and GotoBLAS. On cygwin, trials have been successful for 
ATLAS, multithreaded
ATLAS using posix thread and GotoBLAS.

I have tried on octave-3.0.3.

I have been tried building on mingw platform using gcc-4.3.3-dw2-TDM.
The single threaded ATLAS using libf77blas.a, libcblas.a and libatlas.a made on 
cygwin using
gcc-4.3.2-2 on cygwin has been successful.

One of collegue in octave thread in Japan successfully have reported that the 
GotoBLAS and
multithreaded ATLAS made good results on Core2Duo and Core2Quad.

n=2000; A=randn(n); B=randn(n);tic; C=A*B; t=toc, MFLOPS=2*n^3/t*1e-6
(I do not know the validity of the MFLOPS value for this test but it is near to 
that I have seen at
ATLAS build so that the value is not non-sense.)

Intel Core2Duo 2.4GHz GotoBLAS GFLOPS = 15.75
Core2Duo E8500 3.16GHz Multithreaded ATLAS ca.19 GFLOPS.

Those are suprizing results.

I have tried GotoBLAS on mingw build itself went well, however segmentation 
violation always occur at
test_sparse.m.  I have tried two computers in university (HT-Pentium 3.4GHz 
prescot) and home
(Celeron-M 1.2GHz).

Now I am in home so that the resuls on gdb of GotoBLAS+octave executable is 
only for my home.

The gdb reported

  test_sparse.m ..........................................
Program received signal SIGSEGV, Segmentation fault.
0x64aeda26 in zpotf2_ ()
   from C:\Programs\msys\1.0\local\octave-3.0.3g\bin\liboctave.dll
Current language:  auto; currently c++

This is a just report but it would be grateful for me if somebody give me 
suggestions.


Regards

Tatsuro

--------------------------------------
Power up the Internet with Yahoo! Toolbar.
http://pr.mail.yahoo.co.jp/toolbar/