Re: Octave 3.6.0 on Windows XP plot fails.

[Tatsuro MATSUOKA]

Hello

I have tested full test on the development branch source (changeset 
14388:b5b49f400f9b) using the latest OpenBLAS (alpha2.5 9dcea60).
(Octave and related dependency was linked with Reference BLAS. The libblas.dll 
is replaced with that generated OpenBLAS.)

The OpenBLAS gives only one additional failure (Test in svds.m) for MinGW build.

Regards

Tatsuro


--- On Fri, 2012/2/17, Michael Goffioul  wrote:

> On Fri, Feb 17, 2012 at 11:35 AM, Philip Nienhuis <address@hidden> wrote:
> > This multi-architecture OpenBLAS works OK with MinGW-Octave-3.6.0 on both my
> > PCs:
> > - Core Duo 2.5 Ghz
> > - Pentium-M 1.1 Ghz
> > (both WinXP SP3)
> >
> > That is.... I haven't run any benchmarks, I just verified that the test:
> >>>>>> m=10000;
> >>>>>> n=32;
> >>>>>> temp=randn(m,n)+i*randn(m,n);
> >>>>>> b=temp'*temp;
> > runs without a crash.
> 
> Try to run the full octave test suite. Hopefully that should cover
> enough cases. I tried OpenBLAS recently, compiled under MinGW
> (gcc-4.6.1) [1], and although a simple test like the one above didn't
> crash, more extensive testing still produced a crash.
> 
> Michael.
> 
> [1]: I compiled OpenBLAS fully with gcc then dropped the DLL to
> replace my existing one, then tested it with my octave compiled with
> MSVC. To make sure ABI are compatible, I compiled OpenBLAS with
> -mpreferred-stack-boundary=2 and -ff2c (for gfortran).
>