Re: Any help needed in Fortran related development

[Jaroslav Hajek]

2010/8/25 Jordi Gutiérrez Hermoso <address@hidden>:
> On 24 August 2010 21:29, Jerry DeLisle <address@hidden> wrote:
>> I am curious if there are any particular areas within Octave I may be able
>> to help with?
>
> Yes!
>
> I've been discussing in private a problem with qrupdate[1] that I've
> encountered while packaging Jaroslav's work for Debian.
>
> Briefly, compile qrupdate with the ATLAS implementation of BLAS and
> run the tests. Some of the tests fail. Probably a bug in ATLAS?
>
> Since the Debian packaging seems by default to link with ATLAS, it
> would be nice to know if we're using a buggy implementation of BLAS
> for shipping Octave.
>
>     http://sourceforge.net/projects/qrupdate/files/
>
> Note that the apparent bug happens with the newer rank-1 updates that
> got added in version 1.1.1.
>

No, it doesn't. In fact it doesn't concern the updating routines at
all, as I tried to explain. The problem is that CHERK from BLAS
produces incorrect results (NaNs on diagonal). In version 1.1.1, I
added code to detect this particular problem, warn about it and fix
it. Hence you get all tests passed but a warning about buggy BLAS.

I can see this with ATLAS 3.6.0 which is quite outdated, hence I
wanted to try a newer version before filing a bug, but I never got
round to compiling it :-(

In any case, the bug apparently has nothing to do with qrupdate
(besides being discovered by it), and consequently with Fortran, the
bug is/was in ATLAS (which itself is C+asm).

-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz