octave-maintainers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: compiling development sources


From: Jaroslav Hajek
Subject: Re: compiling development sources
Date: Thu, 11 Feb 2010 12:50:01 +0100

On Thu, Feb 11, 2010 at 12:40 PM, Carlo de Falco
<address@hidden> wrote:
>
> On 11 Feb 2010, at 11:49, Jaroslav Hajek wrote:
>
>>
>> The tests are randomized so random failures like this mean nothing (in
>> theory you can get arbitrarily bad residuals). The single precision
>> routines generally produce much worse residuals. This is the reason
>> why the seed is hard-coded.
>> Anyway I just discovered that I occassionally get failures as well,
>> cause by NaNs compute by CHERK.
>> I'm sending you an unofficial (yet) qrupdate-1.1.1 tarball. Please
>> rebuild and rerun the testsuite, possibly trying multiple seeds as
>> well. Pay attention to possible warnings at the end. I'm interested in
>> your results.
>>
>> Regards
>
> Jaroslav, attached is the result of
> ( make clean && make test ) > result 2>& 1
> I tried to change the seed but I can't get any failures.
> Does it make sense to try to rebuild Octave linking to this new version?
> I never got any segfaults with qrupdate itself, is it possible that
> the problems you found are the cause of the crash in Octave?
> c.
>

It's surely worth trying, although as I said, not a single line of
code in the library routines has changed since 1.1.0, it's just the
test suite, so in theory the binary should be identical to 1.1.0.
Compared to 1.0, only 8 new routines were added. If you start seeing
the test failures in Octave again, try rebuilding and re-running the
qrupdate test suite. For me, the manifestation of the bug seemed to be
dependent on previous calls to BLAS made, and it apparently made a
difference whether the BLAS remained in memory or not.
In any case, it will help if you discover which test was failing for you:
./run-octave
octave:1> test qr verbose

this should dump the tests as they're tried so you'll see which one crashed.


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


reply via email to

[Prev in Thread] Current Thread [Next in Thread]