Re: [Bug-gsl] Possible bug in gsl_multifit_wlinear_svd

[Brian Gough]

At Wed, 8 Aug 2007 23:47:01 +0200,
Marco Lombardi wrote:
> Sorry for the delay of this reply, I was on a trip last week.  I am
> running the test on a Mac, specifically on a MacBook Pro, with Mac OS
> 10.4.10 and gsl 1.9 installed.  The test also fails on various linux
> boxes I tried (e.g., RedHat with kernel 2.6.18, or Debian with kernel
> 2.6.21).  As far as I can tell, the problem is there also in gsl 1.8.
> 
> I suspect that you cannot reproduce the bug because the data file was
> not correctly transfered or read.  To (at least) exclude the first
> possibility, here is the md5 checksum of the file:
> 
> $ md5sum gsl-test.dat
> 768b5a964b8ce98932ed0618ac9f97e1  gsl-test.dat

Thanks, I have the file with the same md5sum.  I have looked again and
found that the problem occurs when the library is compiled with SSE
floating point (and so presumably is something to do with underflow,
which is handled differently in that case) -- I will log that bug to
be fixed in the next release.  Thanks for the bug report.

-- 
Brian Gough

Network Theory Ltd,
Publishing Free Software Manuals --- http://www.network-theory.co.uk/