It is a bug of lapack 3.2 or of gfortran ?

Open discussion regarding features, bugs, issues, vendors, etc.

It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Fri Dec 19, 2008 12:18 am

my make.inc :

FORTRAN = gfortran
LOADER = $(FORTRAN)
OPTS = -fbounds-check
DRVOPTS = $(OPTS)
NOOPT =
LOADOPTS = -fbounds-check

when make lapack_testing ,
$./xlintsts < stest.in > stest.out
At line 47 of file ilaslr.f
Fortran runtime error: Array reference out of bounds for array 'a', lower bound of dimension 1 exceeded (0 < 1)
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Fri Dec 19, 2008 10:45 am

I posted this to google fortran group ,got some answers, please see:

http://groups.google.com/group/comp.lan ... 11beadb93#
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby Julien Langou » Fri Dec 19, 2008 1:10 pm

Thanks a lot for the asking around and getting an answer. Tobbias also sent us an email a few hours ago.
Best wishes,
Julien.
Julien Langou
 
Posts: 727
Joined: Thu Dec 09, 2004 12:32 pm
Location: Denver, CO, USA

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Thu Jan 29, 2009 4:49 am

hi,

during this "spring day" holiday , I read through the fortran95 standard and found these words in chapter 7.1.7.1:

" it is not necessary for a processor to evaluate all of the operands of an expression. or to evaluate entirely each operand.
if the value of the expression acn be determinded otherwise".


so maybe, this problem is of gfortran compiler instead of lapack?
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Thu Jan 29, 2009 4:53 am

the same words also appear in chapter 7.1.7 of fortran2008 draft.
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby Julien Langou » Thu Jan 29, 2009 12:29 pm

Thanks a lot for checking the Fortran95/2008 standard. So if you read:
it is not necessary for a processor to evaluate all of the operands of an expression. or to evaluate entirely each operand.
if the value of the expression can be determinded otherwise

Then this does not help us. "It is not necessary to" does not mean "It is forbidden to". The statement says "it is not necessary for a processor to" however nothing prevents him from doing it if it "wants" to. gfortran flags us because we are calling the routine with an out-of-bound arrray, we were doing it although this array is not useful for our computation.I believe gfortran is correct.
By the way this is patched in the svn repos and waiting for release.
---julien
Julien Langou
 
Posts: 727
Joined: Thu Dec 09, 2004 12:32 pm
Location: Denver, CO, USA

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Fri Jan 30, 2009 1:24 am

thanks for clarification and your work.

for the first time I heard that lapack project use a subversion server, but It seems the server is for developers instead of public?
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby Julien Langou » Fri Jan 30, 2009 11:43 am

Yep, yep, we are internally using svn. The svn repos is not yet open to the public.
The "yet" imply that it might be at some point.
Opinion of users on this subject matter is actually welcome.
Julien.
Julien Langou
 
Posts: 727
Joined: Thu Dec 09, 2004 12:32 pm
Location: Denver, CO, USA

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby linuxl4 » Fri Jan 30, 2009 1:09 pm

personally I like use a svn version so as not to wait. :)

A new release gives users many new characters , and I prefer to enjoy them one by one.
linuxl4
 
Posts: 10
Joined: Fri Dec 19, 2008 12:09 am

Re: It is a bug of lapack 3.2 or of gfortran ?

Postby cottrell » Fri Jan 30, 2009 9:23 pm

There are pluses and minuses of opening your svn repository.

Plus: You may well get some well-informed suggestions for improvements from "the community".
And Lapack enthusiasts get to try out the latest-and-greatest.

Minus: Developers need to have the opportunity to break things from time to time, in order
to achieve substantial improvement. But this could land you with a lot of email complaining
about problems you're well aware of.

But I guess you know these things already ;-)

I'd tend to favour an open repository, but I'm hesitant: Lapack is so basic, and so widely used,
that making the development version publicly available might cause more trouble than it's
worth -- even if you put out the customary warnings.
cottrell
 
Posts: 68
Joined: Thu Jan 15, 2009 1:40 pm


Return to User Discussion

Who is online

Users browsing this forum: No registered users and 3 guests

cron