R devel - Aug 2005 - BLAS and ATLAS

On Mon, Aug 29, 2005 at 11:48:23AM +0100, Prof Brian Ripley
wrote:
> On Mon, 29 Aug 2005, G?ran Brostr?m wrote:
> >LAPACK/BLAS routines call xerbla "if an input parameter has an
> >invalid value" (exact quote from  XERBLA at Netlib). Other types
of errors
> >are not printed (by xerbla), but it is up to the user to check the
return
> >value of the argument 'INFO'. See
> >http://www.netlib.org/lapack/lug/node119.html.
> 
> Maybe, that is not actually true for real-life LAPACK implementations, as 
> I did say.
> 
> >So we can safely give the more informative, original, error messages,
> >together with mentioning LAPACK and eventually BLAS, whatever you
prefer.
> >This would "make clear what these refer to". Given that other
BLAS/LAPACK
> >routines than xerbla aren't changed in R, of course.
> 
> Irrelevant as R is often compiled against other ones implementations.
That confuses me. My R installation (R-devel of yesterday) is compiled
against ATLAS BLAS (from Debian-unstable). Why, then, is xerbla from  R  
sources used anyway? 

G?ran

On Mon, 29 Aug 2005, G?ran Brostr?m wrote:
> On Mon, Aug 29, 2005 at 11:48:23AM +0100, Prof Brian Ripley wrote:
>> On Mon, 29 Aug 2005, G?ran Brostr?m wrote:
>
>>> LAPACK/BLAS routines call xerbla "if an input parameter has an
>>> invalid value" (exact quote from  XERBLA at Netlib). Other
types of errors
>>> are not printed (by xerbla), but it is up to the user to check the
return
>>> value of the argument 'INFO'. See
>>> http://www.netlib.org/lapack/lug/node119.html.
>>
>> Maybe, that is not actually true for real-life LAPACK implementations,
as
>> I did say.
>>
>>> So we can safely give the more informative, original, error
messages,
>>> together with mentioning LAPACK and eventually BLAS, whatever you
prefer.
>>> This would "make clear what these refer to". Given that
other BLAS/LAPACK
>>> routines than xerbla aren't changed in R, of course.
>>
>> Irrelevant as R is often compiled against other ones implementations.
>
> That confuses me. My R installation (R-devel of yesterday) is compiled
> against ATLAS BLAS (from Debian-unstable). Why, then, is xerbla from  R
> sources used anyway?
Because it is the only one seen: ATLAS does not provide it.  There is no 
guarantee that ATLAS or libsunperf or vecLib ... only use xerbla in the 
meaning in that page.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595