Re: [R] linux compile options (64-bit)

From: Prof Brian Ripley <ripley_at_stats.ox.ac.uk>
Date: Tue 02 Aug 2005 - 19:58:31 EST

On Tue, 2 Aug 2005, Göran Broström wrote:

> On Tue, Aug 02, 2005 at 09:33:45AM +0100, Prof Brian Ripley wrote:
>> On Tue, 2 Aug 2005, Göran Broström wrote:
> [...]
>
>>> Incidentally, I have just tried building R on a Fujitsu Amilo amd64 with
>>> debian-amd64 (unstable) and ATLAS. Both 'make' and 'make check' worked
>>> without complaints with gcc-3.4.5,
>>
>> There is no gcc 3.4.5! There are versions 3.3.5 and 3.4.4.
>
> I apologize for the incomplete description; gcc -v gives:
>
> gcc version 3.4.5 20050706 (prerelease) (Debian 3.4.4-5)
>
>>> but when I tried gcc-4.0.1, 'make check' failed with a segmentation
>>> error after
>>
>> I have seen this on FC3. Note that ZAXPY is a BLAS routine, and you are
>> using ATLAS: was that also built with gcc-4.0.1?
>
> Probably not; I just 'apt-get install'ed atlas3-base etc. I'll build atlas
> from source and try again.
>
>> On FC3 I get the failure
>> if I use its BLAS, and not if I configure R with --without-blas. (I did
>> not see the problem using the gcc4 RPM from FC3 update, and I gather
>> others have managed to use gcc4 from FC4.)
>>
>> We do not recommend gcc-4.0.x as yet: we have yet to see any performance
>> gains (only losses) from it, and this is one of a number of known
>> nuisances. More serious are places where it apparently generates
>> incorrect code and so gives incorrect answers.
>
> OK. An additional advantage with gfortran(-4.x) will be the possibility to
> use fortan95 code, n'est ce pas?

Some Fortran95 code, anyway. (It is not yet a complete implementation.)

The problems we are seeing with gcc4 are mainly (but not entirely) with gfortran: it seems not quite ready for production use. -4.0.1 is already a considerable improvement over -4.0.0.

There is another project (www.g95.org) that is using a slightly different version of these sources, and either g95 or gfortran can be used with gcc3 if you want to use F95 sources. However, F95 is unlikely to be suitable for use in distributed R packages for a long time to come.

-- 
Brian D. Ripley,                  ripley@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

______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

Received on Tue Aug 02 20:04:41 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:39:39 EST