Re: compiler discussion

On Thu, Nov 12, 2009 at 02:15:19PM +0000, Anton Shterenlikht wrote:
> Following from the discussion on the system compiler for ia64,
> I tried to list few major ports which I'd love to have on ia64,
> but can't, because of GCC problems:
> 
> - math/blas, lapack, lapack95, arpack, scalapack, atlas, etc.
> - science/hdf5-18 (fortran APIs can't be built)
> - french/aster (industial quality FEA code)
> - cad/calculix (another good FEA code)
> 
> All these depend on lang/gcc44, which doesn't build.

Doesn't build is not a very good description if you're
looking for help.  Post the build log somewhere.
Have you submitted bug reports to gcc.gnu.org?  Bugs
that are unreported are unlikely to be fixed.

> The only fortran compiler I know to build and work
> successfully on ia64 is (correct me if I'm wrong) g95.
> 
> I wonder if it's possible/desirable/easy to use lang/g95 for
> the above and other fortran-dependent ports? 

Given Polyhedron Benchmarks, it may be preferable to determine
why you can't build gcc44 and fix that problem.*

> In principal, lang/g95 looks very good, and it's got
> some features not available in gfortran, e.g. limited
> support for 2003 standard.
> 
> Any comments?

AFAIK, g95 has TR 15580 implemented and gfortran doesn't.
Other than that feature, gfortran has a fairly long list
of Fortran 2003 features implemented, which you can find
partially enumerated at the gfortran wiki.

> Also, any comments on the usability (particularly for fortran)
> of llvm and lang/llvm-gcc4 on ia64?

Last time I checked, Fortran in llvm was based off a very old
gfortran.  The llvm website mentions gcc 4.2.?.  While the 4.2.?
gfortran isn't too bad, you most certainly would rather use gcc44
if you can.  Literally, hundreds of bugs and several new feature
have been add to gfortran in going from 4.2.? to gcc 4.4.2.

Have you checked the Open64 project?

*disclaimer:  I've contributed a few hundred patches to gfortran,
and I'm listed as a gfortran maintainer.
 
-- 
Steve