Re: Fortran and subdir-objects

[Davyd Madeley]

On Mon, Jan 16, 2006 at 04:33:48PM +0100, Stepan Kasal wrote:

> Currently, _Autoconf_ contains macros to test "-c -o" for C, Fortran,
> and Fortran77 languages.  _Automake_ contains a wrapper for the C macro,
> but not for the latter two ones.
> But the latter two macros have the following comment attached:
> # The usefulness of this macro is questionable, as I can't really see
> # why anyone would use it.  The only reason I include it is for
> # completeness, since a similar test exists for the C compiler.
> 
> My impression is that we should remove the artificial macros and write
> new ones if/when there is a real need.

I too had wondered if these were truly helpful.

> So the questions are:
> Are there any Fortran* compilers which don't understand "-c -o"? 
> Are there ones which do understand "-c" yet don't understand "-c -o"?

I can check this on the Pathscale compiler and the Intel compiler.

> Which Fortran* compilers should your project support?  Would it be bad
> if projects using Automake required GNU Compiler for Fortran?

Unfortunately, GNU Fortran is not up the same quality as GCC. In
v4.0, there only exists a Fortran 90/95 compiler (gfortran) which
has some things unimplemented. If you need to compile Fortran 77,
g77 has not been ported 4.0, so you lose out.

Fortran seems to be highly optimisable in the compiler and indeed
our Fortran programmers seem to rely on that. From our crude
benchmarking, we have measured speedups of factor 2 and greater
using the Intel Fortran compiler over gfortran. In the end, this
translates to about a week of time saved on processing a job (we're
doing seismic analysis, for some idea of how much CPU we're using
for a job, 90-odd CPUs across 45 blades are running one of these
jobs for a week).

--d

-- 
Davyd Madeley

http://www.davyd.id.au/
08B0 341A 0B9B 08BB 2118  C060 2EDD BB4F 5191 6CDA