Fwd: Fortran compiler comparison

[François-Xavier Coudert]

Hi gfortranners,

John Appleyard announced today an update to the polyhedron webiste
(http://www.polyhedron.co.uk/) and a refreshing (begun and on-going)
of their compiler comparison pages. Here is a copy of the email I sent
him, both FYI and so that if you have similar suggestions [1], he does
not receive everything in multiple copies (I'm thinking about
-ftrace=full, e.g.)


Ciao,
FX


[1] maybe because of similar interests? ;-)



---------- Forwarded message ----------
From: François-Xavier Coudert <fxcoudert@gmail.com>
Date: Aug 13, 2007 8:48 PM
Subject: Fortran compiler comparison
To: john!news@polyhedron.com


Hi John,

First, congratulations for the new web site, which is really beautiful
(and removes a long-standing javascript error that seemed to upset my
browser)!

There is need for a minor correction to make on the "Intel/Linux
Fortran Compiler Diagnostic Capabilities" page: the options passed to
gfortran shouldn't include "-ftrace=full", as there is no such option
for gfortran. You could use, though (and may have used, I don't know)
the "-g -fbacktrace" combination, which provides runtime backtraces of
both library and user-code errors (similarly the -traceback option of
the Intel compiler). This would probably allow you to put a "Yes" for
gfortran in the last row of the array for diagnostics capabilities:
"Full trace-back from run-time errors".

Since you appear to be thinking of updating other parts of the
compiler comparison, I have a few suggestions to make:

  * The Sun compiler for Linux is really worth including in the
comparison. On a series of run (on the Polyhedron 2005 benchmark
suite; I hope I followed the terms of use satisfactorily) benchmark 6
months ago (results here:
http://www.eleves.ens.fr/home/coudert/benchmark.pdf ; detailed
explanations here:
http://gcc.gnu.org/ml/fortran/2007-02/msg00151.html) it really was a
serious candidate. While it still had a few weaknesses (maybe fixed by
now?) it beat the Intel 9.1 compiler on a fair number of cases.

  * I see gfortran is not included in the Windows benchmarks. It
should be easy to install and run (an installer is available from
http://gcc.gnu.org/wiki/GFortranBinaries ; direct link:
http://quatramaran.ens.fr/~coudert/gfortran/gfortran-windows.exe), but
if you experience the least trouble doing so, I'd be happy to hear
about it and both a) help you and b) fix the problem for future users.

  * Since your position is (as far as I could see) to include stable
versions of compiler, you could update the g95 compiler to its stable
version, v0.9. Downloads are available from http://ftp.g95.org/#V0.9


Well, I thank you again for the service to the community, and wish you the best.

François-Xavier Coudert