It also has documentation!.
Ian
CLAS library wrote:
> On Linux we use the GNU C and FORTRAN compilers (gcc and g77
> respectively). These compilers work fine but g77 has some
> limitations:
>
> a) No access to variables contained in common block from the debugger
> (actually no "easy" access, there is a way that I have yet to
> make work....but I'll let you know when I do)
> b) Lack of support for many VAX extensions, like structures (shudder)
> c) Slow progress, the present version of g77 is 0.5.21 Yes it hasn't
> even made it to beta yet and we are using it for analysis (well at
> least some of us are).
>
> The main problem I see is item a) which will really hinder debugging
> efforts on Linux boxes.
>
> There are several commercial compilers (C and FORTRAN) for Linux and I
> was interested in trying them to see if they address any of the issues
> and I just wanted to see how efficient g77 compiled code is.
>
> I contacted the Portland Group (PGI) for a test suite (actually I was
> attempting to
> purchase a license but the price overran the JLAB credit card, not
> that it is expensive just that we have a low limit on credit card
> purchases) so instead they offered a two week trial.
>
> The installation was a piece of cake and PGI is to be commended for
> making a product that works with and doesn't break existing compilers.
>
> I trial run on a simple prime number generator showed that the PGI
> compiled code was actually a bit slower than g77. I then linked
> the CLAS
> reconstruction code with PGI compilers. The fortran code linked without a
> hitch. The C code barfed and I could not decode the warning to figure
> out which PGI flag to throw. So I returned to gcc for the C code.
>
> On 5000 REAL data events the reconstruction code compiled with the PGI
> compiler was slightly faster than g77
>
> O-flag PGI g77
> -O2 16.149 18.246
> -O3 16.109 18.289 millisec/event
>
> To test item a) I compiled the prime number code with the -g flag and
> tried ddd/gdb on the executible....this failed. One of PGI products
> is a debugger but this was not part of the test suite....
>
> Summary, the PGI FORTRAN compiler offers some advantage over g77, for
> example: Professional documentation, FORTRAN 90 support plus many
> VAX extenstions. Unfortunately the one area they could really blow
> away g77, professional debugger, was untestable. The resulting
> executibles are ~10% faster than those compiled with g77.
>
> Just thought I'd let you all know.....
>
> --