Re: CCL:Re: fortran Compiler for Linux Box. Options?



 "William T. Rankin" writes:
 >> I am willing to go freeware, but I would spend the money on
 >> a good Fortran compiler to gain the performance. (I also believe
 >> there is no freeware fortran compiler for f90 code!)
 >
 >I picked up some information form The Portland Group, who markets
 >auto-parallelizing FORTRAN compilers for Linux.  You may want to
 >look them up at http://www.pgroup.com/
 >
 >Now having said that, I will point out that even with the best
 >tools, simply running a compiler over existing serial FORTRAN
 >codes will most likely not produce highly scalable parallel code
 >without some sort of programmer direction from within the source.
 Hi-
 I've used both g77 the Portland Group compiler for AMBER 5.0 both with and
 without MPI (message passing interface).  I've had excellent results using
 Portland Group's compiler.  I didn't try the auto-parallelization, but I
 did compile MPI with PG's C compiler and AMBER 5.0 with PG's f77 compiler.
 Here are the results I got (times are in seconds) on my dual PPro 200MHz:
 g77 opt -O6, ewald test:
 496
 g77 MPI opt -O6, dual proc, ewald test:
 332
 pgcc full opt, ewald test:
 361
 pgcc MPI full opt, dual proc, ewald test:
 253
 So, dual processors with g77 is faster than one with PGF77, but
 PGF77 with dual processors is definitely the best.   Paralogic
 is going to let me test on their 4 processor platforms-
 if this code scales well on intels, it'd definitely be a much
 cheaper way to get the performance we need.
 Dave
 -----------------------------------------------------------------------------
 Email: dek (- at -) cgl.ucsf.edu   David Konerding   WWW: http://picasso.ucsf.edu/~dek
 -----------------------------------------------------------------------------