On 2/21/2010 4:44 PM, Fuentes, Adolfo wrote:
]$ gcc -O3 -lm -march=nocona -o nbody.x nbody.c
]$ time ./-o nbody.x 50000000
Energy 0: -0.169075164
Energy 1: -0.169059907
Elapsed time: 1m 17.4s
]$ f95 -O3 -lm -march=nocona -o nbody.x nbody.f90
]$ time ./nbody.x 50000000
Energy 0: -0.169075164
Energy 1: -0.169059907
Elapsed time: 2m 31.7s
]$ g95 -O3 -lm -march=nocona -o nbody.x nbody.f90
]$ time ./nbody.x 50000000
Energy 0: -0.169075164
Energy 1: -0.169059907
Elapsed time: 1m 40.3s
gcc is "highly"* optimized, the g95 compiler would have similar optimizations, because they share back ends. The Intel compiler should beat it, however, if you are very familiar with Intel architecture, and are willing to learn the ins and outs. I'm not personally acquainted with other compilers, so can't answer questions about them.
(And my FORTRAN days are behind me. I can only answer in these general terms. I hope someone else can be more specific.)
Hope this helps, Mark Allums*It's been losing ground in some areas, last few releases. gcc has emitted-code-performance regressions due to tighter requirements with floating-point precision and its corresponding standards for some architectures; the g++ people are working-in new code for the upcoming c++ standard. It's always a work in progress. As long as gcc can compile and ld link working Linux kernels, though, most people will be happy.