gfortran speed benchmarks

[Dominique Dhumieres]

> ...
> In the meantime, my inline dot_product intrinsic is nearly ready to submit.
> ...
> the effect is spectacular for induct.f90 and satisfactory for kepler.f90.

I have seen a similar improvement with g95 when the dot_product was inlined.
Note that I have a modified version of induct.f90 that replaces three dot products
by a 3x3 matrix-vector product. On a G5 the result is catastrophic for all
compilers (including xlf).

Dominique