gfortran speed benchmarks

[THOMAS Paul Richard 169137]

>> ... I notice that they have updated the suite. ... 
>
> There is a new test aermod.f90. When compiled with gfortran, it gives the
> "nice" output:
>
....snip.....
>
> It seems that some "read"s miss some keywords, but I did not have the time
> to reduce it to a simple test case. Does it look like a known bug?

Jerry DeLisle and I have been in touch about this over the weekend.  I
isolated
the problem, which is associated with tab formatting during reads.  Jerry
filled
out a bug report(PR24669) and, when last heard of, said that he had a
candidate
patch. It is a bit nasty because the bug is associated with what happens
when
EOR is detected before a tab. Not only is the record updated but the various
counters that track the current position and the maximum righthand position
are
reset too.

Jerry has had aermod.f90 compile and run, so I think that we can look
forward to
a clean sweep on Polyhedron '05 very soon.

In the meantime, my inline dot_product intrinsic is nearly ready to submit.
As
you will note from the updated Wiki http://gcc.gnu.org/wiki/GFortranResults
,
the effect is spectacular for induct.f90 and satisfactory for kepler.f90.

Best regards

Paul