distfile for lang/ifc

[M. L. Dodson]

On Saturday 11 December 2004 07:14 am, Alexander Leidinger wrote:
> On Fri, 10 Dec 2004 08:46:25 -0600
>
> "M. L. Dodson" <bdodson at scms.utmb.EDU> wrote:
> > > > just copied the port, changed the distfile specs in the Makefile
> > > > to point to the one I had, did a "make makesum", and it compiled
> > > > and worked with no problem (this was with fortran, version 7, I
> > > > think).
> > >
> > > This doesn't work with the update from 8.0 to 8.1 (at least this
> > > wasn't the case for icc, and ifc seems to be similar to icc).
> >
> > Well that's good and bad.  I use ifc to compile a molecular
> > mechanics suite called AMBER (not freely available).  A F90 or
> > later compiler is required, so g77 won't suffice.  The AMBER
> > mailing list has documented that only certain versions of ifc v8
> > will (1) compile the key AMBER program, and (2) generate code
> > which can reproduce known good trajectories (generated by earlier
> > AMBER code compiled with many different compilers, not just g77).
> > These new differences from one minor version to another increase
> > the work necessary to work around the Intel compilers
> > "peculiarities".  On the up side, WHEN THEY WORK, the Intel
> > compilers produce easily demonstrably faster code.
>
> From v8.0 to v8.1 it isn't a minor upgrade like the 8.0.x versions... at
> least not when looking at the "internals" of the compiler (I don't have
> access to the source, this is purely based upon my experience with the
> update of lang/icc, and lang/ifc is based upon the icc port). The
> installation procedure changed a lot and some internals (compiler/libs)
> changed too. Since we don't use Intels installation procedure and we
> need to "fixup" some parts to be able to produce native FreeBSD
> binaries, this was a major update from the FreeBSD port perspective.
>
> Bye,
> Alexander.

OK, that makes sense.  BTW, I certainly did not mean to imply that
you or the maintainer were part of the problem.  I was just trying
to issue caveats to people about using the compilers.  I assume
many people that qualify for the noncommercial versions are using
them in an academic environment to do serious (physics, chemistry,
engineering, ...) research.  I do not think the AMBER guys know
what the problem with some versions of the compiler is, but they
know when a compile will not complete or the benchmark gives the
wrong answers.  They have reported their results back to Intel,
but have on some occasions, but not all, apparently been ignored
or told that no fix will be forthcoming.  If it is really
important to you that the compiler "get it right", you need to
have your own validation suite, IMO.

Bud Dodson
-- 
M. L. Dodson                                bdodson at scms.utmb.edu
409-772-2178                                FAX: 409-772-1790