Path: utzoo!utgpu!news-server.csri.toronto.edu!helios.physics.utoronto.ca!ria!uwovax.uwo.ca!2011_552
From: 2011_552@uwovax.uwo.ca (Terry Gaetz (UWO Astronomy))
Newsgroups: comp.lang.fortran
Subject: Re: Running long memory intensive programs!
Message-ID: <1991Jun14.123326.9406@uwovax.uwo.ca>
Date: 14 Jun 91 17:33:26 GMT
References: <1991Jun12.203754.9356@noose.ecn.purdue.edu> <1991Jun13.160602.23726@afit.af.mil> <12844@pucc.Princeton.EDU>
Lines: 52

In article <12844@pucc.Princeton.EDU>, BVAUGHAN@pucc.Princeton.EDU (Barbara Vaughan) writes:
[...] 
> I asked several months ago about Fortran compilers for 386 machines
> and received many replies from happy users of Watcom's compiler.
> I also received a copy of a review published in the April '90 issue
> of Computer Language which mentions four compilers that can run in
> protected mode: Lahey F77LEM/32, OTG FTN77, Microway NDP and SVS F77.
> The latter two compilers got very poor reviews; Watcom was not reviewed
> as it was still in beta test when the review was published.
> All of these can also use virtual memory (some require Phar Lap Tools,
> Lahey requires separate purchase of its own Ergo OS/386.)
> I will send copies
> of these letters and reviews on request.  I must warn that I found
> some inaccuracies and cannot be sure of the correctness of some of
> the letters and comments.  But that's the weakness of the net, as this
> posting shows.

I was singularly unimpressed with the _Computer_Language_ review.  The 
reviewer seemed to be mainly concerned with whether the Fortran could be
made into Turbo Pascal and totally unconcerned with quality/efficiency 
issues important to people interested in large programs with long runtimes.
None of the 'benchmarks' used were standard, so it was impossible to verify 
his results.  (His comparison of runtimes for Microsoft Fortran vs. NDP 
Fortran-386 gave results almost diametrically opposed to my own 
experiences.)  The review included none of the commonly used benchmarks
(LINPACK, Livermore Loops).  In his testing of protected-mode 386 
compilers, the reviewer didn't bother to try out programs which were large 
enough to require more than 640K; why pay more for a protected mode 
compiler if you aren't going to take advantage of its features?  As a 
review of Fortran compilers, the _Computer_Language_ review should be 
taken with a large block of salt.

Somewhat better are Al Cameron's reviews in _MIPS_ (now 
_Personal_Workstation_).  Cameron is a working scientist and at least 
tested the compilers with more realistic code (including, in some cases,
the SPH simulation code he uses in his research).  The main problem I had
with Cameron's reviews was that he didn't pay any attention to error 
trapping.  I tried out SVS Fortran after reading his review; I returned it 
as unacceptable.  Cameron's review had suggested that SVS could be up to
a factor of two faster than NDP; the problem with the NDP compiler had been
fixed in the meantime so that SVS runtimes were only a few percent faster 
than NDP.  More importantly, SVS had no way to handle exceptions (at 
least at that time).  I tried calculating 2./0. and got an answer 
of 0. with _no_ sign that anything was amiss.  Similarly, sqrt(-3.) 
returned -3., with no hint that there was a problem.  I just about 
gagged when I saw this.

I have been using MicroWay's NDP Fortran since about 1987.  The recent 
versions of NDP Fortran allow the user to install an error handler or to
modify the 387 interrupt flags; debugging has become much easier.
-- 
Terry Gaetz         --   gaetz@uwovax.uwo.ca