Ultrix `fort' vs BSD `f77'

[kuo at skatter.UUCP]

In article <5110 at mimsy.UUCP>, chris at mimsy.UUCP (Chris Torek) writes:
> 
> In article <251 at skatter.UUCP> kuo at skatter.UUCP (Dr. Peter Kuo) writes:
> >.... We found that the new Ultrix FORTRAN
> >is at least 3-4 times faster than f77, even though a few bugs have been found
> >in fort.
> 
> Does that mean it *compiles* 3-4 times faster, or that the code *runs*
> 3-4 times faster after having been compiled?  And what code?
> 
> The 4.2BSD f77 compiler was pretty bad; the 4.3 compiler is much
> more correct, and the runtime libraries, now written in Vax assembly,
> are much faster, but I would be greatly surprised to find that even
> the ported VMS compiler could beat even the old 4.2 f77 by a factor
> of four, except on carefully selected benchmarks.
> -- 

I didn't time the compile time, but for sure the execution times for fort
are much improved over f77. The benchmarks we used included ALL the fortran
programs we use in the lab (many of them are ray-tracing type programs and
magnetic field calculation, etc). As one last, but well known test, I used
the infamous savage. The fort is faster than f77 in both single and double
precision modes by a factor of 3-4! This is done under Ultrix on a VAX785.

-------------------------------------------------------------------------------
Peter Kuo                    Bitnet (VMS address)    : KUO at SASK
Accelerator Laboratory              (UUCP address)   : "skatter!kuo at sask.uucp"
(a.k.a. The Beam Warehouse)
Univ. of Saskatchewan          uucp (unix address)   : !ihnp4!sask!skatter!kuo
Saskatoon, Saskatchewan             (Bitnet address) : !ihnp4!sask.BITNET!kuo
CANADA  S7N 0W0
Tel. (306) 966-6059

[Disclaimer: all the standard stuff.]