landers@mabel.uucp (05/04/91)
I've been using the Green Hills FORTRAN and C compilers on the Intergraph
workstations for a number of years. In fact, I'm currently developing and
selling a number of products which have been compiled with the Green Hills
compilers. I can sympathize with your friend!
Compared to FORTRAN compilers on other Unix platforms, the Green Hills
FORTRAN compiler isn't very good. It is slow, produces poor (but mostly
correct!) code and generates huge executable files. To make matter
worse, Green Hills won't help you if you call the phone number on the
inside cover of the manual. Instead, they insist that all problems with
their compilers must be logged with Intergraph via Integraph's 800 number.
To be fair, the FORTRAN standard doesn't say much about the way objects
are stored in memory, how big executable files will be or what size of the
largest source module should be. Although it's been painful, I've been
successful with it.
One problem with the FORTRAN compiler is that Intergraph doesn't use
FORTRAN very much for software development of their workstation products
-- they use C and C++ instead. The Green Hills C product is a little
better, although it's not an ANSI compiler.
A bigger problem with their products, I believe, is that they're basically
the same compilers they've been using since the mid-80's that've been
tweaked a bit over the years. They're using pretty much the same
optimizations and code generation even though the Clipper RISC
architecture is now 10 to 20 times faster than it was when they started
developing the product. Fetches and stores that weren't a problem at 5
mips can be real bottlenecks at 50 mips.
How do compilers support dramatic speed changes even when the underlying
RISC instruction set remains nearly the same? Do you re-write the back
end every time the clock rate gets cranked up? How do developers, say
SPARC compiler developers, manage the change?
BTW, I've read that Intergraph's latest software release contains a new
ANSI C compiler from a different vendor.
Joe Landers
TechniCon Computer Services
444 Spear St., Suite 213 phone: (415)-896-6313
San Francisco, CA. 94105 uucp: ..!uunet!mabel!landers
--
Send compilers articles to compilers@iecc.cambridge.ma.us or
{ima | spdcc | world}!iecc!compilers. Meta-mail to compilers-request.jpff@maths.bath.ac.uk (John ffitch) (05/06/91)
I gave up using the Greenhills C compiler on my Clipper-based machine.
I use the Norcrofy C compiler, which is full ANSI, and the same as
runs on ARM, IBM mainframes, Transputers and other machines. The
Fortran seems to work OK but I have not given it a major work-out.
The C is faster to compile and produces better code. It is hard to be
certain about the Fortran...
==John
I declare an interest; these compilers are developed by Codemist Ltd,
a company of which I am a director. The majority of the systems were
written by Arthur Norman and Alan Mycroft.
--
Send compilers articles to compilers@iecc.cambridge.ma.us or
{ima | spdcc | world}!iecc!compilers. Meta-mail to compilers-request.