Path: utzoo!telly!ddsw1!lll-winken!maddog!brooks
From: brooks@maddog.llnl.gov (Eugene Brooks)
Newsgroups: gnu.gcc
Subject: Re: Gcc on Sequent Balance
Message-ID: <14344@lll-winken.llnl.gov>
Date: 4 Dec 88 07:33:48 GMT
References: <650@ecrcvax.UUCP>
Sender: usenet@lll-winken.llnl.gov
Reply-To: brooks@maddog.UUCP (Eugene Brooks)
Organization: Lawrence Livermore National Laboratory
Lines: 24

In article <650@ecrcvax.UUCP> chassin@ecrcvax.UUCP (J.C de Kergommeaux) writes:
>
>I'd like to hear from people using to compile parallel programs on the Sequent
>Balance 8000.

We use GCC routinely for parallel programs on the Sequent Balance.  GCC produces
much faster code, typical improvements of 30% or more, on computationally
intensive problems as compared to DYNIX CC, SUN CC, or VAX CC. You tend to see
a big improvement in codes where many common subexpressions are to be found.

In order to use GCC on our parallel programs, we added support for the -mshared
flag which causes all static memory to be in shared store.  Our parallel
programming model creates its own private memory using arrays in shared store
so we did not have to add full support for the "shared" and "private" keywords
which appear in DYNIX C.  We have no problem with optimizing parallel programs
although one sometimes finds GCC packing a variable you are busy waiting on
in a register and polling it there, a problem easily solved with the volatile
keyword.  We use our own locally developed Parallel C Preprocessor for parallel
programs, it supports an effient portable environment which also runs on
several other machines.

Anyone interested in the patches to GCC for parallel support on the Balance
or the Parallel C Preprocessor (PCP) only need give me a buzz.

brooks@maddog.uucp, brooks@maddog.llnl.gov