Path: utzoo!news-server.csri.toronto.edu!cs.utexas.edu!usc!elroy.jpl.nasa.gov!decwrl!deccrl!news.crl.dec.com!jg
From: jg@crl.dec.com (Jim Gettys)
Newsgroups: comp.unix.ultrix
Subject: Re: New RISC compilers????
Message-ID: <1991Mar15.162559.3386@crl.dec.com>
Date: 15 Mar 91 16:25:59 GMT
References: <15530012@hpdmd48.boi.hp.com> <1991Mar13.031200.29836@crl.dec.com> <19805@cbmvax.commodore.com> <1991Mar13.125214.7403@crl.dec.com> <19831@cbmvax.commodore.com>
Sender: news@crl.dec.com (USENET News System)
Organization: DEC Cambridge Research Lab
Lines: 60

Hopefully this will clear things up about the new compilers.  Thanks to Kent
Glossop of the GEM group who sent this to me for redistribution.
			- Jim Gettys


----------------------------------------
DEC FORTRAN & DEC C

This is the situation with the 2 new compilers:

  Previous numbers:
    
    FORTRAN:	MIPS f77 front end
		MIPS back end
		called "FORTRAN for RISC V2.n"

    C:		MIPS C front end
		MIPS back end
		called "C for RISC" or something like that

  New numbers:

    FORTRAN:	DEC FORTRAN front end (heavily modified VAX FORTRAN source
		    to have both good compatibility with f77 compilers and
		    with VAX FORTRAN using "-vms")
		GEM (DEC developed) back end
		called "DEC FORTRAN V3.0"

    C:		New DEC-developed ANSI-C compliant "DEC C" front end, which
		    has K&R and PCC modes
		Currently uses the MIPS back end
		called "DEC C V1.0"

In general, "user-visible" properties (source language, etc.) are the
property of the "front end", which generates an intermediate language.
The optimization is performed by the back end.

What happened is:

    - The GEM back end does quite a bit better than the MIPSCO back end
      for a lot of FORTRAN code.  In the case of the SPEC benchmarks,
      this is something like 14% on a DS5000.  Reports on application
      speed ups are generally in the 0-25% range, though there have been
      a few more spectacular ones.)

    - The current DEC C compiler uses the MIPSCO back end, which means that
      there isn't a significant difference in code quality between that
      compiler and the previous C compiler (though they use different
      versions of the "MIPSCO back end".)

So, the summary is that the new FORTRAN compiler represents a significant
change in optimization and code generation technology, while the C compiler
is the same basic technology as the previous version.  (The fact that the
benchmarks happened to get a little slower with DEC C may be due to the
slightly different IL it presumably generates compared to the old C front
end.  It might also be due to the interaction of some new optimizations,
which sometimes happens with new releases of compilers.)

Kent Glossop
(a member of the GEM team...)