Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!stanford.edu!leland.Stanford.EDU!elaine18.Stanford.EDU!dhinds
From: dhinds@elaine18.Stanford.EDU (David Hinds)
Newsgroups: comp.sys.sgi
Subject: Re: GNU c++ bashing (was Re: c++ vs ada results)
Message-ID: <1991Jun16.061857.2271@leland.Stanford.EDU>
Date: 16 Jun 91 06:18:57 GMT
References: <1991Jun12.164741.412@news.larc.nasa.gov> <1991Jun12.201740.16463@netcom.COM> <32088@dime.cs.umass.edu>
Sender: news@leland.Stanford.EDU (Mr News)
Organization: Stanford University - AIR
Lines: 19

In article <32088@dime.cs.umass.edu> eli@smectos.CS.UMASS.EDU (Eli Brandt) writes:
>GCC sidenote:  It appears that gcc -O generates division code which is about
>six times faster than SGI's cc -O on a 4D.  Multiplication is "only" 70%
>faster.  Has anybody else seen numbers like these?  A factor of six seems
>just a bit odd.

What code, exactly, shows this big difference?  It is not believable that
two reasonable decent compilers should show a large difference in generated
code to do a single primitive operation.  Does the difference persist at
-O3 with the MIPS compiler?  It is not completely fair to compare GCC full
optimization with the first level of MIPS optimization.  I was especially
surprised to see this claim, because in my experience, GCC -O3 code is
15-25% slower than than MIPS cc -float -O3 code on my applications.  I
occasionally use GCC to help trace down bugs, because I can get more
warning messages, but I always use the MIPS compiler for compute-intensive
code.  

 -David Hinds
  dhinds@cb-iris.stanford.edu