From owner-freebsd-hackers  Sat Aug 24 01:29:44 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id BAA18140
          for hackers-outgoing; Sat, 24 Aug 1996 01:29:44 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id BAA18131
          for <freebsd-hackers@freefall.freebsd.org>; Sat, 24 Aug 1996 01:29:38 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id SAA20590; Sat, 24 Aug 1996 18:22:55 +1000
Date: Sat, 24 Aug 1996 18:22:55 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199608240822.SAA20590@godzilla.zeta.org.au>
To: bde@zeta.org.au, gclarkii@main.gbdata.com
Subject: Re: max math performance - how?
Cc: freebsd-hackers@freefall.freebsd.org, kuku@gilberto.physik.rwth-aachen.de
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>Maybe we should go back to libm as the default and tell people that if they
>have either the GPL'ed MATH_EMU  --OR-- a real math-co then they should 
>should compile libmsun with HAVE_FPU.

No, msun is slower than libm (in the !HAVE_FPU case) because it is more
correct.  The only reason to keep libm is that msun doesn't work on
machines that don't have IEEE floating point.  I was wrong saying that
msun is slower.  Many parts are faster because they use optimizations
that depend on IEEE format.  I guess only things like sin() for large
args are much slower (because they are more accurate).

Users shouldn't have to manage libraries.

>This brings up another joint subject.  Does GCC still use a dumb version
>of some math routines as default or has this changed?

This changed in gcc-2.6.3 or earlier a long time ago.

Bruce