Subject: Re: Floating point emulation
To: None <current-users@NetBSD.org>
From: Matthias Scheler <firstname.lastname@example.org>
Date: 11/11/2005 03:36:00
In article <200511060440.jA64eAsa017188@guild.plethora.net>,
email@example.com (Peter Seebach) writes:
> On some BSD somewhere, I was once told to use floating point emulation in
> the kernel so that the kernel wouldn't have to save and restore floating
> point registers as often when processing system calls.
I remember that story differently:
From time to time NetBSD-amiga users (and probably users of other m68k
ports) discover that the kernel is compiled with "-msoft-float". In an
attempt to optimize the system performance they change that option to
use "-m68881" because their 680x0 CPU has an FPU. And that is the
occassion where they get told that "-msoft-float" is used to make
sure that any attempt to use FP math in the kernel will create a
link failure instead of building a broken kernel.
> The MATH_EMULATE stuff in options(4) seems to imply that this option is
> useless unless your hardware lacks hardware floating point entirely.
Matthias Scheler http://scheler.de/~matthias/