Re: SUMMARY: pentium optimized debian discussion
On Jul 29, Raul Miller wrote:
> Chris Lawrence <quango@ix.netcom.com> wrote:
> > It might help somewhat with people who don't have FPUs (fairly common
> > on m68k and perhaps other architectures... 386/SX and others?),
> > when the kernel FPU emulator gets straightened out. A m68k-nofpu
> > subarchitecture (with a few things like glibc and FPU-intensive
> > apps in it) would probably improve performance on non-FPU machines
> > dramatically (avoiding the unimplemented FPU instruction traps into
> > kernel space).
>
> This might be a bit premature -- or have the kernel and glibc folks
> given up on trying to arrange it so that the fpu wouldn't have to be
> initialized until it's needed?
At least on 680x0s, there's a dramatic performance penalty when it kicks
into the FP emulation code... the penalty may not be as severe on x86. It's
not an initialization problem, it's a complete lack of FPU problem
(particularly for math intensive apps... which admittedly should be run on
FPU machines, but would prefer to run as integer code in user space than
pretending to run floating code in user space and having the kernel emulate
every FPU instruction). Glibc has most of the higher-level math functions
used by C code and is the natural package for which you'd want a non-FPU
version available.
Chris
--
=============================================================================
| Chris Lawrence | Want to know what's really happening? |
| <quango@ix.netcom.com> | http://www.memphiswatch.org/ |
| | |
| Contract Programmer | Watch Babylon 5 on TNT |
| FedEx Ops Research | <*> http://tnt.turner.com/babylon5/ <*> |
=============================================================================
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: