Subject: Re: FPU Performance on DECstation 5000/125
To: Simon Burge <firstname.lastname@example.org>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 08/27/1997 15:39:36
Simon Burge writes:
>Same goes for a /260 under Ultrix and NetBSD 1.2G. Thinking I'd test
>another FP intensive program (one that tests primality of 1,000,000+
>digit numbers) I got a surprising result - it didn't even run!
>vlad:~ 42> ~/bin/ultrix.mips/mersenne1 1279
>/home/simonb/bin/ultrix.mips/mersenne1: 1: Syntax error: "&" unexpected
>vlad:~ 43> file ~/bin/ultrix.mips/mersenne1
>/home/simonb/bin/ultrix.mips/mersenne1: basic-16 executable not stripped
>This program was compiled under Ultrix using the MIPS compiler with the
>-mips3 option. So I guess the ultrix emulation isn't perfect. Is this
>something I should pr?
Yes, but please send it as a software-change request :).
The kernel context-switch code doesn't do the 64-bit register saves
and restores, so we don't support full mips3 binaries yet -- even with
a 32-bit address space. The same goes for using 32 64-bit FP registers.
(Using the FP registers is an easy kernel change, but the interface to
debuggers &c needs more thought.)
So not executing ECOFF binaries with a mips3 type field is probably
the correct thing to do, for now. I don't know if we want to
short-circuit the exec() and force fail there, or continue.