Path: chuka.playstation.co.uk!chuka.playstation.co.uk!news
From: gil@snsys.com (Gil Jaysmith)
Newsgroups: scee.yaroze.programming.gnu_compiler
Subject: Re: MIPS assembly queries..
Date: Thu, 02 Oct 1997 11:49:25 GMT
Organization: SN Systems
Lines: 29
Message-ID: <343387b1.2092228@news.playstation.co.uk>
References: <3433834F.3248@127.0.0.1>
Reply-To: gil@snsys.com
NNTP-Posting-Host: gil.snsys.com
X-Newsreader: Forte Free Agent 1.11/32.235

Wotcher, Majik...

>for instance the sequence
>	subu $9,$9,1
>	bne  $9,$0,end
>	li   $8,32
>will always load register 8 with 32 whether the branch is taken or not
>or at least this is my finding.. am I wrong, btw I know that preceding
>the 'li' with a nop cures the problem...

Yes, this is correct. Pipelining means that the insn slot after a jump
is scheduled before the jump takes place - in practice, this means
that accidentally placing two routines next to one another without
examining the implications of delay slots may result in strange
effects. You'll certainly get something horrible, probably an
exception, if you write things like two consecutive jumps.

If you want to know more about delay slots and suchlike, pop over to
the MIPS website at www.mips.com and look for the PDFs documenting the
MIPS insn sets. Alternatively I've got the majority of them, should
you want them. I don't have an R3000-specific one, though, but then
the R4600 will do just as well since it supports the R3000 insn set
(and it'll also prepare you for hand-assembly on the Nintendo 64 and
many arcade machines - LOLOL!)


Gil Jaysmith
SN Systems Software Ltd (http://www.snsys.com)
Disclaimer: the ads are my employers' but the opinions are mine.