Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!mhuxt!houxm!whuxl!whuxlm!akgua!gatech!seismo!lll-crg!lll-lcc!qantel!hplabs!oliveb!glacier!mips!mash
From: mash@mips.UUCP
Newsgroups: net.arch
Subject: Re: Addressing modes
Message-ID: <385@mips.UUCP>
Date: Sat, 8-Mar-86 13:37:43 EST
Article-I.D.: mips.385
Posted: Sat Mar  8 13:37:43 1986
Date-Received: Tue, 11-Mar-86 01:24:34 EST
References: <946@garfield.UUCP> <2900001@ztivax.UUCP>
Organization: MIPS Computer Systems, Sunnyvale, CA
Lines: 38

David Smyth writes:
> However, cliff, a good question you have implied:  Does anyone do assembler 
> programming on RISC machines?  
> 
> For example: Is UNIX on RISC entirely in C?  How do 
> the stack recovery mechanisms work?  Interrupted / failed instructions?  
> CPU mode register control manipulation (if applicable)?

1) Does anyone do assembler on RISC machines?
	Yes.  For what it's worth, a well-designed RISC is relatively
	easy to program in assembler when that makes sense.  After all,
	in those cases, what you're doing is writing microcode (more vertical
	than horizontal, but with some idea of overlap).
2) Is UNIX on RISC entirely in C?
	No.  On almost any machine, there are bound to be instructions that
	are hard to get to from C.  ALso, there are a few time-critical
	pieces of code where the programmer knows attributes of the problem
	that no compiler can reasonably deduce.  Example: for string & block
	copy loops, there is usually a tradeoff between a faster loop with
	longer setup time, and a slower loop that gets started quicker.
	It's very hard for a compiler to guess the distribution.
3) Thus, there is still room for selected assembler code.  However, from
experience with assembler (on S/360s, PDP-11s, VAXen, 68K's, and our R2000),
in most cases I'd rather have fast instructions that are easy to use and
give me more access to the machine's pipeline, than most complex ones.
Most memory-intensive multi-cycle instructions (like string instructions)
run just as fast when hand-coded, because they're memory limited anyway;
a lot of them don't really match what you need in some cases: for example,
look at all of the mem* and str* routines in SYstem V, and see how well
the string operations of most machines do them.  About the only win for
complex instructions is when they support a datatype (like floating point)
that is both frequently used in the application at hand, and difficult
to simulate cheaply.
-- 
-john mashey
UUCP: 	{decvax,ucbvax,ihnp4}!decwrl!mips!mash
DDD:  	408-720-1700
USPS: 	MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086