Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site watmum.UUCP
Path: utzoo!watmath!watnot!watmum!cdshaw
From: cdshaw@watmum.UUCP (Chris Shaw)
Newsgroups: net.arch
Subject: Re: risc, cisc, and microprogramming
Message-ID: <171@watmum.UUCP>
Date: Sun, 23-Jun-85 06:18:51 EDT
Article-I.D.: watmum.171
Posted: Sun Jun 23 06:18:51 1985
Date-Received: Mon, 24-Jun-85 02:36:24 EDT
References: <557@hou2b.UUCP> <1078@peora.UUCP> <334@spar.UUCP> <145@mips.UUCP> <350@spar.UUCP>
Reply-To: cdshaw@watmum.UUCP (Chris Shaw)
Organization: U of Waterloo, Ontario
Lines: 29

Does not the average RISC type machine do a purge-pipe action upon
receipt of a branch instruction, or is there something so incredibly obvious
going on here that I am missing?  The reason I ask is that in any but the most
nontrivial loops, throwing in enough NOPs to fill the pipe so that the next 
instruction is the right one strikes me as grossly inefficient.

The alternative with a simple "purge-pipe" is to repeat enough of the beginning
of the loop such that you end up doing the right instructions if you branched
to continue the loop. Otherwise you purge, and continue after the repeated 
instructions without actually executing any of them.

Thus we have a bogus program with a 2-element pipe:

start:	load  a
	store b
start+2:add   c
	store d
	decr  e
	brnz  start+2:
	load  a		\  do these 2 if branch taken, else throw them away
	store b		/   ... (purge) and do the xor
	xor   a     ---> continue here because "load" & "store" were purged
			 ... when the branch was not taken

(Forgive me if this was utterly stupid)

Chris Shaw    watmath!watmum!cdshaw  or  cdshaw@watmath
University of Waterloo
In doubt?  Eat hot high-speed death -- the experts' choice in gastric vileness !