Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!thunder.mcrcim.mcgill.edu!snorkelwacker.mit.edu!apple!agate!dog.ee.lbl.gov!elf.ee.lbl.gov!torek
From: torek@elf.ee.lbl.gov (Chris Torek)
Newsgroups: comp.lang.c
Subject: Re: micro-optimizing loops (was Help with casts)
Message-ID: <10248@dog.ee.lbl.gov>
Date: 25 Feb 91 19:34:23 GMT
References: <339@smds.UUCP> <414@ceco.ceco.com> <10191@dog.ee.lbl.gov> <15396:Feb2419:32:5891@kramden.acf.nyu.edu>
Reply-To: torek@elf.ee.lbl.gov (Chris Torek)
Organization: Lawrence Berkeley Laboratory, Berkeley
Lines: 33
X-Local-Date: Mon, 25 Feb 91 11:34:23 PST

In article <15396:Feb2419:32:5891@kramden.acf.nyu.edu>
brnstnd@kramden.acf.nyu.edu (Dan Bernstein) writes:
>What Chris doesn't point out is that
>
>  i = blah;
>  do
>   {
>    ...
>   }
>  while (--i);
>
>will compile to the optimal code on most machines under most compilers ...

This is, of course, because you have manually moved the loop test to
the bottom.  This almost invariably saves one `jump' instruction, if
nothing else.

>... whether they have exposed jump delays or not ...

Actually,

	do { ... } while (i--);

is a bit easier on some exposed pipeline machines, simply because it
compiles directly to

	jne	r1, r0, loop	# branch if i != 0
	dec	r1		# but do i-- first

without any thought whatsoever.
-- 
In-Real-Life: Chris Torek, Lawrence Berkeley Lab EE div (+1 415 486 5427)
Berkeley, CA		Domain:	torek@ee.lbl.gov