Path: utzoo!attcan!uunet!convex!killer!ames!mailrus!bbn!rochester!pt.cs.cmu.edu!andrew.cmu.edu!jk3k+
From: jk3k+@andrew.cmu.edu (Joe Keane)
Newsgroups: comp.arch
Subject: Re: A simple question on RISC
Message-ID: <oXV-=Wy00V4109jmdK@andrew.cmu.edu>
Date: 18 Nov 88 22:23:30 GMT
References: <6544@xanth.cs.odu.edu> <75577@sun.uucp> <1618@imagine.PAWL.RPI.EDU> <419@augean.OZ> <392@ksr.UUCP>,
	<7723@aw.sei.cmu.edu>
Organization: Carnegie Mellon
Lines: 41
In-Reply-To: <7723@aw.sei.cmu.edu>

Robert Firth writes:
> The only language that defines a FOR loop that is always executed at least
> once is Fortran, and it requires the sign of the step to be known at compile
> time.

C's do-while is uncommon, but they haven't removed it yet.

> For all other languages, you have to hand code a compare at the top of the
> loop anyway, so it's pretty pointless to code it again at the bottom.

The naive way to code `for (i = lower; i < upper; i += step) foo;' is:

        mov     lower,i
test    cmp     i,upper
        blt     end
        foo
        add     step,i
        bra     test
end

> (Well, actually you put the compare at the bottom and have an initial branch
> to it, but you see what I mean.)

Good idea; let's try it:

        mov     lower,i
        bra     entry
body    foo
        add     step,i
entry   cmp     i,upper
        bge     body

Statically, same number of instructions; dynamically, less except in the
zero-trip case.  But look, the add, compare, and branch are consecutive.

Suppose we want to eliminate the unconditional branch.  We can duplicate the
test at the top.  Or the compiler may figure out that the initial test can't
fail (mostly if the loop has constant bounds).  In either of these cases, the
add, compare, and branch are still consecutive.

--Joe