Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!rice!uw-beaver!uw-june!dgb
From: dgb@cs.washington.edu (David Bradlee)
Newsgroups: comp.arch
Subject: Re: i860(was Re: IBM RISC)
Summary: use of T on i860
Message-ID: <10833@june.cs.washington.edu>
Date: 22 Feb 90 23:36:15 GMT
References: <8064@pt.cs.cmu.edu> <1990Feb20.031858.18512@Neon.Stanford.EDU>
Organization: U of Washington, Computer Science, Seattle
Lines: 19

In article <1990Feb20.031858.18512@Neon.Stanford.EDU>, gopi@goretex.Stanford.EDU (K. Gopinath) writes:
> What use are the KR,KT & T  regs? I tried coding for a loop with a stmt
> d = d + a[i]*b[i]*c[i] but the non-orthogonalities just repeatedly
> frustrate any way of keeping both +/* going at the same time (modulo
> 2 mults and 1 add). These regs cannot be used like temps; there are
> too many restrictions.

One use for T is in an expression such as (a*b)+(c*d).  You can
overlap the 2 multiples and save the first in T, then add T directly
to the result of the multiplier.  KR and KI could be used to multiply
by a loop invariant.  I'm sure there are other uses.  The i860 does
have many non-orthogonalities which make it hard write code for and
hard for the compiler to make choices between the different options.

	Dave Bradlee
	Department of Computer Science and Engineering, FR-35
	University of Washington
	Seattle, WA  98195
	dgb@cs.washington.edu