Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: $Revision: 1.6.2.16 $; site ima.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!gamma!epsilon!zeta!sabre!petrus!bellcore!decvax!ima!johnl
From: johnl@ima.UUCP
Newsgroups: net.micro
Subject: Re: Re: 386 Family Products (really :-)
Message-ID: <97800023@ima.UUCP>
Date: Thu, 19-Dec-85 12:14:00 EST
Article-I.D.: ima.97800023
Posted: Thu Dec 19 12:14:00 1985
Date-Received: Sat, 21-Dec-85 05:16:53 EST
References: <183@isieng.UUCP>
Lines: 22
Nf-ID: #R:isieng:-18300:ima:97800023:000:1240
Nf-From: ima!johnl    Dec 19 12:14:00 1985


/* Written  7:03 pm  Dec 16, 1985 by henry@utzoo in ima:net.micro */
> > The 80386 still has too few registers
> It's noteworthy that Bliss-11, at the time the most heavily optimizing
> compiler in the known universe, custom-built to exploit the pdp11 to its
> limit, and, in particular, making a huge effort to use registers as
> effectively as possible, did not manage to use more than 3 or 4 registers
> effectively on most programs.  I'm not aware of any more recent work on
> this sort of thing ...
The IBM 801 project had great success keeping their registers full.  Indeed,
their machine had 32 symmetrical registers, and the compiler used Cocke's
graph coloring scheme for register assignment.  The registers were all the
same, though, and it's not clear to me how well their work would apply to
the 80x86 architecture where the registers are not only different, but
overlay (ah, ax, and eax, for example.)  A fascinating challenge, and no
doubt a problem that will be solved just as Intel announces a RISC chip
with 128 registers to meet the competition from Motorola, National, INMOS,
NCR, Fairchild, NEC, Fujitsu, Siemens, IBM, and AT&T.  ( :-) sort of)

John Levine, ima!johnl

PS:  I miss the PDP-8.  Now THAT was symmetrical.