Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!snorkelwacker!spdcc!mirror!ima!esegue!compilers-sender
From: wendyt@cs.washington.edu (Wendy Thrash)
Newsgroups: comp.compilers
Subject: Re: Register Allocation
Keywords: Pointers
Message-ID: <1990Feb26.214703.10586@esegue.segue.boston.ma.us>
Date: 26 Feb 90 21:47:03 GMT
References: <1990Feb20.155619.3121@esegue.segue.boston.ma.us> <1990Feb24.210757.4202@esegue.segue.boston.ma.us>
Sender: compilers-sender@esegue.segue.boston.ma.us
Reply-To: wendyt@cs.washington.edu (Wendy Thrash)
Organization: University of Washington, Computer Science, Seattle
Lines: 24
Approved: compilers@esegue.segue.boston.ma.us

>[The PDP-10/DEC-20 has addressable registers as well, but as pointed out
>elsewhere that's only a small part of the problem.  As soon as you call
>another routine, it's liable to save the register and put something else
>there, and you lose.  -John]

The Pyramid architecture (Pyramid Technology) combines addressable registers
(which take a bit of work to address, but not _too_ much) with register
windows.  As long as variables that have had their addresses taken are
assigned to parameter registers or local registers, but not temporary
registers, their registers are not accessible to called routines.  (More
precisely, they cannot be accessed in any quasi-legitimate way by a
subroutine, and certainly will not be accessed by the compilers.)

I believe their architecture solves the problems discussed here, but I've been
told that the hardware cost to make this register addressing work is
significant.  (or perhaps it was "enormous")

Disclaimer: Perhaps I should say "our" architecture, since I may technically
still be an employee of Pyramid Technology, on leave.  Maybe my old boss will
see this and let me know whether I still "work" there. :-)
-- 
Send compilers articles to compilers@esegue.segue.boston.ma.us
{spdcc | ima | lotus}!esegue.  Meta-mail to compilers-request@esegue.
Please send responses to the author of the message, not the poster.