From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 12879 invoked by alias); 7 Jan 2003 05:20:04 -0000 Mailing-List: contact gcc-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Archive: List-Post: List-Help: Sender: gcc-owner@gcc.gnu.org Received: (qmail 12562 invoked from network); 7 Jan 2003 05:17:32 -0000 Received: from unknown (HELO gull.mail.pas.earthlink.net) (207.217.120.84) by 209.249.29.67 with SMTP; 7 Jan 2003 05:17:32 -0000 Received: from ilmasc01-67.midwest.net ([209.248.16.77] helo=there) by gull.mail.pas.earthlink.net with smtp (Exim 3.33 #1) id 18Vm71-0003Kv-00; Mon, 06 Jan 2003 21:17:16 -0800 Content-Type: text/plain; charset="iso-8859-1" From: Andy Walker To: dewar@gnat.com (Robert Dewar), lord@emf.net Subject: Re: An unusual Performance approach using Synthetic registers Date: Tue, 07 Jan 2003 05:29:00 -0000 Cc: gcc@gcc.gnu.org References: <20030106202439.01801F2CB6@nile.gnat.com> In-Reply-To: <20030106202439.01801F2CB6@nile.gnat.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Message-Id: X-SW-Source: 2003-01/txt/msg00325.txt.bz2 On Monday 06 January 2003 02:24 pm, Robert Dewar wrote: > > In other words, with synthregs, the CPU can ship some value off to > > memory and not care how long it takes to get there or to get back from > > there -- because it also ships it off to the synthreg, which it > > hypothetically has faster access to. > > But this "hypothesis" is wrong. memory used for spills or locals is exactly > the same as memory used for "synthetic registers" [this fancy term is > nothing more than a fancy name for a local temporary]. So there is no issue > of having faster access to one or the other. It may of course be the case > that in one case you get more competent code than the other, but if so, the > fix is to fix incompetent code :-) An interesting proposition that I had not considered. After Synthetic registers get running well enough to be rationally evaluated, I may be interested in revisiting this idea. Andy