From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-return-57186-listarch-gcc=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 21966 invoked by alias); 26 Jul 2002 11:18:13 -0000
Mailing-List: contact gcc-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Archive: <http://gcc.gnu.org/ml/gcc/>
List-Post: <mailto:gcc@gcc.gnu.org>
List-Help: <http://gcc.gnu.org/ml/>
Sender: gcc-owner@gcc.gnu.org
Received: (qmail 21959 invoked from network); 26 Jul 2002 11:18:11 -0000
Received: from unknown (HELO executor.cambridge.redhat.com) (195.224.55.237)
  by sources.redhat.com with SMTP; 26 Jul 2002 11:18:11 -0000
Received: from host140.cambridge.redhat.com (host140.cambridge.redhat.com [172.16.18.140])
	by executor.cambridge.redhat.com (Postfix) with ESMTP
	id 2F990ABAF8; Fri, 26 Jul 2002 12:18:11 +0100 (BST)
Received: from localhost (bernds@localhost)
	by host140.cambridge.redhat.com (8.11.6/8.11.0) with ESMTP id g6QBIAf16733;
	Fri, 26 Jul 2002 12:18:10 +0100
X-Authentication-Warning: host140.cambridge.redhat.com: bernds owned process doing -bs
Date: Fri, 26 Jul 2002 15:01:00 -0000
From: Bernd Schmidt <bernds@redhat.com>
X-X-Sender: bernds@host140.cambridge.redhat.com
To: Mark Mitchell <mark@codesourcery.com>
Cc: Richard Kenner <kenner@vlsi1.ultra.nyu.edu>,
	"gcc@gcc.gnu.org" <gcc@gcc.gnu.org>
Subject: Re:  Reload question
In-Reply-To: <100950000.1027624769@warlock.codesourcery.com>
Message-ID: <Pine.LNX.4.44.0207261216170.16587-100000@host140.cambridge.redhat.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-SW-Source: 2002-07/txt/msg01275.txt.bz2

On Thu, 25 Jul 2002, Mark Mitchell wrote:

> >     In reload, after doing register allocation, we do, in "reload()":
> >
> >     That mutates the instruction in place, with the possible result that
> >     it is no longer recognizable.
> >
> >     Who is supposed to be responsible for making the instruction
> >     recognizable again?
> >
> > Reload itself.  The insn won't match its constraints.
> 
> But this happens *at the end* of reload.  Search for:
> 
>    PUT_CODE (reg, MEM)
> 
> in reload1.c.
> 
> So, where in reload is the instruction supposed to be checked to see
> that it is safe to make this transmogrification?

find_reloads.  If the modification isn't safe, we'll make a reload for
the operand and thus ensure that when we get to the code that modifies
pseudos into MEMs, the transformation is correct.


Bernd