From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <binutils-return-38005-listarch-binutils=sources.redhat.com@sources.redhat.com>
Received: (qmail 8080 invoked by alias); 16 Feb 2005 18:01:40 -0000
Mailing-List: contact binutils-help@sources.redhat.com; run by ezmlm
Precedence: bulk
List-Subscribe: <mailto:binutils-subscribe@sources.redhat.com>
List-Archive: <http://sources.redhat.com/ml/binutils/>
List-Post: <mailto:binutils@sources.redhat.com>
List-Help: <mailto:binutils-help@sources.redhat.com>, <http://sources.redhat.com/ml/#faqs>
Sender: binutils-owner@sources.redhat.com
Received: (qmail 8029 invoked from network); 16 Feb 2005 18:01:36 -0000
Received: from unknown (HELO pollux.ds.pg.gda.pl) (153.19.208.7)
  by sourceware.org with SMTP; 16 Feb 2005 18:01:36 -0000
Received: from localhost (localhost [127.0.0.1])
	by pollux.ds.pg.gda.pl (Postfix) with ESMTP
	id F0587F598D; Wed, 16 Feb 2005 19:01:30 +0100 (CET)
Received: from pollux.ds.pg.gda.pl ([127.0.0.1])
 by localhost (pollux [127.0.0.1]) (amavisd-new, port 10024) with ESMTP
 id 25304-07; Wed, 16 Feb 2005 19:01:30 +0100 (CET)
Received: from piorun.ds.pg.gda.pl (piorun.ds.pg.gda.pl [153.19.208.8])
	by pollux.ds.pg.gda.pl (Postfix) with ESMTP
	id A752BF5980; Wed, 16 Feb 2005 19:01:30 +0100 (CET)
Received: from blysk.ds.pg.gda.pl (macro@blysk.ds.pg.gda.pl [153.19.208.6])
	by piorun.ds.pg.gda.pl (8.13.1/8.13.1) with ESMTP id j1GI1Ycr026174;
	Wed, 16 Feb 2005 19:01:35 +0100
Date: Thu, 17 Feb 2005 00:25:00 -0000
From: "Maciej W. Rozycki" <macro@linux-mips.org>
To: Thiemo Seufer <ica2_ts@csv.ica.uni-stuttgart.de>
Cc: binutils@sources.redhat.com
Subject: Re: [PATCH] MIPS gas: Fix macro expansions for .set noat
In-Reply-To: <20050216154924.GG1757@rembrandt.csv.ica.uni-stuttgart.de>
Message-ID: <Pine.LNX.4.61L.0502161745010.3450@blysk.ds.pg.gda.pl>
References: <20050216154924.GG1757@rembrandt.csv.ica.uni-stuttgart.de>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-Virus-Status: Clean
X-SW-Source: 2005-02/txt/msg00377.txt.bz2

On Wed, 16 Feb 2005, Thiemo Seufer wrote:

> @@ -5762,11 +5761,19 @@ macro (struct mips_cl_insn *ip)
>      case M_LWU_AB:
>        s = "lwu";
>      ld:
> -      if (breg == treg || coproc || lr)
> +      /* XXX Why don't we try to use AT for all expansions? */

 Because we don't have to and it helps in cases like:

	.set	noat
	ld	k0,foo

> +      if (!mips_opts.noat && (breg == treg || coproc || lr))
>  	{
>  	  tempreg = AT;
>  	  used_at = 1;
>  	}
> +      else if (breg == treg
> +	       && (offset_expr.X_op != O_constant
> +		   || (offset_expr.X_add_number > 0x7fff
> +		       || offset_expr.X_add_number < -0x8000)))
> +	{
> +	  as_bad(_("load expansion needs $at register"));
> +	}
>        else
>  	{
>  	  tempreg = treg;

 This looks broken -- you can only use treg as a scratch pad if it's a 
GPR.

> @@ -5840,8 +5847,23 @@ macro (struct mips_cl_insn *ip)
>      case M_SDR_AB:
>        s = "sdr";
>      st:
> -      tempreg = AT;
> -      used_at = 1;
> +      if (!mips_opts.noat)
> +	{
> +	  tempreg = AT;
> +	  used_at = 1;
> +	}
> +      else if (breg == treg
> +	       && (offset_expr.X_op != O_constant
> +		   || (offset_expr.X_add_number > 0x7fff
> +		       || offset_expr.X_add_number < -0x8000)))
> +	{
> +	  as_bad(_("store expansion needs $at register"));
> +	}
> +      else
> +	{
> +	  tempreg = treg;
> +	  used_at = 0;
> +	}
>      ld_st:
>        /* Itbl support may require additional care here.  */
>        if (mask == M_LWC1_AB

 This is yet more broken -- treg can never be used, because it contains 
the value to be stored!  And of course it's not expected to get trashed 
anyway.

  Maciej