Re: toolchain/45673: gcc on i386 current generates bad code for assembler code in lock.h

[Joerg Sonnenberger <joerg%NetBSD.org@localhost>]

The following reply was made to PR toolchain/45673; it has been noted by GNATS.

From: Joerg Sonnenberger <joerg%homeworld.netbsd.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: toolchain/45673: gcc on i386 current generates bad code for
 assembler code in lock.h
Date: Sun, 11 Mar 2012 23:35:01 +0000

 On Wed, Nov 30, 2011 at 08:55:00PM +0000, kuehro%gmx.de@localhost wrote:
 > The reason seems to be gcc bug 10153. The error with pkgsrc can be fixed
 > by changing r to q in line 108 of /usr/include/x86/lock.h:
 > 
 > _asm volatile ("xchgb %0,(%2)" : 
 >          "=r" (val)
 >          :"0" (val), "r" (lockp));
 >      
 > This restricts allowable registers to a,b,c or d. Of course it would be
 > much better to fix gcc somehow. 
 
 That restricts it more than necessary, doesn't it? I think the correct
 constraint here "=qQ", since both l and h registers are allowed.
 
 Joerg