From mboxrd@z Thu Jan  1 00:00:00 1970
From: Alan Modra <alan@SPRI.Levels.UniSA.Edu.Au>
To: Alexander Sokolov <robocop@amur.rosnet.ru>
Cc: binutils@sourceware.cygnus.com
Subject: Re: Error: suffix or operands invalid for `bswap'
Date: Thu, 01 Jul 1999 00:00:00 -0000
Message-id: <Pine.LNX.4.10.9906230943470.1129-100000@mullet.itr.unisa.edu.au>
References: <Pine.LNX.4.10.9906230844350.4367-100000@ns.amur.rosnet.ru>
X-SW-Source: 1999-q2/msg00337.html

On Wed, 23 Jun 1999, Alexander Sokolov wrote:
> At least openssl and international kernel patch use bswapl. Rejecting
> this syntax causes compiling failure. On the other hand "bswapl %eax"
> is no less legal than "pushl %eax" or "addl %eax,%ebx" allowed in 2.9.4
> and if you disallow l suffix for bswap you should also do it for all the
> instructions with 32-bit regs.

The difference is that there are "pushw", "addw", and "addb" instructions,
but no "bswapw".  In saying this, I'm not arguing against restoring bswapl
to legal status, just pointing out that "bswap" is somewhat special.

I'll check a patch in tonight to fix bswap, unless anyone reading this
list has a violent objection.  I'm also curious to know whether Unixware
or other AT&T format x86 assemblers accept "bswapl".  Would someone with
access to such an assembler please try it out?