ports/95400: [patch] devel/pwlib 1.10.0 does not compile when CPUTYPE?=i686 set

[Kostik Belousov]

The following reply was made to PR ports/95400; it has been noted by GNATS.

From: Kostik Belousov <konstantin.belousov at zoral.com.ua>
To: Steve Ames <steve at energistic.com>
Cc: bug-followup at freebsd.org, edwin at freebsd.org
Subject: Re: ports/95400: [patch] devel/pwlib 1.10.0 does not compile when CPUTYPE?=i686 set
Date: Thu, 6 Apr 2006 18:09:23 +0300

 On Thu, Apr 06, 2006 at 10:38:37AM -0400, Steve Ames wrote:
 > On Thu, Apr 06, 2006 at 11:24:47AM +0000, Edwin Groothuis wrote:
 > > 
 > > The full text of the PR can be found at:
 > >     http://www.freebsd.org/cgi/query-pr.cgi?pr=ports/95400
 > 
 > Kostik Belousov,
 > 
 > Thank you for spotting this. As you noted that assembler snipped is
 > already disabled for x86_64 (for the same reasons and errors you pointed
 > out). I'm afraid my assembler-fu is not up to the task of evaluating of
 > your correction is correct. If it is correct then I'm OK with submitting
 > the change to the assembler especially if it also fixes the x86_64 case.
 I don't have access to the x86_64 machine and don't have
 time to build a cross-compiler, so, I cannot
 check whether change would work for that arch.
 > 
 > However if the assembler cannot be confirmed I'd prefer we just patch
 > jidctflt.cxx so that it always skips that assembler for all architectures.
 
 See, I can't test the code. It looks like openh323 developers did not
 compile it either. The problem is that sar instruction (shift 
 arithmetical right) requires shift count to be supplied in register
 %cl. Constraint supplied allows for shift count to be put in any register.
 If the code compiles on the developers machine, it was by pure
 coincident.
 
 Because of this, I propose to disable the code uncoditionally.