On 3/5/14 7:59 AM, "René J.V. Bertin" wrote:
> On Mar 05, 2014, at 09:55, Ryan Schmidt wrote:
>> environment variables to build each architecture, if desired. It is not 
>> possible to specify different variants for each architecture; variants apply 
>> to the port as a whole.
> But I presume it would be possible to have architecture-specific options for 
> a variant? If so, what I was suggesting as (temporary) fix, was to tweak the 
> i386-specific asm options to render it harmless.
>
>> Is it the case that the asm variant only works for the x86_64 architecture?
> Apparently, yes.
>
>> If so, my proposed fix would be that the asm variant only be selectable if 
>> x86_64 is within the architectures that will be built, and if so, to only 
>> apply to that architecture and not any others.
> Are we saying the same thing?
>
> The other way of looking at it is this: is there a reason NOT to use the asm 
> variant (at least for x86_64)? I suppose it exists for performance reasons, 
> and if justifiable it could in that case be the default. In this case (and 
> until the i386 issue is fixed), it would be activated for the x86_64 part, 
> but not for i386. That would at least get rid of the suggestion that asm is 
> used for all architectures.
> One could keep the +asm variant (which would lead to a pure i386 build), and 
> one might add a +noasm variant (that would disable the asm parts in the 
> x86_64 binaries) if there are usage cases that are incompatible with the asm 
> code.
>
> Again, this would only make sense if the asm option is purelyinternal, 
> without introducing changes to the library's ABI...
> _______________________________________________
>
https://trac.macports.org/changeset/117612

+asm variant dropped and --disable-asm asserted whenever configuring for
i386.  This allows +universal to build with asm optimizations for x86_64
and without for i386.  Also fixes -universal build for i386.  Other
possible archs are handled by the port itself.  No ABI impact. Thanks
for the suggestion.


<https://trac.macports.org/changeset/117612>
_______________________________________________
macports-users mailing list
macports-users@lists.macosforge.org
https://lists.macosforge.org/mailman/listinfo/macports-users

Reply via email to