svn commit: r270099 - in stable: 10/contrib/gcc/config/i386 9/contrib/gcc/config/i386

[Pedro Giffuni]

Hello;

On 08/17/14 17:45, Dimitry Andric wrote:
> On 17 Aug 2014, at 15:45, Alexey Dokuchaev <danfe at FreeBSD.org> wrote:
>> On Sun, Aug 17, 2014 at 03:29:42PM +0200, Dimitry Andric wrote:
>>> In principle it is applicable, but the same file also has other changes
>>> in head which were not MFCd, so just MFCing this one commit does not
>>> make much sense.  For example, the earlier cast fixes were part of a
>>> much larger commit by Pedro Giffuni, adding "experimental support for
>>> amdfam10/barcelona CPUs":
>>>
>>> http://svnweb.freebsd.org/base?view=revision&revision=251212
>> I'm running my stable/8 with Pedro's patches applied, including r251212,
>> no problems so far (although I don't have recent AMD CPUs to play with).
>>
>>> Does it still make sense to backport such experimental changes to an old
>>> stable branch?  Of course I could split off just the changes to
>>> emmintrin.h, and leave the others out, but then we would have a partial
>>> MFC.  I'm not sure if that is the usual way of doing things...
>> Understood.  My goal here is to try to keep stable/8 as alive as possible,
>> since I plan to keep using it beyond its official EOL.  Hence, when I see
>> fixes that potentially help ports to be buildable on it I'd usually ask if
>> they can be MFCed (when it's easy enough to do).

FWIW, I recall the AMD patch was developed on stable/8 and should be
safe to merge. You still need have to teach the build system about the
new CPUs (that was a different change that I didn't do) but it should work.

I personally stopped merging stuff to the stable/8 branch and more recently
to the stable/9 branch as I don't run those anymore. In the case of the 
stable/8
branch I find the ancient version of binutils a real threat/limitation.

I would really suggest people move on to at least stable/9 which has all the
clang cleanups and should be functionally much better.

Pedro.

> Can you please try this diff [1], which merges most of the stable/9 gcc
> changes to stable/8?  I've ran it through a make universe, and the only
> failure I got was with the amd64 XENHVM kernel:
>
> amd64 XENHVM kernel failed, check _.amd64.XENHVM for deatils
>
> but I don't know if this is an expected failure or not.  Tinderbox seems
> to have other trouble with its stable/8 builds.  The actual error is:
>
> In file included from sys/sys/param.h:86,
>                   from sys/compat/ia32/ia32_genassym.c:6:
> sys/sys/types.h:44:28: error: machine/endian.h: No such file or directory
>
> I didn't test any ports yet, though.
>
> -Dimitry
>
> [1] http://www.andric.com/freebsd/sync-stable8-gcc-with-stable9-1.diff.xz
>