[CFT] gcc: support for barcelona

[Warner Losh]

On May 27, 2013, at 2:07 PM, Pedro Giffuni wrote:

> On 27.05.2013 14:38, Dimitry Andric wrote:
>> On May 27, 2013, at 21:12, Rui Paulo <rpaulo at felyko.com> wrote:
>>> On 27 May 2013, at 09:41, Pedro Giffuni <pfg at freebsd.org> wrote:
>>>> Almost a year ago I tried to bring in the support for AMD's barcelona
>>>> chipset into our gcc. This actually filled a lot of holes in that were left
>>>> when similar intel support was brought in.
>>>> 
>>>> Unfortunately I had to revert rapidly such support as it broke building
>>>> some C++ ports even when it was not being used.
>>>> 
>>>> jkim@ did some cleanup of the support and the patch has been
>>>> gathering rust here:
>>>> 
>>>> http://people.freebsd.org/~jkim/reworked-r236962-3.diff
>>>> 
>>>> The patch still applies cleanly and there is a good chance it will work
>>>> since there have been other fixes merged since the last time.
>>>> 
>>>> I did some basic testing and so far it works for me but I don't have
>>>> the specific chipset. Additional testing would be welcome.
>>> I have to question the general direction of this work. We switched to Clang as the default compiler for i386/amd64 some months ago and now you're working on improving our base GCC especially for amd64? I don't really understand how useful this is. It doesn't strike me as a good idea to see people working on things that will eventually be replaced / removed.
>> It is probably a better use of time to work on getting the tree to build
>> with an out-of-tree gcc 4.7 or 4.8 instead.  Why spend more effort on a
>> completely dead branch of gcc?  Newer gcc's have better code generation,
>> support for more modern CPUs, and better diagnostics (including even
>> those controversial carets ;-).
> 
> FWIW, upstream gcc has a bug that affects ctfmerge and they have
> been very slow to fix it. I submitted a bug report and a workaround
> patch for ctfmerge to the Illumos guys but they have been very slow
> to review it as well.
> 
> I do agree having out-of-tree compilers is important though; and
> much preferable than carrying two compilers ;).

Is this patch in the ports version of gcc at least?

Warner