Future atomic operation cleanup

[Carlos O'Donell]

On 08/29/2014 12:33 PM, Chris Metcalf wrote:
> On 8/29/2014 9:32 AM, Adhemerval Zanella wrote:
>> On 29-08-2014 10:27, Torvald Riegel wrote:
>>> *If* we can make 4.7 a requirement, then I'm certainly all for using
>>> just the builtins on architectures where they are available and get
>>> inlined.
>>>
>> And that is exactly what I am asking: are we willing to raise GCC requirements
>> for 2.21?  If we will, I see no issue on rework the patch to use GCC builtins.
>> If we won't, I see to no point in trying to adjust this patch to use conditionally
>> the GCC builtins for GCC 4.7+.
> 
> It seems a little too early still, in my opinion.  For example, RHEL 7 was
> only officially released a couple of months ago, and RHEL 6 was gcc 4.4, so
> Perhaps it's worth adding some mechanism to enable it for testing or
> whatever (e.g. configure with --use-atomic-builtins) in the meanwhile.

Note that RHEL 6 and RHEL 7 have a Developer Toolset (dts) in their software
collections (scl) that allows you to use gcc 4.8 and soon gcc 4.9 on those
systems to build applications.

Shipping a compiler that is decoupled from the system compiler, but still
uses the default system runtimes is becoming the norm for RHEL.

Cheers,
Carlos.