[Bug c/68908] inefficient code for _Atomic operations

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68908

--- Comment #7 from Andrew Macleod <amacleod at redhat dot com> ---
(In reply to Richard Henderson from comment #4)
> I think we should rather handle this in the front end than with
> quite complex pattern matching.
> 
> If we want to do any complex logic with atomics in the middle-end,
> we should change their representation so that we don't have to
> struggle with a sequence of builtins.  Which is clearly a subject
> for gcc7 at minimum.

Yes, I think anything more complex than this should be part of an atomics
optimization framework using a new set of ATOMIC gimple ops rather than
builtins.

For the purpose of this PR we ought to just fix it in the FE.