[Bug target/103066] __sync_val_compare_and_swap/__sync_bool_compare_and_swap aren't optimized

["thiago at kde dot org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103066

--- Comment #10 from Thiago Macieira <thiago at kde dot org> ---
You're right that emitting more penalises those who have done their job and
written proper code.

The problem we're seeing is that such code appears to be the minority. Or,
maybe put differently, the bad code is showing up a lot in our benchmarks,
especially on very big multi-core and multi-socket systems. "Fixing" the
compiler would make a broad update to the industry -- so long as the code is
recompiled with new compilers. Fixing the actual code would make it better even
if used with old ones.

Does anyone have a suggestion on how to get best "bang for buck"? (Biggest
benefit for smallest effort) This is a sincere question. I'm not trying to be
ironic or sarcastic. How can we help the most, the quickest, for the limited
amount of resources we can marshal?

Also, and I've been hitting this key for a few years, how can we do better at
teaching people how to use the tools at their disposal at the proper way? A
very good counter-example is C++11's std::atomic_flag: you MUST NEVER use it
(at least until C++20, where it got a test() member).