[clang] [llvm] [AArch64] Implement the atomic store with hint intrinsic (PR #198316)

[Kerry McLaughlin via llvm-commits]

kmclaughlin-arm wrote:

> This seems to have some compile-time impact at -O0: https://llvm-compile-time-tracker.com/compare.php?from=2ac696e23ca31d68a89060c7a935c75bc4efb846&to=34f5f2c594111e6a3edf65673c3ea2849e96e2cb&stat=instructions:u From a quick glance over the diff it's not super obvious what would cause it.

I have been trying to investigate the compile-time impact using 7zip, as it showed the biggest change after landing this PR according to the link above. I wondered if the fallback I added to `AArch64TargetLowering::fallBackToDAGISel` could be causing the issue at O0, but I didn't find that removing it made a difference. Removing the new patterns I added for the atomic stores with metadata also did not make a difference.

After running a few more experiments, I found that adding only the diagnostic messages to `DiagnosticSemaKinds.td` from this PR caused a similar compile-time increase. I'm interested if anyone has any thoughts on why this might be, as I'm not yet sure why these could be causing an issue?

https://github.com/llvm/llvm-project/pull/198316