[PATCH] D55562: Atomics: support min/max orthogonally

[Tim Northover via Phabricator via cfe-commits]

t.p.northover added a comment.

> What does it do with floating-point inputs?

Same as all the other atomics: run screaming for the hills (or error out, in more reasonable terms). The only way to implement floating atomics in LLVM IR would be with a compare-exchange loop and Clang doesn't expose anything to make that more convenient.

Min/max have at least some pedigree in integer terms. OpenCL has specified them, CPUs have implemented them, and we've had requests for builtins from users.


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D55562/new/

https://reviews.llvm.org/D55562