[PATCH] D52416: Allow FP types for atomicrmw xchg

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In https://reviews.llvm.org/D52416#1243625, @jfb wrote:

> What does this look like with half, fp80, fp128?
>  Does this not change anything on other architectures?


f80 is illegal since it's not a power of two sized type, so that's rejected by the IR verifier. I guess supporting it could somehow be possible, but is beyond the scope of this patch. I added some tests for fp128.

Ideally each target would add support for the FP types directly in its selection patterns, but that's a lot of work. The individual targets avoid work by setting the defaults to the bitcast lowering. For the atomic_load/store FP support, this for some reason was done in the IR AtomicExpand pass.


https://reviews.llvm.org/D52416