[PATCH] D117690: AMDGPU/GlobalISel: Directly diagnose return value use for FP atomics

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D117690#3257315 <https://reviews.llvm.org/D117690#3257315>, @foad wrote:

>> Previously we were falling back to the DAG on selection
>> failure, where it would emit this error and then fail again.
>
> Why is that a problem? Is this just a temporary thing, while you are testing with the fallback to dag enabled?

The permanent solution is to remove the DAG completely so this should have matching behavior

> To put it another way, what's so special about this case that legalization has to fail harder than other cases?

It also saves compile time from compiling the function twice


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D117690/new/

https://reviews.llvm.org/D117690