[llvm] [AMDGPU] Automatic conversion from wave32 to wave64 (PR #137376)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> A few questions to discuss right now:
> 
>     1. May we assume that for the targets which have a wave32 as default correct attribute will be set by the FE?
You always know the wave size as long as you have subtarget. I assume the attribute shall be always set by FE.

>     2. Non-convergent intrinsic calls are left in the IR as a calls. I guess we should change the callee version (.i32/i64) if we set wave64 on the caller.
I am not sure what do you mean, but it sounds you are referring to intrinsics accessing wave size since these need such mangling. That shall not be allowed.



https://github.com/llvm/llvm-project/pull/137376