[Openmp-commits] [openmp] [OpenMP][AMDGPU] Adapt dynamic callstack sizes to HIP behavior (PR #74080)

[Michael Halkenhäuser via Openmp-commits]

mhalk wrote:

Apologies for the long radio silence.

After reaching out to @arsenm the idea of moving the enum `AMDGPUSubtarget::Generation` does not seem attractive anymore.
As Matt suggested I'll try to implement a `SubtargetFeature` (named sth. like `MaxWaveScratchSize`) that may be queried via `TargetParser`, which holds the corresponding callstack size.
This will avoid duplication, if I can get it to work within the two locations where `getMaxWaveScratchSize` is needed.
(These changes should be reasonable within the same PR.)

I plan to be working on this next week.

https://github.com/llvm/llvm-project/pull/74080