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

[Jon Chesterfield via Openmp-commits]

JonChesterfield wrote:

I suggested we just pass too-large values to HSA and catch the error from HSA. Unfortunately it comes in the form of an invocation of callbackError which calls FATAL_MESSAGE which calls abort(), so that's not great. There's quite a lot of calls to FATAL_MESSAGE in the nextgen plugin infra which is a shame given the plugin put effort into returning error codes on other paths.

Leaning slightly towards accepting the copy&paste with string slice and just having things blow up on us later, what do you guys think?

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