[llvm] [Offload] Implement olMemFill (PR #154102)

[Jan Patrick Lehr via llvm-commits]

jplehr wrote:

> I remember there being some weird stuff with the DMA engine on AMDGPU, maybe @jplehr remembers.

This has been fixed with ROCm 6.3

https://github.com/llvm/llvm-project/pull/154102