[llvm-branch-commits] [llvm] [openmp] [offload] Use HSA SVM for AMDGPU shared memory (PR #215801)

[Robert Imschweiler via llvm-branch-commits]

ro-i wrote:

I now completely changed the implementation of this PR, based on #228569, using HSA for managing the SVM allocations instead of managing them myself. Is a lot smaller and cleaner now, I think.
Only issue is that dataFill isn't perfect because hsa_amd_memory_fill was buggy before https://github.com/ROCm/rocm-systems/commit/9ab39639dd

https://github.com/llvm/llvm-project/pull/215801