Allow kernel instrumentation passes to be added to pipeline (#3953) This PR adds the ability to insert passes into the Triton LLVM pass pipeline through the LLVM Pass Plugin capability. This is technically an architecture agnostic patch but as of current AMDGPU is the only backend that would make use of it.
[TritonShared] Fix SME failure type 1 revert merge request: [TritonShared] Fix SME failure type 1 ed1126d4f6bb9d77cd2cc20943c9413519d74aab This revert merge request !177
