[llvm] [AMDGPU] Introduce "amdgpu-sw-lower-lds" pass to lower LDS accesses. (PR #87265)

[via llvm-commits]

skc7 wrote:

Update:

- This PR has been rebased to latest main to use the AMDGPUAsanInstrumentation APIs to instrument LDS instructions lowered to global memory. 

- New tests have been added check the asan instrumented IR. 

- Feature has been enabled in local downstream builds and tested with all the static, dynamic LDS cases and they seem to detect any addressing errors.

https://github.com/llvm/llvm-project/pull/87265