[llvm] [AMDGPU] Add llvm.amdgcn.dead intrinsic (PR #123190)

[Diana Picus via llvm-commits]

================
@@ -3442,4 +3442,11 @@ def int_amdgcn_addrspacecast_nonnull : DefaultAttrsIntrinsic<
   [llvm_anyptr_ty], [llvm_anyptr_ty],
   [IntrNoMem, IntrSpeculatable]
 >;
+
+/// Make it clear to the backend that this value is really dead. For instance,
----------------
rovka wrote:

Because in my current usage it's ok for it to be sunk from the entry block into `shader`. In my very limited understanding, convergent would preclude that (but do correct me if I'm wrong).

https://github.com/llvm/llvm-project/pull/123190