[llvm] [AMDGPU] Add intrinsic for raw atomic buffer loads (PR #97707)

[Jessica Del via llvm-commits]

================
@@ -1378,6 +1380,8 @@ std::optional<Value *> GCNTTIImpl::simplifyDemandedVectorEltsIntrinsic(
   case Intrinsic::amdgcn_raw_ptr_buffer_load:
   case Intrinsic::amdgcn_raw_buffer_load_format:
   case Intrinsic::amdgcn_raw_ptr_buffer_load_format:
+  case Intrinsic::amdgcn_raw_atomic_buffer_load:
+  case Intrinsic::amdgcn_raw_atomic_ptr_buffer_load:
----------------
OutOfCache wrote:

In theory, the test with the `<4 x i32>` load should show the effects, since only one element is used, no? But adding the intrinsics to this unfortunately did not have any effect. When I tried debugging with lldb, it seemed like this method was never called. If there is anything missing, I'd love to know!

https://github.com/llvm/llvm-project/pull/97707