[llvm] [AMDGPU] Change isSingleLaneExecution to account for WWM enabling lanes even if there's only one workitem (PR #188316)

[Matt Arsenault via llvm-commits]

================
@@ -263,6 +264,19 @@ unsigned AMDGPUSubtarget::getMaxWorkitemID(const Function &Kernel,
 }
 
 bool AMDGPUSubtarget::isSingleLaneExecution(const Function &Func) const {
----------------
arsenm wrote:

Any code inspecting the IR like this does not belong in AMDGPUSubtarget. At most anything here should be querying the calling convention and attributes for the function. 

https://github.com/llvm/llvm-project/pull/188316