[llvm] [AMDGPU] Allocate i1 argument to SGPRs (PR #72461)

[Matt Arsenault via llvm-commits]

================
@@ -407,6 +407,12 @@ class CallLowering {
       CallingConv::ID CallConv, bool IsVarArg,
       ArrayRef<Register> ThisReturnRegs = std::nullopt) const;
 
+  bool determineAndHandleAssignments(
----------------
arsenm wrote:

Now that I think of it, I think it would be better to just use separate determineAssignments, and handleAssignments. determineAndHandleAssignments is already the simple case where the CCState doesn't need any custom handling 

https://github.com/llvm/llvm-project/pull/72461