diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index afce1fe6af854..50888b3d9b6d5 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -3808,6 +3808,7 @@ class AMDGPUCooperativeAtomicLoad<LLVMType Ty> : Intrinsic <
   [SDNPMemOperand, SDNPMayLoad]
 >;
 
+// TODO: We may want to drop _relaxed and use an atomic ordering operand instead.
 def int_amdgcn_cooperative_atomic_load_32x4B : AMDGPUCooperativeAtomicLoad<llvm_i32_ty>;
 def int_amdgcn_cooperative_atomic_store_32x4B : AMDGPUCooperativeAtomicStore<llvm_i32_ty>;
 def int_amdgcn_cooperative_atomic_load_16x8B : AMDGPUCooperativeAtomicLoad<llvm_v2i32_ty>;