diff --git a/tools/clang/test/HLSLFileCheck/hlsl/operator_overloading/subscript-operator.hlsl b/tools/clang/test/HLSLFileCheck/hlsl/operator_overloading/subscript-operator.hlsl
new file mode 100644
index 0000000000..063952f126
--- /dev/null
+++ b/tools/clang/test/HLSLFileCheck/hlsl/operator_overloading/subscript-operator.hlsl
@@ -0,0 +1,26 @@
+// RUN: %dxc -T lib_6_4 -HV 2021 -ast-dump %s | FileCheck %s
+
+struct MyVector3 {
+    float3 V;
+
+    float operator[](int x) {
+        return V[x];
+    }
+};
+
+float SomeFun(const MyVector3 V, int Idx) {
+  return V[Idx];
+}
+
+// This test verifies that under HLSL 2021, overload resolution for user-defined
+// subscript operators follows HLSL overload resolution rules and ignores the
+// const-ness of the implicit object parameter.
+
+// CHECK: CXXRecordDecl {{0x[0-9a-fA-F]+}} {{.*}} referenced struct MyVector3 definition
+// CHECK: FieldDecl {{0x[0-9a-fA-F]+}} <line:4:5, col:12> col:12 referenced V 'float3':'vector<float, 3>'
+// CHECK-NEXT: CXXMethodDecl [[Operator:0x[0-9a-fA-F]+]] <line:6:5, line:8:5> line:6:11 used operator[] 'float (int)'
+
+// CHECK: FunctionDecl {{0x[0-9a-fA-F]+}} <line:11:1, line:13:1> line:11:7 SomeFun 'float (const MyVector3, int)'
+// CHECK: CXXOperatorCallExpr {{0x[0-9a-fA-F]+}} <col:10, col:15> 'float'
+// CHECK-NEXT: ImplicitCastExpr {{0x[0-9a-fA-F]+}} <col:11, col:15> 'float (*)(int)' <FunctionToPointerDecay>
+// CHECK-NEXT: DeclRefExpr {{0x[0-9a-fA-F]+}} <col:11, col:15> 'float (int)' lvalue CXXMethod [[Operator]] 'operator[]' 'float (int)'