Fix global phase for PauliI rotation and DumpRegister

This change introduces a new, stdlib-only intrinsic for applying global phase during simulation to make phase behavior more consisten and fixes #1450. This intrinsic only applies the global phase during simulation; the new `GlobalPhase` intrinsic is a no-op for QIR generation, circuit generation, and resource estimation. The change also includes updates to the stdlib and associated tests to account for the corrected global phase behavior.

This also fixes a bug in `DumpRegister` related to iteration through the state vector that ensures the "base" amplitude used for separating the state is consistently chosen, which avoids corner cases that could introduce a phase in the displayed output that wasn't present in the underlying state.

compiler/qsc_codegen/src/qir_base.rs

@@ -462,6 +462,12 @@ impl Backend for BaseProfSim {
         name: &str,
         arg: Value,
     ) -> Option<std::result::Result<Value, String>> {
+        // Global phase is a special case that is non-physical, so there is no need to generate
+        // a call here, just do a shortcut return.
+        if name == "GlobalPhase" {
+            return Some(Ok(Value::unit()));
+        }
+
         match self.write_decl(name, &arg) {
             Ok(()) => {}
             Err(e) => return Some(Err(e)),

compiler/qsc_codegen/src/qir_base/tests.rs

@@ -1592,3 +1592,66 @@ fn custom_intrinsic_fail_on_non_unit_return() {
         "#]],
     );
 }
+
+#[test]
+fn pauli_i_rotation_for_global_phase_is_noop() {
+    check(
+        indoc! {"
+        namespace Test {
+            @EntryPoint()
+            operation Test() : Result {
+                use q = Qubit();
+                R(PauliI, 1.0, q);
+                return MResetZ(q);
+            }
+        }
+        "},
+        None,
+        &expect![[r#"
+            %Result = type opaque
+            %Qubit = type opaque
+
+            define void @ENTRYPOINT__main() #0 {
+              call void @__quantum__qis__mz__body(%Qubit* inttoptr (i64 0 to %Qubit*), %Result* inttoptr (i64 0 to %Result*)) #1
+              call void @__quantum__rt__result_record_output(%Result* inttoptr (i64 0 to %Result*), i8* null)
+              ret void
+            }
+
+            declare void @__quantum__qis__ccx__body(%Qubit*, %Qubit*, %Qubit*)
+            declare void @__quantum__qis__cx__body(%Qubit*, %Qubit*)
+            declare void @__quantum__qis__cy__body(%Qubit*, %Qubit*)
+            declare void @__quantum__qis__cz__body(%Qubit*, %Qubit*)
+            declare void @__quantum__qis__rx__body(double, %Qubit*)
+            declare void @__quantum__qis__rxx__body(double, %Qubit*, %Qubit*)
+            declare void @__quantum__qis__ry__body(double, %Qubit*)
+            declare void @__quantum__qis__ryy__body(double, %Qubit*, %Qubit*)
+            declare void @__quantum__qis__rz__body(double, %Qubit*)
+            declare void @__quantum__qis__rzz__body(double, %Qubit*, %Qubit*)
+            declare void @__quantum__qis__h__body(%Qubit*)
+            declare void @__quantum__qis__s__body(%Qubit*)
+            declare void @__quantum__qis__s__adj(%Qubit*)
+            declare void @__quantum__qis__t__body(%Qubit*)
+            declare void @__quantum__qis__t__adj(%Qubit*)
+            declare void @__quantum__qis__x__body(%Qubit*)
+            declare void @__quantum__qis__y__body(%Qubit*)
+            declare void @__quantum__qis__z__body(%Qubit*)
+            declare void @__quantum__qis__swap__body(%Qubit*, %Qubit*)
+            declare void @__quantum__qis__mz__body(%Qubit*, %Result* writeonly) #1
+            declare void @__quantum__rt__result_record_output(%Result*, i8*)
+            declare void @__quantum__rt__array_record_output(i64, i8*)
+            declare void @__quantum__rt__tuple_record_output(i64, i8*)
+
+            attributes #0 = { "entry_point" "output_labeling_schema" "qir_profiles"="base_profile" "required_num_qubits"="1" "required_num_results"="1" }
+            attributes #1 = { "irreversible" }
+
+            ; module flags
+
+            !llvm.module.flags = !{!0, !1, !2, !3}
+
+            !0 = !{i32 1, !"qir_major_version", i32 1}
+            !1 = !{i32 7, !"qir_minor_version", i32 0}
+            !2 = !{i32 1, !"dynamic_qubit_management", i1 false}
+            !3 = !{i32 1, !"dynamic_result_management", i1 false}
+        "#]],
+    );
+}

compiler/qsc_eval/src/backend.rs

@@ -264,8 +264,28 @@ impl Backend for SparseSim {
         self.sim.qubit_is_zero(q)
     }
 
-    fn custom_intrinsic(&mut self, name: &str, _arg: Value) -> Option<Result<Value, String>> {
+    fn custom_intrinsic(&mut self, name: &str, arg: Value) -> Option<Result<Value, String>> {
         match name {
+            "GlobalPhase" => {
+                // Apply a global phase to the simulation by doing an Rz to a fresh qubit.
+                // The controls list may be empty, in which case the phase is applied unconditionally.
+                let [ctls_val, theta] = &*arg.unwrap_tuple() else {
+                    panic!("tuple arity for GlobalPhase intrinsic should be 2");
+                };
+                let ctls = ctls_val
+                    .clone()
+                    .unwrap_array()
+                    .iter()
+                    .map(|q| q.clone().unwrap_qubit().0)
+                    .collect::<Vec<_>>();
+                let q = self.sim.allocate();
+                // The new qubit is by-definition in the |0⟩ state, so by reversing the sign of the
+                // angle we can apply the phase to the entire state without increasing its size in memory.
+                self.sim
+                    .mcrz(&ctls, -2.0 * theta.clone().unwrap_double(), q);
+                self.sim.release(q);
+                Some(Ok(Value::unit()))
+            }
             "BeginEstimateCaching" => Some(Ok(Value::Bool(true))),
             "EndEstimateCaching"
             | "AccountForEstimatesInternal"

compiler/qsc_eval/src/intrinsic.rs

@@ -53,7 +53,7 @@ pub(crate) fn call(
                 return Err(Error::QubitUniqueness(arg_span));
             }
             let (state, qubit_count) = sim.capture_quantum_state();
-            let state = utils::split_state(&qubits, state, qubit_count)
+            let state = utils::split_state(&qubits, &state, qubit_count)
                 .map_err(|()| Error::QubitsNotSeparable(arg_span))?;
             match out.state(state, qubits.len()) {
                 Ok(()) => Ok(Value::unit()),

compiler/qsc_eval/src/intrinsic/tests.rs

@@ -465,6 +465,45 @@ fn dump_register_qubits_not_unique_fails() {
     );
 }
 
+#[test]
+fn dump_register_target_in_minus_with_other_in_zero() {
+    check_intrinsic_output(
+        "",
+        indoc! {"{
+            use qs = Qubit[2];
+            X(qs[0]);
+            H(qs[0]);
+            Microsoft.Quantum.Diagnostics.DumpRegister([qs[0]]);
+            ResetAll(qs);
+        }"},
+        &expect![[r#"
+            STATE:
+            |0⟩: 0.7071+0.0000𝑖
+            |1⟩: −0.7071+0.0000𝑖
+        "#]],
+    );
+}
+
+#[test]
+fn dump_register_target_in_minus_with_other_in_one() {
+    check_intrinsic_output(
+        "",
+        indoc! {"{
+            use qs = Qubit[2];
+            X(qs[1]);
+            X(qs[0]);
+            H(qs[0]);
+            Microsoft.Quantum.Diagnostics.DumpRegister([qs[0]]);
+            ResetAll(qs);
+        }"},
+        &expect![[r#"
+            STATE:
+            |0⟩: 0.7071+0.0000𝑖
+            |1⟩: −0.7071+0.0000𝑖
+        "#]],
+    );
+}
+
 #[test]
 fn message() {
     check_intrinsic_output(

compiler/qsc_eval/src/intrinsic/utils.rs

@@ -12,10 +12,9 @@ use rustc_hash::FxHashMap;
 /// This function will return an error if the state is not separable using the provided qubit identifiers.
 pub fn split_state(
     qubits: &[usize],
-    state: Vec<(BigUint, Complex64)>,
+    state: &[(BigUint, Complex64)],
     qubit_count: usize,
 ) -> Result<Vec<(BigUint, Complex64)>, ()> {
-    let state = state.into_iter().collect::<FxHashMap<_, _>>();
     let mut dump_state = FxHashMap::default();
     let mut other_state = FxHashMap::default();
 
@@ -25,7 +24,7 @@ pub fn split_state(
     // Try to split out the state for the given qubits from the whole state, detecting any entanglement
     // and returning an error if the qubits are not separable.
     let dump_norm = collect_split_state(
-        &state,
+        state,
         &dump_mask,
         &other_mask,
         &mut dump_state,
@@ -70,13 +69,16 @@ fn compute_mask(qubit_count: usize, qubits: &[usize]) -> (BigUint, BigUint) {
 /// On success, the `dump_state` and `other_state` maps will be populated with the separated states, and the
 /// function returns the accumulated norm of the dump state.
 fn collect_split_state(
-    state: &FxHashMap<BigUint, Complex64>,
+    state: &[(BigUint, Complex64)],
     dump_mask: &BigUint,
     other_mask: &BigUint,
     dump_state: &mut FxHashMap<BigUint, Complex64>,
     other_state: &mut FxHashMap<BigUint, Complex64>,
 ) -> Result<f64, ()> {
+    // To ensure consistent ordering, we iterate over the vector directly (returned from the simulator in a deterministic order),
+    // and not the map used for arbitrary lookup.
     let mut state_iter = state.iter();
+    let state_map = state.iter().cloned().collect::<FxHashMap<_, _>>();
     let (base_label, base_val) = state_iter.next().expect("state should never be empty");
     let dump_base_label = base_label & dump_mask;
     let other_base_label = base_label & other_mask;
@@ -93,10 +95,10 @@ fn collect_split_state(
 
         // If either the state identified by the dump mask or the state identified by the other mask
         // is None, that means it has zero amplitude and we can conclude the state is not separable.
-        let Some(dump_val) = state.get(&(&dump_label | &other_base_label)) else {
+        let Some(dump_val) = state_map.get(&(&dump_label | &other_base_label)) else {
             return Err(());
         };
-        let Some(other_val) = state.get(&(&dump_base_label | &other_label)) else {
+        let Some(other_val) = state_map.get(&(&dump_base_label | &other_label)) else {
             return Err(());
         };
 

compiler/qsc_partial_eval/src/lib.rs

@@ -548,7 +548,8 @@ impl<'a> PartialEvaluator<'a> {
             "DumpRegister"
             | "AccountForEstimatesInternal"
             | "BeginRepeatEstimatesInternal"
-            | "EndRepeatEstimatesInternal" => Value::unit(),
+            | "EndRepeatEstimatesInternal"
+            | "GlobalPhase" => Value::unit(),
             _ => self.eval_expr_call_to_intrinsic_qis(store_item_id, callable_decl, args_value),
         }
     }

compiler/qsc_partial_eval/src/management.rs

@@ -105,7 +105,7 @@ impl Backend for QuantumIntrinsicsChecker {
     ) -> Option<std::result::Result<Value, String>> {
         match name {
             "BeginEstimateCaching" => Some(Ok(Value::Bool(true))),
-            "EndEstimateCaching" => Some(Ok(Value::unit())),
+            "EndEstimateCaching" | "GlobalPhase" => Some(Ok(Value::unit())),
             _ => None,
         }
     }

compiler/qsc_partial_eval/tests/intrinsics.rs

@@ -973,3 +973,26 @@ fn call_to_draw_random_double_panics() {
         "#,
     });
 }
+
+#[test]
+fn call_to_pauli_i_rotation_for_global_phase_is_noop() {
+    let program = get_rir_program(indoc! {
+        r#"
+        namespace Test {
+            @EntryPoint()
+            operation Main() : Unit {
+                use q = Qubit();
+                R(PauliI, 1.0, q);
+            }
+        }
+        "#,
+    });
+    assert_block_instructions(
+        &program,
+        BlockId(0),
+        &expect![[r#"
+        Block:
+            Call id(1), args( Integer(0), Pointer, )
+            Return"#]],
+    );
+}