From 75193850d83285b6b451d2a34406cc61e6dc587e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Luis=20Alfredo=20Nu=C3=B1ez=20Meneses?=
 <alfredo.nunez@xanadu.ai>
Date: Fri, 3 Jan 2025 13:07:26 -0500
Subject: [PATCH] added SX to LK

---
 .../core/src/gates/Constant.hpp               |  2 +-
 pennylane_lightning/core/src/gates/Gates.hpp  |  7 +--
 .../gates/BasicGateFunctors.hpp               | 51 +++++++++++++++++++
 .../cpu_kernels/GateImplementationsLM.hpp     |  4 +-
 4 files changed, 58 insertions(+), 6 deletions(-)

diff --git a/pennylane_lightning/core/src/gates/Constant.hpp b/pennylane_lightning/core/src/gates/Constant.hpp
index 0cb8d98c01..fa850dfafb 100644
--- a/pennylane_lightning/core/src/gates/Constant.hpp
+++ b/pennylane_lightning/core/src/gates/Constant.hpp
@@ -368,8 +368,8 @@ using CGateNParams = typename std::pair<ControlledGateOperation, std::size_t>;
     CGateNParams{ControlledGateOperation::Hadamard, 0},
     CGateNParams{ControlledGateOperation::S, 0},
     CGateNParams{ControlledGateOperation::T, 0},
-    CGateNParams{ControlledGateOperation::PhaseShift, 1},
     CGateNParams{ControlledGateOperation::SX, 0},
+    CGateNParams{ControlledGateOperation::PhaseShift, 1},
     CGateNParams{ControlledGateOperation::RX, 1},
     CGateNParams{ControlledGateOperation::RY, 1},
     CGateNParams{ControlledGateOperation::RZ, 1},
diff --git a/pennylane_lightning/core/src/gates/Gates.hpp b/pennylane_lightning/core/src/gates/Gates.hpp
index 4fee37da40..65bbf0971c 100644
--- a/pennylane_lightning/core/src/gates/Gates.hpp
+++ b/pennylane_lightning/core/src/gates/Gates.hpp
@@ -110,9 +110,10 @@ static constexpr auto getHadamard() -> std::vector<ComplexT<T>> {
  * of SX data.
  */
 template <template <typename...> class ComplexT, typename T>
-static constexpr auto getSX() -> std::vector<ComplexT<T>> {
-    constexpr ComplexT<T> z0{0.5, 0.5};
-    constexpr ComplexT<T> z1{0.5, -0.5};
+static constexpr auto getSX(const bool inverse = false) -> std::vector<ComplexT<T>> {
+    const T half = (inverse) ? -0.5 : 0.5;
+    const ComplexT<T> z0{0.5, half};
+    const ComplexT<T> z1{0.5,-half};
     return {z0, z1, z1, z0};
 }
 
diff --git a/pennylane_lightning/core/src/simulators/lightning_kokkos/gates/BasicGateFunctors.hpp b/pennylane_lightning/core/src/simulators/lightning_kokkos/gates/BasicGateFunctors.hpp
index 5f44391ee0..9f1c356f38 100644
--- a/pennylane_lightning/core/src/simulators/lightning_kokkos/gates/BasicGateFunctors.hpp
+++ b/pennylane_lightning/core/src/simulators/lightning_kokkos/gates/BasicGateFunctors.hpp
@@ -331,6 +331,50 @@ void applyS(Kokkos::View<Kokkos::complex<PrecisionT> *> arr_,
                                          inverse);
 }
 
+template <class ExecutionSpace, class PrecisionT>
+void applyNCSX(Kokkos::View<Kokkos::complex<PrecisionT> *> arr_,
+              std::size_t num_qubits,
+              const std::vector<std::size_t> &controlled_wires,
+              const std::vector<bool> &controlled_values,
+              const std::vector<std::size_t> &wires, bool inverse = false,
+              [[maybe_unused]] const std::vector<PrecisionT> &params = {})
+{
+    constexpr PrecisionT half = 0.5;
+    const Kokkos::complex<PrecisionT> z0{half, (inverse) ? -half : half};
+    const Kokkos::complex<PrecisionT> z1 = Kokkos::conj(z0);
+
+    auto core_function =
+        KOKKOS_LAMBDA(Kokkos::View<Kokkos::complex<PrecisionT> *> arr,
+                      std::size_t i0, std::size_t i1)
+    {
+        const Kokkos::complex<PrecisionT> v0 = arr(i0);
+        const Kokkos::complex<PrecisionT> v1 = arr(i1);
+        arr[i0] = z0 * v0 + z1 * v1;
+        arr[i1] = z1 * v0 + z0 * v1;
+    };
+    if (controlled_wires.empty())
+    {
+        applyNC1Functor<PrecisionT, decltype(core_function), false>(
+            ExecutionSpace{}, arr_, num_qubits, wires, core_function);
+    }
+    else
+    {
+        applyNC1Functor<PrecisionT, decltype(core_function), true>(
+            ExecutionSpace{}, arr_, num_qubits, controlled_wires,
+            controlled_values, wires, core_function);
+    }
+}
+
+template <class ExecutionSpace, class PrecisionT>
+void applySX(Kokkos::View<Kokkos::complex<PrecisionT> *> arr_,
+            std::size_t num_qubits, const std::vector<std::size_t> &wires,
+            bool inverse = false,
+            [[maybe_unused]] const std::vector<PrecisionT> &params = {})
+{
+    applyNCSX<ExecutionSpace, PrecisionT>(arr_, num_qubits, {}, {}, wires,
+                                         inverse);
+}
+
 template <class ExecutionSpace, class PrecisionT>
 void applyNCT(Kokkos::View<Kokkos::complex<PrecisionT> *> arr_,
               std::size_t num_qubits,
@@ -1821,6 +1865,9 @@ void applyNamedOperation(const GateOperation gateop,
     case GateOperation::S:
         applyS<ExecutionSpace>(arr_, num_qubits, wires, inverse, params);
         return;
+    case GateOperation::SX:
+        applySX<ExecutionSpace>(arr_, num_qubits, wires, inverse, params);
+        return;
     case GateOperation::T:
         applyT<ExecutionSpace>(arr_, num_qubits, wires, inverse, params);
         return;
@@ -1956,6 +2003,10 @@ void applyNCNamedOperation(const ControlledGateOperation gateop,
         applyNCS<ExecutionSpace>(arr_, num_qubits, controlled_wires,
                                  controlled_values, wires, inverse, params);
         return;
+    case ControlledGateOperation::SX:
+        applyNCSX<ExecutionSpace>(arr_, num_qubits, controlled_wires,
+                                 controlled_values, wires, inverse, params);
+        return;
     case ControlledGateOperation::T:
         applyNCT<ExecutionSpace>(arr_, num_qubits, controlled_wires,
                                  controlled_values, wires, inverse, params);
diff --git a/pennylane_lightning/core/src/simulators/lightning_qubit/gates/cpu_kernels/GateImplementationsLM.hpp b/pennylane_lightning/core/src/simulators/lightning_qubit/gates/cpu_kernels/GateImplementationsLM.hpp
index 79b97de218..a3fc9f8e04 100644
--- a/pennylane_lightning/core/src/simulators/lightning_qubit/gates/cpu_kernels/GateImplementationsLM.hpp
+++ b/pennylane_lightning/core/src/simulators/lightning_qubit/gates/cpu_kernels/GateImplementationsLM.hpp
@@ -869,9 +869,9 @@ class GateImplementationsLM : public PauliGenerator<GateImplementationsLM> {
     template <class PrecisionT>
     static void
     applyNCSX(std::complex<PrecisionT> *arr, const size_t num_qubits,
-              const std::vector<size_t> &controlled_wires,
+              const std::vector<std::size_t> &controlled_wires,
               const std::vector<bool> &controlled_values,
-              const std::vector<size_t> &wires, const bool inverse) {
+              const std::vector<std::size_t> &wires, const bool inverse) {
         using ParamT = PrecisionT;
 
         constexpr PrecisionT half = 0.5;