Further SIMD optimization and thread number optimization for Pauli operations

benchmark/libcsim_benchmark.cpp

@@ -186,7 +186,7 @@ double benchmark_state_stat_func(UINT n, std::function<T(CTYPE*, ITYPE)>func) {
 };
 
 double benchmark_state_norm(UINT n) {
-	return benchmark_state_func(n, state_norm);
+	return benchmark_state_func(n, state_norm_squared);
 }
 
 double benchmark_gate_single_func(UINT n, UINT target, std::function<void(UINT, CTYPE*, ITYPE)>func) {

python/cppsim_wrapper.cpp

@@ -491,7 +491,8 @@ PYBIND11_MODULE(qulacs, m) {
     py::class_<QuantumCircuitOptimizer>(mcircuit, "QuantumCircuitOptimizer")
         .def(py::init<>(), "Constructor")
         .def("optimize", &QuantumCircuitOptimizer::optimize, "Optimize quantum circuit", py::arg("circuit"), py::arg("block_size"))
-        .def("merge_all", &QuantumCircuitOptimizer::merge_all, pybind11::return_value_policy::take_ownership, py::arg("circuit"))
+		.def("optimize_light", &QuantumCircuitOptimizer::optimize_light, "Optimize quantum circuit with light method", py::arg("circuit"))
+		.def("merge_all", &QuantumCircuitOptimizer::merge_all, pybind11::return_value_policy::take_ownership, py::arg("circuit"))
         ;
 
     py::class_<QuantumCircuitSimulator>(m, "QuantumCircuitSimulator")

src/cppsim/circuit_optimizer.cpp

@@ -104,6 +104,82 @@ void QuantumCircuitOptimizer::optimize(QuantumCircuit* circuit_, UINT max_block_
         }
     }
 }
+/*
+void QuantumCircuitOptimizer::optimize_light(QuantumCircuit* circuit) {
+	UINT qubit_count = circuit->qubit_count;
+	std::vector<std::pair<int, std::vector<UINT>>> current_step(qubit_count, std::make_pair(-1,std::vector<UINT>()));
+	for (UINT ind1 = 0; ind1 < circuit->gate_list.size(); ++ind1) {
+		QuantumGateBase* gate = circuit->gate_list[ind1];
+		std::vector<UINT> target_qubits;
+		for (auto val : gate->get_target_index_list()) target_qubits.push_back(val);
+		for (auto val : gate->get_control_index_list()) target_qubits.push_back(val);
+		std::sort(target_qubits.begin(), target_qubits.end());
+		if (target_qubits.size() == 1) {
+			// merge
+			UINT target_qubit_index = target_qubits[0];
+			UINT target_gate_index = current_step[target_qubit_index].first;
+			if (target_gate_index != -1) {
+				auto merged_gate = gate::merge(circuit->gate_list[target_gate_index], gate);
+				circuit->remove_gate(ind1);
+				circuit->add_gate(merged_gate, target_gate_index+1);
+				circuit->remove_gate(target_gate_index);
+				ind1--;
+			}
+			else {
+				current_step[target_qubit_index] = std::make_pair(ind1, target_qubits);
+			}
+		}
+		else {
+			for (auto target_qubit : target_qubits) {
+				current_step[target_qubit] = make_pair(ind1, target_qubits);
+			}
+		}
+	}
+}
+*/
+
+void QuantumCircuitOptimizer::optimize_light(QuantumCircuit* circuit) {
+	UINT qubit_count = circuit->qubit_count;
+	std::vector<std::pair<int, std::vector<UINT>>> current_step(qubit_count, std::make_pair(-1, std::vector<UINT>()));
+	for (UINT ind1 = 0; ind1 < circuit->gate_list.size(); ++ind1) {
+		QuantumGateBase* gate = circuit->gate_list[ind1];
+		std::vector<UINT> target_qubits;
+		std::vector<UINT> parent_qubits;
+
+		for (auto val : gate->get_target_index_list()) target_qubits.push_back(val);
+		for (auto val : gate->get_control_index_list()) target_qubits.push_back(val);
+		std::sort(target_qubits.begin(), target_qubits.end());
+
+		int pos = -1;
+		int hit = -1;
+		for (UINT target_qubit : target_qubits) {
+			if (current_step[target_qubit].first > pos) {
+				pos = current_step[target_qubit].first;
+				hit = target_qubit;
+			}
+		}
+		if(hit!=-1)
+			parent_qubits = current_step[hit].second;
+		if (std::includes(parent_qubits.begin(), parent_qubits.end(), target_qubits.begin(), target_qubits.end())) {
+			auto merged_gate = gate::merge(circuit->gate_list[pos], gate);
+			circuit->remove_gate(ind1);
+			circuit->add_gate(merged_gate, pos + 1);
+			circuit->remove_gate(pos);
+			ind1--;
+
+			//std::cout << "merge ";
+			//for (auto val : target_qubits) std::cout << val << " ";
+			//std::cout << " into ";
+			//for (auto val : parent_qubits) std::cout << val << " ";
+			//std::cout << std::endl;
+		}
+		else {
+			for (auto target_qubit : target_qubits) {
+				current_step[target_qubit] = make_pair(ind1, target_qubits);
+			}
+		}
+	}
+}
 
 
 QuantumGateMatrix* QuantumCircuitOptimizer::merge_all(const QuantumCircuit* circuit_) {

src/cppsim/circuit_optimizer.hpp

@@ -43,6 +43,17 @@ class DllExport QuantumCircuitOptimizer{
      */
     void optimize(QuantumCircuit* circuit, UINT max_block_size=2);
 
+	/**
+	 * \~japanese-en 与えられた量子回路のゲートを指定されたブロックまで纏める。
+	 *
+	 * 与えられた量子回路において、若い添え字から任意の二つのゲートを選び、二つが他のゲートに影響を与えず合成可能なら合成を行う。
+	 * これを合成可能なペアがなくなるまで繰り返す。
+	 * 二つのゲートが合成可能であるとは、二つのゲートそれぞれについて隣接するゲートとの交換を繰り返し、二つのゲートが隣接した位置まで移動できることを指す。
+	 *
+	 * @param[in] circuit 量子回路のインスタンス
+	 */
+	void optimize_light(QuantumCircuit* circuit);
+
     /**
      * \~japanese-en 量子回路を纏めて一つの巨大な量子ゲートにする
      *