Program Listing for File state.hpp

Return to documentation for file (/home/docs/checkouts/readthedocs.org/user_builds/qulacs-rtd/checkouts/latest/src/cppsim/state.hpp)

#pragma once

#include <csim/init_ops.hpp>
#include <csim/memory_ops.hpp>
#include <csim/stat_ops.hpp>
#include <csim/update_ops.hpp>
#include <iostream>
#include <vector>

#include "csim/MPIutil.hpp"
#include "exception.hpp"
#include "type.hpp"
#include "utility.hpp"

class QuantumStateBase {
protected:
    ITYPE _dim;
    UINT _qubit_count;
    UINT _inner_qc;
    UINT _outer_qc;
    bool _is_state_vector;
    std::vector<UINT> _classical_register;
    UINT _device_number;
    void* _cuda_stream;

public:
    const UINT& qubit_count;
    const UINT& inner_qc;
    const UINT& outer_qc;
    const ITYPE& dim;
    const std::vector<UINT>&
        classical_register;
    const UINT& device_number;
    QuantumStateBase(UINT qubit_count_, bool is_state_vector)
        : qubit_count(_qubit_count),
          inner_qc(_inner_qc),
          outer_qc(_outer_qc),
          dim(_dim),
          classical_register(_classical_register),
          device_number(_device_number) {
        this->_qubit_count = qubit_count_;
        this->_inner_qc = qubit_count_;
        this->_outer_qc = 0;
        this->_dim = 1ULL << qubit_count_;
        this->_is_state_vector = is_state_vector;
        this->_device_number = 0;
    }

    // 本来、QuantumStateBase(UINT, bool, bool) とすべきだが、
    // QuantumStateBase(UINT, bool, UINT) に対し ambiguous error が発生するため
    // QuantumStateBase(UINT, bool, int) とした
    QuantumStateBase(UINT qubit_count_, bool is_state_vector, int use_multi_cpu)
        : qubit_count(_qubit_count),
          inner_qc(_inner_qc),
          outer_qc(_outer_qc),
          dim(_dim),
          classical_register(_classical_register),
          device_number(_device_number) {
        UINT mpirank;
        UINT mpisize;
        if (use_multi_cpu) {
#ifdef _USE_MPI
            MPIutil& mpiutil = MPIutil::get_inst();
            mpirank = mpiutil.get_rank();
            mpisize = mpiutil.get_size();
#else
            mpirank = 0;
            mpisize = 1;
#endif
        } else {
            mpirank = 0;
            mpisize = 1;
        }
        if ((mpisize & (mpisize - 1))) {
            throw MPISizeException(
                "Error: QuantumStateBase::QuantumStateBase(UINT, bool, bool): "
                "mpi-size must be power of 2");
        }
        UINT log_nodes = std::log2(mpisize);
        if (use_multi_cpu &&
            (qubit_count_ >= (log_nodes + 2))) {  // minimum inner_qc=2
            this->_inner_qc = qubit_count_ - log_nodes;
            this->_outer_qc = log_nodes;
        } else {
            this->_inner_qc = qubit_count_;
            this->_outer_qc = 0;
        }

        this->_qubit_count = qubit_count_;
        this->_dim = 1ULL << this->_inner_qc;
        this->_is_state_vector = is_state_vector;
        this->_device_number = mpirank;
    }

    QuantumStateBase(
        UINT qubit_count_, bool is_state_vector, UINT device_number_)
        : qubit_count(_qubit_count),
          inner_qc(_inner_qc),
          outer_qc(_outer_qc),
          dim(_dim),
          classical_register(_classical_register),
          device_number(_device_number) {
        this->_qubit_count = qubit_count_;
        this->_inner_qc = qubit_count_;
        this->_outer_qc = 0;
        this->_dim = 1ULL << qubit_count_;
        this->_is_state_vector = is_state_vector;
        this->_device_number = device_number_;
    }
    virtual ~QuantumStateBase() {}

    virtual bool is_state_vector() const { return this->_is_state_vector; }

    virtual void set_zero_state() = 0;

    virtual void set_zero_norm_state() = 0;

    virtual void set_computational_basis(ITYPE comp_basis) = 0;

    virtual void set_Haar_random_state() = 0;

    virtual void set_Haar_random_state(UINT seed) = 0;

    virtual double get_zero_probability(UINT target_qubit_index) const = 0;

    virtual double get_marginal_probability(
        std::vector<UINT> measured_values) const = 0;

    virtual double get_entropy() const = 0;

    virtual double get_squared_norm() const = 0;

    virtual double get_squared_norm_single_thread() const = 0;

    virtual void normalize(double squared_norm) = 0;

    virtual void normalize_single_thread(double squared_norm) = 0;

    virtual QuantumStateBase* allocate_buffer() const = 0;

    virtual QuantumStateBase* copy() const = 0;

    virtual void load(const QuantumStateBase* state) = 0;

    virtual void load(const std::vector<CPPCTYPE>& state) = 0;

    virtual void load(const CPPCTYPE* state) = 0;

    virtual const std::string get_device_name() const = 0;

    virtual void* data() const = 0;

    virtual CPPCTYPE* data_cpp() const = 0;

    virtual CTYPE* data_c() const = 0;

    virtual CPPCTYPE* duplicate_data_cpp() const = 0;

    virtual CTYPE* duplicate_data_c() const = 0;

    virtual void add_state(const QuantumStateBase* state) = 0;

    virtual void add_state_with_coef(
        CPPCTYPE coef, const QuantumStateBase* state) = 0;

    virtual void add_state_with_coef_single_thread(
        CPPCTYPE coef, const QuantumStateBase* state) = 0;

    virtual void multiply_coef(CPPCTYPE coef) = 0;

    virtual void multiply_elementwise_function(
        const std::function<CPPCTYPE(ITYPE)>& func) = 0;

    virtual UINT get_classical_value(UINT index) {
        if (_classical_register.size() <= index) {
            _classical_register.resize(index + 1, 0);
        }
        return _classical_register[index];
    }

    virtual void set_classical_value(UINT index, UINT val) {
        if (_classical_register.size() <= index) {
            _classical_register.resize(index + 1, 0);
        }
        _classical_register[index] = val;
    }

    virtual const std::vector<UINT> get_classical_register() const {
        return _classical_register;
    }

    virtual std::vector<ITYPE> sampling(UINT sampling_count) = 0;
    virtual std::vector<ITYPE> sampling(
        UINT sampling_count, UINT random_seed) = 0;

    virtual boost::property_tree::ptree to_ptree() const = 0;

    virtual std::string to_string() const {
        std::stringstream os;
        ComplexVector eigen_state(this->dim);
        auto data = this->data_cpp();
        for (ITYPE i = 0; i < this->dim; ++i) eigen_state[i] = data[i];

        os << " *** Quantum State ***" << std::endl;
        UINT myrank = 0;
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            MPIutil& mpiutil = MPIutil::get_inst();
            myrank = mpiutil.get_rank();
        }
#endif
        if (myrank == 0) {
            os << " * Qubit Count : " << this->qubit_count << std::endl;
            os << " * Dimension   : " << this->dim << std::endl;
#ifdef _USE_MPI
            os << " * Local Qubit Count : " << this->inner_qc << std::endl;
            os << " * Global Qubit Count : " << this->outer_qc << std::endl;
#endif
        }
        if (this->outer_qc > 0) {
            os << " * Rank : " << myrank << std::endl;
        }
        os << " * State vector : \n" << eigen_state << std::endl;
        return os.str();
    }

    friend std::ostream& operator<<(
        std::ostream& os, const QuantumStateBase& state) {
        os << state.to_string();
        return os;
    }

    friend std::ostream& operator<<(
        std::ostream& os, const QuantumStateBase* state) {
        os << *state;
        return os;
    }

    virtual void* get_cuda_stream() const { return this->_cuda_stream; }
};

class QuantumStateCpu : public QuantumStateBase {
private:
    CPPCTYPE* _state_vector;
    Random random;

public:
    explicit QuantumStateCpu(UINT qubit_count_)
        : QuantumStateBase(qubit_count_, true) {
        this->_state_vector =
            reinterpret_cast<CPPCTYPE*>(allocate_quantum_state(this->_dim));
        initialize_quantum_state(this->data_c(), _dim);
    }

    explicit QuantumStateCpu(UINT qubit_count_, bool use_multi_cpu)
        : QuantumStateBase(qubit_count_, true, (int)use_multi_cpu) {
        this->_state_vector =
            reinterpret_cast<CPPCTYPE*>(allocate_quantum_state(this->_dim));
#ifdef _USE_MPI
        if (this->outer_qc > 0)
            initialize_quantum_state_mpi(this->data_c(), _dim, this->outer_qc);
        else
#endif
        {
            initialize_quantum_state(this->data_c(), _dim);
        }
    }

    virtual ~QuantumStateCpu() {
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            MPIutil& mpiutil = MPIutil::get_inst();
            mpiutil.release_workarea();
        }
#endif
        release_quantum_state(this->data_c());
    }

    virtual void set_zero_state() override {
#ifdef _USE_MPI
        initialize_quantum_state_mpi(this->data_c(), _dim, this->outer_qc);
#else
        initialize_quantum_state(this->data_c(), _dim);
#endif
    }

    virtual void set_zero_norm_state() override {
        set_zero_state();
        _state_vector[0] = 0;
    }

    virtual void set_computational_basis(ITYPE comp_basis) override {
        if (comp_basis >= (ITYPE)(1ULL << this->qubit_count)) {
            throw MatrixIndexOutOfRangeException(
                "Error: QuantumStateCpu::set_computational_basis(ITYPE): "
                "index of "
                "computational basis must be smaller than 2^qubit_count");
        }
        set_zero_state();
        _state_vector[0] = 0.;
#ifdef _USE_MPI
        ITYPE myrank = 0;
        if (this->outer_qc > 0) {
            MPIutil& mpiutil = MPIutil::get_inst();
            myrank = (ITYPE)mpiutil.get_rank();
        }
        if (this->outer_qc == 0 || (comp_basis >> this->inner_qc) == myrank) {
            _state_vector[comp_basis & (this->_dim - 1)] = 1.;
        }
#else
        _state_vector[comp_basis] = 1.;
#endif
    }
    virtual void set_Haar_random_state() override {
        UINT seed = random.int32();
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            // すべてのrankで同一の結果を得るために、seedを共有する
            MPIutil& mpiutil = MPIutil::get_inst();
            if (mpiutil.get_size() > 1) mpiutil.s_u_bcast(&seed);
        }
#endif
        set_Haar_random_state(seed);
    }
    virtual void set_Haar_random_state(UINT seed) override {
#ifdef _USE_MPI
        // 各rankで異なるseedを用いる必要がある
        if (this->outer_qc > 0) {
            MPIutil& mpiutil = MPIutil::get_inst();
            seed += mpiutil.get_rank();
        }
        initialize_Haar_random_state_mpi_with_seed(
            this->data_c(), _dim, this->outer_qc, seed);
#else
        initialize_Haar_random_state_with_seed(this->data_c(), _dim, seed);
#endif
    }
    virtual double get_zero_probability(
        UINT target_qubit_index) const override {
#ifdef _USE_MPI
        if (this->outer_qc > 0)
            throw NotImplementedException(
                "Error: get_zero_probability does not support multi-cpu");
#endif
        if (target_qubit_index >= this->qubit_count) {
            throw QubitIndexOutOfRangeException(
                "Error: QuantumStateCpu::get_zero_probability(UINT): index "
                "of target qubit must be smaller than qubit_count");
        }
        return M0_prob(target_qubit_index, this->data_c(), _dim);
    }
    virtual double get_marginal_probability(
        std::vector<UINT> measured_values) const override {
#ifdef _USE_MPI
        if (this->outer_qc > 0)
            throw NotImplementedException(
                "Error: get_marginal_probability does not support multi-cpu");
#endif
        if (measured_values.size() != this->qubit_count) {
            throw InvalidQubitCountException(
                "Error: "
                "QuantumStateCpu::get_marginal_probability(vector<UINT>): "
                "the length of measured_values must be equal to qubit_count");
        }

        std::vector<UINT> target_index;
        std::vector<UINT> target_value;
        for (UINT i = 0; i < measured_values.size(); ++i) {
            UINT measured_value = measured_values[i];
            if (measured_value == 0 || measured_value == 1) {
                target_index.push_back(i);
                target_value.push_back(measured_value);
            }
        }
        return marginal_prob(target_index.data(), target_value.data(),
            (UINT)target_index.size(), this->data_c(), _dim);
    }
    virtual double get_entropy() const override {
        double entropy = measurement_distribution_entropy(this->data_c(), _dim);
#ifdef _USE_MPI
        MPIutil& mpiutil = MPIutil::get_inst();
        if (this->outer_qc > 0) mpiutil.s_D_allreduce(&entropy);
#endif
        return entropy;
    }

    virtual double get_squared_norm() const override {
        double norm;
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            norm = state_norm_squared_mpi(this->data_c(), _dim);
        } else
#endif
        {
            norm = state_norm_squared(this->data_c(), _dim);
        }
        return norm;
    }

    virtual double get_squared_norm_single_thread() const override {
        return state_norm_squared_single_thread(this->data_c(), _dim);
    }

    virtual void normalize(double squared_norm) override {
        ::normalize(squared_norm, this->data_c(), _dim);
    }

    virtual void normalize_single_thread(double squared_norm) override {
        ::normalize_single_thread(squared_norm, this->data_c(), _dim);
    }

    virtual QuantumStateCpu* allocate_buffer() const override {
        QuantumStateCpu* new_state;
#ifdef _USE_MPI
        if (this->outer_qc > 0)
            new_state = new QuantumStateCpu(this->_qubit_count, true);
        else
            new_state = new QuantumStateCpu(this->_qubit_count, false);
#else
        new_state = new QuantumStateCpu(this->_qubit_count);
#endif
        return new_state;
    }
    virtual QuantumStateCpu* copy() const override {
        QuantumStateCpu* new_state = this->allocate_buffer();

        memcpy(new_state->data_cpp(), _state_vector,
            (size_t)(sizeof(CPPCTYPE) * _dim));
        for (UINT i = 0; i < _classical_register.size(); ++i) {
            new_state->set_classical_value(i, _classical_register[i]);
        }

        return new_state;
    }
    virtual void load(const QuantumStateBase* _state) override {
        if (_state->qubit_count != this->qubit_count) {
            throw InvalidQubitCountException(
                "Error: QuantumStateCpu::load(const QuantumStateBase*): "
                "invalid qubit count");
        }
        if (!_state->is_state_vector()) {
            throw InoperatableQuantumStateTypeException(
                "Error: QuantumStateCpu::load(const QuantumStateBase*): "
                "cannot load DensityMatrix to StateVector");
        }

        this->_classical_register = _state->classical_register;
        if (_state->get_device_name() == "gpu") {
            auto ptr = _state->duplicate_data_cpp();
            memcpy(this->data_cpp(), ptr, (size_t)(sizeof(CPPCTYPE) * _dim));
            free(ptr);
#ifdef _USE_MPI
        } else if (_state->outer_qc > 0) {
            MPIutil& mpiutil = MPIutil::get_inst();
            if (this->outer_qc > 0) {
                if (_state->qubit_count != this->qubit_count) {
                    throw InvalidQubitCountException(
                        "Error: QuantumStateCpu::load(const QuantumStateBase*)"
                        ": invalid global qubit count");
                }
                // load multicpu to multicpu
                memcpy(this->data_cpp(), _state->data_cpp(),
                    (size_t)(sizeof(CPPCTYPE) * _dim));
            } else {
                // load multicpu to cpu
                mpiutil.m_DC_allgather(_state->data_cpp(), this->data_cpp(),
                    _dim / mpiutil.get_size());
            }
#endif
        } else {
#ifdef _USE_MPI
            if (this->outer_qc > 0) {
                MPIutil& mpiutil = MPIutil::get_inst();
                // load cpu to multicpu
                ITYPE offs = _dim * mpiutil.get_rank();
                memcpy(this->data_cpp(), _state->data_cpp() + offs,
                    (size_t)(sizeof(CPPCTYPE) * _dim));
            } else
#endif
            {
                // load cpu to multicpu
                memcpy(this->data_cpp(), _state->data_cpp(),
                    (size_t)(sizeof(CPPCTYPE) * _dim));
            }
        }
    }
    virtual void load(const std::vector<CPPCTYPE>& _state) override {
        if (_state.size() != _dim) {
            throw InvalidStateVectorSizeException(
                "Error: QuantumStateCpu::load(vector<Complex>&): invalid "
                "length of state");
        }
        memcpy(
            this->data_cpp(), _state.data(), (size_t)(sizeof(CPPCTYPE) * _dim));
    }

    virtual void load(const CPPCTYPE* _state) override {
        memcpy(this->data_cpp(), _state, (size_t)(sizeof(CPPCTYPE) * _dim));
    }

    virtual const std::string get_device_name() const override {
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            return "multi-cpu";
        } else
#endif
        {
            return "cpu";
        }
    }

    virtual void* data() const override {
        return reinterpret_cast<void*>(this->_state_vector);
    }
    virtual CPPCTYPE* data_cpp() const override { return this->_state_vector; }
    virtual CTYPE* data_c() const override {
        return reinterpret_cast<CTYPE*>(this->_state_vector);
    }

    virtual CTYPE* duplicate_data_c() const override {
        CTYPE* new_data = (CTYPE*)malloc(sizeof(CTYPE) * _dim);
        memcpy(new_data, this->data(), (size_t)(sizeof(CTYPE) * _dim));
        return new_data;
    }

    virtual CPPCTYPE* duplicate_data_cpp() const override {
        CPPCTYPE* new_data = (CPPCTYPE*)malloc(sizeof(CPPCTYPE) * _dim);
        memcpy(new_data, this->data(), (size_t)(sizeof(CPPCTYPE) * _dim));
        return new_data;
    }

    virtual void add_state(const QuantumStateBase* state) override {
        if (!state->is_state_vector()) {
            throw InoperatableQuantumStateTypeException(
                "Error: QuantumStateCpu::add_state(const QuantumStateBase*): "
                "cannot add DensityMatrix to StateVector");
        }
        if (state->get_device_name() == "gpu") {
            throw QuantumStateProcessorException(
                "State vector on GPU cannot be added to that on CPU");
        }
        state_add(state->data_c(), this->data_c(), this->dim);
    }

    virtual void add_state_with_coef(
        CPPCTYPE coef, const QuantumStateBase* state) override {
        if (!state->is_state_vector()) {
            throw InoperatableQuantumStateTypeException(
                "Error: QuantumStateCpu::add_state_with_coef(CPPCTYPE, "
                "const QuantumStateBase*): "
                "cannot add DensityMatrix to StateVector");
        }
        if (state->get_device_name() == "gpu") {
            std::cerr << "State vector on GPU cannot be added to that on CPU"
                      << std::endl;
            return;
        }
        state_add_with_coef(coef, state->data_c(), this->data_c(), this->dim);
    }

    virtual void add_state_with_coef_single_thread(
        CPPCTYPE coef, const QuantumStateBase* state) override {
        if (!state->is_state_vector()) {
            throw InoperatableQuantumStateTypeException(
                "Error: "
                "QuantumStateCpu::add_state_with_coef_single_thread(CPPCTYPE, "
                "const QuantumStateBase*): "
                "cannot add DensityMatrix to StateVector");
        }
        if (state->get_device_name() == "gpu") {
            std::cerr << "State vector on GPU cannot be added to that on CPU"
                      << std::endl;
            return;
        }
        state_add_with_coef_single_thread(
            coef, state->data_c(), this->data_c(), this->dim);
    }

    virtual void multiply_coef(CPPCTYPE coef) override {
        state_multiply(coef, this->data_c(), this->dim);
    }

    virtual void multiply_elementwise_function(
        const std::function<CPPCTYPE(ITYPE)>& func) override {
        CPPCTYPE* state = this->data_cpp();
        for (ITYPE idx = 0; idx < dim; ++idx) {
            state[idx] *= (CPPCTYPE)func(idx);
        }
    }

    virtual std::vector<ITYPE> sampling(UINT sampling_count) override {
#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            // すべてのrankで同一の結果を得るために、seedを共有する
            UINT seed = random.int32();
            MPIutil& mpiutil = MPIutil::get_inst();
            if (mpiutil.get_size() > 1) mpiutil.s_u_bcast(&seed);
            return this->sampling(sampling_count, seed);
        }
#endif
        std::vector<double> stacked_prob;
        std::vector<ITYPE> result;
        double sum = 0.;
        auto ptr = this->data_cpp();
        stacked_prob.push_back(0.);
        for (ITYPE i = 0; i < this->dim; ++i) {
            sum += norm(ptr[i]);
            stacked_prob.push_back(sum);
        }

        for (UINT count = 0; count < sampling_count; ++count) {
            double r = random.uniform();
            auto ite =
                std::lower_bound(stacked_prob.begin(), stacked_prob.end(), r);
            auto index = std::distance(stacked_prob.begin(), ite) - 1;
            result.push_back(index);
        }
        return result;
    }

    virtual std::vector<ITYPE> sampling(
        UINT sampling_count, UINT random_seed) override {
        random.set_seed(random_seed);
        std::vector<ITYPE> result;

#ifdef _USE_MPI
        if (this->outer_qc > 0) {
            std::vector<double> stacked_prob;
            MPIutil& mpiutil = MPIutil::get_inst();
            UINT mpirank = mpiutil.get_rank();
            UINT mpisize = mpiutil.get_size();
            double sum = 0.;
            auto ptr = this->data_cpp();
            // resize
            stacked_prob.resize(this->dim + 1);
            result.resize(sampling_count);

            stacked_prob[0] = 0.;
            for (ITYPE i = 0; i < this->dim; ++i) {
                sum += norm(ptr[i]);
                stacked_prob[i + 1] = sum;
            }

            double* sumrank_prob;
            sumrank_prob = new double[mpisize];

            mpiutil.s_D_allgather(sum, sumrank_prob);

            double firstv = 0.;
            for (UINT i = 0; i < mpirank; ++i) {
                firstv += sumrank_prob[i];
            }
            for (ITYPE i = 0; i < this->dim + 1; ++i) {
                stacked_prob[i] += firstv;
            }
            delete[] sumrank_prob;

            for (UINT count = 0; count < sampling_count; ++count) {
                double r = random.uniform();
                auto ite = std::lower_bound(
                    stacked_prob.begin(), stacked_prob.end(), r);
                auto index = std::distance(stacked_prob.begin(), ite) - 1;
                result[count] = index;
            }

            ITYPE geta = mpirank * this->dim;
            for (UINT i = 0; i < sampling_count; ++i) {
                if (result[i] == -1ULL or result[i] == this->dim)
                    result[i] = 0ULL;
                else
                    result[i] += geta;
            }
            mpiutil.m_I_allreduce(result.data(), sampling_count);
        } else
#endif
        {
            result = this->sampling(sampling_count);
        }

        return result;
    }
    virtual boost::property_tree::ptree to_ptree() const override {
        boost::property_tree::ptree pt;
        pt.put("name", "QuantumState");
        pt.put("qubit_count", _qubit_count);
        pt.put_child(
            "classical_register", ptree::to_ptree(_classical_register));
        pt.put_child("state_vector", ptree::to_ptree(std::vector<CPPCTYPE>(
                                         _state_vector, _state_vector + _dim)));
        return pt;
    }
};

using QuantumState = QuantumStateCpu;

namespace state {
CPPCTYPE DllExport inner_product(
    const QuantumState* state_bra, const QuantumState* state_ket);
DllExport QuantumState* tensor_product(
    const QuantumState* state_left, const QuantumState* state_right);
DllExport QuantumState* permutate_qubit(
    const QuantumState* state, std::vector<UINT> qubit_order);
DllExport QuantumState* drop_qubit(const QuantumState* state,
    std::vector<UINT> target, std::vector<UINT> projection);
// create superposition of states of coef1|state1>+coef2|state2>
DllExport QuantumState* make_superposition(CPPCTYPE coef1,
    const QuantumState* state1, CPPCTYPE coef2, const QuantumState* state2);
}  // namespace state