/*!
 * \file
 * \brief   Distributed sort implementation header
 *
 * \author
 *    Christos Choutouridis AEM:8997
 *    <cchoutou@ece.auth.gr>
 */

#ifndef DISTBITONIC_H_
#define DISTBITONIC_H_

#include <vector>
#include <algorithm>
#include <cmath>
#include <cstdint>
#if !defined DEBUG
#define NDEBUG
#endif
#include <cassert>

#include "utils.hpp"
#include "config.h"


/*!
 * Enumerator for the different versions of the sorting method
 */
enum class SortMode {
    Bubbletonic,    //!< The v0.5 of the algorithm where we use a bubble-sort like approach
    Bitonic         //!< The v1.0 of the algorithm where we use the bitonic data-exchange approach
};

/*
 * ============================== Sort utilities ==============================
 */

/*!
 * The primary function template of ascending(). It is DISABLED since , it is explicitly specialized
 * for each of the \c SortMode
 */
template <SortMode Mode> inline bool ascending(mpi_id_t, [[maybe_unused]] size_t) noexcept = delete;

/*!
 * Returns the ascending or descending configuration of the node's sequence based on
 * the current node (MPI process) and the depth of the sorting network
 *
 * @param node      The current node (MPI process)
 * @return          True if we need ascending configuration, false otherwise
 */
template <> inline
bool ascending<SortMode::Bubbletonic>(mpi_id_t node, [[maybe_unused]] size_t depth) noexcept {
    return (node % 2) == 0;
}

/*!
 * Returns the ascending or descending configuration of the node's sequence based on
 * the current node (MPI process) and the depth of the sorting network
 *
 * @param node      The current node (MPI process)
 * @param depth     The total depth of the sorting network (same for each step for a given network)
 *
 * @return          True if we need ascending configuration, false otherwise
 */
template <> inline
bool ascending<SortMode::Bitonic>(mpi_id_t node, size_t depth) noexcept {
    return !(node & (1 << depth));
}

/*!
 * The primary function template of partner(). It is DISABLED since , it is explicitly specialized
 * for each of the \c SortMode
 */
template <SortMode Mode> inline mpi_id_t partner(mpi_id_t, size_t) noexcept = delete;

/*!
 * Returns the node's partner for data exchange during the sorting network iterations
 * of Bubbletonic
 *
 * @param node      The current node
 * @param step      The step of the sorting network
 * @return          The node id of the partner for data exchange
 */
template <> inline
mpi_id_t partner<SortMode::Bubbletonic>(mpi_id_t node, size_t step) noexcept {
    //return (node % 2 == step % 2) ? node + 1 : node - 1;
    return (((node+step) % 2) == 0) ? node + 1 : node - 1;
}

/*!
 * Returns the node's partner for data exchange during the sorting network iterations
 * of Bitonic
 *
 * @param node      The current node
 * @param step      The step of the sorting network
 * @return          The node id of the partner for data exchange
 */
template <> inline
mpi_id_t partner<SortMode::Bitonic>(mpi_id_t node, size_t step) noexcept {
    return (node ^ (1 << step));
}


/*!
 * The primary function template of keepSmall(). It is DISABLED since , it is explicitly specialized
 * for each of the \c SortMode
 */
template<SortMode Mode> inline bool keepSmall(mpi_id_t, mpi_id_t, [[maybe_unused]] size_t) noexcept = delete;

/*!
 * Predicate to check if a node keeps the small numbers during the bubbletonic sort network exchange.
 *
 * @param node      The node for which we check
 * @param partner   The partner of the data exchange
 * @return          True if the node should keep the small values, false otherwise
 */
template <> inline
bool keepSmall<SortMode::Bubbletonic>(mpi_id_t node, mpi_id_t partner, [[maybe_unused]] size_t depth) noexcept {
    assert(node != partner);
    return (node < partner);
}

/*!
 * Predicate to check if a node keeps the small numbers during the bitonic sort network exchange.
 *
 * @param node      The node for which we check
 * @param partner   The partner of the data exchange
 * @param depth     The total depth of the sorting network (same for each step for a given network)
 * @return          True if the node should keep the small values, false otherwise
 */
template <> inline
bool keepSmall<SortMode::Bitonic>(mpi_id_t node, mpi_id_t partner, size_t depth) noexcept {
    assert(node != partner);
    return ascending<SortMode::Bitonic>(node, depth) == (node < partner);
}

/*!
 * Predicate to check if the node is active in the current iteration of the bubbletonic
 * sort exchange.
 *
 * @param node      The node to check
 * @param nodes     The total number of nodes
 * @return          True if the node is active, false otherwise
 */
bool isActive(mpi_id_t node, size_t nodes) noexcept;

/*
 * ============================== Data utilities ==============================
 */

/*!
 *
 * @tparam RangeT
 * @param data
 * @param ascending
 */
template<typename RangeT>
void fullSort(RangeT& data, bool ascending) {
    // Use introsort from stdlib++ here, unless ...
    if (ascending)
        std::sort(data.begin(), data.end(), std::less<>());
    else
        std::sort(data.begin(), data.end(), std::greater<>());
}

/*!
 *
 * @tparam ShadowedT
 * @tparam CompT
 * @param data
 * @param comp
 */
template<typename ShadowedT, typename CompT>
void elbowSortCore(ShadowedT& data, CompT comp) {
    size_t N = data.size();
    auto active = data.getActive();
    auto shadow = data.getShadow();
    size_t left = std::distance(
            active.begin(),
            std::min_element(active.begin(), active.end())
    );
    size_t right = (left == N-1) ? 0 : left + 1;

    for (size_t i = 0 ; i<N ; ++i) {
        if (comp(active[left], active[right])) {
            shadow[i] = active[left];
            left = (left == 0) ? N-1 : left -1;
        }
        else {
            shadow[i] = active[right];
            right = (right + 1) % N;
        }
    }
    data.switch_active();
}

/*!
 *
 * @tparam ShadowedT
 * @param data
 * @param ascending
 */
template<typename ShadowedT>
void elbowSort(ShadowedT& data, bool ascending) {
    if (ascending)
        elbowSortCore(data, std::less<>());
    else
        elbowSortCore(data, std::greater<>());
}

/*!
 *
 * @tparam RangeT
 * @param local
 * @param remote
 * @param keepsmall
 */
template<typename RangeT>
void minmax(RangeT& local, RangeT& remote, bool keepsmall) {
    using value_t = typename RangeT::value_type;
    std::transform(
            local.begin(), local.end(),
            remote.begin(),
            local.begin(),
            [keepsmall](const value_t& a, const value_t& b){
                return (keepsmall) ? std::min(a, b) : std::max(a, b);
            });
}

/*
 * ============================== Sort algorithms ==============================
 */

/*!
 *
 * @tparam ShadowedT
 * @param data
 * @param Processes
 */
template<typename ShadowedT>
void distBubbletonic(ShadowedT& data, mpi_id_t Processes) {
    // Initially sort to create a half part of a bitonic sequence
    fullSort(data, ascending<SortMode::Bubbletonic>(mpi.rank(), 0));

    // Sort network
    for (size_t step = 0; step < Processes-1; ++step) {
        auto part = partner<SortMode::Bubbletonic>(mpi.rank(), step);
        auto ks = keepSmall<SortMode::Bubbletonic>(mpi.rank(), part, Processes);
        if (isActive(mpi.rank(), Processes)) {
            mpi.exchange(part, data.getActive(), data.getShadow(), step);
            minmax(data.getActive(), data.getShadow(), ks);
            elbowSort(data, ascending<SortMode::Bubbletonic>(mpi.rank(), Processes));
        }
    }

    if (!ascending<SortMode::Bubbletonic>(mpi.rank(), 0))
        elbowSort(data, true);

}


/*!
 *
 * @tparam ShadowedT
 * @param data
 * @param Processes
 */
template<typename ShadowedT>
void distBitonic(ShadowedT& data, mpi_id_t Processes) {
    auto p = static_cast<uint32_t>(std::log2(Processes));

    // Initially sort to create a half part of a bitonic sequence
    fullSort(data, ascending<SortMode::Bitonic>(mpi.rank(), 0));

    // Run through sort network using elbow-sort
    for (size_t depth = 1; depth <= p; ++depth) {
        for (size_t step = depth; step > 0;) {
            --step;
            auto part = partner<SortMode::Bitonic>(mpi.rank(), step);
            auto ks = keepSmall<SortMode::Bitonic>(mpi.rank(), part, depth);
            mpi.exchange(part, data.getActive(), data.getShadow(), (depth << 8) | step);
            minmax(data.getActive(), data.getShadow(), ks);
        }
        elbowSort (data, ascending<SortMode::Bitonic>(mpi.rank(), depth));
    }
}

#endif //DISTBITONIC_H_