#ifndef SPINLOCKQUEUE_HPP
#define SPINLOCKQUEUE_HPP 1
#include <cstddef>
#include <algorithm>
#include <atomic>
#include <thread>
#include <mutex>
#include <condition_variable>
#include <utility>
#include <cassert>
template<typename T>
bool operator==(std::atomic<T*> const & a, std::atomic<T*> const & b)
{
    return a.load() == b.load();
}

template <typename T>
bool operator!=(std::atomic<T*> const & a, std::atomic<T*> const & b)
{
    return a.load() != b.load();
}

template <typename T>
class SpinLockQueue {
private:
  struct Node {
    Node( T *val ) : value(val), next(0) { }
    Node( Node && rhs)
    {
    *this = std::move(rhs);
    }
    Node( Node const &) = delete;
    Node& operator=(Node const &) = delete;
    Node& operator=(Node && rhs) {
    if (this != &rhs)
    {
        value = rhs.value;
        rhs.value = 0;
        next = rhs.next;
        rhs.next = 0;
    }

    return *this;
    }

    ~Node() {
    delete value;
    }
    T* value;
    Node* next;
  } __attribute__((aligned(64)));



  std::atomic<std::size_t> mWaiters;
  std::mutex mMutex;
  std::condition_variable mCondition;


  Node* first __attribute__((aligned(64)));

  // shared among consumers
  atomic<bool> consumerLock __attribute__((aligned(64)));

  // for one producer at a time
  Node* last __attribute__((aligned(64)));

  // shared among producers
  atomic<bool> producerLock __attribute__((aligned(64)));


  atomic<Node*> nodeList __attribute__((aligned(64)));


  Node* getNewNode(T const & t) {
    Node* n = nodeList;
    if (n)
    {
        while (n && !nodeList.compare_exchange_weak(n,n->next))
        {
            n = nodeList;
        }

        if (n)
        {
            n->value = new T(t);
            n->next = 0;
            return n;
        } else {
            return new Node(new T(t));
        }

    } else {
        return new Node(new T(t));
    }
  }
public:
  SpinLockQueue() : mWaiters(0) {
    first  = last =
      new Node( T() );           // add dummy separator

    producerLock = false;
    consumerLock = false;
    nodeList = 0;
  }
  ~SpinLockQueue() {
    Node* tmp;
    while( (tmp = first) != 0 ) {   // release the list
      first = tmp->next;
      delete tmp;
    }

    while ( (tmp = nodeList) != 0) {
    nodeList = tmp->next;
    delete tmp;
    }
  }


  void push( const T& t ) {
    Node* newlast = getNewNode(t);
    assert(newlast);
    bool oldvalue = false;
    while (!producerLock.compare_exchange_weak(oldvalue,true)) {}
// own producer lock

    last->next = newlast;
    last = newlast;

    producerLock = false;

    if (mWaiters)
    {
    mCondition.notify_one();
    }
  }

  bool try_pop( T& result ) {
    // No nodes point to front
    bool oldvalue = false;
    while (!consumerLock.compare_exchange_weak(oldvalue,true)) {}
    Node * next = first->next;
    Node * theFirst = first;
    if (next)
    {
    T* val = next->value;
    next->value = 0;
    first = next;
    consumerLock = false;
    result = std::move(*val);
    delete val;
    theFirst->value = 0;
    Node * n(0);
    do {
        n = nodeList;
        theFirst->next = n;

    } while (!nodeList.compare_exchange_weak(n,theFirst));

    return true;
    }

    consumerLock = false;
    return false;
  }


 void pop(T& result) {
    ++mWaiters;

    std::unique_lock<std::mutex> lk(mMutex);
    while (!try_pop(result))
    {
        mCondition.wait(lk);
    }

    --mWaiters;
    }

};

#if 1

void process(SpinLockQueue<int>* queue)
{
#pragma omp parallel for shared(queue)
        for (int i = 0; i < 100000; ++i)
        {
                queue->push(i);
                if (i % 3 == 0)
                {
                        int a;
                        queue->try_pop(a);
                }
        }
#pragma omp barrier
}



int main(int argc, char * argv[])
{
    SpinLockQueue<int> *queue = new SpinLockQueue<int>;
    process(queue);
    delete queue;
    return 0;
}


#endif

#endif