Simple Standalone Single Producer Single Consumer Queue (Based on Facebook/Folly)

spsc_queue.hpp

// based on facebook/folly/ProducerConsumerQueue.h
// main difference is the inclusion of a pop_all() method 
// and more stl-ish method names
// and it uses a vector for the underlying storage (which let's us leave out lots of manual cleanup/memory management)
// this is fine because we never resize the vector

// also the capacity parameter is the usable capacity, this class will automatically allocate one more for the dummy value

// original copyright:
/*
 * Copyright 2014 Facebook, Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

// @author Bo Hu (bhu@fb.com)
// @author Jordan DeLong (delong.j@fb.com)

#ifndef SPSC_QUEUE_H_
#define SPSC_QUEUE_H_

#include <vector>
#include <cstdlib>

template <class T>
class spsc_queue {
public:
    using size_type = std::size_t;
    using difference_type = std::ptrdiff_t;
    using value_type = typename std::decay<T>::type;
    using pointer = typename std::add_pointer<value_type>::type;
    using const_pointer = typename std::add_const<pointer>::type;
    using reference = typename std::add_lvalue_reference<value_type>::type;
    using const_reference = typename std::add_const<reference>::type;
    
    spsc_queue(size_type capacity = 16) :
        m_write_index(0),
        m_read_index(0),
        m_capacity(capacity+1), // +1 capacity for dummy value that indicates "empty"
        m_buffer(m_capacity)
    {
       

    };
    bool try_push(const_reference value) {
        auto const write_index = m_write_index.load(std::memory_order_relaxed);
        auto next_index = write_index + 1;
        if (next_index == m_capacity) {
            next_index = 0;
        }
        if (next_index != m_read_index.load(std::memory_order_acquire)) {
            m_buffer[write_index] = value;
            m_write_index.store(next_index, std::memory_order_release);
            return true;
        }
        
        // queue is full
        return false;
    }
    
    // same as above but takes an R-value reference (doesn't work with lambdas :( )
    bool try_push(value_type&& value) {
        auto const write_index = m_write_index.load(std::memory_order_relaxed);
        auto next_index = write_index + 1;
        if (next_index == m_capacity) {
            next_index = 0;
        }
        if (next_index != m_read_index.load(std::memory_order_acquire)) {
            m_buffer[write_index] = std::move(value);
            m_write_index.store(next_index, std::memory_order_release);
            return true;
        }
        
        // queue is full
        return false;
    }
    // try and pop an element off the queue
    // uses 2 atomic reads and 1 atomic write
    // generally you'll want to use pop_all() instead
    bool try_pop(reference value) {
        auto const read_index = m_read_index.load(std::memory_order_relaxed);
        if (read_index == m_write_index.load(std::memory_order_acquire)) {
            // queue is empty
            return false;
        }
        auto next_index = read_index + 1;
        if (next_index == m_capacity) {
            next_index = 0;
        }
        value = std::move(m_buffer[read_index]);
        m_read_index.store(next_index, std::memory_order_release);
        return true;
    }
    
    // use this for message queues and the like
    // copies the current elements into a vector and returns it
    // only uses 2 atomic reads and 1 atomic write for N elements
    // whereas using try_pop() in a loop you'd use 2N atomic reads and N atomic writes
    // plus we it's faster to move all the elements out of the underlying buffer since
    // they'll likely be in cache already. should be Fast Enough (TM) for most message queues
    std::vector<value_type> pop_all() {
        std::vector<value_type> value;
        using std::make_move_iterator;
        auto const read_index = m_read_index.load(std::memory_order_relaxed);
        auto const write_index = m_write_index.load(std::memory_order_acquire);
        if (read_index == write_index) {
            // queue is empty
            return value;
        }
        const bool is_wrapped = write_index < read_index;
        auto const buffer_begin = m_buffer.begin();
        auto const read_begin = buffer_begin + read_index;
        auto const read_end = buffer_begin + write_index;
        
        // fast case, linear copy
        if (LIKELY(!is_wrapped)) {
            assert(read_begin < read_end);
            assert(read_end <= m_buffer.end());
            value.insert(value.begin(), make_move_iterator(read_begin), make_move_iterator(read_end));
        } else {
            // slow case, copy to end and then wrap around
            // technically we could check if the wrap is false (write_index == 0)
            // but additional branching would slow down the fast path
            
            auto const buffer_end = m_buffer.end();
            size_type initial_size = std::distance(read_begin, buffer_end);
            size_type total_size = initial_size + std::distance(buffer_begin, read_end);
            value.reserve(total_size);
            assert(read_begin < buffer_end);
            value.insert(value.begin(), make_move_iterator(read_begin), make_move_iterator(buffer_end));
            assert(buffer_begin <= read_end);
            assert(read_end <= buffer_end);
            value.insert(value.end(), make_move_iterator(buffer_begin), make_move_iterator(read_end));
            assert(value.size() == total_size);
        }
        // we've emptied the queue (as far as we know)
        m_read_index.store(write_index, std::memory_order_release);
        return value;
    }
    
    bool is_empty() const {
        return m_write_index.load(std::memory_order_consume) == m_read_index.load(std::memory_order_consume);
    }
    
    bool is_full() const {
        auto next_index = m_write_index.load(std::memory_order_consume) + 1;
        if (next_index == m_capacity) {
            next_index = 0;
        }
        if (next_index != m_read_index.load(std::memory_order_consume)) {
            return false;
        }
        return true;
    }
    
    // obviously this is an estimate if you're concurrently writing to the queue
    // can be stupidly inaccurate if you're using it from a producer
    size_type size() const {
        difference_type ret = m_write_index.load(std::memory_order_consume) - m_read_index.load(std::memory_order_consume);
        return ret < 0 ? ret + m_capacity : ret;
    };
    
    // true capacity of the queue is size()-1 because a dummy element is required to check for emptiness
    size_type capacity() const {
        return m_buffer.size() - 1;
    };
    // queue is a fixed size, just return the capacity
    size_type max_size() const {
        return capacity();
    }
    
    
    
private:
    std::atomic<size_t> m_write_index, m_read_index, m_size;
    const size_t m_capacity;
    std::vector<value_type> m_buffer;
};