Add ReaderWriterQueue and update ConcurrentQueue

include/icsneo/third-party/readerwriterqueue/benchmarks/ext/1024cores/spscqueue.h

@@ -0,0 +1,139 @@
+#include "../../../atomicops.h"
+#include <cstdlib>		// For std::size_t
+
+// From http://www.1024cores.net/home/lock-free-algorithms/queues/unbounded-spsc-queue
+// (and http://software.intel.com/en-us/articles/single-producer-single-consumer-queue)
+
+// load with 'consume' (data-dependent) memory ordering 
+template<typename T> 
+T load_consume(T const* addr) 
+{ 
+    // hardware fence is implicit on x86 
+    T v = *const_cast<T const volatile*>(addr); 
+    moodycamel::compiler_fence(moodycamel::memory_order_seq_cst);
+    return v; 
+} 
+
+// store with 'release' memory ordering 
+template<typename T> 
+void store_release(T* addr, T v) 
+{ 
+    // hardware fence is implicit on x86 
+    moodycamel::compiler_fence(moodycamel::memory_order_seq_cst);
+    *const_cast<T volatile*>(addr) = v; 
+} 
+
+// cache line size on modern x86 processors (in bytes) 
+size_t const cache_line_size = 64; 
+// single-producer/single-consumer queue 
+template<typename T> 
+class spsc_queue 
+{ 
+public: 
+  spsc_queue() 
+  { 
+      node* n = new node; 
+      n->next_ = 0; 
+      tail_ = head_ = first_= tail_copy_ = n; 
+  }
+
+  explicit spsc_queue(size_t prealloc)
+  {
+      node* n = new node;
+      n->next_ = 0;
+      tail_ = head_ = first_ = tail_copy_ = n;
+
+      // [CD] Not (at all) the most efficient way to pre-allocate memory, but it works
+      T dummy = T();
+      for (size_t i = 0; i != prealloc; ++i) {
+          enqueue(dummy);
+      }
+      for (size_t i = 0; i != prealloc; ++i) {
+          try_dequeue(dummy);
+      }
+  }
+
+  ~spsc_queue() 
+  { 
+      node* n = first_; 
+      do 
+      { 
+          node* next = n->next_; 
+          delete n; 
+          n = next; 
+      } 
+      while (n); 
+  } 
+
+  void enqueue(T v) 
+  { 
+      node* n = alloc_node(); 
+      n->next_ = 0; 
+      n->value_ = v; 
+      store_release(&head_->next_, n); 
+      head_ = n; 
+  } 
+
+  // returns 'false' if queue is empty 
+  bool try_dequeue(T& v) 
+  { 
+      if (load_consume(&tail_->next_)) 
+      { 
+          v = tail_->next_->value_; 
+          store_release(&tail_, tail_->next_); 
+          return true; 
+      } 
+      else 
+      { 
+          return false; 
+      } 
+  } 
+
+private: 
+  // internal node structure 
+  struct node 
+  { 
+      node* next_; 
+      T value_; 
+  }; 
+
+  // consumer part 
+  // accessed mainly by consumer, infrequently be producer 
+  node* tail_; // tail of the queue 
+
+  // delimiter between consumer part and producer part, 
+  // so that they situated on different cache lines 
+  char cache_line_pad_ [cache_line_size]; 
+
+  // producer part 
+  // accessed only by producer 
+  node* head_; // head of the queue 
+  node* first_; // last unused node (tail of node cache) 
+  node* tail_copy_; // helper (points somewhere between first_ and tail_) 
+
+  node* alloc_node() 
+  { 
+      // first tries to allocate node from internal node cache, 
+      // if attempt fails, allocates node via ::operator new() 
+
+      if (first_ != tail_copy_) 
+      { 
+          node* n = first_; 
+          first_ = first_->next_; 
+          return n; 
+      } 
+      tail_copy_ = load_consume(&tail_); 
+      if (first_ != tail_copy_) 
+      { 
+          node* n = first_; 
+          first_ = first_->next_; 
+          return n; 
+      } 
+      node* n = new node; 
+      return n; 
+  } 
+
+  spsc_queue(spsc_queue const&); 
+  spsc_queue& operator = (spsc_queue const&); 
+
+};

include/icsneo/third-party/readerwriterqueue/benchmarks/ext/folly/ProducerConsumerQueue.h

@@ -0,0 +1,174 @@
+// Adapted from https://github.com/facebook/folly/blob/master/folly/ProducerConsumerQueue.h
+/*
+ * Copyright 2013 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 PRODUCER_CONSUMER_QUEUE_H_
+#define PRODUCER_CONSUMER_QUEUE_H_
+
+#include <new>
+#include <atomic>
+#include <cassert>
+#include <cstdlib>
+#include <stdexcept>
+#include <type_traits>
+#include <utility>
+//#include <boost/noncopyable.hpp>
+
+namespace folly {
+
+/*
+ * ProducerConsumerQueue is a one producer and one consumer queue
+ * without locks.
+ */
+template<class T>
+struct ProducerConsumerQueue {
+  typedef T value_type;
+
+  // size must be >= 1.
+  explicit ProducerConsumerQueue(uint32_t size)
+    : size_(size + 1)    // +1 because one slot is always empty
+    , records_(static_cast<T*>(std::malloc(sizeof(T) * (size + 1))))
+    , readIndex_(0)
+    , writeIndex_(0)
+  {
+    assert(size >= 1);
+    if (!records_) {
+      throw std::bad_alloc();
+    }
+  }
+
+  ~ProducerConsumerQueue() {
+    // We need to destruct anything that may still exist in our queue.
+    // (No real synchronization needed at destructor time: only one
+    // thread can be doing this.)
+    if (!std::is_trivially_destructible<T>::value) {
+      int read = readIndex_;
+      int end = writeIndex_;
+      while (read != end) {
+        records_[read].~T();
+        if (++read == size_) {
+          read = 0;
+        }
+      }
+    }
+
+    std::free(records_);
+  }
+
+  template<class ...Args>
+  bool enqueue(Args&&... recordArgs) {
+    auto const currentWrite = writeIndex_.load(std::memory_order_relaxed);
+    auto nextRecord = currentWrite + 1;
+    if (nextRecord == size_) {
+      nextRecord = 0;
+    }
+    if (nextRecord != readIndex_.load(std::memory_order_acquire)) {
+      new (&records_[currentWrite]) T(std::forward<Args>(recordArgs)...);
+      writeIndex_.store(nextRecord, std::memory_order_release);
+      return true;
+    }
+
+    // queue is full
+    return false;
+  }
+
+  // move (or copy) the value at the front of the queue to given variable
+  bool try_dequeue(T& record) {
+    auto const currentRead = readIndex_.load(std::memory_order_relaxed);
+    if (currentRead == writeIndex_.load(std::memory_order_acquire)) {
+      // queue is empty
+      return false;
+    }
+
+    auto nextRecord = currentRead + 1;
+    if (nextRecord == size_) {
+      nextRecord = 0;
+    }
+    record = std::move(records_[currentRead]);
+    records_[currentRead].~T();
+    readIndex_.store(nextRecord, std::memory_order_release);
+    return true;
+  }
+
+  // pointer to the value at the front of the queue (for use in-place) or
+  // nullptr if empty.
+  T* frontPtr() {
+    auto const currentRead = readIndex_.load(std::memory_order_relaxed);
+    if (currentRead == writeIndex_.load(std::memory_order_acquire)) {
+      // queue is empty
+      return nullptr;
+    }
+    return &records_[currentRead];
+  }
+
+  // queue must not be empty
+  void popFront() {
+    auto const currentRead = readIndex_.load(std::memory_order_relaxed);
+    assert(currentRead != writeIndex_.load(std::memory_order_acquire));
+
+    auto nextRecord = currentRead + 1;
+    if (nextRecord == size_) {
+      nextRecord = 0;
+    }
+    records_[currentRead].~T();
+    readIndex_.store(nextRecord, std::memory_order_release);
+  }
+
+  bool isEmpty() const {
+   return readIndex_.load(std::memory_order_consume) ==
+         writeIndex_.load(std::memory_order_consume);
+  }
+
+  bool isFull() const {
+    auto nextRecord = writeIndex_.load(std::memory_order_consume) + 1;
+    if (nextRecord == size_) {
+      nextRecord = 0;
+    }
+    if (nextRecord != readIndex_.load(std::memory_order_consume)) {
+      return false;
+    }
+    // queue is full
+    return true;
+  }
+
+  // * If called by consumer, then true size may be more (because producer may
+  //   be adding items concurrently).
+  // * If called by producer, then true size may be less (because consumer may
+  //   be removing items concurrently).
+  // * It is undefined to call this from any other thread.
+  size_t sizeGuess() const {
+    int ret = writeIndex_.load(std::memory_order_consume) -
+              readIndex_.load(std::memory_order_consume);
+    if (ret < 0) {
+      ret += size_;
+    }
+    return ret;
+  }
+
+private:
+  const uint32_t size_;
+  T* const records_;
+
+  std::atomic<int> readIndex_;
+  std::atomic<int> writeIndex_;
+};
+
+}
+
+#endif