c++ lock_free_queue example tested more slowly than mutex - c++

Question

i tested the example in book <C++ Concurrency in Action> charpter 7 lock_free_queue and mutex_queue by 2 threads push and 2 threads pop ,the result is using mutex is faster than lock_free queue.here is the lock_free queue code.
so anyone can explain this result? on x86 cpu, the head and tail respectively of lock_free_queue is mutex free,so my first thought is this lock_free_queue will faster than mutex_queue .
here is the mutex queue code:
template<typename T>
class LockFreeQueue
{
public:
struct Node;
struct CountedNodePtr{
int external_count;
Node* ptr;
CountedNodePtr()noexcept:external_count(0),ptr(nullptr){}
};
std::atomic<CountedNodePtr> head;
std::atomic<CountedNodePtr> tail;
struct NodeCounter{
uint32_t internal_count:30;
uint32_t external_counters:2;
};
struct Node{
std::atomic<T*> data;
std::atomic<NodeCounter> count;
CountedNodePtr next;
Node()noexcept{
NodeCounter new_count;
new_count.external_counters = 2;
new_count.internal_count = 0;
count.store(new_count);
data.store(nullptr);
}
void ReleaseRef(){
NodeCounter old_counter = count.load(std::memory_order_relaxed);
NodeCounter new_counter;
do{
new_counter = old_counter;
--new_counter.internal_count;
}while(!count.compare_exchange_strong(old_counter, new_counter,
std::memory_order_acquire,std::memory_order_relaxed));
if(!new_counter.internal_count && !new_counter.external_counters){
delete this;
}
}
};
static void IncreaseExternalCount(std::atomic<CountedNodePtr>& counter,
CountedNodePtr& old_counter){
CountedNodePtr new_counter;
do{
new_counter = old_counter;
++new_counter.external_count;
}while(!counter.compare_exchange_strong(old_counter, new_counter,
std::memory_order_acquire,std::memory_order_relaxed));
old_counter.external_count = new_counter.external_count;
}
static void FreeExternalCounter(CountedNodePtr& old_node_ptr){
Node* const ptr = old_node_ptr.ptr;
int const count_increase = old_node_ptr.external_count-2;
NodeCounter old_counter = ptr->count.load(std::memory_order_relaxed);
NodeCounter new_counter;
do{
new_counter = old_counter;
--new_counter.external_counters;
new_counter.internal_count+=count_increase;
}while(!ptr->count.compare_exchange_strong(old_counter, new_counter,
std::memory_order_acquire,std::memory_order_relaxed));
if(!new_counter.internal_count && !new_counter.external_counters){
delete ptr;
}
}
public:
LockFreeQueue(){
CountedNodePtr ptr ;
ptr.external_count =1;
ptr.ptr = new Node;
head.store(ptr);
tail.store(head.load());
};
~LockFreeQueue(){
while(Node* const old_head = head.load().ptr){
head.store(old_head->next);
delete old_head;
}
}
void Push(T new_value){
std::unique_ptr<T> new_data = std::make_unique<T>(new_value);
CountedNodePtr new_next;
new_next.ptr = new Node;
new_next.external_count = 1;
CountedNodePtr old_tail = tail.load();
for(;;){
IncreaseExternalCount(tail,old_tail);
T* old_data = nullptr;
if(old_tail.ptr->data.compare_exchange_strong(old_data,
new_data.get())){
old_tail.ptr->next = new_next;
old_tail = tail.exchange(new_next);
FreeExternalCounter(old_tail);
new_data.release();
break;
}
old_tail.ptr->ReleaseRef();
}
}
std::unique_ptr<T> Pop(){
CountedNodePtr old_head = head.load(std::memory_order_relaxed);
for(;;){
IncreaseExternalCount(head,old_head);
Node* const ptr = old_head.ptr;
if(ptr==tail.load().ptr){
ptr->ReleaseRef();
return std::unique_ptr<T>();
}
if(head.compare_exchange_strong(old_head, ptr->next)){
T* const res = ptr->data.exchange(nullptr);
FreeExternalCounter(old_head);
return std::unique_ptr<T>(res);
}
ptr->ReleaseRef();
}
}
};
// here is the mutex_queue
template <typename T>
class ThreadSafeQueue {
public:
ThreadSafeQueue() {}
ThreadSafeQueue& operator=(const ThreadSafeQueue& other) = delete;
ThreadSafeQueue(const ThreadSafeQueue& other) = delete;
~ThreadSafeQueue() { BreakAllWait(); }
void Enqueue(const T& element) {
std::lock_guard<std::mutex> lock(mutex_);
queue_.emplace(element);
// cv_.notify_one();
}
bool Dequeue(T* element) {
std::lock_guard<std::mutex> lock(mutex_);
if (queue_.empty()) {
return false;
}
*element = std::move(queue_.front());
queue_.pop();
return true;
}
// bool WaitDequeue(T* element) {
// std::unique_lock<std::mutex> lock(mutex_);
// cv_.wait(lock, [this]() { return break_all_wait_ || !queue_.empty(); });
// if (break_all_wait_) {
// return false;
// }
// *element = std::move(queue_.front());
// queue_.pop();
// return true;
// }
typename std::queue<T>::size_type Size() {
std::lock_guard<std::mutex> lock(mutex_);
return queue_.size();
}
bool Empty() {
std::lock_guard<std::mutex> lock(mutex_);
return queue_.empty();
}
void BreakAllWait() {
break_all_wait_ = true;
// cv_.notify_all();
}
private:
volatile bool break_all_wait_ = false;
std::mutex mutex_;
std::queue<T> queue_;
struct Node{
int data;
Node * next;
Node(const int& d):data(d),next(NULL){}
};
// std::condition_variable cv_;
};
// test code added here
#include <iostream>
#include "lock_free_queue.h"
#include "thread_safe_queue.h"
#include "unbounded_queue.h"
#include <thread>
#include <chrono>
#include <thread>
#include <atomic>
#include <vector>
#include <mutex>
#include <utility>
#include <iomanip>
#include <ucontext.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <stddef.h>
#include <string.h>
#include <stdbool.h>
#include <stdint.h>
//#include <stdatomic.h>
#include <pthread.h>
#include <time.h>
#ifndef MDC_TEST
using namespace rudder::base;
using namespace apollo::cyber::base;
using namespace std;
struct Student
{
int a;
double b;
float c;
};
typedef int swr_ele_t; // 元素
int cycles;
typedef struct swr_queue
{ // 队列结构
std::atomic<size_t> head;
std::atomic<size_t> tail;
swr_ele_t *eles;
size_t size;
pthread_mutex_t lock;
} swr_queue_t;
LockFreeQueue<swr_ele_t> lock_queue_test[7];
ThreadSafeQueue<swr_ele_t> mutex_queue_test[7];
UnboundedQueue<swr_ele_t> unbounde_queue_test[7];
bool LockPush(const swr_ele_t *ele,int i)
{
lock_queue_test[i].Push(*ele);
return true;
};
bool LockPop(swr_ele_t *ele,int i)
{
auto tmp = lock_queue_test[i].Pop();
if(tmp == nullptr)
{
return false;
}
else
{
*ele = *tmp;
// std::cout<<"tmp"<<*tmp<<std::endl;
return true;
}
};
bool MutexPush(const swr_ele_t *ele,int i)
{
mutex_queue_test[i].Enqueue(*ele);
return true;
};
bool MutexPop(swr_ele_t *ele,int i)
{
swr_ele_t tmp;
if(mutex_queue_test[i].Dequeue(&tmp))
{
*ele = tmp;
return true;
}else{
return false;
}
};
bool UnboundedPush(const swr_ele_t *ele,int i)
{
unbounde_queue_test[i].Enqueue(*ele);
return true;
};
bool UnboundedPop(swr_ele_t *ele,int i)
{
swr_ele_t tmp;
if(unbounde_queue_test[i].Dequeue(&tmp))
{
*ele = tmp;
return true;
}else{
return false;
}
};
/***************************测试程序部分******************************/
typedef bool (*push_fn_t)(const swr_ele_t *,int);
typedef bool (*pop_fn_t)(swr_ele_t *,int);
// 线程参数
typedef struct thead_arg
{
//swr_queue_t *queue;
char type;
push_fn_t push;
pop_fn_t pop;
double push_time;
double pop_time;
int i;
} thead_arg_t;
// 计时器: 毫秒
double time_diff(struct timespec *start, struct timespec *end)
{
long sec = end->tv_sec - start->tv_sec;
long nsec = end->tv_nsec - start->tv_nsec;
return sec * 1000.0 + nsec / 1000000.0;
}
void *thread_func(void *a)
{
thead_arg_t *arg = (thead_arg_t *)a;
//swr_queue_t *queue = arg->queue;
struct timespec time1, time2;
clock_gettime(CLOCK_THREAD_CPUTIME_ID, &time1);
if (arg->type == 'W')
{ // Producer
swr_ele_t e = cycles;
while (true)
{
if (arg->push(&e,arg->i))
{
if ((--e) < 0)
break;
}
}
}
else
{ // Consumer
while (true)
{
swr_ele_t e;
// cout<<"arg->pop"<<std::endl;
if (arg->pop(&e,arg->i))
{
// cout<<"pop"<<e <<std::endl;
// if (e % 1000000 == 0) printf("pop ele: %d\n", e);
if (e == 0)
break;
}
}
}
clock_gettime(CLOCK_THREAD_CPUTIME_ID, &time2);
if(arg->type == 'W')
{
arg->push_time = time_diff(&time1, &time2);
}
else
{
arg->pop_time = time_diff(&time1, &time2);
}
printf("%c %d Time=%fms\n", arg->type, arg->i,time_diff(&time1, &time2));
// printf("%f ", arg->type, time_diff(&time1, &time2));
return NULL;
}
int main(int argc, char const *argv[])
{
// std::cout<<"argv"<< argv[1]<<std::endl;;
std::cout<<"core_num is : "<<std::thread::hardware_concurrency()<<std::endl;
std::cout<<"lock free head" <<lock_queue_test[0].head.is_lock_free()<<std::endl;
std::cout<<"lock free tail" <<lock_queue_test[0].tail.is_lock_free()<<std::endl;
int num = 0;
while(num<7)
{
cycles = 10000;
struct test_case
{
push_fn_t push;
pop_fn_t pop;
const char *desc;
} cases[] = {
//{LockPush, LockPop, "LockPush, LockPop"},
{MutexPush, MutexPop, "MutexPush,MutexPop"},
// {UnboundedPush, UnboundedPop, "UnboundedPush,UnboundedPop"},
};
for (unsigned int k = 0; k < sizeof(cases) / sizeof(cases[0]); ++k)
{
printf("----------------------------------------------------------------------------\n");
printf("%s\n", cases[k].desc);
swr_queue_t queue;
const int NUM = 6;
pthread_t tids[NUM];
thead_arg_t args[NUM];
int i;//0 1 2 3
// for (i = 0; i < NUM; ++i)
// {
// args[i].i = num;
// // args[i].queue = &queue;
// args[i].type = (i+1)%2 == 1 ? 'W' : 'R';
// args[i].push = cases[k].push;
// args[i].pop = cases[k].pop;
//// printf("pthread_create start");
// pthread_create(&tids[i], NULL, thread_func, args + i);
// }
for (i = 0; i < 3; ++i)
{
args[i].i = num;
// args[i].queue = &queue;
args[i].type = 'W';
args[i].push = cases[k].push;
args[i].pop = cases[k].pop;
// printf("pthread_create start");
pthread_create(&tids[i], NULL, thread_func, args + i);
// pthread_join(tids[i], NULL);
// printf("pthread_create %d",args[i].i);
}
for (i = 0; i < 3; ++i)
{
pthread_join(tids[i], NULL);
}
for (i = 3; i < 6; ++i)
{
args[i].i = num;
// args[i].queue = &queue;
args[i].type = 'R';
args[i].push = cases[k].push;
args[i].pop = cases[k].pop;
// printf("pthread_create start");
pthread_create(&tids[i], NULL, thread_func, args + i);
// pthread_join(tids[i], NULL);
// printf("pthread_create %d",args[i].i);
}
for (i = 3; i < 6; ++i)
{
pthread_join(tids[i], NULL);
}
printf("\n");
// printf("total time=%fms\n", args[0].push_time + args[1].pop_time);
// printf("%f\n", args[0].push_time + args[1].pop_time);
}
num++;
}
return 0;
}
#endif /* MDC_TEST */