BLT/include/blt/std/thread.h

/*
 *  <Short Description>
 *  Copyright (C) 2023  Brett Terpstra
 *
 *  This program is free software: you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation, either version 3 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program.  If not, see <https://www.gnu.org/licenses/>.
 */

#ifndef BLT_THREAD_H
#define BLT_THREAD_H

#include <blt/std/types.h>
#include <thread>
#include <functional>
#include <string>
#include <vector>
#include <queue>
#include <utility>
#include <variant>
#include <atomic>
#include <mutex>
#include <chrono>
#include <optional>
#include <blt/std/logging.h>
#include <condition_variable>

namespace blt
{
    
    class barrier
    {
        public:
            explicit barrier(blt::size_t threads, std::optional<std::reference_wrapper<std::atomic_bool>> exit_cond = {}):
                    thread_count(threads), threads_waiting(0), use_count(0), exit_cond(exit_cond), count_mutex(), cv()
            {
                if (threads == 0)
                    throw std::runtime_error("Barrier thread count cannot be 0");
            }
            
            barrier(const barrier& copy) = delete;
            
            barrier(barrier&& move) = delete;
            
            barrier& operator=(const barrier& copy) = delete;
            
            barrier& operator=(barrier&& move) = delete;
            
            ~barrier() = default;
            
            void wait()
            {
                // (unique_lock acquires lock)
                std::unique_lock lock(count_mutex);
                std::size_t current_uses = use_count;
                
                if (++threads_waiting == thread_count)
                {
                    threads_waiting = 0;
                    use_count++;
                    cv.notify_all();
                } else
                {
                    if constexpr (BUSY_LOOP_WAIT > 0) // NOLINT
                    {
                        lock.unlock();
                        for (blt::size_t i = 0; i < BUSY_LOOP_WAIT; i++)
                        {
                            if (use_count != current_uses || (exit_cond && exit_cond->get()))
                                return;
                        }
                        lock.lock();
                    }
                    cv.wait(lock, [this, &current_uses]() {
                        return (use_count != current_uses || (exit_cond && exit_cond->get()));
                    });
                }
            }
            
            void notify_all()
            {
                cv.notify_all();
            }
        
        private:
            blt::size_t thread_count;
            blt::size_t threads_waiting;
            blt::size_t use_count;
            std::optional<std::reference_wrapper<std::atomic_bool>> exit_cond;
            std::mutex count_mutex;
            std::condition_variable cv;
            
            // improves performance by not blocking the thread for n iterations of the loop.
            // If the condition is not met by the end of this loop we can block the thread.
            static constexpr blt::size_t BUSY_LOOP_WAIT = 200;
    };
    
    /**
     * @tparam queue should we use a queue or execute the same function over and over?
     */
    template<bool queue = false, typename... Args>
    class thread_pool
    {
        private:
            using thread_function = std::function<void(Args...)>;
            volatile std::atomic_bool should_stop = false;
            volatile std::atomic_uint64_t stopped = 0;
            std::uint64_t number_of_threads = 0;
            std::vector<std::thread*> threads;
            std::variant<std::queue<thread_function>, thread_function> func_queue;
            std::mutex queue_mutex;
            // only used when a queue
            volatile std::atomic_uint64_t tasks = 0;
            volatile std::atomic_uint64_t completed_tasks = 0;
            bool func_loaded = false;
            
            void init()
            {
                for (std::uint64_t i = 0; i < number_of_threads; i++)
                {
                    threads.push_back(new std::thread([this]() {
                        while (!should_stop)
                        {
                            if constexpr (queue)
                            {
                                // should be safe right?
                                std::unique_lock lock(queue_mutex);
                                lock.lock();
                                auto& func_q = std::get<std::queue<thread_function>>(func_queue);
                                if (func_q.empty())
                                {
                                    lock.unlock();
                                    std::this_thread::sleep_for(std::chrono::milliseconds(16));
                                    continue;
                                }
                                auto func = func_q.front();
                                func_q.pop();
                                lock.unlock();
                                func();
                                completed_tasks++;
                            } else
                            {
                                if (!func_loaded)
                                {
                                    std::scoped_lock lock(queue_mutex);
                                    if (std::holds_alternative<std::queue<thread_function>>(func_queue))
                                    {
                                        std::this_thread::sleep_for(std::chrono::milliseconds(16));
                                        //BLT_WARN("Running non queue variant with a queue inside!");
                                        continue;
                                    }
                                    func_loaded = true;
                                }
                                auto& func = std::get<thread_function>(func_queue);
                                func();
                            }
                        }
                        stopped++;
                    }));
                }
            }
            
            void cleanup()
            {
                for (auto* t : threads)
                {
                    if (t->joinable())
                        t->join();
                    delete t;
                }
            }
        
        public:
            explicit thread_pool(std::uint64_t number_of_threads = 8, std::optional<thread_function> default_function = {})
            {
                if (default_function.has_value())
                    func_queue = default_function.value();
                this->number_of_threads = number_of_threads;
                init();
            }
            
            inline void execute(const thread_function& func)
            {
                std::scoped_lock lock(queue_mutex);
                if constexpr (queue)
                {
                    auto& v = std::get<std::queue<thread_function>>(func_queue);
                    v.push(func);
                    tasks++;
                } else
                {
                    func_queue = func;
                }
            }
            
            [[nodiscard]] inline bool tasks_complete() const
            {
                return completed_tasks == tasks;
            }
            
            [[nodiscard]] inline bool complete() const
            {
                return stopped == number_of_threads;
            }
            
            inline void stop()
            {
                should_stop = true;
            }
            
            inline void reset_tasks()
            {
                tasks = 0;
                completed_tasks = 0;
            }
            
            inline void reset()
            {
                stop();
                cleanup();
                stopped = 0;
                init();
            }
            
            ~thread_pool()
            {
                should_stop = true;
                cleanup();
            }
    };
}

#endif //BLT_THREAD_H