PerByte.hpp

Go to the documentation of this file.

#ifndef BYTEME_PERBYTE_HPP
#define BYTEME_PERBYTE_HPP
 
#include <thread>
#include <condition_variable>
#include <mutex>
#include <vector>
#include <algorithm>
#include <exception>
#include <type_traits>
#include <memory>
#include <cstddef>
 
#include "Reader.hpp"
 
namespace byteme {
 
template<class Reader_>
void skip_zero_buffers(Reader_& reader, std::size_t& available) {
    available = 0;
    while (reader.load()) {
        available = reader.available(); // continue collecting bytes if a zero-length buffer is returned without load() returning false.
        if (available) {
            break;
        }
    } 
 
    // If available == 0 on return, then reader->load() must be false,
    // and there are no more bytes left in the source.
}
template<typename Type_>
class PerByteInterface {
public:
    PerByteInterface() = default;
    PerByteInterface(PerByteInterface&&) = delete;
    PerByteInterface(const PerByteInterface&) = delete; // not copyable.
    PerByteInterface& operator=(PerByteInterface&&) = delete;
    PerByteInterface& operator=(const PerByteInterface&) = delete; // not copyable.
    virtual ~PerByteInterface() = default;
protected:
    const Type_* ptr = nullptr;
 
    std::size_t available = 0;
 
    virtual void refill() = 0;
 
private:
    std::size_t my_current = 0;
 
    // Standard guarantees this to be at least 64 bits, which is more than that of size_t.
    // We don't use uint64_t because that might not be defined by the implementation.
    unsigned long long my_overall = 0;
 
public:
    bool advance() {
        ++my_current;
        if (my_current < available) {
            return true;
        }
 
        my_current = 0;
        my_overall += available;
        refill();
        return available > 0; // Check that we haven't reached the end of the reader.
    }
 
    Type_ get() const {
        return ptr[my_current];
    }
 
    unsigned long long position() const {
        return my_overall + my_current;
    }
 
    bool valid() const {
        return my_current < available;
    }
 
public:
    std::pair<std::size_t, bool> extract(std::size_t number, Type_* output) {
        std::size_t original = number;
        bool okay = true;
 
        while (1) {
            auto start = ptr + my_current;
            auto leftover = available - my_current;
 
            if (leftover > number) {
                my_current += number;
                number = 0;
                std::copy(start, ptr + my_current, output);
                break;
 
            } else {
                number -= leftover;
                std::copy(start, ptr + available, output);
 
                my_current = 0;
                my_overall += available;
                refill();
 
                okay = (available > 0);
                if (number == 0 || !okay) {
                    break;
                }
                output += leftover;
            }
        }
 
        return std::make_pair(original - number, okay);
    }
};
 
 
template<typename Type_, class Pointer_ = std::unique_ptr<Reader> >
class PerByteSerial final : public PerByteInterface<Type_> {
public:
    PerByteSerial(Pointer_ reader) : my_reader(std::move(reader)) {
        refill();
    }
 
private:
    Pointer_ my_reader;
 
protected:
    void refill() {
        skip_zero_buffers(*my_reader, this->available);
        this->ptr = reinterpret_cast<const Type_*>(my_reader->buffer());
    }
};
 
template<typename Type_, class Pointer_ = std::unique_ptr<Reader> >
class PerByteParallel final : public PerByteInterface<Type_> {
public:
    PerByteParallel(Pointer_ reader) : my_reader(std::move(reader)) {
        my_ready_input = false;
        my_thread = std::thread([&]() { thread_loop(); });
 
        skip_zero_buffers(*my_reader, my_next_available);
        my_ready_output = true; // run the first iteration of refill().
        refill();
    }
 
    ~PerByteParallel() {
        if (!my_finished) {
            std::unique_lock lck(my_mut);
            my_finished = true;
            my_ready_input = true;
            lck.unlock(); // releasing the lock so that the notified thread doesn't immediately block.
            my_cv.notify_one();
        }
        my_thread.join();
    }
private:
    Pointer_ my_reader;
    std::vector<Type_> my_buffer;
    std::size_t my_next_available = 0;
    bool my_finished = false;
 
private:
    std::thread my_thread;
    std::exception_ptr my_thread_err = nullptr;
    std::mutex my_mut;
    std::condition_variable my_cv;
    bool my_ready_input, my_ready_output;
 
    void thread_loop() {
        while (!my_finished) {
            std::unique_lock lck(my_mut);
            my_cv.wait(lck, [&]() { return my_ready_input; });
            my_ready_input = false;
 
            if (my_finished) { // an explicit kill signal from the destructor.
                break;
            }
 
            try {
                skip_zero_buffers(*my_reader, my_next_available);
                my_finished = my_next_available == 0; // see the definition of skip_zero_buffers().
            } catch (...) {
                my_thread_err = std::current_exception();
                my_finished = true;
            }
 
            my_ready_output = true;
            lck.unlock();
            my_cv.notify_one();
        }
    }
 
protected:
    void refill() {
        if (my_finished) {
            this->ptr = nullptr;
            this->available = 0;
            return;
        }
 
        std::unique_lock lck(my_mut);
        my_cv.wait(lck, [&]() { return my_ready_output; });
        my_ready_output = false;
        if (my_thread_err) {
            std::rethrow_exception(my_thread_err);
        }
 
        auto rptr = reinterpret_cast<const Type_*>(my_reader->buffer());
        this->available = my_next_available;
        my_buffer.resize(this->available);
        std::copy_n(rptr, this->available, my_buffer.data());
        this->ptr = my_buffer.data();
 
        my_ready_input = true;
        lck.unlock();
        if (!my_finished) {
            my_cv.notify_one();
        }
    }
};
 
}
 
#endif