citra/src/core/core_timing.h

// Copyright 2008 Dolphin Emulator Project / 2017 Citra Emulator Project
// Licensed under GPLv2+
// Refer to the license.txt file included.

#pragma once

/**
 * This is a system to schedule events into the emulated machine's future. Time is measured
 * in main CPU clock cycles.
 *
 * To schedule an event, you first have to register its type. This is where you pass in the
 * callback. You then schedule events using the type id you get back.
 *
 * The int cyclesLate that the callbacks get is how many cycles late it was.
 * So to schedule a new event on a regular basis:
 * inside callback:
 *   ScheduleEvent(periodInCycles - cyclesLate, callback, "whatever")
 */

#include <chrono>
#include <functional>
#include <limits>
#include <string>
#include <unordered_map>
#include <vector>
#include <boost/serialization/split_member.hpp>
#include <boost/serialization/vector.hpp>
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/threadsafe_queue.h"

// The timing we get from the assembly is 268,111,855.956 Hz
// It is possible that this number isn't just an integer because the compiler could have
// optimized the multiplication by a multiply-by-constant division.
// Rounding to the nearest integer should be fine
constexpr u64 BASE_CLOCK_RATE_ARM11 = 268111856;
constexpr u64 MAX_VALUE_TO_MULTIPLY = std::numeric_limits<s64>::max() / BASE_CLOCK_RATE_ARM11;

inline s64 msToCycles(int ms) {
    // since ms is int there is no way to overflow
    return BASE_CLOCK_RATE_ARM11 * static_cast<s64>(ms) / 1000;
}

inline s64 msToCycles(float ms) {
    return static_cast<s64>(BASE_CLOCK_RATE_ARM11 * (0.001f) * ms);
}

inline s64 msToCycles(double ms) {
    return static_cast<s64>(BASE_CLOCK_RATE_ARM11 * (0.001) * ms);
}

inline s64 usToCycles(float us) {
    return static_cast<s64>(BASE_CLOCK_RATE_ARM11 * (0.000001f) * us);
}

inline s64 usToCycles(int us) {
    return (BASE_CLOCK_RATE_ARM11 * static_cast<s64>(us) / 1000000);
}

inline s64 usToCycles(s64 us) {
    if (us / 1000000 > static_cast<s64>(MAX_VALUE_TO_MULTIPLY)) {
        LOG_ERROR(Core_Timing, "Integer overflow, use max value");
        return std::numeric_limits<s64>::max();
    }
    if (us > static_cast<s64>(MAX_VALUE_TO_MULTIPLY)) {
        LOG_DEBUG(Core_Timing, "Time very big, do rounding");
        return BASE_CLOCK_RATE_ARM11 * (us / 1000000);
    }
    return (BASE_CLOCK_RATE_ARM11 * us) / 1000000;
}

inline s64 usToCycles(u64 us) {
    if (us / 1000000 > MAX_VALUE_TO_MULTIPLY) {
        LOG_ERROR(Core_Timing, "Integer overflow, use max value");
        return std::numeric_limits<s64>::max();
    }
    if (us > MAX_VALUE_TO_MULTIPLY) {
        LOG_DEBUG(Core_Timing, "Time very big, do rounding");
        return BASE_CLOCK_RATE_ARM11 * static_cast<s64>(us / 1000000);
    }
    return (BASE_CLOCK_RATE_ARM11 * static_cast<s64>(us)) / 1000000;
}

inline s64 nsToCycles(float ns) {
    return static_cast<s64>(BASE_CLOCK_RATE_ARM11 * (0.000000001f) * ns);
}

inline s64 nsToCycles(int ns) {
    return BASE_CLOCK_RATE_ARM11 * static_cast<s64>(ns) / 1000000000;
}

inline s64 nsToCycles(s64 ns) {
    if (ns / 1000000000 > static_cast<s64>(MAX_VALUE_TO_MULTIPLY)) {
        LOG_ERROR(Core_Timing, "Integer overflow, use max value");
        return std::numeric_limits<s64>::max();
    }
    if (ns > static_cast<s64>(MAX_VALUE_TO_MULTIPLY)) {
        LOG_DEBUG(Core_Timing, "Time very big, do rounding");
        return BASE_CLOCK_RATE_ARM11 * (ns / 1000000000);
    }
    return (BASE_CLOCK_RATE_ARM11 * ns) / 1000000000;
}

inline s64 nsToCycles(u64 ns) {
    if (ns / 1000000000 > MAX_VALUE_TO_MULTIPLY) {
        LOG_ERROR(Core_Timing, "Integer overflow, use max value");
        return std::numeric_limits<s64>::max();
    }
    if (ns > MAX_VALUE_TO_MULTIPLY) {
        LOG_DEBUG(Core_Timing, "Time very big, do rounding");
        return BASE_CLOCK_RATE_ARM11 * (static_cast<s64>(ns) / 1000000000);
    }
    return (BASE_CLOCK_RATE_ARM11 * static_cast<s64>(ns)) / 1000000000;
}

inline u64 cyclesToNs(s64 cycles) {
    return cycles * 1000000000 / BASE_CLOCK_RATE_ARM11;
}

inline s64 cyclesToUs(s64 cycles) {
    return cycles * 1000000 / BASE_CLOCK_RATE_ARM11;
}

inline u64 cyclesToMs(s64 cycles) {
    return cycles * 1000 / BASE_CLOCK_RATE_ARM11;
}

namespace Core {

using TimedCallback = std::function<void(u64 userdata, int cycles_late)>;

struct TimingEventType {
    TimedCallback callback;
    const std::string* name;
};

class Timing {
public:
    ~Timing();

    /**
     * This should only be called from the emu thread, if you are calling it any other thread, you
     * are doing something evil
     */
    u64 GetTicks() const;
    u64 GetIdleTicks() const;
    void AddTicks(u64 ticks);

    /**
     * Returns the event_type identifier. if name is not unique, it will assert.
     */
    TimingEventType* RegisterEvent(const std::string& name, TimedCallback callback);

    /**
     * After the first Advance, the slice lengths and the downcount will be reduced whenever an
     * event is scheduled earlier than the current values. Scheduling from a callback will not
     * update the downcount until the Advance() completes.
     */
    void ScheduleEvent(s64 cycles_into_future, const TimingEventType* event_type, u64 userdata = 0);

    /**
     * This is to be called when outside of hle threads, such as the graphics thread, wants to
     * schedule things to be executed on the main thread.
     * Not that this doesn't change slice_length and thus events scheduled by this might be called
     * with a delay of up to MAX_SLICE_LENGTH
     */
    void ScheduleEventThreadsafe(s64 cycles_into_future, const TimingEventType* event_type,
                                 u64 userdata);

    void UnscheduleEvent(const TimingEventType* event_type, u64 userdata);

    /// We only permit one event of each type in the queue at a time.
    void RemoveEvent(const TimingEventType* event_type);
    void RemoveNormalAndThreadsafeEvent(const TimingEventType* event_type);

    /** Advance must be called at the beginning of dispatcher loops, not the end. Advance() ends
     * the previous timing slice and begins the next one, you must Advance from the previous
     * slice to the current one before executing any cycles. CoreTiming starts in slice -1 so an
     * Advance() is required to initialize the slice length before the first cycle of emulated
     * instructions is executed.
     */
    void Advance();
    void MoveEvents();

    /// Pretend that the main CPU has executed enough cycles to reach the next event.
    void Idle();

    void ForceExceptionCheck(s64 cycles);

    std::chrono::microseconds GetGlobalTimeUs() const;

    s64 GetDowncount() const;

private:
    static Timing* deserializing;

    struct Event {
        s64 time;
        u64 fifo_order;
        u64 userdata;
        const TimingEventType* type;

        bool operator>(const Event& right) const;
        bool operator<(const Event& right) const;

    private:
        template <class Archive>
        void save(Archive& ar, const unsigned int) const {
            ar& time;
            ar& fifo_order;
            ar& userdata;
            std::string name = *(type->name);
            ar << name;
        }

        template <class Archive>
        void load(Archive& ar, const unsigned int) {
            ar& time;
            ar& fifo_order;
            ar& userdata;
            std::string name;
            ar >> name;
            type = Timing::deserializing->RegisterEvent(name, nullptr);
        }
        friend class boost::serialization::access;

        BOOST_SERIALIZATION_SPLIT_MEMBER()
    };

    static constexpr int MAX_SLICE_LENGTH = 20000;

    s64 global_timer = 0;
    s64 slice_length = MAX_SLICE_LENGTH;
    s64 downcount = MAX_SLICE_LENGTH;

    // unordered_map stores each element separately as a linked list node so pointers to
    // elements remain stable regardless of rehashes/resizing.
    std::unordered_map<std::string, TimingEventType> event_types;

    // The queue is a min-heap using std::make_heap/push_heap/pop_heap.
    // We don't use std::priority_queue because we need to be able to serialize, unserialize and
    // erase arbitrary events (RemoveEvent()) regardless of the queue order. These aren't
    // accomodated by the standard adaptor class.
    std::vector<Event> event_queue;
    u64 event_fifo_id = 0;
    // the queue for storing the events from other threads threadsafe until they will be added
    // to the event_queue by the emu thread
    Common::MPSCQueue<Event> ts_queue;
    s64 idled_cycles = 0;

    // Are we in a function that has been called from Advance()
    // If events are sheduled from a function that gets called from Advance(),
    // don't change slice_length and downcount.
    // The time between CoreTiming being intialized and the first call to Advance() is considered
    // the slice boundary between slice -1 and slice 0. Dispatcher loops must call Advance() before
    // executing the first cycle of each slice to prepare the slice length and downcount for
    // that slice.
    bool is_global_timer_sane = true;

    template <class Archive>
    void serialize(Archive& ar, const unsigned int) {
        // event_types set during initialization of other things
        deserializing = this;
        MoveEvents();
        ar& global_timer;
        ar& slice_length;
        ar& downcount;
        ar& event_queue;
        ar& event_fifo_id;
        ar& idled_cycles;
        deserializing = nullptr;
    }
    friend class boost::serialization::access;
};

} // namespace Core