AuroraRuntime/Source/Threading/Primitives/AuRWLock.hpp

/***
    Copyright (C) 2021-2024 J Reece Wilson (a/k/a "Reece"). All rights reserved.

    File: AuRWLock.hpp
    Date: 2021-6-12
    Author: Reece
***/
#pragma once

#include "AuConditionVariable.Generic.hpp"
#include "AuConditionMutex.Generic.hpp"
#include "ThreadCookie.hpp"

#if defined(AURORA_IS_LINUX_DERIVED)
    // Disable Windows XP - 7 keyedevent inline optimization (not that we couldn't use AURORA_RWLOCK_IS_CONDLESS on XP/7 with WOA_ALWAYS_DUMB_OS_TARGET).
    // Note that *a lot* of platforms may end up requiring all thread primitives to use more verbose primitives.
    // Keeping the condvar impl around is best in the event we need to optimize for unique platforms of sensitive schedulers.
    // The noisey use-case of the semaphore hash table isn't what we always want for persistent objects with QoS properties.

    // In terms of platform support/what-to-do-where:
    //  Here's one example of a QoS senitive OS without a futex wait queue interface we might care about: > MACOS < 
    //  Now let's ignore the updoooters and support Windows XP - 7...
    //  ...this is a Linux specific thing
    //
    //  Windows XP-7: same as 11 (no such win32 alternative)                               (EOL support)
    //  Windows 11 updoooters can cope with a 56-byte primitive in the worst case scenario (who even cares case?)
    //  Generic: 64-byte (no config, maybe use AURORA_RWLOCK_IS_CONDLESS)                  (expected case)
    //  Linux: AURORA_RWLOCK_IS_CONDLESS best case, 64-byte without                        (best case)
    //  AURORA_RWLOCK_IS_CONDLESS: 48-byte                                                 (best case)
    //     vs couple hundred byte primitives in the STL
    //     in either case AURORA_RWLOCK_IS_CONDLESS doesnt matter so much (delta linux:16, win32: measly 8 bytes)
    //     in either case performance is the same
    #define AURORA_RWLOCK_IS_CONDLESS
#endif

namespace Aurora::Threading::Primitives
{
    template<bool bIsWriteRecursionAllowed>
    struct RWLockImpl;

    template<bool bIsReadView, typename T>
    struct RWLockAccessView final : IWaitable
    {
    #if defined(RWLOCK_VIEW_HAS_PARENT)
        RWLockAccessView(T &impl) :
            parent_(impl)
        {

        }
    #endif

        bool LockMS(AuUInt64 timeout) override;
        bool LockNS(AuUInt64 timeout) override;
        bool LockAbsMS(AuUInt64 timeout) override;
        bool LockAbsNS(AuUInt64 timeout) override;

        bool TryLock() override;

        bool HasOSHandle(AuMach &mach) override
        {
            return false;
        }

        bool HasLockImplementation() override
        {
            return true;
        }

        void Lock() override
        {
            SysAssert(LockNS(0));
        }

        void Unlock() override;

    private:
    #if defined(RWLOCK_VIEW_HAS_PARENT)
        T &parent_;
    #endif
    };

    template<bool bIsWriteRecursionAllowed>
    struct RWLockImpl final : IRWLock
    {
        RWLockImpl();
        ~RWLockImpl();
        
        // i dont think i'll expose a high performance interface yet
        auline bool LockReadNSAbs(AuUInt64 timeout);// override;
        auline bool LockReadNS(AuUInt64 timeout);// override;
        auline bool LockWriteNS(AuUInt64 timeout);// override;
        auline bool LockWriteNSAbs(AuUInt64 timeout);// override;
        auline bool TryLockRead();// override;
        template<bool bCheckWrite>
        auline bool TryLockReadNoSpin();
        auline void UnlockRead();// override;
        auline void UnlockWrite();// override;

        auline void WriterWake();
        auline bool WriterSleep(AuUInt64 qwTimeoutNS);
        auline bool WriterTryLock();
        auline void FutexWriterWake();
        auline bool TryLockWriteMaybeSpin();// override;
        auline bool TryLockWriteNoSpin();// override;
        auline bool LockWriteNSAbsUnlocked(AuUInt64 qwTimeoutNS);// override;

        bool UpgradeReadToWrite(AuUInt64 timeout) override;
        bool DowngradeWriteToRead() override;
        auline bool UpgradeReadToWriteDoUpgrade();
        
        IWaitable *AsReadable() override;
        IWaitable *AsWritable() override;

        bool CheckSelfThreadIsWriter() override;

        auline ConditionVariableInternal &GetCondition();
        auline ConditionVariableInternal &GetConditionWriter();

        auline AuUInt32 *GetFutexCondition();
        auline AuUInt32 *GetFutexConditionWriter();

        auline void SignalOneReader();
        auline void SignalOneWriter();
        auline void SignalManyReader();
        auline void SignalManyWriter(int iBias = 0);
        
        auline AuUInt32 *GetReadSleepCounter();

    public: // "private"
        RWLockAccessView<true, RWLockImpl>  read_;
        RWLockAccessView<false, RWLockImpl> write_;
    
    private:
        ThreadCookie_t reentrantWriteLockHandle_ {};
    #if defined(AURORA_RWLOCK_IS_CONDLESS)
        AuAUInt32 uSemaphore_ {};
        AuAUInt32 uRDCounter_ {};
    #else
        ConditionMutexInternal mutex_;
    #if defined(AURWLOCK_NO_SIZE_OPTIMIZED_CONDVAR)
        ConditionVariableInternal condition_;
        ConditionVariableInternal conditionWriter_;
    #else
        char conditionVariable_[kSizeOfDummyCondVar] {};
        char conditionVariableWriter_[kSizeOfDummyCondVar] {};
    #endif
    #endif
        AuAInt32 iState_ {};
        AuAInt32 dwWritersPending_ {};
    };
}