AuroraSupport/AuroraRuntime
2
1
Fork 0
Code Issues 29 Pull Requests Releases Wiki Activity
e6c2a5c9de
AuroraRuntime/Source/Threading/AuWakeOnAddress.cpp
Jamie Reece Wilson e6c2a5c9de [*] Optimize for legacy 32bit Windows targets "*" 
[*] 32bit compilation regressions

*: It's actually worse for Windows Vista and Windows 7. While netbooks of this era should be running 32bit builds to mitigate oversized words, Windows 7 and Vista had decent enough 64bit install bases; XP, Windows 2000, Windows 98, and NT4 are much more relevant 32bit targets
2024-12-27 14:01:36 +00:00

2367 lines
79 KiB
C++
Raw Blame History

/***
Copyright (C) 2023 J Reece Wilson (a/k/a "Reece"). All rights reserved.
File: AuWakeOnAddress.cpp
Date: 2023-3-10
Author: Reece
***/
#if defined(AURORA_COMPILER_MSVC)
#pragma strict_gs_check(off)
#pragma check_stack(off)
#endif
#include <Source/RuntimeInternal.hpp>
#include "AuWakeOnAddress.hpp"
#include "Primitives/SMTYield.hpp"
#include <Time/Time.hpp>
#define HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD
// WOA_ALWAYS_DUMB_OS_TARGET -> iOS, notarized MacOS, Win9x, Xbox 360, etc
// This will tank the average case:
// > fast paths will be disabled,
// > lock barriers turn into full lock guards,
// > and every sleep will require a condvar or semaphore wakeup.
// Every perf trick will be bypassed.
// Update: partially removed. its not ready yet
//#define WOA_STRICTER_FIFO
//#define WOA_WAIT_SIDE_NO_ABSORB
namespace Aurora::Threading
{
static WaitEntry **GetPBeforeFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress);
static WaitEntry **GetPNextFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress);
static WaitEntry **GetPLastFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress);
static WaitEntry **GetPFirstFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress);
static const void *GetPCompareFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress);
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
static thread_local AuSPtr<WaitEntry> tlsWaitEntry = AuMakeSharedPanic<WaitEntry>();
#else
static thread_local WaitEntry tlsWaitEntry;
#endif
#define DO_OF_METHOD_TYPE(preface, DoOfMethodType, ...) \
switch (eMethod) \
{ \
case EWaitMethod::eNotEqual: \
preface DoOfMethodType<EWaitMethod::eNotEqual>(__VA_ARGS__); \
break; \
case EWaitMethod::eEqual: \
preface DoOfMethodType<EWaitMethod::eEqual>(__VA_ARGS__); \
break; \
case EWaitMethod::eGreaterThanCompare: \
preface DoOfMethodType<EWaitMethod::eGreaterThanCompare>(__VA_ARGS__); \
break; \
case EWaitMethod::eGreaterThanOrEqualsCompare: \
preface DoOfMethodType<EWaitMethod::eGreaterThanOrEqualsCompare>(__VA_ARGS__); \
break; \
case EWaitMethod::eLessThanCompare: \
preface DoOfMethodType<EWaitMethod::eLessThanCompare>(__VA_ARGS__); \
break; \
case EWaitMethod::eLessThanOrEqualsCompare: \
preface DoOfMethodType<EWaitMethod::eLessThanOrEqualsCompare>(__VA_ARGS__); \
break; \
case EWaitMethod::eAnd: \
preface DoOfMethodType<EWaitMethod::eAnd>(__VA_ARGS__); \
break; \
case EWaitMethod::eNotAnd: \
preface DoOfMethodType<EWaitMethod::eNotAnd>(__VA_ARGS__); \
break; \
}
static const int gShouldSpinOnlyInCPU = 1; // TODO: havent decided
// UPDATE: 1 paranoia just in case we get preempted (rare).
template<typename T>
static void DoSpinLockOnVar(T *uPointer)
{
if (gShouldSpinOnlyInCPU == 0)
{
while (!Primitives::DoTryIfAlderLake([&]()
{
return AuAtomicTestAndSet(uPointer, 0) == 0;
}, uPointer))
{
}
}
else if (gShouldSpinOnlyInCPU == 1)
{
while (!Primitives::DoTryIfAlderLake([&]()
{
return AuAtomicTestAndSet(uPointer, 0) == 0;
}, uPointer))
{
ContextYield();
}
}
else if (gShouldSpinOnlyInCPU == 2)
{
while (AuAtomicTestAndSet(uPointer, 0))
{
while (*uPointer)
{
ContextYield();
}
}
}
else
{
SysUnreachable();
}
}
void WaitEntry::Release()
{
AuResetMember(this->uSize);
AuResetMember(this->pAddress);
}
WaitEntry::WaitEntry()
{
}
WaitEntry::~WaitEntry()
{
this->Release();
}
template <EWaitMethod eMethod>
bool WaitEntry::SleepOn(WaitState &state)
{
if (state.qwNanosecondsAbs)
{
#if !defined(WOA_STRICTER_FIFO)
if (!WaitBuffer::Compare2<eMethod, true>(this->pAddress, this->uSize, state.pCompare2))
{
return true;
}
#endif
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Lock();
#endif
auto uNow = AuTime::SteadyClockNS();
auto uEndTime = state.qwNanosecondsAbs.value();
while (uNow < uEndTime)
{
bool bStatus {};
#if defined(AURORA_PLATFORM_WIN32)
Win32DropSchedulerResolution();
#endif
if (!AuAtomicLoad(&this->bAlive))
{
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Unlock();
#endif
(void)gProcessWaitables.WaitBufferFrom<eMethod>(this->pAddress, this->uSize, false, state.pCompare2);
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Lock();
#endif
continue;
}
{
#if defined(WOA_SEMAPHORE_MODE)
bStatus = this->semaphore->LockAbsNS(uEndTime);
#if defined(WOA_WAIT_SIDE_NO_ABSORB) || defined(WOA_STRICTER_FIFO)
if (bStatus)
{
return true;
}
#else
if (!WaitBuffer::Compare2<eMethod, true>(this->pAddress, this->uSize, state.pCompare2))
{
return true;
}
#endif
#else
if (!WaitBuffer::Compare2<eMethod, true>(this->pAddress, this->uSize, state.pCompare2))
{
this->mutex.Unlock();
return true;
}
else
{
auto uTimeRemNS = uEndTime - uNow;
bStatus = this->variable.WaitForSignalNsEx(&this->mutex, uTimeRemNS, false);
#if defined(WOA_WAIT_SIDE_NO_ABSORB)
if (bStatus)
{
return true;
}
#endif
}
#endif
}
uNow = AuTime::SteadyClockNS();
}
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Unlock();
#endif
return !WaitBuffer::Compare2<eMethod, true>(this->pAddress, this->uSize, state.pCompare2);
}
else
{
while (true)
{
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Lock();
#endif
if (!AuAtomicLoad(&this->bAlive))
{
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Unlock();
#endif
return true;
}
{
#if defined(WOA_SEMAPHORE_MODE)
this->semaphore->Lock();
#else
this->variable.WaitForSignalNsEx(&this->mutex, 0, false);
#endif
}
#if !defined(WOA_SEMAPHORE_MODE)
this->mutex.Unlock();
#endif
#if defined(WOA_WAIT_SIDE_NO_ABSORB)
return true;
#else
if (!WaitBuffer::Compare2<eMethod, true>(this->pAddress, this->uSize, state.pCompare2))
{
return true;
}
else
{
(void)gProcessWaitables.WaitBufferFrom<eMethod>(this->pAddress, this->uSize, false, state.pCompare2);
}
#endif
}
}
}
bool WaitEntry::SleepLossy(AuUInt64 qwNanosecondsAbs)
{
#if !defined(WOA_SEMAPHORE_MODE)
AU_LOCK_GUARD(this->mutex);
#endif
if (qwNanosecondsAbs)
{
#if defined(WOA_SEMAPHORE_MODE)
return this->semaphore->LockAbsNS(qwNanosecondsAbs);
#else
auto uNow = AuTime::SteadyClockNS();
while (uNow < qwNanosecondsAbs)
{
if (!AuAtomicLoad(&this->bAlive))
{
return true;
}
auto uTimeRemNS = qwNanosecondsAbs - uNow;
if (this->variable.WaitForSignalNsEx(&this->mutex, uTimeRemNS, false))
{
return true;
}
uNow = AuTime::SteadyClockNS();
}
return false;
#endif
}
else
{
if (!AuAtomicLoad(&this->bAlive))
{
return true;
}
#if defined(WOA_SEMAPHORE_MODE)
this->semaphore->Lock();
#else
this->variable.WaitForSignalNsEx(&this->mutex, 0, false);
#endif
return true;
}
return false;
}
WaitEntry *WaitEntry::GetBefore(const void *pAddress)
{
if (auto pSpecial = this->pSpecial)
{
return *GetPBeforeFromContainer(pSpecial, pAddress);
}
else
{
return this->pBefore;
}
}
WaitEntry *WaitEntry::GetSimilarFirstItr(const void *pAddress)
{
if (auto pSpecial = this->pSpecial)
{
// non-tls double-linked list recovery: traverse the WaitMulipleContainer given the collisions for pAddress
if (auto ppNextHead = GetPFirstFromContainer(pSpecial, pAddress))
{
return *ppNextHead;
}
else
{
return nullptr;
}
}
else
{
// old logic / no wait multiple: traverse backwards
return this->pBefore;
}
}
void WaitEntry::SetBefore(const void *pAddress, WaitEntry *pNext)
{
if (auto pSpecial = this->pSpecial)
{
if (auto pNextEntry = GetPBeforeFromContainer(pSpecial, pAddress))
{
*pNextEntry = pNext;
}
else
{
SysUnreachable();
}
}
else
{
this->pBefore = pNext;
}
}
WaitEntry *WaitEntry::GetNext(const void *pAddress)
{
if (auto pSpecial = this->pSpecial)
{
return *GetPNextFromContainer(pSpecial, pAddress);
}
else
{
return this->pNext;
}
}
WaitEntry *WaitEntry::GetSimilarLastItr(const void *pAddress)
{
if (auto pSpecial = this->pSpecial)
{
if (auto ppNextHead = GetPLastFromContainer(pSpecial, pAddress))
{
return *ppNextHead;
}
else
{
return nullptr;
}
}
else
{
return this->pNext;
}
}
void WaitEntry::SetNext(const void *pAddress, WaitEntry *pNext)
{
if (auto pSpecial = this->pSpecial)
{
if (auto pNextEntry = GetPNextFromContainer(pSpecial, pAddress))
{
*pNextEntry = pNext;
}
else
{
SysUnreachable();
}
}
else
{
this->pNext = pNext;
}
}
bool WaitEntry::TrySignalAddress(const void *pAddress)
{
if (auto &pSpecial = this->pSpecial) AU_EOB_UNLIKELY
{
if (auto pCompare = GetPCompareFromContainer(pSpecial, pAddress))
{
if (WaitBuffer::Compare(pAddress, this->uSize, pCompare, kMax64, this->eWaitMethod))
{
return false;
}
}
if (this->pSpecial->waitArray && gRuntimeConfig.threadingConfig.bEnableWaitOnAddressAndSurpression)
{
auto pFirst = AuReinterpretCast<MultipleInternalContext>(this->pSpecial->waitArray.Begin<WaitMultipleEntry>()->internalContext);
if (AuAtomicAdd(&pFirst->uCounter, 1u) < pFirst->uMinTrigger)
{
// "Successful" - do not dequeue, but consume -1 from "N" of WakeNOnAddress
return true;
}
}
AuAtomicSet(&this->bAlive, 0u);
if (this->bSemaphoreActive &&
this->pSemaphore)
{
this->pSemaphore->AddOne();
}
}
else
{
if (this->pAddress != pAddress)
{
return false;
}
if (WaitBuffer::Compare(pAddress, this->uSize, this->pCompareAddress, kMax64, this->eWaitMethod))
{
return false;
}
AuAtomicSet(&this->bAlive, 0u);
}
#if defined(WOA_SEMAPHORE_MODE)
this->semaphore->Unlock(1);
#else
this->variable.Signal();
#endif
return true;
}
WaitBufferLegacy WaitBufferLegacy::From(const void *pBuf, AuUInt8 uSize)
{
WaitBufferLegacy wait;
AuMemcpy(wait.buffer, pBuf, AuMin(uSize, AuArraySize(wait.buffer)));
wait.uSize = uSize;
return AuMove(wait);
}
bool WaitBuffer::Compare(const void *pHotAddress, AuUInt8 uSize, WaitState &state)
{
return WaitBuffer::Compare(pHotAddress, uSize, state.pCompare2, state.uDownsizeMask, state.eWaitMethod);
}
bool WaitBuffer::Compare(const void *pHotAddress, AuUInt8 uSize, const void *pCompare, AuUInt64 uMask, EWaitMethod eMethod)
{
bool bRet {};
AURORA_COMPILER_VOLATILE_BARRIER();
DO_OF_METHOD_TYPE(return, Compare2, pHotAddress, uSize, pCompare)
return bRet;
}
template <EWaitMethod eMethod, bool bFast>
bool WaitBuffer::Compare2(const void *pHot, AuUInt8 uSize, const void *pBuf2, AuUInt64 uMask)
{
return Compare2<eMethod, bFast>((const volatile void *)pHot, uSize, pBuf2, uMask);
}
template <EWaitMethod eMethod, bool bFast>
bool WaitBuffer::Compare2(const volatile void *pHot, AuUInt8 uSize, const void *pBuf2, AuUInt64 uMask)
{
#if !defined(AURORA_COMPILER_CLANG) && !defined(AURORA_COMPILER_MSVC)
AURORA_COMPILER_VOLATILE_BARRIER();
#endif
if constexpr (!bFast)
{
if constexpr (eMethod == EWaitMethod::eNotEqual)
{
switch (uSize)
{
case 1:
return (AuReadU8(pHot, 0) & uMask) == (AuReadU8(pBuf2, 0) & uMask);
case 2:
return (AuReadU16(pHot, 0) & uMask) == (AuReadU16(pBuf2, 0) & uMask);
case 4:
return (AuReadU32(pHot, 0) & uMask) == (AuReadU32(pBuf2, 0) & uMask);
case 8:
return (AuReadU64(pHot, 0) & uMask) == (AuReadU64(pBuf2, 0) & uMask);
default:
return (AuMemcmp((const void *)pHot, pBuf2, uSize) == 0);
}
}
if constexpr (eMethod == EWaitMethod::eEqual)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) == (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) == (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) == (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) == (AuReadU64(pBuf2, 0) & uMask));
default:
return !(AuMemcmp((const void *)pHot, pBuf2, uSize) == 0);
}
}
if constexpr (eMethod == EWaitMethod::eGreaterThanCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) > (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) > (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) > (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) > (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eGreaterThanOrEqualsCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) >= (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) >= (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) >= (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) >= (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eLessThanCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) < (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) < (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) < (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) < (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eLessThanOrEqualsCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) <= (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) <= (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) <= (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) <= (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eAnd)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0) & uMask) & (AuReadU8(pBuf2, 0) & uMask));
case 2:
return !((AuReadU16(pHot, 0) & uMask) & (AuReadU16(pBuf2, 0) & uMask));
case 4:
return !((AuReadU32(pHot, 0) & uMask) & (AuReadU32(pBuf2, 0) & uMask));
case 8:
return !((AuReadU64(pHot, 0) & uMask) & (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eNotAnd)
{
switch (uSize)
{
case 1:
return ((AuReadU8(pHot, 0) & uMask) & (AuReadU8(pBuf2, 0) & uMask));
case 2:
return ((AuReadU16(pHot, 0) & uMask) & (AuReadU16(pBuf2, 0) & uMask));
case 4:
return ((AuReadU32(pHot, 0) & uMask) & (AuReadU32(pBuf2, 0) & uMask));
case 8:
return ((AuReadU64(pHot, 0) & uMask) & (AuReadU64(pBuf2, 0) & uMask));
default:
return false;
}
}
}
else
{
if constexpr (eMethod == EWaitMethod::eNotEqual)
{
switch (uSize)
{
case 1:
return (AuReadU8(pHot, 0)) == (AuReadU8(pBuf2, 0));
case 2:
return (AuReadU16(pHot, 0)) == (AuReadU16(pBuf2, 0));
case 4:
return (AuReadU32(pHot, 0)) == (AuReadU32(pBuf2, 0));
case 8:
return (AuReadU64(pHot, 0)) == (AuReadU64(pBuf2, 0));
default:
return (AuMemcmp((const void *)pHot, pBuf2, uSize) == 0);
}
}
if constexpr (eMethod == EWaitMethod::eEqual)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0)) == (AuReadU8(pBuf2, 0)));
case 2:
return !((AuReadU16(pHot, 0)) == (AuReadU16(pBuf2, 0)));
case 4:
return !((AuReadU32(pHot, 0)) == (AuReadU32(pBuf2, 0)));
case 8:
return !((AuReadU64(pHot, 0)) == (AuReadU64(pBuf2, 0)));
default:
return !(AuMemcmp((const void *)pHot, pBuf2, uSize) == 0);
}
}
if constexpr (eMethod == EWaitMethod::eGreaterThanCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0)) > (AuReadU8(pBuf2, 0)));
case 2:
return !((AuReadU16(pHot, 0)) > (AuReadU16(pBuf2, 0)));
case 4:
return !((AuReadU32(pHot, 0)) > (AuReadU32(pBuf2, 0)));
case 8:
return !((AuReadU64(pHot, 0)) > (AuReadU64(pBuf2, 0)));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eGreaterThanOrEqualsCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0)) >= (AuReadU8(pBuf2, 0)));
case 2:
return !((AuReadU16(pHot, 0)) >= (AuReadU16(pBuf2, 0)));
case 4:
return !((AuReadU32(pHot, 0)) >= (AuReadU32(pBuf2, 0)));
case 8:
return !((AuReadU64(pHot, 0)) >= (AuReadU64(pBuf2, 0)));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eLessThanCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0)) < (AuReadU8(pBuf2, 0)));
case 2:
return !((AuReadU16(pHot, 0)) < (AuReadU16(pBuf2, 0)));
case 4:
return !((AuReadU32(pHot, 0)) < (AuReadU32(pBuf2, 0)));
case 8:
return !((AuReadU64(pHot, 0)) < (AuReadU64(pBuf2, 0)));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eLessThanOrEqualsCompare)
{
switch (uSize)
{
case 1:
return !((AuReadU8(pHot, 0)) <= (AuReadU8(pBuf2, 0)));
case 2:
return !((AuReadU16(pHot, 0)) <= (AuReadU16(pBuf2, 0)));
case 4:
return !((AuReadU32(pHot, 0)) <= (AuReadU32(pBuf2, 0)));
case 8:
return !((AuReadU64(pHot, 0)) <= (AuReadU64(pBuf2, 0)));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eAnd)
{
switch (uSize)
{
case 1:
return !(AuReadU8(pHot, 0) & AuReadU8(pBuf2, 0));
case 2:
return !(AuReadU16(pHot, 0) & AuReadU16(pBuf2, 0));
case 4:
return !(AuReadU32(pHot, 0) & AuReadU32(pBuf2, 0));
case 8:
return !(AuReadU64(pHot, 0) & AuReadU64(pBuf2, 0));
default:
return false;
}
}
if constexpr (eMethod == EWaitMethod::eNotAnd)
{
switch (uSize)
{
case 1:
return (AuReadU8(pHot, 0) & AuReadU8(pBuf2, 0));
case 2:
return (AuReadU16(pHot, 0) & AuReadU16(pBuf2, 0));
case 4:
return (AuReadU32(pHot, 0) & AuReadU32(pBuf2, 0));
case 8:
return (AuReadU64(pHot, 0) & AuReadU64(pBuf2, 0));
default:
return false;
}
}
}
return false;
}
template <EWaitMethod eMethod>
WaitEntry *ProcessWaitNodeContainer::WaitBufferFrom(const void *pAddress, AuUInt8 uSize, bool bScheduleFirst, const void *pCompareAddress)
{
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pReturn = tlsWaitEntry.get();
#else
auto pReturn = &tlsWaitEntry;
#endif
pReturn->pAddress = pAddress;
pReturn->uSize = uSize;
pReturn->pCompareAddress = pCompareAddress;
pReturn->eWaitMethod = eMethod;
pReturn->pSpecial = nullptr;
Lock();
if (!WaitBuffer::Compare2<eMethod, true>(pAddress, uSize, pCompareAddress))
{
pReturn->bAlive = false;
Unlock();
return nullptr;
}
if (!pReturn->bAlive)
{
pReturn->bAlive = true;
if (bScheduleFirst /*First in, First Out*/)
{
if (auto pLoadFromMemory = this->waitList.pHead)
{
pLoadFromMemory->SetBefore(pAddress, pReturn);
pReturn->SetNext(pAddress, pLoadFromMemory);
}
else
{
this->waitList.pTail = pReturn;
}
this->waitList.pHead = pReturn;
}
else /*Last In, First Out*/
{
if (auto pLoadFromMemory = this->waitList.pTail)
{
pLoadFromMemory->SetNext(pAddress, pReturn);
pReturn->SetBefore(pAddress, pLoadFromMemory);
}
else
{
this->waitList.pHead = pReturn;
}
this->waitList.pTail = pReturn;
}
}
Unlock();
return pReturn;
}
WaitEntry *ProcessWaitNodeContainer::WaitBufferFrom2(const void *pAddress,
AuUInt8 uSize,
const void *pAddressCompare,
EWaitMethod eWaitMethod,
MultipleInternalContext *pContext,
const WaitMulipleContainer *pContainer)
{
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pReturn = tlsWaitEntry.get();
#else
auto pReturn = &tlsWaitEntry;
#endif
pReturn->pAddress = pAddress;
pReturn->uSize = uSize;
pReturn->pCompareAddress = pAddressCompare;
pReturn->eWaitMethod = eWaitMethod;
Lock();
if (!WaitBuffer::Compare(pAddress, uSize, pAddressCompare, kMax64, eWaitMethod))
{
pReturn->bAlive = false;
Unlock();
return nullptr;
}
bool bAddToArray {};
if (!pReturn->bAlive)
{
pReturn->bAlive = true;
bAddToArray = true;
}
else
{
// TODO: traverse list and reject duplicates
bAddToArray = true;
}
if (bAddToArray)
{
if (auto pLoadFromMemory = this->waitList.pHead)
{
pLoadFromMemory->SetBefore(pAddress, pReturn);
pReturn->SetNext(pAddress, pLoadFromMemory);
}
else
{
this->waitList.pTail = pReturn;
}
this->waitList.pHead = pReturn;
}
pReturn->pSpecial = pContainer;
Unlock();
return pReturn;
}
template <typename T>
bool ProcessWaitNodeContainer::IterateWake(const void *pAddress, T callback)
{
bool bRetStatus { true };
if (AuAtomicLoad((AuUInt *)&this->waitList.pTail) == 0)
{
return true;
}
Lock();
{
// FIFO
auto pCurrentHead = this->waitList.pTail;
while (pCurrentHead)
{
decltype(pCurrentHead) pBefore {};
#if !defined(WOA_SEMAPHORE_MODE) && defined(WOA_STRICTER_FIFO)
AU_LOCK_GUARD(pCurrentHead->mutex);
#elif !defined(WOA_SEMAPHORE_MODE) && !defined(WOA_STRICTER_FIFO)
// Condvar wait-list insertion barrier required for binary-semaphore-like conditions.
// We only need to lock against the { lock() if (should Sleep) { *** waitList++; *** unlock(); yield(); lock() } else { ... } unlocks() } condvar pattern.
// I often only care about the order of ->Signal() after the CondVar::waitList++ to ensure one state change is paired with one waitList signal.
// Don't block during the wakeup check, we only care about observing the condvar or semaphore waitlist, no codeguarding mutex required.
{
AU_LOCK_GUARD(pCurrentHead->mutex);
}
#endif
auto [bCont, bRemove] = callback(*pCurrentHead);
pBefore = pCurrentHead->GetBefore(pAddress);
if (bRemove)
{
this->RemoveEntry<true>(pAddress, pCurrentHead);
}
if (!bCont)
{
bRetStatus = false;
break;
}
if (pBefore == pCurrentHead)
{
break;
}
pCurrentHead = pBefore;
}
}
Unlock();
return bRetStatus;
}
template <bool bAllUnderLock>
void ProcessWaitNodeContainer::RemoveEntry(const void *pAddress, WaitEntry *pEntry)
{
#if 0
auto pNext = pEntry->GetNext(pAddress);
auto pBefore = pEntry->GetBefore(pAddress);
if (this->waitList.pHead == pEntry)
{
this->waitList.pHead = pNext;
}
if (this->waitList.pTail == pEntry)
{
this->waitList.pTail = pBefore;
}
// handle shared heads by tranversing collisions within the callers list instead of TLS.
// for the best case and original logic, the above still applies.
#else
WaitEntry *pNext, *pBefore;
if (auto pSpecial = pEntry->pSpecial)
{
pBefore = *GetPBeforeFromContainer(pSpecial, pAddress);
pNext = *GetPNextFromContainer(pSpecial, pAddress);
if (this->waitList.pHead == pEntry)
{
WaitEntry *pNextNextIterator { pEntry }, *pNext {};
while (pNextNextIterator &&
(pNextNextIterator = pNextNextIterator->GetSimilarLastItr(pAddress)))
{
pNext = pNextNextIterator;
}
this->waitList.pHead = pNext;
}
if (this->waitList.pTail == pEntry)
{
WaitEntry *pNextBeforeIterator { pEntry }, *pBefore {};
while (pNextBeforeIterator &&
(pNextBeforeIterator = pNextBeforeIterator->GetSimilarFirstItr(pAddress)))
{
pBefore = pNextBeforeIterator;
}
this->waitList.pTail = pBefore;
}
}
else
{
pBefore = pEntry->pBefore;
pNext = pEntry->pNext;
if (this->waitList.pHead == pEntry)
{
this->waitList.pHead = pNext;
}
if (this->waitList.pTail == pEntry)
{
this->waitList.pTail = pBefore;
}
}
#endif
if (pBefore)
{
pBefore->SetNext(pAddress, pNext);
}
if (pNext)
{
pNext->SetBefore(pAddress, pBefore);
}
if (bAllUnderLock)
{
pEntry->SetNext(pAddress, nullptr);
pEntry->SetBefore(pAddress, nullptr);
//pEntry->bAlive = false; - redundant
}
}
void ProcessWaitNodeContainer::RemoveSelf(const void *pAddress, WaitEntry *pSelf)
{
{
this->Lock();
this->RemoveEntry<false>(pAddress, pSelf);
this->Unlock();
}
pSelf->SetBefore(pAddress, nullptr);
pSelf->SetNext(pAddress, nullptr);
pSelf->bAlive = false;
}
void ProcessWaitNodeContainer::Lock()
{
DoSpinLockOnVar(&this->uAtomic);
}
void ProcessWaitNodeContainer::Unlock()
{
AuAtomicClearU8Lock(&this->uAtomic);
}
#define AddressToIndexOp(pAddress) (AuHashCode(pAddress) & (kDefaultWaitPerProcess - 1))
#define AddressToIndex AddressToIndexOp(pAddress)
template <EWaitMethod eMethod>
WaitEntry *ProcessWaitContainer::WaitBufferFrom(const void *pAddress, AuUInt8 uSize, bool bScheduleFirst, const void *pCompareAddress)
{
return this->list[AddressToIndex].WaitBufferFrom<eMethod>(pAddress, uSize, bScheduleFirst, pCompareAddress);
}
WaitEntry *ProcessWaitContainer::WaitBufferFrom2(const void *pAddress, AuUInt8 uSize, const void *pAddressCompare, EWaitMethod eWaitMethod, MultipleInternalContext *pContext, const WaitMulipleContainer *pContainer)
{
return this->list[AddressToIndex].WaitBufferFrom2(pAddress, uSize, pAddressCompare, eWaitMethod, pContext, pContainer);
}
template <typename T>
bool ProcessWaitContainer::IterateWake(const void *pAddress, T callback)
{
return this->list[AddressToIndex].IterateWake(pAddress, callback);
}
void ProcessWaitContainer::RemoveSelf(const void *pAddress, WaitEntry *pSelf)
{
return this->list[AddressToIndex].RemoveSelf(pAddress, pSelf);
}
bool IsNativeWaitOnSupported()
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
return pWaitOnAddress &&
AuSwInfo::IsWindows8Point1OrGreater();
#elif defined(AURORA_PLATFORM_LINUX)
return true;
#else
return SysNativeWaitOnAddressFutexSupported();
#endif
}
AUKN_SYM bool IsWaitOnRecommended()
{
#if defined(WOA_ALWAYS_DUMB_OS_TARGET)
return false;
#endif
#if defined(WOA_STRICTER_FIFO)
return false;
#endif
static AuOptionalEx<bool> gIsWaitOnRecommendedCache {};
if (gIsWaitOnRecommendedCache)
{
return gIsWaitOnRecommendedCache.value();
}
if (Primitives::ThrdCfg::gPreferEmulatedWakeOnAddress)
{
return false;
}
bool bState = IsNativeWaitOnSupported();
gIsWaitOnRecommendedCache = bState;
return bState;
}
/// @deprecated
AUKN_SYM const AuList<AuUInt8> &GetValidWordSizes()
{
static const AuList<AuUInt8> kArray =
#if defined(AURORA_IS_MODERNNT_DERIVED)
{ 1, 2, 4, 8 };
#else
{ 4 };
#endif
return kArray;
}
template <EWaitMethod T>
bool WaitOnAddressWide(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuOptional<AuUInt64> qwNanoseconds,
AuOptional<AuUInt64> qwNanosecondsAbs,
bool bOSSupportsWait,
const void *pCompareAddress2)
{
WaitState state;
auto pWaitEntry = gProcessWaitables.template WaitBufferFrom<T>(pTargetAddress, uWordSize, true, pCompareAddress2);
if (!pWaitEntry)
{
return true;
}
state.pCompare2 = pCompareAddress2;
if (qwNanoseconds)
{
state.qwNanosecondsAbs = AuTime::SteadyClockNS() + qwNanoseconds.value();
}
else if (qwNanosecondsAbs)
{
state.qwNanosecondsAbs = qwNanosecondsAbs.value();
}
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pTempHoldMe = tlsWaitEntry;
#endif
auto bResult = pWaitEntry->template SleepOn<T>(state);
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
pTempHoldMe.reset();
#endif
if (AuAtomicLoad(&pWaitEntry->bAlive))
{
gProcessWaitables.RemoveSelf(pTargetAddress, pWaitEntry);
}
return bResult;
}
AuTuple<const void *, AuUInt8, AuUInt64> DecodeAddress(const void *pAddress,
AuUInt32 uWordSize)
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
return AuMakeTuple(pAddress, 0, kMax64);
#endif
auto pRounded = AuPageRound(AuUInt(pAddress), AuUInt(4));
auto uDelta = (AuUInt)pAddress - pRounded;
if (uWordSize == 8)
{
return AuMakeTuple((const void *)pRounded, uDelta, kMax64);
}
AuUInt32 uSizeMask = (1ull << (uWordSize * 8)) - 1ull;
switch (uDelta)
{
case 0:
return AuMakeTuple(pAddress, 0, 0xFFFFFFFF & (uSizeMask << 0));
case 1:
return AuMakeTuple(pAddress, 1, 0xFFFFFF00 & (uSizeMask << 8));
case 2:
return AuMakeTuple(pAddress, 2, 0xFFFF0000 & (uSizeMask << 16));
case 3:
return AuMakeTuple(pAddress, 3, 0xFF000000 & (uSizeMask << 24));
default:
SysPanic("Invalid Branch");
}
}
static bool RunOSWaitOnAddressTimed(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 uAbsTimeSteadyClock,
AuUInt64 uRelativeNanoseconds,
AuOptional<AuUInt64> uAbsTimeAltClock /* hint */,
bool bSpun = false)
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
if (pRtlWaitOnAddress)
{
AuUInt64 uNow {};
while (uAbsTimeSteadyClock ?
(uAbsTimeSteadyClock > (uNow = AuTime::SteadyClockNS())) :
true)
{
LARGE_INTEGER word {};
if (uAbsTimeAltClock)
{
word.QuadPart = AuTime::ConvertTimestampNs(uAbsTimeAltClock.value());
}
else if (uAbsTimeSteadyClock)
{
if (uAbsTimeSteadyClock <= uNow)
{
return !WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress);
}
word.QuadPart = -(AuInt64(uAbsTimeSteadyClock - uNow) / 100ull);
if (!word.QuadPart)
{
word.QuadPart = 1;
}
}
if (WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress))
{
if (pRtlWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize, &word))
{
return true;
}
else if (!uAbsTimeSteadyClock)
{
return false;
}
}
else
{
return true;
}
}
return false;
}
else
{
// ~~some paths might miss the uRelativeNanoseconds, like cas loops.~~
// most paths will now skimp on the relative values
if (uAbsTimeSteadyClock && !uRelativeNanoseconds)
{
AuInt64 iDelta = uAbsTimeSteadyClock;
iDelta -= AuTime::SteadyClockNS();
if (iDelta <= 0)
{
return !WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress);
}
uRelativeNanoseconds = iDelta;
}
auto uMaxSwitches = gRuntimeConfig.threadingConfig.uUWPNanosecondEmulationMaxYields;
auto bUWPNanosecondEmulationCheckFirst = Primitives::ThrdCfg::gUWPNanosecondEmulationCheckFirst;
// LockN(<1MS) on a platform without that resolution of yielding... damn
auto uMS = AuNSToMS<AuUInt32>(uRelativeNanoseconds);
if (!uMS)
{
// first: cpu spin to avoid the kernel all together
if (!bSpun)
{
if (TryWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize))
{
return true;
}
}
// second: yield
unsigned uLimit {};
do
{
if (!WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress))
{
break;
}
AuThreading::ContextYield();
if (bUWPNanosecondEmulationCheckFirst)
{
if (uLimit++ > uMaxSwitches)
{
break;
}
}
}
while (uAbsTimeSteadyClock > AuTime::SteadyClockNS()); // ...until times up
}
else // high level lock function was called with ms scale resolution
{
// first: wait on the address with an ms scale timeout
(void)pWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, uMS);
// never trust the error value/status provided by wait addresses - instead, do a quick compare
if (!WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress))
{
// best case: we woke up during the ms-res waitonaddress
return true;
}
// attempt to yield again, potentially context switching a few times to hit any NS remainder
AuUInt64 uNow {};
unsigned uLimit {};
while (uAbsTimeSteadyClock > (uNow = AuTime::SteadyClockNS()))
{
uMS = AuNSToMS<AuUInt32>(uAbsTimeSteadyClock - uNow);
if (Primitives::DoTryIfAlderLake([&]()
{
return !WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress);
}, pTargetAddress))
{
// hit it within the span of 1 << SpinLoopPowerA SMT stalls
return true;
}
if (!uMS)
{
// burn off any remainder cycles by switching contexts (this isnt a very long time usually)
if (uLimit++ < uMaxSwitches)
{
AuThreading::ContextYield();
}
else
{
// do not burn the cpu to meet the timeout. we'll just undershoot.
return false;
}
}
else
{
(void)pWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, uMS);
}
}
}
}
return !WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress);
#else
return SysWaitOnAddressTimed(pTargetAddress,
pCompareAddress,
uWordSize,
uAbsTimeSteadyClock,
uRelativeNanoseconds,
uAbsTimeAltClock,
bSpun);
#endif
}
static void RunOSWaitOnAddressNoTimedNoErrors(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state)
{
while (WaitBuffer::Compare2<EWaitMethod::eNotEqual, kPlatformFutexNoForcedAlignedU32>(pTargetAddress, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
if (!SysWaitOnAddressNoTimed(pTargetAddress, pCompareAddress, state.uWordSize))
{
//AuThreading::ContextYield();
}
}
}
static bool RunOSWaitOnAddressTimedSteady(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state,
bool bSpun = false)
{
#if 1
if (!WaitBuffer::Compare2<EWaitMethod::eNotEqual, kPlatformFutexNoForcedAlignedU32>(pTargetAddress, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
return true;
}
(void)RunOSWaitOnAddressTimed(pTargetAddress, pCompareAddress, state.uWordSize, state.qwNanosecondsAbs.value(), { }, { }, bSpun);
return !WaitBuffer::Compare2<EWaitMethod::eNotEqual, kPlatformFutexNoForcedAlignedU32>(pTargetAddress, state.uWordSize, pCompareAddress, state.uDownsizeMask);
#else
return RunOSWaitOnAddressTimed(pTargetAddress, pCompareAddress, state.uWordSize, state.qwNanosecondsAbs.value(), { }, { }, bSpun);
#endif
}
template <EWaitMethod T>
static void RunOSWaitOnAddressEQNoTimedNoErrors(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state)
{
while (true)
{
WaitBufferLegacy wb = WaitBufferLegacy::From(pTargetAddress, state.uWordSize);
if (!WaitBuffer::Compare2<T, kPlatformFutexNoForcedAlignedU32>(wb.buffer, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
return;
}
(void)SysWaitOnAddressNoTimed(pTargetAddress, wb.buffer, state.uWordSize);
if (WaitBuffer::Compare2<T, kPlatformFutexNoForcedAlignedU32>(pTargetAddress, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
SysWakeOneOnAddress(pTargetAddress);
}
else
{
return;
}
}
}
template <EWaitMethod T>
static bool RunOSWaitOnAddressEQTimedSteady(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state,
bool bSpun = false)
{
while (true)
{
WaitBufferLegacy wb = WaitBufferLegacy::From(pTargetAddress, state.uWordSize);
if (!WaitBuffer::Compare2<T, kPlatformFutexNoForcedAlignedU32>(wb.buffer, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
return true;
}
bool bResult = RunOSWaitOnAddressTimed(pTargetAddress, wb.buffer, state.uWordSize, state.qwNanosecondsAbs.value(), { }, { }, bSpun);
if (WaitBuffer::Compare2<T, kPlatformFutexNoForcedAlignedU32>(pTargetAddress, state.uWordSize, pCompareAddress, state.uDownsizeMask))
{
SysWakeOneOnAddress(pTargetAddress);
if (!bResult)
{
return false;
}
}
else
{
return true;
}
}
}
// Windows 8+ thread primitives might use me instead of the public API
// it does work on Linux and Windows 8+
// it does not, however, work on emulated platforms
// this is intentional
bool InternalLTSWaitOnAddressHighRes(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanosecondsAbs)
{
#if defined(WOA_ALWAYS_DUMB_OS_TARGET)
return WaitOnAddressSteady(pTargetAddress, pCompareAddress, uWordSize, qwNanosecondsAbs, true);
#else
auto [pWaitAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, uWordSize);
auto pCompareAddress2 = AuReinterpretCast<const char *>(pCompareAddress) - uDelta;
WaitState state;
state.uDownsizeMask = uMask;
state.uWordSize = uMask != kMax64 ? 4 : uWordSize;
if (!qwNanosecondsAbs)
{
RunOSWaitOnAddressNoTimedNoErrors(pWaitAddress, pCompareAddress2, state);
return true;
}
else
{
state.qwNanosecondsAbs = qwNanosecondsAbs;
return RunOSWaitOnAddressTimedSteady(pWaitAddress, pCompareAddress2, state, true);
}
#endif
}
void InternalLTSWakeAll(const void *pTargetAddress)
{
#if defined(WOA_ALWAYS_DUMB_OS_TARGET)
WakeAllOnAddress(pTargetAddress);
#else
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
SysWakeAllOnAddress(pWakeAddress);
#endif
}
void InternalLTSWakeOne(const void *pTargetAddress)
{
#if defined(WOA_ALWAYS_DUMB_OS_TARGET)
WakeOnAddress(pTargetAddress);
#else
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
if (uDelta)
{
SysWakeAllOnAddress(pWakeAddress);
}
else
{
SysWakeNOnAddress(pWakeAddress, 1);
}
#endif
}
void InternalLTSWakeCount(const void *pTargetAddress, AuUInt32 uCount)
{
#if defined(WOA_ALWAYS_DUMB_OS_TARGET)
WakeNOnAddress(pTargetAddress, uCount);
#else
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
if (uDelta)
{
SysWakeAllOnAddress(pWakeAddress);
}
else
{
SysWakeNOnAddress(pWakeAddress, uCount);
}
#endif
}
WOAFASTPUB bool WaitOnAddress(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
#if !defined(WOA_STRICTER_FIFO)
// Avoid SteadyTime syscall in the event of HAL retardation (missing KUSER QPC, Linux vDSO, etc)
if (!WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress, kMax64))
{
return true;
}
#endif
return WaitOnAddressSteady(pTargetAddress,
pCompareAddress,
uWordSize,
qwNanoseconds ? qwNanoseconds + AuTime::SteadyClockNS() : 0,
optAlreadySpun);
}
WOAFASTPUB bool WaitOnAddressSpecial(EWaitMethod eMethod,
const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
#if !defined(WOA_STRICTER_FIFO)
// Avoid SteadyTime syscall in the event of HAL retardation (missing KUSER QPC, Linux vDSO, etc)
if (!WaitBuffer::Compare(pTargetAddress, uWordSize, pCompareAddress, kMax64, eMethod))
{
return true;
}
#endif
return WaitOnAddressSpecialSteady(eMethod,
pTargetAddress,
pCompareAddress,
uWordSize,
qwNanoseconds ? qwNanoseconds + AuTime::SteadyClockNS() : 0,
optAlreadySpun);
}
template <EWaitMethod T>
auline bool TryWaitOnAddressSpecialTmpl(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize)
{
#if !defined(WOA_STRICTER_FIFO)
return Primitives::DoTryIfAlderLake([&]()
{
return !WaitBuffer::Compare2<T, true>(pTargetAddress, uWordSize, pCompareAddress);
}, pTargetAddress);
#else
return false;
#endif
}
WOAFASTPUB bool TryWaitOnAddress(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize)
{
return TryWaitOnAddressSpecialTmpl<EWaitMethod::eNotEqual>(pTargetAddress, pCompareAddress, uWordSize);
}
WOAFASTPUB bool TryWaitOnAddressSpecial(EWaitMethod eMethod,
const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize)
{
#if !defined(WOA_STRICTER_FIFO)
DO_OF_METHOD_TYPE(return, TryWaitOnAddressSpecialTmpl, pTargetAddress, pCompareAddress, uWordSize);
#endif
return false;
}
WOAFASTPUB bool TryWaitOnAddressEx(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
const AuFunction<bool(const void *, const void *, AuUInt8)> &check)
{
if (!check)
{
return TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize);
}
return Primitives::DoTryIfAlderLake([&]()
{
if (WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress))
{
return false;
}
return check(pTargetAddress, pCompareAddress, uWordSize);
}, pTargetAddress);
}
template <EWaitMethod T>
bool TryWaitOnAddressSpecialExTmpl(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
const AuFunction<bool(const void *, const void *, AuUInt8)> &check)
{
#if !defined(WOA_STRICTER_FIFO)
return Primitives::DoTryIfAlderLake([&]()
{
if (WaitBuffer::Compare2<T, true>(pTargetAddress, uWordSize, pCompareAddress))
{
return false;
}
return check(pTargetAddress, pCompareAddress, uWordSize);
}, pTargetAddress);
#else
return false;
#endif
}
WOAFASTPUB bool TryWaitOnAddressSpecialEx(EWaitMethod eMethod,
const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
const AuFunction<bool(const void *, const void *, AuUInt8)> &check)
{
if (!check)
{
return TryWaitOnAddressSpecial(eMethod, pTargetAddress, pCompareAddress, uWordSize);
}
#if !defined(WOA_STRICTER_FIFO)
DO_OF_METHOD_TYPE(return, TryWaitOnAddressSpecialExTmpl, pTargetAddress, pCompareAddress, uWordSize, check);
#endif
return false;
}
WOAFASTPUB void WakeNOnAddress(const void *pTargetAddress,
AuUInt8 uNMaximumThreads)
{
if (IsWaitOnRecommended())
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
if (uDelta)
{
SysWakeAllOnAddress(pWakeAddress);
}
else
{
SysWakeNOnAddress(pWakeAddress, uNMaximumThreads);
}
}
else
{
(void)gProcessWaitables.IterateWake(pTargetAddress, [&](WaitEntry &entry) -> AuPair<bool, bool>
{
if (!uNMaximumThreads)
{
return AuMakePair(false, false);
}
bool bWake {};
if (entry.TrySignalAddress(pTargetAddress))
{
bWake = true;
uNMaximumThreads--;
}
bool bCont = uNMaximumThreads != 0;
return AuMakePair(bCont, bWake);
});
}
}
WOAFASTPUB void WakeOnAddress(const void *pTargetAddress)
{
WakeNOnAddress(pTargetAddress, 1);
}
WOAFASTPUB void WakeAllOnAddress(const void *pTargetAddress)
{
if (IsWaitOnRecommended())
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
SysWakeAllOnAddress(pWakeAddress);
}
else
{
(void)gProcessWaitables.IterateWake(pTargetAddress, [&](WaitEntry &entry) -> AuPair<bool, bool>
{
return AuMakePair(true, entry.TrySignalAddress(pTargetAddress));
});
}
}
WOAFASTPUB bool WaitOnAddressSteady(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
#if !defined(WOA_STRICTER_FIFO)
// Avoid emulated path dynamic TLS fetch without TLS section
// or various security checks
// or other such bloated thunks
if (!WaitBuffer::Compare2<EWaitMethod::eNotEqual, true>(pTargetAddress, uWordSize, pCompareAddress, kMax64))
{
return true;
}
#endif
bool bWaitOnAddress = IsWaitOnRecommended();
if (bWaitOnAddress)
{
auto [pWaitAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, uWordSize);
auto pCompareAddress2 = AuReinterpretCast<const char *>(pCompareAddress) - uDelta;
WaitState state;
state.uDownsizeMask = uMask;
state.uWordSize = uMask != kMax64 ? 4 : uWordSize;
bool bSpun {};
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpinNative &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
bSpun = true;
}
if (!qwNanoseconds)
{
RunOSWaitOnAddressNoTimedNoErrors(pWaitAddress, pCompareAddress2, state);
return true;
}
else
{
state.qwNanosecondsAbs = qwNanoseconds;
return RunOSWaitOnAddressTimedSteady(pWaitAddress, pCompareAddress2, state, bSpun);
}
}
else
{
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpin &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
}
return WaitOnAddressWide<EWaitMethod::eNotEqual>(pTargetAddress, pCompareAddress, uWordSize, {}, qwNanoseconds ? qwNanoseconds : AuOptional<AuUInt64>{}, false, pCompareAddress);
}
return false;
}
WOAFASTPUB bool WaitOnAddressSpecialSteady(EWaitMethod eMethod,
const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
#if !defined(WOA_STRICTER_FIFO)
// Avoid emulated path dynamic TLS fetch without TLS section
// or various security checks
// or other such bloated thunks
if (!WaitBuffer::Compare(pTargetAddress, uWordSize, pCompareAddress, kMax64, eMethod))
{
return true;
}
#endif
bool bWaitOnAddress = IsWaitOnRecommended();
if (bWaitOnAddress)
{
auto [pWaitAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, uWordSize);
auto pCompareAddress2 = AuReinterpretCast<const char *>(pCompareAddress) - uDelta;
WaitState state;
state.uDownsizeMask = uMask;
state.uWordSize = uMask != kMax64 ? 4 : uWordSize;
state.pCompare2 = pCompareAddress;
state.eWaitMethod = eMethod;
bool bSpun {};
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpinNative &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddressSpecial(eMethod, pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
bSpun = true;
}
if (!qwNanoseconds)
{
DO_OF_METHOD_TYPE(, RunOSWaitOnAddressEQNoTimedNoErrors, pWaitAddress, pCompareAddress2, state);
return true;
}
else
{
state.qwNanosecondsAbs = qwNanoseconds;
DO_OF_METHOD_TYPE(return, RunOSWaitOnAddressEQTimedSteady, pWaitAddress, pCompareAddress2, state, bSpun);
}
}
else
{
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpin &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddressSpecial(eMethod, pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
}
DO_OF_METHOD_TYPE(return, WaitOnAddressWide, pTargetAddress, pCompareAddress, uWordSize, {}, qwNanoseconds ? qwNanoseconds : AuOptional<AuUInt64> {}, false, pCompareAddress);
}
return false;
}
static WaitEntry **GetPNextFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress)
{
auto uCount = pContainer->waitArray.Count<WaitMultipleEntry>();
auto pBase = pContainer->waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_N(i, uCount))
{
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (pBase[i].pTargetAddress == pAddress)
{
return &pCurrent->pNext;
}
}
return nullptr;
}
static WaitEntry **GetPBeforeFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress)
{
auto uCount = pContainer->waitArray.Count<WaitMultipleEntry>();
auto pBase = pContainer->waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_N(i, uCount))
{
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (pBase[i].pTargetAddress == pAddress)
{
return &pCurrent->pBefore;
}
}
return nullptr;
}
static WaitEntry **GetPLastFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress)
{
auto uCount = pContainer->waitArray.Count<WaitMultipleEntry>();
auto pBase = pContainer->waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_FROM_COUNT(i, uCount))
{
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (AddressToIndexOp(pBase[i].pTargetAddress) == AddressToIndexOp(pAddress))
{
return &pCurrent->pNext;
}
}
return nullptr;
}
static WaitEntry **GetPFirstFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress)
{
auto uCount = pContainer->waitArray.Count<WaitMultipleEntry>();
auto pBase = pContainer->waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_N(i, uCount))
{
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (AddressToIndexOp(pBase[i].pTargetAddress) == AddressToIndexOp(pAddress))
{
return &pCurrent->pBefore;
}
}
return nullptr;
}
static const void *GetPCompareFromContainer(const WaitMulipleContainer *pContainer, const void *pAddress)
{
auto uCount = pContainer->waitArray.Count<WaitMultipleEntry>();
auto pBase = pContainer->waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_N(i, uCount))
{
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (pBase[i].pTargetAddress == pAddress)
{
return pBase[i].pCompareAddress;
}
}
return nullptr;
}
static_assert(sizeof(MultipleInternalContext) <= sizeof(WaitMultipleEntry::internalContext));
static bool WaitForMultipleAddressesOrEx(const WaitMulipleContainer &waitMultipleOnAddress, AuUInt32 uMinTrigger)
{
bool bResult {}, bAny {}, bSleepStatus {};
WaitEntry *pWaitEntryMain {}, *pWaitEntryAux {};
SysAssertDbg(!IsWaitOnRecommended(), "WoA not in emulation mode");
auto uCount = waitMultipleOnAddress.waitArray.Count<WaitMultipleEntry>();
auto pBase = waitMultipleOnAddress.waitArray.Begin<WaitMultipleEntry>();
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pTempHoldMe = tlsWaitEntry;
#endif
do
{
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
auto &state = pCurrent->state;
if (current.bIgnoreCurrentFlag)
{
continue;
}
pCurrent->pBefore = nullptr;
pCurrent->pNext = nullptr;
pCurrent->uMinTrigger = uMinTrigger;
// Prevent inc overflows without interfering with and callers, should these word sizes decrease over time
if (!uMinTrigger)
{
pCurrent->uCounter = 0;
}
pWaitEntryAux = gProcessWaitables.WaitBufferFrom2(current.pTargetAddress, current.uSize, current.pCompareAddress, current.eMethod, pCurrent, &waitMultipleOnAddress);
if (!pWaitEntryAux)
{
break;
}
else
{
pWaitEntryMain = pWaitEntryAux;
}
state.qwNanosecondsAbs = waitMultipleOnAddress.qwNanoseconds;
bAny = true;
}
if (!bAny)
{
return true;
}
bSleepStatus = pWaitEntryAux && pWaitEntryAux->SleepLossy(waitMultipleOnAddress.qwNanoseconds);
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
auto &state = pCurrent->state;
if (current.bIgnoreCurrentFlag)
{
continue;
}
if (!WaitBuffer::Compare(current.pTargetAddress, current.uSize, current.pCompareAddress, kMax64, current.eMethod))
{
current.uHasStateChangedCounter++;
bResult = true;
}
gProcessWaitables.RemoveSelf(current.pTargetAddress, pWaitEntryMain);
}
}
while (!bResult && (!waitMultipleOnAddress.qwNanoseconds || bSleepStatus));
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
pTempHoldMe.reset();
#endif
return bResult;
}
AUKN_SYM bool WaitForMultipleAddressesOr(const WaitMulipleContainer &waitMultipleOnAddress)
{
return WaitForMultipleAddressesOrEx(waitMultipleOnAddress, 0);
}
AUKN_SYM bool WaitForMultipleAddressesAnd(const WaitMulipleContainer &waitMultipleOnAddress)
{
auto uCount = waitMultipleOnAddress.waitArray.Count<WaitMultipleEntry>();
auto pBase = waitMultipleOnAddress.waitArray.Begin<WaitMultipleEntry>();
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
pCurrent->bOldIgnore = current.bIgnoreCurrentFlag;
pCurrent->uOldStateChangedCounter = current.uHasStateChangedCounter;
pCurrent->uMinTrigger = uCount;
pCurrent->uCounter = 0;
}
bool bFoundNotTriggered {}, bTimeout {}, bStatus {};
do
{
bool bRet = WaitForMultipleAddressesOrEx(waitMultipleOnAddress, uCount);
bFoundNotTriggered = false;
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
if (current.bIgnoreCurrentFlag)
{
continue;
}
bool bTriggered = current.uHasStateChangedCounter - pCurrent->uOldStateChangedCounter;
if (bTriggered)
{
current.bIgnoreCurrentFlag = true;
}
else
{
bFoundNotTriggered = true;
}
}
if (waitMultipleOnAddress.qwNanoseconds && !bRet)
{
bTimeout = true;
break;
}
}
while (bFoundNotTriggered);
if (bTimeout)
{
bStatus = false;
}
else
{
bStatus = !bFoundNotTriggered;
}
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
current.bIgnoreCurrentFlag = pCurrent->bOldIgnore;
if (current.bIgnoreCurrentFlag)
{
continue;
}
current.uHasStateChangedCounter = pCurrent->uOldStateChangedCounter + (bStatus ? 1 : 0);
}
return bStatus;
}
AUKN_SYM bool WaitForMultipleAddressesOrWithIO(const WaitMulipleContainerWithIO &waitMultipleOnAddress)
{
AuLoop::WaitForMultipleLoopSourcesParameters parameters;
AuList<AuSPtr<AuLoop::ILoopSource>> loopSourceVec;
AuList<AuLoop::WaitForMultipleLoopSourcesInOutOpt> outVec;
bool bResult {}, bAny {}, bSleepStatus {};
AuUInt32 uTicks {};
WaitEntry *pWaitEntryMain {}, *pWaitEntryAux {};
SysAssertDbg(!IsWaitOnRecommended(), "WoA not in emulation mode");
auto uCountOfIO = waitMultipleOnAddress.lsArray.Count<WaitMultipleIOTrigger>();
auto pBaseOfIO = waitMultipleOnAddress.lsArray.Begin<WaitMultipleIOTrigger>();
if (waitMultipleOnAddress.pTriggeredIndex)
{
*waitMultipleOnAddress.pTriggeredIndex = AuUInt32(-1);
}
if (!AuTryResize(loopSourceVec, uCountOfIO + 1))
{
SysPushErrorMemory();
return false;
}
if (!AuTryResize(outVec, uCountOfIO + 1))
{
SysPushErrorMemory();
return false;
}
for (AU_ITERATE_N(i, uCountOfIO))
{
loopSourceVec[i + 1] = pBaseOfIO[i].pLoopSource;
outVec [i + 1].bSkip = pBaseOfIO[i].bIgnoreCurrentFlag;
}
auto uCount = waitMultipleOnAddress.waitArray.Count<WaitMultipleEntry>();
auto pBase = waitMultipleOnAddress.waitArray.Begin<WaitMultipleEntry>();
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pTempHoldMe = tlsWaitEntry;
auto pHead = pTempHoldMe.get();
#else
auto pHead = &tlsWaitEntry;
#endif
if (!pHead->pSemaphore)
{
pHead->pSemaphore = AuLoop::NewLSSemaphoreSlow(0u);
}
pHead->bSemaphoreActive = true;
loopSourceVec[0] = pHead->pSemaphore;
outVec [0].bSkip = false;
parameters.bIsVectorShared = true;
parameters.uFlags = AuLoop::kWaitMultipleFlagAny;
if (waitMultipleOnAddress.bWaitOnlyOne)
{
parameters.uFlags |= AuLoop::kWaitMultipleFlagBreakAfterOne;
}
parameters.vecArray = loopSourceVec;
parameters.vecArray2 = outVec;
do
{
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
auto &state = pCurrent->state;
if (current.bIgnoreCurrentFlag)
{
continue;
}
pCurrent->pBefore = nullptr;
pCurrent->pNext = nullptr;
pCurrent->uMinTrigger = 0;
pCurrent->uCounter = 0;
pWaitEntryAux = gProcessWaitables.WaitBufferFrom2(current.pTargetAddress, current.uSize, current.pCompareAddress, current.eMethod, pCurrent, &waitMultipleOnAddress);
if (!pWaitEntryAux)
{
break;
}
else
{
pWaitEntryMain = pWaitEntryAux;
}
state.qwNanosecondsAbs = waitMultipleOnAddress.qwNanoseconds;
bAny = true;
}
if (!bAny && !uCountOfIO)
{
return true;
}
if (!pWaitEntryAux)
{
bSleepStatus = false;
}
else if (!AuAtomicLoad(&pWaitEntryAux->bAlive))
{
bSleepStatus = false;
}
else
{
auto uNow = AuTime::SteadyClockNS();
bool bTimeOutEarly {};
if (!waitMultipleOnAddress.qwNanoseconds)
{
parameters.optTimeoutMS = {};
}
else if (uNow >= waitMultipleOnAddress.qwNanoseconds)
{
if (uTicks)
{
bTimeOutEarly = true;
}
else
{
parameters.optTimeoutMS = 0;
}
}
else
{
parameters.optTimeoutMS = AuNSToMS<AuUInt32>(waitMultipleOnAddress.qwNanoseconds - uNow);
}
if (bTimeOutEarly)
{
bSleepStatus = false;
}
else
{
bSleepStatus = AuLoop::WaitMultipleLoopSources2(parameters);
}
}
for (AU_ITERATE_N(i, uCount))
{
auto &current = pBase[i];
auto pCurrent = AuReinterpretCast<MultipleInternalContext>(pBase[i].internalContext);
auto &state = pCurrent->state;
if (current.bIgnoreCurrentFlag)
{
continue;
}
if (!WaitBuffer::Compare(current.pTargetAddress, current.uSize, current.pCompareAddress, kMax64, current.eMethod))
{
current.uHasStateChangedCounter++;
bResult = true;
}
gProcessWaitables.RemoveSelf(current.pTargetAddress, pWaitEntryMain);
}
for (AU_ITERATE_N(i, uCountOfIO))
{
auto &refTriggered = outVec[i + 1].uTriggered;
if (refTriggered)
{
pBaseOfIO[i].uHasStateChangedCounter += refTriggered;
refTriggered = 0;
bResult = true;
if (waitMultipleOnAddress.pTriggeredIndex)
{
*waitMultipleOnAddress.pTriggeredIndex = i;
}
}
}
uTicks++;
}
while (!bResult && (!waitMultipleOnAddress.qwNanoseconds || bSleepStatus));
pHead->bSemaphoreActive = false;
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
pTempHoldMe.reset();
#endif
return bResult;
}
}
Reference in New Issue View Git Blame Copy Permalink