Clang Format.

GPU: Add Fast GPU Time Option.
Maxwell3D: Process Macros on MultiMethod.
2020-04-23 08:52:58 -04:00 · 2020-04-23 08:52:57 -04:00 · 2020-04-23 08:52:56 -04:00 · 2020-04-23 08:52:55 -04:00 · 2020-04-23 02:41:38 -04:00 · 2020-04-23 02:18:52 -04:00
103 changed files with 2889 additions and 871 deletions
@@ -910,14 +910,14 @@ typedef void* (*MicroProfileThreadFunc)(void*);

 #ifndef _WIN32
 typedef pthread_t MicroProfileThread;
-void MicroProfileThreadStart(MicroProfileThread* pThread, MicroProfileThreadFunc Func)
+inline void MicroProfileThreadStart(MicroProfileThread* pThread, MicroProfileThreadFunc Func)
 {
    pthread_attr_t Attr;
    int r  = pthread_attr_init(&Attr);
    MP_ASSERT(r == 0);
    pthread_create(pThread, &Attr, Func, 0);
 }
-void MicroProfileThreadJoin(MicroProfileThread* pThread)
+inline void MicroProfileThreadJoin(MicroProfileThread* pThread)
 {
    int r = pthread_join(*pThread, 0);
    MP_ASSERT(r == 0);
@@ -930,11 +930,11 @@ DWORD _stdcall ThreadTrampoline(void* pFunc)
    return (uint32_t)F(0);
 }

-void MicroProfileThreadStart(MicroProfileThread* pThread, MicroProfileThreadFunc Func)
+inline void MicroProfileThreadStart(MicroProfileThread* pThread, MicroProfileThreadFunc Func)
 {
    *pThread = CreateThread(0, 0, ThreadTrampoline, Func, 0, 0);
 }
-void MicroProfileThreadJoin(MicroProfileThread* pThread)
+inline void MicroProfileThreadJoin(MicroProfileThread* pThread)
 {
    WaitForSingleObject(*pThread, INFINITE);
    CloseHandle(*pThread);
@@ -1131,7 +1131,7 @@ inline void MicroProfileSetThreadLog(MicroProfileThreadLog* pLog)
    pthread_setspecific(g_MicroProfileThreadLogKey, pLog);
 }
 #else
-MicroProfileThreadLog* MicroProfileGetThreadLog()
+inline MicroProfileThreadLog* MicroProfileGetThreadLog()
 {
    return g_MicroProfileThreadLog;
 }
@@ -1247,7 +1247,7 @@ MicroProfileToken MicroProfileFindToken(const char* pGroup, const char* pName)
    return MICROPROFILE_INVALID_TOKEN;
 }

-uint16_t MicroProfileGetGroup(const char* pGroup, MicroProfileTokenType Type)
+inline uint16_t MicroProfileGetGroup(const char* pGroup, MicroProfileTokenType Type)
 {
    for(uint32_t i = 0; i < S.nGroupCount; ++i)
    {
@@ -1276,7 +1276,7 @@ uint16_t MicroProfileGetGroup(const char* pGroup, MicroProfileTokenType Type)
    return nGroupIndex;
 }

-void MicroProfileRegisterGroup(const char* pGroup, const char* pCategory, uint32_t nColor)
+inline void MicroProfileRegisterGroup(const char* pGroup, const char* pCategory, uint32_t nColor)
 {
    int nCategoryIndex = -1;
    for(uint32_t i = 0; i < S.nCategoryCount; ++i)
@@ -1442,7 +1442,7 @@ void MicroProfileGpuLeave(MicroProfileToken nToken_, uint64_t nTickStart)
    }
 }

-void MicroProfileContextSwitchPut(MicroProfileContextSwitch* pContextSwitch)
+inline void MicroProfileContextSwitchPut(MicroProfileContextSwitch* pContextSwitch)
 {
    if(S.nRunning || pContextSwitch->nTicks <= S.nPauseTicks)
    {
@@ -1894,7 +1894,7 @@ void MicroProfileSetEnableAllGroups(bool bEnableAllGroups)
    S.nAllGroupsWanted = bEnableAllGroups ? 1 : 0;
 }

-void MicroProfileEnableCategory(const char* pCategory, bool bEnabled)
+inline void MicroProfileEnableCategory(const char* pCategory, bool bEnabled)
 {
    int nCategoryIndex = -1;
    for(uint32_t i = 0; i < S.nCategoryCount; ++i)
@@ -2004,7 +2004,7 @@ void MicroProfileForceDisableGroup(const char* pGroup, MicroProfileTokenType Typ
 }


-void MicroProfileCalcAllTimers(float* pTimers, float* pAverage, float* pMax, float* pCallAverage, float* pExclusive, float* pAverageExclusive, float* pMaxExclusive, float* pTotal, uint32_t nSize)
+inline void MicroProfileCalcAllTimers(float* pTimers, float* pAverage, float* pMax, float* pCallAverage, float* pExclusive, float* pAverageExclusive, float* pMaxExclusive, float* pTotal, uint32_t nSize)
 {
    for(uint32_t i = 0; i < S.nTotalTimers && i < nSize; ++i)
    {
@@ -417,19 +417,19 @@ void MicroProfileToggleDisplayMode()
 }


-void MicroProfileStringArrayClear(MicroProfileStringArray* pArray)
+inline void MicroProfileStringArrayClear(MicroProfileStringArray* pArray)
 {
    pArray->nNumStrings = 0;
    pArray->pBufferPos = &pArray->Buffer[0];
 }

-void MicroProfileStringArrayAddLiteral(MicroProfileStringArray* pArray, const char* pLiteral)
+inline void MicroProfileStringArrayAddLiteral(MicroProfileStringArray* pArray, const char* pLiteral)
 {
    MP_ASSERT(pArray->nNumStrings < MICROPROFILE_TOOLTIP_MAX_STRINGS);
    pArray->ppStrings[pArray->nNumStrings++] = pLiteral;
 }

-void MicroProfileStringArrayFormat(MicroProfileStringArray* pArray, const char* fmt, ...)
+inline void MicroProfileStringArrayFormat(MicroProfileStringArray* pArray, const char* fmt, ...)
 {
    MP_ASSERT(pArray->nNumStrings < MICROPROFILE_TOOLTIP_MAX_STRINGS);
    pArray->ppStrings[pArray->nNumStrings++] = pArray->pBufferPos;
@@ -439,7 +439,7 @@ void MicroProfileStringArrayFormat(MicroProfileStringArray* pArray, const char*
    va_end(args);
    MP_ASSERT(pArray->pBufferPos < pArray->Buffer + MICROPROFILE_TOOLTIP_STRING_BUFFER_SIZE);
 }
-void MicroProfileStringArrayCopy(MicroProfileStringArray* pDest, MicroProfileStringArray* pSrc)
+inline void MicroProfileStringArrayCopy(MicroProfileStringArray* pDest, MicroProfileStringArray* pSrc)
 {
    memcpy(&pDest->ppStrings[0], &pSrc->ppStrings[0], sizeof(pDest->ppStrings));
    memcpy(&pDest->Buffer[0], &pSrc->Buffer[0], sizeof(pDest->Buffer));
@@ -456,7 +456,7 @@ void MicroProfileStringArrayCopy(MicroProfileStringArray* pDest, MicroProfileStr
    pDest->nNumStrings = pSrc->nNumStrings;
 }

-void MicroProfileFloatWindowSize(const char** ppStrings, uint32_t nNumStrings, uint32_t* pColors, uint32_t& nWidth, uint32_t& nHeight, uint32_t* pStringLengths = 0)
+inline void MicroProfileFloatWindowSize(const char** ppStrings, uint32_t nNumStrings, uint32_t* pColors, uint32_t& nWidth, uint32_t& nHeight, uint32_t* pStringLengths = 0)
 {
    uint32_t* nStringLengths = pStringLengths ? pStringLengths : (uint32_t*)alloca(nNumStrings * sizeof(uint32_t));
    uint32_t nTextCount = nNumStrings/2;
@@ -474,7 +474,7 @@ void MicroProfileFloatWindowSize(const char** ppStrings, uint32_t nNumStrings, u
    nHeight = (MICROPROFILE_TEXT_HEIGHT+1) * nTextCount + 2 * MICROPROFILE_BORDER_SIZE;
 }

-void MicroProfileDrawFloatWindow(uint32_t nX, uint32_t nY, const char** ppStrings, uint32_t nNumStrings, uint32_t nColor, uint32_t* pColors = 0)
+inline void MicroProfileDrawFloatWindow(uint32_t nX, uint32_t nY, const char** ppStrings, uint32_t nNumStrings, uint32_t nColor, uint32_t* pColors = 0)
 {
    uint32_t nWidth = 0, nHeight = 0;
    uint32_t* nStringLengths = (uint32_t*)alloca(nNumStrings * sizeof(uint32_t));
@@ -503,7 +503,7 @@ void MicroProfileDrawFloatWindow(uint32_t nX, uint32_t nY, const char** ppString
        nY += (MICROPROFILE_TEXT_HEIGHT+1);
    }
 }
-void MicroProfileDrawTextBox(uint32_t nX, uint32_t nY, const char** ppStrings, uint32_t nNumStrings, uint32_t nColor, uint32_t* pColors = 0)
+inline void MicroProfileDrawTextBox(uint32_t nX, uint32_t nY, const char** ppStrings, uint32_t nNumStrings, uint32_t nColor, uint32_t* pColors = 0)
 {
    uint32_t nWidth = 0, nHeight = 0;
    uint32_t* nStringLengths = (uint32_t*)alloca(nNumStrings * sizeof(uint32_t));
@@ -529,7 +529,7 @@ void MicroProfileDrawTextBox(uint32_t nX, uint32_t nY, const char** ppStrings, u



-void MicroProfileToolTipMeta(MicroProfileStringArray* pToolTip)
+inline void MicroProfileToolTipMeta(MicroProfileStringArray* pToolTip)
 {
    MicroProfile& S = *MicroProfileGet();
    if(UI.nRangeBeginIndex != UI.nRangeEndIndex && UI.pRangeLog)
@@ -608,7 +608,7 @@ void MicroProfileToolTipMeta(MicroProfileStringArray* pToolTip)
    }
 }

-void MicroProfileDrawFloatTooltip(uint32_t nX, uint32_t nY, uint32_t nToken, uint64_t nTime)
+inline void MicroProfileDrawFloatTooltip(uint32_t nX, uint32_t nY, uint32_t nToken, uint64_t nTime)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -718,7 +718,7 @@ void MicroProfileDrawFloatTooltip(uint32_t nX, uint32_t nY, uint32_t nToken, uin
 }


-void MicroProfileZoomTo(int64_t nTickStart, int64_t nTickEnd)
+inline void MicroProfileZoomTo(int64_t nTickStart, int64_t nTickEnd)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -728,7 +728,7 @@ void MicroProfileZoomTo(int64_t nTickStart, int64_t nTickEnd)
    UI.fDetailedRangeTarget = MicroProfileLogTickDifference(nTickStart, nTickEnd) * fToMs;
 }

-void MicroProfileCenter(int64_t nTickCenter)
+inline void MicroProfileCenter(int64_t nTickCenter)
 {
    MicroProfile& S = *MicroProfileGet();
    int64_t nStart = S.Frames[S.nFrameCurrent].nFrameStartCpu;
@@ -739,7 +739,7 @@ void MicroProfileCenter(int64_t nTickCenter)
 #ifdef MICROPROFILE_DEBUG
 uint64_t* g_pMicroProfileDumpStart = 0;
 uint64_t* g_pMicroProfileDumpEnd = 0;
-void MicroProfileDebugDumpRange()
+inline void MicroProfileDebugDumpRange()
 {
    MicroProfile& S = *MicroProfileGet();
    if(g_pMicroProfileDumpStart != g_pMicroProfileDumpEnd)
@@ -777,7 +777,7 @@ void MicroProfileDebugDumpRange()

 #define MICROPROFILE_HOVER_DIST 0.5f

-void MicroProfileDrawDetailedContextSwitchBars(uint32_t nY, uint32_t nThreadId, uint32_t nContextSwitchStart, uint32_t nContextSwitchEnd, int64_t nBaseTicks, uint32_t nBaseY)
+inline void MicroProfileDrawDetailedContextSwitchBars(uint32_t nY, uint32_t nThreadId, uint32_t nContextSwitchStart, uint32_t nContextSwitchEnd, int64_t nBaseTicks, uint32_t nBaseY)
 {
    MicroProfile& S = *MicroProfileGet();
    int64_t nTickIn = -1;
@@ -841,7 +841,7 @@ void MicroProfileDrawDetailedContextSwitchBars(uint32_t nY, uint32_t nThreadId,
    }
 }

-void MicroProfileDrawDetailedBars(uint32_t nWidth, uint32_t nHeight, int nBaseY, int nSelectedFrame)
+inline void MicroProfileDrawDetailedBars(uint32_t nWidth, uint32_t nHeight, int nBaseY, int nSelectedFrame)
 {
    MicroProfile& S = *MicroProfileGet();
    MP_DEBUG_DUMP_RANGE();
@@ -1325,7 +1325,7 @@ void MicroProfileDrawDetailedBars(uint32_t nWidth, uint32_t nHeight, int nBaseY,
 }


-void MicroProfileDrawDetailedFrameHistory(uint32_t nWidth, uint32_t nHeight, uint32_t nBaseY, uint32_t nSelectedFrame)
+inline void MicroProfileDrawDetailedFrameHistory(uint32_t nWidth, uint32_t nHeight, uint32_t nBaseY, uint32_t nSelectedFrame)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -1379,7 +1379,7 @@ void MicroProfileDrawDetailedFrameHistory(uint32_t nWidth, uint32_t nHeight, uin
    }
    MicroProfileDrawBox(fSelectionStart, nBaseY, fSelectionEnd, nBaseY+MICROPROFILE_FRAME_HISTORY_HEIGHT, MICROPROFILE_FRAME_HISTORY_COLOR_HIGHTLIGHT, MicroProfileBoxTypeFlat);
 }
-void MicroProfileDrawDetailedView(uint32_t nWidth, uint32_t nHeight)
+inline void MicroProfileDrawDetailedView(uint32_t nWidth, uint32_t nHeight)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -1416,11 +1416,11 @@ void MicroProfileDrawDetailedView(uint32_t nWidth, uint32_t nHeight)
    MicroProfileDrawDetailedFrameHistory(nWidth, nHeight, nBaseY, nSelectedFrame);
 }

-void MicroProfileDrawTextRight(uint32_t nX, uint32_t nY, uint32_t nColor, const char* pStr, uint32_t nStrLen)
+inline void MicroProfileDrawTextRight(uint32_t nX, uint32_t nY, uint32_t nColor, const char* pStr, uint32_t nStrLen)
 {
    MicroProfileDrawText(nX - nStrLen * (MICROPROFILE_TEXT_WIDTH+1), nY, nColor, pStr, nStrLen);
 }
-void MicroProfileDrawHeader(int32_t nX, uint32_t nWidth, const char* pName)
+inline void MicroProfileDrawHeader(int32_t nX, uint32_t nWidth, const char* pName)
 {
    if(pName)
    {
@@ -1432,7 +1432,7 @@ void MicroProfileDrawHeader(int32_t nX, uint32_t nWidth, const char* pName)

 typedef void (*MicroProfileLoopGroupCallback)(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pData);

-void MicroProfileLoopActiveGroupsDraw(int32_t nX, int32_t nY, const char* pName, MicroProfileLoopGroupCallback CB, void* pData)
+inline void MicroProfileLoopActiveGroupsDraw(int32_t nX, int32_t nY, const char* pName, MicroProfileLoopGroupCallback CB, void* pData)
 {
    MicroProfile& S = *MicroProfileGet();
    nY += MICROPROFILE_TEXT_HEIGHT + 2;
@@ -1465,7 +1465,7 @@ void MicroProfileLoopActiveGroupsDraw(int32_t nX, int32_t nY, const char* pName,
 }


-void MicroProfileCalcTimers(float* pTimers, float* pAverage, float* pMax, float* pCallAverage, float* pExclusive, float* pAverageExclusive, float* pMaxExclusive, uint64_t nGroup, uint32_t nSize)
+inline void MicroProfileCalcTimers(float* pTimers, float* pAverage, float* pMax, float* pCallAverage, float* pExclusive, float* pAverageExclusive, float* pMaxExclusive, uint64_t nGroup, uint32_t nSize)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -1527,7 +1527,7 @@ void MicroProfileCalcTimers(float* pTimers, float* pAverage, float* pMax, float*

 #define SBUF_MAX 32

-void MicroProfileDrawBarArrayCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
+inline void MicroProfileDrawBarArrayCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
 {
    const uint32_t nHeight = MICROPROFILE_TEXT_HEIGHT;
    const uint32_t nTextWidth = 6 * (1+MICROPROFILE_TEXT_WIDTH);
@@ -1547,7 +1547,7 @@ void MicroProfileDrawBarArrayCallback(uint32_t nTimer, uint32_t nIdx, uint64_t n
 }


-uint32_t MicroProfileDrawBarArray(int32_t nX, int32_t nY, float* pTimers, const char* pName, uint32_t nTotalHeight, float* pTimers2 = NULL)
+inline uint32_t MicroProfileDrawBarArray(int32_t nX, int32_t nY, float* pTimers, const char* pName, uint32_t nTotalHeight, float* pTimers2 = NULL)
 {
    const uint32_t nTextWidth = 6 * (1+MICROPROFILE_TEXT_WIDTH);
    const uint32_t nWidth = MICROPROFILE_BAR_WIDTH;
@@ -1559,7 +1559,7 @@ uint32_t MicroProfileDrawBarArray(int32_t nX, int32_t nY, float* pTimers, const
    return nWidth + 5 + nTextWidth;

 }
-void MicroProfileDrawBarCallCountCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
+inline void MicroProfileDrawBarCallCountCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
 {
    MicroProfile& S = *MicroProfileGet();
    char sBuffer[SBUF_MAX];
@@ -1567,7 +1567,7 @@ void MicroProfileDrawBarCallCountCallback(uint32_t nTimer, uint32_t nIdx, uint64
    MicroProfileDrawText(nX, nY, (uint32_t)-1, sBuffer, nLen);
 }

-uint32_t MicroProfileDrawBarCallCount(int32_t nX, int32_t nY, const char* pName)
+inline uint32_t MicroProfileDrawBarCallCount(int32_t nX, int32_t nY, const char* pName)
 {
    MicroProfileLoopActiveGroupsDraw(nX, nY, pName, MicroProfileDrawBarCallCountCallback, 0);
    const uint32_t nTextWidth = 6 * MICROPROFILE_TEXT_WIDTH;
@@ -1581,7 +1581,7 @@ struct MicroProfileMetaAverageArgs
    float fRcpFrames;
 };

-void MicroProfileDrawBarMetaAverageCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
+inline void MicroProfileDrawBarMetaAverageCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
 {
    MicroProfileMetaAverageArgs* pArgs = (MicroProfileMetaAverageArgs*)pExtra;
    uint64_t* pCounters = pArgs->pCounters;
@@ -1591,7 +1591,7 @@ void MicroProfileDrawBarMetaAverageCallback(uint32_t nTimer, uint32_t nIdx, uint
    MicroProfileDrawText(nX - nLen * (MICROPROFILE_TEXT_WIDTH+1), nY, (uint32_t)-1, sBuffer, nLen);
 }

-uint32_t MicroProfileDrawBarMetaAverage(int32_t nX, int32_t nY, uint64_t* pCounters, const char* pName, uint32_t nTotalHeight)
+inline uint32_t MicroProfileDrawBarMetaAverage(int32_t nX, int32_t nY, uint64_t* pCounters, const char* pName, uint32_t nTotalHeight)
 {
    if(!pName)
        return 0;
@@ -1605,7 +1605,7 @@ uint32_t MicroProfileDrawBarMetaAverage(int32_t nX, int32_t nY, uint64_t* pCount
 }


-void MicroProfileDrawBarMetaCountCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
+inline void MicroProfileDrawBarMetaCountCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
 {
    uint64_t* pCounters = (uint64_t*)pExtra;
    char sBuffer[SBUF_MAX];
@@ -1613,7 +1613,7 @@ void MicroProfileDrawBarMetaCountCallback(uint32_t nTimer, uint32_t nIdx, uint64
    MicroProfileDrawText(nX - nLen * (MICROPROFILE_TEXT_WIDTH+1), nY, (uint32_t)-1, sBuffer, nLen);
 }

-uint32_t MicroProfileDrawBarMetaCount(int32_t nX, int32_t nY, uint64_t* pCounters, const char* pName, uint32_t nTotalHeight)
+inline uint32_t MicroProfileDrawBarMetaCount(int32_t nX, int32_t nY, uint64_t* pCounters, const char* pName, uint32_t nTotalHeight)
 {
    if(!pName)
        return 0;
@@ -1625,7 +1625,7 @@ uint32_t MicroProfileDrawBarMetaCount(int32_t nX, int32_t nY, uint64_t* pCounter
    return 5 + nTextWidth;
 }

-void MicroProfileDrawBarLegendCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
+inline void MicroProfileDrawBarLegendCallback(uint32_t nTimer, uint32_t nIdx, uint64_t nGroupMask, uint32_t nX, uint32_t nY, void* pExtra)
 {
    MicroProfile& S = *MicroProfileGet();
    if (S.TimerInfo[nTimer].bGraph)
@@ -1640,7 +1640,7 @@ void MicroProfileDrawBarLegendCallback(uint32_t nTimer, uint32_t nIdx, uint64_t
    }
 }

-uint32_t MicroProfileDrawBarLegend(int32_t nX, int32_t nY, uint32_t nTotalHeight, uint32_t nMaxWidth)
+inline uint32_t MicroProfileDrawBarLegend(int32_t nX, int32_t nY, uint32_t nTotalHeight, uint32_t nMaxWidth)
 {
    MicroProfileDrawLineVertical(nX-5, nY, nTotalHeight, UI.nOpacityBackground | g_nMicroProfileBackColors[0]|g_nMicroProfileBackColors[1]);
    MicroProfileLoopActiveGroupsDraw(nMaxWidth, nY, 0, MicroProfileDrawBarLegendCallback, 0);
@@ -1807,7 +1807,7 @@ void MicroProfileDumpTimers()
    }
 }

-void MicroProfileDrawBarView(uint32_t nScreenWidth, uint32_t nScreenHeight)
+inline void MicroProfileDrawBarView(uint32_t nScreenWidth, uint32_t nScreenHeight)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -1951,7 +1951,7 @@ typedef const char* (*MicroProfileSubmenuCallback)(int, bool* bSelected);
 typedef void (*MicroProfileClickCallback)(int);


-const char* MicroProfileUIMenuMode(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuMode(int nIndex, bool* bSelected)
 {
    MicroProfile& S = *MicroProfileGet();
    switch(nIndex)
@@ -1979,7 +1979,7 @@ const char* MicroProfileUIMenuMode(int nIndex, bool* bSelected)
    }
 }

-const char* MicroProfileUIMenuGroups(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuGroups(int nIndex, bool* bSelected)
 {
    MicroProfile& S = *MicroProfileGet();
    *bSelected = false;
@@ -2012,7 +2012,7 @@ const char* MicroProfileUIMenuGroups(int nIndex, bool* bSelected)
    }
 }

-const char* MicroProfileUIMenuAggregate(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuAggregate(int nIndex, bool* bSelected)
 {
    MicroProfile& S = *MicroProfileGet();
    if(nIndex < sizeof(g_MicroProfileAggregatePresets)/sizeof(g_MicroProfileAggregatePresets[0]))
@@ -2032,7 +2032,7 @@ const char* MicroProfileUIMenuAggregate(int nIndex, bool* bSelected)

 }

-const char* MicroProfileUIMenuTimers(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuTimers(int nIndex, bool* bSelected)
 {
    MicroProfile& S = *MicroProfileGet();
    *bSelected = 0 != (S.nBars & (1 << nIndex));
@@ -2054,7 +2054,7 @@ const char* MicroProfileUIMenuTimers(int nIndex, bool* bSelected)
    return 0;
 }

-const char* MicroProfileUIMenuOptions(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuOptions(int nIndex, bool* bSelected)
 {
    MicroProfile& S = *MicroProfileGet();
    if(nIndex >= MICROPROFILE_OPTION_SIZE) return 0;
@@ -2094,7 +2094,7 @@ const char* MicroProfileUIMenuOptions(int nIndex, bool* bSelected)
    return UI.Options[nIndex].Text;
 }

-const char* MicroProfileUIMenuPreset(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuPreset(int nIndex, bool* bSelected)
 {
    static char buf[128];
    *bSelected = false;
@@ -2118,7 +2118,7 @@ const char* MicroProfileUIMenuPreset(int nIndex, bool* bSelected)
    }
 }

-const char* MicroProfileUIMenuCustom(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuCustom(int nIndex, bool* bSelected)
 {
    if((uint32_t)-1 == UI.nCustomActive)
    {
@@ -2145,13 +2145,13 @@ const char* MicroProfileUIMenuCustom(int nIndex, bool* bSelected)
    }
 }

-const char* MicroProfileUIMenuEmpty(int nIndex, bool* bSelected)
+inline const char* MicroProfileUIMenuEmpty(int nIndex, bool* bSelected)
 {
    return 0;
 }


-void MicroProfileUIClickMode(int nIndex)
+inline void MicroProfileUIClickMode(int nIndex)
 {
    MicroProfile& S = *MicroProfileGet();
    switch(nIndex)
@@ -2176,7 +2176,7 @@ void MicroProfileUIClickMode(int nIndex)
    }
 }

-void MicroProfileUIClickGroups(int nIndex)
+inline void MicroProfileUIClickGroups(int nIndex)
 {
    MicroProfile& S = *MicroProfileGet();
    if(nIndex == 0)
@@ -2208,7 +2208,7 @@ void MicroProfileUIClickGroups(int nIndex)
    }
 }

-void MicroProfileUIClickAggregate(int nIndex)
+inline void MicroProfileUIClickAggregate(int nIndex)
 {
    MicroProfile& S = *MicroProfileGet();
    S.nAggregateFlip = g_MicroProfileAggregatePresets[nIndex];
@@ -2218,13 +2218,13 @@ void MicroProfileUIClickAggregate(int nIndex)
    }
 }

-void MicroProfileUIClickTimers(int nIndex)
+inline void MicroProfileUIClickTimers(int nIndex)
 {
    MicroProfile& S = *MicroProfileGet();
    S.nBars ^= (1 << nIndex);
 }

-void MicroProfileUIClickOptions(int nIndex)
+inline void MicroProfileUIClickOptions(int nIndex)
 {
    MicroProfile& S = *MicroProfileGet();
    switch(UI.Options[nIndex].nSubType)
@@ -2271,7 +2271,7 @@ void MicroProfileUIClickOptions(int nIndex)
    }
 }

-void MicroProfileUIClickPreset(int nIndex)
+inline void MicroProfileUIClickPreset(int nIndex)
 {
    int nNumPresets = sizeof(g_MicroProfilePresetNames) / sizeof(g_MicroProfilePresetNames[0]);
    int nIndexSave = nIndex - nNumPresets - 1;
@@ -2285,7 +2285,7 @@ void MicroProfileUIClickPreset(int nIndex)
    }
 }

-void MicroProfileUIClickCustom(int nIndex)
+inline void MicroProfileUIClickCustom(int nIndex)
 {
    if(nIndex == 0)
    {
@@ -2298,13 +2298,13 @@ void MicroProfileUIClickCustom(int nIndex)

 }

-void MicroProfileUIClickEmpty(int nIndex)
+inline void MicroProfileUIClickEmpty(int nIndex)
 {

 }


-void MicroProfileDrawMenu(uint32_t nWidth, uint32_t nHeight)
+inline void MicroProfileDrawMenu(uint32_t nWidth, uint32_t nHeight)
 {
    MicroProfile& S = *MicroProfileGet();

@@ -2489,7 +2489,7 @@ void MicroProfileDrawMenu(uint32_t nWidth, uint32_t nHeight)
 }


-void MicroProfileMoveGraph()
+inline void MicroProfileMoveGraph()
 {

    int nZoom = UI.nMouseWheelDelta;
@@ -2536,7 +2536,7 @@ void MicroProfileMoveGraph()
        UI.nOffsetY = 0;
 }

-void MicroProfileDrawCustom(uint32_t nWidth, uint32_t nHeight)
+inline void MicroProfileDrawCustom(uint32_t nWidth, uint32_t nHeight)
 {
    if((uint32_t)-1 != UI.nCustomActive)
    {
@@ -2633,7 +2633,7 @@ void MicroProfileDrawCustom(uint32_t nWidth, uint32_t nHeight)
        }
    }
 }
-void MicroProfileDraw(uint32_t nWidth, uint32_t nHeight)
+inline void MicroProfileDraw(uint32_t nWidth, uint32_t nHeight)
 {
    MICROPROFILE_SCOPE(g_MicroProfileDraw);
    MicroProfile& S = *MicroProfileGet();
@@ -3226,7 +3226,7 @@ void MicroProfileLoadPreset(const char* pSuffix)
    }
 }

-uint32_t MicroProfileCustomGroupFind(const char* pCustomName)
+inline uint32_t MicroProfileCustomGroupFind(const char* pCustomName)
 {
    for(uint32_t i = 0; i < UI.nCustomCount; ++i)
    {
@@ -3238,7 +3238,7 @@ uint32_t MicroProfileCustomGroupFind(const char* pCustomName)
    return (uint32_t)-1;
 }

-uint32_t MicroProfileCustomGroup(const char* pCustomName)
+inline uint32_t MicroProfileCustomGroup(const char* pCustomName)
 {
    for(uint32_t i = 0; i < UI.nCustomCount; ++i)
    {
@@ -3271,7 +3271,7 @@ void MicroProfileCustomGroup(const char* pCustomName, uint32_t nMaxTimers, uint3
    UI.Custom[nIndex].nAggregateFlip = nAggregateFlip;
 }

-void MicroProfileCustomGroupEnable(uint32_t nIndex)
+inline void MicroProfileCustomGroupEnable(uint32_t nIndex)
 {
    if(nIndex < UI.nCustomCount)
    {
@@ -203,7 +203,11 @@ endif()
 target_compile_definitions(opus PRIVATE OPUS_BUILD ENABLE_HARDENING)

 if(NOT MSVC)
-    target_compile_definitions(opus PRIVATE _FORTIFY_SOURCE=2)
+    if(MINGW)
+        target_compile_definitions(opus PRIVATE _FORTIFY_SOURCE=0)
+    else()
+        target_compile_definitions(opus PRIVATE _FORTIFY_SOURCE=2)
+    endif()
 endif()

 # It is strongly recommended to uncomment one of these VAR_ARRAYS: Use C99
@@ -54,8 +54,10 @@ else()
    add_compile_options(
        -Wall
        -Werror=implicit-fallthrough
+        -Werror=missing-declarations
        -Werror=reorder
        -Wextra
+        -Wmissing-declarations
        -Wno-attributes
        -Wno-unused-parameter
    )
@@ -7,9 +7,12 @@ add_library(audio_core STATIC
    audio_out.h
    audio_renderer.cpp
    audio_renderer.h
+    behavior_info.cpp
+    behavior_info.h
    buffer.h
    codec.cpp
    codec.h
+    common.h
    null_sink.h
    sink.h
    sink_details.cpp
@@ -6,6 +6,7 @@
 #include "audio_core/audio_out.h"
 #include "audio_core/audio_renderer.h"
 #include "audio_core/codec.h"
+#include "audio_core/common.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
@@ -79,7 +80,7 @@ AudioRenderer::AudioRenderer(Core::Timing::CoreTiming& core_timing, Core::Memory
                             std::size_t instance_number)
    : worker_params{params}, buffer_event{buffer_event}, voices(params.voice_count),
      effects(params.effect_count), memory{memory_} {
-
+    behavior_info.SetUserRevision(params.revision);
    audio_out = std::make_unique<AudioCore::AudioOut>();
    stream = audio_out->OpenStream(core_timing, STREAM_SAMPLE_RATE, STREAM_NUM_CHANNELS,
                                   fmt::format("AudioRenderer-Instance{}", instance_number),
@@ -109,17 +110,17 @@ Stream::State AudioRenderer::GetStreamState() const {
    return stream->GetState();
 }

-static constexpr u32 VersionFromRevision(u32_le rev) {
-    // "REV7" -> 7
-    return ((rev >> 24) & 0xff) - 0x30;
-}
-
-std::vector<u8> AudioRenderer::UpdateAudioRenderer(const std::vector<u8>& input_params) {
+ResultVal<std::vector<u8>> AudioRenderer::UpdateAudioRenderer(const std::vector<u8>& input_params) {
    // Copy UpdateDataHeader struct
    UpdateDataHeader config{};
    std::memcpy(&config, input_params.data(), sizeof(UpdateDataHeader));
    u32 memory_pool_count = worker_params.effect_count + (worker_params.voice_count * 4);

+    if (!behavior_info.UpdateInput(input_params, sizeof(UpdateDataHeader))) {
+        LOG_ERROR(Audio, "Failed to update behavior info input parameters");
+        return Audren::ERR_INVALID_PARAMETERS;
+    }
+
    // Copy MemoryPoolInfo structs
    std::vector<MemoryPoolInfo> mem_pool_info(memory_pool_count);
    std::memcpy(mem_pool_info.data(),
@@ -173,8 +174,7 @@ std::vector<u8> AudioRenderer::UpdateAudioRenderer(const std::vector<u8>& input_
    // Copy output header
    UpdateDataHeader response_data{worker_params};
    std::vector<u8> output_params(response_data.total_size);
-    const auto audren_revision = VersionFromRevision(config.revision);
-    if (audren_revision >= 5) {
+    if (behavior_info.IsElapsedFrameCountSupported()) {
        response_data.frame_count = 0x10;
        response_data.total_size += 0x10;
    }
@@ -200,7 +200,19 @@ std::vector<u8> AudioRenderer::UpdateAudioRenderer(const std::vector<u8>& input_
                    sizeof(EffectOutStatus));
        effect_out_status_offset += sizeof(EffectOutStatus);
    }
-    return output_params;
+
+    // Update behavior info output
+    const std::size_t behavior_out_status_offset{
+        sizeof(UpdateDataHeader) + response_data.memory_pools_size + response_data.voices_size +
+        response_data.effects_size + response_data.sinks_size +
+        response_data.performance_manager_size};
+
+    if (!behavior_info.UpdateOutput(output_params, behavior_out_status_offset)) {
+        LOG_ERROR(Audio, "Failed to update behavior info output parameters");
+        return Audren::ERR_INVALID_PARAMETERS;
+    }
+
+    return MakeResult(output_params);
 }

 void AudioRenderer::VoiceState::SetWaveIndex(std::size_t index) {
@@ -8,11 +8,13 @@
 #include <memory>
 #include <vector>

+#include "audio_core/behavior_info.h"
 #include "audio_core/stream.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/kernel/object.h"
+#include "core/hle/result.h"

 namespace Core::Timing {
 class CoreTiming;
@@ -226,7 +228,7 @@ public:
                  std::shared_ptr<Kernel::WritableEvent> buffer_event, std::size_t instance_number);
    ~AudioRenderer();

-    std::vector<u8> UpdateAudioRenderer(const std::vector<u8>& input_params);
+    ResultVal<std::vector<u8>> UpdateAudioRenderer(const std::vector<u8>& input_params);
    void QueueMixedBuffer(Buffer::Tag tag);
    void ReleaseAndQueueBuffers();
    u32 GetSampleRate() const;
@@ -237,6 +239,7 @@ public:
 private:
    class EffectState;
    class VoiceState;
+    BehaviorInfo behavior_info{};

    AudioRendererParameter worker_params;
    std::shared_ptr<Kernel::WritableEvent> buffer_event;
@@ -0,0 +1,100 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include "audio_core/behavior_info.h"
+#include "audio_core/common.h"
+#include "common/logging/log.h"
+
+namespace AudioCore {
+
+BehaviorInfo::BehaviorInfo() : process_revision(CURRENT_PROCESS_REVISION) {}
+BehaviorInfo::~BehaviorInfo() = default;
+
+bool BehaviorInfo::UpdateInput(const std::vector<u8>& buffer, std::size_t offset) {
+    if (!CanConsumeBuffer(buffer.size(), offset, sizeof(InParams))) {
+        LOG_ERROR(Audio, "Buffer is an invalid size!");
+        return false;
+    }
+    InParams params{};
+    std::memcpy(&params, buffer.data() + offset, sizeof(InParams));
+
+    if (!IsValidRevision(params.revision)) {
+        LOG_ERROR(Audio, "Invalid input revision, revision=0x{:08X}", params.revision);
+        return false;
+    }
+
+    if (user_revision != params.revision) {
+        LOG_ERROR(Audio,
+                  "User revision differs from input revision, expecting 0x{:08X} but got 0x{:08X}",
+                  user_revision, params.revision);
+        return false;
+    }
+
+    ClearError();
+    UpdateFlags(params.flags);
+
+    // TODO(ogniK): Check input params size when InfoUpdater is used
+
+    return true;
+}
+
+bool BehaviorInfo::UpdateOutput(std::vector<u8>& buffer, std::size_t offset) {
+    if (!CanConsumeBuffer(buffer.size(), offset, sizeof(OutParams))) {
+        LOG_ERROR(Audio, "Buffer is an invalid size!");
+        return false;
+    }
+
+    OutParams params{};
+    std::memcpy(params.errors.data(), errors.data(), sizeof(ErrorInfo) * errors.size());
+    params.error_count = static_cast<u32_le>(error_count);
+    std::memcpy(buffer.data() + offset, &params, sizeof(OutParams));
+    return true;
+}
+
+void BehaviorInfo::ClearError() {
+    error_count = 0;
+}
+
+void BehaviorInfo::UpdateFlags(u64_le dest_flags) {
+    flags = dest_flags;
+}
+
+void BehaviorInfo::SetUserRevision(u32_le revision) {
+    user_revision = revision;
+}
+
+bool BehaviorInfo::IsAdpcmLoopContextBugFixed() const {
+    return IsRevisionSupported(2, user_revision);
+}
+
+bool BehaviorInfo::IsSplitterSupported() const {
+    return IsRevisionSupported(2, user_revision);
+}
+
+bool BehaviorInfo::IsLongSizePreDelaySupported() const {
+    return IsRevisionSupported(3, user_revision);
+}
+
+bool BehaviorInfo::IsAudioRenererProcessingTimeLimit80PercentSupported() const {
+    return IsRevisionSupported(5, user_revision);
+}
+
+bool BehaviorInfo::IsAudioRenererProcessingTimeLimit75PercentSupported() const {
+    return IsRevisionSupported(4, user_revision);
+}
+
+bool BehaviorInfo::IsAudioRenererProcessingTimeLimit70PercentSupported() const {
+    return IsRevisionSupported(1, user_revision);
+}
+
+bool BehaviorInfo::IsElapsedFrameCountSupported() const {
+    return IsRevisionSupported(5, user_revision);
+}
+
+bool BehaviorInfo::IsMemoryPoolForceMappingEnabled() const {
+    return (flags & 1) != 0;
+}
+
+} // namespace AudioCore
@@ -0,0 +1,66 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+
+#include <vector>
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "common/swap.h"
+
+namespace AudioCore {
+class BehaviorInfo {
+public:
+    explicit BehaviorInfo();
+    ~BehaviorInfo();
+
+    bool UpdateInput(const std::vector<u8>& buffer, std::size_t offset);
+    bool UpdateOutput(std::vector<u8>& buffer, std::size_t offset);
+
+    void ClearError();
+    void UpdateFlags(u64_le dest_flags);
+    void SetUserRevision(u32_le revision);
+
+    bool IsAdpcmLoopContextBugFixed() const;
+    bool IsSplitterSupported() const;
+    bool IsLongSizePreDelaySupported() const;
+    bool IsAudioRenererProcessingTimeLimit80PercentSupported() const;
+    bool IsAudioRenererProcessingTimeLimit75PercentSupported() const;
+    bool IsAudioRenererProcessingTimeLimit70PercentSupported() const;
+    bool IsElapsedFrameCountSupported() const;
+    bool IsMemoryPoolForceMappingEnabled() const;
+
+private:
+    u32_le process_revision{};
+    u32_le user_revision{};
+    u64_le flags{};
+
+    struct ErrorInfo {
+        u32_le result{};
+        INSERT_PADDING_WORDS(1);
+        u64_le result_info{};
+    };
+    static_assert(sizeof(ErrorInfo) == 0x10, "ErrorInfo is an invalid size");
+
+    std::array<ErrorInfo, 10> errors{};
+    std::size_t error_count{};
+
+    struct InParams {
+        u32_le revision{};
+        u32_le padding{};
+        u64_le flags{};
+    };
+    static_assert(sizeof(InParams) == 0x10, "InParams is an invalid size");
+
+    struct OutParams {
+        std::array<ErrorInfo, 10> errors{};
+        u32_le error_count{};
+        INSERT_PADDING_BYTES(12);
+    };
+    static_assert(sizeof(OutParams) == 0xb0, "OutParams is an invalid size");
+};
+
+} // namespace AudioCore
@@ -0,0 +1,47 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "common/swap.h"
+#include "core/hle/result.h"
+
+namespace AudioCore {
+namespace Audren {
+constexpr ResultCode ERR_INVALID_PARAMETERS{ErrorModule::Audio, 41};
+}
+
+constexpr u32_le CURRENT_PROCESS_REVISION = Common::MakeMagic('R', 'E', 'V', '8');
+
+static constexpr u32 VersionFromRevision(u32_le rev) {
+    // "REV7" -> 7
+    return ((rev >> 24) & 0xff) - 0x30;
+}
+
+static constexpr bool IsRevisionSupported(u32 required, u32_le user_revision) {
+    const auto base = VersionFromRevision(user_revision);
+    return required <= base;
+}
+
+static constexpr bool IsValidRevision(u32_le revision) {
+    const auto base = VersionFromRevision(revision);
+    constexpr auto max_rev = VersionFromRevision(CURRENT_PROCESS_REVISION);
+    return base <= max_rev;
+}
+
+static constexpr bool CanConsumeBuffer(std::size_t size, std::size_t offset, std::size_t required) {
+    if (offset > size) {
+        return false;
+    }
+    if (size < required) {
+        return false;
+    }
+    if ((size - offset) < required) {
+        return false;
+    }
+    return true;
+}
+
+} // namespace AudioCore
@@ -20,6 +20,7 @@
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/svc.h"
 #include "core/memory.h"
+#include "core/settings.h"

 namespace Core {

@@ -144,6 +145,8 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
    config.page_table_address_space_bits = address_space_bits;
    config.silently_mirror_page_table = false;
    config.absolute_offset_page_table = true;
+    config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128;
+    config.only_detect_misalignment_via_page_table_on_page_boundary = true;

    // Multi-process state
    config.processor_id = core_index;
@@ -159,8 +162,11 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
    // Unpredictable instructions
    config.define_unpredictable_behaviour = true;

-    config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128;
-    config.only_detect_misalignment_via_page_table_on_page_boundary = true;
+    // Optimizations
+    if (Settings::values.disable_cpu_opt) {
+        config.enable_optimizations = false;
+        config.enable_fast_dispatch = false;
+    }

    return std::make_shared<Dynarmic::A64::Jit>(config);
 }
@@ -11,6 +11,7 @@
 #include "core/core_timing.h"
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/svc.h"
+#include "core/memory.h"

 namespace Core {

@@ -171,7 +172,17 @@ MICROPROFILE_DEFINE(ARM_Jit_Unicorn, "ARM JIT", "Unicorn", MP_RGB(255, 64, 64));

 void ARM_Unicorn::ExecuteInstructions(std::size_t num_instructions) {
    MICROPROFILE_SCOPE(ARM_Jit_Unicorn);
+
+    // Temporarily map the code page for Unicorn
+    u64 map_addr{GetPC() & ~Memory::PAGE_MASK};
+    std::vector<u8> page_buffer(Memory::PAGE_SIZE);
+    system.Memory().ReadBlock(map_addr, page_buffer.data(), page_buffer.size());
+
+    CHECKED(uc_mem_map_ptr(uc, map_addr, page_buffer.size(),
+                           UC_PROT_READ | UC_PROT_WRITE | UC_PROT_EXEC, page_buffer.data()));
    CHECKED(uc_emu_start(uc, GetPC(), 1ULL << 63, 0, num_instructions));
+    CHECKED(uc_mem_unmap(uc, map_addr, page_buffer.size()));
+
    system.CoreTiming().AddTicks(num_instructions);
    if (GDBStub::IsServerEnabled()) {
        if (last_bkpt_hit && last_bkpt.type == GDBStub::BreakpointType::Execute) {
@@ -1202,7 +1202,8 @@ const boost::container::flat_map<std::string, KeyIndex<S128KeyType>> KeyManager:
     {S128KeyType::Source, static_cast<u64>(SourceKeyType::KeyAreaKey),
      static_cast<u64>(KeyAreaKeyType::System)}},
    {"titlekek_source", {S128KeyType::Source, static_cast<u64>(SourceKeyType::Titlekek), 0}},
-    {"keyblob_mac_key_source", {S128KeyType::Source, static_cast<u64>(SourceKeyType::KeyblobMAC)}},
+    {"keyblob_mac_key_source",
+     {S128KeyType::Source, static_cast<u64>(SourceKeyType::KeyblobMAC), 0}},
    {"tsec_key", {S128KeyType::TSEC, 0, 0}},
    {"secure_boot_key", {S128KeyType::SecureBoot, 0, 0}},
    {"sd_seed", {S128KeyType::SDSeed, 0, 0}},
@@ -202,8 +202,8 @@ static std::array<Key128, 0x20> FindEncryptedMasterKeyFromHex(const std::vector<
    return out;
 }

-FileSys::VirtualFile FindFileInDirWithNames(const FileSys::VirtualDir& dir,
-                                            const std::string& name) {
+static FileSys::VirtualFile FindFileInDirWithNames(const FileSys::VirtualDir& dir,
+                                                   const std::string& name) {
    const auto upper = Common::ToUpper(name);

    for (const auto& fname : {name, name + ".bin", upper, upper + ".BIN"}) {
@@ -345,8 +345,7 @@ FileSys::VirtualFile PartitionDataManager::GetPackage2Raw(Package2Type type) con
    return package2.at(static_cast<size_t>(type));
 }

-bool AttemptDecrypt(const std::array<u8, 16>& key, Package2Header& header) {
-
+static bool AttemptDecrypt(const std::array<u8, 16>& key, Package2Header& header) {
    const std::vector<u8> iv(header.header_ctr.begin(), header.header_ctr.end());
    Package2Header temp = header;
    AESCipher<Key128> cipher(key, Mode::CTR);
@@ -51,6 +51,17 @@ Loader::ResultStatus ProgramMetadata::Load(VirtualFile file) {
    return Loader::ResultStatus::Success;
 }

+/*static*/ ProgramMetadata ProgramMetadata::GetDefault() {
+    ProgramMetadata result;
+
+    result.LoadManual(
+        true /*is_64_bit*/, FileSys::ProgramAddressSpaceType::Is39Bit /*address_space*/,
+        0x2c /*main_thread_prio*/, 0 /*main_thread_core*/, 0x00100000 /*main_thread_stack_size*/,
+        {}, 0xFFFFFFFFFFFFFFFF /*filesystem_permissions*/, {} /*capabilities*/);
+
+    return result;
+}
+
 void ProgramMetadata::LoadManual(bool is_64_bit, ProgramAddressSpaceType address_space,
                                 s32 main_thread_prio, u32 main_thread_core,
                                 u32 main_thread_stack_size, u64 title_id,
@@ -44,9 +44,13 @@ public:
    ProgramMetadata();
    ~ProgramMetadata();

+    /// Gets a default ProgramMetadata configuration, should only be used for homebrew formats where
+    /// we do not have an NPDM file
+    static ProgramMetadata GetDefault();
+
    Loader::ResultStatus Load(VirtualFile file);

-    // Load from parameters instead of NPDM file, used for KIP
+    /// Load from parameters instead of NPDM file, used for KIP
    void LoadManual(bool is_64_bit, ProgramAddressSpaceType address_space, s32 main_thread_prio,
                    u32 main_thread_core, u32 main_thread_stack_size, u64 title_id,
                    u64 filesystem_permissions, KernelCapabilityDescriptors capabilities);
@@ -1389,10 +1389,9 @@ void SendTrap(Kernel::Thread* thread, int trap) {
        return;
    }

-    if (!halt_loop || current_thread == thread) {
-        current_thread = thread;
-        SendSignal(thread, trap);
-    }
+    current_thread = thread;
+    SendSignal(thread, trap);
+
    halt_loop = true;
    send_trap = false;
 }
@@ -150,8 +150,7 @@ static void ResetThreadContext64(Core::ARM_Interface::ThreadContext64& context,
    context.pc = entry_point;
    context.sp = stack_top;
    // TODO(merry): Perform a hardware test to determine the below value.
-    // AHP = 0, DN = 1, FTZ = 1, RMode = Round towards zero
-    context.fpcr = 0x03C00000;
+    context.fpcr = 0;
 }

 ResultVal<std::shared_ptr<Thread>> Thread::Create(KernelCore& kernel, std::string name,
@@ -903,7 +903,7 @@ private:
    void PopOutData(Kernel::HLERequestContext& ctx) {
        LOG_DEBUG(Service_AM, "called");

-        const auto storage = applet->GetBroker().PopNormalDataToGame();
+        auto storage = applet->GetBroker().PopNormalDataToGame();
        if (storage == nullptr) {
            LOG_ERROR(Service_AM,
                      "storage is a nullptr. There is no data in the current normal channel");
@@ -934,7 +934,7 @@ private:
    void PopInteractiveOutData(Kernel::HLERequestContext& ctx) {
        LOG_DEBUG(Service_AM, "called");

-        const auto storage = applet->GetBroker().PopInteractiveDataToGame();
+        auto storage = applet->GetBroker().PopInteractiveDataToGame();
        if (storage == nullptr) {
            LOG_ERROR(Service_AM,
                      "storage is a nullptr. There is no data in the current interactive channel");
@@ -92,11 +92,16 @@ private:
    }

    void RequestUpdateImpl(Kernel::HLERequestContext& ctx) {
-        LOG_WARNING(Service_Audio, "(STUBBED) called");
+        LOG_DEBUG(Service_Audio, "(STUBBED) called");
+
+        auto result = renderer->UpdateAudioRenderer(ctx.ReadBuffer());
+
+        if (result.Succeeded()) {
+            ctx.WriteBuffer(result.Unwrap());
+        }

-        ctx.WriteBuffer(renderer->UpdateAudioRenderer(ctx.ReadBuffer()));
        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(RESULT_SUCCESS);
+        rb.Push(result.Code());
    }

    void Start(Kernel::HLERequestContext& ctx) {
@@ -252,8 +257,6 @@ private:
    }

    void GetAudioDeviceOutputVolume(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-
        const auto device_name_buffer = ctx.ReadBuffer();
        const std::string name = Common::StringFromBuffer(device_name_buffer);

@@ -18,6 +18,7 @@
 #include "core/hle/service/bcat/backend/boxcat.h"
 #include "core/settings.h"

+namespace Service::BCAT {
 namespace {

 // Prevents conflicts with windows macro called CreateFile
@@ -30,10 +31,6 @@ bool VfsDeleteFileWrap(FileSys::VirtualDir dir, std::string_view name) {
    return dir->DeleteFile(name);
 }

-} // Anonymous namespace
-
-namespace Service::BCAT {
-
 constexpr ResultCode ERROR_GENERAL_BCAT_FAILURE{ErrorModule::BCAT, 1};

 constexpr char BOXCAT_HOSTNAME[] = "api.yuzu-emu.org";
@@ -90,8 +87,6 @@ constexpr u32 PORT = 443;
 constexpr u32 TIMEOUT_SECONDS = 30;
 [[maybe_unused]] constexpr u64 VFS_COPY_BLOCK_SIZE = 1ULL << 24; // 4MB

-namespace {
-
 std::string GetBINFilePath(u64 title_id) {
    return fmt::format("{}bcat/{:016X}/launchparam.bin",
                       FileUtil::GetUserPath(FileUtil::UserPath::CacheDir), title_id);
@@ -4,6 +4,7 @@

 #include "core/crypto/key_manager.h"
 #include "core/hle/ipc_helpers.h"
+#include "core/hle/service/es/es.h"
 #include "core/hle/service/service.h"

 namespace Service::ES {
@@ -76,7 +77,6 @@ private:
    }

    void ImportTicket(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
        const auto ticket = ctx.ReadBuffer();
        const auto cert = ctx.ReadBuffer(1);

@@ -502,7 +502,7 @@ void Controller_NPad::SetNpadMode(u32 npad_id, NPadAssignments assignment_mode)

 void Controller_NPad::VibrateController(const std::vector<u32>& controller_ids,
                                        const std::vector<Vibration>& vibrations) {
-    LOG_WARNING(Service_HID, "(STUBBED) called");
+    LOG_DEBUG(Service_HID, "(STUBBED) called");

    if (!can_controllers_vibrate) {
        return;
@@ -20,8 +20,8 @@ namespace Service::Time {

 class ISystemClock final : public ServiceFramework<ISystemClock> {
 public:
-    ISystemClock(Clock::SystemClockCore& clock_core)
-        : ServiceFramework("ISystemClock"), clock_core{clock_core} {
+    explicit ISystemClock(Clock::SystemClockCore& clock_core, Core::System& system)
+        : ServiceFramework("ISystemClock"), clock_core{clock_core}, system{system} {
        // clang-format off
        static const FunctionInfo functions[] = {
            {0, &ISystemClock::GetCurrentTime, "GetCurrentTime"},
@@ -46,9 +46,8 @@ private:
        }

        s64 posix_time{};
-        if (const ResultCode result{
-                clock_core.GetCurrentTime(Core::System::GetInstance(), posix_time)};
-            result != RESULT_SUCCESS) {
+        if (const ResultCode result{clock_core.GetCurrentTime(system, posix_time)};
+            result.IsError()) {
            IPC::ResponseBuilder rb{ctx, 2};
            rb.Push(result);
            return;
@@ -69,9 +68,8 @@ private:
        }

        Clock::SystemClockContext system_clock_context{};
-        if (const ResultCode result{
-                clock_core.GetClockContext(Core::System::GetInstance(), system_clock_context)};
-            result != RESULT_SUCCESS) {
+        if (const ResultCode result{clock_core.GetClockContext(system, system_clock_context)};
+            result.IsError()) {
            IPC::ResponseBuilder rb{ctx, 2};
            rb.Push(result);
            return;
@@ -83,12 +81,13 @@ private:
    }

    Clock::SystemClockCore& clock_core;
+    Core::System& system;
 };

 class ISteadyClock final : public ServiceFramework<ISteadyClock> {
 public:
-    ISteadyClock(Clock::SteadyClockCore& clock_core)
-        : ServiceFramework("ISteadyClock"), clock_core{clock_core} {
+    explicit ISteadyClock(Clock::SteadyClockCore& clock_core, Core::System& system)
+        : ServiceFramework("ISteadyClock"), clock_core{clock_core}, system{system} {
        static const FunctionInfo functions[] = {
            {0, &ISteadyClock::GetCurrentTimePoint, "GetCurrentTimePoint"},
        };
@@ -105,14 +104,14 @@ private:
            return;
        }

-        const Clock::SteadyClockTimePoint time_point{
-            clock_core.GetCurrentTimePoint(Core::System::GetInstance())};
+        const Clock::SteadyClockTimePoint time_point{clock_core.GetCurrentTimePoint(system)};
        IPC::ResponseBuilder rb{ctx, (sizeof(Clock::SteadyClockTimePoint) / 4) + 2};
        rb.Push(RESULT_SUCCESS);
        rb.PushRaw(time_point);
    }

    Clock::SteadyClockCore& clock_core;
+    Core::System& system;
 };

 ResultCode Module::Interface::GetClockSnapshotFromSystemClockContextInternal(
@@ -134,7 +133,7 @@ ResultCode Module::Interface::GetClockSnapshotFromSystemClockContextInternal(
    }

    const auto current_time_point{
-        time_manager.GetStandardSteadyClockCore().GetCurrentTimePoint(Core::System::GetInstance())};
+        time_manager.GetStandardSteadyClockCore().GetCurrentTimePoint(system)};
    if (const ResultCode result{Clock::ClockSnapshot::GetCurrentTime(
            clock_snapshot.user_time, current_time_point, clock_snapshot.user_context)};
        result != RESULT_SUCCESS) {
@@ -176,21 +175,24 @@ void Module::Interface::GetStandardUserSystemClock(Kernel::HLERequestContext& ct
    LOG_DEBUG(Service_Time, "called");
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardUserSystemClockCore());
+    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardUserSystemClockCore(),
+                                      system);
 }

 void Module::Interface::GetStandardNetworkSystemClock(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_Time, "called");
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardNetworkSystemClockCore());
+    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardNetworkSystemClockCore(),
+                                      system);
 }

 void Module::Interface::GetStandardSteadyClock(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_Time, "called");
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<ISteadyClock>(module->GetTimeManager().GetStandardSteadyClockCore());
+    rb.PushIpcInterface<ISteadyClock>(module->GetTimeManager().GetStandardSteadyClockCore(),
+                                      system);
 }

 void Module::Interface::GetTimeZoneService(Kernel::HLERequestContext& ctx) {
@@ -204,7 +206,8 @@ void Module::Interface::GetStandardLocalSystemClock(Kernel::HLERequestContext& c
    LOG_DEBUG(Service_Time, "called");
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardLocalSystemClockCore());
+    rb.PushIpcInterface<ISystemClock>(module->GetTimeManager().GetStandardLocalSystemClockCore(),
+                                      system);
 }

 void Module::Interface::IsStandardNetworkSystemClockAccuracySufficient(
@@ -228,8 +231,7 @@ void Module::Interface::CalculateMonotonicSystemClockBaseTimePoint(Kernel::HLERe

    IPC::RequestParser rp{ctx};
    const auto context{rp.PopRaw<Clock::SystemClockContext>()};
-    const auto current_time_point{
-        steady_clock_core.GetCurrentTimePoint(Core::System::GetInstance())};
+    const auto current_time_point{steady_clock_core.GetCurrentTimePoint(system)};

    if (current_time_point.clock_source_id == context.steady_time_point.clock_source_id) {
        const auto ticks{Clock::TimeSpanType::FromTicks(
@@ -255,8 +257,8 @@ void Module::Interface::GetClockSnapshot(Kernel::HLERequestContext& ctx) {
    Clock::SystemClockContext user_context{};
    if (const ResultCode result{
            module->GetTimeManager().GetStandardUserSystemClockCore().GetClockContext(
-                Core::System::GetInstance(), user_context)};
-        result != RESULT_SUCCESS) {
+                system, user_context)};
+        result.IsError()) {
        IPC::ResponseBuilder rb{ctx, 2};
        rb.Push(result);
        return;
@@ -264,8 +266,8 @@ void Module::Interface::GetClockSnapshot(Kernel::HLERequestContext& ctx) {
    Clock::SystemClockContext network_context{};
    if (const ResultCode result{
            module->GetTimeManager().GetStandardNetworkSystemClockCore().GetClockContext(
-                Core::System::GetInstance(), network_context)};
-        result != RESULT_SUCCESS) {
+                system, network_context)};
+        result.IsError()) {
        IPC::ResponseBuilder rb{ctx, 2};
        rb.Push(result);
        return;
@@ -274,7 +276,7 @@ void Module::Interface::GetClockSnapshot(Kernel::HLERequestContext& ctx) {
    Clock::ClockSnapshot clock_snapshot{};
    if (const ResultCode result{GetClockSnapshotFromSystemClockContextInternal(
            &ctx.GetThread(), user_context, network_context, type, clock_snapshot)};
-        result != RESULT_SUCCESS) {
+        result.IsError()) {
        IPC::ResponseBuilder rb{ctx, 2};
        rb.Push(result);
        return;
@@ -518,8 +518,8 @@ static bool ParseTimeZoneBinary(TimeZoneRule& time_zone_rule, FileSys::VirtualFi
    constexpr s32 time_zone_max_leaps{50};
    constexpr s32 time_zone_max_chars{50};
    if (!(0 <= header.leap_count && header.leap_count < time_zone_max_leaps &&
-          0 < header.type_count && header.type_count < time_zone_rule.ttis.size() &&
-          0 <= header.time_count && header.time_count < time_zone_rule.ats.size() &&
+          0 < header.type_count && header.type_count < s32(time_zone_rule.ttis.size()) &&
+          0 <= header.time_count && header.time_count < s32(time_zone_rule.ats.size()) &&
          0 <= header.char_count && header.char_count < time_zone_max_chars &&
          (header.ttis_std_count == header.type_count || header.ttis_std_count == 0) &&
          (header.ttis_gmt_count == header.type_count || header.ttis_gmt_count == 0))) {
@@ -398,6 +398,11 @@ AppLoader_ELF::LoadResult AppLoader_ELF::Load(Kernel::Process& process) {
    Kernel::CodeSet codeset = elf_reader.LoadInto(base_address);
    const VAddr entry_point = codeset.entrypoint;

+    // Setup the process code layout
+    if (process.LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), buffer.size()).IsError()) {
+        return {ResultStatus::ErrorNotInitialized, {}};
+    }
+
    process.LoadModule(std::move(codeset), entry_point);

    is_loaded = true;
@@ -131,7 +131,7 @@ static constexpr u32 PageAlignSize(u32 size) {
 }

 static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
-                        const std::string& name, VAddr load_base) {
+                        const std::string& name) {
    if (data.size() < sizeof(NroHeader)) {
        return {};
    }
@@ -187,19 +187,25 @@ static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
    codeset.DataSegment().size += bss_size;
    program_image.resize(static_cast<u32>(program_image.size()) + bss_size);

+    // Setup the process code layout
+    if (process.LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), program_image.size())
+            .IsError()) {
+        return false;
+    }
+
    // Load codeset for current process
    codeset.memory = std::move(program_image);
-    process.LoadModule(std::move(codeset), load_base);
+    process.LoadModule(std::move(codeset), process.PageTable().GetCodeRegionStart());

    // Register module with GDBStub
-    GDBStub::RegisterModule(name, load_base, load_base);
+    GDBStub::RegisterModule(name, process.PageTable().GetCodeRegionStart(),
+                            process.PageTable().GetCodeRegionEnd());

    return true;
 }

-bool AppLoader_NRO::LoadNro(Kernel::Process& process, const FileSys::VfsFile& file,
-                            VAddr load_base) {
-    return LoadNroImpl(process, file.ReadAllBytes(), file.GetName(), load_base);
+bool AppLoader_NRO::LoadNro(Kernel::Process& process, const FileSys::VfsFile& file) {
+    return LoadNroImpl(process, file.ReadAllBytes(), file.GetName());
 }

 AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::Process& process) {
@@ -207,10 +213,7 @@ AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::Process& process) {
        return {ResultStatus::ErrorAlreadyLoaded, {}};
    }

-    // Load NRO
-    const VAddr base_address = process.PageTable().GetCodeRegionStart();
-
-    if (!LoadNro(process, *file, base_address)) {
+    if (!LoadNro(process, *file)) {
        return {ResultStatus::ErrorLoadingNRO, {}};
    }

@@ -47,7 +47,7 @@ public:
    bool IsRomFSUpdatable() const override;

 private:
-    bool LoadNro(Kernel::Process& process, const FileSys::VfsFile& file, VAddr load_base);
+    bool LoadNro(Kernel::Process& process, const FileSys::VfsFile& file);

    std::vector<u8> icon_data;
    std::unique_ptr<FileSys::NACP> nacp;
@@ -92,7 +92,7 @@ void LogSettings() {
    LogSetting("Renderer_UseFrameLimit", Settings::values.use_frame_limit);
    LogSetting("Renderer_FrameLimit", Settings::values.frame_limit);
    LogSetting("Renderer_UseDiskShaderCache", Settings::values.use_disk_shader_cache);
-    LogSetting("Renderer_UseAccurateGpuEmulation", Settings::values.use_accurate_gpu_emulation);
+    LogSetting("Renderer_GPUAccuracyLevel", Settings::values.gpu_accuracy);
    LogSetting("Renderer_UseAsynchronousGpuEmulation",
               Settings::values.use_asynchronous_gpu_emulation);
    LogSetting("Renderer_UseVsync", Settings::values.use_vsync);
@@ -109,4 +109,12 @@ void LogSettings() {
    LogSetting("Services_BCATBoxcatLocal", Settings::values.bcat_boxcat_local);
 }

+bool IsGPULevelExtreme() {
+    return values.gpu_accuracy == GPUAccuracy::Extreme;
+}
+
+bool IsGPULevelHigh() {
+    return values.gpu_accuracy == GPUAccuracy::Extreme || values.gpu_accuracy == GPUAccuracy::High;
+}
+
 } // namespace Settings
@@ -376,6 +376,12 @@ enum class RendererBackend {
    Vulkan = 1,
 };

+enum class GPUAccuracy : u32 {
+    Normal = 0,
+    High = 1,
+    Extreme = 2,
+};
+
 struct Values {
    // System
    bool use_docked_mode;
@@ -436,10 +442,11 @@ struct Values {
    bool use_frame_limit;
    u16 frame_limit;
    bool use_disk_shader_cache;
-    bool use_accurate_gpu_emulation;
+    GPUAccuracy gpu_accuracy;
    bool use_asynchronous_gpu_emulation;
    bool use_vsync;
    bool force_30fps_mode;
+    bool use_fast_gpu_time;

    float bg_red;
    float bg_green;
@@ -464,6 +471,7 @@ struct Values {
    bool dump_nso;
    bool reporting_services;
    bool quest_flag;
+    bool disable_cpu_opt;

    // BCAT
    std::string bcat_backend;
@@ -479,6 +487,9 @@ struct Values {
    std::map<u64, std::vector<std::string>> disabled_addons;
 } extern values;

+bool IsGPULevelExtreme();
+bool IsGPULevelHigh();
+
 void Apply();
 void LogSettings();
 } // namespace Settings
@@ -56,6 +56,18 @@ static const char* TranslateRenderer(Settings::RendererBackend backend) {
    return "Unknown";
 }

+static const char* TranslateGPUAccuracyLevel(Settings::GPUAccuracy backend) {
+    switch (backend) {
+    case Settings::GPUAccuracy::Normal:
+        return "Normal";
+    case Settings::GPUAccuracy::High:
+        return "High";
+    case Settings::GPUAccuracy::Extreme:
+        return "Extreme";
+    }
+    return "Unknown";
+}
+
 u64 GetTelemetryId() {
    u64 telemetry_id{};
    const std::string filename{FileUtil::GetUserPath(FileUtil::UserPath::ConfigDir) +
@@ -184,8 +196,8 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader) {
    AddField(field_type, "Renderer_UseFrameLimit", Settings::values.use_frame_limit);
    AddField(field_type, "Renderer_FrameLimit", Settings::values.frame_limit);
    AddField(field_type, "Renderer_UseDiskShaderCache", Settings::values.use_disk_shader_cache);
-    AddField(field_type, "Renderer_UseAccurateGpuEmulation",
-             Settings::values.use_accurate_gpu_emulation);
+    AddField(field_type, "Renderer_GPUAccuracyLevel",
+             TranslateGPUAccuracyLevel(Settings::values.gpu_accuracy));
    AddField(field_type, "Renderer_UseAsynchronousGpuEmulation",
             Settings::values.use_asynchronous_gpu_emulation);
    AddField(field_type, "Renderer_UseVsync", Settings::values.use_vsync);
@@ -14,13 +14,14 @@
 #include "core/core.h"
 #include "core/core_timing.h"

+namespace {
 // Numbers are chosen randomly to make sure the correct one is given.
-static constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
-static constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals
+constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
+constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals

-static std::bitset<CB_IDS.size()> callbacks_ran_flags;
-static u64 expected_callback = 0;
-static s64 lateness = 0;
+std::bitset<CB_IDS.size()> callbacks_ran_flags;
+u64 expected_callback = 0;
+s64 lateness = 0;

 template <unsigned int IDX>
 void CallbackTemplate(u64 userdata, s64 cycles_late) {
@@ -31,7 +32,7 @@ void CallbackTemplate(u64 userdata, s64 cycles_late) {
    REQUIRE(lateness == cycles_late);
 }

-static u64 callbacks_done = 0;
+u64 callbacks_done = 0;

 void EmptyCallback(u64 userdata, s64 cycles_late) {
    ++callbacks_done;
@@ -48,8 +49,8 @@ struct ScopeInit final {
    Core::Timing::CoreTiming core_timing;
 };

-static void AdvanceAndCheck(Core::Timing::CoreTiming& core_timing, u32 idx, u32 context = 0,
-                            int expected_lateness = 0, int cpu_downcount = 0) {
+void AdvanceAndCheck(Core::Timing::CoreTiming& core_timing, u32 idx, u32 context = 0,
+                     int expected_lateness = 0, int cpu_downcount = 0) {
    callbacks_ran_flags = 0;
    expected_callback = CB_IDS[idx];
    lateness = expected_lateness;
@@ -62,6 +63,7 @@ static void AdvanceAndCheck(Core::Timing::CoreTiming& core_timing, u32 idx, u32

    REQUIRE(decltype(callbacks_ran_flags)().set(idx) == callbacks_ran_flags);
 }
+} // Anonymous namespace

 TEST_CASE("CoreTiming[BasicOrder]", "[core]") {
    ScopeInit guard;
@@ -23,6 +23,7 @@ add_library(video_core STATIC
    engines/shader_bytecode.h
    engines/shader_header.h
    engines/shader_type.h
+    fence_manager.h
    gpu.cpp
    gpu.h
    gpu_asynch.cpp
@@ -51,6 +52,8 @@ add_library(video_core STATIC
    renderer_opengl/gl_buffer_cache.h
    renderer_opengl/gl_device.cpp
    renderer_opengl/gl_device.h
+    renderer_opengl/gl_fence_manager.cpp
+    renderer_opengl/gl_fence_manager.h
    renderer_opengl/gl_framebuffer_cache.cpp
    renderer_opengl/gl_framebuffer_cache.h
    renderer_opengl/gl_rasterizer.cpp
@@ -160,6 +163,8 @@ if (ENABLE_VULKAN)
        renderer_vulkan/fixed_pipeline_state.h
        renderer_vulkan/maxwell_to_vk.cpp
        renderer_vulkan/maxwell_to_vk.h
+        renderer_vulkan/nsight_aftermath_tracker.cpp
+        renderer_vulkan/nsight_aftermath_tracker.h
        renderer_vulkan/renderer_vulkan.h
        renderer_vulkan/renderer_vulkan.cpp
        renderer_vulkan/vk_blit_screen.cpp
@@ -174,6 +179,8 @@ if (ENABLE_VULKAN)
        renderer_vulkan/vk_descriptor_pool.h
        renderer_vulkan/vk_device.cpp
        renderer_vulkan/vk_device.h
+        renderer_vulkan/vk_fence_manager.cpp
+        renderer_vulkan/vk_fence_manager.h
        renderer_vulkan/vk_graphics_pipeline.cpp
        renderer_vulkan/vk_graphics_pipeline.h
        renderer_vulkan/vk_image.cpp
@@ -213,19 +220,30 @@ if (ENABLE_VULKAN)
        renderer_vulkan/wrapper.cpp
        renderer_vulkan/wrapper.h
    )
-
-    target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
-    target_compile_definitions(video_core PRIVATE HAS_VULKAN)
 endif()

 create_target_directory_groups(video_core)

 target_link_libraries(video_core PUBLIC common core)
 target_link_libraries(video_core PRIVATE glad)
+
 if (ENABLE_VULKAN)
+    target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
+    target_compile_definitions(video_core PRIVATE HAS_VULKAN)
    target_link_libraries(video_core PRIVATE sirit)
 endif()

+if (ENABLE_NSIGHT_AFTERMATH)
+    if (NOT DEFINED ENV{NSIGHT_AFTERMATH_SDK})
+        message(ERROR "Environment variable NSIGHT_AFTERMATH_SDK has to be provided")
+    endif()
+    if (NOT WIN32)
+        message(ERROR "Nsight Aftermath doesn't support non-Windows platforms")
+    endif()
+    target_compile_definitions(video_core PRIVATE HAS_NSIGHT_AFTERMATH)
+    target_include_directories(video_core PRIVATE "$ENV{NSIGHT_AFTERMATH_SDK}/include")
+endif()
+
 if (MSVC)
    target_compile_options(video_core PRIVATE /we4267)
 else()
@@ -5,6 +5,7 @@
 #pragma once

 #include <array>
+#include <list>
 #include <memory>
 #include <mutex>
 #include <unordered_map>
@@ -18,8 +19,10 @@

 #include "common/alignment.h"
 #include "common/common_types.h"
+#include "common/logging/log.h"
 #include "core/core.h"
 #include "core/memory.h"
+#include "core/settings.h"
 #include "video_core/buffer_cache/buffer_block.h"
 #include "video_core/buffer_cache/map_interval.h"
 #include "video_core/memory_manager.h"
@@ -79,6 +82,9 @@ public:
        auto map = MapAddress(block, gpu_addr, cpu_addr, size);
        if (is_written) {
            map->MarkAsModified(true, GetModifiedTicks());
+            if (Settings::IsGPULevelHigh() && Settings::values.use_asynchronous_gpu_emulation) {
+                MarkForAsyncFlush(map);
+            }
            if (!map->IsWritten()) {
                map->MarkAsWritten(true);
                MarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
@@ -137,11 +143,22 @@ public:
        });
        for (auto& object : objects) {
            if (object->IsModified() && object->IsRegistered()) {
+                mutex.unlock();
                FlushMap(object);
+                mutex.lock();
            }
        }
    }

+    bool MustFlushRegion(VAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        const std::vector<MapInterval> objects = GetMapsInRange(addr, size);
+        return std::any_of(objects.cbegin(), objects.cend(), [](const MapInterval& map) {
+            return map->IsModified() && map->IsRegistered();
+        });
+    }
+
    /// Mark the specified region as being invalidated
    void InvalidateRegion(VAddr addr, u64 size) {
        std::lock_guard lock{mutex};
@@ -154,6 +171,77 @@ public:
        }
    }

+    void OnCPUWrite(VAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        for (const auto& object : GetMapsInRange(addr, size)) {
+            if (object->IsMemoryMarked() && object->IsRegistered()) {
+                UnmarkMemory(object);
+                object->SetSyncPending(true);
+                marked_for_unregister.emplace_back(object);
+            }
+        }
+    }
+
+    void SyncGuestHost() {
+        std::lock_guard lock{mutex};
+
+        for (const auto& object : marked_for_unregister) {
+            if (object->IsRegistered()) {
+                object->SetSyncPending(false);
+                Unregister(object);
+            }
+        }
+        marked_for_unregister.clear();
+    }
+
+    void CommitAsyncFlushes() {
+        if (uncommitted_flushes) {
+            auto commit_list = std::make_shared<std::list<MapInterval>>();
+            for (auto& map : *uncommitted_flushes) {
+                if (map->IsRegistered() && map->IsModified()) {
+                    // TODO(Blinkhawk): Implement backend asynchronous flushing
+                    // AsyncFlushMap(map)
+                    commit_list->push_back(map);
+                }
+            }
+            if (!commit_list->empty()) {
+                committed_flushes.push_back(commit_list);
+            } else {
+                committed_flushes.emplace_back();
+            }
+        } else {
+            committed_flushes.emplace_back();
+        }
+        uncommitted_flushes.reset();
+    }
+
+    bool ShouldWaitAsyncFlushes() const {
+        return !committed_flushes.empty() && committed_flushes.front() != nullptr;
+    }
+
+    bool HasUncommittedFlushes() const {
+        return uncommitted_flushes != nullptr;
+    }
+
+    void PopAsyncFlushes() {
+        if (committed_flushes.empty()) {
+            return;
+        }
+        auto& flush_list = committed_flushes.front();
+        if (!flush_list) {
+            committed_flushes.pop_front();
+            return;
+        }
+        for (MapInterval& map : *flush_list) {
+            if (map->IsRegistered()) {
+                // TODO(Blinkhawk): Replace this for reading the asynchronous flush
+                FlushMap(map);
+            }
+        }
+        committed_flushes.pop_front();
+    }
+
    virtual BufferType GetEmptyBuffer(std::size_t size) = 0;

 protected:
@@ -196,17 +284,30 @@ protected:
        const IntervalType interval{new_map->GetStart(), new_map->GetEnd()};
        mapped_addresses.insert({interval, new_map});
        rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
+        new_map->SetMemoryMarked(true);
        if (inherit_written) {
            MarkRegionAsWritten(new_map->GetStart(), new_map->GetEnd() - 1);
            new_map->MarkAsWritten(true);
        }
    }

-    /// Unregisters an object from the cache
-    void Unregister(MapInterval& map) {
+    void UnmarkMemory(const MapInterval& map) {
+        if (!map->IsMemoryMarked()) {
+            return;
+        }
        const std::size_t size = map->GetEnd() - map->GetStart();
        rasterizer.UpdatePagesCachedCount(map->GetStart(), size, -1);
+        map->SetMemoryMarked(false);
+    }
+
+    /// Unregisters an object from the cache
+    void Unregister(const MapInterval& map) {
+        UnmarkMemory(map);
        map->MarkAsRegistered(false);
+        if (map->IsSyncPending()) {
+            marked_for_unregister.remove(map);
+            map->SetSyncPending(false);
+        }
        if (map->IsWritten()) {
            UnmarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
        }
@@ -264,6 +365,9 @@ private:
        MapInterval new_map = CreateMap(new_start, new_end, new_gpu_addr);
        if (modified_inheritance) {
            new_map->MarkAsModified(true, GetModifiedTicks());
+            if (Settings::IsGPULevelHigh() && Settings::values.use_asynchronous_gpu_emulation) {
+                MarkForAsyncFlush(new_map);
+            }
        }
        Register(new_map, write_inheritance);
        return new_map;
@@ -450,6 +554,13 @@ private:
        return false;
    }

+    void MarkForAsyncFlush(MapInterval& map) {
+        if (!uncommitted_flushes) {
+            uncommitted_flushes = std::make_shared<std::unordered_set<MapInterval>>();
+        }
+        uncommitted_flushes->insert(map);
+    }
+
    VideoCore::RasterizerInterface& rasterizer;
    Core::System& system;

@@ -479,6 +590,10 @@ private:
    u64 modified_ticks = 0;

    std::vector<u8> staging_buffer;
+    std::list<MapInterval> marked_for_unregister;
+
+    std::shared_ptr<std::unordered_set<MapInterval>> uncommitted_flushes{};
+    std::list<std::shared_ptr<std::list<MapInterval>>> committed_flushes;

    std::recursive_mutex mutex;
 };
@@ -46,6 +46,22 @@ public:
        return is_registered;
    }

+    void SetMemoryMarked(bool is_memory_marked_) {
+        is_memory_marked = is_memory_marked_;
+    }
+
+    bool IsMemoryMarked() const {
+        return is_memory_marked;
+    }
+
+    void SetSyncPending(bool is_sync_pending_) {
+        is_sync_pending = is_sync_pending_;
+    }
+
+    bool IsSyncPending() const {
+        return is_sync_pending;
+    }
+
    VAddr GetStart() const {
        return start;
    }
@@ -83,6 +99,8 @@ private:
    bool is_written{};
    bool is_modified{};
    bool is_registered{};
+    bool is_memory_marked{};
+    bool is_sync_pending{};
    u64 ticks{};
 };

@@ -21,6 +21,7 @@ MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128,
 void DmaPusher::DispatchCalls() {
    MICROPROFILE_SCOPE(DispatchCalls);

+    gpu.SyncGuestHost();
    // On entering GPU code, assume all memory may be touched by the ARM core.
    gpu.Maxwell3D().OnMemoryWrite();

@@ -32,6 +33,8 @@ void DmaPusher::DispatchCalls() {
        }
    }
    gpu.FlushCommands();
+    gpu.SyncGuestHost();
+    gpu.OnCommandListEnd();
 }

 bool DmaPusher::Step() {
@@ -68,16 +71,22 @@ bool DmaPusher::Step() {
    gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(),
                                        command_list_header.size * sizeof(u32));

-    for (const CommandHeader& command_header : command_headers) {
+    for (std::size_t index = 0; index < command_headers.size();) {
+        const CommandHeader& command_header = command_headers[index];

-        // now, see if we're in the middle of a command
-        if (dma_state.length_pending) {
-            // Second word of long non-inc methods command - method count
-            dma_state.length_pending = 0;
-            dma_state.method_count = command_header.method_count_;
-        } else if (dma_state.method_count) {
+        if (dma_state.method_count) {
            // Data word of methods command
-            CallMethod(command_header.argument);
+            if (dma_state.non_incrementing) {
+                const u32 max_write = static_cast<u32>(
+                    std::min<std::size_t>(index + dma_state.method_count, command_headers.size()) -
+                    index);
+                CallMultiMethod(&command_header.argument, max_write);
+                dma_state.method_count -= max_write;
+                index += max_write;
+                continue;
+            } else {
+                CallMethod(command_header.argument);
+            }

            if (!dma_state.non_incrementing) {
                dma_state.method++;
@@ -117,6 +126,7 @@ bool DmaPusher::Step() {
                break;
            }
        }
+        index++;
    }

    if (!non_main) {
@@ -137,4 +147,9 @@ void DmaPusher::CallMethod(u32 argument) const {
    gpu.CallMethod({dma_state.method, argument, dma_state.subchannel, dma_state.method_count});
 }

+void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const {
+    gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods,
+                        dma_state.method_count);
+}
+
 } // namespace Tegra
@@ -75,6 +75,7 @@ private:
    void SetState(const CommandHeader& command_header);

    void CallMethod(u32 argument) const;
+    void CallMultiMethod(const u32* base_start, u32 num_methods) const;

    std::vector<CommandHeader> command_headers; ///< Buffer for list of commands fetched at once

@@ -28,7 +28,13 @@ void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
    }
 }

-std::pair<u32, u32> DelimitLine(u32 src_1, u32 src_2, u32 dst_1, u32 dst_2, u32 src_line) {
+void Fermi2D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) {
+    for (std::size_t i = 0; i < amount; i++) {
+        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+    }
+}
+
+static std::pair<u32, u32> DelimitLine(u32 src_1, u32 src_2, u32 dst_1, u32 dst_2, u32 src_line) {
    const u32 line_a = src_2 - src_1;
    const u32 line_b = dst_2 - dst_1;
    const u32 excess = std::max<s32>(0, line_a - src_line + src_1);
@@ -39,6 +39,9 @@ public:
    /// Write the value to the register identified by method.
    void CallMethod(const GPU::MethodCall& method_call);

+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
    enum class Origin : u32 {
        Center = 0,
        Corner = 1,
@@ -51,6 +51,13 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
    }
 }

+void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                    u32 methods_pending) {
+    for (std::size_t i = 0; i < amount; i++) {
+        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+    }
+}
+
 Texture::FullTextureInfo KeplerCompute::GetTexture(std::size_t offset) const {
    const std::bitset<8> cbuf_mask = launch_description.const_buffer_enable_mask.Value();
    ASSERT(cbuf_mask[regs.tex_cb_index]);
@@ -202,6 +202,9 @@ public:
    /// Write the value to the register identified by method.
    void CallMethod(const GPU::MethodCall& method_call);

+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
    Texture::FullTextureInfo GetTexture(std::size_t offset) const;

    /// Given a texture handle, returns the TSC and TIC entries.
@@ -41,4 +41,11 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
    }
 }

+void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                   u32 methods_pending) {
+    for (std::size_t i = 0; i < amount; i++) {
+        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+    }
+}
+
 } // namespace Tegra::Engines
@@ -40,6 +40,9 @@ public:
    /// Write the value to the register identified by method.
    void CallMethod(const GPU::MethodCall& method_call);

+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
    struct Regs {
        static constexpr size_t NUM_REGS = 0x7F;

@@ -92,6 +92,10 @@ void Maxwell3D::InitializeRegisterDefaults() {
        color_mask.A.Assign(1);
    }

+    for (auto& format : regs.vertex_attrib_format) {
+        format.constant.Assign(1);
+    }
+
    // NVN games expect these values to be enabled at boot
    regs.rasterize_enable = 1;
    regs.rt_separate_frag_data = 1;
@@ -276,6 +280,58 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
    }
 }

+void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                u32 methods_pending) {
+    // Methods after 0xE00 are special, they're actually triggers for some microcode that was
+    // uploaded to the GPU during initialization.
+    if (method >= MacroRegistersStart) {
+        // We're trying to execute a macro
+        if (executing_macro == 0) {
+            // A macro call must begin by writing the macro method's register, not its argument.
+            ASSERT_MSG((method % 2) == 0,
+                       "Can't start macro execution by writing to the ARGS register");
+            executing_macro = method;
+        }
+
+        for (std::size_t i = 0; i < amount; i++) {
+            macro_params.push_back(base_start[i]);
+        }
+
+        // Call the macro when there are no more parameters in the command buffer
+        if (amount == methods_pending) {
+            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            macro_params.clear();
+        }
+        return;
+    }
+    switch (method) {
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
+        ProcessCBMultiData(method, base_start, amount);
+        break;
+    }
+    default: {
+        for (std::size_t i = 0; i < amount; i++) {
+            CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        }
+    }
+    }
+}
+
 void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) {
    if (mme_draw.current_mode == MMEDrawMode::Undefined) {
        if (mme_draw.gl_begin_consume) {
@@ -400,7 +456,11 @@ void Maxwell3D::ProcessQueryGet() {

    switch (regs.query.query_get.operation) {
    case Regs::QueryOperation::Release:
-        StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
+        if (regs.query.query_get.fence == 1) {
+            rasterizer.SignalSemaphore(regs.query.QueryAddress(), regs.query.query_sequence);
+        } else {
+            StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
+        }
        break;
    case Regs::QueryOperation::Acquire:
        // TODO(Blinkhawk): Under this operation, the GPU waits for the CPU to write a value that
@@ -479,7 +539,7 @@ void Maxwell3D::ProcessSyncPoint() {
    const u32 increment = regs.sync_info.increment.Value();
    [[maybe_unused]] const u32 cache_flush = regs.sync_info.unknown.Value();
    if (increment) {
-        system.GPU().IncrementSyncPoint(sync_point);
+        rasterizer.SignalSyncPoint(sync_point);
    }
 }

@@ -562,6 +622,28 @@ void Maxwell3D::StartCBData(u32 method) {
    ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
 }

+void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount) {
+    if (cb_data_state.current != method) {
+        if (cb_data_state.current != null_cb_data) {
+            FinishCBData();
+        }
+        constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+        cb_data_state.start_pos = regs.const_buffer.cb_pos;
+        cb_data_state.id = method - first_cb_data;
+        cb_data_state.current = method;
+        cb_data_state.counter = 0;
+    }
+    const std::size_t id = cb_data_state.id;
+    const std::size_t size = amount;
+    std::size_t i = 0;
+    for (; i < size; i++) {
+        cb_data_state.buffer[id][cb_data_state.counter] = start_base[i];
+        cb_data_state.counter++;
+    }
+    // Increment the current buffer position.
+    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4 * amount;
+}
+
 void Maxwell3D::FinishCBData() {
    // Write the input value to the current const buffer at the current position.
    const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
@@ -1149,7 +1149,7 @@ public:

                    /// Returns whether the vertex array specified by index is supposed to be
                    /// accessed per instance or not.
-                    bool IsInstancingEnabled(u32 index) const {
+                    bool IsInstancingEnabled(std::size_t index) const {
                        return is_instanced[index];
                    }
                } instanced_arrays;
@@ -1358,6 +1358,9 @@ public:
    /// Write the value to the register identified by method.
    void CallMethod(const GPU::MethodCall& method_call);

+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
    /// Write the value to the register identified by method.
    void CallMethodFromMME(const GPU::MethodCall& method_call);

@@ -1511,6 +1514,7 @@ private:
    /// Handles a write to the CB_DATA[i] register.
    void StartCBData(u32 method);
    void ProcessCBData(u32 value);
+    void ProcessCBMultiData(u32 method, const u32* start_base, u32 amount);
    void FinishCBData();

    /// Handles a write to the CB_BIND register.
@@ -36,6 +36,13 @@ void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
 #undef MAXWELLDMA_REG_INDEX
 }

+void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                 u32 methods_pending) {
+    for (std::size_t i = 0; i < amount; i++) {
+        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+    }
+}
+
 void MaxwellDMA::HandleCopy() {
    LOG_TRACE(HW_GPU, "Requested a DMA copy");

@@ -104,8 +111,13 @@ void MaxwellDMA::HandleCopy() {
            write_buffer.resize(dst_size);
        }

-        memory_manager.ReadBlock(source, read_buffer.data(), src_size);
-        memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
+        if (Settings::IsGPULevelExtreme()) {
+            memory_manager.ReadBlock(source, read_buffer.data(), src_size);
+            memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
+        } else {
+            memory_manager.ReadBlockUnsafe(source, read_buffer.data(), src_size);
+            memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
+        }

        Texture::UnswizzleSubrect(
            regs.x_count, regs.y_count, regs.dst_pitch, regs.src_params.size_x, bytes_per_pixel,
@@ -136,7 +148,7 @@ void MaxwellDMA::HandleCopy() {
            write_buffer.resize(dst_size);
        }

-        if (Settings::values.use_accurate_gpu_emulation) {
+        if (Settings::IsGPULevelExtreme()) {
            memory_manager.ReadBlock(source, read_buffer.data(), src_size);
            memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
        } else {
@@ -35,6 +35,9 @@ public:
    /// Write the value to the register identified by method.
    void CallMethod(const GPU::MethodCall& method_call);

+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
    struct Regs {
        static constexpr std::size_t NUM_REGS = 0x1D6;

@@ -0,0 +1,170 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <memory>
+#include <queue>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "core/core.h"
+#include "core/memory.h"
+#include "core/settings.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace VideoCommon {
+
+class FenceBase {
+public:
+    FenceBase(u32 payload, bool is_stubbed)
+        : address{}, payload{payload}, is_semaphore{false}, is_stubbed{is_stubbed} {}
+
+    FenceBase(GPUVAddr address, u32 payload, bool is_stubbed)
+        : address{address}, payload{payload}, is_semaphore{true}, is_stubbed{is_stubbed} {}
+
+    GPUVAddr GetAddress() const {
+        return address;
+    }
+
+    u32 GetPayload() const {
+        return payload;
+    }
+
+    bool IsSemaphore() const {
+        return is_semaphore;
+    }
+
+private:
+    GPUVAddr address;
+    u32 payload;
+    bool is_semaphore;
+
+protected:
+    bool is_stubbed;
+};
+
+template <typename TFence, typename TTextureCache, typename TTBufferCache, typename TQueryCache>
+class FenceManager {
+public:
+    void SignalSemaphore(GPUVAddr addr, u32 value) {
+        TryReleasePendingFences();
+        const bool should_flush = ShouldFlush();
+        CommitAsyncFlushes();
+        TFence new_fence = CreateFence(addr, value, !should_flush);
+        fences.push(new_fence);
+        QueueFence(new_fence);
+        if (should_flush) {
+            rasterizer.FlushCommands();
+        }
+        rasterizer.SyncGuestHost();
+    }
+
+    void SignalSyncPoint(u32 value) {
+        TryReleasePendingFences();
+        const bool should_flush = ShouldFlush();
+        CommitAsyncFlushes();
+        TFence new_fence = CreateFence(value, !should_flush);
+        fences.push(new_fence);
+        QueueFence(new_fence);
+        if (should_flush) {
+            rasterizer.FlushCommands();
+        }
+        rasterizer.SyncGuestHost();
+    }
+
+    void WaitPendingFences() {
+        auto& gpu{system.GPU()};
+        auto& memory_manager{gpu.MemoryManager()};
+        while (!fences.empty()) {
+            TFence& current_fence = fences.front();
+            if (ShouldWait()) {
+                WaitFence(current_fence);
+            }
+            PopAsyncFlushes();
+            if (current_fence->IsSemaphore()) {
+                memory_manager.Write<u32>(current_fence->GetAddress(), current_fence->GetPayload());
+            } else {
+                gpu.IncrementSyncPoint(current_fence->GetPayload());
+            }
+            fences.pop();
+        }
+    }
+
+protected:
+    FenceManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                 TTextureCache& texture_cache, TTBufferCache& buffer_cache,
+                 TQueryCache& query_cache)
+        : system{system}, rasterizer{rasterizer}, texture_cache{texture_cache},
+          buffer_cache{buffer_cache}, query_cache{query_cache} {}
+
+    virtual ~FenceManager() {}
+
+    /// Creates a Sync Point Fence Interface, does not create a backend fence if 'is_stubbed' is
+    /// true
+    virtual TFence CreateFence(u32 value, bool is_stubbed) = 0;
+    /// Creates a Semaphore Fence Interface, does not create a backend fence if 'is_stubbed' is true
+    virtual TFence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) = 0;
+    /// Queues a fence into the backend if the fence isn't stubbed.
+    virtual void QueueFence(TFence& fence) = 0;
+    /// Notifies that the backend fence has been signaled/reached in host GPU.
+    virtual bool IsFenceSignaled(TFence& fence) const = 0;
+    /// Waits until a fence has been signalled by the host GPU.
+    virtual void WaitFence(TFence& fence) = 0;
+
+    Core::System& system;
+    VideoCore::RasterizerInterface& rasterizer;
+    TTextureCache& texture_cache;
+    TTBufferCache& buffer_cache;
+    TQueryCache& query_cache;
+
+private:
+    void TryReleasePendingFences() {
+        auto& gpu{system.GPU()};
+        auto& memory_manager{gpu.MemoryManager()};
+        while (!fences.empty()) {
+            TFence& current_fence = fences.front();
+            if (ShouldWait() && !IsFenceSignaled(current_fence)) {
+                return;
+            }
+            PopAsyncFlushes();
+            if (current_fence->IsSemaphore()) {
+                memory_manager.Write<u32>(current_fence->GetAddress(), current_fence->GetPayload());
+            } else {
+                gpu.IncrementSyncPoint(current_fence->GetPayload());
+            }
+            fences.pop();
+        }
+    }
+
+    bool ShouldWait() const {
+        return texture_cache.ShouldWaitAsyncFlushes() || buffer_cache.ShouldWaitAsyncFlushes() ||
+               query_cache.ShouldWaitAsyncFlushes();
+    }
+
+    bool ShouldFlush() const {
+        return texture_cache.HasUncommittedFlushes() || buffer_cache.HasUncommittedFlushes() ||
+               query_cache.HasUncommittedFlushes();
+    }
+
+    void PopAsyncFlushes() {
+        texture_cache.PopAsyncFlushes();
+        buffer_cache.PopAsyncFlushes();
+        query_cache.PopAsyncFlushes();
+    }
+
+    void CommitAsyncFlushes() {
+        texture_cache.CommitAsyncFlushes();
+        buffer_cache.CommitAsyncFlushes();
+        query_cache.CommitAsyncFlushes();
+    }
+
+    std::queue<TFence> fences;
+};
+
+} // namespace VideoCommon
@@ -9,6 +9,7 @@
 #include "core/core_timing_util.h"
 #include "core/frontend/emu_window.h"
 #include "core/memory.h"
+#include "core/settings.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/kepler_memory.h"
@@ -125,6 +126,28 @@ bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
    return true;
 }

+u64 GPU::RequestFlush(VAddr addr, std::size_t size) {
+    std::unique_lock lck{flush_request_mutex};
+    const u64 fence = ++last_flush_fence;
+    flush_requests.emplace_back(fence, addr, size);
+    return fence;
+}
+
+void GPU::TickWork() {
+    std::unique_lock lck{flush_request_mutex};
+    while (!flush_requests.empty()) {
+        auto& request = flush_requests.front();
+        const u64 fence = request.fence;
+        const VAddr addr = request.addr;
+        const std::size_t size = request.size;
+        flush_requests.pop_front();
+        flush_request_mutex.unlock();
+        renderer->Rasterizer().FlushRegion(addr, size);
+        current_flush_fence.store(fence);
+        flush_request_mutex.lock();
+    }
+}
+
 u64 GPU::GetTicks() const {
    // This values were reversed engineered by fincs from NVN
    // The gpu clock is reported in units of 385/625 nanoseconds
@@ -132,7 +155,10 @@ u64 GPU::GetTicks() const {
    constexpr u64 gpu_ticks_den = 625;

    const u64 cpu_ticks = system.CoreTiming().GetTicks();
-    const u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count();
+    u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count();
+    if (Settings::values.use_fast_gpu_time) {
+        nanoseconds /= 256;
+    }
    const u64 nanoseconds_num = nanoseconds / gpu_ticks_den;
    const u64 nanoseconds_rem = nanoseconds % gpu_ticks_den;
    return nanoseconds_num * gpu_ticks_num + (nanoseconds_rem * gpu_ticks_num) / gpu_ticks_den;
@@ -142,6 +168,13 @@ void GPU::FlushCommands() {
    renderer->Rasterizer().FlushCommands();
 }

+void GPU::SyncGuestHost() {
+    renderer->Rasterizer().SyncGuestHost();
+}
+
+void GPU::OnCommandListEnd() {
+    renderer->Rasterizer().ReleaseFences();
+}
 // Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
 // their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
 // So the values you see in docs might be multiplied by 4.
@@ -180,16 +213,32 @@ void GPU::CallMethod(const MethodCall& method_call) {

    ASSERT(method_call.subchannel < bound_engines.size());

-    if (ExecuteMethodOnEngine(method_call)) {
+    if (ExecuteMethodOnEngine(method_call.method)) {
        CallEngineMethod(method_call);
    } else {
        CallPullerMethod(method_call);
    }
 }

-bool GPU::ExecuteMethodOnEngine(const MethodCall& method_call) {
-    const auto method = static_cast<BufferMethods>(method_call.method);
-    return method >= BufferMethods::NonPullerMethods;
+void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                          u32 methods_pending) {
+    LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel);
+
+    ASSERT(subchannel < bound_engines.size());
+
+    if (ExecuteMethodOnEngine(method)) {
+        CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending);
+    } else {
+        for (std::size_t i = 0; i < amount; i++) {
+            CallPullerMethod(
+                {method, base_start[i], subchannel, methods_pending - static_cast<u32>(i)});
+        }
+    }
+}
+
+bool GPU::ExecuteMethodOnEngine(u32 method) {
+    const auto buffer_method = static_cast<BufferMethods>(method);
+    return buffer_method >= BufferMethods::NonPullerMethods;
 }

 void GPU::CallPullerMethod(const MethodCall& method_call) {
@@ -269,6 +318,31 @@ void GPU::CallEngineMethod(const MethodCall& method_call) {
    }
 }

+void GPU::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                                u32 methods_pending) {
+    const EngineID engine = bound_engines[subchannel];
+
+    switch (engine) {
+    case EngineID::FERMI_TWOD_A:
+        fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending);
+        break;
+    case EngineID::MAXWELL_B:
+        maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending);
+        break;
+    case EngineID::KEPLER_COMPUTE_B:
+        kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending);
+        break;
+    case EngineID::MAXWELL_DMA_COPY_A:
+        maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending);
+        break;
+    case EngineID::KEPLER_INLINE_TO_MEMORY_B:
+        kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented engine");
+    }
+}
+
 void GPU::ProcessBindMethod(const MethodCall& method_call) {
    // Bind the current subchannel to the desired engine id.
    LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
@@ -155,7 +155,27 @@ public:
    /// Calls a GPU method.
    void CallMethod(const MethodCall& method_call);

+    /// Calls a GPU multivalue method.
+    void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                         u32 methods_pending);
+
+    /// Flush all current written commands into the host GPU for execution.
    void FlushCommands();
+    /// Synchronizes CPU writes with Host GPU memory.
+    void SyncGuestHost();
+    /// Signal the ending of command list.
+    virtual void OnCommandListEnd();
+
+    /// Request a host GPU memory flush from the CPU.
+    u64 RequestFlush(VAddr addr, std::size_t size);
+
+    /// Obtains current flush request fence id.
+    u64 CurrentFlushRequestFence() const {
+        return current_flush_fence.load(std::memory_order_relaxed);
+    }
+
+    /// Tick pending requests within the GPU.
+    void TickWork();

    /// Returns a reference to the Maxwell3D GPU engine.
    Engines::Maxwell3D& Maxwell3D();
@@ -293,8 +313,12 @@ private:
    /// Calls a GPU engine method.
    void CallEngineMethod(const MethodCall& method_call);

+    /// Calls a GPU engine multivalue method.
+    void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount,
+                               u32 methods_pending);
+
    /// Determines where the method should be executed.
-    bool ExecuteMethodOnEngine(const MethodCall& method_call);
+    bool ExecuteMethodOnEngine(u32 method);

 protected:
    std::unique_ptr<Tegra::DmaPusher> dma_pusher;
@@ -325,6 +349,19 @@ private:

    std::condition_variable sync_cv;

+    struct FlushRequest {
+        FlushRequest(u64 fence, VAddr addr, std::size_t size)
+            : fence{fence}, addr{addr}, size{size} {}
+        u64 fence;
+        VAddr addr;
+        std::size_t size;
+    };
+
+    std::list<FlushRequest> flush_requests;
+    std::atomic<u64> current_flush_fence{};
+    u64 last_flush_fence{};
+    std::mutex flush_request_mutex;
+
    const bool is_async;
 };

@@ -52,4 +52,8 @@ void GPUAsynch::WaitIdle() const {
    gpu_thread.WaitIdle();
 }

+void GPUAsynch::OnCommandListEnd() {
+    gpu_thread.OnCommandListEnd();
+}
+
 } // namespace VideoCommon
@@ -32,6 +32,8 @@ public:
    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
    void WaitIdle() const override;

+    void OnCommandListEnd() override;
+
 protected:
    void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;

@@ -6,6 +6,7 @@
 #include "common/microprofile.h"
 #include "core/core.h"
 #include "core/frontend/emu_window.h"
+#include "core/settings.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/gpu.h"
 #include "video_core/gpu_thread.h"
@@ -14,8 +15,9 @@
 namespace VideoCommon::GPUThread {

 /// Runs the GPU thread
-static void RunThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context,
-                      Tegra::DmaPusher& dma_pusher, SynchState& state) {
+static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
+                      Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher,
+                      SynchState& state) {
    MicroProfileOnThreadCreate("GpuThread");

    // Wait for first GPU command before acquiring the window context
@@ -37,10 +39,14 @@ static void RunThread(VideoCore::RendererBase& renderer, Core::Frontend::Graphic
            dma_pusher.DispatchCalls();
        } else if (const auto data = std::get_if<SwapBuffersCommand>(&next.data)) {
            renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr);
+        } else if (const auto data = std::get_if<OnCommandListEndCommand>(&next.data)) {
+            renderer.Rasterizer().ReleaseFences();
+        } else if (const auto data = std::get_if<GPUTickCommand>(&next.data)) {
+            system.GPU().TickWork();
        } else if (const auto data = std::get_if<FlushRegionCommand>(&next.data)) {
            renderer.Rasterizer().FlushRegion(data->addr, data->size);
        } else if (const auto data = std::get_if<InvalidateRegionCommand>(&next.data)) {
-            renderer.Rasterizer().InvalidateRegion(data->addr, data->size);
+            renderer.Rasterizer().OnCPUWrite(data->addr, data->size);
        } else if (std::holds_alternative<EndProcessingCommand>(next.data)) {
            return;
        } else {
@@ -65,8 +71,8 @@ ThreadManager::~ThreadManager() {
 void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
                                Core::Frontend::GraphicsContext& context,
                                Tegra::DmaPusher& dma_pusher) {
-    thread = std::thread{RunThread, std::ref(renderer), std::ref(context), std::ref(dma_pusher),
-                         std::ref(state)};
+    thread = std::thread{RunThread,         std::ref(system),     std::ref(renderer),
+                         std::ref(context), std::ref(dma_pusher), std::ref(state)};
 }

 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
@@ -78,16 +84,29 @@ void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
 }

 void ThreadManager::FlushRegion(VAddr addr, u64 size) {
-    PushCommand(FlushRegionCommand(addr, size));
+    if (!Settings::IsGPULevelHigh()) {
+        PushCommand(FlushRegionCommand(addr, size));
+        return;
+    }
+    if (!Settings::IsGPULevelExtreme()) {
+        return;
+    }
+    if (system.Renderer().Rasterizer().MustFlushRegion(addr, size)) {
+        auto& gpu = system.GPU();
+        u64 fence = gpu.RequestFlush(addr, size);
+        PushCommand(GPUTickCommand());
+        while (fence > gpu.CurrentFlushRequestFence()) {
+        }
+    }
 }

 void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
-    system.Renderer().Rasterizer().InvalidateRegion(addr, size);
+    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
 }

 void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
    // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
-    InvalidateRegion(addr, size);
+    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
 }

 void ThreadManager::WaitIdle() const {
@@ -95,6 +114,10 @@ void ThreadManager::WaitIdle() const {
    }
 }

+void ThreadManager::OnCommandListEnd() {
+    PushCommand(OnCommandListEndCommand());
+}
+
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
    const u64 fence{++state.last_fence};
    state.queue.Push(CommandDataContainer(std::move(command_data), fence));
@@ -70,9 +70,16 @@ struct FlushAndInvalidateRegionCommand final {
    u64 size;
 };

+/// Command called within the gpu, to schedule actions after a command list end
+struct OnCommandListEndCommand final {};
+
+/// Command to make the gpu look into pending requests
+struct GPUTickCommand final {};
+
 using CommandData =
    std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
-                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand>;
+                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand, OnCommandListEndCommand,
+                 GPUTickCommand>;

 struct CommandDataContainer {
    CommandDataContainer() = default;
@@ -122,6 +129,8 @@ public:
    // Wait until the gpu thread is idle.
    void WaitIdle() const;

+    void OnCommandListEnd();
+
 private:
    /// Pushes a command to be executed by the GPU thread
    u64 PushCommand(CommandData&& command_data);
@@ -12,10 +12,12 @@
 #include <mutex>
 #include <optional>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>

 #include "common/assert.h"
 #include "core/core.h"
+#include "core/settings.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
@@ -130,6 +132,9 @@ public:
        }

        query->BindCounter(Stream(type).Current(), timestamp);
+        if (Settings::values.use_asynchronous_gpu_emulation) {
+            AsyncFlushQuery(cpu_addr);
+        }
    }

    /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
@@ -170,6 +175,37 @@ public:
        return streams[static_cast<std::size_t>(type)];
    }

+    void CommitAsyncFlushes() {
+        committed_flushes.push_back(uncommitted_flushes);
+        uncommitted_flushes.reset();
+    }
+
+    bool HasUncommittedFlushes() const {
+        return uncommitted_flushes != nullptr;
+    }
+
+    bool ShouldWaitAsyncFlushes() const {
+        if (committed_flushes.empty()) {
+            return false;
+        }
+        return committed_flushes.front() != nullptr;
+    }
+
+    void PopAsyncFlushes() {
+        if (committed_flushes.empty()) {
+            return;
+        }
+        auto& flush_list = committed_flushes.front();
+        if (!flush_list) {
+            committed_flushes.pop_front();
+            return;
+        }
+        for (VAddr query_address : *flush_list) {
+            FlushAndRemoveRegion(query_address, 4);
+        }
+        committed_flushes.pop_front();
+    }
+
 protected:
    std::array<QueryPool, VideoCore::NumQueryTypes> query_pools;

@@ -224,6 +260,13 @@ private:
        return found != std::end(contents) ? &*found : nullptr;
    }

+    void AsyncFlushQuery(VAddr addr) {
+        if (!uncommitted_flushes) {
+            uncommitted_flushes = std::make_shared<std::unordered_set<VAddr>>();
+        }
+        uncommitted_flushes->insert(addr);
+    }
+
    static constexpr std::uintptr_t PAGE_SIZE = 4096;
    static constexpr unsigned PAGE_SHIFT = 12;

@@ -235,6 +278,9 @@ private:
    std::unordered_map<u64, std::vector<CachedQuery>> cached_queries;

    std::array<CounterStream, VideoCore::NumQueryTypes> streams;
+
+    std::shared_ptr<std::unordered_set<VAddr>> uncommitted_flushes{};
+    std::list<std::shared_ptr<std::unordered_set<VAddr>>> committed_flushes;
 };

 template <class QueryCache, class HostCounter>
@@ -49,15 +49,33 @@ public:
    /// Records a GPU query and caches it
    virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0;

+    /// Signal a GPU based semaphore as a fence
+    virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0;
+
+    /// Signal a GPU based syncpoint as a fence
+    virtual void SignalSyncPoint(u32 value) = 0;
+
+    /// Release all pending fences.
+    virtual void ReleaseFences() = 0;
+
    /// Notify rasterizer that all caches should be flushed to Switch memory
    virtual void FlushAll() = 0;

    /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
    virtual void FlushRegion(VAddr addr, u64 size) = 0;

+    /// Check if the the specified memory area requires flushing to CPU Memory.
+    virtual bool MustFlushRegion(VAddr addr, u64 size) = 0;
+
    /// Notify rasterizer that any caches of the specified region should be invalidated
    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;

+    /// Notify rasterizer that any caches of the specified region are desync with guest
+    virtual void OnCPUWrite(VAddr addr, u64 size) = 0;
+
+    /// Sync memory between guest and host.
+    virtual void SyncGuestHost() = 0;
+
    /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
    /// and invalidated
    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
@@ -52,7 +52,7 @@ Buffer OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
 }

 void OGLBufferCache::WriteBarrier() {
-    glMemoryBarrier(GL_ALL_BARRIER_BITS);
+    glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 }

 GLuint OGLBufferCache::ToHandle(const Buffer& buffer) {
@@ -72,6 +72,7 @@ void OGLBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, s
 void OGLBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
                                       u8* data) {
    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
+    glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
    glGetNamedBufferSubData(buffer->GetHandle(), static_cast<GLintptr>(offset),
                            static_cast<GLsizeiptr>(size), data);
 }
@@ -0,0 +1,72 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+
+#include "video_core/renderer_opengl/gl_fence_manager.h"
+
+namespace OpenGL {
+
+GLInnerFence::GLInnerFence(u32 payload, bool is_stubbed)
+    : VideoCommon::FenceBase(payload, is_stubbed), sync_object{} {}
+
+GLInnerFence::GLInnerFence(GPUVAddr address, u32 payload, bool is_stubbed)
+    : VideoCommon::FenceBase(address, payload, is_stubbed), sync_object{} {}
+
+GLInnerFence::~GLInnerFence() = default;
+
+void GLInnerFence::Queue() {
+    if (is_stubbed) {
+        return;
+    }
+    ASSERT(sync_object.handle == 0);
+    sync_object.Create();
+}
+
+bool GLInnerFence::IsSignaled() const {
+    if (is_stubbed) {
+        return true;
+    }
+    ASSERT(sync_object.handle != 0);
+    GLsizei length;
+    GLint sync_status;
+    glGetSynciv(sync_object.handle, GL_SYNC_STATUS, sizeof(GLint), &length, &sync_status);
+    return sync_status == GL_SIGNALED;
+}
+
+void GLInnerFence::Wait() {
+    if (is_stubbed) {
+        return;
+    }
+    ASSERT(sync_object.handle != 0);
+    glClientWaitSync(sync_object.handle, 0, GL_TIMEOUT_IGNORED);
+}
+
+FenceManagerOpenGL::FenceManagerOpenGL(Core::System& system,
+                                       VideoCore::RasterizerInterface& rasterizer,
+                                       TextureCacheOpenGL& texture_cache,
+                                       OGLBufferCache& buffer_cache, QueryCache& query_cache)
+    : GenericFenceManager(system, rasterizer, texture_cache, buffer_cache, query_cache) {}
+
+Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) {
+    return std::make_shared<GLInnerFence>(value, is_stubbed);
+}
+
+Fence FenceManagerOpenGL::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) {
+    return std::make_shared<GLInnerFence>(addr, value, is_stubbed);
+}
+
+void FenceManagerOpenGL::QueueFence(Fence& fence) {
+    fence->Queue();
+}
+
+bool FenceManagerOpenGL::IsFenceSignaled(Fence& fence) const {
+    return fence->IsSignaled();
+}
+
+void FenceManagerOpenGL::WaitFence(Fence& fence) {
+    fence->Wait();
+}
+
+} // namespace OpenGL
@@ -0,0 +1,53 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <glad/glad.h>
+
+#include "common/common_types.h"
+#include "video_core/fence_manager.h"
+#include "video_core/renderer_opengl/gl_buffer_cache.h"
+#include "video_core/renderer_opengl/gl_query_cache.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/renderer_opengl/gl_texture_cache.h"
+
+namespace OpenGL {
+
+class GLInnerFence : public VideoCommon::FenceBase {
+public:
+    GLInnerFence(u32 payload, bool is_stubbed);
+    GLInnerFence(GPUVAddr address, u32 payload, bool is_stubbed);
+    ~GLInnerFence();
+
+    void Queue();
+
+    bool IsSignaled() const;
+
+    void Wait();
+
+private:
+    OGLSync sync_object;
+};
+
+using Fence = std::shared_ptr<GLInnerFence>;
+using GenericFenceManager =
+    VideoCommon::FenceManager<Fence, TextureCacheOpenGL, OGLBufferCache, QueryCache>;
+
+class FenceManagerOpenGL final : public GenericFenceManager {
+public:
+    FenceManagerOpenGL(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                       TextureCacheOpenGL& texture_cache, OGLBufferCache& buffer_cache,
+                       QueryCache& query_cache);
+
+protected:
+    Fence CreateFence(u32 value, bool is_stubbed) override;
+    Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override;
+    void QueueFence(Fence& fence) override;
+    bool IsFenceSignaled(Fence& fence) const override;
+    void WaitFence(Fence& fence) override;
+};
+
+} // namespace OpenGL
@@ -99,9 +99,10 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind
                                   ScreenInfo& info, GLShader::ProgramManager& program_manager,
                                   StateTracker& state_tracker)
    : RasterizerAccelerated{system.Memory()}, texture_cache{system, *this, device, state_tracker},
-      shader_cache{*this, system, emu_window, device}, query_cache{system, *this}, system{system},
-      screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker},
-      buffer_cache{*this, system, device, STREAM_BUFFER_SIZE} {
+      shader_cache{*this, system, emu_window, device}, query_cache{system, *this},
+      buffer_cache{*this, system, device, STREAM_BUFFER_SIZE},
+      fence_manager{system, *this, texture_cache, buffer_cache, query_cache}, system{system},
+      screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} {
    CheckExtensions();
 }

@@ -599,6 +600,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
    EndTransformFeedback();

    ++num_queued_commands;
+
+    system.GPU().TickWork();
 }

 void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
@@ -649,6 +652,13 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
    query_cache.FlushRegion(addr, size);
 }

+bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
+    if (!Settings::IsGPULevelHigh()) {
+        return buffer_cache.MustFlushRegion(addr, size);
+    }
+    return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size);
+}
+
 void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
    if (addr == 0 || size == 0) {
@@ -660,8 +670,52 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
    query_cache.InvalidateRegion(addr, size);
 }

+void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    texture_cache.OnCPUWrite(addr, size);
+    shader_cache.InvalidateRegion(addr, size);
+    buffer_cache.OnCPUWrite(addr, size);
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerOpenGL::SyncGuestHost() {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    texture_cache.SyncGuestHost();
+    buffer_cache.SyncGuestHost();
+}
+
+void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        auto& memory_manager{gpu.MemoryManager()};
+        memory_manager.Write<u32>(addr, value);
+        return;
+    }
+    fence_manager.SignalSemaphore(addr, value);
+}
+
+void RasterizerOpenGL::SignalSyncPoint(u32 value) {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        gpu.IncrementSyncPoint(value);
+        return;
+    }
+    fence_manager.SignalSyncPoint(value);
+}
+
+void RasterizerOpenGL::ReleaseFences() {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        return;
+    }
+    fence_manager.WaitPendingFences();
+}
+
 void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
-    if (Settings::values.use_accurate_gpu_emulation) {
+    if (Settings::IsGPULevelExtreme()) {
        FlushRegion(addr, size);
    }
    InvalidateRegion(addr, size);
@@ -23,6 +23,7 @@
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
+#include "video_core/renderer_opengl/gl_fence_manager.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
 #include "video_core/renderer_opengl/gl_query_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
@@ -66,7 +67,13 @@ public:
    void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
    void FlushAll() override;
    void FlushRegion(VAddr addr, u64 size) override;
+    bool MustFlushRegion(VAddr addr, u64 size) override;
    void InvalidateRegion(VAddr addr, u64 size) override;
+    void OnCPUWrite(VAddr addr, u64 size) override;
+    void SyncGuestHost() override;
+    void SignalSemaphore(GPUVAddr addr, u32 value) override;
+    void SignalSyncPoint(u32 value) override;
+    void ReleaseFences() override;
    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
    void FlushCommands() override;
    void TickFrame() override;
@@ -222,6 +229,8 @@ private:
    SamplerCacheOpenGL sampler_cache;
    FramebufferCacheOpenGL framebuffer_cache;
    QueryCache query_cache;
+    OGLBufferCache buffer_cache;
+    FenceManagerOpenGL fence_manager;

    Core::System& system;
    ScreenInfo& screen_info;
@@ -229,7 +238,6 @@ private:
    StateTracker& state_tracker;

    static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
-    OGLBufferCache buffer_cache;

    GLint vertex_binding = 0;

@@ -448,7 +448,7 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {

    // Look up shader in the cache based on address
    const auto cpu_addr{memory_manager.GpuToCpuAddress(address)};
-    Shader shader{cpu_addr ? TryGet(*cpu_addr) : nullptr};
+    Shader shader{cpu_addr ? TryGet(*cpu_addr) : null_shader};
    if (shader) {
        return last_shaders[static_cast<std::size_t>(program)] = shader;
    }
@@ -477,7 +477,12 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
        const std::size_t size_in_bytes = code.size() * sizeof(u64);
        shader = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
    }
-    Register(shader);
+
+    if (cpu_addr) {
+        Register(shader);
+    } else {
+        null_shader = shader;
+    }

    return last_shaders[static_cast<std::size_t>(program)] = shader;
 }
@@ -486,7 +491,7 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
    auto& memory_manager{system.GPU().MemoryManager()};
    const auto cpu_addr{memory_manager.GpuToCpuAddress(code_addr)};

-    auto kernel = cpu_addr ? TryGet(*cpu_addr) : nullptr;
+    auto kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
    if (kernel) {
        return kernel;
    }
@@ -507,7 +512,11 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
        kernel = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
    }

-    Register(kernel);
+    if (cpu_addr) {
+        Register(kernel);
+    } else {
+        null_kernel = kernel;
+    }
    return kernel;
 }

@@ -125,6 +125,9 @@ private:
    ShaderDiskCacheOpenGL disk_cache;
    std::unordered_map<u64, PrecompiledShader> runtime_cache;

+    Shader null_shader{};
+    Shader null_kernel{};
+
    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
 };

@@ -484,7 +484,7 @@ private:
        code.AddLine("switch (jmp_to) {{");

        for (const auto& pair : ir.GetBasicBlocks()) {
-            const auto [address, bb] = pair;
+            const auto& [address, bb] = pair;
            code.AddLine("case 0x{:X}U: {{", address);
            ++code.scope;

@@ -1484,8 +1484,8 @@ private:
        dy += '(';

        for (std::size_t index = 0; index < components; ++index) {
-            const auto operand_x{derivates.at(index * 2)};
-            const auto operand_y{derivates.at(index * 2 + 1)};
+            const auto& operand_x{derivates.at(index * 2)};
+            const auto& operand_y{derivates.at(index * 2 + 1)};
            dx += Visit(operand_x).AsFloat();
            dy += Visit(operand_y).AsFloat();

@@ -2,10 +2,12 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.

+#include <cstring>
 #include <tuple>

 #include <boost/functional/hash.hpp>

+#include "common/cityhash.h"
 #include "common/common_types.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"

@@ -13,289 +15,352 @@ namespace Vulkan {

 namespace {

-constexpr FixedPipelineState::DepthStencil GetDepthStencilState(const Maxwell& regs) {
-    const FixedPipelineState::StencilFace front_stencil(
-        regs.stencil_front_op_fail, regs.stencil_front_op_zfail, regs.stencil_front_op_zpass,
-        regs.stencil_front_func_func);
-    const FixedPipelineState::StencilFace back_stencil =
-        regs.stencil_two_side_enable
-            ? FixedPipelineState::StencilFace(regs.stencil_back_op_fail, regs.stencil_back_op_zfail,
-                                              regs.stencil_back_op_zpass,
-                                              regs.stencil_back_func_func)
-            : front_stencil;
-    return FixedPipelineState::DepthStencil(
-        regs.depth_test_enable == 1, regs.depth_write_enabled == 1, regs.depth_bounds_enable == 1,
-        regs.stencil_enable == 1, regs.depth_test_func, front_stencil, back_stencil);
-}
-
-constexpr FixedPipelineState::InputAssembly GetInputAssemblyState(const Maxwell& regs) {
-    return FixedPipelineState::InputAssembly(
-        regs.draw.topology, regs.primitive_restart.enabled,
-        regs.draw.topology == Maxwell::PrimitiveTopology::Points ? regs.point_size : 0.0f);
-}
-
-constexpr FixedPipelineState::BlendingAttachment GetBlendingAttachmentState(
-    const Maxwell& regs, std::size_t render_target) {
-    const auto& mask = regs.color_mask[regs.color_mask_common ? 0 : render_target];
-    const std::array components = {mask.R != 0, mask.G != 0, mask.B != 0, mask.A != 0};
-
-    const FixedPipelineState::BlendingAttachment default_blending(
-        false, Maxwell::Blend::Equation::Add, Maxwell::Blend::Factor::One,
-        Maxwell::Blend::Factor::Zero, Maxwell::Blend::Equation::Add, Maxwell::Blend::Factor::One,
-        Maxwell::Blend::Factor::Zero, components);
-    if (render_target >= regs.rt_control.count) {
-        return default_blending;
-    }
-
-    if (!regs.independent_blend_enable) {
-        const auto& src = regs.blend;
-        if (!src.enable[render_target]) {
-            return default_blending;
-        }
-        return FixedPipelineState::BlendingAttachment(
-            true, src.equation_rgb, src.factor_source_rgb, src.factor_dest_rgb, src.equation_a,
-            src.factor_source_a, src.factor_dest_a, components);
-    }
-
-    if (!regs.blend.enable[render_target]) {
-        return default_blending;
-    }
-    const auto& src = regs.independent_blend[render_target];
-    return FixedPipelineState::BlendingAttachment(
-        true, src.equation_rgb, src.factor_source_rgb, src.factor_dest_rgb, src.equation_a,
-        src.factor_source_a, src.factor_dest_a, components);
-}
-
-constexpr FixedPipelineState::ColorBlending GetColorBlendingState(const Maxwell& regs) {
-    return FixedPipelineState::ColorBlending(
-        {regs.blend_color.r, regs.blend_color.g, regs.blend_color.b, regs.blend_color.a},
-        regs.rt_control.count,
-        {GetBlendingAttachmentState(regs, 0), GetBlendingAttachmentState(regs, 1),
-         GetBlendingAttachmentState(regs, 2), GetBlendingAttachmentState(regs, 3),
-         GetBlendingAttachmentState(regs, 4), GetBlendingAttachmentState(regs, 5),
-         GetBlendingAttachmentState(regs, 6), GetBlendingAttachmentState(regs, 7)});
-}
-
-constexpr FixedPipelineState::Tessellation GetTessellationState(const Maxwell& regs) {
-    return FixedPipelineState::Tessellation(regs.patch_vertices, regs.tess_mode.prim,
-                                            regs.tess_mode.spacing, regs.tess_mode.cw != 0);
-}
-
-constexpr std::size_t Point = 0;
-constexpr std::size_t Line = 1;
-constexpr std::size_t Polygon = 2;
-constexpr std::array PolygonOffsetEnableLUT = {
-    Point,   // Points
-    Line,    // Lines
-    Line,    // LineLoop
-    Line,    // LineStrip
-    Polygon, // Triangles
-    Polygon, // TriangleStrip
-    Polygon, // TriangleFan
-    Polygon, // Quads
-    Polygon, // QuadStrip
-    Polygon, // Polygon
-    Line,    // LinesAdjacency
-    Line,    // LineStripAdjacency
-    Polygon, // TrianglesAdjacency
-    Polygon, // TriangleStripAdjacency
-    Polygon, // Patches
+constexpr std::size_t POINT = 0;
+constexpr std::size_t LINE = 1;
+constexpr std::size_t POLYGON = 2;
+constexpr std::array POLYGON_OFFSET_ENABLE_LUT = {
+    POINT,   // Points
+    LINE,    // Lines
+    LINE,    // LineLoop
+    LINE,    // LineStrip
+    POLYGON, // Triangles
+    POLYGON, // TriangleStrip
+    POLYGON, // TriangleFan
+    POLYGON, // Quads
+    POLYGON, // QuadStrip
+    POLYGON, // Polygon
+    LINE,    // LinesAdjacency
+    LINE,    // LineStripAdjacency
+    POLYGON, // TrianglesAdjacency
+    POLYGON, // TriangleStripAdjacency
+    POLYGON, // Patches
 };

-constexpr FixedPipelineState::Rasterizer GetRasterizerState(const Maxwell& regs) {
-    const std::array enabled_lut = {regs.polygon_offset_point_enable,
-                                    regs.polygon_offset_line_enable,
-                                    regs.polygon_offset_fill_enable};
-    const auto topology = static_cast<std::size_t>(regs.draw.topology.Value());
-    const bool depth_bias_enabled = enabled_lut[PolygonOffsetEnableLUT[topology]];
-
-    const auto& clip = regs.view_volume_clip_control;
-    const bool depth_clamp_enabled = clip.depth_clamp_near == 1 || clip.depth_clamp_far == 1;
-
-    Maxwell::FrontFace front_face = regs.front_face;
-    if (regs.screen_y_control.triangle_rast_flip != 0 &&
-        regs.viewport_transform[0].scale_y > 0.0f) {
-        if (front_face == Maxwell::FrontFace::CounterClockWise)
-            front_face = Maxwell::FrontFace::ClockWise;
-        else if (front_face == Maxwell::FrontFace::ClockWise)
-            front_face = Maxwell::FrontFace::CounterClockWise;
-    }
-
-    const bool gl_ndc = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne;
-    return FixedPipelineState::Rasterizer(regs.cull_test_enabled, depth_bias_enabled,
-                                          depth_clamp_enabled, gl_ndc, regs.cull_face, front_face);
-}
-
 } // Anonymous namespace

-std::size_t FixedPipelineState::VertexBinding::Hash() const noexcept {
-    return (index << stride) ^ divisor;
-}
-
-bool FixedPipelineState::VertexBinding::operator==(const VertexBinding& rhs) const noexcept {
-    return std::tie(index, stride, divisor) == std::tie(rhs.index, rhs.stride, rhs.divisor);
-}
-
-std::size_t FixedPipelineState::VertexAttribute::Hash() const noexcept {
-    return static_cast<std::size_t>(index) ^ (static_cast<std::size_t>(buffer) << 13) ^
-           (static_cast<std::size_t>(type) << 22) ^ (static_cast<std::size_t>(size) << 31) ^
-           (static_cast<std::size_t>(offset) << 36);
-}
-
-bool FixedPipelineState::VertexAttribute::operator==(const VertexAttribute& rhs) const noexcept {
-    return std::tie(index, buffer, type, size, offset) ==
-           std::tie(rhs.index, rhs.buffer, rhs.type, rhs.size, rhs.offset);
-}
-
-std::size_t FixedPipelineState::StencilFace::Hash() const noexcept {
-    return static_cast<std::size_t>(action_stencil_fail) ^
-           (static_cast<std::size_t>(action_depth_fail) << 4) ^
-           (static_cast<std::size_t>(action_depth_fail) << 20) ^
-           (static_cast<std::size_t>(action_depth_pass) << 36);
-}
-
-bool FixedPipelineState::StencilFace::operator==(const StencilFace& rhs) const noexcept {
-    return std::tie(action_stencil_fail, action_depth_fail, action_depth_pass, test_func) ==
-           std::tie(rhs.action_stencil_fail, rhs.action_depth_fail, rhs.action_depth_pass,
-                    rhs.test_func);
-}
-
-std::size_t FixedPipelineState::BlendingAttachment::Hash() const noexcept {
-    return static_cast<std::size_t>(enable) ^ (static_cast<std::size_t>(rgb_equation) << 5) ^
-           (static_cast<std::size_t>(src_rgb_func) << 10) ^
-           (static_cast<std::size_t>(dst_rgb_func) << 15) ^
-           (static_cast<std::size_t>(a_equation) << 20) ^
-           (static_cast<std::size_t>(src_a_func) << 25) ^
-           (static_cast<std::size_t>(dst_a_func) << 30) ^
-           (static_cast<std::size_t>(components[0]) << 35) ^
-           (static_cast<std::size_t>(components[1]) << 36) ^
-           (static_cast<std::size_t>(components[2]) << 37) ^
-           (static_cast<std::size_t>(components[3]) << 38);
-}
-
-bool FixedPipelineState::BlendingAttachment::operator==(const BlendingAttachment& rhs) const
-    noexcept {
-    return std::tie(enable, rgb_equation, src_rgb_func, dst_rgb_func, a_equation, src_a_func,
-                    dst_a_func, components) ==
-           std::tie(rhs.enable, rhs.rgb_equation, rhs.src_rgb_func, rhs.dst_rgb_func,
-                    rhs.a_equation, rhs.src_a_func, rhs.dst_a_func, rhs.components);
-}
-
-std::size_t FixedPipelineState::VertexInput::Hash() const noexcept {
-    std::size_t hash = num_bindings ^ (num_attributes << 32);
-    for (std::size_t i = 0; i < num_bindings; ++i) {
-        boost::hash_combine(hash, bindings[i].Hash());
+void FixedPipelineState::DepthStencil::Fill(const Maxwell& regs) noexcept {
+    raw = 0;
+    front.action_stencil_fail.Assign(PackStencilOp(regs.stencil_front_op_fail));
+    front.action_depth_fail.Assign(PackStencilOp(regs.stencil_front_op_zfail));
+    front.action_depth_pass.Assign(PackStencilOp(regs.stencil_front_op_zpass));
+    front.test_func.Assign(PackComparisonOp(regs.stencil_front_func_func));
+    if (regs.stencil_two_side_enable) {
+        back.action_stencil_fail.Assign(PackStencilOp(regs.stencil_back_op_fail));
+        back.action_depth_fail.Assign(PackStencilOp(regs.stencil_back_op_zfail));
+        back.action_depth_pass.Assign(PackStencilOp(regs.stencil_back_op_zpass));
+        back.test_func.Assign(PackComparisonOp(regs.stencil_back_func_func));
+    } else {
+        back.action_stencil_fail.Assign(front.action_stencil_fail);
+        back.action_depth_fail.Assign(front.action_depth_fail);
+        back.action_depth_pass.Assign(front.action_depth_pass);
+        back.test_func.Assign(front.test_func);
    }
-    for (std::size_t i = 0; i < num_attributes; ++i) {
-        boost::hash_combine(hash, attributes[i].Hash());
+    depth_test_enable.Assign(regs.depth_test_enable);
+    depth_write_enable.Assign(regs.depth_write_enabled);
+    depth_bounds_enable.Assign(regs.depth_bounds_enable);
+    stencil_enable.Assign(regs.stencil_enable);
+    depth_test_func.Assign(PackComparisonOp(regs.depth_test_func));
+}
+
+void FixedPipelineState::Rasterizer::Fill(const Maxwell& regs) noexcept {
+    const auto& clip = regs.view_volume_clip_control;
+    const std::array enabled_lut = {regs.polygon_offset_point_enable,
+                                    regs.polygon_offset_line_enable,
+                                    regs.polygon_offset_fill_enable};
+    const u32 topology_index = static_cast<u32>(regs.draw.topology.Value());
+
+    u32 packed_front_face = PackFrontFace(regs.front_face);
+    if (regs.screen_y_control.triangle_rast_flip != 0 &&
+        regs.viewport_transform[0].scale_y > 0.0f) {
+        // Flip front face
+        packed_front_face = 1 - packed_front_face;
    }
-    return hash;
+
+    raw = 0;
+    topology.Assign(topology_index);
+    primitive_restart_enable.Assign(regs.primitive_restart.enabled != 0 ? 1 : 0);
+    cull_enable.Assign(regs.cull_test_enabled != 0 ? 1 : 0);
+    depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0);
+    depth_clamp_enable.Assign(clip.depth_clamp_near == 1 || clip.depth_clamp_far == 1 ? 1 : 0);
+    ndc_minus_one_to_one.Assign(regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1 : 0);
+    cull_face.Assign(PackCullFace(regs.cull_face));
+    front_face.Assign(packed_front_face);
+    polygon_mode.Assign(PackPolygonMode(regs.polygon_mode_front));
+    patch_control_points_minus_one.Assign(regs.patch_vertices - 1);
+    tessellation_primitive.Assign(static_cast<u32>(regs.tess_mode.prim.Value()));
+    tessellation_spacing.Assign(static_cast<u32>(regs.tess_mode.spacing.Value()));
+    tessellation_clockwise.Assign(regs.tess_mode.cw.Value());
+    logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0);
+    logic_op.Assign(PackLogicOp(regs.logic_op.operation));
+    std::memcpy(&point_size, &regs.point_size, sizeof(point_size)); // TODO: C++20 std::bit_cast
 }

-bool FixedPipelineState::VertexInput::operator==(const VertexInput& rhs) const noexcept {
-    return std::equal(bindings.begin(), bindings.begin() + num_bindings, rhs.bindings.begin(),
-                      rhs.bindings.begin() + rhs.num_bindings) &&
-           std::equal(attributes.begin(), attributes.begin() + num_attributes,
-                      rhs.attributes.begin(), rhs.attributes.begin() + rhs.num_attributes);
-}
-
-std::size_t FixedPipelineState::InputAssembly::Hash() const noexcept {
-    std::size_t point_size_int = 0;
-    std::memcpy(&point_size_int, &point_size, sizeof(point_size));
-    return (static_cast<std::size_t>(topology) << 24) ^ (point_size_int << 32) ^
-           static_cast<std::size_t>(primitive_restart_enable);
-}
-
-bool FixedPipelineState::InputAssembly::operator==(const InputAssembly& rhs) const noexcept {
-    return std::tie(topology, primitive_restart_enable, point_size) ==
-           std::tie(rhs.topology, rhs.primitive_restart_enable, rhs.point_size);
-}
-
-std::size_t FixedPipelineState::Tessellation::Hash() const noexcept {
-    return static_cast<std::size_t>(patch_control_points) ^
-           (static_cast<std::size_t>(primitive) << 6) ^ (static_cast<std::size_t>(spacing) << 8) ^
-           (static_cast<std::size_t>(clockwise) << 10);
-}
-
-bool FixedPipelineState::Tessellation::operator==(const Tessellation& rhs) const noexcept {
-    return std::tie(patch_control_points, primitive, spacing, clockwise) ==
-           std::tie(rhs.patch_control_points, rhs.primitive, rhs.spacing, rhs.clockwise);
-}
-
-std::size_t FixedPipelineState::Rasterizer::Hash() const noexcept {
-    return static_cast<std::size_t>(cull_enable) ^
-           (static_cast<std::size_t>(depth_bias_enable) << 1) ^
-           (static_cast<std::size_t>(depth_clamp_enable) << 2) ^
-           (static_cast<std::size_t>(ndc_minus_one_to_one) << 3) ^
-           (static_cast<std::size_t>(cull_face) << 24) ^
-           (static_cast<std::size_t>(front_face) << 48);
-}
-
-bool FixedPipelineState::Rasterizer::operator==(const Rasterizer& rhs) const noexcept {
-    return std::tie(cull_enable, depth_bias_enable, depth_clamp_enable, ndc_minus_one_to_one,
-                    cull_face, front_face) ==
-           std::tie(rhs.cull_enable, rhs.depth_bias_enable, rhs.depth_clamp_enable,
-                    rhs.ndc_minus_one_to_one, rhs.cull_face, rhs.front_face);
-}
-
-std::size_t FixedPipelineState::DepthStencil::Hash() const noexcept {
-    std::size_t hash = static_cast<std::size_t>(depth_test_enable) ^
-                       (static_cast<std::size_t>(depth_write_enable) << 1) ^
-                       (static_cast<std::size_t>(depth_bounds_enable) << 2) ^
-                       (static_cast<std::size_t>(stencil_enable) << 3) ^
-                       (static_cast<std::size_t>(depth_test_function) << 4);
-    boost::hash_combine(hash, front_stencil.Hash());
-    boost::hash_combine(hash, back_stencil.Hash());
-    return hash;
-}
-
-bool FixedPipelineState::DepthStencil::operator==(const DepthStencil& rhs) const noexcept {
-    return std::tie(depth_test_enable, depth_write_enable, depth_bounds_enable, depth_test_function,
-                    stencil_enable, front_stencil, back_stencil) ==
-           std::tie(rhs.depth_test_enable, rhs.depth_write_enable, rhs.depth_bounds_enable,
-                    rhs.depth_test_function, rhs.stencil_enable, rhs.front_stencil,
-                    rhs.back_stencil);
-}
-
-std::size_t FixedPipelineState::ColorBlending::Hash() const noexcept {
-    std::size_t hash = attachments_count << 13;
-    for (std::size_t rt = 0; rt < static_cast<std::size_t>(attachments_count); ++rt) {
-        boost::hash_combine(hash, attachments[rt].Hash());
+void FixedPipelineState::ColorBlending::Fill(const Maxwell& regs) noexcept {
+    for (std::size_t index = 0; index < std::size(attachments); ++index) {
+        attachments[index].Fill(regs, index);
    }
-    return hash;
 }

-bool FixedPipelineState::ColorBlending::operator==(const ColorBlending& rhs) const noexcept {
-    return std::equal(attachments.begin(), attachments.begin() + attachments_count,
-                      rhs.attachments.begin(), rhs.attachments.begin() + rhs.attachments_count);
+void FixedPipelineState::BlendingAttachment::Fill(const Maxwell& regs, std::size_t index) {
+    const auto& mask = regs.color_mask[regs.color_mask_common ? 0 : index];
+
+    raw = 0;
+    mask_r.Assign(mask.R);
+    mask_g.Assign(mask.G);
+    mask_b.Assign(mask.B);
+    mask_a.Assign(mask.A);
+
+    // TODO: C++20 Use templated lambda to deduplicate code
+
+    if (!regs.independent_blend_enable) {
+        const auto& src = regs.blend;
+        if (!src.enable[index]) {
+            return;
+        }
+        equation_rgb.Assign(PackBlendEquation(src.equation_rgb));
+        equation_a.Assign(PackBlendEquation(src.equation_a));
+        factor_source_rgb.Assign(PackBlendFactor(src.factor_source_rgb));
+        factor_dest_rgb.Assign(PackBlendFactor(src.factor_dest_rgb));
+        factor_source_a.Assign(PackBlendFactor(src.factor_source_a));
+        factor_dest_a.Assign(PackBlendFactor(src.factor_dest_a));
+        enable.Assign(1);
+        return;
+    }
+
+    if (!regs.blend.enable[index]) {
+        return;
+    }
+    const auto& src = regs.independent_blend[index];
+    equation_rgb.Assign(PackBlendEquation(src.equation_rgb));
+    equation_a.Assign(PackBlendEquation(src.equation_a));
+    factor_source_rgb.Assign(PackBlendFactor(src.factor_source_rgb));
+    factor_dest_rgb.Assign(PackBlendFactor(src.factor_dest_rgb));
+    factor_source_a.Assign(PackBlendFactor(src.factor_source_a));
+    factor_dest_a.Assign(PackBlendFactor(src.factor_dest_a));
+    enable.Assign(1);
 }

 std::size_t FixedPipelineState::Hash() const noexcept {
-    std::size_t hash = 0;
-    boost::hash_combine(hash, vertex_input.Hash());
-    boost::hash_combine(hash, input_assembly.Hash());
-    boost::hash_combine(hash, tessellation.Hash());
-    boost::hash_combine(hash, rasterizer.Hash());
-    boost::hash_combine(hash, depth_stencil.Hash());
-    boost::hash_combine(hash, color_blending.Hash());
-    return hash;
+    const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
+    return static_cast<std::size_t>(hash);
 }

 bool FixedPipelineState::operator==(const FixedPipelineState& rhs) const noexcept {
-    return std::tie(vertex_input, input_assembly, tessellation, rasterizer, depth_stencil,
-                    color_blending) == std::tie(rhs.vertex_input, rhs.input_assembly,
-                                                rhs.tessellation, rhs.rasterizer, rhs.depth_stencil,
-                                                rhs.color_blending);
+    return std::memcmp(this, &rhs, sizeof *this) == 0;
 }

 FixedPipelineState GetFixedPipelineState(const Maxwell& regs) {
    FixedPipelineState fixed_state;
-    fixed_state.input_assembly = GetInputAssemblyState(regs);
-    fixed_state.tessellation = GetTessellationState(regs);
-    fixed_state.rasterizer = GetRasterizerState(regs);
-    fixed_state.depth_stencil = GetDepthStencilState(regs);
-    fixed_state.color_blending = GetColorBlendingState(regs);
+    fixed_state.rasterizer.Fill(regs);
+    fixed_state.depth_stencil.Fill(regs);
+    fixed_state.color_blending.Fill(regs);
+    fixed_state.padding = {};
    return fixed_state;
 }

+u32 FixedPipelineState::PackComparisonOp(Maxwell::ComparisonOp op) noexcept {
+    // OpenGL enums go from 0x200 to 0x207 and the others from 1 to 8
+    // If we substract 0x200 to OpenGL enums and 1 to the others we get a 0-7 range.
+    // Perfect for a hash.
+    const u32 value = static_cast<u32>(op);
+    return value - (value >= 0x200 ? 0x200 : 1);
+}
+
+Maxwell::ComparisonOp FixedPipelineState::UnpackComparisonOp(u32 packed) noexcept {
+    // Read PackComparisonOp for the logic behind this.
+    return static_cast<Maxwell::ComparisonOp>(packed + 1);
+}
+
+u32 FixedPipelineState::PackStencilOp(Maxwell::StencilOp op) noexcept {
+    switch (op) {
+    case Maxwell::StencilOp::Keep:
+    case Maxwell::StencilOp::KeepOGL:
+        return 0;
+    case Maxwell::StencilOp::Zero:
+    case Maxwell::StencilOp::ZeroOGL:
+        return 1;
+    case Maxwell::StencilOp::Replace:
+    case Maxwell::StencilOp::ReplaceOGL:
+        return 2;
+    case Maxwell::StencilOp::Incr:
+    case Maxwell::StencilOp::IncrOGL:
+        return 3;
+    case Maxwell::StencilOp::Decr:
+    case Maxwell::StencilOp::DecrOGL:
+        return 4;
+    case Maxwell::StencilOp::Invert:
+    case Maxwell::StencilOp::InvertOGL:
+        return 5;
+    case Maxwell::StencilOp::IncrWrap:
+    case Maxwell::StencilOp::IncrWrapOGL:
+        return 6;
+    case Maxwell::StencilOp::DecrWrap:
+    case Maxwell::StencilOp::DecrWrapOGL:
+        return 7;
+    }
+    return 0;
+}
+
+Maxwell::StencilOp FixedPipelineState::UnpackStencilOp(u32 packed) noexcept {
+    static constexpr std::array LUT = {Maxwell::StencilOp::Keep,     Maxwell::StencilOp::Zero,
+                                       Maxwell::StencilOp::Replace,  Maxwell::StencilOp::Incr,
+                                       Maxwell::StencilOp::Decr,     Maxwell::StencilOp::Invert,
+                                       Maxwell::StencilOp::IncrWrap, Maxwell::StencilOp::DecrWrap};
+    return LUT[packed];
+}
+
+u32 FixedPipelineState::PackCullFace(Maxwell::CullFace cull) noexcept {
+    // FrontAndBack is 0x408, by substracting 0x406 in it we get 2.
+    // Individual cull faces are in 0x404 and 0x405, substracting 0x404 we get 0 and 1.
+    const u32 value = static_cast<u32>(cull);
+    return value - (value == 0x408 ? 0x406 : 0x404);
+}
+
+Maxwell::CullFace FixedPipelineState::UnpackCullFace(u32 packed) noexcept {
+    static constexpr std::array LUT = {Maxwell::CullFace::Front, Maxwell::CullFace::Back,
+                                       Maxwell::CullFace::FrontAndBack};
+    return LUT[packed];
+}
+
+u32 FixedPipelineState::PackFrontFace(Maxwell::FrontFace face) noexcept {
+    return static_cast<u32>(face) - 0x900;
+}
+
+Maxwell::FrontFace FixedPipelineState::UnpackFrontFace(u32 packed) noexcept {
+    return static_cast<Maxwell::FrontFace>(packed + 0x900);
+}
+
+u32 FixedPipelineState::PackPolygonMode(Maxwell::PolygonMode mode) noexcept {
+    return static_cast<u32>(mode) - 0x1B00;
+}
+
+Maxwell::PolygonMode FixedPipelineState::UnpackPolygonMode(u32 packed) noexcept {
+    return static_cast<Maxwell::PolygonMode>(packed + 0x1B00);
+}
+
+u32 FixedPipelineState::PackLogicOp(Maxwell::LogicOperation op) noexcept {
+    return static_cast<u32>(op) - 0x1500;
+}
+
+Maxwell::LogicOperation FixedPipelineState::UnpackLogicOp(u32 packed) noexcept {
+    return static_cast<Maxwell::LogicOperation>(packed + 0x1500);
+}
+
+u32 FixedPipelineState::PackBlendEquation(Maxwell::Blend::Equation equation) noexcept {
+    switch (equation) {
+    case Maxwell::Blend::Equation::Add:
+    case Maxwell::Blend::Equation::AddGL:
+        return 0;
+    case Maxwell::Blend::Equation::Subtract:
+    case Maxwell::Blend::Equation::SubtractGL:
+        return 1;
+    case Maxwell::Blend::Equation::ReverseSubtract:
+    case Maxwell::Blend::Equation::ReverseSubtractGL:
+        return 2;
+    case Maxwell::Blend::Equation::Min:
+    case Maxwell::Blend::Equation::MinGL:
+        return 3;
+    case Maxwell::Blend::Equation::Max:
+    case Maxwell::Blend::Equation::MaxGL:
+        return 4;
+    }
+    return 0;
+}
+
+Maxwell::Blend::Equation FixedPipelineState::UnpackBlendEquation(u32 packed) noexcept {
+    static constexpr std::array LUT = {
+        Maxwell::Blend::Equation::Add, Maxwell::Blend::Equation::Subtract,
+        Maxwell::Blend::Equation::ReverseSubtract, Maxwell::Blend::Equation::Min,
+        Maxwell::Blend::Equation::Max};
+    return LUT[packed];
+}
+
+u32 FixedPipelineState::PackBlendFactor(Maxwell::Blend::Factor factor) noexcept {
+    switch (factor) {
+    case Maxwell::Blend::Factor::Zero:
+    case Maxwell::Blend::Factor::ZeroGL:
+        return 0;
+    case Maxwell::Blend::Factor::One:
+    case Maxwell::Blend::Factor::OneGL:
+        return 1;
+    case Maxwell::Blend::Factor::SourceColor:
+    case Maxwell::Blend::Factor::SourceColorGL:
+        return 2;
+    case Maxwell::Blend::Factor::OneMinusSourceColor:
+    case Maxwell::Blend::Factor::OneMinusSourceColorGL:
+        return 3;
+    case Maxwell::Blend::Factor::SourceAlpha:
+    case Maxwell::Blend::Factor::SourceAlphaGL:
+        return 4;
+    case Maxwell::Blend::Factor::OneMinusSourceAlpha:
+    case Maxwell::Blend::Factor::OneMinusSourceAlphaGL:
+        return 5;
+    case Maxwell::Blend::Factor::DestAlpha:
+    case Maxwell::Blend::Factor::DestAlphaGL:
+        return 6;
+    case Maxwell::Blend::Factor::OneMinusDestAlpha:
+    case Maxwell::Blend::Factor::OneMinusDestAlphaGL:
+        return 7;
+    case Maxwell::Blend::Factor::DestColor:
+    case Maxwell::Blend::Factor::DestColorGL:
+        return 8;
+    case Maxwell::Blend::Factor::OneMinusDestColor:
+    case Maxwell::Blend::Factor::OneMinusDestColorGL:
+        return 9;
+    case Maxwell::Blend::Factor::SourceAlphaSaturate:
+    case Maxwell::Blend::Factor::SourceAlphaSaturateGL:
+        return 10;
+    case Maxwell::Blend::Factor::Source1Color:
+    case Maxwell::Blend::Factor::Source1ColorGL:
+        return 11;
+    case Maxwell::Blend::Factor::OneMinusSource1Color:
+    case Maxwell::Blend::Factor::OneMinusSource1ColorGL:
+        return 12;
+    case Maxwell::Blend::Factor::Source1Alpha:
+    case Maxwell::Blend::Factor::Source1AlphaGL:
+        return 13;
+    case Maxwell::Blend::Factor::OneMinusSource1Alpha:
+    case Maxwell::Blend::Factor::OneMinusSource1AlphaGL:
+        return 14;
+    case Maxwell::Blend::Factor::ConstantColor:
+    case Maxwell::Blend::Factor::ConstantColorGL:
+        return 15;
+    case Maxwell::Blend::Factor::OneMinusConstantColor:
+    case Maxwell::Blend::Factor::OneMinusConstantColorGL:
+        return 16;
+    case Maxwell::Blend::Factor::ConstantAlpha:
+    case Maxwell::Blend::Factor::ConstantAlphaGL:
+        return 17;
+    case Maxwell::Blend::Factor::OneMinusConstantAlpha:
+    case Maxwell::Blend::Factor::OneMinusConstantAlphaGL:
+        return 18;
+    }
+    return 0;
+}
+
+Maxwell::Blend::Factor FixedPipelineState::UnpackBlendFactor(u32 packed) noexcept {
+    static constexpr std::array LUT = {
+        Maxwell::Blend::Factor::Zero,
+        Maxwell::Blend::Factor::One,
+        Maxwell::Blend::Factor::SourceColor,
+        Maxwell::Blend::Factor::OneMinusSourceColor,
+        Maxwell::Blend::Factor::SourceAlpha,
+        Maxwell::Blend::Factor::OneMinusSourceAlpha,
+        Maxwell::Blend::Factor::DestAlpha,
+        Maxwell::Blend::Factor::OneMinusDestAlpha,
+        Maxwell::Blend::Factor::DestColor,
+        Maxwell::Blend::Factor::OneMinusDestColor,
+        Maxwell::Blend::Factor::SourceAlphaSaturate,
+        Maxwell::Blend::Factor::Source1Color,
+        Maxwell::Blend::Factor::OneMinusSource1Color,
+        Maxwell::Blend::Factor::Source1Alpha,
+        Maxwell::Blend::Factor::OneMinusSource1Alpha,
+        Maxwell::Blend::Factor::ConstantColor,
+        Maxwell::Blend::Factor::OneMinusConstantColor,
+        Maxwell::Blend::Factor::ConstantAlpha,
+        Maxwell::Blend::Factor::OneMinusConstantAlpha,
+    };
+    return LUT[packed];
+}
+
 } // namespace Vulkan
@@ -7,6 +7,7 @@
 #include <array>
 #include <type_traits>

+#include "common/bit_field.h"
 #include "common/common_types.h"

 #include "video_core/engines/maxwell_3d.h"
@@ -16,93 +17,48 @@ namespace Vulkan {

 using Maxwell = Tegra::Engines::Maxwell3D::Regs;

-// TODO(Rodrigo): Optimize this structure.
+struct alignas(32) FixedPipelineState {
+    static u32 PackComparisonOp(Maxwell::ComparisonOp op) noexcept;
+    static Maxwell::ComparisonOp UnpackComparisonOp(u32 packed) noexcept;

-struct FixedPipelineState {
-    using PixelFormat = VideoCore::Surface::PixelFormat;
+    static u32 PackStencilOp(Maxwell::StencilOp op) noexcept;
+    static Maxwell::StencilOp UnpackStencilOp(u32 packed) noexcept;

-    struct VertexBinding {
-        constexpr VertexBinding(u32 index, u32 stride, u32 divisor)
-            : index{index}, stride{stride}, divisor{divisor} {}
-        VertexBinding() = default;
+    static u32 PackCullFace(Maxwell::CullFace cull) noexcept;
+    static Maxwell::CullFace UnpackCullFace(u32 packed) noexcept;

-        u32 index;
-        u32 stride;
-        u32 divisor;
+    static u32 PackFrontFace(Maxwell::FrontFace face) noexcept;
+    static Maxwell::FrontFace UnpackFrontFace(u32 packed) noexcept;

-        std::size_t Hash() const noexcept;
+    static u32 PackPolygonMode(Maxwell::PolygonMode mode) noexcept;
+    static Maxwell::PolygonMode UnpackPolygonMode(u32 packed) noexcept;

-        bool operator==(const VertexBinding& rhs) const noexcept;
+    static u32 PackLogicOp(Maxwell::LogicOperation op) noexcept;
+    static Maxwell::LogicOperation UnpackLogicOp(u32 packed) noexcept;

-        bool operator!=(const VertexBinding& rhs) const noexcept {
-            return !operator==(rhs);
-        }
-    };
+    static u32 PackBlendEquation(Maxwell::Blend::Equation equation) noexcept;
+    static Maxwell::Blend::Equation UnpackBlendEquation(u32 packed) noexcept;

-    struct VertexAttribute {
-        constexpr VertexAttribute(u32 index, u32 buffer, Maxwell::VertexAttribute::Type type,
-                                  Maxwell::VertexAttribute::Size size, u32 offset)
-            : index{index}, buffer{buffer}, type{type}, size{size}, offset{offset} {}
-        VertexAttribute() = default;
-
-        u32 index;
-        u32 buffer;
-        Maxwell::VertexAttribute::Type type;
-        Maxwell::VertexAttribute::Size size;
-        u32 offset;
-
-        std::size_t Hash() const noexcept;
-
-        bool operator==(const VertexAttribute& rhs) const noexcept;
-
-        bool operator!=(const VertexAttribute& rhs) const noexcept {
-            return !operator==(rhs);
-        }
-    };
-
-    struct StencilFace {
-        constexpr StencilFace(Maxwell::StencilOp action_stencil_fail,
-                              Maxwell::StencilOp action_depth_fail,
-                              Maxwell::StencilOp action_depth_pass, Maxwell::ComparisonOp test_func)
-            : action_stencil_fail{action_stencil_fail}, action_depth_fail{action_depth_fail},
-              action_depth_pass{action_depth_pass}, test_func{test_func} {}
-        StencilFace() = default;
-
-        Maxwell::StencilOp action_stencil_fail;
-        Maxwell::StencilOp action_depth_fail;
-        Maxwell::StencilOp action_depth_pass;
-        Maxwell::ComparisonOp test_func;
-
-        std::size_t Hash() const noexcept;
-
-        bool operator==(const StencilFace& rhs) const noexcept;
-
-        bool operator!=(const StencilFace& rhs) const noexcept {
-            return !operator==(rhs);
-        }
-    };
+    static u32 PackBlendFactor(Maxwell::Blend::Factor factor) noexcept;
+    static Maxwell::Blend::Factor UnpackBlendFactor(u32 packed) noexcept;

    struct BlendingAttachment {
-        constexpr BlendingAttachment(bool enable, Maxwell::Blend::Equation rgb_equation,
-                                     Maxwell::Blend::Factor src_rgb_func,
-                                     Maxwell::Blend::Factor dst_rgb_func,
-                                     Maxwell::Blend::Equation a_equation,
-                                     Maxwell::Blend::Factor src_a_func,
-                                     Maxwell::Blend::Factor dst_a_func,
-                                     std::array<bool, 4> components)
-            : enable{enable}, rgb_equation{rgb_equation}, src_rgb_func{src_rgb_func},
-              dst_rgb_func{dst_rgb_func}, a_equation{a_equation}, src_a_func{src_a_func},
-              dst_a_func{dst_a_func}, components{components} {}
-        BlendingAttachment() = default;
+        union {
+            u32 raw;
+            BitField<0, 1, u32> mask_r;
+            BitField<1, 1, u32> mask_g;
+            BitField<2, 1, u32> mask_b;
+            BitField<3, 1, u32> mask_a;
+            BitField<4, 3, u32> equation_rgb;
+            BitField<7, 3, u32> equation_a;
+            BitField<10, 5, u32> factor_source_rgb;
+            BitField<15, 5, u32> factor_dest_rgb;
+            BitField<20, 5, u32> factor_source_a;
+            BitField<25, 5, u32> factor_dest_a;
+            BitField<30, 1, u32> enable;
+        };

-        bool enable;
-        Maxwell::Blend::Equation rgb_equation;
-        Maxwell::Blend::Factor src_rgb_func;
-        Maxwell::Blend::Factor dst_rgb_func;
-        Maxwell::Blend::Equation a_equation;
-        Maxwell::Blend::Factor src_a_func;
-        Maxwell::Blend::Factor dst_a_func;
-        std::array<bool, 4> components;
+        void Fill(const Maxwell& regs, std::size_t index);

        std::size_t Hash() const noexcept;

@@ -111,136 +67,178 @@ struct FixedPipelineState {
        bool operator!=(const BlendingAttachment& rhs) const noexcept {
            return !operator==(rhs);
        }
+
+        constexpr std::array<bool, 4> Mask() const noexcept {
+            return {mask_r != 0, mask_g != 0, mask_b != 0, mask_a != 0};
+        }
+
+        Maxwell::Blend::Equation EquationRGB() const noexcept {
+            return UnpackBlendEquation(equation_rgb.Value());
+        }
+
+        Maxwell::Blend::Equation EquationAlpha() const noexcept {
+            return UnpackBlendEquation(equation_a.Value());
+        }
+
+        Maxwell::Blend::Factor SourceRGBFactor() const noexcept {
+            return UnpackBlendFactor(factor_source_rgb.Value());
+        }
+
+        Maxwell::Blend::Factor DestRGBFactor() const noexcept {
+            return UnpackBlendFactor(factor_dest_rgb.Value());
+        }
+
+        Maxwell::Blend::Factor SourceAlphaFactor() const noexcept {
+            return UnpackBlendFactor(factor_source_a.Value());
+        }
+
+        Maxwell::Blend::Factor DestAlphaFactor() const noexcept {
+            return UnpackBlendFactor(factor_dest_a.Value());
+        }
    };

    struct VertexInput {
-        std::size_t num_bindings = 0;
-        std::size_t num_attributes = 0;
-        std::array<VertexBinding, Maxwell::NumVertexArrays> bindings;
-        std::array<VertexAttribute, Maxwell::NumVertexAttributes> attributes;
+        union Binding {
+            u16 raw;
+            BitField<0, 1, u16> enabled;
+            BitField<1, 12, u16> stride;
+        };

-        std::size_t Hash() const noexcept;
+        union Attribute {
+            u32 raw;
+            BitField<0, 1, u32> enabled;
+            BitField<1, 5, u32> buffer;
+            BitField<6, 14, u32> offset;
+            BitField<20, 3, u32> type;
+            BitField<23, 6, u32> size;

-        bool operator==(const VertexInput& rhs) const noexcept;
+            constexpr Maxwell::VertexAttribute::Type Type() const noexcept {
+                return static_cast<Maxwell::VertexAttribute::Type>(type.Value());
+            }

-        bool operator!=(const VertexInput& rhs) const noexcept {
-            return !operator==(rhs);
+            constexpr Maxwell::VertexAttribute::Size Size() const noexcept {
+                return static_cast<Maxwell::VertexAttribute::Size>(size.Value());
+            }
+        };
+
+        std::array<Binding, Maxwell::NumVertexArrays> bindings;
+        std::array<u32, Maxwell::NumVertexArrays> binding_divisors;
+        std::array<Attribute, Maxwell::NumVertexAttributes> attributes;
+
+        void SetBinding(std::size_t index, bool enabled, u32 stride, u32 divisor) noexcept {
+            auto& binding = bindings[index];
+            binding.raw = 0;
+            binding.enabled.Assign(enabled ? 1 : 0);
+            binding.stride.Assign(stride);
+            binding_divisors[index] = divisor;
        }
-    };

-    struct InputAssembly {
-        constexpr InputAssembly(Maxwell::PrimitiveTopology topology, bool primitive_restart_enable,
-                                float point_size)
-            : topology{topology}, primitive_restart_enable{primitive_restart_enable},
-              point_size{point_size} {}
-        InputAssembly() = default;
-
-        Maxwell::PrimitiveTopology topology;
-        bool primitive_restart_enable;
-        float point_size;
-
-        std::size_t Hash() const noexcept;
-
-        bool operator==(const InputAssembly& rhs) const noexcept;
-
-        bool operator!=(const InputAssembly& rhs) const noexcept {
-            return !operator==(rhs);
-        }
-    };
-
-    struct Tessellation {
-        constexpr Tessellation(u32 patch_control_points, Maxwell::TessellationPrimitive primitive,
-                               Maxwell::TessellationSpacing spacing, bool clockwise)
-            : patch_control_points{patch_control_points}, primitive{primitive}, spacing{spacing},
-              clockwise{clockwise} {}
-        Tessellation() = default;
-
-        u32 patch_control_points;
-        Maxwell::TessellationPrimitive primitive;
-        Maxwell::TessellationSpacing spacing;
-        bool clockwise;
-
-        std::size_t Hash() const noexcept;
-
-        bool operator==(const Tessellation& rhs) const noexcept;
-
-        bool operator!=(const Tessellation& rhs) const noexcept {
-            return !operator==(rhs);
+        void SetAttribute(std::size_t index, bool enabled, u32 buffer, u32 offset,
+                          Maxwell::VertexAttribute::Type type,
+                          Maxwell::VertexAttribute::Size size) noexcept {
+            auto& attribute = attributes[index];
+            attribute.raw = 0;
+            attribute.enabled.Assign(enabled ? 1 : 0);
+            attribute.buffer.Assign(buffer);
+            attribute.offset.Assign(offset);
+            attribute.type.Assign(static_cast<u32>(type));
+            attribute.size.Assign(static_cast<u32>(size));
        }
    };

    struct Rasterizer {
-        constexpr Rasterizer(bool cull_enable, bool depth_bias_enable, bool depth_clamp_enable,
-                             bool ndc_minus_one_to_one, Maxwell::CullFace cull_face,
-                             Maxwell::FrontFace front_face)
-            : cull_enable{cull_enable}, depth_bias_enable{depth_bias_enable},
-              depth_clamp_enable{depth_clamp_enable}, ndc_minus_one_to_one{ndc_minus_one_to_one},
-              cull_face{cull_face}, front_face{front_face} {}
-        Rasterizer() = default;
+        union {
+            u32 raw;
+            BitField<0, 4, u32> topology;
+            BitField<4, 1, u32> primitive_restart_enable;
+            BitField<5, 1, u32> cull_enable;
+            BitField<6, 1, u32> depth_bias_enable;
+            BitField<7, 1, u32> depth_clamp_enable;
+            BitField<8, 1, u32> ndc_minus_one_to_one;
+            BitField<9, 2, u32> cull_face;
+            BitField<11, 1, u32> front_face;
+            BitField<12, 2, u32> polygon_mode;
+            BitField<14, 5, u32> patch_control_points_minus_one;
+            BitField<19, 2, u32> tessellation_primitive;
+            BitField<21, 2, u32> tessellation_spacing;
+            BitField<23, 1, u32> tessellation_clockwise;
+            BitField<24, 1, u32> logic_op_enable;
+            BitField<25, 4, u32> logic_op;
+        };

-        bool cull_enable;
-        bool depth_bias_enable;
-        bool depth_clamp_enable;
-        bool ndc_minus_one_to_one;
-        Maxwell::CullFace cull_face;
-        Maxwell::FrontFace front_face;
+        // TODO(Rodrigo): Move this to push constants
+        u32 point_size;

-        std::size_t Hash() const noexcept;
+        void Fill(const Maxwell& regs) noexcept;

-        bool operator==(const Rasterizer& rhs) const noexcept;
+        constexpr Maxwell::PrimitiveTopology Topology() const noexcept {
+            return static_cast<Maxwell::PrimitiveTopology>(topology.Value());
+        }

-        bool operator!=(const Rasterizer& rhs) const noexcept {
-            return !operator==(rhs);
+        Maxwell::CullFace CullFace() const noexcept {
+            return UnpackCullFace(cull_face.Value());
+        }
+
+        Maxwell::FrontFace FrontFace() const noexcept {
+            return UnpackFrontFace(front_face.Value());
        }
    };

    struct DepthStencil {
-        constexpr DepthStencil(bool depth_test_enable, bool depth_write_enable,
-                               bool depth_bounds_enable, bool stencil_enable,
-                               Maxwell::ComparisonOp depth_test_function, StencilFace front_stencil,
-                               StencilFace back_stencil)
-            : depth_test_enable{depth_test_enable}, depth_write_enable{depth_write_enable},
-              depth_bounds_enable{depth_bounds_enable}, stencil_enable{stencil_enable},
-              depth_test_function{depth_test_function}, front_stencil{front_stencil},
-              back_stencil{back_stencil} {}
-        DepthStencil() = default;
+        template <std::size_t Position>
+        union StencilFace {
+            BitField<Position + 0, 3, u32> action_stencil_fail;
+            BitField<Position + 3, 3, u32> action_depth_fail;
+            BitField<Position + 6, 3, u32> action_depth_pass;
+            BitField<Position + 9, 3, u32> test_func;

-        bool depth_test_enable;
-        bool depth_write_enable;
-        bool depth_bounds_enable;
-        bool stencil_enable;
-        Maxwell::ComparisonOp depth_test_function;
-        StencilFace front_stencil;
-        StencilFace back_stencil;
+            Maxwell::StencilOp ActionStencilFail() const noexcept {
+                return UnpackStencilOp(action_stencil_fail);
+            }

-        std::size_t Hash() const noexcept;
+            Maxwell::StencilOp ActionDepthFail() const noexcept {
+                return UnpackStencilOp(action_depth_fail);
+            }

-        bool operator==(const DepthStencil& rhs) const noexcept;
+            Maxwell::StencilOp ActionDepthPass() const noexcept {
+                return UnpackStencilOp(action_depth_pass);
+            }

-        bool operator!=(const DepthStencil& rhs) const noexcept {
-            return !operator==(rhs);
+            Maxwell::ComparisonOp TestFunc() const noexcept {
+                return UnpackComparisonOp(test_func);
+            }
+        };
+
+        union {
+            u32 raw;
+            StencilFace<0> front;
+            StencilFace<12> back;
+            BitField<24, 1, u32> depth_test_enable;
+            BitField<25, 1, u32> depth_write_enable;
+            BitField<26, 1, u32> depth_bounds_enable;
+            BitField<27, 1, u32> stencil_enable;
+            BitField<28, 3, u32> depth_test_func;
+        };
+
+        void Fill(const Maxwell& regs) noexcept;
+
+        Maxwell::ComparisonOp DepthTestFunc() const noexcept {
+            return UnpackComparisonOp(depth_test_func);
        }
    };

    struct ColorBlending {
-        constexpr ColorBlending(
-            std::array<float, 4> blend_constants, std::size_t attachments_count,
-            std::array<BlendingAttachment, Maxwell::NumRenderTargets> attachments)
-            : attachments_count{attachments_count}, attachments{attachments} {}
-        ColorBlending() = default;
-
-        std::size_t attachments_count;
        std::array<BlendingAttachment, Maxwell::NumRenderTargets> attachments;

-        std::size_t Hash() const noexcept;
-
-        bool operator==(const ColorBlending& rhs) const noexcept;
-
-        bool operator!=(const ColorBlending& rhs) const noexcept {
-            return !operator==(rhs);
-        }
+        void Fill(const Maxwell& regs) noexcept;
    };

+    VertexInput vertex_input;
+    Rasterizer rasterizer;
+    DepthStencil depth_stencil;
+    ColorBlending color_blending;
+    std::array<u8, 20> padding;
+
    std::size_t Hash() const noexcept;

    bool operator==(const FixedPipelineState& rhs) const noexcept;
@@ -248,25 +246,11 @@ struct FixedPipelineState {
    bool operator!=(const FixedPipelineState& rhs) const noexcept {
        return !operator==(rhs);
    }
-
-    VertexInput vertex_input;
-    InputAssembly input_assembly;
-    Tessellation tessellation;
-    Rasterizer rasterizer;
-    DepthStencil depth_stencil;
-    ColorBlending color_blending;
 };
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::VertexBinding>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::VertexAttribute>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::StencilFace>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::BlendingAttachment>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::VertexInput>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::InputAssembly>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::Tessellation>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::Rasterizer>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::DepthStencil>);
-static_assert(std::is_trivially_copyable_v<FixedPipelineState::ColorBlending>);
+static_assert(std::has_unique_object_representations_v<FixedPipelineState>);
 static_assert(std::is_trivially_copyable_v<FixedPipelineState>);
+static_assert(std::is_trivially_constructible_v<FixedPipelineState>);
+static_assert(sizeof(FixedPipelineState) % 32 == 0, "Size is not aligned");

 FixedPipelineState GetFixedPipelineState(const Maxwell& regs);

@@ -0,0 +1,220 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#ifdef HAS_NSIGHT_AFTERMATH
+
+#include <mutex>
+#include <string>
+#include <string_view>
+#include <utility>
+#include <vector>
+
+#include <fmt/format.h>
+
+#define VK_NO_PROTOTYPES
+#include <vulkan/vulkan.h>
+
+#include <GFSDK_Aftermath.h>
+#include <GFSDK_Aftermath_Defines.h>
+#include <GFSDK_Aftermath_GpuCrashDump.h>
+#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
+
+#include "common/common_paths.h"
+#include "common/common_types.h"
+#include "common/file_util.h"
+#include "common/logging/log.h"
+#include "common/scope_exit.h"
+
+#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
+
+namespace Vulkan {
+
+static constexpr char AFTERMATH_LIB_NAME[] = "GFSDK_Aftermath_Lib.x64.dll";
+
+NsightAftermathTracker::NsightAftermathTracker() = default;
+
+NsightAftermathTracker::~NsightAftermathTracker() {
+    if (initialized) {
+        (void)GFSDK_Aftermath_DisableGpuCrashDumps();
+    }
+}
+
+bool NsightAftermathTracker::Initialize() {
+    if (!dl.Open(AFTERMATH_LIB_NAME)) {
+        LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath DLL");
+        return false;
+    }
+
+    if (!dl.GetSymbol("GFSDK_Aftermath_DisableGpuCrashDumps",
+                      &GFSDK_Aftermath_DisableGpuCrashDumps) ||
+        !dl.GetSymbol("GFSDK_Aftermath_EnableGpuCrashDumps",
+                      &GFSDK_Aftermath_EnableGpuCrashDumps) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GetShaderDebugInfoIdentifier",
+                      &GFSDK_Aftermath_GetShaderDebugInfoIdentifier) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GetShaderHashSpirv", &GFSDK_Aftermath_GetShaderHashSpirv) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_CreateDecoder",
+                      &GFSDK_Aftermath_GpuCrashDump_CreateDecoder) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_DestroyDecoder",
+                      &GFSDK_Aftermath_GpuCrashDump_DestroyDecoder) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GenerateJSON",
+                      &GFSDK_Aftermath_GpuCrashDump_GenerateJSON) ||
+        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GetJSON",
+                      &GFSDK_Aftermath_GpuCrashDump_GetJSON)) {
+        LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath function pointers");
+        return false;
+    }
+
+    dump_dir = FileUtil::GetUserPath(FileUtil::UserPath::LogDir) + "gpucrash";
+
+    (void)FileUtil::DeleteDirRecursively(dump_dir);
+    if (!FileUtil::CreateDir(dump_dir)) {
+        LOG_ERROR(Render_Vulkan, "Failed to create Nsight Aftermath dump directory");
+        return false;
+    }
+
+    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_EnableGpuCrashDumps(
+            GFSDK_Aftermath_Version_API, GFSDK_Aftermath_GpuCrashDumpWatchedApiFlags_Vulkan,
+            GFSDK_Aftermath_GpuCrashDumpFeatureFlags_Default, GpuCrashDumpCallback,
+            ShaderDebugInfoCallback, CrashDumpDescriptionCallback, this))) {
+        LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_EnableGpuCrashDumps failed");
+        return false;
+    }
+
+    LOG_INFO(Render_Vulkan, "Nsight Aftermath dump directory is \"{}\"", dump_dir);
+
+    initialized = true;
+    return true;
+}
+
+void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const {
+    if (!initialized) {
+        return;
+    }
+
+    std::vector<u32> spirv_copy = spirv;
+    GFSDK_Aftermath_SpirvCode shader;
+    shader.pData = spirv_copy.data();
+    shader.size = static_cast<u32>(spirv_copy.size() * 4);
+
+    std::scoped_lock lock{mutex};
+
+    GFSDK_Aftermath_ShaderHash hash;
+    if (!GFSDK_Aftermath_SUCCEED(
+            GFSDK_Aftermath_GetShaderHashSpirv(GFSDK_Aftermath_Version_API, &shader, &hash))) {
+        LOG_ERROR(Render_Vulkan, "Failed to hash SPIR-V module");
+        return;
+    }
+
+    FileUtil::IOFile file(fmt::format("{}/source_{:016x}.spv", dump_dir, hash.hash), "wb");
+    if (!file.IsOpen()) {
+        LOG_ERROR(Render_Vulkan, "Failed to dump SPIR-V module with hash={:016x}", hash.hash);
+        return;
+    }
+    if (file.WriteArray(spirv.data(), spirv.size()) != spirv.size()) {
+        LOG_ERROR(Render_Vulkan, "Failed to write SPIR-V module with hash={:016x}", hash.hash);
+        return;
+    }
+}
+
+void NsightAftermathTracker::OnGpuCrashDumpCallback(const void* gpu_crash_dump,
+                                                    u32 gpu_crash_dump_size) {
+    std::scoped_lock lock{mutex};
+
+    LOG_CRITICAL(Render_Vulkan, "called");
+
+    GFSDK_Aftermath_GpuCrashDump_Decoder decoder;
+    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_CreateDecoder(
+            GFSDK_Aftermath_Version_API, gpu_crash_dump, gpu_crash_dump_size, &decoder))) {
+        LOG_ERROR(Render_Vulkan, "Failed to create decoder");
+        return;
+    }
+    SCOPE_EXIT({ GFSDK_Aftermath_GpuCrashDump_DestroyDecoder(decoder); });
+
+    u32 json_size = 0;
+    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_GenerateJSON(
+            decoder, GFSDK_Aftermath_GpuCrashDumpDecoderFlags_ALL_INFO,
+            GFSDK_Aftermath_GpuCrashDumpFormatterFlags_NONE, nullptr, nullptr, nullptr, nullptr,
+            this, &json_size))) {
+        LOG_ERROR(Render_Vulkan, "Failed to generate JSON");
+        return;
+    }
+    std::vector<char> json(json_size);
+    if (!GFSDK_Aftermath_SUCCEED(
+            GFSDK_Aftermath_GpuCrashDump_GetJSON(decoder, json_size, json.data()))) {
+        LOG_ERROR(Render_Vulkan, "Failed to query JSON");
+        return;
+    }
+
+    const std::string base_name = [this] {
+        const int id = dump_id++;
+        if (id == 0) {
+            return fmt::format("{}/crash.nv-gpudmp", dump_dir);
+        } else {
+            return fmt::format("{}/crash_{}.nv-gpudmp", dump_dir, id);
+        }
+    }();
+
+    std::string_view dump_view(static_cast<const char*>(gpu_crash_dump), gpu_crash_dump_size);
+    if (FileUtil::WriteStringToFile(false, base_name, dump_view) != gpu_crash_dump_size) {
+        LOG_ERROR(Render_Vulkan, "Failed to write dump file");
+        return;
+    }
+    const std::string_view json_view(json.data(), json.size());
+    if (FileUtil::WriteStringToFile(true, base_name + ".json", json_view) != json.size()) {
+        LOG_ERROR(Render_Vulkan, "Failed to write JSON");
+        return;
+    }
+}
+
+void NsightAftermathTracker::OnShaderDebugInfoCallback(const void* shader_debug_info,
+                                                       u32 shader_debug_info_size) {
+    std::scoped_lock lock{mutex};
+
+    GFSDK_Aftermath_ShaderDebugInfoIdentifier identifier;
+    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GetShaderDebugInfoIdentifier(
+            GFSDK_Aftermath_Version_API, shader_debug_info, shader_debug_info_size, &identifier))) {
+        LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_GetShaderDebugInfoIdentifier failed");
+        return;
+    }
+
+    const std::string path =
+        fmt::format("{}/shader_{:016x}{:016x}.nvdbg", dump_dir, identifier.id[0], identifier.id[1]);
+    FileUtil::IOFile file(path, "wb");
+    if (!file.IsOpen()) {
+        LOG_ERROR(Render_Vulkan, "Failed to create file {}", path);
+        return;
+    }
+    if (file.WriteBytes(static_cast<const u8*>(shader_debug_info), shader_debug_info_size) !=
+        shader_debug_info_size) {
+        LOG_ERROR(Render_Vulkan, "Failed to write file {}", path);
+        return;
+    }
+}
+
+void NsightAftermathTracker::OnCrashDumpDescriptionCallback(
+    PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description) {
+    add_description(GFSDK_Aftermath_GpuCrashDumpDescriptionKey_ApplicationName, "yuzu");
+}
+
+void NsightAftermathTracker::GpuCrashDumpCallback(const void* gpu_crash_dump,
+                                                  u32 gpu_crash_dump_size, void* user_data) {
+    static_cast<NsightAftermathTracker*>(user_data)->OnGpuCrashDumpCallback(gpu_crash_dump,
+                                                                            gpu_crash_dump_size);
+}
+
+void NsightAftermathTracker::ShaderDebugInfoCallback(const void* shader_debug_info,
+                                                     u32 shader_debug_info_size, void* user_data) {
+    static_cast<NsightAftermathTracker*>(user_data)->OnShaderDebugInfoCallback(
+        shader_debug_info, shader_debug_info_size);
+}
+
+void NsightAftermathTracker::CrashDumpDescriptionCallback(
+    PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data) {
+    static_cast<NsightAftermathTracker*>(user_data)->OnCrashDumpDescriptionCallback(
+        add_description);
+}
+
+} // namespace Vulkan
+
+#endif // HAS_NSIGHT_AFTERMATH
@@ -0,0 +1,87 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <mutex>
+#include <string>
+#include <vector>
+
+#define VK_NO_PROTOTYPES
+#include <vulkan/vulkan.h>
+
+#ifdef HAS_NSIGHT_AFTERMATH
+#include <GFSDK_Aftermath_Defines.h>
+#include <GFSDK_Aftermath_GpuCrashDump.h>
+#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
+#endif
+
+#include "common/common_types.h"
+#include "common/dynamic_library.h"
+
+namespace Vulkan {
+
+class NsightAftermathTracker {
+public:
+    NsightAftermathTracker();
+    ~NsightAftermathTracker();
+
+    NsightAftermathTracker(const NsightAftermathTracker&) = delete;
+    NsightAftermathTracker& operator=(const NsightAftermathTracker&) = delete;
+
+    // Delete move semantics because Aftermath initialization uses a pointer to this.
+    NsightAftermathTracker(NsightAftermathTracker&&) = delete;
+    NsightAftermathTracker& operator=(NsightAftermathTracker&&) = delete;
+
+    bool Initialize();
+
+    void SaveShader(const std::vector<u32>& spirv) const;
+
+private:
+#ifdef HAS_NSIGHT_AFTERMATH
+    static void GpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size,
+                                     void* user_data);
+
+    static void ShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size,
+                                        void* user_data);
+
+    static void CrashDumpDescriptionCallback(
+        PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data);
+
+    void OnGpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size);
+
+    void OnShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size);
+
+    void OnCrashDumpDescriptionCallback(
+        PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description);
+
+    mutable std::mutex mutex;
+
+    std::string dump_dir;
+    int dump_id = 0;
+
+    bool initialized = false;
+
+    Common::DynamicLibrary dl;
+    PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps;
+    PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps;
+    PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier;
+    PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv;
+    PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder;
+    PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder;
+    PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON;
+    PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON;
+#endif
+};
+
+#ifndef HAS_NSIGHT_AFTERMATH
+inline NsightAftermathTracker::NsightAftermathTracker() = default;
+inline NsightAftermathTracker::~NsightAftermathTracker() = default;
+inline bool NsightAftermathTracker::Initialize() {
+    return false;
+}
+inline void NsightAftermathTracker::SaveShader(const std::vector<u32>&) const {}
+#endif
+
+} // namespace Vulkan
@@ -105,6 +105,8 @@ vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplat
 }

 vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
+    device.SaveShader(code);
+
    VkShaderModuleCreateInfo ci;
    ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
    ci.pNext = nullptr;
@@ -9,6 +9,7 @@
 #include <string_view>
 #include <thread>
 #include <unordered_set>
+#include <utility>
 #include <vector>

 #include "common/assert.h"
@@ -167,6 +168,7 @@ bool VKDevice::Create() {
    VkPhysicalDeviceFeatures2 features2;
    features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
    features2.pNext = nullptr;
+    const void* first_next = &features2;
    void** next = &features2.pNext;

    auto& features = features2.features;
@@ -296,7 +298,19 @@ bool VKDevice::Create() {
        LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
    }

-    logical = vk::Device::Create(physical, queue_cis, extensions, features2, dld);
+    VkDeviceDiagnosticsConfigCreateInfoNV diagnostics_nv;
+    if (nv_device_diagnostics_config) {
+        nsight_aftermath_tracker.Initialize();
+
+        diagnostics_nv.sType = VK_STRUCTURE_TYPE_DEVICE_DIAGNOSTICS_CONFIG_CREATE_INFO_NV;
+        diagnostics_nv.pNext = &features2;
+        diagnostics_nv.flags = VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_SHADER_DEBUG_INFO_BIT_NV |
+                               VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_RESOURCE_TRACKING_BIT_NV |
+                               VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_AUTOMATIC_CHECKPOINTS_BIT_NV;
+        first_next = &diagnostics_nv;
+    }
+
+    logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld);
    if (!logical) {
        LOG_ERROR(Render_Vulkan, "Failed to create logical device");
        return false;
@@ -344,17 +358,12 @@ VkFormat VKDevice::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFla
 void VKDevice::ReportLoss() const {
    LOG_CRITICAL(Render_Vulkan, "Device loss occured!");

-    // Wait some time to let the log flush
-    std::this_thread::sleep_for(std::chrono::seconds{1});
+    // Wait for the log to flush and for Nsight Aftermath to dump the results
+    std::this_thread::sleep_for(std::chrono::seconds{3});
+}

-    if (!nv_device_diagnostic_checkpoints) {
-        return;
-    }
-
-    [[maybe_unused]] const std::vector data = graphics_queue.GetCheckpointDataNV(dld);
-    // Catch here in debug builds (or with optimizations disabled) the last graphics pipeline to be
-    // executed. It can be done on a debugger by evaluating the expression:
-    // *(VKGraphicsPipeline*)data[0]
+void VKDevice::SaveShader(const std::vector<u32>& spirv) const {
+    nsight_aftermath_tracker.SaveShader(spirv);
 }

 bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const {
@@ -527,8 +536,8 @@ std::vector<const char*> VKDevice::LoadExtensions() {
        Test(extension, has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME,
             false);
        if (Settings::values.renderer_debug) {
-            Test(extension, nv_device_diagnostic_checkpoints,
-                 VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true);
+            Test(extension, nv_device_diagnostics_config,
+                 VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true);
        }
    }

@@ -10,6 +10,7 @@
 #include <vector>

 #include "common/common_types.h"
+#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
 #include "video_core/renderer_vulkan/wrapper.h"

 namespace Vulkan {
@@ -43,6 +44,9 @@ public:
    /// Reports a device loss.
    void ReportLoss() const;

+    /// Reports a shader to Nsight Aftermath.
+    void SaveShader(const std::vector<u32>& spirv) const;
+
    /// Returns the dispatch loader with direct function pointers of the device.
    const vk::DeviceDispatch& GetDispatchLoader() const {
        return dld;
@@ -173,11 +177,6 @@ public:
        return ext_transform_feedback;
    }

-    /// Returns true if the device supports VK_NV_device_diagnostic_checkpoints.
-    bool IsNvDeviceDiagnosticCheckpoints() const {
-        return nv_device_diagnostic_checkpoints;
-    }
-
    /// Returns the vendor name reported from Vulkan.
    std::string_view GetVendorName() const {
        return vendor_name;
@@ -233,7 +232,7 @@ private:
    bool ext_depth_range_unrestricted{};       ///< Support for VK_EXT_depth_range_unrestricted.
    bool ext_shader_viewport_index_layer{};    ///< Support for VK_EXT_shader_viewport_index_layer.
    bool ext_transform_feedback{};             ///< Support for VK_EXT_transform_feedback.
-    bool nv_device_diagnostic_checkpoints{};   ///< Support for VK_NV_device_diagnostic_checkpoints.
+    bool nv_device_diagnostics_config{};       ///< Support for VK_NV_device_diagnostics_config.

    // Telemetry parameters
    std::string vendor_name;                      ///< Device's driver name.
@@ -241,6 +240,9 @@ private:

    /// Format properties dictionary.
    std::unordered_map<VkFormat, VkFormatProperties> format_properties;
+
+    /// Nsight Aftermath GPU crash tracker
+    NsightAftermathTracker nsight_aftermath_tracker;
 };

 } // namespace Vulkan
@@ -0,0 +1,101 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <thread>
+
+#include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_fence_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
+#include "video_core/renderer_vulkan/wrapper.h"
+
+namespace Vulkan {
+
+InnerFence::InnerFence(const VKDevice& device, VKScheduler& scheduler, u32 payload, bool is_stubbed)
+    : VideoCommon::FenceBase(payload, is_stubbed), device{device}, scheduler{scheduler} {}
+
+InnerFence::InnerFence(const VKDevice& device, VKScheduler& scheduler, GPUVAddr address,
+                       u32 payload, bool is_stubbed)
+    : VideoCommon::FenceBase(address, payload, is_stubbed), device{device}, scheduler{scheduler} {}
+
+InnerFence::~InnerFence() = default;
+
+void InnerFence::Queue() {
+    if (is_stubbed) {
+        return;
+    }
+    ASSERT(!event);
+
+    event = device.GetLogical().CreateEvent();
+    ticks = scheduler.Ticks();
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([event = *event](vk::CommandBuffer cmdbuf) {
+        cmdbuf.SetEvent(event, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT);
+    });
+}
+
+bool InnerFence::IsSignaled() const {
+    if (is_stubbed) {
+        return true;
+    }
+    ASSERT(event);
+    return IsEventSignalled();
+}
+
+void InnerFence::Wait() {
+    if (is_stubbed) {
+        return;
+    }
+    ASSERT(event);
+
+    if (ticks >= scheduler.Ticks()) {
+        scheduler.Flush();
+    }
+    while (!IsEventSignalled()) {
+        std::this_thread::yield();
+    }
+}
+
+bool InnerFence::IsEventSignalled() const {
+    switch (const VkResult result = event.GetStatus()) {
+    case VK_EVENT_SET:
+        return true;
+    case VK_EVENT_RESET:
+        return false;
+    default:
+        throw vk::Exception(result);
+    }
+}
+
+VKFenceManager::VKFenceManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                               const VKDevice& device, VKScheduler& scheduler,
+                               VKTextureCache& texture_cache, VKBufferCache& buffer_cache,
+                               VKQueryCache& query_cache)
+    : GenericFenceManager(system, rasterizer, texture_cache, buffer_cache, query_cache),
+      device{device}, scheduler{scheduler} {}
+
+Fence VKFenceManager::CreateFence(u32 value, bool is_stubbed) {
+    return std::make_shared<InnerFence>(device, scheduler, value, is_stubbed);
+}
+
+Fence VKFenceManager::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) {
+    return std::make_shared<InnerFence>(device, scheduler, addr, value, is_stubbed);
+}
+
+void VKFenceManager::QueueFence(Fence& fence) {
+    fence->Queue();
+}
+
+bool VKFenceManager::IsFenceSignaled(Fence& fence) const {
+    return fence->IsSignaled();
+}
+
+void VKFenceManager::WaitFence(Fence& fence) {
+    fence->Wait();
+}
+
+} // namespace Vulkan
@@ -0,0 +1,74 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+
+#include "video_core/fence_manager.h"
+#include "video_core/renderer_vulkan/wrapper.h"
+
+namespace Core {
+class System;
+}
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace Vulkan {
+
+class VKBufferCache;
+class VKDevice;
+class VKQueryCache;
+class VKScheduler;
+class VKTextureCache;
+
+class InnerFence : public VideoCommon::FenceBase {
+public:
+    explicit InnerFence(const VKDevice& device, VKScheduler& scheduler, u32 payload,
+                        bool is_stubbed);
+    explicit InnerFence(const VKDevice& device, VKScheduler& scheduler, GPUVAddr address,
+                        u32 payload, bool is_stubbed);
+    ~InnerFence();
+
+    void Queue();
+
+    bool IsSignaled() const;
+
+    void Wait();
+
+private:
+    bool IsEventSignalled() const;
+
+    const VKDevice& device;
+    VKScheduler& scheduler;
+    vk::Event event;
+    u64 ticks = 0;
+};
+using Fence = std::shared_ptr<InnerFence>;
+
+using GenericFenceManager =
+    VideoCommon::FenceManager<Fence, VKTextureCache, VKBufferCache, VKQueryCache>;
+
+class VKFenceManager final : public GenericFenceManager {
+public:
+    explicit VKFenceManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
+                            const VKDevice& device, VKScheduler& scheduler,
+                            VKTextureCache& texture_cache, VKBufferCache& buffer_cache,
+                            VKQueryCache& query_cache);
+
+protected:
+    Fence CreateFence(u32 value, bool is_stubbed) override;
+    Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override;
+    void QueueFence(Fence& fence) override;
+    bool IsFenceSignaled(Fence& fence) const override;
+    void WaitFence(Fence& fence) override;
+
+private:
+    const VKDevice& device;
+    VKScheduler& scheduler;
+};
+
+} // namespace Vulkan
@@ -26,12 +26,13 @@ MICROPROFILE_DECLARE(Vulkan_PipelineCache);

 namespace {

-VkStencilOpState GetStencilFaceState(const FixedPipelineState::StencilFace& face) {
+template <class StencilFace>
+VkStencilOpState GetStencilFaceState(const StencilFace& face) {
    VkStencilOpState state;
-    state.failOp = MaxwellToVK::StencilOp(face.action_stencil_fail);
-    state.passOp = MaxwellToVK::StencilOp(face.action_depth_pass);
-    state.depthFailOp = MaxwellToVK::StencilOp(face.action_depth_fail);
-    state.compareOp = MaxwellToVK::ComparisonOp(face.test_func);
+    state.failOp = MaxwellToVK::StencilOp(face.ActionStencilFail());
+    state.passOp = MaxwellToVK::StencilOp(face.ActionDepthPass());
+    state.depthFailOp = MaxwellToVK::StencilOp(face.ActionDepthFail());
+    state.compareOp = MaxwellToVK::ComparisonOp(face.TestFunc());
    state.compareMask = 0;
    state.writeMask = 0;
    state.reference = 0;
@@ -147,6 +148,8 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
            continue;
        }

+        device.SaveShader(stage->code);
+
        ci.codeSize = stage->code.size() * sizeof(u32);
        ci.pCode = stage->code.data();
        modules.push_back(device.GetLogical().CreateShaderModule(ci));
@@ -157,43 +160,47 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
 vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
                                                const SPIRVProgram& program) const {
    const auto& vi = fixed_state.vertex_input;
-    const auto& ia = fixed_state.input_assembly;
    const auto& ds = fixed_state.depth_stencil;
    const auto& cd = fixed_state.color_blending;
-    const auto& ts = fixed_state.tessellation;
    const auto& rs = fixed_state.rasterizer;

    std::vector<VkVertexInputBindingDescription> vertex_bindings;
    std::vector<VkVertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors;
-    for (std::size_t i = 0; i < vi.num_bindings; ++i) {
-        const auto& binding = vi.bindings[i];
-        const bool instanced = binding.divisor != 0;
+    for (std::size_t index = 0; index < std::size(vi.bindings); ++index) {
+        const auto& binding = vi.bindings[index];
+        if (!binding.enabled) {
+            continue;
+        }
+        const bool instanced = vi.binding_divisors[index] != 0;
        const auto rate = instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;

        auto& vertex_binding = vertex_bindings.emplace_back();
-        vertex_binding.binding = binding.index;
+        vertex_binding.binding = static_cast<u32>(index);
        vertex_binding.stride = binding.stride;
        vertex_binding.inputRate = rate;

        if (instanced) {
            auto& binding_divisor = vertex_binding_divisors.emplace_back();
-            binding_divisor.binding = binding.index;
-            binding_divisor.divisor = binding.divisor;
+            binding_divisor.binding = static_cast<u32>(index);
+            binding_divisor.divisor = vi.binding_divisors[index];
        }
    }

    std::vector<VkVertexInputAttributeDescription> vertex_attributes;
    const auto& input_attributes = program[0]->entries.attributes;
-    for (std::size_t i = 0; i < vi.num_attributes; ++i) {
-        const auto& attribute = vi.attributes[i];
-        if (input_attributes.find(attribute.index) == input_attributes.end()) {
+    for (std::size_t index = 0; index < std::size(vi.attributes); ++index) {
+        const auto& attribute = vi.attributes[index];
+        if (!attribute.enabled) {
+            continue;
+        }
+        if (input_attributes.find(static_cast<u32>(index)) == input_attributes.end()) {
            // Skip attributes not used by the vertex shaders.
            continue;
        }
        auto& vertex_attribute = vertex_attributes.emplace_back();
-        vertex_attribute.location = attribute.index;
+        vertex_attribute.location = static_cast<u32>(index);
        vertex_attribute.binding = attribute.buffer;
-        vertex_attribute.format = MaxwellToVK::VertexFormat(attribute.type, attribute.size);
+        vertex_attribute.format = MaxwellToVK::VertexFormat(attribute.Type(), attribute.Size());
        vertex_attribute.offset = attribute.offset;
    }

@@ -219,15 +226,15 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
    input_assembly_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
    input_assembly_ci.pNext = nullptr;
    input_assembly_ci.flags = 0;
-    input_assembly_ci.topology = MaxwellToVK::PrimitiveTopology(device, ia.topology);
+    input_assembly_ci.topology = MaxwellToVK::PrimitiveTopology(device, rs.Topology());
    input_assembly_ci.primitiveRestartEnable =
-        ia.primitive_restart_enable && SupportsPrimitiveRestart(input_assembly_ci.topology);
+        rs.primitive_restart_enable != 0 && SupportsPrimitiveRestart(input_assembly_ci.topology);

    VkPipelineTessellationStateCreateInfo tessellation_ci;
    tessellation_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO;
    tessellation_ci.pNext = nullptr;
    tessellation_ci.flags = 0;
-    tessellation_ci.patchControlPoints = ts.patch_control_points;
+    tessellation_ci.patchControlPoints = rs.patch_control_points_minus_one.Value() + 1;

    VkPipelineViewportStateCreateInfo viewport_ci;
    viewport_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO;
@@ -246,8 +253,8 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
    rasterization_ci.rasterizerDiscardEnable = VK_FALSE;
    rasterization_ci.polygonMode = VK_POLYGON_MODE_FILL;
    rasterization_ci.cullMode =
-        rs.cull_enable ? MaxwellToVK::CullFace(rs.cull_face) : VK_CULL_MODE_NONE;
-    rasterization_ci.frontFace = MaxwellToVK::FrontFace(rs.front_face);
+        rs.cull_enable ? MaxwellToVK::CullFace(rs.CullFace()) : VK_CULL_MODE_NONE;
+    rasterization_ci.frontFace = MaxwellToVK::FrontFace(rs.FrontFace());
    rasterization_ci.depthBiasEnable = rs.depth_bias_enable;
    rasterization_ci.depthBiasConstantFactor = 0.0f;
    rasterization_ci.depthBiasClamp = 0.0f;
@@ -271,40 +278,38 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
    depth_stencil_ci.flags = 0;
    depth_stencil_ci.depthTestEnable = ds.depth_test_enable;
    depth_stencil_ci.depthWriteEnable = ds.depth_write_enable;
-    depth_stencil_ci.depthCompareOp = ds.depth_test_enable
-                                          ? MaxwellToVK::ComparisonOp(ds.depth_test_function)
-                                          : VK_COMPARE_OP_ALWAYS;
+    depth_stencil_ci.depthCompareOp =
+        ds.depth_test_enable ? MaxwellToVK::ComparisonOp(ds.DepthTestFunc()) : VK_COMPARE_OP_ALWAYS;
    depth_stencil_ci.depthBoundsTestEnable = ds.depth_bounds_enable;
    depth_stencil_ci.stencilTestEnable = ds.stencil_enable;
-    depth_stencil_ci.front = GetStencilFaceState(ds.front_stencil);
-    depth_stencil_ci.back = GetStencilFaceState(ds.back_stencil);
+    depth_stencil_ci.front = GetStencilFaceState(ds.front);
+    depth_stencil_ci.back = GetStencilFaceState(ds.back);
    depth_stencil_ci.minDepthBounds = 0.0f;
    depth_stencil_ci.maxDepthBounds = 0.0f;

    std::array<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
-    const std::size_t num_attachments =
-        std::min(cd.attachments_count, renderpass_params.color_attachments.size());
-    for (std::size_t i = 0; i < num_attachments; ++i) {
-        static constexpr std::array component_table = {
+    const std::size_t num_attachments = renderpass_params.color_attachments.size();
+    for (std::size_t index = 0; index < num_attachments; ++index) {
+        static constexpr std::array COMPONENT_TABLE = {
            VK_COLOR_COMPONENT_R_BIT, VK_COLOR_COMPONENT_G_BIT, VK_COLOR_COMPONENT_B_BIT,
            VK_COLOR_COMPONENT_A_BIT};
-        const auto& blend = cd.attachments[i];
+        const auto& blend = cd.attachments[index];

        VkColorComponentFlags color_components = 0;
-        for (std::size_t j = 0; j < component_table.size(); ++j) {
-            if (blend.components[j]) {
-                color_components |= component_table[j];
+        for (std::size_t i = 0; i < COMPONENT_TABLE.size(); ++i) {
+            if (blend.Mask()[i]) {
+                color_components |= COMPONENT_TABLE[i];
            }
        }

-        VkPipelineColorBlendAttachmentState& attachment = cb_attachments[i];
-        attachment.blendEnable = blend.enable;
-        attachment.srcColorBlendFactor = MaxwellToVK::BlendFactor(blend.src_rgb_func);
-        attachment.dstColorBlendFactor = MaxwellToVK::BlendFactor(blend.dst_rgb_func);
-        attachment.colorBlendOp = MaxwellToVK::BlendEquation(blend.rgb_equation);
-        attachment.srcAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.src_a_func);
-        attachment.dstAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.dst_a_func);
-        attachment.alphaBlendOp = MaxwellToVK::BlendEquation(blend.a_equation);
+        VkPipelineColorBlendAttachmentState& attachment = cb_attachments[index];
+        attachment.blendEnable = blend.enable != 0;
+        attachment.srcColorBlendFactor = MaxwellToVK::BlendFactor(blend.SourceRGBFactor());
+        attachment.dstColorBlendFactor = MaxwellToVK::BlendFactor(blend.DestRGBFactor());
+        attachment.colorBlendOp = MaxwellToVK::BlendEquation(blend.EquationRGB());
+        attachment.srcAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.SourceAlphaFactor());
+        attachment.dstAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.DestAlphaFactor());
+        attachment.alphaBlendOp = MaxwellToVK::BlendEquation(blend.EquationAlpha());
        attachment.colorWriteMask = color_components;
    }

@@ -207,7 +207,7 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
        const GPUVAddr program_addr{GetShaderAddress(system, program)};
        const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
        ASSERT(cpu_addr);
-        auto shader = cpu_addr ? TryGet(*cpu_addr) : nullptr;
+        auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
        if (!shader) {
            const auto host_ptr{memory_manager.GetPointer(program_addr)};

@@ -218,7 +218,11 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {

            shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
                                                    std::move(code), stage_offset);
-            Register(shader);
+            if (cpu_addr) {
+                Register(shader);
+            } else {
+                null_shader = shader;
+            }
        }
        shaders[index] = std::move(shader);
    }
@@ -261,7 +265,7 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
    const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
    ASSERT(cpu_addr);

-    auto shader = cpu_addr ? TryGet(*cpu_addr) : nullptr;
+    auto shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
    if (!shader) {
        // No shader found - create a new one
        const auto host_ptr = memory_manager.GetPointer(program_addr);
@@ -271,7 +275,11 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
        shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
                                                program_addr, *cpu_addr, std::move(code),
                                                kernel_main_offset);
-        Register(shader);
+        if (cpu_addr) {
+            Register(shader);
+        } else {
+            null_kernel = shader;
+        }
    }

    Specialization specialization;
@@ -329,12 +337,14 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
    const auto& gpu = system.GPU().Maxwell3D();

    Specialization specialization;
-    if (fixed_state.input_assembly.topology == Maxwell::PrimitiveTopology::Points) {
-        ASSERT(fixed_state.input_assembly.point_size != 0.0f);
-        specialization.point_size = fixed_state.input_assembly.point_size;
+    if (fixed_state.rasterizer.Topology() == Maxwell::PrimitiveTopology::Points) {
+        float point_size;
+        std::memcpy(&point_size, &fixed_state.rasterizer.point_size, sizeof(float));
+        specialization.point_size = point_size;
+        ASSERT(point_size != 0.0f);
    }
    for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
-        specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].type;
+        specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].Type();
    }
    specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;

@@ -182,6 +182,9 @@ private:
    VKUpdateDescriptorQueue& update_descriptor_queue;
    VKRenderPassCache& renderpass_cache;

+    Shader null_shader{};
+    Shader null_kernel{};
+
    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;

    GraphicsPipelineCacheKey last_graphics_key;
@@ -113,8 +113,19 @@ u64 HostCounter::BlockingQuery() const {
    if (ticks >= cache.Scheduler().Ticks()) {
        cache.Scheduler().Flush();
    }
-    return cache.Device().GetLogical().GetQueryResult<u64>(
-        query.first, query.second, VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
+    u64 data;
+    const VkResult result = cache.Device().GetLogical().GetQueryResults(
+        query.first, query.second, 1, sizeof(data), &data, sizeof(data),
+        VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
+    switch (result) {
+    case VK_SUCCESS:
+        return data;
+    case VK_ERROR_DEVICE_LOST:
+        cache.Device().ReportLoss();
+        [[fallthrough]];
+    default:
+        throw vk::Exception(result);
+    }
 }

 } // namespace Vulkan
@@ -17,6 +17,7 @@
 #include "common/microprofile.h"
 #include "core/core.h"
 #include "core/memory.h"
+#include "core/settings.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
@@ -299,7 +300,9 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
      pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue,
                     renderpass_cache),
      buffer_cache(*this, system, device, memory_manager, scheduler, staging_pool),
-      sampler_cache(device), query_cache(system, *this, device, scheduler) {
+      sampler_cache(device),
+      fence_manager(system, *this, device, scheduler, texture_cache, buffer_cache, query_cache),
+      query_cache(system, *this, device, scheduler) {
    scheduler.SetQueryCache(query_cache);
 }

@@ -347,11 +350,6 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {

    buffer_bindings.Bind(scheduler);

-    if (device.IsNvDeviceDiagnosticCheckpoints()) {
-        scheduler.Record(
-            [&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(&pipeline); });
-    }
-
    BeginTransformFeedback();

    const auto pipeline_layout = pipeline.GetLayout();
@@ -365,6 +363,8 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
    });

    EndTransformFeedback();
+
+    system.GPU().TickWork();
 }

 void RasterizerVulkan::Clear() {
@@ -478,11 +478,6 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
    TransitionImages(image_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                     VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);

-    if (device.IsNvDeviceDiagnosticCheckpoints()) {
-        scheduler.Record(
-            [&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(nullptr); });
-    }
-
    scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
                      grid_z = launch_desc.grid_dim_z, pipeline_handle = pipeline.GetHandle(),
                      layout = pipeline.GetLayout(),
@@ -514,6 +509,13 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
    query_cache.FlushRegion(addr, size);
 }

+bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) {
+    if (!Settings::IsGPULevelHigh()) {
+        return buffer_cache.MustFlushRegion(addr, size);
+    }
+    return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size);
+}
+
 void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
    if (addr == 0 || size == 0) {
        return;
@@ -524,6 +526,47 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
    query_cache.InvalidateRegion(addr, size);
 }

+void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    texture_cache.OnCPUWrite(addr, size);
+    pipeline_cache.InvalidateRegion(addr, size);
+    buffer_cache.OnCPUWrite(addr, size);
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerVulkan::SyncGuestHost() {
+    texture_cache.SyncGuestHost();
+    buffer_cache.SyncGuestHost();
+}
+
+void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        gpu.MemoryManager().Write<u32>(addr, value);
+        return;
+    }
+    fence_manager.SignalSemaphore(addr, value);
+}
+
+void RasterizerVulkan::SignalSyncPoint(u32 value) {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        gpu.IncrementSyncPoint(value);
+        return;
+    }
+    fence_manager.SignalSyncPoint(value);
+}
+
+void RasterizerVulkan::ReleaseFences() {
+    auto& gpu{system.GPU()};
+    if (!gpu.IsAsync()) {
+        return;
+    }
+    fence_manager.WaitPendingFences();
+}
+
 void RasterizerVulkan::FlushAndInvalidateRegion(VAddr addr, u64 size) {
    FlushRegion(addr, size);
    InvalidateRegion(addr, size);
@@ -807,25 +850,29 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex
                                         BufferBindings& buffer_bindings) {
    const auto& regs = system.GPU().Maxwell3D().regs;

-    for (u32 index = 0; index < static_cast<u32>(Maxwell::NumVertexAttributes); ++index) {
+    for (std::size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
        const auto& attrib = regs.vertex_attrib_format[index];
        if (!attrib.IsValid()) {
+            vertex_input.SetAttribute(index, false, 0, 0, {}, {});
            continue;
        }

-        const auto& buffer = regs.vertex_array[attrib.buffer];
+        [[maybe_unused]] const auto& buffer = regs.vertex_array[attrib.buffer];
        ASSERT(buffer.IsEnabled());

-        vertex_input.attributes[vertex_input.num_attributes++] =
-            FixedPipelineState::VertexAttribute(index, attrib.buffer, attrib.type, attrib.size,
-                                                attrib.offset);
+        vertex_input.SetAttribute(index, true, attrib.buffer, attrib.offset, attrib.type.Value(),
+                                  attrib.size.Value());
    }

-    for (u32 index = 0; index < static_cast<u32>(Maxwell::NumVertexArrays); ++index) {
+    for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
        const auto& vertex_array = regs.vertex_array[index];
        if (!vertex_array.IsEnabled()) {
+            vertex_input.SetBinding(index, false, 0, 0);
            continue;
        }
+        vertex_input.SetBinding(
+            index, true, vertex_array.stride,
+            regs.instanced_arrays.IsInstancingEnabled(index) ? vertex_array.divisor : 0);

        const GPUVAddr start{vertex_array.StartAddress()};
        const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
@@ -833,10 +880,6 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex
        ASSERT(end > start);
        const std::size_t size{end - start + 1};
        const auto [buffer, offset] = buffer_cache.UploadMemory(start, size);
-
-        vertex_input.bindings[vertex_input.num_bindings++] = FixedPipelineState::VertexBinding(
-            index, vertex_array.stride,
-            regs.instanced_arrays.IsInstancingEnabled(index) ? vertex_array.divisor : 0);
        buffer_bindings.AddVertexBinding(buffer, offset);
    }
 }
@@ -21,6 +21,7 @@
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_fence_manager.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
 #include "video_core/renderer_vulkan/vk_query_cache.h"
@@ -118,7 +119,13 @@ public:
    void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
    void FlushAll() override;
    void FlushRegion(VAddr addr, u64 size) override;
+    bool MustFlushRegion(VAddr addr, u64 size) override;
    void InvalidateRegion(VAddr addr, u64 size) override;
+    void OnCPUWrite(VAddr addr, u64 size) override;
+    void SyncGuestHost() override;
+    void SignalSemaphore(GPUVAddr addr, u32 value) override;
+    void SignalSyncPoint(u32 value) override;
+    void ReleaseFences() override;
    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
    void FlushCommands() override;
    void TickFrame() override;
@@ -261,6 +268,7 @@ private:
    VKPipelineCache pipeline_cache;
    VKBufferCache buffer_cache;
    VKSamplerCache sampler_cache;
+    VKFenceManager fence_manager;
    VKQueryCache query_cache;

    std::array<View, Maxwell::NumRenderTargets> color_attachments;
@@ -166,7 +166,15 @@ void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
    submit_info.pCommandBuffers = current_cmdbuf.address();
    submit_info.signalSemaphoreCount = semaphore ? 1 : 0;
    submit_info.pSignalSemaphores = &semaphore;
-    device.GetGraphicsQueue().Submit(submit_info, *current_fence);
+    switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info, *current_fence)) {
+    case VK_SUCCESS:
+        break;
+    case VK_ERROR_DEVICE_LOST:
+        device.ReportLoss();
+        [[fallthrough]];
+    default:
+        vk::Check(result);
+    }
 }

 void VKScheduler::AllocateNewContext() {
@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.

 #include <algorithm>
+#include <limits>
 #include <optional>
 #include <tuple>
 #include <vector>
@@ -22,22 +23,38 @@ namespace {
 constexpr u64 WATCHES_INITIAL_RESERVE = 0x4000;
 constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;

-constexpr u64 STREAM_BUFFER_SIZE = 256 * 1024 * 1024;
+constexpr u64 PREFERRED_STREAM_BUFFER_SIZE = 256 * 1024 * 1024;

-std::optional<u32> FindMemoryType(const VKDevice& device, u32 filter,
-                                  VkMemoryPropertyFlags wanted) {
-    const auto properties = device.GetPhysical().GetMemoryProperties();
-    for (u32 i = 0; i < properties.memoryTypeCount; i++) {
-        if (!(filter & (1 << i))) {
-            continue;
-        }
-        if ((properties.memoryTypes[i].propertyFlags & wanted) == wanted) {
+/// Find a memory type with the passed requirements
+std::optional<u32> FindMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
+                                  VkMemoryPropertyFlags wanted,
+                                  u32 filter = std::numeric_limits<u32>::max()) {
+    for (u32 i = 0; i < properties.memoryTypeCount; ++i) {
+        const auto flags = properties.memoryTypes[i].propertyFlags;
+        if ((flags & wanted) == wanted && (filter & (1U << i)) != 0) {
            return i;
        }
    }
    return std::nullopt;
 }

+/// Get the preferred host visible memory type.
+u32 GetMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
+                  u32 filter = std::numeric_limits<u32>::max()) {
+    // Prefer device local host visible allocations. Both AMD and Nvidia now provide one.
+    // Otherwise search for a host visible allocation.
+    static constexpr auto HOST_MEMORY =
+        VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+    static constexpr auto DYNAMIC_MEMORY = HOST_MEMORY | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+
+    std::optional preferred_type = FindMemoryType(properties, DYNAMIC_MEMORY);
+    if (!preferred_type) {
+        preferred_type = FindMemoryType(properties, HOST_MEMORY);
+        ASSERT_MSG(preferred_type, "No host visible and coherent memory type found");
+    }
+    return preferred_type.value_or(0);
+}
+
 } // Anonymous namespace

 VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
@@ -51,7 +68,7 @@ VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
 VKStreamBuffer::~VKStreamBuffer() = default;

 std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
-    ASSERT(size <= STREAM_BUFFER_SIZE);
+    ASSERT(size <= stream_buffer_size);
    mapped_size = size;

    if (alignment > 0) {
@@ -61,7 +78,7 @@ std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
    WaitPendingOperations(offset);

    bool invalidated = false;
-    if (offset + size > STREAM_BUFFER_SIZE) {
+    if (offset + size > stream_buffer_size) {
        // The buffer would overflow, save the amount of used watches and reset the state.
        invalidation_mark = current_watch_cursor;
        current_watch_cursor = 0;
@@ -98,40 +115,37 @@ void VKStreamBuffer::Unmap(u64 size) {
 }

 void VKStreamBuffer::CreateBuffers(VkBufferUsageFlags usage) {
+    const auto memory_properties = device.GetPhysical().GetMemoryProperties();
+    const u32 preferred_type = GetMemoryType(memory_properties);
+    const u32 preferred_heap = memory_properties.memoryTypes[preferred_type].heapIndex;
+
+    // Substract from the preferred heap size some bytes to avoid getting out of memory.
+    const VkDeviceSize heap_size = memory_properties.memoryHeaps[preferred_heap].size;
+    const VkDeviceSize allocable_size = heap_size - 4 * 1024 * 1024;
+
    VkBufferCreateInfo buffer_ci;
    buffer_ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
    buffer_ci.pNext = nullptr;
    buffer_ci.flags = 0;
-    buffer_ci.size = STREAM_BUFFER_SIZE;
+    buffer_ci.size = std::min(PREFERRED_STREAM_BUFFER_SIZE, allocable_size);
    buffer_ci.usage = usage;
    buffer_ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
    buffer_ci.queueFamilyIndexCount = 0;
    buffer_ci.pQueueFamilyIndices = nullptr;

-    const auto& dev = device.GetLogical();
-    buffer = dev.CreateBuffer(buffer_ci);
+    buffer = device.GetLogical().CreateBuffer(buffer_ci);
+
+    const auto requirements = device.GetLogical().GetBufferMemoryRequirements(*buffer);
+    const u32 required_flags = requirements.memoryTypeBits;
+    stream_buffer_size = static_cast<u64>(requirements.size);

-    const auto& dld = device.GetDispatchLoader();
-    const auto requirements = dev.GetBufferMemoryRequirements(*buffer);
-    // Prefer device local host visible allocations (this should hit AMD's pinned memory).
-    auto type =
-        FindMemoryType(device, requirements.memoryTypeBits,
-                       VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
-                           VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
-    if (!type) {
-        // Otherwise search for a host visible allocation.
-        type = FindMemoryType(device, requirements.memoryTypeBits,
-                              VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                                  VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
-        ASSERT_MSG(type, "No host visible and coherent memory type found");
-    }
    VkMemoryAllocateInfo memory_ai;
    memory_ai.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
    memory_ai.pNext = nullptr;
    memory_ai.allocationSize = requirements.size;
-    memory_ai.memoryTypeIndex = *type;
+    memory_ai.memoryTypeIndex = GetMemoryType(memory_properties, required_flags);

-    memory = dev.AllocateMemory(memory_ai);
+    memory = device.GetLogical().AllocateMemory(memory_ai);
    buffer.BindMemory(*memory, 0);
 }

@@ -56,8 +56,9 @@ private:
    const VKDevice& device; ///< Vulkan device manager.
    VKScheduler& scheduler; ///< Command scheduler.

-    vk::Buffer buffer;       ///< Mapped buffer.
-    vk::DeviceMemory memory; ///< Memory allocation.
+    vk::Buffer buffer;        ///< Mapped buffer.
+    vk::DeviceMemory memory;  ///< Memory allocation.
+    u64 stream_buffer_size{}; ///< Stream buffer size.

    u64 offset{};      ///< Buffer iterator.
    u64 mapped_size{}; ///< Size reserved for the current copy.
@@ -61,9 +61,9 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
    X(vkCmdPipelineBarrier);
    X(vkCmdPushConstants);
    X(vkCmdSetBlendConstants);
-    X(vkCmdSetCheckpointNV);
    X(vkCmdSetDepthBias);
    X(vkCmdSetDepthBounds);
+    X(vkCmdSetEvent);
    X(vkCmdSetScissor);
    X(vkCmdSetStencilCompareMask);
    X(vkCmdSetStencilReference);
@@ -76,6 +76,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
    X(vkCreateDescriptorPool);
    X(vkCreateDescriptorSetLayout);
    X(vkCreateDescriptorUpdateTemplateKHR);
+    X(vkCreateEvent);
    X(vkCreateFence);
    X(vkCreateFramebuffer);
    X(vkCreateGraphicsPipelines);
@@ -94,6 +95,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
    X(vkDestroyDescriptorPool);
    X(vkDestroyDescriptorSetLayout);
    X(vkDestroyDescriptorUpdateTemplateKHR);
+    X(vkDestroyEvent);
    X(vkDestroyFence);
    X(vkDestroyFramebuffer);
    X(vkDestroyImage);
@@ -113,10 +115,10 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
    X(vkFreeMemory);
    X(vkGetBufferMemoryRequirements);
    X(vkGetDeviceQueue);
+    X(vkGetEventStatus);
    X(vkGetFenceStatus);
    X(vkGetImageMemoryRequirements);
    X(vkGetQueryPoolResults);
-    X(vkGetQueueCheckpointDataNV);
    X(vkMapMemory);
    X(vkQueueSubmit);
    X(vkResetFences);
@@ -271,6 +273,10 @@ void Destroy(VkDevice device, VkDeviceMemory handle, const DeviceDispatch& dld)
    dld.vkFreeMemory(device, handle, nullptr);
 }

+void Destroy(VkDevice device, VkEvent handle, const DeviceDispatch& dld) noexcept {
+    dld.vkDestroyEvent(device, handle, nullptr);
+}
+
 void Destroy(VkDevice device, VkFence handle, const DeviceDispatch& dld) noexcept {
    dld.vkDestroyFence(device, handle, nullptr);
 }
@@ -409,17 +415,6 @@ DebugCallback Instance::TryCreateDebugCallback(
    return DebugCallback(messenger, handle, *dld);
 }

-std::vector<VkCheckpointDataNV> Queue::GetCheckpointDataNV(const DeviceDispatch& dld) const {
-    if (!dld.vkGetQueueCheckpointDataNV) {
-        return {};
-    }
-    u32 num;
-    dld.vkGetQueueCheckpointDataNV(queue, &num, nullptr);
-    std::vector<VkCheckpointDataNV> checkpoints(num);
-    dld.vkGetQueueCheckpointDataNV(queue, &num, checkpoints.data());
-    return checkpoints;
-}
-
 void Buffer::BindMemory(VkDeviceMemory memory, VkDeviceSize offset) const {
    Check(dld->vkBindBufferMemory(owner, handle, memory, offset));
 }
@@ -469,12 +464,11 @@ std::vector<VkImage> SwapchainKHR::GetImages() const {
 }

 Device Device::Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
-                      Span<const char*> enabled_extensions,
-                      const VkPhysicalDeviceFeatures2& enabled_features,
+                      Span<const char*> enabled_extensions, const void* next,
                      DeviceDispatch& dld) noexcept {
    VkDeviceCreateInfo ci;
    ci.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
-    ci.pNext = &enabled_features;
+    ci.pNext = next;
    ci.flags = 0;
    ci.queueCreateInfoCount = queues_ci.size();
    ci.pQueueCreateInfos = queues_ci.data();
@@ -613,6 +607,16 @@ ShaderModule Device::CreateShaderModule(const VkShaderModuleCreateInfo& ci) cons
    return ShaderModule(object, handle, *dld);
 }

+Event Device::CreateEvent() const {
+    VkEventCreateInfo ci;
+    ci.sType = VK_STRUCTURE_TYPE_EVENT_CREATE_INFO;
+    ci.pNext = nullptr;
+    ci.flags = 0;
+    VkEvent object;
+    Check(dld->vkCreateEvent(handle, &ci, nullptr, &object));
+    return Event(object, handle, *dld);
+}
+
 SwapchainKHR Device::CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const {
    VkSwapchainKHR object;
    Check(dld->vkCreateSwapchainKHR(handle, &ci, nullptr, &object));
@@ -197,9 +197,9 @@ struct DeviceDispatch : public InstanceDispatch {
    PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier;
    PFN_vkCmdPushConstants vkCmdPushConstants;
    PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants;
-    PFN_vkCmdSetCheckpointNV vkCmdSetCheckpointNV;
    PFN_vkCmdSetDepthBias vkCmdSetDepthBias;
    PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds;
+    PFN_vkCmdSetEvent vkCmdSetEvent;
    PFN_vkCmdSetScissor vkCmdSetScissor;
    PFN_vkCmdSetStencilCompareMask vkCmdSetStencilCompareMask;
    PFN_vkCmdSetStencilReference vkCmdSetStencilReference;
@@ -212,6 +212,7 @@ struct DeviceDispatch : public InstanceDispatch {
    PFN_vkCreateDescriptorPool vkCreateDescriptorPool;
    PFN_vkCreateDescriptorSetLayout vkCreateDescriptorSetLayout;
    PFN_vkCreateDescriptorUpdateTemplateKHR vkCreateDescriptorUpdateTemplateKHR;
+    PFN_vkCreateEvent vkCreateEvent;
    PFN_vkCreateFence vkCreateFence;
    PFN_vkCreateFramebuffer vkCreateFramebuffer;
    PFN_vkCreateGraphicsPipelines vkCreateGraphicsPipelines;
@@ -230,6 +231,7 @@ struct DeviceDispatch : public InstanceDispatch {
    PFN_vkDestroyDescriptorPool vkDestroyDescriptorPool;
    PFN_vkDestroyDescriptorSetLayout vkDestroyDescriptorSetLayout;
    PFN_vkDestroyDescriptorUpdateTemplateKHR vkDestroyDescriptorUpdateTemplateKHR;
+    PFN_vkDestroyEvent vkDestroyEvent;
    PFN_vkDestroyFence vkDestroyFence;
    PFN_vkDestroyFramebuffer vkDestroyFramebuffer;
    PFN_vkDestroyImage vkDestroyImage;
@@ -249,10 +251,10 @@ struct DeviceDispatch : public InstanceDispatch {
    PFN_vkFreeMemory vkFreeMemory;
    PFN_vkGetBufferMemoryRequirements vkGetBufferMemoryRequirements;
    PFN_vkGetDeviceQueue vkGetDeviceQueue;
+    PFN_vkGetEventStatus vkGetEventStatus;
    PFN_vkGetFenceStatus vkGetFenceStatus;
    PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements;
    PFN_vkGetQueryPoolResults vkGetQueryPoolResults;
-    PFN_vkGetQueueCheckpointDataNV vkGetQueueCheckpointDataNV;
    PFN_vkMapMemory vkMapMemory;
    PFN_vkQueueSubmit vkQueueSubmit;
    PFN_vkResetFences vkResetFences;
@@ -281,6 +283,7 @@ void Destroy(VkDevice, VkDescriptorPool, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkDescriptorSetLayout, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkDescriptorUpdateTemplateKHR, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkDeviceMemory, const DeviceDispatch&) noexcept;
+void Destroy(VkDevice, VkEvent, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkFence, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkFramebuffer, const DeviceDispatch&) noexcept;
 void Destroy(VkDevice, VkImage, const DeviceDispatch&) noexcept;
@@ -567,12 +570,8 @@ public:
    /// Construct a queue handle.
    constexpr Queue(VkQueue queue, const DeviceDispatch& dld) noexcept : queue{queue}, dld{&dld} {}

-    /// Returns the checkpoint data.
-    /// @note Returns an empty vector when the function pointer is not present.
-    std::vector<VkCheckpointDataNV> GetCheckpointDataNV(const DeviceDispatch& dld) const;
-
-    void Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const {
-        Check(dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence));
+    VkResult Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const noexcept {
+        return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence);
    }

    VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
@@ -654,13 +653,21 @@ public:
    std::vector<VkImage> GetImages() const;
 };

+class Event : public Handle<VkEvent, VkDevice, DeviceDispatch> {
+    using Handle<VkEvent, VkDevice, DeviceDispatch>::Handle;
+
+public:
+    VkResult GetStatus() const noexcept {
+        return dld->vkGetEventStatus(owner, handle);
+    }
+};
+
 class Device : public Handle<VkDevice, NoOwner, DeviceDispatch> {
    using Handle<VkDevice, NoOwner, DeviceDispatch>::Handle;

 public:
    static Device Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
-                         Span<const char*> enabled_extensions,
-                         const VkPhysicalDeviceFeatures2& enabled_features,
+                         Span<const char*> enabled_extensions, const void* next,
                         DeviceDispatch& dld) noexcept;

    Queue GetQueue(u32 family_index) const noexcept;
@@ -702,6 +709,8 @@ public:

    ShaderModule CreateShaderModule(const VkShaderModuleCreateInfo& ci) const;

+    Event CreateEvent() const;
+
    SwapchainKHR CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const;

    DeviceMemory TryAllocateMemory(const VkMemoryAllocateInfo& ai) const noexcept;
@@ -734,18 +743,11 @@ public:
        dld->vkResetQueryPoolEXT(handle, query_pool, first, count);
    }

-    void GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size,
-                         void* data, VkDeviceSize stride, VkQueryResultFlags flags) const {
-        Check(dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride,
-                                         flags));
-    }
-
-    template <typename T>
-    T GetQueryResult(VkQueryPool query_pool, u32 first, VkQueryResultFlags flags) const {
-        static_assert(std::is_trivially_copyable_v<T>);
-        T value;
-        GetQueryResults(query_pool, first, 1, sizeof(T), &value, sizeof(T), flags);
-        return value;
+    VkResult GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size,
+                             void* data, VkDeviceSize stride, VkQueryResultFlags flags) const
+        noexcept {
+        return dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride,
+                                          flags);
    }
 };

@@ -920,10 +922,6 @@ public:
        dld->vkCmdPushConstants(handle, layout, flags, offset, size, values);
    }

-    void SetCheckpointNV(const void* checkpoint_marker) const noexcept {
-        dld->vkCmdSetCheckpointNV(handle, checkpoint_marker);
-    }
-
    void SetViewport(u32 first, Span<VkViewport> viewports) const noexcept {
        dld->vkCmdSetViewport(handle, first, viewports.size(), viewports.data());
    }
@@ -956,6 +954,10 @@ public:
        dld->vkCmdSetDepthBounds(handle, min_depth_bounds, max_depth_bounds);
    }

+    void SetEvent(VkEvent event, VkPipelineStageFlags stage_flags) const noexcept {
+        dld->vkCmdSetEvent(handle, event, stage_flags);
+    }
+
    void BindTransformFeedbackBuffersEXT(u32 first, u32 count, const VkBuffer* buffers,
                                         const VkDeviceSize* offsets,
                                         const VkDeviceSize* sizes) const noexcept {
@@ -587,8 +587,6 @@ bool TryQuery(CFGRebuildState& state) {
    return true;
 }

-} // Anonymous namespace
-
 void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
    const auto get_expr = ([&](const Condition& cond) -> Expr {
        Expr result{};
@@ -655,6 +653,8 @@ void DecompileShader(CFGRebuildState& state) {
    state.manager->Decompile();
 }

+} // Anonymous namespace
+
 std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
                                                const CompilerSettings& settings,
                                                Registry& registry) {
@@ -479,7 +479,7 @@ std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackGlobalMemory(NodeBlock&
    bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset)));

    const GlobalMemoryBase descriptor{index, offset};
-    const auto& [entry, is_new] = used_global_memory.try_emplace(descriptor);
+    const auto& entry = used_global_memory.try_emplace(descriptor).first;
    auto& usage = entry->second;
    usage.is_written |= is_write;
    usage.is_read |= is_read;
@@ -139,7 +139,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
        }
        const Node component = Immediate(static_cast<u32>(instr.tld4s.component));

-        const SamplerInfo info{TextureType::Texture2D, false, is_depth_compare};
+        const SamplerInfo info{TextureType::Texture2D, false, is_depth_compare, false};
        const Sampler& sampler = *GetSampler(instr.sampler, info);

        Node4 values;
@@ -171,13 +171,12 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
        const auto coord_count = GetCoordCount(texture_type);
        Node index_var{};
        const Sampler* sampler =
-            is_bindless ? GetBindlessSampler(base_reg, index_var, {{texture_type, is_array, false}})
-                        : GetSampler(instr.sampler, {{texture_type, is_array, false}});
+            is_bindless
+                ? GetBindlessSampler(base_reg, index_var, {{texture_type, is_array, false, false}})
+                : GetSampler(instr.sampler, {{texture_type, is_array, false, false}});
        Node4 values;
        if (sampler == nullptr) {
-            for (u32 element = 0; element < values.size(); ++element) {
-                values[element] = Immediate(0);
-            }
+            std::generate(values.begin(), values.end(), [] { return Immediate(0); });
            WriteTexInstructionFloat(bb, instr, values);
            break;
        }
@@ -269,7 +268,6 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
                             "NDV is not implemented");

        auto texture_type = instr.tmml.texture_type.Value();
-        const bool is_array = instr.tmml.array != 0;
        Node index_var{};
        const Sampler* sampler =
            is_bindless ? GetBindlessSampler(instr.gpr20, index_var) : GetSampler(instr.sampler);
@@ -593,8 +591,9 @@ Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
        ++parameter_register;
    }

-    const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
-        texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5);
+    const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array,
+                                                              lod_bias_enabled, 4, 5);
+    const auto coord_count = std::get<0>(coord_counts);
    // If enabled arrays index is always stored in the gpr8 field
    const u64 array_register = instr.gpr8.Value();
    // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
@@ -632,8 +631,10 @@ Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
    const bool lod_bias_enabled =
        (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);

-    const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
-        texture_type, depth_compare, is_array, lod_bias_enabled, 4, 4);
+    const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array,
+                                                              lod_bias_enabled, 4, 4);
+    const auto coord_count = std::get<0>(coord_counts);
+
    // If enabled arrays index is always stored in the gpr8 field
    const u64 array_register = instr.gpr8.Value();
    // First coordinate index is stored in gpr8 field or (gpr8 + 1) when arrays are used
@@ -192,6 +192,22 @@ public:
        index = index_;
    }

+    void SetMemoryMarked(bool is_memory_marked_) {
+        is_memory_marked = is_memory_marked_;
+    }
+
+    bool IsMemoryMarked() const {
+        return is_memory_marked;
+    }
+
+    void SetSyncPending(bool is_sync_pending_) {
+        is_sync_pending = is_sync_pending_;
+    }
+
+    bool IsSyncPending() const {
+        return is_sync_pending;
+    }
+
    void MarkAsPicked(bool is_picked_) {
        is_picked = is_picked_;
    }
@@ -303,6 +319,8 @@ private:
    bool is_target{};
    bool is_registered{};
    bool is_picked{};
+    bool is_memory_marked{};
+    bool is_sync_pending{};
    u32 index{NO_RT};
    u64 modification_tick{};
 };
@@ -6,6 +6,7 @@

 #include <algorithm>
 #include <array>
+#include <list>
 #include <memory>
 #include <mutex>
 #include <set>
@@ -62,6 +63,30 @@ public:
        }
    }

+    void OnCPUWrite(VAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        for (const auto& surface : GetSurfacesInRegion(addr, size)) {
+            if (surface->IsMemoryMarked()) {
+                UnmarkMemory(surface);
+                surface->SetSyncPending(true);
+                marked_for_unregister.emplace_back(surface);
+            }
+        }
+    }
+
+    void SyncGuestHost() {
+        std::lock_guard lock{mutex};
+
+        for (const auto& surface : marked_for_unregister) {
+            if (surface->IsRegistered()) {
+                surface->SetSyncPending(false);
+                Unregister(surface);
+            }
+        }
+        marked_for_unregister.clear();
+    }
+
    /**
     * Guarantees that rendertargets don't unregister themselves if the
     * collide. Protection is currently only done on 3D slices.
@@ -85,10 +110,20 @@ public:
            return a->GetModificationTick() < b->GetModificationTick();
        });
        for (const auto& surface : surfaces) {
+            mutex.unlock();
            FlushSurface(surface);
+            mutex.lock();
        }
    }

+    bool MustFlushRegion(VAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        const auto surfaces = GetSurfacesInRegion(addr, size);
+        return std::any_of(surfaces.cbegin(), surfaces.cend(),
+                           [](const TSurface& surface) { return surface->IsModified(); });
+    }
+
    TView GetTextureSurface(const Tegra::Texture::TICEntry& tic,
                            const VideoCommon::Shader::Sampler& entry) {
        std::lock_guard lock{mutex};
@@ -206,8 +241,14 @@ public:

        auto surface_view = GetSurface(gpu_addr, *cpu_addr,
                                       SurfaceParams::CreateForFramebuffer(system, index), true);
-        if (render_targets[index].target)
-            render_targets[index].target->MarkAsRenderTarget(false, NO_RT);
+        if (render_targets[index].target) {
+            auto& surface = render_targets[index].target;
+            surface->MarkAsRenderTarget(false, NO_RT);
+            const auto& cr_params = surface->GetSurfaceParams();
+            if (!cr_params.is_tiled && Settings::values.use_asynchronous_gpu_emulation) {
+                AsyncFlushSurface(surface);
+            }
+        }
        render_targets[index].target = surface_view.first;
        render_targets[index].view = surface_view.second;
        if (render_targets[index].target)
@@ -284,6 +325,34 @@ public:
        return ++ticks;
    }

+    void CommitAsyncFlushes() {
+        committed_flushes.push_back(uncommitted_flushes);
+        uncommitted_flushes.reset();
+    }
+
+    bool HasUncommittedFlushes() const {
+        return uncommitted_flushes != nullptr;
+    }
+
+    bool ShouldWaitAsyncFlushes() const {
+        return !committed_flushes.empty() && committed_flushes.front() != nullptr;
+    }
+
+    void PopAsyncFlushes() {
+        if (committed_flushes.empty()) {
+            return;
+        }
+        auto& flush_list = committed_flushes.front();
+        if (!flush_list) {
+            committed_flushes.pop_front();
+            return;
+        }
+        for (TSurface& surface : *flush_list) {
+            FlushSurface(surface);
+        }
+        committed_flushes.pop_front();
+    }
+
 protected:
    explicit TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                          bool is_astc_supported)
@@ -345,9 +414,20 @@ protected:
        surface->SetCpuAddr(*cpu_addr);
        RegisterInnerCache(surface);
        surface->MarkAsRegistered(true);
+        surface->SetMemoryMarked(true);
        rasterizer.UpdatePagesCachedCount(*cpu_addr, size, 1);
    }

+    void UnmarkMemory(TSurface surface) {
+        if (!surface->IsMemoryMarked()) {
+            return;
+        }
+        const std::size_t size = surface->GetSizeInBytes();
+        const VAddr cpu_addr = surface->GetCpuAddr();
+        rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
+        surface->SetMemoryMarked(false);
+    }
+
    void Unregister(TSurface surface) {
        if (guard_render_targets && surface->IsProtected()) {
            return;
@@ -355,9 +435,11 @@ protected:
        if (!guard_render_targets && surface->IsRenderTarget()) {
            ManageRenderTargetUnregister(surface);
        }
-        const std::size_t size = surface->GetSizeInBytes();
-        const VAddr cpu_addr = surface->GetCpuAddr();
-        rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
+        UnmarkMemory(surface);
+        if (surface->IsSyncPending()) {
+            marked_for_unregister.remove(surface);
+            surface->SetSyncPending(false);
+        }
        UnregisterInnerCache(surface);
        surface->MarkAsRegistered(false);
        ReserveSurface(surface->GetSurfaceParams(), surface);
@@ -417,7 +499,7 @@ private:
     **/
    RecycleStrategy PickStrategy(std::vector<TSurface>& overlaps, const SurfaceParams& params,
                                 const GPUVAddr gpu_addr, const MatchTopologyResult untopological) {
-        if (Settings::values.use_accurate_gpu_emulation) {
+        if (Settings::IsGPULevelExtreme()) {
            return RecycleStrategy::Flush;
        }
        // 3D Textures decision
@@ -461,7 +543,7 @@ private:
        }
        switch (PickStrategy(overlaps, params, gpu_addr, untopological)) {
        case RecycleStrategy::Ignore: {
-            return InitializeSurface(gpu_addr, params, Settings::values.use_accurate_gpu_emulation);
+            return InitializeSurface(gpu_addr, params, Settings::IsGPULevelExtreme());
        }
        case RecycleStrategy::Flush: {
            std::sort(overlaps.begin(), overlaps.end(),
@@ -509,7 +591,7 @@ private:
        }
        const auto& final_params = new_surface->GetSurfaceParams();
        if (cr_params.type != final_params.type) {
-            if (Settings::values.use_accurate_gpu_emulation) {
+            if (Settings::IsGPULevelExtreme()) {
                BufferCopy(current_surface, new_surface);
            }
        } else {
@@ -598,7 +680,7 @@ private:
        if (passed_tests == 0) {
            return {};
            // In Accurate GPU all tests should pass, else we recycle
-        } else if (Settings::values.use_accurate_gpu_emulation && passed_tests != overlaps.size()) {
+        } else if (Settings::IsGPULevelExtreme() && passed_tests != overlaps.size()) {
            return {};
        }
        for (const auto& surface : overlaps) {
@@ -668,7 +750,7 @@ private:
            for (const auto& surface : overlaps) {
                if (!surface->MatchTarget(params.target)) {
                    if (overlaps.size() == 1 && surface->GetCpuAddr() == cpu_addr) {
-                        if (Settings::values.use_accurate_gpu_emulation) {
+                        if (Settings::IsGPULevelExtreme()) {
                            return std::nullopt;
                        }
                        Unregister(surface);
@@ -1106,6 +1188,13 @@ private:
        TView view;
    };

+    void AsyncFlushSurface(TSurface& surface) {
+        if (!uncommitted_flushes) {
+            uncommitted_flushes = std::make_shared<std::list<TSurface>>();
+        }
+        uncommitted_flushes->push_back(surface);
+    }
+
    VideoCore::RasterizerInterface& rasterizer;

    FormatLookupTable format_lookup_table;
@@ -1150,6 +1239,11 @@ private:
    std::unordered_map<u32, TSurface> invalid_cache;
    std::vector<u8> invalid_memory;

+    std::list<TSurface> marked_for_unregister;
+
+    std::shared_ptr<std::list<TSurface>> uncommitted_flushes{};
+    std::list<std::shared_ptr<std::list<TSurface>>> committed_flushes;
+
    StagingCache staging_cache;
    std::recursive_mutex mutex;
 };
@@ -11,6 +11,7 @@
 #include "video_core/textures/texture.h"

 namespace Tegra::Texture {
+namespace {

 /**
 * This table represents the internal swizzle of a gob,
@@ -174,6 +175,8 @@ void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool
    }
 }

+} // Anonymous namespace
+
 void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel,
                      u32 out_bytes_per_pixel, u8* const swizzled_data, u8* const unswizzled_data,
                      bool unswizzle, u32 block_height, u32 block_depth, u32 width_spacing) {
@@ -56,8 +56,7 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32
                      u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
                      u32 offset_x, u32 offset_y);

-void SwizzleKepler(const u32 width, const u32 height, const u32 dst_x, const u32 dst_y,
-                   const u32 block_height, const std::size_t copy_size, const u8* source_data,
-                   u8* swizzle_data);
+void SwizzleKepler(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height,
+                   std::size_t copy_size, const u8* source_data, u8* swizzle_data);

 } // namespace Tegra::Texture
@@ -17,6 +17,7 @@
 #include "yuzu/applets/profile_select.h"
 #include "yuzu/main.h"

+namespace {
 QString FormatUserEntryText(const QString& username, Common::UUID uuid) {
    return QtProfileSelectionDialog::tr(
               "%1\n%2", "%1 is the profile username, %2 is the formatted UUID (e.g. "
@@ -41,6 +42,7 @@ QPixmap GetIcon(Common::UUID uuid) {

    return icon.scaled(64, 64, Qt::IgnoreAspectRatio, Qt::SmoothTransformation);
 }
+} // Anonymous namespace

 QtProfileSelectionDialog::QtProfileSelectionDialog(QWidget* parent)
    : QDialog(parent), profile_manager(std::make_unique<Service::Account::ProfileManager>()) {
@@ -532,6 +532,8 @@ void Config::ReadDebuggingValues() {
    Settings::values.reporting_services =
        ReadSetting(QStringLiteral("reporting_services"), false).toBool();
    Settings::values.quest_flag = ReadSetting(QStringLiteral("quest_flag"), false).toBool();
+    Settings::values.disable_cpu_opt =
+        ReadSetting(QStringLiteral("disable_cpu_opt"), false).toBool();

    qt_config->endGroup();
 }
@@ -637,11 +639,13 @@ void Config::ReadRendererValues() {
    Settings::values.frame_limit = ReadSetting(QStringLiteral("frame_limit"), 100).toInt();
    Settings::values.use_disk_shader_cache =
        ReadSetting(QStringLiteral("use_disk_shader_cache"), true).toBool();
-    Settings::values.use_accurate_gpu_emulation =
-        ReadSetting(QStringLiteral("use_accurate_gpu_emulation"), false).toBool();
+    const int gpu_accuracy_level = ReadSetting(QStringLiteral("gpu_accuracy"), 0).toInt();
+    Settings::values.gpu_accuracy = static_cast<Settings::GPUAccuracy>(gpu_accuracy_level);
    Settings::values.use_asynchronous_gpu_emulation =
        ReadSetting(QStringLiteral("use_asynchronous_gpu_emulation"), false).toBool();
    Settings::values.use_vsync = ReadSetting(QStringLiteral("use_vsync"), true).toBool();
+    Settings::values.use_fast_gpu_time =
+        ReadSetting(QStringLiteral("use_fast_gpu_time"), true).toBool();
    Settings::values.force_30fps_mode =
        ReadSetting(QStringLiteral("force_30fps_mode"), false).toBool();

@@ -1001,6 +1005,7 @@ void Config::SaveDebuggingValues() {
    WriteSetting(QStringLiteral("dump_exefs"), Settings::values.dump_exefs, false);
    WriteSetting(QStringLiteral("dump_nso"), Settings::values.dump_nso, false);
    WriteSetting(QStringLiteral("quest_flag"), Settings::values.quest_flag, false);
+    WriteSetting(QStringLiteral("disable_cpu_opt"), Settings::values.disable_cpu_opt, false);

    qt_config->endGroup();
 }
@@ -1077,11 +1082,12 @@ void Config::SaveRendererValues() {
    WriteSetting(QStringLiteral("frame_limit"), Settings::values.frame_limit, 100);
    WriteSetting(QStringLiteral("use_disk_shader_cache"), Settings::values.use_disk_shader_cache,
                 true);
-    WriteSetting(QStringLiteral("use_accurate_gpu_emulation"),
-                 Settings::values.use_accurate_gpu_emulation, false);
+    WriteSetting(QStringLiteral("gpu_accuracy"), static_cast<int>(Settings::values.gpu_accuracy),
+                 0);
    WriteSetting(QStringLiteral("use_asynchronous_gpu_emulation"),
                 Settings::values.use_asynchronous_gpu_emulation, false);
    WriteSetting(QStringLiteral("use_vsync"), Settings::values.use_vsync, true);
+    WriteSetting(QStringLiteral("use_fast_gpu_time"), Settings::values.use_fast_gpu_time, true);
    WriteSetting(QStringLiteral("force_30fps_mode"), Settings::values.force_30fps_mode, false);

    // Cast to double because Qt's written float values are not human-readable
@@ -36,6 +36,7 @@ void ConfigureDebug::SetConfiguration() {
    ui->homebrew_args_edit->setText(QString::fromStdString(Settings::values.program_args));
    ui->reporting_services->setChecked(Settings::values.reporting_services);
    ui->quest_flag->setChecked(Settings::values.quest_flag);
+    ui->disable_cpu_opt->setChecked(Settings::values.disable_cpu_opt);
    ui->enable_graphics_debugging->setEnabled(!Core::System::GetInstance().IsPoweredOn());
    ui->enable_graphics_debugging->setChecked(Settings::values.renderer_debug);
 }
@@ -48,6 +49,7 @@ void ConfigureDebug::ApplyConfiguration() {
    Settings::values.program_args = ui->homebrew_args_edit->text().toStdString();
    Settings::values.reporting_services = ui->reporting_services->isChecked();
    Settings::values.quest_flag = ui->quest_flag->isChecked();
+    Settings::values.disable_cpu_opt = ui->disable_cpu_opt->isChecked();
    Settings::values.renderer_debug = ui->enable_graphics_debugging->isChecked();
    Debugger::ToggleConsole();
    Log::Filter filter;
@@ -215,6 +215,13 @@
        </property>
       </widget>
      </item>
+      <item>
+       <widget class="QCheckBox" name="disable_cpu_opt">
+        <property name="text">
+         <string>Disable CPU JIT optimizations</string>
+        </property>
+       </widget>
+      </item>
     </layout>
    </widget>
   </item>
@@ -19,9 +19,10 @@ ConfigureGraphicsAdvanced::~ConfigureGraphicsAdvanced() = default;

 void ConfigureGraphicsAdvanced::SetConfiguration() {
    const bool runtime_lock = !Core::System::GetInstance().IsPoweredOn();
-    ui->use_accurate_gpu_emulation->setChecked(Settings::values.use_accurate_gpu_emulation);
+    ui->gpu_accuracy->setCurrentIndex(static_cast<int>(Settings::values.gpu_accuracy));
    ui->use_vsync->setEnabled(runtime_lock);
    ui->use_vsync->setChecked(Settings::values.use_vsync);
+    ui->use_fast_gpu_time->setChecked(Settings::values.use_fast_gpu_time);
    ui->force_30fps_mode->setEnabled(runtime_lock);
    ui->force_30fps_mode->setChecked(Settings::values.force_30fps_mode);
    ui->anisotropic_filtering_combobox->setEnabled(runtime_lock);
@@ -29,8 +30,10 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
 }

 void ConfigureGraphicsAdvanced::ApplyConfiguration() {
-    Settings::values.use_accurate_gpu_emulation = ui->use_accurate_gpu_emulation->isChecked();
+    auto gpu_accuracy = static_cast<Settings::GPUAccuracy>(ui->gpu_accuracy->currentIndex());
+    Settings::values.gpu_accuracy = gpu_accuracy;
    Settings::values.use_vsync = ui->use_vsync->isChecked();
+    Settings::values.use_fast_gpu_time = ui->use_fast_gpu_time->isChecked();
    Settings::values.force_30fps_mode = ui->force_30fps_mode->isChecked();
    Settings::values.max_anisotropy = ui->anisotropic_filtering_combobox->currentIndex();
 }
@@ -23,11 +23,34 @@
       </property>
       <layout class="QVBoxLayout" name="verticalLayout_3">
        <item>
-         <widget class="QCheckBox" name="use_accurate_gpu_emulation">
-          <property name="text">
-           <string>Use accurate GPU emulation (slow)</string>
-          </property>
-         </widget>
+         <layout class="QHBoxLayout" name="horizontalLayout_2">
+          <item>
+           <widget class="QLabel" name="label_gpu_accuracy">
+            <property name="text">
+             <string>Accuracy Level:</string>
+            </property>
+           </widget>
+          </item>
+          <item>
+           <widget class="QComboBox" name="gpu_accuracy">
+            <item>
+             <property name="text">
+              <string notr="true">Normal</string>
+             </property>
+            </item>
+            <item>
+             <property name="text">
+              <string notr="true">High</string>
+             </property>
+            </item>
+            <item>
+             <property name="text">
+              <string notr="true">Extreme(very slow)</string>
+             </property>
+            </item>
+           </widget>
+          </item>
+         </layout>
        </item>
        <item>
         <widget class="QCheckBox" name="use_vsync">
@@ -46,6 +69,13 @@
          </property>
         </widget>
        </item>
+        <item>
+         <widget class="QCheckBox" name="use_fast_gpu_time">
+          <property name="text">
+           <string>Use Fast GPU Time</string>
+          </property>
+         </widget>
+        </item>
        <item>
         <layout class="QHBoxLayout" name="horizontalLayout_1">
          <item>
@@ -388,12 +388,14 @@ void Config::ReadValues() {
        static_cast<u16>(sdl2_config->GetInteger("Renderer", "frame_limit", 100));
    Settings::values.use_disk_shader_cache =
        sdl2_config->GetBoolean("Renderer", "use_disk_shader_cache", false);
-    Settings::values.use_accurate_gpu_emulation =
-        sdl2_config->GetBoolean("Renderer", "use_accurate_gpu_emulation", false);
+    const int gpu_accuracy_level = sdl2_config->GetInteger("Renderer", "gpu_accuracy", 0);
+    Settings::values.gpu_accuracy = static_cast<Settings::GPUAccuracy>(gpu_accuracy_level);
    Settings::values.use_asynchronous_gpu_emulation =
        sdl2_config->GetBoolean("Renderer", "use_asynchronous_gpu_emulation", false);
    Settings::values.use_vsync =
        static_cast<u16>(sdl2_config->GetInteger("Renderer", "use_vsync", 1));
+    Settings::values.use_fast_gpu_time =
+        sdl2_config->GetBoolean("Renderer", "use_fast_gpu_time", true);

    Settings::values.bg_red = static_cast<float>(sdl2_config->GetReal("Renderer", "bg_red", 0.0));
    Settings::values.bg_green =
@@ -425,6 +427,8 @@ void Config::ReadValues() {
    Settings::values.reporting_services =
        sdl2_config->GetBoolean("Debugging", "reporting_services", false);
    Settings::values.quest_flag = sdl2_config->GetBoolean("Debugging", "quest_flag", false);
+    Settings::values.disable_cpu_opt =
+        sdl2_config->GetBoolean("Debugging", "disable_cpu_opt", false);

    const auto title_list = sdl2_config->Get("AddOns", "title_ids", "");
    std::stringstream ss(title_list);
@@ -146,9 +146,9 @@ frame_limit =
 # 0 (default): Off, 1 : On
 use_disk_shader_cache =

-# Whether to use accurate GPU emulation
-# 0 (default): Off (fast), 1 : On (slow)
-use_accurate_gpu_emulation =
+# Which gpu accuracy level to use
+# 0 (Normal), 1 (High), 2 (Extreme)
+gpu_accuracy =

 # Whether to use asynchronous GPU emulation
 # 0 : Off (slow), 1 (default): On (fast)
@@ -280,6 +280,9 @@ dump_nso=false
 # Determines whether or not yuzu will report to the game that the emulated console is in Kiosk Mode
 # false: Retail/Normal Mode (default), true: Kiosk Mode
 quest_flag =
+# Determines whether or not JIT CPU optimizations are enabled
+# false: Optimizations Enabled, true: Optimizations Disabled
+disable_cpu_opt =

 [WebService]
 # Whether or not to enable telemetry
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Fernando Sahmkow	5c9feaebb6	Clang Format.	2020-04-23 08:52:58 -04:00
Fernando Sahmkow	b8aef40c56	GPU: Add Fast GPU Time Option.	2020-04-23 08:52:57 -04:00
Fernando Sahmkow	18a88d19dc	Maxwell3D: Process Macros on MultiMethod.	2020-04-23 08:52:56 -04:00
Fernando Sahmkow	3fedcc2f6e	DMAPusher: Propagate multimethod writes into the engines.	2020-04-23 08:52:55 -04:00
bunnei	eb26e9e711	Merge pull request #3730 from lioncash/time service/time: Remove reliance on the global system accessor	2020-04-23 02:41:38 -04:00
bunnei	2409fedacf	Merge pull request #3697 from lioncash/declarations CMakeLists: Enable -Wmissing-declarations on Linux builds	2020-04-23 02:18:52 -04:00
bunnei	bf2ddb8fd5	Merge pull request #3677 from FernandoS27/better-sync Introduce Predictive Flushing and Improve ASYNC GPU	2020-04-22 22:09:38 -04:00
Mat M	2c806c5fd3	Merge pull request #3767 from ReinUsesLisp/point-size-pipeline vk_pipeline_cache: Fix unintentional memcpy into optional	2020-04-22 21:20:14 -04:00
ReinUsesLisp	d9463f4562	vk_pipeline_cache: Fix unintentional memcpy into optional The intention behind this was to assign a float to from an uint32_t, but it was unintentionally being copied directly into the std::optional. Copy to a temporary and assign that temporary to std::optional. This can be replaced with std::bit_cast<float> once we are in C++20.	2020-04-22 21:36:05 -03:00
Fernando Sahmkow	c043ac4f13	GL_Fence_Manager: use GL_TIMEOUT_IGNORED instead of a loop,	2020-04-22 20:34:32 -04:00
bunnei	bee2c64b34	Merge pull request #3725 from MerryMage/fpcr thread: FPCR.FZ is likely not 1 (and FPCR.RMode = TieAway and FPCR.DN = 0)	2020-04-22 19:49:13 -04:00
Mat M	6ce3d174b7	Merge pull request #3759 from H27CK/opus-mingw-w64 Set _FORTIFY_SOURCE=0 if building Opus with mingw-w64	2020-04-22 17:45:44 -04:00
Fernando Sahmkow	63d2ba4f69	Merge pull request #3763 from H27CK/vk-cmd Add missing ;	2020-04-22 17:43:42 -04:00
H27CK	4d74578d35	Add missing ;	2020-04-22 23:36:21 +02:00
bunnei	5ed13304e1	Merge pull request #3758 from H27CK/vk-cmd Introduce dummy context for yuzu-cmd VK support	2020-04-22 12:37:01 -04:00
Fernando Sahmkow	afae40a99e	Merge pull request #3653 from ReinUsesLisp/nsight-aftermath renderer_vulkan: Integrate Nvidia Nsight Aftermath on Windows	2020-04-22 11:39:01 -04:00
Fernando Sahmkow	4e37f1b113	Address Feedback.	2020-04-22 11:36:27 -04:00
Fernando Sahmkow	39e5b72948	Async GPU: Correct flushing behavior to be similar to old async GPU behavior.	2020-04-22 11:36:26 -04:00
Fernando Sahmkow	1b3be8a8f8	MaxwellDMA: Correct copying on accuracy level.	2020-04-22 11:36:25 -04:00
Fernando Sahmkow	644588fd88	ShaderCache/PipelineCache: Cache null shaders.	2020-04-22 11:36:25 -04:00
Fernando Sahmkow	f616dc0b59	Address Feedback.	2020-04-22 11:36:24 -04:00
Fernando Sahmkow	ec2f3e48e1	Fix GCC error.	2020-04-22 11:36:23 -04:00
Fernando Sahmkow	7f44f22451	Correct Linux Compile Error.	2020-04-22 11:36:22 -04:00
Fernando Sahmkow	d2d4a6cbcf	Clang format.	2020-04-22 11:36:22 -04:00
Fernando Sahmkow	b3e5f177ba	QueryCache: Only do async flushes on async gpu.	2020-04-22 11:36:21 -04:00
Fernando Sahmkow	f4ab223ef0	Async GPU: Only do reactive flushing on Extreme Level.	2020-04-22 11:36:20 -04:00
ReinUsesLisp	b752faf2d3	vk_fence_manager: Initial implementation	2020-04-22 11:36:19 -04:00
Fernando Sahmkow	0649f05900	QueryCache: Implement Async Flushes.	2020-04-22 11:36:18 -04:00
Fernando Sahmkow	131b342130	OpenGL: Guarantee writes to Buffers.	2020-04-22 11:36:18 -04:00
Fernando Sahmkow	1fb516cd97	GPU: Implement Flush Requests for Async mode.	2020-04-22 11:36:17 -04:00
Fernando Sahmkow	b7bc3c2549	FenceManager: Manage syncpoints and rename fences to semaphores.	2020-04-22 11:36:16 -04:00
Fernando Sahmkow	96bb961a64	BufferCache: Refactor async managing.	2020-04-22 11:36:15 -04:00
Fernando Sahmkow	b10db7e4a5	FenceManager: Implement async buffer cache flushes on High settings	2020-04-22 11:36:15 -04:00
Fernando Sahmkow	4adfc9bb08	Rasterizer: Document SignalFence & ReleaseFences and setup skeletons on Vulkan.	2020-04-22 11:36:14 -04:00
Fernando Sahmkow	a081a7c855	GPU: Fix rebase errors.	2020-04-22 11:36:13 -04:00
Fernando Sahmkow	e84eb64e51	Rasterizer: Disable fence managing in synchronous gpu.	2020-04-22 11:36:12 -04:00
Fernando Sahmkow	165ae823f5	ThreadManager: Sync async reads on accurate gpu.	2020-04-22 11:36:12 -04:00
Fernando Sahmkow	57fdbd9b89	FenceManager: Implement should wait.	2020-04-22 11:36:11 -04:00
Fernando Sahmkow	1f345ebe3a	GPU: Implement a Fence Manager.	2020-04-22 11:36:10 -04:00
Fernando Sahmkow	487379c593	OpenGL: Implement Fencing backend.	2020-04-22 11:36:10 -04:00
Fernando Sahmkow	ed7e965712	TextureCache: Flush linear textures after finishing rendering.	2020-04-22 11:36:09 -04:00
Fernando Sahmkow	339d0d9d6c	GPU: Delay Fences.	2020-04-22 11:36:08 -04:00
Fernando Sahmkow	8b1eb44b3e	BufferCache: Implement OnCPUWrite and SyncGuestHost	2020-04-22 11:36:07 -04:00
Fernando Sahmkow	da8f17715d	GPU: Refactor synchronization on Async GPU	2020-04-22 11:36:06 -04:00
Fernando Sahmkow	a60a22d9c2	Texture Cache: Implement OnCPUWrite and SyncGuestHost	2020-04-22 11:36:05 -04:00
Fernando Sahmkow	084ceb925a	UI: Replasce accurate GPU option for GPU Accuracy Level	2020-04-22 11:36:04 -04:00
H27CK	52e66779e7	Set _FORTIFY_SOURCE=0 if building Opus with mingw-w64	2020-04-22 08:20:12 +02:00
H27CK	c883cd103e	Init SDL info structure and add dummy context	2020-04-22 07:47:21 +02:00
bunnei	e84f82a028	Merge pull request #3699 from FearlessTobi/port-5185 Port citra-emu/citra#5185: "gdbstub: Fix some gdbstub jankiness"	2020-04-21 22:26:10 -04:00
bunnei	d64290884a	Merge pull request #3714 from lioncash/copies gl_shader_decompiler: Avoid copies where applicable	2020-04-21 20:16:02 -04:00
bunnei	cd47ccec49	Merge pull request #3745 from bunnei/fix-homebrew-load Fix process memory initialization for ELF and NRO	2020-04-21 18:59:16 -04:00
Rodrigo Locatti	f293b15611	Merge pull request #3718 from ReinUsesLisp/better-pipeline-state fixed_pipeline_state: Pack structure, use memcmp and CityHash on it	2020-04-21 18:17:58 -03:00
bunnei	9bf3abcb63	Merge pull request #3698 from lioncash/warning General: Resolve minor assorted warnings	2020-04-21 14:11:18 -04:00
bunnei	48b670d132	Merge pull request #3724 from bunnei/fix-unicorn core: arm_unicorn: Fix interpret fallback by temporarily mapping instruction page.	2020-04-20 23:28:23 -04:00
David	11c63ca969	audio_renderer: Preliminary BehaviorInfo (#3736 ) * audio_renderer: Preliminary BehaviorInfo * clang format * Fixed IsRevisionSupported * fixed IsValidRevision * Fixed logic error & spelling errors & crash * Addressed issues	2020-04-20 22:57:30 -04:00
bunnei	d3e0cefa60	Merge pull request #3695 from ReinUsesLisp/default-attributes maxwell_3d: Initialize format attributes constant as one	2020-04-20 21:40:18 -04:00
Mat M	cb5b8ca886	Merge pull request #3733 from ambasta/patch-2 Initialize quad_indexed_pass before uint8_pass	2020-04-20 20:36:46 -04:00
bunnei	9c12aef2f8	loader: nro: Fix process initialization using ProgramMetadata default.	2020-04-20 18:19:45 -04:00
bunnei	68039addbc	loader: elf: Fix process initialization using ProgramMetadata default.	2020-04-20 18:19:45 -04:00
bunnei	f0a7f05070	file_sys: program_metadata: Add a helper function for generating reasonable default metadata. - We need this for homebrew process initialization.	2020-04-20 18:19:45 -04:00
Mat M	200f69d2ff	Merge pull request #3739 from MerryMage/disable_cpu_opt dynarmic: Add option to disable CPU JIT optimizations	2020-04-20 14:19:18 -04:00
bunnei	fe2173429a	Merge pull request #3741 from FearlessTobi/silence-warnings Try to reduce log spam a bit by lowering log levels to Debug	2020-04-20 13:01:49 -04:00
FearlessTobi	4e99a06c70	npad: Lower log level for VibrateController to Debug	2020-04-20 18:44:57 +02:00
FearlessTobi	6ce0f3575a	audren: Lower log level for RequestUpdateImpl to Debug	2020-04-20 18:44:41 +02:00
Fernando Sahmkow	ec2f8f4272	Merge pull request #3700 from ReinUsesLisp/stream-buffer-sizes vk_stream_buffer: Fix out of memory on boot on recent Nvidia drivers	2020-04-20 09:37:42 -04:00
MerryMage	a3a12deecc	dynarmic: Add option to disable CPU JIT optimizations	2020-04-20 13:36:26 +01:00
bunnei	6de36f0b61	Merge pull request #3712 from lioncash/remove service: Remove unused RequestParser instances	2020-04-20 01:04:04 -04:00
bunnei	e3977243e2	Merge pull request #3709 from lioncash/am am: Resolve ineffective moves	2020-04-20 00:15:00 -04:00
Lioncash	bfee33cce3	service/time: Remove reliance on the global system accessor Eliminates usages of the global system accessor and instead passes the existing system instance into the interfaces.	2020-04-19 16:31:28 -04:00
MerryMage	2bfac7b61d	thread: FPCR.FZ is likely not 1	2020-04-19 08:37:20 +01:00
bunnei	74c27fd1b5	core: arm_unicorn: Fix interpret fallback by temporarily mapping instruction page.	2020-04-19 00:53:23 -04:00
ReinUsesLisp	d62f57cf5a	fixed_pipeline_state: Hash and compare the whole structure Pad FixedPipelineState's size to 384 bytes to be a multiple of 16. Compare the whole struct with std::memcmp and hash with CityHash. Using CityHash instead of a naive hash should reduce the number of collisions. Improve used type traits to ensure this operation is safe. With these changes the improvements to the hashable pipeline state are: Optimized structure Hash: 89 ns Comparison: 103 ns Construction: 164 ns Struct size: 384 bytes Original structure Hash: 148 ns Equal: 174 ns Construction: 281 ns Size: 1384 bytes * Attribute state initialization is not measured These measures are averages taken with std::chrono::high_accuracy_clock on MSVC shipped on Visual Studio 16.6.0 Preview 2.1.	2020-04-18 19:57:26 -03:00
ReinUsesLisp	b571c92dfd	fixed_pipeline_state: Pack blending state Reduce FixedPipelineState's size to 364 bytes.	2020-04-18 19:23:35 -03:00
ReinUsesLisp	548dd27f45	fixed_pipeline_state: Pack rasterizer state Reduce FixedPipelineState's size to 600 bytes.	2020-04-18 19:22:57 -03:00
ReinUsesLisp	7790144a55	fixed_pipeline_state: Pack depth stencil state Reduce FixedPipelineState's size to 632 bytes.	2020-04-18 19:22:11 -03:00
ReinUsesLisp	ab6704f20c	fixed_pipeline_state: Pack attribute state Reduce FixedPipelineState's size from 1384 to 664 bytes	2020-04-18 19:21:19 -03:00
Lioncash	bf328ed35a	gl_shader_decompiler: Avoid copies where applicable Avoids unnecessary reference count increments where applicable and also avoids reallocating a vector. Unlikely to make a huge difference, but given how trivial of an amendment it is, why not?	2020-04-17 20:48:52 -04:00
Lioncash	4d8a8a8033	service: Remove unused RequestParser instances These aren't used, so they should be removed to reduce compilation warnings.	2020-04-17 19:47:43 -04:00
Lioncash	64f226889c	am: Resolve ineffective moves Previously const objects were being std::moved, which results in no move actually occurring. This resolves that.	2020-04-17 19:22:46 -04:00
ReinUsesLisp	a7b6bd56d7	vk_stream_buffer: Fix out of memory on boot on recent Nvidia drivers Nvidia recently introduced a new memory type for data streaming (awesome!), but yuzu was assuming that all heaps had enough memory for the assumed stream buffer size (256 MiB). This worked fine on AMD but Nvidia's new memory heap was smaller than 256 MiB. This commit changes this assumption and allocates a bit less than the size of the preferred heap, with a maximum of 256 MiB (to avoid allocating all system memory on integrated devices). - Fixes a crash on NVIDIA 450.82.0.0	2020-04-17 18:12:48 -03:00
Lioncash	dcbb39cdae	CMakeLists: Make missing declarations a compile-time error Ensures that our code always has its linkage explicit.	2020-04-16 23:43:41 -04:00
Lioncash	e2d8be1ca2	General: Resolve warnings related to missing declarations	2020-04-16 23:43:34 -04:00
MerryMage	1cc9507bc5	gdbstub: Fix some gdbstub jankiness 1. Ensure that register information available to gdbstub is most up-to-date. 2. There's no reason to check for current_thread == thread when emitting a trap. Doing this results in random hangs whenever a step happens upon a thread switch.	2020-04-17 05:41:43 +02:00
Lioncash	8f9c599c9f	key_manager: Resolve missing field initializer warning	2020-04-16 22:45:44 -04:00
Lioncash	678ac54749	decode/memory: Resolve unused variable warning Only the first element of the returned pair is ever used.	2020-04-16 22:45:44 -04:00
Lioncash	d159643fd7	decode/texture: Resolve unused variable warnings. Some variables aren't used, so we can remove these. Unfortunately, diagnostics are still reported on structured bindings even when annotated with [[maybe_unused]], so we need to unpack the elements that we want to use manually.	2020-04-16 22:45:41 -04:00
Lioncash	f522abd8ab	decode/texture: Collapse loop down into std::generate Same behavior, less code.	2020-04-16 22:29:07 -04:00
Lioncash	7e2d60de26	decode/texture: Eliminate trivial missing field initializer warnings We can just specify the initializers.	2020-04-16 22:27:21 -04:00
Lioncash	337f2dc11f	time_zone_manager: Resolve sign conversion warnings ttis and ats will never exceed the length of INT32_MAX in our case, so this is safe.	2020-04-16 22:23:59 -04:00
Lioncash	fc5df84581	CMakeLists: Enable -Wmissing-declarations on Linux builds Allows catching cases where internal linkage isn't specified for helper functions when they should be marked as such.	2020-04-16 22:07:16 -04:00
ReinUsesLisp	238c6016f9	maxwell_3d: Initialize format attributes constant as one nouveau expects this to be true but it doesn't set it.	2020-04-16 21:15:07 -03:00
ReinUsesLisp	0e232cfdc1	renderer_vulkan: Integrate Nvidia Nsight Aftermath on Windows Adds optional support for Nsight Aftermath. It is enabled through ENABLE_NSIGHT_AFTERMATH in cmake. A path to the SDK has to be provided by the environment variable NSIGHT_AFTERMATH_SDK. Nsight Aftermath allows an application to generate "minidumps" of the GPU state when a device loss happens. By analysing these on Nsight we can know what a game was doing and why it triggered a device loss. The dump is generated inside %APPDATA%\yuzu\log\gpucrash and this directory is deleted every time a new instance is initialized with Nsight enabled. To enable it on yuzu there has a to be a driver and device capable of running Nsight Aftermath on Vulkan. That means only Turing based GPUs on the latest stable driver, beta drivers won't work for now. It is manually enabled in Configuration>Debug>Enable Graphics Debugging because when using all debugging capabilities there is a runtime cost.	2020-04-14 00:39:21 -03:00
ReinUsesLisp	6cfe2a7246	renderer_vulkan: Remove Nvidia checkpoints	2020-04-13 17:33:59 -03:00
ReinUsesLisp	16105c6a66	renderer_vulkan: Catch device losses in more places	2020-04-13 17:33:59 -03:00