Agent/fix gen5 thread agc compat (#130)

* Fix Gen5 thread and AGC compatibility

* Trim compatibility comments

* Report selected Vulkan GPU

* Clean up CPU title label

* Improve emulator frame pacing and performance

* Regenerate package locks with pinned SDK

---------

Co-authored-by: Spooks4576 <Spooks4576@users.noreply.github.com>

src/SharpEmu.CLI/SharpEmu.CLI.csproj

@@ -24,6 +24,9 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <ImplicitUsings>enable</ImplicitUsings>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <Version>0.0.1</Version>
+    <ServerGarbageCollection>true</ServerGarbageCollection>
+    <ConcurrentGarbageCollection>true</ConcurrentGarbageCollection>
+    <TieredPGO>true</TieredPGO>
   </PropertyGroup>
 
   <PropertyGroup Condition="'$(Configuration)' == 'Release'">

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.Imports.cs

@@ -161,7 +161,7 @@ public sealed partial class DirectExecutionBackend
 				*(ulong*)(xmmSlot + 8));
 		}
 		cpuContext[CpuRegister.Rsp] = (ulong)argPackPtr + 96uL;
-		if (string.Equals(importStubEntry.Nid, RuntimeStubNids.BootstrapBridge, StringComparison.Ordinal))
+		if (importStubEntry.Kind == ImportStubKind.BootstrapBridge)
 		{
 			NormalizeKernelDynlibDlsymArguments(cpuContext, out _, out _);
 			*(ulong*)argPackPtr = cpuContext[CpuRegister.Rdi];
@@ -245,12 +245,13 @@ public sealed partial class DirectExecutionBackend
 		bool flag4 = !string.IsNullOrWhiteSpace(_importFilter);
 		bool flag5 = false;
 		ExportedFunction? matchedExport = importStubEntry.Export;
+		var traceFlags = importStubEntry.TraceFlags;
 		bool periodicTrace = num <= 128 ||
 			(num >= 240 && num <= 400) ||
 			(num >= 900 && num <= 1300) ||
 			num % 100000 == 0L ||
-			(importStubEntry.Nid == "tsvEmnenz48" && (num <= 256 || num % 1000 == 0L)) ||
-			(importStubEntry.Nid == "rTXw65xmLIA" && (num <= 256 || num % 128 == 0)) ||
+			((traceFlags & ImportStubTraceFlags.PeriodicEvery1000) != 0 && (num <= 256 || num % 1000 == 0L)) ||
+			((traceFlags & ImportStubTraceFlags.PeriodicEvery128) != 0 && (num <= 256 || num % 128 == 0)) ||
 			flag ||
 			flag2 ||
 			flag3;
@@ -311,15 +312,15 @@ public sealed partial class DirectExecutionBackend
 				cpuContext[CpuRegister.Rsi],
 				cpuContext[CpuRegister.Rdx]);
 		}
-		if (importStubEntry.Nid == "8zTFvBIAIN8" && num <= 256)
+		if ((traceFlags & ImportStubTraceFlags.Memset) != 0 && num <= 256)
 		{
 			Console.Error.WriteLine($"[LOADER][TRACE] memset#{num}: dst=0x{cpuContext[CpuRegister.Rdi]:X16} val=0x{cpuContext[CpuRegister.Rsi] & 0xFF:X2} len=0x{cpuContext[CpuRegister.Rdx]:X16} ret=0x{num7:X16}");
 		}
-		if (importStubEntry.Nid == "tsvEmnenz48" && num <= 64)
+		if ((traceFlags & ImportStubTraceFlags.CxaAtexit) != 0 && num <= 64)
 		{
 			Console.Error.WriteLine($"[LOADER][TRACE] __cxa_atexit#{num}: func=0x{cpuContext[CpuRegister.Rdi]:X16} arg=0x{cpuContext[CpuRegister.Rsi]:X16} dso=0x{cpuContext[CpuRegister.Rdx]:X16} ret=0x{num7:X16}");
 		}
-		if (importStubEntry.Nid == "bzQExy189ZI" || importStubEntry.Nid == "8G2LB+A3rzg")
+		if ((traceFlags & ImportStubTraceFlags.RawArgs) != 0)
 		{
 			Console.Error.WriteLine($"[LOADER][TRACE] {importStubEntry.Nid}#{num}: rdi=0x{cpuContext[CpuRegister.Rdi]:X16} rsi=0x{cpuContext[CpuRegister.Rsi]:X16} rdx=0x{cpuContext[CpuRegister.Rdx]:X16} ret=0x{num7:X16}");
 		}
@@ -349,7 +350,7 @@ public sealed partial class DirectExecutionBackend
 				Console.Error.Flush();
 			}
 		}
-		if (importStubEntry.Nid == "Ou3iL1abvng")
+		if ((traceFlags & ImportStubTraceFlags.StackChkFail) != 0)
 		{
 			if (_logStackCheck)
 			{
@@ -381,7 +382,7 @@ public sealed partial class DirectExecutionBackend
 				ActiveGuestReturnSlotAddress);
 			try
 			{
-				if (string.Equals(importStubEntry.Nid, RuntimeStubNids.BootstrapBridge, StringComparison.Ordinal))
+				if (importStubEntry.Kind == ImportStubKind.BootstrapBridge)
 				{
 					if (_logBootstrap)
 					{
@@ -390,12 +391,11 @@ public sealed partial class DirectExecutionBackend
 
 					orbisGen2Result = DispatchBootstrapBridge();
 				}
-				else if (string.Equals(importStubEntry.Nid, RuntimeStubNids.KernelDynlibDlsym, StringComparison.Ordinal) ||
-					string.Equals(importStubEntry.Nid, "LwG8g3niqwA", StringComparison.Ordinal))
+				else if (importStubEntry.Kind == ImportStubKind.KernelDynlibDlsym)
 				{
 					orbisGen2Result = DispatchKernelDynlibDlsym();
 				}
-				else if (string.Equals(importStubEntry.Nid, "r8mvOaWdi28", StringComparison.Ordinal))
+				else if (importStubEntry.Kind == ImportStubKind.Il2CppApiLookupSymbol)
 				{
 					orbisGen2Result = DispatchIl2CppApiLookupSymbol();
 				}
@@ -818,6 +818,7 @@ public sealed partial class DirectExecutionBackend
 			"2Z+PpY6CaJg" or // pthread_mutex_unlock
 			"EgmLo6EWgso" or // pthread_rwlock_unlock
 			"+L98PIbGttk" or // scePthreadRwlockUnlock
+			"q1cHNfGycLI" or // scePadRead
 			"8aI7R7WaOlc" or // sceAmprCommandBufferConstructor
 			"zgXifHT9ErY" or // sceVideoOutIsFlipPending
 			"V++UgBtQhn0" or // sceAgcGetDataPacketPayloadAddress
@@ -1076,9 +1077,15 @@ public sealed partial class DirectExecutionBackend
 	}
 
 	private static bool IsImportLoopGuardBoundary(string nid) =>
-		string.Equals(nid, "1jfXLRVzisc", StringComparison.Ordinal) ||
-		string.Equals(nid, "Zxa0VhQVTsk", StringComparison.Ordinal) ||
-		string.Equals(nid, "T72hz6ffq08", StringComparison.Ordinal);
+    nid switch
+    {
+        "1jfXLRVzisc" => true, // sceKernelUsleep
+        "QcteRwbsnV0" => true, // usleep
+        "n88vx3C5nW8" => true, // gettimeofday
+        "Zxa0VhQVIsk" => true,
+        "T72hz6ffq08" => true, // scePthreadYield
+        _ => false
+    };
 
 	private void ResetImportLoopPattern()
 	{

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.cs

@@ -25,6 +25,26 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private const int DefaultImportLoopGuardSeconds = 5;
 
+	private enum ImportStubKind : byte
+	{
+		Normal = 0,
+		BootstrapBridge = 1,
+		KernelDynlibDlsym = 2,
+		Il2CppApiLookupSymbol = 3,
+	}
+
+	[Flags]
+	private enum ImportStubTraceFlags : byte
+	{
+		None = 0,
+		Memset = 1 << 0,
+		CxaAtexit = 1 << 1,
+		RawArgs = 1 << 2,
+		StackChkFail = 1 << 3,
+		PeriodicEvery1000 = 1 << 4,
+		PeriodicEvery128 = 1 << 5,
+	}
+
 	private readonly struct ImportStubEntry
 	{
 		public ulong Address { get; }
@@ -33,12 +53,36 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 		public ExportedFunction? Export { get; }
 
+		public ImportStubKind Kind { get; }
+
+		public ImportStubTraceFlags TraceFlags { get; }
+
 		public ImportStubEntry(ulong address, string nid, ExportedFunction? export)
 		{
 			Address = address;
 			Nid = nid;
 			Export = export;
+			Kind = ClassifyKind(nid);
+			TraceFlags = ClassifyTraceFlags(nid);
 		}
+
+		private static ImportStubKind ClassifyKind(string nid) => nid switch
+		{
+			RuntimeStubNids.BootstrapBridge => ImportStubKind.BootstrapBridge,
+			RuntimeStubNids.KernelDynlibDlsym or "LwG8g3niqwA" => ImportStubKind.KernelDynlibDlsym,
+			"r8mvOaWdi28" => ImportStubKind.Il2CppApiLookupSymbol,
+			_ => ImportStubKind.Normal,
+		};
+
+		private static ImportStubTraceFlags ClassifyTraceFlags(string nid) => nid switch
+		{
+			"8zTFvBIAIN8" => ImportStubTraceFlags.Memset,
+			"tsvEmnenz48" => ImportStubTraceFlags.CxaAtexit | ImportStubTraceFlags.PeriodicEvery1000,
+			"bzQExy189ZI" or "8G2LB+A3rzg" => ImportStubTraceFlags.RawArgs,
+			"Ou3iL1abvng" => ImportStubTraceFlags.StackChkFail,
+			"rTXw65xmLIA" => ImportStubTraceFlags.PeriodicEvery128,
+			_ => ImportStubTraceFlags.None,
+		};
 	}
 
 	private readonly record struct RecentImportTraceEntry(
@@ -125,6 +169,8 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private const ulong GuestThreadRegionStride = 0x0100_0000UL;
 
+	private const int GuestThreadRegionSlotCount = 256;
+
 	private const uint PAGE_EXECUTE_READWRITE = 64u;
 
 	private const uint PAGE_READWRITE = 4u;
@@ -2757,7 +2803,21 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			if (hostThread is not null &&
 				!ReferenceEquals(hostThread, Thread.CurrentThread))
 			{
-				hostThread.Join(1);
+				// The handle is published before the host thread starts.
+				if ((hostThread.ThreadState & System.Threading.ThreadState.Unstarted) != 0)
+				{
+					Thread.Sleep(1);
+					continue;
+				}
+
+				try
+				{
+					hostThread.Join(1);
+				}
+				catch (ThreadStateException)
+				{
+					Thread.Sleep(1);
+				}
 			}
 			else
 			{
@@ -3446,7 +3506,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		out ulong mappedBase,
 		out string? error)
 	{
-		for (int i = 0; i < 64; i++)
+		for (int i = 0; i < GuestThreadRegionSlotCount; i++)
 		{
 			var candidateBase = baseAddress - ((ulong)i * GuestThreadRegionStride);
 			if (!IsGuestThreadRegionFree(virtualMemory, candidateBase, size))
@@ -3480,7 +3540,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		out ulong tlsBase,
 		out string? error)
 	{
-		for (int i = 0; i < 64; i++)
+		for (int i = 0; i < GuestThreadRegionSlotCount; i++)
 		{
 			var candidateBase = GuestThreadTlsBaseAddress - ((ulong)i * GuestThreadRegionStride);
 			var mappedBase = candidateBase - GuestThreadTlsPrefixSize;

src/SharpEmu.GUI/MainWindow.axaml.cs

@@ -25,6 +25,7 @@ namespace SharpEmu.GUI;
 public partial class MainWindow : Window
 {
     private const int MaxConsoleLines = 4000;
+    private const int MaxConsoleLinesPerFlush = 500;
 
     private static readonly IBrush DefaultLineBrush = new SolidColorBrush(Color.Parse("#C7CFDE"));
     private static readonly IBrush DimLineBrush = new SolidColorBrush(Color.Parse("#6B7488"));
@@ -1504,7 +1505,8 @@ public partial class MainWindow : Window
         }
 
         var incoming = new List<LogLine>();
-        while (_pendingLines.TryDequeue(out var pending))
+        while (incoming.Count < MaxConsoleLinesPerFlush &&
+               _pendingLines.TryDequeue(out var pending))
         {
             WriteFileLog(pending.Line);
             incoming.Add(new LogLine(pending.Line, BrushForLine(pending.Line)));

src/SharpEmu.Libs/Agc/AgcExports.cs

@@ -52,6 +52,8 @@ public static class AgcExports
     private const uint RFlip = 0x17;
     private const uint RReleaseMem = 0x18;
     private const uint RDmaData = 0x19;
+    private const uint RPredication = 0x1A;
+    private const uint RJump = 0x1B;
     private const uint SpiShaderPgmLoPs = 0x8;
     private const uint SpiShaderPgmHiPs = 0x9;
     private const uint SpiShaderPgmLoEs = 0xC8;
@@ -389,6 +391,7 @@ public static class AgcExports
         public Dictionary<uint, SubmittedDcbState> ComputeQueues { get; } = new();
         public Dictionary<ulong, ComputeImageWriter> ComputeImageWriters { get; } = new();
         public ulong WorkSequence { get; set; }
+        public ulong FlipSequence { get; set; }
     }
 
     private readonly record struct RegisterDefaultValue(uint Offset, uint Value);
@@ -1923,6 +1926,90 @@ public static class AgcExports
         return ReturnPointer(ctx, commandAddress);
     }
 
+    [SysAbiExport(
+        Nid = "bbFueFP+J4k",
+        ExportName = "sceAgcDcbSetPredication",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAgc")]
+    public static int DcbSetPredication(CpuContext ctx)
+    {
+        var commandBufferAddress = ctx[CpuRegister.Rdi];
+        var enable = (uint)ctx[CpuRegister.Rsi];
+        var predicateType = (uint)ctx[CpuRegister.Rdx];
+        var continuePredicate = (uint)ctx[CpuRegister.Rcx];
+        var predicateAddress = ctx[CpuRegister.R8];
+        if (commandBufferAddress == 0 || enable > 1 || continuePredicate > 1)
+        {
+            return ReturnPointer(ctx, 0);
+        }
+
+        if (!TryAllocateCommandDwords(ctx, commandBufferAddress, 5, out var commandAddress) ||
+            !ctx.TryWriteUInt32(commandAddress, Pm4(5, ItNop, RPredication)) ||
+            !ctx.TryWriteUInt32(commandAddress + 4, enable) ||
+            !ctx.TryWriteUInt32(commandAddress + 8, predicateType) ||
+            !ctx.TryWriteUInt32(commandAddress + 12, continuePredicate) ||
+            !ctx.TryWriteUInt32(commandAddress + 16, unchecked((uint)predicateAddress)))
+        {
+            return ReturnPointer(ctx, 0);
+        }
+
+        TraceAgc(
+            $"agc.dcb_set_predication buf=0x{commandBufferAddress:X16} cmd=0x{commandAddress:X16} " +
+            $"enable={enable} type={predicateType} continue={continuePredicate} addr=0x{predicateAddress:X16}");
+        return ReturnPointer(ctx, commandAddress);
+    }
+
+    [SysAbiExport(
+        Nid = "xSAR0LTcRKM",
+        ExportName = "sceAgcDcbJump",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAgc")]
+    public static int DcbJump(CpuContext ctx)
+    {
+        var commandBufferAddress = ctx[CpuRegister.Rdi];
+        var targetAddress = ctx[CpuRegister.Rsi];
+        if (commandBufferAddress == 0)
+        {
+            return ReturnPointer(ctx, 0);
+        }
+
+        if (!TryAllocateCommandDwords(ctx, commandBufferAddress, 3, out var commandAddress) ||
+            !ctx.TryWriteUInt32(commandAddress, Pm4(3, ItNop, RJump)) ||
+            !ctx.TryWriteUInt32(commandAddress + 4, unchecked((uint)targetAddress)) ||
+            !ctx.TryWriteUInt32(commandAddress + 8, unchecked((uint)(targetAddress >> 32))))
+        {
+            return ReturnPointer(ctx, 0);
+        }
+
+        TraceAgc(
+            $"agc.dcb_jump buf=0x{commandBufferAddress:X16} cmd=0x{commandAddress:X16} " +
+            $"target=0x{targetAddress:X16}");
+        return ReturnPointer(ctx, commandAddress);
+    }
+
+    [SysAbiExport(
+        Nid = "w6Dj1VJt5qY",
+        ExportName = "sceAgcSetPacketPredication",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAgc")]
+    public static int SetPacketPredication(CpuContext ctx)
+    {
+        var packetAddress = ctx[CpuRegister.Rdi];
+        var enabled = ctx[CpuRegister.Rsi] != 0;
+        if (packetAddress == 0 || !ctx.TryReadUInt32(packetAddress, out var packetHeader))
+        {
+            return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        packetHeader = enabled ? packetHeader | 1u : packetHeader & ~1u;
+        if (!ctx.TryWriteUInt32(packetAddress, packetHeader))
+        {
+            return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
     [SysAbiExport(
         Nid = "w2rJhmD+dsE",
         ExportName = "sceAgcDriverAddEqEvent",
@@ -1975,9 +2062,22 @@ public static class AgcExports
     public static int DriverSubmitDcb(CpuContext ctx)
     {
         var packetAddress = ctx[CpuRegister.Rdi];
-        if (packetAddress == 0 ||
-            !ctx.TryReadUInt64(packetAddress, out var commandAddress) ||
-            !ctx.TryReadUInt32(packetAddress + 8, out var dwordCount))
+        // Gen5 supports direct RSI/RDX and packed RDI submissions.
+        var commandAddress = ctx[CpuRegister.Rsi];
+        var directDwordCount = ctx[CpuRegister.Rdx];
+        uint dwordCount;
+        var directArguments = commandAddress != 0 &&
+            directDwordCount is > 0 and <= 1_000_000UL;
+        if (directArguments)
+        {
+            dwordCount = (uint)directDwordCount;
+        }
+        else if (packetAddress == 0 ||
+                 !ctx.TryReadUInt64(packetAddress, out commandAddress) ||
+                 !ctx.TryReadUInt32(packetAddress + 8, out dwordCount) ||
+                 commandAddress == 0 ||
+                 dwordCount == 0 ||
+                 dwordCount > 1_000_000)
         {
             return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
         }
@@ -1993,16 +2093,29 @@ public static class AgcExports
 
         if (tracePackets)
         {
-            TraceAgc($"agc.driver_submit_dcb packet=0x{packetAddress:X16} addr=0x{commandAddress:X16} dwords={dwordCount}");
+            TraceAgc(
+                $"agc.driver_submit_dcb abi={(directArguments ? "direct" : "packed")} " +
+                $"packet=0x{packetAddress:X16} addr=0x{commandAddress:X16} dwords={dwordCount}");
         }
 
         var gpuState = _submittedGpuStates.GetValue(ctx.Memory, static _ => new SubmittedGpuState());
+        ulong flipSequenceBefore;
         lock (gpuState.Gate)
         {
+            flipSequenceBefore = gpuState.FlipSequence;
             ParseSubmittedDcb(ctx, gpuState, gpuState.Graphics, commandAddress, dwordCount, tracePackets);
             DrainResumableDcbs(ctx, gpuState, tracePackets);
         }
 
+        // Flip submissions are completed through VideoOut pacing.
+        if (gpuState.FlipSequence == flipSequenceBefore)
+        {
+            KernelEventQueueCompatExports.TriggerRegisteredEvents(
+                0,
+                KernelEventQueueCompatExports.KernelEventFilterGraphics,
+                0);
+        }
+
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -2028,8 +2141,10 @@ public static class AgcExports
             Environment.GetEnvironmentVariable("SHARPEMU_LOG_AGC"), "1", StringComparison.Ordinal);
 
         var gpuState = _submittedGpuStates.GetValue(ctx.Memory, static _ => new SubmittedGpuState());
+        ulong flipSequenceBefore;
         lock (gpuState.Gate)
         {
+            flipSequenceBefore = gpuState.FlipSequence;
             for (uint i = 0; i < bufferCount; i++)
             {
                 if (!ctx.TryReadUInt64(addressArray + i * 8, out var commandAddress) ||
@@ -2053,6 +2168,14 @@ public static class AgcExports
             DrainResumableDcbs(ctx, gpuState, tracePackets);
         }
 
+        if (gpuState.FlipSequence == flipSequenceBefore)
+        {
+            KernelEventQueueCompatExports.TriggerRegisteredEvents(
+                0,
+                KernelEventQueueCompatExports.KernelEventFilterGraphics,
+                0);
+        }
+
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -2301,6 +2424,40 @@ public static class AgcExports
             }
 
             var packetType = header >> 30;
+            if (packetType == 0)
+            {
+                // Consume alignment padding one dword at a time.
+                if (header == 0)
+                {
+                    offset++;
+                    continue;
+                }
+
+                // Preserve stream alignment across type-0 packets.
+                var type0Length = ((header >> 16) & 0x3FFFu) + 2u;
+                if (offset + type0Length > dwordCount)
+                {
+                    offset++;
+                    continue;
+                }
+
+                offset += type0Length;
+                continue;
+            }
+
+            if (packetType == 1)
+            {
+                const uint type1Length = 3;
+                if (offset + type1Length > dwordCount)
+                {
+                    offset++;
+                    continue;
+                }
+
+                offset += type1Length;
+                continue;
+            }
+
             if (packetType == 2)
             {
                 if (tracePackets)
@@ -2315,13 +2472,15 @@ public static class AgcExports
 
             if (packetType != 3)
             {
-                return;
+                offset++;
+                continue;
             }
 
             var length = Pm4Length(header);
             if (length == 0 || offset + length > dwordCount)
             {
-                return;
+                offset++;
+                continue;
             }
 
             var op = (header >> 8) & 0xFFu;
@@ -2559,6 +2718,7 @@ public static class AgcExports
 
             if (op == ItNop && register == RFlip && length >= 6)
             {
+                gpuState.FlipSequence++;
                 if (!ctx.TryReadUInt32(currentAddress + 4, out var videoOutHandle) ||
                     !ctx.TryReadUInt32(currentAddress + 8, out var displayBufferIndexRaw) ||
                     !ctx.TryReadUInt32(currentAddress + 12, out var flipMode) ||
@@ -2571,23 +2731,8 @@ public static class AgcExports
                 var flipArg = unchecked((long)(((ulong)flipArgHi << 32) | flipArgLo));
                 var displayBufferIndex = unchecked((int)displayBufferIndexRaw);
                 var handle = unchecked((int)videoOutHandle);
-                if (VideoOutExports.TryGetDisplayBufferInfo(
-                        handle,
-                        displayBufferIndex,
-                        out var cachedDisplayBuffer) &&
-                    VulkanVideoPresenter.TrySubmitGuestImage(
-                        cachedDisplayBuffer.Address,
-                        cachedDisplayBuffer.Width,
-                        cachedDisplayBuffer.Height,
-                        cachedDisplayBuffer.PitchInPixel))
-                {
-                    TraceDisplayBuffer(
-                        handle,
-                        displayBufferIndex,
-                        cachedDisplayBuffer,
-                        "gpu-cache");
-                }
-                else if (state.SawIndexedDraw &&
+                // Prefer output decoded from the current DCB.
+                if (state.SawIndexedDraw &&
                     state.TranslatedDraw is { } translatedDraw &&
                     VideoOutExports.TryGetDisplayBufferInfo(
                         handle,
@@ -2651,6 +2796,22 @@ public static class AgcExports
                         displayBuffer.Width,
                         displayBuffer.Height);
                 }
+                else if (VideoOutExports.TryGetDisplayBufferInfo(
+                             handle,
+                             displayBufferIndex,
+                             out var cachedDisplayBuffer) &&
+                         VulkanVideoPresenter.TrySubmitGuestImage(
+                             cachedDisplayBuffer.Address,
+                             cachedDisplayBuffer.Width,
+                             cachedDisplayBuffer.Height,
+                             cachedDisplayBuffer.PitchInPixel))
+                {
+                    TraceDisplayBuffer(
+                        handle,
+                        displayBufferIndex,
+                        cachedDisplayBuffer,
+                        "gpu-cache");
+                }
 
                 _ = VideoOutExports.SubmitFlipFromAgc(ctx, handle, displayBufferIndex, unchecked((int)flipMode), flipArg);
                 state.SawIndexedDraw = false;
@@ -3972,25 +4133,28 @@ public static class AgcExports
 
         TraceTextureHash(descriptor, source);
 
-        var nonZero = 0;
-        for (var i = 0; i < source.Length; i++)
+        if (_traceAgcShader)
         {
-            if (source[i] != 0)
+            var nonZero = 0;
+            for (var i = 0; i < source.Length; i++)
             {
-                nonZero++;
-                if (nonZero >= 64)
+                if (source[i] != 0)
                 {
-                    break;
+                    nonZero++;
+                    if (nonZero >= 64)
+                    {
+                        break;
+                    }
                 }
             }
-        }
 
-        TraceAgcShader(
-            $"agc.texture_source addr=0x{descriptor.Address:X16} " +
-            $"fmt={descriptor.Format} num={descriptor.NumberType} tile={descriptor.TileMode} " +
-            $"size={descriptor.Width}x{descriptor.Height} pitch={descriptor.Pitch} " +
-            $"dst=0x{descriptor.DstSelect:X3} " +
-            $"bytes={source.Length} nonzero64={nonZero}");
+            TraceAgcShader(
+                $"agc.texture_source addr=0x{descriptor.Address:X16} " +
+                $"fmt={descriptor.Format} num={descriptor.NumberType} tile={descriptor.TileMode} " +
+                $"size={descriptor.Width}x{descriptor.Height} pitch={descriptor.Pitch} " +
+                $"dst=0x{descriptor.DstSelect:X3} " +
+                $"bytes={source.Length} nonzero64={nonZero}");
+        }
 
         var rgba = source;
         texture = new VulkanGuestDrawTexture(
@@ -5701,6 +5865,16 @@ public static class AgcExports
         Console.Error.WriteLine($"[LOADER][TRACE] {message}");
     }
 
+    private static void TraceAgc(ref AgcTraceHandler message)
+    {
+        if (!_traceAgc)
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] {message.ToStringAndClear()}");
+    }
+
     private static void TraceAgcShader(string message)
     {
         if (!_traceAgcShader)
@@ -5711,6 +5885,62 @@ public static class AgcExports
         Console.Error.WriteLine($"[LOADER][TRACE] {message}");
     }
 
+    private static void TraceAgcShader(ref AgcShaderTraceHandler message)
+    {
+        if (!_traceAgcShader)
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] {message.ToStringAndClear()}");
+    }
+
+    [InterpolatedStringHandler]
+    internal ref struct AgcTraceHandler
+    {
+        private DefaultInterpolatedStringHandler _inner;
+
+        public AgcTraceHandler(int literalLength, int formattedCount, out bool isEnabled)
+        {
+            isEnabled = _traceAgc;
+            _inner = isEnabled
+                ? new DefaultInterpolatedStringHandler(literalLength, formattedCount)
+                : default;
+        }
+
+        public void AppendLiteral(string value) => _inner.AppendLiteral(value);
+
+        public void AppendFormatted<T>(T value) => _inner.AppendFormatted(value);
+
+        public void AppendFormatted<T>(T value, string? format) =>
+            _inner.AppendFormatted(value, format);
+
+        public string ToStringAndClear() => _inner.ToStringAndClear();
+    }
+
+    [InterpolatedStringHandler]
+    internal ref struct AgcShaderTraceHandler
+    {
+        private DefaultInterpolatedStringHandler _inner;
+
+        public AgcShaderTraceHandler(int literalLength, int formattedCount, out bool isEnabled)
+        {
+            isEnabled = _traceAgcShader;
+            _inner = isEnabled
+                ? new DefaultInterpolatedStringHandler(literalLength, formattedCount)
+                : default;
+        }
+
+        public void AppendLiteral(string value) => _inner.AppendLiteral(value);
+
+        public void AppendFormatted<T>(T value) => _inner.AppendFormatted(value);
+
+        public void AppendFormatted<T>(T value, string? format) =>
+            _inner.AppendFormatted(value, format);
+
+        public string ToStringAndClear() => _inner.ToStringAndClear();
+    }
+
     private static string FormatShaderDwords(IReadOnlyList<uint> values) =>
         values.Count == 0
             ? "none"

src/SharpEmu.Libs/Ajm/AjmExports.cs

@@ -0,0 +1,37 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Ajm;
+
+public static class AjmExports
+{
+    private const int OrbisAjmErrorInvalidParameter = unchecked((int)0x80930005);
+    private static int _nextContextId;
+
+    [SysAbiExport(
+        Nid = "dl+4eHSzUu4",
+        ExportName = "sceAjmInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceAjm")]
+    public static int Initialize(CpuContext ctx)
+    {
+        var reserved = ctx[CpuRegister.Rdi];
+        var outContextAddress = ctx[CpuRegister.Rsi];
+
+        // AJM requires a zero reserved argument.
+        if (reserved != 0 || outContextAddress == 0)
+        {
+            return ctx.SetReturn(OrbisAjmErrorInvalidParameter);
+        }
+
+        var contextId = unchecked((uint)Interlocked.Increment(ref _nextContextId));
+        if (!ctx.TryWriteUInt32(outContextAddress, contextId))
+        {
+            return ctx.SetReturn(OrbisAjmErrorInvalidParameter);
+        }
+
+        return ctx.SetReturn(0);
+    }
+}

src/SharpEmu.Libs/Audio/AudioOut2Exports.cs

@@ -202,7 +202,7 @@ public static class AudioOut2Exports
     {
         var userId = unchecked((int)ctx[CpuRegister.Rdi]);
         var outUserAddress = ctx[CpuRegister.Rsi];
-        if ((userId != 0 && userId != 1 && userId != 255) || outUserAddress == 0)
+        if ((userId != 0 && userId != 1 && userId != 1000 && userId != 255) || outUserAddress == 0)
         {
             return ctx.SetReturn((int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
         }

src/SharpEmu.Libs/HostTimerResolution.cs

@@ -0,0 +1,46 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+using System.Runtime.Versioning;
+
+namespace SharpEmu.Libs;
+
+public static class HostTimerResolution
+{
+    private const uint TargetPeriodMilliseconds = 1;
+
+    private static int _requested;
+
+    public static void Request()
+    {
+        if (Interlocked.Exchange(ref _requested, 1) != 0)
+        {
+            return;
+        }
+
+        if (!OperatingSystem.IsWindows())
+        {
+            return;
+        }
+
+        try
+        {
+            if (TimeBeginPeriod(TargetPeriodMilliseconds) != 0)
+            {
+                Console.Error.WriteLine(
+                    "[LOADER][WARN] Host timer resolution request rejected; " +
+                    "timed waits keep the default ~15.6 ms granularity.");
+            }
+        }
+        catch (DllNotFoundException exception)
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][WARN] Host timer resolution unavailable: {exception.Message}");
+        }
+    }
+
+    [SupportedOSPlatform("windows")]
+    [DllImport("winmm.dll", EntryPoint = "timeBeginPeriod", ExactSpelling = true)]
+    private static extern uint TimeBeginPeriod(uint uPeriod);
+}

src/SharpEmu.Libs/Json/JsonExports.cs

@@ -78,6 +78,58 @@ public static class JsonExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "WSOuge5IsCg",
+        ExportName = "_ZN3sce4Json15InitParameter2C2Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson2")]
+    public static int InitParameter2Constructor(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("InitParameter2.ctor", thisAddress, ctx[CpuRegister.Rsi]);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "I2QC8PYhJWY",
+        ExportName = "_ZN3sce4Json15InitParameter212setAllocatorERNS0_12MemAllocatorE",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson2")]
+    public static int InitParameter2SetAllocator(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("InitParameter2.setAllocator", thisAddress, ctx[CpuRegister.Rsi]);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "Eu95jmqn5Rw",
+        ExportName = "_ZN3sce4Json15InitParameter217setFileBufferSizeEm",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson2")]
+    public static int InitParameter2SetFileBufferSize(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("InitParameter2.setFileBufferSize", thisAddress, ctx[CpuRegister.Rsi]);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "IXW-z8pggfg",
+        ExportName = "_ZN3sce4Json12Initializer2C1Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson2")]
+    public static int Initializer2Constructor(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("Initializer2.ctor", thisAddress, ctx[CpuRegister.Rsi]);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     private static void TraceJson(string operation, ulong thisAddress, ulong argument)
     {
         if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_JSON"), "1", StringComparison.Ordinal))

src/SharpEmu.Libs/Kernel/KernelAprCompatExports.cs

@@ -80,14 +80,8 @@ public static class KernelAprCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
 
-        var resultAddress = ResolveWaitResultAddress(waitArg1, waitArg2, submission.ResultAddress);
-        if (resultAddress != 0 && !TryWriteAprResult(ctx, resultAddress))
-        {
-            TraceAprWaitFailure(ctx, "wait_result_fault", submissionId, submission.CommandBuffer, waitArg1, waitArg2);
-            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
-        }
-
-        TraceApr(ctx, "wait", submissionId, submission.CommandBuffer, waitArg1, resultAddress);
+        // Completion output was written when the command was submitted.
+        TraceApr(ctx, "wait", submissionId, submission.CommandBuffer, waitArg1, waitArg2);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -169,27 +163,6 @@ public static class KernelAprCompatExports
         return ctx.Memory.TryWrite(resultAddress, result);
     }
 
-    private static ulong ResolveWaitResultAddress(ulong waitArg1, ulong waitArg2, ulong submittedResultAddress)
-    {
-        if (waitArg2 == 0)
-        {
-            return submittedResultAddress;
-        }
-
-        if (IsAmprCompletionToken(waitArg1) && waitArg2 <= 0xFFFF)
-        {
-            return submittedResultAddress;
-        }
-
-        return waitArg2;
-    }
-
-    private static bool IsAmprCompletionToken(ulong value)
-    {
-        var tag = value >> 56;
-        return tag is 0x0C or 0x10;
-    }
-
     private static void TraceApr(
         CpuContext ctx,
         string operation,

src/SharpEmu.Libs/Kernel/KernelEventQueueCompatExports.cs

@@ -11,6 +11,7 @@ public static class KernelEventQueueCompatExports
 {
     private const int KernelEventSize = 0x20;
     public const short KernelEventFilterGraphics = -14;
+    public const short KernelEventFilterUser = -11;
     public const short KernelEventFilterAmpr = -16;
     public const short KernelEventFilterAmprSystem = -17;
 
@@ -89,8 +90,16 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelAddUserEventEdge(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "add_user_edge", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var registered = RegisterEvent(
+            handle,
+            ctx[CpuRegister.Rsi],
+            KernelEventFilterUser,
+            0);
+        TraceEventQueue(ctx, "add_user_edge", handle);
+        return registered
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -100,8 +109,16 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelAddUserEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "add_user", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var registered = RegisterEvent(
+            handle,
+            ctx[CpuRegister.Rsi],
+            KernelEventFilterUser,
+            0);
+        TraceEventQueue(ctx, "add_user", handle);
+        return registered
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -111,8 +128,15 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelDeleteUserEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "delete_user", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var deleted = DeleteRegisteredEvent(
+            handle,
+            ctx[CpuRegister.Rsi],
+            KernelEventFilterUser);
+        TraceEventQueue(ctx, "delete_user", handle);
+        return deleted
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -122,8 +146,18 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelTriggerUserEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "trigger_user", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var triggered = TriggerRegisteredEvent(
+            handle,
+            ctx[CpuRegister.Rsi],
+            KernelEventFilterUser,
+            flags: 0x21,
+            fflags: 0,
+            data: ctx[CpuRegister.Rdx]);
+        TraceEventQueue(ctx, "trigger_user", handle);
+        return triggered
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -475,6 +509,43 @@ public static class KernelEventQueueCompatExports
         return triggeredCount;
     }
 
+    private static bool TriggerRegisteredEvent(
+        ulong handle,
+        ulong ident,
+        short filter,
+        ushort flags,
+        uint fflags,
+        ulong data)
+    {
+        lock (_eventQueueGate)
+        {
+            if (!_registeredEvents.TryGetValue(handle, out var registrations) ||
+                !registrations.TryGetValue((ident, filter), out var registration))
+            {
+                return false;
+            }
+
+            if (!_pendingEvents.TryGetValue(handle, out var queue))
+            {
+                queue = new LinkedList<KernelQueuedEvent>();
+                _pendingEvents[handle] = queue;
+            }
+
+            QueueOrUpdateEvent(
+                queue,
+                new KernelQueuedEvent(
+                    registration.Ident,
+                    registration.Filter,
+                    flags,
+                    fflags,
+                    data,
+                    registration.UserData));
+        }
+
+        WakeEventQueue(handle);
+        return true;
+    }
+
     public static bool TriggerDisplayEvent(
         ulong handle,
         ulong ident,
@@ -656,6 +727,6 @@ public static class KernelEventQueueCompatExports
 
         _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out ulong returnRip);
         Console.Error.WriteLine(
-            $"[LOADER][TRACE] equeue.{operation}: handle=0x{handle:X16} rsi=0x{ctx[CpuRegister.Rsi]:X16} rdx=0x{ctx[CpuRegister.Rdx]:X16} ret=0x{returnRip:X16}");
+            $"[LOADER][TRACE] equeue.{operation}: thread=0x{KernelPthreadState.GetCurrentThreadHandle():X16} handle=0x{handle:X16} rsi=0x{ctx[CpuRegister.Rsi]:X16} rdx=0x{ctx[CpuRegister.Rdx]:X16} ret=0x{returnRip:X16}");
     }
 }

src/SharpEmu.Libs/Kernel/KernelPthreadCompatExports.cs

@@ -3,6 +3,7 @@
 
 using System.Collections.Concurrent;
 using SharpEmu.HLE;
+using SharpEmu.Libs.Fiber;
 using System.Diagnostics.CodeAnalysis;
 
 namespace SharpEmu.Libs.Kernel;
@@ -44,6 +45,8 @@ public static class KernelPthreadCompatExports
     // type initializer that first runs on a guest thread and fail-fasts the CLR.
     private static long _nextMutexWakeId;
     private static long _nextCondWakeId;
+    private static readonly ConcurrentDictionary<ulong, long> _posixTimedWaitCounts = new();
+    private static readonly ConcurrentDictionary<ulong, byte> _reportedContendedMutexes = new();
 
     private sealed class PthreadMutexState
     {
@@ -351,29 +354,95 @@ public static class KernelPthreadCompatExports
         LibraryName = "libKernel")]
     public static int PosixPthreadCondTimedwait(CpuContext ctx)
     {
-        // POSIX passes `const struct timespec *abstime` (absolute deadline), not relative microseconds (#113).
-        var abstimeAddress = ctx[CpuRegister.Rdx];
-        if (abstimeAddress == 0)
+        var absoluteTimeAddress = ctx[CpuRegister.Rdx];
+        if (absoluteTimeAddress == 0 ||
+            !ctx.TryReadUInt64(absoluteTimeAddress, out var secondsRaw) ||
+            !ctx.TryReadUInt64(absoluteTimeAddress + sizeof(ulong), out var nanosecondsRaw))
         {
-            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+            return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
         }
 
-        if (!ctx.TryReadUInt64(abstimeAddress, out var deadlineSeconds) ||
-            !ctx.TryReadUInt64(abstimeAddress + 8, out var deadlineNanoseconds))
+        var deadlineSeconds = unchecked((long)secondsRaw);
+        var deadlineNanoseconds = unchecked((long)nanosecondsRaw);
+        if (deadlineSeconds < 0 || deadlineNanoseconds is < 0 or >= 1_000_000_000)
         {
-            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
         }
 
-        var nowMicroseconds = DateTimeOffset.UtcNow.ToUnixTimeMilliseconds() * 1000L;
-        var deadlineMicroseconds =
-            deadlineSeconds >= long.MaxValue / 1_000_000UL
-                ? long.MaxValue
-                : (long)(deadlineSeconds * 1_000_000UL + Math.Min(deadlineNanoseconds, 999_999_999UL) / 1000UL);
-        var remainingMicroseconds = deadlineMicroseconds - nowMicroseconds;
-        var timeoutUsec = remainingMicroseconds <= 0
-            ? 1u
-            : (uint)Math.Min(remainingMicroseconds, uint.MaxValue);
-        return PthreadCondWaitCore(ctx, ctx[CpuRegister.Rdi], ctx[CpuRegister.Rsi], timed: true, timeoutUsec: timeoutUsec);
+        long nowSeconds;
+        long nowNanoseconds;
+        if (deadlineSeconds >= 1_000_000_000)
+        {
+            var now = DateTimeOffset.UtcNow;
+            nowSeconds = now.ToUnixTimeSeconds();
+            nowNanoseconds = (now.Ticks % TimeSpan.TicksPerSecond) * 100;
+        }
+        else
+        {
+            // Engine condition variables use monotonic deadlines.
+            KernelRuntimeCompatExports.GetProcessMonotonicTime(out nowSeconds, out nowNanoseconds);
+        }
+
+        var remainingNanoseconds =
+            ((Int128)deadlineSeconds * 1_000_000_000 + deadlineNanoseconds) -
+            ((Int128)nowSeconds * 1_000_000_000 + nowNanoseconds);
+        var timeoutUsec = remainingNanoseconds <= 0
+            ? 0u
+            : (uint)Int128.Min(uint.MaxValue, (remainingNanoseconds + 999) / 1_000);
+
+        TracePosixTimedWait(
+            ctx,
+            deadlineSeconds,
+            deadlineNanoseconds,
+            nowSeconds,
+            nowNanoseconds,
+            remainingNanoseconds,
+            timeoutUsec);
+
+        return PthreadCondWaitCore(
+            ctx,
+            ctx[CpuRegister.Rdi],
+            ctx[CpuRegister.Rsi],
+            timed: true,
+            timeoutUsec,
+            posixResult: true);
+    }
+
+    private static void TracePosixTimedWait(
+        CpuContext ctx,
+        long deadlineSeconds,
+        long deadlineNanoseconds,
+        long nowSeconds,
+        long nowNanoseconds,
+        Int128 remainingNanoseconds,
+        uint timeoutUsec)
+    {
+        var condAddress = ctx[CpuRegister.Rdi];
+        var count = _posixTimedWaitCounts.AddOrUpdate(condAddress, 1, static (_, current) => current + 1);
+        if (count != 1 && count % 1_000_000 != 0)
+        {
+            return;
+        }
+
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out var returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][DIAG] pthread_cond_timedwait#{count}: thread=0x{KernelPthreadState.GetCurrentThreadHandle():X16} " +
+            $"cond=0x{condAddress:X16} mutex=0x{ctx[CpuRegister.Rsi]:X16} abstime=0x{ctx[CpuRegister.Rdx]:X16} " +
+            $"deadline={deadlineSeconds}.{deadlineNanoseconds:D9} now={nowSeconds}.{nowNanoseconds:D9} " +
+            $"remaining_ns={remainingNanoseconds} timeout_us={timeoutUsec} ret=0x{returnRip:X16}");
+
+        if (GuestThreadExecution.Scheduler is not { } scheduler)
+        {
+            return;
+        }
+
+        foreach (var snapshot in scheduler.SnapshotThreads())
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][DIAG] guest_thread handle=0x{snapshot.ThreadHandle:X16} name='{snapshot.Name}' " +
+                $"state={snapshot.State} imports={snapshot.ImportCount} nid={snapshot.LastImportNid ?? "none"} " +
+                $"ret=0x{snapshot.LastReturnRip:X16} block={snapshot.BlockReason ?? "none"}");
+        }
     }
 
     [SysAbiExport(
@@ -654,11 +723,12 @@ public static class KernelPthreadCompatExports
         var acquired = state.Semaphore.Wait(0);
         if (!acquired)
         {
+            TraceContendedMutex(ctx, mutexAddress, resolvedAddress, state, currentThreadId);
             var waiter = new PthreadMutexWaiter { ThreadId = currentThreadId };
-            // Cooperative deschedule on a contended lock corrupts fiber state
-            // (Demon's Souls sceFiberSwitch -> ESRCH). Off by default: fall through to the
-            // synchronous host-thread wait below.
-            if (_enableMutexLockBlocking &&
+            // Fibers retain the synchronous fallback to preserve switch state.
+            var currentFiber = FiberExports.GetCurrentFiberAddressForDiagnostics(ctx);
+            var canCooperativelyBlock = _enableMutexLockBlocking || currentFiber == 0;
+            if (canCooperativelyBlock &&
                 !tryOnly &&
                 GuestThreadExecution.IsGuestThread &&
                 GuestThreadExecution.TryGetCurrentImportCallFrame(out _) &&
@@ -697,6 +767,53 @@ public static class KernelPthreadCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    private static void TraceContendedMutex(
+        CpuContext ctx,
+        ulong mutexAddress,
+        ulong resolvedAddress,
+        PthreadMutexState state,
+        ulong currentThreadId)
+    {
+        if (!_reportedContendedMutexes.TryAdd(resolvedAddress, 0))
+        {
+            return;
+        }
+
+        ulong ownerThreadId;
+        int recursionCount;
+        int type;
+        lock (state)
+        {
+            ownerThreadId = state.OwnerThreadId;
+            recursionCount = state.RecursionCount;
+            type = state.Type;
+        }
+
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out var returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][DIAG] pthread_mutex_contended: waiter=0x{currentThreadId:X16} owner=0x{ownerThreadId:X16} " +
+            $"mutex=0x{mutexAddress:X16} resolved=0x{resolvedAddress:X16} recursion={recursionCount} " +
+            $"type={type} semaphore_count={state.Semaphore.CurrentCount} ret=0x{returnRip:X16}");
+
+        if (GuestThreadExecution.Scheduler is not { } scheduler)
+        {
+            return;
+        }
+
+        foreach (var snapshot in scheduler.SnapshotThreads())
+        {
+            if (snapshot.ThreadHandle != currentThreadId && snapshot.ThreadHandle != ownerThreadId)
+            {
+                continue;
+            }
+
+            Console.Error.WriteLine(
+                $"[LOADER][DIAG] pthread_mutex_party handle=0x{snapshot.ThreadHandle:X16} name='{snapshot.Name}' " +
+                $"state={snapshot.State} imports={snapshot.ImportCount} nid={snapshot.LastImportNid ?? "none"} " +
+                $"ret=0x{snapshot.LastReturnRip:X16} block={snapshot.BlockReason ?? "none"}");
+        }
+    }
+
     private static int PthreadMutexUnlockCore(CpuContext ctx, ulong mutexAddress, bool requireOwner)
     {
         if (mutexAddress == 0)
@@ -1076,6 +1193,13 @@ public static class KernelPthreadCompatExports
 
         lock (_stateGate)
         {
+            if (_condStates.TryGetValue(condAddress, out var raced))
+            {
+                resolvedAddress = condAddress;
+                state = raced;
+                return true;
+            }
+
             _condStates[condAddress] = createdState;
             _condStates[handle] = createdState;
         }
@@ -1172,7 +1296,13 @@ public static class KernelPthreadCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
-    private static int PthreadCondWaitCore(CpuContext ctx, ulong condAddress, ulong mutexAddress, bool timed, uint timeoutUsec = 0)
+    private static int PthreadCondWaitCore(
+        CpuContext ctx,
+        ulong condAddress,
+        ulong mutexAddress,
+        bool timed,
+        uint timeoutUsec = 0,
+        bool posixResult = false)
     {
         if (condAddress == 0 || mutexAddress == 0)
         {
@@ -1210,9 +1340,10 @@ public static class KernelPthreadCompatExports
 
             if (consumedPendingSignal)
             {
-                // Leave SyncRoot before relocking the mutex to avoid lock-order inversion with cond-signal (#113).
                 state.Waiters = Math.Max(0, state.Waiters - 1);
                 TracePthreadCond("wait-wake-pending", condAddress, mutexAddress, state, timed, waitResult);
+
+                // Relock outside SyncRoot to preserve lock ordering.
                 Monitor.Exit(state.SyncRoot);
                 try
                 {
@@ -1221,6 +1352,7 @@ public static class KernelPthreadCompatExports
                 }
                 finally
                 {
+                    // Balance the surrounding lock statement.
                     Monitor.Enter(state.SyncRoot);
                 }
             }
@@ -1231,7 +1363,7 @@ public static class KernelPthreadCompatExports
                     ctx,
                     timed ? "pthread_cond_timedwait" : "pthread_cond_wait",
                     state.WakeKey,
-                    () => ResumePthreadCondWait(ctx, condAddress, mutexAddress, state, observedEpoch, timed, releasedRecursion),
+                    () => ResumePthreadCondWait(ctx, condAddress, mutexAddress, state, observedEpoch, timed, releasedRecursion, posixResult),
                     () => state.SignalEpoch != observedEpoch,
                     timed ? GuestThreadExecution.ComputeDeadlineTimestamp(GetCondWaitTimeout(timeoutUsec)) : 0))
             {
@@ -1315,7 +1447,7 @@ public static class KernelPthreadCompatExports
             state,
             timed,
             waitResult);
-        return waitResult;
+        return TranslatePthreadResult(waitResult, posixResult);
     }
 
     private static int PthreadCondSignalCore(CpuContext ctx, ulong condAddress, bool broadcast)
@@ -1372,7 +1504,8 @@ public static class KernelPthreadCompatExports
         PthreadCondState state,
         ulong observedEpoch,
         bool timed,
-        int releasedRecursion)
+        int releasedRecursion,
+        bool posixResult)
     {
         var waitResult = (int)OrbisGen2Result.ORBIS_GEN2_OK;
         lock (state.SyncRoot)
@@ -1393,7 +1526,32 @@ public static class KernelPthreadCompatExports
         }
 
         var lockResult = PthreadMutexRelockForCondWait(ctx, mutexAddress, releasedRecursion);
-        return lockResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? waitResult : lockResult;
+        return TranslatePthreadResult(
+            lockResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? waitResult : lockResult,
+            posixResult);
+    }
+
+    private static int TranslatePthreadResult(int result, bool posixResult)
+    {
+        if (!posixResult || result == (int)OrbisGen2Result.ORBIS_GEN2_OK)
+        {
+            return result;
+        }
+
+        return result switch
+        {
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_PERMISSION_DENIED => 1, // EPERM
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND => 2,        // ENOENT
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT => 22, // EINVAL
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_DEADLOCK => 11,        // EDEADLK
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_DELETED => 13,         // EACCES
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY => 16,            // EBUSY
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN => 35,       // EAGAIN
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_TIMED_OUT => 60,       // ETIMEDOUT
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_CANCELED => 85,        // ECANCELED
+            (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT => 14,    // EFAULT
+            _ => result
+        };
     }
 
     private static string GetCondWakeKey(ulong resolvedCondAddress) =>
@@ -1559,7 +1717,10 @@ public static class KernelPthreadCompatExports
             return TimeSpan.Zero;
         }
 
-        return TimeSpan.FromTicks((long)timeoutUsec * 10L);
+        // Round positive sub-millisecond waits to the host sleep quantum.
+        return timeoutUsec < 1_000
+            ? TimeSpan.FromMilliseconds(1)
+            : TimeSpan.FromTicks((long)timeoutUsec * 10L);
     }
 
     private static TimeSpan GetCondSpuriousWakeTimeout()
@@ -1701,7 +1862,8 @@ public static class KernelPthreadCompatExports
         }
 
         Console.Error.WriteLine(
-            $"[LOADER][TRACE] pthread_cond_{operation}: cond=0x{condAddress:X16} mutex=0x{mutexAddress:X16} " +
+            $"[LOADER][TRACE] pthread_cond_{operation}: thread=0x{KernelPthreadState.GetCurrentThreadHandle():X16} " +
+            $"cond=0x{condAddress:X16} key={state?.WakeKey ?? "none"} mutex=0x{mutexAddress:X16} " +
             $"waiters={(state?.Waiters ?? 0)} epoch=0x{(state?.SignalEpoch ?? 0):X} timed={timed} result=0x{unchecked((uint)result):X8}");
     }
 

src/SharpEmu.Libs/Kernel/KernelRuntimeCompatExports.cs

@@ -213,6 +213,13 @@ public static class KernelRuntimeCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "QcteRwbsnV0",
+        ExportName = "usleep",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixUsleep(CpuContext ctx) => KernelUsleep(ctx);
+
     private static void TraceUsleepSpin(CpuContext ctx, ulong micros)
     {
         if (!_traceUsleep)

src/SharpEmu.Libs/Np/NpManagerExports.cs

@@ -53,8 +53,19 @@ public static class NpManagerExports
         LibraryName = "libSceNpManager")]
     public static int NpGetOnlineId(CpuContext ctx)
     {
-        ctx[CpuRegister.Rax] = 0;
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        // Gen5 ABI: user ID, then output structure.
+        return WriteOfflineOnlineId(ctx, ctx[CpuRegister.Rsi]);
+    }
+
+    [SysAbiExport(
+        Nid = "rbknaUjpqWo",
+        ExportName = "sceNpGetOnlineIdA",
+        Target = Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpGetOnlineIdA(CpuContext ctx)
+    {
+        // The A variant takes a user ID before OnlineId.
+        return WriteOfflineOnlineId(ctx, ctx[CpuRegister.Rsi]);
     }
 
     [SysAbiExport(
@@ -169,4 +180,19 @@ public static class NpManagerExports
 
         Console.Error.WriteLine($"[LOADER][TRACE] np.{message}");
     }
+
+    private static int WriteOfflineOnlineId(CpuContext ctx, ulong address)
+    {
+        if (address == 0)
+        {
+            return ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        // SceNpOnlineId is a 16-byte handle plus four trailing bytes.
+        Span<byte> onlineId = stackalloc byte[20];
+        "Player"u8.CopyTo(onlineId);
+        return ctx.Memory.TryWrite(address, onlineId)
+            ? ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_OK)
+            : ctx.SetReturn(OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
 }

src/SharpEmu.Libs/Np/NpWebApi2Exports.cs

@@ -31,6 +31,18 @@ public static class NpWebApi2Exports
         return ctx.SetReturn(0);
     }
 
+    [SysAbiExport(
+        Nid = "WV1GwM32NgY",
+        ExportName = "sceNpWebApi2InitializeForToolkit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpWebApi2")]
+    public static int NpWebApi2InitializeAlt(CpuContext ctx)
+    {
+        Interlocked.Exchange(ref _initialized, 1);
+        TraceNpWebApi2("init-alt", unchecked((int)ctx[CpuRegister.Rdi]), ctx[CpuRegister.Rsi]);
+        return ctx.SetReturn(0);
+    }
+
     [SysAbiExport(
         Nid = "bEvXpcEk200",
         ExportName = "sceNpWebApi2Terminate",

src/SharpEmu.Libs/Pad/PadExports.cs

@@ -14,11 +14,18 @@ public static class PadExports
     private const int OrbisPadErrorNotInitialized = unchecked((int)0x80920005);
     private const int OrbisPadErrorDeviceNotConnected = unchecked((int)0x80920007);
     private const int OrbisPadErrorDeviceNoHandle = unchecked((int)0x80920008);
-    private const int PrimaryUserId = 1;
+    private const int PrimaryUserId = 1000;
     private const int StandardPortType = 0;
     private const int PrimaryPadHandle = 1;
     private const int ControllerInformationSize = 0x1C;
     private const int PadDataSize = 0x78;
+    private static readonly long InputSampleIntervalTicks = Math.Max(1, Stopwatch.Frequency / 1000);
+
+    [ThreadStatic]
+    private static long _lastInputSampleTicks;
+
+    [ThreadStatic]
+    private static PadState _cachedInputState;
 
     private static bool _initialized;
 
@@ -270,6 +277,44 @@ public static class PadExports
     {
         Span<byte> data = stackalloc byte[PadDataSize];
         data.Clear();
+        var input = ReadHostInputState();
+        var buttons = input.Buttons;
+        var leftX = input.LeftX;
+        var leftY = input.LeftY;
+        var rightX = input.RightX;
+        var rightY = input.RightY;
+        var l2 = input.L2;
+        var r2 = input.R2;
+
+        BinaryPrimitives.WriteUInt32LittleEndian(data[0x00..], buttons);
+        data[0x04] = leftX;
+        data[0x05] = leftY;
+        data[0x06] = rightX;
+        data[0x07] = rightY;
+        data[0x08] = l2;
+        data[0x09] = r2;
+        BinaryPrimitives.WriteSingleLittleEndian(data[0x18..], 1.0f);
+        data[0x4C] = 1;
+        var timestampTicks = Stopwatch.GetTimestamp();
+        var timestampMicroseconds =
+            ((ulong)(timestampTicks / Stopwatch.Frequency) * 1_000_000UL) +
+            ((ulong)(timestampTicks % Stopwatch.Frequency) * 1_000_000UL / (ulong)Stopwatch.Frequency);
+        BinaryPrimitives.WriteUInt64LittleEndian(
+            data[0x50..],
+            timestampMicroseconds);
+        data[0x68] = 1;
+
+        return ctx.Memory.TryWrite(dataAddress, data);
+    }
+
+    private static PadState ReadHostInputState()
+    {
+        var now = Stopwatch.GetTimestamp();
+        if (_lastInputSampleTicks != 0 && now - _lastInputSampleTicks < InputSampleIntervalTicks)
+        {
+            return _cachedInputState;
+        }
+
         var acceptsKeyboardInput = IsEmulatorWindowFocused();
         var buttons = acceptsKeyboardInput ? ReadKeyboardButtons() : 0;
         var leftX = acceptsKeyboardInput ? ReadAnalogStick(IsKeyDown(0x41), IsKeyDown(0x44)) : (byte)128;
@@ -303,25 +348,17 @@ public static class PadExports
             r2 = Math.Max(r2, xpad.R2);
         }
 
-        BinaryPrimitives.WriteUInt32LittleEndian(data[0x00..], buttons);
-        data[0x04] = leftX;
-        data[0x05] = leftY;
-        data[0x06] = rightX;
-        data[0x07] = rightY;
-        data[0x08] = l2;
-        data[0x09] = r2;
-        BinaryPrimitives.WriteSingleLittleEndian(data[0x18..], 1.0f);
-        data[0x4C] = 1;
-        var timestampTicks = Stopwatch.GetTimestamp();
-        var timestampMicroseconds =
-            ((ulong)(timestampTicks / Stopwatch.Frequency) * 1_000_000UL) +
-            ((ulong)(timestampTicks % Stopwatch.Frequency) * 1_000_000UL / (ulong)Stopwatch.Frequency);
-        BinaryPrimitives.WriteUInt64LittleEndian(
-            data[0x50..],
-            timestampMicroseconds);
-        data[0x68] = 1;
-
-        return ctx.Memory.TryWrite(dataAddress, data);
+        _cachedInputState = new PadState(
+            Connected: true,
+            Buttons: buttons,
+            LeftX: leftX,
+            LeftY: leftY,
+            RightX: rightX,
+            RightY: rightY,
+            L2: l2,
+            R2: r2);
+        _lastInputSampleTicks = now;
+        return _cachedInputState;
     }
 
     [DllImport("user32.dll")]

src/SharpEmu.Libs/PlayGo/PlayGoExports.cs

@@ -682,7 +682,9 @@ public static class PlayGoExports
         var hasMetadata = File.Exists(playGoDat) || File.Exists(scenarioJson) || File.Exists(chunkDefsXml);
         if (!hasMetadata)
         {
-            return PlayGoMetadata.Empty;
+            // Full installs may omit PlayGo sidecar metadata.
+            TracePlayGo("metadata_missing; using fully-installed default chunk");
+            return new PlayGoMetadata(true, [(ushort)0]);
         }
 
         var chunkIds = LoadChunkIds(chunkDefsXml);

src/SharpEmu.Libs/SaveData/SaveDataExports.cs

@@ -244,7 +244,6 @@ public static class SaveDataExports
         }
     }
 
-    private static int _nextTransactionResource;
     [SysAbiExport(
         Nid = "gjRZNnw0JPE",
         ExportName = "sceSaveDataCreateTransactionResource",
@@ -252,24 +251,25 @@ public static class SaveDataExports
         LibraryName = "libSceSaveData")]
     public static int SaveDataCreateTransactionResource(CpuContext ctx)
     {
-        var userId = unchecked((int)ctx[CpuRegister.Rdi]);
-        var reserved = ctx[CpuRegister.Rsi];
-        var resourceAddress = ctx[CpuRegister.Rdx];
+        // Gen5 ABI: memory size, resource output, reserved.
+        var memorySize = ctx[CpuRegister.Rdi];
+        var resourceAddress = ctx[CpuRegister.Rsi];
+        var reserved = ctx[CpuRegister.Rdx];
 
         if (resourceAddress == 0)
         {
             return ctx.SetReturn(OrbisSaveDataErrorParameter);
         }
 
-        var id = (uint)Interlocked.Increment(ref _nextTransactionResource);
-
-        if (!ctx.TryWriteUInt32(resourceAddress, id))
+        // Offline HLE operations carry no transaction state.
+        if (!ctx.TryWriteUInt64(resourceAddress, 0))
         {
             return ctx.SetReturn((int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
         }
 
         TraceSaveData(
-            $"create_transaction_resource user={userId} reserved=0x{reserved:X} resource_addr=0x{resourceAddress:X} id={id}");
+            $"create_transaction_resource memory_size=0x{memorySize:X} reserved=0x{reserved:X} " +
+            $"resource_addr=0x{resourceAddress:X} resource=0x0");
 
         return ctx.SetReturn(0);
     }

src/SharpEmu.Libs/UserService/UserServiceExports.cs

@@ -13,7 +13,8 @@ public static class UserServiceExports
     private const int OrbisUserServiceErrorNoEvent = unchecked((int)0x80960007);
     private const int OrbisUserServiceErrorInvalidParameter = unchecked((int)0x80960009);
     private const int OrbisUserServiceErrorBufferTooShort = unchecked((int)0x8096000A);
-    private const int PrimaryUserId = 1;
+    // Retail user-service IDs begin at 0x3E8.
+    private const int PrimaryUserId = 1000;
     private const int InvalidUserId = -1;
     private const string PrimaryUserName = "SharpEmu";
     private static int _loginEventDelivered;
@@ -111,7 +112,8 @@ public static class UserServiceExports
         var userId = unchecked((int)ctx[CpuRegister.Rdi]);
         var nameAddress = ctx[CpuRegister.Rsi];
         var capacity = ctx[CpuRegister.Rdx];
-        if (userId != PrimaryUserId)
+        // Zero selects the current user.
+        if (userId != 0 && userId != PrimaryUserId)
         {
             return ctx.SetReturn(OrbisUserServiceErrorInvalidParameter);
         }
@@ -137,6 +139,28 @@ public static class UserServiceExports
         return ctx.SetReturn(result);
     }
 
+    [SysAbiExport(
+        Nid = "-sD02mFDBh4",
+        ExportName = "sceUserServiceGetGamePresets",
+        Target = Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetGamePresets(CpuContext ctx)
+    {
+        // Return deterministic defaults for the offline profile.
+        var userId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var resultAddress = ctx[CpuRegister.Rcx];
+        Span<byte> defaults = stackalloc byte[0x18];
+        if (resultAddress == 0 || !ctx.Memory.TryWrite(resultAddress, defaults))
+        {
+            return ctx.SetReturn(OrbisUserServiceErrorInvalidArgument);
+        }
+
+        TraceUserService(
+            $"get_game_presets user={userId} title=0x{ctx[CpuRegister.Rsi]:X16} " +
+            $"key=0x{ctx[CpuRegister.R9]:X} out=0x{resultAddress:X16} result=0x00000000");
+        return ctx.SetReturn(0);
+    }
+
     [SysAbiExport(
         Nid = "D-CzAxQL0XI",
         ExportName = "sceUserServiceGetPlatformPrivacySetting",

src/SharpEmu.Libs/VideoOut/VideoOutExports.cs

@@ -48,54 +48,106 @@ public static class VideoOutExports
     private static readonly Dictionary<int, VideoOutPortState> _ports = new();
 
     // Hardware raises vblank autonomously; UE blocks its frame loop on it.
-    private static Timer? _vblankPumpTimer;
-    private static int _vblankPumpActive;
-    private const int VblankPumpIntervalMs = 16;
+    private const double VblankHz = 60.0;
+    private const int VblankWaitTimeoutMilliseconds = 100;
+    private static Thread? _vblankPumpThread;
+    private static int _vblankPumpStarted;
+
+    private static readonly object _vblankEdgeGate = new();
+    private static ulong _vblankEdgeSequence;
+    private static long _vblankMissedEdges;
 
     private static void EnsureVblankPumpStarted()
     {
-        if (_vblankPumpTimer is not null)
+        if (Interlocked.Exchange(ref _vblankPumpStarted, 1) != 0)
         {
             return;
         }
 
-        _vblankPumpTimer = new Timer(
-            static _ => PumpVblanks(),
-            null,
-            VblankPumpIntervalMs,
-            VblankPumpIntervalMs);
+        HostTimerResolution.Request();
+
+        _vblankPumpThread = new Thread(VblankPumpLoop)
+        {
+            IsBackground = true,
+            Name = "SharpEmu VideoOut vblank",
+            Priority = ThreadPriority.AboveNormal,
+        };
+        _vblankPumpThread.Start();
+    }
+
+    private static void VblankPumpLoop()
+    {
+        var intervalTicks = Math.Max(1L, (long)(Stopwatch.Frequency / VblankHz));
+        var nextEdge = Stopwatch.GetTimestamp() + intervalTicks;
+
+        while (true)
+        {
+            WaitUntilTimestamp(nextEdge);
+            PumpVblanks();
+
+            nextEdge += intervalTicks;
+
+            var now = Stopwatch.GetTimestamp();
+            if (nextEdge < now)
+            {
+                var missed = (now - nextEdge) / intervalTicks + 1;
+                Interlocked.Add(ref _vblankMissedEdges, missed);
+                nextEdge = now + intervalTicks;
+            }
+        }
+    }
+
+    private static void WaitUntilTimestamp(long deadlineTicks)
+    {
+        var spinThresholdTicks = Stopwatch.Frequency * 2L / 1000L;
+
+        while (true)
+        {
+            var remaining = deadlineTicks - Stopwatch.GetTimestamp();
+            if (remaining <= 0)
+            {
+                return;
+            }
+
+            if (remaining > spinThresholdTicks)
+            {
+                var sleepMilliseconds =
+                    (int)((remaining - spinThresholdTicks) * 1000L / Stopwatch.Frequency);
+                if (sleepMilliseconds > 0)
+                {
+                    Thread.Sleep(sleepMilliseconds);
+                    continue;
+                }
+            }
+
+            Thread.SpinWait(64);
+        }
     }
 
     private static void PumpVblanks()
     {
-        if (Interlocked.Exchange(ref _vblankPumpActive, 1) != 0)
+        lock (_vblankEdgeGate)
         {
-            return;
+            _vblankEdgeSequence++;
+            Monitor.PulseAll(_vblankEdgeGate);
         }
 
-        try
+        VideoOutPortState[] ports;
+        lock (_stateGate)
         {
-            VideoOutPortState[] ports;
-            lock (_stateGate)
+            if (_ports.Count == 0)
             {
-                if (_ports.Count == 0)
-                {
-                    return;
-                }
-
-                // Signalling reaches WakeBlockedThreads -> Pump(), which serialises on one global
-                // flag. Waking an unwatched queue would hold it 60x/sec and starve guest threads.
-                ports = _ports.Values.Where(static port => port.VblankEvents.Count != 0).ToArray();
+                return;
             }
 
-            foreach (var port in ports)
-            {
-                SignalVblank(port);
-            }
+            // Signalling reaches WakeBlockedThreads -> Pump(), which serialises on one global
+            // flag. Waking an unwatched queue would hold it 60x/sec and starve guest threads.
+            ports = _ports.Values.Where(static port => port.VblankEvents.Count != 0).ToArray();
         }
-        finally
+
+        foreach (var port in ports)
         {
-            Volatile.Write(ref _vblankPumpActive, 0);
+            SignalVblank(port);
         }
     }
 
@@ -103,6 +155,7 @@ public static class VideoOutExports
     private static int _nextHandle = 1;
     private static int _frameDumpCount;
     private static long _nextFrameDumpIndex;
+    private static string _windowTitleBase = "SharpEmu VideoOut";
     private static string _windowTitle = "SharpEmu VideoOut";
     private static readonly bool _logFrameRate = string.Equals(
         Environment.GetEnvironmentVariable("SHARPEMU_LOG_VIDEOOUT_FPS"),
@@ -112,6 +165,10 @@ public static class VideoOutExports
         Environment.GetEnvironmentVariable("SHARPEMU_LOG_VIDEOOUT_SYNC"),
         "1",
         StringComparison.Ordinal);
+    private static readonly bool _dumpVideoOut = string.Equals(
+        Environment.GetEnvironmentVariable("SHARPEMU_DUMP_VIDEOOUT"),
+        "1",
+        StringComparison.Ordinal);
     private static long _frameRateWindowStart = Stopwatch.GetTimestamp();
     private static long _submittedFrameCount;
     private static long _presentedFrameCount;
@@ -134,10 +191,24 @@ public static class VideoOutExports
         var application = parts.Count == 0 ? "VideoOut" : string.Join(' ', parts);
         var versionSuffix = string.IsNullOrWhiteSpace(version) ? string.Empty : $" v{version.Trim()}";
         var commitSuffix = string.IsNullOrWhiteSpace(emulatorCommitSha) ? string.Empty : $" \u00b7 {emulatorCommitSha.Trim()}";
-        var hardwareSuffix = $" \u00b7 {HostSystemInfo.CpuName} \u00b7 {HostSystemInfo.GpuName}";
+        var hardwareSuffix = $" \u00b7 {HostSystemInfo.CpuName}";
         lock (_stateGate)
         {
-            _windowTitle = $"SharpEmu{commitSuffix} - {application}{versionSuffix}{hardwareSuffix}";
+            _windowTitleBase = $"SharpEmu{commitSuffix} - {application}{versionSuffix}{hardwareSuffix}";
+            _windowTitle = $"{_windowTitleBase} \u00b7 {HostSystemInfo.GpuName}";
+        }
+    }
+
+    internal static void SetSelectedGpuName(string gpuName)
+    {
+        if (string.IsNullOrWhiteSpace(gpuName))
+        {
+            return;
+        }
+
+        lock (_stateGate)
+        {
+            _windowTitle = $"{_windowTitleBase} \u00b7 {gpuName.Trim()}";
         }
     }
 
@@ -376,7 +447,20 @@ public static class VideoOutExports
             return OrbisVideoOutErrorInvalidHandle;
         }
 
-        Thread.Sleep(1);
+        EnsureVblankPumpStarted();
+
+        lock (_vblankEdgeGate)
+        {
+            var entryEdge = _vblankEdgeSequence;
+            while (_vblankEdgeSequence == entryEdge)
+            {
+                if (!Monitor.Wait(_vblankEdgeGate, VblankWaitTimeoutMilliseconds))
+                {
+                    break;
+                }
+            }
+        }
+
         SignalVblank(port);
 
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -987,10 +1071,7 @@ public static class VideoOutExports
                 displayBuffer.PitchInPixel);
         }
 
-        if (string.Equals(
-                Environment.GetEnvironmentVariable("SHARPEMU_DUMP_VIDEOOUT"),
-                "1",
-                StringComparison.Ordinal))
+        if (_dumpVideoOut)
         {
             _ = TryDumpFrame(ctx, port, bufferIndex, flipMode, flipArg);
         }
@@ -1050,9 +1131,11 @@ public static class VideoOutExports
         var elapsedSeconds = (double)elapsedTicks / Stopwatch.Frequency;
         var submitted = Interlocked.Exchange(ref _submittedFrameCount, 0);
         var presentedCount = Interlocked.Exchange(ref _presentedFrameCount, 0);
+        var missedEdges = Interlocked.Exchange(ref _vblankMissedEdges, 0);
         Console.Error.WriteLine(
             $"[LOADER][PERF] videoout submitted_fps={submitted / elapsedSeconds:F1} " +
-            $"presented_fps={presentedCount / elapsedSeconds:F1}");
+            $"presented_fps={presentedCount / elapsedSeconds:F1} " +
+            $"vblank_missed={missedEdges}");
     }
 
     private static int RegisterBufferRange(VideoOutPortState port, int startIndex, ReadOnlySpan<ulong> addresses, BufferAttribute attribute, int requestedGroupIndex = -1)

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.cs

@@ -11,6 +11,7 @@ using Silk.NET.Vulkan.Extensions.EXT;
 using Silk.NET.Windowing;
 using System.Numerics;
 using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 using System.Security.Cryptography;
 using System.Text;
 using VkBuffer = Silk.NET.Vulkan.Buffer;
@@ -600,7 +601,8 @@ internal static unsafe class VulkanVideoPresenter
         var traceSubmission = false;
         lock (_gate)
         {
-            var known = _availableGuestImages.ContainsKey(address);
+            // VideoOut registration does not imply a rendered Vulkan image.
+            var known = _gpuGuestImages.ContainsKey(address);
             if (ShouldTracePresentedGuestImageContentsForDiagnostics())
             {
                 Console.Error.WriteLine(
@@ -622,7 +624,7 @@ internal static unsafe class VulkanVideoPresenter
                 {
                     Console.Error.WriteLine(
                         $"[LOADER][WARN] vk.submit_guest_image_unknown addr=0x{address:X16} " +
-                        $"{width}x{height} - flip target was never registered as a render output");
+                        $"{width}x{height} - flip target has no completed GPU render output");
                 }
 
                 return false;
@@ -933,6 +935,8 @@ internal static unsafe class VulkanVideoPresenter
         private ExtDebugUtils? _debugUtils;
         private PhysicalDevice _physicalDevice;
         private Device _device;
+        private PipelineCache _pipelineCache;
+        private string? _pipelineCachePath;
         private Queue _queue;
         private uint _queueFamilyIndex;
         private SwapchainKHR _swapchain;
@@ -950,6 +954,9 @@ internal static unsafe class VulkanVideoPresenter
         private CommandBuffer _presentationCommandBuffer;
         private VkSemaphore _imageAvailable;
         private VkSemaphore _renderFinished;
+        private Fence _presentationFence;
+        private bool _presentationInFlight;
+        private TranslatedDrawResources? _pendingPresentationResources;
         private VkBuffer _stagingBuffer;
         private DeviceMemory _stagingMemory;
         private ulong _stagingSize;
@@ -966,6 +973,10 @@ internal static unsafe class VulkanVideoPresenter
         private int _directPresentationCount;
         private readonly Dictionary<ulong, GuestImageResource> _guestImages = new();
         private readonly HashSet<(ulong Address, uint Width, uint Height, Format Format)> _tracedTextureCacheHits = new();
+        private static readonly bool _dumpTextures = string.Equals(
+            Environment.GetEnvironmentVariable("SHARPEMU_DUMP_TEXTURES"),
+            "1",
+            StringComparison.Ordinal);
         private readonly HashSet<(ulong Address, uint Width, uint Height, Format Format)> _tracedTextureUploads = new();
         private readonly HashSet<(ulong Address, uint Width, uint Height, uint Format)> _dumpedTextures = new();
         private readonly HashSet<(ulong Address, int Size)> _tracedGlobalBuffers = new();
@@ -988,7 +999,7 @@ internal static unsafe class VulkanVideoPresenter
         private readonly record struct GraphicsPipelineKey(
             string VertexShader,
             string FragmentShader,
-            ulong RenderPass,
+            Format RenderTargetFormat,
             PrimitiveTopology Topology,
             VulkanGuestBlendState Blend,
             string ResourceLayout,
@@ -1150,6 +1161,7 @@ internal static unsafe class VulkanVideoPresenter
             CreateSurface();
             SelectPhysicalDevice();
             CreateDevice();
+            CreatePipelineCache();
             CreateSwapchain();
             CreateCommandResources();
             CreateGuestDrawResources();
@@ -1559,8 +1571,11 @@ internal static unsafe class VulkanVideoPresenter
             }
 
             _vk.GetPhysicalDeviceProperties(_physicalDevice, out var selected);
+            var selectedName = SilkMarshal.PtrToString((nint)selected.DeviceName) ?? "unknown";
             Console.Error.WriteLine(
-                $"[LOADER][INFO] Vulkan device: {SilkMarshal.PtrToString((nint)selected.DeviceName)} ({selected.DeviceType})");
+                $"[LOADER][INFO] Vulkan device: {selectedName} ({selected.DeviceType})");
+            VideoOutExports.SetSelectedGpuName(selectedName);
+            _window.Title = VideoOutExports.GetWindowTitle();
         }
 
         private static int ScorePhysicalDevice(
@@ -1743,6 +1758,110 @@ internal static unsafe class VulkanVideoPresenter
             }
         }
 
+        private void CreatePipelineCache()
+        {
+            _vk.GetPhysicalDeviceProperties(_physicalDevice, out var properties);
+            var directory = Path.Combine(AppContext.BaseDirectory, "pipeline-cache");
+            _pipelineCachePath = Path.Combine(
+                directory,
+                $"vulkan-{properties.VendorID:X4}-{properties.DeviceID:X4}-{properties.DriverVersion:X8}.bin");
+
+            byte[] initialData = [];
+            try
+            {
+                if (File.Exists(_pipelineCachePath))
+                {
+                    initialData = File.ReadAllBytes(_pipelineCachePath);
+                }
+            }
+            catch (IOException exception)
+            {
+                Console.Error.WriteLine(
+                    $"[LOADER][WARN] Vulkan pipeline cache could not be read: {exception.Message}");
+            }
+
+            fixed (byte* initialDataPointer = initialData)
+            {
+                var createInfo = new PipelineCacheCreateInfo
+                {
+                    SType = StructureType.PipelineCacheCreateInfo,
+                    InitialDataSize = (nuint)initialData.Length,
+                    PInitialData = initialData.Length == 0 ? null : initialDataPointer,
+                };
+                var result = _vk.CreatePipelineCache(
+                    _device,
+                    &createInfo,
+                    null,
+                    out _pipelineCache);
+                if (result != Result.Success && initialData.Length != 0)
+                {
+                    createInfo.InitialDataSize = 0;
+                    createInfo.PInitialData = null;
+                    Check(
+                        _vk.CreatePipelineCache(
+                            _device,
+                            &createInfo,
+                            null,
+                            out _pipelineCache),
+                        "vkCreatePipelineCache(empty)");
+                    initialData = [];
+                }
+                else
+                {
+                    Check(result, "vkCreatePipelineCache");
+                }
+            }
+
+            Console.Error.WriteLine(
+                $"[LOADER][INFO] Vulkan pipeline cache: " +
+                $"{(initialData.Length == 0 ? "new" : $"loaded {initialData.Length} bytes")}");
+        }
+
+        private void SavePipelineCache()
+        {
+            if (_pipelineCache.Handle == 0 || string.IsNullOrWhiteSpace(_pipelineCachePath))
+            {
+                return;
+            }
+
+            try
+            {
+                nuint size = 0;
+                Check(
+                    _vk.GetPipelineCacheData(_device, _pipelineCache, &size, null),
+                    "vkGetPipelineCacheData(size)");
+                if (size == 0 || size > int.MaxValue)
+                {
+                    return;
+                }
+
+                var data = new byte[(int)size];
+                fixed (byte* dataPointer = data)
+                {
+                    Check(
+                        _vk.GetPipelineCacheData(
+                            _device,
+                            _pipelineCache,
+                            &size,
+                            dataPointer),
+                        "vkGetPipelineCacheData");
+                }
+
+                var directory = Path.GetDirectoryName(_pipelineCachePath);
+                if (!string.IsNullOrWhiteSpace(directory))
+                {
+                    Directory.CreateDirectory(directory);
+                }
+                File.WriteAllBytes(_pipelineCachePath, data.AsSpan(0, (int)size).ToArray());
+            }
+            catch (Exception exception) when (
+                exception is IOException or UnauthorizedAccessException or InvalidOperationException)
+            {
+                Console.Error.WriteLine(
+                    $"[LOADER][WARN] Vulkan pipeline cache could not be saved: {exception.Message}");
+            }
+        }
+
         private void CreateSwapchain()
         {
             Check(
@@ -1835,9 +1954,44 @@ internal static unsafe class VulkanVideoPresenter
             Check(_vk.CreateSemaphore(_device, &semaphoreInfo, null, out _imageAvailable), "vkCreateSemaphore");
             Check(_vk.CreateSemaphore(_device, &semaphoreInfo, null, out _renderFinished), "vkCreateSemaphore");
 
+            var fenceInfo = new FenceCreateInfo
+            {
+                SType = StructureType.FenceCreateInfo,
+            };
+            Check(
+                _vk.CreateFence(_device, &fenceInfo, null, out _presentationFence),
+                "vkCreateFence(presentation)");
+
             CreateStagingBuffer((ulong)_extent.Width * _extent.Height * 4);
         }
 
+        private void CompletePendingPresentation(bool wait)
+        {
+            if (!_presentationInFlight)
+            {
+                return;
+            }
+
+            var fence = _presentationFence;
+            var result = wait
+                ? _vk.WaitForFences(_device, 1, &fence, true, ulong.MaxValue)
+                : _vk.GetFenceStatus(_device, fence);
+            if (!wait && result == Result.NotReady)
+            {
+                return;
+            }
+
+            Check(result, "vkWaitForFences(presentation)");
+            _presentationInFlight = false;
+            if (_pendingPresentationResources is not null)
+            {
+                MarkSampledImagesInitialized(_pendingPresentationResources);
+                MarkStorageImagesInitialized(_pendingPresentationResources);
+                DestroyTranslatedDrawResources(_pendingPresentationResources);
+                _pendingPresentationResources = null;
+            }
+        }
+
         private CommandBuffer AllocateGuestCommandBuffer()
         {
             var allocateInfo = new CommandBufferAllocateInfo
@@ -2184,7 +2338,7 @@ internal static unsafe class VulkanVideoPresenter
                 Check(
                     _vk.CreateGraphicsPipelines(
                         _device,
-                        default,
+                        _pipelineCache,
                         1,
                         &pipelineInfo,
                         null,
@@ -2219,6 +2373,7 @@ internal static unsafe class VulkanVideoPresenter
         private TranslatedDrawResources CreateTranslatedDrawResources(
             VulkanTranslatedGuestDraw draw,
             RenderPass renderPass,
+            Format renderTargetFormat,
             Extent2D extent)
         {
             var vertexSpirv = draw.VertexSpirv;
@@ -2285,7 +2440,13 @@ internal static unsafe class VulkanVideoPresenter
                 CreateTranslatedDescriptorResources(
                     resources,
                     ShaderStageFlags.VertexBit | ShaderStageFlags.FragmentBit);
-                CreateTranslatedPipeline(resources, vertexSpirv, draw.PixelSpirv, renderPass, extent);
+                CreateTranslatedPipeline(
+                    resources,
+                    vertexSpirv,
+                    draw.PixelSpirv,
+                    renderPass,
+                    renderTargetFormat,
+                    extent);
                 return resources;
             }
             catch
@@ -2581,12 +2742,13 @@ internal static unsafe class VulkanVideoPresenter
             byte[] vertexSpirv,
             byte[] fragmentSpirv,
             RenderPass renderPass,
+            Format renderTargetFormat,
             Extent2D extent)
         {
             var pipelineKey = new GraphicsPipelineKey(
                 GetShaderDigest(vertexSpirv),
                 GetShaderDigest(fragmentSpirv),
-                renderPass.Handle,
+                renderTargetFormat,
                 resources.Topology,
                 resources.Blend,
                 GetResourceLayoutKey(resources),
@@ -2742,7 +2904,7 @@ internal static unsafe class VulkanVideoPresenter
                     Check(
                         _vk.CreateGraphicsPipelines(
                             _device,
-                            default,
+                            _pipelineCache,
                             1,
                             &pipelineInfo,
                             null,
@@ -2922,7 +3084,7 @@ internal static unsafe class VulkanVideoPresenter
                 Check(
                     _vk.CreateComputePipelines(
                         _device,
-                        default,
+                        _pipelineCache,
                         1,
                         &pipelineInfo,
                         null,
@@ -3448,11 +3610,36 @@ internal static unsafe class VulkanVideoPresenter
         {
             const ulong offsetBasis = 14695981039346656037;
             const ulong prime = 1099511628211;
-            var hash = offsetBasis;
-            foreach (var value in pixels)
+
+            var h0 = offsetBasis ^ (ulong)pixels.Length;
+            var h1 = offsetBasis;
+            var h2 = offsetBasis;
+            var h3 = offsetBasis;
+
+            var words = MemoryMarshal.Cast<byte, ulong>(pixels);
+            var index = 0;
+            var blockEnd = words.Length - (words.Length & 3);
+            for (; index < blockEnd; index += 4)
             {
-                hash ^= value;
-                hash *= prime;
+                h0 = (h0 ^ words[index]) * prime;
+                h1 = (h1 ^ words[index + 1]) * prime;
+                h2 = (h2 ^ words[index + 2]) * prime;
+                h3 = (h3 ^ words[index + 3]) * prime;
+            }
+
+            for (; index < words.Length; index++)
+            {
+                h0 = (h0 ^ words[index]) * prime;
+            }
+
+            var hash = h0;
+            hash = (hash ^ h1) * prime;
+            hash = (hash ^ h2) * prime;
+            hash = (hash ^ h3) * prime;
+
+            foreach (var value in pixels[(words.Length * sizeof(ulong))..])
+            {
+                hash = (hash ^ value) * prime;
             }
 
             return hash;
@@ -3465,10 +3652,7 @@ internal static unsafe class VulkanVideoPresenter
             uint width,
             uint height)
         {
-            if (!string.Equals(
-                    Environment.GetEnvironmentVariable("SHARPEMU_DUMP_TEXTURES"),
-                    "1",
-                    StringComparison.Ordinal) ||
+            if (!_dumpTextures ||
                 texture.IsFallback ||
                 texture.IsStorage ||
                 GetTextureBytesPerPixel(texture.Format) != 4 ||
@@ -4489,6 +4673,7 @@ internal static unsafe class VulkanVideoPresenter
                 resources = CreateTranslatedDrawResources(
                     work.Draw,
                     target.RenderPass,
+                    target.Format,
                     extent);
                 resources.DebugName =
                     $"SharpEmu offscreen rt=0x{work.Target.Address:X16} " +
@@ -5125,6 +5310,8 @@ internal static unsafe class VulkanVideoPresenter
                 return;
             }
 
+            CompletePendingPresentation(wait: true);
+
             byte[]? pixels = null;
             if (presentation.Pixels is { } sourcePixels)
             {
@@ -5172,6 +5359,7 @@ internal static unsafe class VulkanVideoPresenter
                     translatedResources = CreateTranslatedDrawResources(
                         translatedDraw,
                         _renderPass,
+                        _swapchainFormat,
                         _extent);
                     if (ShouldTracePresentedGuestImageContentsForDiagnostics() &&
                         !_firstGuestDrawPresented)
@@ -5303,7 +5491,16 @@ internal static unsafe class VulkanVideoPresenter
                 SignalSemaphoreCount = 1,
                 PSignalSemaphores = &renderFinished,
             };
-            Check(_vk.QueueSubmit(_queue, 1, &submitInfo, default), "vkQueueSubmit");
+            var presentationFence = _presentationFence;
+            Check(
+                _vk.ResetFences(_device, 1, &presentationFence),
+                "vkResetFences(presentation)");
+            Check(
+                _vk.QueueSubmit(_queue, 1, &submitInfo, presentationFence),
+                "vkQueueSubmit");
+            _presentationInFlight = true;
+            _pendingPresentationResources = translatedResources;
+            translatedResources = null;
 
             var swapchain = _swapchain;
             var presentInfo = new PresentInfoKHR
@@ -5319,6 +5516,7 @@ internal static unsafe class VulkanVideoPresenter
             if (presentResult == Result.ErrorOutOfDateKhr)
             {
                 Check(_vk.QueueWaitIdle(_queue), "vkQueueWaitIdle");
+                CompletePendingPresentation(wait: false);
                 CollectCompletedGuestSubmissions(waitForOldest: false);
                 if (translatedResources is not null)
                 {
@@ -5331,19 +5529,13 @@ internal static unsafe class VulkanVideoPresenter
 
             CheckSwapchainResult(presentResult, "vkQueuePresentKHR");
             recreateAfterPresent |= presentResult == Result.SuboptimalKhr;
-            Check(_vk.QueueWaitIdle(_queue), "vkQueueWaitIdle");
             VideoOutExports.ReportPresentedFrame();
             if (_swapchainReadbackPending)
             {
+                CompletePendingPresentation(wait: true);
                 TraceSwapchainReadback();
             }
             CollectCompletedGuestSubmissions(waitForOldest: false);
-            if (translatedResources is not null)
-            {
-                MarkSampledImagesInitialized(translatedResources);
-                MarkStorageImagesInitialized(translatedResources);
-                DestroyTranslatedDrawResources(translatedResources);
-            }
 
             _imageInitialized[imageIndex] = true;
             _presentedSequence = presentation.Sequence;
@@ -6597,7 +6789,9 @@ internal static unsafe class VulkanVideoPresenter
             }
             _vulkanReady = false;
             _vk.DeviceWaitIdle(_device);
+            CompletePendingPresentation(wait: false);
             CollectCompletedGuestSubmissions(waitForOldest: false);
+            SavePipelineCache();
             foreach (var pipeline in _computePipelines.Values)
             {
                 _vk.DestroyPipeline(_device, pipeline, null);
@@ -6608,6 +6802,11 @@ internal static unsafe class VulkanVideoPresenter
                 _vk.DestroyPipeline(_device, pipeline, null);
             }
             _graphicsPipelines.Clear();
+            if (_pipelineCache.Handle != 0)
+            {
+                _vk.DestroyPipelineCache(_device, _pipelineCache, null);
+                _pipelineCache = default;
+            }
             foreach (var layout in _descriptorLayouts.Values)
             {
                 _vk.DestroyPipelineLayout(_device, layout.PipelineLayout, null);
@@ -6699,6 +6898,7 @@ internal static unsafe class VulkanVideoPresenter
             Console.Error.WriteLine(
                 $"[LOADER][INFO] Vulkan VideoOut recreating swapchain after {operation}: {result}");
             _vk.DeviceWaitIdle(_device);
+            CompletePendingPresentation(wait: false);
             CollectCompletedGuestSubmissions(waitForOldest: false);
             DestroySwapchainResources();
             CreateSwapchain();
@@ -6732,6 +6932,11 @@ internal static unsafe class VulkanVideoPresenter
                 _vk.DestroySemaphore(_device, _renderFinished, null);
                 _renderFinished = default;
             }
+            if (_presentationFence.Handle != 0)
+            {
+                _vk.DestroyFence(_device, _presentationFence, null);
+                _presentationFence = default;
+            }
             if (_barycentricPipeline.Handle != 0)
             {
                 _vk.DestroyPipeline(_device, _barycentricPipeline, null);

src/SharpEmu.Libs/Voice/VoiceQoSExports.cs

@@ -0,0 +1,19 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Voice;
+
+public static class VoiceQoSExports
+{
+    [SysAbiExport(
+        Nid = "U8IfNl6-Css",
+        ExportName = "sceVoiceQoSInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVoiceQoS")]
+    public static int VoiceQoSInit(CpuContext ctx)
+    {
+        return ctx.SetReturn(0);
+    }
+}

src/SharpEmu.Logging/FileLogSink.cs

@@ -6,19 +6,15 @@ using System.Text;
 
 namespace SharpEmu.Logging;
 
-/// <summary>
-/// Writes log entries to a file. Thread-safe via an internal lock.
-/// <see cref="StreamWriter.AutoFlush"/> is enabled so entries survive a crash.
-/// </summary>
 public sealed class FileLogSink : ISharpEmuLogSink, IDisposable
 {
+    private static readonly TimeSpan FlushInterval = TimeSpan.FromMilliseconds(500);
+
     private readonly object _sync = new();
     private readonly StreamWriter _writer;
+    private readonly Timer _flushTimer;
     private bool _disposed;
 
-    /// <param name="path">Absolute or relative file path. Parent directories are created if missing.</param>
-    /// <param name="append"><see langword="true"/> to append to an existing file; <see langword="false"/> to overwrite.</param>
-    /// <param name="includeTimestamp">Always recommended for file logs — entries include a full date-time prefix.</param>
     public FileLogSink(string path, bool append = true, bool includeTimestamp = true)
     {
         ArgumentException.ThrowIfNullOrWhiteSpace(path);
@@ -34,14 +30,20 @@ public sealed class FileLogSink : ISharpEmuLogSink, IDisposable
             append ? FileMode.Append : FileMode.Create,
             FileAccess.Write,
             FileShare.Read,
-            bufferSize: 4096,
+            bufferSize: 65536,
             FileOptions.SequentialScan);
-        _writer = new StreamWriter(fileStream, Encoding.UTF8)
+        _writer = new StreamWriter(fileStream, Encoding.UTF8, bufferSize: 65536)
         {
-            AutoFlush = true
+            AutoFlush = false
         };
 
         IncludeTimestamp = includeTimestamp;
+
+        _flushTimer = new Timer(
+            static state => ((FileLogSink)state!).FlushBuffered(),
+            this,
+            FlushInterval,
+            FlushInterval);
     }
 
     public bool IncludeTimestamp { get; set; }
@@ -84,11 +86,31 @@ public sealed class FileLogSink : ISharpEmuLogSink, IDisposable
             {
                 _writer.WriteLine(entry.Exception);
             }
+
+            if (entry.Level >= LogLevel.Error)
+            {
+                _writer.Flush();
+            }
+        }
+    }
+
+    private void FlushBuffered()
+    {
+        lock (_sync)
+        {
+            if (_disposed)
+            {
+                return;
+            }
+
+            _writer.Flush();
         }
     }
 
     public void Dispose()
     {
+        _flushTimer.Dispose();
+
         lock (_sync)
         {
             if (_disposed)

src/SharpEmu.Logging/HostSystemInfo.cs

@@ -36,11 +36,16 @@ public static class HostSystemInfo
             using var key = Registry.LocalMachine.OpenSubKey(@"HARDWARE\DESCRIPTION\System\CentralProcessor\0");
             if (key?.GetValue("ProcessorNameString") is string name && !string.IsNullOrWhiteSpace(name))
             {
-                return Regex.Replace(
+                name = Regex.Replace(
                     name.Trim(),
                     @"\s+\d+-Core Processor$",
                     string.Empty,
                     RegexOptions.IgnoreCase | RegexOptions.CultureInvariant);
+                return Regex.Replace(
+                    name,
+                    @"\s+with Radeon Graphics$",
+                    string.Empty,
+                    RegexOptions.IgnoreCase | RegexOptions.CultureInvariant);
             }
         }
         catch (Exception)
@@ -103,8 +108,12 @@ public static class HostSystemInfo
         }
 
         if (name.Contains("nvidia", StringComparison.OrdinalIgnoreCase) ||
-            name.Contains("geforce", StringComparison.OrdinalIgnoreCase) ||
-            name.Contains("amd", StringComparison.OrdinalIgnoreCase) ||
+            name.Contains("geforce", StringComparison.OrdinalIgnoreCase))
+        {
+            return 3;
+        }
+
+        if (name.Contains("amd", StringComparison.OrdinalIgnoreCase) ||
             name.Contains("radeon", StringComparison.OrdinalIgnoreCase))
         {
             return 2;