[video] added sdl window and host display plumbing

src/SharpEmu.Libs/Gpu/IGuestGpuBackend.cs

@@ -237,7 +237,21 @@ internal interface IGuestGpuBackend
 
     void SubmitGuestImageFill(ulong address, uint fillValue);
 
-    void SubmitGuestImageWrite(ulong address, byte[] pixels);
+    /// <summary>
+    /// Uploads guest-authored pixels into a live guest image. <paramref name="rowOffset"/>
+    /// is the first image row the buffer covers, so a caller that knows only part
+    /// of the surface changed can send that band instead of the whole thing; the
+    /// untouched rows on the host already hold the same bytes.
+    /// </summary>
+    void SubmitGuestImageWrite(ulong address, byte[] pixels, uint rowOffset = 0);
+
+    /// <summary>
+    /// Whether a non-zero <c>rowOffset</c> is honoured. Backends that cannot
+    /// upload a sub-range must report false so callers keep sending the whole
+    /// surface: a dropped band would leave the host copy stale, which is worse
+    /// than an oversized upload.
+    /// </summary>
+    bool SupportsPartialImageWrite => false;
 
     /// <summary>
     /// Asks the presenter to refresh CPU-dirty guest images on its render/present

src/SharpEmu.Libs/Gpu/Metal/MetalGuestGpuBackend.cs

@@ -398,7 +398,7 @@ internal sealed class MetalGuestGpuBackend : IGuestGpuBackend
     public void SubmitGuestImageFill(ulong address, uint fillValue) =>
         MetalVideoPresenter.SubmitGuestImageFill(address, fillValue);
 
-    public void SubmitGuestImageWrite(ulong address, byte[] pixels) =>
+    public void SubmitGuestImageWrite(ulong address, byte[] pixels, uint rowOffset = 0) =>
         MetalVideoPresenter.SubmitGuestImageWrite(address, pixels);
 
     public void RequestCpuWrittenGuestImageSync(ulong scopeAddress = 0, ulong scopeByteCount = ulong.MaxValue) =>

src/SharpEmu.Libs/Gpu/Metal/MetalNative.cs

@@ -5,19 +5,6 @@ using System.Runtime.InteropServices;
 
 namespace SharpEmu.Libs.Gpu.Metal;
 
-// Core Graphics / Metal ABI structs passed by value through objc_msgSend. Struct
-// *returns* are deliberately never used: on x86-64 (this process runs under Rosetta
-// on Apple silicon) large struct returns switch to objc_msgSend_stret, and avoiding
-// them entirely keeps one calling convention everywhere.
-[StructLayout(LayoutKind.Sequential)]
-internal struct CGRect
-{
-    public double X;
-    public double Y;
-    public double Width;
-    public double Height;
-}
-
 [StructLayout(LayoutKind.Sequential)]
 internal struct CGSize
 {
@@ -93,31 +80,21 @@ internal struct MtlViewport
 }
 
 /// <summary>
-/// Objective-C runtime access for the Metal presenter: AppKit, QuartzCore, and Metal
+/// Objective-C runtime access for the Metal presenter: QuartzCore and Metal
 /// through objc_msgSend, with one LibraryImport overload per distinct native
 /// signature. Dependency-free by design — this plus the OS frameworks is the entire
 /// Metal path, which is what keeps it NativeAOT-clean.
 /// </summary>
 internal static partial class MetalNative
 {
-    private const string CoreFoundation =
-        "/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation";
-
-    [LibraryImport(CoreFoundation)]
-    public static partial nint CFRunLoopGetMain();
-
-    [LibraryImport(CoreFoundation)]
-    public static partial void CFRunLoopStop(nint runLoop);
-
     private const string ObjCLibrary = "/usr/lib/libobjc.A.dylib";
     private const string MetalFramework = "/System/Library/Frameworks/Metal.framework/Metal";
-    private const string AppKitFramework = "/System/Library/Frameworks/AppKit.framework/AppKit";
     private const string QuartzCoreFramework = "/System/Library/Frameworks/QuartzCore.framework/QuartzCore";
 
     private static bool _frameworksLoaded;
 
     /// <summary>
-    /// Makes the AppKit and QuartzCore classes visible to objc_getClass; Metal is
+    /// Makes QuartzCore classes visible to objc_getClass; Metal is
     /// pulled in by its own LibraryImport. Call once before any Class() lookup.
     /// </summary>
     public static void EnsureFrameworksLoaded()
@@ -127,7 +104,6 @@ internal static partial class MetalNative
             return;
         }
 
-        NativeLibrary.Load(AppKitFramework);
         NativeLibrary.Load(QuartzCoreFramework);
         _frameworksLoaded = true;
     }
@@ -141,16 +117,6 @@ internal static partial class MetalNative
     [LibraryImport(ObjCLibrary, StringMarshalling = StringMarshalling.Utf8)]
     private static partial nint sel_registerName(string name);
 
-    [LibraryImport(ObjCLibrary, StringMarshalling = StringMarshalling.Utf8)]
-    public static partial nint objc_allocateClassPair(nint superclass, string name, nuint extraBytes);
-
-    [LibraryImport(ObjCLibrary)]
-    public static partial void objc_registerClassPair(nint cls);
-
-    [LibraryImport(ObjCLibrary, StringMarshalling = StringMarshalling.Utf8)]
-    [return: MarshalAs(UnmanagedType.I1)]
-    public static partial bool class_addMethod(nint cls, nint name, nint imp, string types);
-
     [LibraryImport(ObjCLibrary)]
     public static partial nint objc_autoreleasePoolPush();
 
@@ -169,14 +135,6 @@ internal static partial class MetalNative
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial nint Send(nint receiver, nint selector, nint argument);
 
-
-    /// <summary>objc_msgSend for a CGRect-returning selector (e.g. -bounds).
-    /// A 32-byte struct is returned via the x86-64 stret ABI — a hidden
-    /// pointer to caller storage passed ahead of self/_cmd — so this must not
-    /// be folded into the plain objc_msgSend overloads.</summary>
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend_stret")]
-    public static partial void SendStretRect(out CGRect result, nint receiver, nint selector);
-
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial nint Send(nint receiver, nint selector, nint argument, ref nint error);
 
@@ -209,17 +167,6 @@ internal static partial class MetalNative
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial void SendVoid(nint receiver, nint selector, nint argument0, nint argument1);
 
-    /// <summary>performSelectorOnMainThread:withObject:waitUntilDone: — the SEL
-    /// to perform is itself an argument, followed by the object and the wait
-    /// flag.</summary>
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial void SendVoidPerformSelector(
-        nint receiver,
-        nint selector,
-        nint performedSelector,
-        nint argument,
-        [MarshalAs(UnmanagedType.I1)] bool waitUntilDone);
-
     /// <summary>setSwizzle: on MTLTextureDescriptor. Four one-byte
     /// MTLTextureSwizzle values, passed packed like the framework expects.</summary>
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
@@ -237,9 +184,6 @@ internal static partial class MetalNative
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial void SendVoidSize(nint receiver, nint selector, CGSize size);
 
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial void SendVoidRect(nint receiver, nint selector, CGRect rect);
-
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial void SendVoidClearColor(nint receiver, nint selector, MtlClearColor color);
 
@@ -328,37 +272,6 @@ internal static partial class MetalNative
         nuint indexBufferOffset,
         nuint instanceCount);
 
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial nint SendTimer(
-        nint receiver,
-        nint selector,
-        double interval,
-        nint target,
-        nint timerSelector,
-        nint userInfo,
-        [MarshalAs(UnmanagedType.I1)] bool repeats);
-
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial nint SendInitFrame(nint receiver, nint selector, CGRect frame);
-
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial nint SendInitWindow(
-        nint receiver,
-        nint selector,
-        CGRect contentRect,
-        nuint styleMask,
-        nuint backing,
-        [MarshalAs(UnmanagedType.I1)] bool defer);
-
-    [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
-    public static partial nint SendNextEvent(
-        nint receiver,
-        nint selector,
-        ulong eventMask,
-        nint untilDate,
-        nint inMode,
-        [MarshalAs(UnmanagedType.I1)] bool dequeue);
-
     [LibraryImport(ObjCLibrary, EntryPoint = "objc_msgSend")]
     public static partial nint SendTextureDescriptor(
         nint receiver,

src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.cs

@@ -1,8 +1,6 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-using System.Runtime.CompilerServices;
-using System.Runtime.InteropServices;
 using SharpEmu.HLE;
 using SharpEmu.Libs.VideoOut;
 using SharpEmu.ShaderCompiler;
@@ -11,28 +9,12 @@ using SharpEmu.ShaderCompiler.Metal;
 namespace SharpEmu.Libs.Gpu.Metal;
 
 /// <summary>
-/// The Metal presenter: an AppKit window hosting a CAMetalLayer. A CADisplayLink
+/// The Metal presenter uses SDL for its host window, events and input, then renders
-/// requested from the content view drives <see cref="RenderFrame"/> in sync with the
+/// into the CAMetalLayer owned by SDL's Metal view. Windowing and input therefore
-/// display refresh, on the main run loop — the loop whose Core Animation observer
+/// share the Vulkan path while all Metal command encoding remains native.
-/// commits presented drawables to the window server, so it must be a real running
-/// run loop (a hand-pumped event drain never fires that observer and the window
-/// stays black). Everything AppKit runs on the process main thread via
-/// <see cref="HostMainThread"/> (AppKit traps off-main), which the CLI parks for us.
 /// </summary>
 internal static partial class MetalVideoPresenter
 {
-    private const uint DefaultWindowWidth = 1280;
-    private const uint DefaultWindowHeight = 720;
-
-    // NSWindow style: Titled | Closable | Miniaturizable | Resizable. Resizable
-    // both lets the user drag the window edges and turns the green zoom button
-    // into the full-screen toggle (paired with the collection behavior below).
-    private const nuint WindowStyleMask = 1 | 2 | 4 | 8;
-
-    // NSWindowCollectionBehaviorFullScreenPrimary: opt this window into native
-    // full-screen, so the green button enters full-screen rather than zooming.
-    private const nuint FullScreenPrimaryBehavior = 1 << 7;
-    private const nuint BackingStoreBuffered = 2;
     private const nuint PixelFormatBgra8Unorm = (nuint)MtlPixelFormat.Bgra8Unorm;
     private const nuint LoadActionLoad = 1;
     private const nuint LoadActionClear = 2;
@@ -69,17 +51,14 @@ internal static partial class MetalVideoPresenter
     private static readonly byte[] _overlayPixels =
         new byte[PerfOverlay.PanelWidth * PerfOverlay.PanelHeight * 4];
 
-    // Presenter objects and per-frame present state, shared between window setup
+    // Presenter objects and per-frame present state, all used on the SDL host thread.
-    // and the display-link RenderFrame callback (both on the main thread).
     private static nint _device;
     private static nint _commandQueue;
     private static nint _metalLayer;
     private static nint _presentPipeline;
     private static nint _presentSampler;
-    private static nint _window;
+    private static SdlHostWindow? _hostWindow;
-    private static nint _application;
+    private static HostVideoOptions _videoOptions = HostVideoOptions.Default;
-    private static nint _renderTimer;
-    private static nint _renderTimerTarget;
     private static double _drawableWidth;
     private static double _drawableHeight;
     private static nint _frameTexture;
@@ -91,10 +70,23 @@ internal static partial class MetalVideoPresenter
     private static nint _ownedVersionTexture;
     private static ulong _presentGuestAddress;
     private static long _presentedSequence = -1;
-    private static bool _userClosed;
     private static uint _windowWidth;
     private static uint _windowHeight;
 
+    public static bool TryConfigureVideo(HostVideoOptions options)
+    {
+        lock (_gate)
+        {
+            if (_thread is not null)
+            {
+                return false;
+            }
+
+            _videoOptions = options.Normalize();
+            return true;
+        }
+    }
+
     public static void EnsureStarted(uint width, uint height)
     {
         if (width == 0 || height == 0)
@@ -183,6 +175,7 @@ internal static partial class MetalVideoPresenter
     public static void RequestClose()
     {
         Volatile.Write(ref _closeRequested, true);
+        Volatile.Read(ref _hostWindow)?.Close();
     }
 
     private static void StartPresenterLocked()
@@ -247,33 +240,23 @@ internal static partial class MetalVideoPresenter
             VideoOutExports.SetSelectedGpuName(deviceName);
         }
 
-        // Fixed window like the Vulkan presenter: guest frames letterbox into
+        HostVideoOptions videoOptions;
-        // it. Sizing the window from the guest's display mode (4K) exceeds the
+        lock (_gate)
-        // screen — macOS clamps the window while the layer keeps the requested
+        {
-        // geometry, leaving the visible region showing nothing but clear.
+            videoOptions = _videoOptions;
-        const uint width = DefaultWindowWidth;
+        }
-        const uint height = DefaultWindowHeight;
 
+        using var hostWindow = new SdlHostWindow(
+            VideoOutExports.GetWindowTitle(),
+            videoOptions,
+            SdlGraphicsApi.Metal,
+            ToggleMetalPerformanceHud);
+        Volatile.Write(ref _hostWindow, hostWindow);
         var setupPool = MetalNative.objc_autoreleasePoolPush();
         try
         {
-            _application = MetalNative.Send(
+            _metalLayer = hostWindow.CreateMetalLayer();
-                MetalNative.Class("NSApplication"), MetalNative.Selector("sharedApplication"));
+            ConfigureMetalLayer(_device, _metalLayer);
-            // NSApplicationActivationPolicyRegular: dock icon + key window like any app.
-            MetalNative.Send(_application, MetalNative.Selector("setActivationPolicy:"), 0);
-            MetalNative.SendVoid(_application, MetalNative.Selector("finishLaunching"));
-
-            _window = CreateWindow(width, height);
-
-            // Swap in the key-capturing view before the metal layer attaches so
-            // the layer lands on the input-aware content view.
-            var keyView = MetalNative.SendInitFrame(
-                MetalNative.Send(CreateKeyViewClass(), MetalNative.Selector("alloc")),
-                MetalNative.Selector("initWithFrame:"),
-                new CGRect { X = 0, Y = 0, Width = width, Height = height });
-            MetalNative.SendVoid(_window, MetalNative.Selector("setContentView:"), keyView);
-
-            _metalLayer = CreateLayer(_device, _window, out _drawableWidth, out _drawableHeight);
             _commandQueue = MetalNative.Send(_device, MetalNative.Selector("newCommandQueue"));
             if (!TryCreatePresentPipeline(_device, out _presentPipeline, out var pipelineError))
             {
@@ -282,31 +265,7 @@ internal static partial class MetalVideoPresenter
             }
 
             _presentSampler = CreateLinearSampler(_device);
-
+            SyncDrawableSizeToLayer();
-            MetalNative.SendVoid(_window, MetalNative.Selector("makeKeyAndOrderFront:"), 0);
-            MetalNative.SendVoidBool(
-                _application, MetalNative.Selector("activateIgnoringOtherApps:"), true);
-            MetalNative.SendVoid(_window, MetalNative.Selector("makeFirstResponder:"), keyView);
-            MetalHostInput.Attach();
-
-            // A repeating NSTimer on this (main) run loop fires onFrame: at the
-            // display rate. CADisplayLink (NSView.displayLinkWithTarget:selector:)
-            // is the natural choice but its callback never fires under the x86-64
-            // Rosetta process this emulator runs as — proven in isolation against
-            // a bare AppKit harness, where a timer fires and composites and the
-            // display link does not. nextDrawable still throttles presentation to
-            // the display, so the timer only needs to keep up, not pace precisely.
-            _renderTimerTarget = CreateRenderTimerTarget();
-            _renderTimer = MetalNative.Send(
-                MetalNative.SendTimer(
-                    MetalNative.Class("NSTimer"),
-                    MetalNative.Selector("scheduledTimerWithTimeInterval:target:selector:userInfo:repeats:"),
-                    1.0 / 60.0,
-                    _renderTimerTarget,
-                    MetalNative.Selector("onFrame:"),
-                    0,
-                    repeats: true),
-                MetalNative.Selector("retain"));
         }
         finally
         {
@@ -314,25 +273,20 @@ internal static partial class MetalVideoPresenter
         }
 
         Console.Error.WriteLine("[LOADER][INFO] Metal VideoOut presenter started.");
-
-        // [NSApp run] runs the main run loop (its Core Animation observer commits
-        // presented drawables to the window server) AND fully activates the app,
-        // which a bare CFRunLoopRun does not — the CADisplayLink is only serviced
-        // once the app is running, and NSApp dispatches window events itself.
-        // Returns once RenderFrame stops it.
-        MetalNative.SendVoid(_application, MetalNative.Selector("run"));
-
-        var closePool = MetalNative.objc_autoreleasePoolPush();
         try
         {
-            MetalNative.SendVoid(_window, MetalNative.Selector("close"));
+            hostWindow.Run(
+                static () => { },
+                static _ => RenderFrameSafely(),
+                static () => { });
         }
         finally
         {
-            MetalNative.objc_autoreleasePoolPop(closePool);
+            Volatile.Write(ref _hostWindow, null);
+            _metalLayer = 0;
         }
 
-        if (_userClosed)
+        if (hostWindow.ClosedByUser)
         {
             Console.Error.WriteLine(
                 "[LOADER][WARN] Metal VideoOut window closed; requesting emulator shutdown.");
@@ -340,115 +294,8 @@ internal static partial class MetalVideoPresenter
         }
     }
 
-    /// <summary>
-    /// An NSView subclass that records key events for pad emulation. Overriding
-    /// keyDown:/keyUp: (instead of an event monitor) needs no ObjC blocks, and
-    /// swallowing the events also silences the system alert beep AppKit plays
-    /// for unhandled keys. Registered once per process.
-    /// </summary>
-    private static unsafe nint CreateKeyViewClass()
-    {
-        var cls = MetalNative.objc_allocateClassPair(
-            MetalNative.Class("NSView"), "SharpEmuMetalView", 0);
-        if (cls == 0)
-        {
-            return MetalNative.Class("SharpEmuMetalView");
-        }
-
-        var keyDown = (nint)(delegate* unmanaged[Cdecl]<nint, nint, nint, void>)&OnKeyDown;
-        MetalNative.class_addMethod(cls, MetalNative.Selector("keyDown:"), keyDown, "v@:@");
-        var keyUp = (nint)(delegate* unmanaged[Cdecl]<nint, nint, nint, void>)&OnKeyUp;
-        MetalNative.class_addMethod(cls, MetalNative.Selector("keyUp:"), keyUp, "v@:@");
-        // Command-modified keys never reach keyDown: — AppKit routes them through
-        // performKeyEquivalent:, so Cmd+F1 (Metal Performance HUD) hooks in here.
-        var keyEquivalent = (nint)(delegate* unmanaged[Cdecl]<nint, nint, nint, byte>)&OnPerformKeyEquivalent;
-        MetalNative.class_addMethod(
-            cls, MetalNative.Selector("performKeyEquivalent:"), keyEquivalent, "c@:@");
-        // First responder status is what routes key events to this view.
-        var accepts = (nint)(delegate* unmanaged[Cdecl]<nint, nint, byte>)&AcceptsFirstResponder;
-        MetalNative.class_addMethod(
-            cls, MetalNative.Selector("acceptsFirstResponder"), accepts, "c@:");
-        MetalNative.objc_registerClassPair(cls);
-        return cls;
-    }
-
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
-    private static void OnKeyDown(nint self, nint cmd, nint nsEvent)
-    {
-        try
-        {
-            var keyCode = (ushort)(MetalNative.Send(nsEvent, MetalNative.Selector("keyCode")) & 0xFFFF);
-            var isRepeat = MetalNative.SendBool(nsEvent, MetalNative.Selector("isARepeat"));
-
-            // Function keys can arrive here even with Command held (AppKit only
-            // reroutes some chords through the key-equivalent path), so catch
-            // Cmd+F1 in both places — and keep it away from MetalHostInput so it
-            // never toggles the plain-F1 perf overlay.
-            var modifiers = (ulong)MetalNative.Send(nsEvent, MetalNative.Selector("modifierFlags"));
-            if (keyCode == KeyCodeF1 && (modifiers & NsEventModifierFlagCommand) != 0)
-            {
-                if (!isRepeat)
-                {
-                    ToggleMetalPerformanceHud();
-                }
-
-                return;
-            }
-
-            MetalHostInput.KeyDown(keyCode, isRepeat);
-        }
-        catch (Exception exception)
-        {
-            Console.Error.WriteLine($"[LOADER][WARN] Metal key-down handler failed: {exception.Message}");
-        }
-    }
-
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
-    private static void OnKeyUp(nint self, nint cmd, nint nsEvent)
-    {
-        try
-        {
-            var keyCode = (ushort)(MetalNative.Send(nsEvent, MetalNative.Selector("keyCode")) & 0xFFFF);
-            MetalHostInput.KeyUp(keyCode);
-        }
-        catch (Exception exception)
-        {
-            Console.Error.WriteLine($"[LOADER][WARN] Metal key-up handler failed: {exception.Message}");
-        }
-    }
-
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
-    private static byte AcceptsFirstResponder(nint self, nint cmd) => 1;
-
-    private const ushort KeyCodeF1 = 0x7A;
-    private const ulong NsEventModifierFlagCommand = 1UL << 20;
     private static bool _metalHudVisible;
 
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
-    private static byte OnPerformKeyEquivalent(nint self, nint cmd, nint nsEvent)
-    {
-        try
-        {
-            var keyCode = (ushort)(MetalNative.Send(nsEvent, MetalNative.Selector("keyCode")) & 0xFFFF);
-            var modifiers = (ulong)MetalNative.Send(nsEvent, MetalNative.Selector("modifierFlags"));
-            if (keyCode == KeyCodeF1 && (modifiers & NsEventModifierFlagCommand) != 0)
-            {
-                if (!MetalNative.SendBool(nsEvent, MetalNative.Selector("isARepeat")))
-                {
-                    ToggleMetalPerformanceHud();
-                }
-
-                return 1; // handled: no system beep, no further routing
-            }
-        }
-        catch (Exception exception)
-        {
-            Console.Error.WriteLine($"[LOADER][WARN] Metal key-equivalent handler failed: {exception.Message}");
-        }
-
-        return 0;
-    }
-
     /// <summary>
     /// Cmd+F1: Apple's Metal Performance HUD on the CAMetalLayer (plain F1 keeps
     /// the built-in CPU-rasterized perf overlay). Configured per Apple's
@@ -456,7 +303,7 @@ internal static partial class MetalVideoPresenter
     /// mode=default|disabled and logging=default, plus any MTL_HUD_* environment
     /// keys directly in the dictionary — all three HUD flags (enabled, per-frame
     /// logging, shader-compile logging) ride in one property set. Runs on the
-    /// AppKit main thread (the key-equivalent path), same thread as the render loop.
+    /// SDL host thread, the same thread as the render loop.
     /// </summary>
     private static void ToggleMetalPerformanceHud()
     {
@@ -508,33 +355,13 @@ internal static partial class MetalVideoPresenter
             $"[LOADER][INFO] Metal Performance HUD {(_metalHudVisible ? "shown" : "hidden")} (Cmd+F1).");
     }
 
-    private static unsafe nint CreateRenderTimerTarget()
+    /// <summary>The SDL host loop drains guest work continuously. The method
+    /// remains as the enqueue-side hook used by guest image synchronization.</summary>
+    internal static void ScheduleGuestWorkDrain()
     {
-        // A minimal NSObject subclass whose onFrame: is our unmanaged callback —
-        // the dependency-free way to hand a target/selector to NSTimer without a
-        // binding library. Registered once per process.
-        var cls = MetalNative.objc_allocateClassPair(
-            MetalNative.Class("NSObject"), "SharpEmuRenderTimerTarget", 0);
-        if (cls != 0)
-        {
-            var imp = (nint)(delegate* unmanaged[Cdecl]<nint, nint, nint, void>)&OnRenderTimer;
-            // "v@:@": void return, self, _cmd, one object argument (the timer).
-            MetalNative.class_addMethod(cls, MetalNative.Selector("onFrame:"), imp, "v@:@");
-            var wakeImp = (nint)(delegate* unmanaged[Cdecl]<nint, nint, nint, void>)&OnGuestWorkWake;
-            MetalNative.class_addMethod(cls, MetalNative.Selector("onGuestWork:"), wakeImp, "v@:@");
-            MetalNative.objc_registerClassPair(cls);
-        }
-        else
-        {
-            cls = MetalNative.Class("SharpEmuRenderTimerTarget");
-        }
-
-        return MetalNative.Send(
-            MetalNative.Send(cls, MetalNative.Selector("alloc")), MetalNative.Selector("init"));
     }
 
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
+    private static void RenderFrameSafely()
-    private static void OnRenderTimer(nint self, nint cmd, nint timer)
     {
         try
         {
@@ -546,78 +373,13 @@ internal static partial class MetalVideoPresenter
         }
     }
 
-    /// <summary>Set while an onGuestWork: wake is scheduled on the main run
-    /// loop; coalesces enqueue-side wake requests to one in-flight message.</summary>
-    private static int _guestWorkWakeScheduled;
-
-    /// <summary>Wakes the main run loop to drain guest work now instead of at
-    /// the next render tick. Guest submit→wait round-trips (release-mem labels,
-    /// CPU-visible write-backs) otherwise cost a full frame interval each —
-    /// games that chain several per frame crawl at a fraction of the display
-    /// rate. Safe from any thread; no-op until the presenter starts.</summary>
-    internal static void ScheduleGuestWorkDrain()
-    {
-        if (Interlocked.CompareExchange(ref _guestWorkWakeScheduled, 1, 0) != 0)
-        {
-            return;
-        }
-
-        var target = _renderTimerTarget;
-        if (target == 0)
-        {
-            Volatile.Write(ref _guestWorkWakeScheduled, 0);
-            return;
-        }
-
-        MetalNative.SendVoidPerformSelector(
-            target,
-            MetalNative.Selector("performSelectorOnMainThread:withObject:waitUntilDone:"),
-            MetalNative.Selector("onGuestWork:"),
-            0,
-            waitUntilDone: false);
-    }
-
-    [UnmanagedCallersOnly(CallConvs = [typeof(CallConvCdecl)])]
-    private static void OnGuestWorkWake(nint self, nint cmd, nint argument)
-    {
-        Volatile.Write(ref _guestWorkWakeScheduled, 0);
-        if (_device == 0 || _commandQueue == 0 || Volatile.Read(ref _closeRequested))
-        {
-            return;
-        }
-
-        var pool = MetalNative.objc_autoreleasePoolPush();
-        try
-        {
-            DrainGuestWork(_device, _commandQueue);
-        }
-        catch (Exception exception)
-        {
-            Console.Error.WriteLine($"[LOADER][ERROR] Metal guest work wake failed: {exception}");
-        }
-        finally
-        {
-            MetalNative.objc_autoreleasePoolPop(pool);
-        }
-    }
-
     private static void RenderFrame()
     {
-        MetalHostInput.PumpAutoKeys();
         var pool = MetalNative.objc_autoreleasePoolPush();
         try
         {
-            // NSApp.run dispatches window events itself, so there is no manual
+            if (Volatile.Read(ref _closeRequested))
-            // event drain here.
-            var visible = MetalNative.SendBool(_window, MetalNative.Selector("isVisible"));
-            if (Volatile.Read(ref _closeRequested) || !visible)
             {
-                _userClosed = !visible && !Volatile.Read(ref _closeRequested);
-                MetalNative.SendVoid(_renderTimer, MetalNative.Selector("invalidate"));
-                // Stop both the AppKit loop and the underlying CFRunLoop so
-                // [NSApp run] returns.
-                MetalNative.SendVoid(_application, MetalNative.Selector("stop:"), 0);
-                MetalNative.CFRunLoopStop(MetalNative.CFRunLoopGetMain());
                 return;
             }
 
@@ -715,8 +477,7 @@ internal static partial class MetalVideoPresenter
                 if (!string.Equals(title, _lastWindowTitle, StringComparison.Ordinal))
                 {
                     _lastWindowTitle = title;
-                    MetalNative.SendVoid(
+                    Volatile.Read(ref _hostWindow)?.SetTitle(title);
-                        _window, MetalNative.Selector("setTitle:"), MetalNative.NsString(title));
                 }
             }
 
@@ -725,8 +486,20 @@ internal static partial class MetalVideoPresenter
             // for a drawable, or nextDrawable keeps handing back the original
             // resolution and Core Animation stretches it (blurry, mis-scaled
             // overlay). No-op when the size is unchanged, i.e. almost every tick.
+            var hostWindow = Volatile.Read(ref _hostWindow);
+            if (hostWindow?.ConsumeSurfaceRestore() == true)
+            {
+                _drawableWidth = 0;
+                _drawableHeight = 0;
+            }
+
             SyncDrawableSizeToLayer();
 
+            if (hostWindow?.IsMinimized == true)
+            {
+                return;
+            }
+
             var drawable = MetalNative.Send(_metalLayer, MetalNative.Selector("nextDrawable"));
             if (drawable == 0)
             {
@@ -846,43 +619,33 @@ internal static partial class MetalVideoPresenter
             3);
     }
 
-    private static nint CreateWindow(uint width, uint height)
+    private static void ConfigureMetalLayer(nint device, nint layer)
     {
-        var window = MetalNative.SendInitWindow(
+        MetalNative.SendVoid(layer, MetalNative.Selector("setDevice:"), device);
-            MetalNative.Send(MetalNative.Class("NSWindow"), MetalNative.Selector("alloc")),
+        MetalNative.Send(layer, MetalNative.Selector("setPixelFormat:"), (nint)PixelFormatBgra8Unorm);
-            MetalNative.Selector("initWithContentRect:styleMask:backing:defer:"),
+        MetalNative.SendVoidBool(layer, MetalNative.Selector("setOpaque:"), true);
-            new CGRect { X = 0, Y = 0, Width = width, Height = height },
+        MetalNative.SendVoidBool(layer, MetalNative.Selector("setFramebufferOnly:"), true);
-            WindowStyleMask,
+
-            BackingStoreBuffered,
+        var setDisplaySync = MetalNative.Selector("setDisplaySyncEnabled:");
-            defer: false);
+        if (MetalNative.SendBool(layer, MetalNative.Selector("respondsToSelector:"), setDisplaySync))
-        // The presenter owns the handle; AppKit must not free it on user close.
+        {
-        MetalNative.SendVoidBool(window, MetalNative.Selector("setReleasedWhenClosed:"), false);
+            MetalNative.SendVoidBool(layer, setDisplaySync, _videoOptions.VSync);
-        MetalNative.Send(
+        }
-            window, MetalNative.Selector("setCollectionBehavior:"), (nint)FullScreenPrimaryBehavior);
-        MetalNative.SendVoid(
-            window,
-            MetalNative.Selector("setTitle:"),
-            MetalNative.NsString(VideoOutExports.GetWindowTitle()));
-        MetalNative.SendVoid(window, MetalNative.Selector("center"));
-        // makeKeyAndOrderFront happens after the metal layer is attached.
-        return window;
     }
 
-    /// <summary>Keeps the CAMetalLayer's drawable size (pixels) matched to its
+    /// <summary>Keeps the SDL-owned CAMetalLayer drawable in host pixels as the
-    /// current bounds (points) × scale as the window resizes or moves between
+    /// window resizes or moves between displays with different scale factors.</summary>
-    /// displays. CAMetalLayer never updates drawableSize on its own, even as a
-    /// view's backing layer, so the render loop drives it.</summary>
     private static void SyncDrawableSizeToLayer()
     {
-        MetalNative.SendStretRect(out var bounds, _metalLayer, MetalNative.Selector("bounds"));
+        var hostWindow = Volatile.Read(ref _hostWindow);
-        var scale = MetalNative.SendDouble(_metalLayer, MetalNative.Selector("contentsScale"));
+        if (hostWindow is null || _metalLayer == 0)
-        if (scale <= 0)
         {
-            scale = 1;
+            return;
         }
 
-        var width = Math.Max(1, Math.Round(bounds.Width * scale));
+        var pixelSize = hostWindow.PixelSize;
-        var height = Math.Max(1, Math.Round(bounds.Height * scale));
+        var width = (double)pixelSize.Width;
+        var height = (double)pixelSize.Height;
         if (width == _drawableWidth && height == _drawableHeight)
         {
             return;
@@ -896,57 +659,6 @@ internal static partial class MetalVideoPresenter
             new CGSize { Width = width, Height = height });
     }
 
-    private static nint CreateLayer(nint device, nint window, out double drawableWidth, out double drawableHeight)
-    {
-        var contentView = MetalNative.Send(window, MetalNative.Selector("contentView"));
-        var scale = MetalNative.SendDouble(window, MetalNative.Selector("backingScaleFactor"));
-        if (scale <= 0)
-        {
-            scale = 1;
-        }
-
-        const uint width = DefaultWindowWidth;
-        const uint height = DefaultWindowHeight;
-        drawableWidth = width * scale;
-        drawableHeight = height * scale;
-
-        var layer = MetalNative.Send(
-            MetalNative.Send(MetalNative.Class("CAMetalLayer"), MetalNative.Selector("alloc")),
-            MetalNative.Selector("init"));
-        MetalNative.SendVoid(layer, MetalNative.Selector("setDevice:"), device);
-        MetalNative.Send(layer, MetalNative.Selector("setPixelFormat:"), (nint)PixelFormatBgra8Unorm);
-        // A Core Animation layer composites with its alpha channel by default, so
-        // a presented frame whose guest alpha is zero would show through as the
-        // window background (black). The presenter output is a finished opaque
-        // frame; mark the layer opaque so alpha never reaches the compositor.
-        MetalNative.SendVoidBool(layer, MetalNative.Selector("setOpaque:"), true);
-        MetalNative.SendVoidBool(layer, MetalNative.Selector("setFramebufferOnly:"), true);
-        MetalNative.SendVoidDouble(layer, MetalNative.Selector("setContentsScale:"), scale);
-        MetalNative.SendVoidSize(
-            layer,
-            MetalNative.Selector("setDrawableSize:"),
-            new CGSize { Width = drawableWidth, Height = drawableHeight });
-
-        // A manually created layer defaults to a zero-size frame, and a hosted
-        // layer's geometry is the caller's job: without this the presenter
-        // happily presents every drawable into a layer with no on-screen
-        // extent — a permanently black window.
-        MetalNative.SendVoidRect(
-            layer,
-            MetalNative.Selector("setFrame:"),
-            new CGRect { X = 0, Y = 0, Width = width, Height = height });
-
-        // wantsLayer FIRST, then the layer: that makes the metal layer the
-        // view's AppKit-managed BACKING layer (geometry and window-server
-        // commits handled by AppKit) — the SDL/GLFW pattern. The reverse order
-        // creates a layer-hosting view whose tree the app must commit itself,
-        // which never composites under a manually pumped run loop.
-        MetalNative.SendVoidBool(contentView, MetalNative.Selector("setWantsLayer:"), true);
-        MetalNative.SendVoid(contentView, MetalNative.Selector("setLayer:"), layer);
-        MetalNative.SendVoid(MetalNative.Class("CATransaction"), MetalNative.Selector("flush"));
-        return layer;
-    }
-
     private static bool TryCreatePresentPipeline(nint device, out nint pipeline, out string error)
     {
         pipeline = 0;
@@ -1105,10 +817,24 @@ internal static partial class MetalVideoPresenter
     {
         MetalNative.SendVoid(encoder, MetalNative.Selector("setRenderPipelineState:"), pipeline);
 
-        // Aspect-fit letterbox: scale the frame into the drawable via the viewport.
+        var scaleX = drawableWidth / frameWidth;
-        var scale = Math.Min(drawableWidth / frameWidth, drawableHeight / frameHeight);
+        var scaleY = drawableHeight / frameHeight;
-        var viewportWidth = frameWidth * scale;
+        var viewportWidth = drawableWidth;
-        var viewportHeight = frameHeight * scale;
+        var viewportHeight = drawableHeight;
+        if (_videoOptions.ScalingMode != HostScalingMode.Stretch)
+        {
+            var scale = _videoOptions.ScalingMode == HostScalingMode.Cover
+                ? Math.Max(scaleX, scaleY)
+                : Math.Min(scaleX, scaleY);
+            if (_videoOptions.ScalingMode == HostScalingMode.Integer && scale >= 1)
+            {
+                scale = Math.Floor(scale);
+            }
+
+            viewportWidth = frameWidth * scale;
+            viewportHeight = frameHeight * scale;
+        }
+
         MetalNative.SendVoidViewport(
             encoder,
             MetalNative.Selector("setViewport:"),

src/SharpEmu.Libs/Gpu/Vulkan/VulkanGuestGpuBackend.cs

@@ -378,8 +378,10 @@ internal sealed class VulkanGuestGpuBackend : IGuestGpuBackend
     public void SubmitGuestImageFill(ulong address, uint fillValue) =>
         VulkanVideoPresenter.SubmitGuestImageFill(address, fillValue);
 
-    public void SubmitGuestImageWrite(ulong address, byte[] pixels) =>
+    public void SubmitGuestImageWrite(ulong address, byte[] pixels, uint rowOffset = 0) =>
-        VulkanVideoPresenter.SubmitGuestImageWrite(address, pixels);
+        VulkanVideoPresenter.SubmitGuestImageWrite(address, pixels, rowOffset);
+
+    public bool SupportsPartialImageWrite => true;
 
     public void RequestCpuWrittenGuestImageSync(ulong scopeAddress = 0, ulong scopeByteCount = ulong.MaxValue) =>
         VulkanVideoPresenter.RequestCpuWrittenGuestImageSync(scopeAddress, scopeByteCount);

src/SharpEmu.Libs/VideoOut/HostDisplayCatalog.cs

@@ -0,0 +1,129 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SDL;
+using static SDL.SDL3;
+
+namespace SharpEmu.Libs.VideoOut;
+
+public sealed record HostDisplayMode(int Width, int Height, int RefreshRate);
+
+public sealed record HostDisplayInfo(
+    int Index,
+    string Name,
+    IReadOnlyList<HostDisplayMode> Modes);
+
+public static unsafe class HostDisplayCatalog
+{
+    private const SDL_InitFlags VideoFlag = SDL_InitFlags.SDL_INIT_VIDEO;
+    private static int _queryFailureLogged;
+
+    public static IReadOnlyList<HostDisplayInfo> Query()
+    {
+        var initializedHere = false;
+        var videoReady = false;
+        try
+        {
+            initializedHere = (SDL_WasInit(VideoFlag) & VideoFlag) == 0;
+            if (initializedHere && !SDL_InitSubSystem(VideoFlag))
+            {
+                LogQueryFailure(SDL_GetError() ?? "unknown SDL error");
+                return CreateFallback();
+            }
+            videoReady = true;
+
+            using var displays = SDL_GetDisplays();
+            if (displays is null || displays.Count == 0)
+            {
+                LogQueryFailure("SDL reported no displays");
+                return CreateFallback();
+            }
+
+            var result = new List<HostDisplayInfo>(displays.Count);
+            for (var index = 0; index < displays.Count; index++)
+            {
+                var display = displays[index];
+                var name = SDL_GetDisplayName(display);
+                var modes = ReadModes(display);
+                result.Add(new HostDisplayInfo(
+                    index,
+                    string.IsNullOrWhiteSpace(name) ? $"Display {index + 1}" : name,
+                    modes));
+            }
+
+            return result;
+        }
+        catch (Exception exception)
+        {
+            LogQueryFailure(exception.Message);
+            return CreateFallback();
+        }
+        finally
+        {
+            if (initializedHere && videoReady)
+            {
+                SDL_QuitSubSystem(VideoFlag);
+            }
+        }
+    }
+
+    private static IReadOnlyList<HostDisplayMode> ReadModes(SDL_DisplayID display)
+    {
+        var modes = new HashSet<HostDisplayMode>();
+        using (var fullscreenModes = SDL_GetFullscreenDisplayModes(display))
+        {
+            if (fullscreenModes is not null)
+            {
+                for (var index = 0; index < fullscreenModes.Count; index++)
+                {
+                    var mode = fullscreenModes[index];
+                    AddMode(modes, &mode);
+                }
+            }
+        }
+
+        AddMode(modes, SDL_GetDesktopDisplayMode(display));
+        if (modes.Count == 0)
+        {
+            return CreateFallbackModes();
+        }
+
+        return modes
+            .OrderByDescending(mode => (long)mode.Width * mode.Height)
+            .ThenByDescending(mode => mode.Width)
+            .ThenByDescending(mode => mode.RefreshRate)
+            .ToArray();
+    }
+
+    private static void AddMode(HashSet<HostDisplayMode> modes, SDL_DisplayMode* mode)
+    {
+        if (mode is null || mode->w <= 0 || mode->h <= 0)
+        {
+            return;
+        }
+
+        var refreshRate = mode->refresh_rate > 0
+            ? Math.Max(1, (int)Math.Round(mode->refresh_rate, MidpointRounding.AwayFromZero))
+            : 0;
+        modes.Add(new HostDisplayMode(mode->w, mode->h, refreshRate));
+    }
+
+    private static IReadOnlyList<HostDisplayInfo> CreateFallback() =>
+        [new HostDisplayInfo(0, "Display 1", CreateFallbackModes())];
+
+    private static IReadOnlyList<HostDisplayMode> CreateFallbackModes() =>
+        [
+            new HostDisplayMode(3840, 2160, 60),
+            new HostDisplayMode(2560, 1440, 60),
+            new HostDisplayMode(1920, 1080, 60),
+            new HostDisplayMode(1280, 720, 60),
+        ];
+
+    private static void LogQueryFailure(string message)
+    {
+        if (Interlocked.Exchange(ref _queryFailureLogged, 1) == 0)
+        {
+            Console.Error.WriteLine($"[GUI][WARN] SDL display query failed: {message}");
+        }
+    }
+}

src/SharpEmu.Libs/VideoOut/HostVideoOptions.cs

@@ -0,0 +1,68 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.Libs.VideoOut;
+
+using SharpEmu.Libs.Gpu.Metal;
+
+public enum HostWindowMode
+{
+    Windowed,
+    Borderless,
+    ExclusiveFullscreen,
+}
+
+public enum HostScalingMode
+{
+    Fit,
+    Cover,
+    Stretch,
+    Integer,
+}
+
+public enum HostHdrMode
+{
+    Auto,
+    On,
+    Off,
+}
+
+public sealed record HostVideoOptions
+{
+    public static HostVideoOptions Default { get; } = new();
+
+    public HostWindowMode WindowMode { get; init; } = HostWindowMode.Windowed;
+
+    public HostScalingMode ScalingMode { get; init; } = HostScalingMode.Fit;
+
+    public int Width { get; init; } = 1920;
+
+    public int Height { get; init; } = 1080;
+
+    public int DisplayIndex { get; init; }
+
+    public int RefreshRate { get; init; }
+
+    public bool VSync { get; init; } = true;
+
+    public HostHdrMode HdrMode { get; init; } = HostHdrMode.Auto;
+
+    public HostVideoOptions Normalize() => this with
+    {
+        Width = Math.Clamp(Width, 640, 16384),
+        Height = Math.Clamp(Height, 360, 16384),
+        DisplayIndex = Math.Max(0, DisplayIndex),
+        RefreshRate = Math.Clamp(RefreshRate, 0, 1000),
+        HdrMode = Enum.IsDefined(HdrMode) ? HdrMode : HostHdrMode.Auto,
+    };
+}
+
+public static class HostVideoHost
+{
+    public static bool TryConfigureVideo(HostVideoOptions options)
+    {
+        var normalized = options.Normalize();
+        return VulkanVideoPresenter.TryConfigureVideo(normalized) &
+               MetalVideoPresenter.TryConfigureVideo(normalized);
+    }
+}

src/SharpEmu.Libs/VideoOut/RenderPhaseProfile.cs

@@ -0,0 +1,252 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Diagnostics;
+
+namespace SharpEmu.Libs.VideoOut;
+
+/// <summary>
+/// Self-time accounting for the render thread, enabled with
+/// SHARPEMU_PROFILE_RENDER=1. The existing videoout counters report how much
+/// work was done (draws, pipelines, SPIR-V) but not where the render thread's
+/// second went, which is the number that decides whether a low frame rate is
+/// the emulator recording commands, the GPU executing them, or neither.
+///
+/// Scopes nest: entering a phase suspends the enclosing one and resumes it on
+/// dispose, so a <see cref="Phase.QueueSubmit"/> inside
+/// <see cref="Phase.Flush"/> is never counted twice.
+/// </summary>
+internal static class RenderPhaseProfile
+{
+    internal enum Phase
+    {
+        /// <summary>Outside any measured phase — loop overhead.</summary>
+        Unattributed = 0,
+        /// <summary>Parked because no guest work and no newer flip exist.</summary>
+        Idle,
+        /// <summary>Blocked on the frame slot's fence: the GPU is behind.</summary>
+        FrameSlotWait,
+        /// <summary>Reaping completed guest submissions (fence polls).</summary>
+        Collect,
+        Evict,
+        /// <summary>Dequeuing the next guest work item.</summary>
+        TakeWork,
+        /// <summary>Building the diagnostic label for a work item.</summary>
+        Describe,
+        /// <summary>Publishing a work item's completion to its waiters.</summary>
+        CompleteWork,
+        /// <summary>Selecting the presentation to show this iteration.</summary>
+        TakePresentation,
+        Draw,
+        Compute,
+        ColorClear,
+        ImageWrite,
+        OrderedAction,
+        Flip,
+        /// <summary>Closing and submitting the batched guest command buffer.</summary>
+        Flush,
+        /// <summary>vkQueueSubmit itself.</summary>
+        QueueSubmit,
+        /// <summary>vkAcquireNextImageKHR.</summary>
+        Acquire,
+        /// <summary>Recording + submitting the presentation command buffer.</summary>
+        Present,
+        /// <summary>vkQueuePresentKHR.</summary>
+        QueuePresent,
+        Count,
+    }
+
+    public static readonly bool Enabled = string.Equals(
+        Environment.GetEnvironmentVariable("SHARPEMU_PROFILE_RENDER"),
+        "1",
+        StringComparison.Ordinal);
+
+    /// <summary>
+    /// Breaks down the CPU-visible actions which are deliberately serialized
+    /// behind guest GPU work. This stays opt-in because it is diagnostic data,
+    /// not a normal render-thread cost.
+    /// </summary>
+    public static readonly bool OrderedActionDetailsEnabled = string.Equals(
+        Environment.GetEnvironmentVariable("SHARPEMU_PROFILE_ORDERED_ACTION"),
+        "1",
+        StringComparison.Ordinal);
+
+    private static readonly double _reportSeconds =
+        double.TryParse(
+            Environment.GetEnvironmentVariable("SHARPEMU_PROFILE_RENDER_REPORT_S"),
+            System.Globalization.CultureInfo.InvariantCulture,
+            out var seconds) && seconds > 0
+            ? seconds
+            : 5.0;
+
+    private static readonly long[] _ticks = new long[(int)Phase.Count];
+    private static readonly long[] _entries = new long[(int)Phase.Count];
+    private static long _frames;
+    private static long _windowStart = Stopwatch.GetTimestamp();
+    private static readonly Dictionary<string, OrderedActionStats> _orderedActions =
+        new(StringComparer.Ordinal);
+
+    // The render loop is single-threaded, so plain fields are enough and keep
+    // the per-scope cost to two timestamp reads.
+    [ThreadStatic] private static Phase _current;
+    [ThreadStatic] private static long _lastTimestamp;
+
+    internal readonly ref struct Scope
+    {
+        private readonly Phase _previous;
+        private readonly bool _active;
+
+        internal Scope(Phase previous)
+        {
+            _previous = previous;
+            _active = true;
+        }
+
+        public void Dispose()
+        {
+            if (!_active)
+            {
+                return;
+            }
+
+            Charge(_previous);
+        }
+    }
+
+    public static Scope Measure(Phase phase)
+    {
+        if (!Enabled)
+        {
+            return default;
+        }
+
+        var previous = Charge(phase);
+        _entries[(int)phase]++;
+        return new Scope(previous);
+    }
+
+    public static void RecordOrderedAction(string debugName, bool completed)
+    {
+        if (!OrderedActionDetailsEnabled)
+        {
+            return;
+        }
+
+        var category = GetOrderedActionCategory(debugName);
+        ref var stats = ref System.Runtime.InteropServices.CollectionsMarshal.GetValueRefOrAddDefault(
+            _orderedActions,
+            category,
+            out _);
+        stats.Executed += completed ? 1 : 0;
+        stats.Deferred += completed ? 0 : 1;
+    }
+
+    /// <summary>
+    /// Closes out the running phase and switches to <paramref name="next"/>,
+    /// returning the phase that was running.
+    /// </summary>
+    private static Phase Charge(Phase next)
+    {
+        var now = Stopwatch.GetTimestamp();
+        var previous = _current;
+        if (_lastTimestamp != 0)
+        {
+            _ticks[(int)previous] += now - _lastTimestamp;
+        }
+
+        _lastTimestamp = now;
+        _current = next;
+        return previous;
+    }
+
+    /// <summary>Called once per presented frame; also drives the report.</summary>
+    public static void RecordFrame()
+    {
+        if (!Enabled)
+        {
+            return;
+        }
+
+        _frames++;
+        var now = Stopwatch.GetTimestamp();
+        var elapsedTicks = now - _windowStart;
+        if (elapsedTicks < _reportSeconds * Stopwatch.Frequency)
+        {
+            return;
+        }
+
+        _windowStart = now;
+        var seconds = elapsedTicks / (double)Stopwatch.Frequency;
+        var frames = _frames;
+        _frames = 0;
+
+        var parts = new List<(Phase Phase, double Percent, long Entries)>((int)Phase.Count);
+        var accounted = 0L;
+        for (var index = 0; index < (int)Phase.Count; index++)
+        {
+            var phaseTicks = _ticks[index];
+            _ticks[index] = 0;
+            var entries = _entries[index];
+            _entries[index] = 0;
+            accounted += phaseTicks;
+            if (phaseTicks <= 0)
+            {
+                continue;
+            }
+
+            parts.Add(((Phase)index, phaseTicks * 100.0 / elapsedTicks, entries));
+        }
+
+        parts.Sort(static (left, right) => right.Percent.CompareTo(left.Percent));
+        Console.Error.WriteLine(
+            $"[PERF][RENDER] {seconds:F1}s fps={frames / seconds:F1} " +
+            $"covered={accounted * 100.0 / elapsedTicks:F0}% " +
+            string.Join(
+                " ",
+                parts.Select(part =>
+                    $"{part.Phase}={part.Percent:F1}%" +
+                    (part.Entries > 0 ? $"/n{part.Entries}" : string.Empty))));
+
+        if (OrderedActionDetailsEnabled && _orderedActions.Count != 0)
+        {
+            var ordered = _orderedActions
+                .OrderByDescending(static pair => pair.Value.Executed + pair.Value.Deferred)
+                .Take(12)
+                .Select(static pair =>
+                    $"{pair.Key}=ok{pair.Value.Executed}/defer{pair.Value.Deferred}");
+            Console.Error.WriteLine($"[PERF][ORDERED] {string.Join(" ", ordered)}");
+            _orderedActions.Clear();
+        }
+    }
+
+    private static string GetOrderedActionCategory(string debugName)
+    {
+        if (debugName.EndsWith(" completion", StringComparison.Ordinal))
+        {
+            return "completion";
+        }
+
+        var firstSpace = debugName.IndexOf(' ');
+        if (firstSpace < 0)
+        {
+            return debugName;
+        }
+
+        // AGC labels conventionally begin with "agc <packet>". Keeping the
+        // packet token separates DMA, submit and register traffic without
+        // retaining guest addresses in the diagnostic key.
+        if (debugName.StartsWith("agc ", StringComparison.Ordinal))
+        {
+            var secondSpace = debugName.IndexOf(' ', firstSpace + 1);
+            return secondSpace < 0 ? debugName : debugName[..secondSpace];
+        }
+
+        return debugName[..firstSpace];
+    }
+
+    private struct OrderedActionStats
+    {
+        public long Executed;
+        public long Deferred;
+    }
+}

src/SharpEmu.Libs/VideoOut/SdlHostWindow.cs

@@ -0,0 +1,832 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Diagnostics;
+using System.Runtime.InteropServices;
+using SharpEmu.HLE.Host;
+using SharpEmu.Libs.Pad;
+using SDL;
+using Silk.NET.Vulkan;
+using static SDL.SDL3;
+
+namespace SharpEmu.Libs.VideoOut;
+
+internal enum SdlGraphicsApi
+{
+    Vulkan,
+    Metal,
+}
+
+internal readonly record struct SdlHdrState(
+    bool Enabled,
+    float SdrWhiteLevel,
+    float Headroom);
+
+internal sealed unsafe class SdlHostWindow : IDisposable, IHostGamepadOutput
+{
+    private const SDL_InitFlags InitFlags = SDL_InitFlags.SDL_INIT_VIDEO | SDL_InitFlags.SDL_INIT_GAMEPAD;
+    private static readonly long CursorHideDelayTicks = 2 * Stopwatch.Frequency;
+    private const uint OutputDurationMs = 5_000;
+
+    private readonly HostVideoOptions _options;
+    private readonly SdlGraphicsApi _graphicsApi;
+    private readonly Action? _toggleBackendHud;
+    private readonly object _gamepadGate = new();
+    private SDL_Window* _window;
+    private nint _metalView;
+    private SDL_Gamepad* _gamepad;
+    private HostGamepadType _gamepadType;
+    private byte _leftTriggerRumble;
+    private byte _rightTriggerRumble;
+    private int _closeRequested;
+    private bool _closedByUser;
+    private bool _fullscreen;
+    private bool _focused = true;
+    private bool _cursorVisible = true;
+    private long _cursorHideDeadline;
+    private bool _surfaceRestorePending;
+    private bool _hdrStateChangePending;
+    private bool _disposed;
+
+    public SdlHostWindow(
+        string title,
+        HostVideoOptions options,
+        SdlGraphicsApi graphicsApi,
+        Action? toggleBackendHud = null)
+    {
+        _options = options.Normalize();
+        _graphicsApi = graphicsApi;
+        _toggleBackendHud = toggleBackendHud;
+        SdlGamepadStateReader.EnableSonyHidApi();
+        if (!SDL_InitSubSystem(InitFlags))
+        {
+            throw new InvalidOperationException($"SDL video initialization failed: {GetError()}");
+        }
+
+        var graphicsFlag = graphicsApi == SdlGraphicsApi.Vulkan
+            ? SDL_WindowFlags.SDL_WINDOW_VULKAN
+            : SDL_WindowFlags.SDL_WINDOW_METAL;
+        var flags = graphicsFlag |
+                    SDL_WindowFlags.SDL_WINDOW_RESIZABLE |
+                    SDL_WindowFlags.SDL_WINDOW_HIGH_PIXEL_DENSITY |
+                    SDL_WindowFlags.SDL_WINDOW_HIDDEN;
+        _window = CreateWindow(title, _options.Width, _options.Height, flags);
+        if (_window is null)
+        {
+            SDL_QuitSubSystem(InitFlags);
+            throw new InvalidOperationException($"SDL window creation failed: {GetError()}");
+        }
+
+        MoveToConfiguredDisplay();
+        ApplyConfiguredMode(_options.WindowMode);
+        SetIcon();
+        SDL_ShowWindow(_window);
+        SDL_RaiseWindow(_window);
+        if (_fullscreen && !SDL_SyncWindow(_window))
+        {
+            Console.Error.WriteLine($"[LOADER][WARN] SDL initial fullscreen sync failed: {GetError()}");
+        }
+        HostWindowInput.Connect(this);
+        OpenFirstGamepad();
+
+        Console.Error.WriteLine(
+            $"[LOADER][INFO] SDL3 window ready: mode={_options.WindowMode} " +
+            $"size={_options.Width}x{_options.Height} display={_options.DisplayIndex} " +
+            $"refresh={(_options.RefreshRate == 0 ? "auto" : _options.RefreshRate)}");
+        var hdr = HdrState;
+        Console.Error.WriteLine(
+            $"[LOADER][INFO] SDL3 display HDR: enabled={hdr.Enabled} " +
+            $"sdr_white={hdr.SdrWhiteLevel:F3} headroom={hdr.Headroom:F3}");
+    }
+
+    public (int Width, int Height) PixelSize
+    {
+        get
+        {
+            var width = 0;
+            var height = 0;
+            if (_window is not null)
+            {
+                SDL_GetWindowSizeInPixels(_window, &width, &height);
+            }
+
+            return (Math.Max(width, 1), Math.Max(height, 1));
+        }
+    }
+
+    public bool IsMinimized =>
+        _window is not null &&
+        (SDL_GetWindowFlags(_window) & SDL_WindowFlags.SDL_WINDOW_MINIMIZED) != 0;
+
+    public bool ClosedByUser => _closedByUser;
+
+    public SdlHdrState HdrState
+    {
+        get
+        {
+            if (_window is null)
+            {
+                return default;
+            }
+
+            var properties = SDL_GetWindowProperties(_window);
+            fixed (byte* hdrEnabledName = SDL_PROP_WINDOW_HDR_ENABLED_BOOLEAN)
+            fixed (byte* sdrWhiteName = SDL_PROP_WINDOW_SDR_WHITE_LEVEL_FLOAT)
+            fixed (byte* headroomName = SDL_PROP_WINDOW_HDR_HEADROOM_FLOAT)
+            {
+                return new SdlHdrState(
+                    (bool)SDL_GetBooleanProperty(properties, hdrEnabledName, false),
+                    SDL_GetFloatProperty(properties, sdrWhiteName, 1f),
+                    SDL_GetFloatProperty(properties, headroomName, 1f));
+            }
+        }
+    }
+
+    public bool ConsumeSurfaceRestore()
+    {
+        var pending = _surfaceRestorePending;
+        _surfaceRestorePending = false;
+        return pending;
+    }
+
+    public bool ConsumeHdrStateChange()
+    {
+        var pending = _hdrStateChangePending;
+        _hdrStateChangePending = false;
+        return pending;
+    }
+
+    public byte** GetRequiredVulkanInstanceExtensions(out uint count)
+    {
+        EnsureGraphicsApi(SdlGraphicsApi.Vulkan);
+        uint extensionCount = 0;
+        var extensions = SDL_Vulkan_GetInstanceExtensions(&extensionCount);
+        count = extensionCount;
+        if (extensions is null || count == 0)
+        {
+            throw new InvalidOperationException($"SDL did not provide Vulkan instance extensions: {GetError()}");
+        }
+
+        return extensions;
+    }
+
+    public SurfaceKHR CreateVulkanSurface(Instance instance)
+    {
+        EnsureGraphicsApi(SdlGraphicsApi.Vulkan);
+        VkSurfaceKHR_T* surface = null;
+        if (!SDL_Vulkan_CreateSurface(
+                _window,
+                (VkInstance_T*)instance.Handle,
+                null,
+                &surface) || surface is null)
+        {
+            throw new InvalidOperationException($"SDL Vulkan surface creation failed: {GetError()}");
+        }
+
+        return new SurfaceKHR(unchecked((ulong)surface));
+    }
+
+    public nint CreateMetalLayer()
+    {
+        EnsureGraphicsApi(SdlGraphicsApi.Metal);
+        if (_metalView == 0)
+        {
+            _metalView = SDL_Metal_CreateView(_window);
+            if (_metalView == 0)
+            {
+                throw new InvalidOperationException($"SDL Metal view creation failed: {GetError()}");
+            }
+        }
+
+        var layer = SDL_Metal_GetLayer(_metalView);
+        if (layer == 0)
+        {
+            throw new InvalidOperationException($"SDL Metal layer lookup failed: {GetError()}");
+        }
+
+        return layer;
+    }
+
+    public void SetTitle(string title)
+    {
+        if (_window is null)
+        {
+            return;
+        }
+
+        var utf8 = Marshal.StringToCoTaskMemUTF8(title);
+        try
+        {
+            SDL_SetWindowTitle(_window, (byte*)utf8);
+        }
+        finally
+        {
+            Marshal.FreeCoTaskMem(utf8);
+        }
+    }
+
+    public void Close() => Volatile.Write(ref _closeRequested, 1);
+
+    public void SetRumble(byte largeMotor, byte smallMotor)
+    {
+        lock (_gamepadGate)
+        {
+            if (IsGamepadConnected())
+            {
+                SDL_RumbleGamepad(
+                    _gamepad,
+                    ExpandByte(largeMotor),
+                    ExpandByte(smallMotor),
+                    OutputDurationMs);
+            }
+        }
+    }
+
+    public void SetTriggerRumble(byte? leftTrigger, byte? rightTrigger)
+    {
+        lock (_gamepadGate)
+        {
+            if (IsGamepadConnected())
+            {
+                if (leftTrigger is { } left)
+                {
+                    _leftTriggerRumble = left;
+                }
+                if (rightTrigger is { } right)
+                {
+                    _rightTriggerRumble = right;
+                }
+
+                SDL_RumbleGamepadTriggers(
+                    _gamepad,
+                    ExpandByte(_leftTriggerRumble),
+                    ExpandByte(_rightTriggerRumble),
+                    OutputDurationMs);
+            }
+        }
+    }
+
+    public void SetAdaptiveTriggerEffect(
+        HostAdaptiveTriggerEffect? leftTrigger,
+        HostAdaptiveTriggerEffect? rightTrigger)
+    {
+        lock (_gamepadGate)
+        {
+            if (!IsGamepadConnected())
+            {
+                return;
+            }
+
+            if (_gamepadType != HostGamepadType.DualSense)
+            {
+                if (leftTrigger is { } fallbackLeft)
+                {
+                    _leftTriggerRumble = fallbackLeft.FallbackStrength;
+                }
+                if (rightTrigger is { } fallbackRight)
+                {
+                    _rightTriggerRumble = fallbackRight.FallbackStrength;
+                }
+
+                SDL_RumbleGamepadTriggers(
+                    _gamepad,
+                    ExpandByte(_leftTriggerRumble),
+                    ExpandByte(_rightTriggerRumble),
+                    OutputDurationMs);
+                return;
+            }
+
+            Span<byte> state = stackalloc byte[47];
+            state.Clear();
+            if (rightTrigger is { } nativeRight)
+            {
+                state[0] |= 0x04;
+                nativeRight.CopyTo(state[10..21]);
+            }
+            if (leftTrigger is { } nativeLeft)
+            {
+                state[0] |= 0x08;
+                nativeLeft.CopyTo(state[21..32]);
+            }
+
+            if (state[0] == 0)
+            {
+                return;
+            }
+
+            fixed (byte* data = state)
+            {
+                SDL_SendGamepadEffect(_gamepad, (nint)data, state.Length);
+            }
+        }
+    }
+
+    public void SetLightbar(byte red, byte green, byte blue)
+    {
+        lock (_gamepadGate)
+        {
+            if (IsGamepadConnected())
+            {
+                SDL_SetGamepadLED(_gamepad, red, green, blue);
+            }
+        }
+    }
+
+    public void ResetLightbar() => SetLightbar(0, 0, 64);
+
+    public void Run(
+        Action initialize,
+        Action<double> render,
+        Action closing,
+        Action? idle = null)
+    {
+        initialize();
+        var timer = Stopwatch.StartNew();
+        var last = timer.Elapsed.TotalSeconds;
+        try
+        {
+            while (Volatile.Read(ref _closeRequested) == 0)
+            {
+                PumpEvents();
+                if (Volatile.Read(ref _closeRequested) != 0)
+                {
+                    break;
+                }
+
+                UpdateCursorAutoHide();
+                SampleGamepad();
+                var now = timer.Elapsed.TotalSeconds;
+                render(now - last);
+                last = now;
+                if (IsMinimized)
+                {
+                    SDL_Delay(10);
+                }
+                else
+                {
+                    idle?.Invoke();
+                }
+            }
+        }
+        finally
+        {
+            closing();
+        }
+    }
+
+    public void Dispose()
+    {
+        if (_disposed)
+        {
+            return;
+        }
+
+        _disposed = true;
+        SDL_ShowCursor();
+        HostWindowInput.Disconnect();
+        CloseGamepad();
+        if (_metalView != 0)
+        {
+            SDL_Metal_DestroyView(_metalView);
+            _metalView = 0;
+        }
+
+        if (_window is not null)
+        {
+            SDL_DestroyWindow(_window);
+            _window = null;
+        }
+
+        SDL_QuitSubSystem(InitFlags);
+    }
+
+    private void PumpEvents()
+    {
+        SDL_Event windowEvent;
+        while (SDL_PollEvent(&windowEvent))
+        {
+            switch (windowEvent.Type)
+            {
+                case SDL_EventType.SDL_EVENT_QUIT:
+                case SDL_EventType.SDL_EVENT_WINDOW_CLOSE_REQUESTED:
+                    _closedByUser = true;
+                    Volatile.Write(ref _closeRequested, 1);
+                    break;
+                case SDL_EventType.SDL_EVENT_WINDOW_FOCUS_GAINED:
+                    _focused = true;
+                    UpdateCursorVisibility();
+                    HostWindowInput.SetFocused(true);
+                    break;
+                case SDL_EventType.SDL_EVENT_WINDOW_FOCUS_LOST:
+                    _focused = false;
+                    UpdateCursorVisibility();
+                    HostWindowInput.SetFocused(false);
+                    break;
+                case SDL_EventType.SDL_EVENT_WINDOW_MINIMIZED:
+                    _focused = false;
+                    UpdateCursorVisibility();
+                    HostWindowInput.SetFocused(false);
+                    break;
+                case SDL_EventType.SDL_EVENT_WINDOW_RESTORED:
+                case SDL_EventType.SDL_EVENT_WINDOW_RESIZED:
+                case SDL_EventType.SDL_EVENT_WINDOW_PIXEL_SIZE_CHANGED:
+                case SDL_EventType.SDL_EVENT_WINDOW_MAXIMIZED:
+                case SDL_EventType.SDL_EVENT_WINDOW_ENTER_FULLSCREEN:
+                case SDL_EventType.SDL_EVENT_WINDOW_LEAVE_FULLSCREEN:
+                case SDL_EventType.SDL_EVENT_WINDOW_EXPOSED:
+                case SDL_EventType.SDL_EVENT_WINDOW_DISPLAY_CHANGED:
+                case SDL_EventType.SDL_EVENT_WINDOW_DISPLAY_SCALE_CHANGED:
+                    _surfaceRestorePending = true;
+                    break;
+                case SDL_EventType.SDL_EVENT_WINDOW_HDR_STATE_CHANGED:
+                    _hdrStateChangePending = true;
+                    break;
+                case SDL_EventType.SDL_EVENT_KEY_DOWN:
+                case SDL_EventType.SDL_EVENT_KEY_UP:
+                    HandleKey(windowEvent.key);
+                    break;
+                case SDL_EventType.SDL_EVENT_MOUSE_MOTION:
+                    ShowCursorTemporarily();
+                    break;
+                case SDL_EventType.SDL_EVENT_MOUSE_BUTTON_DOWN:
+                    ShowCursorTemporarily();
+                    if (windowEvent.button.button == 1 && windowEvent.button.clicks == 2)
+                    {
+                        ToggleFullscreen();
+                    }
+                    break;
+                case SDL_EventType.SDL_EVENT_GAMEPAD_ADDED:
+                    if (_gamepad is null)
+                    {
+                        OpenFirstGamepad();
+                    }
+                    break;
+                case SDL_EventType.SDL_EVENT_GAMEPAD_REMOVED:
+                    if (_gamepad is not null && !SDL_GamepadConnected(_gamepad))
+                    {
+                        CloseGamepad();
+                        OpenFirstGamepad();
+                    }
+                    break;
+            }
+        }
+    }
+
+    private void HandleKey(SDL_KeyboardEvent keyEvent)
+    {
+        var down = keyEvent.type == SDL_EventType.SDL_EVENT_KEY_DOWN;
+        if (down && !keyEvent.repeat)
+        {
+            if (keyEvent.key == SDL_Keycode.SDLK_F1 &&
+                (keyEvent.mod & SDL_Keymod.SDL_KMOD_GUI) != 0 &&
+                _toggleBackendHud is not null)
+            {
+                _toggleBackendHud();
+            }
+            else if (keyEvent.key == SDL_Keycode.SDLK_F1)
+            {
+                PerfOverlay.Toggle();
+            }
+            else if (keyEvent.key == SDL_Keycode.SDLK_F11)
+            {
+                ToggleFullscreen();
+            }
+        }
+
+        if (TryMapVirtualKey(keyEvent.key, out var virtualKey))
+        {
+            HostWindowInput.SetKey(virtualKey, down);
+        }
+    }
+
+    private void ToggleFullscreen()
+    {
+        if (_fullscreen)
+        {
+            SDL_SetWindowFullscreen(_window, false);
+            SDL_SetWindowFullscreenMode(_window, null);
+            SDL_SetWindowResizable(_window, true);
+            SDL_SetWindowSize(_window, _options.Width, _options.Height);
+            MoveToConfiguredDisplay();
+            _fullscreen = false;
+            UpdateCursorVisibility();
+            return;
+        }
+
+        ApplyConfiguredMode(
+            _options.WindowMode == HostWindowMode.ExclusiveFullscreen
+                ? HostWindowMode.ExclusiveFullscreen
+                : HostWindowMode.Borderless);
+    }
+
+    private void ApplyConfiguredMode(HostWindowMode mode)
+    {
+        if (mode == HostWindowMode.Windowed)
+        {
+            _fullscreen = false;
+            UpdateCursorVisibility();
+            return;
+        }
+
+        if (mode == HostWindowMode.ExclusiveFullscreen)
+        {
+            var display = GetConfiguredDisplay();
+            SDL_DisplayMode closest;
+            if (SDL_GetClosestFullscreenDisplayMode(
+                    display,
+                    _options.Width,
+                    _options.Height,
+                    _options.RefreshRate,
+                    true,
+                    &closest))
+            {
+                SDL_SetWindowFullscreenMode(_window, &closest);
+            }
+            else
+            {
+                Console.Error.WriteLine(
+                    $"[LOADER][WARN] SDL exclusive mode unavailable; using borderless: {GetError()}");
+                SDL_SetWindowFullscreenMode(_window, null);
+            }
+        }
+        else
+        {
+            SDL_SetWindowFullscreenMode(_window, null);
+        }
+
+        SDL_SetWindowFullscreen(_window, true);
+        _fullscreen = true;
+        UpdateCursorVisibility();
+    }
+
+    private void UpdateCursorVisibility()
+    {
+        _cursorHideDeadline = 0;
+        SetCursorVisible(!_fullscreen || !_focused);
+    }
+
+    private void ShowCursorTemporarily()
+    {
+        if (!_fullscreen || !_focused)
+        {
+            return;
+        }
+
+        SetCursorVisible(true);
+        _cursorHideDeadline = Stopwatch.GetTimestamp() + CursorHideDelayTicks;
+    }
+
+    private void UpdateCursorAutoHide()
+    {
+        if (!_fullscreen || !_focused || !_cursorVisible || _cursorHideDeadline == 0 ||
+            Stopwatch.GetTimestamp() < _cursorHideDeadline)
+        {
+            return;
+        }
+
+        _cursorHideDeadline = 0;
+        SetCursorVisible(false);
+    }
+
+    private void SetCursorVisible(bool visible)
+    {
+        if (_cursorVisible == visible)
+        {
+            return;
+        }
+
+        if (visible)
+        {
+            SDL_ShowCursor();
+        }
+        else
+        {
+            SDL_HideCursor();
+        }
+
+        _cursorVisible = visible;
+    }
+
+    private void MoveToConfiguredDisplay()
+    {
+        var display = GetConfiguredDisplay();
+        SDL_Rect bounds;
+        if (SDL_GetDisplayBounds(display, &bounds))
+        {
+            var x = bounds.x + Math.Max(0, (bounds.w - _options.Width) / 2);
+            var y = bounds.y + Math.Max(0, (bounds.h - _options.Height) / 2);
+            SDL_SetWindowPosition(_window, x, y);
+        }
+    }
+
+    private SDL_DisplayID GetConfiguredDisplay()
+    {
+        using var displays = SDL_GetDisplays();
+        if (displays is null || displays.Count == 0)
+        {
+            return SDL_GetPrimaryDisplay();
+        }
+
+        return displays[Math.Clamp(_options.DisplayIndex, 0, displays.Count - 1)];
+    }
+
+    private void OpenFirstGamepad()
+    {
+        lock (_gamepadGate)
+        {
+            _gamepad = SdlGamepadStateReader.OpenPreferredGamepad();
+            if (_gamepad is null)
+            {
+                return;
+            }
+
+            _gamepadType = SdlGamepadStateReader.MapGamepadType(SDL_GetRealGamepadType(_gamepad));
+            EnableSensor(SDL_SensorType.SDL_SENSOR_ACCEL);
+            EnableSensor(SDL_SensorType.SDL_SENSOR_GYRO);
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][INFO] SDL gamepad connected: {DescribeGamepad()} " +
+            $"type={_gamepadType} connection={SdlGamepadStateReader.GetConnection(_gamepad)}");
+        SampleGamepad();
+    }
+
+    private void CloseGamepad()
+    {
+        lock (_gamepadGate)
+        {
+            if (_gamepad is not null)
+            {
+                SDL_CloseGamepad(_gamepad);
+                _gamepad = null;
+            }
+
+            _gamepadType = HostGamepadType.Generic;
+            _leftTriggerRumble = 0;
+            _rightTriggerRumble = 0;
+        }
+
+        HostWindowInput.ClearGamepad();
+    }
+
+    private void SampleGamepad()
+    {
+        lock (_gamepadGate)
+        {
+            if (!IsGamepadConnected())
+            {
+                HostWindowInput.ClearGamepad();
+                return;
+            }
+
+            SDL_UpdateGamepads();
+            var state = SdlGamepadStateReader.Read(_gamepad) with
+            {
+                Motion = ReadMotion(),
+                Touch = ReadTouch(),
+            };
+            HostWindowInput.SetGamepad(
+                DescribeGamepad(),
+                state);
+        }
+    }
+
+    private void EnableSensor(SDL_SensorType sensor)
+    {
+        if (SDL_GamepadHasSensor(_gamepad, sensor))
+        {
+            SDL_SetGamepadSensorEnabled(_gamepad, sensor, true);
+        }
+    }
+
+    private HostMotionState ReadMotion()
+    {
+        Span<float> acceleration = stackalloc float[3];
+        Span<float> angularVelocity = stackalloc float[3];
+        acceleration.Clear();
+        angularVelocity.Clear();
+        var hasAcceleration = false;
+        var hasAngularVelocity = false;
+        fixed (float* data = acceleration)
+        {
+            hasAcceleration = SDL_GamepadHasSensor(_gamepad, SDL_SensorType.SDL_SENSOR_ACCEL) &&
+                SDL_GetGamepadSensorData(_gamepad, SDL_SensorType.SDL_SENSOR_ACCEL, data, acceleration.Length);
+        }
+        fixed (float* data = angularVelocity)
+        {
+            hasAngularVelocity = SDL_GamepadHasSensor(_gamepad, SDL_SensorType.SDL_SENSOR_GYRO) &&
+                SDL_GetGamepadSensorData(_gamepad, SDL_SensorType.SDL_SENSOR_GYRO, data, angularVelocity.Length);
+        }
+
+        return new HostMotionState(
+            hasAcceleration || hasAngularVelocity,
+            acceleration[0],
+            acceleration[1],
+            acceleration[2],
+            angularVelocity[0],
+            angularVelocity[1],
+            angularVelocity[2]);
+    }
+
+    private HostTouchState ReadTouch()
+    {
+        if (SDL_GetNumGamepadTouchpads(_gamepad) <= 0)
+        {
+            return default;
+        }
+
+        return new HostTouchState(ReadTouchPoint(0), ReadTouchPoint(1));
+    }
+
+    private HostTouchPoint ReadTouchPoint(int finger)
+    {
+        if (SDL_GetNumGamepadTouchpadFingers(_gamepad, 0) <= finger)
+        {
+            return default;
+        }
+
+        SDLBool down = false;
+        float x = 0;
+        float y = 0;
+        float pressure = 0;
+        if (!SDL_GetGamepadTouchpadFinger(_gamepad, 0, finger, &down, &x, &y, &pressure))
+        {
+            return default;
+        }
+
+        return new HostTouchPoint(down, (byte)finger, Math.Clamp(x, 0, 1), Math.Clamp(y, 0, 1));
+    }
+
+    private bool IsGamepadConnected() => _gamepad is not null && SDL_GamepadConnected(_gamepad);
+
+    private string DescribeGamepad() =>
+        Marshal.PtrToStringUTF8((nint)Unsafe_SDL_GetGamepadName(_gamepad)) ?? "SDL gamepad";
+
+    private static ushort ExpandByte(byte value) => (ushort)(value * 257);
+
+    private static bool TryMapVirtualKey(SDL_Keycode key, out int virtualKey)
+    {
+        virtualKey = key switch
+        {
+            SDL_Keycode.SDLK_BACKSPACE => 0x08,
+            SDL_Keycode.SDLK_TAB => 0x09,
+            SDL_Keycode.SDLK_RETURN => 0x0D,
+            SDL_Keycode.SDLK_ESCAPE => 0x1B,
+            SDL_Keycode.SDLK_LEFT => 0x25,
+            SDL_Keycode.SDLK_UP => 0x26,
+            SDL_Keycode.SDLK_RIGHT => 0x27,
+            SDL_Keycode.SDLK_DOWN => 0x28,
+            >= SDL_Keycode.SDLK_A and <= SDL_Keycode.SDLK_Z => 0x41 + (int)(key - SDL_Keycode.SDLK_A),
+            _ => 0,
+        };
+        return virtualKey != 0;
+    }
+
+    private void SetIcon()
+    {
+        if (!PngSplashLoader.TryLoadIcon(out var pixels, out var width, out var height))
+        {
+            return;
+        }
+
+        fixed (byte* data = pixels)
+        {
+            var surface = SDL_CreateSurfaceFrom(
+                (int)width,
+                (int)height,
+                SDL_PixelFormat.SDL_PIXELFORMAT_ABGR8888,
+                (nint)data,
+                checked((int)width * 4));
+            if (surface is not null)
+            {
+                SDL_SetWindowIcon(_window, surface);
+                SDL_DestroySurface(surface);
+            }
+        }
+    }
+
+    private static SDL_Window* CreateWindow(string title, int width, int height, SDL_WindowFlags flags)
+    {
+        var utf8 = Marshal.StringToCoTaskMemUTF8(title);
+        try
+        {
+            return SDL_CreateWindow((byte*)utf8, width, height, flags);
+        }
+        finally
+        {
+            Marshal.FreeCoTaskMem(utf8);
+        }
+    }
+
+    private static string GetError() =>
+        Marshal.PtrToStringUTF8((nint)Unsafe_SDL_GetError()) ?? "unknown SDL error";
+
+    private void EnsureGraphicsApi(SdlGraphicsApi expected)
+    {
+        if (_graphicsApi != expected)
+        {
+            throw new InvalidOperationException(
+                $"SDL host window uses {_graphicsApi}, not {expected}.");
+        }
+    }
+}

src/SharpEmu.Libs/VideoOut/VideoOutExports.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Logging;
 using SharpEmu.HLE.Host;
 using SharpEmu.Libs.Diagnostics;
 using SharpEmu.Libs.Gpu;
@@ -69,11 +70,14 @@ public static class VideoOutExports
     private static readonly Dictionary<int, VideoOutPortState> _ports = new();
     private static int _presentationWindowCloseNotified;
     private static int _vblankStopRequested;
+    private static int _hdrOutputRequested;
     private static readonly Dictionary<(int Handle, int BufferIndex, ulong Address), ulong> _lastFrameFingerprints = new();
     private static int _nextHandle = 1;
     private static int _frameDumpCount;
     private static long _nextFrameDumpIndex;
-    private static string _windowTitle = "SharpEmu VideoOut";
+    private static string _applicationWindowTitle = "VideoOut";
+    private static string _selectedGpuName = string.Empty;
+    private static string _applicationTitleId = "UNKNOWN";
     private static readonly bool _logFrameRate = string.Equals(
         Environment.GetEnvironmentVariable("SHARPEMU_LOG_VIDEOOUT_FPS"),
         "1",
@@ -113,7 +117,18 @@ public static class VideoOutExports
         var versionSuffix = string.IsNullOrWhiteSpace(version) ? string.Empty : $" v{version.Trim()}";
         lock (_stateGate)
         {
-            _windowTitle = $"SharpEmu - {application}{versionSuffix}";
+            _applicationTitleId = string.IsNullOrWhiteSpace(titleId)
+                ? "UNKNOWN"
+                : titleId.Trim();
+            _applicationWindowTitle = $"{application}{versionSuffix}";
+        }
+    }
+
+    internal static string GetApplicationTitleId()
+    {
+        lock (_stateGate)
+        {
+            return _applicationTitleId;
         }
     }
 
@@ -121,7 +136,10 @@ public static class VideoOutExports
     {
         lock (_stateGate)
         {
-            return _windowTitle;
+            var gpuSuffix = string.IsNullOrWhiteSpace(_selectedGpuName)
+                ? string.Empty
+                : $" · {_selectedGpuName}";
+            return $"SharpEmu · {BuildInfo.CommitSha ?? "dev"} - {_applicationWindowTitle}{gpuSuffix}";
         }
     }
 
@@ -139,7 +157,7 @@ public static class VideoOutExports
             : string.Empty;
         lock (_stateGate)
         {
-            _windowTitle = $"{_windowTitle} · {gpuName.Trim()}{backendSuffix}";
+            _selectedGpuName = $"{gpuName.Trim()}{backendSuffix}";
         }
     }
 
@@ -166,30 +184,17 @@ public static class VideoOutExports
     private static void RequestHostShutdown(string reason)
     {
         Console.Error.WriteLine($"[LOADER][INFO] Host shutdown requested: {reason}");
-        var embedded = VulkanVideoHost.IsEmbedded;
         AudioOutExports.ShutdownAllPorts();
         Interlocked.Exchange(ref _vblankStopRequested, 1);
         HostSessionControl.RequestShutdown(reason);
+        GuestGpu.Current.RequestClose();
 
-        // A hosted game can still be issuing AGC work after it requests its
+        // Give guest and GPU threads a bounded window to leave cooperatively.
-        // own shutdown. Keep the presenter's resources alive until the GUI
+        ThreadPool.QueueUserWorkItem(static _ =>
-        // session reaches its guest-safe exit path and disposes the host
-        // surface.
-        if (!embedded)
         {
-            GuestGpu.Current.RequestClose();
+            Thread.Sleep(2000);
-        }
+            Environment.Exit(0);
-
+        });
-        // The embedded GUI owns the process lifetime. A guest shutdown should
-        // end only that session rather than terminating the launcher itself.
-        if (!embedded)
-        {
-            ThreadPool.QueueUserWorkItem(static _ =>
-            {
-                Thread.Sleep(2000);
-                Environment.Exit(0);
-            });
-        }
     }
 
     private sealed class VideoOutPortState
@@ -871,6 +876,8 @@ public static class VideoOutExports
         }
     }
 
+    internal static bool IsHdrOutputRequested => Volatile.Read(ref _hdrOutputRequested) != 0;
+
     [SysAbiExport(
         Nid = "MTxxrOCeSig",
         ExportName = "sceVideoOutSetWindowModeMargins",
@@ -1175,16 +1182,21 @@ public static class VideoOutExports
 
         var guestImageSubmitted = false;
         ulong guestImageAddress = 0;
-        if (submitGpuImage &&
+        if (bufferIndex >= 0 &&
-            bufferIndex >= 0 &&
             TryGetDisplayBufferInfo(handle, bufferIndex, out var displayBuffer))
         {
+            Interlocked.Exchange(
+                ref _hdrOutputRequested,
+                IsHdrPixelFormat(displayBuffer.PixelFormat) ? 1 : 0);
             guestImageAddress = displayBuffer.Address;
-            guestImageSubmitted = GuestGpu.Current.TrySubmitGuestImage(
+            if (submitGpuImage)
-                displayBuffer.Address,
+            {
-                displayBuffer.Width,
+                guestImageSubmitted = GuestGpu.Current.TrySubmitGuestImage(
-                displayBuffer.Height,
+                    displayBuffer.Address,
-                displayBuffer.PitchInPixel);
+                    displayBuffer.Width,
+                    displayBuffer.Height,
+                    displayBuffer.PitchInPixel);
+            }
         }
 
         if (_dumpVideoOut)
@@ -1711,6 +1723,12 @@ public static class VideoOutExports
     internal static bool IsPacked10BitPixelFormat(ulong pixelFormat) =>
         IsPacked10BitPixelFormatNormalized(NormalizePixelFormat(pixelFormat));
 
+    internal static bool IsHdrPixelFormat(ulong pixelFormat) =>
+        NormalizePixelFormat(pixelFormat) is
+            SceVideoOutPixelFormatA2R10G10B10Bt2020Pq or
+            SceVideoOutPixelFormat2R10G10B10A2Bt2100Pq or
+            SceVideoOutPixelFormat2B10G10R10A2Bt2100Pq;
+
     private static bool IsPacked10BitPixelFormatNormalized(ulong pixelFormat) =>
         pixelFormat is
             SceVideoOutPixelFormatA2R10G10B10 or

src/SharpEmu.Libs/VideoOut/VulkanHostSurface.cs

@@ -1,270 +0,0 @@
-// Copyright (C) 2026 SharpEmu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-namespace SharpEmu.Libs.VideoOut;
-
-/// <summary>
-/// A native child surface owned by a host UI. The presenter consumes this
-/// directly, avoiding a second top-level GLFW window when the GUI is active.
-/// </summary>
-public enum VulkanHostSurfaceKind
-{
-    Win32,
-    Xlib,
-    Metal,
-}
-
-/// <summary>
-/// Platform-native handles required to create a Vulkan presentation surface.
-/// The GUI owns their lifetime and updates the physical pixel size on resize.
-/// </summary>
-public sealed class VulkanHostSurface : IDisposable
-{
-    private int _pixelWidth;
-    private int _pixelHeight;
-    private int _resizeGeneration;
-    private readonly bool _ownsDisplay;
-    private readonly bool _pollNativeSize;
-    private long _nextNativeSizePoll;
-
-    public VulkanHostSurface(
-        VulkanHostSurfaceKind kind,
-        nint windowHandle,
-        nint displayHandle = 0,
-        nint metalLayerHandle = 0,
-        bool ownsDisplay = false,
-        bool pollNativeSize = false)
-    {
-        Kind = kind;
-        WindowHandle = windowHandle;
-        DisplayHandle = displayHandle;
-        MetalLayerHandle = metalLayerHandle;
-        _ownsDisplay = ownsDisplay;
-        _pollNativeSize = pollNativeSize;
-    }
-
-    public VulkanHostSurfaceKind Kind { get; }
-
-    public nint WindowHandle { get; }
-
-    /// <summary>X11 Display* when <see cref="Kind"/> is <see cref="VulkanHostSurfaceKind.Xlib"/>.</summary>
-    public nint DisplayHandle { get; }
-
-    /// <summary>CAMetalLayer* when <see cref="Kind"/> is <see cref="VulkanHostSurfaceKind.Metal"/>.</summary>
-    public nint MetalLayerHandle { get; }
-
-    public int PixelWidth => Volatile.Read(ref _pixelWidth);
-
-    public int PixelHeight => Volatile.Read(ref _pixelHeight);
-
-    internal int ResizeGeneration => Volatile.Read(ref _resizeGeneration);
-
-    public void UpdatePixelSize(int width, int height)
-    {
-        width = Math.Max(width, 1);
-        height = Math.Max(height, 1);
-        if (Volatile.Read(ref _pixelWidth) == width && Volatile.Read(ref _pixelHeight) == height)
-        {
-            return;
-        }
-
-        Volatile.Write(ref _pixelWidth, width);
-        Volatile.Write(ref _pixelHeight, height);
-        Interlocked.Increment(ref _resizeGeneration);
-    }
-
-    /// <summary>
-    /// The child emulator cannot receive Avalonia resize notifications. Poll
-    /// the native host at a bounded rate so embedded child swapchains still
-    /// follow normal resize and F11 transitions.
-    /// </summary>
-    internal void RefreshChildProcessPixelSize()
-    {
-        if (!_pollNativeSize)
-        {
-            return;
-        }
-
-        var now = System.Diagnostics.Stopwatch.GetTimestamp();
-        var due = Volatile.Read(ref _nextNativeSizePoll);
-        if (now < due || Interlocked.CompareExchange(
-                ref _nextNativeSizePoll,
-                now + (System.Diagnostics.Stopwatch.Frequency / 8),
-                due) != due)
-        {
-            return;
-        }
-
-        if (Kind == VulkanHostSurfaceKind.Win32 && GetClientRect(WindowHandle, out var rect))
-        {
-            UpdatePixelSize(rect.Right - rect.Left, rect.Bottom - rect.Top);
-            return;
-        }
-
-        if (Kind == VulkanHostSurfaceKind.Xlib && DisplayHandle != 0 &&
-            XGetGeometry(
-                DisplayHandle,
-                WindowHandle,
-                out _,
-                out _,
-                out _,
-                out var width,
-                out var height,
-                out _,
-                out _) != 0)
-        {
-            UpdatePixelSize(unchecked((int)width), unchecked((int)height));
-        }
-    }
-
-    /// <summary>
-    /// Serializes a native child handle for a separately hosted emulator
-    /// process. Metal object pointers are process-local, so macOS falls back
-    /// to a standalone child window until an IPC Metal host is implemented.
-    /// </summary>
-    public bool TryGetChildProcessDescriptor(out string descriptor)
-    {
-        descriptor = string.Empty;
-        if (Kind == VulkanHostSurfaceKind.Metal || WindowHandle == 0)
-        {
-            return false;
-        }
-
-        var kind = Kind == VulkanHostSurfaceKind.Win32 ? "win32" : "xlib";
-        descriptor = string.Create(
-            System.Globalization.CultureInfo.InvariantCulture,
-            $"{kind}:{unchecked((ulong)WindowHandle):X}:{Math.Max(PixelWidth, 1)}:{Math.Max(PixelHeight, 1)}:{unchecked((ulong)DisplayHandle):X}");
-        return true;
-    }
-
-    /// <summary>
-    /// Reconstructs a surface in the isolated emulator process. X11 clients
-    /// must open their own Display connection; Display* values cannot cross a
-    /// process boundary.
-    /// </summary>
-    public static bool TryCreateChildProcessSurface(
-        string descriptor,
-        out VulkanHostSurface? surface,
-        out string? error)
-    {
-        surface = null;
-        error = null;
-        var parts = descriptor.Split(':');
-        if (parts.Length != 5 ||
-            !ulong.TryParse(parts[1], System.Globalization.NumberStyles.AllowHexSpecifier, System.Globalization.CultureInfo.InvariantCulture, out var window) ||
-            !int.TryParse(parts[2], System.Globalization.NumberStyles.Integer, System.Globalization.CultureInfo.InvariantCulture, out var width) ||
-            !int.TryParse(parts[3], System.Globalization.NumberStyles.Integer, System.Globalization.CultureInfo.InvariantCulture, out var height) ||
-            !ulong.TryParse(parts[4], System.Globalization.NumberStyles.AllowHexSpecifier, System.Globalization.CultureInfo.InvariantCulture, out var nativeDisplay))
-        {
-            error = "invalid host-surface descriptor";
-            return false;
-        }
-
-        if (window == 0 || width <= 0 || height <= 0)
-        {
-            error = "host-surface descriptor has an invalid size or handle";
-            return false;
-        }
-
-        if (string.Equals(parts[0], "win32", StringComparison.OrdinalIgnoreCase))
-        {
-            surface = new VulkanHostSurface(
-                VulkanHostSurfaceKind.Win32,
-                unchecked((nint)window),
-                unchecked((nint)nativeDisplay),
-                pollNativeSize: true);
-        }
-        else if (string.Equals(parts[0], "xlib", StringComparison.OrdinalIgnoreCase))
-        {
-            var display = XOpenDisplay(0);
-            if (display == 0)
-            {
-                error = "could not open an X11 display for the host surface";
-                return false;
-            }
-
-            surface = new VulkanHostSurface(
-                VulkanHostSurfaceKind.Xlib,
-                unchecked((nint)window),
-                display,
-                ownsDisplay: true,
-                pollNativeSize: true);
-        }
-        else
-        {
-            error = $"unsupported host-surface kind '{parts[0]}'";
-            return false;
-        }
-
-        surface.UpdatePixelSize(width, height);
-        return true;
-    }
-
-    public void Dispose()
-    {
-        if (_ownsDisplay && DisplayHandle != 0 && OperatingSystem.IsLinux())
-        {
-            _ = XCloseDisplay(DisplayHandle);
-        }
-    }
-
-    [System.Runtime.InteropServices.DllImport("libX11.so.6", EntryPoint = "XOpenDisplay")]
-    private static extern nint XOpenDisplay(nint displayName);
-
-    [System.Runtime.InteropServices.DllImport("libX11.so.6", EntryPoint = "XCloseDisplay")]
-    private static extern int XCloseDisplay(nint display);
-
-    [System.Runtime.InteropServices.DllImport("user32.dll", EntryPoint = "GetClientRect", SetLastError = true)]
-    [return: System.Runtime.InteropServices.MarshalAs(System.Runtime.InteropServices.UnmanagedType.Bool)]
-    private static extern bool GetClientRect(nint window, out Rect rect);
-
-    [System.Runtime.InteropServices.DllImport("libX11.so.6", EntryPoint = "XGetGeometry")]
-    private static extern int XGetGeometry(
-        nint display,
-        nint drawable,
-        out nint root,
-        out int x,
-        out int y,
-        out uint width,
-        out uint height,
-        out uint borderWidth,
-        out uint depth);
-
-    [System.Runtime.InteropServices.StructLayout(System.Runtime.InteropServices.LayoutKind.Sequential)]
-    private struct Rect
-    {
-        public int Left;
-        public int Top;
-        public int Right;
-        public int Bottom;
-    }
-}
-
-/// <summary>
-/// Small public bridge between a desktop UI and the internal Vulkan
-/// presenter. Launchers can host a surface without depending on renderer
-/// submission internals.
-/// </summary>
-public static class VulkanVideoHost
-{
-    /// <summary>
-    /// Raised after the first successful Vulkan present to an embedded host
-    /// surface. UI hosts use this to retire their launch affordance only once
-    /// a real frame can be seen.
-    /// </summary>
-    public static event Action<VulkanHostSurface>? FirstFramePresented
-    {
-        add => VulkanVideoPresenter.FirstHostFramePresented += value;
-        remove => VulkanVideoPresenter.FirstHostFramePresented -= value;
-    }
-
-    public static bool TryAttachSurface(VulkanHostSurface surface) =>
-        VulkanVideoPresenter.TryAttachHostSurface(surface);
-
-    public static void DetachSurface(VulkanHostSurface surface) =>
-        VulkanVideoPresenter.DetachHostSurface(surface);
-
-    public static void RequestClose() => VulkanVideoPresenter.RequestClose();
-
-    public static bool IsEmbedded => VulkanVideoPresenter.UsesHostSurface;
-}

src/SharpEmu.Libs/VideoOut/VulkanPipelineCacheStorage.cs

@@ -0,0 +1,96 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.Libs.VideoOut;
+
+internal static class VulkanPipelineCacheStorage
+{
+    private const string CacheFileName = "vulkan-pipeline-cache.bin";
+
+    internal static string ResolvePath(string? titleId, string? configuredPath)
+    {
+        if (!string.IsNullOrWhiteSpace(configuredPath))
+        {
+            return Path.GetFullPath(
+                Environment.ExpandEnvironmentVariables(configuredPath));
+        }
+
+        return Path.GetFullPath(Path.Combine(
+            AppContext.BaseDirectory,
+            "user",
+            "pipeline_cache",
+            SanitizeTitleId(titleId),
+            CacheFileName));
+    }
+
+    internal static string GetLegacyPath()
+    {
+        var root = OperatingSystem.IsMacOS()
+            ? Path.Combine(
+                Environment.GetFolderPath(Environment.SpecialFolder.UserProfile),
+                "Library",
+                "Caches")
+            : Environment.GetFolderPath(Environment.SpecialFolder.LocalApplicationData);
+        return Path.Combine(root, "SharpEmu", CacheFileName);
+    }
+
+    internal static bool ImportLegacyCache(string legacyPath, string destinationPath)
+    {
+        if (File.Exists(destinationPath) || !File.Exists(legacyPath))
+        {
+            return false;
+        }
+
+        var directory = Path.GetDirectoryName(destinationPath);
+        if (!string.IsNullOrWhiteSpace(directory))
+        {
+            Directory.CreateDirectory(directory);
+        }
+
+        try
+        {
+            File.Copy(legacyPath, destinationPath, overwrite: false);
+            try
+            {
+                File.Delete(legacyPath);
+            }
+            catch (IOException)
+            {
+                // The destination is valid; a locked legacy cache can remain
+                // as an unused artifact without affecting future writes.
+            }
+            catch (UnauthorizedAccessException)
+            {
+                // See above. Cache migration must not block game startup.
+            }
+
+            return true;
+        }
+        catch (IOException) when (File.Exists(destinationPath))
+        {
+            return false;
+        }
+    }
+
+    private static string SanitizeTitleId(string? titleId)
+    {
+        if (string.IsNullOrWhiteSpace(titleId))
+        {
+            return "UNKNOWN";
+        }
+
+        var source = titleId.Trim();
+        Span<char> sanitized = source.Length <= 128
+            ? stackalloc char[source.Length]
+            : new char[source.Length];
+        for (var index = 0; index < source.Length; index++)
+        {
+            var value = source[index];
+            sanitized[index] = char.IsAsciiLetterOrDigit(value) || value is '-' or '_'
+                ? char.ToUpperInvariant(value)
+                : '_';
+        }
+
+        return new string(sanitized);
+    }
+}

src/SharpEmu.ShaderCompiler.Vulkan/SpirvFixedShaders.cs

@@ -92,7 +92,7 @@ public static class SpirvFixedShaders
         return module.Build();
     }
 
-    public static byte[] CreateCopyFragment()
+    public static byte[] CreateCopyFragment(float colorScale = 1f)
     {
         var module = new SpirvModuleBuilder();
         module.AddCapability(SpirvCapability.Shader);
@@ -152,6 +152,16 @@ public static class SpirvFixedShaders
             coordinates,
             2,
             lod);
+        if (colorScale != 1f)
+        {
+            var scale = module.ConstantComposite(
+                vec4Type,
+                module.ConstantFloat(floatType, colorScale),
+                module.ConstantFloat(floatType, colorScale),
+                module.ConstantFloat(floatType, colorScale),
+                module.ConstantFloat(floatType, 1f));
+            color = module.AddInstruction(SpirvOp.FMul, vec4Type, color, scale);
+        }
         module.AddStatement(SpirvOp.Store, output, color);
         module.AddStatement(SpirvOp.Return);
         module.EndFunction();
@@ -165,6 +175,155 @@ public static class SpirvFixedShaders
         return module.Build();
     }
 
+    public static byte[] CreatePqToScRgbFragment()
+    {
+        const float inverseM1 = 16384f / 2610f;
+        const float inverseM2 = 32f / 2523f;
+        const float c1 = 3424f / 4096f;
+        const float c2 = 2413f / 128f;
+        const float c3 = 2392f / 128f;
+
+        var module = new SpirvModuleBuilder();
+        module.AddCapability(SpirvCapability.Shader);
+        var glsl = module.ImportExtInst("GLSL.std.450");
+        var voidType = module.TypeVoid();
+        var floatType = module.TypeFloat(32);
+        var vec2Type = module.TypeVector(floatType, 2);
+        var vec3Type = module.TypeVector(floatType, 3);
+        var vec4Type = module.TypeVector(floatType, 4);
+        var inputVec4Pointer = module.TypePointer(SpirvStorageClass.Input, vec4Type);
+        var outputVec4Pointer = module.TypePointer(SpirvStorageClass.Output, vec4Type);
+        var imageType = module.TypeImage(
+            floatType,
+            SpirvImageDim.Dim2D,
+            depth: false,
+            arrayed: false,
+            multisampled: false,
+            sampled: 1,
+            SpirvImageFormat.Unknown);
+        var sampledImageType = module.TypeSampledImage(imageType);
+        var sampledImagePointer =
+            module.TypePointer(SpirvStorageClass.UniformConstant, sampledImageType);
+        var attribute = module.AddGlobalVariable(inputVec4Pointer, SpirvStorageClass.Input);
+        module.AddDecoration(attribute, SpirvDecoration.Location, 0);
+        var texture = module.AddGlobalVariable(
+            sampledImagePointer,
+            SpirvStorageClass.UniformConstant);
+        module.AddDecoration(texture, SpirvDecoration.DescriptorSet, 0);
+        module.AddDecoration(texture, SpirvDecoration.Binding, 1);
+        var output = module.AddGlobalVariable(outputVec4Pointer, SpirvStorageClass.Output);
+        module.AddDecoration(output, SpirvDecoration.Location, 0);
+
+        uint Float(float value) => module.ConstantFloat(floatType, value);
+        uint Vec3(float value) => module.ConstantComposite(
+            vec3Type,
+            Float(value),
+            Float(value),
+            Float(value));
+        uint Ext(uint operation, uint resultType, params uint[] operands)
+        {
+            var values = new uint[2 + operands.Length];
+            values[0] = glsl;
+            values[1] = operation;
+            operands.CopyTo(values, 2);
+            return module.AddInstruction(SpirvOp.ExtInst, resultType, values);
+        }
+        uint Component(uint vector, uint index) =>
+            module.AddInstruction(SpirvOp.CompositeExtract, floatType, vector, index);
+        uint Multiply(uint left, float right) =>
+            module.AddInstruction(SpirvOp.FMul, floatType, left, Float(right));
+        uint Add3(uint first, uint second, uint third) =>
+            module.AddInstruction(
+                SpirvOp.FAdd,
+                floatType,
+                module.AddInstruction(SpirvOp.FAdd, floatType, first, second),
+                third);
+
+        var functionType = module.TypeFunction(voidType);
+        var main = module.BeginFunction(voidType, functionType);
+        module.AddLabel();
+        var attributeValue = module.AddInstruction(SpirvOp.Load, vec4Type, attribute);
+        var coordinates = module.AddInstruction(
+            SpirvOp.VectorShuffle,
+            vec2Type,
+            attributeValue,
+            attributeValue,
+            0,
+            1);
+        var sampledImage = module.AddInstruction(SpirvOp.Load, sampledImageType, texture);
+        var color = module.AddInstruction(
+            SpirvOp.ImageSampleExplicitLod,
+            vec4Type,
+            sampledImage,
+            coordinates,
+            2,
+            Float(0));
+        var pq = module.AddInstruction(
+            SpirvOp.VectorShuffle,
+            vec3Type,
+            color,
+            color,
+            0,
+            1,
+            2);
+
+        // SMPTE ST 2084 converts normalized PQ code values to absolute luminance.
+        var powered = Ext(26, vec3Type, pq, Vec3(inverseM2));
+        var numerator = Ext(
+            40,
+            vec3Type,
+            module.AddInstruction(SpirvOp.FSub, vec3Type, powered, Vec3(c1)),
+            Vec3(0));
+        var denominator = module.AddInstruction(
+            SpirvOp.FSub,
+            vec3Type,
+            Vec3(c2),
+            module.AddInstruction(SpirvOp.FMul, vec3Type, Vec3(c3), powered));
+        var normalizedLuminance = Ext(
+            26,
+            vec3Type,
+            module.AddInstruction(SpirvOp.FDiv, vec3Type, numerator, denominator),
+            Vec3(inverseM1));
+        var scRgb2020 = module.AddInstruction(
+            SpirvOp.FMul,
+            vec3Type,
+            normalizedLuminance,
+            Vec3(10000f / 80f));
+
+        var red2020 = Component(scRgb2020, 0);
+        var green2020 = Component(scRgb2020, 1);
+        var blue2020 = Component(scRgb2020, 2);
+        var red = Add3(
+            Multiply(red2020, 1.660491f),
+            Multiply(green2020, -0.587641f),
+            Multiply(blue2020, -0.072850f));
+        var green = Add3(
+            Multiply(red2020, -0.124550f),
+            Multiply(green2020, 1.132900f),
+            Multiply(blue2020, -0.008349f));
+        var blue = Add3(
+            Multiply(red2020, -0.018151f),
+            Multiply(green2020, -0.100579f),
+            Multiply(blue2020, 1.118730f));
+        var converted = module.AddInstruction(
+            SpirvOp.CompositeConstruct,
+            vec4Type,
+            red,
+            green,
+            blue,
+            Component(color, 3));
+        module.AddStatement(SpirvOp.Store, output, converted);
+        module.AddStatement(SpirvOp.Return);
+        module.EndFunction();
+        module.AddEntryPoint(
+            SpirvExecutionModel.Fragment,
+            main,
+            "main",
+            [attribute, texture, output]);
+        module.AddExecutionMode(main, SpirvExecutionMode.OriginUpperLeft);
+        return module.Build();
+    }
+
     public static byte[] CreateSolidFragment(float red, float green, float blue, float alpha)
     {
         var module = new SpirvModuleBuilder();

tests/SharpEmu.Libs.Tests/VideoOut/HostVideoOptionsTests.cs

@@ -0,0 +1,29 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.Libs.VideoOut;
+using Xunit;
+
+namespace SharpEmu.Libs.Tests.VideoOut;
+
+public sealed class HostVideoOptionsTests
+{
+    [Fact]
+    public void NormalizeClampsUnsafeHostValues()
+    {
+        var options = new HostVideoOptions
+        {
+            Width = 1,
+            Height = 99_999,
+            DisplayIndex = -2,
+            RefreshRate = 5000,
+            HdrMode = (HostHdrMode)99,
+        }.Normalize();
+
+        Assert.Equal(640, options.Width);
+        Assert.Equal(16384, options.Height);
+        Assert.Equal(0, options.DisplayIndex);
+        Assert.Equal(1000, options.RefreshRate);
+        Assert.Equal(HostHdrMode.Auto, options.HdrMode);
+    }
+}

tests/SharpEmu.Libs.Tests/VideoOut/VideoOutPixelFormatTests.cs

@@ -97,6 +97,24 @@ public sealed class VideoOutPixelFormatTests
         Assert.Equal(56u, InvokeMapPixelFormat(0xDEADBEEFUL));
     }
 
+    [Theory]
+    [InlineData(0x88740000UL)]
+    [InlineData(0x8100070422000000UL)]
+    [InlineData(0x8100070400000000UL)]
+    public void IsHdrPixelFormat_PqFormats_ReturnTrue(ulong pixelFormat)
+    {
+        Assert.True(VideoOutExports.IsHdrPixelFormat(pixelFormat));
+    }
+
+    [Theory]
+    [InlineData(0x80000000UL)]
+    [InlineData(0x88060000UL)]
+    [InlineData(0x8100000622000000UL)]
+    public void IsHdrPixelFormat_SdrFormats_ReturnFalse(ulong pixelFormat)
+    {
+        Assert.False(VideoOutExports.IsHdrPixelFormat(pixelFormat));
+    }
+
     // ---- Self-check activation ----
 
     [Fact]

tests/SharpEmu.Libs.Tests/VideoOut/VulkanPipelineCacheStorageTests.cs

@@ -0,0 +1,74 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.Libs.VideoOut;
+using Xunit;
+
+namespace SharpEmu.Libs.Tests.VideoOut;
+
+public sealed class VulkanPipelineCacheStorageTests
+{
+    [Fact]
+    public void ResolvePathUsesExecutableUserDirectoryAndTitleId()
+    {
+        var path = VulkanPipelineCacheStorage.ResolvePath("PPSA02929", configuredPath: null);
+
+        Assert.Equal(
+            Path.GetFullPath(Path.Combine(
+                AppContext.BaseDirectory,
+                "user",
+                "pipeline_cache",
+                "PPSA02929",
+                "vulkan-pipeline-cache.bin")),
+            path);
+    }
+
+    [Fact]
+    public void ResolvePathSanitizesTitleId()
+    {
+        var path = VulkanPipelineCacheStorage.ResolvePath("ppsa/02:929", configuredPath: null);
+
+        Assert.EndsWith(
+            Path.Combine("PPSA_02_929", "vulkan-pipeline-cache.bin"),
+            path,
+            StringComparison.Ordinal);
+    }
+
+    [Fact]
+    public void ResolvePathPreservesConfiguredOverride()
+    {
+        var configured = Path.Combine(Path.GetTempPath(), "SharpEmuTests", "custom-cache.bin");
+
+        Assert.Equal(
+            Path.GetFullPath(configured),
+            VulkanPipelineCacheStorage.ResolvePath("PPSA02929", configured));
+    }
+
+    [Fact]
+    public void ImportLegacyCacheDoesNotOverwriteExistingTitleCache()
+    {
+        var root = Path.Combine(Path.GetTempPath(), "SharpEmuTests", Guid.NewGuid().ToString("N"));
+        var legacyPath = Path.Combine(root, "legacy.bin");
+        var destinationPath = Path.Combine(root, "user", "pipeline_cache", "PPSA02929", "cache.bin");
+        try
+        {
+            Directory.CreateDirectory(root);
+            File.WriteAllBytes(legacyPath, [1, 2, 3]);
+
+            Assert.True(VulkanPipelineCacheStorage.ImportLegacyCache(legacyPath, destinationPath));
+            Assert.Equal(new byte[] { 1, 2, 3 }, File.ReadAllBytes(destinationPath));
+            Assert.False(File.Exists(legacyPath));
+
+            File.WriteAllBytes(legacyPath, [4, 5, 6]);
+            Assert.False(VulkanPipelineCacheStorage.ImportLegacyCache(legacyPath, destinationPath));
+            Assert.Equal(new byte[] { 1, 2, 3 }, File.ReadAllBytes(destinationPath));
+        }
+        finally
+        {
+            if (Directory.Exists(root))
+            {
+                Directory.Delete(root, recursive: true);
+            }
+        }
+    }
+}