Start building more accurate vertex as compute usage info

- Fixes

Directory.Packages.props

@@ -8,14 +8,14 @@
     <PackageVersion Include="Avalonia.Desktop" Version="12.1.1" />
     <PackageVersion Include="AvaloniaUI.DiagnosticsSupport" Version="2.2.3" />
     <PackageVersion Include="Avalonia.Markup.Xaml.Loader" Version="12.1.1" />
-    <PackageVersion Include="Svg.Controls.Avalonia" Version="12.0.0.13" />
+    <PackageVersion Include="Svg.Controls.Avalonia" Version="12.0.0.15" />
-    <PackageVersion Include="Svg.Controls.Skia.Avalonia" Version="12.0.0.13" />
+    <PackageVersion Include="Svg.Controls.Skia.Avalonia" Version="12.0.0.15" />
     <PackageVersion Include="CommandLineParser" Version="2.9.1" />
     <PackageVersion Include="CommunityToolkit.Mvvm" Version="8.4.2" />
     <PackageVersion Include="Concentus" Version="2.2.2" />
     <PackageVersion Include="DiscordRichPresence" Version="1.6.1.70" />
     <PackageVersion Include="DynamicData" Version="9.4.33" />
-    <PackageVersion Include="FluentAvaloniaUI" Version="3.0.2" />
+    <PackageVersion Include="FluentAvaloniaUI" Version="3.1.0" />
     <PackageVersion Include="Gommon" Version="2.8.1.2" />
     <PackageVersion Include="Humanizer" Version="3.0.10" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
@@ -48,7 +48,7 @@
     <PackageVersion Include="SkiaSharp" Version="3.119.4" />
     <PackageVersion Include="SkiaSharp.NativeAssets.Linux" Version="3.119.4" />
     <PackageVersion Include="SPB" Version="0.0.4-build32" />
-    <PackageVersion Include="System.IO.Hashing" Version="10.0.10" />
+    <PackageVersion Include="System.IO.Hashing" Version="10.0.11" />
     <PackageVersion Include="UnicornEngine.Unicorn" Version="2.1.0" />
   </ItemGroup>
 </Project>

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -32,14 +32,11 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 7020; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 7031; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";
 
-        private const string TitleIdTextDefault = "0000000000000000";
-        private const string DisplayVersionDefault = "0";
-
         public static readonly Symbol PageTableSymbol = new(SymbolType.Special, 1);
         public static readonly Symbol CountTableSymbol = new(SymbolType.Special, 2);
         public static readonly Symbol DispatchStubSymbol = new(SymbolType.Special, 3);
@@ -67,8 +64,7 @@ namespace ARMeilleure.Translation.PTC
 
         private bool _disposed;
 
-        public string TitleIdText { get; private set; }
+        public PtcCacheInfo CacheInfo { get; private set; }
-        public string DisplayVersion { get; private set; }
 
         private MemoryManagerType _memoryMode;
 
@@ -97,8 +93,7 @@ namespace ARMeilleure.Translation.PTC
 
             _disposed = false;
 
-            TitleIdText = TitleIdTextDefault;
+            CacheInfo = new PtcCacheInfo(0, null, null, 0, null, "Unknown", "default");
-            DisplayVersion = DisplayVersionDefault;
 
             CachePathActual = string.Empty;
             CachePathBackup = string.Empty;
@@ -106,20 +101,24 @@ namespace ARMeilleure.Translation.PTC
             Disable();
         }
 
-        public void Initialize(string titleIdText, string displayVersion, bool enabled, MemoryManagerType memoryMode, string cacheSelector)
+        public void Initialize(PtcCacheInfo cacheInfo, bool enabled, MemoryManagerType memoryMode)
         {
             Wait();
 
             Profiler.Wait();
             Profiler.ClearEntries();
 
-            Logger.Info?.Print(LogClass.Ptc, $"Initializing Profiled Persistent Translation Cache v{InternalVersion}\n\t\t (title: {titleIdText}, version: '{displayVersion}', selector: '{cacheSelector}', enabled: {enabled}).");
+            CacheInfo = cacheInfo;
 
-            if (!enabled || string.IsNullOrEmpty(titleIdText) || titleIdText == TitleIdTextDefault)
+            Logger.Info?.Print(
+                LogClass.Ptc,
+                $"Initializing Profiled Persistent Translation Cache v{InternalVersion}\n\t\t " +
+                $"(pid: {cacheInfo.ProcessId}, title: {cacheInfo.TitleIdText}, application: {cacheInfo.ApplicationIdText}, " +
+                $"programIndex: {cacheInfo.ProgramIndex}, version: '{cacheInfo.DisplayVersion}', kind: {cacheInfo.ProcessKind}, " +
+                $"selector: '{cacheInfo.CacheSelector}', key: '{cacheInfo.CacheKey}', enabled: {enabled}).");
+
+            if (!enabled || cacheInfo.TitleIdText == PtcCacheInfo.TitleIdTextDefault)
             {
-                TitleIdText = TitleIdTextDefault;
-                DisplayVersion = DisplayVersionDefault;
-
                 CachePathActual = string.Empty;
                 CachePathBackup = string.Empty;
 
@@ -128,12 +127,10 @@ namespace ARMeilleure.Translation.PTC
                 return;
             }
 
-            TitleIdText = titleIdText;
-            DisplayVersion = !string.IsNullOrEmpty(displayVersion) ? displayVersion : DisplayVersionDefault;
             _memoryMode = memoryMode;
 
-            string workPathActual = Path.Combine(AppDataManager.GamesDirPath, TitleIdText, "cache", "cpu", ActualDir);
+            string workPathActual = Path.Combine(AppDataManager.GamesDirPath, CacheInfo.TitleIdText, "cache", "cpu", ActualDir);
-            string workPathBackup = Path.Combine(AppDataManager.GamesDirPath, TitleIdText, "cache", "cpu", BackupDir);
+            string workPathBackup = Path.Combine(AppDataManager.GamesDirPath, CacheInfo.TitleIdText, "cache", "cpu", BackupDir);
 
             if (!Directory.Exists(workPathActual))
             {
@@ -145,8 +142,14 @@ namespace ARMeilleure.Translation.PTC
                 Directory.CreateDirectory(workPathBackup);
             }
 
-            CachePathActual = Path.Combine(workPathActual, DisplayVersion) + "-" + cacheSelector;
+            CachePathActual = Path.Combine(workPathActual, CacheInfo.DisplayVersion) + "-" + CacheInfo.CacheSelector;
-            CachePathBackup = Path.Combine(workPathBackup, DisplayVersion) + "-" + cacheSelector;
+            CachePathBackup = Path.Combine(workPathBackup, CacheInfo.DisplayVersion) + "-" + CacheInfo.CacheSelector;
+
+            Logger.Info?.Print(
+                LogClass.Ptc,
+                $"PPTC cache owner selected (pid: {CacheInfo.ProcessId}, title: {CacheInfo.TitleIdText}, application: {CacheInfo.ApplicationIdText}, " +
+                $"version: '{CacheInfo.DisplayVersion}', kind: {CacheInfo.ProcessKind}, selector: '{CacheInfo.CacheSelector}', " +
+                $"key: '{CacheInfo.CacheKey}', path: '{CachePathActual}').");
 
             PreLoad();
             Profiler.PreLoad();
@@ -370,7 +373,12 @@ namespace ARMeilleure.Translation.PTC
 
             long fileSize = new FileInfo(fileName).Length;
 
-            Logger.Info?.Print(LogClass.Ptc, $"{(isBackup ? "Loaded Backup Translation Cache" : "Loaded Translation Cache")} (size: {fileSize} bytes, translated functions: {GetEntriesCount()}).");
+            Logger.Info?.Print(
+                LogClass.Ptc,
+                $"{(isBackup ? "Loaded Backup Translation Cache" : "Loaded Translation Cache")} " +
+                $"(pid: {CacheInfo.ProcessId}, title: {CacheInfo.TitleIdText}, version: '{CacheInfo.DisplayVersion}', kind: {CacheInfo.ProcessKind}, " +
+                $"selector: '{CacheInfo.CacheSelector}', key: '{CacheInfo.CacheKey}', path: '{fileName}', " +
+                $"size: {fileSize} bytes, translated functions: {GetEntriesCount()}).");
 
             return true;
         }

src/ARMeilleure/Translation/PTC/PtcCacheInfo.cs

@@ -0,0 +1,37 @@
+namespace ARMeilleure.Translation.PTC
+{
+    public readonly struct PtcCacheInfo
+    {
+        public const string TitleIdTextDefault = "0000000000000000";
+        public const string ApplicationIdTextDefault = "0000000000000000";
+        public const string DisplayVersionDefault = "0";
+
+        public ulong ProcessId { get; }
+        public string TitleIdText { get; }
+        public string ApplicationIdText { get; }
+        public byte ProgramIndex { get; }
+        public string DisplayVersion { get; }
+        public string ProcessKind { get; }
+        public string CacheSelector { get; }
+
+        public string CacheKey => $"{DisplayVersion}-{CacheSelector}";
+
+        public PtcCacheInfo(
+            ulong processId,
+            string titleIdText,
+            string applicationIdText,
+            byte programIndex,
+            string displayVersion,
+            string processKind,
+            string cacheSelector)
+        {
+            ProcessId = processId;
+            TitleIdText = !string.IsNullOrEmpty(titleIdText) ? titleIdText : TitleIdTextDefault;
+            ApplicationIdText = !string.IsNullOrEmpty(applicationIdText) ? applicationIdText : ApplicationIdTextDefault;
+            ProgramIndex = programIndex;
+            DisplayVersion = !string.IsNullOrEmpty(displayVersion) ? displayVersion : DisplayVersionDefault;
+            ProcessKind = processKind ?? string.Empty;
+            CacheSelector = string.IsNullOrEmpty(cacheSelector) ? "default" : cacheSelector;
+        }
+    }
+}

src/ARMeilleure/Translation/PTC/PtcProfiler.cs

@@ -23,7 +23,7 @@ namespace ARMeilleure.Translation.PTC
     {
         private const string OuterHeaderMagicString = "Pohd\0\0\0\0";
 
-        private const uint InternalVersion = 6698; //! Not to be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 7031; //! Not to be incremented manually for each change to the ARMeilleure project.
 
         private static readonly uint[] _migrateInternalVersions = 
         [
@@ -254,7 +254,12 @@ namespace ARMeilleure.Translation.PTC
 
             long fileSize = new FileInfo(fileName).Length;
 
-            Logger.Info?.Print(LogClass.Ptc, $"{(isBackup ? "Loaded Backup Profiling Info" : "Loaded Profiling Info")} (size: {fileSize} bytes, profiled functions: {ProfiledFuncs.Count}).");
+            Logger.Info?.Print(
+                LogClass.Ptc,
+                $"{(isBackup ? "Loaded Backup Profiling Info" : "Loaded Profiling Info")} " +
+                $"(pid: {_ptc.CacheInfo.ProcessId}, title: {_ptc.CacheInfo.TitleIdText}, version: '{_ptc.CacheInfo.DisplayVersion}', " +
+                $"kind: {_ptc.CacheInfo.ProcessKind}, selector: '{_ptc.CacheInfo.CacheSelector}', key: '{_ptc.CacheInfo.CacheKey}', " +
+                $"path: '{fileName}', size: {fileSize} bytes, profiled functions: {ProfiledFuncs.Count}).");
 
             return true;
         }
@@ -375,7 +380,11 @@ namespace ARMeilleure.Translation.PTC
 
             if (fileSize != 0L)
             {
-                Logger.Info?.Print(LogClass.Ptc, $"Saved Profiling Info (size: {fileSize} bytes, profiled functions: {profiledFuncsCount}).");
+                Logger.Info?.Print(
+                    LogClass.Ptc,
+                    $"Saved Profiling Info (pid: {_ptc.CacheInfo.ProcessId}, title: {_ptc.CacheInfo.TitleIdText}, version: '{_ptc.CacheInfo.DisplayVersion}', " +
+                    $"kind: {_ptc.CacheInfo.ProcessKind}, selector: '{_ptc.CacheInfo.CacheSelector}', key: '{_ptc.CacheInfo.CacheKey}', " +
+                    $"path: '{fileName}', size: {fileSize} bytes, profiled functions: {profiledFuncsCount}).");
             }
         }
 

src/ARMeilleure/Translation/Translator.cs

@@ -61,9 +61,9 @@ namespace ARMeilleure.Translation
             FunctionTable.Fill = (ulong)Stubs.SlowDispatchStub;
         }
 
-        public IPtcLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector)
+        public IPtcLoadState LoadDiskCache(PtcCacheInfo cacheInfo, bool enabled)
         {
-            _ptc.Initialize(titleIdText, displayVersion, enabled, Memory.Type, cacheSelector);
+            _ptc.Initialize(cacheInfo, enabled, Memory.Type);
             return _ptc;
         }
 
@@ -413,6 +413,10 @@ namespace ARMeilleure.Translation
 
                         context.CurrOp = opCode;
 
+                        Operand nativeContext = context.LoadArgument(OperandType.I64, 0);
+                        Operand dispAddressAddr = context.Add(nativeContext, Const((ulong)NativeContext.GetDispatchAddressOffset()));
+                        context.Store(dispAddressAddr, Const(opCode.Address));
+
                         bool isLastOp = opcIndex == block.OpCodes.Count - 1;
 
                         if (isLastOp)

src/Ryujinx.Cpu/AppleHv/HvCpuContext.cs

@@ -1,4 +1,5 @@
 using ARMeilleure.Memory;
+using ARMeilleure.Translation.PTC;
 using System.Runtime.Versioning;
 
 namespace Ryujinx.Cpu.AppleHv
@@ -32,7 +33,7 @@ namespace Ryujinx.Cpu.AppleHv
         {
         }
 
-        public IDiskCacheLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector)
+        public IDiskCacheLoadState LoadDiskCache(PtcCacheInfo cacheInfo, bool enabled)
         {
             return new DummyDiskCacheLoadState();
         }

src/Ryujinx.Cpu/ICpuContext.cs

@@ -1,4 +1,5 @@
 using System;
+using ARMeilleure.Translation.PTC;
 
 namespace Ryujinx.Cpu
 {
@@ -44,11 +45,10 @@ namespace Ryujinx.Cpu
         /// <remarks>
         /// If the execution engine is recompiling guest code, this can be used to load cached code from disk.
         /// </remarks>
-        /// <param name="titleIdText">Title ID of the application in padded hex form</param>
+        /// <param name="cacheInfo">Identity and selector for the process-owned disk cache</param>
-        /// <param name="displayVersion">Version of the application</param>
         /// <param name="enabled">True if the cache should be loaded from disk if it exists, false otherwise</param>
         /// <returns>Disk cache load progress reporter and manager</returns>
-        IDiskCacheLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector);
+        IDiskCacheLoadState LoadDiskCache(PtcCacheInfo cacheInfo, bool enabled);
 
         /// <summary>
         /// Indicates that code has been loaded into guest memory, and that it might be executed in the future.

src/Ryujinx.Cpu/Jit/JitCpuContext.cs

@@ -1,6 +1,7 @@
 using ARMeilleure.Common;
 using ARMeilleure.Memory;
 using ARMeilleure.Translation;
+using ARMeilleure.Translation.PTC;
 using Ryujinx.Cpu.Signal;
 
 namespace Ryujinx.Cpu.Jit
@@ -51,9 +52,9 @@ namespace Ryujinx.Cpu.Jit
         }
 
         /// <inheritdoc/>
-        public IDiskCacheLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector)
+        public IDiskCacheLoadState LoadDiskCache(PtcCacheInfo cacheInfo, bool enabled)
         {
-            return new JitDiskCacheLoadState(_translator.LoadDiskCache(titleIdText, displayVersion, enabled, cacheSelector));
+            return new JitDiskCacheLoadState(_translator.LoadDiskCache(cacheInfo, enabled));
         }
 
         /// <inheritdoc/>

src/Ryujinx.Cpu/LightningJit/Arm32/Decoder.cs

@@ -8,7 +8,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32
 {
     static class Decoder<T> where T : IInstEmit
     {
-        public static MultiBlock DecodeMulti(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address, bool isThumb)
+        public static MultiBlock DecodeMulti(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address, bool singleBlock, bool isThumb)
         {
             List<Block> blocks = [];
             List<ulong> branchTargets = [];
@@ -24,7 +24,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32
 
                 blocks.Add(block);
 
-                if (block.IsTruncated || !HasNextBlock(block, block.EndAddress - 4UL, branchTargets))
+                if (block.IsTruncated || (singleBlock && block.EndsWithBranch) || !HasNextBlock(block, block.EndAddress - 4UL, branchTargets))
                 {
                     break;
                 }

src/Ryujinx.Cpu/LightningJit/Arm32/Target/Arm64/Compiler.cs

@@ -227,7 +227,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
 
         public static CompiledFunction Compile(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address, IAddressTable<ulong> funcTable, nint dispatchStubPtr, bool isThumb)
         {
-            MultiBlock multiBlock = Decoder<InstEmit>.DecodeMulti(cpuPreset, memoryManager, address, isThumb);
+            MultiBlock multiBlock = Decoder<InstEmit>.DecodeMulti(cpuPreset, memoryManager, address, singleBlock: true, isThumb);
 
             Dictionary<ulong, int> targets = new();
 

src/Ryujinx.Cpu/LightningJit/Arm64/Target/Arm64/Compiler.cs

@@ -305,7 +305,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm64.Target.Arm64
 
         public static CompiledFunction Compile(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address, IAddressTable<ulong> funcTable, nint dispatchStubPtr)
         {
-            MultiBlock multiBlock = Decoder.DecodeMulti(cpuPreset, memoryManager, address);
+            MultiBlock multiBlock = Decoder.DecodeMulti(cpuPreset, memoryManager, address, singleBlock: true);
 
             Dictionary<ulong, int> targets = new();
             List<PendingBranch> pendingBranches = [];

src/Ryujinx.Cpu/LightningJit/Arm64/Target/Arm64/Decoder.cs

@@ -13,7 +13,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm64.Target.Arm64
         private const uint NzcvFlags = 0xfu << 28;
         private const uint CFlag = 0x1u << 29;
 
-        public static MultiBlock DecodeMulti(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address)
+        public static MultiBlock DecodeMulti(CpuPreset cpuPreset, IMemoryManager memoryManager, ulong address, bool singleBlock)
         {
             List<Block> blocks = [];
             List<ulong> branchTargets = [];
@@ -35,7 +35,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm64.Target.Arm64
 
                 blocks.Add(block);
 
-                if (block.IsTruncated || !HasNextBlock(block, block.EndAddress - 4UL, branchTargets))
+                if (block.IsTruncated || (singleBlock && block.EndsWithBranch) || !HasNextBlock(block, block.EndAddress - 4UL, branchTargets))
                 {
                     break;
                 }

src/Ryujinx.Cpu/LightningJit/LightningJitCpuContext.cs

@@ -1,5 +1,6 @@
 using ARMeilleure.Common;
 using ARMeilleure.Memory;
+using ARMeilleure.Translation.PTC;
 using Ryujinx.Cpu.Jit;
 using Ryujinx.Cpu.LightningJit.State;
 
@@ -46,7 +47,7 @@ namespace Ryujinx.Cpu.LightningJit
         }
 
         /// <inheritdoc/>
-        public IDiskCacheLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector)
+        public IDiskCacheLoadState LoadDiskCache(PtcCacheInfo cacheInfo, bool enabled)
         {
             return new DummyDiskCacheLoadState();
         }

src/Ryujinx.Graphics.GAL/Capabilities.cs

@@ -32,6 +32,7 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool SupportsGeometryShader;
         public readonly bool SupportsGeometryShaderPassthrough;
         public readonly bool SupportsTransformFeedback;
+        public readonly bool SupportsImageBufferPixelAlignment;
         public readonly bool SupportsImageLoadFormatted;
         public readonly bool SupportsLayerVertexTessellation;
         public readonly bool SupportsMismatchingViewFormat;
@@ -43,6 +44,7 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool SupportsShaderBarrierDivergence;
         public readonly bool SupportsShaderFloat64;
         public readonly bool SupportsShaderNonUniformIndexing;
+        public readonly bool SupportsTextureBufferPixelAlignment;
         public readonly bool SupportsTextureGatherOffsets;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsVertexStoreAndAtomics;
@@ -101,6 +103,7 @@ namespace Ryujinx.Graphics.GAL
             bool supportsGeometryShader,
             bool supportsGeometryShaderPassthrough,
             bool supportsTransformFeedback,
+            bool supportsImageBufferPixelAlignment,
             bool supportsImageLoadFormatted,
             bool supportsLayerVertexTessellation,
             bool supportsMismatchingViewFormat,
@@ -112,6 +115,7 @@ namespace Ryujinx.Graphics.GAL
             bool supportsShaderBarrierDivergence,
             bool supportsShaderFloat64,
             bool supportsShaderNonUniformIndexing,
+            bool supportsTextureBufferPixelAlignment,
             bool supportsTextureGatherOffsets,
             bool supportsTextureShadowLod,
             bool supportsVertexStoreAndAtomics,
@@ -164,6 +168,7 @@ namespace Ryujinx.Graphics.GAL
             SupportsGeometryShader = supportsGeometryShader;
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
             SupportsTransformFeedback = supportsTransformFeedback;
+            SupportsImageBufferPixelAlignment = supportsImageBufferPixelAlignment;
             SupportsImageLoadFormatted = supportsImageLoadFormatted;
             SupportsLayerVertexTessellation = supportsLayerVertexTessellation;
             SupportsMismatchingViewFormat = supportsMismatchingViewFormat;
@@ -175,6 +180,7 @@ namespace Ryujinx.Graphics.GAL
             SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
             SupportsShaderFloat64 = supportsShaderFloat64;
             SupportsShaderNonUniformIndexing = supportsShaderNonUniformIndexing;
+            SupportsTextureBufferPixelAlignment = supportsTextureBufferPixelAlignment;
             SupportsTextureGatherOffsets = supportsTextureGatherOffsets;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsVertexStoreAndAtomics = supportsVertexStoreAndAtomics;

src/Ryujinx.Graphics.GAL/Target.cs

@@ -16,19 +16,18 @@ namespace Ryujinx.Graphics.GAL
 
     public static class TargetExtensions
     {
-        public static bool IsMultisample(this Target target)
+        extension(Target target)
         {
-            return target is Target.Texture2DMultisample or Target.Texture2DMultisampleArray;
+            public bool IsMultisample => target is Target.Texture2DMultisample or Target.Texture2DMultisampleArray;
-        }
 
-        public static bool HasDepthOrLayers(this Target target)
+            public bool HasDepthOrLayers =>
-        {
+                target is 
-            return target is Target.Texture3D
+                    Target.Texture3D or
-                or Target.Texture1DArray
+                    Target.Texture1DArray or
-                or Target.Texture2DArray
+                    Target.Texture2DArray or
-                or Target.Texture2DMultisampleArray
+                    Target.Texture2DMultisampleArray or
-                or Target.Cubemap
+                    Target.Cubemap or
-                or Target.CubemapArray;
+                    Target.CubemapArray;
         }
     }
 }

src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs

@@ -466,6 +466,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             engine.UpdateState(ulong.MaxValue & ~(1UL << StateUpdater.ShaderStateIndex));
 
+            _channel.TextureManager.SignalRenderTargetsModifiable();
             _channel.TextureManager.UpdateRenderTargets();
 
             int textureId = _state.State.DrawTextureTextureId;
@@ -803,7 +804,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             int index = (argument >> 6) & 0xf;
             int layer = (argument >> 10) & 0x3ff;
 
-            RenderTargetUpdateFlags updateFlags = RenderTargetUpdateFlags.SingleColor;
+            RenderTargetUpdateFlags updateFlags = RenderTargetUpdateFlags.SingleColor | RenderTargetUpdateFlags.ForClear;
 
             if (layer != 0 || layerCount > 1)
             {

src/Ryujinx.Graphics.Gpu/Engine/Threed/RenderTargetUpdateFlags.cs

@@ -38,6 +38,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         DiscardClip = 1 << 4,
 
+        /// <summary>
+        /// Indicates that the render target will be used for a clear operation.
+        /// </summary>
+        ForClear = 1 << 5,
+
         /// <summary>
         /// Default update flags for draw.
         /// </summary>

src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs

@@ -45,6 +45,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         private ProgramPipelineState _pipeline;
 
         private bool _fsReadsFragCoord;
+        private bool _fsAlwaysDiscards;
         private bool _vsUsesDrawParameters;
         private bool _vtgWritesRtLayer;
         private byte _vsClipDistancesWritten;
@@ -491,6 +492,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             Span<RtColorState> rtColorStateSpan = _state.State.RtColorState.AsSpan();
 
+            bool rtModifiable = updateFlags.HasFlag(RenderTargetUpdateFlags.ForClear) || !_fsAlwaysDiscards;
+
             for (int index = 0; index < Constants.TotalRenderTargets; index++)
             {
                 int rtIndex = useControl ? rtControl.UnpackPermutationIndex(index) : index;
@@ -499,7 +502,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
                 if (index >= count || !IsRtEnabled(colorState) || (singleColor && index != singleUse))
                 {
-                    changedScale |= _channel.TextureManager.SetRenderTargetColor(index, null);
+                    changedScale |= _channel.TextureManager.SetRenderTargetColor(index, null, rtModifiable);
 
                     continue;
                 }
@@ -518,7 +521,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                     samplesInY,
                     sizeHint);
 
-                changedScale |= _channel.TextureManager.SetRenderTargetColor(index, color);
+                changedScale |= _channel.TextureManager.SetRenderTargetColor(index, color, rtModifiable);
 
                 if (color != null)
                 {
@@ -572,7 +575,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 }
             }
 
-            changedScale |= _channel.TextureManager.SetRenderTargetDepthStencil(depthStencil);
+            changedScale |= _channel.TextureManager.SetRenderTargetDepthStencil(depthStencil, rtModifiable);
 
             if (changedScale)
             {
@@ -774,6 +777,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 float width = scaleX * 2;
                 float height = scaleY * 2;
 
+                if (yNegate)
+                {
+                    y += _state.State.ScreenScissorState.Height - MathF.Abs(height);
+                }
+
                 float scale = _channel.TextureManager.RenderTargetScale;
                 if (scale != 1f)
                 {
@@ -1517,20 +1525,38 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _currentProgramInfo[stageIndex] = info;
             }
 
-            if (gs.Shaders[5]?.Info.UsesFragCoord == true)
+            _fsReadsFragCoord = false;
-            {
-                // Make sure we update the viewport size on the support buffer if it will be consumed on the new shader.
 
-                if (!_fsReadsFragCoord && (_state.State.YControl & YControl.NegateY) != 0)
+            ShaderProgramInfo fragmentShaderInfo = gs.Shaders[5]?.Info;
+
+            if (fragmentShaderInfo != null)
+            {
+                if (fragmentShaderInfo.UsesFragCoord)
                 {
-                    UpdateSupportBufferViewportSize();
+                    // Make sure we update the viewport size on the support buffer if it will be consumed on the new shader.
+
+                    if (!_fsReadsFragCoord && (_state.State.YControl & YControl.NegateY) != 0)
+                    {
+                        UpdateSupportBufferViewportSize();
+                    }
+
+                    _fsReadsFragCoord = true;
                 }
 
-                _fsReadsFragCoord = true;
+                if (_fsAlwaysDiscards != fragmentShaderInfo.HasUnconditionalDiscard)
+                {
+                    _fsAlwaysDiscards = fragmentShaderInfo.HasUnconditionalDiscard;
+
+                    if (!_fsAlwaysDiscards)
+                    {
+                        _channel.TextureManager.RefreshModifiedTextures();
+                        _channel.TextureManager.SignalRenderTargetsModifiable();
+                    }
+                }
             }
             else
             {
-                _fsReadsFragCoord = false;
+                _fsAlwaysDiscards = false;
             }
 
             if (gs.VertexAsCompute != null)

src/Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -1117,7 +1117,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <returns>True if data was flushed, false otherwise</returns>
         public bool FlushModified(bool tracked = true)
         {
-            return TextureCompatibility.CanTextureFlush(this, _context.Capabilities) && Group.FlushModified(this, tracked);
+            return TextureCompatibility.CanTextureFlush(Info, _context.Capabilities) && Group.FlushModified(this, tracked);
         }
 
         /// <summary>
@@ -1131,7 +1131,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="tracked">Whether or not the flush triggers write tracking. If it doesn't, the texture will not be blacklisted for scaling either.</param>
         public void Flush(bool tracked)
         {
-            if (TextureCompatibility.CanTextureFlush(this, _context.Capabilities))
+            if (TextureCompatibility.CanTextureFlush(Info, _context.Capabilities))
             {
                 FlushTextureDataToGuest(tracked);
             }
@@ -1336,7 +1336,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             {
                 result = TextureCompatibility.PropagateViewCompatibility(result, TextureCompatibility.ViewTargetCompatible(Info, info, ref caps));
 
-                bool bothMs = Info.Target.IsMultisample() && info.Target.IsMultisample();
+                bool bothMs = Info.Target.IsMultisample && info.Target.IsMultisample;
                 if (bothMs && (Info.SamplesInX != info.SamplesInX || Info.SamplesInY != info.SamplesInY))
                 {
                     result = TextureViewCompatibility.Incompatible;
@@ -1572,11 +1572,6 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="bound">True if the texture has been bound, false if it has been unbound</param>
         public void SignalModifying(bool bound)
         {
-            if (bound)
-            {
-                _scaledSetScore = Math.Max(0, _scaledSetScore - 1);
-            }
-
             if (_modifiedStale || Group.HasCopyDependencies || Group.HasFlushBuffer)
             {
                 _modifiedStale = false;
@@ -1584,9 +1579,18 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
 
             _physicalMemory.TextureCache.Lift(this);
+        }
 
+        /// <summary>
+        /// Signals that a render target texture has been either bound or unbound.
+        /// </summary>
+        /// <param name="bound">True if the texture has been bound, false if it has been unbound</param>
+        public void SignalBindingChange(bool bound)
+        {
             if (bound)
             {
+                _scaledSetScore = Math.Max(0, _scaledSetScore - 1);
+
                 IncrementReferenceCount();
             }
             else

src/Ryujinx.Graphics.Gpu/Image/TextureBindingsManager.cs

@@ -4,6 +4,7 @@ using Ryujinx.Graphics.Gpu.Engine.Types;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Gpu.Shader;
 using Ryujinx.Graphics.Shader;
+using Ryujinx.Memory.Range;
 using System;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
@@ -67,6 +68,9 @@ namespace Ryujinx.Graphics.Gpu.Image
 
         private int _lastFragmentTotal;
 
+        private readonly int _bufferTextureAlignment;
+        private readonly int _bufferImageAlignment;
+
         /// <summary>
         /// Constructs a new instance of the texture bindings manager.
         /// </summary>
@@ -108,6 +112,10 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
 
             _textureCounts = [];
+
+            int alignment = context.Capabilities.TextureBufferOffsetAlignment;
+            _bufferTextureAlignment = context.Capabilities.SupportsTextureBufferPixelAlignment ? 0 : alignment;
+            _bufferImageAlignment = context.Capabilities.SupportsImageBufferPixelAlignment ? 0 : alignment;
         }
 
         /// <summary>
@@ -521,10 +529,12 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                 if (hostTexture != null && texture.Target == Target.TextureBuffer)
                 {
+                    MultiRange range = GetAlignedBufferTextureRange(texture, index, stageIndex, false);
+
                     // Ensure that the buffer texture is using the correct buffer as storage.
                     // Buffers are frequently re-created to accommodate larger data, so we need to re-bind
                     // to ensure we're not using a old buffer that was already deleted.
-                    _channel.BufferManager.SetBufferTextureStorage(stage, hostTexture, texture.Range, bindingInfo, false);
+                    _channel.BufferManager.SetBufferTextureStorage(stage, hostTexture, range, bindingInfo, false);
 
                     // Cache is not used for buffer texture, it must always rebind.
                     state.CachedTexture = null;
@@ -659,7 +669,9 @@ namespace Ryujinx.Graphics.Gpu.Image
                     // Buffers are frequently re-created to accommodate larger data, so we need to re-bind
                     // to ensure we're not using a old buffer that was already deleted.
 
-                    _channel.BufferManager.SetBufferTextureStorage(stage, hostTexture, texture.Range, bindingInfo, true);
+                    MultiRange range = GetAlignedBufferTextureRange(texture, index, stageIndex, true);
+
+                    _channel.BufferManager.SetBufferTextureStorage(stage, hostTexture, range, bindingInfo, true);
 
                     // Cache is not used for buffer texture, it must always rebind.
                     state.CachedTexture = null;
@@ -692,6 +704,61 @@ namespace Ryujinx.Graphics.Gpu.Image
             return specStateMatches;
         }
 
+        /// <summary>
+        /// Gets the aligned address of the texture according to the host requirements.
+        /// </summary>
+        /// <param name="texture">Texture to have its address aligned</param>
+        /// <param name="index">Index of the texture in the shader</param>
+        /// <param name="stageIndex">Index of the shader stage</param>
+        /// <param name="isImage">True if the texture is bound as image, false for sampled textures</param>
+        /// <returns>Aligned address and size of the buffer texture</returns>
+        private MultiRange GetAlignedBufferTextureRange(Texture texture, int index, int stageIndex, bool isImage)
+        {
+            MultiRange range = texture.Range;
+            int alignment = isImage ? _bufferImageAlignment : _bufferTextureAlignment;
+
+            if (alignment != 0)
+            {
+                MemoryRange firstRange = range.GetSubRange(0);
+                ulong misalign = firstRange.Address & ((ulong)alignment - 1);
+                int offset = misalign != 0 ? (int)misalign / texture.Info.FormatInfo.BytesPerPixel : 0;
+
+                if (misalign != 0)
+                {
+                    firstRange = new MemoryRange(firstRange.Address - misalign, firstRange.Size + misalign);
+
+                    if (range.Count > 1)
+                    {
+                        MemoryRange[] ranges = new MemoryRange[range.Count];
+
+                        ranges[0] = firstRange;
+
+                        for (int i = 1; i < range.Count; i++)
+                        {
+                            ranges[i] = range.GetSubRange(i);
+                        }
+
+                        range = new MultiRange(ranges);
+                    }
+                    else
+                    {
+                        range = new MultiRange(firstRange.Address, firstRange.Size);
+                    }
+                }
+
+                if (isImage)
+                {
+                    _context.SupportBufferUpdater.UpdateBufferImageOffset(stageIndex, index, offset);
+                }
+                else
+                {
+                    _context.SupportBufferUpdater.UpdateBufferTextureOffset(stageIndex, index, offset);
+                }
+            }
+
+            return range;
+        }
+
         /// <summary>
         /// Gets the texture descriptor for a given texture handle.
         /// </summary>

src/Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs

@@ -195,16 +195,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             return true;
         }
 
-        /// <summary>
-        /// Determines whether a texture can flush its data back to guest memory.
-        /// </summary>
-        /// <param name="info">Texture that will have its data flushed</param>
-        /// <param name="caps">Host GPU Capabilities</param>
-        /// <returns>True if the texture can flush, false otherwise</returns>
-        public static bool CanTextureFlush(Texture texture, in Capabilities caps)
-        {
-            return !texture.HasImportOverride() && CanTextureFlush(texture.Info, caps);
-        }
 
         /// <summary>
         /// Determines whether a texture can flush its data back to guest memory.
@@ -212,14 +202,15 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="info">Texture information</param>
         /// <param name="caps">Host GPU Capabilities</param>
         /// <returns>True if the texture can flush, false otherwise</returns>
-        private static bool CanTextureFlush(TextureInfo info, in Capabilities caps)
+       public static bool CanTextureFlush(TextureInfo info, Capabilities caps)
         {
-            if (IsFormatHostIncompatible(info, in caps))
+            if (IsFormatHostIncompatible(info, caps))
             {
                 return false; // Flushing this format is not supported, as it may have been converted to another host format.
             }
 
-            if (info.Target is Target.Texture2DMultisample or Target.Texture2DMultisampleArray)
+            if (info.Target is Target.Texture2DMultisample or
+                Target.Texture2DMultisampleArray)
             {
                 return false; // Flushing multisample textures is not supported, the host does not allow getting their data.
             }
@@ -381,12 +372,15 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
 
             // Some APIs align the width for copy and render target textures,
-            // so the width may not match in this case for different uses of the same texture.
+            // so the width may not match for different uses of the same texture.
             // To account for this, we compare the aligned width here.
-            // We expect height to always match exactly, if the texture is the same.
+            // However, matching block footprints are not sufficient for compressed textures;
+            // their logical dimensions must also match.
             if (alignedWidthMatches && lhsSize.Height == rhsSize.Height)
             {
-                return (exact && lhsSize.Width != rhsSize.Width) || lhsSize.Width < rhsSize.Width
+                return ((lhs.FormatInfo.IsCompressed || rhs.FormatInfo.IsCompressed) &&
+                        (Math.Max(1, lhs.Width >> level) != rhs.Width || Math.Max(1, lhs.Height >> level) != rhs.Height)) ||
+                       (exact && lhsSize.Width != rhsSize.Width) || lhsSize.Width < rhsSize.Width
                     ? TextureViewCompatibility.CopyOnly
                     : result;
             }
@@ -397,7 +391,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                 return stride == rhs.Stride ? TextureViewCompatibility.CopyOnly : TextureViewCompatibility.LayoutIncompatible;
             }
-            else if (lhs.Target.IsMultisample() != rhs.Target.IsMultisample() && alignedWidthMatches && lhsAlignedSize.Height == rhsAlignedSize.Height)
+            else if (lhs.Target.IsMultisample != rhs.Target.IsMultisample && alignedWidthMatches && lhsAlignedSize.Height == rhsAlignedSize.Height)
             {
                 // Copy between multisample and non-multisample textures with mismatching size is allowed,
                 // as long aligned size matches.

src/Ryujinx.Graphics.Gpu/Image/TextureGroup.cs

@@ -147,7 +147,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             _allOffsets = size.AllOffsets;
             _sliceSizes = size.SliceSizes;
 
-            if (Storage.Target.HasDepthOrLayers() && Storage.Info.GetSlices() > GranularLayerThreshold)
+            if (Storage.Target.HasDepthOrLayers && Storage.Info.GetSlices() > GranularLayerThreshold)
             {
                 _hasLayerViews = true;
                 _hasMipViews = true;
@@ -182,7 +182,11 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             foreach (TextureIncompatibleOverlap overlap in _incompatibleOverlaps)
             {
-                if (overlap.Compatibility <= TextureViewCompatibility.LayoutIncompatible)
+                // LayoutIncompatible and better may still use a regular texture copy dependency.
+                // Fully Incompatible pairs are not copy compatible in general, but may still qualify for an
+                // exact raw byte copy dependency (checked internally by CreateCopyDependency) when they map
+                // to exactly the same guest memory, such as differently typed/sized aliases of the same data.
+                if (overlap.Compatibility <= TextureViewCompatibility.Incompatible)
                 {
                     CreateCopyDependency(overlap.Group, false, overlap.Compatibility);
                 }
@@ -226,7 +230,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
         }
 
-
         /// <summary>
         /// Flushes incompatible overlaps if the storage format requires it, and they have been modified.
         /// This allows unsupported host formats to accept data written to format aliased textures.
@@ -324,7 +327,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             FlushIncompatibleOverlapsIfNeeded();
 
-            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, _) =>
+            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, bound) =>
             {
                 bool dirty = false;
                 bool anyModified = false;
@@ -479,7 +482,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="texture">The texture to synchronize dependents of</param>
         public void SynchronizeDependents(Texture texture)
         {
-            EvaluateRelevantHandles(texture, (baseHandle, regionCount, _, _) =>
+            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, bound) =>
             {
                 for (int i = 0; i < regionCount; i++)
                 {
@@ -571,7 +574,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             tracked = tracked || ShouldFlushTriggerTracking();
             bool flushed = false;
 
-            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, _) =>
+            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, bound) =>
             {
                 int startSlice = 0;
                 int endSlice = 0;
@@ -652,14 +655,14 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             if (_flushBuffer == BufferHandle.Null)
             {
-                if (!TextureCompatibility.CanTextureFlush(Storage, _context.Capabilities))
+                if (!TextureCompatibility.CanTextureFlush(Storage.Info, _context.Capabilities))
                 {
                     return;
                 }
 
                 bool canImport = Storage.Info.IsLinear && Storage.Info.Stride >= Storage.Info.Width * Storage.Info.FormatInfo.BytesPerPixel;
 
-                IntPtr hostPointer = canImport ? _physicalMemory.GetHostPointer(Storage.Range) : 0;
+                nint hostPointer = canImport ? _physicalMemory.GetHostPointer(Storage.Range) : 0;
 
                 if (hostPointer != 0 && _context.Renderer.PrepareHostMapping(hostPointer, Storage.Size))
                 {
@@ -716,7 +719,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             ClearIncompatibleOverlaps(texture);
 
-            EvaluateRelevantHandles(texture, (baseHandle, regionCount, _, _) =>
+            EvaluateRelevantHandles(texture, (baseHandle, regionCount, split, bound) =>
             {
                 for (int i = 0; i < regionCount; i++)
                 {
@@ -1049,7 +1052,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             int endOffset = _allOffsets[viewEnd] + _sliceSizes[lastLevel];
             int size = endOffset - offset;
 
-            List<RegionHandle> result = new();
+            List<RegionHandle> result = [];
 
             for (int i = 0; i < TextureRange.Count; i++)
             {
@@ -1163,7 +1166,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             SignalAllDirty();
         }
 
-
         /// <summary>
         /// Removes a view from the group, removing it from all overlap lists.
         /// </summary>
@@ -1385,7 +1387,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                 if (_is3D)
                 {
-                    List<TextureGroupHandle> handlesList = new();
+                    List<TextureGroupHandle> handlesList = [];
 
                     for (int i = 0; i < levelHandles; i++)
                     {
@@ -1468,15 +1470,15 @@ namespace Ryujinx.Graphics.Gpu.Image
             // Get the location of each texture within its storage, so we can find the handles to apply the dependency to.
             // This can consist of multiple disjoint regions, for example if this is a mip slice of an array texture.
 
-            List<(int BaseHandle, int RegionCount)> targetRange = new();
+            List<(int BaseHandle, int RegionCount)> targetRange = [];
-            List<(int BaseHandle, int RegionCount)> otherRange = new();
+            List<(int BaseHandle, int RegionCount)> otherRange = [];
 
-            EvaluateRelevantHandles(firstLayer, firstLevel, other.Info.GetSlices(), other.Info.Levels, (baseHandle, regionCount, _, _) =>
+            EvaluateRelevantHandles(firstLayer, firstLevel, other.Info.GetSlices(), other.Info.Levels, (baseHandle, regionCount, split, specialData) =>
             {
                 targetRange.Add((baseHandle, regionCount));
                 return true;
             }, out _);
-            otherGroup.EvaluateRelevantHandles(other, (baseHandle, regionCount, _, _) =>
+            otherGroup.EvaluateRelevantHandles(other, (baseHandle, regionCount, split, specialData) =>
             {
                 otherRange.Add((baseHandle, regionCount));
                 return true;
@@ -1601,7 +1603,15 @@ namespace Ryujinx.Graphics.Gpu.Image
                         TextureInfo info = Storage.Info;
                         TextureInfo otherInfo = other.Storage.Info;
 
-                        bool textureCopy = TextureCompatibility.ViewLayoutCompatible(info, otherInfo, level, otherLevel) &&
+                        // ViewLayoutCompatible/CopySizeMatches only reason about textures with some genuine
+                        // format relationship (LayoutIncompatible or better) - they are not aware of, and must
+                        // never be used to justify, a plain texture-to-texture copy (which some backends
+                        // implement via a reinterpreting view) between fully Incompatible aliases such as a
+                        // depth format and an unrelated color format. For Incompatible pairs, only the strict
+                        // raw byte copy dependency below (which explicitly excludes depth/stencil formats) may
+                        // be used.
+                        bool textureCopy = compatibility != TextureViewCompatibility.Incompatible &&
+                            TextureCompatibility.ViewLayoutCompatible(info, otherInfo, level, otherLevel) &&
                             TextureCompatibility.CopySizeMatches(info, otherInfo, level, otherLevel);
 
                         if (textureCopy || rawCopy)
@@ -1659,9 +1669,12 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             if (!_incompatibleOverlaps.Any(overlap => overlap.Group == other.Group))
             {
-                if (copy && other.Compatibility <= TextureViewCompatibility.LayoutIncompatible)
+                if (copy && other.Compatibility <= TextureViewCompatibility.Incompatible)
                 {
                     // Any of the group's views may share compatibility, even if the parents do not fully.
+                    // Fully Incompatible groups are also let through here, since CreateCopyDependency will
+                    // fall back to an exact raw byte copy dependency for them when the strict requirements
+                    // for one are met (see CanCreateRawCopyDependency).
                     CreateCopyDependency(other.Group, false, other.Compatibility);
                 }
 
@@ -1763,7 +1776,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                     }
                 }
 
-                if (TextureCompatibility.CanTextureFlush(Storage, _context.Capabilities) && !(inBuffer && _flushBufferImported))
+                if (TextureCompatibility.CanTextureFlush(Storage.Info, _context.Capabilities) && !(inBuffer && _flushBufferImported))
                 {
                     FlushSliceRange(false, handle.BaseSlice, handle.BaseSlice + handle.SliceCount, inBuffer, Storage.GetFlushTexture());
                 }
@@ -1803,4 +1816,3 @@ namespace Ryujinx.Graphics.Gpu.Image
         }
     }
 }
-

src/Ryujinx.Graphics.Gpu/Image/TextureGroupHandle.cs

@@ -717,6 +717,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
 
                 DeferredCopy = old.DeferredCopy;
+                DeferredCopyRaw = old.DeferredCopyRaw;
             }
         }
 

src/Ryujinx.Graphics.Gpu/Image/TextureManager.cs

@@ -19,12 +19,39 @@ namespace Ryujinx.Graphics.Gpu.Image
         private readonly TexturePoolCache _texturePoolCache;
         private readonly SamplerPoolCache _samplerPoolCache;
 
+        /// <summary>
+        /// Bound render target texture modification report state.
+        /// </summary>
+        [Flags]
+        private enum BindState : byte
+        {
+            /// <summary>
+            /// Render target texture has not been signalled for modification, and can't be modified on the next render operations.
+            /// </summary>
+            None = 0,
+
+            /// <summary>
+            /// Render target texture has been signalled for modification.
+            /// </summary>
+            Bound = 1 << 0,
+
+            /// <summary>
+            /// Render target texture might be modified on the next render operations.
+            /// </summary>
+            Modified = 1 << 1,
+
+            /// <summary>
+            /// Render target texture has been signalled for modification and might be modified on the next render operations.
+            /// </summary>
+            BoundModified = Bound | Modified,
+        }
+
         private readonly Texture[] _rtColors;
         private readonly ITexture[] _rtHostColors;
-        private readonly bool[] _rtColorsBound;
+        private readonly BindState[] _rtColorsBound;
         private Texture _rtDepthStencil;
         private ITexture _rtHostDs;
-        private bool _rtDsBound;
+        private BindState _rtDsBound;
 
         public int ClipRegionWidth { get; private set; }
         public int ClipRegionHeight { get; private set; }
@@ -55,7 +82,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             _rtColors = new Texture[Constants.TotalRenderTargets];
             _rtHostColors = new ITexture[Constants.TotalRenderTargets];
-            _rtColorsBound = new bool[Constants.TotalRenderTargets];
+            _rtColorsBound = new BindState[Constants.TotalRenderTargets];
         }
 
         /// <summary>
@@ -151,27 +178,39 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         /// <param name="index">The index of the color buffer to set (up to 8)</param>
         /// <param name="color">The color buffer texture</param>
+        /// <param name="modified">Indicates if the following render operations will modidify <paramref name="color"/> contents</param>
         /// <returns>True if render target scale must be updated.</returns>
-        public bool SetRenderTargetColor(int index, Texture color)
+        public bool SetRenderTargetColor(int index, Texture color, bool modified)
         {
             bool hasValue = color != null;
             bool changesScale = (hasValue != (_rtColors[index] != null)) || (hasValue && RenderTargetScale != color.ScaleFactor);
 
             if (_rtColors[index] != color)
             {
-                if (_rtColorsBound[index])
+                Texture oldColor = _rtColors[index];
+
+                if (oldColor != null)
                 {
-                    _rtColors[index]?.SignalModifying(false);
+                    if (_rtColorsBound[index].HasFlag(BindState.Bound))
-                }
+                    {
-                else
+                        oldColor.SignalModifying(false);
-                {
+                    }
-                    _rtColorsBound[index] = true;
+
+                    oldColor.SignalBindingChange(false);
                 }
 
+                _rtColorsBound[index] = modified ? BindState.BoundModified : BindState.None;
+
                 if (color != null)
                 {
                     color.SynchronizeMemory();
-                    color.SignalModifying(true);
+
+                    if (modified)
+                    {
+                        color.SignalModifying(true);
+                    }
+
+                    color.SignalBindingChange(true);
                 }
 
                 _rtColors[index] = color;
@@ -184,27 +223,39 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Sets the render target depth-stencil buffer.
         /// </summary>
         /// <param name="depthStencil">The depth-stencil buffer texture</param>
+        /// <param name="modified">Indicates if the following render operations will modidify <paramref name="depthStencil"/> contents</param>
         /// <returns>True if render target scale must be updated.</returns>
-        public bool SetRenderTargetDepthStencil(Texture depthStencil)
+        public bool SetRenderTargetDepthStencil(Texture depthStencil, bool modified)
         {
             bool hasValue = depthStencil != null;
             bool changesScale = (hasValue != (_rtDepthStencil != null)) || (hasValue && RenderTargetScale != depthStencil.ScaleFactor);
 
             if (_rtDepthStencil != depthStencil)
             {
-                if (_rtDsBound)
+                Texture oldDepthStencil = _rtDepthStencil;
+
+                if (oldDepthStencil != null)
                 {
-                    _rtDepthStencil?.SignalModifying(false);
+                    if (_rtDsBound.HasFlag(BindState.Bound))
-                }
+                    {
-                else
+                        oldDepthStencil.SignalModifying(false);
-                {
+                    }
-                    _rtDsBound = true;
+
+                    oldDepthStencil.SignalBindingChange(false);
                 }
 
+                _rtDsBound = modified ? BindState.BoundModified : BindState.None;
+
                 if (depthStencil != null)
                 {
                     depthStencil.SynchronizeMemory();
-                    depthStencil.SignalModifying(true);
+
+                    if (modified)
+                    {
+                        depthStencil.SignalModifying(true);
+                    }
+
+                    depthStencil.SignalBindingChange(true);
                 }
 
                 _rtDepthStencil = depthStencil;
@@ -443,10 +494,10 @@ namespace Ryujinx.Graphics.Gpu.Image
             {
                 hostDsTexture = dsTexture.HostTexture;
 
-                if (!_rtDsBound)
+                if (_rtDsBound == BindState.Modified)
                 {
                     dsTexture.SignalModifying(true);
-                    _rtDsBound = true;
+                    _rtDsBound |= BindState.Bound;
                 }
             }
 
@@ -470,10 +521,10 @@ namespace Ryujinx.Graphics.Gpu.Image
                 {
                     hostTexture = texture.HostTexture;
 
-                    if (!_rtColorsBound[index])
+                    if (_rtColorsBound[index] == BindState.Modified)
                     {
                         texture.SignalModifying(true);
-                        _rtColorsBound[index] = true;
+                        _rtColorsBound[index] |= BindState.Bound;
                     }
                 }
 
@@ -518,24 +569,37 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             Texture dsTexture = _rtDepthStencil;
 
-            if (dsTexture != null && _rtDsBound)
+            if (dsTexture != null && _rtDsBound.HasFlag(BindState.Bound))
             {
                 dsTexture.SignalModifying(false);
-                _rtDsBound = false;
+                _rtDsBound &= ~BindState.Bound;
             }
 
             for (int index = 0; index < _rtColors.Length; index++)
             {
                 Texture texture = _rtColors[index];
 
-                if (texture != null && _rtColorsBound[index])
+                if (texture != null && _rtColorsBound[index].HasFlag(BindState.Bound))
                 {
                     texture.SignalModifying(false);
-                    _rtColorsBound[index] = false;
+                    _rtColorsBound[index] &= ~BindState.Bound;
                 }
             }
         }
 
+        /// <summary>
+        /// Indicates that the currently bound render targets might be modified, if they are used on the next render operation.
+        /// </summary>
+        public void SignalRenderTargetsModifiable()
+        {
+            _rtDsBound |= BindState.Modified;
+
+            for (int index = 0; index < _rtColorsBound.Length; index++)
+            {
+                _rtColorsBound[index] |= BindState.Modified;
+            }
+        }
+
         /// <summary>
         /// Forces the texture and sampler pools to be re-loaded from the cache on next use.
         /// </summary>
@@ -572,19 +636,11 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             for (int i = 0; i < _rtColors.Length; i++)
             {
-                if (_rtColorsBound[i])
+                _rtColors[i]?.DecrementReferenceCount();
-                {
-                    _rtColors[i]?.DecrementReferenceCount();
-                }
-
                 _rtColors[i] = null;
             }
 
-            if (_rtDsBound)
+            _rtDepthStencil?.DecrementReferenceCount();
-            {
-                _rtDepthStencil?.DecrementReferenceCount();
-            }
-
             _rtDepthStencil = null;
         }
     }

src/Ryujinx.Graphics.Gpu/Memory/SupportBufferUpdater.cs

@@ -129,6 +129,38 @@ namespace Ryujinx.Graphics.Gpu.Memory
             }
         }
 
+        /// <summary>
+        /// Updates the offset used for accessing buffer textures that are not aligned to the host requirements.
+        /// </summary>
+        /// <param name="stageIndex">Index of the shader stage where the texture is used</param>
+        /// <param name="bindingIndex">Index of the texture binding in the shader</param>
+        /// <param name="offset">Offset for the misaligned part of the address</param>
+        public void UpdateBufferTextureOffset(int stageIndex, int bindingIndex, int offset)
+        {
+            if (_data.BufferTextureOffset[stageIndex][bindingIndex].X != offset)
+            {
+                _data.BufferTextureOffset[stageIndex][bindingIndex].X = offset;
+                int index = stageIndex * SupportBuffer.TextureCount + bindingIndex;
+                MarkDirty(SupportBuffer.BufferTextureOffsetOffset + index * sizeof(int) * 4, sizeof(int));
+            }
+        }
+
+        /// <summary>
+        /// Updates the offset used for accessing buffer images that are not aligned to the host requirements.
+        /// </summary>
+        /// <param name="stageIndex">Index of the shader stage where the image is used</param>
+        /// <param name="bindingIndex">Index of the image binding in the shader</param>
+        /// <param name="offset">Offset for the misaligned part of the address</param>
+        public void UpdateBufferImageOffset(int stageIndex, int bindingIndex, int offset)
+        {
+            if (_data.BufferTextureOffset[stageIndex][bindingIndex].Y != offset)
+            {
+                _data.BufferTextureOffset[stageIndex][bindingIndex].Y = offset;
+                int index = stageIndex * SupportBuffer.TextureCount + bindingIndex;
+                MarkDirty(SupportBuffer.BufferTextureOffsetOffset + index * sizeof(int) * 4 + sizeof(int), sizeof(int));
+            }
+        }
+
         /// <summary>
         /// Sets whether the format of a given render target is a BGRA format.
         /// </summary>

src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs

@@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         private const ushort FileFormatVersionMajor = 1;
         private const ushort FileFormatVersionMinor = 2;
         private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
-        private const uint CodeGenVersion = 7354;
+        private const uint CodeGenVersion = 6371;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";
@@ -170,6 +170,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             /// </summary>
             public bool UsesRtLayer;
 
+            /// <summary>
+            /// Indicates that the fragment shader always discards the fragment, not producing any output for the bound render targets.
+            /// </summary>
+            public bool HasUnconditionalDiscard;
+
             /// <summary>
             /// Bit mask with the clip distances written on the vertex stage.
             /// </summary>
@@ -806,6 +811,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 dataInfo.UsesInstanceId,
                 dataInfo.UsesDrawParameters,
                 dataInfo.UsesRtLayer,
+                dataInfo.HasUnconditionalDiscard,
                 dataInfo.ClipDistancesWritten,
                 dataInfo.FragmentOutputMap);
         }
@@ -836,6 +842,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 UsesInstanceId = info.UsesInstanceId,
                 UsesDrawParameters = info.UsesDrawParameters,
                 UsesRtLayer = info.UsesRtLayer,
+                HasUnconditionalDiscard = info.HasUnconditionalDiscard,
                 ClipDistancesWritten = info.ClipDistancesWritten,
                 FragmentOutputMap = info.FragmentOutputMap,
             };

src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs

@@ -207,6 +207,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         public bool QueryHostSupportsBgraFormat() => _context.Capabilities.SupportsBgraFormat;
 
+        public bool QueryHostSupportsBufferImagePixelAlignment() => _context.Capabilities.SupportsImageBufferPixelAlignment;
+
+        public bool QueryHostSupportsBufferTexturePixelAlignment() => _context.Capabilities.SupportsTextureBufferPixelAlignment;
+
         public bool QueryHostSupportsFragmentShaderInterlock() => _context.Capabilities.SupportsFragmentShaderInterlock;
 
         public bool QueryHostSupportsFragmentShaderOrderingIntel() => _context.Capabilities.SupportsFragmentShaderOrderingIntel;

src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs

@@ -430,7 +430,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
                             TranslatorContext lastInVertexPipeline = geometryToCompute ? translatorContexts[4] ?? currentStage : currentStage;
 
-                            program = lastInVertexPipeline.GenerateVertexPassthroughForCompute();
+                            (program, ShaderProgramInfo vacInfo) = lastInVertexPipeline.GenerateVertexPassthroughForCompute();
+                            infoBuilder.AddStageInfoVac(vacInfo);
                         }
                         else
                         {
@@ -535,7 +536,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private ShaderAsCompute CreateHostVertexAsComputeProgram(ShaderProgram program, TranslatorContext context, bool tfEnabled)
         {
             ShaderSource source = new(program.Code, program.BinaryCode, ShaderStage.Compute, program.Language);
-            ShaderInfo info = ShaderInfoBuilder.BuildForVertexAsCompute(_context, program.Info, tfEnabled);
+            ShaderInfo info = ShaderInfoBuilder.BuildForVertexAsCompute(_context, program.Info, context.GetVertexAsComputeInfo(), tfEnabled);
 
             return new(_context.Renderer.CreateProgram([source], info), program.Info, context.GetResourceReservations());
         }

src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs

@@ -95,7 +95,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, int setIndex, int start, int count, bool write = false)
         {
             AddDescriptor(stages, type, setIndex, start, count);
-            AddUsage(stages, type, setIndex, start, count, write);
+            // AddUsage(stages, type, setIndex, start, count, write);
         }
 
         /// <summary>
@@ -159,6 +159,25 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AddUsage(info.Images, stages, isImage: true);
         }
 
+        public void AddStageInfoVac(ShaderProgramInfo info)
+        {
+            ResourceStages stages = info.Stage switch
+            {
+                ShaderStage.Compute => ResourceStages.Compute,
+                ShaderStage.Vertex => ResourceStages.Vertex,
+                ShaderStage.TessellationControl => ResourceStages.TessellationControl,
+                ShaderStage.TessellationEvaluation => ResourceStages.TessellationEvaluation,
+                ShaderStage.Geometry => ResourceStages.Geometry,
+                ShaderStage.Fragment => ResourceStages.Fragment,
+                _ => ResourceStages.None,
+            };
+
+            AddUsage(info.CBuffers, stages, isStorage: false);
+            AddUsage(info.SBuffers, stages, isStorage: true);
+            AddUsage(info.Textures, stages, isImage: false);
+            AddUsage(info.Images, stages, isImage: true);
+        }
+
         /// <summary>
         /// Adds a resource descriptor to the list of descriptors.
         /// </summary>
@@ -422,10 +441,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
         /// <param name="fromCache">True if the compute shader comes from a disk cache, false otherwise</param>
         /// <returns>Shader information</returns>
-        public static ShaderInfo BuildForVertexAsCompute(GpuContext context, ShaderProgramInfo info, bool tfEnabled, bool fromCache = false)
+        public static ShaderInfo BuildForVertexAsCompute(GpuContext context, ShaderProgramInfo info, ShaderProgramInfo info2, bool tfEnabled, bool fromCache = false)
         {
             ShaderInfoBuilder builder = new(context, tfEnabled, vertexAsCompute: true);
 
+            builder.AddStageInfoVac(info2);
             builder.AddStageInfo(info, vertexAsCompute: true);
 
             return builder.Build(null, fromCache);

src/Ryujinx.Graphics.OpenGL/Framebuffer.cs

@@ -20,6 +20,27 @@ namespace Ryujinx.Graphics.OpenGL
         private int _colorsCount;
         private bool _dualSourceBlend;
 
+        public bool HasAttachments
+        {
+            get
+            {
+                if (_depthStencil != null)
+                {
+                    return true;
+                }
+
+                for (int index = 0; index < _colors.Length; index++)
+                {
+                    if (_colors[index] != null)
+                    {
+                        return true;
+                    }
+                }
+
+                return false;
+            }
+        }
+
         public Framebuffer()
         {
             Handle = GL.GenFramebuffer();
@@ -34,6 +55,12 @@ namespace Ryujinx.Graphics.OpenGL
             return Handle;
         }
 
+        public void SetDefaultSize(int width, int height)
+        {
+            GL.FramebufferParameter(FramebufferTarget.Framebuffer, FramebufferDefaultParameter.FramebufferDefaultWidth, Math.Max(1, width));
+            GL.FramebufferParameter(FramebufferTarget.Framebuffer, FramebufferDefaultParameter.FramebufferDefaultHeight, Math.Max(1, height));
+        }
+
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public void AttachColor(int index, TextureView color)
         {
@@ -105,13 +132,20 @@ namespace Ryujinx.Graphics.OpenGL
             _colorsCount = colorsCount;
         }
 
-        private static void SetDrawBuffersImpl(int colorsCount)
+        private void SetDrawBuffersImpl(int colorsCount)
         {
             DrawBuffersEnum[] drawBuffers = new DrawBuffersEnum[colorsCount];
 
             for (int index = 0; index < colorsCount; index++)
             {
-                drawBuffers[index] = DrawBuffersEnum.ColorAttachment0 + index;
+                if (_colors[index] != null)
+                {
+                    drawBuffers[index] = DrawBuffersEnum.ColorAttachment0 + index;
+                }
+                else
+                {
+                    drawBuffers[index] = DrawBuffersEnum.None;
+                }
             }
 
             GL.DrawBuffers(colorsCount, drawBuffers);

src/Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -116,8 +116,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
         {
             TextureView destinationView = (TextureView)destination;
 
-            bool srcIsMultisample = Target.IsMultisample();
+            bool srcIsMultisample = Target.IsMultisample;
-            bool dstIsMultisample = destinationView.Target.IsMultisample();
+            bool dstIsMultisample = destinationView.Target.IsMultisample;
 
             if (dstIsMultisample != srcIsMultisample && Info.Format.IsDepthOrStencil())
             {
@@ -172,8 +172,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
         {
             TextureView destinationView = (TextureView)destination;
 
-            bool srcIsMultisample = Target.IsMultisample();
+            bool srcIsMultisample = Target.IsMultisample;
-            bool dstIsMultisample = destinationView.Target.IsMultisample();
+            bool dstIsMultisample = destinationView.Target.IsMultisample;
 
             if (dstIsMultisample != srcIsMultisample && Info.Format.IsDepthOrStencil())
             {
@@ -216,7 +216,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
             Extents2D srcRegion = new(0, 0, Width, Height);
             Extents2D dstRegion = new(0, 0, destinationView.Width, destinationView.Height);
 
-            if (destinationView.Target.IsMultisample())
+            if (destinationView.Target.IsMultisample)
             {
                 TextureView intermmediate = _renderer.TextureCopy.IntermediatePool.GetOrCreateWithAtLeast(
                     Info.Target,

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -174,6 +174,7 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsGeometryShader: true,
                 supportsGeometryShaderPassthrough: HwCapabilities.SupportsGeometryShaderPassthrough,
                 supportsTransformFeedback: true,
+                supportsImageBufferPixelAlignment: false,
                 supportsImageLoadFormatted: HwCapabilities.SupportsImageLoadFormatted,
                 supportsLayerVertexTessellation: HwCapabilities.SupportsShaderViewportLayerArray,
                 supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat,
@@ -185,6 +186,7 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsShaderBarrierDivergence: !(intelWindows || intelUnix),
                 supportsShaderFloat64: true,
                 supportsShaderNonUniformIndexing: false,
+                supportsTextureBufferPixelAlignment: false,
                 supportsTextureGatherOffsets: true,
                 supportsTextureShadowLod: HwCapabilities.SupportsTextureShadowLod,
                 supportsVertexStoreAndAtomics: true,

src/Ryujinx.Graphics.OpenGL/Pipeline.cs

@@ -1537,6 +1537,11 @@ namespace Ryujinx.Graphics.OpenGL
         {
             DrawCount++;
 
+            if (!_framebuffer.HasAttachments && _viewportArray.Length >= 4)
+            {
+                _framebuffer.SetDefaultSize((int)_viewportArray[2], (int)_viewportArray[3]);
+            }
+
             _unit0Texture?.Bind(0);
         }
 

src/Ryujinx.Graphics.Shader/Decoders/InstDecoders.cs

@@ -522,6 +522,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
 
     enum PMode
     {
+        Idx = 0,
         F4e = 1,
         B4e = 2,
         Rc8 = 3,

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -234,6 +234,24 @@ namespace Ryujinx.Graphics.Shader
             return true;
         }
 
+        /// <summary>
+        /// Queries host support for buffer image access with the buffer offset aligned to a single pixel rather than a fixed alignment.
+        /// </summary>
+        /// <returns>True if the host supports buffer image access with pixel alignment, false otherwise</returns>
+        bool QueryHostSupportsBufferImagePixelAlignment()
+        {
+            return true;
+        }
+
+        /// <summary>
+        /// Queries host support for buffer texture access with the buffer offset aligned to a single pixel rather than a fixed alignment.
+        /// </summary>
+        /// <returns>True if the host supports buffer texture access with pixel alignment, false otherwise</returns>
+        bool QueryHostSupportsBufferTexturePixelAlignment()
+        {
+            return true;
+        }
+
         /// <summary>
         /// Queries host support for fragment shader ordering critical sections on the shader code.
         /// </summary>

src/Ryujinx.Graphics.Shader/Instructions/InstEmit.cs

@@ -215,34 +215,6 @@ namespace Ryujinx.Graphics.Shader.Instructions
             context.TranslatorContext.GpuAccessor.Log("Shader instruction Pret is not implemented.");
         }
 
-        public static void PrmtR(EmitterContext context)
-        {
-            context.GetOp<InstPrmtR>();
-
-            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtR is not implemented.");
-        }
-
-        public static void PrmtI(EmitterContext context)
-        {
-            context.GetOp<InstPrmtI>();
-
-            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtI is not implemented.");
-        }
-
-        public static void PrmtC(EmitterContext context)
-        {
-            context.GetOp<InstPrmtC>();
-
-            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtC is not implemented.");
-        }
-
-        public static void PrmtRc(EmitterContext context)
-        {
-            context.GetOp<InstPrmtRc>();
-
-            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtRc is not implemented.");
-        }
-
         public static void R2b(EmitterContext context)
         {
             context.GetOp<InstR2b>();

src/Ryujinx.Graphics.Shader/Instructions/InstEmitMove.cs

@@ -1,7 +1,7 @@
 using Ryujinx.Graphics.Shader.Decoders;
 using Ryujinx.Graphics.Shader.IntermediateRepresentation;
 using Ryujinx.Graphics.Shader.Translation;
-
+using System;
 using static Ryujinx.Graphics.Shader.Instructions.InstEmitHelper;
 using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
 
@@ -37,6 +37,38 @@ namespace Ryujinx.Graphics.Shader.Instructions
             context.Copy(GetDest(op.Dest), GetSrcImm(context, op.Imm32));
         }
 
+        public static void PrmtR(EmitterContext context)
+        {
+            context.GetOp<InstPrmtR>();
+
+            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtR is not implemented.");
+        }
+
+        public static void PrmtI(EmitterContext context)
+        {
+            InstPrmtI op = context.GetOp<InstPrmtI>();
+
+            Operand op1 = GetSrcReg(context, op.SrcA);
+            Operand control = GetSrcImm(context, op.Imm20);
+            Operand op2 = GetSrcReg(context, op.SrcC);
+
+            EmitPrmt(context, op1, control, op2, op.PMode, op.Dest);
+        }
+
+        public static void PrmtC(EmitterContext context)
+        {
+            context.GetOp<InstPrmtC>();
+
+            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtC is not implemented.");
+        }
+
+        public static void PrmtRc(EmitterContext context)
+        {
+            context.GetOp<InstPrmtRc>();
+
+            context.TranslatorContext.GpuAccessor.Log("Shader instruction PrmtRc is not implemented.");
+        }
+
         public static void R2pR(EmitterContext context)
         {
             InstR2pR op = context.GetOp<InstR2pR>();
@@ -169,6 +201,39 @@ namespace Ryujinx.Graphics.Shader.Instructions
             context.Copy(GetDest(op.Dest), src);
         }
 
+        public static void SelR(EmitterContext context)
+        {
+            InstSelR op = context.GetOp<InstSelR>();
+
+            Operand srcA = GetSrcReg(context, op.SrcA);
+            Operand srcB = GetSrcReg(context, op.SrcB);
+            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
+
+            EmitSel(context, srcA, srcB, srcPred, op.Dest);
+        }
+
+        public static void SelI(EmitterContext context)
+        {
+            InstSelI op = context.GetOp<InstSelI>();
+
+            Operand srcA = GetSrcReg(context, op.SrcA);
+            Operand srcB = GetSrcImm(context, Imm20ToSInt(op.Imm20));
+            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
+
+            EmitSel(context, srcA, srcB, srcPred, op.Dest);
+        }
+
+        public static void SelC(EmitterContext context)
+        {
+            InstSelC op = context.GetOp<InstSelC>();
+
+            Operand srcA = GetSrcReg(context, op.SrcA);
+            Operand srcB = GetSrcCbuf(context, op.CbufSlot, op.CbufOffset);
+            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
+
+            EmitSel(context, srcA, srcB, srcPred, op.Dest);
+        }
+
         private static Operand EmitLoadSubgroupLaneId(EmitterContext context)
         {
             if (context.TranslatorContext.GpuAccessor.QueryHostSubgroupSize() <= 32)
@@ -215,37 +280,34 @@ namespace Ryujinx.Graphics.Shader.Instructions
             }
         }
 
-        public static void SelR(EmitterContext context)
+        private static void EmitPrmt(EmitterContext context, Operand op1, Operand control, Operand op2, PMode pMode, int rd)
         {
-            InstSelR op = context.GetOp<InstSelR>();
+            if (pMode == PMode.Idx)
+            {
+                Operand res = Const(0);
 
-            Operand srcA = GetSrcReg(context, op.SrcA);
+                for (int b = 0; b < 4; b++)
-            Operand srcB = GetSrcReg(context, op.SrcB);
+                {
-            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
+                    Operand sel = context.ShiftRightU32(control, Const(b * 4));
+                    Operand byteSel = context.BitwiseAnd(sel, Const(7));
+                    Operand copySign = context.BitwiseAnd(sel, Const(8));
 
-            EmitSel(context, srcA, srcB, srcPred, op.Dest);
+                    Operand srcOp = context.ConditionalSelect(context.BitwiseAnd(byteSel, Const(4)), op2, op1);
-        }
+                    Operand srcValue = context.ShiftRightU32(srcOp, context.ShiftLeft(context.BitwiseAnd(byteSel, Const(3)), Const(3)));
+                    Operand srcSign = context.ShiftRightS32(context.ShiftLeft(srcValue, Const(24)), Const(31));
 
-        public static void SelI(EmitterContext context)
+                    srcValue = context.ConditionalSelect(copySign, srcSign, srcValue);
-        {
+                    srcValue = context.BitwiseAnd(srcValue, Const(0xff));
-            InstSelI op = context.GetOp<InstSelI>();
 
-            Operand srcA = GetSrcReg(context, op.SrcA);
+                    res = context.BitfieldInsert(res, srcValue, Const(b * 8), Const(8));
-            Operand srcB = GetSrcImm(context, Imm20ToSInt(op.Imm20));
+                }
-            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
 
-            EmitSel(context, srcA, srcB, srcPred, op.Dest);
+                context.Copy(GetDest(rd), res);
-        }
+            }
-
+            else
-        public static void SelC(EmitterContext context)
+            {
-        {
+                throw new NotImplementedException(pMode.ToString());
-            InstSelC op = context.GetOp<InstSelC>();
+            }
-
-            Operand srcA = GetSrcReg(context, op.SrcA);
-            Operand srcB = GetSrcCbuf(context, op.CbufSlot, op.CbufOffset);
-            Operand srcPred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
-
-            EmitSel(context, srcA, srcB, srcPred, op.Dest);
         }
 
         private static void EmitR2p(EmitterContext context, Operand value, Operand mask, ByteSel byteSel, bool ccpr)

src/Ryujinx.Graphics.Shader/IntermediateRepresentation/BasicBlock.cs

@@ -23,6 +23,7 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
             set => _branch = AddSuccessor(_branch, value);
         }
 
+        public bool HasSuccessor => _branch != null || _next != null;
         public bool HasBranch => _branch != null;
         public bool Reachable => Index == 0 || Predecessors.Count != 0;
 

src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Shader
         public bool UsesInstanceId { get; }
         public bool UsesDrawParameters { get; }
         public bool UsesRtLayer { get; }
+        public bool HasUnconditionalDiscard { get; }
         public byte ClipDistancesWritten { get; }
         public int FragmentOutputMap { get; }
 
@@ -34,6 +35,7 @@ namespace Ryujinx.Graphics.Shader
             bool usesInstanceId,
             bool usesDrawParameters,
             bool usesRtLayer,
+            bool hasUnconditionalDiscard,
             byte clipDistancesWritten,
             int fragmentOutputMap)
         {
@@ -50,6 +52,7 @@ namespace Ryujinx.Graphics.Shader
             UsesInstanceId = usesInstanceId;
             UsesDrawParameters = usesDrawParameters;
             UsesRtLayer = usesRtLayer;
+            HasUnconditionalDiscard = hasUnconditionalDiscard;
             ClipDistancesWritten = clipDistancesWritten;
             FragmentOutputMap = fragmentOutputMap;
         }

src/Ryujinx.Graphics.Shader/SupportBuffer.cs

@@ -24,6 +24,7 @@ namespace Ryujinx.Graphics.Shader
         RenderScale,
         TfeOffset,
         TfeVertexCount,
+        BufferTextureOffset,
     }
 
     public struct SupportBuffer
@@ -42,10 +43,13 @@ namespace Ryujinx.Graphics.Shader
         public static readonly int ComputeRenderScaleOffset;
         public static readonly int TfeOffsetOffset;
         public static readonly int TfeVertexCountOffset;
+        public static readonly int BufferTextureOffsetOffset;
 
         public const int FragmentIsBgraCount = 8;
+        public const int TextureCount = 64;
+
         // One for the render target, 64 for the textures, and 8 for the images.
-        public const int RenderScaleMaxCount = 1 + 64 + 8;
+        public const int RenderScaleMaxCount = 1 + TextureCount + 8;
 
         private static int OffsetOf<T>(ref SupportBuffer storage, ref T target)
         {
@@ -68,6 +72,7 @@ namespace Ryujinx.Graphics.Shader
             ComputeRenderScaleOffset = GraphicsRenderScaleOffset + FieldSize;
             TfeOffsetOffset = OffsetOf(ref instance, ref instance.TfeOffset);
             TfeVertexCountOffset = OffsetOf(ref instance, ref instance.TfeVertexCount);
+            BufferTextureOffsetOffset = OffsetOf(ref instance, ref instance.BufferTextureOffset);
         }
 
         internal static StructureType GetStructureType()
@@ -80,7 +85,8 @@ namespace Ryujinx.Graphics.Shader
                 new StructureField(AggregateType.S32, "frag_scale_count"),
                 new StructureField(AggregateType.Array | AggregateType.FP32, "render_scale", RenderScaleMaxCount),
                 new StructureField(AggregateType.Vector4 | AggregateType.S32, "tfe_offset"),
-                new StructureField(AggregateType.S32, "tfe_vertex_count")
+                new StructureField(AggregateType.S32, "tfe_vertex_count"),
+                new StructureField(AggregateType.Array | AggregateType.Vector2 | AggregateType.S32, "buffer_texture_offset")
             ]);
         }
 
@@ -95,5 +101,7 @@ namespace Ryujinx.Graphics.Shader
 
         public Vector4<int> TfeOffset;
         public Vector4<int> TfeVertexCount;
+
+        public Array5<Array64<Vector4<int>>> BufferTextureOffset;
     }
 }

src/Ryujinx.Graphics.Shader/Translation/FeatureFlags.cs

@@ -26,5 +26,6 @@ namespace Ryujinx.Graphics.Shader.Translation
         SharedMemory = 1 << 11,
         Store = 1 << 12,
         VtgAsCompute = 1 << 13,
+        UnconditionalDiscard = 1 << 14,
     }
 }

src/Ryujinx.Graphics.Shader/Translation/FeatureIdentification.cs

@@ -0,0 +1,38 @@
+using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+
+namespace Ryujinx.Graphics.Shader.Translation
+{
+    static class FeatureIdentification
+    {
+        public static void RunPass(BasicBlock[] blocks, ShaderStage stage, ref FeatureFlags usedFeatures)
+        {
+            if (stage == ShaderStage.Fragment)
+            {
+                bool endsWithDiscardOnly = true;
+
+                for (int blockIndex = 0; blockIndex < blocks.Length; blockIndex++)
+                {
+                    BasicBlock block = blocks[blockIndex];
+
+                    if (block.HasSuccessor)
+                    {
+                        continue;
+                    }
+
+                    if (block.Operations.Count == 0 ||
+                        block.Operations.Last.Value is not Operation operation ||
+                        operation.Inst != Instruction.Discard)
+                    {
+                        endsWithDiscardOnly = false;
+                        break;
+                    }
+                }
+
+                if (endsWithDiscardOnly)
+                {
+                    usedFeatures |= FeatureFlags.UnconditionalDiscard;
+                }
+            }
+        }
+    }
+}

src/Ryujinx.Graphics.Shader/Translation/Optimizations/BindlessElimination.cs

@@ -70,19 +70,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
             Operand bindlessHandle = texOp.GetSource(0);
 
-            if (bindlessHandle.AsgOp is PhiNode phi)
+            if (!IsBindlessAccessAllowed(bindlessHandle))
-            {
-                for (int srcIndex = 0; srcIndex < phi.SourcesCount; srcIndex++)
-                {
-                    Operand phiSource = phi.GetSource(srcIndex);
-
-                    if (phiSource.AsgOp is not PhiNode && !IsBindlessAccessAllowed(phiSource))
-                    {
-                        return false;
-                    }
-                }
-            }
-            else if (!IsBindlessAccessAllowed(bindlessHandle))
             {
                 return false;
             }
@@ -100,7 +88,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
             texOp.SetSource(0, textureIndex);
 
-            bool hasSampler = !texOp.Inst.IsImage();
+            // Buffer textures use a TIC/BufferView only and never consume a sampler.
+            // Avoiding a sampler descriptor array here is especially important for bindless
+            // texture buffers, where the array can span the entire sampler pool for no benefit.
+            bool hasSampler = !texOp.Inst.IsImage() && texOp.Type != SamplerType.TextureBuffer;
 
             SetBindingPair textureSetAndBinding = resourceManager.GetTextureOrImageBinding(
                 texOp.Inst,
@@ -143,26 +134,112 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
         private static bool IsBindlessAccessAllowed(Operand bindlessHandle)
         {
-            if (bindlessHandle.Type == OperandType.ConstantBuffer)
+            // Walk only SSA merges and integer operations that can transparently construct
-            {
+            // or select a packed texture/sampler handle. Do not walk arbitrary operations:
-                // Bindless access with handles from constant buffer is allowed.
+            // finding an unrelated resource load elsewhere in the SSA graph must not enable
+            // descriptor-array access for this handle.
+            const int MaxVisitedOperands = 256;
+            const int MaxDepth = 64;
 
-                return true;
+            Stack<(Operand Operand, int Depth)> work = new();
+            HashSet<Operand> visited = new();
+
+            work.Push((bindlessHandle, 0));
+
+            while (work.Count != 0 && visited.Count < MaxVisitedOperands)
+            {
+                (Operand operand, int depth) = work.Pop();
+
+                if (operand == null || depth > MaxDepth || !visited.Add(operand))
+                {
+                    continue;
+                }
+
+                if (operand.Type == OperandType.ConstantBuffer)
+                {
+                    // Constant buffers are accepted by the existing direct-handle path and
+                    // remain accepted when SSA merges or integer handle arithmetic obscure them.
+                    return true;
+                }
+
+                if (operand.AsgOp is PhiNode phi)
+                {
+                    for (int index = 0; index < phi.SourcesCount; index++)
+                    {
+                        Operand source = phi.GetSource(index);
+
+                        if (source.Type != OperandType.Undefined)
+                        {
+                            work.Push((source, depth + 1));
+                        }
+                    }
+
+                    continue;
+                }
+
+                if (operand.AsgOp is not Operation operation)
+                {
+                    continue;
+                }
+
+                Instruction inst = operation.Inst & Instruction.Mask;
+
+                if (inst == Instruction.Load)
+                {
+                    // Preserve the performance restriction. Static CBUF operands are already
+                    // accepted above; accept their dynamically indexed Load form as the same
+                    // resource class, plus the existing shader-input and storage-buffer sources.
+                    // Other loads remain traversal barriers and cannot authorize a pool array.
+                    if (operation.StorageKind == StorageKind.ConstantBuffer ||
+                        operation.StorageKind == StorageKind.Input ||
+                        operation.StorageKind == StorageKind.StorageBuffer)
+                    {
+                        return true;
+                    }
+
+                    continue;
+                }
+
+                if (!IsHandleConstructionOperation(inst))
+                {
+                    // Texture/image operations, arbitrary loads, floating-point conversions,
+                    // calls and other unrelated calculations are deliberate traversal barriers.
+                    continue;
+                }
+
+                for (int index = 0; index < operation.SourcesCount; index++)
+                {
+                    work.Push((operation.GetSource(index), depth + 1));
+                }
             }
 
-            if (bindlessHandle.AsgOp is not Operation handleOp ||
+            return false;
-                handleOp.Inst != Instruction.Load ||
+        }
-                (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
-            {
-                // Right now, we only allow bindless access when the handle comes from a shader input or storage buffer.
-                // This is an artificial limitation to prevent it from being used in cases where it
-                // would have a large performance impact of loading all textures in the pool.
-                // It might be removed in the future, if we can mitigate the performance impact.
 
-                return false;
+        private static bool IsHandleConstructionOperation(Instruction inst)
-            }
+        {
-
+            return inst is
-            return true;
+                Instruction.Add or
+                Instruction.Subtract or
+                Instruction.Multiply or
+                Instruction.MultiplyHighS32 or
+                Instruction.MultiplyHighU32 or
+                Instruction.BitwiseAnd or
+                Instruction.BitwiseExclusiveOr or
+                Instruction.BitwiseNot or
+                Instruction.BitwiseOr or
+                Instruction.BitfieldExtractS32 or
+                Instruction.BitfieldExtractU32 or
+                Instruction.BitfieldInsert or
+                Instruction.ShiftLeft or
+                Instruction.ShiftRightS32 or
+                Instruction.ShiftRightU32 or
+                Instruction.MinimumU32 or
+                Instruction.MaximumU32 or
+                Instruction.ClampU32 or
+                Instruction.ConditionalSelect or
+                Instruction.Copy or
+                Instruction.VectorExtract;
         }
 
         private static bool TryConvertBindless(BasicBlock block, ResourceManager resourceManager, IGpuAccessor gpuAccessor, TextureOperation texOp)

src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs

@@ -43,6 +43,11 @@ namespace Ryujinx.Graphics.Shader.Translation
         private readonly Dictionary<TextureInfo, TextureMeta> _usedTextures;
         private readonly Dictionary<TextureInfo, TextureMeta> _usedImages;
 
+        private readonly List<BufferDefinition> _vacConstantBuffers;
+        private readonly List<BufferDefinition> _vacStorageBuffers;
+        private readonly List<TextureDefinition> _vacTextures;
+        private readonly List<TextureDefinition> _vacImages;
+
         public int LocalMemoryId { get; private set; }
         public int SharedMemoryId { get; private set; }
 
@@ -78,6 +83,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             _usedTextures = new();
             _usedImages = new();
 
+            _vacConstantBuffers = new();
+            _vacStorageBuffers = new();
+            _vacTextures = new();
+            _vacImages = new();
+
             Properties.AddOrUpdateConstantBuffer(new(BufferLayout.Std140, 0, SupportBuffer.Binding, "support_buffer", SupportBuffer.GetStructureType()));
 
             LocalMemoryId = -1;
@@ -563,6 +573,76 @@ namespace Ryujinx.Graphics.Shader.Translation
             return descriptors.ToArray();
         }
 
+        public ShaderProgramInfo GetVertexAsComputeInfo(bool isVertex = false)
+        {
+            var cbDescriptors = new BufferDescriptor[_vacConstantBuffers.Count];
+            int cbDescriptorIndex = 0;
+
+            foreach (BufferDefinition definition in _vacConstantBuffers)
+            {
+                cbDescriptors[cbDescriptorIndex++] = new BufferDescriptor(definition.Set, definition.Binding, 0, 0, 0, BufferUsageFlags.None);
+            }
+
+            var sbDescriptors = new BufferDescriptor[_vacStorageBuffers.Count];
+            int sbDescriptorIndex = 0;
+
+            foreach (BufferDefinition definition in _vacStorageBuffers)
+            {
+                sbDescriptors[sbDescriptorIndex++] = new BufferDescriptor(definition.Set, definition.Binding, 0, 0, 0, BufferUsageFlags.Write);
+            }
+
+            var tDescriptors = new TextureDescriptor[_vacTextures.Count];
+            int tDescriptorIndex = 0;
+
+            foreach (TextureDefinition definition in _vacTextures)
+            {
+                tDescriptors[tDescriptorIndex++] = new TextureDescriptor(
+                    definition.Set,
+                    definition.Binding,
+                    definition.Type,
+                    definition.Format,
+                    0,
+                    0,
+                    definition.ArrayLength,
+                    definition.Separate,
+                    definition.Flags);
+            }
+
+            var iDescriptors = new TextureDescriptor[_vacImages.Count];
+            int iDescriptorIndex = 0;
+
+            foreach (TextureDefinition definition in _vacImages)
+            {
+                iDescriptors[iDescriptorIndex++] = new TextureDescriptor(
+                    definition.Set,
+                    definition.Binding,
+                    definition.Type,
+                    definition.Format,
+                    0,
+                    0,
+                    definition.ArrayLength,
+                    definition.Separate,
+                    definition.Flags);
+            }
+
+            return new ShaderProgramInfo(
+                cbDescriptors,
+                sbDescriptors,
+                tDescriptors,
+                iDescriptors,
+                isVertex ? ShaderStage.Vertex : ShaderStage.Compute,
+                0,
+                0,
+                0,
+                false,
+                false,
+                false,
+                false,
+                false,
+                0,
+                0);
+        }
+
         public bool TryGetCbufSlotAndHandleForTexture(int binding, out int cbufSlot, out int handle)
         {
             foreach ((TextureInfo info, TextureMeta meta) in _usedTextures)
@@ -627,6 +707,30 @@ namespace Ryujinx.Graphics.Shader.Translation
             Properties.AddOrUpdateStorageBuffer(new(BufferLayout.Std430, setIndex, binding, name, type));
         }
 
+        public void AddVertexAsComputeConstantBuffer(BufferDefinition definition)
+        {
+            _vacConstantBuffers.Add(definition);
+            Properties.AddOrUpdateConstantBuffer(definition);
+        }
+
+        public void AddVertexAsComputeStorageBuffer(BufferDefinition definition)
+        {
+            _vacStorageBuffers.Add(definition);
+            Properties.AddOrUpdateStorageBuffer(definition);
+        }
+
+        public void AddVertexAsComputeTexture(TextureDefinition definition)
+        {
+            _vacTextures.Add(definition);
+            Properties.AddOrUpdateTexture(definition);
+        }
+
+        public void AddVertexAsComputeImage(TextureDefinition definition)
+        {
+            _vacImages.Add(definition);
+            Properties.AddOrUpdateImage(definition);
+        }
+
         public static string GetShaderStagePrefix(ShaderStage stage)
         {
             uint index = (uint)stage;

src/Ryujinx.Graphics.Shader/Translation/Transforms/TexturePass.cs

@@ -1,4 +1,5 @@
 using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+using Ryujinx.Graphics.Shader.Translation.Optimizations;
 using System.Collections.Generic;
 using System.Linq;
 using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
@@ -27,7 +28,19 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
 
                     if (texOp.Type == SamplerType.TextureBuffer && !context.GpuAccessor.QueryHostSupportsSnormBufferTextureFormat())
                     {
-                        node = InsertSnormNormalization(node, context.ResourceManager, context.GpuAccessor);
+                        if (!context.GpuAccessor.QueryHostSupportsSnormBufferTextureFormat())
+                        {
+                            node = InsertSnormNormalization(node, context.ResourceManager, context.GpuAccessor);
+                        }
+
+                        if (!context.GpuAccessor.QueryHostSupportsBufferTexturePixelAlignment())
+                        {
+                            node = InsertCoordOffset(node, context.ResourceManager, context.Stage, isImage: false);
+                        }
+                    }
+                    else if (!context.GpuAccessor.QueryHostSupportsBufferTexturePixelAlignment() && texOp.Inst.IsImage())
+                    {
+                        node = InsertCoordOffset(node, context.ResourceManager, context.Stage, isImage: true);
                     }
                 }
             }
@@ -278,6 +291,87 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
             return node;
         }
 
+        private static LinkedListNode<INode> InsertCoordOffset(LinkedListNode<INode> node, ResourceManager resourceManager, ShaderStage stage, bool isImage)
+        {
+            // Some GPUs have fixed alignment requirements for buffer textures.
+            // For those cases, we bind the aligned buffer offset, and apply the remaining offset on the shader.
+
+            TextureOperation texOp = (TextureOperation)node.Value;
+
+            if ((texOp.Type & SamplerType.Mask) != SamplerType.TextureBuffer)
+            {
+                return node;
+            }
+
+            Operand[] sources = new Operand[texOp.SourcesCount];
+
+            for (int i = 0; i < texOp.SourcesCount; i++)
+            {
+                sources[i] = texOp.GetSource(i);
+            }
+
+            bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
+            bool isIndexed = resourceManager.IsArrayOfTexturesOrImages(texOp.Binding, isImage);
+
+            int coordsIndex = isBindless || isIndexed ? 1 : 0;
+
+            Operand[] dests = new Operand[texOp.DestsCount];
+
+            for (int i = 0; i < texOp.DestsCount; i++)
+            {
+                dests[i] = texOp.GetDest(i);
+            }
+
+            LinkedListNode<INode> oldNode = node;
+
+            Operand source = sources[coordsIndex];
+            Operand offset = Local();
+            Operand coordPlusOffset = Local();
+
+            int stageIndex = stage switch
+            {
+                ShaderStage.TessellationControl => 1,
+                ShaderStage.TessellationEvaluation => 2,
+                ShaderStage.Geometry => 3,
+                ShaderStage.Fragment => 4,
+                _ => 0,
+            };
+
+            int bindingIndex = isImage
+                ? resourceManager.FindImageDescriptorIndex(texOp.Binding)
+                : resourceManager.FindTextureDescriptorIndex(texOp.Binding);
+
+            node.List.AddBefore(node, new Operation(
+                Instruction.Load,
+                StorageKind.ConstantBuffer,
+                offset,
+                Const(SupportBuffer.Binding),
+                Const((int)SupportBufferField.BufferTextureOffset),
+                Const(stageIndex * SupportBuffer.TextureCount + bindingIndex),
+                Const(isImage ? 1 : 0)));
+
+            node.List.AddBefore(node, new Operation(Instruction.Add, coordPlusOffset, source, offset));
+
+            sources[coordsIndex] = coordPlusOffset;
+
+            TextureOperation newTexOp = new(
+                texOp.Inst,
+                texOp.Type,
+                texOp.Format,
+                texOp.Flags,
+                texOp.Set,
+                texOp.Binding,
+                texOp.Index,
+                dests,
+                sources);
+
+            node = node.List.AddBefore(node, newTexOp);
+
+            Utils.DeleteNode(oldNode, texOp);
+
+            return node;
+        }
+
         private static LinkedListNode<INode> InsertConstOffsets(LinkedListNode<INode> node, ResourceManager resourceManager, IGpuAccessor gpuAccessor, ShaderStage stage)
         {
             // Non-constant texture offsets are not allowed (according to the spec),

src/Ryujinx.Graphics.Shader/Translation/TranslatorContext.cs

@@ -301,6 +301,11 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                     Optimizer.RunPass(context);
                     TransformPasses.RunPass(context);
+
+                    if (i == 0)
+                    {
+                        FeatureIdentification.RunPass(cfg.Blocks, Definitions.Stage, ref usedFeatures);
+                    }
                 }
 
                 funcs[i] = new Function(cfg.Blocks, $"fun{i}", false, inArgumentsCount, outArgumentsCount);
@@ -353,6 +358,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 usedFeatures.HasFlag(FeatureFlags.InstanceId),
                 usedFeatures.HasFlag(FeatureFlags.DrawParameters),
                 usedFeatures.HasFlag(FeatureFlags.RtLayer),
+                usedFeatures.HasFlag(FeatureFlags.UnconditionalDiscard),
                 clipDistancesWritten,
                 originalDefinitions.OmapTargets);
 
@@ -393,7 +399,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 {
                     int binding = resourceManager.Reservations.GetTfeBufferStorageBufferBinding(i);
                     BufferDefinition tfeDataBuffer = new(BufferLayout.Std430, 1, binding, $"tfe_data{i}", tfeDataStruct);
-                    resourceManager.Properties.AddOrUpdateStorageBuffer(tfeDataBuffer);
+                    resourceManager.AddVertexAsComputeStorageBuffer(tfeDataBuffer);
                 }
             }
 
@@ -401,7 +407,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             {
                 int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
                 BufferDefinition vertexInfoBuffer = new(BufferLayout.Std140, 0, vertexInfoCbBinding, "vb_info", VertexInfoBuffer.GetStructureType());
-                resourceManager.Properties.AddOrUpdateConstantBuffer(vertexInfoBuffer);
+                resourceManager.AddVertexAsComputeConstantBuffer(vertexInfoBuffer);
 
                 StructureType vertexOutputStruct = new([
                     new StructureField(AggregateType.Array | AggregateType.FP32, "data", 0)
@@ -409,13 +415,13 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 int vertexOutputSbBinding = resourceManager.Reservations.VertexOutputStorageBufferBinding;
                 BufferDefinition vertexOutputBuffer = new(BufferLayout.Std430, 1, vertexOutputSbBinding, "vertex_output", vertexOutputStruct);
-                resourceManager.Properties.AddOrUpdateStorageBuffer(vertexOutputBuffer);
+                resourceManager.AddVertexAsComputeStorageBuffer(vertexOutputBuffer);
 
                 if (Stage == ShaderStage.Vertex)
                 {
                     SetBindingPair ibSetAndBinding = resourceManager.Reservations.GetIndexBufferTextureSetAndBinding();
                     TextureDefinition indexBuffer = new(ibSetAndBinding.SetIndex, ibSetAndBinding.Binding, "ib_data", SamplerType.TextureBuffer);
-                    resourceManager.Properties.AddOrUpdateTexture(indexBuffer);
+                    resourceManager.AddVertexAsComputeTexture(indexBuffer);
 
                     int inputMap = _program.AttributeUsage.UsedInputAttributes;
 
@@ -424,7 +430,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                         int location = BitOperations.TrailingZeroCount(inputMap);
                         SetBindingPair setAndBinding = resourceManager.Reservations.GetVertexBufferTextureSetAndBinding(location);
                         TextureDefinition vaBuffer = new(setAndBinding.SetIndex, setAndBinding.Binding, $"vb_data{location}", SamplerType.TextureBuffer);
-                        resourceManager.Properties.AddOrUpdateTexture(vaBuffer);
+                        resourceManager.AddVertexAsComputeTexture(vaBuffer);
 
                         inputMap &= ~(1 << location);
                     }
@@ -433,11 +439,11 @@ namespace Ryujinx.Graphics.Shader.Translation
                 {
                     SetBindingPair trbSetAndBinding = resourceManager.Reservations.GetTopologyRemapBufferTextureSetAndBinding();
                     TextureDefinition remapBuffer = new(trbSetAndBinding.SetIndex, trbSetAndBinding.Binding, "trb_data", SamplerType.TextureBuffer);
-                    resourceManager.Properties.AddOrUpdateTexture(remapBuffer);
+                    resourceManager.AddVertexAsComputeTexture(remapBuffer);
 
                     int geometryVbOutputSbBinding = resourceManager.Reservations.GeometryVertexOutputStorageBufferBinding;
                     BufferDefinition geometryVbOutputBuffer = new(BufferLayout.Std430, 1, geometryVbOutputSbBinding, "geometry_vb_output", vertexOutputStruct);
-                    resourceManager.Properties.AddOrUpdateStorageBuffer(geometryVbOutputBuffer);
+                    resourceManager.AddVertexAsComputeStorageBuffer(geometryVbOutputBuffer);
 
                     StructureType geometryIbOutputStruct = new([
                         new StructureField(AggregateType.Array | AggregateType.U32, "data", 0)
@@ -445,7 +451,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                     int geometryIbOutputSbBinding = resourceManager.Reservations.GeometryIndexOutputStorageBufferBinding;
                     BufferDefinition geometryIbOutputBuffer = new(BufferLayout.Std430, 1, geometryIbOutputSbBinding, "geometry_ib_output", geometryIbOutputStruct);
-                    resourceManager.Properties.AddOrUpdateStorageBuffer(geometryIbOutputBuffer);
+                    resourceManager.AddVertexAsComputeStorageBuffer(geometryIbOutputBuffer);
                 }
 
                 resourceManager.SetVertexAsComputeLocalMemories(Definitions.Stage, Definitions.InputTopology);
@@ -478,12 +484,17 @@ namespace Ryujinx.Graphics.Shader.Translation
             return new ResourceReservations(GpuAccessor, IsTransformFeedbackEmulated, vertexAsCompute: true, _vertexOutput, ioUsage);
         }
 
+        public ShaderProgramInfo GetVertexAsComputeInfo()
+        {
+            return CreateResourceManager(true).GetVertexAsComputeInfo();
+        }
+
         public void SetVertexOutputMapForGeometryAsCompute(TranslatorContext vertexContext)
         {
             _vertexOutput = vertexContext._program.GetIoUsage();
         }
 
-        public ShaderProgram GenerateVertexPassthroughForCompute()
+        public (ShaderProgram, ShaderProgramInfo) GenerateVertexPassthroughForCompute()
         {
             AttributeUsage attributeUsage = new(GpuAccessor);
             ResourceManager resourceManager = new(ShaderStage.Vertex, GpuAccessor);
@@ -495,7 +506,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             if (Stage == ShaderStage.Vertex)
             {
                 BufferDefinition vertexInfoBuffer = new(BufferLayout.Std140, 0, vertexInfoCbBinding, "vb_info", VertexInfoBuffer.GetStructureType());
-                resourceManager.Properties.AddOrUpdateConstantBuffer(vertexInfoBuffer);
+                resourceManager.AddVertexAsComputeConstantBuffer(vertexInfoBuffer);
             }
 
             StructureType vertexInputStruct = new([
@@ -504,7 +515,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
             int vertexDataSbBinding = reservations.VertexOutputStorageBufferBinding;
             BufferDefinition vertexOutputBuffer = new(BufferLayout.Std430, 1, vertexDataSbBinding, "vb_input", vertexInputStruct);
-            resourceManager.Properties.AddOrUpdateStorageBuffer(vertexOutputBuffer);
+            resourceManager.AddVertexAsComputeStorageBuffer(vertexOutputBuffer);
 
             EmitterContext context = new();
 
@@ -562,14 +573,14 @@ namespace Ryujinx.Graphics.Shader.Translation
                 LastInVertexPipeline = true
             };
 
-            return Generate(
+            return (Generate(
                 [function],
                 attributeUsage,
                 definitions,
                 definitions,
                 resourceManager,
                 FeatureFlags.None,
-                0);
+                0), resourceManager.GetVertexAsComputeInfo(isVertex: true));
         }
 
         public ShaderProgram GenerateGeometryPassthrough()

src/Ryujinx.Graphics.Vulkan/BackgroundResources.cs

@@ -36,6 +36,7 @@ namespace Ryujinx.Graphics.Vulkan
                         queueLock,
                         _gd.QueueFamilyIndex,
                         _gd.IsQualcommProprietary,
+                        _gd.IsTurnip,
                         isLight: true);
                 }
             }

src/Ryujinx.Graphics.Vulkan/CommandBufferPool.cs

@@ -19,6 +19,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Queue _queue;
         private readonly Lock _queueLock;
         private readonly bool _concurrentFenceWaitUnsupported;
+        private readonly bool _fenceAlwaysWaits;
         private readonly CommandPool _pool;
         private readonly Thread _owner;
 
@@ -66,6 +67,7 @@ namespace Ryujinx.Graphics.Vulkan
             Lock queueLock,
             uint queueFamilyIndex,
             bool concurrentFenceWaitUnsupported,
+            bool fenceAlwaysWaits,
             bool isLight = false)
         {
             _api = api;
@@ -73,6 +75,7 @@ namespace Ryujinx.Graphics.Vulkan
             _queue = queue;
             _queueLock = queueLock;
             _concurrentFenceWaitUnsupported = concurrentFenceWaitUnsupported;
+            _fenceAlwaysWaits = fenceAlwaysWaits;
             _owner = Thread.CurrentThread;
 
             CommandPoolCreateInfo commandPoolCreateInfo = new()
@@ -207,7 +210,7 @@ namespace Ryujinx.Graphics.Vulkan
 
                 ref ReservedCommandBuffer entry = ref _commandBuffers[index];
 
-                if (wait || !entry.InConsumption || entry.Fence.IsSignaled())
+                if (wait || !entry.InConsumption || entry.Fence.IsSignaledLazy())
                 {
                     WaitAndDecrementRef(index);
 
@@ -349,7 +352,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (refreshFence)
             {
-                entry.Fence = new FenceHolder(_api, _device, _concurrentFenceWaitUnsupported);
+                entry.Fence = new FenceHolder(_api, _device, _concurrentFenceWaitUnsupported, _fenceAlwaysWaits);
             }
             else
             {

src/Ryujinx.Graphics.Vulkan/FenceHolder.cs

@@ -12,13 +12,15 @@ namespace Ryujinx.Graphics.Vulkan
         private int _referenceCount;
         private int _lock;
         private readonly bool _concurrentWaitUnsupported;
+        private readonly bool _alwaysWaits;
         private bool _disposed;
 
-        public unsafe FenceHolder(Vk api, Device device, bool concurrentWaitUnsupported)
+        public unsafe FenceHolder(Vk api, Device device, bool concurrentWaitUnsupported, bool alwaysWaits)
         {
             _api = api;
             _device = device;
             _concurrentWaitUnsupported = concurrentWaitUnsupported;
+            _alwaysWaits = alwaysWaits;
 
             FenceCreateInfo fenceCreateInfo = new()
             {
@@ -123,6 +125,16 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        public bool IsSignaledLazy()
+        {
+            if (_alwaysWaits)
+            {
+                return false;
+            }
+
+            return IsSignaled();
+        }
+
         public bool IsSignaled()
         {
             if (_concurrentWaitUnsupported)

src/Ryujinx.Graphics.Vulkan/FramebufferParams.cs

@@ -32,6 +32,29 @@ namespace Ryujinx.Graphics.Vulkan
         public bool HasDepthStencil { get; private set; }
         public int ColorAttachmentsCount => AttachmentsCount - (HasDepthStencil ? 1 : 0);
 
+        public bool SetVirtualSize(uint width, uint height, uint layers)
+        {
+            if (AttachmentsCount != 0)
+            {
+                return false;
+            }
+
+            width = Math.Max(1u, width);
+            height = Math.Max(1u, height);
+            layers = Math.Max(1u, layers);
+
+            if (Width == width && Height == height && Layers == layers)
+            {
+                return false;
+            }
+
+            Width = width;
+            Height = height;
+            Layers = layers;
+
+            return true;
+        }
+
         public FramebufferParams(Device device, TextureView view, uint width, uint height)
         {
             Format format = view.Info.Format;

src/Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -406,10 +406,10 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (dstIsDepthOrStencil)
             {
-                _pipeline.SetProgram(src.Info.Target.IsMultisample() ? _programDepthBlitMs : _programDepthBlit);
+                _pipeline.SetProgram(src.Info.Target.IsMultisample ? _programDepthBlitMs : _programDepthBlit);
                 _pipeline.SetDepthTest(new DepthTestDescriptor(true, true, CompareOp.Always));
             }
-            else if (src.Info.Target.IsMultisample())
+            else if (src.Info.Target.IsMultisample)
             {
                 _pipeline.SetProgram(_programColorBlitMs);
             }
@@ -566,12 +566,12 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (isDepth)
             {
-                _pipeline.SetProgram(src.Info.Target.IsMultisample() ? _programDepthBlitMs : _programDepthBlit);
+                _pipeline.SetProgram(src.Info.Target.IsMultisample ? _programDepthBlitMs : _programDepthBlit);
                 _pipeline.SetDepthTest(new DepthTestDescriptor(true, true, CompareOp.Always));
             }
             else
             {
-                _pipeline.SetProgram(src.Info.Target.IsMultisample() ? _programStencilBlitMs : _programStencilBlit);
+                _pipeline.SetProgram(src.Info.Target.IsMultisample ? _programStencilBlitMs : _programStencilBlit);
                 _pipeline.SetStencilTest(CreateStencilTestDescriptor(true));
             }
 

src/Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -843,6 +843,11 @@ namespace Ryujinx.Graphics.Vulkan
         public void SetImage(ShaderStage stage, int binding, ITexture image)
         {
             _descriptorSetUpdater.SetImage(Cbs, stage, binding, image);
+
+            if (stage == ShaderStage.Fragment && image is TextureView view)
+            {
+                FramebufferParams?.SetVirtualSize((uint)view.Width, (uint)view.Height, (uint)view.Layers);
+            }
         }
 
         public void SetImage(int binding, Auto<DisposableImageView> image)
@@ -1607,6 +1612,24 @@ namespace Ryujinx.Graphics.Vulkan
 
         private bool RecreateGraphicsPipelineIfNeeded()
         {
+            if (FramebufferParams != null &&
+                FramebufferParams.AttachmentsCount == 0 &&
+                FramebufferParams.Width == 1 &&
+                FramebufferParams.Height == 1 &&
+                DynamicState.ViewportsCount != 0)
+            {
+                // An attachmentless fragment pass can reach its first draw before the storage
+                // image descriptor is rebound. At that point the null framebuffer still has
+                // its constructor fallback of 1x1, even though the guest viewport already
+                // describes the real render area. Seed the virtual framebuffer from that
+                // viewport so the first storage-image draw is not clipped to one pixel.
+                Silk.NET.Vulkan.Viewport viewport = DynamicState.Viewports[0];
+                uint width = (uint)Math.Max(1f, Math.Abs(viewport.Width));
+                uint height = (uint)Math.Max(1f, Math.Abs(viewport.Height));
+
+                FramebufferParams.SetVirtualSize(width, height, 1);
+            }
+
             if (AutoFlush.ShouldFlushDraw(DrawCount))
             {
                 Gd.FlushAllCommands();

src/Ryujinx.Graphics.Vulkan/ShaderCollection.cs

@@ -60,7 +60,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         private ProgramPipelineState _state;
         private DisposableRenderPass _dummyRenderPass;
-        private readonly Task _compileTask;
+        private ShaderCompilationRequest _compileRequest;
         private bool _firstBackgroundUse;
 
         public ShaderCollection(
@@ -139,7 +139,7 @@ namespace Ryujinx.Graphics.Vulkan
             // Updating buffer texture bindings using template updates crashes the Adreno driver on Windows.
             UpdateTexturesWithoutTemplate = gd.IsQualcommProprietary && usesBufferTextures;
 
-            _compileTask = Task.CompletedTask;
+            _compileRequest = new ShaderCompilationRequest(Task.CompletedTask);
             _firstBackgroundUse = false;
         }
 
@@ -153,7 +153,9 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _state = state;
 
-            _compileTask = BackgroundCompilation();
+            _compileRequest = gd.ShaderCompilationQueue != null
+                ? gd.ShaderCompilationQueue.Add(BackgroundCompilation)
+                : new ShaderCompilationRequest(BackgroundCompilationAsync());
             _firstBackgroundUse = !fromCache;
         }
 
@@ -458,10 +460,25 @@ namespace Ryujinx.Graphics.Vulkan
             return (buffer, texture);
         }
 
-        private async Task BackgroundCompilation()
+        private async Task BackgroundCompilationAsync()
         {
             await Task.WhenAll(_shaders.Select(shader => shader.CompileTask));
 
+            BackgroundCompilationImpl();
+        }
+
+        private void BackgroundCompilation()
+        {
+            foreach (var shader in _shaders)
+            {
+                shader.CompileTask.Wait();
+            }
+
+            BackgroundCompilationImpl();
+        }
+
+        private void BackgroundCompilationImpl()
+        {
             if (Array.Exists(_shaders, shader => shader.CompileStatus == ProgramLinkStatus.Failure))
             {
                 LinkStatus = ProgramLinkStatus.Failure;
@@ -604,11 +621,11 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                 }
 
-                if (!_compileTask.IsCompleted)
+                if (!_compileRequest.IsCompleted)
                 {
                     if (blocking)
                     {
-                        _compileTask.Wait();
+                        _compileRequest.Wait();
 
                         if (LinkStatus == ProgramLinkStatus.Failure)
                         {

src/Ryujinx.Graphics.Vulkan/ShaderCompilationQueue.cs

@@ -0,0 +1,131 @@
+using System;
+using System.Collections.Concurrent;
+using System.Threading;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    class ShaderCompilationQueue
+    {
+        private const int MaxParallelCompilations = 8;
+        private const int MaxThreadStackSize = 2 * 1024 * 1024; // MB
+
+        private struct Request
+        {
+            public readonly ulong Id;
+            public readonly Action Callback;
+
+            public Request(ulong id, Action callback)
+            {
+                Id = id;
+                Callback = callback;
+            }
+        }
+
+        private readonly Thread[] _workerThreads;
+        private readonly CancellationTokenSource _cts;
+        private readonly BlockingCollection<Request>[] _queues;
+        private readonly ulong[] _finishedIds;
+        private ulong _currentId;
+        private int _currentQueueIndex;
+
+        public ShaderCompilationQueue()
+        {
+            _workerThreads = new Thread[MaxParallelCompilations];
+            _queues = new BlockingCollection<Request>[MaxParallelCompilations];
+            _finishedIds = new ulong[MaxParallelCompilations];
+
+            _cts = new CancellationTokenSource();
+
+            for (int i = 0; i < MaxParallelCompilations; i++)
+            {
+                _queues[i] = new BlockingCollection<Request>();
+
+                Thread thread = new Thread(DoWork, MaxThreadStackSize) { Name = $"BackgroundShaderCompiler.{i}" };
+                thread.IsBackground = true;
+                thread.Start(i);
+
+                _workerThreads[i] = thread;
+            }
+        }
+
+        private void DoWork(object threadId)
+        {
+            int queueIndex = (int)threadId;
+
+            try
+            {
+                var queue = _queues[queueIndex];
+
+                foreach (var request in queue.GetConsumingEnumerable(_cts.Token))
+                {
+                    request.Callback();
+
+                    lock (queue)
+                    {
+                        _finishedIds[queueIndex] = request.Id;
+
+                        Monitor.PulseAll(queue);
+                    }
+                }
+            }
+            catch (OperationCanceledException)
+            {
+            }
+        }
+
+        public ShaderCompilationRequest Add(Action callback)
+        {
+            ulong newId = Interlocked.Increment(ref _currentId);
+
+            // Let's keep rotating between the queues to increase the chances
+            // that the selected queue thread is currently idle.
+            int queueIndex = Interlocked.Increment(ref _currentQueueIndex) % MaxParallelCompilations;
+
+            _queues[queueIndex].Add(new Request(newId, callback));
+
+            return new ShaderCompilationRequest(this, queueIndex, newId);
+        }
+
+        public void Wait(int queueIndex, ulong id)
+        {
+            var queue = _queues[queueIndex];
+
+            lock (queue)
+            {
+                while (_finishedIds[queueIndex] < id)
+                {
+                    Monitor.Wait(queue);
+                }
+            }
+        }
+
+        public bool IsCompleted(int queueIndex, ulong id)
+        {
+            var queue = _queues[queueIndex];
+
+            lock (queue)
+            {
+                return _finishedIds[queueIndex] >= id;
+            }
+        }
+
+        public void Dispose()
+        {
+            for (int i = 0; i < MaxParallelCompilations; i++)
+            {
+                _queues[i].CompleteAdding();
+            }
+
+            _cts.Cancel();
+
+            for (int i = 0; i < MaxParallelCompilations; i++)
+            {
+                _workerThreads[i].Join();
+
+                _queues[i].Dispose();
+            }
+
+            _cts.Dispose();
+        }
+    }
+}

src/Ryujinx.Graphics.Vulkan/ShaderCompilationRequest.cs

@@ -0,0 +1,55 @@
+using System.Threading.Tasks;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    struct ShaderCompilationRequest
+    {
+        private readonly Task _task;
+        private readonly ShaderCompilationQueue _queue;
+        private readonly int _queueIndex;
+        private readonly ulong _requestId;
+
+        public bool IsCompleted
+        {
+            get
+            {
+                if (_task != null)
+                {
+                    return _task.IsCompleted;
+                }
+                else
+                {
+                    return _queue.IsCompleted(_queueIndex, _requestId);
+                }
+            }
+        }
+
+        public ShaderCompilationRequest(Task task)
+        {
+            _task = task;
+            _queue = null;
+            _queueIndex = 0;
+            _requestId = 0;
+        }
+
+        public ShaderCompilationRequest(ShaderCompilationQueue queue, int queueIndex, ulong requestId)
+        {
+            _task = null;
+            _queue = queue;
+            _queueIndex = queueIndex;
+            _requestId = requestId;
+        }
+
+        public void Wait()
+        {
+            if (_task != null)
+            {
+                _task.Wait();
+            }
+            else
+            {
+                _queue.Wait(_queueIndex, _requestId);
+            }
+        }
+    }
+}

src/Ryujinx.Graphics.Vulkan/StagingBuffer.cs

@@ -266,7 +266,7 @@ namespace Ryujinx.Graphics.Vulkan
         public void FreeCompleted()
         {
             FenceHolder signalledFence = null;
-            while (_pendingCopies.TryPeek(out PendingCopy pc) && pc.Fence != null && (pc.Fence == signalledFence || pc.Fence.IsSignaled()))
+            while (_pendingCopies.TryPeek(out PendingCopy pc) && pc.Fence != null && (pc.Fence == signalledFence || pc.Fence.IsSignaledLazy()))
             {
                 signalledFence = pc.Fence; // Already checked - don't need to do it again.
                 PendingCopy dequeued = _pendingCopies.Dequeue();

src/Ryujinx.Graphics.Vulkan/TextureStorage.cs

@@ -79,7 +79,7 @@ namespace Ryujinx.Graphics.Vulkan
             _device = device;
             _info = info;
 
-            bool isMsImageStorageSupported = gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample();
+            bool isMsImageStorageSupported = gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample;
 
             VkFormat format = _gd.FormatCapabilities.ConvertToVkFormat(info.Format, isMsImageStorageSupported);
             uint levels = (uint)info.Levels;
@@ -323,7 +323,7 @@ namespace Ryujinx.Graphics.Vulkan
                 usage |= ImageUsageFlags.ColorAttachmentBit;
             }
 
-            if (format.IsImageCompatible() && (isMsImageStorageSupported || !target.IsMultisample()))
+            if (format.IsImageCompatible() && (isMsImageStorageSupported || !target.IsMultisample))
             {
                 usage |= ImageUsageFlags.StorageBit;
             }

src/Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -61,7 +61,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.Textures.Add(this);
 
-            bool isMsImageStorageSupported = gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample();
+            bool isMsImageStorageSupported = gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample;
 
             VkFormat format = _gd.FormatCapabilities.ConvertToVkFormat(info.Format, isMsImageStorageSupported);
             ImageUsageFlags usage = TextureStorage.GetImageUsage(info.Format, info.Target, gd.Capabilities, isMsImageStorageSupported) & storage.UsageFlags;
@@ -126,7 +126,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             ImageUsageFlags shaderUsage = ImageUsageFlags.SampledBit;
 
-            if (info.Format.IsImageCompatible() && (_gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample()))
+            if (info.Format.IsImageCompatible() && (_gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample))
             {
                 shaderUsage |= ImageUsageFlags.StorageBit;
             }
@@ -225,12 +225,12 @@ namespace Ryujinx.Graphics.Vulkan
             Image srcImage = src.GetImage().Get(cbs).Value;
             Image dstImage = dst.GetImage().Get(cbs).Value;
 
-            if (!dst.Info.Target.IsMultisample() && Info.Target.IsMultisample())
+            if (!dst.Info.Target.IsMultisample && Info.Target.IsMultisample)
             {
                 int layers = Math.Min(Info.GetLayers(), dst.Info.GetLayers() - firstLayer);
                 _gd.HelperShader.CopyMSToNonMS(_gd, cbs, src, dst, 0, firstLayer, layers);
             }
-            else if (dst.Info.Target.IsMultisample() && !Info.Target.IsMultisample())
+            else if (dst.Info.Target.IsMultisample && !Info.Target.IsMultisample)
             {
                 int layers = Math.Min(Info.GetLayers(), dst.Info.GetLayers() - firstLayer);
                 _gd.HelperShader.CopyNonMSToMS(_gd, cbs, src, dst, 0, firstLayer, layers);
@@ -287,11 +287,11 @@ namespace Ryujinx.Graphics.Vulkan
             Image srcImage = src.GetImage().Get(cbs).Value;
             Image dstImage = dst.GetImage().Get(cbs).Value;
 
-            if (!dst.Info.Target.IsMultisample() && Info.Target.IsMultisample())
+            if (!dst.Info.Target.IsMultisample && Info.Target.IsMultisample)
             {
                 _gd.HelperShader.CopyMSToNonMS(_gd, cbs, src, dst, srcLayer, dstLayer, 1);
             }
-            else if (dst.Info.Target.IsMultisample() && !Info.Target.IsMultisample())
+            else if (dst.Info.Target.IsMultisample && !Info.Target.IsMultisample)
             {
                 _gd.HelperShader.CopyNonMSToMS(_gd, cbs, src, dst, srcLayer, dstLayer, 1);
             }

src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -55,6 +55,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal CommandBufferPool CommandBufferPool { get; private set; }
         internal PipelineLayoutCache PipelineLayoutCache { get; private set; }
         internal BackgroundResources BackgroundResources { get; private set; }
+        internal ShaderCompilationQueue ShaderCompilationQueue { get; private set; }
         internal Action<Action> InterruptAction { get; private set; }
         internal SyncManager SyncManager { get; private set; }
 
@@ -96,6 +97,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal bool IsNvidiaPreTuring { get; private set; }
         internal bool IsIntelArc { get; private set; }
         internal bool IsQualcommProprietary { get; private set; }
+        internal bool IsTurnip { get; private set; }
         internal bool IsMoltenVk { get; private set; }
         internal bool SupportsMTL31 { get; private set; }
         internal bool IsTBDR { get; private set; }
@@ -127,6 +129,12 @@ namespace Ryujinx.Graphics.Vulkan
 
                 // Any device running on MacOS is using MoltenVK, even Intel and AMD vendors.
                 IsMoltenVk = true;
+
+                // The default thread stack size on MacOS is low, and can cause stack overflow
+                // on SPIR-V Cross during shader compilation.
+                // As a workaround, we use this custom queue which allows us to specify the stack
+                // size of the threads used for compilation.
+                ShaderCompilationQueue = new ShaderCompilationQueue();
             }
 
             SupportsMTL31 = OperatingSystem.IsMacOSVersionAtLeast(14);
@@ -396,6 +404,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             IsFeedbackLoopDevice = IsAmdRdna3;
 
+            IsTurnip = GpuRenderer.StartsWith("Turnip");
+
             if (Vendor == Vendor.Nvidia)
             {
                 Match match = VendorUtils.NvidiaConsumerClassRegex().Match(GpuRenderer);
@@ -476,7 +486,7 @@ namespace Ryujinx.Graphics.Vulkan
             Api.TryGetDeviceExtension(_instance.Instance, _device, out ExtExternalMemoryHost hostMemoryApi);
             HostMemoryAllocator = new HostMemoryAllocator(MemoryAllocator, Api, hostMemoryApi, _device);
 
-            CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex, IsQualcommProprietary);
+            CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex, IsQualcommProprietary, IsTurnip);
 
             PipelineLayoutCache = new PipelineLayoutCache();
 
@@ -777,6 +787,7 @@ namespace Ryujinx.Graphics.Vulkan
                 supportsGeometryShader: Capabilities.SupportsGeometryShader,
                 supportsGeometryShaderPassthrough: Capabilities.SupportsGeometryShaderPassthrough,
                 supportsTransformFeedback: Capabilities.SupportsTransformFeedback,
+                supportsImageBufferPixelAlignment: false,
                 supportsImageLoadFormatted: features2.Features.ShaderStorageImageReadWithoutFormat,
                 supportsLayerVertexTessellation: featuresVk12.ShaderOutputLayer,
                 supportsMismatchingViewFormat: true,
@@ -790,6 +801,7 @@ namespace Ryujinx.Graphics.Vulkan
                 supportsShaderNonUniformIndexing:
                     featuresVk12.ShaderSampledImageArrayNonUniformIndexing &&
                     featuresVk12.ShaderStorageImageArrayNonUniformIndexing,
+                supportsTextureBufferPixelAlignment: false,
                 supportsTextureGatherOffsets: features2.Features.ShaderImageGatherExtended,
                 supportsTextureShadowLod: false,
                 supportsVertexStoreAndAtomics: features2.Features.VertexPipelineStoresAndAtomics,
@@ -1112,6 +1124,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             SurfaceApi.DestroySurface(_instance.Instance, _surface, null);
 
+            ShaderCompilationQueue?.Dispose();
+
             Api.DestroyDevice(_device, null);
 
             _debugMessenger.Dispose();

src/Ryujinx.HLE/HOS/ArmProcessContext.cs

@@ -1,4 +1,5 @@
 using ARMeilleure.Memory;
+using ARMeilleure.Translation.PTC;
 using Ryujinx.Cpu;
 using Ryujinx.Graphics.Gpu;
 using Ryujinx.HLE.HOS.Kernel.Process;
@@ -9,12 +10,10 @@ namespace Ryujinx.HLE.HOS
     interface IArmProcessContext : IProcessContext
     {
         IDiskCacheLoadState Initialize(
-            string titleIdText,
+            PtcCacheInfo cacheInfo,
-            string displayVersion,
             bool diskCacheEnabled,
             ulong codeAddress,
-            ulong codeSize,
+            ulong codeSize);
-            string cacheSelector);
     }
 
     class ArmProcessContext<T> : IArmProcessContext where T : class, IVirtualMemoryManagerTracked, IMemoryManager
@@ -64,15 +63,13 @@ namespace Ryujinx.HLE.HOS
         }
 
         public IDiskCacheLoadState Initialize(
-            string titleIdText,
+            PtcCacheInfo cacheInfo,
-            string displayVersion,
             bool diskCacheEnabled,
             ulong codeAddress,
-            ulong codeSize,
+            ulong codeSize)
-            string cacheSelector)
         {
             _cpuContext.PrepareCodeRange(codeAddress, codeSize);
-            return _cpuContext.LoadDiskCache(titleIdText, displayVersion, diskCacheEnabled, cacheSelector);
+            return _cpuContext.LoadDiskCache(cacheInfo, diskCacheEnabled);
         }
 
         public void InvalidateCacheRegion(ulong address, ulong size)

src/Ryujinx.HLE/HOS/ArmProcessContextFactory.cs

@@ -1,3 +1,4 @@
+using ARMeilleure.Translation.PTC;
 using Ryujinx.Common.Configuration;
 using Ryujinx.Common.Logging;
 using Ryujinx.Cpu;
@@ -7,6 +8,7 @@ using Ryujinx.Cpu.LightningJit;
 using Ryujinx.Graphics.Gpu;
 using Ryujinx.HLE.HOS.Kernel;
 using Ryujinx.HLE.HOS.Kernel.Process;
+using Ryujinx.HLE.Loaders.Processes;
 using Ryujinx.Memory;
 using System;
 using System.Runtime.InteropServices;
@@ -17,8 +19,10 @@ namespace Ryujinx.HLE.HOS
     {
         private readonly ITickSource _tickSource;
         private readonly GpuContext _gpu;
-        private readonly string _titleIdText;
+        private readonly ulong _programId;
+        private readonly byte _programIndex;
         private readonly string _displayVersion;
+        private readonly ProcessKind _processKind;
         private readonly bool _diskCacheEnabled;
         private readonly string _diskCacheSelector;
         private readonly ulong _codeAddress;
@@ -29,8 +33,10 @@ namespace Ryujinx.HLE.HOS
         public ArmProcessContextFactory(
             ITickSource tickSource,
             GpuContext gpu,
-            string titleIdText,
+            ulong programId,
+            byte programIndex,
             string displayVersion,
+            ProcessKind processKind,
             bool diskCacheEnabled,
             string diskCacheSelector,
             ulong codeAddress,
@@ -38,8 +44,10 @@ namespace Ryujinx.HLE.HOS
         {
             _tickSource = tickSource;
             _gpu = gpu;
-            _titleIdText = titleIdText;
+            _programId = programId;
+            _programIndex = programIndex;
             _displayVersion = displayVersion;
+            _processKind = processKind;
             _diskCacheEnabled = diskCacheEnabled;
             _diskCacheSelector = diskCacheSelector;
             _codeAddress = codeAddress;
@@ -121,8 +129,18 @@ namespace Ryujinx.HLE.HOS
             }
 
             string cacheSelector = _diskCacheSelector ?? "default";
+            string programIdText = _programId == 0 ? string.Empty : $"{_programId:x16}";
+            string applicationIdText = _programId == 0 ? string.Empty : $"{_programId & ~0xFul:x16}";
+            PtcCacheInfo cacheInfo = new(
+                pid,
+                programIdText,
+                applicationIdText,
+                _programIndex,
+                _displayVersion,
+                _processKind.ToString(),
+                cacheSelector);
 
-            DiskCacheLoadState = processContext.Initialize(_titleIdText, _displayVersion, _diskCacheEnabled, _codeAddress, _codeSize, cacheSelector);
+            DiskCacheLoadState = processContext.Initialize(cacheInfo, _diskCacheEnabled, _codeAddress, _codeSize);
 
             return processContext;
         }

src/Ryujinx.HLE/HOS/Services/Am/AppletAE/AllSystemAppletProxiesService/SystemAppletProxy/ICommonStateGetter.cs

@@ -12,12 +12,23 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
 {
     class ICommonStateGetter : DisposableIpcService
     {
+        // Nintendo serves the VR goggle commands based on Title ID, apparently.
+        private static ReadOnlySpan<ulong> _supportedVRGoggleTitles =>
+        [
+            0x0100CA001D972000, // Nintendo Classics: Virtual Boy NSO
+            0x0100BFC01D976000  // Nintendo Classics: Virtual Boy NSO
+        ];
+        
+        private int _resolutionWidth = 1280;
+        private int _resolutionHeight = 720;
+        
         private readonly ServiceCtx _context;
 
         private readonly Apm.ManagerServer _apmManagerServer;
         private readonly Apm.SystemManagerServer _apmSystemManagerServer;
 
         private bool _vrModeEnabled;
+        private bool _vrMode3dEnabled;
 #pragma warning disable CS0414, IDE0052 // Remove unread private member
         private bool _lcdBacklighOffEnabled;
         private bool _requestExitToLibraryAppletAtExecuteNextProgramEnabled;
@@ -153,6 +164,47 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
             return ResultCode.Success;
         }
 
+        [CommandCmif(30)]
+        // GetHomeButtonReaderLockAccessor() -> nn::am::service::ILockAccessor
+        public ResultCode GetHomeButtonReaderLockAccessor(ServiceCtx context)
+        {
+            // We currently do not have any home button functionality, so it's fine to stub this for now.
+            // Similar to using GetReaderLockAccessorEx() with inval=0.
+            Logger.Stub?.PrintStub(LogClass.ServiceAm);
+            MakeObject(context, new ILockAccessor(context));
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(31)] // 2.0.0 +
+        // GetReaderLockAccessorEx(uint unknown) -> nn::am::service::ILockAccessor
+        public ResultCode GetReaderLockAccessorEx(ServiceCtx context)
+        {
+            uint unknown = context.RequestData.ReadUInt32();
+            Logger.Stub?.PrintStub(LogClass.ServiceAm, new { unknown });
+            if (unknown < 0 || unknown > 3)
+            {
+                throw new ArgumentOutOfRangeException(nameof(unknown));
+            }
+            
+            MakeObject(context, new ILockAccessor(context));
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(32)] // 2.0.0+
+        // GetWriterLockAccessorEx(uint unknown) -> nn::am::service::ILockAccessor
+        public ResultCode GetWriterLockAccessorEx(ServiceCtx context)
+        {
+            uint unknown = context.RequestData.ReadUInt32();
+            Logger.Stub?.PrintStub(LogClass.ServiceAm, new { unknown });
+            if (unknown < 0 || unknown > 3)
+            {
+                throw new ArgumentOutOfRangeException(nameof(unknown));
+            }
+            
+            MakeObject(context, new ILockAccessor(context));
+            return ResultCode.Success;
+        }
+
         [CommandCmif(50)] // 3.0.0+
         // IsVrModeEnabled() -> b8
         public ResultCode IsVrModeEnabled(ServiceCtx context)
@@ -177,8 +229,8 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
         // SetLcdBacklighOffEnabled(b8)
         public ResultCode SetLcdBacklighOffEnabled(ServiceCtx context)
         {
-            // NOTE: Service sets a private field here, maybe this field is used somewhere else to turned off the backlight.
+            // NOTE: Service sets a private field here, maybe this field is used somewhere else to turn off the backlight.
-            //       Since we don't support backlight, it's fine to do nothing.
+            //       Since we don't support the backlight feature, it's fine to stub it.
 
             _lcdBacklighOffEnabled = context.RequestData.ReadBoolean();
 
@@ -298,18 +350,43 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
             return (ResultCode)_apmSystemManagerServer.GetCurrentPerformanceConfiguration(context);
         }
 
+        [CommandCmif(130)] // 21.0.0+
+        // EnableStartupLogoDisappearedMessage()
+        public ResultCode EnableStartupLogoDisappearedMessage(ServiceCtx context)
+        {
+            // NOTE: Service only delivers the message once the startup logo has actually
+            //       disappeared. We never display one, so it is already deliverable. Callers
+            //       gate their first frame on this message and hang without it.
+
+            AppletStateMgr appletState = context.Device.System.AppletState;
+
+            appletState.Messages.Enqueue(AppletMessage.StartupLogoDisappeared);
+            appletState.MessageEvent.ReadableEvent.Signal();
+
+            return ResultCode.Success;
+        }
+
         [CommandCmif(300)] // 9.0.0+
         // GetSettingsPlatformRegion() -> u8
         public ResultCode GetSettingsPlatformRegion(ServiceCtx context)
         {
             PlatformRegion platformRegion = context.Device.System.State.DesiredRegionCode == (uint)RegionCode.China ? PlatformRegion.China : PlatformRegion.Global;
-
+            
             // FIXME: Call set:sys GetPlatformRegion
             context.ResponseData.Write((byte)platformRegion);
 
             return ResultCode.Success;
         }
 
+        [CommandCmif(610)] // 21.0.0+
+        // UnknownCommand610(long unknown)
+        public ResultCode UnknownCommand610(ServiceCtx context)
+        {
+            long unknown = context.RequestData.ReadInt64();
+            Logger.Stub?.PrintStub(LogClass.ServiceAm, new { unknown });
+            return ResultCode.Success;
+        }
+
         [CommandCmif(900)] // 11.0.0+
         // SetRequestExitToLibraryAppletAtExecuteNextProgramEnabled()
         public ResultCode SetRequestExitToLibraryAppletAtExecuteNextProgramEnabled(ServiceCtx context)
@@ -320,6 +397,100 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
             return ResultCode.Success;
         }
 
+        [CommandCmif(1000)] // 19.0.0+
+        // BeginVrMode3d()
+        public ResultCode BeginVrMode3d(ServiceCtx context)
+        {
+            // NOTE: Service also applies a stereo scale to the goggle display, which we don't
+            //       model, so only the state is kept here.
+
+            _vrMode3dEnabled = true;
+
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(1001)] // 19.0.0+
+        // EndVrMode3d()
+        public ResultCode EndVrMode3d(ServiceCtx context)
+        {
+            _vrMode3dEnabled = false;
+
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(1002)] // 19.0.0+
+        // IsVrModeEnabled3d() -> b8
+        public ResultCode IsVrModeEnabled3d(ServiceCtx context)
+        {
+            context.ResponseData.Write(_vrMode3dEnabled);
+
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(1003)] // 21.0.0+
+        // GetVrLaboGoggleViewport() -> (s32 x, s32 y, s32 width, s32 height)
+        public ResultCode GetVrLaboGoggleViewport(ServiceCtx context)
+        {
+            if (!IsVrLaboGoggleSupportedTitle(context))
+            {
+                return ResultCode.ObjectInvalid;
+            }
+            
+            int VrDisplayCoordinateX = 0;
+            int VrDisplayCoordinateY = 0;
+            int VrDisplayWidth = _resolutionWidth;
+            int VrDisplayHeight = _resolutionHeight;
+
+            context.ResponseData.Write(VrDisplayCoordinateX);
+            context.ResponseData.Write(VrDisplayCoordinateY);
+            context.ResponseData.Write(VrDisplayWidth);
+            context.ResponseData.Write(VrDisplayHeight);
+
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(1004)] // 21.0.0+
+        // GetPanelPhysicalSizeForSpecificTitle() -> (f32 width, f32 height)
+        public ResultCode GetPanelPhysicalSizeForSpecificTitle(ServiceCtx context)
+        {
+            if (!IsVrLaboGoggleSupportedTitle(context))
+            {
+                return ResultCode.ObjectInvalid;
+            }
+            
+            // The switch provides micrometres, but the command reports millimetres.
+            // 6.2 inch 16:9 panel.
+            float PanelPhysicalWidthMicroMeters = 137250f;
+            float PanelPhysicalHeightMicroMeters = 77200f;
+            
+            context.ResponseData.Write(PanelPhysicalWidthMicroMeters / 1000f);
+            context.ResponseData.Write(PanelPhysicalHeightMicroMeters / 1000f);
+
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(1005)] // 21.0.0+
+        // GetPanelResolutionForSpecificTitle() -> (s32 width, s32 height)
+        public ResultCode GetPanelResolutionForSpecificTitle(ServiceCtx context)
+        {
+            if (!IsVrLaboGoggleSupportedTitle(context))
+            {
+                return ResultCode.ObjectInvalid;
+            }
+            
+            context.ResponseData.Write(_resolutionWidth);
+            context.ResponseData.Write(_resolutionHeight);
+
+            return ResultCode.Success;
+        }
+
+        private static bool IsVrLaboGoggleSupportedTitle(ServiceCtx context)
+        {
+            ulong programId = context.Device.Processes.ActiveApplication.ProgramId;
+
+            return _supportedVRGoggleTitles.Contains(programId);
+        }
+
         protected override void Dispose(bool isDisposing)
         {
             if (isDisposing)

src/Ryujinx.HLE/HOS/Services/Am/AppletAE/AllSystemAppletProxiesService/SystemAppletProxy/ILockAccessor.cs

@@ -0,0 +1,80 @@
+using Ryujinx.Common.Logging;
+using Ryujinx.HLE.HOS.Ipc;
+using Ryujinx.HLE.HOS.Kernel.Threading;
+using Ryujinx.Horizon.Common;
+using System;
+
+namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.SystemAppletProxy
+{
+    class ILockAccessor : IpcService
+    {
+        private readonly ServiceCtx _context;
+
+        private readonly Apm.ManagerServer _apmManagerServer;
+        private readonly Apm.SystemManagerServer _apmSystemManagerServer;
+        
+        private readonly KEvent _lockEvent;
+        private int _lockEventHandle;
+
+        public ILockAccessor(ServiceCtx context)
+        {
+            
+            _context = context;
+            
+            _apmManagerServer = new Apm.ManagerServer(context);
+            _apmSystemManagerServer = new Apm.SystemManagerServer(context);
+
+            _lockEvent = new KEvent(context.Device.System.KernelContext);
+        }
+
+        [CommandCmif(1)]
+        // TryLock(unknown u8 bool) -> unknown u8 bool
+        public ResultCode TryLock(ServiceCtx context)
+        {
+            // Official sw only uses inflag=false.
+            // Official sw just closes the output handle.
+            // The input flag controls whether this returns the output handle. 
+            bool unknown = context.RequestData.ReadBoolean();
+            Logger.Stub?.PrintStub(LogClass.ServiceAm, new { unknown });
+            
+            context.ResponseData.Write(false);
+            context.Response.HandleDesc = IpcHandleDesc.MakeCopy(_lockEventHandle);
+            
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(2)]
+        // Unlock()
+        public ResultCode Unlock(ServiceCtx context)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceAm);
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(3)]
+        // GetEvent() -> EventHandle w/ autoclear = false
+        public ResultCode GetEvent(ServiceCtx context)
+        {
+            if (_lockEventHandle == 0)
+            {
+                if (context.Process.HandleTable.GenerateHandle(_lockEvent.ReadableEvent, out _lockEventHandle) != Result.Success)
+                {
+                    throw new InvalidOperationException("Out of handles!");
+                }
+            }
+            
+            context.Response.HandleDesc = IpcHandleDesc.MakeCopy(_lockEventHandle);
+            Logger.Stub?.PrintStub(LogClass.ServiceAm);
+            return ResultCode.Success;
+        }
+
+        [CommandCmif(4)] // 10.0.0+
+        // IsLocked() -> unknown u8 bool
+        public ResultCode IsLocked(ServiceCtx context)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceAm);
+            context.ResponseData.Write(false);
+            return ResultCode.Success;
+        }
+    }
+}

src/Ryujinx.HLE/HOS/Services/Am/AppletAE/AllSystemAppletProxiesService/SystemAppletProxy/Types/AppletMessage.cs

@@ -32,5 +32,6 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.AllSystemAppletProxiesService.Sys
         DetectShortPressingCaptureButton = 90,
         AlbumScreenShotTaken = 92,
         AlbumRecordingSaved = 93,
+        StartupLogoDisappeared = 95, // 21.0.0+
     }
 }

src/Ryujinx.HLE/Loaders/Processes/ProcessLoaderHelper.cs

@@ -182,8 +182,10 @@ namespace Ryujinx.HLE.Loaders.Processes
             ArmProcessContextFactory processContextFactory = new(
                 context.Device.System.TickSource,
                 context.Device.Gpu,
-                string.Empty,
+                kip.ProgramId,
-                string.Empty,
+                0,
+                kip.Version.ToString(),
+                ProcessResult.GetProcessKind(kip.ProgramId),
                 false,
                 null,
                 codeAddress,
@@ -377,8 +379,10 @@ namespace Ryujinx.HLE.Loaders.Processes
             ArmProcessContextFactory processContextFactory = new(
                 context.Device.System.TickSource,
                 context.Device.Gpu,
-                $"{programId:x16}",
+                programId,
+                programIndex,
                 displayVersion,
+                ProcessResult.GetProcessKind(programId),
                 diskCacheEnabled,
                 diskCacheSelector,
                 codeStart,

src/Ryujinx.HLE/Loaders/Processes/ProcessResult.cs

@@ -87,7 +87,7 @@ namespace Ryujinx.HLE.Loaders.Processes
             AllowCodeMemoryForJit = allowCodeMemoryForJit;
         }
 
-        private static ProcessKind GetProcessKind(ulong programId)
+        internal static ProcessKind GetProcessKind(ulong programId)
         {
             if (programId == 0)
             {

src/Ryujinx.ShaderTools/Program.cs

@@ -116,7 +116,7 @@ namespace Ryujinx.ShaderTools
 
             if (options.VertexPassthrough)
             {
-                program = translatorContext.GenerateVertexPassthroughForCompute();
+                (program, _) = translatorContext.GenerateVertexPassthroughForCompute();
             }
             else
             {

src/Ryujinx/AppHost.cs

@@ -45,6 +45,7 @@ using Ryujinx.UI.Common.Configuration;
 using Ryujinx.UI.Common.Helper;
 using Silk.NET.Vulkan;
 using SkiaSharp;
+using SPB.Graphics.Exceptions;
 using SPB.Graphics.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -654,8 +655,22 @@ namespace Ryujinx.Ava
 
             if (RendererHost.EmbeddedWindow is EmbeddedWindowOpenGL openGlWindow)
             {
-                // Try to bind the OpenGL context before calling the shutdown event.
+                try
-                openGlWindow.MakeCurrent(false, false);
+                {
+                    // Try to bind the OpenGL context before disposing GPU resources.
+                    openGlWindow.MakeCurrent();
+                }
+                catch (ContextException e) when (_userChannelPersistence.ShouldRestart)
+                {
+                    // ExecuteProgram may detach the old native window before GPU
+                    // disposal. Allow the old context to be released with the window
+                    // and continue the requested program relaunch.
+                    Logger.Warning?.Print(
+                        LogClass.UI,
+                        $"Failed to bind OpenGL context during program relaunch: {e}");
+
+                    return;
+                }
 
                 Device.DisposeGpu();
 
@@ -1075,9 +1090,10 @@ namespace Ryujinx.Ava
         {
             Dispatcher.UIThread.InvokeAsync(() =>
             {
-                if (_viewModel.StartGamesInFullscreen)
+                if (_viewModel.StartGamesInFullscreen && _viewModel.WindowState is not WindowState.FullScreen)
                 {
-                    _viewModel.WindowState = WindowState.FullScreen;
+                    // Use the view model toggle so decoration ordering matches user toggles.
+                    _viewModel.ToggleFullscreen();
                 }
 
                 if (_viewModel.WindowState == WindowState.FullScreen || _viewModel.StartGamesWithoutUI)

src/Ryujinx/UI/Windows/XCITrimmerWindow.axaml.cs

@@ -6,6 +6,8 @@ using Ryujinx.Ava.Common.Locale;
 using Ryujinx.Ava.UI.ViewModels;
 using Ryujinx.UI.Common.Models;
 using System;
+using System.Collections.Generic;
+using System.Linq;
 using System.Threading.Tasks;
 
 namespace Ryujinx.Ava.UI.Windows
@@ -81,20 +83,17 @@ namespace Ryujinx.Ava.UI.Windows
 
         private void OnSelectionChanged(object sender, SelectionChangedEventArgs e)
         {
-            foreach (object content in e.AddedItems)
+            List<XCITrimmerFileModel> added = [.. e.AddedItems.OfType<XCITrimmerFileModel>()];
+            List<XCITrimmerFileModel> removed = [.. e.RemovedItems.OfType<XCITrimmerFileModel>()];
+
+            foreach (XCITrimmerFileModel applicationData in added)
             {
-                if (content is XCITrimmerFileModel applicationData)
+                ViewModel.Select(applicationData);
-                {
-                    ViewModel.Select(applicationData);
-                }
             }
 
-            foreach (object content in e.RemovedItems)
+            foreach (XCITrimmerFileModel applicationData in removed)
             {
-                if (content is XCITrimmerFileModel applicationData)
+                ViewModel.Deselect(applicationData);
-                {
-                    ViewModel.Deselect(applicationData);
-                }
             }
         }
     }