Force Vulkan swapchain re-creation when window size changes (#7003)

* Fix some validation errors

* Whitespace correction

* Resolve some runtime validation errors.

* Whitespace

* Properly fix usage realted validation error by setting Extended Usage image creation flag.

* Only if supported

* Remove checking extension for features that are core functionality of Vulkan 1.2

Directory.Packages.props

@@ -20,9 +20,9 @@
     <PackageVersion Include="LibHac" Version="0.19.0" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.9.2" />
-    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="7.6.0" />
+    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="7.6.2" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.9.0" />
-    <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.0" />
+    <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.1" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />
     <PackageVersion Include="NetCoreServer" Version="8.0.7" />
     <PackageVersion Include="NUnit" Version="3.13.3" />

src/ARMeilleure/CodeGen/RegisterAllocators/LinearScanAllocator.cs

@@ -251,7 +251,20 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                 }
             }
 
-            int selectedReg = GetHighestValueIndex(freePositions);
+            // If this is a copy destination variable, we prefer the register used for the copy source.
+            // If the register is available, then the copy can be eliminated later as both source
+            // and destination will use the same register.
+            int selectedReg;
+
+            if (current.TryGetCopySourceRegister(out int preferredReg) && freePositions[preferredReg] >= current.GetEnd())
+            {
+                selectedReg = preferredReg;
+            }
+            else
+            {
+                selectedReg = GetHighestValueIndex(freePositions);
+            }
+
             int selectedNextUse = freePositions[selectedReg];
 
             // Intervals starts and ends at odd positions, unless they span an entire
@@ -431,7 +444,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             }
         }
 
-        private static int GetHighestValueIndex(Span<int> span)
+        private static int GetHighestValueIndex(ReadOnlySpan<int> span)
         {
             int highest = int.MinValue;
 
@@ -798,12 +811,12 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             // The "visited" state is stored in the MSB of the local's value.
             const ulong VisitedMask = 1ul << 63;
 
-            bool IsVisited(Operand local)
+            static bool IsVisited(Operand local)
             {
                 return (local.GetValueUnsafe() & VisitedMask) != 0;
             }
 
-            void SetVisited(Operand local)
+            static void SetVisited(Operand local)
             {
                 local.GetValueUnsafe() |= VisitedMask;
             }
@@ -826,9 +839,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                         {
                             dest.NumberLocal(_intervals.Count);
 
-                            _intervals.Add(new LiveInterval(dest));
+                            LiveInterval interval = new LiveInterval(dest);
+                            _intervals.Add(interval);
 
                             SetVisited(dest);
+
+                            // If this is a copy (or copy-like operation), set the copy source interval as well.
+                            // This is used for register preferencing later on, which allows the copy to be eliminated
+                            // in some cases.
+                            if (node.Instruction == Instruction.Copy || node.Instruction == Instruction.ZeroExtend32)
+                            {
+                                Operand source = node.GetSource(0);
+
+                                if (source.Kind == OperandKind.LocalVariable &&
+                                    source.GetLocalNumber() > 0 &&
+                                    (node.Instruction == Instruction.Copy || source.Type == OperandType.I32))
+                                {
+                                    interval.SetCopySource(_intervals[source.GetLocalNumber()]);
+                                }
+                            }
                         }
                     }
                 }

src/ARMeilleure/CodeGen/RegisterAllocators/LiveInterval.cs

@@ -19,6 +19,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             public LiveRange CurrRange;
 
             public LiveInterval Parent;
+            public LiveInterval CopySource;
 
             public UseList Uses;
             public LiveIntervalList Children;
@@ -37,6 +38,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
         private ref LiveRange CurrRange => ref _data->CurrRange;
         private ref LiveRange PrevRange => ref _data->PrevRange;
         private ref LiveInterval Parent => ref _data->Parent;
+        private ref LiveInterval CopySource => ref _data->CopySource;
         private ref UseList Uses => ref _data->Uses;
         private ref LiveIntervalList Children => ref _data->Children;
 
@@ -78,6 +80,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             Register = register;
         }
 
+        public void SetCopySource(LiveInterval copySource)
+        {
+            CopySource = copySource;
+        }
+
+        public bool TryGetCopySourceRegister(out int copySourceRegIndex)
+        {
+            if (CopySource._data != null)
+            {
+                copySourceRegIndex = CopySource.Register.Index;
+
+                return true;
+            }
+
+            copySourceRegIndex = 0;
+
+            return false;
+        }
+
         public void Reset()
         {
             PrevRange = default;

src/ARMeilleure/Translation/ControlFlowGraph.cs

@@ -11,7 +11,7 @@ namespace ARMeilleure.Translation
         private int[] _postOrderMap;
 
         public int LocalsCount { get; private set; }
-        public BasicBlock Entry { get; }
+        public BasicBlock Entry { get; private set; }
         public IntrusiveList<BasicBlock> Blocks { get; }
         public BasicBlock[] PostOrderBlocks => _postOrderBlocks;
         public int[] PostOrderMap => _postOrderMap;
@@ -34,6 +34,15 @@ namespace ARMeilleure.Translation
             return result;
         }
 
+        public void UpdateEntry(BasicBlock newEntry)
+        {
+            newEntry.AddSuccessor(Entry);
+
+            Entry = newEntry;
+            Blocks.AddFirst(newEntry);
+            Update();
+        }
+
         public void Update()
         {
             RemoveUnreachableBlocks(Blocks);

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -29,7 +29,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 6634; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 6950; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

src/ARMeilleure/Translation/RegisterUsage.cs

@@ -89,6 +89,17 @@ namespace ARMeilleure.Translation
 
         public static void RunPass(ControlFlowGraph cfg, ExecutionMode mode)
         {
+            if (cfg.Entry.Predecessors.Count != 0)
+            {
+                // We expect the entry block to have no predecessors.
+                // This is required because we have a implicit context load at the start of the function,
+                // but if there is a jump to the start of the function, the context load would trash the modified values.
+                // Here we insert a new entry block that will jump to the existing entry block.
+                BasicBlock newEntry = new BasicBlock(cfg.Blocks.Count);
+
+                cfg.UpdateEntry(newEntry);
+            }
+
             // Compute local register inputs and outputs used inside blocks.
             RegisterMask[] localInputs = new RegisterMask[cfg.Blocks.Count];
             RegisterMask[] localOutputs = new RegisterMask[cfg.Blocks.Count];
@@ -201,7 +212,7 @@ namespace ARMeilleure.Translation
 
                 // The only block without any predecessor should be the entry block.
                 // It always needs a context load as it is the first block to run.
-                if (block.Predecessors.Count == 0 || hasContextLoad)
+                if (block == cfg.Entry || hasContextLoad)
                 {
                     long vecMask = globalInputs[block.Index].VecMask;
                     long intMask = globalInputs[block.Index].IntMask;

src/Ryujinx.Audio.Backends.SDL2/SDL2HardwareDeviceSession.cs

@@ -89,9 +89,9 @@ namespace Ryujinx.Audio.Backends.SDL2
                 return;
             }
 
-            using IMemoryOwner<byte> samplesOwner = ByteMemoryPool.Rent(frameCount * _bytesPerFrame);
+            using SpanOwner<byte> samplesOwner = SpanOwner<byte>.Rent(frameCount * _bytesPerFrame);
 
-            Span<byte> samples = samplesOwner.Memory.Span;
+            Span<byte> samples = samplesOwner.Span;
 
             _ringBuffer.Read(samples, 0, samples.Length);
 

src/Ryujinx.Audio.Backends.SoundIo/SoundIoHardwareDeviceSession.cs

@@ -122,9 +122,9 @@ namespace Ryujinx.Audio.Backends.SoundIo
 
             int channelCount = areas.Length;
 
-            using IMemoryOwner<byte> samplesOwner = ByteMemoryPool.Rent(frameCount * bytesPerFrame);
+            using SpanOwner<byte> samplesOwner = SpanOwner<byte>.Rent(frameCount * bytesPerFrame);
 
-            Span<byte> samples = samplesOwner.Memory.Span;
+            Span<byte> samples = samplesOwner.Span;
 
             _ringBuffer.Read(samples, 0, samples.Length);
 

src/Ryujinx.Audio/Backends/Common/DynamicRingBuffer.cs

@@ -14,7 +14,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         private readonly object _lock = new();
 
-        private IMemoryOwner<byte> _bufferOwner;
+        private MemoryOwner<byte> _bufferOwner;
         private Memory<byte> _buffer;
         private int _size;
         private int _headOffset;
@@ -24,7 +24,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         public DynamicRingBuffer(int initialCapacity = RingBufferAlignment)
         {
-            _bufferOwner = ByteMemoryPool.RentCleared(initialCapacity);
+            _bufferOwner = MemoryOwner<byte>.RentCleared(initialCapacity);
             _buffer = _bufferOwner.Memory;
         }
 
@@ -62,7 +62,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         private void SetCapacityLocked(int capacity)
         {
-            IMemoryOwner<byte> newBufferOwner = ByteMemoryPool.RentCleared(capacity);
+            MemoryOwner<byte> newBufferOwner = MemoryOwner<byte>.RentCleared(capacity);
             Memory<byte> newBuffer = newBufferOwner.Memory;
 
             if (_size > 0)

src/Ryujinx.Common/Memory/MemoryOwner.cs

@@ -124,7 +124,7 @@ namespace Ryujinx.Common.Memory
 
             if (array is not null)
             {
-                ArrayPool<T>.Shared.Return(array);
+                ArrayPool<T>.Shared.Return(array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
             }
         }
 

src/Ryujinx.Common/Memory/SpanOwner.cs

@@ -108,7 +108,7 @@ namespace Ryujinx.Common.Memory
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public void Dispose()
         {
-            ArrayPool<T>.Shared.Return(_array);
+            ArrayPool<T>.Shared.Return(_array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
         }
     }
 }

src/Ryujinx.Graphics.GAL/IImageArray.cs

@@ -1,6 +1,8 @@
+using System;
+
 namespace Ryujinx.Graphics.GAL
 {
-    public interface IImageArray
+    public interface IImageArray : IDisposable
     {
         void SetFormats(int index, Format[] imageFormats);
         void SetImages(int index, ITexture[] images);

src/Ryujinx.Graphics.GAL/ITextureArray.cs

@@ -1,6 +1,8 @@
+using System;
+
 namespace Ryujinx.Graphics.GAL
 {
-    public interface ITextureArray
+    public interface ITextureArray : IDisposable
     {
         void SetSamplers(int index, ISampler[] samplers);
         void SetTextures(int index, ITexture[] textures);

src/Ryujinx.Graphics.GAL/Multithreading/CommandHelper.cs

@@ -66,6 +66,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             Register<CounterEventDisposeCommand>(CommandType.CounterEventDispose);
             Register<CounterEventFlushCommand>(CommandType.CounterEventFlush);
 
+            Register<ImageArrayDisposeCommand>(CommandType.ImageArrayDispose);
             Register<ImageArraySetFormatsCommand>(CommandType.ImageArraySetFormats);
             Register<ImageArraySetImagesCommand>(CommandType.ImageArraySetImages);
 
@@ -88,6 +89,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             Register<TextureSetDataSliceRegionCommand>(CommandType.TextureSetDataSliceRegion);
             Register<TextureSetStorageCommand>(CommandType.TextureSetStorage);
 
+            Register<TextureArrayDisposeCommand>(CommandType.TextureArrayDispose);
             Register<TextureArraySetSamplersCommand>(CommandType.TextureArraySetSamplers);
             Register<TextureArraySetTexturesCommand>(CommandType.TextureArraySetTextures);
 

src/Ryujinx.Graphics.GAL/Multithreading/CommandType.cs

@@ -26,6 +26,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
         CounterEventDispose,
         CounterEventFlush,
 
+        ImageArrayDispose,
         ImageArraySetFormats,
         ImageArraySetImages,
 
@@ -48,6 +49,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
         TextureSetDataSliceRegion,
         TextureSetStorage,
 
+        TextureArrayDispose,
         TextureArraySetSamplers,
         TextureArraySetTextures,
 

src/Ryujinx.Graphics.GAL/Multithreading/Commands/ImageArray/ImageArrayDisposeCommand.cs

@@ -0,0 +1,21 @@
+using Ryujinx.Graphics.GAL.Multithreading.Model;
+using Ryujinx.Graphics.GAL.Multithreading.Resources;
+
+namespace Ryujinx.Graphics.GAL.Multithreading.Commands.ImageArray
+{
+    struct ImageArrayDisposeCommand : IGALCommand, IGALCommand<ImageArrayDisposeCommand>
+    {
+        public readonly CommandType CommandType => CommandType.ImageArrayDispose;
+        private TableRef<ThreadedImageArray> _imageArray;
+
+        public void Set(TableRef<ThreadedImageArray> imageArray)
+        {
+            _imageArray = imageArray;
+        }
+
+        public static void Run(ref ImageArrayDisposeCommand command, ThreadedRenderer threaded, IRenderer renderer)
+        {
+            command._imageArray.Get(threaded).Base.Dispose();
+        }
+    }
+}

src/Ryujinx.Graphics.GAL/Multithreading/Commands/TextureArray/TextureArrayDisposeCommand.cs

@@ -0,0 +1,21 @@
+using Ryujinx.Graphics.GAL.Multithreading.Model;
+using Ryujinx.Graphics.GAL.Multithreading.Resources;
+
+namespace Ryujinx.Graphics.GAL.Multithreading.Commands.TextureArray
+{
+    struct TextureArrayDisposeCommand : IGALCommand, IGALCommand<TextureArrayDisposeCommand>
+    {
+        public readonly CommandType CommandType => CommandType.TextureArrayDispose;
+        private TableRef<ThreadedTextureArray> _textureArray;
+
+        public void Set(TableRef<ThreadedTextureArray> textureArray)
+        {
+            _textureArray = textureArray;
+        }
+
+        public static void Run(ref TextureArrayDisposeCommand command, ThreadedRenderer threaded, IRenderer renderer)
+        {
+            command._textureArray.Get(threaded).Base.Dispose();
+        }
+    }
+}

src/Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedImageArray.cs

@@ -21,6 +21,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             return new TableRef<T>(_renderer, reference);
         }
 
+        public void Dispose()
+        {
+            _renderer.New<ImageArrayDisposeCommand>().Set(Ref(this));
+            _renderer.QueueCommand();
+        }
+
         public void SetFormats(int index, Format[] imageFormats)
         {
             _renderer.New<ImageArraySetFormatsCommand>().Set(Ref(this), index, Ref(imageFormats));

src/Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedTextureArray.cs

@@ -22,6 +22,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             return new TableRef<T>(_renderer, reference);
         }
 
+        public void Dispose()
+        {
+            _renderer.New<TextureArrayDisposeCommand>().Set(Ref(this));
+            _renderer.QueueCommand();
+        }
+
         public void SetSamplers(int index, ISampler[] samplers)
         {
             _renderer.New<TextureArraySetSamplersCommand>().Set(Ref(this), index, Ref(samplers.ToArray()));

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

@@ -1113,6 +1113,15 @@ namespace Ryujinx.Graphics.Gpu.Image
                 nextNode = nextNode.Next;
                 _cacheFromBuffer.Remove(toRemove.Value.Key);
                 _lruCache.Remove(toRemove);
+
+                if (toRemove.Value.Key.IsImage)
+                {
+                    toRemove.Value.ImageArray.Dispose();
+                }
+                else
+                {
+                    toRemove.Value.TextureArray.Dispose();
+                }
             }
         }
 
@@ -1124,11 +1133,20 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             List<CacheEntryFromPoolKey> keysToRemove = null;
 
-            foreach (CacheEntryFromPoolKey key in _cacheFromPool.Keys)
+            foreach ((CacheEntryFromPoolKey key, CacheEntry entry) in _cacheFromPool)
             {
                 if (key.MatchesPool(pool))
                 {
                     (keysToRemove ??= new()).Add(key);
+
+                    if (key.IsImage)
+                    {
+                        entry.ImageArray.Dispose();
+                    }
+                    else
+                    {
+                        entry.TextureArray.Dispose();
+                    }
                 }
             }
 

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 = 6852;
+        private const uint CodeGenVersion = 6921;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

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

@@ -63,5 +63,9 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 }
             }
         }
+
+        public void Dispose()
+        {
+        }
     }
 }

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

@@ -48,5 +48,9 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 }
             }
         }
+
+        public void Dispose()
+        {
+        }
     }
 }

src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs

@@ -98,11 +98,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Logger = parameters.Logger;
             TargetApi = parameters.TargetApi;
 
-            AddCapability(Capability.Shader);
-            AddCapability(Capability.Float64);
-
-            SetMemoryModel(AddressingModel.Logical, MemoryModel.GLSL450);
-
             Delegates = new SpirvDelegates(this);
         }
 

src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs

@@ -43,6 +43,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             CodeGenContext context = new(info, parameters, instPool, integerPool);
 
+            context.AddCapability(Capability.Shader);
+
+            context.SetMemoryModel(AddressingModel.Logical, MemoryModel.GLSL450);
+
             context.AddCapability(Capability.GroupNonUniformBallot);
             context.AddCapability(Capability.GroupNonUniformShuffle);
             context.AddCapability(Capability.GroupNonUniformVote);
@@ -51,6 +55,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             context.AddCapability(Capability.ImageQuery);
             context.AddCapability(Capability.SampledBuffer);
 
+            if (parameters.HostCapabilities.SupportsShaderFloat64)
+            {
+                context.AddCapability(Capability.Float64);
+            }
+
             if (parameters.Definitions.TransformFeedbackEnabled && parameters.Definitions.LastInVertexPipeline)
             {
                 context.AddCapability(Capability.TransformFeedback);
@@ -58,7 +67,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             if (parameters.Definitions.Stage == ShaderStage.Fragment)
             {
-                if (context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.Layer)))
+                if (context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.Layer)) ||
+                    context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.PrimitiveId)))
                 {
                     context.AddCapability(Capability.Geometry);
                 }

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

@@ -24,7 +24,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             if (op.BVal)
             {
-                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), Const(0)));
+                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), ConstF(0)));
             }
             else
             {

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

@@ -156,6 +156,26 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
             return false;
         }
 
+        public static bool IsComparison(this Instruction inst)
+        {
+            switch (inst & Instruction.Mask)
+            {
+                case Instruction.CompareEqual:
+                case Instruction.CompareGreater:
+                case Instruction.CompareGreaterOrEqual:
+                case Instruction.CompareGreaterOrEqualU32:
+                case Instruction.CompareGreaterU32:
+                case Instruction.CompareLess:
+                case Instruction.CompareLessOrEqual:
+                case Instruction.CompareLessOrEqualU32:
+                case Instruction.CompareLessU32:
+                case Instruction.CompareNotEqual:
+                    return true;
+            }
+
+            return false;
+        }
+
         public static bool IsTextureQuery(this Instruction inst)
         {
             inst &= Instruction.Mask;

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

@@ -8,6 +8,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         public readonly bool SupportsGeometryShaderPassthrough;
         public readonly bool SupportsShaderBallot;
         public readonly bool SupportsShaderBarrierDivergence;
+        public readonly bool SupportsShaderFloat64;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsViewportMask;
 
@@ -18,6 +19,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             bool supportsGeometryShaderPassthrough,
             bool supportsShaderBallot,
             bool supportsShaderBarrierDivergence,
+            bool supportsShaderFloat64,
             bool supportsTextureShadowLod,
             bool supportsViewportMask)
         {
@@ -27,6 +29,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
             SupportsShaderBallot = supportsShaderBallot;
             SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
+            SupportsShaderFloat64 = supportsShaderFloat64;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsViewportMask = supportsViewportMask;
         }

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

@@ -141,16 +141,16 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static bool IsBindlessAccessAllowed(Operand nvHandle)
+        private static bool IsBindlessAccessAllowed(Operand bindlessHandle)
         {
-            if (nvHandle.Type == OperandType.ConstantBuffer)
+            if (bindlessHandle.Type == OperandType.ConstantBuffer)
             {
                 // Bindless access with handles from constant buffer is allowed.
 
                 return true;
             }
 
-            if (nvHandle.AsgOp is not Operation handleOp ||
+            if (bindlessHandle.AsgOp is not Operation handleOp ||
                 handleOp.Inst != Instruction.Load ||
                 (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
             {
@@ -300,7 +300,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                         resourceManager,
                         gpuAccessor,
                         texOp,
-                        TextureHandle.PackOffsets(src0.GetCbufOffset(), ((src1.Value >> 20) & 0xfff), handleType),
+                        TextureHandle.PackOffsets(src0.GetCbufOffset(), (src1.Value >> 20) & 0xfff, handleType),
                         TextureHandle.PackSlots(src0.GetCbufSlot(), 0),
                         rewriteSamplerType,
                         isImage: false);

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

@@ -126,7 +126,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     continue;
                 }
 
-                if (texOp.GetSource(0).AsgOp is not Operation handleAsgOp)
+                Operand bindlessHandle = Utils.FindLastOperation(texOp.GetSource(0), block);
+
+                if (bindlessHandle.AsgOp is not Operation handleAsgOp)
                 {
                     continue;
                 }
@@ -137,8 +139,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
                 if (handleAsgOp.Inst == Instruction.BitwiseOr)
                 {
-                    Operand src0 = handleAsgOp.GetSource(0);
-                    Operand src1 = handleAsgOp.GetSource(1);
+                    Operand src0 = Utils.FindLastOperation(handleAsgOp.GetSource(0), block);
+                    Operand src1 = Utils.FindLastOperation(handleAsgOp.GetSource(1), block);
 
                     if (src0.Type == OperandType.ConstantBuffer && src1.AsgOp is Operation)
                     {

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

@@ -152,18 +152,14 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
         {
             // If all phi sources are the same, we can propagate it and remove the phi.
 
-            Operand firstSrc = phi.GetSource(0);
-
-            for (int index = 1; index < phi.SourcesCount; index++)
-            {
-                if (!IsSameOperand(firstSrc, phi.GetSource(index)))
+            if (!Utils.AreAllSourcesTheSameOperand(phi))
             {
                 return false;
             }
-            }
 
             // All sources are equal, we can propagate the value.
 
+            Operand firstSrc = phi.GetSource(0);
             Operand dest = phi.Dest;
 
             INode[] uses = dest.UseOps.ToArray();
@@ -182,17 +178,6 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static bool IsSameOperand(Operand x, Operand y)
-        {
-            if (x.Type != y.Type || x.Value != y.Value)
-            {
-                return false;
-            }
-
-            // TODO: Handle Load operations with the same storage and the same constant parameters.
-            return x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
-        }
-
         private static bool PropagatePack(Operation packOp)
         {
             // Propagate pack source operands to uses by unpack

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

@@ -31,6 +31,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     TryEliminateBitwiseOr(operation);
                     break;
 
+                case Instruction.CompareNotEqual:
+                    TryEliminateCompareNotEqual(operation);
+                    break;
+
                 case Instruction.ConditionalSelect:
                     TryEliminateConditionalSelect(operation);
                     break;
@@ -174,6 +178,32 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             }
         }
 
+        private static void TryEliminateCompareNotEqual(Operation operation)
+        {
+            // Comparison instruction returns 0 if the result is false, and -1 if true.
+            // Doing a not equal zero comparison on the result is redundant, so we can just copy the first result in this case.
+
+            Operand lhs = operation.GetSource(0);
+            Operand rhs = operation.GetSource(1);
+
+            if (lhs.Type == OperandType.Constant)
+            {
+                (lhs, rhs) = (rhs, lhs);
+            }
+
+            if (rhs.Type != OperandType.Constant || rhs.Value != 0)
+            {
+                return;
+            }
+
+            if (lhs.AsgOp is not Operation compareOp || !compareOp.Inst.IsComparison())
+            {
+                return;
+            }
+
+            operation.TurnIntoCopy(lhs);
+        }
+
         private static void TryEliminateConditionalSelect(Operation operation)
         {
             Operand cond = operation.GetSource(0);

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

@@ -34,6 +34,50 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return elemIndexSrc.Type == OperandType.Constant && elemIndexSrc.Value == elemIndex;
         }
 
+        private static bool IsSameOperand(Operand x, Operand y)
+        {
+            if (x.Type != y.Type || x.Value != y.Value)
+            {
+                return false;
+            }
+
+            // TODO: Handle Load operations with the same storage and the same constant parameters.
+            return x == y || x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
+        }
+
+        private static bool AreAllSourcesEqual(INode node, INode otherNode)
+        {
+            if (node.SourcesCount != otherNode.SourcesCount)
+            {
+                return false;
+            }
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(node.GetSource(index), otherNode.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
+        public static bool AreAllSourcesTheSameOperand(INode node)
+        {
+            Operand firstSrc = node.GetSource(0);
+
+            for (int index = 1; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(firstSrc, node.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
         private static Operation FindBranchSource(BasicBlock block)
         {
             foreach (BasicBlock sourceBlock in block.Predecessors)
@@ -55,6 +99,19 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return inst == Instruction.BranchIfFalse || inst == Instruction.BranchIfTrue;
         }
 
+        private static bool IsSameCondition(Operand currentCondition, Operand queryCondition)
+        {
+            if (currentCondition == queryCondition)
+            {
+                return true;
+            }
+
+            return currentCondition.AsgOp is Operation currentOperation &&
+                queryCondition.AsgOp is Operation queryOperation &&
+                currentOperation.Inst == queryOperation.Inst &&
+                AreAllSourcesEqual(currentOperation, queryOperation);
+        }
+
         private static bool BlockConditionsMatch(BasicBlock currentBlock, BasicBlock queryBlock)
         {
             // Check if all the conditions for the query block are satisfied by the current block.
@@ -70,10 +127,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
             return currentBranch != null && queryBranch != null &&
                    currentBranch.Inst == queryBranch.Inst &&
-                   currentCondition == queryCondition;
+                   IsSameCondition(currentCondition, queryCondition);
         }
 
-        public static Operand FindLastOperation(Operand source, BasicBlock block)
+        public static Operand FindLastOperation(Operand source, BasicBlock block, bool recurse = true)
         {
             if (source.AsgOp is PhiNode phiNode)
             {
@@ -84,10 +141,23 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 for (int i = phiNode.SourcesCount - 1; i >= 0; i--)
                 {
                     BasicBlock phiBlock = phiNode.GetBlock(i);
+                    Operand phiSource = phiNode.GetSource(i);
 
                     if (BlockConditionsMatch(block, phiBlock))
                     {
-                        return phiNode.GetSource(i);
+                        return phiSource;
+                    }
+                    else if (recurse && phiSource.AsgOp is PhiNode)
+                    {
+                        // Phi source is another phi.
+                        // Let's check if that phi has a block that matches our condition.
+
+                        Operand match = FindLastOperation(phiSource, block, false);
+
+                        if (match != phiSource)
+                        {
+                            return match;
+                        }
                     }
                 }
             }

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

@@ -363,6 +363,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 GpuAccessor.QueryHostSupportsGeometryShaderPassthrough(),
                 GpuAccessor.QueryHostSupportsShaderBallot(),
                 GpuAccessor.QueryHostSupportsShaderBarrierDivergence(),
+                GpuAccessor.QueryHostSupportsShaderFloat64(),
                 GpuAccessor.QueryHostSupportsTextureShadowLod(),
                 GpuAccessor.QueryHostSupportsViewportMask());
 

src/Ryujinx.Graphics.Vulkan/BufferManager.cs

@@ -103,12 +103,19 @@ namespace Ryujinx.Graphics.Vulkan
                 usage |= BufferUsageFlags.IndirectBufferBit;
             }
 
+            var externalMemoryBuffer = new ExternalMemoryBufferCreateInfo
+            {
+                SType = StructureType.ExternalMemoryBufferCreateInfo,
+                HandleTypes = ExternalMemoryHandleTypeFlags.HostAllocationBitExt,
+            };
+
             var bufferCreateInfo = new BufferCreateInfo
             {
                 SType = StructureType.BufferCreateInfo,
                 Size = (ulong)size,
                 Usage = usage,
                 SharingMode = SharingMode.Exclusive,
+                PNext = &externalMemoryBuffer,
             };
 
             gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();

src/Ryujinx.Graphics.Vulkan/DescriptorSetUpdater.cs

@@ -291,8 +291,9 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                     else
                     {
-                        PipelineStageFlags stageFlags = _textureArrayRefs[segment.Binding].Stage.ConvertToPipelineStageFlags();
-                        _textureArrayRefs[segment.Binding].Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                        ref var arrayRef = ref _textureArrayRefs[segment.Binding];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
                     }
                 }
             }
@@ -311,8 +312,40 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                     else
                     {
-                        PipelineStageFlags stageFlags = _imageArrayRefs[segment.Binding].Stage.ConvertToPipelineStageFlags();
-                        _imageArrayRefs[segment.Binding].Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                        ref var arrayRef = ref _imageArrayRefs[segment.Binding];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                    }
+                }
+            }
+
+            for (int setIndex = PipelineBase.DescriptorSetLayouts; setIndex < _program.BindingSegments.Length; setIndex++)
+            {
+                var bindingSegments = _program.BindingSegments[setIndex];
+
+                if (bindingSegments.Length == 0)
+                {
+                    continue;
+                }
+
+                ResourceBindingSegment segment = bindingSegments[0];
+
+                if (segment.IsArray)
+                {
+                    if (segment.Type == ResourceType.Texture ||
+                        segment.Type == ResourceType.Sampler ||
+                        segment.Type == ResourceType.TextureAndSampler ||
+                        segment.Type == ResourceType.BufferTexture)
+                    {
+                        ref var arrayRef = ref _textureArrayExtraRefs[setIndex - PipelineBase.DescriptorSetLayouts];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                    }
+                    else if (segment.Type == ResourceType.Image || segment.Type == ResourceType.BufferImage)
+                    {
+                        ref var arrayRef = ref _imageArrayExtraRefs[setIndex - PipelineBase.DescriptorSetLayouts];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
                     }
                 }
             }

src/Ryujinx.Graphics.Vulkan/ImageArray.cs

@@ -2,11 +2,10 @@ using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
-using System.Diagnostics;
 
 namespace Ryujinx.Graphics.Vulkan
 {
-    class ImageArray : IImageArray
+    class ImageArray : ResourceArray, IImageArray
     {
         private readonly VulkanRenderer _gd;
 
@@ -25,19 +24,11 @@ namespace Ryujinx.Graphics.Vulkan
 
         private HashSet<TextureStorage> _storages;
 
-        private DescriptorSet[] _cachedDescriptorSets;
-
         private int _cachedCommandBufferIndex;
         private int _cachedSubmissionCount;
 
-        private ShaderCollection _cachedDscProgram;
-        private int _cachedDscSetIndex;
-        private int _cachedDscIndex;
-
         private readonly bool _isBuffer;
 
-        private int _bindCount;
-
         public ImageArray(VulkanRenderer gd, int size, bool isBuffer)
         {
             _gd = gd;
@@ -104,12 +95,7 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _cachedCommandBufferIndex = -1;
             _storages = null;
-            _cachedDescriptorSets = null;
-
-            if (_bindCount != 0)
-            {
-                _gd.PipelineInternal.ForceImageDirty();
-            }
+            SetDirty(_gd);
         }
 
         public void QueueWriteToReadBarriers(CommandBufferScoped cbs, PipelineStageFlags stageFlags)
@@ -195,7 +181,7 @@ namespace Ryujinx.Graphics.Vulkan
             int setIndex,
             TextureView dummyTexture)
         {
-            if (_cachedDescriptorSets != null)
+            if (TryGetCachedDescriptorSets(cbs, program, setIndex, out DescriptorSet[] sets))
             {
                 // We still need to ensure the current command buffer holds a reference to all used textures.
 
@@ -208,12 +194,9 @@ namespace Ryujinx.Graphics.Vulkan
                     GetBufferViews(cbs);
                 }
 
-                return _cachedDescriptorSets;
+                return sets;
             }
 
-            _cachedDscProgram?.ReleaseManualDescriptorSetCollection(_cachedDscSetIndex, _cachedDscIndex);
-            var dsc = program.GetNewManualDescriptorSetCollection(cbs.CommandBufferIndex, setIndex, out _cachedDscIndex).Get(cbs);
-
             DescriptorSetTemplate template = program.Templates[setIndex];
 
             DescriptorSetTemplateWriter tu = templateUpdater.Begin(template);
@@ -227,24 +210,9 @@ namespace Ryujinx.Graphics.Vulkan
                 tu.Push(GetBufferViews(cbs));
             }
 
-            var sets = dsc.GetSets();
             templateUpdater.Commit(_gd, device, sets[0]);
-            _cachedDescriptorSets = sets;
-            _cachedDscProgram = program;
-            _cachedDscSetIndex = setIndex;
 
             return sets;
         }
-
-        public void IncrementBindCount()
-        {
-            _bindCount++;
-        }
-
-        public void DecrementBindCount()
-        {
-            int newBindCount = --_bindCount;
-            Debug.Assert(newBindCount >= 0);
-        }
     }
 }

src/Ryujinx.Graphics.Vulkan/PipelineLayoutCacheEntry.cs

@@ -3,6 +3,7 @@ using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
 using System.Collections.ObjectModel;
+using System.Diagnostics;
 using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Vulkan
@@ -15,6 +16,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
 
         public DescriptorSetLayout[] DescriptorSetLayouts { get; }
+        public bool[] DescriptorSetLayoutsUpdateAfterBind { get; }
         public PipelineLayout PipelineLayout { get; }
 
         private readonly int[] _consumedDescriptorsPerSet;
@@ -31,20 +33,37 @@ namespace Ryujinx.Graphics.Vulkan
         private struct ManualDescriptorSetEntry
         {
             public Auto<DescriptorSetCollection> DescriptorSet;
-            public int CbIndex;
-            public int CbSubmissionCount;
+            public uint CbRefMask;
             public bool InUse;
 
-            public ManualDescriptorSetEntry(Auto<DescriptorSetCollection> descriptorSet, int cbIndex, int cbSubmissionCount, bool inUse)
+            public ManualDescriptorSetEntry(Auto<DescriptorSetCollection> descriptorSet, int cbIndex)
             {
                 DescriptorSet = descriptorSet;
-                CbIndex = cbIndex;
-                CbSubmissionCount = cbSubmissionCount;
-                InUse = inUse;
+                CbRefMask = 1u << cbIndex;
+                InUse = true;
+            }
+        }
+
+        private readonly struct PendingManualDsConsumption
+        {
+            public FenceHolder Fence { get; }
+            public int CommandBufferIndex { get; }
+            public int SetIndex { get; }
+            public int CacheIndex { get; }
+
+            public PendingManualDsConsumption(FenceHolder fence, int commandBufferIndex, int setIndex, int cacheIndex)
+            {
+                Fence = fence;
+                CommandBufferIndex = commandBufferIndex;
+                SetIndex = setIndex;
+                CacheIndex = cacheIndex;
+                fence.Get();
             }
         }
 
         private readonly List<ManualDescriptorSetEntry>[] _manualDsCache;
+        private readonly Queue<PendingManualDsConsumption> _pendingManualDsConsumptions;
+        private readonly Queue<int>[] _freeManualDsCacheEntries;
 
         private readonly Dictionary<long, DescriptorSetTemplate> _pdTemplates;
         private readonly ResourceDescriptorCollection _pdDescriptors;
@@ -70,6 +89,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             _dsCacheCursor = new int[setsCount];
             _manualDsCache = new List<ManualDescriptorSetEntry>[setsCount];
+            _pendingManualDsConsumptions = new Queue<PendingManualDsConsumption>();
+            _freeManualDsCacheEntries = new Queue<int>[setsCount];
         }
 
         public PipelineLayoutCacheEntry(
@@ -78,7 +99,11 @@ namespace Ryujinx.Graphics.Vulkan
             ReadOnlyCollection<ResourceDescriptorCollection> setDescriptors,
             bool usePushDescriptors) : this(gd, device, setDescriptors.Count)
         {
-            (DescriptorSetLayouts, PipelineLayout) = PipelineLayoutFactory.Create(gd, device, setDescriptors, usePushDescriptors);
+            ResourceLayouts layouts = PipelineLayoutFactory.Create(gd, device, setDescriptors, usePushDescriptors);
+
+            DescriptorSetLayouts = layouts.DescriptorSetLayouts;
+            DescriptorSetLayoutsUpdateAfterBind = layouts.DescriptorSetLayoutsUpdateAfterBind;
+            PipelineLayout = layouts.PipelineLayout;
 
             _consumedDescriptorsPerSet = new int[setDescriptors.Count];
             _poolSizes = new DescriptorPoolSize[setDescriptors.Count][];
@@ -133,7 +158,7 @@ namespace Ryujinx.Graphics.Vulkan
                     _poolSizes[setIndex],
                     setIndex,
                     _consumedDescriptorsPerSet[setIndex],
-                    false);
+                    DescriptorSetLayoutsUpdateAfterBind[setIndex]);
 
                 list.Add(dsc);
                 isNew = true;
@@ -144,49 +169,99 @@ namespace Ryujinx.Graphics.Vulkan
             return list[index];
         }
 
-        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(int commandBufferIndex, int setIndex, out int cacheIndex)
+        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(CommandBufferScoped cbs, int setIndex, out int cacheIndex)
         {
-            int submissionCount = _gd.CommandBufferPool.GetSubmissionCount(commandBufferIndex);
+            FreeCompletedManualDescriptorSets();
 
             var list = _manualDsCache[setIndex] ??= new();
             var span = CollectionsMarshal.AsSpan(list);
 
-            for (int index = 0; index < span.Length; index++)
-            {
-                ref ManualDescriptorSetEntry entry = ref span[index];
+            Queue<int> freeQueue = _freeManualDsCacheEntries[setIndex];
 
-                if (!entry.InUse && (entry.CbIndex != commandBufferIndex || entry.CbSubmissionCount != submissionCount))
+            // Do we have at least one freed descriptor set? If so, just use that.
+            if (freeQueue != null && freeQueue.TryDequeue(out int freeIndex))
             {
+                ref ManualDescriptorSetEntry entry = ref span[freeIndex];
+
+                Debug.Assert(!entry.InUse && entry.CbRefMask == 0);
+
                 entry.InUse = true;
-                    entry.CbIndex = commandBufferIndex;
-                    entry.CbSubmissionCount = submissionCount;
+                entry.CbRefMask = 1u << cbs.CommandBufferIndex;
+                cacheIndex = freeIndex;
 
-                    cacheIndex = index;
+                _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, freeIndex));
 
                 return entry.DescriptorSet;
             }
-            }
 
+            // Otherwise create a new descriptor set, and add to our pending queue for command buffer consumption tracking.
             var dsc = _descriptorSetManager.AllocateDescriptorSet(
                 _gd.Api,
                 DescriptorSetLayouts[setIndex],
                 _poolSizes[setIndex],
                 setIndex,
                 _consumedDescriptorsPerSet[setIndex],
-                false);
+                DescriptorSetLayoutsUpdateAfterBind[setIndex]);
 
             cacheIndex = list.Count;
-            list.Add(new ManualDescriptorSetEntry(dsc, commandBufferIndex, submissionCount, inUse: true));
+            list.Add(new ManualDescriptorSetEntry(dsc, cbs.CommandBufferIndex));
+            _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, cacheIndex));
 
             return dsc;
         }
 
+        public void UpdateManualDescriptorSetCollectionOwnership(CommandBufferScoped cbs, int setIndex, int cacheIndex)
+        {
+            FreeCompletedManualDescriptorSets();
+
+            var list = _manualDsCache[setIndex];
+            var span = CollectionsMarshal.AsSpan(list);
+            ref var entry = ref span[cacheIndex];
+
+            uint cbMask = 1u << cbs.CommandBufferIndex;
+
+            if ((entry.CbRefMask & cbMask) == 0)
+            {
+                entry.CbRefMask |= cbMask;
+
+                _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, cacheIndex));
+            }
+        }
+
+        private void FreeCompletedManualDescriptorSets()
+        {
+            FenceHolder signalledFence = null;
+            while (_pendingManualDsConsumptions.TryPeek(out var pds) && (pds.Fence == signalledFence || pds.Fence.IsSignaled()))
+            {
+                signalledFence = pds.Fence; // Already checked - don't need to do it again.
+                var dequeued = _pendingManualDsConsumptions.Dequeue();
+                Debug.Assert(dequeued.Fence == pds.Fence);
+                pds.Fence.Put();
+
+                var span = CollectionsMarshal.AsSpan(_manualDsCache[dequeued.SetIndex]);
+                ref var entry = ref span[dequeued.CacheIndex];
+                entry.CbRefMask &= ~(1u << dequeued.CommandBufferIndex);
+
+                if (!entry.InUse && entry.CbRefMask == 0)
+                {
+                    // If not in use by any array, and not bound to any command buffer, the descriptor set can be re-used immediately.
+                    (_freeManualDsCacheEntries[dequeued.SetIndex] ??= new()).Enqueue(dequeued.CacheIndex);
+                }
+            }
+        }
+
         public void ReleaseManualDescriptorSetCollection(int setIndex, int cacheIndex)
         {
             var list = _manualDsCache[setIndex];
             var span = CollectionsMarshal.AsSpan(list);
 
             span[cacheIndex].InUse = false;
+
+            if (span[cacheIndex].CbRefMask == 0)
+            {
+                // This is no longer in use by any array, so if not bound to any command buffer, the descriptor set can be re-used immediately.
+                (_freeManualDsCacheEntries[setIndex] ??= new()).Enqueue(cacheIndex);
+            }
         }
 
         private static Span<DescriptorPoolSize> GetDescriptorPoolSizes(Span<DescriptorPoolSize> output, ResourceDescriptorCollection setDescriptor, uint multiplier)
@@ -291,6 +366,11 @@ namespace Ryujinx.Graphics.Vulkan
                     _gd.Api.DestroyDescriptorSetLayout(_device, DescriptorSetLayouts[i], null);
                 }
 
+                while (_pendingManualDsConsumptions.TryDequeue(out var pds))
+                {
+                    pds.Fence.Put();
+                }
+
                 _descriptorSetManager.Dispose();
             }
         }

src/Ryujinx.Graphics.Vulkan/PipelineLayoutFactory.cs

@@ -1,18 +1,23 @@
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
+using System;
 using System.Collections.ObjectModel;
 
 namespace Ryujinx.Graphics.Vulkan
 {
+    record struct ResourceLayouts(DescriptorSetLayout[] DescriptorSetLayouts, bool[] DescriptorSetLayoutsUpdateAfterBind, PipelineLayout PipelineLayout);
+
     static class PipelineLayoutFactory
     {
-        public static unsafe (DescriptorSetLayout[], PipelineLayout) Create(
+        public static unsafe ResourceLayouts Create(
             VulkanRenderer gd,
             Device device,
             ReadOnlyCollection<ResourceDescriptorCollection> setDescriptors,
             bool usePushDescriptors)
         {
             DescriptorSetLayout[] layouts = new DescriptorSetLayout[setDescriptors.Count];
+            bool[] updateAfterBindFlags = new bool[setDescriptors.Count];
 
             bool isMoltenVk = gd.IsMoltenVk;
 
@@ -32,10 +37,11 @@ namespace Ryujinx.Graphics.Vulkan
 
                 DescriptorSetLayoutBinding[] layoutBindings = new DescriptorSetLayoutBinding[rdc.Descriptors.Count];
 
+                bool hasArray = false;
+
                 for (int descIndex = 0; descIndex < rdc.Descriptors.Count; descIndex++)
                 {
                     ResourceDescriptor descriptor = rdc.Descriptors[descIndex];
-
                     ResourceStages stages = descriptor.Stages;
 
                     if (descriptor.Type == ResourceType.StorageBuffer && isMoltenVk)
@@ -52,16 +58,37 @@ namespace Ryujinx.Graphics.Vulkan
                         DescriptorCount = (uint)descriptor.Count,
                         StageFlags = stages.Convert(),
                     };
+
+                    if (descriptor.Count > 1)
+                    {
+                        hasArray = true;
+                    }
                 }
 
                 fixed (DescriptorSetLayoutBinding* pLayoutBindings = layoutBindings)
                 {
+                    DescriptorSetLayoutCreateFlags flags = DescriptorSetLayoutCreateFlags.None;
+
+                    if (usePushDescriptors && setIndex == 0)
+                    {
+                        flags = DescriptorSetLayoutCreateFlags.PushDescriptorBitKhr;
+                    }
+
+                    if (gd.Vendor == Vendor.Intel && hasArray)
+                    {
+                        // Some vendors (like Intel) have low per-stage limits.
+                        // We must set the flag if we exceed those limits.
+                        flags |= DescriptorSetLayoutCreateFlags.UpdateAfterBindPoolBit;
+
+                        updateAfterBindFlags[setIndex] = true;
+                    }
+
                     var descriptorSetLayoutCreateInfo = new DescriptorSetLayoutCreateInfo
                     {
                         SType = StructureType.DescriptorSetLayoutCreateInfo,
                         PBindings = pLayoutBindings,
                         BindingCount = (uint)layoutBindings.Length,
-                        Flags = usePushDescriptors && setIndex == 0 ? DescriptorSetLayoutCreateFlags.PushDescriptorBitKhr : DescriptorSetLayoutCreateFlags.None,
+                        Flags = flags,
                     };
 
                     gd.Api.CreateDescriptorSetLayout(device, descriptorSetLayoutCreateInfo, null, out layouts[setIndex]).ThrowOnError();
@@ -82,7 +109,7 @@ namespace Ryujinx.Graphics.Vulkan
                 gd.Api.CreatePipelineLayout(device, &pipelineLayoutCreateInfo, null, out layout).ThrowOnError();
             }
 
-            return (layouts, layout);
+            return new ResourceLayouts(layouts, updateAfterBindFlags, layout);
         }
     }
 }

src/Ryujinx.Graphics.Vulkan/ResourceArray.cs

@@ -0,0 +1,74 @@
+using Silk.NET.Vulkan;
+using System;
+using System.Diagnostics;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    class ResourceArray : IDisposable
+    {
+        private DescriptorSet[] _cachedDescriptorSets;
+
+        private ShaderCollection _cachedDscProgram;
+        private int _cachedDscSetIndex;
+        private int _cachedDscIndex;
+
+        private int _bindCount;
+
+        protected void SetDirty(VulkanRenderer gd)
+        {
+            ReleaseDescriptorSet();
+
+            if (_bindCount != 0)
+            {
+                gd.PipelineInternal.ForceTextureDirty();
+            }
+        }
+
+        public bool TryGetCachedDescriptorSets(CommandBufferScoped cbs, ShaderCollection program, int setIndex, out DescriptorSet[] sets)
+        {
+            if (_cachedDescriptorSets != null)
+            {
+                _cachedDscProgram.UpdateManualDescriptorSetCollectionOwnership(cbs, _cachedDscSetIndex, _cachedDscIndex);
+
+                sets = _cachedDescriptorSets;
+
+                return true;
+            }
+
+            var dsc = program.GetNewManualDescriptorSetCollection(cbs, setIndex, out _cachedDscIndex).Get(cbs);
+
+            sets = dsc.GetSets();
+
+            _cachedDescriptorSets = sets;
+            _cachedDscProgram = program;
+            _cachedDscSetIndex = setIndex;
+
+            return false;
+        }
+
+        public void IncrementBindCount()
+        {
+            _bindCount++;
+        }
+
+        public void DecrementBindCount()
+        {
+            int newBindCount = --_bindCount;
+            Debug.Assert(newBindCount >= 0);
+        }
+
+        private void ReleaseDescriptorSet()
+        {
+            if (_cachedDescriptorSets != null)
+            {
+                _cachedDscProgram.ReleaseManualDescriptorSetCollection(_cachedDscSetIndex, _cachedDscIndex);
+                _cachedDescriptorSets = null;
+            }
+        }
+
+        public void Dispose()
+        {
+            ReleaseDescriptorSet();
+        }
+    }
+}

src/Ryujinx.Graphics.Vulkan/ShaderCollection.cs

@@ -604,9 +604,14 @@ namespace Ryujinx.Graphics.Vulkan
             return _plce.GetNewDescriptorSetCollection(setIndex, out isNew);
         }
 
-        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(int commandBufferIndex, int setIndex, out int cacheIndex)
+        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(CommandBufferScoped cbs, int setIndex, out int cacheIndex)
         {
-            return _plce.GetNewManualDescriptorSetCollection(commandBufferIndex, setIndex, out cacheIndex);
+            return _plce.GetNewManualDescriptorSetCollection(cbs, setIndex, out cacheIndex);
+        }
+
+        public void UpdateManualDescriptorSetCollectionOwnership(CommandBufferScoped cbs, int setIndex, int cacheIndex)
+        {
+            _plce.UpdateManualDescriptorSetCollectionOwnership(cbs, setIndex, cacheIndex);
         }
 
         public void ReleaseManualDescriptorSetCollection(int setIndex, int cacheIndex)

src/Ryujinx.Graphics.Vulkan/TextureArray.cs

@@ -2,11 +2,10 @@ using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
-using System.Diagnostics;
 
 namespace Ryujinx.Graphics.Vulkan
 {
-    class TextureArray : ITextureArray
+    class TextureArray : ResourceArray, ITextureArray
     {
         private readonly VulkanRenderer _gd;
 
@@ -25,19 +24,11 @@ namespace Ryujinx.Graphics.Vulkan
 
         private HashSet<TextureStorage> _storages;
 
-        private DescriptorSet[] _cachedDescriptorSets;
-
         private int _cachedCommandBufferIndex;
         private int _cachedSubmissionCount;
 
-        private ShaderCollection _cachedDscProgram;
-        private int _cachedDscSetIndex;
-        private int _cachedDscIndex;
-
         private readonly bool _isBuffer;
 
-        private int _bindCount;
-
         public TextureArray(VulkanRenderer gd, int size, bool isBuffer)
         {
             _gd = gd;
@@ -113,12 +104,7 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _cachedCommandBufferIndex = -1;
             _storages = null;
-            _cachedDescriptorSets = null;
-
-            if (_bindCount != 0)
-            {
-                _gd.PipelineInternal.ForceTextureDirty();
-            }
+            SetDirty(_gd);
         }
 
         public void QueueWriteToReadBarriers(CommandBufferScoped cbs, PipelineStageFlags stageFlags)
@@ -211,7 +197,7 @@ namespace Ryujinx.Graphics.Vulkan
             TextureView dummyTexture,
             SamplerHolder dummySampler)
         {
-            if (_cachedDescriptorSets != null)
+            if (TryGetCachedDescriptorSets(cbs, program, setIndex, out DescriptorSet[] sets))
             {
                 // We still need to ensure the current command buffer holds a reference to all used textures.
 
@@ -224,12 +210,9 @@ namespace Ryujinx.Graphics.Vulkan
                     GetBufferViews(cbs);
                 }
 
-                return _cachedDescriptorSets;
+                return sets;
             }
 
-            _cachedDscProgram?.ReleaseManualDescriptorSetCollection(_cachedDscSetIndex, _cachedDscIndex);
-            var dsc = program.GetNewManualDescriptorSetCollection(cbs.CommandBufferIndex, setIndex, out _cachedDscIndex).Get(cbs);
-
             DescriptorSetTemplate template = program.Templates[setIndex];
 
             DescriptorSetTemplateWriter tu = templateUpdater.Begin(template);
@@ -243,24 +226,9 @@ namespace Ryujinx.Graphics.Vulkan
                 tu.Push(GetBufferViews(cbs));
             }
 
-            var sets = dsc.GetSets();
             templateUpdater.Commit(_gd, device, sets[0]);
-            _cachedDescriptorSets = sets;
-            _cachedDscProgram = program;
-            _cachedDscSetIndex = setIndex;
 
             return sets;
         }
-
-        public void IncrementBindCount()
-        {
-            _bindCount++;
-        }
-
-        public void DecrementBindCount()
-        {
-            int newBindCount = --_bindCount;
-            Debug.Assert(newBindCount >= 0);
-        }
     }
 }

src/Ryujinx.Graphics.Vulkan/TextureStorage.cs

@@ -80,7 +80,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             var usage = GetImageUsage(info.Format, info.Target, gd.Capabilities.SupportsShaderStorageImageMultisample);
 
-            var flags = ImageCreateFlags.CreateMutableFormatBit;
+            var flags = ImageCreateFlags.CreateMutableFormatBit | ImageCreateFlags.CreateExtendedUsageBit;
 
             // This flag causes mipmapped texture arrays to break on AMD GCN, so for that copy dependencies are forced for aliasing as cube.
             bool isCube = info.Target == Target.Cubemap || info.Target == Target.CubemapArray;

src/Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -100,7 +100,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             unsafe Auto<DisposableImageView> CreateImageView(ComponentMapping cm, ImageSubresourceRange sr, ImageViewType viewType, ImageUsageFlags usageFlags)
             {
-                var usage = new ImageViewUsageCreateInfo
+                var imageViewUsage = new ImageViewUsageCreateInfo
                 {
                     SType = StructureType.ImageViewUsageCreateInfo,
                     Usage = usageFlags,
@@ -114,7 +114,7 @@ namespace Ryujinx.Graphics.Vulkan
                     Format = format,
                     Components = cm,
                     SubresourceRange = sr,
-                    PNext = &usage,
+                    PNext = &imageViewUsage,
                 };
 
                 gd.Api.CreateImageView(device, imageCreateInfo, null, out var imageView).ThrowOnError();
@@ -123,7 +123,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             ImageUsageFlags shaderUsage = ImageUsageFlags.SampledBit;
 
-            if (info.Format.IsImageCompatible())
+            if (info.Format.IsImageCompatible() && (_gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample()))
             {
                 shaderUsage |= ImageUsageFlags.StorageBit;
             }
@@ -154,7 +154,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
                 else
                 {
-                    subresourceRange = new ImageSubresourceRange(aspectFlags, (uint)firstLevel, levels, (uint)firstLayer, (uint)info.Depth);
+                    subresourceRange = new ImageSubresourceRange(aspectFlags, (uint)firstLevel, 1, (uint)firstLayer, (uint)info.Depth);
 
                     _imageView2dArray = CreateImageView(identityComponentMapping, subresourceRange, ImageViewType.Type2DArray, usage);
                 }

src/Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -42,6 +42,8 @@ namespace Ryujinx.Graphics.Vulkan
             "VK_EXT_depth_clip_control",
             "VK_KHR_portability_subset", // As per spec, we should enable this if present.
             "VK_EXT_4444_formats",
+            "VK_KHR_8bit_storage",
+            "VK_KHR_maintenance2",
         };
 
         private static readonly string[] _requiredExtensions = {
@@ -355,6 +357,14 @@ namespace Ryujinx.Graphics.Vulkan
                 features2.PNext = &supportedFeaturesDepthClipControl;
             }
 
+            PhysicalDeviceVulkan12Features supportedPhysicalDeviceVulkan12Features = new()
+            {
+                SType = StructureType.PhysicalDeviceVulkan12Features,
+                PNext = features2.PNext,
+            };
+
+            features2.PNext = &supportedPhysicalDeviceVulkan12Features;
+
             api.GetPhysicalDeviceFeatures2(physicalDevice.PhysicalDevice, &features2);
 
             var supportedFeatures = features2.Features;
@@ -382,6 +392,7 @@ namespace Ryujinx.Graphics.Vulkan
                 TessellationShader = supportedFeatures.TessellationShader,
                 VertexPipelineStoresAndAtomics = supportedFeatures.VertexPipelineStoresAndAtomics,
                 RobustBufferAccess = useRobustBufferAccess,
+                SampleRateShading = supportedFeatures.SampleRateShading,
             };
 
             void* pExtendedFeatures = null;
@@ -451,9 +462,11 @@ namespace Ryujinx.Graphics.Vulkan
             {
                 SType = StructureType.PhysicalDeviceVulkan12Features,
                 PNext = pExtendedFeatures,
-                DescriptorIndexing = physicalDevice.IsDeviceExtensionPresent("VK_EXT_descriptor_indexing"),
-                DrawIndirectCount = physicalDevice.IsDeviceExtensionPresent(KhrDrawIndirectCount.ExtensionName),
-                UniformBufferStandardLayout = physicalDevice.IsDeviceExtensionPresent("VK_KHR_uniform_buffer_standard_layout"),
+                DescriptorIndexing = supportedPhysicalDeviceVulkan12Features.DescriptorIndexing,
+                DrawIndirectCount = supportedPhysicalDeviceVulkan12Features.DrawIndirectCount,
+                UniformBufferStandardLayout = supportedPhysicalDeviceVulkan12Features.UniformBufferStandardLayout,
+                UniformAndStorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.UniformAndStorageBuffer8BitAccess,
+                StorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.StorageBuffer8BitAccess,
             };
 
             pExtendedFeatures = &featuresVk12;

src/Ryujinx.Graphics.Vulkan/Window.cs

@@ -623,7 +623,8 @@ namespace Ryujinx.Graphics.Vulkan
 
         public override void SetSize(int width, int height)
         {
-            // Not needed as we can get the size from the surface.
+            // We don't need to use width and height as we can get the size from the surface.
+            _swapchainIsDirty = true;
         }
 
         public override void ChangeVSyncMode(bool vsyncEnabled)

src/Ryujinx.HLE/HOS/Kernel/SupervisorCall/Syscall.cs

@@ -616,7 +616,7 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
                 }
             }
 
-            ArrayPool<KSynchronizationObject>.Shared.Return(syncObjsArray);
+            ArrayPool<KSynchronizationObject>.Shared.Return(syncObjsArray, true);
 
             return result;
         }
@@ -1546,8 +1546,8 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
 #pragma warning disable CA1822 // Mark member as static
         public Result SetProcessMemoryPermission(
             int handle,
-            [PointerSized] ulong src,
-            [PointerSized] ulong size,
+            ulong src,
+            ulong size,
             KMemoryPermission permission)
         {
             if (!PageAligned(src))

src/Ryujinx.HLE/HOS/Kernel/Threading/KSynchronization.cs

@@ -104,7 +104,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                     }
                 }
 
-                ArrayPool<LinkedListNode<KThread>>.Shared.Return(syncNodesArray);
+                ArrayPool<LinkedListNode<KThread>>.Shared.Return(syncNodesArray, true);
             }
 
             _context.CriticalSection.Leave();

src/Ryujinx.UI.Common/DiscordIntegrationModule.cs

@@ -104,8 +104,13 @@ namespace Ryujinx.UI.Common
             // Find the length to trim the string to guarantee we have space for the trailing ellipsis.
             int trimLimit = byteLimit - Encoding.UTF8.GetByteCount(Ellipsis);
 
+            // Make sure the string is long enough to perform the basic trim.
+            // Amount of bytes != Length of the string
+            if (input.Length > trimLimit)
+            {
                 // Basic trim to best case scenario of 1 byte characters.
                 input = input[..trimLimit];
+            }
 
             while (Encoding.UTF8.GetByteCount(input) > trimLimit)
             {