Rewrite kernel memory allocator (#3316)

* Rewrite kernel memory allocator

* Remove unused using

* Adjust private static field naming

* Change UlongBitSize to UInt64BitSize

* Fix unused argument, change argument order to be inline with official code and disable random allocation

ARMeilleure/IntermediateRepresentation/Operand.cs

@@ -14,10 +14,11 @@ namespace ARMeilleure.IntermediateRepresentation
             public byte Kind;
             public byte Type;
             public byte SymbolType;
+            public byte Padding; // Unused space.
             public ushort AssignmentsCount;
             public ushort AssignmentsCapacity;
-            public ushort UsesCount;
-            public ushort UsesCapacity;
+            public uint UsesCount;
+            public uint UsesCapacity;
             public Operation* Assignments;
             public Operation* Uses;
             public ulong Value;
@@ -84,11 +85,11 @@ namespace ARMeilleure.IntermediateRepresentation
             {
                 Debug.Assert(Kind != OperandKind.Memory);
 
-                return new ReadOnlySpan<Operation>(_data->Uses, _data->UsesCount);
+                return new ReadOnlySpan<Operation>(_data->Uses, (int)_data->UsesCount);
             }
         }
 
-        public int UsesCount => _data->UsesCount;
+        public int UsesCount => (int)_data->UsesCount;
         public int AssignmentsCount => _data->AssignmentsCount;
 
         public bool Relocatable => Symbol.Type != SymbolType.None;
@@ -265,6 +266,13 @@ namespace ARMeilleure.IntermediateRepresentation
             data = Allocators.References.Allocate<T>(initialCapacity);
         }
 
+        private static void New<T>(ref T* data, ref uint count, ref uint capacity, uint initialCapacity) where T : unmanaged
+        {
+            count = 0;
+            capacity = initialCapacity;
+            data = Allocators.References.Allocate<T>(initialCapacity);
+        }
+
         private static void Add<T>(T item, ref T* data, ref ushort count, ref ushort capacity) where T : unmanaged
         {
             if (count < capacity)
@@ -294,6 +302,40 @@ namespace ARMeilleure.IntermediateRepresentation
             }
         }
 
+        private static void Add<T>(T item, ref T* data, ref uint count, ref uint capacity) where T : unmanaged
+        {
+            if (count < capacity)
+            {
+                data[count++] = item;
+
+                return;
+            }
+
+            // Could not add item in the fast path, fallback onto the slow path.
+            ExpandAdd(item, ref data, ref count, ref capacity);
+
+            static void ExpandAdd(T item, ref T* data, ref uint count, ref uint capacity)
+            {
+                uint newCount = checked(count + 1);
+                uint newCapacity = (uint)Math.Min(capacity * 2, int.MaxValue);
+
+                if (newCapacity <= capacity)
+                {
+                    throw new OverflowException();
+                }
+
+                var oldSpan = new Span<T>(data, (int)count);
+
+                capacity = newCapacity;
+                data = Allocators.References.Allocate<T>(capacity);
+
+                oldSpan.CopyTo(new Span<T>(data, (int)count));
+
+                data[count] = item;
+                count = newCount;
+            }
+        }
+
         private static void Remove<T>(in T item, ref T* data, ref ushort count) where T : unmanaged
         {
             var span = new Span<T>(data, count);
@@ -314,6 +356,26 @@ namespace ARMeilleure.IntermediateRepresentation
             }
         }
 
+        private static void Remove<T>(in T item, ref T* data, ref uint count) where T : unmanaged
+        {
+            var span = new Span<T>(data, (int)count);
+
+            for (int i = 0; i < span.Length; i++)
+            {
+                if (EqualityComparer<T>.Default.Equals(span[i], item))
+                {
+                    if (i + 1 < count)
+                    {
+                        span.Slice(i + 1).CopyTo(span.Slice(i));
+                    }
+
+                    count--;
+
+                    return;
+                }
+            }
+        }
+
         public override int GetHashCode()
         {
             if (Kind == OperandKind.LocalVariable)

Ryujinx.Common/Utilities/BitUtils.cs

@@ -22,7 +22,17 @@ namespace Ryujinx.Common
 
         public static long AlignUp(long value, int size)
         {
-            return (value + (size - 1)) & -(long)size;
+            return AlignUp(value, (long)size);
+        }
+
+        public static ulong AlignUp(ulong value, ulong size)
+        {
+            return (ulong)AlignUp((long)value, (long)size);
+        }
+
+        public static long AlignUp(long value, long size)
+        {
+            return (value + (size - 1)) & -size;
         }
 
         public static uint AlignDown(uint value, int size)
@@ -42,7 +52,17 @@ namespace Ryujinx.Common
 
         public static long AlignDown(long value, int size)
         {
-            return value & -(long)size;
+            return AlignDown(value, (long)size);
+        }
+
+        public static ulong AlignDown(ulong value, ulong size)
+        {
+            return (ulong)AlignDown((long)value, (long)size);
+        }
+
+        public static long AlignDown(long value, long size)
+        {
+            return value & -size;
         }
 
         public static int DivRoundUp(int value, int dividend)

Ryujinx.Graphics.GAL/IPipeline.cs

@@ -10,9 +10,10 @@ namespace Ryujinx.Graphics.GAL
 
         void ClearBuffer(BufferHandle destination, int offset, int size, uint value);
 
-        void ClearRenderTargetColor(int index, uint componentMask, ColorF color);
+        void ClearRenderTargetColor(int index, int layer, uint componentMask, ColorF color);
 
         void ClearRenderTargetDepthStencil(
+            int layer,
             float depthValue,
             bool depthMask,
             int stencilValue,

Ryujinx.Graphics.GAL/Multithreading/Commands/ClearRenderTargetColorCommand.cs

@@ -4,19 +4,21 @@
     {
         public CommandType CommandType => CommandType.ClearRenderTargetColor;
         private int _index;
+        private int _layer;
         private uint _componentMask;
         private ColorF _color;
 
-        public void Set(int index, uint componentMask, ColorF color)
+        public void Set(int index, int layer, uint componentMask, ColorF color)
         {
             _index = index;
+            _layer = layer;
             _componentMask = componentMask;
             _color = color;
         }
 
         public static void Run(ref ClearRenderTargetColorCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            renderer.Pipeline.ClearRenderTargetColor(command._index, command._componentMask, command._color);
+            renderer.Pipeline.ClearRenderTargetColor(command._index, command._layer, command._componentMask, command._color);
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/Commands/ClearRenderTargetDepthStencilCommand.cs

@@ -3,13 +3,15 @@
     struct ClearRenderTargetDepthStencilCommand : IGALCommand
     {
         public CommandType CommandType => CommandType.ClearRenderTargetDepthStencil;
+        private int _layer;
         private float _depthValue;
         private bool _depthMask;
         private int _stencilValue;
         private int _stencilMask;
 
-        public void Set(float depthValue, bool depthMask, int stencilValue, int stencilMask)
+        public void Set(int layer, float depthValue, bool depthMask, int stencilValue, int stencilMask)
         {
+            _layer = layer;
             _depthValue = depthValue;
             _depthMask = depthMask;
             _stencilValue = stencilValue;
@@ -18,7 +20,7 @@
 
         public static void Run(ref ClearRenderTargetDepthStencilCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            renderer.Pipeline.ClearRenderTargetDepthStencil(command._depthValue, command._depthMask, command._stencilValue, command._stencilMask);
+            renderer.Pipeline.ClearRenderTargetDepthStencil(command._layer, command._depthValue, command._depthMask, command._stencilValue, command._stencilMask);
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/ThreadedPipeline.cs

@@ -40,15 +40,15 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             _renderer.QueueCommand();
         }
 
-        public void ClearRenderTargetColor(int index, uint componentMask, ColorF color)
+        public void ClearRenderTargetColor(int index, int layer, uint componentMask, ColorF color)
         {
-            _renderer.New<ClearRenderTargetColorCommand>().Set(index, componentMask, color);
+            _renderer.New<ClearRenderTargetColorCommand>().Set(index, layer, componentMask, color);
             _renderer.QueueCommand();
         }
 
-        public void ClearRenderTargetDepthStencil(float depthValue, bool depthMask, int stencilValue, int stencilMask)
+        public void ClearRenderTargetDepthStencil(int layer, float depthValue, bool depthMask, int stencilValue, int stencilMask)
         {
-            _renderer.New<ClearRenderTargetDepthStencilCommand>().Set(depthValue, depthMask, stencilValue, stencilMask);
+            _renderer.New<ClearRenderTargetDepthStencilCommand>().Set(layer, depthValue, depthMask, stencilValue, stencilMask);
             _renderer.QueueCommand();
         }
 

Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs

@@ -188,6 +188,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
             _channel.BufferManager.SetComputeStorageBufferBindings(info.SBuffers);
             _channel.BufferManager.SetComputeUniformBufferBindings(info.CBuffers);
 
+            int maxTextureBinding = -1;
+            int maxImageBinding = -1;
+
             TextureBindingInfo[] textureBindings = _channel.TextureManager.RentComputeTextureBindings(info.Textures.Count);
 
             for (int index = 0; index < info.Textures.Count; index++)
@@ -202,6 +205,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
                     descriptor.CbufSlot,
                     descriptor.HandleIndex,
                     descriptor.Flags);
+
+                if (descriptor.Binding > maxTextureBinding)
+                {
+                    maxTextureBinding = descriptor.Binding;
+                }
             }
 
             TextureBindingInfo[] imageBindings = _channel.TextureManager.RentComputeImageBindings(info.Images.Count);
@@ -220,9 +228,18 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
                     descriptor.CbufSlot,
                     descriptor.HandleIndex,
                     descriptor.Flags);
+
+                if (descriptor.Binding > maxImageBinding)
+                {
+                    maxImageBinding = descriptor.Binding;
+                }
             }
 
-            _channel.TextureManager.CommitComputeBindings();
+            _channel.TextureManager.SetComputeMaxBindings(maxTextureBinding, maxImageBinding);
+
+            // Should never return false for mismatching spec state, since the shader was fetched above.
+            _channel.TextureManager.CommitComputeBindings(cs.SpecializationState); 
+            
             _channel.BufferManager.CommitComputeBindings();
 
             _context.Renderer.Pipeline.DispatchCompute(qmd.CtaRasterWidth, qmd.CtaRasterHeight, qmd.CtaRasterDepth);

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

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
         private bool _instancedDrawPending;
         private bool _instancedIndexed;
+        private bool _instancedIndexedInline;
 
         private int _instancedFirstIndex;
         private int _instancedFirstVertex;
@@ -134,13 +135,16 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             {
                 _instancedDrawPending = true;
 
+                int ibCount = _drawState.IbStreamer.InlineIndexCount;
+
                 _instancedIndexed = _drawState.DrawIndexed;
+                _instancedIndexedInline = ibCount != 0;
 
                 _instancedFirstIndex = firstIndex;
                 _instancedFirstVertex = (int)_state.State.FirstVertex;
                 _instancedFirstInstance = (int)_state.State.FirstInstance;
 
-                _instancedIndexCount = indexCount;
+                _instancedIndexCount = ibCount != 0 ? ibCount : indexCount;
 
                 var drawState = _state.State.VertexBufferDrawState;
 
@@ -451,8 +455,18 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             {
                 _instancedDrawPending = false;
 
-                if (_instancedIndexed)
+                bool indexedInline = _instancedIndexedInline;
+
+                if (_instancedIndexed || indexedInline)
                 {
+                    if (indexedInline)
+                    {
+                        int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount();
+                        BufferRange br = new BufferRange(_drawState.IbStreamer.GetInlineIndexBuffer(), 0, inlineIndexCount * 4);
+
+                        _channel.BufferManager.SetIndexBuffer(br, IndexType.UInt);
+                    }
+
                     _context.Renderer.Pipeline.DrawIndexed(
                         _instancedIndexCount,
                         _instanceIndex + 1,
@@ -491,8 +505,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             }
 
             int index = (argument >> 6) & 0xf;
+            int layer = (argument >> 10) & 0x3ff;
 
-            engine.UpdateRenderTargetState(useControl: false, singleUse: index);
+            engine.UpdateRenderTargetState(useControl: false, layered: layer != 0, singleUse: index);
 
             // If there is a mismatch on the host clip region and the one explicitly defined by the guest
             // on the screen scissor state, then we need to force only one texture to be bound to avoid
@@ -567,7 +582,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
                 ColorF color = new ColorF(clearColor.Red, clearColor.Green, clearColor.Blue, clearColor.Alpha);
 
-                _context.Renderer.Pipeline.ClearRenderTargetColor(index, componentMask, color);
+                _context.Renderer.Pipeline.ClearRenderTargetColor(index, layer, componentMask, color);
             }
 
             if (clearDepth || clearStencil)
@@ -588,6 +603,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 }
 
                 _context.Renderer.Pipeline.ClearRenderTargetDepthStencil(
+                    layer,
                     depthValue,
                     clearDepth,
                     stencilValue,

Ryujinx.Graphics.Gpu/Engine/Threed/IbStreamer.cs

@@ -20,6 +20,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         public bool HasInlineIndexData => _inlineIndexCount != 0;
 
+        /// <summary>
+        /// Total numbers of indices that have been pushed.
+        /// </summary>
+        public int InlineIndexCount => _inlineIndexCount;
+
         /// <summary>
         /// Gets the handle for the host buffer currently holding the inline index buffer data.
         /// </summary>

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

@@ -201,7 +201,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             // of the shader for the new state.
             if (_shaderSpecState != null)
             {
-                if (!_shaderSpecState.MatchesGraphics(_channel, GetPoolState(), GetGraphicsState()))
+                if (!_shaderSpecState.MatchesGraphics(_channel, GetPoolState(), GetGraphicsState(), false))
                 {
                     ForceShaderUpdate();
                 }
@@ -275,7 +275,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         {
             UpdateStorageBuffers();
 
-            _channel.TextureManager.CommitGraphicsBindings();
+            if (!_channel.TextureManager.CommitGraphicsBindings(_shaderSpecState))
+            {
+                // Shader must be reloaded.
+                UpdateShaderState();
+            }
+
             _channel.BufferManager.CommitGraphicsBindings();
         }
 
@@ -362,8 +367,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// Updates render targets (color and depth-stencil buffers) based on current render target state.
         /// </summary>
         /// <param name="useControl">Use draw buffers information from render target control register</param>
+        /// <param name="layered">Indicates if the texture is layered</param>
         /// <param name="singleUse">If this is not -1, it indicates that only the given indexed target will be used.</param>
-        public void UpdateRenderTargetState(bool useControl, int singleUse = -1)
+        public void UpdateRenderTargetState(bool useControl, bool layered = false, int singleUse = -1)
         {
             var memoryManager = _channel.MemoryManager;
             var rtControl = _state.State.RtControl;
@@ -399,7 +405,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 Image.Texture color = memoryManager.Physical.TextureCache.FindOrCreateTexture(
                     memoryManager,
                     colorState,
-                    _vtgWritesRtLayer,
+                    _vtgWritesRtLayer || layered,
                     samplesInX,
                     samplesInY,
                     sizeHint);
@@ -433,6 +439,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                     memoryManager,
                     dsState,
                     dsSize,
+                    _vtgWritesRtLayer || layered,
                     samplesInX,
                     samplesInY,
                     sizeHint);
@@ -1148,6 +1155,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 return;
             }
 
+            int maxTextureBinding = -1;
+            int maxImageBinding = -1;
+
             Span<TextureBindingInfo> textureBindings = _channel.TextureManager.RentGraphicsTextureBindings(stage, info.Textures.Count);
 
             if (info.UsesRtLayer)
@@ -1167,6 +1177,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                     descriptor.CbufSlot,
                     descriptor.HandleIndex,
                     descriptor.Flags);
+
+                if (descriptor.Binding > maxTextureBinding)
+                {
+                    maxTextureBinding = descriptor.Binding;
+                }
             }
 
             TextureBindingInfo[] imageBindings = _channel.TextureManager.RentGraphicsImageBindings(stage, info.Images.Count);
@@ -1185,7 +1200,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                     descriptor.CbufSlot,
                     descriptor.HandleIndex,
                     descriptor.Flags);
+
+                if (descriptor.Binding > maxImageBinding)
+                {
+                    maxImageBinding = descriptor.Binding;
                 }
+            }
+
+            _channel.TextureManager.SetGraphicsMaxBindings(maxTextureBinding, maxImageBinding);
 
             _channel.BufferManager.SetGraphicsStorageBufferBindings(stage, info.SBuffers);
             _channel.BufferManager.SetGraphicsUniformBufferBindings(stage, info.CBuffers);

Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs

@@ -131,10 +131,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// Updates render targets (color and depth-stencil buffers) based on current render target state.
         /// </summary>
         /// <param name="useControl">Use draw buffers information from render target control register</param>
+        /// <param name="layered">Indicates if the texture is layered</param>
         /// <param name="singleUse">If this is not -1, it indicates that only the given indexed target will be used.</param>
-        public void UpdateRenderTargetState(bool useControl, int singleUse = -1)
+        public void UpdateRenderTargetState(bool useControl, bool layered = false, int singleUse = -1)
         {
-            _stateUpdater.UpdateRenderTargetState(useControl, singleUse);
+            _stateUpdater.UpdateRenderTargetState(useControl, layered, singleUse);
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Image/Pool.cs

@@ -1,6 +1,7 @@
 using Ryujinx.Cpu.Tracking;
 using Ryujinx.Graphics.Gpu.Memory;
 using System;
+using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Gpu.Image
 {
@@ -16,6 +17,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         protected GpuContext Context;
         protected PhysicalMemory PhysicalMemory;
         protected int SequenceNumber;
+        protected int ModifiedSequenceNumber;
 
         protected T1[] Items;
         protected T2[] DescriptorCache;
@@ -41,6 +43,9 @@ namespace Ryujinx.Graphics.Gpu.Image
         private readonly CpuMultiRegionHandle _memoryTracking;
         private readonly Action<ulong, ulong> _modifiedDelegate;
 
+        private int _modifiedSequenceOffset;
+        private bool _modified;
+
         /// <summary>
         /// Creates a new instance of the GPU resource pool.
         /// </summary>
@@ -79,6 +84,16 @@ namespace Ryujinx.Graphics.Gpu.Image
             return PhysicalMemory.Read<T2>(Address + (ulong)id * DescriptorSize);
         }
 
+        /// <summary>
+        /// Gets a reference to the descriptor for a given ID.
+        /// </summary>
+        /// <param name="id">ID of the descriptor. This is effectively a zero-based index</param>
+        /// <returns>A reference to the descriptor</returns>
+        public ref readonly T2 GetDescriptorRef(int id)
+        {
+            return ref MemoryMarshal.Cast<byte, T2>(PhysicalMemory.GetSpan(Address + (ulong)id * DescriptorSize, DescriptorSize))[0];
+        }
+
         /// <summary>
         /// Gets the GPU resource with the given ID.
         /// </summary>
@@ -93,7 +108,13 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         public void SynchronizeMemory()
         {
+            _modified = false;
             _memoryTracking.QueryModified(_modifiedDelegate);
+
+            if (_modified)
+            {
+                UpdateModifiedSequence();
+            }
         }
 
         /// <summary>
@@ -103,6 +124,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="mSize">Size of the modified region</param>
         private void RegionModified(ulong mAddress, ulong mSize)
         {
+            _modified = true;
+
             if (mAddress < Address)
             {
                 mAddress = Address;
@@ -118,6 +141,15 @@ namespace Ryujinx.Graphics.Gpu.Image
             InvalidateRangeImpl(mAddress, mSize);
         }
 
+        /// <summary>
+        /// Updates the modified sequence number using the current sequence number and offset,
+        /// indicating that it has been modified.
+        /// </summary>
+        protected void UpdateModifiedSequence()
+        {
+            ModifiedSequenceNumber = SequenceNumber + _modifiedSequenceOffset;
+        }
+
         /// <summary>
         /// An action to be performed when a precise memory access occurs to this resource.
         /// Makes sure that the dirty flags are checked.
@@ -129,6 +161,16 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             if (write && Context.SequenceNumber == SequenceNumber)
             {
+                if (ModifiedSequenceNumber == SequenceNumber + _modifiedSequenceOffset)
+                {
+                    // The modified sequence number is offset when PreciseActions occur so that
+                    // users checking it will see an increment and know the pool has changed since
+                    // their last look, even though the main SequenceNumber has not been changed.
+
+                    _modifiedSequenceOffset++;
+                }
+
+                // Force the pool to be checked again the next time it is used.
                 SequenceNumber--;
             }
 

Ryujinx.Graphics.Gpu/Image/Sampler.cs

@@ -8,6 +8,11 @@ namespace Ryujinx.Graphics.Gpu.Image
     /// </summary>
     class Sampler : IDisposable
     {
+        /// <summary>
+        /// True if the sampler is disposed, false otherwise.
+        /// </summary>
+        public bool IsDisposed { get; private set; }
+
         /// <summary>
         /// Host sampler object.
         /// </summary>
@@ -101,6 +106,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         public void Dispose()
         {
+            IsDisposed = true;
+
             _hostSampler.Dispose();
             _anisoSampler?.Dispose();
         }

Ryujinx.Graphics.Gpu/Image/SamplerPool.cs

@@ -48,6 +48,8 @@ namespace Ryujinx.Graphics.Gpu.Image
                             Items[i] = null;
                         }
                     }
+
+                    UpdateModifiedSequence();
                 }
 
                 SequenceNumber = Context.SequenceNumber;
@@ -71,6 +73,39 @@ namespace Ryujinx.Graphics.Gpu.Image
             return sampler;
         }
 
+        /// <summary>
+        /// Checks if the pool was modified, and returns the last sequence number where a modification was detected.
+        /// </summary>
+        /// <returns>A number that increments each time a modification is detected</returns>
+        public int CheckModified()
+        {
+            if (SequenceNumber != Context.SequenceNumber)
+            {
+                SequenceNumber = Context.SequenceNumber;
+
+                if (_forcedAnisotropy != GraphicsConfig.MaxAnisotropy)
+                {
+                    _forcedAnisotropy = GraphicsConfig.MaxAnisotropy;
+
+                    for (int i = 0; i < Items.Length; i++)
+                    {
+                        if (Items[i] != null)
+                        {
+                            Items[i].Dispose();
+
+                            Items[i] = null;
+                        }
+                    }
+
+                    UpdateModifiedSequence();
+                }
+
+                SynchronizeMemory();
+            }
+
+            return ModifiedSequenceNumber;
+        }
+
         /// <summary>
         /// Implementation of the sampler pool range invalidation.
         /// </summary>

Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -100,6 +100,11 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         public bool AlwaysFlushOnOverlap { get; private set; }
 
+        /// <summary>
+        /// Increments when the host texture is swapped, or when the texture is removed from all pools.
+        /// </summary>
+        public int InvalidatedSequence { get; private set; }
+
         private int _depth;
         private int _layers;
         public int FirstLayer { get; private set; }
@@ -1407,6 +1412,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             DisposeTextures();
 
             HostTexture = hostTexture;
+            InvalidatedSequence++;
         }
 
         /// <summary>
@@ -1535,6 +1541,8 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                 _poolOwners.Clear();
             }
+
+            InvalidatedSequence++;
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Image/TextureBindingsManager.cs

@@ -1,8 +1,12 @@
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Engine.Types;
+using Ryujinx.Graphics.Gpu.Memory;
+using Ryujinx.Graphics.Gpu.Shader;
 using Ryujinx.Graphics.Shader;
 using System;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Gpu.Image
 {
@@ -31,21 +35,30 @@ namespace Ryujinx.Graphics.Gpu.Image
         private readonly TextureBindingInfo[][] _textureBindings;
         private readonly TextureBindingInfo[][] _imageBindings;
 
-        private struct TextureStatePerStage
+        private struct TextureState
         {
             public ITexture Texture;
             public ISampler Sampler;
+
+            public int TextureHandle;
+            public int SamplerHandle;
+            public int InvalidatedSequence;
+            public Texture CachedTexture;
+            public Sampler CachedSampler;
+            public int ScaleIndex;
+            public TextureUsageFlags UsageFlags;
         }
 
-        private readonly TextureStatePerStage[][] _textureState;
-        private readonly TextureStatePerStage[][] _imageState;
+        private TextureState[] _textureState;
+        private TextureState[] _imageState;
 
         private int[] _textureBindingsCount;
         private int[] _imageBindingsCount;
 
-        private int _textureBufferIndex;
+        private int _texturePoolSequence;
+        private int _samplerPoolSequence;
 
-        private bool _rebind;
+        private int _textureBufferIndex;
 
         private readonly float[] _scales;
         private bool _scaleChanged;
@@ -72,8 +85,8 @@ namespace Ryujinx.Graphics.Gpu.Image
             _textureBindings = new TextureBindingInfo[stages][];
             _imageBindings   = new TextureBindingInfo[stages][];
 
-            _textureState = new TextureStatePerStage[stages][];
-            _imageState   = new TextureStatePerStage[stages][];
+            _textureState = new TextureState[InitialTextureStateSize];
+            _imageState   = new TextureState[InitialImageStateSize];
 
             _textureBindingsCount = new int[stages];
             _imageBindingsCount = new int[stages];
@@ -82,9 +95,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             {
                 _textureBindings[stage] = new TextureBindingInfo[InitialTextureStateSize];
                 _imageBindings[stage] = new TextureBindingInfo[InitialImageStateSize];
-
-                _textureState[stage] = new TextureStatePerStage[InitialTextureStateSize];
-                _imageState[stage] = new TextureStatePerStage[InitialImageStateSize];
             }
         }
 
@@ -99,15 +109,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             if (count > _textureBindings[stage].Length)
             {
                 Array.Resize(ref _textureBindings[stage], count);
-                Array.Resize(ref _textureState[stage], count);
-            }
-
-            int toClear = Math.Max(_textureBindingsCount[stage], count);
-            TextureStatePerStage[] state = _textureState[stage];
-
-            for (int i = 0; i < toClear; i++)
-            {
-                state[i] = new TextureStatePerStage();
             }
 
             _textureBindingsCount[stage] = count;
@@ -126,15 +127,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             if (count > _imageBindings[stage].Length)
             {
                 Array.Resize(ref _imageBindings[stage], count);
-                Array.Resize(ref _imageState[stage], count);
-            }
-
-            int toClear = Math.Max(_imageBindingsCount[stage], count);
-            TextureStatePerStage[] state = _imageState[stage];
-
-            for (int i = 0; i < toClear; i++)
-            {
-                state[i] = new TextureStatePerStage();
             }
 
             _imageBindingsCount[stage] = count;
@@ -142,6 +134,24 @@ namespace Ryujinx.Graphics.Gpu.Image
             return _imageBindings[stage];
         }
 
+        /// <summary>
+        /// Sets the max binding indexes for textures and images.
+        /// </summary>
+        /// <param name="maxTextureBinding">The maximum texture binding</param>
+        /// <param name="maxImageBinding">The maximum image binding</param>
+        public void SetMaxBindings(int maxTextureBinding, int maxImageBinding)
+        {
+            if (maxTextureBinding >= _textureState.Length)
+            {
+                Array.Resize(ref _textureState, maxTextureBinding + 1);
+            }
+
+            if (maxImageBinding >= _imageState.Length)
+            {
+                Array.Resize(ref _imageState, maxImageBinding + 1);
+            }
+        }
+
         /// <summary>
         /// Sets the textures constant buffer index.
         /// The constant buffer specified holds the texture handles.
@@ -222,18 +232,18 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Updates the texture scale for a given texture or image.
         /// </summary>
         /// <param name="texture">Start GPU virtual address of the pool</param>
-        /// <param name="binding">The related texture binding</param>
+        /// <param name="usageFlags">The related texture usage flags</param>
         /// <param name="index">The texture/image binding index</param>
         /// <param name="stage">The active shader stage</param>
         /// <returns>True if the given texture has become blacklisted, indicating that its host texture may have changed.</returns>
-        private bool UpdateScale(Texture texture, TextureBindingInfo binding, int index, ShaderStage stage)
+        private bool UpdateScale(Texture texture, TextureUsageFlags usageFlags, int index, ShaderStage stage)
         {
             float result = 1f;
             bool changed = false;
 
-            if ((binding.Flags & TextureUsageFlags.NeedsScaleValue) != 0 && texture != null)
+            if ((usageFlags & TextureUsageFlags.NeedsScaleValue) != 0 && texture != null)
             {
-                if ((binding.Flags & TextureUsageFlags.ResScaleUnsupported) != 0)
+                if ((usageFlags & TextureUsageFlags.ResScaleUnsupported) != 0)
                 {
                     changed = texture.ScaleMode != TextureScaleMode.Blacklisted;
                     texture.BlacklistScale();
@@ -323,7 +333,9 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Ensures that the bindings are visible to the host GPU.
         /// Note: this actually performs the binding using the host graphics API.
         /// </summary>
-        public void CommitBindings()
+        /// <param name="specState">Specialization state for the bound shader</param>
+        /// <returns>True if all bound textures match the current shader specialiation state, false otherwise</returns>
+        public bool CommitBindings(ShaderSpecializationState specState)
         {
             ulong texturePoolAddress = _texturePoolAddress;
 
@@ -331,10 +343,38 @@ namespace Ryujinx.Graphics.Gpu.Image
                 ? _texturePoolCache.FindOrCreate(_channel, texturePoolAddress, _texturePoolMaximumId)
                 : null;
 
+            // Check if the texture pool has been modified since bindings were last committed.
+            // If it wasn't, then it's possible to avoid looking up textures again when the handle remains the same.
+            bool poolModified = false;
+
+            if (texturePool != null)
+            {
+                int texturePoolSequence = texturePool.CheckModified();
+
+                if (_texturePoolSequence != texturePoolSequence)
+                {
+                    poolModified = true;
+                    _texturePoolSequence = texturePoolSequence;
+                }
+            }
+
+            if (_samplerPool != null)
+            {
+                int samplerPoolSequence = _samplerPool.CheckModified();
+
+                if (_samplerPoolSequence != samplerPoolSequence)
+                {
+                    poolModified = true;
+                    _samplerPoolSequence = samplerPoolSequence;
+                }
+            }
+
+            bool specStateMatches = true;
+
             if (_isCompute)
             {
-                CommitTextureBindings(texturePool, ShaderStage.Compute, 0);
-                CommitImageBindings  (texturePool, ShaderStage.Compute, 0);
+                specStateMatches &= CommitTextureBindings(texturePool, ShaderStage.Compute, 0, poolModified, specState);
+                specStateMatches &= CommitImageBindings(texturePool, ShaderStage.Compute, 0, poolModified, specState);
             }
             else
             {
@@ -342,14 +382,57 @@ namespace Ryujinx.Graphics.Gpu.Image
                 {
                     int stageIndex = (int)stage - 1;
 
-                    CommitTextureBindings(texturePool, stage, stageIndex);
-                    CommitImageBindings  (texturePool, stage, stageIndex);
+                    specStateMatches &= CommitTextureBindings(texturePool, stage, stageIndex, poolModified, specState);
+                    specStateMatches &= CommitImageBindings(texturePool, stage, stageIndex, poolModified, specState);
                 }
             }
 
             CommitRenderScale();
 
-            _rebind = false;
+            return specStateMatches;
+        }
+
+        /// <summary>
+        /// Fetch the constant buffers used for a texture to cache.
+        /// </summary>
+        /// <param name="stageIndex">Stage index of the constant buffer</param>
+        /// <param name="cachedTextureBufferIndex">The currently cached texture buffer index</param>
+        /// <param name="cachedSamplerBufferIndex">The currently cached sampler buffer index</param>
+        /// <param name="cachedTextureBuffer">The currently cached texture buffer data</param>
+        /// <param name="cachedSamplerBuffer">The currently cached sampler buffer data</param>
+        /// <param name="textureBufferIndex">The new texture buffer index</param>
+        /// <param name="samplerBufferIndex">The new sampler buffer index</param>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private void UpdateCachedBuffer(
+            int stageIndex,
+            ref int cachedTextureBufferIndex,
+            ref int cachedSamplerBufferIndex,
+            ref ReadOnlySpan<int> cachedTextureBuffer,
+            ref ReadOnlySpan<int> cachedSamplerBuffer,
+            int textureBufferIndex,
+            int samplerBufferIndex)
+        {
+            if (textureBufferIndex != cachedTextureBufferIndex)
+            {
+                ref BufferBounds bounds = ref _channel.BufferManager.GetUniformBufferBounds(_isCompute, stageIndex, textureBufferIndex);
+
+                cachedTextureBuffer = MemoryMarshal.Cast<byte, int>(_channel.MemoryManager.Physical.GetSpan(bounds.Address, (int)bounds.Size));
+                cachedTextureBufferIndex = textureBufferIndex;
+
+                if (samplerBufferIndex == textureBufferIndex)
+                {
+                    cachedSamplerBuffer = cachedTextureBuffer;
+                    cachedSamplerBufferIndex = samplerBufferIndex;
+                }
+            }
+
+            if (samplerBufferIndex != cachedSamplerBufferIndex)
+            {
+                ref BufferBounds bounds = ref _channel.BufferManager.GetUniformBufferBounds(_isCompute, stageIndex, samplerBufferIndex);
+
+                cachedSamplerBuffer = MemoryMarshal.Cast<byte, int>(_channel.MemoryManager.Physical.GetSpan(bounds.Address, (int)bounds.Size));
+                cachedSamplerBufferIndex = samplerBufferIndex;
+            }
         }
 
         /// <summary>
@@ -358,13 +441,16 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         /// <param name="pool">The current texture pool</param>
         /// <param name="stage">The shader stage using the textures to be bound</param>
-        /// <param name="stageIndex">The stage number of the specified shader stage</param>
-        private void CommitTextureBindings(TexturePool pool, ShaderStage stage, int stageIndex)
+        /// <param name="stageIndex">The stage number of the specified shader stage</param
+        /// <param name="poolModified">True if either the texture or sampler pool was modified, false otherwise</param>
+        /// <param name="specState">Specialization state for the bound shader</param>
+        /// <returns>True if all bound textures match the current shader specialiation state, false otherwise</returns>
+        private bool CommitTextureBindings(TexturePool pool, ShaderStage stage, int stageIndex, bool poolModified, ShaderSpecializationState specState)
         {
             int textureCount = _textureBindingsCount[stageIndex];
             if (textureCount == 0)
             {
-                return;
+                return true;
             }
 
             var samplerPool = _samplerPool;
@@ -372,17 +458,27 @@ namespace Ryujinx.Graphics.Gpu.Image
             if (pool == null)
             {
                 Logger.Error?.Print(LogClass.Gpu, $"Shader stage \"{stage}\" uses textures, but texture pool was not set.");
-                return;
+                return true;
             }
 
+            bool specStateMatches = true;
+
+            int cachedTextureBufferIndex = -1;
+            int cachedSamplerBufferIndex = -1;
+            ReadOnlySpan<int> cachedTextureBuffer = Span<int>.Empty;
+            ReadOnlySpan<int> cachedSamplerBuffer = Span<int>.Empty;
+
             for (int index = 0; index < textureCount; index++)
             {
                 TextureBindingInfo bindingInfo = _textureBindings[stageIndex][index];
+                TextureUsageFlags usageFlags = bindingInfo.Flags;
 
                 (int textureBufferIndex, int samplerBufferIndex) = TextureHandle.UnpackSlots(bindingInfo.CbufSlot, _textureBufferIndex);
 
-                int packedId = ReadPackedId(stageIndex, bindingInfo.Handle, textureBufferIndex, samplerBufferIndex);
-                int textureId = UnpackTextureId(packedId);
+                UpdateCachedBuffer(stageIndex, ref cachedTextureBufferIndex, ref cachedSamplerBufferIndex, ref cachedTextureBuffer, ref cachedSamplerBuffer, textureBufferIndex, samplerBufferIndex);
+
+                int packedId = TextureHandle.ReadPackedId(bindingInfo.Handle, cachedTextureBuffer, cachedSamplerBuffer);
+                int textureId = TextureHandle.UnpackTextureId(packedId);
                 int samplerId;
 
                 if (_samplerIndex == SamplerIndex.ViaHeaderIndex)
@@ -391,10 +487,42 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
                 else
                 {
-                    samplerId = UnpackSamplerId(packedId);
+                    samplerId = TextureHandle.UnpackSamplerId(packedId);
                 }
 
-                Texture texture = pool.Get(textureId);
+                ref TextureState state = ref _textureState[bindingInfo.Binding];
+
+                if (!poolModified &&
+                    state.TextureHandle == textureId &&
+                    state.SamplerHandle == samplerId &&
+                    state.CachedTexture != null &&
+                    state.CachedTexture.InvalidatedSequence == state.InvalidatedSequence &&
+                    state.CachedSampler?.IsDisposed != true)
+                {
+                    // The texture is already bound.
+                    state.CachedTexture.SynchronizeMemory();
+
+                    if ((state.ScaleIndex != index || state.UsageFlags != usageFlags) &&
+                        UpdateScale(state.CachedTexture, usageFlags, index, stage))
+                    {
+                        ITexture hostTextureRebind = state.CachedTexture.GetTargetTexture(bindingInfo.Target);
+
+                        state.Texture = hostTextureRebind;
+                        state.ScaleIndex = index;
+                        state.UsageFlags = usageFlags;
+
+                        _context.Renderer.Pipeline.SetTexture(bindingInfo.Binding, hostTextureRebind);
+                    }
+
+                    continue;
+                }
+
+                state.TextureHandle = textureId;
+                state.SamplerHandle = samplerId;
+
+                ref readonly TextureDescriptor descriptor = ref pool.GetForBinding(textureId, out Texture texture);
+
+                specStateMatches &= specState.MatchesTexture(stage, index, descriptor);
 
                 ITexture hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
 
@@ -407,30 +535,38 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
                 else
                 {
-                    if (_textureState[stageIndex][index].Texture != hostTexture || _rebind)
+                    if (state.Texture != hostTexture)
                     {
-                        if (UpdateScale(texture, bindingInfo, index, stage))
+                        if (UpdateScale(texture, usageFlags, index, stage))
                         {
                             hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
                         }
 
-                        _textureState[stageIndex][index].Texture = hostTexture;
+                        state.Texture = hostTexture;
+                        state.ScaleIndex = index;
+                        state.UsageFlags = usageFlags;
 
                         _context.Renderer.Pipeline.SetTexture(bindingInfo.Binding, hostTexture);
                     }
 
                     Sampler sampler = samplerPool?.Get(samplerId);
+                    state.CachedSampler = sampler;
 
                     ISampler hostSampler = sampler?.GetHostSampler(texture);
 
-                    if (_textureState[stageIndex][index].Sampler != hostSampler || _rebind)
+                    if (state.Sampler != hostSampler)
                     {
-                        _textureState[stageIndex][index].Sampler = hostSampler;
+                        state.Sampler = hostSampler;
 
                         _context.Renderer.Pipeline.SetSampler(bindingInfo.Binding, hostSampler);
                     }
+
+                    state.CachedTexture = texture;
+                    state.InvalidatedSequence = texture?.InvalidatedSequence ?? 0;
                 }
             }
+
+            return specStateMatches;
         }
 
         /// <summary>
@@ -440,38 +576,90 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="pool">The current texture pool</param>
         /// <param name="stage">The shader stage using the textures to be bound</param>
         /// <param name="stageIndex">The stage number of the specified shader stage</param>
-        private void CommitImageBindings(TexturePool pool, ShaderStage stage, int stageIndex)
+        /// <param name="poolModified">True if either the texture or sampler pool was modified, false otherwise</param>
+        /// <param name="specState">Specialization state for the bound shader</param>
+        /// <returns>True if all bound images match the current shader specialiation state, false otherwise</returns>
+        private bool CommitImageBindings(TexturePool pool, ShaderStage stage, int stageIndex, bool poolModified, ShaderSpecializationState specState)
         {
             int imageCount = _imageBindingsCount[stageIndex];
             if (imageCount == 0)
             {
-                return;
+                return true;
             }
 
             if (pool == null)
             {
                 Logger.Error?.Print(LogClass.Gpu, $"Shader stage \"{stage}\" uses images, but texture pool was not set.");
-                return;
+                return true;
             }
 
             // Scales for images appear after the texture ones.
             int baseScaleIndex = _textureBindingsCount[stageIndex];
 
+            int cachedTextureBufferIndex = -1;
+            int cachedSamplerBufferIndex = -1;
+            ReadOnlySpan<int> cachedTextureBuffer = Span<int>.Empty;
+            ReadOnlySpan<int> cachedSamplerBuffer = Span<int>.Empty;
+
+            bool specStateMatches = true;
+
             for (int index = 0; index < imageCount; index++)
             {
                 TextureBindingInfo bindingInfo = _imageBindings[stageIndex][index];
+                TextureUsageFlags usageFlags = bindingInfo.Flags;
+                int scaleIndex = baseScaleIndex + index;
 
                 (int textureBufferIndex, int samplerBufferIndex) = TextureHandle.UnpackSlots(bindingInfo.CbufSlot, _textureBufferIndex);
 
-                int packedId = ReadPackedId(stageIndex, bindingInfo.Handle, textureBufferIndex, samplerBufferIndex);
-                int textureId = UnpackTextureId(packedId);
+                UpdateCachedBuffer(stageIndex, ref cachedTextureBufferIndex, ref cachedSamplerBufferIndex, ref cachedTextureBuffer, ref cachedSamplerBuffer, textureBufferIndex, samplerBufferIndex);
 
-                Texture texture = pool.Get(textureId);
+                int packedId = TextureHandle.ReadPackedId(bindingInfo.Handle, cachedTextureBuffer, cachedSamplerBuffer);
+                int textureId = TextureHandle.UnpackTextureId(packedId);
 
-                ITexture hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
+                ref TextureState state = ref _imageState[bindingInfo.Binding];
 
                 bool isStore = bindingInfo.Flags.HasFlag(TextureUsageFlags.ImageStore);
 
+                if (!poolModified &&
+                    state.TextureHandle == textureId &&
+                    state.CachedTexture != null &&
+                    state.CachedTexture.InvalidatedSequence == state.InvalidatedSequence)
+                {
+                    Texture cachedTexture = state.CachedTexture;
+
+                    // The texture is already bound.
+                    cachedTexture.SynchronizeMemory();
+
+                    if (isStore)
+                    {
+                        cachedTexture?.SignalModified();
+                    }
+
+                    if ((state.ScaleIndex != index || state.UsageFlags != usageFlags) &&
+                        UpdateScale(state.CachedTexture, usageFlags, scaleIndex, stage))
+                    {
+                        ITexture hostTextureRebind = state.CachedTexture.GetTargetTexture(bindingInfo.Target);
+
+                        Format format = bindingInfo.Format == 0 ? cachedTexture.Format : bindingInfo.Format;
+
+                        state.Texture = hostTextureRebind;
+                        state.ScaleIndex = scaleIndex;
+                        state.UsageFlags = usageFlags;
+
+                        _context.Renderer.Pipeline.SetImage(bindingInfo.Binding, hostTextureRebind, format);
+                    }
+
+                    continue;
+                }
+
+                state.TextureHandle = textureId;
+
+                ref readonly TextureDescriptor descriptor = ref pool.GetForBinding(textureId, out Texture texture);
+
+                specStateMatches &= specState.MatchesImage(stage, index, descriptor);
+
+                ITexture hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
+
                 if (hostTexture != null && texture.Target == Target.TextureBuffer)
                 {
                     // Ensure that the buffer texture is using the correct buffer as storage.
@@ -494,14 +682,16 @@ namespace Ryujinx.Graphics.Gpu.Image
                         texture?.SignalModified();
                     }
 
-                    if (_imageState[stageIndex][index].Texture != hostTexture || _rebind)
+                    if (state.Texture != hostTexture)
                     {
-                        if (UpdateScale(texture, bindingInfo, baseScaleIndex + index, stage))
+                        if (UpdateScale(texture, usageFlags, scaleIndex, stage))
                         {
                             hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
                         }
 
-                        _imageState[stageIndex][index].Texture = hostTexture;
+                        state.Texture = hostTexture;
+                        state.ScaleIndex = scaleIndex;
+                        state.UsageFlags = usageFlags;
 
                         Format format = bindingInfo.Format;
 
@@ -512,8 +702,13 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                         _context.Renderer.Pipeline.SetImage(bindingInfo.Binding, hostTexture, format);
                     }
+
+                    state.CachedTexture = texture;
+                    state.InvalidatedSequence = texture?.InvalidatedSequence ?? 0;
                 }
             }
+
+            return specStateMatches;
         }
 
         /// <summary>
@@ -537,7 +732,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             (int textureBufferIndex, int samplerBufferIndex) = TextureHandle.UnpackSlots(cbufSlot, bufferIndex);
 
             int packedId = ReadPackedId(stageIndex, handle, textureBufferIndex, samplerBufferIndex);
-            int textureId = UnpackTextureId(packedId);
+            int textureId = TextureHandle.UnpackTextureId(packedId);
 
             ulong poolAddress = _channel.MemoryManager.Translate(poolGpuVa);
 
@@ -555,6 +750,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="textureBufferIndex">Index of the constant buffer holding the texture handles</param>
         /// <param name="samplerBufferIndex">Index of the constant buffer holding the sampler handles</param>
         /// <returns>The packed texture and sampler ID (the real texture handle)</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
         private int ReadPackedId(int stageIndex, int wordOffset, int textureBufferIndex, int samplerBufferIndex)
         {
             (int textureWordOffset, int samplerWordOffset, TextureHandleType handleType) = TextureHandle.UnpackOffsets(wordOffset);
@@ -590,32 +786,13 @@ namespace Ryujinx.Graphics.Gpu.Image
             return handle;
         }
 
-        /// <summary>
-        /// Unpacks the texture ID from the real texture handle.
-        /// </summary>
-        /// <param name="packedId">The real texture handle</param>
-        /// <returns>The texture ID</returns>
-        private static int UnpackTextureId(int packedId)
-        {
-            return (packedId >> 0) & 0xfffff;
-        }
-
-        /// <summary>
-        /// Unpacks the sampler ID from the real texture handle.
-        /// </summary>
-        /// <param name="packedId">The real texture handle</param>
-        /// <returns>The sampler ID</returns>
-        private static int UnpackSamplerId(int packedId)
-        {
-            return (packedId >> 20) & 0xfff;
-        }
-
         /// <summary>
         /// Force all bound textures and images to be rebound the next time CommitBindings is called.
         /// </summary>
         public void Rebind()
         {
-            _rebind = true;
+            Array.Clear(_textureState);
+            Array.Clear(_imageState);
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Image/TextureCache.cs

@@ -349,6 +349,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="memoryManager">GPU memory manager where the texture is mapped</param>
         /// <param name="dsState">Depth-stencil buffer texture to find or create</param>
         /// <param name="size">Size of the depth-stencil texture</param>
+        /// <param name="layered">Indicates if the texture might be accessed with a non-zero layer index</param>
         /// <param name="samplesInX">Number of samples in the X direction, for MSAA</param>
         /// <param name="samplesInY">Number of samples in the Y direction, for MSAA</param>
         /// <param name="sizeHint">A hint indicating the minimum used size for the texture</param>
@@ -357,6 +358,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             MemoryManager memoryManager,
             RtDepthStencilState dsState,
             Size3D size,
+            bool layered,
             int samplesInX,
             int samplesInY,
             Size sizeHint)
@@ -364,9 +366,24 @@ namespace Ryujinx.Graphics.Gpu.Image
             int gobBlocksInY = dsState.MemoryLayout.UnpackGobBlocksInY();
             int gobBlocksInZ = dsState.MemoryLayout.UnpackGobBlocksInZ();
 
-            Target target = (samplesInX | samplesInY) != 1
-                ? Target.Texture2DMultisample
+            Target target;
+
+            if (dsState.MemoryLayout.UnpackIsTarget3D())
+            {
+                target = Target.Texture3D;
+            }
+            else if ((samplesInX | samplesInY) != 1)
+            {
+                target = size.Depth > 1 && layered
+                    ? Target.Texture2DMultisampleArray
+                    : Target.Texture2DMultisample;
+            }
+            else
+            {
+                target = size.Depth > 1 && layered
+                    ? Target.Texture2DArray
                     : Target.Texture2D;
+            }
 
             FormatInfo formatInfo = dsState.Format.Convert();
 

Ryujinx.Graphics.Gpu/Image/TextureManager.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Engine.Types;
+using Ryujinx.Graphics.Gpu.Shader;
 using System;
 
 namespace Ryujinx.Graphics.Gpu.Image
@@ -10,9 +11,11 @@ namespace Ryujinx.Graphics.Gpu.Image
     class TextureManager : IDisposable
     {
         private readonly GpuContext _context;
+        private readonly GpuChannel _channel;
 
         private readonly TextureBindingsManager _cpBindingsManager;
         private readonly TextureBindingsManager _gpBindingsManager;
+        private readonly TexturePoolCache _texturePoolCache;
 
         private readonly Texture[] _rtColors;
         private readonly ITexture[] _rtHostColors;
@@ -35,6 +38,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         public TextureManager(GpuContext context, GpuChannel channel)
         {
             _context = context;
+            _channel = channel;
 
             TexturePoolCache texturePoolCache = new TexturePoolCache(context);
 
@@ -43,6 +47,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             _cpBindingsManager = new TextureBindingsManager(context, channel, texturePoolCache, scales, isCompute: true);
             _gpBindingsManager = new TextureBindingsManager(context, channel, texturePoolCache, scales, isCompute: false);
+            _texturePoolCache = texturePoolCache;
 
             _rtColors = new Texture[Constants.TotalRenderTargets];
             _rtHostColors = new ITexture[Constants.TotalRenderTargets];
@@ -99,6 +104,16 @@ namespace Ryujinx.Graphics.Gpu.Image
             _cpBindingsManager.SetTextureBufferIndex(index);
         }
 
+        /// <summary>
+        /// Sets the max binding indexes on the compute pipeline.
+        /// </summary>
+        /// <param name="maxTextureBinding">The maximum texture binding</param>
+        /// <param name="maxImageBinding">The maximum image binding</param>
+        public void SetComputeMaxBindings(int maxTextureBinding, int maxImageBinding)
+        {
+            _cpBindingsManager.SetMaxBindings(maxTextureBinding, maxImageBinding);
+        }
+
         /// <summary>
         /// Sets the texture constant buffer index on the graphics pipeline.
         /// </summary>
@@ -108,6 +123,16 @@ namespace Ryujinx.Graphics.Gpu.Image
             _gpBindingsManager.SetTextureBufferIndex(index);
         }
 
+        /// <summary>
+        /// Sets the max binding indexes on the graphics pipeline.
+        /// </summary>
+        /// <param name="maxTextureBinding">The maximum texture binding</param>
+        /// <param name="maxImageBinding">The maximum image binding</param>
+        public void SetGraphicsMaxBindings(int maxTextureBinding, int maxImageBinding)
+        {
+            _gpBindingsManager.SetMaxBindings(maxTextureBinding, maxImageBinding);
+        }
+
         /// <summary>
         /// Sets the current sampler pool on the compute pipeline.
         /// </summary>
@@ -335,25 +360,48 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <summary>
         /// Commits bindings on the compute pipeline.
         /// </summary>
-        public void CommitComputeBindings()
+        /// <param name="specState">Specialization state for the bound shader</param>
+        /// <returns>True if all bound textures match the current shader specialization state, false otherwise</returns>
+        public bool CommitComputeBindings(ShaderSpecializationState specState)
         {
             // Every time we switch between graphics and compute work,
             // we must rebind everything.
             // Since compute work happens less often, we always do that
             // before and after the compute dispatch.
             _cpBindingsManager.Rebind();
-            _cpBindingsManager.CommitBindings();
+            bool result = _cpBindingsManager.CommitBindings(specState);
             _gpBindingsManager.Rebind();
+
+            return result;
         }
 
         /// <summary>
         /// Commits bindings on the graphics pipeline.
         /// </summary>
-        public void CommitGraphicsBindings()
+        /// <param name="specState">Specialization state for the bound shader</param>
+        /// <returns>True if all bound textures match the current shader specialization state, false otherwise</returns>
+        public bool CommitGraphicsBindings(ShaderSpecializationState specState)
         {
-            _gpBindingsManager.CommitBindings();
+            bool result = _gpBindingsManager.CommitBindings(specState);
 
             UpdateRenderTargets();
+
+            return result;
+        }
+
+        /// <summary>
+        /// Returns a texture pool from the cache, with the given address and maximum id.
+        /// </summary>
+        /// <param name="poolGpuVa">GPU virtual address of the texture pool</param>
+        /// <param name="maximumId">Maximum ID of the texture pool</param>
+        /// <returns>The texture pool</returns>
+        public TexturePool GetTexturePool(ulong poolGpuVa, int maximumId)
+        {
+            ulong poolAddress = _channel.MemoryManager.Translate(poolGpuVa);
+
+            TexturePool texturePool = _texturePoolCache.FindOrCreate(_channel, poolAddress, maximumId);
+
+            return texturePool;
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Image/TexturePool.cs

@@ -14,6 +14,7 @@ namespace Ryujinx.Graphics.Gpu.Image
     {
         private readonly GpuChannel _channel;
         private readonly ConcurrentQueue<Texture> _dereferenceQueue = new ConcurrentQueue<Texture>();
+        private TextureDescriptor _defaultDescriptor;
 
         /// <summary>
         /// Intrusive linked list node used on the texture pool cache.
@@ -32,6 +33,62 @@ namespace Ryujinx.Graphics.Gpu.Image
             _channel = channel;
         }
 
+        /// <summary>
+        /// Gets the texture descripor and texture with the given ID with no bounds check or synchronization.
+        /// </summary>
+        /// <param name="id">ID of the texture. This is effectively a zero-based index</param>
+        /// <param name="texture">The texture with the given ID</param>
+        /// <returns>The texture descriptor with the given ID</returns>
+        private ref readonly TextureDescriptor GetInternal(int id, out Texture texture)
+        {
+            texture = Items[id];
+
+            ref readonly TextureDescriptor descriptor = ref GetDescriptorRef(id);
+
+            if (texture == null)
+            {
+                TextureInfo info = GetInfo(descriptor, out int layerSize);
+
+                ProcessDereferenceQueue();
+
+                texture = PhysicalMemory.TextureCache.FindOrCreateTexture(_channel.MemoryManager, TextureSearchFlags.ForSampler, info, layerSize);
+
+                // If this happens, then the texture address is invalid, we can't add it to the cache.
+                if (texture == null)
+                {
+                    return ref descriptor;
+                }
+
+                texture.IncrementReferenceCount(this, id);
+
+                Items[id] = texture;
+
+                DescriptorCache[id] = descriptor;
+            }
+            else
+            {
+                if (texture.ChangedSize)
+                {
+                    // Texture changed size at one point - it may be a different size than the sampler expects.
+                    // This can be triggered when the size is changed by a size hint on copy or draw, but the texture has been sampled before.
+
+                    int baseLevel = descriptor.UnpackBaseLevel();
+                    int width = Math.Max(1, descriptor.UnpackWidth() >> baseLevel);
+                    int height = Math.Max(1, descriptor.UnpackHeight() >> baseLevel);
+
+                    if (texture.Info.Width != width || texture.Info.Height != height)
+                    {
+                        texture.ChangeSize(width, height, texture.Info.DepthOrLayers);
+                    }
+                }
+
+                // Memory is automatically synchronized on texture creation.
+                texture.SynchronizeMemory();
+            }
+
+            return ref descriptor;
+        }
+
         /// <summary>
         /// Gets the texture with the given ID.
         /// </summary>
@@ -51,56 +108,49 @@ namespace Ryujinx.Graphics.Gpu.Image
                 SynchronizeMemory();
             }
 
-            Texture texture = Items[id];
-
-            if (texture == null)
-            {
-                TextureDescriptor descriptor = GetDescriptor(id);
-
-                TextureInfo info = GetInfo(descriptor, out int layerSize);
-
-                ProcessDereferenceQueue();
-
-                texture = PhysicalMemory.TextureCache.FindOrCreateTexture(_channel.MemoryManager, TextureSearchFlags.ForSampler, info, layerSize);
-
-                // If this happens, then the texture address is invalid, we can't add it to the cache.
-                if (texture == null)
-                {
-                    return null;
-                }
-
-                texture.IncrementReferenceCount(this, id);
-
-                Items[id] = texture;
-
-                DescriptorCache[id] = descriptor;
-            }
-            else
-            {
-                if (texture.ChangedSize)
-                {
-                    // Texture changed size at one point - it may be a different size than the sampler expects.
-                    // This can be triggered when the size is changed by a size hint on copy or draw, but the texture has been sampled before.
-
-                    TextureDescriptor descriptor = GetDescriptor(id);
-
-                    int baseLevel = descriptor.UnpackBaseLevel();
-                    int width = Math.Max(1, descriptor.UnpackWidth() >> baseLevel);
-                    int height = Math.Max(1, descriptor.UnpackHeight() >> baseLevel);
-
-                    if (texture.Info.Width != width || texture.Info.Height != height)
-                    {
-                        texture.ChangeSize(width, height, texture.Info.DepthOrLayers);
-                    }
-                }
-
-                // Memory is automatically synchronized on texture creation.
-                texture.SynchronizeMemory();
-            }
+            GetInternal(id, out Texture texture);
 
             return texture;
         }
 
+        /// <summary>
+        /// Gets the texture descriptor and texture with the given ID.
+        /// </summary>
+        /// <remarks>
+        /// This method assumes that the pool has been manually synchronized before doing binding.
+        /// </remarks>
+        /// <param name="id">ID of the texture. This is effectively a zero-based index</param>
+        /// <param name="texture">The texture with the given ID</param>
+        /// <returns>The texture descriptor with the given ID</returns>
+        public ref readonly TextureDescriptor GetForBinding(int id, out Texture texture)
+        {
+            if ((uint)id >= Items.Length)
+            {
+                texture = null;
+                return ref _defaultDescriptor;
+            }
+
+            // When getting for binding, assume the pool has already been synchronized.
+
+            return ref GetInternal(id, out texture);
+        }
+
+        /// <summary>
+        /// Checks if the pool was modified, and returns the last sequence number where a modification was detected.
+        /// </summary>
+        /// <returns>A number that increments each time a modification is detected</returns>
+        public int CheckModified()
+        {
+            if (SequenceNumber != Context.SequenceNumber)
+            {
+                SequenceNumber = Context.SequenceNumber;
+
+                SynchronizeMemory();
+            }
+
+            return ModifiedSequenceNumber;
+        }
+
         /// <summary>
         /// Forcibly remove a texture from this pool's items.
         /// If deferred, the dereference will be queued to occur on the render thread.
@@ -175,7 +225,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="descriptor">The texture descriptor</param>
         /// <param name="layerSize">Layer size for textures using a sub-range of mipmap levels, otherwise 0</param>
         /// <returns>The texture information</returns>
-        private TextureInfo GetInfo(TextureDescriptor descriptor, out int layerSize)
+        private TextureInfo GetInfo(in TextureDescriptor descriptor, out int layerSize)
         {
             int depthOrLayers = descriptor.UnpackDepth();
             int levels        = descriptor.UnpackLevels();

Ryujinx.Graphics.Gpu/Memory/BufferManager.cs

@@ -378,6 +378,25 @@ namespace Ryujinx.Graphics.Gpu.Memory
             return _gpUniformBuffers[stage].Buffers[index].Address;
         }
 
+        /// <summary>
+        /// Gets the bounds of the uniform buffer currently bound at the given index.
+        /// </summary>
+        /// <param name="isCompute">Indicates whenever the uniform is requested by the 3D or compute engine</param>
+        /// <param name="stage">Index of the shader stage, if the uniform is for the 3D engine</param>
+        /// <param name="index">Index of the uniform buffer binding</param>
+        /// <returns>The uniform buffer bounds, or an undefined value if the buffer is not currently bound</returns>
+        public ref BufferBounds GetUniformBufferBounds(bool isCompute, int stage, int index)
+        {
+            if (isCompute)
+            {
+                return ref _cpUniformBuffers.Buffers[index];
+            }
+            else
+            {
+                return ref _gpUniformBuffers[stage].Buffers[index];
+            }
+        }
+
         /// <summary>
         /// Ensures that the compute engine bindings are visible to the host GPU.
         /// Note: this actually performs the binding using the host graphics API.

Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs

@@ -35,6 +35,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             HostProgram = hostProgram;
             SpecializationState = specializationState;
             Shaders = shaders;
+
+            SpecializationState.Prepare(shaders);
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs

@@ -418,7 +418,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             if (IsShaderEqual(channel.MemoryManager, cpShader.Shaders[0], gpuVa))
             {
-                return cpShader.SpecializationState.MatchesCompute(channel, poolState);
+                return cpShader.SpecializationState.MatchesCompute(channel, poolState, true);
             }
 
             return false;
@@ -454,7 +454,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 }
             }
 
-            return gpShaders.SpecializationState.MatchesGraphics(channel, poolState, graphicsState);
+            return gpShaders.SpecializationState.MatchesGraphics(channel, poolState, graphicsState, true);
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs

@@ -35,7 +35,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             foreach (var entry in _entries)
             {
-                if (entry.SpecializationState.MatchesGraphics(channel, poolState, graphicsState))
+                if (entry.SpecializationState.MatchesGraphics(channel, poolState, graphicsState, true))
                 {
                     program = entry;
                     return true;
@@ -57,7 +57,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             foreach (var entry in _entries)
             {
-                if (entry.SpecializationState.MatchesCompute(channel, poolState))
+                if (entry.SpecializationState.MatchesCompute(channel, poolState, true))
                 {
                     program = entry;
                     return true;

Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs

@@ -1,9 +1,14 @@
 using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.Gpu.Image;
+using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Gpu.Shader.DiskCache;
 using Ryujinx.Graphics.Shader;
 using System;
 using System.Collections.Generic;
+using System.Linq;
 using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Gpu.Shader
 {
@@ -158,6 +163,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
         }
 
         private readonly Dictionary<TextureKey, Box<TextureSpecializationState>> _textureSpecialization;
+        private KeyValuePair<TextureKey, Box<TextureSpecializationState>>[] _allTextures;
+        private Box<TextureSpecializationState>[][] _textureByBinding;
+        private Box<TextureSpecializationState>[][] _imageByBinding;
 
         /// <summary>
         /// Creates a new instance of the shader specialization state.
@@ -194,6 +202,48 @@ namespace Ryujinx.Graphics.Gpu.Shader
             }
         }
 
+        /// <summary>
+        /// Prepare the shader specialization state for quick binding lookups.
+        /// </summary>
+        /// <param name="stages">The shader stages</param>
+        public void Prepare(CachedShaderStage[] stages)
+        {
+            _allTextures = _textureSpecialization.ToArray();
+
+            _textureByBinding = new Box<TextureSpecializationState>[stages.Length][];
+            _imageByBinding = new Box<TextureSpecializationState>[stages.Length][];
+
+            for (int i = 0; i < stages.Length; i++)
+            {
+                CachedShaderStage stage = stages[i];
+                if (stage?.Info != null)
+                {
+                    var textures = stage.Info.Textures;
+                    var images = stage.Info.Images;
+
+                    var texBindings = new Box<TextureSpecializationState>[textures.Count];
+                    var imageBindings = new Box<TextureSpecializationState>[images.Count];
+
+                    int stageIndex = Math.Max(i - 1, 0); // Don't count VertexA for looking up spec state. No-Op for compute.
+
+                    for (int j = 0; j < textures.Count; j++)
+                    {
+                        var texture = textures[j];
+                        texBindings[j] = GetTextureSpecState(stageIndex, texture.HandleIndex, texture.CbufSlot);
+                    }
+
+                    for (int j = 0; j < images.Count; j++)
+                    {
+                        var image = images[j];
+                        imageBindings[j] = GetTextureSpecState(stageIndex, image.HandleIndex, image.CbufSlot);
+                    }
+
+                    _textureByBinding[i] = texBindings;
+                    _imageByBinding[i] = imageBindings;
+                }
+            }
+        }
+
         /// <summary>
         /// Indicates that the shader accesses the early Z force state.
         /// </summary>
@@ -396,15 +446,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
         /// <param name="graphicsState">Graphics state</param>
+        /// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
         /// <returns>True if the state matches, false otherwise</returns>
-        public bool MatchesGraphics(GpuChannel channel, GpuChannelPoolState poolState, GpuChannelGraphicsState graphicsState)
+        public bool MatchesGraphics(GpuChannel channel, GpuChannelPoolState poolState, GpuChannelGraphicsState graphicsState, bool checkTextures)
         {
             if (graphicsState.ViewportTransformDisable != GraphicsState.ViewportTransformDisable)
             {
                 return false;
             }
 
-            return Matches(channel, poolState, isCompute: false);
+            return Matches(channel, poolState, checkTextures, isCompute: false);
         }
 
         /// <summary>
@@ -412,10 +463,64 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
+        /// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
         /// <returns>True if the state matches, false otherwise</returns>
-        public bool MatchesCompute(GpuChannel channel, GpuChannelPoolState poolState)
+        public bool MatchesCompute(GpuChannel channel, GpuChannelPoolState poolState, bool checkTextures)
         {
-            return Matches(channel, poolState, isCompute: true);
+            return Matches(channel, poolState, checkTextures, isCompute: true);
+        }
+
+        /// <summary>
+        /// Fetch the constant buffers used for a texture to cache.
+        /// </summary>
+        /// <param name="channel">GPU channel</param>
+        /// <param name="isCompute">Indicates whenever the check is requested by the 3D or compute engine</param>
+        /// <param name="cachedTextureBufferIndex">The currently cached texture buffer index</param>
+        /// <param name="cachedSamplerBufferIndex">The currently cached sampler buffer index</param>
+        /// <param name="cachedTextureBuffer">The currently cached texture buffer data</param>
+        /// <param name="cachedSamplerBuffer">The currently cached sampler buffer data</param>
+        /// <param name="cachedStageIndex">The currently cached stage</param>
+        /// <param name="textureBufferIndex">The new texture buffer index</param>
+        /// <param name="samplerBufferIndex">The new sampler buffer index</param>
+        /// <param name="stageIndex">Stage index of the constant buffer</param>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void UpdateCachedBuffer(
+            GpuChannel channel,
+            bool isCompute,
+            ref int cachedTextureBufferIndex,
+            ref int cachedSamplerBufferIndex,
+            ref ReadOnlySpan<int> cachedTextureBuffer,
+            ref ReadOnlySpan<int> cachedSamplerBuffer,
+            ref int cachedStageIndex,
+            int textureBufferIndex,
+            int samplerBufferIndex,
+            int stageIndex)
+        {
+            bool stageChange = stageIndex != cachedStageIndex;
+
+            if (stageChange || textureBufferIndex != cachedTextureBufferIndex)
+            {
+                ref BufferBounds bounds = ref channel.BufferManager.GetUniformBufferBounds(isCompute, stageIndex, textureBufferIndex);
+
+                cachedTextureBuffer = MemoryMarshal.Cast<byte, int>(channel.MemoryManager.Physical.GetSpan(bounds.Address, (int)bounds.Size));
+                cachedTextureBufferIndex = textureBufferIndex;
+
+                if (samplerBufferIndex == textureBufferIndex)
+                {
+                    cachedSamplerBuffer = cachedTextureBuffer;
+                    cachedSamplerBufferIndex = samplerBufferIndex;
+                }
+            }
+
+            if (stageChange || samplerBufferIndex != cachedSamplerBufferIndex)
+            {
+                ref BufferBounds bounds = ref channel.BufferManager.GetUniformBufferBounds(isCompute, stageIndex, samplerBufferIndex);
+
+                cachedSamplerBuffer = MemoryMarshal.Cast<byte, int>(channel.MemoryManager.Physical.GetSpan(bounds.Address, (int)bounds.Size));
+                cachedSamplerBufferIndex = samplerBufferIndex;
+            }
+
+            cachedStageIndex = stageIndex;
         }
 
         /// <summary>
@@ -423,9 +528,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
+        /// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
         /// <param name="isCompute">Indicates whenever the check is requested by the 3D or compute engine</param>
         /// <returns>True if the state matches, false otherwise</returns>
-        private bool Matches(GpuChannel channel, GpuChannelPoolState poolState, bool isCompute)
+        private bool Matches(GpuChannel channel, GpuChannelPoolState poolState, bool checkTextures, bool isCompute)
         {
             int constantBufferUsePerStageMask = _constantBufferUsePerStage;
 
@@ -445,55 +551,60 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 constantBufferUsePerStageMask &= ~(1 << index);
             }
 
-            foreach (var kv in _textureSpecialization)
+            if (checkTextures)
+            {
+                TexturePool pool = channel.TextureManager.GetTexturePool(poolState.TexturePoolGpuVa, poolState.TexturePoolMaximumId);
+
+                int cachedTextureBufferIndex = -1;
+                int cachedSamplerBufferIndex = -1;
+                int cachedStageIndex = -1;
+                ReadOnlySpan<int> cachedTextureBuffer = Span<int>.Empty;
+                ReadOnlySpan<int> cachedSamplerBuffer = Span<int>.Empty;
+
+                foreach (var kv in _allTextures)
                 {
                     TextureKey textureKey = kv.Key;
 
                     (int textureBufferIndex, int samplerBufferIndex) = TextureHandle.UnpackSlots(textureKey.CbufSlot, poolState.TextureBufferIndex);
 
-                ulong textureCbAddress;
-                ulong samplerCbAddress;
+                    UpdateCachedBuffer(channel,
+                        isCompute,
+                        ref cachedTextureBufferIndex,
+                        ref cachedSamplerBufferIndex,
+                        ref cachedTextureBuffer,
+                        ref cachedSamplerBuffer,
+                        ref cachedStageIndex,
+                        textureBufferIndex,
+                        samplerBufferIndex,
+                        textureKey.StageIndex);
 
-                if (isCompute)
+                    int packedId = TextureHandle.ReadPackedId(textureKey.Handle, cachedTextureBuffer, cachedSamplerBuffer);
+
+                    int textureId = TextureHandle.UnpackTextureId(packedId);
+
+                    ref readonly Image.TextureDescriptor descriptor = ref pool.GetDescriptorRef(textureId);
+
+                    if (!MatchesTexture(kv.Value, descriptor))
                     {
-                    textureCbAddress = channel.BufferManager.GetComputeUniformBufferAddress(textureBufferIndex);
-                    samplerCbAddress = channel.BufferManager.GetComputeUniformBufferAddress(samplerBufferIndex);
+                        return false;
+                    }
                 }
-                else
-                {
-                    textureCbAddress = channel.BufferManager.GetGraphicsUniformBufferAddress(textureKey.StageIndex, textureBufferIndex);
-                    samplerCbAddress = channel.BufferManager.GetGraphicsUniformBufferAddress(textureKey.StageIndex, samplerBufferIndex);
             }
 
-                if (!channel.MemoryManager.Physical.IsMapped(textureCbAddress) || !channel.MemoryManager.Physical.IsMapped(samplerCbAddress))
-                {
-                    continue;
+            return true;
         }
 
-                Image.TextureDescriptor descriptor;
-
-                if (isCompute)
+        /// <summary>
+        /// Checks if the recorded texture state matches the given texture descriptor.
+        /// </summary>
+        /// <param name="specializationState">Texture specialization state</param>
+        /// <param name="descriptor">Texture descriptor</param>
+        /// <returns>True if the state matches, false otherwise</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private bool MatchesTexture(Box<TextureSpecializationState> specializationState, in Image.TextureDescriptor descriptor)
         {
-                    descriptor = channel.TextureManager.GetComputeTextureDescriptor(
-                        poolState.TexturePoolGpuVa,
-                        poolState.TextureBufferIndex,
-                        poolState.TexturePoolMaximumId,
-                        textureKey.Handle,
-                        textureKey.CbufSlot);
-                }
-                else
+            if (specializationState != null)
             {
-                    descriptor = channel.TextureManager.GetGraphicsTextureDescriptor(
-                        poolState.TexturePoolGpuVa,
-                        poolState.TextureBufferIndex,
-                        poolState.TexturePoolMaximumId,
-                        textureKey.StageIndex,
-                        textureKey.Handle,
-                        textureKey.CbufSlot);
-                }
-
-                Box<TextureSpecializationState> specializationState = kv.Value;
-
                 if (specializationState.Value.QueriedFlags.HasFlag(QueriedTextureStateFlags.CoordNormalized) &&
                     specializationState.Value.CoordNormalized != descriptor.UnpackTextureCoordNormalized())
                 {
@@ -504,6 +615,34 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return true;
         }
 
+        /// <summary>
+        /// Checks if the recorded texture state for a given texture binding matches a texture descriptor.
+        /// </summary>
+        /// <param name="stage">The shader stage</param>
+        /// <param name="index">The texture index</param>
+        /// <param name="descriptor">Texture descriptor</param>
+        /// <returns>True if the state matches, false otherwise</returns>
+        public bool MatchesTexture(ShaderStage stage, int index, in Image.TextureDescriptor descriptor)
+        {
+            Box<TextureSpecializationState> specializationState = _textureByBinding[(int)stage][index];
+
+            return MatchesTexture(specializationState, descriptor);
+        }
+
+        /// <summary>
+        /// Checks if the recorded texture state for a given image binding matches a texture descriptor.
+        /// </summary>
+        /// <param name="stage">The shader stage</param>
+        /// <param name="index">The texture index</param>
+        /// <param name="descriptor">Texture descriptor</param>
+        /// <returns>True if the state matches, false otherwise</returns>
+        public bool MatchesImage(ShaderStage stage, int index, in Image.TextureDescriptor descriptor)
+        {
+            Box<TextureSpecializationState> specializationState = _imageByBinding[(int)stage][index];
+
+            return MatchesTexture(specializationState, descriptor);
+        }
+
         /// <summary>
         /// Reads shader specialization state that has been serialized.
         /// </summary>

Ryujinx.Graphics.OpenGL/Framebuffer.cs

@@ -9,10 +9,13 @@ namespace Ryujinx.Graphics.OpenGL
     class Framebuffer : IDisposable
     {
         public int Handle { get; private set; }
+        private int _clearFbHandle;
+        private bool _clearFbInitialized;
 
         private FramebufferAttachment _lastDsAttachment;
 
         private readonly TextureView[] _colors;
+        private TextureView _depthStencil;
 
         private int _colorsCount;
         private bool _dualSourceBlend;
@@ -20,6 +23,7 @@ namespace Ryujinx.Graphics.OpenGL
         public Framebuffer()
         {
             Handle = GL.GenFramebuffer();
+            _clearFbHandle = GL.GenFramebuffer();
 
             _colors = new TextureView[8];
         }
@@ -55,20 +59,7 @@ namespace Ryujinx.Graphics.OpenGL
 
             if (depthStencil != null)
             {
-                FramebufferAttachment attachment;
-
-                if (IsPackedDepthStencilFormat(depthStencil.Format))
-                {
-                    attachment = FramebufferAttachment.DepthStencilAttachment;
-                }
-                else if (IsDepthOnlyFormat(depthStencil.Format))
-                {
-                    attachment = FramebufferAttachment.DepthAttachment;
-                }
-                else
-                {
-                    attachment = FramebufferAttachment.StencilAttachment;
-                }
+                FramebufferAttachment attachment = GetAttachment(depthStencil.Format);
 
                 GL.FramebufferTexture(
                     FramebufferTarget.Framebuffer,
@@ -82,6 +73,8 @@ namespace Ryujinx.Graphics.OpenGL
             {
                 _lastDsAttachment = 0;
             }
+
+            _depthStencil = depthStencil;
         }
 
         public void SetDualSourceBlend(bool enable)
@@ -124,6 +117,22 @@ namespace Ryujinx.Graphics.OpenGL
             GL.DrawBuffers(colorsCount, drawBuffers);
         }
 
+        private static FramebufferAttachment GetAttachment(Format format)
+        {
+            if (IsPackedDepthStencilFormat(format))
+            {
+                return FramebufferAttachment.DepthStencilAttachment;
+            }
+            else if (IsDepthOnlyFormat(format))
+            {
+                return FramebufferAttachment.DepthAttachment;
+            }
+            else
+            {
+                return FramebufferAttachment.StencilAttachment;
+            }
+        }
+
         private static bool IsPackedDepthStencilFormat(Format format)
         {
             return format == Format.D24UnormS8Uint ||
@@ -136,6 +145,78 @@ namespace Ryujinx.Graphics.OpenGL
             return format == Format.D16Unorm || format == Format.D32Float;
         }
 
+        public void AttachColorLayerForClear(int index, int layer)
+        {
+            TextureView color = _colors[index];
+
+            if (!IsLayered(color))
+            {
+                return;
+            }
+
+            BindClearFb();
+            GL.FramebufferTextureLayer(FramebufferTarget.Framebuffer, FramebufferAttachment.ColorAttachment0 + index, color.Handle, 0, layer);
+        }
+
+        public void DetachColorLayerForClear(int index)
+        {
+            TextureView color = _colors[index];
+
+            if (!IsLayered(color))
+            {
+                return;
+            }
+
+            GL.FramebufferTexture(FramebufferTarget.Framebuffer, FramebufferAttachment.ColorAttachment0 + index, 0, 0);
+            Bind();
+        }
+
+        public void AttachDepthStencilLayerForClear(int layer)
+        {
+            TextureView depthStencil = _depthStencil;
+
+            if (!IsLayered(depthStencil))
+            {
+                return;
+            }
+
+            BindClearFb();
+            GL.FramebufferTextureLayer(FramebufferTarget.Framebuffer, GetAttachment(depthStencil.Format), depthStencil.Handle, 0, layer);
+        }
+
+        public void DetachDepthStencilLayerForClear()
+        {
+            TextureView depthStencil = _depthStencil;
+
+            if (!IsLayered(depthStencil))
+            {
+                return;
+            }
+
+            GL.FramebufferTexture(FramebufferTarget.Framebuffer, GetAttachment(depthStencil.Format), 0, 0);
+            Bind();
+        }
+
+        private void BindClearFb()
+        {
+            GL.BindFramebuffer(FramebufferTarget.Framebuffer, _clearFbHandle);
+
+            if (!_clearFbInitialized)
+            {
+                SetDrawBuffersImpl(Constants.MaxRenderTargets);
+                _clearFbInitialized = true;
+            }
+        }
+
+        private static bool IsLayered(TextureView view)
+        {
+            return view != null &&
+                   view.Target != Target.Texture1D &&
+                   view.Target != Target.Texture2D &&
+                   view.Target != Target.Texture2DMultisample &&
+                   view.Target != Target.TextureBuffer;
+        }
+
         public void Dispose()
         {
             if (Handle != 0)
@@ -144,6 +225,13 @@ namespace Ryujinx.Graphics.OpenGL
 
                 Handle = 0;
             }
+
+            if (_clearFbHandle != 0)
+            {
+                GL.DeleteFramebuffer(_clearFbHandle);
+
+                _clearFbHandle = 0;
+            }
         }
     }
 }

Ryujinx.Graphics.OpenGL/Pipeline.cs

@@ -110,7 +110,7 @@ namespace Ryujinx.Graphics.OpenGL
             Buffer.Clear(destination, offset, size, value);
         }
 
-        public void ClearRenderTargetColor(int index, uint componentMask, ColorF color)
+        public void ClearRenderTargetColor(int index, int layer, uint componentMask, ColorF color)
         {
             GL.ColorMask(
                 index,
@@ -119,14 +119,18 @@ namespace Ryujinx.Graphics.OpenGL
                 (componentMask & 4) != 0,
                 (componentMask & 8) != 0);
 
+            _framebuffer.AttachColorLayerForClear(index, layer);
+
             float[] colors = new float[] { color.Red, color.Green, color.Blue, color.Alpha };
 
             GL.ClearBuffer(OpenTK.Graphics.OpenGL.ClearBuffer.Color, index, colors);
 
+            _framebuffer.DetachColorLayerForClear(index);
+
             RestoreComponentMask(index);
         }
 
-        public void ClearRenderTargetDepthStencil(float depthValue, bool depthMask, int stencilValue, int stencilMask)
+        public void ClearRenderTargetDepthStencil(int layer, float depthValue, bool depthMask, int stencilValue, int stencilMask)
         {
             bool stencilMaskChanged =
                 stencilMask != 0 &&
@@ -144,6 +148,8 @@ namespace Ryujinx.Graphics.OpenGL
                 GL.DepthMask(depthMask);
             }
 
+            _framebuffer.AttachDepthStencilLayerForClear(layer);
+
             if (depthMask && stencilMask != 0)
             {
                 GL.ClearBuffer(ClearBufferCombined.DepthStencil, 0, depthValue, stencilValue);
@@ -157,6 +163,8 @@ namespace Ryujinx.Graphics.OpenGL
                 GL.ClearBuffer(OpenTK.Graphics.OpenGL.ClearBuffer.Stencil, 0, ref stencilValue);
             }
 
+            _framebuffer.DetachDepthStencilLayerForClear();
+
             if (stencilMaskChanged)
             {
                 GL.StencilMaskSeparate(StencilFace.Front, _stencilFrontMask);
@@ -597,6 +605,8 @@ namespace Ryujinx.Graphics.OpenGL
                         GL.EndTransformFeedback();
                     }
 
+                    GL.ClipControl(ClipOrigin.UpperLeft, ClipDepthMode.NegativeOneToOne);
+
                     _drawTexture.Draw(
                         view,
                         samp,
@@ -627,6 +637,8 @@ namespace Ryujinx.Graphics.OpenGL
                     {
                         GL.BeginTransformFeedback(_tfTopology);
                     }
+
+                    RestoreClipControl();
                 }
             }
         }

Ryujinx.Graphics.Shader/TextureHandle.cs

@@ -1,3 +1,4 @@
+using System;
 using System.Runtime.CompilerServices;
 
 namespace Ryujinx.Graphics.Shader
@@ -50,5 +51,63 @@ namespace Ryujinx.Graphics.Shader
         {
             return (handle & 0x3fff, (handle >> 14) & 0x3fff, (TextureHandleType)((uint)handle >> 28));
         }
+
+        /// <summary>
+        /// Unpacks the texture ID from the real texture handle.
+        /// </summary>
+        /// <param name="packedId">The real texture handle</param>
+        /// <returns>The texture ID</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static int UnpackTextureId(int packedId)
+        {
+            return (packedId >> 0) & 0xfffff;
+        }
+
+        /// <summary>
+        /// Unpacks the sampler ID from the real texture handle.
+        /// </summary>
+        /// <param name="packedId">The real texture handle</param>
+        /// <returns>The sampler ID</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static int UnpackSamplerId(int packedId)
+        {
+            return (packedId >> 20) & 0xfff;
+        }
+
+        /// <summary>
+        /// Reads a packed texture and sampler ID (basically, the real texture handle)
+        /// from a given texture/sampler constant buffer.
+        /// </summary>
+        /// <param name="wordOffset">A word offset of the handle on the buffer (the "fake" shader handle)</param>
+        /// <param name="cachedTextureBuffer">The constant buffer to fetch texture IDs from</param>
+        /// <param name="cachedSamplerBuffer">The constant buffer to fetch sampler IDs from</param>
+        /// <returns>The packed texture and sampler ID (the real texture handle)</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static int ReadPackedId(int wordOffset, ReadOnlySpan<int> cachedTextureBuffer, ReadOnlySpan<int> cachedSamplerBuffer)
+        {
+            (int textureWordOffset, int samplerWordOffset, TextureHandleType handleType) = UnpackOffsets(wordOffset);
+
+            int handle = cachedTextureBuffer[textureWordOffset];
+
+            // The "wordOffset" (which is really the immediate value used on texture instructions on the shader)
+            // is a 13-bit value. However, in order to also support separate samplers and textures (which uses
+            // bindless textures on the shader), we extend it with another value on the higher 16 bits with
+            // another offset for the sampler.
+            // The shader translator has code to detect separate texture and sampler uses with a bindless texture,
+            // turn that into a regular texture access and produce those special handles with values on the higher 16 bits.
+            if (handleType != TextureHandleType.CombinedSampler)
+            {
+                int samplerHandle = cachedSamplerBuffer[samplerWordOffset];
+
+                if (handleType == TextureHandleType.SeparateSamplerId)
+                {
+                    samplerHandle <<= 20;
+                }
+
+                handle |= samplerHandle;
+            }
+
+            return handle;
+        }
     }
 }

Ryujinx.Graphics.Vic/Blender.cs

@@ -10,17 +10,22 @@ namespace Ryujinx.Graphics.Vic
 {
     static class Blender
     {
-        public static void BlendOne(Surface dst, Surface src, ref SlotStruct slot)
+        public static void BlendOne(Surface dst, Surface src, ref SlotStruct slot, Rectangle targetRect)
         {
-            if (Sse41.IsSupported && (dst.Width & 3) == 0)
+            int x1 = targetRect.X;
+            int y1 = targetRect.Y;
+            int x2 = Math.Min(src.Width, x1 + targetRect.Width);
+            int y2 = Math.Min(src.Height, y1 + targetRect.Height);
+
+            if (Sse41.IsSupported && ((x1 | x2) & 3) == 0)
             {
-                BlendOneSse41(dst, src, ref slot);
+                BlendOneSse41(dst, src, ref slot, x1, y1, x2, y2);
                 return;
             }
 
-            for (int y = 0; y < dst.Height; y++)
+            for (int y = y1; y < y2; y++)
             {
-                for (int x = 0; x < dst.Width; x++)
+                for (int x = x1; x < x2; x++)
                 {
                     int inR = src.GetR(x, y);
                     int inG = src.GetG(x, y);
@@ -40,9 +45,9 @@ namespace Ryujinx.Graphics.Vic
             }
         }
 
-        private unsafe static void BlendOneSse41(Surface dst, Surface src, ref SlotStruct slot)
+        private unsafe static void BlendOneSse41(Surface dst, Surface src, ref SlotStruct slot, int x1, int y1, int x2, int y2)
         {
-            Debug.Assert((dst.Width & 3) == 0);
+            Debug.Assert(((x1 | x2) & 3) == 0);
 
             ref MatrixStruct mtx = ref slot.ColorMatrixStruct;
 
@@ -62,9 +67,9 @@ namespace Ryujinx.Graphics.Vic
                 Pixel* ip = srcPtr;
                 Pixel* op = dstPtr;
 
-                for (int y = 0; y < dst.Height; y++, ip += src.Width, op += dst.Width)
+                for (int y = y1; y < y2; y++, ip += src.Width, op += dst.Width)
                 {
-                    for (int x = 0; x < dst.Width; x += 4)
+                    for (int x = x1; x < x2; x += 4)
                     {
                         Vector128<int> pixel1 = Sse41.ConvertToVector128Int32((ushort*)(ip + (uint)x));
                         Vector128<int> pixel2 = Sse41.ConvertToVector128Int32((ushort*)(ip + (uint)x + 1));

Ryujinx.Graphics.Vic/Rectangle.cs

@@ -0,0 +1,18 @@
+namespace Ryujinx.Graphics.Vic
+{
+    struct Rectangle
+    {
+        public readonly int X;
+        public readonly int Y;
+        public readonly int Width;
+        public readonly int Height;
+
+        public Rectangle(int x, int y, int width, int height)
+        {
+            X = x;
+            Y = y;
+            Width = width;
+            Height = height;
+        }
+    }
+}

Ryujinx.Graphics.Vic/VicDevice.cs

@@ -2,6 +2,7 @@
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Vic.Image;
 using Ryujinx.Graphics.Vic.Types;
+using System;
 using System.Collections.Generic;
 
 namespace Ryujinx.Graphics.Vic
@@ -47,7 +48,19 @@ namespace Ryujinx.Graphics.Vic
 
                 using Surface src = SurfaceReader.Read(_rm, ref slot.SlotConfig, ref slot.SlotSurfaceConfig, ref offsets);
 
-                Blender.BlendOne(output, src, ref slot);
+                int x1 = config.OutputConfig.TargetRectLeft;
+                int y1 = config.OutputConfig.TargetRectTop;
+                int x2 = config.OutputConfig.TargetRectRight + 1;
+                int y2 = config.OutputConfig.TargetRectBottom + 1;
+
+                int targetX = Math.Min(x1, x2);
+                int targetY = Math.Min(y1, y2);
+                int targetW = Math.Min(output.Width - targetX, Math.Abs(x2 - x1));
+                int targetH = Math.Min(output.Height - targetY, Math.Abs(y2 - y1));
+
+                Rectangle targetRect = new Rectangle(targetX, targetY, targetW, targetH);
+
+                Blender.BlendOne(output, src, ref slot, targetRect);
             }
 
             SurfaceWriter.Write(_rm, output, ref config.OutputSurfaceConfig, ref _state.State.SetOutputSurface);

Ryujinx.HLE/HOS/Kernel/Memory/KMemoryRegionBlock.cs

@@ -1,87 +0,0 @@
-namespace Ryujinx.HLE.HOS.Kernel.Memory
-{
-    class KMemoryRegionBlock
-    {
-        public long[][] Masks;
-
-        public ulong FreeCount;
-        public int   MaxLevel;
-        public ulong StartAligned;
-        public ulong SizeInBlocksTruncated;
-        public ulong SizeInBlocksRounded;
-        public int   Order;
-        public int   NextOrder;
-
-        public bool TryCoalesce(int index, int count)
-        {
-            long mask = ((1L << count) - 1) << (index & 63);
-
-            index /= 64;
-
-            if (count >= 64)
-            {
-                int remaining = count;
-                int tempIdx = index;
-
-                do
-                {
-                    if (Masks[MaxLevel - 1][tempIdx++] != -1L)
-                    {
-                        return false;
-                    }
-
-                    remaining -= 64;
-                }
-                while (remaining != 0);
-
-                remaining = count;
-                tempIdx = index;
-
-                do
-                {
-                    Masks[MaxLevel - 1][tempIdx] = 0;
-
-                    ClearMaskBit(MaxLevel - 2, tempIdx++);
-
-                    remaining -= 64;
-                }
-                while (remaining != 0);
-            }
-            else
-            {
-                long value = Masks[MaxLevel - 1][index];
-
-                if ((mask & ~value) != 0)
-                {
-                    return false;
-                }
-
-                value &= ~mask;
-
-                Masks[MaxLevel - 1][index] = value;
-
-                if (value == 0)
-                {
-                    ClearMaskBit(MaxLevel - 2, index);
-                }
-            }
-
-            FreeCount -= (ulong)count;
-
-            return true;
-        }
-
-        public void ClearMaskBit(int startLevel, int index)
-        {
-            for (int level = startLevel; level >= 0; level--, index /= 64)
-            {
-                Masks[level][index / 64] &= ~(1L << (index & 63));
-
-                if (Masks[level][index / 64] != 0)
-                {
-                    break;
-                }
-            }
-        }
-    }
-}

Ryujinx.HLE/HOS/Kernel/Memory/KMemoryRegionManager.cs

@@ -1,102 +1,42 @@
-using Ryujinx.Common;
 using Ryujinx.HLE.HOS.Kernel.Common;
 using System.Diagnostics;
-using System.Numerics;
 
 namespace Ryujinx.HLE.HOS.Kernel.Memory
 {
     class KMemoryRegionManager
     {
-        private static readonly int[] BlockOrders = new int[] { 12, 16, 21, 22, 25, 29, 30 };
+        private readonly KPageHeap _pageHeap;
 
-        public ulong Address { get; private set; }
-        public ulong EndAddr { get; private set; }
-        public ulong Size    { get; private set; }
-
-        private int _blockOrdersCount;
-
-        private readonly KMemoryRegionBlock[] _blocks;
+        public ulong Address { get; }
+        public ulong Size { get; }
+        public ulong EndAddr => Address + Size;
 
         private readonly ushort[] _pageReferenceCounts;
 
         public KMemoryRegionManager(ulong address, ulong size, ulong endAddr)
         {
-            _blocks = new KMemoryRegionBlock[BlockOrders.Length];
-
             Address = address;
             Size = size;
-            EndAddr = endAddr;
-
-            _blockOrdersCount = BlockOrders.Length;
-
-            for (int blockIndex = 0; blockIndex < _blockOrdersCount; blockIndex++)
-            {
-                _blocks[blockIndex] = new KMemoryRegionBlock();
-
-                _blocks[blockIndex].Order = BlockOrders[blockIndex];
-
-                int nextOrder = blockIndex == _blockOrdersCount - 1 ? 0 : BlockOrders[blockIndex + 1];
-
-                _blocks[blockIndex].NextOrder = nextOrder;
-
-                int currBlockSize = 1 << BlockOrders[blockIndex];
-                int nextBlockSize = currBlockSize;
-
-                if (nextOrder != 0)
-                {
-                    nextBlockSize = 1 << nextOrder;
-                }
-
-                ulong startAligned   = BitUtils.AlignDown(address, nextBlockSize);
-                ulong endAddrAligned = BitUtils.AlignDown(endAddr, currBlockSize);
-
-                ulong sizeInBlocksTruncated = (endAddrAligned - startAligned) >> BlockOrders[blockIndex];
-
-                ulong endAddrRounded = BitUtils.AlignUp(address + size, nextBlockSize);
-
-                ulong sizeInBlocksRounded = (endAddrRounded - startAligned) >> BlockOrders[blockIndex];
-
-                _blocks[blockIndex].StartAligned          = startAligned;
-                _blocks[blockIndex].SizeInBlocksTruncated = sizeInBlocksTruncated;
-                _blocks[blockIndex].SizeInBlocksRounded   = sizeInBlocksRounded;
-
-                ulong currSizeInBlocks = sizeInBlocksRounded;
-
-                int maxLevel = 0;
-
-                do
-                {
-                    maxLevel++;
-                }
-                while ((currSizeInBlocks /= 64) != 0);
-
-                _blocks[blockIndex].MaxLevel = maxLevel;
-
-                _blocks[blockIndex].Masks = new long[maxLevel][];
-
-                currSizeInBlocks = sizeInBlocksRounded;
-
-                for (int level = maxLevel - 1; level >= 0; level--)
-                {
-                    currSizeInBlocks = (currSizeInBlocks + 63) / 64;
-
-                    _blocks[blockIndex].Masks[level] = new long[currSizeInBlocks];
-                }
-            }
 
             _pageReferenceCounts = new ushort[size / KPageTableBase.PageSize];
 
-            if (size != 0)
-            {
-                FreePages(address, size / KPageTableBase.PageSize);
-            }
+            _pageHeap = new KPageHeap(address, size);
+            _pageHeap.Free(address, size / KPageTableBase.PageSize);
+            _pageHeap.UpdateUsedSize();
         }
 
-        public KernelResult AllocatePages(ulong pagesCount, bool backwards, out KPageList pageList)
+        public KernelResult AllocatePages(out KPageList pageList, ulong pagesCount)
         {
-            lock (_blocks)
+            if (pagesCount == 0)
             {
-                KernelResult result = AllocatePagesImpl(pagesCount, backwards, out pageList);
+                pageList = new KPageList();
+
+                return KernelResult.Success;
+            }
+
+            lock (_pageHeap)
+            {
+                KernelResult result = AllocatePagesImpl(out pageList, pagesCount, false);
 
                 if (result == KernelResult.Success)
                 {
@@ -112,9 +52,14 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
 
         public ulong AllocatePagesContiguous(KernelContext context, ulong pagesCount, bool backwards)
         {
-            lock (_blocks)
+            if (pagesCount == 0)
             {
-                ulong address = AllocatePagesContiguousImpl(pagesCount, backwards);
+                return 0;
+            }
+
+            lock (_pageHeap)
+            {
+                ulong address = AllocatePagesContiguousImpl(pagesCount, 1, backwards);
 
                 if (address != 0)
                 {
@@ -126,373 +71,110 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
             }
         }
 
-        private KernelResult AllocatePagesImpl(ulong pagesCount, bool backwards, out KPageList pageList)
+        private KernelResult AllocatePagesImpl(out KPageList pageList, ulong pagesCount, bool random)
         {
             pageList = new KPageList();
 
-            if (_blockOrdersCount > 0)
-            {
-                if (GetFreePagesImpl() < pagesCount)
-                {
-                    return KernelResult.OutOfMemory;
-                }
-            }
-            else if (pagesCount != 0)
+            int heapIndex = KPageHeap.GetBlockIndex(pagesCount);
+
+            if (heapIndex < 0)
             {
                 return KernelResult.OutOfMemory;
             }
 
-            for (int blockIndex = _blockOrdersCount - 1; blockIndex >= 0; blockIndex--)
+            for (int index = heapIndex; index >= 0; index--)
             {
-                KMemoryRegionBlock block = _blocks[blockIndex];
+                ulong pagesPerAlloc = KPageHeap.GetBlockPagesCount(index);
 
-                ulong bestFitBlockSize = 1UL << block.Order;
-
-                ulong blockPagesCount = bestFitBlockSize / KPageTableBase.PageSize;
-
-                // Check if this is the best fit for this page size.
-                // If so, try allocating as much requested pages as possible.
-                while (blockPagesCount <= pagesCount)
+                while (pagesCount >= pagesPerAlloc)
                 {
-                    ulong address = AllocatePagesForOrder(blockIndex, backwards, bestFitBlockSize);
+                    ulong allocatedBlock = _pageHeap.AllocateBlock(index, random);
 
-                    // The address being zero means that no free space was found on that order,
-                    // just give up and try with the next one.
-                    if (address == 0)
+                    if (allocatedBlock == 0)
                     {
                         break;
                     }
 
-                    // Add new allocated page(s) to the pages list.
-                    // If an error occurs, then free all allocated pages and fail.
-                    KernelResult result = pageList.AddRange(address, blockPagesCount);
+                    KernelResult result = pageList.AddRange(allocatedBlock, pagesPerAlloc);
 
                     if (result != KernelResult.Success)
                     {
-                        FreePages(address, blockPagesCount);
-
-                        foreach (KPageNode pageNode in pageList)
-                        {
-                            FreePages(pageNode.Address, pageNode.PagesCount);
-                        }
+                        FreePages(pageList);
+                        _pageHeap.Free(allocatedBlock, pagesPerAlloc);
 
                         return result;
                     }
 
-                    pagesCount -= blockPagesCount;
+                    pagesCount -= pagesPerAlloc;
                 }
             }
 
-            // Success case, all requested pages were allocated successfully.
-            if (pagesCount == 0)
+            if (pagesCount != 0)
             {
-                return KernelResult.Success;
-            }
-
-            // Error case, free allocated pages and return out of memory.
-            foreach (KPageNode pageNode in pageList)
-            {
-                FreePages(pageNode.Address, pageNode.PagesCount);
-            }
-
-            pageList = null;
+                FreePages(pageList);
 
                 return KernelResult.OutOfMemory;
             }
 
-        private ulong AllocatePagesContiguousImpl(ulong pagesCount, bool backwards)
+            return KernelResult.Success;
+        }
+
+        private ulong AllocatePagesContiguousImpl(ulong pagesCount, ulong alignPages, bool random)
         {
-            if (pagesCount == 0 || _blocks.Length < 1)
+            int heapIndex = KPageHeap.GetAlignedBlockIndex(pagesCount, alignPages);
+
+            ulong allocatedBlock = _pageHeap.AllocateBlock(heapIndex, random);
+
+            if (allocatedBlock == 0)
             {
                 return 0;
             }
 
-            int blockIndex = 0;
+            ulong allocatedPages = KPageHeap.GetBlockPagesCount(heapIndex);
 
-            while ((1UL << _blocks[blockIndex].Order) / KPageTableBase.PageSize < pagesCount)
+            if (allocatedPages > pagesCount)
             {
-                if (++blockIndex >= _blocks.Length)
+                _pageHeap.Free(allocatedBlock + pagesCount * KPageTableBase.PageSize, allocatedPages - pagesCount);
+            }
+
+            return allocatedBlock;
+        }
+
+        public void FreePage(ulong address)
         {
-                    return 0;
+            lock (_pageHeap)
+            {
+                _pageHeap.Free(address, 1);
             }
         }
 
-            ulong tightestFitBlockSize = 1UL << _blocks[blockIndex].Order;
-
-            ulong address = AllocatePagesForOrder(blockIndex, backwards, tightestFitBlockSize);
-
-            ulong requiredSize = pagesCount * KPageTableBase.PageSize;
-
-            if (address != 0 && tightestFitBlockSize > requiredSize)
+        public void FreePages(KPageList pageList)
         {
-                FreePages(address + requiredSize, (tightestFitBlockSize - requiredSize) / KPageTableBase.PageSize);
+            lock (_pageHeap)
+            {
+                foreach (KPageNode pageNode in pageList)
+                {
+                    _pageHeap.Free(pageNode.Address, pageNode.PagesCount);
                 }
-
-            return address;
-        }
-
-        private ulong AllocatePagesForOrder(int blockIndex, bool backwards, ulong bestFitBlockSize)
-        {
-            ulong address = 0;
-
-            KMemoryRegionBlock block = null;
-
-            for (int currBlockIndex = blockIndex;
-                     currBlockIndex < _blockOrdersCount && address == 0;
-                     currBlockIndex++)
-            {
-                block = _blocks[currBlockIndex];
-
-                int index = 0;
-
-                bool zeroMask = false;
-
-                for (int level = 0; level < block.MaxLevel; level++)
-                {
-                    long mask = block.Masks[level][index];
-
-                    if (mask == 0)
-                    {
-                        zeroMask = true;
-
-                        break;
-                    }
-
-                    if (backwards)
-                    {
-                        index = (index * 64 + 63) - BitOperations.LeadingZeroCount((ulong)mask);
-                    }
-                    else
-                    {
-                        index = index * 64 + BitOperations.LeadingZeroCount((ulong)BitUtils.ReverseBits64(mask));
             }
         }
 
-                if (block.SizeInBlocksTruncated <= (ulong)index || zeroMask)
+        public void FreePages(ulong address, ulong pagesCount)
         {
-                    continue;
-                }
-
-                block.FreeCount--;
-
-                int tempIdx = index;
-
-                for (int level = block.MaxLevel - 1; level >= 0; level--, tempIdx /= 64)
+            lock (_pageHeap)
             {
-                    block.Masks[level][tempIdx / 64] &= ~(1L << (tempIdx & 63));
-
-                    if (block.Masks[level][tempIdx / 64] != 0)
-                    {
-                        break;
-                    }
-                }
-
-                address = block.StartAligned + ((ulong)index << block.Order);
-            }
-
-            for (int currBlockIndex = blockIndex;
-                     currBlockIndex < _blockOrdersCount && address == 0;
-                     currBlockIndex++)
-            {
-                block = _blocks[currBlockIndex];
-
-                int index = 0;
-
-                bool zeroMask = false;
-
-                for (int level = 0; level < block.MaxLevel; level++)
-                {
-                    long mask = block.Masks[level][index];
-
-                    if (mask == 0)
-                    {
-                        zeroMask = true;
-
-                        break;
-                    }
-
-                    if (backwards)
-                    {
-                        index = index * 64 + BitOperations.LeadingZeroCount((ulong)BitUtils.ReverseBits64(mask));
-                    }
-                    else
-                    {
-                        index = (index * 64 + 63) - BitOperations.LeadingZeroCount((ulong)mask);
-                    }
-                }
-
-                if (block.SizeInBlocksTruncated <= (ulong)index || zeroMask)
-                {
-                    continue;
-                }
-
-                block.FreeCount--;
-
-                int tempIdx = index;
-
-                for (int level = block.MaxLevel - 1; level >= 0; level--, tempIdx /= 64)
-                {
-                    block.Masks[level][tempIdx / 64] &= ~(1L << (tempIdx & 63));
-
-                    if (block.Masks[level][tempIdx / 64] != 0)
-                    {
-                        break;
-                    }
-                }
-
-                address = block.StartAligned + ((ulong)index << block.Order);
-            }
-
-            if (address != 0)
-            {
-                // If we are using a larger order than best fit, then we should
-                // split it into smaller blocks.
-                ulong firstFreeBlockSize = 1UL << block.Order;
-
-                if (firstFreeBlockSize > bestFitBlockSize)
-                {
-                    FreePages(address + bestFitBlockSize, (firstFreeBlockSize - bestFitBlockSize) / KPageTableBase.PageSize);
-                }
-            }
-
-            return address;
-        }
-
-        private void FreePages(ulong address, ulong pagesCount)
-        {
-            lock (_blocks)
-            {
-                ulong endAddr = address + pagesCount * KPageTableBase.PageSize;
-
-                int blockIndex = _blockOrdersCount - 1;
-
-                ulong addressRounded   = 0;
-                ulong endAddrTruncated = 0;
-
-                for (; blockIndex >= 0; blockIndex--)
-                {
-                    KMemoryRegionBlock allocInfo = _blocks[blockIndex];
-
-                    int blockSize = 1 << allocInfo.Order;
-
-                    addressRounded   = BitUtils.AlignUp  (address, blockSize);
-                    endAddrTruncated = BitUtils.AlignDown(endAddr, blockSize);
-
-                    if (addressRounded < endAddrTruncated)
-                    {
-                        break;
-                    }
-                }
-
-                void FreeRegion(ulong currAddress)
-                {
-                    for (int currBlockIndex = blockIndex;
-                            currBlockIndex < _blockOrdersCount && currAddress != 0;
-                            currBlockIndex++)
-                    {
-                        KMemoryRegionBlock block = _blocks[currBlockIndex];
-
-                        block.FreeCount++;
-
-                        ulong freedBlocks = (currAddress - block.StartAligned) >> block.Order;
-
-                        int index = (int)freedBlocks;
-
-                        for (int level = block.MaxLevel - 1; level >= 0; level--, index /= 64)
-                        {
-                            long mask = block.Masks[level][index / 64];
-
-                            block.Masks[level][index / 64] = mask | (1L << (index & 63));
-
-                            if (mask != 0)
-                            {
-                                break;
-                            }
-                        }
-
-                        int blockSizeDelta = 1 << (block.NextOrder - block.Order);
-
-                        int freedBlocksTruncated = BitUtils.AlignDown((int)freedBlocks, blockSizeDelta);
-
-                        if (!block.TryCoalesce(freedBlocksTruncated, blockSizeDelta))
-                        {
-                            break;
-                        }
-
-                        currAddress = block.StartAligned + ((ulong)freedBlocksTruncated << block.Order);
-                    }
-                }
-
-                // Free inside aligned region.
-                ulong baseAddress = addressRounded;
-
-                while (baseAddress < endAddrTruncated)
-                {
-                    ulong blockSize = 1UL << _blocks[blockIndex].Order;
-
-                    FreeRegion(baseAddress);
-
-                    baseAddress += blockSize;
-                }
-
-                int nextBlockIndex = blockIndex - 1;
-
-                // Free region between Address and aligned region start.
-                baseAddress = addressRounded;
-
-                for (blockIndex = nextBlockIndex; blockIndex >= 0; blockIndex--)
-                {
-                    ulong blockSize = 1UL << _blocks[blockIndex].Order;
-
-                    while (baseAddress - blockSize >= address)
-                    {
-                        baseAddress -= blockSize;
-
-                        FreeRegion(baseAddress);
-                    }
-                }
-
-                // Free region between aligned region end and End Address.
-                baseAddress = endAddrTruncated;
-
-                for (blockIndex = nextBlockIndex; blockIndex >= 0; blockIndex--)
-                {
-                    ulong blockSize = 1UL << _blocks[blockIndex].Order;
-
-                    while (baseAddress + blockSize <= endAddr)
-                    {
-                        FreeRegion(baseAddress);
-
-                        baseAddress += blockSize;
-                    }
-                }
+                _pageHeap.Free(address, pagesCount);
             }
         }
 
         public ulong GetFreePages()
         {
-            lock (_blocks)
+            lock (_pageHeap)
             {
-                return GetFreePagesImpl();
+                return _pageHeap.GetFreePagesCount();
             }
         }
 
-        private ulong GetFreePagesImpl()
-        {
-            ulong availablePages = 0;
-
-            for (int blockIndex = 0; blockIndex < _blockOrdersCount; blockIndex++)
-            {
-                KMemoryRegionBlock block = _blocks[blockIndex];
-
-                ulong blockPagesCount = (1UL << block.Order) / KPageTableBase.PageSize;
-
-                availablePages += blockPagesCount * block.FreeCount;
-            }
-
-            return availablePages;
-        }
-
         public void IncrementPagesReferenceCount(ulong address, ulong pagesCount)
         {
             ulong index = GetPageOffset(address);

Ryujinx.HLE/HOS/Kernel/Memory/KPageBitmap.cs

@@ -0,0 +1,298 @@
+using Ryujinx.Common;
+using System;
+using System.Numerics;
+
+namespace Ryujinx.HLE.HOS.Kernel.Memory
+{
+    class KPageBitmap
+    {
+        private struct RandomNumberGenerator
+        {
+            private uint _entropy;
+            private uint _bitsAvailable;
+
+            private void RefreshEntropy()
+            {
+                _entropy = 0;
+                _bitsAvailable = sizeof(uint) * 8;
+            }
+
+            private bool GenerateRandomBit()
+            {
+                if (_bitsAvailable == 0)
+                {
+                    RefreshEntropy();
+                }
+
+                bool bit = (_entropy & 1) != 0;
+
+                _entropy >>= 1;
+                _bitsAvailable--;
+
+                return bit;
+            }
+
+            public int SelectRandomBit(ulong bitmap)
+            {
+                int selected = 0;
+
+                int bitsCount = UInt64BitSize / 2;
+                ulong mask = (1UL << bitsCount) - 1;
+
+                while (bitsCount != 0)
+                {
+                    ulong low = bitmap & mask;
+                    ulong high = (bitmap >> bitsCount) & mask;
+
+                    bool chooseLow;
+
+                    if (high == 0)
+                    {
+                        chooseLow = true;
+                    }
+                    else if (low == 0)
+                    {
+                        chooseLow = false;
+                    }
+                    else
+                    {
+                        chooseLow = GenerateRandomBit();
+                    }
+
+                    if (chooseLow)
+                    {
+                        bitmap = low;
+                    }
+                    else
+                    {
+                        bitmap = high;
+                        selected += bitsCount;
+                    }
+
+                    bitsCount /= 2;
+                    mask >>= bitsCount;
+                }
+
+                return selected;
+            }
+        }
+
+        private const int UInt64BitSize = sizeof(ulong) * 8;
+        private const int MaxDepth = 4;
+
+        private readonly RandomNumberGenerator _rng;
+        private readonly ArraySegment<ulong>[] _bitStorages;
+        private int _usedDepths;
+
+        public int BitsCount { get; private set; }
+
+        public int HighestDepthIndex => _usedDepths - 1;
+
+        public KPageBitmap()
+        {
+            _rng = new RandomNumberGenerator();
+            _bitStorages = new ArraySegment<ulong>[MaxDepth];
+        }
+
+        public ArraySegment<ulong> Initialize(ArraySegment<ulong> storage, ulong size)
+        {
+            _usedDepths = GetRequiredDepth(size);
+
+            for (int depth = HighestDepthIndex; depth >= 0; depth--)
+            {
+                _bitStorages[depth] = storage;
+                size = BitUtils.DivRoundUp(size, UInt64BitSize);
+                storage = storage.Slice((int)size);
+            }
+
+            return storage;
+        }
+
+        public ulong FindFreeBlock(bool random)
+        {
+            ulong offset = 0;
+            int depth = 0;
+
+            if (random)
+            {
+                do
+                {
+                    ulong v = _bitStorages[depth][(int)offset];
+
+                    if (v == 0)
+                    {
+                        return ulong.MaxValue;
+                    }
+
+                    offset = offset * UInt64BitSize + (ulong)_rng.SelectRandomBit(v);
+                }
+                while (++depth < _usedDepths);
+            }
+            else
+            {
+                do
+                {
+                    ulong v = _bitStorages[depth][(int)offset];
+
+                    if (v == 0)
+                    {
+                        return ulong.MaxValue;
+                    }
+
+                    offset = offset * UInt64BitSize + (ulong)BitOperations.TrailingZeroCount(v);
+                }
+                while (++depth < _usedDepths);
+            }
+
+            return offset;
+        }
+
+        public void SetBit(ulong offset)
+        {
+            SetBit(HighestDepthIndex, offset);
+            BitsCount++;
+        }
+
+        public void ClearBit(ulong offset)
+        {
+            ClearBit(HighestDepthIndex, offset);
+            BitsCount--;
+        }
+
+        public bool ClearRange(ulong offset, int count)
+        {
+            int depth = HighestDepthIndex;
+            var bits = _bitStorages[depth];
+
+            int bitInd = (int)(offset / UInt64BitSize);
+
+            if (count < UInt64BitSize)
+            {
+                int shift = (int)(offset % UInt64BitSize);
+
+                ulong mask = ((1UL << count) - 1) << shift;
+
+                ulong v = bits[bitInd];
+
+                if ((v & mask) != mask)
+                {
+                    return false;
+                }
+
+                v &= ~mask;
+                bits[bitInd] = v;
+
+                if (v == 0)
+                {
+                    ClearBit(depth - 1, (ulong)bitInd);
+                }
+            }
+            else
+            {
+                int remaining = count;
+                int i = 0;
+
+                do
+                {
+                    if (bits[bitInd + i++] != ulong.MaxValue)
+                    {
+                        return false;
+                    }
+
+                    remaining -= UInt64BitSize;
+                }
+                while (remaining > 0);
+
+                remaining = count;
+                i = 0;
+
+                do
+                {
+                    bits[bitInd + i] = 0;
+                    ClearBit(depth - 1, (ulong)(bitInd + i));
+                    i++;
+                    remaining -= UInt64BitSize;
+                }
+                while (remaining > 0);
+            }
+
+            BitsCount -= count;
+            return true;
+        }
+
+        private void SetBit(int depth, ulong offset)
+        {
+            while (depth >= 0)
+            {
+                int ind   = (int)(offset / UInt64BitSize);
+                int which = (int)(offset % UInt64BitSize);
+
+                ulong mask = 1UL << which;
+
+                ulong v = _bitStorages[depth][ind];
+
+                _bitStorages[depth][ind] = v | mask;
+
+                if (v != 0)
+                {
+                    break;
+                }
+
+                offset = (ulong)ind;
+                depth--;
+            }
+        }
+
+        private void ClearBit(int depth, ulong offset)
+        {
+            while (depth >= 0)
+            {
+                int ind   = (int)(offset / UInt64BitSize);
+                int which = (int)(offset % UInt64BitSize);
+
+                ulong mask = 1UL << which;
+
+                ulong v = _bitStorages[depth][ind];
+
+                v &= ~mask;
+
+                _bitStorages[depth][ind] = v;
+
+                if (v != 0)
+                {
+                    break;
+                }
+
+                offset = (ulong)ind;
+                depth--;
+            }
+        }
+
+        private static int GetRequiredDepth(ulong regionSize)
+        {
+            int depth = 0;
+
+            do
+            {
+                regionSize /= UInt64BitSize;
+                depth++;
+            }
+            while (regionSize != 0);
+
+            return depth;
+        }
+
+        public static int CalculateManagementOverheadSize(ulong regionSize)
+        {
+            int overheadBits = 0;
+
+            for (int depth = GetRequiredDepth(regionSize) - 1; depth >= 0; depth--)
+            {
+                regionSize = BitUtils.DivRoundUp(regionSize, UInt64BitSize);
+                overheadBits += (int)regionSize;
+            }
+
+            return overheadBits * sizeof(ulong);
+        }
+    }
+}

Ryujinx.HLE/HOS/Kernel/Memory/KPageHeap.cs

@@ -0,0 +1,283 @@
+using Ryujinx.Common;
+using System;
+
+namespace Ryujinx.HLE.HOS.Kernel.Memory
+{
+    class KPageHeap
+    {
+        private class Block
+        {
+            private KPageBitmap _bitmap = new KPageBitmap();
+            private ulong _heapAddress;
+            private ulong _endOffset;
+
+            public int Shift { get; private set; }
+            public int NextShift { get; private set; }
+            public ulong Size => 1UL << Shift;
+            public int PagesCount => (int)(Size / KPageTableBase.PageSize);
+            public int FreeBlocksCount => _bitmap.BitsCount;
+            public int FreePagesCount => FreeBlocksCount * PagesCount;
+
+            public ArraySegment<ulong> Initialize(ulong address, ulong size, int blockShift, int nextBlockShift, ArraySegment<ulong> bitStorage)
+            {
+                Shift = blockShift;
+                NextShift = nextBlockShift;
+
+                ulong endAddress = address + size;
+
+                ulong align = nextBlockShift != 0
+                    ? 1UL << nextBlockShift
+                    : 1UL << blockShift;
+
+                address    = BitUtils.AlignDown(address,    align);
+                endAddress = BitUtils.AlignUp  (endAddress, align);
+
+                _heapAddress = address;
+                _endOffset = (endAddress - address) / (1UL << blockShift);
+
+                return _bitmap.Initialize(bitStorage, _endOffset);
+            }
+
+            public ulong PushBlock(ulong address)
+            {
+                ulong offset = (address - _heapAddress) >> Shift;
+
+                _bitmap.SetBit(offset);
+
+                if (NextShift != 0)
+                {
+                    int diff = 1 << (NextShift - Shift);
+
+                    offset = BitUtils.AlignDown(offset, diff);
+
+                    if (_bitmap.ClearRange(offset, diff))
+                    {
+                        return _heapAddress + (offset << Shift);
+                    }
+                }
+
+                return 0;
+            }
+
+            public ulong PopBlock(bool random)
+            {
+                long sOffset = (long)_bitmap.FindFreeBlock(random);
+
+                if (sOffset < 0L)
+                {
+                    return 0;
+                }
+
+                ulong offset = (ulong)sOffset;
+
+                _bitmap.ClearBit(offset);
+
+                return _heapAddress + (offset << Shift);
+            }
+
+            public static int CalculateManagementOverheadSize(ulong regionSize, int currBlockShift, int nextBlockShift)
+            {
+                ulong currBlockSize = 1UL << currBlockShift;
+                ulong nextBlockSize = 1UL << nextBlockShift;
+                ulong align = nextBlockShift != 0 ? nextBlockSize : currBlockSize;
+                return KPageBitmap.CalculateManagementOverheadSize((align * 2 + BitUtils.AlignUp(regionSize, align)) / currBlockSize);
+            }
+        }
+
+        private static readonly int[] _memoryBlockPageShifts = new int[] { 12, 16, 21, 22, 25, 29, 30 };
+
+        private readonly ulong _heapAddress;
+        private readonly ulong _heapSize;
+        private ulong _usedSize;
+        private readonly int _blocksCount;
+        private readonly Block[] _blocks;
+
+        public KPageHeap(ulong address, ulong size) : this(address, size, _memoryBlockPageShifts)
+        {
+        }
+
+        public KPageHeap(ulong address, ulong size, int[] blockShifts)
+        {
+            _heapAddress = address;
+            _heapSize = size;
+            _blocksCount = blockShifts.Length;
+            _blocks = new Block[_memoryBlockPageShifts.Length];
+
+            var currBitmapStorage = new ArraySegment<ulong>(new ulong[CalculateManagementOverheadSize(size, blockShifts)]);
+
+            for (int i = 0; i < blockShifts.Length; i++)
+            {
+                int currBlockShift = blockShifts[i];
+                int nextBlockShift = i != blockShifts.Length - 1 ? blockShifts[i + 1] : 0;
+
+                _blocks[i] = new Block();
+
+                currBitmapStorage = _blocks[i].Initialize(address, size, currBlockShift, nextBlockShift, currBitmapStorage);
+            }
+        }
+
+        public void UpdateUsedSize()
+        {
+            _usedSize = _heapSize - (GetFreePagesCount() * KPageTableBase.PageSize);
+        }
+
+        public ulong GetFreePagesCount()
+        {
+            ulong freeCount = 0;
+
+            for (int i = 0; i < _blocksCount; i++)
+            {
+                freeCount += (ulong)_blocks[i].FreePagesCount;
+            }
+
+            return freeCount;
+        }
+
+        public ulong AllocateBlock(int index, bool random)
+        {
+            ulong neededSize = _blocks[index].Size;
+
+            for (int i = index; i < _blocksCount; i++)
+            {
+                ulong address = _blocks[i].PopBlock(random);
+
+                if (address != 0)
+                {
+                    ulong allocatedSize = _blocks[i].Size;
+
+                    if (allocatedSize > neededSize)
+                    {
+                        Free(address + neededSize, (allocatedSize - neededSize) / KPageTableBase.PageSize);
+                    }
+
+                    return address;
+                }
+            }
+
+            return 0;
+        }
+
+        private void FreeBlock(ulong block, int index)
+        {
+            do
+            {
+                block = _blocks[index++].PushBlock(block);
+            }
+            while (block != 0);
+        }
+
+        public void Free(ulong address, ulong pagesCount)
+        {
+            if (pagesCount == 0)
+            {
+                return;
+            }
+
+            int bigIndex = _blocksCount - 1;
+
+            ulong start       = address;
+            ulong end         = address + pagesCount * KPageTableBase.PageSize;
+            ulong beforeStart = start;
+            ulong beforeEnd   = start;
+            ulong afterStart  = end;
+            ulong afterEnd    = end;
+
+            while (bigIndex >= 0)
+            {
+                ulong blockSize = _blocks[bigIndex].Size;
+
+                ulong bigStart = BitUtils.AlignUp  (start, blockSize);
+                ulong bigEnd   = BitUtils.AlignDown(end,   blockSize);
+
+                if (bigStart < bigEnd)
+                {
+                    for (ulong block = bigStart; block < bigEnd; block += blockSize)
+                    {
+                        FreeBlock(block, bigIndex);
+                    }
+
+                    beforeEnd  = bigStart;
+                    afterStart = bigEnd;
+
+                    break;
+                }
+
+                bigIndex--;
+            }
+
+            for (int i = bigIndex - 1; i >= 0; i--)
+            {
+                ulong blockSize = _blocks[i].Size;
+
+                while (beforeStart + blockSize <= beforeEnd)
+                {
+                    beforeEnd -= blockSize;
+                    FreeBlock(beforeEnd, i);
+                }
+            }
+
+            for (int i = bigIndex - 1; i >= 0; i--)
+            {
+                ulong blockSize = _blocks[i].Size;
+
+                while (afterStart + blockSize <= afterEnd)
+                {
+                    FreeBlock(afterStart, i);
+                    afterStart += blockSize;
+                }
+            }
+        }
+
+        public static int GetAlignedBlockIndex(ulong pagesCount, ulong alignPages)
+        {
+            ulong targetPages = Math.Max(pagesCount, alignPages);
+
+            for (int i = 0; i < _memoryBlockPageShifts.Length; i++)
+            {
+                if (targetPages <= GetBlockPagesCount(i))
+                {
+                    return i;
+                }
+            }
+
+            return -1;
+        }
+
+        public static int GetBlockIndex(ulong pagesCount)
+        {
+            for (int i = _memoryBlockPageShifts.Length - 1; i >= 0; i--)
+            {
+                if (pagesCount >= GetBlockPagesCount(i))
+                {
+                    return i;
+                }
+            }
+
+            return -1;
+        }
+
+        public static ulong GetBlockSize(int index)
+        {
+            return 1UL << _memoryBlockPageShifts[index];
+        }
+
+        public static ulong GetBlockPagesCount(int index)
+        {
+            return GetBlockSize(index) / KPageTableBase.PageSize;
+        }
+
+        private static int CalculateManagementOverheadSize(ulong regionSize, int[] blockShifts)
+        {
+            int overheadSize = 0;
+
+            for (int i = 0; i < blockShifts.Length; i++)
+            {
+                int currBlockShift = blockShifts[i];
+                int nextBlockShift = i != blockShifts.Length - 1 ? blockShifts[i + 1] : 0;
+                overheadSize += Block.CalculateManagementOverheadSize(regionSize, currBlockShift, nextBlockShift);
+            }
+
+            return BitUtils.AlignUp(overheadSize, KPageTableBase.PageSize);
+        }
+    }
+}

Ryujinx.HLE/HOS/Kernel/Memory/KPageTableBase.cs

@@ -555,7 +555,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
         {
             KMemoryRegionManager region = GetMemoryRegionManager();
 
-            KernelResult result = region.AllocatePages(pagesCount, _aslrDisabled, out KPageList pageList);
+            KernelResult result = region.AllocatePages(out KPageList pageList, pagesCount);
 
             if (result != KernelResult.Success)
             {
@@ -712,7 +712,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
 
                     KMemoryRegionManager region = GetMemoryRegionManager();
 
-                    KernelResult result = region.AllocatePages(pagesCount, _aslrDisabled, out KPageList pageList);
+                    KernelResult result = region.AllocatePages(out KPageList pageList, pagesCount);
 
                     using var _ = new OnScopeExit(() => pageList.DecrementPagesReferenceCount(Context.MemoryManager));
 
@@ -1276,7 +1276,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
 
                 KMemoryRegionManager region = GetMemoryRegionManager();
 
-                KernelResult result = region.AllocatePages(remainingPages, _aslrDisabled, out KPageList pageList);
+                KernelResult result = region.AllocatePages(out KPageList pageList, remainingPages);
 
                 using var _ = new OnScopeExit(() => pageList.DecrementPagesReferenceCount(Context.MemoryManager));
 

Ryujinx.HLE/HOS/Kernel/Process/KProcess.cs

@@ -735,11 +735,12 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
             ulong argsPtr,
             ulong stackTop,
             int priority,
-            int cpuCore)
+            int cpuCore,
+            ThreadStart customThreadStart = null)
         {
             lock (_processLock)
             {
-                return thread.Initialize(entrypoint, argsPtr, stackTop, priority, cpuCore, this, ThreadType.User, null);
+                return thread.Initialize(entrypoint, argsPtr, stackTop, priority, cpuCore, this, ThreadType.User, customThreadStart);
             }
         }
 

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

@@ -2350,6 +2350,18 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
             [PointerSized] ulong stackTop,
             int priority,
             int cpuCore)
+        {
+            return CreateThread(out handle, entrypoint, argsPtr, stackTop, priority, cpuCore, null);
+        }
+
+        public KernelResult CreateThread(
+            out int handle,
+            ulong entrypoint,
+            ulong argsPtr,
+            ulong stackTop,
+            int priority,
+            int cpuCore,
+            ThreadStart customThreadStart)
         {
             handle = 0;
 
@@ -2386,7 +2398,8 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
                 argsPtr,
                 stackTop,
                 priority,
-                cpuCore);
+                cpuCore,
+                customThreadStart);
 
             if (result == KernelResult.Success)
             {

Ryujinx.HLE/HOS/ProgramLoader.cs

@@ -90,7 +90,7 @@ namespace Ryujinx.HLE.HOS
 
             KMemoryRegionManager region = context.MemoryManager.MemoryRegions[(int)memoryRegion];
 
-            KernelResult result = region.AllocatePages((ulong)codePagesCount, false, out KPageList pageList);
+            KernelResult result = region.AllocatePages(out KPageList pageList, (ulong)codePagesCount);
 
             if (result != KernelResult.Success)
             {

Ryujinx.HLE/HOS/Services/ServerBase.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Common.Logging;
 using Ryujinx.HLE.HOS.Ipc;
 using Ryujinx.HLE.HOS.Kernel;
 using Ryujinx.HLE.HOS.Kernel.Common;
@@ -38,15 +39,15 @@ namespace Ryujinx.HLE.HOS.Services
         private readonly Dictionary<int, Func<IpcService>> _ports = new Dictionary<int, Func<IpcService>>();
 
         public ManualResetEvent InitDone { get; }
-        public Func<IpcService> SmObjectFactory { get; }
         public string Name { get; }
+        public Func<IpcService> SmObjectFactory { get; }
 
         public ServerBase(KernelContext context, string name, Func<IpcService> smObjectFactory = null)
         {
             InitDone = new ManualResetEvent(false);
+            _context = context;
             Name = name;
             SmObjectFactory = smObjectFactory;
-            _context = context;
 
             const ProcessCreationFlags flags =
                 ProcessCreationFlags.EnableAslr |
@@ -56,7 +57,7 @@ namespace Ryujinx.HLE.HOS.Services
 
             ProcessCreationInfo creationInfo = new ProcessCreationInfo("Service", 1, 0, 0x8000000, 1, flags, 0, 0);
 
-            KernelStatic.StartInitialProcess(context, creationInfo, DefaultCapabilities, 44, ServerLoop);
+            KernelStatic.StartInitialProcess(context, creationInfo, DefaultCapabilities, 44, Main);
         }
 
         private void AddPort(int serverPortHandle, Func<IpcService> objectFactory)
@@ -80,6 +81,11 @@ namespace Ryujinx.HLE.HOS.Services
             _sessions.Add(serverSessionHandle, obj);
         }
 
+        private void Main()
+        {
+            ServerLoop();
+        }
+
         private void ServerLoop()
         {
             _selfProcess = KernelStatic.GetCurrentProcess();

Ryujinx.HLE/HOS/Services/Sockets/Bsd/Impl/EventFileDescriptor.cs

@@ -8,7 +8,6 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
     {
         private ulong _value;
         private readonly EventFdFlags _flags;
-        private AutoResetEvent _event;
 
         private object _lock = new object();
 
@@ -19,9 +18,13 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
 
         public EventFileDescriptor(ulong value, EventFdFlags flags)
         {
+            // FIXME: We should support blocking operations.
+            // Right now they can't be supported because it would cause the
+            // service to lock up as we only have one thread processing requests.
+            flags |= EventFdFlags.NonBlocking;
+
             _value = value;
             _flags = flags;
-            _event = new AutoResetEvent(false);
 
             WriteEvent = new ManualResetEvent(true);
             ReadEvent = new ManualResetEvent(true);
@@ -31,7 +34,6 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
 
         public void Dispose()
         {
-            _event.Dispose();
             WriteEvent.Dispose();
             ReadEvent.Dispose();
         }
@@ -57,7 +59,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
                     {
                         while (_value == 0)
                         {
-                            _event.WaitOne();
+                            Monitor.Wait(_lock);
                         }
                     }
                     else
@@ -106,7 +108,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
                 {
                     if (Blocking)
                     {
-                        _event.WaitOne();
+                        Monitor.Wait(_lock);
                     }
                     else
                     {
@@ -119,7 +121,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
                 writeSize = sizeof(ulong);
 
                 _value += count;
-                _event.Set();
+                Monitor.Pulse(_lock);
 
                 WriteEvent.Set();
 

Ryujinx.Input.SDL2/SDL2GamepadDriver.cs

@@ -79,6 +79,13 @@ namespace Ryujinx.Input.SDL2
                     return;
                 }
 
+                // Sometimes a JoyStick connected event fires after the app starts even though it was connected before
+                // so it is rejected to avoid doubling the entries.
+                if (_gamepadsIds.Contains(id))
+                {
+                    return;
+                }
+
                 if (_gamepadsInstanceIdsMapping.TryAdd(joystickInstanceId, id))
                 {
                     _gamepadsIds.Add(id);

Ryujinx.Memory/MemoryBlock.cs

@@ -19,6 +19,8 @@ namespace Ryujinx.Memory
         private ConcurrentDictionary<MemoryBlock, byte> _viewStorages;
         private int _viewCount;
 
+        internal bool ForceWindows4KBView => _forceWindows4KBView;
+
         /// <summary>
         /// Pointer to the memory block data.
         /// </summary>
@@ -145,7 +147,7 @@ namespace Ryujinx.Memory
                 srcBlock.IncrementViewCount();
             }
 
-            MemoryManagement.MapView(srcBlock._sharedMemory, srcOffset, GetPointerInternal(dstOffset, size), size, _forceWindows4KBView);
+            MemoryManagement.MapView(srcBlock._sharedMemory, srcOffset, GetPointerInternal(dstOffset, size), size, this);
         }
 
         /// <summary>
@@ -156,7 +158,7 @@ namespace Ryujinx.Memory
         /// <param name="size">Size of the range to be unmapped</param>
         public void UnmapView(MemoryBlock srcBlock, ulong offset, ulong size)
         {
-            MemoryManagement.UnmapView(srcBlock._sharedMemory, GetPointerInternal(offset, size), size, _forceWindows4KBView);
+            MemoryManagement.UnmapView(srcBlock._sharedMemory, GetPointerInternal(offset, size), size, this);
         }
 
         /// <summary>

Ryujinx.Memory/MemoryManagement.cs

@@ -68,17 +68,17 @@ namespace Ryujinx.Memory
             }
         }
 
-        public static void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr address, ulong size, bool force4KBMap)
+        public static void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr address, ulong size, MemoryBlock owner)
         {
             if (OperatingSystem.IsWindows())
             {
-                if (force4KBMap)
+                if (owner.ForceWindows4KBView)
                 {
                     MemoryManagementWindows.MapView4KB(sharedMemory, srcOffset, address, (IntPtr)size);
                 }
                 else
                 {
-                    MemoryManagementWindows.MapView(sharedMemory, srcOffset, address, (IntPtr)size);
+                    MemoryManagementWindows.MapView(sharedMemory, srcOffset, address, (IntPtr)size, owner);
                 }
             }
             else if (OperatingSystem.IsLinux() || OperatingSystem.IsMacOS())
@@ -91,17 +91,17 @@ namespace Ryujinx.Memory
             }
         }
 
-        public static void UnmapView(IntPtr sharedMemory, IntPtr address, ulong size, bool force4KBMap)
+        public static void UnmapView(IntPtr sharedMemory, IntPtr address, ulong size, MemoryBlock owner)
         {
             if (OperatingSystem.IsWindows())
             {
-                if (force4KBMap)
+                if (owner.ForceWindows4KBView)
                 {
                     MemoryManagementWindows.UnmapView4KB(address, (IntPtr)size);
                 }
                 else
                 {
-                    MemoryManagementWindows.UnmapView(sharedMemory, address, (IntPtr)size);
+                    MemoryManagementWindows.UnmapView(sharedMemory, address, (IntPtr)size, owner);
                 }
             }
             else if (OperatingSystem.IsLinux() || OperatingSystem.IsMacOS())

Ryujinx.Memory/MemoryManagementWindows.cs

@@ -68,9 +68,9 @@ namespace Ryujinx.Memory
             return WindowsApi.VirtualFree(location, size, AllocationType.Decommit);
         }
 
-        public static void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr location, IntPtr size)
+        public static void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr location, IntPtr size, MemoryBlock owner)
         {
-            _placeholders.MapView(sharedMemory, srcOffset, location, size);
+            _placeholders.MapView(sharedMemory, srcOffset, location, size, owner);
         }
 
         public static void MapView4KB(IntPtr sharedMemory, ulong srcOffset, IntPtr location, IntPtr size)
@@ -106,9 +106,9 @@ namespace Ryujinx.Memory
             }
         }
 
-        public static void UnmapView(IntPtr sharedMemory, IntPtr location, IntPtr size)
+        public static void UnmapView(IntPtr sharedMemory, IntPtr location, IntPtr size, MemoryBlock owner)
         {
-            _placeholders.UnmapView(sharedMemory, location, size);
+            _placeholders.UnmapView(sharedMemory, location, size, owner);
         }
 
         public static void UnmapView4KB(IntPtr location, IntPtr size)
@@ -154,7 +154,7 @@ namespace Ryujinx.Memory
             }
             else
             {
-                _placeholders.UnmapView(IntPtr.Zero, address, size);
+                _placeholders.UnreserveRange((ulong)address, (ulong)size);
             }
 
             return WindowsApi.VirtualFree(address, IntPtr.Zero, AllocationType.Release);

Ryujinx.Memory/WindowsShared/PlaceholderManager.cs

@@ -44,6 +44,50 @@ namespace Ryujinx.Memory.WindowsShared
             }
         }
 
+        /// <summary>
+        /// Unreserves a range of memory that has been previously reserved with <see cref="ReserveRange"/>.
+        /// </summary>
+        /// <param name="address">Start address of the region to unreserve</param>
+        /// <param name="size">Size in bytes of the region to unreserve</param>
+        /// <exception cref="WindowsApiException">Thrown when the Windows API returns an error unreserving the memory</exception>
+        public void UnreserveRange(ulong address, ulong size)
+        {
+            ulong endAddress = address + size;
+
+            var overlaps = Array.Empty<IntervalTreeNode<ulong, ulong>>();
+            int count;
+
+            lock (_mappings)
+            {
+                count = _mappings.Get(address, endAddress, ref overlaps);
+
+                for (int index = 0; index < count; index++)
+                {
+                    var overlap = overlaps[index];
+
+                    if (IsMapped(overlap.Value))
+                    {
+                        if (!WindowsApi.UnmapViewOfFile2(WindowsApi.CurrentProcessHandle, (IntPtr)overlap.Start, 2))
+                        {
+                            throw new WindowsApiException("UnmapViewOfFile2");
+                        }
+                    }
+
+                    _mappings.Remove(overlap);
+                }
+            }
+
+            if (count > 1)
+            {
+                CheckFreeResult(WindowsApi.VirtualFree(
+                    (IntPtr)address,
+                    (IntPtr)size,
+                    AllocationType.Release | AllocationType.CoalescePlaceholders));
+            }
+
+            RemoveProtection(address, size);
+        }
+
         /// <summary>
         /// Maps a shared memory view on a previously reserved memory region.
         /// </summary>
@@ -51,13 +95,14 @@ namespace Ryujinx.Memory.WindowsShared
         /// <param name="srcOffset">Offset in the shared memory to map</param>
         /// <param name="location">Address to map the view into</param>
         /// <param name="size">Size of the view in bytes</param>
-        public void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr location, IntPtr size)
+        /// <param name="owner">Memory block that owns the mapping</param>
+        public void MapView(IntPtr sharedMemory, ulong srcOffset, IntPtr location, IntPtr size, MemoryBlock owner)
         {
             _partialUnmapLock.AcquireReaderLock(Timeout.Infinite);
 
             try
             {
-                UnmapViewInternal(sharedMemory, location, size);
+                UnmapViewInternal(sharedMemory, location, size, owner);
                 MapViewInternal(sharedMemory, srcOffset, location, size);
             }
             finally
@@ -173,13 +218,14 @@ namespace Ryujinx.Memory.WindowsShared
         /// <param name="sharedMemory">Shared memory that the view being unmapped belongs to</param>
         /// <param name="location">Address to unmap</param>
         /// <param name="size">Size of the region to unmap in bytes</param>
-        public void UnmapView(IntPtr sharedMemory, IntPtr location, IntPtr size)
+        /// <param name="owner">Memory block that owns the mapping</param>
+        public void UnmapView(IntPtr sharedMemory, IntPtr location, IntPtr size, MemoryBlock owner)
         {
             _partialUnmapLock.AcquireReaderLock(Timeout.Infinite);
 
             try
             {
-                UnmapViewInternal(sharedMemory, location, size);
+                UnmapViewInternal(sharedMemory, location, size, owner);
             }
             finally
             {
@@ -197,8 +243,9 @@ namespace Ryujinx.Memory.WindowsShared
         /// <param name="sharedMemory">Shared memory that the view being unmapped belongs to</param>
         /// <param name="location">Address to unmap</param>
         /// <param name="size">Size of the region to unmap in bytes</param>
+        /// <param name="owner">Memory block that owns the mapping</param>
         /// <exception cref="WindowsApiException">Thrown when the Windows API returns an error unmapping or remapping the memory</exception>
-        private void UnmapViewInternal(IntPtr sharedMemory, IntPtr location, IntPtr size)
+        private void UnmapViewInternal(IntPtr sharedMemory, IntPtr location, IntPtr size, MemoryBlock owner)
         {
             ulong startAddress = (ulong)location;
             ulong unmapSize = (ulong)size;
@@ -272,7 +319,7 @@ namespace Ryujinx.Memory.WindowsShared
                 }
             }
 
-            CoalesceForUnmap(startAddress, unmapSize);
+            CoalesceForUnmap(startAddress, unmapSize, owner);
             RemoveProtection(startAddress, unmapSize);
         }
 
@@ -281,15 +328,21 @@ namespace Ryujinx.Memory.WindowsShared
         /// </summary>
         /// <param name="address">Address of the region that was unmapped</param>
         /// <param name="size">Size of the region that was unmapped in bytes</param>
-        private void CoalesceForUnmap(ulong address, ulong size)
+        /// <param name="owner">Memory block that owns the mapping</param>
+        private void CoalesceForUnmap(ulong address, ulong size, MemoryBlock owner)
         {
             ulong endAddress = address + size;
+            ulong blockAddress = (ulong)owner.Pointer;
+            ulong blockEnd = blockAddress + owner.Size;
             var overlaps = Array.Empty<IntervalTreeNode<ulong, ulong>>();
             int unmappedCount = 0;
 
             lock (_mappings)
             {
-                int count = _mappings.Get(address - MinimumPageSize, endAddress + MinimumPageSize, ref overlaps);
+                int count = _mappings.Get(
+                    Math.Max(address - MinimumPageSize, blockAddress),
+                    Math.Min(endAddress + MinimumPageSize, blockEnd), ref overlaps);
+
                 if (count < 2)
                 {
                     // Nothing to coalesce if we only have 1 or no overlaps.