Vulkan: Implement multisample <-> non-multisample copies and depth-stencil resolve (#3723)

* Vulkan: Implement multisample <-> non-multisample copies and depth-stencil resolve

* FramebufferParams is no longer required there

* Implement Specialization Constants and merge CopyMS Shaders (#15)

* Vulkan: Initial Specialization Constants

* Replace with specialized helper shader

* Reimplement everything

Fix nonexistant interaction with Ryu pipeline caching
Decouple specialization info from data and relocate them
Generalize mapping and add type enum to better match spv types
Use local fixed scopes instead of global unmanaged allocs

* Fix misses in initial implementation

Use correct info variable in Create2DLayerView
Add ShaderStorageImageMultisample to required feature set

* Use texture for source image

* No point in using ReadOnlyMemory

* Apply formatting feedback

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

* Apply formatting suggestions on shader source

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

* Support conversion with samples count that does not match the requested count, other minor changes

Co-authored-by: mageven <62494521+mageven@users.noreply.github.com>

.github/workflows/build.yml

@@ -57,6 +57,7 @@ jobs:
       - name: Get git short hash
         id: git_short_hash
         run: echo "result=$(git rev-parse --short "${{ github.sha }}")" >> $GITHUB_OUTPUT
+        shell: bash
       - name: Clear
         run: dotnet clean && dotnet nuget locals all --clear
       - name: Build

Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs

@@ -710,7 +710,12 @@ namespace Ryujinx.Graphics.Gpu.Image
                     break;
                 case Target.Texture2DMultisample:
                 case Target.Texture2DMultisampleArray:
-                    if (rhs.Target == Target.Texture2D || rhs.Target == Target.Texture2DArray)
+                    // We don't support copy between multisample and non-multisample depth-stencil textures
+                    // because there's no way to emulate that since most GPUs don't support writing a
+                    // custom stencil value into the texture, among several other API limitations.
+
+                    if ((rhs.Target == Target.Texture2D || rhs.Target == Target.Texture2DArray) &&
+                        !rhs.FormatInfo.Format.IsDepthOrStencil())
                     {
                         return TextureViewCompatibility.CopyOnly;
                     }

Ryujinx.Graphics.Nvdec.FFmpeg/FFmpegContext.cs

@@ -7,7 +7,9 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg
 {
     unsafe class FFmpegContext : IDisposable
     {
-        private readonly FFCodec.AVCodec_decode _decodeFrame;
+        private unsafe delegate int AVCodec_decode(AVCodecContext* avctx, void* outdata, int* got_frame_ptr, AVPacket* avpkt);
+
+        private readonly AVCodec_decode _decodeFrame;
         private static readonly FFmpegApi.av_log_set_callback_callback _logFunc;
         private readonly AVCodec* _codec;
         private AVPacket* _packet;
@@ -53,17 +55,17 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg
             // libavcodec 59.24 changed AvCodec to move its private API and also move the codec function to an union.
             if (avCodecMajorVersion > 59 || (avCodecMajorVersion == 59 && avCodecMinorVersion > 24))
             {
-                _decodeFrame = Marshal.GetDelegateForFunctionPointer<FFCodec.AVCodec_decode>(((FFCodec*)_codec)->CodecCallback);
+                _decodeFrame = Marshal.GetDelegateForFunctionPointer<AVCodec_decode>(((FFCodec<AVCodec>*)_codec)->CodecCallback);
             }
             // libavcodec 59.x changed AvCodec private API layout.
             else if (avCodecMajorVersion == 59)
             {
-                _decodeFrame = Marshal.GetDelegateForFunctionPointer<FFCodec.AVCodec_decode>(((FFCodecLegacy<AVCodec>*)_codec)->Decode);
+                _decodeFrame = Marshal.GetDelegateForFunctionPointer<AVCodec_decode>(((FFCodecLegacy<AVCodec501>*)_codec)->Decode);
             }
             // libavcodec 58.x and lower
             else
             {
-                _decodeFrame = Marshal.GetDelegateForFunctionPointer<FFCodec.AVCodec_decode>(((FFCodecLegacy<AVCodecLegacy>*)_codec)->Decode);
+                _decodeFrame = Marshal.GetDelegateForFunctionPointer<AVCodec_decode>(((FFCodecLegacy<AVCodec>*)_codec)->Decode);
             }
         }
 

Ryujinx.Graphics.Nvdec.FFmpeg/Native/AVCodec.cs

@@ -20,6 +20,7 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.Native
         public unsafe IntPtr PrivClass;
         public IntPtr Profiles;
         public unsafe byte* WrapperName;
+        public IntPtr ChLayouts;
 #pragma warning restore CS0649
     }
 }

Ryujinx.Graphics.Nvdec.FFmpeg/Native/AVCodec501.cs

@@ -2,7 +2,7 @@
 
 namespace Ryujinx.Graphics.Nvdec.FFmpeg.Native
 {
-    struct AVCodecLegacy
+    struct AVCodec501
     {
 #pragma warning disable CS0649
         public unsafe byte* Name;
@@ -20,7 +20,6 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.Native
         public unsafe IntPtr PrivClass;
         public IntPtr Profiles;
         public unsafe byte* WrapperName;
-        public IntPtr ChLayouts;
 #pragma warning restore CS0649
     }
 }

Ryujinx.Graphics.Nvdec.FFmpeg/Native/AVCodecContext.cs

@@ -9,7 +9,7 @@ namespace Ryujinx.Graphics.Nvdec.FFmpeg.Native
         public unsafe IntPtr AvClass;
         public int LogLevelOffset;
         public int CodecType;
-        public unsafe AVCodecLegacy* Codec;
+        public unsafe AVCodec* Codec;
         public AVCodecID CodecId;
         public uint CodecTag;
         public IntPtr PrivData;

Ryujinx.Graphics.Nvdec.FFmpeg/Native/FFCodec.cs

@@ -2,12 +2,10 @@
 
 namespace Ryujinx.Graphics.Nvdec.FFmpeg.Native
 {
-    struct FFCodec
+    struct FFCodec<T> where T: struct
     {
-        public unsafe delegate int AVCodec_decode(AVCodecContext* avctx, void* outdata, int* got_frame_ptr, AVPacket* avpkt);
-
 #pragma warning disable CS0649
-        public AVCodec Base;
+        public T Base;
         public int CapsInternalOrCbType;
         public int PrivDataSize;
         public IntPtr UpdateThreadContext;

Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -139,7 +139,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
 
             if (!destinationView.Target.IsMultisample() && Target.IsMultisample())
             {
-                _renderer.TextureCopyMS.CopyMSToNonMS(this, destinationView, srcLayer, dstLayer,1);
+                _renderer.TextureCopyMS.CopyMSToNonMS(this, destinationView, srcLayer, dstLayer, 1);
             }
             else if (destinationView.Target.IsMultisample() && !Target.IsMultisample())
             {

Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly IProgram _programColorBlitClearAlpha;
         private readonly IProgram _programColorClear;
         private readonly IProgram _programStrideChange;
+        private readonly IProgram _programColorCopyBetweenMsNonMs;
 
         public HelperShader(VulkanRenderer gd, Device device)
         {
@@ -73,6 +74,20 @@ namespace Ryujinx.Graphics.Vulkan
             {
                 new ShaderSource(ShaderBinaries.ChangeBufferStrideShaderSource, strideChangeBindings, ShaderStage.Compute, TargetLanguage.Spirv),
             });
+
+            var colorCopyMSBindings = new ShaderBindings(
+                new[] { 0 },
+                Array.Empty<int>(),
+                new[] { 0 },
+                new[] { 0 });
+
+            _programColorCopyBetweenMsNonMs = gd.CreateProgramWithMinimalLayout(new[]
+            {
+                new ShaderSource(ShaderBinaries.ColorCopyBetweenMsNonMs, colorCopyMSBindings, ShaderStage.Compute, TargetLanguage.Spirv),
+            }, new[]
+            {
+                new SpecDescription((0, SpecConstType.Int32))
+            });
         }
 
         public void Blit(
@@ -136,11 +151,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, region);
 
-            Span<BufferRange> bufferRanges = stackalloc BufferRange[1];
+            _pipeline.SetUniformBuffers(1, stackalloc[] { new BufferRange(bufferHandle, 0, RegionBufferSize) });
-
-            bufferRanges[0] = new BufferRange(bufferHandle, 0, RegionBufferSize);
-
-            _pipeline.SetUniformBuffers(1, bufferRanges);
 
             Span<GAL.Viewport> viewports = stackalloc GAL.Viewport[1];
 
@@ -203,11 +214,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, clearColor);
 
-            Span<BufferRange> bufferRanges = stackalloc BufferRange[1];
+            _pipeline.SetUniformBuffers(1, stackalloc[] { new BufferRange(bufferHandle, 0, ClearColorBufferSize) });
-
-            bufferRanges[0] = new BufferRange(bufferHandle, 0, ClearColorBufferSize);
-
-            _pipeline.SetUniformBuffers(1, bufferRanges);
 
             Span<GAL.Viewport> viewports = stackalloc GAL.Viewport[1];
 
@@ -269,11 +276,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, region);
 
-            Span<BufferRange> bufferRanges = stackalloc BufferRange[1];
+            pipeline.SetUniformBuffers(1, stackalloc[] { new BufferRange(bufferHandle, 0, RegionBufferSize) });
-
-            bufferRanges[0] = new BufferRange(bufferHandle, 0, RegionBufferSize);
-
-            pipeline.SetUniformBuffers(1, bufferRanges);
 
             Span<GAL.Viewport> viewports = stackalloc GAL.Viewport[1];
 
@@ -351,11 +354,7 @@ namespace Ryujinx.Graphics.Vulkan
 
                 _pipeline.SetCommandBuffer(cbs);
 
-                Span<BufferRange> cbRanges = stackalloc BufferRange[1];
+                _pipeline.SetUniformBuffers(0, stackalloc[] { new BufferRange(bufferHandle, 0, ParamsBufferSize) });
-
-                cbRanges[0] = new BufferRange(bufferHandle, 0, ParamsBufferSize);
-
-                _pipeline.SetUniformBuffers(0, cbRanges);
 
                 Span<Auto<DisposableBuffer>> sbRanges = new Auto<DisposableBuffer>[2];
 
@@ -480,12 +479,207 @@ namespace Ryujinx.Graphics.Vulkan
                 convertedCount * outputIndexSize);
         }
 
+        public void CopyMSToNonMS(VulkanRenderer gd, CommandBufferScoped cbs, TextureView src, TextureView dst, int srcLayer, int dstLayer, int depth)
+        {
+            CopyMS(gd, cbs, src, dst, srcLayer, dstLayer, depth, src.Info.Samples, dst.Info.Width, dst.Info.Height);
+        }
+
+        public void CopyNonMSToMS(VulkanRenderer gd, CommandBufferScoped cbs, TextureView src, TextureView dst, int srcLayer, int dstLayer, int depth)
+        {
+            CopyMS(gd, cbs, src, dst, srcLayer, dstLayer, depth, dst.Info.Samples, src.Info.Width, src.Info.Height);
+        }
+
+        private void CopyMS(
+            VulkanRenderer gd,
+            CommandBufferScoped cbs,
+            TextureView src,
+            TextureView dst,
+            int srcLayer,
+            int dstLayer,
+            int depth,
+            int samples,
+            int nonMSWidth,
+            int nonMSHeight)
+        {
+            const int ParamsBufferSize = 16;
+
+            Span<int> shaderParams = stackalloc int[ParamsBufferSize / sizeof(int)];
+
+            // X and Y are the expected texture samples.
+            // Z and W are the actual texture samples used.
+            // They may differ if the GPU does not support the samples count requested and we had to use a lower amount.
+            (shaderParams[0], shaderParams[1]) = GetSampleCountXYLog2(samples);
+            (shaderParams[2], shaderParams[3]) = GetSampleCountXYLog2((int)TextureStorage.ConvertToSampleCountFlags((uint)samples));
+
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false);
+
+            gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
+
+            TextureView.InsertImageBarrier(
+                gd.Api,
+                cbs.CommandBuffer,
+                src.GetImage().Get(cbs).Value,
+                TextureStorage.DefaultAccessMask,
+                AccessFlags.AccessShaderReadBit,
+                PipelineStageFlags.PipelineStageAllCommandsBit,
+                PipelineStageFlags.PipelineStageComputeShaderBit,
+                ImageAspectFlags.ImageAspectColorBit,
+                src.FirstLayer + srcLayer,
+                src.FirstLevel,
+                depth,
+                1);
+
+            _pipeline.SetCommandBuffer(cbs);
+
+            _pipeline.SetProgram(_programColorCopyBetweenMsNonMs);
+
+            var format = GetFormat(src.Info.BytesPerPixel);
+
+            int dispatchX = (nonMSWidth + 31) / 32;
+            int dispatchY = (nonMSHeight + 31) / 32;
+
+            // Specialize shader.
+            bool srcIsMs = src.Info.Target.IsMultisample();
+            int conversionType = srcIsMs ? src.Info.BytesPerPixel : -src.Info.BytesPerPixel;
+            _pipeline.Specialize(conversionType);
+
+            _pipeline.SetUniformBuffers(0, stackalloc[] { new BufferRange(bufferHandle, 0, ParamsBufferSize) });
+
+            if (src.Info.Target == Target.Texture2DMultisampleArray ||
+                dst.Info.Target == Target.Texture2DMultisampleArray)
+            {
+                for (int z = 0; z < depth; z++)
+                {
+                    var srcView = Create2DLayerView(src, srcLayer + z, format);
+                    var dstView = Create2DLayerView(dst, dstLayer + z);
+
+                    _pipeline.SetTextureAndSampler(ShaderStage.Compute, 0, srcView, null);
+                    _pipeline.SetImage(0, dstView, format);
+
+                    _pipeline.DispatchCompute(dispatchX, dispatchY, 1);
+
+                    srcView.Release();
+                    dstView.Release();
+                }
+            }
+            else
+            {
+                var srcView = Create2DLayerView(src, srcLayer, format);
+
+                _pipeline.SetTextureAndSampler(ShaderStage.Compute, 0, srcView, null);
+                _pipeline.SetImage(0, dst, format);
+
+                _pipeline.DispatchCompute(dispatchX, dispatchY, 1);
+
+                srcView.Release();
+            }
+
+            gd.BufferManager.Delete(bufferHandle);
+
+            _pipeline.Finish(gd, cbs);
+
+            TextureView.InsertImageBarrier(
+                gd.Api,
+                cbs.CommandBuffer,
+                dst.GetImage().Get(cbs).Value,
+                AccessFlags.AccessShaderWriteBit,
+                TextureStorage.DefaultAccessMask,
+                PipelineStageFlags.PipelineStageComputeShaderBit,
+                PipelineStageFlags.PipelineStageAllCommandsBit,
+                ImageAspectFlags.ImageAspectColorBit,
+                dst.FirstLayer + dstLayer,
+                dst.FirstLevel,
+                depth,
+                1);
+        }
+
+        private static (int, int) GetSampleCountXYLog2(int samples)
+        {
+            int samplesInXLog2 = 0;
+            int samplesInYLog2 = 0;
+
+            switch (samples)
+            {
+                case 2: // 2x1
+                    samplesInXLog2 = 1;
+                    break;
+                case 4: // 2x2
+                    samplesInXLog2 = 1;
+                    samplesInYLog2 = 1;
+                    break;
+                case 8: // 4x2
+                    samplesInXLog2 = 2;
+                    samplesInYLog2 = 1;
+                    break;
+                case 16: // 4x4
+                    samplesInXLog2 = 2;
+                    samplesInYLog2 = 2;
+                    break;
+                case 32: // 8x4
+                    samplesInXLog2 = 3;
+                    samplesInYLog2 = 2;
+                    break;
+                case 64: // 8x8
+                    samplesInXLog2 = 3;
+                    samplesInYLog2 = 3;
+                    break;
+            }
+
+            return (samplesInXLog2, samplesInYLog2);
+        }
+
+        private static ITexture Create2DLayerView(TextureView from, int layer, GAL.Format? format = null)
+        {
+            var target = from.Info.Target switch
+            {
+                Target.Texture1DArray => Target.Texture1D,
+                Target.Texture2DArray => Target.Texture2D,
+                Target.Texture2DMultisampleArray => Target.Texture2DMultisample,
+                _ => from.Info.Target
+            };
+
+            var info = new TextureCreateInfo(
+                from.Info.Width,
+                from.Info.Height,
+                from.Info.Depth,
+                1,
+                from.Info.Samples,
+                from.Info.BlockWidth,
+                from.Info.BlockHeight,
+                from.Info.BytesPerPixel,
+                format ?? from.Info.Format,
+                from.Info.DepthStencilMode,
+                target,
+                from.Info.SwizzleR,
+                from.Info.SwizzleG,
+                from.Info.SwizzleB,
+                from.Info.SwizzleA);
+
+            return from.CreateView(info, layer, 0);
+        }
+
+        private static GAL.Format GetFormat(int bytesPerPixel)
+        {
+            return bytesPerPixel switch
+            {
+                1 => GAL.Format.R8Uint,
+                2 => GAL.Format.R16Uint,
+                4 => GAL.Format.R32Uint,
+                8 => GAL.Format.R32G32Uint,
+                16 => GAL.Format.R32G32B32A32Uint,
+                _ => throw new ArgumentException($"Invalid bytes per pixel {bytesPerPixel}.")
+            };
+        }
+
         protected virtual void Dispose(bool disposing)
         {
             if (disposing)
             {
                 _programColorBlitClearAlpha.Dispose();
                 _programColorBlit.Dispose();
+                _programColorClear.Dispose();
+                _programStrideChange.Dispose();
+                _programColorCopyBetweenMsNonMs.Dispose();
                 _samplerNearest.Dispose();
                 _samplerLinear.Dispose();
                 _pipeline.Dispose();

Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -3,6 +3,8 @@ using Ryujinx.Graphics.Shader;
 using Silk.NET.Vulkan;
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Vulkan
 {
@@ -697,6 +699,18 @@ namespace Ryujinx.Graphics.Vulkan
             SignalStateChange();
         }
 
+        public void Specialize<T>(in T data) where T : unmanaged
+        {
+            var dataSpan = MemoryMarshal.AsBytes(MemoryMarshal.CreateReadOnlySpan(ref Unsafe.AsRef(in data), 1));
+
+            if (!dataSpan.SequenceEqual(_newState.SpecializationData.Span))
+            {
+                _newState.SpecializationData = new SpecData(dataSpan);
+                
+                SignalStateChange();
+            }
+        }
+
         protected virtual void SignalAttachmentChange()
         {
         }
@@ -1188,14 +1202,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            var subpassDependency = new SubpassDependency(
+            var subpassDependency = PipelineConverter.CreateSubpassDependency();
-                0,
-                0,
-                PipelineStageFlags.PipelineStageAllGraphicsBit,
-                PipelineStageFlags.PipelineStageAllGraphicsBit,
-                AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit | AccessFlags.AccessColorAttachmentWriteBit,
-                AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit | AccessFlags.AccessShaderReadBit,
-                0);
 
             fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
             {

Ryujinx.Graphics.Vulkan/PipelineConverter.cs

@@ -6,6 +6,9 @@ namespace Ryujinx.Graphics.Vulkan
 {
     static class PipelineConverter
     {
+        private const AccessFlags SubpassSrcAccessMask = AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit | AccessFlags.AccessColorAttachmentWriteBit;
+        private const AccessFlags SubpassDstAccessMask = AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit | AccessFlags.AccessShaderReadBit;
+
         public static unsafe DisposableRenderPass ToRenderPass(this ProgramPipelineState state, VulkanRenderer gd, Device device)
         {
             const int MaxAttachments = Constants.MaxRenderTargets + 1;
@@ -100,14 +103,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            var subpassDependency = new SubpassDependency(
+            var subpassDependency = CreateSubpassDependency();
-                0,
-                0,
-                PipelineStageFlags.PipelineStageAllGraphicsBit,
-                PipelineStageFlags.PipelineStageAllGraphicsBit,
-                AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit,
-                AccessFlags.AccessMemoryReadBit | AccessFlags.AccessMemoryWriteBit,
-                0);
 
             fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
             {
@@ -128,6 +124,32 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        public static SubpassDependency CreateSubpassDependency()
+        {
+            return new SubpassDependency(
+                0,
+                0,
+                PipelineStageFlags.PipelineStageAllGraphicsBit,
+                PipelineStageFlags.PipelineStageAllGraphicsBit,
+                SubpassSrcAccessMask,
+                SubpassDstAccessMask,
+                0);
+        }
+
+        public unsafe static SubpassDependency2 CreateSubpassDependency2()
+        {
+            return new SubpassDependency2(
+                StructureType.SubpassDependency2,
+                null,
+                0,
+                0,
+                PipelineStageFlags.PipelineStageAllGraphicsBit,
+                PipelineStageFlags.PipelineStageAllGraphicsBit,
+                SubpassSrcAccessMask,
+                SubpassDstAccessMask,
+                0);
+        }
+
         public static PipelineState ToVulkanPipelineState(this ProgramPipelineState state, VulkanRenderer gd)
         {
             PipelineState pipeline = new PipelineState();

Ryujinx.Graphics.Vulkan/PipelineState.cs

@@ -312,6 +312,7 @@ namespace Ryujinx.Graphics.Vulkan
         public NativeArray<PipelineShaderStageCreateInfo> Stages;
         public NativeArray<PipelineShaderStageRequiredSubgroupSizeCreateInfoEXT> StageRequiredSubgroupSizes;
         public PipelineLayout PipelineLayout;
+        public SpecData SpecializationData;
 
         public void Initialize()
         {
@@ -334,7 +335,7 @@ namespace Ryujinx.Graphics.Vulkan
             ShaderCollection program,
             PipelineCache cache)
         {
-            if (program.TryGetComputePipeline(out var pipeline))
+            if (program.TryGetComputePipeline(ref SpecializationData, out var pipeline))
             {
                 return pipeline;
             }
@@ -354,20 +355,36 @@ namespace Ryujinx.Graphics.Vulkan
 
             Pipeline pipelineHandle = default;
 
-            gd.Api.CreateComputePipelines(device, cache, 1, &pipelineCreateInfo, null, &pipelineHandle).ThrowOnError();
+            bool hasSpec = program.SpecDescriptions != null;
+                
+            var desc = hasSpec ? program.SpecDescriptions[0] : SpecDescription.Empty;
+
+            if (hasSpec && SpecializationData.Length < (int)desc.Info.DataSize)
+            {
+                throw new InvalidOperationException("Specialization data size does not match description");
+            }
+
+            fixed (SpecializationInfo* info = &desc.Info)
+            fixed (SpecializationMapEntry* map = desc.Map)
+            fixed (byte* data = SpecializationData.Span)
+            {
+                if (hasSpec)
+                {
+                    info->PMapEntries = map;
+                    info->PData = data;
+                    pipelineCreateInfo.Stage.PSpecializationInfo = info;
+                }
+
+                gd.Api.CreateComputePipelines(device, cache, 1, &pipelineCreateInfo, null, &pipelineHandle).ThrowOnError();
+            }
 
             pipeline = new Auto<DisposablePipeline>(new DisposablePipeline(gd.Api, device, pipelineHandle));
 
-            program.AddComputePipeline(pipeline);
+            program.AddComputePipeline(ref SpecializationData, pipeline);
 
             return pipeline;
         }
 
-        public unsafe void DestroyComputePipeline(ShaderCollection program)
-        {
-            program.RemoveComputePipeline();
-        }
-
         public unsafe Auto<DisposablePipeline> CreateGraphicsPipeline(
             VulkanRenderer gd,
             Device device,

Ryujinx.Graphics.Vulkan/ShaderCollection.cs

@@ -26,6 +26,8 @@ namespace Ryujinx.Graphics.Vulkan
 
         public ProgramLinkStatus LinkStatus { get; private set; }
 
+        public readonly SpecDescription[] SpecDescriptions;
+
         public bool IsLinked
         {
             get
@@ -40,7 +42,7 @@ namespace Ryujinx.Graphics.Vulkan
         }
 
         private HashTableSlim<PipelineUid, Auto<DisposablePipeline>> _graphicsPipelineCache;
-        private Auto<DisposablePipeline> _computePipeline;
+        private HashTableSlim<SpecData, Auto<DisposablePipeline>> _computePipelineCache;
 
         private VulkanRenderer _gd;
         private Device _device;
@@ -52,17 +54,24 @@ namespace Ryujinx.Graphics.Vulkan
         private Task _compileTask;
         private bool _firstBackgroundUse;
 
-        public ShaderCollection(VulkanRenderer gd, Device device, ShaderSource[] shaders, bool isMinimal = false)
+        public ShaderCollection(VulkanRenderer gd, Device device, ShaderSource[] shaders, SpecDescription[] specDescription = null, bool isMinimal = false)
         {
             _gd = gd;
             _device = device;
 
+            if (specDescription != null && specDescription.Length != shaders.Length)
+            {
+                throw new ArgumentException($"{nameof(specDescription)} array length must match {nameof(shaders)} array if provided");
+            }
+
             gd.Shaders.Add(this);
 
             var internalShaders = new Shader[shaders.Length];
 
             _infos = new PipelineShaderStageCreateInfo[shaders.Length];
 
+            SpecDescriptions = specDescription;
+
             LinkStatus = ProgramLinkStatus.Incomplete;
 
             uint stages = 0;
@@ -314,14 +323,9 @@ namespace Ryujinx.Graphics.Vulkan
             return null;
         }
 
-        public void AddComputePipeline(Auto<DisposablePipeline> pipeline)
+        public void AddComputePipeline(ref SpecData key, Auto<DisposablePipeline> pipeline)
         {
-            _computePipeline = pipeline;
+            (_computePipelineCache ??= new()).Add(ref key, pipeline);
-        }
-
-        public void RemoveComputePipeline()
-        {
-            _computePipeline = null;
         }
 
         public void AddGraphicsPipeline(ref PipelineUid key, Auto<DisposablePipeline> pipeline)
@@ -329,10 +333,20 @@ namespace Ryujinx.Graphics.Vulkan
             (_graphicsPipelineCache ??= new()).Add(ref key, pipeline);
         }
 
-        public bool TryGetComputePipeline(out Auto<DisposablePipeline> pipeline)
+        public bool TryGetComputePipeline(ref SpecData key, out Auto<DisposablePipeline> pipeline)
         {
-            pipeline = _computePipeline;
+            if (_computePipelineCache == null)
-            return pipeline != null;
+            {
+                pipeline = default;
+                return false;
+            }
+
+            if (_computePipelineCache.TryGetValue(ref key, out pipeline))
+            {
+                return true;
+            }
+
+            return false;
         }
 
         public bool TryGetGraphicsPipeline(ref PipelineUid key, out Auto<DisposablePipeline> pipeline)
@@ -390,7 +404,14 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                 }
 
-                _computePipeline?.Dispose();
+                if (_computePipelineCache != null)
+                {
+                    foreach (Auto<DisposablePipeline> pipeline in _computePipelineCache.Values)
+                    {
+                        pipeline.Dispose();
+                    }
+                }
+
                 if (_dummyRenderPass.Value.Handle != 0)
                 {
                     _dummyRenderPass.Dispose();

Ryujinx.Graphics.Vulkan/Shaders/ColorCopyBetweenMsNonMs.comp

@@ -0,0 +1,89 @@
+#version 450 core
+
+// +ve for MsToNonMs, -ve for reverse
+layout (constant_id = 0) const int convType = 0;
+
+layout (std140, binding = 0) uniform sample_counts_log2_in
+{
+    ivec4 sample_counts_log2;
+};
+
+#define R8_ID       1
+#define R16_ID      2
+#define R32_ID      4
+#define RG32_ID     8
+#define RGBA32_ID   16
+
+#define R8_TYPE     r8ui
+#define R16_TYPE    r16ui
+#define R32_TYPE    r32ui
+#define RG32_TYPE   rg32ui
+#define RGBA32_TYPE rgba32ui
+
+#define DECLARE_BINDINGS(type) layout (set = 3, binding = 0, type##_TYPE) uniform uimage2DMS dstMS ## type; \
+                               layout (set = 3, binding = 0, type##_TYPE) uniform uimage2D dst ## type;
+
+#define CASE_SIZE(type) case type##_ID: imageSz = imageSize(dst ## type); break;
+
+#define CASE_CONVERT(type) case  type##_ID: imageStore(dst ## type, ivec2(coords), texelFetch(srcMS, shiftedCoords, sampleIdx)); break; \
+                           case -type##_ID: imageStore(dstMS ## type, shiftedCoords, sampleIdx, texelFetch(src, ivec2(coords), 0)); break;
+
+// src tex
+layout (set = 2, binding = 0) uniform usampler2DMS srcMS;
+layout (set = 2, binding = 0) uniform usampler2D src;
+
+// dst img
+DECLARE_BINDINGS(R8)
+DECLARE_BINDINGS(R16)
+DECLARE_BINDINGS(R32)
+DECLARE_BINDINGS(RG32)
+DECLARE_BINDINGS(RGBA32)
+
+layout (local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+
+void main()
+{
+    uvec2 coords = gl_GlobalInvocationID.xy;
+
+    ivec2 imageSz = ivec2(0, 0);
+
+    switch (convType)
+    {
+        case 0: break;
+        CASE_SIZE(R8    )
+        CASE_SIZE(R16   )
+        CASE_SIZE(R32   )
+        CASE_SIZE(RG32  )
+        CASE_SIZE(RGBA32)
+        default:
+            imageSz = textureSize(src, 0);
+            break;
+    }
+
+    if (int(coords.x) >= imageSz.x || int(coords.y) >= imageSz.y)
+    {
+        return;
+    }
+
+    int deltaX = sample_counts_log2.x - sample_counts_log2.z;
+    int deltaY = sample_counts_log2.y - sample_counts_log2.w;
+    int samplesInXLog2 = sample_counts_log2.z;
+    int samplesInYLog2 = sample_counts_log2.w;
+    int samplesInX = 1 << samplesInXLog2;
+    int samplesInY = 1 << samplesInYLog2;
+    int sampleIdx = ((int(coords.x) >> deltaX) & (samplesInX - 1)) | (((int(coords.y) >> deltaY) & (samplesInY - 1)) << samplesInXLog2);
+
+    samplesInXLog2 = sample_counts_log2.x;
+    samplesInYLog2 = sample_counts_log2.y;
+
+    ivec2 shiftedCoords = ivec2(int(coords.x) >> samplesInXLog2, int(coords.y) >> samplesInYLog2);
+
+    switch (convType)
+    {
+        CASE_CONVERT(R8    )
+        CASE_CONVERT(R16   )
+        CASE_CONVERT(R32   )
+        CASE_CONVERT(RG32  )
+        CASE_CONVERT(RGBA32)
+    }
+}

Ryujinx.Graphics.Vulkan/Shaders/ShaderBinaries.cs

@@ -551,5 +551,324 @@ namespace Ryujinx.Graphics.Vulkan.Shaders
             0x21, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
             0x35, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0xFD, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
         };
+
+        public static readonly byte[] ColorCopyBetweenMsNonMs = new byte[]
+        {
+            0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x0A, 0x00, 0x08, 0x00, 0x2D, 0x01, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
+            0x1B, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x31, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
+            0x32, 0x00, 0x00, 0x00, 0x0B, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x4C, 0x53, 0x4C,
+            0x2E, 0x73, 0x74, 0x64, 0x2E, 0x34, 0x35, 0x30, 0x00, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x03, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0F, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00,
+            0x04, 0x00, 0x00, 0x00, 0x6D, 0x61, 0x69, 0x6E, 0x00, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
+            0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x03, 0x00, 0x03, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0xC2, 0x01, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6D, 0x61, 0x69, 0x6E,
+            0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x67, 0x6C, 0x5F, 0x47,
+            0x6C, 0x6F, 0x62, 0x61, 0x6C, 0x49, 0x6E, 0x76, 0x6F, 0x63, 0x61, 0x74, 0x69, 0x6F, 0x6E, 0x49,
+            0x44, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00, 0x15, 0x00, 0x00, 0x00, 0x63, 0x6F, 0x6E, 0x76,
+            0x54, 0x79, 0x70, 0x65, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x52, 0x38, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x27, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x52, 0x31, 0x36, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x2D, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x52, 0x33, 0x32, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x33, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x52, 0x47, 0x33, 0x32, 0x00, 0x05, 0x00, 0x05, 0x00, 0x39, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x52, 0x47, 0x42, 0x41, 0x33, 0x32, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x73, 0x72, 0x63, 0x00, 0x05, 0x00, 0x08, 0x00, 0x60, 0x00, 0x00, 0x00,
+            0x73, 0x61, 0x6D, 0x70, 0x6C, 0x65, 0x5F, 0x63, 0x6F, 0x75, 0x6E, 0x74, 0x73, 0x5F, 0x6C, 0x6F,
+            0x67, 0x32, 0x5F, 0x69, 0x6E, 0x00, 0x00, 0x00, 0x06, 0x00, 0x08, 0x00, 0x60, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x73, 0x61, 0x6D, 0x70, 0x6C, 0x65, 0x5F, 0x63, 0x6F, 0x75, 0x6E, 0x74,
+            0x73, 0x5F, 0x6C, 0x6F, 0x67, 0x32, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00, 0x62, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x73, 0x72, 0x63, 0x4D,
+            0x53, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0xBD, 0x00, 0x00, 0x00, 0x64, 0x73, 0x74, 0x4D,
+            0x53, 0x52, 0x38, 0x00, 0x05, 0x00, 0x05, 0x00, 0xD2, 0x00, 0x00, 0x00, 0x64, 0x73, 0x74, 0x4D,
+            0x53, 0x52, 0x31, 0x36, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00, 0xE7, 0x00, 0x00, 0x00,
+            0x64, 0x73, 0x74, 0x4D, 0x53, 0x52, 0x33, 0x32, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00,
+            0xFC, 0x00, 0x00, 0x00, 0x64, 0x73, 0x74, 0x4D, 0x53, 0x52, 0x47, 0x33, 0x32, 0x00, 0x00, 0x00,
+            0x05, 0x00, 0x05, 0x00, 0x11, 0x01, 0x00, 0x00, 0x64, 0x73, 0x74, 0x4D, 0x53, 0x52, 0x47, 0x42,
+            0x41, 0x33, 0x32, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x0B, 0x00, 0x00, 0x00,
+            0x1C, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x27, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x27, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x33, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x33, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x39, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x39, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x40, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x40, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x60, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x60, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x62, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x62, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xBD, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xBD, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xD2, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xD2, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xE7, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xE7, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xFC, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0xFC, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x11, 0x01, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x11, 0x01, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1D, 0x01, 0x00, 0x00, 0x0B, 0x00, 0x00, 0x00,
+            0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
+            0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0A, 0x00, 0x00, 0x00,
+            0x06, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0B, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x0B, 0x00, 0x00, 0x00,
+            0x0C, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x05, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0x14, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00,
+            0x1F, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0x00, 0x00, 0x00,
+            0x3B, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x19, 0x00, 0x09, 0x00, 0x25, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x26, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x25, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0x2B, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x2C, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x2C, 0x00, 0x00, 0x00,
+            0x2D, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0x31, 0x00, 0x00, 0x00,
+            0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+            0x32, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
+            0x32, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00,
+            0x37, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x04, 0x00, 0x38, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00,
+            0x3B, 0x00, 0x04, 0x00, 0x38, 0x00, 0x00, 0x00, 0x39, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x19, 0x00, 0x09, 0x00, 0x3D, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x03, 0x00, 0x3E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x04, 0x00, 0x3F, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0x3B, 0x00, 0x04, 0x00, 0x3F, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x14, 0x00, 0x02, 0x00, 0x46, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x47, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x53, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x5F, 0x00, 0x00, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x03, 0x00, 0x60, 0x00, 0x00, 0x00,
+            0x5F, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x61, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x60, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x61, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x63, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x66, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x6D, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x78, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0xB0, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x03, 0x00, 0xB1, 0x00, 0x00, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0xB2, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0xB2, 0x00, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0xB8, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x04, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0xBB, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0xBC, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0xBB, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0xBC, 0x00, 0x00, 0x00,
+            0xBD, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0xD0, 0x00, 0x00, 0x00,
+            0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+            0xD1, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD0, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
+            0xD1, 0x00, 0x00, 0x00, 0xD2, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00,
+            0xE5, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+            0x20, 0x00, 0x04, 0x00, 0xE6, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xE5, 0x00, 0x00, 0x00,
+            0x3B, 0x00, 0x04, 0x00, 0xE6, 0x00, 0x00, 0x00, 0xE7, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x19, 0x00, 0x09, 0x00, 0xFA, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x23, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0xFB, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0xFA, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0xFB, 0x00, 0x00, 0x00, 0xFC, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0x0F, 0x01, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x10, 0x01, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x0F, 0x01, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x10, 0x01, 0x00, 0x00,
+            0x11, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x1C, 0x01, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x06, 0x00, 0x0A, 0x00, 0x00, 0x00,
+            0x1D, 0x01, 0x00, 0x00, 0x1C, 0x01, 0x00, 0x00, 0x1C, 0x01, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00,
+            0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x03, 0x00,
+            0x1E, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFB, 0x00, 0x03, 0x00, 0x47, 0x00, 0x00, 0x00,
+            0x1F, 0x01, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x1F, 0x01, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x0A, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x07, 0x00,
+            0x07, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x03, 0x00, 0x1D, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0xFB, 0x00, 0x0F, 0x00, 0x15, 0x00, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+            0x08, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0x41, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0x42, 0x00, 0x00, 0x00, 0x41, 0x00, 0x00, 0x00, 0x67, 0x00, 0x05, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0x43, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x16, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x17, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x1F, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x68, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x18, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x25, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x68, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x19, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x2B, 0x00, 0x00, 0x00, 0x2E, 0x00, 0x00, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x68, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x2F, 0x00, 0x00, 0x00, 0x2E, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x31, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x68, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x1B, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x37, 0x00, 0x00, 0x00, 0x3A, 0x00, 0x00, 0x00, 0x39, 0x00, 0x00, 0x00, 0x68, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x3A, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1D, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x1D, 0x00, 0x00, 0x00, 0xF5, 0x00, 0x11, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x2C, 0x01, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00,
+            0x14, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00,
+            0x29, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x2F, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+            0x35, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x00, 0x00,
+            0x51, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x4A, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00,
+            0x4A, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x4E, 0x00, 0x00, 0x00,
+            0x2C, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xAF, 0x00, 0x05, 0x00, 0x46, 0x00, 0x00, 0x00,
+            0x4F, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00, 0x4E, 0x00, 0x00, 0x00, 0xA8, 0x00, 0x04, 0x00,
+            0x46, 0x00, 0x00, 0x00, 0x50, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x03, 0x00,
+            0x52, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFA, 0x00, 0x04, 0x00, 0x50, 0x00, 0x00, 0x00,
+            0x51, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x51, 0x00, 0x00, 0x00,
+            0x51, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x55, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x56, 0x00, 0x00, 0x00,
+            0x55, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00,
+            0x2C, 0x01, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0xAF, 0x00, 0x05, 0x00, 0x46, 0x00, 0x00, 0x00,
+            0x59, 0x00, 0x00, 0x00, 0x56, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x52, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x52, 0x00, 0x00, 0x00, 0xF5, 0x00, 0x07, 0x00,
+            0x46, 0x00, 0x00, 0x00, 0x5A, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x1D, 0x00, 0x00, 0x00,
+            0x59, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x03, 0x00, 0x5C, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0xFA, 0x00, 0x04, 0x00, 0x5A, 0x00, 0x00, 0x00, 0x5B, 0x00, 0x00, 0x00,
+            0x5C, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x5B, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1E, 0x01, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x5C, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
+            0x63, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
+            0x47, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00,
+            0x64, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x63, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00,
+            0x62, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x66, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x69, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00,
+            0x41, 0x00, 0x06, 0x00, 0x63, 0x00, 0x00, 0x00, 0x6B, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00,
+            0x6C, 0x00, 0x00, 0x00, 0x6B, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x63, 0x00, 0x00, 0x00,
+            0x6E, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x6D, 0x00, 0x00, 0x00,
+            0x3D, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x6F, 0x00, 0x00, 0x00, 0x6E, 0x00, 0x00, 0x00,
+            0x82, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00, 0x6C, 0x00, 0x00, 0x00,
+            0x6F, 0x00, 0x00, 0x00, 0xC4, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x7A, 0x00, 0x00, 0x00,
+            0x78, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0xC4, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00,
+            0x7D, 0x00, 0x00, 0x00, 0x78, 0x00, 0x00, 0x00, 0x6F, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x05, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x83, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00, 0x69, 0x00, 0x00, 0x00,
+            0x82, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x85, 0x00, 0x00, 0x00, 0x7A, 0x00, 0x00, 0x00,
+            0x78, 0x00, 0x00, 0x00, 0xC7, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x86, 0x00, 0x00, 0x00,
+            0x83, 0x00, 0x00, 0x00, 0x85, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x88, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x89, 0x00, 0x00, 0x00, 0x88, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x05, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x8B, 0x00, 0x00, 0x00, 0x89, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
+            0x82, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x8D, 0x00, 0x00, 0x00, 0x7D, 0x00, 0x00, 0x00,
+            0x78, 0x00, 0x00, 0x00, 0xC7, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x8E, 0x00, 0x00, 0x00,
+            0x8B, 0x00, 0x00, 0x00, 0x8D, 0x00, 0x00, 0x00, 0xC4, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00,
+            0x90, 0x00, 0x00, 0x00, 0x8E, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0xC5, 0x00, 0x05, 0x00,
+            0x0F, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x86, 0x00, 0x00, 0x00, 0x90, 0x00, 0x00, 0x00,
+            0xC3, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x9B, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00,
+            0x65, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00, 0xA0, 0x00, 0x00, 0x00,
+            0x89, 0x00, 0x00, 0x00, 0x6C, 0x00, 0x00, 0x00, 0x50, 0x00, 0x05, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0xA1, 0x00, 0x00, 0x00, 0x9B, 0x00, 0x00, 0x00, 0xA0, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x03, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFB, 0x00, 0x17, 0x00, 0x15, 0x00, 0x00, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0xA2, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
+            0xA3, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0xA4, 0x00, 0x00, 0x00, 0xFE, 0xFF, 0xFF, 0xFF,
+            0xA5, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xA6, 0x00, 0x00, 0x00, 0xFC, 0xFF, 0xFF, 0xFF,
+            0xA7, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0xA8, 0x00, 0x00, 0x00, 0xF8, 0xFF, 0xFF, 0xFF,
+            0xA9, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0xAA, 0x00, 0x00, 0x00, 0xF0, 0xFF, 0xFF, 0xFF,
+            0xAB, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xA2, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x1F, 0x00, 0x00, 0x00, 0xAD, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0xAF, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0xB4, 0x00, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0xB7, 0x00, 0x00, 0x00, 0xB4, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00,
+            0xB8, 0x00, 0x00, 0x00, 0xB9, 0x00, 0x00, 0x00, 0xB7, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x63, 0x00, 0x04, 0x00, 0xAD, 0x00, 0x00, 0x00,
+            0xAF, 0x00, 0x00, 0x00, 0xB9, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0xA3, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0xBB, 0x00, 0x00, 0x00,
+            0xBE, 0x00, 0x00, 0x00, 0xBD, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0xC1, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0xC3, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0xC4, 0x00, 0x00, 0x00, 0xC1, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00, 0xB8, 0x00, 0x00, 0x00,
+            0xC5, 0x00, 0x00, 0x00, 0xC4, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x63, 0x00, 0x06, 0x00, 0xBE, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0xC5, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xA4, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x25, 0x00, 0x00, 0x00, 0xC7, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0xC9, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0xCA, 0x00, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0xCD, 0x00, 0x00, 0x00, 0xCA, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00,
+            0xB8, 0x00, 0x00, 0x00, 0xCE, 0x00, 0x00, 0x00, 0xCD, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x63, 0x00, 0x04, 0x00, 0xC7, 0x00, 0x00, 0x00,
+            0xC9, 0x00, 0x00, 0x00, 0xCE, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0xA5, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0xD0, 0x00, 0x00, 0x00,
+            0xD3, 0x00, 0x00, 0x00, 0xD2, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0xD6, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0xD8, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0xD9, 0x00, 0x00, 0x00, 0xD6, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00, 0xB8, 0x00, 0x00, 0x00,
+            0xDA, 0x00, 0x00, 0x00, 0xD9, 0x00, 0x00, 0x00, 0xD8, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x63, 0x00, 0x06, 0x00, 0xD3, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0xDA, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xA6, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x2B, 0x00, 0x00, 0x00, 0xDC, 0x00, 0x00, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0xDE, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0xDF, 0x00, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0xE2, 0x00, 0x00, 0x00, 0xDF, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00,
+            0xB8, 0x00, 0x00, 0x00, 0xE3, 0x00, 0x00, 0x00, 0xE2, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x63, 0x00, 0x04, 0x00, 0xDC, 0x00, 0x00, 0x00,
+            0xDE, 0x00, 0x00, 0x00, 0xE3, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0xA7, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0xE5, 0x00, 0x00, 0x00,
+            0xE8, 0x00, 0x00, 0x00, 0xE7, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0xEB, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0xED, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0xEE, 0x00, 0x00, 0x00, 0xEB, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00, 0xB8, 0x00, 0x00, 0x00,
+            0xEF, 0x00, 0x00, 0x00, 0xEE, 0x00, 0x00, 0x00, 0xED, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x63, 0x00, 0x06, 0x00, 0xE8, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0xEF, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xA8, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x31, 0x00, 0x00, 0x00, 0xF1, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0xF3, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0xF4, 0x00, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x00, 0x00, 0xF4, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00,
+            0xB8, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x00, 0x00, 0xF7, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x63, 0x00, 0x04, 0x00, 0xF1, 0x00, 0x00, 0x00,
+            0xF3, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0xA9, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0xFA, 0x00, 0x00, 0x00,
+            0xFD, 0x00, 0x00, 0x00, 0xFC, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0x00, 0x01, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0x02, 0x01, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0x03, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00, 0xB8, 0x00, 0x00, 0x00,
+            0x04, 0x01, 0x00, 0x00, 0x03, 0x01, 0x00, 0x00, 0x02, 0x01, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x63, 0x00, 0x06, 0x00, 0xFD, 0x00, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x04, 0x01, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xAA, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0x37, 0x00, 0x00, 0x00, 0x06, 0x01, 0x00, 0x00, 0x39, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x08, 0x01, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
+            0xB1, 0x00, 0x00, 0x00, 0x09, 0x01, 0x00, 0x00, 0xB3, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00,
+            0xB0, 0x00, 0x00, 0x00, 0x0C, 0x01, 0x00, 0x00, 0x09, 0x01, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00,
+            0xB8, 0x00, 0x00, 0x00, 0x0D, 0x01, 0x00, 0x00, 0x0C, 0x01, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0x63, 0x00, 0x04, 0x00, 0x06, 0x01, 0x00, 0x00,
+            0x08, 0x01, 0x00, 0x00, 0x0D, 0x01, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00,
+            0xF8, 0x00, 0x02, 0x00, 0xAB, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x0F, 0x01, 0x00, 0x00,
+            0x12, 0x01, 0x00, 0x00, 0x11, 0x01, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x3E, 0x00, 0x00, 0x00,
+            0x15, 0x01, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+            0x17, 0x01, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x64, 0x00, 0x04, 0x00, 0x3D, 0x00, 0x00, 0x00,
+            0x18, 0x01, 0x00, 0x00, 0x15, 0x01, 0x00, 0x00, 0x5F, 0x00, 0x07, 0x00, 0xB8, 0x00, 0x00, 0x00,
+            0x19, 0x01, 0x00, 0x00, 0x18, 0x01, 0x00, 0x00, 0x17, 0x01, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x13, 0x00, 0x00, 0x00, 0x63, 0x00, 0x06, 0x00, 0x12, 0x01, 0x00, 0x00, 0xA1, 0x00, 0x00, 0x00,
+            0x19, 0x01, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x91, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0xAC, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0xAC, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
+            0x1E, 0x01, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x1E, 0x01, 0x00, 0x00, 0xFD, 0x00, 0x01, 0x00,
+            0x38, 0x00, 0x01, 0x00,
+        };
     }
 }

Ryujinx.Graphics.Vulkan/SpecInfo.cs

@@ -0,0 +1,103 @@
+using Silk.NET.Vulkan;
+using System;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    public enum SpecConstType
+    {
+        Bool32,
+        Int16,
+        Int32,
+        Int64,
+        Float16,
+        Float32,
+        Float64
+    }
+
+    sealed class SpecDescription
+    {
+        public readonly SpecializationInfo Info;
+        public readonly SpecializationMapEntry[] Map;
+
+        // For mapping a simple packed struct or single entry
+        public SpecDescription(params (uint Id, SpecConstType Type)[] description)
+        {
+            int count = description.Length;
+            Map = new SpecializationMapEntry[count];
+
+            uint structSize = 0;
+
+            for (int i = 0; i < count; ++i)
+            {
+                var typeSize = SizeOf(description[i].Type);
+                Map[i] = new SpecializationMapEntry(description[i].Id, structSize, typeSize);
+                structSize += typeSize;
+            }
+
+            Info = new SpecializationInfo()
+            {
+                DataSize = structSize,
+                MapEntryCount = (uint)count
+            };
+        }
+
+        // For advanced mapping with overlapping or staggered fields
+        public SpecDescription(SpecializationMapEntry[] map)
+        {
+            int count = map.Length;
+            Map = map;
+
+            uint structSize = 0;
+            for (int i = 0; i < count; ++i)
+            {
+                structSize = Math.Max(structSize, map[i].Offset + (uint)map[i].Size);
+            }
+
+            Info = new SpecializationInfo()
+            {
+                DataSize = structSize,
+                MapEntryCount = (uint)map.Length
+            };
+        }
+
+        private static uint SizeOf(SpecConstType type) => type switch
+        {
+            SpecConstType.Int16 or SpecConstType.Float16 => 2,
+            SpecConstType.Bool32 or SpecConstType.Int32 or SpecConstType.Float32 => 4,
+            SpecConstType.Int64 or SpecConstType.Float64 => 8,
+            _ => throw new ArgumentOutOfRangeException(nameof(type))
+        };
+
+        private SpecDescription()
+        {
+            Info = new();
+        }
+
+        public static readonly SpecDescription Empty = new();
+    }
+
+    readonly struct SpecData : IRefEquatable<SpecData>
+    {
+        private readonly byte[] _data;
+        private readonly int _hash;
+
+        public int Length => _data.Length;
+        public ReadOnlySpan<byte> Span => _data.AsSpan();
+        public override int GetHashCode() => _hash;
+
+        public SpecData(ReadOnlySpan<byte> data)
+        {
+            _data = new byte[data.Length];
+            data.CopyTo(_data);
+
+            var hc = new HashCode();
+            hc.AddBytes(data);
+            _hash = hc.ToHashCode();
+        }
+
+        public override bool Equals(object obj) => obj is SpecData other && Equals(other);
+        public bool Equals(ref SpecData other) => _data.AsSpan().SequenceEqual(other._data);
+    }
+}

Ryujinx.Graphics.Vulkan/TextureCopy.cs

@@ -355,5 +355,122 @@ namespace Ryujinx.Graphics.Vulkan
                 dstLayers,
                 levels);
         }
+
+        public unsafe static void ResolveDepthStencil(
+            VulkanRenderer gd,
+            Device device,
+            CommandBufferScoped cbs,
+            TextureView src,
+            TextureView dst)
+        {
+            var dsAttachmentReference = new AttachmentReference2(StructureType.AttachmentReference2, null, 0, ImageLayout.General);
+            var dsResolveAttachmentReference = new AttachmentReference2(StructureType.AttachmentReference2, null, 1, ImageLayout.General);
+
+            var subpassDsResolve = new SubpassDescriptionDepthStencilResolve()
+            {
+                SType = StructureType.SubpassDescriptionDepthStencilResolve,
+                PDepthStencilResolveAttachment = &dsResolveAttachmentReference,
+                DepthResolveMode = ResolveModeFlags.ResolveModeSampleZeroBit,
+                StencilResolveMode = ResolveModeFlags.ResolveModeSampleZeroBit
+            };
+
+            var subpass = new SubpassDescription2()
+            {
+                SType = StructureType.SubpassDescription2,
+                PipelineBindPoint = PipelineBindPoint.Graphics,
+                PDepthStencilAttachment = &dsAttachmentReference,
+                PNext = &subpassDsResolve
+            };
+
+            AttachmentDescription2[] attachmentDescs = new AttachmentDescription2[2];
+
+            attachmentDescs[0] = new AttachmentDescription2(
+                StructureType.AttachmentDescription2,
+                null,
+                0,
+                src.VkFormat,
+                TextureStorage.ConvertToSampleCountFlags((uint)src.Info.Samples),
+                AttachmentLoadOp.Load,
+                AttachmentStoreOp.Store,
+                AttachmentLoadOp.Load,
+                AttachmentStoreOp.Store,
+                ImageLayout.General,
+                ImageLayout.General);
+
+            attachmentDescs[1] = new AttachmentDescription2(
+                StructureType.AttachmentDescription2,
+                null,
+                0,
+                dst.VkFormat,
+                TextureStorage.ConvertToSampleCountFlags((uint)dst.Info.Samples),
+                AttachmentLoadOp.Load,
+                AttachmentStoreOp.Store,
+                AttachmentLoadOp.Load,
+                AttachmentStoreOp.Store,
+                ImageLayout.General,
+                ImageLayout.General);
+
+            var subpassDependency = PipelineConverter.CreateSubpassDependency2();
+
+            fixed (AttachmentDescription2* pAttachmentDescs = attachmentDescs)
+            {
+                var renderPassCreateInfo = new RenderPassCreateInfo2()
+                {
+                    SType = StructureType.RenderPassCreateInfo2,
+                    PAttachments = pAttachmentDescs,
+                    AttachmentCount = (uint)attachmentDescs.Length,
+                    PSubpasses = &subpass,
+                    SubpassCount = 1,
+                    PDependencies = &subpassDependency,
+                    DependencyCount = 1
+                };
+
+                gd.Api.CreateRenderPass2(device, renderPassCreateInfo, null, out var renderPass).ThrowOnError();
+
+                using var rp = new Auto<DisposableRenderPass>(new DisposableRenderPass(gd.Api, device, renderPass));
+
+                ImageView* attachments = stackalloc ImageView[2];
+
+                var srcView = src.GetImageViewForAttachment();
+                var dstView = dst.GetImageViewForAttachment();
+
+                attachments[0] = srcView.Get(cbs).Value;
+                attachments[1] = dstView.Get(cbs).Value;
+
+                var framebufferCreateInfo = new FramebufferCreateInfo()
+                {
+                    SType = StructureType.FramebufferCreateInfo,
+                    RenderPass = rp.Get(cbs).Value,
+                    AttachmentCount = 2,
+                    PAttachments = attachments,
+                    Width = (uint)src.Width,
+                    Height = (uint)src.Height,
+                    Layers = (uint)src.Layers
+                };
+
+                gd.Api.CreateFramebuffer(device, framebufferCreateInfo, null, out var framebuffer).ThrowOnError();
+                using var fb = new Auto<DisposableFramebuffer>(new DisposableFramebuffer(gd.Api, device, framebuffer), null, new[] { srcView, dstView });
+
+                var renderArea = new Rect2D(null, new Extent2D((uint)src.Info.Width, (uint)src.Info.Height));
+                var clearValue = new ClearValue();
+
+                var renderPassBeginInfo = new RenderPassBeginInfo()
+                {
+                    SType = StructureType.RenderPassBeginInfo,
+                    RenderPass = rp.Get(cbs).Value,
+                    Framebuffer = fb.Get(cbs).Value,
+                    RenderArea = renderArea,
+                    PClearValues = &clearValue,
+                    ClearValueCount = 1
+                };
+
+                // The resolve operation happens at the end of the subpass, so let's just do a begin/end
+                // to resolve the depth-stencil texture.
+                // TODO: Do speculative resolve and part of the same render pass as the draw to avoid
+                // ending the current render pass?
+                gd.Api.CmdBeginRenderPass(cbs.CommandBuffer, renderPassBeginInfo, SubpassContents.Inline);
+                gd.Api.CmdEndRenderPass(cbs.CommandBuffer);
+            }
+        }
     }
 }

Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -169,12 +169,15 @@ namespace Ryujinx.Graphics.Vulkan
             var srcImage = src.GetImage().Get(cbs).Value;
             var dstImage = dst.GetImage().Get(cbs).Value;
 
-            if (src.Info.Target.IsMultisample())
+            if (!dst.Info.Target.IsMultisample() && Info.Target.IsMultisample())
             {
-                int depth = Math.Min(src.Info.Depth, dst.Info.Depth - firstLayer);
+                int layers = Math.Min(Info.GetLayers(), dst.Info.GetLayers() - firstLayer);
-                int levels = Math.Min(src.Info.Levels, dst.Info.Levels - firstLevel);
+                _gd.HelperShader.CopyMSToNonMS(_gd, cbs, src, dst, 0, firstLayer, layers);
-
+            }
-                CopyMSToNonMS(_gd, cbs, src, dst, srcImage, dstImage, 0, firstLayer, 0, firstLevel, depth, levels);
+            else if (dst.Info.Target.IsMultisample() && !Info.Target.IsMultisample())
+            {
+                int layers = Math.Min(Info.GetLayers(), dst.Info.GetLayers() - firstLayer);
+                _gd.HelperShader.CopyNonMSToMS(_gd, cbs, src, dst, 0, firstLayer, layers);
             }
             else
             {
@@ -213,9 +216,13 @@ namespace Ryujinx.Graphics.Vulkan
             var srcImage = src.GetImage().Get(cbs).Value;
             var dstImage = dst.GetImage().Get(cbs).Value;
 
-            if (src.Info.Target.IsMultisample())
+            if (!dst.Info.Target.IsMultisample() && Info.Target.IsMultisample())
             {
-                CopyMSToNonMS(_gd, cbs, src, dst, srcImage, dstImage, srcLayer, dstLayer, srcLevel, dstLevel, 1, 1);
+                _gd.HelperShader.CopyMSToNonMS(_gd, cbs, src, dst, srcLayer, dstLayer, 1);
+            }
+            else if (dst.Info.Target.IsMultisample() && !Info.Target.IsMultisample())
+            {
+                _gd.HelperShader.CopyNonMSToMS(_gd, cbs, src, dst, srcLayer, dstLayer, 1);
             }
             else
             {
@@ -239,142 +246,6 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        private static void CopyMSToNonMS(
-            VulkanRenderer gd,
-            CommandBufferScoped cbs,
-            TextureView src,
-            TextureView dst,
-            Image srcImage,
-            Image dstImage,
-            int srcLayer,
-            int dstLayer,
-            int srcLevel,
-            int dstLevel,
-            int layers,
-            int levels)
-        {
-            bool differentFormats = src.Info.Format != dst.Info.Format;
-
-            var target = src.Info.Target switch
-            {
-                Target.Texture2D => Target.Texture2DMultisample,
-                Target.Texture2DArray => Target.Texture2DMultisampleArray,
-                Target.Texture2DMultisampleArray => Target.Texture2DArray,
-                _ => Target.Texture2D
-            };
-
-            var intermmediateTarget = differentFormats ? dst.Info.Target : target;
-            using var intermmediate = CreateIntermmediateTexture(gd, src, ref dst._info, intermmediateTarget, layers, levels);
-            var intermmediateImage = intermmediate.GetImage().Get(cbs).Value;
-
-            if (differentFormats)
-            {
-                // If the formats are different, the resolve would perform format conversion.
-                // So we need yet another intermmediate texture and do a copy to reinterpret the
-                // data into the correct (destination) format, without doing any sort of conversion.
-                using var intermmediate2 = CreateIntermmediateTexture(gd, src, ref src._info, target, layers, levels);
-                var intermmediate2Image = intermmediate2.GetImage().Get(cbs).Value;
-
-                TextureCopy.Copy(
-                    gd.Api,
-                    cbs.CommandBuffer,
-                    srcImage,
-                    intermmediate2Image,
-                    src.Info,
-                    intermmediate2.Info,
-                    src.FirstLayer,
-                    0,
-                    src.FirstLevel,
-                    0,
-                    srcLayer,
-                    0,
-                    srcLevel,
-                    0,
-                    layers,
-                    levels);
-
-                TextureCopy.Copy(
-                    gd.Api,
-                    cbs.CommandBuffer,
-                    intermmediate2Image,
-                    intermmediateImage,
-                    intermmediate2.Info,
-                    intermmediate.Info,
-                    0,
-                    0,
-                    0,
-                    0,
-                    0,
-                    0,
-                    0,
-                    0,
-                    layers,
-                    levels);
-            }
-            else
-            {
-                TextureCopy.Copy(
-                    gd.Api,
-                    cbs.CommandBuffer,
-                    srcImage,
-                    intermmediateImage,
-                    src.Info,
-                    intermmediate.Info,
-                    src.FirstLayer,
-                    0,
-                    src.FirstLevel,
-                    0,
-                    srcLayer,
-                    0,
-                    srcLevel,
-                    0,
-                    layers,
-                    levels);
-            }
-
-            var srcRegion = new Extents2D(0, 0, src.Width, src.Height);
-            var dstRegion = new Extents2D(0, 0, dst.Width, dst.Height);
-
-            TextureCopy.Blit(
-                gd.Api,
-                cbs.CommandBuffer,
-                intermmediateImage,
-                dstImage,
-                intermmediate.Info,
-                dst.Info,
-                srcRegion,
-                dstRegion,
-                0,
-                dst.FirstLevel + dstLevel,
-                0,
-                dst.FirstLayer + dstLayer,
-                layers,
-                levels,
-                true,
-                ImageAspectFlags.ImageAspectColorBit,
-                ImageAspectFlags.ImageAspectColorBit);
-        }
-
-        private static TextureView CreateIntermmediateTexture(VulkanRenderer gd, TextureView src, ref TextureCreateInfo formatInfo, Target target, int depth, int levels)
-        {
-            return gd.CreateTextureView(new GAL.TextureCreateInfo(
-                src.Width,
-                src.Height,
-                depth,
-                levels,
-                1,
-                formatInfo.BlockWidth,
-                formatInfo.BlockHeight,
-                formatInfo.BytesPerPixel,
-                formatInfo.Format,
-                DepthStencilMode.Depth,
-                target,
-                SwizzleComponent.Red,
-                SwizzleComponent.Green,
-                SwizzleComponent.Blue,
-                SwizzleComponent.Alpha), 1f);
-        }
-
         public void CopyTo(ITexture destination, Extents2D srcRegion, Extents2D dstRegion, bool linearFilter)
         {
             var dst = (TextureView)destination;
@@ -422,23 +293,32 @@ namespace Ryujinx.Graphics.Vulkan
                     src.Height == dst.Height &&
                     src.VkFormat == dst.VkFormat)
                 {
-                    TextureCopy.Copy(
+                    if (src.Info.Samples > 1 && src.Info.Samples != dst.Info.Samples && src.Info.Format.IsDepthOrStencil())
-                        _gd.Api,
+                    {
-                        cbs.CommandBuffer,
+                        // CmdResolveImage does not support depth-stencil resolve, so we need to use an alternative path
-                        src.GetImage().Get(cbs).Value,
+                        // for those textures.
-                        dst.GetImage().Get(cbs).Value,
+                        TextureCopy.ResolveDepthStencil(_gd, _device, cbs, src, dst);
-                        src.Info,
+                    }
-                        dst.Info,
+                    else
-                        src.FirstLayer,
+                    {
-                        dst.FirstLayer,
+                        TextureCopy.Copy(
-                        src.FirstLevel,
+                            _gd.Api,
-                        dst.FirstLevel,
+                            cbs.CommandBuffer,
-                        0,
+                            src.GetImage().Get(cbs).Value,
-                        0,
+                            dst.GetImage().Get(cbs).Value,
-                        0,
+                            src.Info,
-                        0,
+                            dst.Info,
-                        layers,
+                            src.FirstLayer,
-                        levels);
+                            dst.FirstLayer,
+                            src.FirstLevel,
+                            dst.FirstLevel,
+                            0,
+                            0,
+                            0,
+                            0,
+                            layers,
+                            levels);
+                    }
 
                     return;
                 }

Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -1,4 +1,4 @@
-using Ryujinx.Common.Configuration;
+using Ryujinx.Common.Configuration;
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
@@ -53,7 +53,7 @@ namespace Ryujinx.Graphics.Vulkan
             "VUID-VkSubpassDependency-srcSubpass-00867"
         };
 
-        internal static Instance CreateInstance(Vk api, GraphicsDebugLevel logLevel, string[] requiredExtensions, out ExtDebugReport debugReport, out DebugReportCallbackEXT debugReportCallback)
+        internal static Instance CreateInstance(Vk api, GraphicsDebugLevel logLevel, string[] requiredExtensions, out ExtDebugUtils debugUtils, out DebugUtilsMessengerEXT debugUtilsMessenger)
         {
             var enabledLayers = new List<string>();
 
@@ -89,7 +89,7 @@ namespace Ryujinx.Graphics.Vulkan
                 AddAvailableLayer("VK_LAYER_KHRONOS_validation");
             }
 
-            var enabledExtensions = requiredExtensions.Append(ExtDebugReport.ExtensionName).ToArray();
+            var enabledExtensions = requiredExtensions.Append(ExtDebugUtils.ExtensionName).ToArray();
 
             var appName = Marshal.StringToHGlobalAnsi(AppName);
 
@@ -139,22 +139,18 @@ namespace Ryujinx.Graphics.Vulkan
                 Marshal.FreeHGlobal(ppEnabledLayers[i]);
             }
 
-            CreateDebugCallbacks(api, logLevel, instance, out debugReport, out debugReportCallback);
+            CreateDebugMessenger(api, logLevel, instance, out debugUtils, out debugUtilsMessenger);
 
             return instance;
         }
 
-        private unsafe static uint DebugReport(
+        private unsafe static uint DebugMessenger(
-            uint flags,
+            DebugUtilsMessageSeverityFlagsEXT messageSeverity,
-            DebugReportObjectTypeEXT objectType,
+            DebugUtilsMessageTypeFlagsEXT messageTypes,
-            ulong @object,
+            DebugUtilsMessengerCallbackDataEXT* pCallbackData,
-            nuint location,
+            void* pUserData)
-            int messageCode,
-            byte* layerPrefix,
-            byte* message,
-            void* userData)
         {
-            var msg = Marshal.PtrToStringAnsi((IntPtr)message);
+            var msg = Marshal.PtrToStringAnsi((IntPtr)pCallbackData->PMessage);
 
             foreach (string excludedMessagePart in _excludedMessages)
             {
@@ -164,26 +160,20 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            DebugReportFlagsEXT debugFlags = (DebugReportFlagsEXT)flags;
+            if (messageSeverity.HasFlag(DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityErrorBitExt))
-
-            if (debugFlags.HasFlag(DebugReportFlagsEXT.DebugReportErrorBitExt))
             {
                 Logger.Error?.Print(LogClass.Gpu, msg);
                 //throw new Exception(msg);
             }
-            else if (debugFlags.HasFlag(DebugReportFlagsEXT.DebugReportWarningBitExt))
+            else if (messageSeverity.HasFlag(DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityWarningBitExt))
             {
                 Logger.Warning?.Print(LogClass.Gpu, msg);
             }
-            else if (debugFlags.HasFlag(DebugReportFlagsEXT.DebugReportInformationBitExt))
+            else if (messageSeverity.HasFlag(DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityInfoBitExt))
             {
                 Logger.Info?.Print(LogClass.Gpu, msg);
             }
-            else if (debugFlags.HasFlag(DebugReportFlagsEXT.DebugReportPerformanceWarningBitExt))
+            else // if (messageSeverity.HasFlag(DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityVerboseBitExt))
-            {
-                Logger.Warning?.Print(LogClass.Gpu, msg);
-            }
-            else
             {
                 Logger.Debug?.Print(LogClass.Gpu, msg);
             }
@@ -403,6 +393,7 @@ namespace Ryujinx.Graphics.Vulkan
                 ShaderClipDistance = true,
                 ShaderFloat64 = supportedFeatures.ShaderFloat64,
                 ShaderImageGatherExtended = true,
+                ShaderStorageImageMultisample = supportedFeatures.ShaderStorageImageMultisample,
                 // ShaderStorageImageReadWithoutFormat = true,
                 // ShaderStorageImageWriteWithoutFormat = true,
                 TessellationShader = true,
@@ -551,46 +542,59 @@ namespace Ryujinx.Graphics.Vulkan
             return new CommandBufferPool(api, device, queue, queueLock, queueFamilyIndex);
         }
 
-        internal unsafe static void CreateDebugCallbacks(
+        internal unsafe static void CreateDebugMessenger(
             Vk api,
             GraphicsDebugLevel logLevel,
             Instance instance,
-            out ExtDebugReport debugReport,
+            out ExtDebugUtils debugUtils,
-            out DebugReportCallbackEXT debugReportCallback)
+            out DebugUtilsMessengerEXT debugUtilsMessenger)
         {
-            debugReport = default;
+            debugUtils = default;
 
             if (logLevel != GraphicsDebugLevel.None)
             {
-                if (!api.TryGetInstanceExtension(instance, out debugReport))
+                if (!api.TryGetInstanceExtension(instance, out debugUtils))
                 {
-                    debugReportCallback = default;
+                    debugUtilsMessenger = default;
                     return;
                 }
 
-                var flags = logLevel switch
+                var filterLogType = logLevel switch
                 {
-                    GraphicsDebugLevel.Error => DebugReportFlagsEXT.DebugReportErrorBitExt,
+                    GraphicsDebugLevel.Error => DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypeValidationBitExt,
-                    GraphicsDebugLevel.Slowdowns => DebugReportFlagsEXT.DebugReportErrorBitExt | DebugReportFlagsEXT.DebugReportPerformanceWarningBitExt,
+                    GraphicsDebugLevel.Slowdowns => DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypeValidationBitExt |
-                    GraphicsDebugLevel.All => DebugReportFlagsEXT.DebugReportInformationBitExt |
+                                                    DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypePerformanceBitExt,
-                                              DebugReportFlagsEXT.DebugReportWarningBitExt |
+                    GraphicsDebugLevel.All => DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypeGeneralBitExt |
-                                              DebugReportFlagsEXT.DebugReportPerformanceWarningBitExt |
+                                              DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypeValidationBitExt |
-                                              DebugReportFlagsEXT.DebugReportErrorBitExt |
+                                              DebugUtilsMessageTypeFlagsEXT.DebugUtilsMessageTypePerformanceBitExt,
-                                              DebugReportFlagsEXT.DebugReportDebugBitExt,
                     _ => throw new ArgumentException($"Invalid log level \"{logLevel}\".")
                 };
-                var debugReportCallbackCreateInfo = new DebugReportCallbackCreateInfoEXT()
+
+                var filterLogSeverity = logLevel switch
                 {
-                    SType = StructureType.DebugReportCallbackCreateInfoExt,
+                    GraphicsDebugLevel.Error => DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityErrorBitExt,
-                    Flags = flags,
+                    GraphicsDebugLevel.Slowdowns => DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityErrorBitExt |
-                    PfnCallback = new PfnDebugReportCallbackEXT(DebugReport)
+                                                    DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityWarningBitExt,
+                    GraphicsDebugLevel.All => DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityInfoBitExt |
+                                              DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityWarningBitExt |
+                                              DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityVerboseBitExt |
+                                              DebugUtilsMessageSeverityFlagsEXT.DebugUtilsMessageSeverityErrorBitExt,
+                    _ => throw new ArgumentException($"Invalid log level \"{logLevel}\".")
                 };
 
-                debugReport.CreateDebugReportCallback(instance, in debugReportCallbackCreateInfo, null, out debugReportCallback).ThrowOnError();
+                var debugUtilsMessengerCreateInfo = new DebugUtilsMessengerCreateInfoEXT()
+                {
+                    SType = StructureType.DebugUtilsMessengerCreateInfoExt,
+                    MessageType = filterLogType,
+                    MessageSeverity = filterLogSeverity,
+                    PfnUserCallback = new PfnDebugUtilsMessengerCallbackEXT(DebugMessenger)
+                };
+
+                debugUtils.CreateDebugUtilsMessenger(instance, in debugUtilsMessengerCreateInfo, null, out debugUtilsMessenger).ThrowOnError();
             }
             else
             {
-                debugReportCallback = default;
+                debugUtilsMessenger = default;
             }
         }
     }

Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -33,7 +33,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal KhrPushDescriptor PushDescriptorApi { get; private set; }
         internal ExtTransformFeedback TransformFeedbackApi { get; private set; }
         internal KhrDrawIndirectCount DrawIndirectCountApi { get; private set; }
-        internal ExtDebugReport DebugReportApi { get; private set; }
+        internal ExtDebugUtils DebugUtilsApi { get; private set; }
 
         internal uint QueueFamilyIndex { get; private set; }
         internal Queue Queue { get; private set; }
@@ -57,7 +57,7 @@ namespace Ryujinx.Graphics.Vulkan
         private SyncManager _syncManager;
 
         private PipelineFull _pipeline;
-        private DebugReportCallbackEXT _debugReportCallback;
+        private DebugUtilsMessengerEXT _debugUtilsMessenger;
 
         internal HelperShader HelperShader { get; private set; }
         internal PipelineFull PipelineInternal => _pipeline;
@@ -237,9 +237,9 @@ namespace Ryujinx.Graphics.Vulkan
 
             Api = api;
 
-            _instance = VulkanInitialization.CreateInstance(api, logLevel, _getRequiredExtensions(), out ExtDebugReport debugReport, out _debugReportCallback);
+            _instance = VulkanInitialization.CreateInstance(api, logLevel, _getRequiredExtensions(), out ExtDebugUtils debugUtils, out _debugUtilsMessenger);
 
-            DebugReportApi = debugReport;
+            DebugUtilsApi = debugUtils;
 
             if (api.TryGetInstanceExtension(_instance, out KhrSurface surfaceApi))
             {
@@ -287,9 +287,9 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        internal ShaderCollection CreateProgramWithMinimalLayout(ShaderSource[] sources)
+        internal ShaderCollection CreateProgramWithMinimalLayout(ShaderSource[] sources, SpecDescription[] specDescription = null)
         {
-            return new ShaderCollection(this, _device, sources, isMinimal: true);
+            return new ShaderCollection(this, _device, sources, specDescription: specDescription, isMinimal: true);
         }
 
         public ISampler CreateSampler(GAL.SamplerCreateInfo info)
@@ -584,9 +584,9 @@ namespace Ryujinx.Graphics.Vulkan
 
             MemoryAllocator.Dispose();
 
-            if (_debugReportCallback.Handle != 0)
+            if (_debugUtilsMessenger.Handle != 0)
             {
-                DebugReportApi.DestroyDebugReportCallback(_instance, _debugReportCallback, null);
+                DebugUtilsApi.DestroyDebugUtilsMessenger(_instance, _debugUtilsMessenger, null);
             }
 
             foreach (var shader in Shaders)

Ryujinx.Memory/Tracking/BitMap.cs

@@ -0,0 +1,199 @@
+using System.Runtime.CompilerServices;
+
+namespace Ryujinx.Memory.Tracking
+{
+    /// <summary>
+    /// A bitmap that can check or set large ranges of true/false values at once.
+    /// </summary>
+    struct BitMap
+    {
+        public const int IntSize = 64;
+
+        private const int IntShift = 6;
+        private const int IntMask = IntSize - 1;
+
+        /// <summary>
+        /// Masks representing the bitmap. Least significant bit first, 64-bits per mask.
+        /// </summary>
+        public readonly long[] Masks;
+
+        /// <summary>
+        /// Create a new bitmap.
+        /// </summary>
+        /// <param name="count">The number of bits to reserve</param>
+        public BitMap(int count)
+        {
+            Masks = new long[(count + IntMask) / IntSize];
+        }
+
+        /// <summary>
+        /// Check if any bit in the bitmap is set.
+        /// </summary>
+        /// <returns>True if any bits are set, false otherwise</returns>
+        public bool AnySet()
+        {
+            for (int i = 0; i < Masks.Length; i++)
+            {
+                if (Masks[i] != 0)
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        /// <summary>
+        /// Check if a bit in the bitmap is set.
+        /// </summary>
+        /// <param name="bit">The bit index to check</param>
+        /// <returns>True if the bit is set, false otherwise</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public bool IsSet(int bit)
+        {
+            int wordIndex = bit >> IntShift;
+            int wordBit = bit & IntMask;
+
+            long wordMask = 1L << wordBit;
+
+            return (Masks[wordIndex] & wordMask) != 0;
+        }
+
+        /// <summary>
+        /// Check if any bit in a range of bits in the bitmap are set. (inclusive)
+        /// </summary>
+        /// <param name="start">The first bit index to check</param>
+        /// <param name="end">The last bit index to check</param>
+        /// <returns>True if a bit is set, false otherwise</returns>
+        public bool IsSet(int start, int end)
+        {
+            if (start == end)
+            {
+                return IsSet(start);
+            }
+
+            int startIndex = start >> IntShift;
+            int startBit = start & IntMask;
+            long startMask = -1L << startBit;
+
+            int endIndex = end >> IntShift;
+            int endBit = end & IntMask;
+            long endMask = (long)(ulong.MaxValue >> (IntMask - endBit));
+
+            if (startIndex == endIndex)
+            {
+                return (Masks[startIndex] & startMask & endMask) != 0;
+            }
+
+            if ((Masks[startIndex] & startMask) != 0)
+            {
+                return true;
+            }
+
+            for (int i = startIndex + 1; i < endIndex; i++)
+            {
+                if (Masks[i] != 0)
+                {
+                    return true;
+                }
+            }
+
+            if ((Masks[endIndex] & endMask) != 0)
+            {
+                return true;
+            }
+
+            return false;
+        }
+
+        /// <summary>
+        /// Set a bit at a specific index to 1.
+        /// </summary>
+        /// <param name="bit">The bit index to set</param>
+        /// <returns>True if the bit is set, false if it was already set</returns>
+        public bool Set(int bit)
+        {
+            int wordIndex = bit >> IntShift;
+            int wordBit = bit & IntMask;
+
+            long wordMask = 1L << wordBit;
+
+            if ((Masks[wordIndex] & wordMask) != 0)
+            {
+                return false;
+            }
+
+            Masks[wordIndex] |= wordMask;
+
+            return true;
+        }
+
+        /// <summary>
+        /// Set a range of bits in the bitmap to 1.
+        /// </summary>
+        /// <param name="start">The first bit index to set</param>
+        /// <param name="end">The last bit index to set</param>
+        public void SetRange(int start, int end)
+        {
+            if (start == end)
+            {
+                Set(start);
+                return;
+            }
+
+            int startIndex = start >> IntShift;
+            int startBit = start & IntMask;
+            long startMask = -1L << startBit;
+
+            int endIndex = end >> IntShift;
+            int endBit = end & IntMask;
+            long endMask = (long)(ulong.MaxValue >> (IntMask - endBit));
+
+            if (startIndex == endIndex)
+            {
+                Masks[startIndex] |= startMask & endMask;
+            }
+            else
+            {
+                Masks[startIndex] |= startMask;
+
+                for (int i = startIndex + 1; i < endIndex; i++)
+                {
+                    Masks[i] |= -1;
+                }
+
+                Masks[endIndex] |= endMask;
+            }
+        }
+
+        /// <summary>
+        /// Clear a bit at a specific index to 0.
+        /// </summary>
+        /// <param name="bit">The bit index to clear</param>
+        /// <returns>True if the bit was set, false if it was not</returns>
+        public bool Clear(int bit)
+        {
+            int wordIndex = bit >> IntShift;
+            int wordBit = bit & IntMask;
+
+            long wordMask = 1L << wordBit;
+
+            bool wasSet = (Masks[wordIndex] & wordMask) != 0;
+
+            Masks[wordIndex] &= ~wordMask;
+
+            return wasSet;
+        }
+
+        /// <summary>
+        /// Clear the bitmap entirely, setting all bits to 0.
+        /// </summary>
+        public void Clear()
+        {
+            for (int i = 0; i < Masks.Length; i++)
+            {
+                Masks[i] = 0;
+            }
+        }
+    }
+}

Ryujinx.Memory/Tracking/ConcurrentBitmap.cs

@@ -0,0 +1,161 @@
+using System;
+using System.Threading;
+
+namespace Ryujinx.Memory.Tracking
+{
+    /// <summary>
+    /// A bitmap that can be safely modified from multiple threads.
+    /// </summary>
+    internal class ConcurrentBitmap
+    {
+        public const int IntSize = 64;
+
+        public const int IntShift = 6;
+        public const int IntMask = IntSize - 1;
+
+        /// <summary>
+        /// Masks representing the bitmap. Least significant bit first, 64-bits per mask.
+        /// </summary>
+        public readonly long[] Masks;
+
+        /// <summary>
+        /// Create a new multithreaded bitmap.
+        /// </summary>
+        /// <param name="count">The number of bits to reserve</param>
+        /// <param name="set">Whether the bits should be initially set or not</param>
+        public ConcurrentBitmap(int count, bool set)
+        {
+            Masks = new long[(count + IntMask) / IntSize];
+
+            if (set)
+            {
+                Array.Fill(Masks, -1L);
+            }
+        }
+
+        /// <summary>
+        /// Check if any bit in the bitmap is set.
+        /// </summary>
+        /// <returns>True if any bits are set, false otherwise</returns>
+        public bool AnySet()
+        {
+            for (int i = 0; i < Masks.Length; i++)
+            {
+                if (Volatile.Read(ref Masks[i]) != 0)
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        /// <summary>
+        /// Check if a bit in the bitmap is set.
+        /// </summary>
+        /// <param name="bit">The bit index to check</param>
+        /// <returns>True if the bit is set, false otherwise</returns>
+        public bool IsSet(int bit)
+        {
+            int wordIndex = bit >> IntShift;
+            int wordBit = bit & IntMask;
+
+            long wordMask = 1L << wordBit;
+
+            return (Volatile.Read(ref Masks[wordIndex]) & wordMask) != 0;
+        }
+
+        /// <summary>
+        /// Check if any bit in a range of bits in the bitmap are set. (inclusive)
+        /// </summary>
+        /// <param name="start">The first bit index to check</param>
+        /// <param name="end">The last bit index to check</param>
+        /// <returns>True if a bit is set, false otherwise</returns>
+        public bool IsSet(int start, int end)
+        {
+            if (start == end)
+            {
+                return IsSet(start);
+            }
+
+            int startIndex = start >> IntShift;
+            int startBit = start & IntMask;
+            long startMask = -1L << startBit;
+
+            int endIndex = end >> IntShift;
+            int endBit = end & IntMask;
+            long endMask = (long)(ulong.MaxValue >> (IntMask - endBit));
+
+            long startValue = Volatile.Read(ref Masks[startIndex]);
+
+            if (startIndex == endIndex)
+            {
+                return (startValue & startMask & endMask) != 0;
+            }
+
+            if ((startValue & startMask) != 0)
+            {
+                return true;
+            }
+
+            for (int i = startIndex + 1; i < endIndex; i++)
+            {
+                if (Volatile.Read(ref Masks[i]) != 0)
+                {
+                    return true;
+                }
+            }
+
+            long endValue = Volatile.Read(ref Masks[endIndex]);
+
+            if ((endValue & endMask) != 0)
+            {
+                return true;
+            }
+
+            return false;
+        }
+
+        /// <summary>
+        /// Set a bit at a specific index to either true or false.
+        /// </summary>
+        /// <param name="bit">The bit index to set</param>
+        /// <param name="value">Whether the bit should be set or not</param>
+        public void Set(int bit, bool value)
+        {
+            int wordIndex = bit >> IntShift;
+            int wordBit = bit & IntMask;
+
+            long wordMask = 1L << wordBit;
+
+            long existing;
+            long newValue;
+
+            do
+            {
+                existing = Volatile.Read(ref Masks[wordIndex]);
+
+                if (value)
+                {
+                    newValue = existing | wordMask;
+                }
+                else
+                {
+                    newValue = existing & ~wordMask;
+                }
+            }
+            while (Interlocked.CompareExchange(ref Masks[wordIndex], newValue, existing) != existing);
+        }
+
+        /// <summary>
+        /// Clear the bitmap entirely, setting all bits to 0.
+        /// </summary>
+        public void Clear()
+        {
+            for (int i = 0; i < Masks.Length; i++)
+            {
+                Volatile.Write(ref Masks[i], 0);
+            }
+        }
+    }
+}

Ryujinx.Memory/Tracking/MemoryTracking.cs

@@ -176,6 +176,26 @@ namespace Ryujinx.Memory.Tracking
             }
         }
 
+        /// <summary>
+        /// Obtains a memory tracking handle for the given virtual region. This should be disposed when finished with.
+        /// </summary>
+        /// <param name="address">CPU virtual address of the region</param>
+        /// <param name="size">Size of the region</param>
+        /// <param name="bitmap">The bitmap owning the dirty flag for this handle</param>
+        /// <param name="bit">The bit of this handle within the dirty flag</param>
+        /// <returns>The memory tracking handle</returns>
+        internal RegionHandle BeginTrackingBitmap(ulong address, ulong size, ConcurrentBitmap bitmap, int bit)
+        {
+            (address, size) = PageAlign(address, size);
+
+            lock (TrackingLock)
+            {
+                RegionHandle handle = new RegionHandle(this, address, size, bitmap, bit, _memoryManager.IsRangeMapped(address, size));
+
+                return handle;
+            }
+        }
+
         /// <summary>
         /// Signal that a virtual memory event happened at the given location (one byte).
         /// </summary>

Ryujinx.Memory/Tracking/MultiRegionHandle.cs

@@ -1,11 +1,15 @@
 using System;
 using System.Collections.Generic;
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Threading;
 
 namespace Ryujinx.Memory.Tracking
 {
     /// <summary>
     /// A region handle that tracks a large region using many smaller handles, to provide
-    /// granular tracking that can be used to track partial updates.
+    /// granular tracking that can be used to track partial updates. Backed by a bitmap
+    /// to improve performance when scanning large regions.
     /// </summary>
     public class MultiRegionHandle : IMultiRegionHandle
     {
@@ -17,6 +21,12 @@ namespace Ryujinx.Memory.Tracking
         private readonly ulong Granularity;
         private readonly ulong Size;
 
+        private ConcurrentBitmap _dirtyBitmap;
+
+        private int _sequenceNumber;
+        private BitMap _sequenceNumberBitmap;
+        private int _uncheckedHandles;
+
         public bool Dirty { get; private set; } = true;
 
         internal MultiRegionHandle(MemoryTracking tracking, ulong address, ulong size, IEnumerable<IRegionHandle> handles, ulong granularity)
@@ -24,6 +34,9 @@ namespace Ryujinx.Memory.Tracking
             _handles = new RegionHandle[size / granularity];
             Granularity = granularity;
 
+            _dirtyBitmap = new ConcurrentBitmap(_handles.Length, true);
+            _sequenceNumberBitmap = new BitMap(_handles.Length);
+
             int i = 0;
 
             if (handles != null)
@@ -40,37 +53,43 @@ namespace Ryujinx.Memory.Tracking
                     // Fill any gap left before this handle.
                     while (i < startIndex)
                     {
-                        RegionHandle fillHandle = tracking.BeginTracking(address + (ulong)i * granularity, granularity);
+                        RegionHandle fillHandle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
                         fillHandle.Parent = this;
                         _handles[i++] = fillHandle;
                     }
 
-                    if (handle.Size == granularity)
+                    lock (tracking.TrackingLock)
                     {
-                        handle.Parent = this;
+                        if (handle is RegionHandle bitHandle && handle.Size == granularity)
-                        _handles[i++] = handle;
-                    }
-                    else
-                    {
-                        int endIndex = (int)((handle.EndAddress - address) / granularity);
-
-                        while (i < endIndex)
                         {
-                            RegionHandle splitHandle = tracking.BeginTracking(address + (ulong)i * granularity, granularity);
+                            handle.Parent = this;
-                            splitHandle.Parent = this;
 
-                            splitHandle.Reprotect(handle.Dirty);
+                            bitHandle.ReplaceBitmap(_dirtyBitmap, i);
 
-                            RegionSignal signal = handle.PreAction;
+                            _handles[i++] = bitHandle;
-                            if (signal != null)
+                        }
+                        else
+                        {
+                            int endIndex = (int)((handle.EndAddress - address) / granularity);
+
+                            while (i < endIndex)
                             {
-                                splitHandle.RegisterAction(signal);
+                                RegionHandle splitHandle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
+                                splitHandle.Parent = this;
+
+                                splitHandle.Reprotect(handle.Dirty);
+
+                                RegionSignal signal = handle.PreAction;
+                                if (signal != null)
+                                {
+                                    splitHandle.RegisterAction(signal);
+                                }
+
+                                _handles[i++] = splitHandle;
                             }
 
-                            _handles[i++] = splitHandle;
+                            handle.Dispose();
                         }
-
-                        handle.Dispose();
                     }
                 }
             }
@@ -78,15 +97,27 @@ namespace Ryujinx.Memory.Tracking
             // Fill any remaining space with new handles.
             while (i < _handles.Length)
             {
-                RegionHandle handle = tracking.BeginTracking(address + (ulong)i * granularity, granularity);
+                RegionHandle handle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
                 handle.Parent = this;
                 _handles[i++] = handle;
             }
 
+            _uncheckedHandles = _handles.Length;
+
             Address = address;
             Size = size;
         }
 
+        public void SignalWrite()
+        {
+            Dirty = true;
+        }
+
+        public IEnumerable<RegionHandle> GetHandles()
+        {
+            return _handles;
+        }
+
         public void ForceDirty(ulong address, ulong size)
         {
             Dirty = true;
@@ -96,21 +127,15 @@ namespace Ryujinx.Memory.Tracking
 
             for (int i = startHandle; i <= lastHandle; i++)
             {
-                _handles[i].SequenceNumber--;
+                if (_sequenceNumberBitmap.Clear(i))
+                {
+                    _uncheckedHandles++;
+                }
+
                 _handles[i].ForceDirty();
             }
         }
 
-        public IEnumerable<RegionHandle> GetHandles()
-        {
-            return _handles;
-        }
-
-        public void SignalWrite()
-        {
-            Dirty = true;
-        }
-
         public void QueryModified(Action<ulong, ulong> modifiedAction)
         {
             if (!Dirty)
@@ -123,33 +148,95 @@ namespace Ryujinx.Memory.Tracking
             QueryModified(Address, Size, modifiedAction);
         }
 
-        public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction)
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private void ParseDirtyBits(long dirtyBits, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
         {
-            int startHandle = (int)((address - Address) / Granularity);
+            while (dirtyBits != 0)
-            int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
-
-            ulong rgStart = _handles[startHandle].Address;
-            ulong rgSize = 0;
-
-            for (int i = startHandle; i <= lastHandle; i++)
             {
-                RegionHandle handle = _handles[i];
+                int bit = BitOperations.TrailingZeroCount(dirtyBits);
+
+                dirtyBits &= ~(1L << bit);
+
+                int handleIndex = baseBit + bit;
+
+                RegionHandle handle = _handles[handleIndex];
+
+                if (handleIndex != prevHandle + 1)
+                {
+                    // Submit handles scanned until the gap as dirty
+                    if (rgSize != 0)
+                    {
+                        modifiedAction(rgStart, rgSize);
+                        rgSize = 0;
+                    }
+                    rgStart = handle.Address;
+                }
 
                 if (handle.Dirty)
                 {
                     rgSize += handle.Size;
                     handle.Reprotect();
                 }
-                else
+
+                prevHandle = handleIndex;
+            }
+
+            baseBit += ConcurrentBitmap.IntSize;
+        }
+
+        public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction)
+        {
+            int startHandle = (int)((address - Address) / Granularity);
+            int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
+
+            ulong rgStart = _handles[startHandle].Address;
+
+            if (startHandle == lastHandle)
+            {
+                RegionHandle handle = _handles[startHandle];
+
+                if (handle.Dirty)
                 {
-                    // Submit the region scanned so far as dirty
+                    handle.Reprotect();
-                    if (rgSize != 0)
+                    modifiedAction(rgStart, handle.Size);
-                    {
-                        modifiedAction(rgStart, rgSize);
-                        rgSize = 0;
-                    }
-                    rgStart = handle.EndAddress;
                 }
+
+                return;
+            }
+
+            ulong rgSize = 0;
+
+            long[] masks = _dirtyBitmap.Masks;
+
+            int startIndex = startHandle >> ConcurrentBitmap.IntShift;
+            int startBit = startHandle & ConcurrentBitmap.IntMask;
+            long startMask = -1L << startBit;
+
+            int endIndex = lastHandle >> ConcurrentBitmap.IntShift;
+            int endBit = lastHandle & ConcurrentBitmap.IntMask;
+            long endMask = (long)(ulong.MaxValue >> (ConcurrentBitmap.IntMask - endBit));
+
+            long startValue = Volatile.Read(ref masks[startIndex]);
+
+            int baseBit = startIndex << ConcurrentBitmap.IntShift;
+            int prevHandle = startHandle - 1;
+
+            if (startIndex == endIndex)
+            {
+                ParseDirtyBits(startValue & startMask & endMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+            }
+            else
+            {
+                ParseDirtyBits(startValue & startMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+
+                for (int i = startIndex + 1; i < endIndex; i++)
+                {
+                    ParseDirtyBits(Volatile.Read(ref masks[i]), ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                }
+
+                long endValue = Volatile.Read(ref masks[endIndex]);
+
+                ParseDirtyBits(endValue & endMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
             }
 
             if (rgSize != 0)
@@ -158,35 +245,120 @@ namespace Ryujinx.Memory.Tracking
             }
         }
 
-        public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction, int sequenceNumber)
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private void ParseDirtyBits(long dirtyBits, long mask, int index, long[] seqMasks, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
         {
-            int startHandle = (int)((address - Address) / Granularity);
+            long seqMask = mask & ~seqMasks[index];
-            int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
+            dirtyBits &= seqMask;
 
-            ulong rgStart = _handles[startHandle].Address;
+            while (dirtyBits != 0)
-            ulong rgSize = 0;
-
-            for (int i = startHandle; i <= lastHandle; i++)
             {
-                RegionHandle handle = _handles[i];
+                int bit = BitOperations.TrailingZeroCount(dirtyBits);
 
-                if (sequenceNumber != handle.SequenceNumber && handle.DirtyOrVolatile())
+                dirtyBits &= ~(1L << bit);
+
+                int handleIndex = baseBit + bit;
+
+                RegionHandle handle = _handles[handleIndex];
+
+                if (handleIndex != prevHandle + 1)
                 {
-                    rgSize += handle.Size;
+                    // Submit handles scanned until the gap as dirty
-                    handle.Reprotect();
-                }
-                else
-                {
-                    // Submit the region scanned so far as dirty
                     if (rgSize != 0)
                     {
                         modifiedAction(rgStart, rgSize);
                         rgSize = 0;
                     }
-                    rgStart = handle.EndAddress;
+                    rgStart = handle.Address;
                 }
 
-                handle.SequenceNumber = sequenceNumber;
+                rgSize += handle.Size;
+                handle.Reprotect();
+
+                prevHandle = handleIndex;
+            }
+
+            seqMasks[index] |= mask;
+            _uncheckedHandles -= BitOperations.PopCount((ulong)seqMask);
+
+            baseBit += ConcurrentBitmap.IntSize;
+        }
+
+        public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction, int sequenceNumber)
+        {
+            int startHandle = (int)((address - Address) / Granularity);
+            int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
+
+            ulong rgStart = Address + (ulong)startHandle * Granularity;
+
+            if (sequenceNumber != _sequenceNumber)
+            {
+                if (_uncheckedHandles != _handles.Length)
+                {
+                    _sequenceNumberBitmap.Clear();
+                    _uncheckedHandles = _handles.Length;
+                }
+
+                _sequenceNumber = sequenceNumber;
+            }
+
+            if (startHandle == lastHandle)
+            {
+                var handle = _handles[startHandle];
+                if (_sequenceNumberBitmap.Set(startHandle))
+                {
+                    _uncheckedHandles--;
+
+                    if (handle.DirtyOrVolatile())
+                    {
+                        handle.Reprotect();
+
+                        modifiedAction(rgStart, handle.Size);
+                    }
+                }
+
+                return;
+            }
+
+            if (_uncheckedHandles == 0)
+            {
+                return;
+            }
+
+            ulong rgSize = 0;
+
+            long[] seqMasks = _sequenceNumberBitmap.Masks;
+            long[] masks = _dirtyBitmap.Masks;
+
+            int startIndex = startHandle >> ConcurrentBitmap.IntShift;
+            int startBit = startHandle & ConcurrentBitmap.IntMask;
+            long startMask = -1L << startBit;
+
+            int endIndex = lastHandle >> ConcurrentBitmap.IntShift;
+            int endBit = lastHandle & ConcurrentBitmap.IntMask;
+            long endMask = (long)(ulong.MaxValue >> (ConcurrentBitmap.IntMask - endBit));
+
+            long startValue = Volatile.Read(ref masks[startIndex]);
+
+            int baseBit = startIndex << ConcurrentBitmap.IntShift;
+            int prevHandle = startHandle - 1;
+
+            if (startIndex == endIndex)
+            {
+                ParseDirtyBits(startValue, startMask & endMask, startIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+            }
+            else
+            {
+                ParseDirtyBits(startValue, startMask, startIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+
+                for (int i = startIndex + 1; i < endIndex; i++)
+                {
+                    ParseDirtyBits(Volatile.Read(ref masks[i]), -1L, i, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                }
+
+                long endValue = Volatile.Read(ref masks[endIndex]);
+
+                ParseDirtyBits(endValue, endMask, endIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
             }
 
             if (rgSize != 0)

Ryujinx.Memory/Tracking/RegionHandle.cs

@@ -1,5 +1,4 @@
-using Ryujinx.Memory.Range;
+using System;
-using System;
 using System.Collections.Generic;
 using System.Linq;
 using System.Threading;
@@ -10,7 +9,7 @@ namespace Ryujinx.Memory.Tracking
     /// A tracking handle for a given region of virtual memory. The Dirty flag is updated whenever any changes are made,
     /// and an action can be performed when the region is read to or written from.
     /// </summary>
-    public class RegionHandle : IRegionHandle, IRange
+    public class RegionHandle : IRegionHandle
     {
         /// <summary>
         /// If more than this number of checks have been performed on a dirty flag since its last reprotect,
@@ -23,7 +22,20 @@ namespace Ryujinx.Memory.Tracking
         /// </summary>
         private static int VolatileThreshold = 5;
 
-        public bool Dirty { get; private set; }
+        public bool Dirty
+        {
+            get
+            {
+                return Bitmap.IsSet(DirtyBit);
+            }
+            protected set
+            {
+                Bitmap.Set(DirtyBit, value);
+            }
+        }
+
+        internal int SequenceNumber { get; set; }
+
         public bool Unmapped { get; private set; }
 
         public ulong Address { get; }
@@ -31,7 +43,6 @@ namespace Ryujinx.Memory.Tracking
         public ulong EndAddress { get; }
 
         internal IMultiRegionHandle Parent { get; set; }
-        internal int SequenceNumber { get; set; }
 
         private event Action _onDirty;
 
@@ -68,17 +79,26 @@ namespace Ryujinx.Memory.Tracking
 
         internal RegionSignal PreAction => _preAction;
 
+        internal ConcurrentBitmap Bitmap;
+        internal int DirtyBit;
+
         /// <summary>
-        /// Create a new region handle. The handle is registered with the given tracking object,
+        /// Create a new bitmap backed region handle. The handle is registered with the given tracking object,
         /// and will be notified of any changes to the specified region.
         /// </summary>
         /// <param name="tracking">Tracking object for the target memory block</param>
         /// <param name="address">Virtual address of the region to track</param>
         /// <param name="size">Size of the region to track</param>
+        /// <param name="bitmap">The bitmap the dirty flag for this handle is stored in</param>
+        /// <param name="bit">The bit index representing the dirty flag for this handle</param>
         /// <param name="mapped">True if the region handle starts mapped</param>
-        internal RegionHandle(MemoryTracking tracking, ulong address, ulong size, bool mapped = true)
+        internal RegionHandle(MemoryTracking tracking, ulong address, ulong size, ConcurrentBitmap bitmap, int bit, bool mapped = true)
         {
+            Bitmap = bitmap;
+            DirtyBit = bit;
+
             Dirty = mapped;
+
             Unmapped = !mapped;
             Address = address;
             Size = size;
@@ -92,6 +112,54 @@ namespace Ryujinx.Memory.Tracking
             }
         }
 
+        /// <summary>
+        /// Create a new region handle. The handle is registered with the given tracking object,
+        /// and will be notified of any changes to the specified region.
+        /// </summary>
+        /// <param name="tracking">Tracking object for the target memory block</param>
+        /// <param name="address">Virtual address of the region to track</param>
+        /// <param name="size">Size of the region to track</param>
+        /// <param name="mapped">True if the region handle starts mapped</param>
+        internal RegionHandle(MemoryTracking tracking, ulong address, ulong size, bool mapped = true)
+        {
+            Bitmap = new ConcurrentBitmap(1, mapped);
+
+            Unmapped = !mapped;
+            Address = address;
+            Size = size;
+            EndAddress = address + size;
+
+            _tracking = tracking;
+            _regions = tracking.GetVirtualRegionsForHandle(address, size);
+            foreach (var region in _regions)
+            {
+                region.Handles.Add(this);
+            }
+        }
+
+        /// <summary>
+        /// Replace the bitmap and bit index used to track dirty state.
+        /// </summary>
+        /// <remarks>
+        /// The tracking lock should be held when this is called, to ensure neither bitmap is modified.
+        /// </remarks>
+        /// <param name="bitmap">The bitmap the dirty flag for this handle is stored in</param>
+        /// <param name="bit">The bit index representing the dirty flag for this handle</param>
+        internal void ReplaceBitmap(ConcurrentBitmap bitmap, int bit)
+        {
+            // Assumes the tracking lock is held, so nothing else can signal right now.
+
+            var oldBitmap = Bitmap;
+            var oldBit = DirtyBit;
+
+            bitmap.Set(bit, Dirty);
+
+            Bitmap = bitmap;
+            DirtyBit = bit;
+
+            Dirty |= oldBitmap.IsSet(oldBit);
+        }
+
         /// <summary>
         /// Clear the volatile state of this handle.
         /// </summary>
@@ -108,7 +176,7 @@ namespace Ryujinx.Memory.Tracking
         public bool DirtyOrVolatile()
         {
             _checkCount++;
-            return Dirty || _volatile;
+            return _volatile || Dirty;
         }
 
         /// <summary>
@@ -195,6 +263,8 @@ namespace Ryujinx.Memory.Tracking
         /// </summary>
         public void ForceDirty()
         {
+            _checkCount++;
+
             Dirty = true;
         }