Start building more accurate vertex as compute usage info

- Fixes

Directory.Packages.props

@@ -3,13 +3,13 @@
     <ManagePackageVersionsCentrally>true</ManagePackageVersionsCentrally>
   </PropertyGroup>
   <ItemGroup>
-    <PackageVersion Include="Avalonia" Version="12.1.1" />
+    <PackageVersion Include="Avalonia" Version="12.1.2" />
     <PackageVersion Include="Avalonia.Controls.DataGrid" Version="12.1.2" />
-    <PackageVersion Include="Avalonia.Desktop" Version="12.1.1" />
+    <PackageVersion Include="Avalonia.Desktop" Version="12.1.2" />
     <PackageVersion Include="AvaloniaUI.DiagnosticsSupport" Version="2.2.3" />
-    <PackageVersion Include="Avalonia.Markup.Xaml.Loader" Version="12.1.1" />
-    <PackageVersion Include="Svg.Controls.Avalonia" Version="12.0.0.15" />
-    <PackageVersion Include="Svg.Controls.Skia.Avalonia" Version="12.0.0.15" />
+    <PackageVersion Include="Avalonia.Markup.Xaml.Loader" Version="12.1.2" />
+    <PackageVersion Include="Svg.Controls.Avalonia" Version="12.0.0.17" />
+    <PackageVersion Include="Svg.Controls.Skia.Avalonia" Version="12.0.0.17" />
     <PackageVersion Include="CommandLineParser" Version="2.9.1" />
     <PackageVersion Include="CommunityToolkit.Mvvm" Version="8.4.2" />
     <PackageVersion Include="Concentus" Version="2.2.2" />
@@ -20,8 +20,8 @@
     <PackageVersion Include="Humanizer" Version="3.0.10" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.12.0" />
-    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="8.21.0" />
-    <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="18.8.1" />
+    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="8.22.0" />
+    <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="18.10.0" />
     <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.1" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />
     <PackageVersion Include="NetCoreServer" Version="8.0.7" />
@@ -36,7 +36,7 @@
     <PackageVersion Include="Ryujinx.Graphics.Nvdec.Dependencies.Linux" Version="6.1.4-build6" />
     <PackageVersion Include="Ryujinx.Graphics.Nvdec.Dependencies.macOS" Version="5.0.3-build14" />
     <PackageVersion Include="Ryujinx.Graphics.Nvdec.Dependencies.Windows" Version="6.1.4-build6" />
-    <PackageVersion Include="Ryujinx.Graphics.Vulkan.MoltenVK" Version="1.4.2-ryujinx.6" />
+    <PackageVersion Include="Ryujinx.Graphics.Vulkan.MoltenVK" Version="1.4.3-ryujinx.1" />
     <PackageVersion Include="Ryujinx.LibHac" Version="0.21.0-alpha.133" />
     <PackageVersion Include="Ryujinx.SDL3-CS" Version="2026.707.0" />
     <PackageVersion Include="securifybv.ShellLink" Version="0.1.0" />
@@ -48,7 +48,7 @@
     <PackageVersion Include="SkiaSharp" Version="3.119.4" />
     <PackageVersion Include="SkiaSharp.NativeAssets.Linux" Version="3.119.4" />
     <PackageVersion Include="SPB" Version="0.0.4-build32" />
-    <PackageVersion Include="System.IO.Hashing" Version="10.0.11" />
+    <PackageVersion Include="System.IO.Hashing" Version="10.0.12" />
     <PackageVersion Include="UnicornEngine.Unicorn" Version="2.1.0" />
   </ItemGroup>
-</Project>
+</Project>

Ryujinx.slnx

@@ -1,10 +1,4 @@
 <Solution>
-  <Configurations>
-    <BuildType Name="Debug" />
-    <BuildType Name="Debug AOT" />
-    <BuildType Name="Release" />
-    <BuildType Name="Release AOT" />
-  </Configurations>
   <Folder Name="/Solution Items/">
     <File Path=".editorconfig" />
     <File Path="Directory.Build.props" />
@@ -27,6 +21,7 @@
   <Project Path="src/Ryujinx.Graphics.Nvdec.Vp9/Ryujinx.Graphics.Nvdec.Vp9.csproj" />
   <Project Path="src/Ryujinx.Graphics.Nvdec/Ryujinx.Graphics.Nvdec.csproj" />
   <Project Path="src/Ryujinx.Graphics.OpenGL/Ryujinx.Graphics.OpenGL.csproj" />
+  <Project Path="src/Ryujinx.Graphics.RenderDocApi/Ryujinx.Graphics.RenderDocApi.csproj" />
   <Project Path="src/Ryujinx.Graphics.Shader/Ryujinx.Graphics.Shader.csproj" />
   <Project Path="src/Ryujinx.Graphics.Texture/Ryujinx.Graphics.Texture.csproj" />
   <Project Path="src/Ryujinx.Graphics.Vic/Ryujinx.Graphics.Vic.csproj" />

src/ARMeilleure/ARMeilleure.csproj

@@ -4,8 +4,6 @@
     <RuntimeIdentifiers>osx-arm64</RuntimeIdentifiers>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs

@@ -406,7 +406,7 @@ namespace ARMeilleure.Instructions
                 {
                     Operand res = EmitSoftFloatCall(context, nameof(SoftFloat32.FPSub), op1, op2);
 
-                    return EmitUnaryMathCall(context, nameof(Math.Abs), res);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Abs), res);
                 });
             }
         }
@@ -451,7 +451,7 @@ namespace ARMeilleure.Instructions
                 {
                     Operand res = EmitSoftFloatCall(context, nameof(SoftFloat32.FPSub), op1, op2);
 
-                    return EmitUnaryMathCall(context, nameof(Math.Abs), res);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Abs), res);
                 });
             }
         }
@@ -483,7 +483,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitScalarUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Abs), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Abs), op1);
                 });
             }
         }
@@ -522,7 +522,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitVectorUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Abs), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Abs), op1);
                 });
             }
         }
@@ -2246,7 +2246,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitScalarUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Floor), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Floor), op1);
                 });
             }
         }
@@ -2265,7 +2265,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitVectorUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Floor), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Floor), op1);
                 });
             }
         }
@@ -2322,7 +2322,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitScalarUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Ceiling), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), op1);
                 });
             }
         }
@@ -2341,7 +2341,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitVectorUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Ceiling), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), op1);
                 });
             }
         }
@@ -2390,7 +2390,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitScalarUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Truncate), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Truncate), op1);
                 });
             }
         }
@@ -2409,7 +2409,7 @@ namespace ARMeilleure.Instructions
             {
                 EmitVectorUnaryOpF(context, (op1) =>
                 {
-                    return EmitUnaryMathCall(context, nameof(Math.Truncate), op1);
+                    return EmitUnaryMathCall(context, nameof(MathHelper.Truncate), op1);
                 });
             }
         }

src/ARMeilleure/Instructions/InstEmitSimdArithmetic32.cs

@@ -43,7 +43,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Abs), op1));
+                EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Abs), op1));
             }
         }
 
@@ -66,7 +66,7 @@ namespace ARMeilleure.Instructions
                 }
                 else
                 {
-                    EmitVectorUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Abs), op1));
+                    EmitVectorUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Abs), op1));
                 }
             }
             else

src/ARMeilleure/Instructions/InstEmitSimdCmp.cs

@@ -726,8 +726,8 @@ namespace ARMeilleure.Instructions
 
                 if (absolute)
                 {
-                    ne = EmitUnaryMathCall(context, nameof(Math.Abs), ne);
-                    me = EmitUnaryMathCall(context, nameof(Math.Abs), me);
+                    ne = EmitUnaryMathCall(context, nameof(MathHelper.Abs), ne);
+                    me = EmitUnaryMathCall(context, nameof(MathHelper.Abs), me);
                 }
 
                 Operand e = EmitSoftFloatCall(context, name, ne, me);

src/ARMeilleure/Instructions/InstEmitSimdCvt.cs

@@ -333,7 +333,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Floor), op1));
+                EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Floor), op1));
             }
         }
 
@@ -349,7 +349,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitFcvt(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Floor), op1), signed: true, scalar: false);
+                EmitFcvt(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Floor), op1), signed: true, scalar: false);
             }
         }
 
@@ -365,7 +365,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Floor), op1));
+                EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Floor), op1));
             }
         }
 
@@ -538,7 +538,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Ceiling), op1));
+                EmitFcvt_s_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), op1));
             }
         }
 
@@ -554,7 +554,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Ceiling), op1));
+                EmitFcvt_u_Gp(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), op1));
             }
         }
 

src/ARMeilleure/Instructions/InstEmitSimdCvt32.cs

@@ -357,10 +357,10 @@ namespace ARMeilleure.Instructions
                         toConvert = EmitRoundMathCall(context, MidpointRounding.ToEven, toConvert);
                         break;
                     case 0b10: // Towards positive infinity
-                        toConvert = EmitUnaryMathCall(context, nameof(Math.Ceiling), toConvert);
+                        toConvert = EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), toConvert);
                         break;
                     case 0b11: // Towards negative infinity
-                        toConvert = EmitUnaryMathCall(context, nameof(Math.Floor), toConvert);
+                        toConvert = EmitUnaryMathCall(context, nameof(MathHelper.Floor), toConvert);
                         break;
                 }
 
@@ -494,10 +494,10 @@ namespace ARMeilleure.Instructions
                         toConvert = EmitRoundMathCall(context, MidpointRounding.ToEven, toConvert);
                         break;
                     case 0b10: // Towards positive infinity
-                        toConvert = EmitUnaryMathCall(context, nameof(Math.Ceiling), toConvert);
+                        toConvert = EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), toConvert);
                         break;
                     case 0b11: // Towards negative infinity
-                        toConvert = EmitUnaryMathCall(context, nameof(Math.Floor), toConvert);
+                        toConvert = EmitUnaryMathCall(context, nameof(MathHelper.Floor), toConvert);
                         break;
                 }
 
@@ -534,7 +534,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitVectorUnaryOpF32(context, (m) => EmitUnaryMathCall(context, nameof(Math.Floor), m));
+                EmitVectorUnaryOpF32(context, (m) => EmitUnaryMathCall(context, nameof(MathHelper.Floor), m));
             }
         }
 
@@ -574,7 +574,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitVectorUnaryOpF32(context, (m) => EmitUnaryMathCall(context, nameof(Math.Ceiling), m));
+                EmitVectorUnaryOpF32(context, (m) => EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), m));
             }
         }
 
@@ -613,7 +613,7 @@ namespace ARMeilleure.Instructions
             }
             else
             {
-                EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Truncate), op1));
+                EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(MathHelper.Truncate), op1));
             }
         }
 

src/ARMeilleure/Instructions/InstEmitSimdHelper.cs

@@ -460,8 +460,8 @@ namespace ARMeilleure.Instructions
             IOpCodeSimd op = (IOpCodeSimd)context.CurrOp;
 
             MethodInfo info = (op.Size & 1) == 0
-                ? typeof(MathF).GetMethod(name, [typeof(float)])
-                : typeof(Math).GetMethod(name, [typeof(double)]);
+                ? typeof(MathHelperF).GetMethod(name, [typeof(float)])
+                : typeof(MathHelper).GetMethod(name, [typeof(double)]);
 
             return context.Call(info, n);
         }
@@ -470,11 +470,11 @@ namespace ARMeilleure.Instructions
         {
             IOpCodeSimd op = (IOpCodeSimd)context.CurrOp;
 
-            string name = nameof(Math.Round);
+            string name = nameof(MathHelper.Round);
 
             MethodInfo info = (op.Size & 1) == 0
-                ? typeof(MathF).GetMethod(name, [typeof(float), typeof(MidpointRounding)])
-                : typeof(Math).GetMethod(name, [typeof(double), typeof(MidpointRounding)]);
+                ? typeof(MathHelperF).GetMethod(name, [typeof(float), typeof(int)])
+                : typeof(MathHelper).GetMethod(name, [typeof(double), typeof(int)]);
 
             return context.Call(info, n, Const((int)roundMode));
         }
@@ -510,16 +510,16 @@ namespace ARMeilleure.Instructions
 
             context.MarkLabel(lbl1);
             context.BranchIf(lbl2, rMode, rP, Comparison.NotEqual);
-            context.Copy(res, EmitUnaryMathCall(context, nameof(Math.Ceiling), op));
+            context.Copy(res, EmitUnaryMathCall(context, nameof(MathHelper.Ceiling), op));
             context.Branch(lblEnd);
 
             context.MarkLabel(lbl2);
             context.BranchIf(lbl3, rMode, rM, Comparison.NotEqual);
-            context.Copy(res, EmitUnaryMathCall(context, nameof(Math.Floor), op));
+            context.Copy(res, EmitUnaryMathCall(context, nameof(MathHelper.Floor), op));
             context.Branch(lblEnd);
 
             context.MarkLabel(lbl3);
-            context.Copy(res, EmitUnaryMathCall(context, nameof(Math.Truncate), op));
+            context.Copy(res, EmitUnaryMathCall(context, nameof(MathHelper.Truncate), op));
             context.Branch(lblEnd);
 
             context.MarkLabel(lblEnd);

src/ARMeilleure/Instructions/MathHelper.cs

@@ -0,0 +1,71 @@
+using System;
+using System.Runtime.InteropServices;
+
+namespace ARMeilleure.Instructions
+{
+    static class MathHelper
+    {
+        [UnmanagedCallersOnly]
+        public static double Abs(double value)
+        {
+            return Math.Abs(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static double Ceiling(double value)
+        {
+            return Math.Ceiling(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static double Floor(double value)
+        {
+            return Math.Floor(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static double Round(double value, int mode)
+        {
+            return Math.Round(value, (MidpointRounding)mode);
+        }
+
+        [UnmanagedCallersOnly]
+        public static double Truncate(double value)
+        {
+            return Math.Truncate(value);
+        }
+    }
+
+    static class MathHelperF
+    {
+        [UnmanagedCallersOnly]
+        public static float Abs(float value)
+        {
+            return MathF.Abs(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static float Ceiling(float value)
+        {
+            return MathF.Ceiling(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static float Floor(float value)
+        {
+            return MathF.Floor(value);
+        }
+
+        [UnmanagedCallersOnly]
+        public static float Round(float value, int mode)
+        {
+            return MathF.Round(value, (MidpointRounding)mode);
+        }
+
+        [UnmanagedCallersOnly]
+        public static float Truncate(float value)
+        {
+            return MathF.Truncate(value);
+        }
+    }
+}

src/ARMeilleure/Instructions/NativeInterface.cs

@@ -2,6 +2,7 @@ using ARMeilleure.Memory;
 using ARMeilleure.State;
 using ARMeilleure.Translation;
 using System;
+using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Instructions
 {
@@ -34,6 +35,7 @@ namespace ARMeilleure.Instructions
             Context = null;
         }
 
+        [UnmanagedCallersOnly]
         public static void Break(ulong address, int imm)
         {
             Statistics.PauseTimer();
@@ -43,6 +45,7 @@ namespace ARMeilleure.Instructions
             Statistics.ResumeTimer();
         }
 
+        [UnmanagedCallersOnly]
         public static void SupervisorCall(ulong address, int imm)
         {
             Statistics.PauseTimer();
@@ -52,6 +55,7 @@ namespace ARMeilleure.Instructions
             Statistics.ResumeTimer();
         }
 
+        [UnmanagedCallersOnly]
         public static void Undefined(ulong address, int opCode)
         {
             Statistics.PauseTimer();
@@ -62,26 +66,31 @@ namespace ARMeilleure.Instructions
         }
 
         #region "System registers"
+        [UnmanagedCallersOnly]
         public static ulong GetCtrEl0()
         {
             return GetContext().CtrEl0;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong GetDczidEl0()
         {
             return GetContext().DczidEl0;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong GetCntfrqEl0()
         {
             return GetContext().CntfrqEl0;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong GetCntpctEl0()
         {
             return GetContext().CntpctEl0;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong GetCntvctEl0()
         {
             return GetContext().CntvctEl0;
@@ -89,26 +98,31 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Read"
+        [UnmanagedCallersOnly]
         public static byte ReadByte(ulong address)
         {
             return GetMemoryManager().ReadGuest<byte>(address);
         }
 
+        [UnmanagedCallersOnly]
         public static ushort ReadUInt16(ulong address)
         {
             return GetMemoryManager().ReadGuest<ushort>(address);
         }
 
+        [UnmanagedCallersOnly]
         public static uint ReadUInt32(ulong address)
         {
             return GetMemoryManager().ReadGuest<uint>(address);
         }
 
+        [UnmanagedCallersOnly]
         public static ulong ReadUInt64(ulong address)
         {
             return GetMemoryManager().ReadGuest<ulong>(address);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 ReadVector128(ulong address)
         {
             return GetMemoryManager().ReadGuest<V128>(address);
@@ -116,47 +130,56 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Write"
+        [UnmanagedCallersOnly]
         public static void WriteByte(ulong address, byte value)
         {
             GetMemoryManager().WriteGuest(address, value);
         }
 
+        [UnmanagedCallersOnly]
         public static void WriteUInt16(ulong address, ushort value)
         {
             GetMemoryManager().WriteGuest(address, value);
         }
 
+        [UnmanagedCallersOnly]
         public static void WriteUInt32(ulong address, uint value)
         {
             GetMemoryManager().WriteGuest(address, value);
         }
 
+        [UnmanagedCallersOnly]
         public static void WriteUInt64(ulong address, ulong value)
         {
             GetMemoryManager().WriteGuest(address, value);
         }
 
+        [UnmanagedCallersOnly]
         public static void WriteVector128(ulong address, V128 value)
         {
             GetMemoryManager().WriteGuest(address, value);
         }
         #endregion
 
+        [UnmanagedCallersOnly]
         public static void EnqueueForRejit(ulong address)
         {
             Context.Translator.EnqueueForRejit(address, GetContext().ExecutionMode);
         }
 
-        public static void SignalMemoryTracking(ulong address, ulong size, bool write)
+        [UnmanagedCallersOnly]
+        public static void SignalMemoryTracking(ulong address, ulong size, byte write)
         {
-            GetMemoryManager().SignalMemoryTracking(address, size, write);
+            GetMemoryManager().SignalMemoryTracking(address, size, write == 1);
         }
 
+        [UnmanagedCallersOnly]
         public static void ThrowInvalidMemoryAccess(ulong address)
         {
             throw new InvalidAccessException(address);
         }
 
+        [UnmanagedCallersOnly]
         public static ulong GetFunctionAddress(ulong address)
         {
             TranslatedFunction function = Context.Translator.GetOrTranslate(address, GetContext().ExecutionMode);
@@ -164,12 +187,14 @@ namespace ARMeilleure.Instructions
             return (ulong)function.FuncPointer.ToInt64();
         }
 
+        [UnmanagedCallersOnly]
         public static void InvalidateCacheLine(ulong address)
         {
             Context.Translator.InvalidateJitCacheRegion(address, InstEmit.DczSizeInBytes);
         }
 
-        public static bool CheckSynchronization()
+        [UnmanagedCallersOnly]
+        public static byte CheckSynchronization()
         {
             Statistics.PauseTimer();
 
@@ -179,7 +204,7 @@ namespace ARMeilleure.Instructions
 
             Statistics.ResumeTimer();
 
-            return context.Running;
+            return (byte)(context.Running ? 1 : 0);
         }
 
         public static ExecutionContext GetContext()

src/ARMeilleure/Instructions/SoftFallback.cs

@@ -1,11 +1,13 @@
 using ARMeilleure.State;
 using System;
+using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Instructions
 {
     static class SoftFallback
     {
         #region "ShrImm64"
+        [UnmanagedCallersOnly]
         public static long SignedShrImm64(long value, long roundConst, int shift)
         {
             if (roundConst == 0L)
@@ -48,6 +50,7 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        [UnmanagedCallersOnly]
         public static ulong UnsignedShrImm64(ulong value, long roundConst, int shift)
         {
             if (roundConst == 0L)
@@ -92,6 +95,7 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Saturation"
+        [UnmanagedCallersOnly]
         public static int SatF32ToS32(float value)
         {
             if (float.IsNaN(value))
@@ -103,6 +107,7 @@ namespace ARMeilleure.Instructions
                    value <= int.MinValue ? int.MinValue : (int)value;
         }
 
+        [UnmanagedCallersOnly]
         public static long SatF32ToS64(float value)
         {
             if (float.IsNaN(value))
@@ -114,6 +119,7 @@ namespace ARMeilleure.Instructions
                    value <= long.MinValue ? long.MinValue : (long)value;
         }
 
+        [UnmanagedCallersOnly]
         public static uint SatF32ToU32(float value)
         {
             if (float.IsNaN(value))
@@ -125,6 +131,7 @@ namespace ARMeilleure.Instructions
                    value <= uint.MinValue ? uint.MinValue : (uint)value;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong SatF32ToU64(float value)
         {
             if (float.IsNaN(value))
@@ -136,6 +143,7 @@ namespace ARMeilleure.Instructions
                    value <= ulong.MinValue ? ulong.MinValue : (ulong)value;
         }
 
+        [UnmanagedCallersOnly]
         public static int SatF64ToS32(double value)
         {
             if (double.IsNaN(value))
@@ -147,6 +155,7 @@ namespace ARMeilleure.Instructions
                    value <= int.MinValue ? int.MinValue : (int)value;
         }
 
+        [UnmanagedCallersOnly]
         public static long SatF64ToS64(double value)
         {
             if (double.IsNaN(value))
@@ -158,6 +167,7 @@ namespace ARMeilleure.Instructions
                    value <= long.MinValue ? long.MinValue : (long)value;
         }
 
+        [UnmanagedCallersOnly]
         public static uint SatF64ToU32(double value)
         {
             if (double.IsNaN(value))
@@ -169,6 +179,7 @@ namespace ARMeilleure.Instructions
                    value <= uint.MinValue ? uint.MinValue : (uint)value;
         }
 
+        [UnmanagedCallersOnly]
         public static ulong SatF64ToU64(double value)
         {
             if (double.IsNaN(value))
@@ -182,6 +193,7 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Count"
+        [UnmanagedCallersOnly]
         public static ulong CountLeadingSigns(ulong value, int size) // size is 8, 16, 32 or 64 (SIMD&FP or Base Inst.).
         {
             value ^= value >> 1;
@@ -201,6 +213,7 @@ namespace ARMeilleure.Instructions
 
         private static ReadOnlySpan<byte> ClzNibbleTbl => [4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0];
 
+        [UnmanagedCallersOnly]
         public static ulong CountLeadingZeros(ulong value, int size) // size is 8, 16, 32 or 64 (SIMD&FP or Base Inst.).
         {
             if (value == 0ul)
@@ -224,41 +237,49 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Table"
+        [UnmanagedCallersOnly]
         public static V128 Tbl1(V128 vector, int bytes, V128 tb0)
         {
             return TblOrTbx(default, vector, bytes, tb0);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbl2(V128 vector, int bytes, V128 tb0, V128 tb1)
         {
             return TblOrTbx(default, vector, bytes, tb0, tb1);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbl3(V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2)
         {
             return TblOrTbx(default, vector, bytes, tb0, tb1, tb2);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbl4(V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2, V128 tb3)
         {
             return TblOrTbx(default, vector, bytes, tb0, tb1, tb2, tb3);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbx1(V128 dest, V128 vector, int bytes, V128 tb0)
         {
             return TblOrTbx(dest, vector, bytes, tb0);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbx2(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1)
         {
             return TblOrTbx(dest, vector, bytes, tb0, tb1);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbx3(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2)
         {
             return TblOrTbx(dest, vector, bytes, tb0, tb1, tb2);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Tbx4(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2, V128 tb3)
         {
             return TblOrTbx(dest, vector, bytes, tb0, tb1, tb2, tb3);
@@ -300,14 +321,22 @@ namespace ARMeilleure.Instructions
         private const uint Crc32RevPoly = 0xedb88320;
         private const uint Crc32cRevPoly = 0x82f63b78;
 
+        [UnmanagedCallersOnly]
         public static uint Crc32b(uint crc, byte value) => Crc32(crc, Crc32RevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32h(uint crc, ushort value) => Crc32h(crc, Crc32RevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32w(uint crc, uint value) => Crc32w(crc, Crc32RevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32x(uint crc, ulong value) => Crc32x(crc, Crc32RevPoly, value);
 
+        [UnmanagedCallersOnly]
         public static uint Crc32cb(uint crc, byte value) => Crc32(crc, Crc32cRevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32ch(uint crc, ushort value) => Crc32h(crc, Crc32cRevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32cw(uint crc, uint value) => Crc32w(crc, Crc32cRevPoly, value);
+        [UnmanagedCallersOnly]
         public static uint Crc32cx(uint crc, ulong value) => Crc32x(crc, Crc32cRevPoly, value);
 
         private static uint Crc32h(uint crc, uint poly, ushort val)
@@ -358,21 +387,25 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Aes"
+        [UnmanagedCallersOnly]
         public static V128 Decrypt(V128 value, V128 roundKey)
         {
             return CryptoHelper.AesInvSubBytes(CryptoHelper.AesInvShiftRows(value ^ roundKey));
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Encrypt(V128 value, V128 roundKey)
         {
             return CryptoHelper.AesSubBytes(CryptoHelper.AesShiftRows(value ^ roundKey));
         }
 
+        [UnmanagedCallersOnly]
         public static V128 InverseMixColumns(V128 value)
         {
             return CryptoHelper.AesInvMixColumns(value);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 MixColumns(V128 value)
         {
             return CryptoHelper.AesMixColumns(value);
@@ -380,6 +413,7 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Sha1"
+        [UnmanagedCallersOnly]
         public static V128 HashChoose(V128 hash_abcd, uint hash_e, V128 wk)
         {
             for (int e = 0; e <= 3; e++)
@@ -400,11 +434,13 @@ namespace ARMeilleure.Instructions
             return hash_abcd;
         }
 
+        [UnmanagedCallersOnly]
         public static uint FixedRotate(uint hash_e)
         {
             return hash_e.Rol(30);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 HashMajority(V128 hash_abcd, uint hash_e, V128 wk)
         {
             for (int e = 0; e <= 3; e++)
@@ -425,6 +461,7 @@ namespace ARMeilleure.Instructions
             return hash_abcd;
         }
 
+        [UnmanagedCallersOnly]
         public static V128 HashParity(V128 hash_abcd, uint hash_e, V128 wk)
         {
             for (int e = 0; e <= 3; e++)
@@ -445,6 +482,7 @@ namespace ARMeilleure.Instructions
             return hash_abcd;
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Sha1SchedulePart1(V128 w0_3, V128 w4_7, V128 w8_11)
         {
             ulong t2 = w4_7.Extract<ulong>(0);
@@ -455,6 +493,7 @@ namespace ARMeilleure.Instructions
             return result ^ (w0_3 ^ w8_11);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Sha1SchedulePart2(V128 tw0_3, V128 w12_15)
         {
             V128 t = tw0_3 ^ (w12_15 >> 32);
@@ -499,16 +538,19 @@ namespace ARMeilleure.Instructions
         #endregion
 
         #region "Sha256"
+        [UnmanagedCallersOnly]
         public static V128 HashLower(V128 hash_abcd, V128 hash_efgh, V128 wk)
         {
             return Sha256Hash(hash_abcd, hash_efgh, wk, part1: true);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 HashUpper(V128 hash_abcd, V128 hash_efgh, V128 wk)
         {
             return Sha256Hash(hash_abcd, hash_efgh, wk, part1: false);
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Sha256SchedulePart1(V128 w0_3, V128 w4_7)
         {
             V128 result = new();
@@ -527,6 +569,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static V128 Sha256SchedulePart2(V128 w0_3, V128 w8_11, V128 w12_15)
         {
             V128 result = new();
@@ -628,6 +671,7 @@ namespace ARMeilleure.Instructions
         }
         #endregion
 
+        [UnmanagedCallersOnly]
         public static V128 PolynomialMult64_128(ulong op1, ulong op2)
         {
             V128 result = V128.Zero;

src/ARMeilleure/Instructions/SoftFloat.cs

@@ -1,6 +1,7 @@
 using ARMeilleure.State;
 using System;
 using System.Diagnostics;
+using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Instructions
 {
@@ -312,6 +313,7 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat16_32
     {
+        [UnmanagedCallersOnly]
         public static float FPConvert(ushort valueBits)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -487,6 +489,7 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat16_64
     {
+        [UnmanagedCallersOnly]
         public static double FPConvert(ushort valueBits)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -662,6 +665,7 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat32_16
     {
+        [UnmanagedCallersOnly]
         public static ushort FPConvert(float value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -781,12 +785,19 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat32
     {
+        [UnmanagedCallersOnly]
         public static float FPAdd(float value1, float value2)
         {
-            return FPAddFpscr(value1, value2, false);
+            return FPAddFpscrImpl(value1, value2, false);
         }
 
-        public static float FPAddFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPAddFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPAddFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPAddFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -837,7 +848,8 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
-        public static int FPCompare(float value1, float value2, bool signalNaNs)
+        [UnmanagedCallersOnly]
+        public static int FPCompare(float value1, float value2, byte signalNaNs)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = context.Fpcr;
@@ -851,7 +863,7 @@ namespace ARMeilleure.Instructions
             {
                 result = 0b0011;
 
-                if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs)
+                if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs == 1)
                 {
                     SoftFloat.FPProcessException(FPException.InvalidOp, context, fpcr);
                 }
@@ -875,12 +887,13 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPCompareEQ(float value1, float value2)
         {
-            return FPCompareEQFpscr(value1, value2, false);
+            return FPCompareEQFpscrImpl(value1, value2, false);
         }
 
-        public static float FPCompareEQFpscr(float value1, float value2, bool standardFpscr)
+        private static float FPCompareEQFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -907,12 +920,25 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
-        public static float FPCompareGE(float value1, float value2)
+        [UnmanagedCallersOnly]
+        public static float FPCompareEQFpscr(float value1, float value2, byte standardFpscr)
         {
-            return FPCompareGEFpscr(value1, value2, false);
+            return FPCompareEQFpscrImpl(value1, value2, standardFpscr == 1);
         }
 
-        public static float FPCompareGEFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPCompareGE(float value1, float value2)
+        {
+            return FPCompareGEFpscrImpl(value1, value2, false);
+        }
+
+        [UnmanagedCallersOnly]
+        public static float FPCompareGEFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPCompareGEFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPCompareGEFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -936,12 +962,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPCompareGT(float value1, float value2)
         {
-            return FPCompareGTFpscr(value1, value2, false);
+            return FPCompareGTFpscrImpl(value1, value2, false);
         }
 
-        public static float FPCompareGTFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPCompareGTFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPCompareGTFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPCompareGTFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -965,26 +998,31 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPCompareLE(float value1, float value2)
         {
-            return FPCompareGE(value2, value1);
+            return FPCompareGEFpscrImpl(value2, value1, false);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPCompareLT(float value1, float value2)
         {
-            return FPCompareGT(value2, value1);
+            return FPCompareGTFpscrImpl(value2, value1, false);
         }
 
-        public static float FPCompareLEFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPCompareLEFpscr(float value1, float value2, byte standardFpscr)
         {
-            return FPCompareGEFpscr(value2, value1, standardFpscr);
+            return FPCompareGEFpscrImpl(value2, value1, standardFpscr == 1);
         }
 
-        public static float FPCompareLTFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPCompareLTFpscr(float value1, float value2, byte standardFpscr)
         {
-            return FPCompareGTFpscr(value2, value1, standardFpscr);
+            return FPCompareGEFpscrImpl(value2, value1, standardFpscr == 1);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPDiv(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1037,12 +1075,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMax(float value1, float value2)
         {
-            return FPMaxFpscr(value1, value2, false);
+            return FPMaxFpscrImpl(value1, value2, false);
         }
 
-        public static float FPMaxFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMaxFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPMaxFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPMaxFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1103,12 +1148,13 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMaxNum(float value1, float value2)
         {
-            return FPMaxNumFpscr(value1, value2, false);
+            return FPMaxNumFpscrImpl(value1, value2, false);
         }
 
-        public static float FPMaxNumFpscr(float value1, float value2, bool standardFpscr)
+        private static float FPMaxNumFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1125,15 +1171,28 @@ namespace ARMeilleure.Instructions
                 value2 = FPInfinity(true);
             }
 
-            return FPMaxFpscr(value1, value2, standardFpscr);
+            return FPMaxFpscrImpl(value1, value2, standardFpscr);
         }
 
+        [UnmanagedCallersOnly]
+        public static float FPMaxNumFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPMaxNumFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        [UnmanagedCallersOnly]
         public static float FPMin(float value1, float value2)
         {
-            return FPMinFpscr(value1, value2, false);
+            return FPMinFpscrImpl(value1, value2, false);
         }
 
-        public static float FPMinFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMinFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPMinFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPMinFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1194,12 +1253,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMinNum(float value1, float value2)
         {
-            return FPMinNumFpscr(value1, value2, false);
+            return FPMinNumFpscrImpl(value1, value2, false);
         }
 
-        public static float FPMinNumFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMinNumFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPMinNumFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPMinNumFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1216,15 +1282,22 @@ namespace ARMeilleure.Instructions
                 value2 = FPInfinity(false);
             }
 
-            return FPMinFpscr(value1, value2, standardFpscr);
+            return FPMinFpscrImpl(value1, value2, standardFpscr);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMul(float value1, float value2)
         {
-            return FPMulFpscr(value1, value2, false);
+            return FPMulFpscrImpl(value1, value2, false);
         }
 
-        public static float FPMulFpscr(float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMulFpscr(float value1, float value2, byte standardFpscr)
+        {
+            return FPMulFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPMulFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1271,12 +1344,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMulAdd(float valueA, float value1, float value2)
         {
-            return FPMulAddFpscr(valueA, value1, value2, false);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
-        public static float FPMulAddFpscr(float valueA, float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMulAddFpscr(float valueA, float value1, float value2, byte standardFpscr)
+        {
+            return FPMulAddFpscrImpl(valueA, value1, value2, standardFpscr == 1);
+        }
+
+        private static float FPMulAddFpscrImpl(float valueA, float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1342,20 +1422,23 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMulSub(float valueA, float value1, float value2)
         {
             value1 = value1.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
-        public static float FPMulSubFpscr(float valueA, float value1, float value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPMulSubFpscr(float valueA, float value1, float value2, byte standardFpscr)
         {
             value1 = value1.FPNeg();
 
-            return FPMulAddFpscr(valueA, value1, value2, standardFpscr);
+            return FPMulAddFpscrImpl(valueA, value1, value2, standardFpscr == 1);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPMulX(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1401,27 +1484,36 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPNegMulAdd(float valueA, float value1, float value2)
         {
             valueA = valueA.FPNeg();
             value1 = value1.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPNegMulSub(float valueA, float value1, float value2)
         {
             valueA = valueA.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRecipEstimate(float value)
         {
-            return FPRecipEstimateFpscr(value, false);
+            return FPRecipEstimateFpscrImpl(value, false);
         }
 
-        public static float FPRecipEstimateFpscr(float value, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPRecipEstimateFpscr(float value, byte standardFpscr)
+        {
+            return FPRecipEstimateFpscrImpl(value, standardFpscr == 1);
+        }
+
+        private static float FPRecipEstimateFpscrImpl(float value, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1508,6 +1600,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRecipStep(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1533,15 +1626,16 @@ namespace ARMeilleure.Instructions
                 }
                 else
                 {
-                    product = FPMulFpscr(value1, value2, true);
+                    product = FPMulFpscrImpl(value1, value2, true);
                 }
 
-                result = FPSubFpscr(FPTwo(false), product, true);
+                result = FPSubFpscrImpl(FPTwo(false), product, true);
             }
 
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRecipStepFused(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1585,6 +1679,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRecpX(float value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1610,12 +1705,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRSqrtEstimate(float value)
         {
-            return FPRSqrtEstimateFpscr(value, false);
+            return FPRSqrtEstimateFpscrImpl(value, false);
         }
 
-        public static float FPRSqrtEstimateFpscr(float value, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static float FPRSqrtEstimateFpscr(float value, byte standardFpscr)
+        {
+            return FPRSqrtEstimateFpscrImpl(value, standardFpscr == 1);
+        }
+
+        private static float FPRSqrtEstimateFpscrImpl(float value, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -1729,6 +1831,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRSqrtStep(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1754,7 +1857,7 @@ namespace ARMeilleure.Instructions
                 }
                 else
                 {
-                    product = FPMulFpscr(value1, value2, true);
+                    product = FPMulFpscrImpl(value1, value2, true);
                 }
 
                 result = FPHalvedSub(FPThree(false), product, context, fpcr);
@@ -1763,6 +1866,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPRSqrtStepFused(float value1, float value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1806,6 +1910,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPSqrt(float value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -1848,12 +1953,13 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static float FPSub(float value1, float value2)
         {
-            return FPSubFpscr(value1, value2, false);
+            return FPSubFpscrImpl(value1, value2, false);
         }
 
-        public static float FPSubFpscr(float value1, float value2, bool standardFpscr)
+        private static float FPSubFpscrImpl(float value1, float value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2094,6 +2200,7 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat64_16
     {
+        [UnmanagedCallersOnly]
         public static ushort FPConvert(double value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -2213,12 +2320,19 @@ namespace ARMeilleure.Instructions
 
     static class SoftFloat64
     {
+        [UnmanagedCallersOnly]
         public static double FPAdd(double value1, double value2)
         {
-            return FPAddFpscr(value1, value2, false);
+            return FPAddFpscrImpl(value1, value2, false);
         }
 
-        public static double FPAddFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPAddFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPAddFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPAddFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2269,7 +2383,8 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
-        public static int FPCompare(double value1, double value2, bool signalNaNs)
+        [UnmanagedCallersOnly]
+        public static int FPCompare(double value1, double value2, byte signalNaNs)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = context.Fpcr;
@@ -2283,7 +2398,7 @@ namespace ARMeilleure.Instructions
             {
                 result = 0b0011;
 
-                if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs)
+                if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs == 1)
                 {
                     SoftFloat.FPProcessException(FPException.InvalidOp, context, fpcr);
                 }
@@ -2307,12 +2422,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPCompareEQ(double value1, double value2)
         {
-            return FPCompareEQFpscr(value1, value2, false);
+            return FPCompareEQFpscrImpl(value1, value2, false);
         }
 
-        public static double FPCompareEQFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPCompareEQFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPCompareEQFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPCompareEQFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2339,12 +2461,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPCompareGE(double value1, double value2)
         {
-            return FPCompareGEFpscr(value1, value2, false);
+            return FPCompareGEFpscrImpl(value1, value2, false);
         }
 
-        public static double FPCompareGEFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPCompareGEFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPCompareGEFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPCompareGEFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2368,12 +2497,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPCompareGT(double value1, double value2)
         {
-            return FPCompareGTFpscr(value1, value2, false);
+            return FPCompareGTFpscrImpl(value1, value2, false);
         }
 
-        public static double FPCompareGTFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPCompareGTFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPCompareGTFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPCompareGTFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2397,26 +2533,31 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPCompareLE(double value1, double value2)
         {
-            return FPCompareGE(value2, value1);
+            return FPCompareGEFpscrImpl(value2, value1, false);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPCompareLT(double value1, double value2)
         {
-            return FPCompareGT(value2, value1);
+            return FPCompareGTFpscrImpl(value2, value1, false);
         }
 
-        public static double FPCompareLEFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPCompareLEFpscr(double value1, double value2, byte standardFpscr)
         {
-            return FPCompareGEFpscr(value2, value1, standardFpscr);
+            return FPCompareGEFpscrImpl(value2, value1, standardFpscr == 1);
         }
 
-        public static double FPCompareLTFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPCompareLTFpscr(double value1, double value2, byte standardFpscr)
         {
-            return FPCompareGTFpscr(value2, value1, standardFpscr);
+            return FPCompareGTFpscrImpl(value2, value1, standardFpscr == 1);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPDiv(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -2469,12 +2610,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMax(double value1, double value2)
         {
-            return FPMaxFpscr(value1, value2, false);
+            return FPMaxFpscrImpl(value1, value2, false);
         }
 
-        public static double FPMaxFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMaxFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPMaxFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMaxFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2535,12 +2683,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMaxNum(double value1, double value2)
         {
-            return FPMaxNumFpscr(value1, value2, false);
+            return FPMaxNumFpscrImpl(value1, value2, false);
         }
 
-        public static double FPMaxNumFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMaxNumFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPMaxNumFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMaxNumFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2557,15 +2712,22 @@ namespace ARMeilleure.Instructions
                 value2 = FPInfinity(true);
             }
 
-            return FPMaxFpscr(value1, value2, standardFpscr);
+            return FPMaxFpscrImpl(value1, value2, standardFpscr);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMin(double value1, double value2)
         {
-            return FPMinFpscr(value1, value2, false);
+            return FPMinFpscrImpl(value1, value2, false);
         }
 
-        public static double FPMinFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMinFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPMinFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMinFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2626,12 +2788,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMinNum(double value1, double value2)
         {
-            return FPMinNumFpscr(value1, value2, false);
+            return FPMinNumFpscrImpl(value1, value2, false);
         }
 
-        public static double FPMinNumFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMinNumFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPMinNumFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMinNumFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2648,15 +2817,22 @@ namespace ARMeilleure.Instructions
                 value2 = FPInfinity(false);
             }
 
-            return FPMinFpscr(value1, value2, standardFpscr);
+            return FPMinFpscrImpl(value1, value2, standardFpscr);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMul(double value1, double value2)
         {
-            return FPMulFpscr(value1, value2, false);
+            return FPMulFpscrImpl(value1, value2, false);
         }
 
-        public static double FPMulFpscr(double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMulFpscr(double value1, double value2, byte standardFpscr)
+        {
+            return FPMulFpscrImpl(value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMulFpscrImpl(double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2703,12 +2879,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMulAdd(double valueA, double value1, double value2)
         {
-            return FPMulAddFpscr(valueA, value1, value2, false);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
-        public static double FPMulAddFpscr(double valueA, double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMulAddFpscr(double valueA, double value1, double value2, byte standardFpscr)
+        {
+            return FPMulAddFpscrImpl(valueA, value1, value2, standardFpscr == 1);
+        }
+
+        private static double FPMulAddFpscrImpl(double valueA, double value1, double value2, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2774,20 +2957,23 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMulSub(double valueA, double value1, double value2)
         {
             value1 = value1.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
-        public static double FPMulSubFpscr(double valueA, double value1, double value2, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPMulSubFpscr(double valueA, double value1, double value2, byte standardFpscr)
         {
             value1 = value1.FPNeg();
 
-            return FPMulAddFpscr(valueA, value1, value2, standardFpscr);
+            return FPMulAddFpscrImpl(valueA, value1, value2, standardFpscr == 1);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPMulX(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -2833,27 +3019,36 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPNegMulAdd(double valueA, double value1, double value2)
         {
             valueA = valueA.FPNeg();
             value1 = value1.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPNegMulSub(double valueA, double value1, double value2)
         {
             valueA = valueA.FPNeg();
 
-            return FPMulAdd(valueA, value1, value2);
+            return FPMulAddFpscrImpl(valueA, value1, value2, false);
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRecipEstimate(double value)
         {
-            return FPRecipEstimateFpscr(value, false);
+            return FPRecipEstimateFpscrImpl(value, false);
         }
 
-        public static double FPRecipEstimateFpscr(double value, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPRecipEstimateFpscr(double value, byte standardFpscr)
+        {
+            return FPRecipEstimateFpscrImpl(value, standardFpscr == 1);
+        }
+
+        private static double FPRecipEstimateFpscrImpl(double value, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -2940,6 +3135,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRecipStep(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -2965,7 +3161,7 @@ namespace ARMeilleure.Instructions
                 }
                 else
                 {
-                    product = FPMulFpscr(value1, value2, true);
+                    product = FPMulFpscrImpl(value1, value2, true);
                 }
 
                 result = FPSubFpscr(FPTwo(false), product, true);
@@ -2974,6 +3170,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRecipStepFused(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -3017,6 +3214,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRecpX(double value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -3042,12 +3240,19 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRSqrtEstimate(double value)
         {
-            return FPRSqrtEstimateFpscr(value, false);
+            return FPRSqrtEstimateFpscrImpl(value, false);
         }
 
-        public static double FPRSqrtEstimateFpscr(double value, bool standardFpscr)
+        [UnmanagedCallersOnly]
+        public static double FPRSqrtEstimateFpscr(double value, byte standardFpscr)
+        {
+            return FPRSqrtEstimateFpscrImpl(value, standardFpscr == 1);
+        }
+
+        private static double FPRSqrtEstimateFpscrImpl(double value, bool standardFpscr)
         {
             ExecutionContext context = NativeInterface.GetContext();
             FPCR fpcr = standardFpscr ? context.StandardFpcrValue : context.Fpcr;
@@ -3161,6 +3366,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRSqrtStep(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -3186,7 +3392,7 @@ namespace ARMeilleure.Instructions
                 }
                 else
                 {
-                    product = FPMulFpscr(value1, value2, true);
+                    product = FPMulFpscrImpl(value1, value2, true);
                 }
 
                 result = FPHalvedSub(FPThree(false), product, context, fpcr);
@@ -3195,6 +3401,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPRSqrtStepFused(double value1, double value2)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -3238,6 +3445,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPSqrt(double value)
         {
             ExecutionContext context = NativeInterface.GetContext();
@@ -3280,6 +3488,7 @@ namespace ARMeilleure.Instructions
             return result;
         }
 
+        [UnmanagedCallersOnly]
         public static double FPSub(double value1, double value2)
         {
             return FPSubFpscr(value1, value2, false);

src/ARMeilleure/Translation/DelegateInfo.cs

@@ -1,18 +1,11 @@
-using System;
-
 namespace ARMeilleure.Translation
 {
     class DelegateInfo
     {
-#pragma warning disable IDE0052 // Remove unread private member
-        private readonly Delegate _dlg; // Ensure that this delegate will not be garbage collected.
-#pragma warning restore IDE0052
-
         public nint FuncPtr { get; }
 
-        public DelegateInfo(Delegate dlg, nint funcPtr)
+        public DelegateInfo(nint funcPtr)
         {
-            _dlg = dlg;
             FuncPtr = funcPtr;
         }
     }

src/ARMeilleure/Translation/Delegates.cs

@@ -1,9 +1,7 @@
 using ARMeilleure.Instructions;
-using ARMeilleure.State;
 using System;
 using System.Collections.Generic;
 using System.Reflection;
-using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Translation
 {
@@ -35,20 +33,6 @@ namespace ARMeilleure.Translation
             return _delegates.Values[index].FuncPtr; // O(1).
         }
 
-        public static IntPtr GetDelegateFuncPtr(MethodInfo info)
-        {
-            ArgumentNullException.ThrowIfNull(info);
-
-            string key = GetKey(info);
-
-            if (!_delegates.TryGetValue(key, out DelegateInfo dlgInfo)) // O(log(n)).
-            {
-                throw new KeyNotFoundException($"({nameof(key)} = {key})");
-            }
-
-            return dlgInfo.FuncPtr;
-        }
-
         public static int GetDelegateIndex(MethodInfo info)
         {
             ArgumentNullException.ThrowIfNull(info);
@@ -65,11 +49,11 @@ namespace ARMeilleure.Translation
             return index;
         }
 
-        private static void SetDelegateInfo(Delegate dlg, IntPtr funcPtr)
+        private static void SetDelegateInfo(MethodInfo method)
         {
-            string key = GetKey(dlg.Method);
+            string key = GetKey(method);
 
-            _delegates.Add(key, new DelegateInfo(dlg, funcPtr)); // ArgumentException (key).
+            _delegates.Add(key, new DelegateInfo(method.MethodHandle.GetFunctionPointer())); // ArgumentException (key).
         }
 
         private static string GetKey(MethodInfo info)
@@ -79,532 +63,183 @@ namespace ARMeilleure.Translation
 
         private static readonly SortedList<string, DelegateInfo> _delegates;
 
-        static Delegates()
+        unsafe static Delegates()
         {
             _delegates = new SortedList<string, DelegateInfo>();
 
-            var dlgMathAbs = new MathAbs(Math.Abs);
-            var dlgMathCeiling = new MathCeiling(Math.Ceiling);
-            var dlgMathFloor = new MathFloor(Math.Floor);
-            var dlgMathRound = new MathRound(Math.Round);
-            var dlgMathTruncate = new MathTruncate(Math.Truncate);
+            SetDelegateInfo(typeof(MathHelper).GetMethod(nameof(MathHelper.Abs)));
+            SetDelegateInfo(typeof(MathHelper).GetMethod(nameof(MathHelper.Ceiling)));
+            SetDelegateInfo(typeof(MathHelper).GetMethod(nameof(MathHelper.Floor)));
+            SetDelegateInfo(typeof(MathHelper).GetMethod(nameof(MathHelper.Round)));
+            SetDelegateInfo(typeof(MathHelper).GetMethod(nameof(MathHelper.Truncate)));
 
-            var dlgMathFAbs = new MathFAbs(MathF.Abs);
-            var dlgMathFCeiling = new MathFCeiling(MathF.Ceiling);
-            var dlgMathFFloor = new MathFFloor(MathF.Floor);
-            var dlgMathFRound = new MathFRound(MathF.Round);
-            var dlgMathFTruncate = new MathFTruncate(MathF.Truncate);
+            SetDelegateInfo(typeof(MathHelperF).GetMethod(nameof(MathHelperF.Abs)));
+            SetDelegateInfo(typeof(MathHelperF).GetMethod(nameof(MathHelperF.Ceiling)));
+            SetDelegateInfo(typeof(MathHelperF).GetMethod(nameof(MathHelperF.Floor)));
+            SetDelegateInfo(typeof(MathHelperF).GetMethod(nameof(MathHelperF.Round)));
+            SetDelegateInfo(typeof(MathHelperF).GetMethod(nameof(MathHelperF.Truncate)));
 
-            var dlgNativeInterfaceBreak = new NativeInterfaceBreak(NativeInterface.Break);
-            var dlgNativeInterfaceCheckSynchronization = new NativeInterfaceCheckSynchronization(NativeInterface.CheckSynchronization);
-            var dlgNativeInterfaceEnqueueForRejit = new NativeInterfaceEnqueueForRejit(NativeInterface.EnqueueForRejit);
-            var dlgNativeInterfaceGetCntfrqEl0 = new NativeInterfaceGetCntfrqEl0(NativeInterface.GetCntfrqEl0);
-            var dlgNativeInterfaceGetCntpctEl0 = new NativeInterfaceGetCntpctEl0(NativeInterface.GetCntpctEl0);
-            var dlgNativeInterfaceGetCntvctEl0 = new NativeInterfaceGetCntvctEl0(NativeInterface.GetCntvctEl0);
-            var dlgNativeInterfaceGetCtrEl0 = new NativeInterfaceGetCtrEl0(NativeInterface.GetCtrEl0);
-            var dlgNativeInterfaceGetDczidEl0 = new NativeInterfaceGetDczidEl0(NativeInterface.GetDczidEl0);
-            var dlgNativeInterfaceGetFunctionAddress = new NativeInterfaceGetFunctionAddress(NativeInterface.GetFunctionAddress);
-            var dlgNativeInterfaceInvalidateCacheLine = new NativeInterfaceInvalidateCacheLine(NativeInterface.InvalidateCacheLine);
-            var dlgNativeInterfaceReadByte = new NativeInterfaceReadByte(NativeInterface.ReadByte);
-            var dlgNativeInterfaceReadUInt16 = new NativeInterfaceReadUInt16(NativeInterface.ReadUInt16);
-            var dlgNativeInterfaceReadUInt32 = new NativeInterfaceReadUInt32(NativeInterface.ReadUInt32);
-            var dlgNativeInterfaceReadUInt64 = new NativeInterfaceReadUInt64(NativeInterface.ReadUInt64);
-            var dlgNativeInterfaceReadVector128 = new NativeInterfaceReadVector128(NativeInterface.ReadVector128);
-            var dlgNativeInterfaceSignalMemoryTracking = new NativeInterfaceSignalMemoryTracking(NativeInterface.SignalMemoryTracking);
-            var dlgNativeInterfaceSupervisorCall = new NativeInterfaceSupervisorCall(NativeInterface.SupervisorCall);
-            var dlgNativeInterfaceThrowInvalidMemoryAccess = new NativeInterfaceThrowInvalidMemoryAccess(NativeInterface.ThrowInvalidMemoryAccess);
-            var dlgNativeInterfaceUndefined = new NativeInterfaceUndefined(NativeInterface.Undefined);
-            var dlgNativeInterfaceWriteByte = new NativeInterfaceWriteByte(NativeInterface.WriteByte);
-            var dlgNativeInterfaceWriteUInt16 = new NativeInterfaceWriteUInt16(NativeInterface.WriteUInt16);
-            var dlgNativeInterfaceWriteUInt32 = new NativeInterfaceWriteUInt32(NativeInterface.WriteUInt32);
-            var dlgNativeInterfaceWriteUInt64 = new NativeInterfaceWriteUInt64(NativeInterface.WriteUInt64);
-            var dlgNativeInterfaceWriteVector128 = new NativeInterfaceWriteVector128(NativeInterface.WriteVector128);
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.Break)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.CheckSynchronization)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.EnqueueForRejit)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetCntfrqEl0)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetCntpctEl0)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetCntvctEl0)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetCtrEl0)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetDczidEl0)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetFunctionAddress)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.InvalidateCacheLine)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ReadByte)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ReadUInt16)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ReadUInt32)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ReadUInt64)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ReadVector128)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.SignalMemoryTracking)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.SupervisorCall)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.ThrowInvalidMemoryAccess)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.Undefined)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.WriteByte)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.WriteUInt16)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.WriteUInt32)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.WriteUInt64)));
+            SetDelegateInfo(typeof(NativeInterface).GetMethod(nameof(NativeInterface.WriteVector128)));
 
-            var dlgSoftFallbackCountLeadingSigns = new SoftFallbackCountLeadingSigns(SoftFallback.CountLeadingSigns);
-            var dlgSoftFallbackCountLeadingZeros = new SoftFallbackCountLeadingZeros(SoftFallback.CountLeadingZeros);
-            var dlgSoftFallbackCrc32b = new SoftFallbackCrc32b(SoftFallback.Crc32b);
-            var dlgSoftFallbackCrc32cb = new SoftFallbackCrc32cb(SoftFallback.Crc32cb);
-            var dlgSoftFallbackCrc32ch = new SoftFallbackCrc32ch(SoftFallback.Crc32ch);
-            var dlgSoftFallbackCrc32cw = new SoftFallbackCrc32cw(SoftFallback.Crc32cw);
-            var dlgSoftFallbackCrc32cx = new SoftFallbackCrc32cx(SoftFallback.Crc32cx);
-            var dlgSoftFallbackCrc32h = new SoftFallbackCrc32h(SoftFallback.Crc32h);
-            var dlgSoftFallbackCrc32w = new SoftFallbackCrc32w(SoftFallback.Crc32w);
-            var dlgSoftFallbackCrc32x = new SoftFallbackCrc32x(SoftFallback.Crc32x);
-            var dlgSoftFallbackDecrypt = new SoftFallbackDecrypt(SoftFallback.Decrypt);
-            var dlgSoftFallbackEncrypt = new SoftFallbackEncrypt(SoftFallback.Encrypt);
-            var dlgSoftFallbackFixedRotate = new SoftFallbackFixedRotate(SoftFallback.FixedRotate);
-            var dlgSoftFallbackHashChoose = new SoftFallbackHashChoose(SoftFallback.HashChoose);
-            var dlgSoftFallbackHashLower = new SoftFallbackHashLower(SoftFallback.HashLower);
-            var dlgSoftFallbackHashMajority = new SoftFallbackHashMajority(SoftFallback.HashMajority);
-            var dlgSoftFallbackHashParity = new SoftFallbackHashParity(SoftFallback.HashParity);
-            var dlgSoftFallbackHashUpper = new SoftFallbackHashUpper(SoftFallback.HashUpper);
-            var dlgSoftFallbackInverseMixColumns = new SoftFallbackInverseMixColumns(SoftFallback.InverseMixColumns);
-            var dlgSoftFallbackMixColumns = new SoftFallbackMixColumns(SoftFallback.MixColumns);
-            var dlgSoftFallbackPolynomialMult64_128 = new SoftFallbackPolynomialMult64_128(SoftFallback.PolynomialMult64_128);
-            var dlgSoftFallbackSatF32ToS32 = new SoftFallbackSatF32ToS32(SoftFallback.SatF32ToS32);
-            var dlgSoftFallbackSatF32ToS64 = new SoftFallbackSatF32ToS64(SoftFallback.SatF32ToS64);
-            var dlgSoftFallbackSatF32ToU32 = new SoftFallbackSatF32ToU32(SoftFallback.SatF32ToU32);
-            var dlgSoftFallbackSatF32ToU64 = new SoftFallbackSatF32ToU64(SoftFallback.SatF32ToU64);
-            var dlgSoftFallbackSatF64ToS32 = new SoftFallbackSatF64ToS32(SoftFallback.SatF64ToS32);
-            var dlgSoftFallbackSatF64ToS64 = new SoftFallbackSatF64ToS64(SoftFallback.SatF64ToS64);
-            var dlgSoftFallbackSatF64ToU32 = new SoftFallbackSatF64ToU32(SoftFallback.SatF64ToU32);
-            var dlgSoftFallbackSatF64ToU64 = new SoftFallbackSatF64ToU64(SoftFallback.SatF64ToU64);
-            var dlgSoftFallbackSha1SchedulePart1 = new SoftFallbackSha1SchedulePart1(SoftFallback.Sha1SchedulePart1);
-            var dlgSoftFallbackSha1SchedulePart2 = new SoftFallbackSha1SchedulePart2(SoftFallback.Sha1SchedulePart2);
-            var dlgSoftFallbackSha256SchedulePart1 = new SoftFallbackSha256SchedulePart1(SoftFallback.Sha256SchedulePart1);
-            var dlgSoftFallbackSha256SchedulePart2 = new SoftFallbackSha256SchedulePart2(SoftFallback.Sha256SchedulePart2);
-            var dlgSoftFallbackSignedShrImm64 = new SoftFallbackSignedShrImm64(SoftFallback.SignedShrImm64);
-            var dlgSoftFallbackTbl1 = new SoftFallbackTbl1(SoftFallback.Tbl1);
-            var dlgSoftFallbackTbl2 = new SoftFallbackTbl2(SoftFallback.Tbl2);
-            var dlgSoftFallbackTbl3 = new SoftFallbackTbl3(SoftFallback.Tbl3);
-            var dlgSoftFallbackTbl4 = new SoftFallbackTbl4(SoftFallback.Tbl4);
-            var dlgSoftFallbackTbx1 = new SoftFallbackTbx1(SoftFallback.Tbx1);
-            var dlgSoftFallbackTbx2 = new SoftFallbackTbx2(SoftFallback.Tbx2);
-            var dlgSoftFallbackTbx3 = new SoftFallbackTbx3(SoftFallback.Tbx3);
-            var dlgSoftFallbackTbx4 = new SoftFallbackTbx4(SoftFallback.Tbx4);
-            var dlgSoftFallbackUnsignedShrImm64 = new SoftFallbackUnsignedShrImm64(SoftFallback.UnsignedShrImm64);
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.CountLeadingSigns)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.CountLeadingZeros)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32b)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32cb)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32ch)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32cw)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32cx)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32h)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32w)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Crc32x)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Decrypt)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Encrypt)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.FixedRotate)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashChoose)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashLower)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashMajority)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashParity)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashUpper)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.InverseMixColumns)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.MixColumns)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.PolynomialMult64_128)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF32ToS32)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF32ToS64)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF32ToU32)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF32ToU64)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF64ToS32)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF64ToS64)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF64ToU32)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SatF64ToU64)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha1SchedulePart1)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha1SchedulePart2)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart1)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart2)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.SignedShrImm64)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl1)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl2)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl3)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbl4)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx1)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx2)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx3)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Tbx4)));
+            SetDelegateInfo(typeof(SoftFallback).GetMethod(nameof(SoftFallback.UnsignedShrImm64)));
 
-            var dlgSoftFloat16_32FPConvert = new SoftFloat16_32FPConvert(SoftFloat16_32.FPConvert);
-            var dlgSoftFloat16_64FPConvert = new SoftFloat16_64FPConvert(SoftFloat16_64.FPConvert);
+            SetDelegateInfo(typeof(SoftFloat16_32).GetMethod(nameof(SoftFloat16_32.FPConvert)));
+            SetDelegateInfo(typeof(SoftFloat16_64).GetMethod(nameof(SoftFloat16_64.FPConvert)));
 
-            var dlgSoftFloat32FPAdd = new SoftFloat32FPAdd(SoftFloat32.FPAdd);
-            var dlgSoftFloat32FPAddFpscr = new SoftFloat32FPAddFpscr(SoftFloat32.FPAddFpscr); // A32 only.
-            var dlgSoftFloat32FPCompare = new SoftFloat32FPCompare(SoftFloat32.FPCompare);
-            var dlgSoftFloat32FPCompareEQ = new SoftFloat32FPCompareEQ(SoftFloat32.FPCompareEQ);
-            var dlgSoftFloat32FPCompareEQFpscr = new SoftFloat32FPCompareEQFpscr(SoftFloat32.FPCompareEQFpscr); // A32 only.
-            var dlgSoftFloat32FPCompareGE = new SoftFloat32FPCompareGE(SoftFloat32.FPCompareGE);
-            var dlgSoftFloat32FPCompareGEFpscr = new SoftFloat32FPCompareGEFpscr(SoftFloat32.FPCompareGEFpscr); // A32 only.
-            var dlgSoftFloat32FPCompareGT = new SoftFloat32FPCompareGT(SoftFloat32.FPCompareGT);
-            var dlgSoftFloat32FPCompareGTFpscr = new SoftFloat32FPCompareGTFpscr(SoftFloat32.FPCompareGTFpscr); // A32 only.
-            var dlgSoftFloat32FPCompareLE = new SoftFloat32FPCompareLE(SoftFloat32.FPCompareLE);
-            var dlgSoftFloat32FPCompareLEFpscr = new SoftFloat32FPCompareLEFpscr(SoftFloat32.FPCompareLEFpscr); // A32 only.
-            var dlgSoftFloat32FPCompareLT = new SoftFloat32FPCompareLT(SoftFloat32.FPCompareLT);
-            var dlgSoftFloat32FPCompareLTFpscr = new SoftFloat32FPCompareLTFpscr(SoftFloat32.FPCompareLTFpscr); // A32 only.
-            var dlgSoftFloat32FPDiv = new SoftFloat32FPDiv(SoftFloat32.FPDiv);
-            var dlgSoftFloat32FPMax = new SoftFloat32FPMax(SoftFloat32.FPMax);
-            var dlgSoftFloat32FPMaxFpscr = new SoftFloat32FPMaxFpscr(SoftFloat32.FPMaxFpscr); // A32 only.
-            var dlgSoftFloat32FPMaxNum = new SoftFloat32FPMaxNum(SoftFloat32.FPMaxNum);
-            var dlgSoftFloat32FPMaxNumFpscr = new SoftFloat32FPMaxNumFpscr(SoftFloat32.FPMaxNumFpscr); // A32 only.
-            var dlgSoftFloat32FPMin = new SoftFloat32FPMin(SoftFloat32.FPMin);
-            var dlgSoftFloat32FPMinFpscr = new SoftFloat32FPMinFpscr(SoftFloat32.FPMinFpscr); // A32 only.
-            var dlgSoftFloat32FPMinNum = new SoftFloat32FPMinNum(SoftFloat32.FPMinNum);
-            var dlgSoftFloat32FPMinNumFpscr = new SoftFloat32FPMinNumFpscr(SoftFloat32.FPMinNumFpscr); // A32 only.
-            var dlgSoftFloat32FPMul = new SoftFloat32FPMul(SoftFloat32.FPMul);
-            var dlgSoftFloat32FPMulFpscr = new SoftFloat32FPMulFpscr(SoftFloat32.FPMulFpscr); // A32 only.
-            var dlgSoftFloat32FPMulAdd = new SoftFloat32FPMulAdd(SoftFloat32.FPMulAdd);
-            var dlgSoftFloat32FPMulAddFpscr = new SoftFloat32FPMulAddFpscr(SoftFloat32.FPMulAddFpscr); // A32 only.
-            var dlgSoftFloat32FPMulSub = new SoftFloat32FPMulSub(SoftFloat32.FPMulSub);
-            var dlgSoftFloat32FPMulSubFpscr = new SoftFloat32FPMulSubFpscr(SoftFloat32.FPMulSubFpscr); // A32 only.
-            var dlgSoftFloat32FPMulX = new SoftFloat32FPMulX(SoftFloat32.FPMulX);
-            var dlgSoftFloat32FPNegMulAdd = new SoftFloat32FPNegMulAdd(SoftFloat32.FPNegMulAdd);
-            var dlgSoftFloat32FPNegMulSub = new SoftFloat32FPNegMulSub(SoftFloat32.FPNegMulSub);
-            var dlgSoftFloat32FPRecipEstimate = new SoftFloat32FPRecipEstimate(SoftFloat32.FPRecipEstimate);
-            var dlgSoftFloat32FPRecipEstimateFpscr = new SoftFloat32FPRecipEstimateFpscr(SoftFloat32.FPRecipEstimateFpscr); // A32 only.
-            var dlgSoftFloat32FPRecipStep = new SoftFloat32FPRecipStep(SoftFloat32.FPRecipStep); // A32 only.
-            var dlgSoftFloat32FPRecipStepFused = new SoftFloat32FPRecipStepFused(SoftFloat32.FPRecipStepFused);
-            var dlgSoftFloat32FPRecpX = new SoftFloat32FPRecpX(SoftFloat32.FPRecpX);
-            var dlgSoftFloat32FPRSqrtEstimate = new SoftFloat32FPRSqrtEstimate(SoftFloat32.FPRSqrtEstimate);
-            var dlgSoftFloat32FPRSqrtEstimateFpscr = new SoftFloat32FPRSqrtEstimateFpscr(SoftFloat32.FPRSqrtEstimateFpscr); // A32 only.
-            var dlgSoftFloat32FPRSqrtStep = new SoftFloat32FPRSqrtStep(SoftFloat32.FPRSqrtStep); // A32 only.
-            var dlgSoftFloat32FPRSqrtStepFused = new SoftFloat32FPRSqrtStepFused(SoftFloat32.FPRSqrtStepFused);
-            var dlgSoftFloat32FPSqrt = new SoftFloat32FPSqrt(SoftFloat32.FPSqrt);
-            var dlgSoftFloat32FPSub = new SoftFloat32FPSub(SoftFloat32.FPSub);
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPAdd)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPAddFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompare)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareEQ)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareEQFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareGE)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareGEFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareGT)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareGTFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareLE)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareLEFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareLT)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompareLTFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPDiv)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMax)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMaxFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMaxNum)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMaxNumFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMin)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMinFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMinNum)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMinNumFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMul)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulAdd)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulAddFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulSub)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulSubFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPMulX)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPNegMulAdd)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPNegMulSub)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRecipEstimate)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRecipEstimateFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRecipStep))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRecipStepFused)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRecpX)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRSqrtEstimate)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRSqrtEstimateFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRSqrtStep))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPRSqrtStepFused)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPSqrt)));
+            SetDelegateInfo(typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPSub)));
 
-            var dlgSoftFloat32_16FPConvert = new SoftFloat32_16FPConvert(SoftFloat32_16.FPConvert);
+            SetDelegateInfo(typeof(SoftFloat32_16).GetMethod(nameof(SoftFloat32_16.FPConvert)));
 
-            var dlgSoftFloat64FPAdd = new SoftFloat64FPAdd(SoftFloat64.FPAdd);
-            var dlgSoftFloat64FPAddFpscr = new SoftFloat64FPAddFpscr(SoftFloat64.FPAddFpscr); // A32 only.
-            var dlgSoftFloat64FPCompare = new SoftFloat64FPCompare(SoftFloat64.FPCompare);
-            var dlgSoftFloat64FPCompareEQ = new SoftFloat64FPCompareEQ(SoftFloat64.FPCompareEQ);
-            var dlgSoftFloat64FPCompareEQFpscr = new SoftFloat64FPCompareEQFpscr(SoftFloat64.FPCompareEQFpscr); // A32 only.
-            var dlgSoftFloat64FPCompareGE = new SoftFloat64FPCompareGE(SoftFloat64.FPCompareGE);
-            var dlgSoftFloat64FPCompareGEFpscr = new SoftFloat64FPCompareGEFpscr(SoftFloat64.FPCompareGEFpscr); // A32 only.
-            var dlgSoftFloat64FPCompareGT = new SoftFloat64FPCompareGT(SoftFloat64.FPCompareGT);
-            var dlgSoftFloat64FPCompareGTFpscr = new SoftFloat64FPCompareGTFpscr(SoftFloat64.FPCompareGTFpscr); // A32 only.
-            var dlgSoftFloat64FPCompareLE = new SoftFloat64FPCompareLE(SoftFloat64.FPCompareLE);
-            var dlgSoftFloat64FPCompareLEFpscr = new SoftFloat64FPCompareLEFpscr(SoftFloat64.FPCompareLEFpscr); // A32 only.
-            var dlgSoftFloat64FPCompareLT = new SoftFloat64FPCompareLT(SoftFloat64.FPCompareLT);
-            var dlgSoftFloat64FPCompareLTFpscr = new SoftFloat64FPCompareLTFpscr(SoftFloat64.FPCompareLTFpscr); // A32 only.
-            var dlgSoftFloat64FPDiv = new SoftFloat64FPDiv(SoftFloat64.FPDiv);
-            var dlgSoftFloat64FPMax = new SoftFloat64FPMax(SoftFloat64.FPMax);
-            var dlgSoftFloat64FPMaxFpscr = new SoftFloat64FPMaxFpscr(SoftFloat64.FPMaxFpscr); // A32 only.
-            var dlgSoftFloat64FPMaxNum = new SoftFloat64FPMaxNum(SoftFloat64.FPMaxNum);
-            var dlgSoftFloat64FPMaxNumFpscr = new SoftFloat64FPMaxNumFpscr(SoftFloat64.FPMaxNumFpscr); // A32 only.
-            var dlgSoftFloat64FPMin = new SoftFloat64FPMin(SoftFloat64.FPMin);
-            var dlgSoftFloat64FPMinFpscr = new SoftFloat64FPMinFpscr(SoftFloat64.FPMinFpscr); // A32 only.
-            var dlgSoftFloat64FPMinNum = new SoftFloat64FPMinNum(SoftFloat64.FPMinNum);
-            var dlgSoftFloat64FPMinNumFpscr = new SoftFloat64FPMinNumFpscr(SoftFloat64.FPMinNumFpscr); // A32 only.
-            var dlgSoftFloat64FPMul = new SoftFloat64FPMul(SoftFloat64.FPMul);
-            var dlgSoftFloat64FPMulFpscr = new SoftFloat64FPMulFpscr(SoftFloat64.FPMulFpscr); // A32 only.
-            var dlgSoftFloat64FPMulAdd = new SoftFloat64FPMulAdd(SoftFloat64.FPMulAdd);
-            var dlgSoftFloat64FPMulAddFpscr = new SoftFloat64FPMulAddFpscr(SoftFloat64.FPMulAddFpscr); // A32 only.
-            var dlgSoftFloat64FPMulSub = new SoftFloat64FPMulSub(SoftFloat64.FPMulSub);
-            var dlgSoftFloat64FPMulSubFpscr = new SoftFloat64FPMulSubFpscr(SoftFloat64.FPMulSubFpscr); // A32 only.
-            var dlgSoftFloat64FPMulX = new SoftFloat64FPMulX(SoftFloat64.FPMulX);
-            var dlgSoftFloat64FPNegMulAdd = new SoftFloat64FPNegMulAdd(SoftFloat64.FPNegMulAdd);
-            var dlgSoftFloat64FPNegMulSub = new SoftFloat64FPNegMulSub(SoftFloat64.FPNegMulSub);
-            var dlgSoftFloat64FPRecipEstimate = new SoftFloat64FPRecipEstimate(SoftFloat64.FPRecipEstimate);
-            var dlgSoftFloat64FPRecipEstimateFpscr = new SoftFloat64FPRecipEstimateFpscr(SoftFloat64.FPRecipEstimateFpscr); // A32 only.
-            var dlgSoftFloat64FPRecipStep = new SoftFloat64FPRecipStep(SoftFloat64.FPRecipStep); // A32 only.
-            var dlgSoftFloat64FPRecipStepFused = new SoftFloat64FPRecipStepFused(SoftFloat64.FPRecipStepFused);
-            var dlgSoftFloat64FPRecpX = new SoftFloat64FPRecpX(SoftFloat64.FPRecpX);
-            var dlgSoftFloat64FPRSqrtEstimate = new SoftFloat64FPRSqrtEstimate(SoftFloat64.FPRSqrtEstimate);
-            var dlgSoftFloat64FPRSqrtEstimateFpscr = new SoftFloat64FPRSqrtEstimateFpscr(SoftFloat64.FPRSqrtEstimateFpscr); // A32 only.
-            var dlgSoftFloat64FPRSqrtStep = new SoftFloat64FPRSqrtStep(SoftFloat64.FPRSqrtStep); // A32 only.
-            var dlgSoftFloat64FPRSqrtStepFused = new SoftFloat64FPRSqrtStepFused(SoftFloat64.FPRSqrtStepFused);
-            var dlgSoftFloat64FPSqrt = new SoftFloat64FPSqrt(SoftFloat64.FPSqrt);
-            var dlgSoftFloat64FPSub = new SoftFloat64FPSub(SoftFloat64.FPSub);
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPAdd)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPAddFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompare)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareEQ)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareEQFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareGE)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareGEFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareGT)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareGTFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareLE)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareLEFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareLT)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompareLTFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPDiv)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMax)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMaxFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMaxNum)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMaxNumFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMin)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMinFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMinNum)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMinNumFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMul)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulAdd)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulAddFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulSub)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulSubFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPMulX)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPNegMulAdd)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPNegMulSub)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRecipEstimate)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRecipEstimateFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRecipStep))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRecipStepFused)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRecpX)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRSqrtEstimate)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRSqrtEstimateFpscr))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRSqrtStep))); // A32 only.
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPRSqrtStepFused)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPSqrt)));
+            SetDelegateInfo(typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPSub)));
 
-            var dlgSoftFloat64_16FPConvert = new SoftFloat64_16FPConvert(SoftFloat64_16.FPConvert);
-
-            SetDelegateInfo(dlgMathAbs, Marshal.GetFunctionPointerForDelegate<MathAbs>(dlgMathAbs));
-            SetDelegateInfo(dlgMathCeiling, Marshal.GetFunctionPointerForDelegate<MathCeiling>(dlgMathCeiling));
-            SetDelegateInfo(dlgMathFloor, Marshal.GetFunctionPointerForDelegate<MathFloor>(dlgMathFloor));
-            SetDelegateInfo(dlgMathRound, Marshal.GetFunctionPointerForDelegate<MathRound>(dlgMathRound));
-            SetDelegateInfo(dlgMathTruncate, Marshal.GetFunctionPointerForDelegate<MathTruncate>(dlgMathTruncate));
-
-            SetDelegateInfo(dlgMathFAbs, Marshal.GetFunctionPointerForDelegate<MathFAbs>(dlgMathFAbs));
-            SetDelegateInfo(dlgMathFCeiling, Marshal.GetFunctionPointerForDelegate<MathFCeiling>(dlgMathFCeiling));
-            SetDelegateInfo(dlgMathFFloor, Marshal.GetFunctionPointerForDelegate<MathFFloor>(dlgMathFFloor));
-            SetDelegateInfo(dlgMathFRound, Marshal.GetFunctionPointerForDelegate<MathFRound>(dlgMathFRound));
-            SetDelegateInfo(dlgMathFTruncate, Marshal.GetFunctionPointerForDelegate<MathFTruncate>(dlgMathFTruncate));
-
-            SetDelegateInfo(dlgNativeInterfaceBreak, Marshal.GetFunctionPointerForDelegate<NativeInterfaceBreak>(dlgNativeInterfaceBreak));
-            SetDelegateInfo(dlgNativeInterfaceCheckSynchronization, Marshal.GetFunctionPointerForDelegate<NativeInterfaceCheckSynchronization>(dlgNativeInterfaceCheckSynchronization));
-            SetDelegateInfo(dlgNativeInterfaceEnqueueForRejit, Marshal.GetFunctionPointerForDelegate<NativeInterfaceEnqueueForRejit>(dlgNativeInterfaceEnqueueForRejit));
-            SetDelegateInfo(dlgNativeInterfaceGetCntfrqEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntfrqEl0>(dlgNativeInterfaceGetCntfrqEl0));
-            SetDelegateInfo(dlgNativeInterfaceGetCntpctEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntpctEl0>(dlgNativeInterfaceGetCntpctEl0));
-            SetDelegateInfo(dlgNativeInterfaceGetCntvctEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntvctEl0>(dlgNativeInterfaceGetCntvctEl0));
-            SetDelegateInfo(dlgNativeInterfaceGetCtrEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCtrEl0>(dlgNativeInterfaceGetCtrEl0));
-            SetDelegateInfo(dlgNativeInterfaceGetDczidEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetDczidEl0>(dlgNativeInterfaceGetDczidEl0));
-            SetDelegateInfo(dlgNativeInterfaceGetFunctionAddress, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetFunctionAddress>(dlgNativeInterfaceGetFunctionAddress));
-            SetDelegateInfo(dlgNativeInterfaceInvalidateCacheLine, Marshal.GetFunctionPointerForDelegate<NativeInterfaceInvalidateCacheLine>(dlgNativeInterfaceInvalidateCacheLine));
-            SetDelegateInfo(dlgNativeInterfaceReadByte, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadByte>(dlgNativeInterfaceReadByte));
-            SetDelegateInfo(dlgNativeInterfaceReadUInt16, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt16>(dlgNativeInterfaceReadUInt16));
-            SetDelegateInfo(dlgNativeInterfaceReadUInt32, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt32>(dlgNativeInterfaceReadUInt32));
-            SetDelegateInfo(dlgNativeInterfaceReadUInt64, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt64>(dlgNativeInterfaceReadUInt64));
-            SetDelegateInfo(dlgNativeInterfaceReadVector128, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadVector128>(dlgNativeInterfaceReadVector128));
-            SetDelegateInfo(dlgNativeInterfaceSignalMemoryTracking, Marshal.GetFunctionPointerForDelegate<NativeInterfaceSignalMemoryTracking>(dlgNativeInterfaceSignalMemoryTracking));
-            SetDelegateInfo(dlgNativeInterfaceSupervisorCall, Marshal.GetFunctionPointerForDelegate<NativeInterfaceSupervisorCall>(dlgNativeInterfaceSupervisorCall));
-            SetDelegateInfo(dlgNativeInterfaceThrowInvalidMemoryAccess, Marshal.GetFunctionPointerForDelegate<NativeInterfaceThrowInvalidMemoryAccess>(dlgNativeInterfaceThrowInvalidMemoryAccess));
-            SetDelegateInfo(dlgNativeInterfaceUndefined, Marshal.GetFunctionPointerForDelegate<NativeInterfaceUndefined>(dlgNativeInterfaceUndefined));
-            SetDelegateInfo(dlgNativeInterfaceWriteByte, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteByte>(dlgNativeInterfaceWriteByte));
-            SetDelegateInfo(dlgNativeInterfaceWriteUInt16, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt16>(dlgNativeInterfaceWriteUInt16));
-            SetDelegateInfo(dlgNativeInterfaceWriteUInt32, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt32>(dlgNativeInterfaceWriteUInt32));
-            SetDelegateInfo(dlgNativeInterfaceWriteUInt64, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt64>(dlgNativeInterfaceWriteUInt64));
-            SetDelegateInfo(dlgNativeInterfaceWriteVector128, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteVector128>(dlgNativeInterfaceWriteVector128));
-
-            SetDelegateInfo(dlgSoftFallbackCountLeadingSigns, Marshal.GetFunctionPointerForDelegate<SoftFallbackCountLeadingSigns>(dlgSoftFallbackCountLeadingSigns));
-            SetDelegateInfo(dlgSoftFallbackCountLeadingZeros, Marshal.GetFunctionPointerForDelegate<SoftFallbackCountLeadingZeros>(dlgSoftFallbackCountLeadingZeros));
-            SetDelegateInfo(dlgSoftFallbackCrc32b, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32b>(dlgSoftFallbackCrc32b));
-            SetDelegateInfo(dlgSoftFallbackCrc32cb, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cb>(dlgSoftFallbackCrc32cb));
-            SetDelegateInfo(dlgSoftFallbackCrc32ch, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32ch>(dlgSoftFallbackCrc32ch));
-            SetDelegateInfo(dlgSoftFallbackCrc32cw, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cw>(dlgSoftFallbackCrc32cw));
-            SetDelegateInfo(dlgSoftFallbackCrc32cx, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cx>(dlgSoftFallbackCrc32cx));
-            SetDelegateInfo(dlgSoftFallbackCrc32h, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32h>(dlgSoftFallbackCrc32h));
-            SetDelegateInfo(dlgSoftFallbackCrc32w, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32w>(dlgSoftFallbackCrc32w));
-            SetDelegateInfo(dlgSoftFallbackCrc32x, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32x>(dlgSoftFallbackCrc32x));
-            SetDelegateInfo(dlgSoftFallbackDecrypt, Marshal.GetFunctionPointerForDelegate<SoftFallbackDecrypt>(dlgSoftFallbackDecrypt));
-            SetDelegateInfo(dlgSoftFallbackEncrypt, Marshal.GetFunctionPointerForDelegate<SoftFallbackEncrypt>(dlgSoftFallbackEncrypt));
-            SetDelegateInfo(dlgSoftFallbackFixedRotate, Marshal.GetFunctionPointerForDelegate<SoftFallbackFixedRotate>(dlgSoftFallbackFixedRotate));
-            SetDelegateInfo(dlgSoftFallbackHashChoose, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashChoose>(dlgSoftFallbackHashChoose));
-            SetDelegateInfo(dlgSoftFallbackHashLower, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashLower>(dlgSoftFallbackHashLower));
-            SetDelegateInfo(dlgSoftFallbackHashMajority, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashMajority>(dlgSoftFallbackHashMajority));
-            SetDelegateInfo(dlgSoftFallbackHashParity, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashParity>(dlgSoftFallbackHashParity));
-            SetDelegateInfo(dlgSoftFallbackHashUpper, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashUpper>(dlgSoftFallbackHashUpper));
-            SetDelegateInfo(dlgSoftFallbackInverseMixColumns, Marshal.GetFunctionPointerForDelegate<SoftFallbackInverseMixColumns>(dlgSoftFallbackInverseMixColumns));
-            SetDelegateInfo(dlgSoftFallbackMixColumns, Marshal.GetFunctionPointerForDelegate<SoftFallbackMixColumns>(dlgSoftFallbackMixColumns));
-            SetDelegateInfo(dlgSoftFallbackPolynomialMult64_128, Marshal.GetFunctionPointerForDelegate<SoftFallbackPolynomialMult64_128>(dlgSoftFallbackPolynomialMult64_128));
-            SetDelegateInfo(dlgSoftFallbackSatF32ToS32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToS32>(dlgSoftFallbackSatF32ToS32));
-            SetDelegateInfo(dlgSoftFallbackSatF32ToS64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToS64>(dlgSoftFallbackSatF32ToS64));
-            SetDelegateInfo(dlgSoftFallbackSatF32ToU32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToU32>(dlgSoftFallbackSatF32ToU32));
-            SetDelegateInfo(dlgSoftFallbackSatF32ToU64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToU64>(dlgSoftFallbackSatF32ToU64));
-            SetDelegateInfo(dlgSoftFallbackSatF64ToS32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToS32>(dlgSoftFallbackSatF64ToS32));
-            SetDelegateInfo(dlgSoftFallbackSatF64ToS64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToS64>(dlgSoftFallbackSatF64ToS64));
-            SetDelegateInfo(dlgSoftFallbackSatF64ToU32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToU32>(dlgSoftFallbackSatF64ToU32));
-            SetDelegateInfo(dlgSoftFallbackSatF64ToU64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToU64>(dlgSoftFallbackSatF64ToU64));
-            SetDelegateInfo(dlgSoftFallbackSha1SchedulePart1, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha1SchedulePart1>(dlgSoftFallbackSha1SchedulePart1));
-            SetDelegateInfo(dlgSoftFallbackSha1SchedulePart2, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha1SchedulePart2>(dlgSoftFallbackSha1SchedulePart2));
-            SetDelegateInfo(dlgSoftFallbackSha256SchedulePart1, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha256SchedulePart1>(dlgSoftFallbackSha256SchedulePart1));
-            SetDelegateInfo(dlgSoftFallbackSha256SchedulePart2, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha256SchedulePart2>(dlgSoftFallbackSha256SchedulePart2));
-            SetDelegateInfo(dlgSoftFallbackSignedShrImm64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSignedShrImm64>(dlgSoftFallbackSignedShrImm64));
-            SetDelegateInfo(dlgSoftFallbackTbl1, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl1>(dlgSoftFallbackTbl1));
-            SetDelegateInfo(dlgSoftFallbackTbl2, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl2>(dlgSoftFallbackTbl2));
-            SetDelegateInfo(dlgSoftFallbackTbl3, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl3>(dlgSoftFallbackTbl3));
-            SetDelegateInfo(dlgSoftFallbackTbl4, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl4>(dlgSoftFallbackTbl4));
-            SetDelegateInfo(dlgSoftFallbackTbx1, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx1>(dlgSoftFallbackTbx1));
-            SetDelegateInfo(dlgSoftFallbackTbx2, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx2>(dlgSoftFallbackTbx2));
-            SetDelegateInfo(dlgSoftFallbackTbx3, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx3>(dlgSoftFallbackTbx3));
-            SetDelegateInfo(dlgSoftFallbackTbx4, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx4>(dlgSoftFallbackTbx4));
-            SetDelegateInfo(dlgSoftFallbackUnsignedShrImm64, Marshal.GetFunctionPointerForDelegate<SoftFallbackUnsignedShrImm64>(dlgSoftFallbackUnsignedShrImm64));
-
-            SetDelegateInfo(dlgSoftFloat16_32FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat16_32FPConvert>(dlgSoftFloat16_32FPConvert));
-            SetDelegateInfo(dlgSoftFloat16_64FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat16_64FPConvert>(dlgSoftFloat16_64FPConvert));
-
-            SetDelegateInfo(dlgSoftFloat32FPAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPAdd>(dlgSoftFloat32FPAdd));
-            SetDelegateInfo(dlgSoftFloat32FPAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPAddFpscr>(dlgSoftFloat32FPAddFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPCompare, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompare>(dlgSoftFloat32FPCompare));
-            SetDelegateInfo(dlgSoftFloat32FPCompareEQ, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareEQ>(dlgSoftFloat32FPCompareEQ));
-            SetDelegateInfo(dlgSoftFloat32FPCompareEQFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareEQFpscr>(dlgSoftFloat32FPCompareEQFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPCompareGE, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGE>(dlgSoftFloat32FPCompareGE));
-            SetDelegateInfo(dlgSoftFloat32FPCompareGEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGEFpscr>(dlgSoftFloat32FPCompareGEFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPCompareGT, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGT>(dlgSoftFloat32FPCompareGT));
-            SetDelegateInfo(dlgSoftFloat32FPCompareGTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGTFpscr>(dlgSoftFloat32FPCompareGTFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPCompareLE, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLE>(dlgSoftFloat32FPCompareLE));
-            SetDelegateInfo(dlgSoftFloat32FPCompareLEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLEFpscr>(dlgSoftFloat32FPCompareLEFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPCompareLT, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLT>(dlgSoftFloat32FPCompareLT));
-            SetDelegateInfo(dlgSoftFloat32FPCompareLTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLTFpscr>(dlgSoftFloat32FPCompareLTFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPDiv, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPDiv>(dlgSoftFloat32FPDiv));
-            SetDelegateInfo(dlgSoftFloat32FPMax, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMax>(dlgSoftFloat32FPMax));
-            SetDelegateInfo(dlgSoftFloat32FPMaxFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxFpscr>(dlgSoftFloat32FPMaxFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMaxNum, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxNum>(dlgSoftFloat32FPMaxNum));
-            SetDelegateInfo(dlgSoftFloat32FPMaxNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxNumFpscr>(dlgSoftFloat32FPMaxNumFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMin, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMin>(dlgSoftFloat32FPMin));
-            SetDelegateInfo(dlgSoftFloat32FPMinFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinFpscr>(dlgSoftFloat32FPMinFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMinNum, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinNum>(dlgSoftFloat32FPMinNum));
-            SetDelegateInfo(dlgSoftFloat32FPMinNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinNumFpscr>(dlgSoftFloat32FPMinNumFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMul, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMul>(dlgSoftFloat32FPMul));
-            SetDelegateInfo(dlgSoftFloat32FPMulFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulFpscr>(dlgSoftFloat32FPMulFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulAdd>(dlgSoftFloat32FPMulAdd));
-            SetDelegateInfo(dlgSoftFloat32FPMulAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulAddFpscr>(dlgSoftFloat32FPMulAddFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulSub>(dlgSoftFloat32FPMulSub));
-            SetDelegateInfo(dlgSoftFloat32FPMulSubFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulSubFpscr>(dlgSoftFloat32FPMulSubFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPMulX, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulX>(dlgSoftFloat32FPMulX));
-            SetDelegateInfo(dlgSoftFloat32FPNegMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPNegMulAdd>(dlgSoftFloat32FPNegMulAdd));
-            SetDelegateInfo(dlgSoftFloat32FPNegMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPNegMulSub>(dlgSoftFloat32FPNegMulSub));
-            SetDelegateInfo(dlgSoftFloat32FPRecipEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipEstimate>(dlgSoftFloat32FPRecipEstimate));
-            SetDelegateInfo(dlgSoftFloat32FPRecipEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipEstimateFpscr>(dlgSoftFloat32FPRecipEstimateFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPRecipStep, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipStep>(dlgSoftFloat32FPRecipStep)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPRecipStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipStepFused>(dlgSoftFloat32FPRecipStepFused));
-            SetDelegateInfo(dlgSoftFloat32FPRecpX, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecpX>(dlgSoftFloat32FPRecpX));
-            SetDelegateInfo(dlgSoftFloat32FPRSqrtEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtEstimate>(dlgSoftFloat32FPRSqrtEstimate));
-            SetDelegateInfo(dlgSoftFloat32FPRSqrtEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtEstimateFpscr>(dlgSoftFloat32FPRSqrtEstimateFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPRSqrtStep, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtStep>(dlgSoftFloat32FPRSqrtStep)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat32FPRSqrtStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtStepFused>(dlgSoftFloat32FPRSqrtStepFused));
-            SetDelegateInfo(dlgSoftFloat32FPSqrt, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPSqrt>(dlgSoftFloat32FPSqrt));
-            SetDelegateInfo(dlgSoftFloat32FPSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPSub>(dlgSoftFloat32FPSub));
-
-            SetDelegateInfo(dlgSoftFloat32_16FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat32_16FPConvert>(dlgSoftFloat32_16FPConvert));
-
-            SetDelegateInfo(dlgSoftFloat64FPAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPAdd>(dlgSoftFloat64FPAdd));
-            SetDelegateInfo(dlgSoftFloat64FPAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPAddFpscr>(dlgSoftFloat64FPAddFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPCompare, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompare>(dlgSoftFloat64FPCompare));
-            SetDelegateInfo(dlgSoftFloat64FPCompareEQ, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareEQ>(dlgSoftFloat64FPCompareEQ));
-            SetDelegateInfo(dlgSoftFloat64FPCompareEQFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareEQFpscr>(dlgSoftFloat64FPCompareEQFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPCompareGE, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGE>(dlgSoftFloat64FPCompareGE));
-            SetDelegateInfo(dlgSoftFloat64FPCompareGEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGEFpscr>(dlgSoftFloat64FPCompareGEFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPCompareGT, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGT>(dlgSoftFloat64FPCompareGT));
-            SetDelegateInfo(dlgSoftFloat64FPCompareGTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGTFpscr>(dlgSoftFloat64FPCompareGTFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPCompareLE, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLE>(dlgSoftFloat64FPCompareLE));
-            SetDelegateInfo(dlgSoftFloat64FPCompareLEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLEFpscr>(dlgSoftFloat64FPCompareLEFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPCompareLT, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLT>(dlgSoftFloat64FPCompareLT));
-            SetDelegateInfo(dlgSoftFloat64FPCompareLTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLTFpscr>(dlgSoftFloat64FPCompareLTFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPDiv, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPDiv>(dlgSoftFloat64FPDiv));
-            SetDelegateInfo(dlgSoftFloat64FPMax, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMax>(dlgSoftFloat64FPMax));
-            SetDelegateInfo(dlgSoftFloat64FPMaxFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxFpscr>(dlgSoftFloat64FPMaxFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMaxNum, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxNum>(dlgSoftFloat64FPMaxNum));
-            SetDelegateInfo(dlgSoftFloat64FPMaxNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxNumFpscr>(dlgSoftFloat64FPMaxNumFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMin, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMin>(dlgSoftFloat64FPMin));
-            SetDelegateInfo(dlgSoftFloat64FPMinFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinFpscr>(dlgSoftFloat64FPMinFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMinNum, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinNum>(dlgSoftFloat64FPMinNum));
-            SetDelegateInfo(dlgSoftFloat64FPMinNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinNumFpscr>(dlgSoftFloat64FPMinNumFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMul, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMul>(dlgSoftFloat64FPMul));
-            SetDelegateInfo(dlgSoftFloat64FPMulFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulFpscr>(dlgSoftFloat64FPMulFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulAdd>(dlgSoftFloat64FPMulAdd));
-            SetDelegateInfo(dlgSoftFloat64FPMulAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulAddFpscr>(dlgSoftFloat64FPMulAddFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulSub>(dlgSoftFloat64FPMulSub));
-            SetDelegateInfo(dlgSoftFloat64FPMulSubFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulSubFpscr>(dlgSoftFloat64FPMulSubFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPMulX, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulX>(dlgSoftFloat64FPMulX));
-            SetDelegateInfo(dlgSoftFloat64FPNegMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPNegMulAdd>(dlgSoftFloat64FPNegMulAdd));
-            SetDelegateInfo(dlgSoftFloat64FPNegMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPNegMulSub>(dlgSoftFloat64FPNegMulSub));
-            SetDelegateInfo(dlgSoftFloat64FPRecipEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipEstimate>(dlgSoftFloat64FPRecipEstimate));
-            SetDelegateInfo(dlgSoftFloat64FPRecipEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipEstimateFpscr>(dlgSoftFloat64FPRecipEstimateFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPRecipStep, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipStep>(dlgSoftFloat64FPRecipStep)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPRecipStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipStepFused>(dlgSoftFloat64FPRecipStepFused));
-            SetDelegateInfo(dlgSoftFloat64FPRecpX, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecpX>(dlgSoftFloat64FPRecpX));
-            SetDelegateInfo(dlgSoftFloat64FPRSqrtEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtEstimate>(dlgSoftFloat64FPRSqrtEstimate));
-            SetDelegateInfo(dlgSoftFloat64FPRSqrtEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtEstimateFpscr>(dlgSoftFloat64FPRSqrtEstimateFpscr)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPRSqrtStep, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtStep>(dlgSoftFloat64FPRSqrtStep)); // A32 only.
-            SetDelegateInfo(dlgSoftFloat64FPRSqrtStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtStepFused>(dlgSoftFloat64FPRSqrtStepFused));
-            SetDelegateInfo(dlgSoftFloat64FPSqrt, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPSqrt>(dlgSoftFloat64FPSqrt));
-            SetDelegateInfo(dlgSoftFloat64FPSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPSub>(dlgSoftFloat64FPSub));
-
-            SetDelegateInfo(dlgSoftFloat64_16FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat64_16FPConvert>(dlgSoftFloat64_16FPConvert));
+            SetDelegateInfo(typeof(SoftFloat64_16).GetMethod(nameof(SoftFloat64_16.FPConvert)));
         }
-
-        private delegate double MathAbs(double value);
-        private delegate double MathCeiling(double a);
-        private delegate double MathFloor(double d);
-        private delegate double MathRound(double value, MidpointRounding mode);
-        private delegate double MathTruncate(double d);
-
-        private delegate float MathFAbs(float x);
-        private delegate float MathFCeiling(float x);
-        private delegate float MathFFloor(float x);
-        private delegate float MathFRound(float x, MidpointRounding mode);
-        private delegate float MathFTruncate(float x);
-
-        private delegate void NativeInterfaceBreak(ulong address, int imm);
-        private delegate bool NativeInterfaceCheckSynchronization();
-        private delegate void NativeInterfaceEnqueueForRejit(ulong address);
-        private delegate ulong NativeInterfaceGetCntfrqEl0();
-        private delegate ulong NativeInterfaceGetCntpctEl0();
-        private delegate ulong NativeInterfaceGetCntvctEl0();
-        private delegate ulong NativeInterfaceGetCtrEl0();
-        private delegate ulong NativeInterfaceGetDczidEl0();
-        private delegate ulong NativeInterfaceGetFunctionAddress(ulong address);
-        private delegate void NativeInterfaceInvalidateCacheLine(ulong address);
-        private delegate byte NativeInterfaceReadByte(ulong address);
-        private delegate ushort NativeInterfaceReadUInt16(ulong address);
-        private delegate uint NativeInterfaceReadUInt32(ulong address);
-        private delegate ulong NativeInterfaceReadUInt64(ulong address);
-        private delegate V128 NativeInterfaceReadVector128(ulong address);
-        private delegate void NativeInterfaceSignalMemoryTracking(ulong address, ulong size, bool write);
-        private delegate void NativeInterfaceSupervisorCall(ulong address, int imm);
-        private delegate void NativeInterfaceThrowInvalidMemoryAccess(ulong address);
-        private delegate void NativeInterfaceUndefined(ulong address, int opCode);
-        private delegate void NativeInterfaceWriteByte(ulong address, byte value);
-        private delegate void NativeInterfaceWriteUInt16(ulong address, ushort value);
-        private delegate void NativeInterfaceWriteUInt32(ulong address, uint value);
-        private delegate void NativeInterfaceWriteUInt64(ulong address, ulong value);
-        private delegate void NativeInterfaceWriteVector128(ulong address, V128 value);
-
-        private delegate ulong SoftFallbackCountLeadingSigns(ulong value, int size);
-        private delegate ulong SoftFallbackCountLeadingZeros(ulong value, int size);
-        private delegate uint SoftFallbackCrc32b(uint crc, byte value);
-        private delegate uint SoftFallbackCrc32cb(uint crc, byte value);
-        private delegate uint SoftFallbackCrc32ch(uint crc, ushort value);
-        private delegate uint SoftFallbackCrc32cw(uint crc, uint value);
-        private delegate uint SoftFallbackCrc32cx(uint crc, ulong value);
-        private delegate uint SoftFallbackCrc32h(uint crc, ushort value);
-        private delegate uint SoftFallbackCrc32w(uint crc, uint value);
-        private delegate uint SoftFallbackCrc32x(uint crc, ulong value);
-        private delegate V128 SoftFallbackDecrypt(V128 value, V128 roundKey);
-        private delegate V128 SoftFallbackEncrypt(V128 value, V128 roundKey);
-        private delegate uint SoftFallbackFixedRotate(uint hash_e);
-        private delegate V128 SoftFallbackHashChoose(V128 hash_abcd, uint hash_e, V128 wk);
-        private delegate V128 SoftFallbackHashLower(V128 hash_abcd, V128 hash_efgh, V128 wk);
-        private delegate V128 SoftFallbackHashMajority(V128 hash_abcd, uint hash_e, V128 wk);
-        private delegate V128 SoftFallbackHashParity(V128 hash_abcd, uint hash_e, V128 wk);
-        private delegate V128 SoftFallbackHashUpper(V128 hash_abcd, V128 hash_efgh, V128 wk);
-        private delegate V128 SoftFallbackInverseMixColumns(V128 value);
-        private delegate V128 SoftFallbackMixColumns(V128 value);
-        private delegate V128 SoftFallbackPolynomialMult64_128(ulong op1, ulong op2);
-        private delegate int SoftFallbackSatF32ToS32(float value);
-        private delegate long SoftFallbackSatF32ToS64(float value);
-        private delegate uint SoftFallbackSatF32ToU32(float value);
-        private delegate ulong SoftFallbackSatF32ToU64(float value);
-        private delegate int SoftFallbackSatF64ToS32(double value);
-        private delegate long SoftFallbackSatF64ToS64(double value);
-        private delegate uint SoftFallbackSatF64ToU32(double value);
-        private delegate ulong SoftFallbackSatF64ToU64(double value);
-        private delegate V128 SoftFallbackSha1SchedulePart1(V128 w0_3, V128 w4_7, V128 w8_11);
-        private delegate V128 SoftFallbackSha1SchedulePart2(V128 tw0_3, V128 w12_15);
-        private delegate V128 SoftFallbackSha256SchedulePart1(V128 w0_3, V128 w4_7);
-        private delegate V128 SoftFallbackSha256SchedulePart2(V128 w0_3, V128 w8_11, V128 w12_15);
-        private delegate long SoftFallbackSignedShrImm64(long value, long roundConst, int shift);
-        private delegate V128 SoftFallbackTbl1(V128 vector, int bytes, V128 tb0);
-        private delegate V128 SoftFallbackTbl2(V128 vector, int bytes, V128 tb0, V128 tb1);
-        private delegate V128 SoftFallbackTbl3(V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2);
-        private delegate V128 SoftFallbackTbl4(V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2, V128 tb3);
-        private delegate V128 SoftFallbackTbx1(V128 dest, V128 vector, int bytes, V128 tb0);
-        private delegate V128 SoftFallbackTbx2(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1);
-        private delegate V128 SoftFallbackTbx3(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2);
-        private delegate V128 SoftFallbackTbx4(V128 dest, V128 vector, int bytes, V128 tb0, V128 tb1, V128 tb2, V128 tb3);
-        private delegate ulong SoftFallbackUnsignedShrImm64(ulong value, long roundConst, int shift);
-
-        private delegate float SoftFloat16_32FPConvert(ushort valueBits);
-
-        private delegate double SoftFloat16_64FPConvert(ushort valueBits);
-
-        private delegate float SoftFloat32FPAdd(float value1, float value2);
-        private delegate float SoftFloat32FPAddFpscr(float value1, float value2, bool standardFpscr);
-        private delegate int SoftFloat32FPCompare(float value1, float value2, bool signalNaNs);
-        private delegate float SoftFloat32FPCompareEQ(float value1, float value2);
-        private delegate float SoftFloat32FPCompareEQFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPCompareGE(float value1, float value2);
-        private delegate float SoftFloat32FPCompareGEFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPCompareGT(float value1, float value2);
-        private delegate float SoftFloat32FPCompareGTFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPCompareLE(float value1, float value2);
-        private delegate float SoftFloat32FPCompareLEFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPCompareLT(float value1, float value2);
-        private delegate float SoftFloat32FPCompareLTFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPDiv(float value1, float value2);
-        private delegate float SoftFloat32FPMax(float value1, float value2);
-        private delegate float SoftFloat32FPMaxFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMaxNum(float value1, float value2);
-        private delegate float SoftFloat32FPMaxNumFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMin(float value1, float value2);
-        private delegate float SoftFloat32FPMinFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMinNum(float value1, float value2);
-        private delegate float SoftFloat32FPMinNumFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMul(float value1, float value2);
-        private delegate float SoftFloat32FPMulFpscr(float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMulAdd(float valueA, float value1, float value2);
-        private delegate float SoftFloat32FPMulAddFpscr(float valueA, float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMulSub(float valueA, float value1, float value2);
-        private delegate float SoftFloat32FPMulSubFpscr(float valueA, float value1, float value2, bool standardFpscr);
-        private delegate float SoftFloat32FPMulX(float value1, float value2);
-        private delegate float SoftFloat32FPNegMulAdd(float valueA, float value1, float value2);
-        private delegate float SoftFloat32FPNegMulSub(float valueA, float value1, float value2);
-        private delegate float SoftFloat32FPRecipEstimate(float value);
-        private delegate float SoftFloat32FPRecipEstimateFpscr(float value, bool standardFpscr);
-        private delegate float SoftFloat32FPRecipStep(float value1, float value2);
-        private delegate float SoftFloat32FPRecipStepFused(float value1, float value2);
-        private delegate float SoftFloat32FPRecpX(float value);
-        private delegate float SoftFloat32FPRSqrtEstimate(float value);
-        private delegate float SoftFloat32FPRSqrtEstimateFpscr(float value, bool standardFpscr);
-        private delegate float SoftFloat32FPRSqrtStep(float value1, float value2);
-        private delegate float SoftFloat32FPRSqrtStepFused(float value1, float value2);
-        private delegate float SoftFloat32FPSqrt(float value);
-        private delegate float SoftFloat32FPSub(float value1, float value2);
-
-        private delegate ushort SoftFloat32_16FPConvert(float value);
-
-        private delegate double SoftFloat64FPAdd(double value1, double value2);
-        private delegate double SoftFloat64FPAddFpscr(double value1, double value2, bool standardFpscr);
-        private delegate int SoftFloat64FPCompare(double value1, double value2, bool signalNaNs);
-        private delegate double SoftFloat64FPCompareEQ(double value1, double value2);
-        private delegate double SoftFloat64FPCompareEQFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPCompareGE(double value1, double value2);
-        private delegate double SoftFloat64FPCompareGEFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPCompareGT(double value1, double value2);
-        private delegate double SoftFloat64FPCompareGTFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPCompareLE(double value1, double value2);
-        private delegate double SoftFloat64FPCompareLEFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPCompareLT(double value1, double value2);
-        private delegate double SoftFloat64FPCompareLTFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPDiv(double value1, double value2);
-        private delegate double SoftFloat64FPMax(double value1, double value2);
-        private delegate double SoftFloat64FPMaxFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMaxNum(double value1, double value2);
-        private delegate double SoftFloat64FPMaxNumFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMin(double value1, double value2);
-        private delegate double SoftFloat64FPMinFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMinNum(double value1, double value2);
-        private delegate double SoftFloat64FPMinNumFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMul(double value1, double value2);
-        private delegate double SoftFloat64FPMulFpscr(double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMulAdd(double valueA, double value1, double value2);
-        private delegate double SoftFloat64FPMulAddFpscr(double valueA, double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMulSub(double valueA, double value1, double value2);
-        private delegate double SoftFloat64FPMulSubFpscr(double valueA, double value1, double value2, bool standardFpscr);
-        private delegate double SoftFloat64FPMulX(double value1, double value2);
-        private delegate double SoftFloat64FPNegMulAdd(double valueA, double value1, double value2);
-        private delegate double SoftFloat64FPNegMulSub(double valueA, double value1, double value2);
-        private delegate double SoftFloat64FPRecipEstimate(double value);
-        private delegate double SoftFloat64FPRecipEstimateFpscr(double value, bool standardFpscr);
-        private delegate double SoftFloat64FPRecipStep(double value1, double value2);
-        private delegate double SoftFloat64FPRecipStepFused(double value1, double value2);
-        private delegate double SoftFloat64FPRecpX(double value);
-        private delegate double SoftFloat64FPRSqrtEstimate(double value);
-        private delegate double SoftFloat64FPRSqrtEstimateFpscr(double value, bool standardFpscr);
-        private delegate double SoftFloat64FPRSqrtStep(double value1, double value2);
-        private delegate double SoftFloat64FPRSqrtStepFused(double value1, double value2);
-        private delegate double SoftFloat64FPSqrt(double value);
-        private delegate double SoftFloat64FPSub(double value1, double value2);
-
-        private delegate ushort SoftFloat64_16FPConvert(double value);
     }
 }

src/ARMeilleure/Translation/EmitterContext.cs

@@ -97,7 +97,7 @@ namespace ARMeilleure.Translation
 
         public virtual Operand Call(MethodInfo info, params Operand[] callArgs)
         {
-            nint funcPtr = Delegates.GetDelegateFuncPtr(info);
+            nint funcPtr = info.MethodHandle.GetFunctionPointer();
 
             OperandType returnType = GetOperandType(info.ReturnType);
 

src/ARMeilleure/Translation/Translator.cs

@@ -413,6 +413,10 @@ namespace ARMeilleure.Translation
 
                         context.CurrOp = opCode;
 
+                        Operand nativeContext = context.LoadArgument(OperandType.I64, 0);
+                        Operand dispAddressAddr = context.Add(nativeContext, Const((ulong)NativeContext.GetDispatchAddressOffset()));
+                        context.Store(dispAddressAddr, Const(opCode.Address));
+
                         bool isLastOp = opcIndex == block.OpCodes.Count - 1;
 
                         if (isLastOp)

src/Ryujinx.Audio.Backends.OpenAL/Ryujinx.Audio.Backends.OpenAL.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Audio.Backends.SDL3/Ryujinx.Audio.Backends.SDL3.csproj

@@ -3,8 +3,6 @@
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Audio.Backends.SoundIo/Ryujinx.Audio.Backends.SoundIo.csproj

@@ -4,8 +4,6 @@
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <RuntimeIdentifiers>win-x64;osx-x64;linux-x64;win-arm64;osx-arm64;linux-arm64</RuntimeIdentifiers>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Audio/Ryujinx.Audio.csproj

@@ -3,8 +3,6 @@
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Common/Configuration/Hid/Controller/RumbleConfigController.cs

@@ -16,5 +16,10 @@ namespace Ryujinx.Common.Configuration.Hid.Controller
         /// Enable Rumble
         /// </summary>
         public bool EnableRumble { get; set; }
+        
+        /// <summary>
+        /// Enable HD Rumble support
+        /// </summary
+        public bool UseHDRumble { get; set; }
     }
 }

src/Ryujinx.Common/Ryujinx.Common.csproj

@@ -3,8 +3,6 @@
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefineConstants Condition=" '$(ExtraDefineConstants)' != '' ">$(DefineConstants);$(ExtraDefineConstants)</DefineConstants>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Common/Utilities/EmbeddedResources.cs

@@ -126,8 +126,13 @@ namespace Ryujinx.Common
 
         public static string[] GetAllAvailableResources(string path, string ext = "")
         {
-            return ResolveManifestPath(path).Item1.GetManifestResourceNames()
-                .Where(r => r.EndsWith(ext))
+            (Assembly assembly, string resourcePath) = ResolveManifestPath(path);
+
+            string manifestPath = assembly.GetName().Name + "." + resourcePath.Replace('/', '.');
+
+            return assembly.GetManifestResourceNames()
+                .Where(r => r.StartsWith(manifestPath + ".", StringComparison.Ordinal))
+                .Where(r => r.EndsWith(ext, StringComparison.Ordinal))
                 .ToArray();
         }
 

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

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

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

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

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

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

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

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

src/Ryujinx.Cpu/Ryujinx.Cpu.csproj

@@ -3,8 +3,6 @@
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Device/Ryujinx.Graphics.Device.csproj

@@ -1,8 +1,6 @@
 <Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.GAL/Capabilities.cs

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

src/Ryujinx.Graphics.GAL/Ryujinx.Graphics.GAL.csproj

@@ -1,10 +1,9 @@
-<Project Sdk="Microsoft.NET.Sdk">
+<Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
+    <Configurations>Debug;Release</Configurations>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

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

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

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

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

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

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

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

@@ -1572,11 +1572,6 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="bound">True if the texture has been bound, false if it has been unbound</param>
         public void SignalModifying(bool bound)
         {
-            if (bound)
-            {
-                _scaledSetScore = Math.Max(0, _scaledSetScore - 1);
-            }
-
             if (_modifiedStale || Group.HasCopyDependencies || Group.HasFlushBuffer)
             {
                 _modifiedStale = false;
@@ -1584,9 +1579,18 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
 
             _physicalMemory.TextureCache.Lift(this);
+        }
 
+        /// <summary>
+        /// Signals that a render target texture has been either bound or unbound.
+        /// </summary>
+        /// <param name="bound">True if the texture has been bound, false if it has been unbound</param>
+        public void SignalBindingChange(bool bound)
+        {
             if (bound)
             {
+                _scaledSetScore = Math.Max(0, _scaledSetScore - 1);
+
                 IncrementReferenceCount();
             }
             else

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

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

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

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

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

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

src/Ryujinx.Graphics.Gpu/Ryujinx.Graphics.Gpu.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

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

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

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

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

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

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

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

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

src/Ryujinx.Graphics.Host1x/Ryujinx.Graphics.Host1x.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Nvdec.FFmpeg/Ryujinx.Graphics.Nvdec.FFmpeg.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Nvdec.Vp9/Ryujinx.Graphics.Nvdec.Vp9.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Nvdec/Ryujinx.Graphics.Nvdec.csproj

@@ -3,8 +3,6 @@
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

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

src/Ryujinx.Graphics.OpenGL/Ryujinx.Graphics.OpenGL.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.RenderDocApi/Capture.cs

@@ -0,0 +1,12 @@
+using System;
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    public readonly record struct Capture(int Index, string FileName, DateTime Timestamp)
+    {
+        public void SetComments(string comments)
+        {
+            RenderDoc.SetCaptureFileComments(FileName, comments);
+        }
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/CaptureOption.cs

@@ -0,0 +1,100 @@
+// ReSharper disable UnusedMember.Global
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    public enum CaptureOption
+    {
+        /// <summary>
+        /// specifies whether the application is allowed to enable vsync. Default is on.
+        /// </summary>
+        AllowVsync = 0,
+        /// <summary>
+        /// specifies whether the application is allowed to enter exclusive fullscreen. Default is on.
+        /// </summary>
+        AllowFullscreen = 1,
+        /// <summary>
+        /// specifies whether (where possible) API-specific debugging is enabled. Default is off.
+        /// </summary>
+        ApiValidation = 2,
+        /// <summary>
+        /// specifies whether each API call should save a callstack. Default is off.
+        /// </summary>
+        CaptureCallstacks = 3,
+        /// <summary>
+        /// specifies whether, if <see cref="CaptureCallstacks"/> is enabled, callstacks are only saved on actions. Default is off.
+        /// </summary>
+        CaptureCallstacksOnlyDraws = 4,
+        /// <summary>
+        /// specifies a delay in seconds after launching a process to pause, to allow debuggers to attach. <br/>
+        /// This will only apply to child processes since the delay happens at process startup. Default is 0.
+        /// </summary>
+        DelayForDebugger = 5,
+        /// <summary>
+        /// specifies whether any mapped memory updates should be bounds-checked for overruns,
+        /// and uninitialised buffers are initialized to <code>0xDDDDDDDD</code> to catch use of uninitialised data.
+        /// Only supported on D3D11 and OpenGL. Default is off.
+        /// </summary>
+        /// <remarks>
+        /// This option is only valid for OpenGL and D3D11. Explicit APIs such as D3D12 and Vulkan do
+        /// not do the same kind of interception &amp; checking, and undefined contents are really undefined.
+        /// </remarks>
+        VerifyBufferAccess = 6,
+        /// <summary>
+        /// Hooks any system API calls that create child processes, and injects
+        /// RenderDoc into them recursively with the same options.
+        /// </summary>
+        HookIntoChildren = 7,
+        /// <summary>
+        /// specifies whether all live resources at the time of capture should be included in the capture,
+        /// even if they are not referenced by the frame. Default is off.
+        /// </summary>
+        RefAllSources = 8,
+        /// <summary>
+        /// By default, RenderDoc skips saving initial states for resources where the
+        /// previous contents don't appear to be used, assuming that writes before
+        /// reads indicate previous contents aren't used.
+        /// </summary>
+        /// <remarks>
+        /// **NOTE**: As of RenderDoc v1.1 this option has been deprecated. Setting or
+        /// getting it will be ignored, to allow compatibility with older versions.
+        /// In v1.1 the option acts as if it's always enabled.
+        /// </remarks>
+        SaveAllInitials = 9,
+        /// <summary>
+        /// In APIs that allow for the recording of command lists to be replayed later,
+        /// RenderDoc may choose to not capture command lists before a frame capture is
+        /// triggered, to reduce overheads. This means any command lists recorded once
+        /// and replayed many times will not be available and may cause a failure to
+        /// capture.
+        /// </summary>
+        /// <remarks>
+        /// NOTE: This is only true for APIs where multithreading is difficult or
+        /// discouraged. Newer APIs like Vulkan and D3D12 will ignore this option
+        /// and always capture all command lists since the API is heavily oriented
+        /// around it and the overheads have been reduced by API design.
+        /// </remarks>
+        CaptureAllCmdLists = 10,
+        /// <summary>
+        /// Mute API debugging output when the <see cref="ApiValidation"/> option is enabled.
+        /// </summary>
+        DebugOutputMute = 11,
+        /// <summary>
+        /// Allow vendor extensions to be used even when they may be
+        /// incompatible with RenderDoc and cause corrupted replays or crashes.
+        /// </summary>
+        AllowUnsupportedVendorExtensions = 12,
+        /// <summary>
+        /// Define a soft memory limit which some APIs may aim to keep overhead under where
+        /// possible. Anything above this limit will where possible be saved directly to disk during
+        /// capture.<br/>
+        /// This will cause increased disk space use (which may cause a capture to fail if disk space is
+        /// exhausted) as well as slower capture times.
+        /// <br/><br/>
+        /// Not all memory allocations may be deferred like this so it is not a guarantee of a memory
+        /// limit.
+        /// <br/><br/>
+        /// Units are in MBs, suggested values would range from 200MB to 1000MB.
+        /// </summary>
+        SoftMemoryLimit = 13,
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/InputButton.cs

@@ -0,0 +1,83 @@
+
+// ReSharper disable UnusedMember.Global
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    public enum InputButton
+    {
+        // '0' - '9' matches ASCII values
+        Key0 = 0x30,
+        Key1 = 0x31,
+        Key2 = 0x32,
+        Key3 = 0x33,
+        Key4 = 0x34,
+        Key5 = 0x35,
+        Key6 = 0x36,
+        Key7 = 0x37,
+        Key8 = 0x38,
+        Key9 = 0x39,
+
+        // 'A' - 'Z' matches ASCII values
+        A = 0x41,
+        B = 0x42,
+        C = 0x43,
+        D = 0x44,
+        E = 0x45,
+        F = 0x46,
+        G = 0x47,
+        H = 0x48,
+        I = 0x49,
+        J = 0x4A,
+        K = 0x4B,
+        L = 0x4C,
+        M = 0x4D,
+        N = 0x4E,
+        O = 0x4F,
+        P = 0x50,
+        Q = 0x51,
+        R = 0x52,
+        S = 0x53,
+        T = 0x54,
+        U = 0x55,
+        V = 0x56,
+        W = 0x57,
+        X = 0x58,
+        Y = 0x59,
+        Z = 0x5A,
+
+        // leave the rest of the ASCII range free
+        // in case we want to use it later
+        NonPrintable = 0x100,
+
+        Divide,
+        Multiply,
+        Subtract,
+        Plus,
+
+        F1,
+        F2,
+        F3,
+        F4,
+        F5,
+        F6,
+        F7,
+        F8,
+        F9,
+        F10,
+        F11,
+        F12,
+
+        Home,
+        End,
+        Insert,
+        Delete,
+        PageUp,
+        PageDn,
+
+        Backspace,
+        Tab,
+        PrtScrn,
+        Pause,
+
+        Max,
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/OverlayBits.cs

@@ -0,0 +1,39 @@
+// ReSharper disable UnusedMember.Global
+
+using System;
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    [Flags]
+    public enum OverlayBits
+    {
+        /// <summary>
+        /// This single bit controls whether the overlay is enabled or disabled globally
+        /// </summary>
+        Enabled = 1 << 0,
+        /// <summary>
+        /// Show the average framerate over several seconds as well as min/max
+        /// </summary>
+        FrameRate = 1 << 1,
+        /// <summary>
+        /// Show the current frame number
+        /// </summary>
+        FrameNumber = 1 << 2,
+        /// <summary>
+        /// Show a list of recent captures, and how many captures have been made
+        /// </summary>
+        CaptureList = 1 << 3,
+        /// <summary>
+        /// Default values for the overlay mask
+        /// </summary>
+        Default = Enabled | FrameRate | FrameNumber | CaptureList,
+        /// <summary>
+        /// Enable all bits
+        /// </summary>
+        All = ~0,
+        /// <summary>
+        /// Disable all bits
+        /// </summary>
+        None = 0
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/README.md

@@ -0,0 +1,5 @@
+# Ryujinx.Graphics.RenderDocApi
+
+This is a C# binding for RenderDoc's application API.
+This is a source-inclusion of https://github.com/utkumaden/RenderdocSharp. 
+I didn't use the NuGet package as I had a few minor changes I wanted to make, and I want to learn from it as well via hands-on experience.

src/Ryujinx.Graphics.RenderDocApi/RenderDoc.cs

@@ -0,0 +1,639 @@
+using System;
+using System.Collections.Immutable;
+using System.Diagnostics;
+using System.Diagnostics.CodeAnalysis;
+using System.Reflection;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+using System.Text;
+using System.Text.RegularExpressions;
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    public static unsafe partial class RenderDoc
+    {
+        /// <summary>
+        /// True if the API is available.
+        /// </summary>
+        public static bool IsAvailable => Api != null;
+
+        /// <summary>
+        /// Set the minimum version of the API you require.
+        /// </summary>
+        /// <remarks>Set this before you do anything else with the RenderDoc API, including <see cref="IsAvailable"/>.</remarks>
+        public static RenderDocVersion MinimumRequired { get; set; } = RenderDocVersion.Version_1_0_0;
+
+        /// <summary>
+        /// Set to true to assert versions.
+        /// </summary>
+        public static bool AssertVersionEnabled { get; set; } = true;
+
+        /// <summary>
+        /// Version of the API available.
+        /// </summary>
+        [MemberNotNullWhen(true, nameof(IsAvailable))]
+        public static Version? Version
+        {
+            get
+            {
+                if (!IsAvailable)
+                    return null;
+
+                int major, minor, build;
+                Api->GetApiVersion(&major, &minor, &build);
+                return new Version(major, minor, build);
+            }
+        }
+
+        /// <summary>
+        /// The current mask which determines what sections of the overlay render on each window.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static OverlayBits OverlayBits
+        {
+            get => Api->GetOverlayBits();
+            set
+            {
+                Api->MaskOverlayBits(~value, value);
+            }
+        }
+
+        /// <summary>
+        /// The template for new captures.<br/>
+        /// The template can either be a relative or absolute path, which determines where captures will be saved and how they will be named.
+        /// If the path template is 'my_captures/example', then captures saved will be e.g.
+        /// 'my_captures/example_frame123.rdc' and 'my_captures/example_frame456.rdc'.<br/>
+        /// Relative paths will be saved relative to the process’s current working directory.<br/>
+        /// </summary>
+        /// <remarks>The default template is in a folder controlled by the UI - initially the system temporary folder, and the filename is the executable’s filename.</remarks>
+        [RenderDocApiVersion(1, 0)]
+        public static string CaptureFilePathTemplate
+        {
+            get
+            {
+                byte* ptr = Api->GetCaptureFilePathTemplate();
+                return Marshal.PtrToStringUTF8((nint)ptr)!;
+            }
+            set
+            {
+                fixed (byte* ptr = value.ToNullTerminatedByteArray())
+                {
+                    Api->SetCaptureFilePathTemplate(ptr);
+                }
+            }
+        }
+
+        /// <summary>
+        /// The amount of frame captures that have been made.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static int CaptureCount => Api->GetNumCaptures();
+
+        /// <summary>
+        /// Checks if the RenderDoc UI is currently connected to this process.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static bool IsTargetControlConnected => Api is not null && Api->IsTargetControlConnected() != 0;
+
+        /// <summary>
+        /// Checks if the current frame is capturing.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static bool IsFrameCapturing => Api is not null && Api->IsFrameCapturing() != 0;
+
+        /// <summary>
+        ///     Set one of the options for tweaking some behaviors of capturing.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be set.</param>
+        /// <param name="integer">the unsigned integer value to set for the option.</param>
+        /// <remarks>Note that each option only takes effect from after it is set - so it is advised to set these options as early as possible, ideally before any graphics API has been initialized.</remarks>
+        /// <returns>
+        /// true, if the <paramref name="option"/> is valid, and the value set on the option is within valid ranges.<br/>
+        /// false, if the option is not a <see cref="CaptureOption"/>, or the value is not valid for the option.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool SetCaptureOption(CaptureOption option, uint integer)
+        {
+            return Api is not null && Api->SetCaptureOptionU32(option, integer) != 0;
+        }
+
+        /// <summary>
+        /// Set one of the options for tweaking some behaviors of capturing.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be set.</param>
+        /// <param name="boolean">the value to set for the option, converted to a 0 or 1 before setting.</param>
+        /// <remarks>Note that each option only takes effect from after it is set - so it is advised to set these options as early as possible, ideally before any graphics API has been initialized.</remarks>
+        /// <returns>
+        /// true, if the <paramref name="option"/> is valid, and the value set on the option is within valid ranges.<br/>
+        /// false, if the option is not a <see cref="CaptureOption"/>, or the value is not valid for the option.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool SetCaptureOption(CaptureOption option, bool boolean)
+            => SetCaptureOption(option, boolean ? 1 : 0);
+
+        /// <summary>
+        /// Set one of the options for tweaking some behaviors of capturing.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be set.</param>
+        /// <param name="single">the floating point value to set for the option.</param>
+        /// <remarks>Note that each option only takes effect from after it is set - so it is advised to set these options as early as possible, ideally before any graphics API has been initialized.</remarks>
+        /// <returns>
+        /// true, if the <paramref name="option"/> is valid, and the value set on the option is within valid ranges.<br/>
+        /// false, if the option is not a <see cref="CaptureOption"/>, or the value is not valid for the option.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool SetCaptureOption(CaptureOption option, float single)
+        {
+            return Api is not null && Api->SetCaptureOptionF32(option, single) != 0;
+        }
+
+        /// <summary>
+        /// Gets the current value of one of the different options in <see cref="CaptureOption"/>, writing it to an out parameter.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be retrieved.</param>
+        /// <param name="integer">the value of the capture option, if the option is a valid <see cref="CaptureOption"/> enum. Otherwise, <see cref="int.MaxValue"/>.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void GetCaptureOption(CaptureOption option, out uint integer)
+        {
+            integer = Api->GetCaptureOptionU32(option);
+        }
+
+        /// <summary>
+        ///     Gets the current value of one of the different options in <see cref="CaptureOption"/>, writing it to an out parameter.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be retrieved.</param>
+        /// <param name="single">the value of the capture option, if the option is a valid <see cref="CaptureOption"/> enum. Otherwise, -<see cref="float.MaxValue"/>.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void GetCaptureOption(CaptureOption option, out float single)
+        {
+            single = Api->GetCaptureOptionF32(option);
+        }
+
+        /// <summary>
+        /// Gets the current value of one of the different options in <see cref="CaptureOption"/>,
+        /// converted to a boolean.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be retrieved.</param>
+        /// <returns>
+        /// the value of the capture option, converted to bool, if the option is a valid <see cref="CaptureOption"/> enum.
+        /// Otherwise, returns null.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool? GetCaptureOptionBool(CaptureOption option)
+        {
+            if (Api is null) return false;
+
+            uint returnVal = GetCaptureOptionU32(option);
+            if (returnVal == uint.MaxValue)
+                return null;
+
+            return returnVal is not 0;
+        }
+
+        /// <summary>
+        /// Gets the current value of one of the different options in <see cref="CaptureOption"/>.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be retrieved.</param>
+        /// <returns>
+        /// the value of the capture option, if the option is a valid <see cref="CaptureOption"/> enum.
+        /// Otherwise, returns <see cref="int.MaxValue"/>.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static uint GetCaptureOptionU32(CaptureOption option) => Api->GetCaptureOptionU32(option);
+
+        /// <summary>
+        /// Gets the current value of one of the different options in <see cref="CaptureOption"/>.
+        /// </summary>
+        /// <param name="option">specifies which capture option should be retrieved.</param>
+        /// <returns>
+        /// the value of the capture option, if the option is a valid <see cref="CaptureOption"/> enum.
+        /// Otherwise, returns -<see cref="float.MaxValue"/>.
+        /// </returns>
+        [RenderDocApiVersion(1, 0)]
+        public static float GetCaptureOptionF32(CaptureOption option) => Api->GetCaptureOptionF32(option);
+
+        /// <summary>
+        /// Changes the key bindings in-application for changing the focussed window.
+        /// </summary>
+        /// <param name="buttons">lists the keys to bind.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void SetFocusToggleKeys(ReadOnlySpan<InputButton> buttons)
+        {
+            if (Api is null) return;
+
+            fixed (InputButton* ptr = buttons)
+            {
+                Api->SetFocusToggleKeys(ptr, buttons.Length);
+            }
+        }
+
+        /// <summary>
+        /// Changes the key bindings in-application for triggering a capture on the current window.
+        /// </summary>
+        /// <param name="buttons">lists the keys to bind.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void SetCaptureKeys(ReadOnlySpan<InputButton> buttons)
+        {
+            if (Api is null) return;
+
+            fixed (InputButton* ptr = buttons)
+            {
+                Api->SetCaptureKeys(ptr, buttons.Length);
+            }
+        }
+
+        /// <summary>
+        /// Attempts to remove RenderDoc and its hooks from the target process.<br/>
+        /// It must be called as early as possible in the process, and will have undefined results
+        /// if any graphics API functions have been called.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static void RemoveHooks()
+        {
+            if (Api is null) return;
+
+            Api->RemoveHooks();
+        }
+
+        /// <summary>
+        /// Remove RenderDoc’s crash handler from the target process.<br/>
+        /// If you have your own crash handler that you want to handle any exceptions,
+        /// RenderDoc’s handler could interfere; so it can be disabled.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static void UnloadCrashHandler()
+        {
+            if (Api is null) return;
+
+            Api->UnloadCrashHandler();
+        }
+
+        /// <summary>
+        /// Trigger a capture as if the user had pressed one of the capture hotkeys.<br/>
+        /// The capture will be taken from the next frame presented to whichever window is considered current.
+        /// </summary>
+        [RenderDocApiVersion(1, 0)]
+        public static void TriggerCapture()
+        {
+            if (Api is null) return;
+
+            Api->TriggerCapture();
+        }
+
+
+        /// <summary>
+        /// Gets the details of all frame capture in the current session.
+        /// This simply calls <see cref="GetCapture"/> for each index available as specified by <see cref="CaptureCount"/>.
+        /// </summary>
+        /// <returns>An immutable array of structs representing RenderDoc Captures.</returns>
+        public static ImmutableArray<Capture> GetCaptures()
+        {
+            if (Api is null) return [];
+            int captureCount = CaptureCount;
+            if (captureCount is 0) return [];
+
+            ImmutableArray<Capture>.Builder captures = ImmutableArray.CreateBuilder<Capture>(captureCount);
+
+            for (int captureIndex = 0; captureIndex < captureCount; captureIndex++)
+            {
+                if (GetCapture(captureIndex) is { } capture)
+                    captures.Add(capture);
+            }
+
+            return captures.DrainToImmutable();
+        }
+
+        /// <summary>
+        /// Gets the details of a particular frame capture, as specified by an index from 0 to <see cref="CaptureCount"/> - 1.
+        /// </summary>
+        /// <param name="index">specifies which capture to return the details of. Must be less than the value returned by <see cref="CaptureCount"/>.</param>
+        /// <returns>A struct representing a RenderDoc Capture.</returns>
+        [RenderDocApiVersion(1, 0)]
+        public static Capture? GetCapture(int index)
+        {
+            if (Api is null) return null;
+
+            int length = 0;
+            if (Api->GetCapture(index, null, &length, null) == 0)
+            {
+                return null;
+            }
+
+            Span<byte> bytes = stackalloc byte[length + 1];
+            long timestamp;
+
+            fixed (byte* ptr = bytes)
+                Api->GetCapture(index, ptr, &length, &timestamp);
+
+            string fileName = Encoding.UTF8.GetString(bytes[length..]);
+            return new Capture(index, fileName, DateTimeOffset.FromUnixTimeSeconds(timestamp).DateTime);
+        }
+
+        /// <summary>
+        /// Determine the closest matching replay UI executable for the current RenderDoc module, and launch it.
+        /// </summary>
+        /// <param name="connectTargetControl">if the UI should immediately connect to the application.</param>
+        /// <param name="commandLine">string to be appended to the command line, e.g. a capture filename. If this parameter is null, the command line will be unmodified.</param>
+        /// <returns>true if the UI was successfully launched; false otherwise.</returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool LaunchReplayUI(bool connectTargetControl, string? commandLine = null)
+        {
+            if (Api is null) return false;
+
+            if (commandLine == null)
+            {
+                return Api->LaunchReplayUI(connectTargetControl ? 1u : 0u, null) != 0;
+            }
+
+            fixed (byte* ptr = commandLine.ToNullTerminatedByteArray())
+            {
+                return Api->LaunchReplayUI(connectTargetControl ? 1u : 0u, ptr) != 0;
+            }
+        }
+
+        /// <summary>
+        /// Explicitly sets which window is considered active.<br/>
+        /// The active window is the one that will be captured when the keybind to trigger a capture is pressed.
+        /// </summary>
+        /// <param name="hDevice">a handle to the API ‘device’ object that will be set active. Must be valid.</param>
+        /// <param name="hWindow">a handle to the platform window handle that will be set active. Must be valid.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void SetActiveWindow(nint hDevice, nint hWindow)
+        {
+            if (Api is null) return;
+
+            Api->SetActiveWindow((void*)hDevice, (void*)hWindow);
+        }
+
+        /// <summary>
+        /// Immediately begin a capture for the specified device/window combination.
+        /// </summary>
+        /// <param name="hDevice">a handle to the API ‘device’ object that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        /// <param name="hWindow">a handle to the platform window handle that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        [RenderDocApiVersion(1, 0)]
+        public static void StartFrameCapture(nint hDevice, nint hWindow)
+        {
+            if (Api is null) return;
+
+            Api->StartFrameCapture((void*)hDevice, (void*)hWindow);
+        }
+
+        /// <summary>
+        /// Immediately end an active capture for the specified device/window combination.
+        /// </summary>
+        /// <param name="hDevice">a handle to the API ‘device’ object that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        /// <param name="hWindow">a handle to the platform window handle that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        /// <returns>true if the capture succeeded; false otherwise.</returns>
+        [RenderDocApiVersion(1, 0)]
+        public static bool EndFrameCapture(nint hDevice, nint hWindow)
+        {
+            if (Api is null) return false;
+
+            return Api->EndFrameCapture((void*)hDevice, (void*)hWindow) != 0;
+        }
+
+        /// <summary>
+        /// Trigger multiple sequential frame captures as if the user had pressed one of the capture hotkeys before each frame.<br/>
+        /// The captures will be taken from the next frames presented to whichever window is considered current.<br/>
+        /// Each capture will be taken independently and saved to a separate file, with no reference to the other frames.
+        /// </summary>
+        /// <param name="numFrames">the number of frames to capture.</param>
+        /// <remarks>Requires RenderDoc API version 1.1</remarks>
+        [RenderDocApiVersion(1, 1)]
+        public static void TriggerMultiFrameCapture(uint numFrames)
+        {
+            if (Api is null) return;
+
+            AssertAtLeast(1, 1);
+            Api->TriggerMultiFrameCapture(numFrames);
+        }
+
+        /// <summary>
+        /// Adds an arbitrary comments field to the most recent capture,
+        /// which will then be displayed in the UI to anyone opening the capture.
+        /// <br/><br/>
+        /// This is equivalent to calling <see cref="SetCaptureFileComments"/> with a null first (fileName) parameter.
+        /// </summary>
+        /// <param name="comments">the comments to set in the capture file.</param>
+        /// <remarks>Requires RenderDoc API version 1.2</remarks>
+        public static void SetMostRecentCaptureFileComments(string comments)
+        {
+            if (Api is null) return;
+
+            AssertAtLeast(1, 2);
+
+            byte[] commentBytes = comments.ToNullTerminatedByteArray();
+
+            fixed (byte* pcomment = commentBytes)
+            {
+                Api->SetCaptureFileComments((byte*)nint.Zero, pcomment);
+            }
+        }
+
+        /// <summary>
+        /// Adds an arbitrary comments field to an existing capture on disk,
+        /// which will then be displayed in the UI to anyone opening the capture.
+        /// </summary>
+        /// <param name="fileName">the path to the capture file to set comments in. If this path is null or an empty string, the most recent capture file that has been created will be used.</param>
+        /// <param name="comments">the comments to set in the capture file.</param>
+        /// <remarks>Requires RenderDoc API version 1.2</remarks>
+        [RenderDocApiVersion(1, 2)]
+        public static void SetCaptureFileComments(string? fileName, string comments)
+        {
+            if (Api is null) return;
+
+            AssertAtLeast(1, 2);
+
+            byte[] commentBytes = comments.ToNullTerminatedByteArray();
+
+            fixed (byte* pcomment = commentBytes)
+            {
+                if (fileName is null)
+                {
+                    Api->SetCaptureFileComments((byte*)nint.Zero, pcomment);
+                }
+                else
+                {
+                    byte[] fileBytes = fileName.ToNullTerminatedByteArray();
+
+                    fixed (byte* pfile = fileBytes)
+                    {
+                        Api->SetCaptureFileComments(pfile, pcomment);
+                    }
+                }
+            }
+        }
+
+        /// <summary>
+        /// Similar to <see cref="EndFrameCapture"/>, but the capture contents will be discarded immediately, and not processed and written to disk.<br/>
+        /// This will be more efficient than <see cref="EndFrameCapture"/> if the frame capture is not needed.
+        /// </summary>
+        /// <param name="hDevice">a handle to the API ‘device’ object that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        /// <param name="hWindow">a handle to the platform window handle that will be set active. May be <see cref="nint.Zero"/> to wildcard match.</param>
+        /// <returns>true if the capture was discarded; false if there was an error or no capture was in progress.</returns>
+        /// <remarks>Requires RenderDoc API version 1.4</remarks>
+        [RenderDocApiVersion(1, 4)]
+        public static bool DiscardFrameCapture(nint hDevice, nint hWindow)
+        {
+            if (Api is null) return false;
+
+            AssertAtLeast(1, 4);
+            return Api->DiscardFrameCapture((void*)hDevice, (void*)hWindow) != 0;
+        }
+
+
+        /// <summary>
+        /// Requests that the currently connected replay UI raise its window to the top.<br/>
+        /// This is only possible if an instance of the replay UI is currently connected, otherwise this method does nothing.<br/>
+        /// This can be used in conjunction with <see cref="IsTargetControlConnected"/> and <see cref="LaunchReplayUI"/>,<br/> to intelligently handle showing the UI after making a capture.<br/><br/>
+        /// Given OS differences, it is not guaranteed that the UI will be successfully raised even if the request is passed on.<br/>
+        /// On some systems it may only be highlighted or otherwise indicated to the user.
+        /// </summary>
+        /// <returns>true if the request was passed onto the UI successfully; false if there is no UI connected or some other error occurred.</returns>
+        /// <remarks>Requires RenderDoc API version 1.5</remarks>
+        [RenderDocApiVersion(1, 5)]
+        public static bool ShowReplayUI()
+        {
+            if (Api is null) return false;
+
+            AssertAtLeast(1, 5);
+            return Api->ShowReplayUI() != 0;
+        }
+
+        /// <summary>
+        /// Sets a given title for the currently in-progress capture, which will be displayed in the UI.<br/>
+        /// This can be used either with a user-defined capture using a manual start and end,
+        /// or an automatic capture triggered by <see cref="TriggerCapture"/> or a keypress.<br/>
+        /// If multiple captures are ongoing at once, the title will be applied to the first capture to end only.<br/>
+        /// Any subsequent captures will not get any title unless the function is called again.
+        /// This function can only be called while a capture is in-progress,
+        /// after <see cref="StartFrameCapture"/> and before <see cref="EndFrameCapture"/>.<br/>
+        /// If it is called elsewhere it will have no effect.
+        /// If it is called multiple times within a capture, only the last title will have any effect.
+        /// </summary>
+        /// <param name="title">The title to set for the in-progress capture.</param>
+        /// <remarks>Requires RenderDoc API version 1.6</remarks>
+        [RenderDocApiVersion(1, 6)]
+        public static void SetCaptureTitle(string title)
+        {
+            if (Api is null) return;
+
+            AssertAtLeast(1, 6);
+            fixed (byte* ptr = title.ToNullTerminatedByteArray())
+                Api->SetCaptureTitle(ptr);
+        }
+
+        #region Dynamic Library loading
+
+        /// <summary>
+        /// Reload the internal RenderDoc API structure. Useful for manually refreshing <see cref="Api"/> while using process injection.
+        /// </summary>
+        /// <param name="ignoreAlreadyLoaded">Ignores the existing API function structure and overwrites it with a re-request.</param>
+        /// <param name="requiredVersion">The version of the RenderDoc API required by your application.</param>
+        public static void ReloadApi(bool ignoreAlreadyLoaded = false, RenderDocVersion? requiredVersion = null)
+        {
+            if (_loaded && !ignoreAlreadyLoaded)
+                return;
+
+            lock (typeof(RenderDoc))
+            {
+                // Prevent double loads.
+                if (_loaded && !ignoreAlreadyLoaded)
+                    return;
+
+                if (requiredVersion.HasValue)
+                    MinimumRequired = requiredVersion.Value;
+
+                _loaded = true;
+                _api = GetApi(MinimumRequired);
+
+                if (_api != null)
+                    AssertAtLeast(MinimumRequired);
+            }
+        }
+
+        private static RenderDocApi* _api = null;
+        private static bool _loaded;
+
+        private static RenderDocApi* Api
+        {
+            get
+            {
+                ReloadApi();
+                return _api;
+            }
+        }
+
+        private static readonly Regex _dynamicLibraryPattern = RenderDocApiDynamicLibraryRegex();
+
+        private static RenderDocApi* GetApi(RenderDocVersion minimumRequired = RenderDocVersion.Version_1_0_0)
+        {
+            foreach (ProcessModule module in Process.GetCurrentProcess().Modules)
+            {
+                string moduleName = module.FileName ?? string.Empty;
+
+                if (!_dynamicLibraryPattern.IsMatch(moduleName))
+                    continue;
+
+                if (!NativeLibrary.TryLoad(moduleName, out nint moduleHandle))
+                    return null;
+
+                if (!NativeLibrary.TryGetExport(moduleHandle, "RENDERDOC_GetAPI", out nint procAddress))
+                    return null;
+
+                var RENDERDOC_GetApi = (delegate* unmanaged[Cdecl]<RenderDocVersion, RenderDocApi**, int>)procAddress;
+
+                RenderDocApi* api;
+                return RENDERDOC_GetApi(minimumRequired, &api) != 0 ? api : null;
+            }
+
+            return null;
+        }
+
+        private static void AssertAtLeast(RenderDocVersion rdv, [CallerMemberName] string callee = "")
+        {
+            Version ver = rdv.SystemVersion;
+            AssertAtLeast(ver.Major, ver.Minor, ver.Build, callee);
+        }
+
+        private static void AssertAtLeast(int major, int minor, int patch = 0, [CallerMemberName] string callee = "")
+        {
+            if (!AssertVersionEnabled)
+                return;
+
+            if (Version!.Major < major)
+                goto fail;
+
+            if (Version.Major > major)
+                goto success;
+            if (Version.Minor < minor)
+                goto fail;
+            if (Version.Minor > minor)
+                goto success;
+            if (Version.Build < patch)
+                goto fail;
+
+            success:
+            return;
+
+            fail:
+            Version minVersion =
+                typeof(RenderDoc).GetMethod(callee)!.GetCustomAttribute<RenderDocApiVersionAttribute>()!.MinVersion;
+            throw new NotSupportedException(
+                $"This API was introduced in RenderDoc API {minVersion}. Current API version is {Version}.");
+        }
+
+        private static byte[] ToNullTerminatedByteArray(this string str, Encoding? encoding = null)
+        {
+            encoding ??= Encoding.UTF8;
+
+            return encoding.GetBytes(str + '\0');
+        }
+
+        [GeneratedRegex(@"(lib)?renderdoc(\.dll|\.so|\.dylib)(\.\d+)?",
+            RegexOptions.IgnoreCase | RegexOptions.CultureInvariant)]
+        private static partial Regex RenderDocApiDynamicLibraryRegex();
+
+        #endregion
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/RenderDocApi.cs

@@ -0,0 +1,51 @@
+namespace Ryujinx.Graphics.RenderDocApi
+{
+#pragma warning disable CS0649
+    internal unsafe struct RenderDocApi
+    {
+        public delegate* unmanaged[Cdecl]<int*, int*, int*, void> GetApiVersion;
+
+        public delegate* unmanaged[Cdecl]<CaptureOption, uint, int> SetCaptureOptionU32;
+        public delegate* unmanaged[Cdecl]<CaptureOption, float, int> SetCaptureOptionF32;
+        public delegate* unmanaged[Cdecl]<CaptureOption, uint> GetCaptureOptionU32;
+        public delegate* unmanaged[Cdecl]<CaptureOption, float> GetCaptureOptionF32;
+
+        public delegate* unmanaged[Cdecl]<InputButton*, int, void> SetFocusToggleKeys;
+        public delegate* unmanaged[Cdecl]<InputButton*, int, void> SetCaptureKeys;
+
+        public delegate* unmanaged[Cdecl]<OverlayBits> GetOverlayBits;
+        public delegate* unmanaged[Cdecl]<OverlayBits, OverlayBits, void> MaskOverlayBits;
+
+        public delegate* unmanaged[Cdecl]<void> RemoveHooks;
+        public delegate* unmanaged[Cdecl]<void> UnloadCrashHandler;
+        public delegate* unmanaged[Cdecl]<byte*, void> SetCaptureFilePathTemplate;
+        public delegate* unmanaged[Cdecl]<byte*> GetCaptureFilePathTemplate;
+
+        public delegate* unmanaged[Cdecl]<int> GetNumCaptures;
+        public delegate* unmanaged[Cdecl]<int, byte*, int*, long*, uint> GetCapture;
+        public delegate* unmanaged[Cdecl]<void> TriggerCapture;
+        public delegate* unmanaged[Cdecl]<uint> IsTargetControlConnected;
+        public delegate* unmanaged[Cdecl]<uint, byte*, uint> LaunchReplayUI;
+
+        public delegate* unmanaged[Cdecl]<void*, void*, void> SetActiveWindow;
+        public delegate* unmanaged[Cdecl]<void*, void*, void> StartFrameCapture;
+        public delegate* unmanaged[Cdecl]<uint> IsFrameCapturing;
+        public delegate* unmanaged[Cdecl]<void*, void*, uint> EndFrameCapture;
+
+        // 1.1
+        public delegate* unmanaged[Cdecl]<uint, void> TriggerMultiFrameCapture;
+
+        // 1.2
+        public delegate* unmanaged[Cdecl]<byte*, byte*, void> SetCaptureFileComments;
+
+        // 1.3
+        public delegate* unmanaged[Cdecl]<void*, void*, uint> DiscardFrameCapture;
+
+        // 1.5
+        public delegate* unmanaged[Cdecl]<uint> ShowReplayUI;
+
+        // 1.6
+        public delegate* unmanaged[Cdecl]<byte*, void> SetCaptureTitle;
+    }
+#pragma warning restore CS0649
+}

src/Ryujinx.Graphics.RenderDocApi/RenderDocApiVersionAttribute.cs

@@ -0,0 +1,16 @@
+
+using System;
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    [AttributeUsage(AttributeTargets.Method | AttributeTargets.Property)]
+    public sealed class RenderDocApiVersionAttribute : Attribute
+    {
+        public Version MinVersion { get; }
+
+        public RenderDocApiVersionAttribute(int major, int minor, int patch = 0)
+        {
+            MinVersion = new Version(major, minor, patch);
+        }
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/RenderDocVersion.cs

@@ -0,0 +1,47 @@
+using System;
+
+namespace Ryujinx.Graphics.RenderDocApi
+{
+    public enum RenderDocVersion
+    {
+        Version_1_0_0 = 10000,
+        Version_1_0_1 = 10001,
+        Version_1_0_2 = 10002,
+        Version_1_1_0 = 10100,
+        Version_1_1_1 = 10101,
+        Version_1_1_2 = 10102,
+        Version_1_2_0 = 10200,
+        Version_1_3_0 = 10300,
+        Version_1_4_0 = 10400,
+        Version_1_4_1 = 10401,
+        Version_1_4_2 = 10402,
+        Version_1_5_0 = 10500,
+        Version_1_6_0 = 10600,
+    }
+    
+    public static partial class Helpers
+    {
+        extension(RenderDocVersion rdv)
+        {
+            public Version SystemVersion
+            {
+                get
+                {
+                    int i = (int)rdv;
+                    return new (i / 10000, (i % 10000) / 100, i % 100);
+                }
+            }
+        }
+
+        extension(Version sv)
+        {
+            public RenderDocVersion RenderDocVersion
+            {
+                get
+                {
+                    return (RenderDocVersion)(sv.Major * 10000 + sv.Minor * 100 + sv.Build);
+                }
+            }
+        }
+    }
+}

src/Ryujinx.Graphics.RenderDocApi/Ryujinx.Graphics.RenderDocApi.csproj

@@ -0,0 +1,9 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+    <PropertyGroup>
+        <TargetFramework>net10.0</TargetFramework>
+        <ImplicitUsings>disable</ImplicitUsings>
+        <Nullable>enable</Nullable>
+        <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+    </PropertyGroup>
+</Project>

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

@@ -1790,7 +1790,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                     {
                         (_, varType) = IoMap.GetSpirvBuiltIn(ioVariable);
 
-                        if (IoMap.IsPerVertexBuiltIn(ioVariable))
+                        if (context.Definitions.Stage.IsVtg() && IoMap.IsPerVertexBuiltIn(ioVariable))
                         {
                             perVertexBuiltIn = ioVariable;
                             ioVariable = IoVariable.Position;

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

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

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

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

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

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

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

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

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

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

src/Ryujinx.Graphics.Shader/Ryujinx.Graphics.Shader.csproj

@@ -1,8 +1,6 @@
 <Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs

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

src/Ryujinx.Graphics.Shader/SupportBuffer.cs

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

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

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

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

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

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

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

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

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

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

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

src/Ryujinx.Graphics.Texture/Ryujinx.Graphics.Texture.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Vic/Ryujinx.Graphics.Vic.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Video/Ryujinx.Graphics.Video.csproj

@@ -1,8 +1,6 @@
 <Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Vulkan/BackgroundResources.cs

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

src/Ryujinx.Graphics.Vulkan/CommandBufferPool.cs

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

src/Ryujinx.Graphics.Vulkan/FenceHolder.cs

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

src/Ryujinx.Graphics.Vulkan/Helpers.cs

@@ -0,0 +1,32 @@
+using Silk.NET.Vulkan;
+using System.Runtime.CompilerServices;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    public static class Helpers
+    {
+        extension(Vk api)
+        {
+            /// <summary>
+            ///     C# implementation of the RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE macro from the RenderDoc API header, since we cannot use macros from C#.
+            /// </summary>
+            /// <returns>The dispatch table pointer, which sits as the first pointer-sized object in the memory pointed to by the <see cref="Vk"/>'s <see cref="Instance"/> pointer.</returns>
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            public unsafe void* GetRenderDocDevicePointer() =>
+                api.CurrentInstance is not null
+                    ? api.CurrentInstance.Value.GetRenderDocDevicePointer()
+                    : null;
+        }
+
+        extension(Instance instance)
+        {
+            /// <summary>
+            ///     C# implementation of the RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE macro from the RenderDoc API header, since we cannot use macros from C#.
+            /// </summary>
+            /// <returns>The dispatch table pointer, which sits as the first pointer-sized object in the memory pointed to by the <see cref="Instance"/>'s pointer.</returns>
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            public unsafe void* GetRenderDocDevicePointer()
+                => (*((void**)(instance.Handle)));
+        }
+    }
+}

src/Ryujinx.Graphics.Vulkan/Ryujinx.Graphics.Vulkan.csproj

@@ -2,8 +2,7 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
+    <Configurations>Debug;Release</Configurations>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.Vulkan/ShaderCollection.cs

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

src/Ryujinx.Graphics.Vulkan/ShaderCompilationQueue.cs

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

src/Ryujinx.Graphics.Vulkan/ShaderCompilationRequest.cs

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

src/Ryujinx.Graphics.Vulkan/StagingBuffer.cs

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

src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

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

src/Ryujinx.HLE.Generators/Ryujinx.HLE.Generators.csproj

@@ -6,8 +6,6 @@
         <EmitCompilerGeneratedFiles>true</EmitCompilerGeneratedFiles>
         <CompilerGeneratedFilesOutputPath>Generated</CompilerGeneratedFilesOutputPath>
         <IsRoslynComponent>true</IsRoslynComponent>
-        <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-        <Platforms>AnyCPU</Platforms>
     </PropertyGroup>
 
     <ItemGroup>

src/Ryujinx.HLE/HOS/Applets/Error/ErrorApplet.cs

@@ -166,13 +166,15 @@ namespace Ryujinx.HLE.HOS.Applets.Error
 
             string[] buttons = GetButtonsText(module, description, "DlgBtn");
 
-            bool showDetails = _horizon.Device.UIHandler.DisplayErrorAppletDialog($"Error Code: {module}-{description:0000}", "\n" + message, buttons);
+            (uint Module, uint Description) errorCodeTuple = (module, uint.Parse(description.ToString("0000")));
+
+            bool showDetails = _horizon.Device.UIHandler.DisplayErrorAppletDialog($"Error Code: {module}-{description:0000}", "\n" + message, buttons, errorCodeTuple);
             if (showDetails)
             {
                 message = GetMessageText(module, description, "FlvMsg");
                 buttons = GetButtonsText(module, description, "FlvBtn");
 
-                _horizon.Device.UIHandler.DisplayErrorAppletDialog($"Details: {module}-{description:0000}", "\n" + message, buttons);
+                _horizon.Device.UIHandler.DisplayErrorAppletDialog($"Details: {module}-{description:0000}", "\n" + message, buttons, errorCodeTuple);
             }
         }
 

src/Ryujinx.HLE/HOS/Applets/PlayerSelect/PlayerSelectApplet.cs

@@ -27,9 +27,19 @@ namespace Ryujinx.HLE.HOS.Applets
             _normalSession = normalSession;
             _interactiveSession = interactiveSession;
 
-            // TODO(jduncanator): Parse PlayerSelectConfig from input data
-            _normalSession.Push(BuildResponse());
-
+            UserProfile selected = _system.Device.UIHandler.ShowPlayerSelectDialog();
+            if (selected == null)
+            {
+                _normalSession.Push(BuildResponse());
+            }
+            else if (selected.UserId == new UserId("00000000000000000000000000000080"))
+            {
+                _normalSession.Push(BuildGuestResponse());
+            }
+            else
+            {
+                _normalSession.Push(BuildResponse(selected));
+            }
             AppletStateChanged?.Invoke(this, null);
 
             _system.ReturnFocus();
@@ -37,16 +47,34 @@ namespace Ryujinx.HLE.HOS.Applets
             return ResultCode.Success;
         }
 
-        private byte[] BuildResponse()
+        private byte[] BuildResponse(UserProfile selectedUser)
         {
-            UserProfile currentUser = _system.AccountManager.LastOpenedUser;
-
             using MemoryStream stream = MemoryStreamManager.Shared.GetStream();
             using BinaryWriter writer = new(stream);
 
             writer.Write((ulong)PlayerSelectResult.Success);
 
-            currentUser.UserId.Write(writer);
+            selectedUser.UserId.Write(writer);
+
+            return stream.ToArray();
+        }
+
+        private byte[] BuildGuestResponse()
+        {
+            using MemoryStream stream = MemoryStreamManager.Shared.GetStream();
+            using BinaryWriter writer = new(stream);
+
+            writer.Write(new byte());
+
+            return stream.ToArray();
+        }
+        
+        private byte[] BuildResponse()
+        {
+            using MemoryStream stream = MemoryStreamManager.Shared.GetStream();
+            using BinaryWriter writer = new(stream);
+
+            writer.Write((ulong)PlayerSelectResult.Failure);
 
             return stream.ToArray();
         }

src/Ryujinx.HLE/HOS/Kernel/Common/KSystemControl.cs

@@ -26,12 +26,12 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
             {
                 MemoryArrange.MemoryArrange4GiB or
                 MemoryArrange.MemoryArrange4GiBSystemDev or
-                MemoryArrange.MemoryArrange6GiBAppletDev => 3152 * MiB,
+                MemoryArrange.MemoryArrange6GiBAppletDev => 3173 * MiB,
                 MemoryArrange.MemoryArrange4GiBAppletDev => 2048 * MiB,
-                MemoryArrange.MemoryArrange6GiB => 4783 * MiB,
-                MemoryArrange.MemoryArrange8GiB => 6831 * MiB,
-                MemoryArrange.MemoryArrange10GiB => 8879 * MiB,
-                MemoryArrange.MemoryArrange12GiB => 10927 * MiB,
+                MemoryArrange.MemoryArrange6GiB => 4804 * MiB,
+                MemoryArrange.MemoryArrange8GiB => 6852 * MiB,
+                MemoryArrange.MemoryArrange10GiB => 8900 * MiB,
+                MemoryArrange.MemoryArrange12GiB => 10948 * MiB,
                 _ => throw new ArgumentException($"Invalid memory arrange \"{arrange}\"."),
             };
         }

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

@@ -15,7 +15,6 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
         private readonly Dictionary<ulong, List<KThread>> _condVarThreads;
         private readonly Dictionary<ulong, List<KThread>> _arbiterThreads;
-        private readonly ByDynamicPriority _byDynamicPriority;
 
         public KAddressArbiter(KernelContext context)
         {
@@ -23,7 +22,6 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
             _condVarThreads = [];
             _arbiterThreads = [];
-            _byDynamicPriority = new ByDynamicPriority();
         }
 
         public Result ArbitrateLock(int ownerHandle, ulong mutexAddress, int requesterHandle)
@@ -142,14 +140,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
             if (_condVarThreads.TryGetValue(condVarAddress, out List<KThread> threads))
             {
-                int i = 0;
+                int i = FindDynamicPriorityFifoInsertionIndex(threads, currentThread);
+
 
-                if (threads.Count > 0)
-                {
-                    i = threads.BinarySearch(currentThread, _byDynamicPriority);
-                    if (i < 0) i = ~i;
-                }
-                
                 threads.Insert(i, currentThread);
             }
             else
@@ -332,14 +325,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                 
                 if (_arbiterThreads.TryGetValue(address, out List<KThread> threads))
                 {
-                    int i = 0;
+                    int i = FindDynamicPriorityFifoInsertionIndex(threads, currentThread);
+
 
-                    if (threads.Count > 0)
-                    {
-                        i = threads.BinarySearch(currentThread, _byDynamicPriority);
-                        if (i < 0) i = ~i;
-                    }
-                
                     threads.Insert(i, currentThread);
                 }
                 else
@@ -424,14 +412,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                 
                 if (_arbiterThreads.TryGetValue(address, out List<KThread> threads))
                 {
-                    int i = 0;
+                    int i = FindDynamicPriorityFifoInsertionIndex(threads, currentThread);
+
 
-                    if (threads.Count > 0)
-                    {
-                        i = threads.BinarySearch(currentThread, _byDynamicPriority);
-                        if (i < 0) i = ~i;
-                    }
-                
                     threads.Insert(i, currentThread);
                 }
                 else
@@ -627,12 +610,28 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
             return validCount;
         }
         
-        private class ByDynamicPriority : IComparer<KThread>
+        private static int FindDynamicPriorityFifoInsertionIndex(List<KThread> threads, KThread currentThread)
         {
-            public int Compare(KThread x, KThread y)
+            int low = 0;
+            int high = threads.Count;
+
+            // Lower numeric values represent higher priorities. Use upper-bound insertion
+            // to preserve FIFO order among waiters with the same dynamic priority.
+            while (low < high)
             {
-                return x!.DynamicPriority.CompareTo(y!.DynamicPriority);
+                int middle = low + ((high - low) >> 1);
+
+                if (threads[middle].DynamicPriority <= currentThread.DynamicPriority)
+                {
+                    low = middle + 1;
+                }
+                else
+                {
+                    high = middle;
+                }
             }
+
+            return low;
         }
     }
 }

src/Ryujinx.HLE/HOS/ModLoader.cs

@@ -783,8 +783,28 @@ namespace Ryujinx.HLE.HOS
             }
 
             List<Cheat> cheats = mods.Cheats;
-            Dictionary<string, ulong> processExes = tamperInfo.BuildIds.Zip(tamperInfo.CodeAddresses, (k, v) => new { k, v })
-                .ToDictionary(x => x.k[..Math.Min(Cheat.CheatIdSize, x.k.Length)], x => x.v);
+            Dictionary<string, ulong> processExes = new();
+
+            foreach ((string buildId, ulong codeAddress) in tamperInfo.BuildIds.Zip(tamperInfo.CodeAddresses))
+            {
+                string normalizedBuildId = buildId[..Math.Min(Cheat.CheatIdSize, buildId.Length)];
+
+                if (processExes.TryGetValue(normalizedBuildId, out ulong existingAddress))
+                {
+                    if (existingAddress != codeAddress)
+                    {
+                        Logger.Warning?.Print(
+                            LogClass.ModLoader,
+                            $"Duplicate BuildId prefix '{normalizedBuildId}' has different code addresses. " +
+                            $"Existing: 0x{existingAddress:X}, duplicate: 0x{codeAddress:X}. " +
+                            $"Keeping the first one.");
+                    }
+
+                    continue;
+                }
+
+                processExes.Add(normalizedBuildId, codeAddress);
+            }
 
             foreach (Cheat cheat in cheats)
             {

src/Ryujinx.HLE/HOS/Services/Account/Acc/IAccountServiceForApplication.cs

@@ -71,7 +71,14 @@ namespace Ryujinx.HLE.HOS.Services.Account.Acc
             return _applicationServiceServer.IsUserRegistrationRequestPermitted(context);
         }
 
-        [CommandCmif(51)]
+        [CommandCmif(51)] // 1.0.0 – 18.1.0
+        // TrySelectUserWithoutInteraction(bool) -> nn::account::Uid
+        public ResultCode TrySelectUserWithoutInteractionDeprecated(ServiceCtx context)
+        {
+            return _applicationServiceServer.TrySelectUserWithoutInteraction(context);
+        }
+
+        [CommandCmif(52)] // 19.0.0+
         // TrySelectUserWithoutInteraction(bool) -> nn::account::Uid
         public ResultCode TrySelectUserWithoutInteraction(ServiceCtx context)
         {

src/Ryujinx.HLE/HOS/Services/Hid/HidDevices/NpadDevices.cs

@@ -23,7 +23,7 @@ namespace Ryujinx.HLE.HOS.Services.Hid
         private readonly bool[] _supportedPlayers;
         private VibrationValue _neutralVibrationValue = new()
         {
-            AmplitudeLow = 0f,
+            AmplitudeLow = 0.01f,
             FrequencyLow = 160f,
             AmplitudeHigh = 0f,
             FrequencyHigh = 320f,

src/Ryujinx.HLE/Ryujinx.HLE.csproj

@@ -2,8 +2,6 @@
 
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>
@@ -61,6 +59,7 @@
     <EmbeddedResource Include="HOS\Applets\SoftwareKeyboard\Resources\Icon_BtnB.png" />
     <EmbeddedResource Include="HOS\Applets\SoftwareKeyboard\Resources\Icon_KeyF6.png" />
     <EmbeddedResource Include="HOS\Services\Account\Acc\DefaultUserImage.jpg" />
+    <EmbeddedResource Include="HOS\Services\Account\Acc\GuestUserImage.jpg" />
   </ItemGroup>
 
 </Project>

src/Ryujinx.HLE/UI/IHostUIHandler.cs

@@ -1,4 +1,5 @@
 using Ryujinx.HLE.HOS.Applets;
+using Ryujinx.HLE.HOS.Services.Account.Acc;
 using Ryujinx.HLE.HOS.Services.Am.AppletOE.ApplicationProxyService.ApplicationProxy.Types;
 
 namespace Ryujinx.HLE.UI
@@ -48,7 +49,8 @@ namespace Ryujinx.HLE.UI
         /// Displays a Message Dialog box specific to Error Applet and blocks until it is closed.
         /// </summary>
         /// <returns>False when OK is pressed, True when another button (Details) is pressed.</returns>
-        bool DisplayErrorAppletDialog(string title, string message, string[] buttonsText);
+        // ReSharper disable once UnusedParameter.Global
+        bool DisplayErrorAppletDialog(string title, string message, string[] buttonsText, (uint Module, uint Description)? errorCode = null);
 
         /// <summary>
         /// Creates a handler to process keyboard inputs into text strings.
@@ -65,5 +67,10 @@ namespace Ryujinx.HLE.UI
         /// Takes a screenshot from the current renderer and saves it in the screenshots folder.
         /// </summary>
         void TakeScreenshot();
+
+        /// <summary>
+        /// Displays the player select dialog and returns the selected profile.
+        /// </summary>
+        UserProfile ShowPlayerSelectDialog();
     }
 }

src/Ryujinx.Horizon.Common/Ryujinx.Horizon.Common.csproj

@@ -1,8 +1,6 @@
 <Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>

src/Ryujinx.Horizon.Generators/Ryujinx.Horizon.Generators.csproj

@@ -4,8 +4,6 @@
     <TargetFramework>netstandard2.0</TargetFramework>
     <EnforceExtendedAnalyzerRules>true</EnforceExtendedAnalyzerRules>
     <DefaultItemExcludes>$(DefaultItemExcludes);._*</DefaultItemExcludes>
-    <Configurations>Debug;Release;Debug AOT;Release AOT</Configurations>
-    <Platforms>AnyCPU</Platforms>
   </PropertyGroup>
 
   <ItemGroup>