misc: Implement address space size workarounds (#5191)

* misc: Implement address space size workarounds

This adds code to support userland with less than 39 bits of address
space available by testing reserving multiple sizes and reducing
guess address space when needed.

This is required for ARM64 support when the kernel is
configured to use 63..39 bits for kernel space.(meaning only 38 bits is available to userland)

* Address comments

* Fix 32 bits address space support and address more comments

.github/assign/audio.yml

@@ -0,0 +1,8 @@
+addReviewers: true
+
+reviewers:
+  - marysaka
+
+filterLabels:
+  include:
+    - audio

.github/assign/cpu.yml

@@ -0,0 +1,11 @@
+addReviewers: true
+
+reviewers:
+  - gdkchan
+  - riperiperi
+  - marysaka
+  - LDj3SNuD
+
+filterLabels:
+  include:
+    - cpu

.github/assign/global.yml

@@ -0,0 +1,4 @@
+addReviewers: true
+
+reviewers:
+  - Ryujinx/developers

.github/assign/gpu.yml

@@ -0,0 +1,10 @@
+addReviewers: true
+
+reviewers:
+  - gdkchan
+  - riperiperi
+  - marysaka
+
+filterLabels:
+  include:
+    - gpu

.github/assign/gui.yml

@@ -0,0 +1,11 @@
+addReviewers: true
+
+reviewers:
+  - Ack77
+  - emmauss
+  - TSRBerry
+  - marysaka
+
+filterLabels:
+  include:
+    - gui

.github/assign/horizon.yml

@@ -0,0 +1,11 @@
+addReviewers: true
+
+reviewers:
+  - gdkchan
+  - Ack77
+  - marysaka
+  - TSRBerry
+
+filterLabels:
+  include:
+    - horizon

.github/assign/infra.yml

@@ -0,0 +1,9 @@
+addReviewers: true
+
+reviewers:
+  - marysaka
+  - TSRBerry
+
+filterLabels:
+  include:
+    - infra

.github/labeler.yml

@@ -0,0 +1,33 @@
+audio: 'src/Ryujinx.Audio*/**'
+
+cpu:
+  - 'src/ARMeilleure/**'
+  - 'src/Ryujinx.Cpu/**'
+  - 'src/Ryujinx.Memory/**'
+
+gpu:
+  - 'src/Ryujinx.Graphics.*/**'
+  - 'src/Spv.Generator/**'
+  - 'src/Ryujinx.ShaderTools/**'
+
+'graphics-backend:opengl': 'src/Ryujinx.Graphics.OpenGL/**'
+'graphics-backend:vulkan':
+  - 'src/Ryujinx.Graphics.Vulkan/**'
+  - 'src/Spv.Generator/**'
+
+gui:
+  - 'src/Ryujinx/**'
+  - 'src/Ryujinx.Ui.Common/**'
+  - 'src/Ryujinx.Ui.LocaleGenerator/**'
+  - 'src/Ryujinx.Ava/**'
+
+horizon:
+  - 'src/Ryujinx.HLE/**'
+  - 'src/Ryujinx.Horizon*/**'
+
+kernel: 'src/Ryujinx.HLE/HOS/Kernel/**'
+
+infra:
+  - '.github/**'
+  - 'distribution/**'
+  - 'Directory.Packages.props'

.github/workflows/build.yml

@@ -3,19 +3,13 @@ name: Build job
 on:
   workflow_dispatch:
     inputs: {}
-  #push:
-  #  branches: [ master ]
-  #  paths-ignore:
-  #    - '.github/*'
-  #    - '.github/ISSUE_TEMPLATE/**'
-  #    - '*.yml'
-  #    - 'README.md'
   pull_request:
     branches: [ master ]
     paths-ignore:
-      - '.github/*'
+      - '.github/**'
-      - '.github/ISSUE_TEMPLATE/**'
       - '*.yml'
+      - '*.json'
+      - '*.config'
       - 'README.md'
 
 concurrency:

.github/workflows/pr_triage.yml

@@ -0,0 +1,54 @@
+name: "Pull Request Triage"
+on:
+  pull_request_target:
+    types: [opened, ready_for_review]
+
+jobs:
+  triage:
+    permissions:
+      contents: read
+      pull-requests: write
+
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Update labels based on changes
+        uses: actions/labeler@v4
+        with:
+          sync-labels: true
+          dot: true
+
+      - name: Auto Assign [Audio]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/audio.yml'
+
+      - name: Auto Assign [CPU]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/cpu.yml'
+
+      - name: Auto Assign [GPU]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/gpu.yml'
+
+      - name: Auto Assign [GUI]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/gui.yml'
+
+      - name: Auto Assign [Horizon]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/horizon.yml'
+
+      - name: Auto Assign [Infra]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/infra.yml'
+
+      - name: Auto Assign [Global]
+        uses: kentaro-m/auto-assign-action@v1.2.5
+        with:
+          configuration-path: '.github/assign/global.yml'

.github/workflows/release.yml

@@ -6,9 +6,10 @@ on:
   push:
     branches: [ master ]
     paths-ignore:
-      - '.github/*'
+      - '.github/**'
-      - '.github/ISSUE_TEMPLATE/**'
       - '*.yml'
+      - '*.json'
+      - '*.config'
       - 'README.md'
 
 concurrency: release

Directory.Packages.props

@@ -21,7 +21,7 @@
     <PackageVersion Include="LibHac" Version="0.18.0" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.5.0" />
-    <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.6.1" />
+    <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.6.2" />
     <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="2.3.2" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />
     <PackageVersion Include="NUnit" Version="3.13.3" />
@@ -44,9 +44,9 @@
     <PackageVersion Include="SixLabors.ImageSharp.Drawing" Version="1.0.0-beta11" />
     <PackageVersion Include="SPB" Version="0.0.4-build28" />
     <PackageVersion Include="System.Drawing.Common" Version="7.0.0" />
-    <PackageVersion Include="System.IdentityModel.Tokens.Jwt" Version="6.30.1" />
+    <PackageVersion Include="System.IdentityModel.Tokens.Jwt" Version="6.31.0" />
     <PackageVersion Include="System.IO.Hashing" Version="7.0.0" />
-    <PackageVersion Include="System.Management" Version="7.0.1" />
+    <PackageVersion Include="System.Management" Version="7.0.2" />
     <PackageVersion Include="UnicornEngine.Unicorn" Version="2.0.2-rc1-fb78016" />
     <PackageVersion Include="XamlNameReferenceGenerator" Version="1.6.1" />
   </ItemGroup>

src/ARMeilleure/CodeGen/Arm64/CodeGenerator.cs

@@ -168,8 +168,6 @@ namespace ARMeilleure.CodeGen.Arm64
 
             Logger.StartPass(PassName.CodeGeneration);
 
-            //Console.Error.WriteLine(IRDumper.GetDump(cfg));
-
             bool relocatable = (cctx.Options & CompilerOptions.Relocatable) != 0;
 
             CodeGenContext context = new(allocResult, maxCallArgs, cfg.Blocks.Count, relocatable);

src/ARMeilleure/CodeGen/Arm64/CodeGeneratorIntrinsic.cs

@@ -179,6 +179,35 @@ namespace ARMeilleure.CodeGen.Arm64
                         (uint)operation.GetSource(2).AsInt32());
                     break;
 
+                case IntrinsicType.Vector128Unary:
+                    GenerateVectorUnary(
+                        context,
+                        1,
+                        0,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.Vector128Binary:
+                    GenerateVectorBinary(
+                        context,
+                        1,
+                        0,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.Vector128BinaryRd:
+                    GenerateVectorUnary(
+                        context,
+                        1,
+                        0,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1));
+                    break;
+
                 case IntrinsicType.VectorUnary:
                     GenerateVectorUnary(
                         context,

src/ARMeilleure/CodeGen/Arm64/IntrinsicTable.cs

@@ -19,8 +19,8 @@ namespace ARMeilleure.CodeGen.Arm64
             Add(Intrinsic.Arm64AddvV,         new IntrinsicInfo(0x0e31b800u, IntrinsicType.VectorUnary));
             Add(Intrinsic.Arm64AddS,          new IntrinsicInfo(0x5e208400u, IntrinsicType.ScalarBinary));
             Add(Intrinsic.Arm64AddV,          new IntrinsicInfo(0x0e208400u, IntrinsicType.VectorBinary));
-            Add(Intrinsic.Arm64AesdV,         new IntrinsicInfo(0x4e285800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AesdV,         new IntrinsicInfo(0x4e285800u, IntrinsicType.Vector128BinaryRd));
-            Add(Intrinsic.Arm64AeseV,         new IntrinsicInfo(0x4e284800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AeseV,         new IntrinsicInfo(0x4e284800u, IntrinsicType.Vector128BinaryRd));
             Add(Intrinsic.Arm64AesimcV,       new IntrinsicInfo(0x4e287800u, IntrinsicType.Vector128Unary));
             Add(Intrinsic.Arm64AesmcV,        new IntrinsicInfo(0x4e286800u, IntrinsicType.Vector128Unary));
             Add(Intrinsic.Arm64AndV,          new IntrinsicInfo(0x0e201c00u, IntrinsicType.VectorBinaryBitwise));

src/ARMeilleure/CodeGen/Arm64/IntrinsicType.cs

@@ -23,6 +23,10 @@ namespace ARMeilleure.CodeGen.Arm64
         ScalarTernaryShlRd,
         ScalarTernaryShrRd,
 
+        Vector128Unary,
+        Vector128Binary,
+        Vector128BinaryRd,
+
         VectorUnary,
         VectorUnaryBitwise,
         VectorUnaryByElem,
@@ -50,9 +54,6 @@ namespace ARMeilleure.CodeGen.Arm64
         VectorTernaryShlRd,
         VectorTernaryShrRd,
 
-        Vector128Unary,
-        Vector128Binary,
-
         GetRegister,
         SetRegister
     }

src/ARMeilleure/CodeGen/Arm64/PreAllocator.cs

@@ -746,6 +746,7 @@ namespace ARMeilleure.CodeGen.Arm64
                    info.Type == IntrinsicType.ScalarTernaryFPRdByElem ||
                    info.Type == IntrinsicType.ScalarTernaryShlRd ||
                    info.Type == IntrinsicType.ScalarTernaryShrRd ||
+                   info.Type == IntrinsicType.Vector128BinaryRd ||
                    info.Type == IntrinsicType.VectorBinaryRd ||
                    info.Type == IntrinsicType.VectorInsertByElem ||
                    info.Type == IntrinsicType.VectorTernaryRd ||

src/ARMeilleure/Decoders/OpCode32SimdSel.cs

@@ -13,7 +13,7 @@
         }
     }
 
-    enum OpCode32SimdSelMode : int
+    enum OpCode32SimdSelMode
     {
         Eq = 0,
         Vs,

src/ARMeilleure/Instructions/InstEmitSimdCrypto.cs

@@ -17,7 +17,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesdV, d, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesdeclast, context.AddIntrinsic(Intrinsic.X86Xorpd, d, n), context.VectorZero());
             }
@@ -38,7 +42,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AeseV, d, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesenclast, context.AddIntrinsic(Intrinsic.X86Xorpd, d, n), context.VectorZero());
             }
@@ -58,7 +66,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesimcV, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesimc, n);
             }
@@ -78,7 +90,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesmcV, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 Operand roundKey = context.VectorZero();
 

src/ARMeilleure/Instructions/InstEmitSimdCrypto32.cs

@@ -17,7 +17,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesdV, d, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesdeclast, context.AddIntrinsic(Intrinsic.X86Xorpd, d, n), context.VectorZero());
             }
@@ -38,7 +42,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AeseV, d, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesenclast, context.AddIntrinsic(Intrinsic.X86Xorpd, d, n), context.VectorZero());
             }
@@ -58,7 +66,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesimcV, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 res = context.AddIntrinsic(Intrinsic.X86Aesimc, n);
             }
@@ -78,7 +90,11 @@ namespace ARMeilleure.Instructions
 
             Operand res;
 
-            if (Optimizations.UseAesni)
+            if (Optimizations.UseArm64Aes)
+            {
+                res = context.AddIntrinsic(Intrinsic.Arm64AesmcV, n);
+            }
+            else if (Optimizations.UseAesni)
             {
                 Operand roundKey = context.VectorZero();
 

src/ARMeilleure/Instructions/InstEmitSimdCvt32.cs

@@ -165,7 +165,7 @@ namespace ARMeilleure.Instructions
                     {
                         Operand m = GetVecA32(op.Vm >> 1);
 
-                        Operand toConvert = InstEmitSimdHelper32Arm64.EmitExtractScalar(context, m, op.Vm, doubleSize);
+                        Operand toConvert = InstEmitSimdHelper32Arm64.EmitExtractScalar(context, m, op.Vm, true);
 
                         Intrinsic inst = (unsigned ? Intrinsic.Arm64FcvtzuGp : Intrinsic.Arm64FcvtzsGp) | Intrinsic.Arm64VDouble;
 
@@ -175,7 +175,7 @@ namespace ARMeilleure.Instructions
                     }
                     else
                     {
-                        InstEmitSimdHelper32Arm64.EmitScalarUnaryOpF32(context, unsigned ? Intrinsic.Arm64FcvtzuS : Intrinsic.Arm64FcvtzsS);
+                        InstEmitSimdHelper32Arm64.EmitScalarUnaryOpF32(context, unsigned ? Intrinsic.Arm64FcvtzuS : Intrinsic.Arm64FcvtzsS, false);
                     }
                 }
                 else if (!roundWithFpscr && Optimizations.UseSse41)
@@ -260,28 +260,64 @@ namespace ARMeilleure.Instructions
 
             if (Optimizations.UseAdvSimd)
             {
-                if (unsigned)
+                bool doubleSize = floatSize == OperandType.FP64;
+
+                if (doubleSize)
                 {
-                    inst = rm switch {
+                    Operand m = GetVecA32(op.Vm >> 1);
-                        0b00 => Intrinsic.Arm64FcvtauS,
+
-                        0b01 => Intrinsic.Arm64FcvtnuS,
+                    Operand toConvert = InstEmitSimdHelper32Arm64.EmitExtractScalar(context, m, op.Vm, true);
-                        0b10 => Intrinsic.Arm64FcvtpuS,
+
-                        0b11 => Intrinsic.Arm64FcvtmuS,
+                    if (unsigned)
-                        _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                    {
-                    };
+                        inst = rm switch {
+                            0b00 => Intrinsic.Arm64FcvtauGp,
+                            0b01 => Intrinsic.Arm64FcvtnuGp,
+                            0b10 => Intrinsic.Arm64FcvtpuGp,
+                            0b11 => Intrinsic.Arm64FcvtmuGp,
+                            _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                        };
+                    }
+                    else
+                    {
+                        inst = rm switch {
+                            0b00 => Intrinsic.Arm64FcvtasGp,
+                            0b01 => Intrinsic.Arm64FcvtnsGp,
+                            0b10 => Intrinsic.Arm64FcvtpsGp,
+                            0b11 => Intrinsic.Arm64FcvtmsGp,
+                            _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                        };
+                    }
+
+                    Operand asInteger = context.AddIntrinsicInt(inst | Intrinsic.Arm64VDouble, toConvert);
+
+                    InsertScalar(context, op.Vd, asInteger);
                 }
                 else
                 {
-                    inst = rm switch {
+                    if (unsigned)
-                        0b00 => Intrinsic.Arm64FcvtasS,
+                    {
-                        0b01 => Intrinsic.Arm64FcvtnsS,
+                        inst = rm switch {
-                        0b10 => Intrinsic.Arm64FcvtpsS,
+                            0b00 => Intrinsic.Arm64FcvtauS,
-                        0b11 => Intrinsic.Arm64FcvtmsS,
+                            0b01 => Intrinsic.Arm64FcvtnuS,
-                        _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                            0b10 => Intrinsic.Arm64FcvtpuS,
-                    };
+                            0b11 => Intrinsic.Arm64FcvtmuS,
-                }
+                            _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                        };
+                    }
+                    else
+                    {
+                        inst = rm switch {
+                            0b00 => Intrinsic.Arm64FcvtasS,
+                            0b01 => Intrinsic.Arm64FcvtnsS,
+                            0b10 => Intrinsic.Arm64FcvtpsS,
+                            0b11 => Intrinsic.Arm64FcvtmsS,
+                            _ => throw new ArgumentOutOfRangeException(nameof(rm))
+                        };
+                    }
 
-                InstEmitSimdHelper32Arm64.EmitScalarUnaryOpF32(context, inst);
+                    InstEmitSimdHelper32Arm64.EmitScalarUnaryOpF32(context, inst);
+                }
             }
             else if (Optimizations.UseSse41)
             {

src/ARMeilleure/Instructions/InstEmitSimdHelper32Arm64.cs

@@ -192,11 +192,10 @@ namespace ARMeilleure.Instructions
             EmitVectorTernaryOpSimd32(context, (d, n, m) => context.AddIntrinsic(inst, d, n, m));
         }
 
-        public static void EmitScalarUnaryOpSimd32(ArmEmitterContext context, Func1I scalarFunc)
+        public static void EmitScalarUnaryOpSimd32(ArmEmitterContext context, Func1I scalarFunc, bool doubleSize)
         {
             OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
 
-            bool doubleSize = (op.Size & 1) != 0;
             int shift = doubleSize ? 1 : 2;
             Operand m = GetVecA32(op.Vm >> shift);
             Operand d = GetVecA32(op.Vd >> shift);
@@ -215,8 +214,13 @@ namespace ARMeilleure.Instructions
         {
             OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
 
-            inst |= ((op.Size & 1) != 0 ? Intrinsic.Arm64VDouble : Intrinsic.Arm64VFloat) | Intrinsic.Arm64V128;
+            EmitScalarUnaryOpF32(context, inst, (op.Size & 1) != 0);
-            EmitScalarUnaryOpSimd32(context, (m) => (inst == 0) ? m : context.AddIntrinsic(inst, m));
+        }
+
+        public static void EmitScalarUnaryOpF32(ArmEmitterContext context, Intrinsic inst, bool doubleSize)
+        {
+            inst |= (doubleSize ? Intrinsic.Arm64VDouble : Intrinsic.Arm64VFloat) | Intrinsic.Arm64V128;
+            EmitScalarUnaryOpSimd32(context, (m) => (inst == 0) ? m : context.AddIntrinsic(inst, m), doubleSize);
         }
 
         public static void EmitScalarBinaryOpSimd32(ArmEmitterContext context, Func2I scalarFunc)

src/ARMeilleure/IntermediateRepresentation/Intrinsic.cs

@@ -1,5 +1,8 @@
+using System;
+
 namespace ARMeilleure.IntermediateRepresentation
 {
+    [Flags]
     enum Intrinsic : ushort
     {
         // X86 (SSE and AVX)

src/ARMeilleure/Optimizations.cs

@@ -13,6 +13,7 @@ namespace ARMeilleure
         public static bool UseUnmanagedDispatchLoop { get; set; } = true;
 
         public static bool UseAdvSimdIfAvailable    { get; set; } = true;
+        public static bool UseArm64AesIfAvailable   { get; set; } = true;
         public static bool UseArm64PmullIfAvailable { get; set; } = true;
 
         public static bool UseSseIfAvailable       { get; set; } = true;
@@ -41,6 +42,7 @@ namespace ARMeilleure
         }
 
         internal static bool UseAdvSimd    => UseAdvSimdIfAvailable    && Arm64HardwareCapabilities.SupportsAdvSimd;
+        internal static bool UseArm64Aes   => UseArm64AesIfAvailable   && Arm64HardwareCapabilities.SupportsAes;
         internal static bool UseArm64Pmull => UseArm64PmullIfAvailable && Arm64HardwareCapabilities.SupportsPmull;
 
         internal static bool UseSse       => UseSseIfAvailable       && X86HardwareCapabilities.SupportsSse;

src/ARMeilleure/Signal/NativeSignalHandler.cs

@@ -78,7 +78,7 @@ namespace ARMeilleure.Signal
         private static IntPtr _signalHandlerPtr;
         private static IntPtr _signalHandlerHandle;
 
-        private static readonly object _lock = new object();
+        private static readonly object _lock = new();
         private static bool _initialized;
 
         static NativeSignalHandler()

src/ARMeilleure/State/ExecutionMode.cs

@@ -1,6 +1,6 @@
 namespace ARMeilleure.State
 {
-    enum ExecutionMode : int
+    enum ExecutionMode
     {
         Aarch32Arm = 0,
         Aarch32Thumb = 1,

src/ARMeilleure/Translation/Cache/JitCache.cs

@@ -2,6 +2,7 @@ using ARMeilleure.CodeGen;
 using ARMeilleure.CodeGen.Unwinding;
 using ARMeilleure.Memory;
 using ARMeilleure.Native;
+using Ryujinx.Memory;
 using System;
 using System.Collections.Generic;
 using System.Diagnostics;
@@ -12,8 +13,8 @@ namespace ARMeilleure.Translation.Cache
 {
     static partial class JitCache
     {
-        private const int PageSize = 4 * 1024;
+        private static readonly int PageSize = (int)MemoryBlock.GetPageSize();
-        private const int PageMask = PageSize - 1;
+        private static readonly int PageMask = PageSize - 1;
 
         private const int CodeAlignment = 4; // Bytes.
         private const int CacheSize = 2047 * 1024 * 1024;
@@ -25,7 +26,7 @@ namespace ARMeilleure.Translation.Cache
 
         private static readonly List<CacheEntry> _cacheEntries = new List<CacheEntry>();
 
-        private static readonly object _lock = new object();
+        private static readonly object _lock = new();
         private static bool _initialized;
 
         [SupportedOSPlatform("windows")]

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -30,7 +30,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 4661; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 5292; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

src/Ryujinx.Audio.Backends.OpenAL/OpenALHardwareDeviceSession.cs

@@ -17,7 +17,7 @@ namespace Ryujinx.Audio.Backends.OpenAL
         private Queue<OpenALAudioBuffer> _queuedBuffers;
         private ulong _playedSampleCount;
 
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         public OpenALHardwareDeviceSession(OpenALHardwareDeviceDriver driver, IVirtualMemoryManager memoryManager, SampleFormat requestedSampleFormat, uint requestedSampleRate, uint requestedChannelCount, float requestedVolume) : base(memoryManager, requestedSampleFormat, requestedSampleRate, requestedChannelCount)
         {

src/Ryujinx.Audio.Backends.SoundIo/Native/SoundIoBackend.cs

@@ -1,6 +1,6 @@
 namespace Ryujinx.Audio.Backends.SoundIo.Native
 {
-    public enum SoundIoBackend : int
+    public enum SoundIoBackend
     {
         None = 0,
         Jack = 1,

src/Ryujinx.Audio/AudioManager.cs

@@ -11,7 +11,7 @@ namespace Ryujinx.Audio
         /// <summary>
         /// Lock used to control the waiters registration.
         /// </summary>
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         /// <summary>
         /// Events signaled when the driver played audio buffers.

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

@@ -10,7 +10,7 @@ namespace Ryujinx.Audio.Backends.Common
     {
         private const int RingBufferAlignment = 2048;
 
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         private byte[] _buffer;
         private int _size;

src/Ryujinx.Audio/Input/AudioInputManager.cs

@@ -14,12 +14,12 @@ namespace Ryujinx.Audio.Input
     /// </summary>
     public class AudioInputManager : IDisposable
     {
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         /// <summary>
         /// Lock used for session allocation.
         /// </summary>
-        private object _sessionLock = new object();
+        private readonly object _sessionLock = new();
 
         /// <summary>
         /// The session ids allocation table.

src/Ryujinx.Audio/Input/AudioInputSystem.cs

@@ -48,7 +48,7 @@ namespace Ryujinx.Audio.Input
         /// <summary>
         /// The lock of the parent.
         /// </summary>
-        private object _parentLock;
+        private readonly object _parentLock;
 
         /// <summary>
         /// The dispose state.

src/Ryujinx.Audio/Output/AudioOutputManager.cs

@@ -14,12 +14,12 @@ namespace Ryujinx.Audio.Output
     /// </summary>
     public class AudioOutputManager : IDisposable
     {
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         /// <summary>
         /// Lock used for session allocation.
         /// </summary>
-        private object _sessionLock = new object();
+        private readonly object _sessionLock = new();
 
         /// <summary>
         /// The session ids allocation table.

src/Ryujinx.Audio/Output/AudioOutputSystem.cs

@@ -48,7 +48,7 @@ namespace Ryujinx.Audio.Output
         /// <summary>
         /// THe lock of the parent.
         /// </summary>
-        private object _parentLock;
+        private readonly object _parentLock;
 
         /// <summary>
         /// The dispose state.

src/Ryujinx.Audio/Renderer/Server/AudioRenderSystem.cs

@@ -26,7 +26,7 @@ namespace Ryujinx.Audio.Renderer.Server
 {
     public class AudioRenderSystem : IDisposable
     {
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         private AudioRendererRenderingDevice _renderingDevice;
         private AudioRendererExecutionMode _executionMode;

src/Ryujinx.Audio/Renderer/Server/AudioRendererManager.cs

@@ -19,12 +19,12 @@ namespace Ryujinx.Audio.Renderer.Server
         /// <summary>
         /// Lock used for session allocation.
         /// </summary>
-        private object _sessionLock = new object();
+        private readonly object _sessionLock = new();
 
         /// <summary>
         /// Lock used to control the <see cref="AudioProcessor"/> running state.
         /// </summary>
-        private object _audioProcessorLock = new object();
+        private readonly object _audioProcessorLock = new();
 
         /// <summary>
         /// The session ids allocation table.

src/Ryujinx.Audio/Renderer/Server/Upsampler/UpsamplerManager.cs

@@ -16,7 +16,7 @@ namespace Ryujinx.Audio.Renderer.Server.Upsampler
         /// <summary>
         /// Global lock of the object.
         /// </summary>
-        private object Lock = new object();
+        private readonly object Lock = new();
 
         /// <summary>
         /// The upsamplers instances.

src/Ryujinx.Ava/AppHost.cs

@@ -40,6 +40,7 @@ using SixLabors.ImageSharp;
 using SixLabors.ImageSharp.Formats.Png;
 using SixLabors.ImageSharp.PixelFormats;
 using SixLabors.ImageSharp.Processing;
+using SPB.Graphics.Exceptions;
 using SPB.Graphics.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -475,11 +476,20 @@ namespace Ryujinx.Ava
                 _windowsMultimediaTimerResolution = null;
             }
 
-            (_rendererHost.EmbeddedWindow as EmbeddedWindowOpenGL)?.MakeCurrent();
+            if (_rendererHost.EmbeddedWindow is EmbeddedWindowOpenGL openGlWindow)
+            {
+                // Try to bind the OpenGL context before calling the shutdown event.
+                openGlWindow.MakeCurrent(false, false);
 
-            Device.DisposeGpu();
+                Device.DisposeGpu();
 
-            (_rendererHost.EmbeddedWindow as EmbeddedWindowOpenGL)?.MakeCurrent(null);
+                // Unbind context and destroy everything.
+                openGlWindow.MakeCurrent(true, false);
+            }
+            else
+            {
+                Device.DisposeGpu();
+            }
         }
 
         private void HideCursorState_Changed(object sender, ReactiveEventArgs<HideCursorMode> state)
@@ -930,7 +940,7 @@ namespace Ryujinx.Ava
                 _gpuDoneEvent.Set();
             });
 
-            (_rendererHost.EmbeddedWindow as EmbeddedWindowOpenGL)?.MakeCurrent(null);
+            (_rendererHost.EmbeddedWindow as EmbeddedWindowOpenGL)?.MakeCurrent(true);
         }
 
         public void UpdateStatus()
@@ -1044,7 +1054,7 @@ namespace Ryujinx.Ava
                             ScreenshotRequested = true;
                             break;
                         case KeyboardHotkeyState.ShowUi:
-                            _viewModel.ShowMenuAndStatusBar = true;
+                            _viewModel.ShowMenuAndStatusBar = !_viewModel.ShowMenuAndStatusBar;
                             break;
                         case KeyboardHotkeyState.Pause:
                             if (_viewModel.IsPaused)

src/Ryujinx.Ava/Assets/Locales/en_US.json

@@ -544,6 +544,9 @@
   "SwkbdMinCharacters": "Must be at least {0} characters long",
   "SwkbdMinRangeCharacters": "Must be {0}-{1} characters long",
   "SoftwareKeyboard": "Software Keyboard",
+  "SoftwareKeyboardModeNumbersOnly": "Must be numbers only",
+  "SoftwareKeyboardModeAlphabet": "Must be non CJK-characters only",
+  "SoftwareKeyboardModeASCII": "Must be ASCII text only",
   "DialogControllerAppletMessagePlayerRange": "Application requests {0} player(s) with:\n\nTYPES: {1}\n\nPLAYERS: {2}\n\n{3}Please open Settings and reconfigure Input now or press Close.",
   "DialogControllerAppletMessage": "Application requests exactly {0} player(s) with:\n\nTYPES: {1}\n\nPLAYERS: {2}\n\n{3}Please open Settings and reconfigure Input now or press Close.",
   "DialogControllerAppletDockModeSet": "Docked mode set. Handheld is also invalid.\n\n",

src/Ryujinx.Ava/Assets/Locales/zh_CN.json

@@ -527,6 +527,9 @@
   "SwkbdMinCharacters": "至少应为 {0} 个字长",
   "SwkbdMinRangeCharacters": "必须为 {0}-{1} 个字长",
   "SoftwareKeyboard": "软件键盘",
+  "SoftwareKeyboardModeNumbersOnly": "只接受数字",
+  "SoftwareKeyboardModeAlphabet": "只接受英文字母",
+  "SoftwareKeyboardModeASCII": "只接受 ASCII 符号",
   "DialogControllerAppletMessagePlayerRange": "游戏需要 {0} 个玩家并满足以下要求：\n\n手柄类型：{1}\n\n玩家类型：{2}\n\n{3}请打开设置窗口，重新配置手柄输入；或者关闭返回。",
   "DialogControllerAppletMessage": "游戏需要刚好 {0} 个玩家并满足以下要求：\n\n手柄类型：{1}\n\n玩家类型：{2}\n\n{3}请打开设置窗口，重新配置手柄输入；或者关闭返回。",
   "DialogControllerAppletDockModeSet": "目前处于主机模式，无法使用掌机操作方式",

src/Ryujinx.Ava/Assets/Locales/zh_TW.json

@@ -527,6 +527,9 @@
   "SwkbdMinCharacters": "至少應為 {0} 個字長",
   "SwkbdMinRangeCharacters": "必須為 {0}-{1} 個字長",
   "SoftwareKeyboard": "軟體鍵盤",
+  "SoftwareKeyboardModeNumbersOnly": "只接受數字",
+  "SoftwareKeyboardModeAlphabet": "只接受英文字母",
+  "SoftwareKeyboardModeASCII": "只接受 ASCII 符號",
   "DialogControllerAppletMessagePlayerRange": "本遊戲需要 {0} 個玩家持有：\n\n類型：{1}\n\n玩家：{2}\n\n{3}請打開設定畫面，配置手把，或者關閉本視窗。",
   "DialogControllerAppletMessage": "本遊戲需要剛好 {0} 個玩家持有：\n\n類型：{1}\n\n玩家：{2}\n\n{3}請打開設定畫面，配置手把，或者關閉本視窗。",
   "DialogControllerAppletDockModeSet": "現在處於主機模式，無法使用掌機操作方式\n\n",

src/Ryujinx.Ava/Modules/Updater/Updater.cs

@@ -740,6 +740,18 @@ namespace Ryujinx.Modules
         {
             var files = Directory.EnumerateFiles(HomeDir); // All files directly in base dir.
 
+            // Determine and exclude user files only when the updater is running, not when cleaning old files
+            if (_running && !OperatingSystem.IsMacOS())
+            {
+                // Compare the loose files in base directory against the loose files from the incoming update, and store foreign ones in a user list.
+                var oldFiles = Directory.EnumerateFiles(HomeDir, "*", SearchOption.TopDirectoryOnly).Select(Path.GetFileName);
+                var newFiles = Directory.EnumerateFiles(UpdatePublishDir, "*", SearchOption.TopDirectoryOnly).Select(Path.GetFileName);
+                var userFiles = oldFiles.Except(newFiles).Select(filename => Path.Combine(HomeDir, filename));
+
+                // Remove user files from the paths in files.
+                files = files.Except(userFiles);
+            }
+
             if (OperatingSystem.IsWindows())
             {
                 foreach (string dir in WindowsDependencyDirs)

src/Ryujinx.Ava/UI/Applet/SwkbdAppletDialog.axaml.cs

@@ -9,14 +9,17 @@ using Ryujinx.Ava.Common.Locale;
 using Ryujinx.Ava.UI.Helpers;
 using Ryujinx.Ava.UI.Windows;
 using Ryujinx.HLE.HOS.Applets;
+using Ryujinx.HLE.HOS.Applets.SoftwareKeyboard;
 using System;
+using System.Linq;
 using System.Threading.Tasks;
 
 namespace Ryujinx.Ava.UI.Controls
 {
     internal partial class SwkbdAppletDialog : UserControl
     {
-        private Predicate<int> _checkLength;
+        private Predicate<int> _checkLength = _ => true;
+        private Predicate<string> _checkInput = _ => true;
         private int _inputMax;
         private int _inputMin;
         private string _placeholder;
@@ -35,8 +38,6 @@ namespace Ryujinx.Ava.UI.Controls
             Input.Watermark = _placeholder;
 
             Input.AddHandler(TextInputEvent, Message_TextInput, RoutingStrategies.Tunnel, true);
-
-            SetInputLengthValidation(0, int.MaxValue); // Disable by default.
         }
 
         public SwkbdAppletDialog()
@@ -67,6 +68,7 @@ namespace Ryujinx.Ava.UI.Controls
             string input = string.Empty;
 
             content.SetInputLengthValidation(args.StringLengthMin, args.StringLengthMax);
+            content.SetInputValidation(args.KeyboardMode);
 
             content._host = contentDialog;
             contentDialog.Title = title;
@@ -91,6 +93,12 @@ namespace Ryujinx.Ava.UI.Controls
             return (result, input);
         }
 
+        private void ApplyValidationInfo(string text)
+        {
+            Error.IsVisible = !string.IsNullOrEmpty(text);
+            Error.Text = text;
+        }
+
         public void SetInputLengthValidation(int min, int max)
         {
             _inputMin = Math.Min(min, max);
@@ -99,6 +107,8 @@ namespace Ryujinx.Ava.UI.Controls
             Error.IsVisible = false;
             Error.FontStyle = FontStyle.Italic;
 
+            string validationInfoText = "";
+
             if (_inputMin <= 0 && _inputMax == int.MaxValue) // Disable.
             {
                 Error.IsVisible = false;
@@ -107,21 +117,48 @@ namespace Ryujinx.Ava.UI.Controls
             }
             else if (_inputMin > 0 && _inputMax == int.MaxValue)
             {
-                Error.IsVisible = true;
+                validationInfoText = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SwkbdMinCharacters, _inputMin);
-
-                Error.Text = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SwkbdMinCharacters, _inputMin);
 
                 _checkLength = length => _inputMin <= length;
             }
             else
             {
-                Error.IsVisible = true;
+                validationInfoText = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SwkbdMinRangeCharacters, _inputMin, _inputMax);
-
-                Error.Text = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SwkbdMinRangeCharacters, _inputMin, _inputMax);
 
                 _checkLength = length => _inputMin <= length && length <= _inputMax;
             }
 
+            ApplyValidationInfo(validationInfoText);
+            Message_TextInput(this, new TextInputEventArgs());
+        }
+
+        private void SetInputValidation(KeyboardMode mode)
+        {
+            string validationInfoText = Error.Text;
+            string localeText;
+            switch (mode)
+            {
+                case KeyboardMode.NumbersOnly:
+                    localeText = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SoftwareKeyboardModeNumbersOnly);
+                    validationInfoText = string.IsNullOrEmpty(validationInfoText) ? localeText : string.Join("\n", validationInfoText, localeText);
+                    _checkInput = text => text.All(char.IsDigit);
+                    break;
+                case KeyboardMode.Alphabet:
+                    localeText = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SoftwareKeyboardModeAlphabet);
+                    validationInfoText = string.IsNullOrEmpty(validationInfoText) ? localeText : string.Join("\n", validationInfoText, localeText);
+                    _checkInput = text => text.All(value => !CJKCharacterValidation.IsCJK(value));
+                    break;
+                case KeyboardMode.ASCII:
+                    localeText = LocaleManager.Instance.UpdateAndGetDynamicValue(LocaleKeys.SoftwareKeyboardModeASCII);
+                    validationInfoText = string.IsNullOrEmpty(validationInfoText) ? localeText : string.Join("\n", validationInfoText, localeText);
+                    _checkInput = text => text.All(char.IsAscii);
+                    break;
+                default:
+                    _checkInput = _ => true;
+                    break;
+            }
+
+            ApplyValidationInfo(validationInfoText);
             Message_TextInput(this, new TextInputEventArgs());
         }
 
@@ -129,7 +166,7 @@ namespace Ryujinx.Ava.UI.Controls
         {
             if (_host != null)
             {
-                _host.IsPrimaryButtonEnabled = _checkLength(Message.Length);
+                _host.IsPrimaryButtonEnabled = _checkLength(Message.Length) && _checkInput(Message);
             }
         }
 
@@ -141,7 +178,7 @@ namespace Ryujinx.Ava.UI.Controls
             }
             else
             {
-                _host.IsPrimaryButtonEnabled = _checkLength(Message.Length);
+                _host.IsPrimaryButtonEnabled = _checkLength(Message.Length) && _checkInput(Message);
             }
         }
     }

src/Ryujinx.Ava/UI/Controls/ApplicationGridView.axaml

@@ -32,10 +32,10 @@
             <ListBox.ItemsPanel>
                 <ItemsPanelTemplate>
                     <flex:FlexPanel
-                        HorizontalAlignment="Stretch"
+                        HorizontalAlignment="Center"
                         VerticalAlignment="Stretch"
                         AlignContent="FlexStart"
-                        JustifyContent="Center" />
+                        JustifyContent="FlexStart" />
                 </ItemsPanelTemplate>
             </ListBox.ItemsPanel>
             <ListBox.Styles>

src/Ryujinx.Ava/UI/Renderer/EmbeddedWindow.cs

@@ -123,7 +123,7 @@ namespace Ryujinx.Ava.UI.Renderer
             }
             else
             {
-                X11Window = PlatformHelper.CreateOpenGLWindow(FramebufferFormat.Default, 0, 0, 100, 100) as GLXWindow;
+                X11Window = PlatformHelper.CreateOpenGLWindow(new FramebufferFormat(new ColorFormat(8, 8, 8, 0), 16, 0, ColorFormat.Zero, 0, 2, false), 0, 0, 100, 100) as GLXWindow;
             }
 
             WindowHandle = X11Window.WindowHandle.RawHandle;

src/Ryujinx.Ava/UI/Renderer/EmbeddedWindowOpenGL.cs

@@ -1,9 +1,11 @@
 using OpenTK.Graphics.OpenGL;
 using Ryujinx.Common.Configuration;
+using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.OpenGL;
 using Ryujinx.Ui.Common.Configuration;
 using SPB.Graphics;
+using SPB.Graphics.Exceptions;
 using SPB.Graphics.OpenGL;
 using SPB.Platform;
 using SPB.Platform.WGL;
@@ -18,8 +20,6 @@ namespace Ryujinx.Ava.UI.Renderer
 
         public OpenGLContextBase Context { get; set; }
 
-        public EmbeddedWindowOpenGL() { }
-
         protected override void OnWindowDestroying()
         {
             Context.Dispose();
@@ -62,14 +62,21 @@ namespace Ryujinx.Ava.UI.Renderer
             Context.MakeCurrent(null);
         }
 
-        public void MakeCurrent()
+        public void MakeCurrent(bool unbind = false, bool shouldThrow = true)
         {
-            Context?.MakeCurrent(_window);
+            try
-        }
+            {
+                Context?.MakeCurrent(!unbind ? _window : null);
+            }
+            catch (ContextException e)
+            {
+                if (shouldThrow)
+                {
+                    throw;
+                }
 
-        public void MakeCurrent(NativeWindowBase window)
+                Logger.Warning?.Print(LogClass.Ui, $"Failed to {(!unbind ? "bind" : "unbind")} OpenGL context: {e}");
-        {
+            }
-            Context?.MakeCurrent(window);
         }
 
         public void SwapBuffers()

src/Ryujinx.Common/Configuration/AppDataManager.cs

@@ -96,7 +96,7 @@ namespace Ryujinx.Common.Configuration
             if (OperatingSystem.IsMacOS() && Mode == LaunchMode.UserProfile)
             {
                 string oldConfigPath = Path.Combine(Environment.GetFolderPath(Environment.SpecialFolder.ApplicationData), DefaultBaseDir);
-                if (Path.Exists(oldConfigPath) && !Path.Exists(BaseDirPath))
+                if (Path.Exists(oldConfigPath) && !IsPathSymlink(oldConfigPath) && !Path.Exists(BaseDirPath))
                 {
                     CopyDirectory(oldConfigPath, BaseDirPath);
                     Directory.Delete(oldConfigPath, true);
@@ -115,6 +115,14 @@ namespace Ryujinx.Common.Configuration
             Directory.CreateDirectory(KeysDirPath = Path.Combine(BaseDirPath, KeysDir));
         }
 
+        // Check if existing old baseDirPath is a symlink, to prevent possible errors.
+        // Should be removed, when the existance of the old directory isn't checked anymore.
+        private static bool IsPathSymlink(string path)
+        {
+            FileAttributes attributes = File.GetAttributes(path);
+            return (attributes & FileAttributes.ReparsePoint) == FileAttributes.ReparsePoint;
+        }
+
         private static void CopyDirectory(string sourceDir, string destinationDir)
         {
             var dir = new DirectoryInfo(sourceDir);

src/Ryujinx.Common/Configuration/Hid/ControllerType.cs

@@ -7,7 +7,7 @@ namespace Ryujinx.Common.Configuration.Hid
     // This enum was duplicated from Ryujinx.HLE.HOS.Services.Hid.PlayerIndex and should be kept identical
     [Flags]
     [JsonConverter(typeof(TypedStringEnumConverter<ControllerType>))]
-    public enum ControllerType : int
+    public enum ControllerType
     {
         None,
         ProController  = 1 << 0,

src/Ryujinx.Common/Configuration/Hid/PlayerIndex.cs

@@ -5,7 +5,7 @@ namespace Ryujinx.Common.Configuration.Hid
 {
     // This enum was duplicated from Ryujinx.HLE.HOS.Services.Hid.PlayerIndex and should be kept identical
     [JsonConverter(typeof(TypedStringEnumConverter<PlayerIndex>))]
-    public enum PlayerIndex : int
+    public enum PlayerIndex
     {
         Player1  = 0,
         Player2  = 1,

src/Ryujinx.Common/Memory/SpanReader.cs

@@ -24,6 +24,24 @@ namespace Ryujinx.Common.Memory
             return value;
         }
 
+        public bool TryRead<T>(out T value) where T : unmanaged
+        {
+            int valueSize = Unsafe.SizeOf<T>();
+
+            if (valueSize > _input.Length)
+            {
+                value = default;
+
+                return false;
+            }
+
+            value = MemoryMarshal.Cast<byte, T>(_input)[0];
+
+            _input = _input.Slice(valueSize);
+
+            return true;
+        }
+
         public ReadOnlySpan<byte> GetSpan(int size)
         {
             ReadOnlySpan<byte> data = _input.Slice(0, size);

src/Ryujinx.Common/Utilities/TypedStringEnumConverter.cs

@@ -1,4 +1,5 @@
 #nullable enable
+using Ryujinx.Common.Logging;
 using System;
 using System.Text.Json;
 using System.Text.Json.Serialization;
@@ -18,12 +19,14 @@ namespace Ryujinx.Common.Utilities
         public override TEnum Read(ref Utf8JsonReader reader, Type typeToConvert, JsonSerializerOptions options)
         {
             var enumValue = reader.GetString();
-            if (string.IsNullOrEmpty(enumValue))
+
+            if (Enum.TryParse(enumValue, out TEnum value))
             {
-                return default;
+                return value;
             }
 
-            return Enum.Parse<TEnum>(enumValue);
+            Logger.Warning?.Print(LogClass.Configuration, $"Failed to parse enum value \"{enumValue}\" for {typeof(TEnum)}, using default \"{default(TEnum)}\"");
+            return default;
         }
 
         public override void Write(Utf8JsonWriter writer, TEnum value, JsonSerializerOptions options)
@@ -31,4 +34,4 @@ namespace Ryujinx.Common.Utilities
             writer.WriteStringValue(value.ToString());
         }
     }
-}
+}

src/Ryujinx.Cpu/AddressSpace.cs

@@ -5,7 +5,7 @@ using System;
 
 namespace Ryujinx.Cpu
 {
-    class AddressSpace : IDisposable
+    public class AddressSpace : IDisposable
     {
         private const ulong PageSize = 0x1000;
 
@@ -154,7 +154,9 @@ namespace Ryujinx.Cpu
         public MemoryBlock Base { get; }
         public MemoryBlock Mirror { get; }
 
-        public AddressSpace(MemoryBlock backingMemory, ulong asSize, bool supports4KBPages)
+        public ulong AddressSpaceSize { get; }
+
+        public AddressSpace(MemoryBlock backingMemory, MemoryBlock baseMemory, MemoryBlock mirrorMemory, ulong addressSpaceSize, bool supports4KBPages)
         {
             if (!supports4KBPages)
             {
@@ -163,17 +165,48 @@ namespace Ryujinx.Cpu
                 _privateTree = new IntrusiveRedBlackTree<PrivateMapping>();
                 _treeLock = new object();
 
-                _mappingTree.Add(new Mapping(0UL, asSize, MappingType.None));
+                _mappingTree.Add(new Mapping(0UL, addressSpaceSize, MappingType.None));
-                _privateTree.Add(new PrivateMapping(0UL, asSize, default));
+                _privateTree.Add(new PrivateMapping(0UL, addressSpaceSize, default));
             }
 
             _backingMemory = backingMemory;
             _supports4KBPages = supports4KBPages;
 
+            Base = baseMemory;
+            Mirror = mirrorMemory;
+            AddressSpaceSize = addressSpaceSize;
+        }
+
+        public static bool TryCreate(MemoryBlock backingMemory, ulong asSize, bool supports4KBPages, out AddressSpace addressSpace)
+        {
+            addressSpace = null;
+
             MemoryAllocationFlags asFlags = MemoryAllocationFlags.Reserve | MemoryAllocationFlags.ViewCompatible;
 
-            Base = new MemoryBlock(asSize, asFlags);
+            ulong minAddressSpaceSize = Math.Min(asSize, 1UL << 36);
-            Mirror = new MemoryBlock(asSize, asFlags);
+
+            // Attempt to create the address space with expected size or try to reduce it until it succeed.
+            for (ulong addressSpaceSize = asSize; addressSpaceSize >= minAddressSpaceSize; addressSpaceSize >>= 1)
+            {
+                MemoryBlock baseMemory = null;
+                MemoryBlock mirrorMemory = null;
+
+                try
+                {
+                    baseMemory = new MemoryBlock(addressSpaceSize, asFlags);
+                    mirrorMemory = new MemoryBlock(addressSpaceSize, asFlags);
+                    addressSpace = new AddressSpace(backingMemory, baseMemory, mirrorMemory, addressSpaceSize, supports4KBPages);
+
+                    break;
+                }
+                catch (OutOfMemoryException)
+                {
+                    baseMemory?.Dispose();
+                    mirrorMemory?.Dispose();
+                }
+            }
+
+            return addressSpace != null;
         }
 
         public void Map(ulong va, ulong pa, ulong size, MemoryMapFlags flags)

src/Ryujinx.Cpu/Jit/MemoryManagerHostMapped.cs

@@ -38,7 +38,8 @@ namespace Ryujinx.Cpu.Jit
         private readonly bool _unsafeMode;
 
         private readonly AddressSpace _addressSpace;
-        private readonly ulong _addressSpaceSize;
+
+        public ulong AddressSpaceSize { get; }
 
         private readonly PageTable<ulong> _pageTable;
 
@@ -62,21 +63,21 @@ namespace Ryujinx.Cpu.Jit
         /// <summary>
         /// Creates a new instance of the host mapped memory manager.
         /// </summary>
-        /// <param name="backingMemory">Physical backing memory where virtual memory will be mapped to</param>
+        /// <param name="addressSpace">Address space instance to use</param>
-        /// <param name="addressSpaceSize">Size of the address space</param>
         /// <param name="unsafeMode">True if unmanaged access should not be masked (unsafe), false otherwise.</param>
         /// <param name="invalidAccessHandler">Optional function to handle invalid memory accesses</param>
-        public MemoryManagerHostMapped(MemoryBlock backingMemory, ulong addressSpaceSize, bool unsafeMode, InvalidAccessHandler invalidAccessHandler = null)
+        public MemoryManagerHostMapped(AddressSpace addressSpace, bool unsafeMode, InvalidAccessHandler invalidAccessHandler)
         {
+            _addressSpace = addressSpace;
             _pageTable = new PageTable<ulong>();
             _invalidAccessHandler = invalidAccessHandler;
             _unsafeMode = unsafeMode;
-            _addressSpaceSize = addressSpaceSize;
+            AddressSpaceSize = addressSpace.AddressSpaceSize;
 
             ulong asSize = PageSize;
             int asBits = PageBits;
 
-            while (asSize < addressSpaceSize)
+            while (asSize < AddressSpaceSize)
             {
                 asSize <<= 1;
                 asBits++;
@@ -86,8 +87,6 @@ namespace Ryujinx.Cpu.Jit
 
             _pageBitmap = new ulong[1 << (AddressSpaceBits - (PageBits + PageToPteShift))];
 
-            _addressSpace = new AddressSpace(backingMemory, asSize, Supports4KBPages);
-
             Tracking = new MemoryTracking(this, (int)MemoryBlock.GetPageSize(), invalidAccessHandler);
             _memoryEh = new MemoryEhMeilleure(_addressSpace.Base, _addressSpace.Mirror, Tracking);
         }
@@ -99,7 +98,7 @@ namespace Ryujinx.Cpu.Jit
         /// <returns>True if the virtual address is part of the addressable space</returns>
         private bool ValidateAddress(ulong va)
         {
-            return va < _addressSpaceSize;
+            return va < AddressSpaceSize;
         }
 
         /// <summary>
@@ -111,7 +110,7 @@ namespace Ryujinx.Cpu.Jit
         private bool ValidateAddressAndSize(ulong va, ulong size)
         {
             ulong endVa = va + size;
-            return endVa >= va && endVa >= size && endVa <= _addressSpaceSize;
+            return endVa >= va && endVa >= size && endVa <= AddressSpaceSize;
         }
 
         /// <summary>

src/Ryujinx.Graphics.GAL/Capabilities.cs

@@ -28,12 +28,15 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool SupportsFragmentShaderOrderingIntel;
         public readonly bool SupportsGeometryShader;
         public readonly bool SupportsGeometryShaderPassthrough;
+        public readonly bool SupportsTransformFeedback;
         public readonly bool SupportsImageLoadFormatted;
         public readonly bool SupportsLayerVertexTessellation;
         public readonly bool SupportsMismatchingViewFormat;
         public readonly bool SupportsCubemapView;
         public readonly bool SupportsNonConstantTextureOffset;
         public readonly bool SupportsShaderBallot;
+        public readonly bool SupportsShaderBarrierDivergence;
+        public readonly bool SupportsShaderFloat64;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsViewportIndexVertexTessellation;
         public readonly bool SupportsViewportMask;
@@ -75,12 +78,15 @@ namespace Ryujinx.Graphics.GAL
             bool supportsFragmentShaderOrderingIntel,
             bool supportsGeometryShader,
             bool supportsGeometryShaderPassthrough,
+            bool supportsTransformFeedback,
             bool supportsImageLoadFormatted,
             bool supportsLayerVertexTessellation,
             bool supportsMismatchingViewFormat,
             bool supportsCubemapView,
             bool supportsNonConstantTextureOffset,
             bool supportsShaderBallot,
+            bool supportsShaderBarrierDivergence,
+            bool supportsShaderFloat64,
             bool supportsTextureShadowLod,
             bool supportsViewportIndexVertexTessellation,
             bool supportsViewportMask,
@@ -118,12 +124,15 @@ namespace Ryujinx.Graphics.GAL
             SupportsFragmentShaderOrderingIntel = supportsFragmentShaderOrderingIntel;
             SupportsGeometryShader = supportsGeometryShader;
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
+            SupportsTransformFeedback = supportsTransformFeedback;
             SupportsImageLoadFormatted = supportsImageLoadFormatted;
             SupportsLayerVertexTessellation = supportsLayerVertexTessellation;
             SupportsMismatchingViewFormat = supportsMismatchingViewFormat;
             SupportsCubemapView = supportsCubemapView;
             SupportsNonConstantTextureOffset = supportsNonConstantTextureOffset;
             SupportsShaderBallot = supportsShaderBallot;
+            SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
+            SupportsShaderFloat64 = supportsShaderFloat64;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsViewportIndexVertexTessellation = supportsViewportIndexVertexTessellation;
             SupportsViewportMask = supportsViewportMask;

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

@@ -56,7 +56,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
         private int _refConsumerPtr;
 
         private Action _interruptAction;
-        private object _interruptLock = new();
+        private readonly object _interruptLock = new();
 
         public event EventHandler<ScreenCaptureImageInfo> ScreenCaptured;
 

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

@@ -151,8 +151,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
 
             ShaderProgramInfo info = cs.Shaders[0].Info;
 
-            bool hasUnaligned = _channel.BufferManager.HasUnalignedStorageBuffers;
-
             for (int index = 0; index < info.SBuffers.Count; index++)
             {
                 BufferDescriptor sb = info.SBuffers[index];
@@ -177,9 +175,17 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
                 _channel.BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), size, sb.Flags);
             }
 
-            if ((_channel.BufferManager.HasUnalignedStorageBuffers) != hasUnaligned)
+            if (_channel.BufferManager.HasUnalignedStorageBuffers != computeState.HasUnalignedStorageBuffer)
             {
                 // Refetch the shader, as assumptions about storage buffer alignment have changed.
+                computeState = new GpuChannelComputeState(
+                    qmd.CtaThreadDimension0,
+                    qmd.CtaThreadDimension1,
+                    qmd.CtaThreadDimension2,
+                    localMemorySize,
+                    sharedMemorySize,
+                    _channel.BufferManager.HasUnalignedStorageBuffers);
+
                 cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, poolState, computeState, shaderGpuVa);
 
                 _context.Renderer.Pipeline.SetProgram(cs.HostProgram);

src/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClassState.cs

@@ -100,22 +100,22 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
     {
 #pragma warning disable CS0649
         public uint SetObject;
-        public int SetObjectClassId => (int)((SetObject >> 0) & 0xFFFF);
+        public int SetObjectClassId => (int)(SetObject & 0xFFFF);
         public int SetObjectEngineId => (int)((SetObject >> 16) & 0x1F);
         public fixed uint Reserved04[63];
         public uint NoOperation;
         public uint SetNotifyA;
-        public int SetNotifyAAddressUpper => (int)((SetNotifyA >> 0) & 0xFF);
+        public int SetNotifyAAddressUpper => (int)(SetNotifyA & 0xFF);
         public uint SetNotifyB;
         public uint Notify;
         public NotifyType NotifyType => (NotifyType)(Notify);
         public uint WaitForIdle;
         public fixed uint Reserved114[7];
         public uint SetGlobalRenderEnableA;
-        public int SetGlobalRenderEnableAOffsetUpper => (int)((SetGlobalRenderEnableA >> 0) & 0xFF);
+        public int SetGlobalRenderEnableAOffsetUpper => (int)(SetGlobalRenderEnableA & 0xFF);
         public uint SetGlobalRenderEnableB;
         public uint SetGlobalRenderEnableC;
-        public int SetGlobalRenderEnableCMode => (int)((SetGlobalRenderEnableC >> 0) & 0x7);
+        public int SetGlobalRenderEnableCMode => (int)(SetGlobalRenderEnableC & 0x7);
         public uint SendGoIdle;
         public uint PmTrigger;
         public uint PmTriggerWfi;
@@ -126,11 +126,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint LineLengthIn;
         public uint LineCount;
         public uint OffsetOutUpper;
-        public int OffsetOutUpperValue => (int)((OffsetOutUpper >> 0) & 0xFF);
+        public int OffsetOutUpperValue => (int)(OffsetOutUpper & 0xFF);
         public uint OffsetOut;
         public uint PitchOut;
         public uint SetDstBlockSize;
-        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)((SetDstBlockSize >> 0) & 0xF);
+        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)(SetDstBlockSize & 0xF);
         public SetDstBlockSizeHeight SetDstBlockSizeHeight => (SetDstBlockSizeHeight)((SetDstBlockSize >> 4) & 0xF);
         public SetDstBlockSizeDepth SetDstBlockSizeDepth => (SetDstBlockSizeDepth)((SetDstBlockSize >> 8) & 0xF);
         public uint SetDstWidth;
@@ -138,11 +138,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetDstDepth;
         public uint SetDstLayer;
         public uint SetDstOriginBytesX;
-        public int SetDstOriginBytesXV => (int)((SetDstOriginBytesX >> 0) & 0xFFFFF);
+        public int SetDstOriginBytesXV => (int)(SetDstOriginBytesX & 0xFFFFF);
         public uint SetDstOriginSamplesY;
-        public int SetDstOriginSamplesYV => (int)((SetDstOriginSamplesY >> 0) & 0xFFFF);
+        public int SetDstOriginSamplesYV => (int)(SetDstOriginSamplesY & 0xFFFF);
         public uint LaunchDma;
-        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)((LaunchDma >> 0) & 0x1);
+        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)(LaunchDma & 0x1);
         public LaunchDmaCompletionType LaunchDmaCompletionType => (LaunchDmaCompletionType)((LaunchDma >> 4) & 0x3);
         public LaunchDmaInterruptType LaunchDmaInterruptType => (LaunchDmaInterruptType)((LaunchDma >> 8) & 0x3);
         public LaunchDmaSemaphoreStructSize LaunchDmaSemaphoreStructSize => (LaunchDmaSemaphoreStructSize)((LaunchDma >> 12) & 0x1);
@@ -153,7 +153,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint LoadInlineData;
         public fixed uint Reserved1B8[9];
         public uint SetI2mSemaphoreA;
-        public int SetI2mSemaphoreAOffsetUpper => (int)((SetI2mSemaphoreA >> 0) & 0xFF);
+        public int SetI2mSemaphoreAOffsetUpper => (int)(SetI2mSemaphoreA & 0xFF);
         public uint SetI2mSemaphoreB;
         public uint SetI2mSemaphoreC;
         public fixed uint Reserved1E8[2];
@@ -162,7 +162,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetI2mSpareNoop02;
         public uint SetI2mSpareNoop03;
         public uint SetValidSpanOverflowAreaA;
-        public int SetValidSpanOverflowAreaAAddressUpper => (int)((SetValidSpanOverflowAreaA >> 0) & 0xFF);
+        public int SetValidSpanOverflowAreaAAddressUpper => (int)(SetValidSpanOverflowAreaA & 0xFF);
         public uint SetValidSpanOverflowAreaB;
         public uint SetValidSpanOverflowAreaC;
         public uint SetCoalesceWaitingPeriodUnit;
@@ -185,12 +185,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetReservedSwMethod06;
         public uint SetReservedSwMethod07;
         public uint SetCwdControl;
-        public SetCwdControlSmSelection SetCwdControlSmSelection => (SetCwdControlSmSelection)((SetCwdControl >> 0) & 0x1);
+        public SetCwdControlSmSelection SetCwdControlSmSelection => (SetCwdControlSmSelection)(SetCwdControl & 0x1);
         public uint InvalidateTextureHeaderCacheNoWfi;
-        public InvalidateCacheLines InvalidateTextureHeaderCacheNoWfiLines => (InvalidateCacheLines)((InvalidateTextureHeaderCacheNoWfi >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateTextureHeaderCacheNoWfiLines => (InvalidateCacheLines)(InvalidateTextureHeaderCacheNoWfi & 0x1);
         public int InvalidateTextureHeaderCacheNoWfiTag => (int)((InvalidateTextureHeaderCacheNoWfi >> 4) & 0x3FFFFF);
         public uint SetCwdRefCounter;
-        public int SetCwdRefCounterSelect => (int)((SetCwdRefCounter >> 0) & 0x3F);
+        public int SetCwdRefCounterSelect => (int)(SetCwdRefCounter & 0x3F);
         public int SetCwdRefCounterValue => (int)((SetCwdRefCounter >> 8) & 0xFFFF);
         public uint SetReservedSwMethod08;
         public uint SetReservedSwMethod09;
@@ -201,28 +201,28 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetReservedSwMethod14;
         public uint SetReservedSwMethod15;
         public uint SetGwcScgType;
-        public SetGwcScgTypeScgType SetGwcScgTypeScgType => (SetGwcScgTypeScgType)((SetGwcScgType >> 0) & 0x1);
+        public SetGwcScgTypeScgType SetGwcScgTypeScgType => (SetGwcScgTypeScgType)(SetGwcScgType & 0x1);
         public uint SetScgControl;
-        public int SetScgControlCompute1MaxSmCount => (int)((SetScgControl >> 0) & 0x1FF);
+        public int SetScgControlCompute1MaxSmCount => (int)(SetScgControl & 0x1FF);
         public uint InvalidateConstantBufferCacheA;
-        public int InvalidateConstantBufferCacheAAddressUpper => (int)((InvalidateConstantBufferCacheA >> 0) & 0xFF);
+        public int InvalidateConstantBufferCacheAAddressUpper => (int)(InvalidateConstantBufferCacheA & 0xFF);
         public uint InvalidateConstantBufferCacheB;
         public uint InvalidateConstantBufferCacheC;
-        public int InvalidateConstantBufferCacheCByteCount => (int)((InvalidateConstantBufferCacheC >> 0) & 0x1FFFF);
+        public int InvalidateConstantBufferCacheCByteCount => (int)(InvalidateConstantBufferCacheC & 0x1FFFF);
         public bool InvalidateConstantBufferCacheCThruL2 => (InvalidateConstantBufferCacheC & 0x80000000) != 0;
         public uint SetComputeClassVersion;
-        public int SetComputeClassVersionCurrent => (int)((SetComputeClassVersion >> 0) & 0xFFFF);
+        public int SetComputeClassVersionCurrent => (int)(SetComputeClassVersion & 0xFFFF);
         public int SetComputeClassVersionOldestSupported => (int)((SetComputeClassVersion >> 16) & 0xFFFF);
         public uint CheckComputeClassVersion;
-        public int CheckComputeClassVersionCurrent => (int)((CheckComputeClassVersion >> 0) & 0xFFFF);
+        public int CheckComputeClassVersionCurrent => (int)(CheckComputeClassVersion & 0xFFFF);
         public int CheckComputeClassVersionOldestSupported => (int)((CheckComputeClassVersion >> 16) & 0xFFFF);
         public uint SetQmdVersion;
-        public int SetQmdVersionCurrent => (int)((SetQmdVersion >> 0) & 0xFFFF);
+        public int SetQmdVersionCurrent => (int)(SetQmdVersion & 0xFFFF);
         public int SetQmdVersionOldestSupported => (int)((SetQmdVersion >> 16) & 0xFFFF);
         public uint SetWfiConfig;
         public bool SetWfiConfigEnableScgTypeWfi => (SetWfiConfig & 0x1) != 0;
         public uint CheckQmdVersion;
-        public int CheckQmdVersionCurrent => (int)((CheckQmdVersion >> 0) & 0xFFFF);
+        public int CheckQmdVersionCurrent => (int)(CheckQmdVersion & 0xFFFF);
         public int CheckQmdVersionOldestSupported => (int)((CheckQmdVersion >> 16) & 0xFFFF);
         public uint WaitForIdleScgType;
         public uint InvalidateSkedCaches;
@@ -231,28 +231,28 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public bool SetScgRenderEnableControlCompute1UsesRenderEnable => (SetScgRenderEnableControl & 0x1) != 0;
         public fixed uint Reserved2A0[4];
         public uint SetCwdSlotCount;
-        public int SetCwdSlotCountV => (int)((SetCwdSlotCount >> 0) & 0xFF);
+        public int SetCwdSlotCountV => (int)(SetCwdSlotCount & 0xFF);
         public uint SendPcasA;
         public uint SendPcasB;
-        public int SendPcasBFrom => (int)((SendPcasB >> 0) & 0xFFFFFF);
+        public int SendPcasBFrom => (int)(SendPcasB & 0xFFFFFF);
         public int SendPcasBDelta => (int)((SendPcasB >> 24) & 0xFF);
         public uint SendSignalingPcasB;
         public bool SendSignalingPcasBInvalidate => (SendSignalingPcasB & 0x1) != 0;
         public bool SendSignalingPcasBSchedule => (SendSignalingPcasB & 0x2) != 0;
         public fixed uint Reserved2C0[9];
         public uint SetShaderLocalMemoryNonThrottledA;
-        public int SetShaderLocalMemoryNonThrottledASizeUpper => (int)((SetShaderLocalMemoryNonThrottledA >> 0) & 0xFF);
+        public int SetShaderLocalMemoryNonThrottledASizeUpper => (int)(SetShaderLocalMemoryNonThrottledA & 0xFF);
         public uint SetShaderLocalMemoryNonThrottledB;
         public uint SetShaderLocalMemoryNonThrottledC;
-        public int SetShaderLocalMemoryNonThrottledCMaxSmCount => (int)((SetShaderLocalMemoryNonThrottledC >> 0) & 0x1FF);
+        public int SetShaderLocalMemoryNonThrottledCMaxSmCount => (int)(SetShaderLocalMemoryNonThrottledC & 0x1FF);
         public uint SetShaderLocalMemoryThrottledA;
-        public int SetShaderLocalMemoryThrottledASizeUpper => (int)((SetShaderLocalMemoryThrottledA >> 0) & 0xFF);
+        public int SetShaderLocalMemoryThrottledASizeUpper => (int)(SetShaderLocalMemoryThrottledA & 0xFF);
         public uint SetShaderLocalMemoryThrottledB;
         public uint SetShaderLocalMemoryThrottledC;
-        public int SetShaderLocalMemoryThrottledCMaxSmCount => (int)((SetShaderLocalMemoryThrottledC >> 0) & 0x1FF);
+        public int SetShaderLocalMemoryThrottledCMaxSmCount => (int)(SetShaderLocalMemoryThrottledC & 0x1FF);
         public fixed uint Reserved2FC[5];
         public uint SetSpaVersion;
-        public int SetSpaVersionMinor => (int)((SetSpaVersion >> 0) & 0xFF);
+        public int SetSpaVersionMinor => (int)(SetSpaVersion & 0xFF);
         public int SetSpaVersionMajor => (int)((SetSpaVersion >> 8) & 0xFF);
         public fixed uint Reserved314[123];
         public uint SetFalcon00;
@@ -291,14 +291,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetShaderLocalMemoryWindow;
         public fixed uint Reserved780[4];
         public uint SetShaderLocalMemoryA;
-        public int SetShaderLocalMemoryAAddressUpper => (int)((SetShaderLocalMemoryA >> 0) & 0xFF);
+        public int SetShaderLocalMemoryAAddressUpper => (int)(SetShaderLocalMemoryA & 0xFF);
         public uint SetShaderLocalMemoryB;
         public fixed uint Reserved798[383];
         public uint SetShaderCacheControl;
         public bool SetShaderCacheControlIcachePrefetchEnable => (SetShaderCacheControl & 0x1) != 0;
         public fixed uint ReservedD98[19];
         public uint SetSmTimeoutInterval;
-        public int SetSmTimeoutIntervalCounterBit => (int)((SetSmTimeoutInterval >> 0) & 0x3F);
+        public int SetSmTimeoutIntervalCounterBit => (int)(SetSmTimeoutInterval & 0x3F);
         public fixed uint ReservedDE8[87];
         public uint SetSpareNoop12;
         public uint SetSpareNoop13;
@@ -324,48 +324,48 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public bool InvalidateTextureHeaderCacheAllV => (InvalidateTextureHeaderCacheAll & 0x1) != 0;
         public fixed uint Reserved1214[29];
         public uint InvalidateTextureDataCacheNoWfi;
-        public InvalidateCacheLines InvalidateTextureDataCacheNoWfiLines => (InvalidateCacheLines)((InvalidateTextureDataCacheNoWfi >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateTextureDataCacheNoWfiLines => (InvalidateCacheLines)(InvalidateTextureDataCacheNoWfi & 0x1);
         public int InvalidateTextureDataCacheNoWfiTag => (int)((InvalidateTextureDataCacheNoWfi >> 4) & 0x3FFFFF);
         public fixed uint Reserved128C[7];
         public uint ActivatePerfSettingsForComputeContext;
         public bool ActivatePerfSettingsForComputeContextAll => (ActivatePerfSettingsForComputeContext & 0x1) != 0;
         public fixed uint Reserved12AC[33];
         public uint InvalidateSamplerCache;
-        public InvalidateCacheLines InvalidateSamplerCacheLines => (InvalidateCacheLines)((InvalidateSamplerCache >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateSamplerCacheLines => (InvalidateCacheLines)(InvalidateSamplerCache & 0x1);
         public int InvalidateSamplerCacheTag => (int)((InvalidateSamplerCache >> 4) & 0x3FFFFF);
         public uint InvalidateTextureHeaderCache;
-        public InvalidateCacheLines InvalidateTextureHeaderCacheLines => (InvalidateCacheLines)((InvalidateTextureHeaderCache >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateTextureHeaderCacheLines => (InvalidateCacheLines)(InvalidateTextureHeaderCache & 0x1);
         public int InvalidateTextureHeaderCacheTag => (int)((InvalidateTextureHeaderCache >> 4) & 0x3FFFFF);
         public uint InvalidateTextureDataCache;
-        public InvalidateCacheLines InvalidateTextureDataCacheLines => (InvalidateCacheLines)((InvalidateTextureDataCache >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateTextureDataCacheLines => (InvalidateCacheLines)(InvalidateTextureDataCache & 0x1);
         public int InvalidateTextureDataCacheTag => (int)((InvalidateTextureDataCache >> 4) & 0x3FFFFF);
         public fixed uint Reserved133C[58];
         public uint InvalidateSamplerCacheNoWfi;
-        public InvalidateCacheLines InvalidateSamplerCacheNoWfiLines => (InvalidateCacheLines)((InvalidateSamplerCacheNoWfi >> 0) & 0x1);
+        public InvalidateCacheLines InvalidateSamplerCacheNoWfiLines => (InvalidateCacheLines)(InvalidateSamplerCacheNoWfi & 0x1);
         public int InvalidateSamplerCacheNoWfiTag => (int)((InvalidateSamplerCacheNoWfi >> 4) & 0x3FFFFF);
         public fixed uint Reserved1428[64];
         public uint SetShaderExceptions;
         public bool SetShaderExceptionsEnable => (SetShaderExceptions & 0x1) != 0;
         public fixed uint Reserved152C[9];
         public uint SetRenderEnableA;
-        public int SetRenderEnableAOffsetUpper => (int)((SetRenderEnableA >> 0) & 0xFF);
+        public int SetRenderEnableAOffsetUpper => (int)(SetRenderEnableA & 0xFF);
         public uint SetRenderEnableB;
         public uint SetRenderEnableC;
-        public int SetRenderEnableCMode => (int)((SetRenderEnableC >> 0) & 0x7);
+        public int SetRenderEnableCMode => (int)(SetRenderEnableC & 0x7);
         public uint SetTexSamplerPoolA;
-        public int SetTexSamplerPoolAOffsetUpper => (int)((SetTexSamplerPoolA >> 0) & 0xFF);
+        public int SetTexSamplerPoolAOffsetUpper => (int)(SetTexSamplerPoolA & 0xFF);
         public uint SetTexSamplerPoolB;
         public uint SetTexSamplerPoolC;
-        public int SetTexSamplerPoolCMaximumIndex => (int)((SetTexSamplerPoolC >> 0) & 0xFFFFF);
+        public int SetTexSamplerPoolCMaximumIndex => (int)(SetTexSamplerPoolC & 0xFFFFF);
         public fixed uint Reserved1568[3];
         public uint SetTexHeaderPoolA;
-        public int SetTexHeaderPoolAOffsetUpper => (int)((SetTexHeaderPoolA >> 0) & 0xFF);
+        public int SetTexHeaderPoolAOffsetUpper => (int)(SetTexHeaderPoolA & 0xFF);
         public uint SetTexHeaderPoolB;
         public uint SetTexHeaderPoolC;
-        public int SetTexHeaderPoolCMaximumIndex => (int)((SetTexHeaderPoolC >> 0) & 0x3FFFFF);
+        public int SetTexHeaderPoolCMaximumIndex => (int)(SetTexHeaderPoolC & 0x3FFFFF);
         public fixed uint Reserved1580[34];
         public uint SetProgramRegionA;
-        public int SetProgramRegionAAddressUpper => (int)((SetProgramRegionA >> 0) & 0xFF);
+        public int SetProgramRegionAAddressUpper => (int)(SetProgramRegionA & 0xFF);
         public uint SetProgramRegionB;
         public fixed uint Reserved1610[34];
         public uint InvalidateShaderCachesNoWfi;
@@ -374,7 +374,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public bool InvalidateShaderCachesNoWfiConstant => (InvalidateShaderCachesNoWfi & 0x1000) != 0;
         public fixed uint Reserved169C[170];
         public uint SetRenderEnableOverride;
-        public SetRenderEnableOverrideMode SetRenderEnableOverrideMode => (SetRenderEnableOverrideMode)((SetRenderEnableOverride >> 0) & 0x3);
+        public SetRenderEnableOverrideMode SetRenderEnableOverrideMode => (SetRenderEnableOverrideMode)(SetRenderEnableOverride & 0x3);
         public fixed uint Reserved1948[57];
         public uint PipeNop;
         public uint SetSpare00;
@@ -383,11 +383,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public uint SetSpare03;
         public fixed uint Reserved1A40[48];
         public uint SetReportSemaphoreA;
-        public int SetReportSemaphoreAOffsetUpper => (int)((SetReportSemaphoreA >> 0) & 0xFF);
+        public int SetReportSemaphoreAOffsetUpper => (int)(SetReportSemaphoreA & 0xFF);
         public uint SetReportSemaphoreB;
         public uint SetReportSemaphoreC;
         public uint SetReportSemaphoreD;
-        public SetReportSemaphoreDOperation SetReportSemaphoreDOperation => (SetReportSemaphoreDOperation)((SetReportSemaphoreD >> 0) & 0x3);
+        public SetReportSemaphoreDOperation SetReportSemaphoreDOperation => (SetReportSemaphoreDOperation)(SetReportSemaphoreD & 0x3);
         public bool SetReportSemaphoreDAwakenEnable => (SetReportSemaphoreD & 0x100000) != 0;
         public SetReportSemaphoreDStructureSize SetReportSemaphoreDStructureSize => (SetReportSemaphoreDStructureSize)((SetReportSemaphoreD >> 28) & 0x1);
         public bool SetReportSemaphoreDFlushDisable => (SetReportSemaphoreD & 0x4) != 0;
@@ -396,7 +396,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public SetReportSemaphoreDReductionFormat SetReportSemaphoreDReductionFormat => (SetReportSemaphoreDReductionFormat)((SetReportSemaphoreD >> 17) & 0x3);
         public fixed uint Reserved1B10[702];
         public uint SetBindlessTexture;
-        public int SetBindlessTextureConstantBufferSlotSelect => (int)((SetBindlessTexture >> 0) & 0x7);
+        public int SetBindlessTextureConstantBufferSlotSelect => (int)(SetBindlessTexture & 0x7);
         public uint SetTrapHandler;
         public fixed uint Reserved2610[843];
         public Array8<uint> SetShaderPerformanceCounterValueUpper;
@@ -423,11 +423,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
         public bool SetShaderPerformanceCounterControlBWindowed(int i) => (SetShaderPerformanceCounterControlB[i] & 0x8) != 0;
         public int SetShaderPerformanceCounterControlBFunc(int i) => (int)((SetShaderPerformanceCounterControlB[i] >> 4) & 0xFFFF);
         public uint SetShaderPerformanceCounterTrapControl;
-        public int SetShaderPerformanceCounterTrapControlMask => (int)((SetShaderPerformanceCounterTrapControl >> 0) & 0xFF);
+        public int SetShaderPerformanceCounterTrapControlMask => (int)(SetShaderPerformanceCounterTrapControl & 0xFF);
         public uint StartShaderPerformanceCounter;
-        public int StartShaderPerformanceCounterCounterMask => (int)((StartShaderPerformanceCounter >> 0) & 0xFF);
+        public int StartShaderPerformanceCounterCounterMask => (int)(StartShaderPerformanceCounter & 0xFF);
         public uint StopShaderPerformanceCounter;
-        public int StopShaderPerformanceCounterCounterMask => (int)((StopShaderPerformanceCounter >> 0) & 0xFF);
+        public int StopShaderPerformanceCounterCounterMask => (int)(StopShaderPerformanceCounter & 0xFF);
         public fixed uint Reserved33E8[6];
         public MmeShadowScratch SetMmeShadowScratch;
 #pragma warning restore CS0649

src/Ryujinx.Graphics.Gpu/Engine/Dma/DmaClassState.cs

@@ -186,22 +186,22 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public uint PmTrigger;
         public fixed uint Reserved144[63];
         public uint SetSemaphoreA;
-        public int SetSemaphoreAUpper => (int)((SetSemaphoreA >> 0) & 0xFF);
+        public int SetSemaphoreAUpper => (int)(SetSemaphoreA & 0xFF);
         public uint SetSemaphoreB;
         public uint SetSemaphorePayload;
         public fixed uint Reserved24C[2];
         public uint SetRenderEnableA;
-        public int SetRenderEnableAUpper => (int)((SetRenderEnableA >> 0) & 0xFF);
+        public int SetRenderEnableAUpper => (int)(SetRenderEnableA & 0xFF);
         public uint SetRenderEnableB;
         public uint SetRenderEnableC;
-        public int SetRenderEnableCMode => (int)((SetRenderEnableC >> 0) & 0x7);
+        public int SetRenderEnableCMode => (int)(SetRenderEnableC & 0x7);
         public uint SetSrcPhysMode;
-        public SetPhysModeTarget SetSrcPhysModeTarget => (SetPhysModeTarget)((SetSrcPhysMode >> 0) & 0x3);
+        public SetPhysModeTarget SetSrcPhysModeTarget => (SetPhysModeTarget)(SetSrcPhysMode & 0x3);
         public uint SetDstPhysMode;
-        public SetPhysModeTarget SetDstPhysModeTarget => (SetPhysModeTarget)((SetDstPhysMode >> 0) & 0x3);
+        public SetPhysModeTarget SetDstPhysModeTarget => (SetPhysModeTarget)(SetDstPhysMode & 0x3);
         public fixed uint Reserved268[38];
         public uint LaunchDma;
-        public LaunchDmaDataTransferType LaunchDmaDataTransferType => (LaunchDmaDataTransferType)((LaunchDma >> 0) & 0x3);
+        public LaunchDmaDataTransferType LaunchDmaDataTransferType => (LaunchDmaDataTransferType)(LaunchDma & 0x3);
         public bool LaunchDmaFlushEnable => (LaunchDma & 0x4) != 0;
         public LaunchDmaSemaphoreType LaunchDmaSemaphoreType => (LaunchDmaSemaphoreType)((LaunchDma >> 3) & 0x3);
         public LaunchDmaInterruptType LaunchDmaInterruptType => (LaunchDmaInterruptType)((LaunchDma >> 5) & 0x3);
@@ -218,10 +218,10 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public LaunchDmaBypassL2 LaunchDmaBypassL2 => (LaunchDmaBypassL2)((LaunchDma >> 20) & 0x1);
         public fixed uint Reserved304[63];
         public uint OffsetInUpper;
-        public int OffsetInUpperUpper => (int)((OffsetInUpper >> 0) & 0xFF);
+        public int OffsetInUpperUpper => (int)(OffsetInUpper & 0xFF);
         public uint OffsetInLower;
         public uint OffsetOutUpper;
-        public int OffsetOutUpperUpper => (int)((OffsetOutUpper >> 0) & 0xFF);
+        public int OffsetOutUpperUpper => (int)(OffsetOutUpper & 0xFF);
         public uint OffsetOutLower;
         public uint PitchIn;
         public uint PitchOut;
@@ -231,7 +231,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public uint SetRemapConstA;
         public uint SetRemapConstB;
         public uint SetRemapComponents;
-        public SetRemapComponentsDst SetRemapComponentsDstX => (SetRemapComponentsDst)((SetRemapComponents >> 0) & 0x7);
+        public SetRemapComponentsDst SetRemapComponentsDstX => (SetRemapComponentsDst)(SetRemapComponents & 0x7);
         public SetRemapComponentsDst SetRemapComponentsDstY => (SetRemapComponentsDst)((SetRemapComponents >> 4) & 0x7);
         public SetRemapComponentsDst SetRemapComponentsDstZ => (SetRemapComponentsDst)((SetRemapComponents >> 8) & 0x7);
         public SetRemapComponentsDst SetRemapComponentsDstW => (SetRemapComponentsDst)((SetRemapComponents >> 12) & 0x7);
@@ -239,7 +239,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public SetRemapComponentsNumComponents SetRemapComponentsNumSrcComponents => (SetRemapComponentsNumComponents)((SetRemapComponents >> 20) & 0x3);
         public SetRemapComponentsNumComponents SetRemapComponentsNumDstComponents => (SetRemapComponentsNumComponents)((SetRemapComponents >> 24) & 0x3);
         public uint SetDstBlockSize;
-        public SetBlockSizeWidth SetDstBlockSizeWidth => (SetBlockSizeWidth)((SetDstBlockSize >> 0) & 0xF);
+        public SetBlockSizeWidth SetDstBlockSizeWidth => (SetBlockSizeWidth)(SetDstBlockSize & 0xF);
         public SetBlockSizeHeight SetDstBlockSizeHeight => (SetBlockSizeHeight)((SetDstBlockSize >> 4) & 0xF);
         public SetBlockSizeDepth SetDstBlockSizeDepth => (SetBlockSizeDepth)((SetDstBlockSize >> 8) & 0xF);
         public SetBlockSizeGobHeight SetDstBlockSizeGobHeight => (SetBlockSizeGobHeight)((SetDstBlockSize >> 12) & 0xF);
@@ -248,11 +248,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public uint SetDstDepth;
         public uint SetDstLayer;
         public uint SetDstOrigin;
-        public int SetDstOriginX => (int)((SetDstOrigin >> 0) & 0xFFFF);
+        public int SetDstOriginX => (int)(SetDstOrigin & 0xFFFF);
         public int SetDstOriginY => (int)((SetDstOrigin >> 16) & 0xFFFF);
         public uint Reserved724;
         public uint SetSrcBlockSize;
-        public SetBlockSizeWidth SetSrcBlockSizeWidth => (SetBlockSizeWidth)((SetSrcBlockSize >> 0) & 0xF);
+        public SetBlockSizeWidth SetSrcBlockSizeWidth => (SetBlockSizeWidth)(SetSrcBlockSize & 0xF);
         public SetBlockSizeHeight SetSrcBlockSizeHeight => (SetBlockSizeHeight)((SetSrcBlockSize >> 4) & 0xF);
         public SetBlockSizeDepth SetSrcBlockSizeDepth => (SetBlockSizeDepth)((SetSrcBlockSize >> 8) & 0xF);
         public SetBlockSizeGobHeight SetSrcBlockSizeGobHeight => (SetBlockSizeGobHeight)((SetSrcBlockSize >> 12) & 0xF);
@@ -261,7 +261,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
         public uint SetSrcDepth;
         public uint SetSrcLayer;
         public uint SetSrcOrigin;
-        public int SetSrcOriginX => (int)((SetSrcOrigin >> 0) & 0xFFFF);
+        public int SetSrcOriginX => (int)(SetSrcOrigin & 0xFFFF);
         public int SetSrcOriginY => (int)((SetSrcOrigin >> 16) & 0xFFFF);
         public fixed uint Reserved740[629];
         public uint PmTriggerEnd;

src/Ryujinx.Graphics.Gpu/Engine/GPFifo/CompressedMethod.cs

@@ -29,7 +29,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
         public uint Method;
 #pragma warning restore CS0649
         public int MethodAddressOld => (int)((Method >> 2) & 0x7FF);
-        public int MethodAddress => (int)((Method >> 0) & 0xFFF);
+        public int MethodAddress => (int)(Method & 0xFFF);
         public int SubdeviceMask => (int)((Method >> 4) & 0xFFF);
         public int MethodSubchannel => (int)((Method >> 13) & 0x7);
         public TertOp TertOp => (TertOp)((Method >> 16) & 0x3);

src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPEntry.cs

@@ -39,17 +39,17 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
 #pragma warning disable CS0649
         public uint Entry0;
 #pragma warning restore CS0649
-        public Entry0Fetch Entry0Fetch => (Entry0Fetch)((Entry0 >> 0) & 0x1);
+        public Entry0Fetch Entry0Fetch => (Entry0Fetch)(Entry0 & 0x1);
         public int Entry0Get => (int)((Entry0 >> 2) & 0x3FFFFFFF);
         public int Entry0Operand => (int)(Entry0);
 #pragma warning disable CS0649
         public uint Entry1;
 #pragma warning restore CS0649
-        public int Entry1GetHi => (int)((Entry1 >> 0) & 0xFF);
+        public int Entry1GetHi => (int)(Entry1 & 0xFF);
         public Entry1Priv Entry1Priv => (Entry1Priv)((Entry1 >> 8) & 0x1);
         public Entry1Level Entry1Level => (Entry1Level)((Entry1 >> 9) & 0x1);
         public int Entry1Length => (int)((Entry1 >> 10) & 0x1FFFFF);
         public Entry1Sync Entry1Sync => (Entry1Sync)((Entry1 >> 31) & 0x1);
-        public Entry1Opcode Entry1Opcode => (Entry1Opcode)((Entry1 >> 0) & 0xFF);
+        public Entry1Opcode Entry1Opcode => (Entry1Opcode)(Entry1 & 0xFF);
     }
 }

src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoClassState.cs

@@ -153,7 +153,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
     {
 #pragma warning disable CS0649
         public uint SetObject;
-        public int SetObjectNvclass => (int)((SetObject >> 0) & 0xFFFF);
+        public int SetObjectNvclass => (int)(SetObject & 0xFFFF);
         public int SetObjectEngine => (int)((SetObject >> 16) & 0x1F);
         public uint Illegal;
         public int IllegalHandle => (int)(Illegal);
@@ -161,13 +161,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
         public int NopHandle => (int)(Nop);
         public uint Reserved0C;
         public uint Semaphorea;
-        public int SemaphoreaOffsetUpper => (int)((Semaphorea >> 0) & 0xFF);
+        public int SemaphoreaOffsetUpper => (int)(Semaphorea & 0xFF);
         public uint Semaphoreb;
         public int SemaphorebOffsetLower => (int)((Semaphoreb >> 2) & 0x3FFFFFFF);
         public uint Semaphorec;
         public int SemaphorecPayload => (int)(Semaphorec);
         public uint Semaphored;
-        public SemaphoredOperation SemaphoredOperation => (SemaphoredOperation)((Semaphored >> 0) & 0x1F);
+        public SemaphoredOperation SemaphoredOperation => (SemaphoredOperation)(Semaphored & 0x1F);
         public SemaphoredAcquireSwitch SemaphoredAcquireSwitch => (SemaphoredAcquireSwitch)((Semaphored >> 12) & 0x1);
         public SemaphoredReleaseWfi SemaphoredReleaseWfi => (SemaphoredReleaseWfi)((Semaphored >> 20) & 0x1);
         public SemaphoredReleaseSize SemaphoredReleaseSize => (SemaphoredReleaseSize)((Semaphored >> 24) & 0x1);
@@ -181,14 +181,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
         public uint Reserved2C;
         public uint MemOpC;
         public int MemOpCOperandLow => (int)((MemOpC >> 2) & 0x3FFFFFFF);
-        public MemOpCTlbInvalidatePdb MemOpCTlbInvalidatePdb => (MemOpCTlbInvalidatePdb)((MemOpC >> 0) & 0x1);
+        public MemOpCTlbInvalidatePdb MemOpCTlbInvalidatePdb => (MemOpCTlbInvalidatePdb)(MemOpC & 0x1);
         public MemOpCTlbInvalidateGpc MemOpCTlbInvalidateGpc => (MemOpCTlbInvalidateGpc)((MemOpC >> 1) & 0x1);
         public MemOpCTlbInvalidateTarget MemOpCTlbInvalidateTarget => (MemOpCTlbInvalidateTarget)((MemOpC >> 10) & 0x3);
         public int MemOpCTlbInvalidateAddrLo => (int)((MemOpC >> 12) & 0xFFFFF);
         public uint MemOpD;
-        public int MemOpDOperandHigh => (int)((MemOpD >> 0) & 0xFF);
+        public int MemOpDOperandHigh => (int)(MemOpD & 0xFF);
         public MemOpDOperation MemOpDOperation => (MemOpDOperation)((MemOpD >> 27) & 0x1F);
-        public int MemOpDTlbInvalidateAddrHi => (int)((MemOpD >> 0) & 0xFF);
+        public int MemOpDTlbInvalidateAddrHi => (int)(MemOpD & 0xFF);
         public uint Reserved38;
         public uint Reserved3C;
         public uint Reserved40;
@@ -207,15 +207,15 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
         public uint Syncpointa;
         public int SyncpointaPayload => (int)(Syncpointa);
         public uint Syncpointb;
-        public SyncpointbOperation SyncpointbOperation => (SyncpointbOperation)((Syncpointb >> 0) & 0x1);
+        public SyncpointbOperation SyncpointbOperation => (SyncpointbOperation)(Syncpointb & 0x1);
         public SyncpointbWaitSwitch SyncpointbWaitSwitch => (SyncpointbWaitSwitch)((Syncpointb >> 4) & 0x1);
         public int SyncpointbSyncptIndex => (int)((Syncpointb >> 8) & 0xFFF);
         public uint Wfi;
-        public WfiScope WfiScope => (WfiScope)((Wfi >> 0) & 0x1);
+        public WfiScope WfiScope => (WfiScope)(Wfi & 0x1);
         public uint CrcCheck;
         public int CrcCheckValue => (int)(CrcCheck);
         public uint Yield;
-        public YieldOp YieldOp => (YieldOp)((Yield >> 0) & 0x3);
+        public YieldOp YieldOp => (YieldOp)(Yield & 0x3);
         // TODO: Eventually move this to per-engine state.
         public Array31<uint> Reserved84;
         public uint NoOperation;

src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClassState.cs

@@ -113,22 +113,22 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
     {
 #pragma warning disable CS0649
         public uint SetObject;
-        public int SetObjectClassId => (int)((SetObject >> 0) & 0xFFFF);
+        public int SetObjectClassId => (int)(SetObject & 0xFFFF);
         public int SetObjectEngineId => (int)((SetObject >> 16) & 0x1F);
         public fixed uint Reserved04[63];
         public uint NoOperation;
         public uint SetNotifyA;
-        public int SetNotifyAAddressUpper => (int)((SetNotifyA >> 0) & 0xFF);
+        public int SetNotifyAAddressUpper => (int)(SetNotifyA & 0xFF);
         public uint SetNotifyB;
         public uint Notify;
         public NotifyType NotifyType => (NotifyType)(Notify);
         public uint WaitForIdle;
         public fixed uint Reserved114[7];
         public uint SetGlobalRenderEnableA;
-        public int SetGlobalRenderEnableAOffsetUpper => (int)((SetGlobalRenderEnableA >> 0) & 0xFF);
+        public int SetGlobalRenderEnableAOffsetUpper => (int)(SetGlobalRenderEnableA & 0xFF);
         public uint SetGlobalRenderEnableB;
         public uint SetGlobalRenderEnableC;
-        public int SetGlobalRenderEnableCMode => (int)((SetGlobalRenderEnableC >> 0) & 0x7);
+        public int SetGlobalRenderEnableCMode => (int)(SetGlobalRenderEnableC & 0x7);
         public uint SendGoIdle;
         public uint PmTrigger;
         public uint PmTriggerWfi;
@@ -139,11 +139,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
         public uint LineLengthIn;
         public uint LineCount;
         public uint OffsetOutUpper;
-        public int OffsetOutUpperValue => (int)((OffsetOutUpper >> 0) & 0xFF);
+        public int OffsetOutUpperValue => (int)(OffsetOutUpper & 0xFF);
         public uint OffsetOut;
         public uint PitchOut;
         public uint SetDstBlockSize;
-        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)((SetDstBlockSize >> 0) & 0xF);
+        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)(SetDstBlockSize & 0xF);
         public SetDstBlockSizeHeight SetDstBlockSizeHeight => (SetDstBlockSizeHeight)((SetDstBlockSize >> 4) & 0xF);
         public SetDstBlockSizeDepth SetDstBlockSizeDepth => (SetDstBlockSizeDepth)((SetDstBlockSize >> 8) & 0xF);
         public uint SetDstWidth;
@@ -151,11 +151,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
         public uint SetDstDepth;
         public uint SetDstLayer;
         public uint SetDstOriginBytesX;
-        public int SetDstOriginBytesXV => (int)((SetDstOriginBytesX >> 0) & 0xFFFFF);
+        public int SetDstOriginBytesXV => (int)(SetDstOriginBytesX & 0xFFFFF);
         public uint SetDstOriginSamplesY;
-        public int SetDstOriginSamplesYV => (int)((SetDstOriginSamplesY >> 0) & 0xFFFF);
+        public int SetDstOriginSamplesYV => (int)(SetDstOriginSamplesY & 0xFFFF);
         public uint LaunchDma;
-        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)((LaunchDma >> 0) & 0x1);
+        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)(LaunchDma & 0x1);
         public LaunchDmaCompletionType LaunchDmaCompletionType => (LaunchDmaCompletionType)((LaunchDma >> 4) & 0x3);
         public LaunchDmaInterruptType LaunchDmaInterruptType => (LaunchDmaInterruptType)((LaunchDma >> 8) & 0x3);
         public LaunchDmaSemaphoreStructSize LaunchDmaSemaphoreStructSize => (LaunchDmaSemaphoreStructSize)((LaunchDma >> 12) & 0x1);
@@ -166,7 +166,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
         public uint LoadInlineData;
         public fixed uint Reserved1B8[9];
         public uint SetI2mSemaphoreA;
-        public int SetI2mSemaphoreAOffsetUpper => (int)((SetI2mSemaphoreA >> 0) & 0xFF);
+        public int SetI2mSemaphoreAOffsetUpper => (int)(SetI2mSemaphoreA & 0xFF);
         public uint SetI2mSemaphoreB;
         public uint SetI2mSemaphoreC;
         public fixed uint Reserved1E8[2];

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

@@ -539,6 +539,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             engine.UpdateState();
 
+            if (instanceCount > 1)
+            {
+                // Must be called after UpdateState as it assumes the shader state
+                // has already been set, and that bindings have been updated already.
+
+                _channel.BufferManager.SetInstancedDrawVertexCount(count);
+            }
+
             if (indexed)
             {
                 _context.Renderer.Pipeline.DrawIndexed(count, instanceCount, firstIndex, firstVertex, firstInstance);
@@ -676,6 +684,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                         _channel.BufferManager.SetIndexBuffer(br, IndexType.UInt);
                     }
 
+                    _channel.BufferManager.SetInstancedDrawVertexCount(_instancedIndexCount);
+
                     _context.Renderer.Pipeline.DrawIndexed(
                         _instancedIndexCount,
                         _instanceIndex + 1,
@@ -685,6 +695,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 }
                 else
                 {
+                    _channel.BufferManager.SetInstancedDrawVertexCount(_instancedDrawStateCount);
+
                     _context.Renderer.Pipeline.Draw(
                         _instancedDrawStateCount,
                         _instanceIndex + 1,

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

@@ -269,7 +269,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _prevFirstVertex = _state.State.FirstVertex;
             }
 
-            bool tfEnable = _state.State.TfEnable;
+            bool tfEnable = _state.State.TfEnable && _context.Capabilities.SupportsTransformFeedback;
 
             if (!tfEnable && _prevTfEnable)
             {
@@ -1367,6 +1367,22 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             _vsUsesDrawParameters = gs.Shaders[1]?.Info.UsesDrawParameters ?? false;
             _vsClipDistancesWritten = gs.Shaders[1]?.Info.ClipDistancesWritten ?? 0;
 
+            bool hasTransformFeedback = gs.SpecializationState.TransformFeedbackDescriptors != null;
+            if (hasTransformFeedback != _channel.BufferManager.HasTransformFeedbackOutputs)
+            {
+                if (!_context.Capabilities.SupportsTransformFeedback)
+                {
+                    // If host does not support transform feedback, and the shader changed,
+                    // we might need to update bindings as transform feedback emulation
+                    // uses storage buffer bindings that might have been used for something
+                    // else in a previous draw.
+
+                    _channel.BufferManager.ForceTransformFeedbackAndStorageBuffersDirty();
+                }
+
+                _channel.BufferManager.HasTransformFeedbackOutputs = hasTransformFeedback;
+            }
+
             if (oldVsClipDistancesWritten != _vsClipDistancesWritten)
             {
                 UpdateUserClipState();

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

@@ -746,12 +746,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
     {
 #pragma warning disable CS0649
         public uint SetObject;
-        public int SetObjectClassId => (int)((SetObject >> 0) & 0xFFFF);
+        public int SetObjectClassId => (int)(SetObject & 0xFFFF);
         public int SetObjectEngineId => (int)((SetObject >> 16) & 0x1F);
         public fixed uint Reserved04[63];
         public uint NoOperation;
         public uint SetNotifyA;
-        public int SetNotifyAAddressUpper => (int)((SetNotifyA >> 0) & 0xFF);
+        public int SetNotifyAAddressUpper => (int)(SetNotifyA & 0xFF);
         public uint SetNotifyB;
         public uint Notify;
         public NotifyType NotifyType => (NotifyType)(Notify);
@@ -761,13 +761,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         public uint LoadMmeStartAddressRamPointer;
         public uint LoadMmeStartAddressRam;
         public uint SetMmeShadowRamControl;
-        public SetMmeShadowRamControlMode SetMmeShadowRamControlMode => (SetMmeShadowRamControlMode)((SetMmeShadowRamControl >> 0) & 0x3);
+        public SetMmeShadowRamControlMode SetMmeShadowRamControlMode => (SetMmeShadowRamControlMode)(SetMmeShadowRamControl & 0x3);
         public fixed uint Reserved128[2];
         public uint SetGlobalRenderEnableA;
-        public int SetGlobalRenderEnableAOffsetUpper => (int)((SetGlobalRenderEnableA >> 0) & 0xFF);
+        public int SetGlobalRenderEnableAOffsetUpper => (int)(SetGlobalRenderEnableA & 0xFF);
         public uint SetGlobalRenderEnableB;
         public uint SetGlobalRenderEnableC;
-        public int SetGlobalRenderEnableCMode => (int)((SetGlobalRenderEnableC >> 0) & 0x7);
+        public int SetGlobalRenderEnableCMode => (int)(SetGlobalRenderEnableC & 0x7);
         public uint SendGoIdle;
         public uint PmTrigger;
         public uint PmTriggerWfi;
@@ -778,11 +778,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         public uint LineLengthIn;
         public uint LineCount;
         public uint OffsetOutUpper;
-        public int OffsetOutUpperValue => (int)((OffsetOutUpper >> 0) & 0xFF);
+        public int OffsetOutUpperValue => (int)(OffsetOutUpper & 0xFF);
         public uint OffsetOut;
         public uint PitchOut;
         public uint SetDstBlockSize;
-        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)((SetDstBlockSize >> 0) & 0xF);
+        public SetDstBlockSizeWidth SetDstBlockSizeWidth => (SetDstBlockSizeWidth)(SetDstBlockSize & 0xF);
         public SetDstBlockSizeHeight SetDstBlockSizeHeight => (SetDstBlockSizeHeight)((SetDstBlockSize >> 4) & 0xF);
         public SetDstBlockSizeDepth SetDstBlockSizeDepth => (SetDstBlockSizeDepth)((SetDstBlockSize >> 8) & 0xF);
         public uint SetDstWidth;
@@ -790,11 +790,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         public uint SetDstDepth;
         public uint SetDstLayer;
         public uint SetDstOriginBytesX;
-        public int SetDstOriginBytesXV => (int)((SetDstOriginBytesX >> 0) & 0xFFFFF);
+        public int SetDstOriginBytesXV => (int)(SetDstOriginBytesX & 0xFFFFF);
         public uint SetDstOriginSamplesY;
-        public int SetDstOriginSamplesYV => (int)((SetDstOriginSamplesY >> 0) & 0xFFFF);
+        public int SetDstOriginSamplesYV => (int)(SetDstOriginSamplesY & 0xFFFF);
         public uint LaunchDma;
-        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)((LaunchDma >> 0) & 0x1);
+        public LaunchDmaDstMemoryLayout LaunchDmaDstMemoryLayout => (LaunchDmaDstMemoryLayout)(LaunchDma & 0x1);
         public LaunchDmaCompletionType LaunchDmaCompletionType => (LaunchDmaCompletionType)((LaunchDma >> 4) & 0x3);
         public LaunchDmaInterruptType LaunchDmaInterruptType => (LaunchDmaInterruptType)((LaunchDma >> 8) & 0x3);
         public LaunchDmaSemaphoreStructSize LaunchDmaSemaphoreStructSize => (LaunchDmaSemaphoreStructSize)((LaunchDma >> 12) & 0x1);

src/Ryujinx.Graphics.Gpu/Engine/Twod/TwodClassState.cs

@@ -499,12 +499,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
     {
 #pragma warning disable CS0649
         public uint SetObject;
-        public int SetObjectClassId => (int)((SetObject >> 0) & 0xFFFF);
+        public int SetObjectClassId => (int)(SetObject & 0xFFFF);
         public int SetObjectEngineId => (int)((SetObject >> 16) & 0x1F);
         public fixed uint Reserved04[63];
         public uint NoOperation;
         public uint SetNotifyA;
-        public int SetNotifyAAddressUpper => (int)((SetNotifyA >> 0) & 0x1FFFFFF);
+        public int SetNotifyAAddressUpper => (int)(SetNotifyA & 0x1FFFFFF);
         public uint SetNotifyB;
         public uint Notify;
         public NotifyType NotifyType => (NotifyType)(Notify);
@@ -514,13 +514,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public uint LoadMmeStartAddressRamPointer;
         public uint LoadMmeStartAddressRam;
         public uint SetMmeShadowRamControl;
-        public SetMmeShadowRamControlMode SetMmeShadowRamControlMode => (SetMmeShadowRamControlMode)((SetMmeShadowRamControl >> 0) & 0x3);
+        public SetMmeShadowRamControlMode SetMmeShadowRamControlMode => (SetMmeShadowRamControlMode)(SetMmeShadowRamControl & 0x3);
         public fixed uint Reserved128[2];
         public uint SetGlobalRenderEnableA;
-        public int SetGlobalRenderEnableAOffsetUpper => (int)((SetGlobalRenderEnableA >> 0) & 0xFF);
+        public int SetGlobalRenderEnableAOffsetUpper => (int)(SetGlobalRenderEnableA & 0xFF);
         public uint SetGlobalRenderEnableB;
         public uint SetGlobalRenderEnableC;
-        public int SetGlobalRenderEnableCMode => (int)((SetGlobalRenderEnableC >> 0) & 0x7);
+        public int SetGlobalRenderEnableCMode => (int)(SetGlobalRenderEnableC & 0x7);
         public uint SendGoIdle;
         public uint PmTrigger;
         public fixed uint Reserved144[3];
@@ -533,9 +533,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public int SetMmeSwitchStateRestoreMacro => (int)((SetMmeSwitchState >> 12) & 0xFF);
         public fixed uint Reserved1F0[4];
         public uint SetDstFormat;
-        public SetDstFormatV SetDstFormatV => (SetDstFormatV)((SetDstFormat >> 0) & 0xFF);
+        public SetDstFormatV SetDstFormatV => (SetDstFormatV)(SetDstFormat & 0xFF);
         public uint SetDstMemoryLayout;
-        public SetDstMemoryLayoutV SetDstMemoryLayoutV => (SetDstMemoryLayoutV)((SetDstMemoryLayout >> 0) & 0x1);
+        public SetDstMemoryLayoutV SetDstMemoryLayoutV => (SetDstMemoryLayoutV)(SetDstMemoryLayout & 0x1);
         public uint SetDstBlockSize;
         public SetDstBlockSizeHeight SetDstBlockSizeHeight => (SetDstBlockSizeHeight)((SetDstBlockSize >> 4) & 0x7);
         public SetDstBlockSizeDepth SetDstBlockSizeDepth => (SetDstBlockSizeDepth)((SetDstBlockSize >> 8) & 0x7);
@@ -545,37 +545,37 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public uint SetDstWidth;
         public uint SetDstHeight;
         public uint SetDstOffsetUpper;
-        public int SetDstOffsetUpperV => (int)((SetDstOffsetUpper >> 0) & 0xFF);
+        public int SetDstOffsetUpperV => (int)(SetDstOffsetUpper & 0xFF);
         public uint SetDstOffsetLower;
         public uint FlushAndInvalidateRopMiniCache;
         public bool FlushAndInvalidateRopMiniCacheV => (FlushAndInvalidateRopMiniCache & 0x1) != 0;
         public uint SetSpareNoop06;
         public uint SetSrcFormat;
-        public SetSrcFormatV SetSrcFormatV => (SetSrcFormatV)((SetSrcFormat >> 0) & 0xFF);
+        public SetSrcFormatV SetSrcFormatV => (SetSrcFormatV)(SetSrcFormat & 0xFF);
         public uint SetSrcMemoryLayout;
-        public SetSrcMemoryLayoutV SetSrcMemoryLayoutV => (SetSrcMemoryLayoutV)((SetSrcMemoryLayout >> 0) & 0x1);
+        public SetSrcMemoryLayoutV SetSrcMemoryLayoutV => (SetSrcMemoryLayoutV)(SetSrcMemoryLayout & 0x1);
         public uint SetSrcBlockSize;
         public SetSrcBlockSizeHeight SetSrcBlockSizeHeight => (SetSrcBlockSizeHeight)((SetSrcBlockSize >> 4) & 0x7);
         public SetSrcBlockSizeDepth SetSrcBlockSizeDepth => (SetSrcBlockSizeDepth)((SetSrcBlockSize >> 8) & 0x7);
         public uint SetSrcDepth;
         public uint TwodInvalidateTextureDataCache;
-        public TwodInvalidateTextureDataCacheV TwodInvalidateTextureDataCacheV => (TwodInvalidateTextureDataCacheV)((TwodInvalidateTextureDataCache >> 0) & 0x3);
+        public TwodInvalidateTextureDataCacheV TwodInvalidateTextureDataCacheV => (TwodInvalidateTextureDataCacheV)(TwodInvalidateTextureDataCache & 0x3);
         public uint SetSrcPitch;
         public uint SetSrcWidth;
         public uint SetSrcHeight;
         public uint SetSrcOffsetUpper;
-        public int SetSrcOffsetUpperV => (int)((SetSrcOffsetUpper >> 0) & 0xFF);
+        public int SetSrcOffsetUpperV => (int)(SetSrcOffsetUpper & 0xFF);
         public uint SetSrcOffsetLower;
         public uint SetPixelsFromMemorySectorPromotion;
-        public SetPixelsFromMemorySectorPromotionV SetPixelsFromMemorySectorPromotionV => (SetPixelsFromMemorySectorPromotionV)((SetPixelsFromMemorySectorPromotion >> 0) & 0x3);
+        public SetPixelsFromMemorySectorPromotionV SetPixelsFromMemorySectorPromotionV => (SetPixelsFromMemorySectorPromotionV)(SetPixelsFromMemorySectorPromotion & 0x3);
         public uint SetSpareNoop12;
         public uint SetNumProcessingClusters;
-        public SetNumProcessingClustersV SetNumProcessingClustersV => (SetNumProcessingClustersV)((SetNumProcessingClusters >> 0) & 0x1);
+        public SetNumProcessingClustersV SetNumProcessingClustersV => (SetNumProcessingClustersV)(SetNumProcessingClusters & 0x1);
         public uint SetRenderEnableA;
-        public int SetRenderEnableAOffsetUpper => (int)((SetRenderEnableA >> 0) & 0xFF);
+        public int SetRenderEnableAOffsetUpper => (int)(SetRenderEnableA & 0xFF);
         public uint SetRenderEnableB;
         public uint SetRenderEnableC;
-        public int SetRenderEnableCMode => (int)((SetRenderEnableC >> 0) & 0x7);
+        public int SetRenderEnableCMode => (int)(SetRenderEnableC & 0x7);
         public uint SetSpareNoop08;
         public uint SetSpareNoop01;
         public uint SetSpareNoop11;
@@ -587,25 +587,25 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public uint SetClipEnable;
         public bool SetClipEnableV => (SetClipEnable & 0x1) != 0;
         public uint SetColorKeyFormat;
-        public SetColorKeyFormatV SetColorKeyFormatV => (SetColorKeyFormatV)((SetColorKeyFormat >> 0) & 0x7);
+        public SetColorKeyFormatV SetColorKeyFormatV => (SetColorKeyFormatV)(SetColorKeyFormat & 0x7);
         public uint SetColorKey;
         public uint SetColorKeyEnable;
         public bool SetColorKeyEnableV => (SetColorKeyEnable & 0x1) != 0;
         public uint SetRop;
-        public int SetRopV => (int)((SetRop >> 0) & 0xFF);
+        public int SetRopV => (int)(SetRop & 0xFF);
         public uint SetBeta1;
         public uint SetBeta4;
-        public int SetBeta4B => (int)((SetBeta4 >> 0) & 0xFF);
+        public int SetBeta4B => (int)(SetBeta4 & 0xFF);
         public int SetBeta4G => (int)((SetBeta4 >> 8) & 0xFF);
         public int SetBeta4R => (int)((SetBeta4 >> 16) & 0xFF);
         public int SetBeta4A => (int)((SetBeta4 >> 24) & 0xFF);
         public uint SetOperation;
-        public SetOperationV SetOperationV => (SetOperationV)((SetOperation >> 0) & 0x7);
+        public SetOperationV SetOperationV => (SetOperationV)(SetOperation & 0x7);
         public uint SetPatternOffset;
-        public int SetPatternOffsetX => (int)((SetPatternOffset >> 0) & 0x3F);
+        public int SetPatternOffsetX => (int)(SetPatternOffset & 0x3F);
         public int SetPatternOffsetY => (int)((SetPatternOffset >> 8) & 0x3F);
         public uint SetPatternSelect;
-        public SetPatternSelectV SetPatternSelectV => (SetPatternSelectV)((SetPatternSelect >> 0) & 0x3);
+        public SetPatternSelectV SetPatternSelectV => (SetPatternSelectV)(SetPatternSelect & 0x3);
         public uint SetDstColorRenderToZetaSurface;
         public bool SetDstColorRenderToZetaSurfaceV => (SetDstColorRenderToZetaSurface & 0x1) != 0;
         public uint SetSpareNoop04;
@@ -618,15 +618,15 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public bool SetCompressionEnable => (SetCompression & 0x1) != 0;
         public uint SetSpareNoop09;
         public uint SetRenderEnableOverride;
-        public SetRenderEnableOverrideMode SetRenderEnableOverrideMode => (SetRenderEnableOverrideMode)((SetRenderEnableOverride >> 0) & 0x3);
+        public SetRenderEnableOverrideMode SetRenderEnableOverrideMode => (SetRenderEnableOverrideMode)(SetRenderEnableOverride & 0x3);
         public uint SetPixelsFromMemoryDirection;
-        public SetPixelsFromMemoryDirectionHorizontal SetPixelsFromMemoryDirectionHorizontal => (SetPixelsFromMemoryDirectionHorizontal)((SetPixelsFromMemoryDirection >> 0) & 0x3);
+        public SetPixelsFromMemoryDirectionHorizontal SetPixelsFromMemoryDirectionHorizontal => (SetPixelsFromMemoryDirectionHorizontal)(SetPixelsFromMemoryDirection & 0x3);
         public SetPixelsFromMemoryDirectionVertical SetPixelsFromMemoryDirectionVertical => (SetPixelsFromMemoryDirectionVertical)((SetPixelsFromMemoryDirection >> 4) & 0x3);
         public uint SetSpareNoop10;
         public uint SetMonochromePatternColorFormat;
-        public SetMonochromePatternColorFormatV SetMonochromePatternColorFormatV => (SetMonochromePatternColorFormatV)((SetMonochromePatternColorFormat >> 0) & 0x7);
+        public SetMonochromePatternColorFormatV SetMonochromePatternColorFormatV => (SetMonochromePatternColorFormatV)(SetMonochromePatternColorFormat & 0x7);
         public uint SetMonochromePatternFormat;
-        public SetMonochromePatternFormatV SetMonochromePatternFormatV => (SetMonochromePatternFormatV)((SetMonochromePatternFormat >> 0) & 0x1);
+        public SetMonochromePatternFormatV SetMonochromePatternFormatV => (SetMonochromePatternFormatV)(SetMonochromePatternFormat & 0x1);
         public uint SetMonochromePatternColor0;
         public uint SetMonochromePatternColor1;
         public uint SetMonochromePattern0;
@@ -662,26 +662,26 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public uint SetRenderSolidPrimColor2;
         public uint SetRenderSolidPrimColor3;
         public uint SetMmeMemAddressA;
-        public int SetMmeMemAddressAUpper => (int)((SetMmeMemAddressA >> 0) & 0x1FFFFFF);
+        public int SetMmeMemAddressAUpper => (int)(SetMmeMemAddressA & 0x1FFFFFF);
         public uint SetMmeMemAddressB;
         public uint SetMmeDataRamAddress;
         public uint MmeDmaRead;
         public uint MmeDmaReadFifoed;
         public uint MmeDmaWrite;
         public uint MmeDmaReduction;
-        public MmeDmaReductionReductionOp MmeDmaReductionReductionOp => (MmeDmaReductionReductionOp)((MmeDmaReduction >> 0) & 0x7);
+        public MmeDmaReductionReductionOp MmeDmaReductionReductionOp => (MmeDmaReductionReductionOp)(MmeDmaReduction & 0x7);
         public MmeDmaReductionReductionFormat MmeDmaReductionReductionFormat => (MmeDmaReductionReductionFormat)((MmeDmaReduction >> 4) & 0x3);
         public MmeDmaReductionReductionSize MmeDmaReductionReductionSize => (MmeDmaReductionReductionSize)((MmeDmaReduction >> 8) & 0x1);
         public uint MmeDmaSysmembar;
         public bool MmeDmaSysmembarV => (MmeDmaSysmembar & 0x1) != 0;
         public uint MmeDmaSync;
         public uint SetMmeDataFifoConfig;
-        public SetMmeDataFifoConfigFifoSize SetMmeDataFifoConfigFifoSize => (SetMmeDataFifoConfigFifoSize)((SetMmeDataFifoConfig >> 0) & 0x7);
+        public SetMmeDataFifoConfigFifoSize SetMmeDataFifoConfigFifoSize => (SetMmeDataFifoConfigFifoSize)(SetMmeDataFifoConfig & 0x7);
         public fixed uint Reserved578[2];
         public uint RenderSolidPrimMode;
-        public RenderSolidPrimModeV RenderSolidPrimModeV => (RenderSolidPrimModeV)((RenderSolidPrimMode >> 0) & 0x7);
+        public RenderSolidPrimModeV RenderSolidPrimModeV => (RenderSolidPrimModeV)(RenderSolidPrimMode & 0x7);
         public uint SetRenderSolidPrimColorFormat;
-        public SetRenderSolidPrimColorFormatV SetRenderSolidPrimColorFormatV => (SetRenderSolidPrimColorFormatV)((SetRenderSolidPrimColorFormat >> 0) & 0xFF);
+        public SetRenderSolidPrimColorFormatV SetRenderSolidPrimColorFormatV => (SetRenderSolidPrimColorFormatV)(SetRenderSolidPrimColorFormat & 0xFF);
         public uint SetRenderSolidPrimColor;
         public uint SetRenderSolidLineTieBreakBits;
         public bool SetRenderSolidLineTieBreakBitsXmajXincYinc => (SetRenderSolidLineTieBreakBits & 0x1) != 0;
@@ -690,24 +690,24 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public bool SetRenderSolidLineTieBreakBitsYmajXdecYinc => (SetRenderSolidLineTieBreakBits & 0x1000) != 0;
         public fixed uint Reserved590[20];
         public uint RenderSolidPrimPointXY;
-        public int RenderSolidPrimPointXYX => (int)((RenderSolidPrimPointXY >> 0) & 0xFFFF);
+        public int RenderSolidPrimPointXYX => (int)(RenderSolidPrimPointXY & 0xFFFF);
         public int RenderSolidPrimPointXYY => (int)((RenderSolidPrimPointXY >> 16) & 0xFFFF);
         public fixed uint Reserved5E4[7];
         public Array64<RenderSolidPrimPoint> RenderSolidPrimPoint;
         public uint SetPixelsFromCpuDataType;
-        public SetPixelsFromCpuDataTypeV SetPixelsFromCpuDataTypeV => (SetPixelsFromCpuDataTypeV)((SetPixelsFromCpuDataType >> 0) & 0x1);
+        public SetPixelsFromCpuDataTypeV SetPixelsFromCpuDataTypeV => (SetPixelsFromCpuDataTypeV)(SetPixelsFromCpuDataType & 0x1);
         public uint SetPixelsFromCpuColorFormat;
-        public SetPixelsFromCpuColorFormatV SetPixelsFromCpuColorFormatV => (SetPixelsFromCpuColorFormatV)((SetPixelsFromCpuColorFormat >> 0) & 0xFF);
+        public SetPixelsFromCpuColorFormatV SetPixelsFromCpuColorFormatV => (SetPixelsFromCpuColorFormatV)(SetPixelsFromCpuColorFormat & 0xFF);
         public uint SetPixelsFromCpuIndexFormat;
-        public SetPixelsFromCpuIndexFormatV SetPixelsFromCpuIndexFormatV => (SetPixelsFromCpuIndexFormatV)((SetPixelsFromCpuIndexFormat >> 0) & 0x3);
+        public SetPixelsFromCpuIndexFormatV SetPixelsFromCpuIndexFormatV => (SetPixelsFromCpuIndexFormatV)(SetPixelsFromCpuIndexFormat & 0x3);
         public uint SetPixelsFromCpuMonoFormat;
-        public SetPixelsFromCpuMonoFormatV SetPixelsFromCpuMonoFormatV => (SetPixelsFromCpuMonoFormatV)((SetPixelsFromCpuMonoFormat >> 0) & 0x1);
+        public SetPixelsFromCpuMonoFormatV SetPixelsFromCpuMonoFormatV => (SetPixelsFromCpuMonoFormatV)(SetPixelsFromCpuMonoFormat & 0x1);
         public uint SetPixelsFromCpuWrap;
-        public SetPixelsFromCpuWrapV SetPixelsFromCpuWrapV => (SetPixelsFromCpuWrapV)((SetPixelsFromCpuWrap >> 0) & 0x3);
+        public SetPixelsFromCpuWrapV SetPixelsFromCpuWrapV => (SetPixelsFromCpuWrapV)(SetPixelsFromCpuWrap & 0x3);
         public uint SetPixelsFromCpuColor0;
         public uint SetPixelsFromCpuColor1;
         public uint SetPixelsFromCpuMonoOpacity;
-        public SetPixelsFromCpuMonoOpacityV SetPixelsFromCpuMonoOpacityV => (SetPixelsFromCpuMonoOpacityV)((SetPixelsFromCpuMonoOpacity >> 0) & 0x1);
+        public SetPixelsFromCpuMonoOpacityV SetPixelsFromCpuMonoOpacityV => (SetPixelsFromCpuMonoOpacityV)(SetPixelsFromCpuMonoOpacity & 0x1);
         public fixed uint Reserved820[6];
         public uint SetPixelsFromCpuSrcWidth;
         public uint SetPixelsFromCpuSrcHeight;
@@ -753,13 +753,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
         public bool SetBigEndianControlOverride => (SetBigEndianControl & 0x10000000) != 0;
         public fixed uint Reserved874[3];
         public uint SetPixelsFromMemoryBlockShape;
-        public SetPixelsFromMemoryBlockShapeV SetPixelsFromMemoryBlockShapeV => (SetPixelsFromMemoryBlockShapeV)((SetPixelsFromMemoryBlockShape >> 0) & 0x7);
+        public SetPixelsFromMemoryBlockShapeV SetPixelsFromMemoryBlockShapeV => (SetPixelsFromMemoryBlockShapeV)(SetPixelsFromMemoryBlockShape & 0x7);
         public uint SetPixelsFromMemoryCorralSize;
-        public int SetPixelsFromMemoryCorralSizeV => (int)((SetPixelsFromMemoryCorralSize >> 0) & 0x3FF);
+        public int SetPixelsFromMemoryCorralSizeV => (int)(SetPixelsFromMemoryCorralSize & 0x3FF);
         public uint SetPixelsFromMemorySafeOverlap;
         public bool SetPixelsFromMemorySafeOverlapV => (SetPixelsFromMemorySafeOverlap & 0x1) != 0;
         public uint SetPixelsFromMemorySampleMode;
-        public SetPixelsFromMemorySampleModeOrigin SetPixelsFromMemorySampleModeOrigin => (SetPixelsFromMemorySampleModeOrigin)((SetPixelsFromMemorySampleMode >> 0) & 0x1);
+        public SetPixelsFromMemorySampleModeOrigin SetPixelsFromMemorySampleModeOrigin => (SetPixelsFromMemorySampleModeOrigin)(SetPixelsFromMemorySampleMode & 0x1);
         public SetPixelsFromMemorySampleModeFilter SetPixelsFromMemorySampleModeFilter => (SetPixelsFromMemorySampleModeFilter)((SetPixelsFromMemorySampleMode >> 4) & 0x1);
         public fixed uint Reserved890[8];
         public uint SetPixelsFromMemoryDstX0;

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

@@ -541,7 +541,8 @@ namespace Ryujinx.Graphics.Gpu.Image
                     depth,
                     lhs.FormatInfo.BlockHeight,
                     lhs.GobBlocksInY,
-                    lhs.GobBlocksInZ);
+                    lhs.GobBlocksInZ,
+                    level);
 
                 return gobBlocksInY == rhs.GobBlocksInY &&
                        gobBlocksInZ == rhs.GobBlocksInZ;
@@ -587,7 +588,8 @@ namespace Ryujinx.Graphics.Gpu.Image
                     lhsDepth,
                     lhs.FormatInfo.BlockHeight,
                     lhs.GobBlocksInY,
-                    lhs.GobBlocksInZ);
+                    lhs.GobBlocksInZ,
+                    lhsLevel);
 
                 int rhsHeight = Math.Max(1, rhs.Height >> rhsLevel);
                 int rhsDepth = Math.Max(1, rhs.GetDepth() >> rhsLevel);
@@ -597,7 +599,8 @@ namespace Ryujinx.Graphics.Gpu.Image
                     rhsDepth,
                     rhs.FormatInfo.BlockHeight,
                     rhs.GobBlocksInY,
-                    rhs.GobBlocksInZ);
+                    rhs.GobBlocksInZ,
+                    rhsLevel);
 
                 return lhsGobBlocksInY == rhsGobBlocksInY &&
                        lhsGobBlocksInZ == rhsGobBlocksInZ;

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

@@ -484,7 +484,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                         depthOrLayers = Math.Max(1, depthOrLayers >> minLod);
                     }
 
-                    (gobBlocksInY, gobBlocksInZ) = SizeCalculator.GetMipGobBlockSizes(height, depth, formatInfo.BlockHeight, gobBlocksInY, gobBlocksInZ);
+                    (gobBlocksInY, gobBlocksInZ) = SizeCalculator.GetMipGobBlockSizes(height, depth, formatInfo.BlockHeight, gobBlocksInY, gobBlocksInZ, minLod);
                 }
 
                 levels = (maxLod - minLod) + 1;

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

@@ -6,6 +6,7 @@ using Ryujinx.Graphics.Shader;
 using System;
 using System.Collections.Generic;
 using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Gpu.Memory
 {
@@ -14,12 +15,17 @@ namespace Ryujinx.Graphics.Gpu.Memory
     /// </summary>
     class BufferManager
     {
+        private const int TfInfoVertexCountOffset = Constants.TotalTransformFeedbackBuffers * sizeof(int);
+        private const int TfInfoBufferSize = TfInfoVertexCountOffset + sizeof(int);
+
         private readonly GpuContext _context;
         private readonly GpuChannel _channel;
 
         private int _unalignedStorageBuffers;
         public bool HasUnalignedStorageBuffers => _unalignedStorageBuffers > 0;
 
+        public bool HasTransformFeedbackOutputs { get; set; }
+
         private IndexBuffer _indexBuffer;
         private readonly VertexBuffer[] _vertexBuffers;
         private readonly BufferBounds[] _transformFeedbackBuffers;
@@ -98,6 +104,9 @@ namespace Ryujinx.Graphics.Gpu.Memory
         private readonly BuffersPerStage[] _gpStorageBuffers;
         private readonly BuffersPerStage[] _gpUniformBuffers;
 
+        private BufferHandle _tfInfoBuffer;
+        private int[] _tfInfoData;
+
         private bool _gpStorageBuffersDirty;
         private bool _gpUniformBuffersDirty;
 
@@ -137,6 +146,11 @@ namespace Ryujinx.Graphics.Gpu.Memory
             _bufferTextures = new List<BufferTextureBinding>();
 
             _ranges = new BufferAssignment[Constants.TotalGpUniformBuffers * Constants.ShaderStages];
+
+            if (!context.Capabilities.SupportsTransformFeedback)
+            {
+                _tfInfoData = new int[Constants.TotalTransformFeedbackBuffers];
+            }
         }
 
 
@@ -319,6 +333,31 @@ namespace Ryujinx.Graphics.Gpu.Memory
             _gpUniformBuffersDirty = true;
         }
 
+        /// <summary>
+        /// Sets the number of vertices per instance on a instanced draw. Used for transform feedback emulation.
+        /// </summary>
+        /// <param name="vertexCount">Vertex count per instance</param>
+        public void SetInstancedDrawVertexCount(int vertexCount)
+        {
+            if (!_context.Capabilities.SupportsTransformFeedback &&
+                HasTransformFeedbackOutputs &&
+                _tfInfoBuffer != BufferHandle.Null)
+            {
+                Span<byte> data = stackalloc byte[sizeof(int)];
+                MemoryMarshal.Cast<byte, int>(data)[0] = vertexCount;
+                _context.Renderer.SetBufferData(_tfInfoBuffer, TfInfoVertexCountOffset, data);
+            }
+        }
+
+        /// <summary>
+        /// Forces transform feedback and storage buffers to be updated on the next draw.
+        /// </summary>
+        public void ForceTransformFeedbackAndStorageBuffersDirty()
+        {
+            _transformFeedbackBuffersDirty = true;
+            _gpStorageBuffersDirty = true;
+        }
+
         /// <summary>
         /// Sets the binding points for the storage buffers bound on the compute pipeline.
         /// </summary>
@@ -537,22 +576,75 @@ namespace Ryujinx.Graphics.Gpu.Memory
             {
                 _transformFeedbackBuffersDirty = false;
 
-                Span<BufferRange> tfbs = stackalloc BufferRange[Constants.TotalTransformFeedbackBuffers];
+                if (_context.Capabilities.SupportsTransformFeedback)
-
-                for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
                 {
-                    BufferBounds tfb = _transformFeedbackBuffers[index];
+                    Span<BufferRange> tfbs = stackalloc BufferRange[Constants.TotalTransformFeedbackBuffers];
 
-                    if (tfb.Address == 0)
+                    for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
                     {
-                        tfbs[index] = BufferRange.Empty;
+                        BufferBounds tfb = _transformFeedbackBuffers[index];
-                        continue;
+
+                        if (tfb.Address == 0)
+                        {
+                            tfbs[index] = BufferRange.Empty;
+                            continue;
+                        }
+
+                        tfbs[index] = bufferCache.GetBufferRange(tfb.Address, tfb.Size, write: true);
                     }
 
-                    tfbs[index] = bufferCache.GetBufferRange(tfb.Address, tfb.Size, write: true);
+                    _context.Renderer.Pipeline.SetTransformFeedbackBuffers(tfbs);
                 }
+                else if (HasTransformFeedbackOutputs)
+                {
+                    Span<int> info = _tfInfoData.AsSpan();
+                    Span<BufferAssignment> buffers = stackalloc BufferAssignment[Constants.TotalTransformFeedbackBuffers + 1];
 
-                _context.Renderer.Pipeline.SetTransformFeedbackBuffers(tfbs);
+                    bool needsDataUpdate = false;
+
+                    if (_tfInfoBuffer == BufferHandle.Null)
+                    {
+                        _tfInfoBuffer = _context.Renderer.CreateBuffer(TfInfoBufferSize);
+                    }
+
+                    buffers[0] = new BufferAssignment(0, new BufferRange(_tfInfoBuffer, 0, TfInfoBufferSize));
+
+                    int alignment = _context.Capabilities.StorageBufferOffsetAlignment;
+
+                    for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
+                    {
+                        BufferBounds tfb = _transformFeedbackBuffers[index];
+
+                        if (tfb.Address == 0)
+                        {
+                            buffers[1 + index] = new BufferAssignment(1 + index, BufferRange.Empty);
+                        }
+                        else
+                        {
+                            ulong endAddress = tfb.Address + tfb.Size;
+                            ulong address = BitUtils.AlignDown(tfb.Address, (ulong)alignment);
+                            ulong size = endAddress - address;
+
+                            int tfeOffset = ((int)tfb.Address & (alignment - 1)) / 4;
+
+                            if (info[index] != tfeOffset)
+                            {
+                                info[index] = tfeOffset;
+                                needsDataUpdate = true;
+                            }
+
+                            buffers[1 + index] = new BufferAssignment(1 + index, bufferCache.GetBufferRange(address, size, write: true));
+                        }
+                    }
+
+                    if (needsDataUpdate)
+                    {
+                        Span<byte> infoData = MemoryMarshal.Cast<int, byte>(info);
+                        _context.Renderer.SetBufferData(_tfInfoBuffer, 0, infoData);
+                    }
+
+                    _context.Renderer.Pipeline.SetStorageBuffers(buffers);
+                }
             }
             else
             {

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

@@ -78,7 +78,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
         private List<BufferMigration> _sources;
         private BufferMigration _migrationTarget;
 
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         /// <summary>
         /// Whether the modified range list has any entries or not.
@@ -125,7 +125,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
                 for (int i = 0; i < count; i++)
                 {
                     BufferModifiedRange overlap = overlaps[i];
-                    
+
                     if (overlap.Address > address)
                     {
                         // The start of the remaining region is uncovered by this overlap. Call the action for it.

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

@@ -37,7 +37,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             ShaderSpecializationState oldSpecState,
             ShaderSpecializationState newSpecState,
             ResourceCounts counts,
-            int stageIndex) : base(context, counts, stageIndex)
+            int stageIndex) : base(context, counts, stageIndex, oldSpecState.TransformFeedbackDescriptors != null)
         {
             _data = data;
             _cb1Data = cb1Data;

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 = 4992;
+        private const uint CodeGenVersion = 5241;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";
@@ -368,7 +368,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
 
                         if (hostCode != null)
                         {
-                            ShaderInfo shaderInfo = ShaderInfoBuilder.BuildForCache(context, shaders, specState.PipelineState);
+                            ShaderInfo shaderInfo = ShaderInfoBuilder.BuildForCache(
+                                context,
+                                shaders,
+                                specState.PipelineState,
+                                specState.TransformFeedbackDescriptors != null);
 
                             IProgram hostProgram;
 

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

@@ -491,7 +491,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             {
                 ShaderSource[] shaderSources = new ShaderSource[compilation.TranslatedStages.Length];
 
-                ShaderInfoBuilder shaderInfoBuilder = new ShaderInfoBuilder(_context);
+                ShaderInfoBuilder shaderInfoBuilder = new ShaderInfoBuilder(_context, compilation.SpecializationState.TransformFeedbackDescriptors != null);
 
                 for (int index = 0; index < compilation.TranslatedStages.Length; index++)
                 {

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

@@ -30,7 +30,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             GpuContext context,
             GpuChannel channel,
             GpuAccessorState state,
-            int stageIndex) : base(context, state.ResourceCounts, stageIndex)
+            int stageIndex) : base(context, state.ResourceCounts, stageIndex, state.TransformFeedbackDescriptors != null)
         {
             _isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
             _channel = channel;
@@ -44,7 +44,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="context">GPU context</param>
         /// <param name="channel">GPU channel</param>
         /// <param name="state">Current GPU state</param>
-        public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0)
+        public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0, false)
         {
             _channel = channel;
             _state = state;

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

@@ -17,40 +17,56 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private readonly ResourceCounts _resourceCounts;
         private readonly int _stageIndex;
 
+        private readonly int _reservedConstantBuffers;
+        private readonly int _reservedStorageBuffers;
+
         /// <summary>
         /// Creates a new GPU accessor.
         /// </summary>
         /// <param name="context">GPU context</param>
-        public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex)
+        /// <param name="resourceCounts">Counter of GPU resources used by the shader</param>
+        /// <param name="stageIndex">Index of the shader stage, 0 for compute</param>
+        /// <param name="tfEnabled">Indicates if the current graphics shader is used with transform feedback enabled</param>
+        public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex, bool tfEnabled)
         {
             _context = context;
             _resourceCounts = resourceCounts;
             _stageIndex = stageIndex;
+
+            _reservedConstantBuffers = 1; // For the support buffer.
+            _reservedStorageBuffers = !context.Capabilities.SupportsTransformFeedback && tfEnabled ? 5 : 0;
         }
 
         public int QueryBindingConstantBuffer(int index)
         {
+            int binding;
+
             if (_context.Capabilities.Api == TargetApi.Vulkan)
             {
-                // We need to start counting from 1 since binding 0 is reserved for the support uniform buffer.
+                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumUniformBuffersPerStage, "Uniform buffer");
-                return GetBindingFromIndex(index, _context.Capabilities.MaximumUniformBuffersPerStage, "Uniform buffer") + 1;
             }
             else
             {
-                return _resourceCounts.UniformBuffersCount++;
+                binding = _resourceCounts.UniformBuffersCount++;
             }
+
+            return binding + _reservedConstantBuffers;
         }
 
         public int QueryBindingStorageBuffer(int index)
         {
+            int binding;
+
             if (_context.Capabilities.Api == TargetApi.Vulkan)
             {
-                return GetBindingFromIndex(index, _context.Capabilities.MaximumStorageBuffersPerStage, "Storage buffer");
+                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumStorageBuffersPerStage, "Storage buffer");
             }
             else
             {
-                return _resourceCounts.StorageBuffersCount++;
+                binding = _resourceCounts.StorageBuffersCount++;
             }
+
+            return binding + _reservedStorageBuffers;
         }
 
         public int QueryBindingTexture(int index, bool isBuffer)
@@ -141,10 +157,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         public bool QueryHostSupportsShaderBallot() => _context.Capabilities.SupportsShaderBallot;
 
+        public bool QueryHostSupportsShaderBarrierDivergence() => _context.Capabilities.SupportsShaderBarrierDivergence;
+
+        public bool QueryHostSupportsShaderFloat64() => _context.Capabilities.SupportsShaderFloat64;
+
         public bool QueryHostSupportsSnormBufferTextureFormat() => _context.Capabilities.SupportsSnormBufferTextureFormat;
 
         public bool QueryHostSupportsTextureShadowLod() => _context.Capabilities.SupportsTextureShadowLod;
 
+        public bool QueryHostSupportsTransformFeedback() => _context.Capabilities.SupportsTransformFeedback;
+
         public bool QueryHostSupportsViewportIndexVertexTessellation() => _context.Capabilities.SupportsViewportIndexVertexTessellation;
 
         public bool QueryHostSupportsViewportMask() => _context.Capabilities.SupportsViewportMask;

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

@@ -24,13 +24,5 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// Total of images used by the shaders.
         /// </summary>
         public int ImagesCount;
-
-        /// <summary>
-        /// Creates a new instance of the shader resource counts class.
-        /// </summary>
-        public ResourceCounts()
-        {
-            UniformBuffersCount = 1; // The first binding is reserved for the support buffer.
-        }
     }
 }

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

@@ -362,7 +362,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
             TranslatorContext previousStage = null;
 
-            ShaderInfoBuilder infoBuilder = new ShaderInfoBuilder(_context);
+            ShaderInfoBuilder infoBuilder = new ShaderInfoBuilder(_context, transformFeedbackDescriptors != null);
 
             for (int stageIndex = 0; stageIndex < Constants.ShaderStages; stageIndex++)
             {

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

@@ -16,15 +16,24 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private const int TextureSetIndex = 2;
         private const int ImageSetIndex = 3;
 
-        private const ResourceStages SupportBufferStags =
+        private const ResourceStages SupportBufferStages =
             ResourceStages.Compute |
             ResourceStages.Vertex |
             ResourceStages.Fragment;
 
+        private const ResourceStages VtgStages =
+            ResourceStages.Vertex |
+            ResourceStages.TessellationControl |
+            ResourceStages.TessellationEvaluation |
+            ResourceStages.Geometry;
+
         private readonly GpuContext _context;
 
         private int _fragmentOutputMap;
 
+        private readonly int _reservedConstantBuffers;
+        private readonly int _reservedStorageBuffers;
+
         private readonly List<ResourceDescriptor>[] _resourceDescriptors;
         private readonly List<ResourceUsage>[] _resourceUsages;
 
@@ -32,7 +41,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// Creates a new shader info builder.
         /// </summary>
         /// <param name="context">GPU context that owns the shaders that will be added to the builder</param>
-        public ShaderInfoBuilder(GpuContext context)
+        /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
+        public ShaderInfoBuilder(GpuContext context, bool tfEnabled)
         {
             _context = context;
 
@@ -47,7 +57,22 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 _resourceUsages[index] = new();
             }
 
-            AddDescriptor(SupportBufferStags, ResourceType.UniformBuffer, UniformSetIndex, 0, 1);
+            AddDescriptor(SupportBufferStages, ResourceType.UniformBuffer, UniformSetIndex, 0, 1);
+
+            _reservedConstantBuffers = 1; // For the support buffer.
+
+            if (!context.Capabilities.SupportsTransformFeedback && tfEnabled)
+            {
+                _reservedStorageBuffers = 5;
+
+                AddDescriptor(VtgStages, ResourceType.StorageBuffer, StorageSetIndex, 0, 5);
+                AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Read, StorageSetIndex, 0, 1);
+                AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Write, StorageSetIndex, 1, 4);
+            }
+            else
+            {
+                _reservedStorageBuffers = 0;
+            }
         }
 
         /// <summary>
@@ -86,8 +111,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             int texturesPerStage = (int)_context.Capabilities.MaximumTexturesPerStage;
             int imagesPerStage = (int)_context.Capabilities.MaximumImagesPerStage;
 
-            int uniformBinding = 1 + stageIndex * uniformsPerStage;
+            int uniformBinding = _reservedConstantBuffers + stageIndex * uniformsPerStage;
-            int storageBinding = stageIndex * storagesPerStage;
+            int storageBinding = _reservedStorageBuffers + stageIndex * storagesPerStage;
             int textureBinding = stageIndex * texturesPerStage * 2;
             int imageBinding = stageIndex * imagesPerStage * 2;
 
@@ -133,6 +158,23 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AddDescriptor(stages, type2, setIndex, binding + count, count);
         }
 
+        /// <summary>
+        /// Adds buffer usage information to the list of usages.
+        /// </summary>
+        /// <param name="stages">Shader stages where the resource is used</param>
+        /// <param name="type">Type of the resource</param>
+        /// <param name="access">How the resource is accessed by the shader stages where it is used</param>
+        /// <param name="setIndex">Descriptor set number where the resource will be bound</param>
+        /// <param name="binding">Binding number where the resource will be bound</param>
+        /// <param name="count">Number of resources bound at the binding location</param>
+        private void AddUsage(ResourceStages stages, ResourceType type, ResourceAccess access, int setIndex, int binding, int count)
+        {
+            for (int index = 0; index < count; index++)
+            {
+                _resourceUsages[setIndex].Add(new ResourceUsage(binding + index, type, stages, access));
+            }
+        }
+
         /// <summary>
         /// Adds buffer usage information to the list of usages.
         /// </summary>
@@ -212,10 +254,15 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="context">GPU context that owns the shaders</param>
         /// <param name="programs">Shaders from the disk cache</param>
         /// <param name="pipeline">Optional pipeline for background compilation</param>
+        /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
         /// <returns>Shader information</returns>
-        public static ShaderInfo BuildForCache(GpuContext context, IEnumerable<CachedShaderStage> programs, ProgramPipelineState? pipeline)
+        public static ShaderInfo BuildForCache(
+            GpuContext context,
+            IEnumerable<CachedShaderStage> programs,
+            ProgramPipelineState? pipeline,
+            bool tfEnabled)
         {
-            ShaderInfoBuilder builder = new ShaderInfoBuilder(context);
+            ShaderInfoBuilder builder = new ShaderInfoBuilder(context, tfEnabled);
 
             foreach (CachedShaderStage program in programs)
             {
@@ -237,7 +284,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <returns>Shader information</returns>
         public static ShaderInfo BuildForCompute(GpuContext context, ShaderProgramInfo info, bool fromCache = false)
         {
-            ShaderInfoBuilder builder = new ShaderInfoBuilder(context);
+            ShaderInfoBuilder builder = new ShaderInfoBuilder(context, tfEnabled: false);
 
             builder.AddStageInfo(info);
 

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -127,6 +127,7 @@ namespace Ryujinx.Graphics.OpenGL
         public Capabilities GetCapabilities()
         {
             bool intelWindows = HwCapabilities.Vendor == HwCapabilities.GpuVendor.IntelWindows;
+            bool intelUnix = HwCapabilities.Vendor == HwCapabilities.GpuVendor.IntelUnix;
             bool amdWindows = HwCapabilities.Vendor == HwCapabilities.GpuVendor.AmdWindows;
 
             return new Capabilities(
@@ -152,12 +153,15 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsFragmentShaderOrderingIntel: HwCapabilities.SupportsFragmentShaderOrdering,
                 supportsGeometryShader: true,
                 supportsGeometryShaderPassthrough: HwCapabilities.SupportsGeometryShaderPassthrough,
+                supportsTransformFeedback: true,
                 supportsImageLoadFormatted: HwCapabilities.SupportsImageLoadFormatted,
                 supportsLayerVertexTessellation: HwCapabilities.SupportsShaderViewportLayerArray,
                 supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat,
                 supportsCubemapView: true,
                 supportsNonConstantTextureOffset: HwCapabilities.SupportsNonConstantTextureOffset,
                 supportsShaderBallot: HwCapabilities.SupportsShaderBallot,
+                supportsShaderBarrierDivergence: !(intelWindows || intelUnix),
+                supportsShaderFloat64: true,
                 supportsTextureShadowLod: HwCapabilities.SupportsTextureShadowLod,
                 supportsViewportIndexVertexTessellation: HwCapabilities.SupportsShaderViewportLayerArray,
                 supportsViewportMask: HwCapabilities.SupportsViewportArray2,

src/Ryujinx.Graphics.OpenGL/Queries/CounterQueue.cs

@@ -21,7 +21,7 @@ namespace Ryujinx.Graphics.OpenGL.Queries
         private ulong _accumulatedCounter;
         private int _waiterCount;
 
-        private object _lock = new object();
+        private readonly object _lock = new();
 
         private Queue<BufferedQuery> _queryPool;
         private AutoResetEvent _queuedEvent = new AutoResetEvent(false);

src/Ryujinx.Graphics.OpenGL/Queries/CounterQueueEvent.cs

@@ -24,7 +24,7 @@ namespace Ryujinx.Graphics.OpenGL.Queries
         private bool _hostAccessReserved = false;
         private int _refCount = 1; // Starts with a reference from the counter queue.
 
-        private object _lock = new object();
+        private readonly object _lock = new();
         private ulong _result = ulong.MaxValue;
         private double _divisor = 1f;
 

src/Ryujinx.Graphics.OpenGL/ResourcePool.cs

@@ -20,7 +20,7 @@ namespace Ryujinx.Graphics.OpenGL
     {
         private const int DisposedLiveFrames = 2;
 
-        private readonly object _lock = new object();
+        private readonly object _lock = new();
         private readonly Dictionary<TextureCreateInfo, List<DisposedTexture>> _textures = new Dictionary<TextureCreateInfo, List<DisposedTexture>>();
 
         /// <summary>

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs

@@ -71,40 +71,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             context.AppendLine($"const int {DefaultNames.UndefinedName} = 0;");
             context.AppendLine();
 
-            if (context.Config.Stage == ShaderStage.Compute)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeLocalMemorySize(), 4);
-
-                if (localMemorySize != 0)
-                {
-                    string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
-
-                    context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
-                    context.AppendLine();
-                }
-
-                int sharedMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeSharedMemorySize(), 4);
-
-                if (sharedMemorySize != 0)
-                {
-                    string sharedMemorySizeStr = NumberFormatter.FormatInt(sharedMemorySize);
-
-                    context.AppendLine($"shared uint {DefaultNames.SharedMemoryName}[{sharedMemorySizeStr}];");
-                    context.AppendLine();
-                }
-            }
-            else if (context.Config.LocalMemorySize != 0)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
-
-                string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
-
-                context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
-                context.AppendLine();
-            }
-
             DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
             DeclareStorageBuffers(context, context.Config.Properties.StorageBuffers.Values);
+            DeclareMemories(context, context.Config.Properties.LocalMemories.Values, isShared: false);
+            DeclareMemories(context, context.Config.Properties.SharedMemories.Values, isShared: true);
 
             var textureDescriptors = context.Config.GetTextureDescriptors();
             if (textureDescriptors.Length != 0)
@@ -238,11 +208,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 context.AppendLine();
             }
 
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.AtomicMinMaxS32Shared) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl");
-            }
-
             if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighS32) != 0)
             {
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighS32.glsl");
@@ -273,11 +238,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleXor.glsl");
             }
 
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.StoreSharedSmallInt) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl");
-            }
-
             if ((info.HelperFunctionsMask & HelperFunctionsMask.SwizzleAdd) != 0)
             {
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/SwizzleAdd.glsl");
@@ -358,7 +318,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                     _ => "std430"
                 };
 
-                context.AppendLine($"layout (binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
+                string set = string.Empty;
+
+                if (context.Config.Options.TargetApi == TargetApi.Vulkan)
+                {
+                    set = $"set = {buffer.Set}, ";
+                }
+
+                context.AppendLine($"layout ({set}binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
                 context.EnterScope();
 
                 foreach (StructureField field in buffer.Type.Fields)
@@ -391,6 +358,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             }
         }
 
+        private static void DeclareMemories(CodeGenContext context, IEnumerable<MemoryDefinition> memories, bool isShared)
+        {
+            string prefix = isShared ? "shared " : string.Empty;
+
+            foreach (MemoryDefinition memory in memories)
+            {
+                string typeName = GetVarTypeName(context, memory.Type & ~AggregateType.Array);
+
+                if (memory.ArrayLength > 0)
+                {
+                    string arraySize = memory.ArrayLength.ToString(CultureInfo.InvariantCulture);
+
+                    context.AppendLine($"{prefix}{typeName} {memory.Name}[{arraySize}];");
+                }
+                else
+                {
+                    context.AppendLine($"{prefix}{typeName} {memory.Name}[];");
+                }
+            }
+        }
+
         private static void DeclareSamplers(CodeGenContext context, TextureDescriptor[] descriptors)
         {
             int arraySize = 0;
@@ -717,7 +705,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             string code = EmbeddedResources.ReadAllText(filename);
 
             code = code.Replace("\t", CodeGenContext.Tab);
-            code = code.Replace("$SHARED_MEM$", DefaultNames.SharedMemoryName);
 
             if (context.Config.GpuAccessor.QueryHostSupportsShaderBallot())
             {

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs

@@ -11,9 +11,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
         public const string IAttributePrefix = "in_attr";
         public const string OAttributePrefix = "out_attr";
 
-        public const string LocalMemoryName  = "local_mem";
-        public const string SharedMemoryName = "shared_mem";
-
         public const string ArgumentNamePrefix = "a";
 
         public const string UndefinedName = "undef";

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/GlslGenerator.cs

@@ -28,18 +28,18 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
                 for (int i = 1; i < info.Functions.Count; i++)
                 {
-                    PrintFunction(context, info, info.Functions[i]);
+                    PrintFunction(context, info.Functions[i]);
 
                     context.AppendLine();
                 }
             }
 
-            PrintFunction(context, info, info.Functions[0], MainFunctionName);
+            PrintFunction(context, info.Functions[0], MainFunctionName);
 
             return context.GetCode();
         }
 
-        private static void PrintFunction(CodeGenContext context, StructuredProgramInfo info, StructuredFunction function, string funcName = null)
+        private static void PrintFunction(CodeGenContext context, StructuredFunction function, string funcName = null)
         {
             context.CurrentFunction = function;
 
@@ -48,7 +48,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
             Declarations.DeclareLocals(context, function);
 
-            PrintBlock(context, function.MainBlock);
+            PrintBlock(context, function.MainBlock, funcName == MainFunctionName);
 
             context.LeaveScope();
         }
@@ -72,7 +72,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             return $"{Declarations.GetVarTypeName(context, function.ReturnType)} {funcName ?? function.Name}({string.Join(", ", args)})";
         }
 
-        private static void PrintBlock(CodeGenContext context, AstBlock block)
+        private static void PrintBlock(CodeGenContext context, AstBlock block, bool isMainFunction)
         {
             AstBlockVisitor visitor = new AstBlockVisitor(block);
 
@@ -112,10 +112,32 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 }
             };
 
+            bool supportsBarrierDivergence = context.Config.GpuAccessor.QueryHostSupportsShaderBarrierDivergence();
+            bool mayHaveReturned = false;
+
             foreach (IAstNode node in visitor.Visit())
             {
                 if (node is AstOperation operation)
                 {
+                    if (!supportsBarrierDivergence)
+                    {
+                        if (operation.Inst == IntermediateRepresentation.Instruction.Barrier)
+                        {
+                            // Barrier on divergent control flow paths may cause the GPU to hang,
+                            // so skip emitting the barrier for those cases.
+                            if (visitor.Block.Type != AstBlockType.Main || mayHaveReturned || !isMainFunction)
+                            {
+                                context.Config.GpuAccessor.Log($"Shader has barrier on potentially divergent block, the barrier will be removed.");
+
+                                continue;
+                            }
+                        }
+                        else if (operation.Inst == IntermediateRepresentation.Instruction.Return)
+                        {
+                            mayHaveReturned = true;
+                        }
+                    }
+
                     string expr = InstGen.GetExpression(context, operation);
 
                     if (expr != null)

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl

@@ -1,21 +0,0 @@
-int Helper_AtomicMaxS32(int offset, int value)
-{
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[offset];
-        newValue = uint(max(int(oldValue), value));
-    } while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
-    return int(oldValue);
-}
-
-int Helper_AtomicMinS32(int offset, int value)
-{
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[offset];
-        newValue = uint(min(int(oldValue), value));
-    } while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
-    return int(oldValue);
-}

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs

@@ -2,9 +2,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 {
     static class HelperFunctionNames
     {
-        public static string AtomicMaxS32 = "Helper_AtomicMaxS32";
-        public static string AtomicMinS32 = "Helper_AtomicMinS32";
-
         public static string MultiplyHighS32 = "Helper_MultiplyHighS32";
         public static string MultiplyHighU32 = "Helper_MultiplyHighU32";
 
@@ -13,10 +10,5 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
         public static string ShuffleUp   = "Helper_ShuffleUp";
         public static string ShuffleXor  = "Helper_ShuffleXor";
         public static string SwizzleAdd  = "Helper_SwizzleAdd";
-
-        public static string StoreShared16  = "Helper_StoreShared16";
-        public static string StoreShared8   = "Helper_StoreShared8";
-        public static string StoreStorage16 = "Helper_StoreStorage16";
-        public static string StoreStorage8  = "Helper_StoreStorage8";
     }
 }

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl

@@ -1,23 +0,0 @@
-void Helper_StoreShared16(int offset, uint value)
-{
-    int wordOffset = offset >> 2;
-    int bitOffset = (offset & 3) * 8;
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[wordOffset];
-        newValue = bitfieldInsert(oldValue, value, bitOffset, 16);
-    } while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
-}
-
-void Helper_StoreShared8(int offset, uint value)
-{
-    int wordOffset = offset >> 2;
-    int bitOffset = (offset & 3) * 8;
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[wordOffset];
-        newValue = bitfieldInsert(oldValue, value, bitOffset, 8);
-    } while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
-}

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs

@@ -68,7 +68,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
 
                 string args = string.Empty;
 
-                if (atomic && operation.StorageKind == StorageKind.StorageBuffer)
+                if (atomic && (operation.StorageKind == StorageKind.StorageBuffer || operation.StorageKind == StorageKind.SharedMemory))
                 {
                     args = GenerateLoadOrStore(context, operation, isStore: false);
 
@@ -81,23 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                         args += ", " + GetSoureExpr(context, operation.GetSource(argIndex), dstType);
                     }
                 }
-                else if (atomic && operation.StorageKind == StorageKind.SharedMemory)
-                {
-                    args = LoadShared(context, operation);
-
-                    // For shared memory access, the second argument is unused and should be ignored.
-                    // It is there to make both storage and shared access have the same number of arguments.
-                    // For storage, both inputs are consumed when the argument index is 0, so we should skip it here.
-
-                    for (int argIndex = 2; argIndex < arity; argIndex++)
-                    {
-                        args += ", ";
-
-                        AggregateType dstType = GetSrcVarType(inst, argIndex);
-
-                        args += GetSoureExpr(context, operation.GetSource(argIndex), dstType);
-                    }
-                }
                 else
                 {
                     for (int argIndex = 0; argIndex < arity; argIndex++)
@@ -179,12 +162,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     case Instruction.Load:
                         return Load(context, operation);
 
-                    case Instruction.LoadLocal:
-                        return LoadLocal(context, operation);
-
-                    case Instruction.LoadShared:
-                        return LoadShared(context, operation);
-
                     case Instruction.Lod:
                         return Lod(context, operation);
 
@@ -200,18 +177,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     case Instruction.Store:
                         return Store(context, operation);
 
-                    case Instruction.StoreLocal:
-                        return StoreLocal(context, operation);
-
-                    case Instruction.StoreShared:
-                        return StoreShared(context, operation);
-
-                    case Instruction.StoreShared16:
-                        return StoreShared16(context, operation);
-
-                    case Instruction.StoreShared8:
-                        return StoreShared8(context, operation);
-
                     case Instruction.TextureSample:
                         return TextureSample(context, operation);
 

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs

@@ -17,9 +17,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.AtomicAdd,                InstType.AtomicBinary,   "atomicAdd");
             Add(Instruction.AtomicAnd,                InstType.AtomicBinary,   "atomicAnd");
             Add(Instruction.AtomicCompareAndSwap,     InstType.AtomicTernary,  "atomicCompSwap");
-            Add(Instruction.AtomicMaxS32,             InstType.CallTernary,    HelperFunctionNames.AtomicMaxS32);
             Add(Instruction.AtomicMaxU32,             InstType.AtomicBinary,   "atomicMax");
-            Add(Instruction.AtomicMinS32,             InstType.CallTernary,    HelperFunctionNames.AtomicMinS32);
             Add(Instruction.AtomicMinU32,             InstType.AtomicBinary,   "atomicMin");
             Add(Instruction.AtomicOr,                 InstType.AtomicBinary,   "atomicOr");
             Add(Instruction.AtomicSwap,               InstType.AtomicBinary,   "atomicExchange");
@@ -83,8 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.ImageAtomic,              InstType.Special);
             Add(Instruction.IsNan,                    InstType.CallUnary,      "isnan");
             Add(Instruction.Load,                     InstType.Special);
-            Add(Instruction.LoadLocal,                InstType.Special);
-            Add(Instruction.LoadShared,               InstType.Special);
             Add(Instruction.Lod,                      InstType.Special);
             Add(Instruction.LogarithmB2,              InstType.CallUnary,      "log2");
             Add(Instruction.LogicalAnd,               InstType.OpBinaryCom,    "&&",              9);
@@ -118,10 +114,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.Sine,                     InstType.CallUnary,      "sin");
             Add(Instruction.SquareRoot,               InstType.CallUnary,      "sqrt");
             Add(Instruction.Store,                    InstType.Special);
-            Add(Instruction.StoreLocal,               InstType.Special);
-            Add(Instruction.StoreShared,              InstType.Special);
-            Add(Instruction.StoreShared16,            InstType.Special);
-            Add(Instruction.StoreShared8,             InstType.Special);
             Add(Instruction.Subtract,                 InstType.OpBinary,       "-",               2);
             Add(Instruction.SwizzleAdd,               InstType.CallTernary,    HelperFunctionNames.SwizzleAdd);
             Add(Instruction.TextureSample,            InstType.Special);

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs

@@ -191,25 +191,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             return GenerateLoadOrStore(context, operation, isStore: false);
         }
 
-        public static string LoadLocal(CodeGenContext context, AstOperation operation)
-        {
-            return LoadLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
-        }
-
-        public static string LoadShared(CodeGenContext context, AstOperation operation)
-        {
-            return LoadLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
-        }
-
-        private static string LoadLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
-        {
-            IAstNode src1 = operation.GetSource(0);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            return $"{arrayName}[{offsetExpr}]";
-        }
-
         public static string Lod(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -263,58 +244,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             return GenerateLoadOrStore(context, operation, isStore: true);
         }
 
-        public static string StoreLocal(CodeGenContext context, AstOperation operation)
-        {
-            return StoreLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
-        }
-
-        public static string StoreShared(CodeGenContext context, AstOperation operation)
-        {
-            return StoreLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
-        }
-
-        private static string StoreLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{arrayName}[{offsetExpr}] = {src}";
-        }
-
-        public static string StoreShared16(CodeGenContext context, AstOperation operation)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{HelperFunctionNames.StoreShared16}({offsetExpr}, {src})";
-        }
-
-        public static string StoreShared8(CodeGenContext context, AstOperation operation)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{HelperFunctionNames.StoreShared8}({offsetExpr}, {src})";
-        }
-
         public static string TextureSample(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -675,6 +604,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     varType = field.Type;
                     break;
 
+                case StorageKind.LocalMemory:
+                case StorageKind.SharedMemory:
+                    if (!(operation.GetSource(srcIndex++) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
+                    {
+                        throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
+                    }
+
+                    MemoryDefinition memory = storageKind == StorageKind.LocalMemory
+                        ? context.Config.Properties.LocalMemories[bindingId.Value]
+                        : context.Config.Properties.SharedMemories[bindingId.Value];
+
+                    varName = memory.Name;
+                    varType = memory.Type;
+                    break;
+
                 case StorageKind.Input:
                 case StorageKind.InputPerPatch:
                 case StorageKind.Output:

src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs

@@ -113,7 +113,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
             if (node is AstOperation operation)
             {
-                if (operation.Inst == Instruction.Load)
+                if (operation.Inst == Instruction.Load || operation.Inst.IsAtomic())
                 {
                     switch (operation.StorageKind)
                     {
@@ -136,6 +136,19 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
                             return field.Type & AggregateType.ElementTypeMask;
 
+                        case StorageKind.LocalMemory:
+                        case StorageKind.SharedMemory:
+                            if (!(operation.GetSource(0) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
+                            {
+                                throw new InvalidOperationException($"First input of {operation.Inst} with {operation.StorageKind} storage must be a constant operand.");
+                            }
+
+                            MemoryDefinition memory = operation.StorageKind == StorageKind.LocalMemory
+                                ? context.Config.Properties.LocalMemories[bindingId.Value]
+                                : context.Config.Properties.SharedMemories[bindingId.Value];
+
+                            return memory.Type & AggregateType.ElementTypeMask;
+
                         case StorageKind.Input:
                         case StorageKind.InputPerPatch:
                         case StorageKind.Output:

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

@@ -25,8 +25,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         public Dictionary<int, Instruction> ConstantBuffers { get; } = new Dictionary<int, Instruction>();
         public Dictionary<int, Instruction> StorageBuffers { get; } = new Dictionary<int, Instruction>();
-        public Instruction LocalMemory { get; set; }
+        public Dictionary<int, Instruction> LocalMemories { get; } = new Dictionary<int, Instruction>();
-        public Instruction SharedMemory { get; set; }
+        public Dictionary<int, Instruction> SharedMemories { get; } = new Dictionary<int, Instruction>();
         public Dictionary<TextureMeta, SamplerType> SamplersTypes { get; } = new Dictionary<TextureMeta, SamplerType>();
         public Dictionary<TextureMeta, (Instruction, Instruction, Instruction)> Samplers { get; } = new Dictionary<TextureMeta, (Instruction, Instruction, Instruction)>();
         public Dictionary<TextureMeta, (Instruction, Instruction)> Images { get; } = new Dictionary<TextureMeta, (Instruction, Instruction)>();
@@ -35,7 +35,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
         public Dictionary<IoDefinition, Instruction> InputsPerPatch { get; } = new Dictionary<IoDefinition, Instruction>();
         public Dictionary<IoDefinition, Instruction> OutputsPerPatch { get; } = new Dictionary<IoDefinition, Instruction>();
 
-        public Instruction CoordTemp { get; set; }
         public StructuredFunction CurrentFunction { get; set; }
         private readonly Dictionary<AstOperand, Instruction> _locals = new Dictionary<AstOperand, Instruction>();
         private readonly Dictionary<int, Instruction[]> _localForArgs = new Dictionary<int, Instruction[]>();
@@ -76,6 +75,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         public SpirvDelegates Delegates { get; }
 
+        public bool IsMainFunction { get; private set; }
+        public bool MayHaveReturned { get; set; }
+
         public CodeGenContext(
             StructuredProgramInfo info,
             ShaderConfig config,
@@ -108,8 +110,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Delegates = new SpirvDelegates(this);
         }
 
-        public void StartFunction()
+        public void StartFunction(bool isMainFunction)
         {
+            IsMainFunction = isMainFunction;
+            MayHaveReturned = false;
             _locals.Clear();
             _localForArgs.Clear();
             _funcArgs.Clear();

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

@@ -6,7 +6,6 @@ using Spv.Generator;
 using System;
 using System.Collections.Generic;
 using System.Diagnostics;
-using System.Linq;
 using System.Numerics;
 using static Spv.Specification;
 using SpvInstruction = Spv.Generator.Instruction;
@@ -44,13 +43,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 context.AddLocalVariable(spvLocal);
                 context.DeclareLocal(local, spvLocal);
             }
-
-            var ivector2Type = context.TypeVector(context.TypeS32(), 2);
-            var coordTempPointerType = context.TypePointer(StorageClass.Function, ivector2Type);
-            var coordTemp = context.Variable(coordTempPointerType, StorageClass.Function);
-
-            context.AddLocalVariable(coordTemp);
-            context.CoordTemp = coordTemp;
         }
 
         public static void DeclareLocalForArgs(CodeGenContext context, List<StructuredFunction> functions)
@@ -77,54 +69,30 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         public static void DeclareAll(CodeGenContext context, StructuredProgramInfo info)
         {
-            if (context.Config.Stage == ShaderStage.Compute)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeLocalMemorySize(), 4);
-
-                if (localMemorySize != 0)
-                {
-                    DeclareLocalMemory(context, localMemorySize);
-                }
-
-                int sharedMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeSharedMemorySize(), 4);
-
-                if (sharedMemorySize != 0)
-                {
-                    DeclareSharedMemory(context, sharedMemorySize);
-                }
-            }
-            else if (context.Config.LocalMemorySize != 0)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
-                DeclareLocalMemory(context, localMemorySize);
-            }
-
             DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
             DeclareStorageBuffers(context, context.Config.Properties.StorageBuffers.Values);
+            DeclareMemories(context, context.Config.Properties.LocalMemories, context.LocalMemories, StorageClass.Private);
+            DeclareMemories(context, context.Config.Properties.SharedMemories, context.SharedMemories, StorageClass.Workgroup);
             DeclareSamplers(context, context.Config.GetTextureDescriptors());
             DeclareImages(context, context.Config.GetImageDescriptors());
             DeclareInputsAndOutputs(context, info);
         }
 
-        private static void DeclareLocalMemory(CodeGenContext context, int size)
+        private static void DeclareMemories(
+            CodeGenContext context,
+            IReadOnlyDictionary<int, MemoryDefinition> memories,
+            Dictionary<int, SpvInstruction> dict,
+            StorageClass storage)
         {
-            context.LocalMemory = DeclareMemory(context, StorageClass.Private, size);
+            foreach ((int id, MemoryDefinition memory) in memories)
-        }
+            {
+                var pointerType = context.TypePointer(storage, context.GetType(memory.Type, memory.ArrayLength));
+                var variable = context.Variable(pointerType, storage);
 
-        private static void DeclareSharedMemory(CodeGenContext context, int size)
+                context.AddGlobalVariable(variable);
-        {
-            context.SharedMemory = DeclareMemory(context, StorageClass.Workgroup, size);
-        }
 
-        private static SpvInstruction DeclareMemory(CodeGenContext context, StorageClass storage, int size)
+                dict.Add(id, variable);
-        {
+            }
-            var arrayType = context.TypeArray(context.TypeU32(), context.Constant(context.TypeU32(), size));
-            var pointerType = context.TypePointer(storage, arrayType);
-            var variable = context.Variable(pointerType, storage);
-
-            context.AddGlobalVariable(variable);
-
-            return variable;
         }
 
         private static void DeclareConstantBuffers(CodeGenContext context, IEnumerable<BufferDefinition> buffers)

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

@@ -97,8 +97,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Add(Instruction.ImageStore,               GenerateImageStore);
             Add(Instruction.IsNan,                    GenerateIsNan);
             Add(Instruction.Load,                     GenerateLoad);
-            Add(Instruction.LoadLocal,                GenerateLoadLocal);
-            Add(Instruction.LoadShared,               GenerateLoadShared);
             Add(Instruction.Lod,                      GenerateLod);
             Add(Instruction.LogarithmB2,              GenerateLogarithmB2);
             Add(Instruction.LogicalAnd,               GenerateLogicalAnd);
@@ -132,10 +130,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Add(Instruction.Sine,                     GenerateSine);
             Add(Instruction.SquareRoot,               GenerateSquareRoot);
             Add(Instruction.Store,                    GenerateStore);
-            Add(Instruction.StoreLocal,               GenerateStoreLocal);
-            Add(Instruction.StoreShared,              GenerateStoreShared);
-            Add(Instruction.StoreShared16,            GenerateStoreShared16);
-            Add(Instruction.StoreShared8,             GenerateStoreShared8);
             Add(Instruction.Subtract,                 GenerateSubtract);
             Add(Instruction.SwizzleAdd,               GenerateSwizzleAdd);
             Add(Instruction.TextureSample,            GenerateTextureSample);
@@ -242,6 +236,16 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         private static OperationResult GenerateBarrier(CodeGenContext context, AstOperation operation)
         {
+            // Barrier on divergent control flow paths may cause the GPU to hang,
+            // so skip emitting the barrier for those cases.
+            if (!context.Config.GpuAccessor.QueryHostSupportsShaderBarrierDivergence() &&
+                (context.CurrentBlock.Type != AstBlockType.Main || context.MayHaveReturned || !context.IsMainFunction))
+            {
+                context.Config.GpuAccessor.Log($"Shader has barrier on potentially divergent block, the barrier will be removed.");
+
+                return OperationResult.Invalid;
+            }
+
             context.ControlBarrier(
                 context.Constant(context.TypeU32(), Scope.Workgroup),
                 context.Constant(context.TypeU32(), Scope.Workgroup),
@@ -861,30 +865,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             return GenerateLoadOrStore(context, operation, isStore: false);
         }
 
-        private static OperationResult GenerateLoadLocal(CodeGenContext context, AstOperation operation)
-        {
-            return GenerateLoadLocalOrShared(context, operation, StorageClass.Private, context.LocalMemory);
-        }
-
-        private static OperationResult GenerateLoadShared(CodeGenContext context, AstOperation operation)
-        {
-            return GenerateLoadLocalOrShared(context, operation, StorageClass.Workgroup, context.SharedMemory);
-        }
-
-        private static OperationResult GenerateLoadLocalOrShared(
-            CodeGenContext context,
-            AstOperation operation,
-            StorageClass storageClass,
-            SpvInstruction memory)
-        {
-            var offset = context.Get(AggregateType.S32, operation.GetSource(0));
-
-            var elemPointer = context.AccessChain(context.TypePointer(storageClass, context.TypeU32()), memory, offset);
-            var value = context.Load(context.TypeU32(), elemPointer);
-
-            return new OperationResult(AggregateType.U32, value);
-        }
-
         private static OperationResult GenerateLod(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -1092,6 +1072,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         private static OperationResult GenerateReturn(CodeGenContext context, AstOperation operation)
         {
+            context.MayHaveReturned = true;
+
             if (operation.SourcesCount != 0)
             {
                 context.ReturnValue(context.Get(context.CurrentFunction.ReturnType, operation.GetSource(0)));
@@ -1256,45 +1238,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             return GenerateLoadOrStore(context, operation, isStore: true);
         }
 
-        private static OperationResult GenerateStoreLocal(CodeGenContext context, AstOperation operation)
-        {
-            return GenerateStoreLocalOrShared(context, operation, StorageClass.Private, context.LocalMemory);
-        }
-
-        private static OperationResult GenerateStoreShared(CodeGenContext context, AstOperation operation)
-        {
-            return GenerateStoreLocalOrShared(context, operation, StorageClass.Workgroup, context.SharedMemory);
-        }
-
-        private static OperationResult GenerateStoreLocalOrShared(
-            CodeGenContext context,
-            AstOperation operation,
-            StorageClass storageClass,
-            SpvInstruction memory)
-        {
-            var offset = context.Get(AggregateType.S32, operation.GetSource(0));
-            var value = context.Get(AggregateType.U32, operation.GetSource(1));
-
-            var elemPointer = context.AccessChain(context.TypePointer(storageClass, context.TypeU32()), memory, offset);
-            context.Store(elemPointer, value);
-
-            return OperationResult.Invalid;
-        }
-
-        private static OperationResult GenerateStoreShared16(CodeGenContext context, AstOperation operation)
-        {
-            GenerateStoreSharedSmallInt(context, operation, 16);
-
-            return OperationResult.Invalid;
-        }
-
-        private static OperationResult GenerateStoreShared8(CodeGenContext context, AstOperation operation)
-        {
-            GenerateStoreSharedSmallInt(context, operation, 8);
-
-            return OperationResult.Invalid;
-        }
-
         private static OperationResult GenerateSubtract(CodeGenContext context, AstOperation operation)
         {
             return GenerateBinary(context, operation, context.Delegates.FSub, context.Delegates.ISub);
@@ -1815,55 +1758,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             AstOperation operation,
             Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
         {
-            var value = context.GetU32(operation.GetSource(operation.SourcesCount - 1));
+            SpvInstruction elemPointer = GetStoragePointer(context, operation, out AggregateType varType);
 
-            SpvInstruction elemPointer;
+            var value = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
-
-            if (operation.StorageKind == StorageKind.StorageBuffer)
-            {
-                elemPointer = GetStoragePointer(context, operation, out _);
-            }
-            else if (operation.StorageKind == StorageKind.SharedMemory)
-            {
-                var offset = context.GetU32(operation.GetSource(0));
-                elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
-            }
-            else
-            {
-                throw new InvalidOperationException($"Invalid storage kind \"{operation.StorageKind}\".");
-            }
 
             var one = context.Constant(context.TypeU32(), 1);
             var zero = context.Constant(context.TypeU32(), 0);
 
-            return new OperationResult(AggregateType.U32, emitU(context.TypeU32(), elemPointer, one, zero, value));
+            return new OperationResult(varType, emitU(context.GetType(varType), elemPointer, one, zero, value));
         }
 
         private static OperationResult GenerateAtomicMemoryCas(CodeGenContext context, AstOperation operation)
         {
-            var value0 = context.GetU32(operation.GetSource(operation.SourcesCount - 2));
+            SpvInstruction elemPointer = GetStoragePointer(context, operation, out AggregateType varType);
-            var value1 = context.GetU32(operation.GetSource(operation.SourcesCount - 1));
 
-            SpvInstruction elemPointer;
+            var value0 = context.Get(varType, operation.GetSource(operation.SourcesCount - 2));
-
+            var value1 = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
-            if (operation.StorageKind == StorageKind.StorageBuffer)
-            {
-                elemPointer = GetStoragePointer(context, operation, out _);
-            }
-            else if (operation.StorageKind == StorageKind.SharedMemory)
-            {
-                var offset = context.GetU32(operation.GetSource(0));
-                elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
-            }
-            else
-            {
-                throw new InvalidOperationException($"Invalid storage kind \"{operation.StorageKind}\".");
-            }
 
             var one = context.Constant(context.TypeU32(), 1);
             var zero = context.Constant(context.TypeU32(), 0);
 
-            return new OperationResult(AggregateType.U32, context.AtomicCompareExchange(context.TypeU32(), elemPointer, one, zero, zero, value1, value0));
+            return new OperationResult(varType, context.AtomicCompareExchange(context.GetType(varType), elemPointer, one, zero, zero, value1, value0));
         }
 
         private static OperationResult GenerateLoadOrStore(CodeGenContext context, AstOperation operation, bool isStore)
@@ -1916,6 +1831,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                         : context.StorageBuffers[bindingIndex.Value];
                     break;
 
+                case StorageKind.LocalMemory:
+                case StorageKind.SharedMemory:
+                    if (!(operation.GetSource(srcIndex++) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
+                    {
+                        throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
+                    }
+
+                    if (storageKind == StorageKind.LocalMemory)
+                    {
+                        storageClass = StorageClass.Private;
+                        varType = context.Config.Properties.LocalMemories[bindingId.Value].Type & AggregateType.ElementTypeMask;
+                        baseObj = context.LocalMemories[bindingId.Value];
+                    }
+                    else
+                    {
+                        storageClass = StorageClass.Workgroup;
+                        varType = context.Config.Properties.SharedMemories[bindingId.Value].Type & AggregateType.ElementTypeMask;
+                        baseObj = context.SharedMemories[bindingId.Value];
+                    }
+                    break;
+
                 case StorageKind.Input:
                 case StorageKind.InputPerPatch:
                 case StorageKind.Output:
@@ -2036,50 +1972,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             return context.Load(context.GetType(varType), context.Inputs[ioDefinition]);
         }
 
-        private static void GenerateStoreSharedSmallInt(CodeGenContext context, AstOperation operation, int bitSize)
-        {
-            var offset = context.Get(AggregateType.U32, operation.GetSource(0));
-            var value = context.Get(AggregateType.U32, operation.GetSource(1));
-
-            var wordOffset = context.ShiftRightLogical(context.TypeU32(), offset, context.Constant(context.TypeU32(), 2));
-            var bitOffset = context.BitwiseAnd(context.TypeU32(), offset, context.Constant(context.TypeU32(), 3));
-            bitOffset = context.ShiftLeftLogical(context.TypeU32(), bitOffset, context.Constant(context.TypeU32(), 3));
-
-            var memory = context.SharedMemory;
-
-            var elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), memory, wordOffset);
-
-            GenerateStoreSmallInt(context, elemPointer, bitOffset, value, bitSize);
-        }
-
-        private static void GenerateStoreSmallInt(
-            CodeGenContext context,
-            SpvInstruction elemPointer,
-            SpvInstruction bitOffset,
-            SpvInstruction value,
-            int bitSize)
-        {
-            var loopStart = context.Label();
-            var loopEnd = context.Label();
-
-            context.Branch(loopStart);
-            context.AddLabel(loopStart);
-
-            var oldValue = context.Load(context.TypeU32(), elemPointer);
-            var newValue = context.BitFieldInsert(context.TypeU32(), oldValue, value, bitOffset, context.Constant(context.TypeU32(), bitSize));
-
-            var one = context.Constant(context.TypeU32(), 1);
-            var zero = context.Constant(context.TypeU32(), 0);
-
-            var result = context.AtomicCompareExchange(context.TypeU32(), elemPointer, one, zero, zero, newValue, oldValue);
-            var failed = context.INotEqual(context.TypeBool(), result, oldValue);
-
-            context.LoopMerge(loopEnd, loopStart, LoopControlMask.MaskNone);
-            context.BranchConditional(failed, loopStart, loopEnd);
-
-            context.AddLabel(loopEnd);
-        }
-
         private static OperationResult GetZeroOperationResult(
             CodeGenContext context,
             AstTextureOperation texOp,

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

@@ -148,7 +148,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             context.CurrentFunction = function;
             context.AddFunction(spvFunc);
-            context.StartFunction();
+            context.StartFunction(isMainFunction: funcIndex == 0);
 
             Declarations.DeclareParameters(context, function);
 

src/Ryujinx.Graphics.Shader/Constants.cs

@@ -10,5 +10,11 @@ namespace Ryujinx.Graphics.Shader
         public const int NvnBaseVertexByteOffset = 0x640;
         public const int NvnBaseInstanceByteOffset = 0x644;
         public const int NvnDrawIndexByteOffset = 0x648;
+
+        // Transform Feedback emulation.
+
+        public const int TfeInfoBinding = 0;
+        public const int TfeBufferBaseBinding = 1;
+        public const int TfeBuffersCount = 4;
     }
 }

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

@@ -247,6 +247,17 @@ namespace Ryujinx.Graphics.Shader.Decoders
                 {
                     block.AddPushOp(op);
                 }
+                else if (op.Name == InstName.Ldl || op.Name == InstName.Stl)
+                {
+                    config.SetUsedFeature(FeatureFlags.LocalMemory);
+                }
+                else if (op.Name == InstName.Atoms ||
+                         op.Name == InstName.AtomsCas ||
+                         op.Name == InstName.Lds ||
+                         op.Name == InstName.Sts)
+                {
+                    config.SetUsedFeature(FeatureFlags.SharedMemory);
+                }
 
                 block.OpCodes.Add(op);