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/ISSUE_TEMPLATE/**'
+      - '.github/**'
       - '*.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/ISSUE_TEMPLATE/**'
+      - '.github/**'
       - '*.yml'
+      - '*.json'
+      - '*.config'
       - 'README.md'
 
 concurrency: release

Directory.Packages.props

@@ -46,7 +46,7 @@
     <PackageVersion Include="System.Drawing.Common" Version="7.0.0" />
     <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.Arm64AeseV,         new IntrinsicInfo(0x4e284800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AesdV,         new IntrinsicInfo(0x4e285800u, IntrinsicType.Vector128BinaryRd));
+            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 {
-                        0b00 => Intrinsic.Arm64FcvtauS,
-                        0b01 => Intrinsic.Arm64FcvtnuS,
-                        0b10 => Intrinsic.Arm64FcvtpuS,
-                        0b11 => Intrinsic.Arm64FcvtmuS,
-                        _ => throw new ArgumentOutOfRangeException(nameof(rm))
-                    };
+                    Operand m = GetVecA32(op.Vm >> 1);
+
+                    Operand toConvert = InstEmitSimdHelper32Arm64.EmitExtractScalar(context, m, op.Vm, true);
+
+                    if (unsigned)
+                    {
+                        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 {
-                        0b00 => Intrinsic.Arm64FcvtasS,
-                        0b01 => Intrinsic.Arm64FcvtnsS,
-                        0b10 => Intrinsic.Arm64FcvtpsS,
-                        0b11 => Intrinsic.Arm64FcvtmsS,
-                        _ => throw new ArgumentOutOfRangeException(nameof(rm))
-                    };
-                }
+                    if (unsigned)
+                    {
+                        inst = rm switch {
+                            0b00 => Intrinsic.Arm64FcvtauS,
+                            0b01 => Intrinsic.Arm64FcvtnuS,
+                            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;
-            EmitScalarUnaryOpSimd32(context, (m) => (inst == 0) ? m : context.AddIntrinsic(inst, m));
+            EmitScalarUnaryOpF32(context, inst, (op.Size & 1) != 0);
+        }
+
+        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/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 const int PageMask = PageSize - 1;
+        private static readonly int PageSize = (int)MemoryBlock.GetPageSize();
+        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/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)
-        {
-            Context?.MakeCurrent(window);
+                Logger.Warning?.Print(LogClass.Ui, $"Failed to {(!unbind ? "bind" : "unbind")} OpenGL context: {e}");
+            }
         }
 
         public void SwapBuffers()

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.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));
-                _privateTree.Add(new PrivateMapping(0UL, asSize, default));
+                _mappingTree.Add(new Mapping(0UL, addressSpaceSize, MappingType.None));
+                _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);
-            Mirror = new MemoryBlock(asSize, asFlags);
+            ulong minAddressSpaceSize = Math.Min(asSize, 1UL << 36);
+
+            // 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="addressSpaceSize">Size of the address space</param>
+        /// <param name="addressSpace">Address space instance to use</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

@@ -28,6 +28,7 @@ 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;
@@ -77,6 +78,7 @@ namespace Ryujinx.Graphics.GAL
             bool supportsFragmentShaderOrderingIntel,
             bool supportsGeometryShader,
             bool supportsGeometryShaderPassthrough,
+            bool supportsTransformFeedback,
             bool supportsImageLoadFormatted,
             bool supportsLayerVertexTessellation,
             bool supportsMismatchingViewFormat,
@@ -122,6 +124,7 @@ namespace Ryujinx.Graphics.GAL
             SupportsFragmentShaderOrderingIntel = supportsFragmentShaderOrderingIntel;
             SupportsGeometryShader = supportsGeometryShader;
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
+            SupportsTransformFeedback = supportsTransformFeedback;
             SupportsImageLoadFormatted = supportsImageLoadFormatted;
             SupportsLayerVertexTessellation = supportsLayerVertexTessellation;
             SupportsMismatchingViewFormat = supportsMismatchingViewFormat;

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/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/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];
-
-                for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
+                if (_context.Capabilities.SupportsTransformFeedback)
                 {
-                    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;
-                        continue;
+                        BufferBounds tfb = _transformFeedbackBuffers[index];
+
+                        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.

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 = 5044;
+        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.
-                return GetBindingFromIndex(index, _context.Capabilities.MaximumUniformBuffersPerStage, "Uniform buffer") + 1;
+                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumUniformBuffersPerStage, "Uniform buffer");
             }
             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)
@@ -149,6 +165,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         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 storageBinding = stageIndex * storagesPerStage;
+            int uniformBinding = _reservedConstantBuffers + stageIndex * uniformsPerStage;
+            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

@@ -153,6 +153,7 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsFragmentShaderOrderingIntel: HwCapabilities.SupportsFragmentShaderOrdering,
                 supportsGeometryShader: true,
                 supportsGeometryShaderPassthrough: HwCapabilities.SupportsGeometryShaderPassthrough,
+                supportsTransformFeedback: true,
                 supportsImageLoadFormatted: HwCapabilities.SupportsImageLoadFormatted,
                 supportsLayerVertexTessellation: HwCapabilities.SupportsShaderViewportLayerArray,
                 supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat,

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/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 Instruction SharedMemory { get; set; }
+        public Dictionary<int, Instruction> LocalMemories { get; } = new Dictionary<int, Instruction>();
+        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[]>();

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.SharedMemory = DeclareMemory(context, StorageClass.Workgroup, size);
-        }
+                context.AddGlobalVariable(variable);
 
-        private static SpvInstruction DeclareMemory(CodeGenContext context, StorageClass storage, int size)
-        {
-            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;
+                dict.Add(id, 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);
@@ -871,30 +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;
@@ -1268,45 +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);
@@ -1827,55 +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;
-
-            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 value = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
 
             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));
-            var value1 = context.GetU32(operation.GetSource(operation.SourcesCount - 1));
+            SpvInstruction elemPointer = GetStoragePointer(context, operation, out AggregateType varType);
 
-            SpvInstruction elemPointer;
-
-            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 value0 = context.Get(varType, operation.GetSource(operation.SourcesCount - 2));
+            var value1 = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
 
             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)
@@ -1928,6 +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:
@@ -2048,50 +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/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);
 

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -367,6 +367,15 @@ namespace Ryujinx.Graphics.Shader
             return true;
         }
 
+        /// <summary>
+        /// Queries host GPU transform feedback support.
+        /// </summary>
+        /// <returns>True if the GPU and driver supports transform feedback, false otherwise</returns>
+        bool QueryHostSupportsTransformFeedback()
+        {
+            return true;
+        }
+
         /// <summary>
         /// Queries host support for writes to the viewport index from vertex or tessellation shader stages.
         /// </summary>

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

@@ -10,12 +10,6 @@ namespace Ryujinx.Graphics.Shader.Instructions
 {
     static partial class InstEmit
     {
-        private enum MemoryRegion
-        {
-            Local,
-            Shared
-        }
-
         public static void Atom(EmitterContext context)
         {
             InstAtom op = context.GetOp<InstAtom>();
@@ -33,6 +27,12 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
         public static void Atoms(EmitterContext context)
         {
+            if (context.Config.Stage != ShaderStage.Compute)
+            {
+                context.Config.GpuAccessor.Log($"Atoms instruction is not valid on \"{context.Config.Stage}\" stage.");
+                return;
+            }
+
             InstAtoms op = context.GetOp<InstAtoms>();
 
             Operand offset = context.ShiftRightU32(GetSrcReg(context, op.SrcA), Const(2));
@@ -51,7 +51,8 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 _ => AtomSize.U32
             };
 
-            Operand res = EmitAtomicOp(context, StorageKind.SharedMemory, op.AtomOp, size, offset, Const(0), value);
+            Operand id = Const(context.Config.ResourceManager.SharedMemoryId);
+            Operand res = EmitAtomicOp(context, StorageKind.SharedMemory, op.AtomOp, size, id, offset, value);
 
             context.Copy(GetDest(op.Dest), res);
         }
@@ -114,14 +115,20 @@ namespace Ryujinx.Graphics.Shader.Instructions
         {
             InstLdl op = context.GetOp<InstLdl>();
 
-            EmitLoad(context, MemoryRegion.Local, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
+            EmitLoad(context, StorageKind.LocalMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
         }
 
         public static void Lds(EmitterContext context)
         {
+            if (context.Config.Stage != ShaderStage.Compute)
+            {
+                context.Config.GpuAccessor.Log($"Lds instruction is not valid on \"{context.Config.Stage}\" stage.");
+                return;
+            }
+
             InstLds op = context.GetOp<InstLds>();
 
-            EmitLoad(context, MemoryRegion.Shared, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
+            EmitLoad(context, StorageKind.SharedMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
         }
 
         public static void Red(EmitterContext context)
@@ -144,14 +151,20 @@ namespace Ryujinx.Graphics.Shader.Instructions
         {
             InstStl op = context.GetOp<InstStl>();
 
-            EmitStore(context, MemoryRegion.Local, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
+            EmitStore(context, StorageKind.LocalMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
         }
 
         public static void Sts(EmitterContext context)
         {
+            if (context.Config.Stage != ShaderStage.Compute)
+            {
+                context.Config.GpuAccessor.Log($"Sts instruction is not valid on \"{context.Config.Stage}\" stage.");
+                return;
+            }
+
             InstSts op = context.GetOp<InstSts>();
 
-            EmitStore(context, MemoryRegion.Shared, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
+            EmitStore(context, StorageKind.SharedMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
         }
 
         private static Operand EmitLoadConstant(EmitterContext context, Operand slot, Operand offset)
@@ -192,8 +205,8 @@ namespace Ryujinx.Graphics.Shader.Instructions
             StorageKind storageKind,
             AtomOp op,
             AtomSize type,
-            Operand addrLow,
-            Operand addrHigh,
+            Operand e0,
+            Operand e1,
             Operand value)
         {
             Operand res = Const(0);
@@ -203,7 +216,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.Add:
                     if (type == AtomSize.S32 || type == AtomSize.U32)
                     {
-                        res = context.AtomicAdd(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicAdd(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -213,7 +226,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.And:
                     if (type == AtomSize.S32 || type == AtomSize.U32)
                     {
-                        res = context.AtomicAnd(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicAnd(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -223,7 +236,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.Xor:
                     if (type == AtomSize.S32 || type == AtomSize.U32)
                     {
-                        res = context.AtomicXor(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicXor(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -233,7 +246,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.Or:
                     if (type == AtomSize.S32 || type == AtomSize.U32)
                     {
-                        res = context.AtomicOr(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicOr(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -243,11 +256,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.Max:
                     if (type == AtomSize.S32)
                     {
-                        res = context.AtomicMaxS32(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicMaxS32(storageKind, e0, e1, value);
                     }
                     else if (type == AtomSize.U32)
                     {
-                        res = context.AtomicMaxU32(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicMaxU32(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -257,11 +270,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 case AtomOp.Min:
                     if (type == AtomSize.S32)
                     {
-                        res = context.AtomicMinS32(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicMinS32(storageKind, e0, e1, value);
                     }
                     else if (type == AtomSize.U32)
                     {
-                        res = context.AtomicMinU32(storageKind, addrLow, addrHigh, value);
+                        res = context.AtomicMinU32(storageKind, e0, e1, value);
                     }
                     else
                     {
@@ -275,7 +288,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
         private static void EmitLoad(
             EmitterContext context,
-            MemoryRegion region,
+            StorageKind storageKind,
             LsSize2 size,
             Operand srcA,
             int rd,
@@ -287,19 +300,19 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 return;
             }
 
+            int id = storageKind == StorageKind.LocalMemory
+                ? context.Config.ResourceManager.LocalMemoryId
+                : context.Config.ResourceManager.SharedMemoryId;
             bool isSmallInt = size < LsSize2.B32;
 
-            int count = 1;
-
-            switch (size)
+            int count = size switch
             {
-                case LsSize2.B64: count = 2; break;
-                case LsSize2.B128: count = 4; break;
-            }
+                LsSize2.B64 => 2,
+                LsSize2.B128 => 4,
+                _ => 1
+            };
 
-            Operand baseOffset = context.IAdd(srcA, Const(offset));
-            Operand wordOffset = context.ShiftRightU32(baseOffset, Const(2)); // Word offset = byte offset / 4 (one word = 4 bytes).
-            Operand bitOffset = GetBitOffset(context, baseOffset);
+            Operand baseOffset = context.Copy(srcA);
 
             for (int index = 0; index < count; index++)
             {
@@ -310,14 +323,10 @@ namespace Ryujinx.Graphics.Shader.Instructions
                     break;
                 }
 
-                Operand elemOffset = context.IAdd(wordOffset, Const(index));
-                Operand value = null;
-
-                switch (region)
-                {
-                    case MemoryRegion.Local: value = context.LoadLocal(elemOffset); break;
-                    case MemoryRegion.Shared: value = context.LoadShared(elemOffset); break;
-                }
+                Operand byteOffset = context.IAdd(baseOffset, Const(offset + index * 4));
+                Operand wordOffset = context.ShiftRightU32(byteOffset, Const(2)); // Word offset = byte offset / 4 (one word = 4 bytes).
+                Operand bitOffset = GetBitOffset(context, byteOffset);
+                Operand value = context.Load(storageKind, id, wordOffset);
 
                 if (isSmallInt)
                 {
@@ -360,7 +369,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
         private static void EmitStore(
             EmitterContext context,
-            MemoryRegion region,
+            StorageKind storageKind,
             LsSize2 size,
             Operand srcA,
             int rd,
@@ -372,52 +381,54 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 return;
             }
 
+            int id = storageKind == StorageKind.LocalMemory
+                ? context.Config.ResourceManager.LocalMemoryId
+                : context.Config.ResourceManager.SharedMemoryId;
             bool isSmallInt = size < LsSize2.B32;
 
-            int count = 1;
-
-            switch (size)
+            int count = size switch
             {
-                case LsSize2.B64: count = 2; break;
-                case LsSize2.B128: count = 4; break;
-            }
+                LsSize2.B64 => 2,
+                LsSize2.B128 => 4,
+                _ => 1
+            };
 
-            Operand baseOffset = context.IAdd(srcA, Const(offset));
-            Operand wordOffset = context.ShiftRightU32(baseOffset, Const(2));
-            Operand bitOffset = GetBitOffset(context, baseOffset);
+            Operand baseOffset = context.Copy(srcA);
 
             for (int index = 0; index < count; index++)
             {
                 bool isRz = rd + index >= RegisterConsts.RegisterZeroIndex;
 
                 Operand value = Register(isRz ? rd : rd + index, RegisterType.Gpr);
-                Operand elemOffset = context.IAdd(wordOffset, Const(index));
+                Operand byteOffset = context.IAdd(baseOffset, Const(offset + index * 4));
+                Operand wordOffset = context.ShiftRightU32(byteOffset, Const(2));
+                Operand bitOffset = GetBitOffset(context, byteOffset);
 
-                if (isSmallInt && region == MemoryRegion.Local)
+                if (isSmallInt && storageKind == StorageKind.LocalMemory)
                 {
-                    Operand word = context.LoadLocal(elemOffset);
+                    Operand word = context.Load(storageKind, id, wordOffset);
 
                     value = InsertSmallInt(context, (LsSize)size, bitOffset, word, value);
                 }
 
-                if (region == MemoryRegion.Local)
+                if (storageKind == StorageKind.LocalMemory)
                 {
-                    context.StoreLocal(elemOffset, value);
+                    context.Store(storageKind, id, wordOffset, value);
                 }
-                else if (region == MemoryRegion.Shared)
+                else if (storageKind == StorageKind.SharedMemory)
                 {
                     switch (size)
                     {
                         case LsSize2.U8:
                         case LsSize2.S8:
-                            context.StoreShared8(baseOffset, value);
+                            context.Store(StorageKind.SharedMemory8, id, byteOffset, value);
                             break;
                         case LsSize2.U16:
                         case LsSize2.S16:
-                            context.StoreShared16(baseOffset, value);
+                            context.Store(StorageKind.SharedMemory16, id, byteOffset, value);
                             break;
                         default:
-                            context.StoreShared(elemOffset, value);
+                            context.Store(storageKind, id, wordOffset, value);
                             break;
                     }
                 }

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

@@ -79,8 +79,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
         ImageAtomic,
         IsNan,
         Load,
-        LoadLocal,
-        LoadShared,
         Lod,
         LogarithmB2,
         LogicalAnd,
@@ -115,10 +113,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
         Sine,
         SquareRoot,
         Store,
-        StoreLocal,
-        StoreShared,
-        StoreShared16,
-        StoreShared8,
         Subtract,
         SwizzleAdd,
         TextureSample,

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

@@ -11,12 +11,13 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
         StorageBuffer,
         LocalMemory,
         SharedMemory,
+        SharedMemory8, // TODO: Remove this and store type as a field on the Operation class itself.
+        SharedMemory16, // TODO: Remove this and store type as a field on the Operation class itself.
         GlobalMemory,
-        // TODO: Remove those and store type as a field on the Operation class itself.
-        GlobalMemoryS8,
-        GlobalMemoryS16,
-        GlobalMemoryU8,
-        GlobalMemoryU16
+        GlobalMemoryS8, // TODO: Remove this and store type as a field on the Operation class itself.
+        GlobalMemoryS16, // TODO: Remove this and store type as a field on the Operation class itself.
+        GlobalMemoryU8, // TODO: Remove this and store type as a field on the Operation class itself.
+        GlobalMemoryU16 // TODO: Remove this and store type as a field on the Operation class itself.
     }
 
     static class StorageKindExtensions

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

@@ -10,14 +10,12 @@
   </ItemGroup>
 
   <ItemGroup>
-    <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\AtomicMinMaxS32Shared.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\MultiplyHighS32.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\MultiplyHighU32.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\Shuffle.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleDown.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleUp.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleXor.glsl" />
-    <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\StoreSharedSmallInt.glsl" />
     <EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\SwizzleAdd.glsl" />
   </ItemGroup>
 

src/Ryujinx.Graphics.Shader/StructuredIr/HelperFunctionsMask.cs

@@ -5,15 +5,13 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
     [Flags]
     enum HelperFunctionsMask
     {
-        AtomicMinMaxS32Shared  = 1 << 0,
-        MultiplyHighS32        = 1 << 2,
-        MultiplyHighU32        = 1 << 3,
-        Shuffle                = 1 << 4,
-        ShuffleDown            = 1 << 5,
-        ShuffleUp              = 1 << 6,
-        ShuffleXor             = 1 << 7,
-        StoreSharedSmallInt    = 1 << 8,
-        SwizzleAdd             = 1 << 10,
-        FSI                    = 1 << 11
+        MultiplyHighS32 = 1 << 2,
+        MultiplyHighU32 = 1 << 3,
+        Shuffle         = 1 << 4,
+        ShuffleDown     = 1 << 5,
+        ShuffleUp       = 1 << 6,
+        ShuffleXor      = 1 << 7,
+        SwizzleAdd      = 1 << 10,
+        FSI             = 1 << 11
     }
 }

src/Ryujinx.Graphics.Shader/StructuredIr/InstructionInfo.cs

@@ -90,8 +90,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             Add(Instruction.ImageAtomic,              AggregateType.S32);
             Add(Instruction.IsNan,                    AggregateType.Bool,   AggregateType.Scalar);
             Add(Instruction.Load,                     AggregateType.FP32);
-            Add(Instruction.LoadLocal,                AggregateType.U32,    AggregateType.S32);
-            Add(Instruction.LoadShared,               AggregateType.U32,    AggregateType.S32);
             Add(Instruction.Lod,                      AggregateType.FP32);
             Add(Instruction.LogarithmB2,              AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.LogicalAnd,               AggregateType.Bool,   AggregateType.Bool,    AggregateType.Bool);
@@ -121,10 +119,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             Add(Instruction.Sine,                     AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.SquareRoot,               AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.Store,                    AggregateType.Void);
-            Add(Instruction.StoreLocal,               AggregateType.Void,   AggregateType.S32,     AggregateType.U32);
-            Add(Instruction.StoreShared,              AggregateType.Void,   AggregateType.S32,     AggregateType.U32);
-            Add(Instruction.StoreShared16,            AggregateType.Void,   AggregateType.S32,     AggregateType.U32);
-            Add(Instruction.StoreShared8,             AggregateType.Void,   AggregateType.S32,     AggregateType.U32);
             Add(Instruction.Subtract,                 AggregateType.Scalar, AggregateType.Scalar,  AggregateType.Scalar);
             Add(Instruction.SwizzleAdd,               AggregateType.FP32,   AggregateType.FP32,    AggregateType.FP32,    AggregateType.S32);
             Add(Instruction.TextureSample,            AggregateType.FP32);

src/Ryujinx.Graphics.Shader/StructuredIr/MemoryDefinition.cs

@@ -0,0 +1,18 @@
+using Ryujinx.Graphics.Shader.Translation;
+
+namespace Ryujinx.Graphics.Shader.StructuredIr
+{
+    readonly struct MemoryDefinition
+    {
+        public string Name { get; }
+        public AggregateType Type { get; }
+        public int ArrayLength { get; }
+
+        public MemoryDefinition(string name, AggregateType type, int arrayLength = 1)
+        {
+            Name = name;
+            Type = type;
+            ArrayLength = arrayLength;
+        }
+    }
+}

src/Ryujinx.Graphics.Shader/StructuredIr/ShaderProperties.cs

@@ -6,14 +6,20 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
     {
         private readonly Dictionary<int, BufferDefinition> _constantBuffers;
         private readonly Dictionary<int, BufferDefinition> _storageBuffers;
+        private readonly Dictionary<int, MemoryDefinition> _localMemories;
+        private readonly Dictionary<int, MemoryDefinition> _sharedMemories;
 
         public IReadOnlyDictionary<int, BufferDefinition> ConstantBuffers => _constantBuffers;
         public IReadOnlyDictionary<int, BufferDefinition> StorageBuffers => _storageBuffers;
+        public IReadOnlyDictionary<int, MemoryDefinition> LocalMemories => _localMemories;
+        public IReadOnlyDictionary<int, MemoryDefinition> SharedMemories => _sharedMemories;
 
         public ShaderProperties()
         {
             _constantBuffers = new Dictionary<int, BufferDefinition>();
             _storageBuffers = new Dictionary<int, BufferDefinition>();
+            _localMemories = new Dictionary<int, MemoryDefinition>();
+            _sharedMemories = new Dictionary<int, MemoryDefinition>();
         }
 
         public void AddConstantBuffer(int binding, BufferDefinition definition)
@@ -25,5 +31,21 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
         {
             _storageBuffers[binding] = definition;
         }
+
+        public int AddLocalMemory(MemoryDefinition definition)
+        {
+            int id = _localMemories.Count;
+            _localMemories.Add(id, definition);
+
+            return id;
+        }
+
+        public int AddSharedMemory(MemoryDefinition definition)
+        {
+            int id = _sharedMemories.Count;
+            _sharedMemories.Add(id, definition);
+
+            return id;
+        }
     }
 }

src/Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs

@@ -274,13 +274,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             // decide which helper functions are needed on the final generated code.
             switch (operation.Inst)
             {
-                case Instruction.AtomicMaxS32:
-                case Instruction.AtomicMinS32:
-                    if (operation.StorageKind == StorageKind.SharedMemory)
-                    {
-                        context.Info.HelperFunctionsMask |= HelperFunctionsMask.AtomicMinMaxS32Shared;
-                    }
-                    break;
                 case Instruction.MultiplyHighS32:
                     context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighS32;
                     break;
@@ -299,10 +292,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
                 case Instruction.ShuffleXor:
                     context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleXor;
                     break;
-                case Instruction.StoreShared16:
-                case Instruction.StoreShared8:
-                    context.Info.HelperFunctionsMask |= HelperFunctionsMask.StoreSharedSmallInt;
-                    break;
                 case Instruction.SwizzleAdd:
                     context.Info.HelperFunctionsMask |= HelperFunctionsMask.SwizzleAdd;
                     break;

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

@@ -234,6 +234,45 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public void PrepareForVertexReturn()
         {
+            if (!Config.GpuAccessor.QueryHostSupportsTransformFeedback() && Config.GpuAccessor.QueryTransformFeedbackEnabled())
+            {
+                Operand vertexCount = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(1));
+
+                for (int tfbIndex = 0; tfbIndex < Constants.TfeBuffersCount; tfbIndex++)
+                {
+                    var locations = Config.GpuAccessor.QueryTransformFeedbackVaryingLocations(tfbIndex);
+                    var stride = Config.GpuAccessor.QueryTransformFeedbackStride(tfbIndex);
+
+                    Operand baseOffset = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(0), Const(tfbIndex));
+                    Operand baseVertex = this.Load(StorageKind.Input, IoVariable.BaseVertex);
+                    Operand baseInstance = this.Load(StorageKind.Input, IoVariable.BaseInstance);
+                    Operand vertexIndex = this.Load(StorageKind.Input, IoVariable.VertexIndex);
+                    Operand instanceIndex = this.Load(StorageKind.Input, IoVariable.InstanceIndex);
+
+                    Operand outputVertexOffset = this.ISubtract(vertexIndex, baseVertex);
+                    Operand outputInstanceOffset = this.ISubtract(instanceIndex, baseInstance);
+
+                    Operand outputBaseVertex = this.IMultiply(outputInstanceOffset, vertexCount);
+
+                    Operand vertexOffset = this.IMultiply(this.IAdd(outputBaseVertex, outputVertexOffset), Const(stride / 4));
+                    baseOffset = this.IAdd(baseOffset, vertexOffset);
+
+                    for (int j = 0; j < locations.Length; j++)
+                    {
+                        byte location = locations[j];
+                        if (location == 0xff)
+                        {
+                            continue;
+                        }
+
+                        Operand offset = this.IAdd(baseOffset, Const(j));
+                        Operand value = Instructions.AttributeMap.GenerateAttributeLoad(this, null, location * 4, isOutput: true, isPerPatch: false);
+
+                        this.Store(StorageKind.StorageBuffer, Constants.TfeBufferBaseBinding + tfbIndex, Const(0), offset, value);
+                    }
+                }
+            }
+
             if (Config.GpuAccessor.QueryViewportTransformDisable())
             {
                 Operand x = this.Load(StorageKind.Output, IoVariable.Position, null, Const(0));

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

@@ -67,6 +67,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.AtomicAnd, storageKind, Local(), Const(binding), e0, e1, value);
         }
 
+        public static Operand AtomicCompareAndSwap(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand compare, Operand value)
+        {
+            return context.Add(Instruction.AtomicCompareAndSwap, storageKind, Local(), Const(binding), e0, compare, value);
+        }
+
         public static Operand AtomicCompareAndSwap(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1, Operand compare, Operand value)
         {
             return context.Add(Instruction.AtomicCompareAndSwap, storageKind, Local(), Const(binding), e0, e1, compare, value);
@@ -661,16 +666,6 @@ namespace Ryujinx.Graphics.Shader.Translation
                 : context.Load(storageKind, (int)ioVariable, arrayIndex, elemIndex);
         }
 
-        public static Operand LoadLocal(this EmitterContext context, Operand a)
-        {
-            return context.Add(Instruction.LoadLocal, Local(), a);
-        }
-
-        public static Operand LoadShared(this EmitterContext context, Operand a)
-        {
-            return context.Add(Instruction.LoadShared, Local(), a);
-        }
-
         public static Operand MemoryBarrier(this EmitterContext context)
         {
             return context.Add(Instruction.MemoryBarrier);
@@ -753,6 +748,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.Store, storageKind, null, e0, e1, value);
         }
 
+        public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand value)
+        {
+            return context.Add(Instruction.Store, storageKind, null, Const(binding), e0, value);
+        }
+
         public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1, Operand value)
         {
             return context.Add(Instruction.Store, storageKind, null, Const(binding), e0, e1, value);
@@ -797,26 +797,6 @@ namespace Ryujinx.Graphics.Shader.Translation
                 : context.Add(Instruction.Store, storageKind, null, Const((int)ioVariable), arrayIndex, elemIndex, value);
         }
 
-        public static Operand StoreLocal(this EmitterContext context, Operand a, Operand b)
-        {
-            return context.Add(Instruction.StoreLocal, null, a, b);
-        }
-
-        public static Operand StoreShared(this EmitterContext context, Operand a, Operand b)
-        {
-            return context.Add(Instruction.StoreShared, null, a, b);
-        }
-
-        public static Operand StoreShared16(this EmitterContext context, Operand a, Operand b)
-        {
-            return context.Add(Instruction.StoreShared16, null, a, b);
-        }
-
-        public static Operand StoreShared8(this EmitterContext context, Operand a, Operand b)
-        {
-            return context.Add(Instruction.StoreShared8, null, a, b);
-        }
-
         public static Operand UnpackDouble2x32High(this EmitterContext context, Operand a)
         {
             return UnpackDouble2x32(context, a, 1);

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

@@ -21,6 +21,8 @@ namespace Ryujinx.Graphics.Shader.Translation
         RtLayer = 1 << 5,
         IaIndexing = 1 << 7,
         OaIndexing = 1 << 8,
-        FixedFuncAttr = 1 << 9
+        FixedFuncAttr = 1 << 9,
+        LocalMemory = 1 << 10,
+        SharedMemory = 1 << 11
     }
 }

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

@@ -9,13 +9,13 @@ namespace Ryujinx.Graphics.Shader.Translation
     class HelperFunctionManager
     {
         private readonly List<Function> _functionList;
-        private readonly Dictionary<HelperFunctionName, int> _functionIds;
+        private readonly Dictionary<int, int> _functionIds;
         private readonly ShaderStage _stage;
 
         public HelperFunctionManager(List<Function> functionList, ShaderStage stage)
         {
             _functionList = functionList;
-            _functionIds = new Dictionary<HelperFunctionName, int>();
+            _functionIds = new Dictionary<int, int>();
             _stage = stage;
         }
 
@@ -29,14 +29,30 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public int GetOrCreateFunctionId(HelperFunctionName functionName)
         {
-            if (_functionIds.TryGetValue(functionName, out int functionId))
+            if (_functionIds.TryGetValue((int)functionName, out int functionId))
             {
                 return functionId;
             }
 
             Function function = GenerateFunction(functionName);
             functionId = AddFunction(function);
-            _functionIds.Add(functionName, functionId);
+            _functionIds.Add((int)functionName, functionId);
+
+            return functionId;
+        }
+
+        public int GetOrCreateFunctionId(HelperFunctionName functionName, int id)
+        {
+            int key = (int)functionName | (id << 16);
+
+            if (_functionIds.TryGetValue(key, out int functionId))
+            {
+                return functionId;
+            }
+
+            Function function = GenerateFunction(functionName, id);
+            functionId = AddFunction(function);
+            _functionIds.Add(key, functionId);
 
             return functionId;
         }
@@ -140,6 +156,67 @@ namespace Ryujinx.Graphics.Shader.Translation
             return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "ConvertFloatToDouble", false, 1, 2);
         }
 
+        private static Function GenerateFunction(HelperFunctionName functionName, int id)
+        {
+            return functionName switch
+            {
+                HelperFunctionName.SharedAtomicMaxS32 => GenerateSharedAtomicSigned(id, isMin: false),
+                HelperFunctionName.SharedAtomicMinS32 => GenerateSharedAtomicSigned(id, isMin: true),
+                HelperFunctionName.SharedStore8 => GenerateSharedStore8(id),
+                HelperFunctionName.SharedStore16 => GenerateSharedStore16(id),
+                _ => throw new ArgumentException($"Invalid function name {functionName}")
+            };
+        }
+
+        private static Function GenerateSharedAtomicSigned(int id, bool isMin)
+        {
+            EmitterContext context = new EmitterContext();
+
+            Operand wordOffset = Argument(0);
+            Operand value = Argument(1);
+
+            Operand result = GenerateSharedAtomicCasLoop(context, wordOffset, id, (memValue) =>
+            {
+                return isMin
+                    ? context.IMinimumS32(memValue, value)
+                    : context.IMaximumS32(memValue, value);
+            });
+
+            context.Return(result);
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, $"SharedAtomic{(isMin ? "Min" : "Max")}_{id}", true, 2, 0);
+        }
+
+        private static Function GenerateSharedStore8(int id)
+        {
+            return GenerateSharedStore(id, 8);
+        }
+
+        private static Function GenerateSharedStore16(int id)
+        {
+            return GenerateSharedStore(id, 16);
+        }
+
+        private static Function GenerateSharedStore(int id, int bitSize)
+        {
+            EmitterContext context = new EmitterContext();
+
+            Operand offset = Argument(0);
+            Operand value = Argument(1);
+
+            Operand wordOffset = context.ShiftRightU32(offset, Const(2));
+            Operand bitOffset = GetBitOffset(context, offset);
+
+            GenerateSharedAtomicCasLoop(context, wordOffset, id, (memValue) =>
+            {
+                return context.BitfieldInsert(memValue, value, bitOffset, Const(bitSize));
+            });
+
+            context.Return();
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, $"SharedStore{bitSize}_{id}", false, 2, 0);
+        }
+
         private Function GenerateTexelFetchScaleFunction()
         {
             EmitterContext context = new EmitterContext();
@@ -226,5 +303,29 @@ namespace Ryujinx.Graphics.Shader.Translation
                     return context.IAdd(Const(1), index);
             }
         }
+
+        public static Operand GetBitOffset(EmitterContext context, Operand offset)
+        {
+            return context.ShiftLeft(context.BitwiseAnd(offset, Const(3)), Const(3));
+        }
+
+        private static Operand GenerateSharedAtomicCasLoop(EmitterContext context, Operand wordOffset, int id, Func<Operand, Operand> opCallback)
+        {
+            Operand lblLoopHead = Label();
+
+            context.MarkLabel(lblLoopHead);
+
+            Operand oldValue = context.Load(StorageKind.SharedMemory, id, wordOffset);
+            Operand newValue = opCallback(oldValue);
+
+            Operand casResult = context.AtomicCompareAndSwap(StorageKind.SharedMemory, id, wordOffset, oldValue, newValue);
+
+            Operand casFail = context.ICompareNotEqual(casResult, oldValue);
+
+            context.BranchIfTrue(lblLoopHead, casFail);
+
+            return oldValue;
+        }
+
     }
 }

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

@@ -4,6 +4,10 @@ namespace Ryujinx.Graphics.Shader.Translation
     {
         ConvertDoubleToFloat,
         ConvertFloatToDouble,
+        SharedAtomicMaxS32,
+        SharedAtomicMinS32,
+        SharedStore8,
+        SharedStore16,
         TexelFetchScale,
         TextureSizeUnscale
     }

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

@@ -244,7 +244,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                             node = nextNode;
                         }
                     }
-                    else if (operation.Inst == Instruction.StoreShared || operation.Inst == Instruction.StoreLocal)
+                    else if (operation.Inst == Instruction.Store &&
+                        (operation.StorageKind == StorageKind.SharedMemory ||
+                        operation.StorageKind == StorageKind.LocalMemory))
                     {
                         // The NVIDIA compiler can sometimes use shared or local memory as temporary
                         // storage to place the base address and size on, so we need
@@ -874,7 +876,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
                 if (bitSize < 32)
                 {
-                    Operand bitOffset = GetBitOffset(context, offset);
+                    Operand bitOffset = HelperFunctionManager.GetBitOffset(context, offset);
 
                     GenerateAtomicCasLoop(context, wordOffset, binding, (memValue) =>
                     {
@@ -892,7 +894,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
                 if (IsSmallInt(storageKind))
                 {
-                    Operand bitOffset = GetBitOffset(context, offset);
+                    Operand bitOffset = HelperFunctionManager.GetBitOffset(context, offset);
 
                     switch (storageKind)
                     {
@@ -921,11 +923,6 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static Operand GetBitOffset(EmitterContext context, Operand offset)
-        {
-            return context.ShiftLeft(context.BitwiseAnd(offset, Const(3)), Const(3));
-        }
-
         private static Operand GenerateAtomicCasLoop(EmitterContext context, Operand wordOffset, int binding, Func<Operand, Operand> opCallback)
         {
             Operand lblLoopHead = Label();
@@ -1070,15 +1067,18 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
         {
             baseOffset = null;
 
-            if (operation.Inst == Instruction.LoadShared || operation.Inst == Instruction.StoreShared)
+            if (operation.Inst == Instruction.Load || operation.Inst == Instruction.Store)
             {
-                type = LsMemoryType.Shared;
-                return TryGetSharedMemoryOffsets(operation, out baseOffset, out constOffset);
-            }
-            else if (operation.Inst == Instruction.LoadLocal || operation.Inst == Instruction.StoreLocal)
-            {
-                type = LsMemoryType.Local;
-                return TryGetLocalMemoryOffset(operation, out constOffset);
+                if (operation.StorageKind == StorageKind.SharedMemory)
+                {
+                    type = LsMemoryType.Shared;
+                    return TryGetSharedMemoryOffsets(operation, out baseOffset, out constOffset);
+                }
+                else if (operation.StorageKind == StorageKind.LocalMemory)
+                {
+                    type = LsMemoryType.Local;
+                    return TryGetLocalMemoryOffset(operation, out constOffset);
+                }
             }
 
             type = default;

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

@@ -1,3 +1,4 @@
+using Ryujinx.Common;
 using Ryujinx.Graphics.Shader.StructuredIr;
 using System;
 using System.Collections.Generic;
@@ -7,6 +8,11 @@ namespace Ryujinx.Graphics.Shader.Translation
 {
     class ResourceManager
     {
+        // Those values are used if the shader as local or shared memory access,
+        // but for some reason the supplied size was 0.
+        private const int DefaultLocalMemorySize = 128;
+        private const int DefaultSharedMemorySize = 4096;
+
         private static readonly string[] _stagePrefixes = new string[] { "cp", "vp", "tcp", "tep", "gp", "fp" };
 
         private readonly IGpuAccessor _gpuAccessor;
@@ -22,6 +28,9 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         private readonly HashSet<int> _usedConstantBufferBindings;
 
+        public int LocalMemoryId { get; private set; }
+        public int SharedMemoryId { get; private set; }
+
         public ShaderProperties Properties => _properties;
 
         public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor, ShaderProperties properties)
@@ -41,6 +50,47 @@ namespace Ryujinx.Graphics.Shader.Translation
             _usedConstantBufferBindings = new HashSet<int>();
 
             properties.AddConstantBuffer(0, new BufferDefinition(BufferLayout.Std140, 0, 0, "support_buffer", SupportBuffer.GetStructureType()));
+
+            LocalMemoryId = -1;
+            SharedMemoryId = -1;
+        }
+
+        public void SetCurrentLocalMemory(int size, bool isUsed)
+        {
+            if (isUsed)
+            {
+                if (size <= 0)
+                {
+                    size = DefaultLocalMemorySize;
+                }
+
+                var lmem = new MemoryDefinition("local_memory", AggregateType.Array | AggregateType.U32, BitUtils.DivRoundUp(size, sizeof(uint)));
+
+                LocalMemoryId = Properties.AddLocalMemory(lmem);
+            }
+            else
+            {
+                LocalMemoryId = -1;
+            }
+        }
+
+        public void SetCurrentSharedMemory(int size, bool isUsed)
+        {
+            if (isUsed)
+            {
+                if (size <= 0)
+                {
+                    size = DefaultSharedMemorySize;
+                }
+
+                var smem = new MemoryDefinition("shared_memory", AggregateType.Array | AggregateType.U32,  BitUtils.DivRoundUp(size, sizeof(uint)));
+
+                SharedMemoryId = Properties.AddSharedMemory(smem);
+            }
+            else
+            {
+                SharedMemoryId = -1;
+            }
         }
 
         public int GetConstantBufferBinding(int slot)

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

@@ -1,6 +1,8 @@
 using Ryujinx.Graphics.Shader.IntermediateRepresentation;
 using Ryujinx.Graphics.Shader.StructuredIr;
+using Ryujinx.Graphics.Shader.Translation.Optimizations;
 using System.Collections.Generic;
+using System.Diagnostics;
 using System.Linq;
 
 using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
@@ -70,6 +72,15 @@ namespace Ryujinx.Graphics.Shader.Translation
                             }
                         }
                     }
+                    else
+                    {
+                        node = InsertSharedStoreSmallInt(hfm, node);
+
+                        if (config.Options.TargetLanguage != TargetLanguage.Spirv)
+                        {
+                            node = InsertSharedAtomicSigned(hfm, node);
+                        }
+                    }
                 }
             }
         }
@@ -171,6 +182,87 @@ namespace Ryujinx.Graphics.Shader.Translation
             operation.TurnIntoCopy(result);
         }
 
+        private static LinkedListNode<INode> InsertSharedStoreSmallInt(HelperFunctionManager hfm, LinkedListNode<INode> node)
+        {
+            Operation operation = (Operation)node.Value;
+            HelperFunctionName name;
+
+            if (operation.StorageKind == StorageKind.SharedMemory8)
+            {
+                name = HelperFunctionName.SharedStore8;
+            }
+            else if (operation.StorageKind == StorageKind.SharedMemory16)
+            {
+                name = HelperFunctionName.SharedStore16;
+            }
+            else
+            {
+                return node;
+            }
+
+            if (operation.Inst != Instruction.Store)
+            {
+                return node;
+            }
+
+            Operand memoryId = operation.GetSource(0);
+            Operand byteOffset = operation.GetSource(1);
+            Operand value = operation.GetSource(2);
+
+            Debug.Assert(memoryId.Type == OperandType.Constant);
+
+            int functionId = hfm.GetOrCreateFunctionId(name, memoryId.Value);
+
+            Operand[] callArgs = new Operand[] { Const(functionId), byteOffset, value };
+
+            LinkedListNode<INode> newNode = node.List.AddBefore(node, new Operation(Instruction.Call, 0, (Operand)null, callArgs));
+
+            Utils.DeleteNode(node, operation);
+
+            return newNode;
+        }
+
+        private static LinkedListNode<INode> InsertSharedAtomicSigned(HelperFunctionManager hfm, LinkedListNode<INode> node)
+        {
+            Operation operation = (Operation)node.Value;
+            HelperFunctionName name;
+
+            if (operation.Inst == Instruction.AtomicMaxS32)
+            {
+                name = HelperFunctionName.SharedAtomicMaxS32;
+            }
+            else if (operation.Inst == Instruction.AtomicMinS32)
+            {
+                name = HelperFunctionName.SharedAtomicMinS32;
+            }
+            else
+            {
+                return node;
+            }
+
+            if (operation.StorageKind != StorageKind.SharedMemory)
+            {
+                return node;
+            }
+
+            Operand result = operation.Dest;
+            Operand memoryId = operation.GetSource(0);
+            Operand byteOffset = operation.GetSource(1);
+            Operand value = operation.GetSource(2);
+
+            Debug.Assert(memoryId.Type == OperandType.Constant);
+
+            int functionId = hfm.GetOrCreateFunctionId(name, memoryId.Value);
+
+            Operand[] callArgs = new Operand[] { Const(functionId), byteOffset, value };
+
+            LinkedListNode<INode> newNode = node.List.AddBefore(node, new Operation(Instruction.Call, 0, result, callArgs));
+
+            Utils.DeleteNode(node, operation);
+
+            return newNode;
+        }
+
         private static LinkedListNode<INode> InsertTexelFetchScale(HelperFunctionManager hfm, LinkedListNode<INode> node, ShaderConfig config)
         {
             TextureOperation texOp = (TextureOperation)node.Value;

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

@@ -124,14 +124,20 @@ namespace Ryujinx.Graphics.Shader.Translation
         private TextureDescriptor[] _cachedTextureDescriptors;
         private TextureDescriptor[] _cachedImageDescriptors;
 
-        public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options)
+        public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options, int localMemorySize)
         {
-            Stage       = stage;
-            GpuAccessor = gpuAccessor;
-            Options     = options;
+            Stage           = stage;
+            GpuAccessor     = gpuAccessor;
+            Options         = options;
+            LocalMemorySize = localMemorySize;
 
             _transformFeedbackDefinitions = new Dictionary<TransformFeedbackVariable, TransformFeedbackOutput>();
 
+            TransformFeedbackEnabled =
+                stage != ShaderStage.Compute &&
+                gpuAccessor.QueryTransformFeedbackEnabled() &&
+                gpuAccessor.QueryHostSupportsTransformFeedback();
+
             UsedInputAttributesPerPatch  = new HashSet<int>();
             UsedOutputAttributesPerPatch = new HashSet<int>();
 
@@ -139,6 +145,31 @@ namespace Ryujinx.Graphics.Shader.Translation
             _usedImages   = new Dictionary<TextureInfo, TextureMeta>();
 
             ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties());
+
+            if (!gpuAccessor.QueryHostSupportsTransformFeedback() && gpuAccessor.QueryTransformFeedbackEnabled())
+            {
+                StructureType tfeInfoStruct = new StructureType(new StructureField[]
+                {
+                    new StructureField(AggregateType.Array | AggregateType.U32, "base_offset", 4),
+                    new StructureField(AggregateType.U32, "vertex_count")
+                });
+
+                BufferDefinition tfeInfoBuffer = new BufferDefinition(BufferLayout.Std430, 1, Constants.TfeInfoBinding, "tfe_info", tfeInfoStruct);
+
+                Properties.AddStorageBuffer(Constants.TfeInfoBinding, tfeInfoBuffer);
+
+                StructureType tfeDataStruct = new StructureType(new StructureField[]
+                {
+                    new StructureField(AggregateType.Array | AggregateType.U32, "data", 0)
+                });
+
+                for (int i = 0; i < Constants.TfeBuffersCount; i++)
+                {
+                    int binding = Constants.TfeBufferBaseBinding + i;
+                    BufferDefinition tfeDataBuffer = new BufferDefinition(BufferLayout.Std430, 1, binding, $"tfe_data{i}", tfeDataStruct);
+                    Properties.AddStorageBuffer(binding, tfeDataBuffer);
+                }
+            }
         }
 
         public ShaderConfig(
@@ -146,29 +177,34 @@ namespace Ryujinx.Graphics.Shader.Translation
             OutputTopology outputTopology,
             int maxOutputVertices,
             IGpuAccessor gpuAccessor,
-            TranslationOptions options) : this(stage, gpuAccessor, options)
+            TranslationOptions options) : this(stage, gpuAccessor, options, 0)
         {
             ThreadsPerInputPrimitive = 1;
             OutputTopology           = outputTopology;
             MaxOutputVertices        = maxOutputVertices;
-            TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
         }
 
-        public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(header.Stage, gpuAccessor, options)
+        public ShaderConfig(
+            ShaderHeader header,
+            IGpuAccessor gpuAccessor,
+            TranslationOptions options) : this(header.Stage, gpuAccessor, options, GetLocalMemorySize(header))
         {
             GpPassthrough            = header.Stage == ShaderStage.Geometry && header.GpPassthrough;
             ThreadsPerInputPrimitive = header.ThreadsPerInputPrimitive;
             OutputTopology           = header.OutputTopology;
             MaxOutputVertices        = header.MaxOutputVertexCount;
-            LocalMemorySize          = header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize + (header.ShaderLocalMemoryCrsSize / ThreadsPerWarp);
             ImapTypes                = header.ImapTypes;
             OmapTargets              = header.OmapTargets;
             OmapSampleMask           = header.OmapSampleMask;
             OmapDepth                = header.OmapDepth;
-            TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
             LastInVertexPipeline     = header.Stage < ShaderStage.Fragment;
         }
 
+        private static int GetLocalMemorySize(ShaderHeader header)
+        {
+            return header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize + (header.ShaderLocalMemoryCrsSize / ThreadsPerWarp);
+        }
+
         private void EnsureTransformFeedbackInitialized()
         {
             if (HasTransformFeedbackOutputs() && _transformFeedbackOutputs == null)

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

@@ -107,7 +107,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
             if (options.Flags.HasFlag(TranslationFlags.Compute))
             {
-                config = new ShaderConfig(ShaderStage.Compute, gpuAccessor, options);
+                config = new ShaderConfig(ShaderStage.Compute, gpuAccessor, options, gpuAccessor.QueryComputeLocalMemorySize());
 
                 program = Decoder.Decode(config, address);
             }

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

@@ -149,6 +149,17 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public ShaderProgram Translate(TranslatorContext other = null)
         {
+            bool usesLocalMemory = _config.UsedFeatures.HasFlag(FeatureFlags.LocalMemory);
+
+            _config.ResourceManager.SetCurrentLocalMemory(_config.LocalMemorySize, usesLocalMemory);
+
+            if (_config.Stage == ShaderStage.Compute)
+            {
+                bool usesSharedMemory = _config.UsedFeatures.HasFlag(FeatureFlags.SharedMemory);
+
+                _config.ResourceManager.SetCurrentSharedMemory(GpuAccessor.QueryComputeSharedMemorySize(), usesSharedMemory);
+            }
+
             FunctionCode[] code = EmitShader(_program, _config, initializeOutputs: other == null, out _);
 
             if (other != null)
@@ -157,6 +168,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 // We need to share the resource manager since both shaders accesses the same constant buffers.
                 other._config.ResourceManager = _config.ResourceManager;
+                other._config.ResourceManager.SetCurrentLocalMemory(other._config.LocalMemorySize, other._config.UsedFeatures.HasFlag(FeatureFlags.LocalMemory));
 
                 FunctionCode[] otherCode = EmitShader(other._program, other._config, initializeOutputs: true, out int aStart);
 

src/Ryujinx.Graphics.Vulkan/DescriptorSetCollection.cs

@@ -16,9 +16,9 @@ namespace Ryujinx.Graphics.Vulkan
             _descriptorSets = descriptorSets;
         }
 
-        public void InitializeBuffers(int setIndex, int baseBinding, int countPerUnit, DescriptorType type, VkBuffer dummyBuffer)
+        public void InitializeBuffers(int setIndex, int baseBinding, int count, DescriptorType type, VkBuffer dummyBuffer)
         {
-            Span<DescriptorBufferInfo> infos = stackalloc DescriptorBufferInfo[countPerUnit];
+            Span<DescriptorBufferInfo> infos = stackalloc DescriptorBufferInfo[count];
 
             infos.Fill(new DescriptorBufferInfo()
             {