Implement support for masked stencil clears on Vulkan (#5589)

* Implement support for masked stencil clears on Vulkan

* PR feedback

.github/reviewers.yml

@@ -29,4 +29,4 @@ infra:
   - TSRBerry
 
 default:
-  - @developers
+  - '@developers'

.github/update_reviewers.py

@@ -62,9 +62,9 @@ if __name__ == "__main__":
         sys.stderr.write("usage: <app_id> <private_key_env_name> <installation_id> <repo_path> <pr_id> <config_path>\n")
         sys.exit(1)
 
-    app_id = sys.argv[1]
+    app_id = int(sys.argv[1])
     private_key = os.environ[sys.argv[2]]
-    installation_id = sys.argv[3]
+    installation_id = int(sys.argv[3])
     repo_path = sys.argv[4]
     pr_id = int(sys.argv[5])
     config_path = Path(sys.argv[6])

src/Ryujinx.Graphics.GAL/Capabilities.cs

@@ -52,6 +52,7 @@ namespace Ryujinx.Graphics.GAL
 
         public readonly int MaximumComputeSharedMemorySize;
         public readonly float MaximumSupportedAnisotropy;
+        public readonly int ShaderSubgroupSize;
         public readonly int StorageBufferOffsetAlignment;
 
         public readonly int GatherBiasPrecision;
@@ -101,6 +102,7 @@ namespace Ryujinx.Graphics.GAL
             uint maximumImagesPerStage,
             int maximumComputeSharedMemorySize,
             float maximumSupportedAnisotropy,
+            int shaderSubgroupSize,
             int storageBufferOffsetAlignment,
             int gatherBiasPrecision)
         {
@@ -148,6 +150,7 @@ namespace Ryujinx.Graphics.GAL
             MaximumImagesPerStage = maximumImagesPerStage;
             MaximumComputeSharedMemorySize = maximumComputeSharedMemorySize;
             MaximumSupportedAnisotropy = maximumSupportedAnisotropy;
+            ShaderSubgroupSize = shaderSubgroupSize;
             StorageBufferOffsetAlignment = storageBufferOffsetAlignment;
             GatherBiasPrecision = gatherBiasPrecision;
         }

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 = 5576;
+        private const uint CodeGenVersion = 5540;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

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

@@ -137,6 +137,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         public int QueryHostStorageBufferOffsetAlignment() => _context.Capabilities.StorageBufferOffsetAlignment;
 
+        public int QueryHostSubgroupSize() => _context.Capabilities.ShaderSubgroupSize;
+
         public bool QueryHostSupportsBgraFormat() => _context.Capabilities.SupportsBgraFormat;
 
         public bool QueryHostSupportsFragmentShaderInterlock() => _context.Capabilities.SupportsFragmentShaderInterlock;

src/Ryujinx.Graphics.OpenGL/Constants.cs

@@ -7,5 +7,6 @@
         public const int MaxVertexAttribs = 16;
         public const int MaxVertexBuffers = 16;
         public const int MaxTransformFeedbackBuffers = 4;
+        public const int MaxSubgroupSize = 64;
     }
 }

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -175,6 +175,7 @@ namespace Ryujinx.Graphics.OpenGL
                 maximumImagesPerStage: 8,
                 maximumComputeSharedMemorySize: HwCapabilities.MaximumComputeSharedMemorySize,
                 maximumSupportedAnisotropy: HwCapabilities.MaximumSupportedAnisotropy,
+                shaderSubgroupSize: Constants.MaxSubgroupSize,
                 storageBufferOffsetAlignment: HwCapabilities.StorageBufferOffsetAlignment,
                 gatherBiasPrecision: intelWindows || amdWindows ? 8 : 0); // Precision is 8 for these vendors on Vulkan.
         }

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

@@ -25,6 +25,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             {
                 context.AppendLine("#extension GL_KHR_shader_subgroup_basic : enable");
                 context.AppendLine("#extension GL_KHR_shader_subgroup_ballot : enable");
+                context.AppendLine("#extension GL_KHR_shader_subgroup_shuffle : enable");
             }
 
             context.AppendLine("#extension GL_ARB_shader_group_vote : enable");
@@ -201,26 +202,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighU32.glsl");
             }
 
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.Shuffle) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/Shuffle.glsl");
-            }
-
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.ShuffleDown) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleDown.glsl");
-            }
-
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.ShuffleUp) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleUp.glsl");
-            }
-
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.ShuffleXor) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleXor.glsl");
-            }
-
             if ((info.HelperFunctionsMask & HelperFunctionsMask.SwizzleAdd) != 0)
             {
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/SwizzleAdd.glsl");

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

@@ -5,10 +5,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
         public static string MultiplyHighS32 = "Helper_MultiplyHighS32";
         public static string MultiplyHighU32 = "Helper_MultiplyHighU32";
 
-        public static string Shuffle = "Helper_Shuffle";
-        public static string ShuffleDown = "Helper_ShuffleDown";
-        public static string ShuffleUp = "Helper_ShuffleUp";
-        public static string ShuffleXor = "Helper_ShuffleXor";
         public static string SwizzleAdd = "Helper_SwizzleAdd";
     }
 }

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

@@ -1,11 +0,0 @@
-float Helper_Shuffle(float x, uint index, uint mask, out bool valid)
-{
-    uint clamp = mask & 0x1fu;
-    uint segMask = (mask >> 8) & 0x1fu;
-    uint minThreadId = $SUBGROUP_INVOCATION$ & segMask;
-    uint maxThreadId = minThreadId | (clamp & ~segMask);
-    uint srcThreadId = (index & ~segMask) | minThreadId;
-    valid = srcThreadId <= maxThreadId;
-    float v = $SUBGROUP_BROADCAST$(x, srcThreadId);
-    return valid ? v : x;
-}

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

@@ -1,11 +0,0 @@
-float Helper_ShuffleDown(float x, uint index, uint mask, out bool valid)
-{
-    uint clamp = mask & 0x1fu;
-    uint segMask = (mask >> 8) & 0x1fu;
-    uint minThreadId = $SUBGROUP_INVOCATION$ & segMask;
-    uint maxThreadId = minThreadId | (clamp & ~segMask);
-    uint srcThreadId = $SUBGROUP_INVOCATION$ + index;
-    valid = srcThreadId <= maxThreadId;
-    float v = $SUBGROUP_BROADCAST$(x, srcThreadId);
-    return valid ? v : x;
-}

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

@@ -1,9 +0,0 @@
-float Helper_ShuffleUp(float x, uint index, uint mask, out bool valid)
-{
-    uint segMask = (mask >> 8) & 0x1fu;
-    uint minThreadId = $SUBGROUP_INVOCATION$ & segMask;
-    uint srcThreadId = $SUBGROUP_INVOCATION$ - index;
-    valid = int(srcThreadId) >= int(minThreadId);
-    float v = $SUBGROUP_BROADCAST$(x, srcThreadId);
-    return valid ? v : x;
-}

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

@@ -1,11 +0,0 @@
-float Helper_ShuffleXor(float x, uint index, uint mask, out bool valid)
-{
-    uint clamp = mask & 0x1fu;
-    uint segMask = (mask >> 8) & 0x1fu;
-    uint minThreadId = $SUBGROUP_INVOCATION$ & segMask;
-    uint maxThreadId = minThreadId | (clamp & ~segMask);
-    uint srcThreadId = $SUBGROUP_INVOCATION$ ^ index;
-    valid = srcThreadId <= maxThreadId;
-    float v = $SUBGROUP_BROADCAST$(x, srcThreadId);
-    return valid ? v : x;
-}

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

@@ -9,6 +9,7 @@ using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenFSI;
 using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
 using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenMemory;
 using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenPacking;
+using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenShuffle;
 using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenVector;
 using static Ryujinx.Graphics.Shader.StructuredIr.InstructionInfo;
 
@@ -174,6 +175,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     case Instruction.PackHalf2x16:
                         return PackHalf2x16(context, operation);
 
+                    case Instruction.Shuffle:
+                        return Shuffle(context, operation);
+
                     case Instruction.Store:
                         return Store(context, operation);
 

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

@@ -13,14 +13,15 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             AggregateType dstType = GetSrcVarType(operation.Inst, 0);
 
             string arg = GetSoureExpr(context, operation.GetSource(0), dstType);
+            char component = "xyzw"[operation.Index];
 
             if (context.HostCapabilities.SupportsShaderBallot)
             {
-                return $"unpackUint2x32(ballotARB({arg})).x";
+                return $"unpackUint2x32(ballotARB({arg})).{component}";
             }
             else
             {
-                return $"subgroupBallot({arg}).x";
+                return $"subgroupBallot({arg}).{component}";
             }
         }
     }

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

@@ -108,10 +108,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.ShiftLeft,                InstType.OpBinary,       "<<",              3);
             Add(Instruction.ShiftRightS32,            InstType.OpBinary,       ">>",              3);
             Add(Instruction.ShiftRightU32,            InstType.OpBinary,       ">>",              3);
-            Add(Instruction.Shuffle,                  InstType.CallQuaternary, HelperFunctionNames.Shuffle);
-            Add(Instruction.ShuffleDown,              InstType.CallQuaternary, HelperFunctionNames.ShuffleDown);
-            Add(Instruction.ShuffleUp,                InstType.CallQuaternary, HelperFunctionNames.ShuffleUp);
-            Add(Instruction.ShuffleXor,               InstType.CallQuaternary, HelperFunctionNames.ShuffleXor);
+            Add(Instruction.Shuffle,                  InstType.Special);
+            Add(Instruction.ShuffleDown,              InstType.CallBinary,     "subgroupShuffleDown");
+            Add(Instruction.ShuffleUp,                InstType.CallBinary,     "subgroupShuffleUp");
+            Add(Instruction.ShuffleXor,               InstType.CallBinary,     "subgroupShuffleXor");
             Add(Instruction.Sine,                     InstType.CallUnary,      "sin");
             Add(Instruction.SquareRoot,               InstType.CallUnary,      "sqrt");
             Add(Instruction.Store,                    InstType.Special);

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

@@ -0,0 +1,25 @@
+using Ryujinx.Graphics.Shader.StructuredIr;
+using Ryujinx.Graphics.Shader.Translation;
+
+using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
+
+namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
+{
+    static class InstGenShuffle
+    {
+        public static string Shuffle(CodeGenContext context, AstOperation operation)
+        {
+            string value = GetSoureExpr(context, operation.GetSource(0), AggregateType.FP32);
+            string index = GetSoureExpr(context, operation.GetSource(1), AggregateType.U32);
+
+            if (context.HostCapabilities.SupportsShaderBallot)
+            {
+                return $"readInvocationARB({value}, {index})";
+            }
+            else
+            {
+                return $"subgroupShuffle({value}, {index})";
+            }
+        }
+    }
+}

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

@@ -231,7 +231,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             var execution = context.Constant(context.TypeU32(), Scope.Subgroup);
 
             var maskVector = context.GroupNonUniformBallot(uvec4Type, execution, context.Get(AggregateType.Bool, source));
-            var mask = context.CompositeExtract(context.TypeU32(), maskVector, (SpvLiteralInteger)0);
+            var mask = context.CompositeExtract(context.TypeU32(), maskVector, (SpvLiteralInteger)operation.Index);
 
             return new OperationResult(AggregateType.U32, mask);
         }
@@ -1100,117 +1100,40 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         private static OperationResult GenerateShuffle(CodeGenContext context, AstOperation operation)
         {
-            var x = context.GetFP32(operation.GetSource(0));
+            var value = context.GetFP32(operation.GetSource(0));
             var index = context.GetU32(operation.GetSource(1));
-            var mask = context.GetU32(operation.GetSource(2));
 
-            var const31 = context.Constant(context.TypeU32(), 31);
-            var const8 = context.Constant(context.TypeU32(), 8);
-
-            var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
-            var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
-            var notSegMask = context.Not(context.TypeU32(), segMask);
-            var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
-            var indexNotSegMask = context.BitwiseAnd(context.TypeU32(), index, notSegMask);
-
-            var threadId = GetScalarInput(context, IoVariable.SubgroupLaneId);
-
-            var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
-            var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
-            var srcThreadId = context.BitwiseOr(context.TypeU32(), indexNotSegMask, minThreadId);
-            var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
-            var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
-            var result = context.Select(context.TypeFP32(), valid, value, x);
-
-            var validLocal = (AstOperand)operation.GetSource(3);
-
-            context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
+            var result = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), value, index);
 
             return new OperationResult(AggregateType.FP32, result);
         }
 
         private static OperationResult GenerateShuffleDown(CodeGenContext context, AstOperation operation)
         {
-            var x = context.GetFP32(operation.GetSource(0));
+            var value = context.GetFP32(operation.GetSource(0));
             var index = context.GetU32(operation.GetSource(1));
-            var mask = context.GetU32(operation.GetSource(2));
 
-            var const31 = context.Constant(context.TypeU32(), 31);
-            var const8 = context.Constant(context.TypeU32(), 8);
-
-            var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
-            var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
-            var notSegMask = context.Not(context.TypeU32(), segMask);
-            var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
-
-            var threadId = GetScalarInput(context, IoVariable.SubgroupLaneId);
-
-            var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
-            var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
-            var srcThreadId = context.IAdd(context.TypeU32(), threadId, index);
-            var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
-            var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
-            var result = context.Select(context.TypeFP32(), valid, value, x);
-
-            var validLocal = (AstOperand)operation.GetSource(3);
-
-            context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
+            var result = context.GroupNonUniformShuffleDown(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), value, index);
 
             return new OperationResult(AggregateType.FP32, result);
         }
 
         private static OperationResult GenerateShuffleUp(CodeGenContext context, AstOperation operation)
         {
-            var x = context.GetFP32(operation.GetSource(0));
+            var value = context.GetFP32(operation.GetSource(0));
             var index = context.GetU32(operation.GetSource(1));
-            var mask = context.GetU32(operation.GetSource(2));
 
-            var const31 = context.Constant(context.TypeU32(), 31);
-            var const8 = context.Constant(context.TypeU32(), 8);
-
-            var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
-
-            var threadId = GetScalarInput(context, IoVariable.SubgroupLaneId);
-
-            var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
-            var srcThreadId = context.ISub(context.TypeU32(), threadId, index);
-            var valid = context.SGreaterThanEqual(context.TypeBool(), srcThreadId, minThreadId);
-            var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
-            var result = context.Select(context.TypeFP32(), valid, value, x);
-
-            var validLocal = (AstOperand)operation.GetSource(3);
-
-            context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
+            var result = context.GroupNonUniformShuffleUp(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), value, index);
 
             return new OperationResult(AggregateType.FP32, result);
         }
 
         private static OperationResult GenerateShuffleXor(CodeGenContext context, AstOperation operation)
         {
-            var x = context.GetFP32(operation.GetSource(0));
+            var value = context.GetFP32(operation.GetSource(0));
             var index = context.GetU32(operation.GetSource(1));
-            var mask = context.GetU32(operation.GetSource(2));
 
-            var const31 = context.Constant(context.TypeU32(), 31);
-            var const8 = context.Constant(context.TypeU32(), 8);
-
-            var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
-            var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
-            var notSegMask = context.Not(context.TypeU32(), segMask);
-            var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
-
-            var threadId = GetScalarInput(context, IoVariable.SubgroupLaneId);
-
-            var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
-            var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
-            var srcThreadId = context.BitwiseXor(context.TypeU32(), threadId, index);
-            var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
-            var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
-            var result = context.Select(context.TypeFP32(), valid, value, x);
-
-            var validLocal = (AstOperand)operation.GetSource(3);
-
-            context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
+            var result = context.GroupNonUniformShuffleXor(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), value, index);
 
             return new OperationResult(AggregateType.FP32, result);
         }

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

@@ -28,12 +28,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             _poolLock = new object();
         }
 
-        private const HelperFunctionsMask NeedsInvocationIdMask =
-            HelperFunctionsMask.Shuffle |
-            HelperFunctionsMask.ShuffleDown |
-            HelperFunctionsMask.ShuffleUp |
-            HelperFunctionsMask.ShuffleXor |
-            HelperFunctionsMask.SwizzleAdd;
+        private const HelperFunctionsMask NeedsInvocationIdMask = HelperFunctionsMask.SwizzleAdd;
 
         public static byte[] Generate(StructuredProgramInfo info, CodeGenParameters parameters)
         {

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

@@ -307,6 +307,9 @@ namespace Ryujinx.Graphics.Shader.Decoders
                     case InstName.Sts:
                         context.SetUsedFeature(FeatureFlags.SharedMemory);
                         break;
+                    case InstName.Shfl:
+                        context.SetUsedFeature(FeatureFlags.Shuffle);
+                        break;
                 }
 
                 block.OpCodes.Add(op);

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -194,6 +194,15 @@ namespace Ryujinx.Graphics.Shader
             return 16;
         }
 
+        /// <summary>
+        /// Queries host shader subgroup size.
+        /// </summary>
+        /// <returns>Host shader subgroup size in invocations</returns>
+        int QueryHostSubgroupSize()
+        {
+            return 32;
+        }
+
         /// <summary>
         /// Queries host support for texture formats with BGRA component order (such as BGRA8).
         /// </summary>

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

@@ -76,7 +76,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
             switch (op.SReg)
             {
                 case SReg.LaneId:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+                    src = EmitLoadSubgroupLaneId(context);
                     break;
 
                 case SReg.InvocationId:
@@ -146,19 +146,19 @@ namespace Ryujinx.Graphics.Shader.Instructions
                     break;
 
                 case SReg.EqMask:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupEqMask, null, Const(0));
+                    src = EmitLoadSubgroupMask(context, IoVariable.SubgroupEqMask);
                     break;
                 case SReg.LtMask:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupLtMask, null, Const(0));
+                    src = EmitLoadSubgroupMask(context, IoVariable.SubgroupLtMask);
                     break;
                 case SReg.LeMask:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupLeMask, null, Const(0));
+                    src = EmitLoadSubgroupMask(context, IoVariable.SubgroupLeMask);
                     break;
                 case SReg.GtMask:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupGtMask, null, Const(0));
+                    src = EmitLoadSubgroupMask(context, IoVariable.SubgroupGtMask);
                     break;
                 case SReg.GeMask:
-                    src = context.Load(StorageKind.Input, IoVariable.SubgroupGeMask, null, Const(0));
+                    src = EmitLoadSubgroupMask(context, IoVariable.SubgroupGeMask);
                     break;
 
                 default:
@@ -169,6 +169,52 @@ namespace Ryujinx.Graphics.Shader.Instructions
             context.Copy(GetDest(op.Dest), src);
         }
 
+        private static Operand EmitLoadSubgroupLaneId(EmitterContext context)
+        {
+            if (context.TranslatorContext.GpuAccessor.QueryHostSubgroupSize() <= 32)
+            {
+                return context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+            }
+
+            return context.BitwiseAnd(context.Load(StorageKind.Input, IoVariable.SubgroupLaneId), Const(0x1f));
+        }
+
+        private static Operand EmitLoadSubgroupMask(EmitterContext context, IoVariable ioVariable)
+        {
+            int subgroupSize = context.TranslatorContext.GpuAccessor.QueryHostSubgroupSize();
+
+            if (subgroupSize <= 32)
+            {
+                return context.Load(StorageKind.Input, ioVariable, null, Const(0));
+            }
+            else if (subgroupSize == 64)
+            {
+                Operand laneId = context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+                Operand low = context.Load(StorageKind.Input, ioVariable, null, Const(0));
+                Operand high = context.Load(StorageKind.Input, ioVariable, null, Const(1));
+
+                return context.ConditionalSelect(context.BitwiseAnd(laneId, Const(32)), high, low);
+            }
+            else
+            {
+                Operand laneId = context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+                Operand element = context.ShiftRightU32(laneId, Const(5));
+
+                Operand res = context.Load(StorageKind.Input, ioVariable, null, Const(0));
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(1)),
+                    context.Load(StorageKind.Input, ioVariable, null, Const(1)), res);
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(2)),
+                    context.Load(StorageKind.Input, ioVariable, null, Const(2)), res);
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(3)),
+                    context.Load(StorageKind.Input, ioVariable, null, Const(3)), res);
+
+                return res;
+            }
+        }
+
         public static void SelR(EmitterContext context)
         {
             InstSelR op = context.GetOp<InstSelR>();

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

@@ -10,10 +10,10 @@ namespace Ryujinx.Graphics.Shader.Instructions
 {
     static partial class InstEmit
     {
-        private static readonly int[,] _maskLut = new int[,]
+        private static readonly int[][] _maskLut = new int[][]
         {
-            { 0b0001, 0b0010, 0b0100, 0b1000, 0b0011, 0b1001, 0b1010, 0b1100 },
-            { 0b0111, 0b1011, 0b1101, 0b1110, 0b1111, 0b0000, 0b0000, 0b0000 },
+            new int[] { 0b0001, 0b0010, 0b0100, 0b1000, 0b0011, 0b1001, 0b1010, 0b1100 },
+            new int[] { 0b0111, 0b1011, 0b1101, 0b1110, 0b1111, 0b0000, 0b0000, 0b0000 },
         };
 
         public const bool Sample1DAs2D = true;
@@ -605,7 +605,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
             Operand[] rd1 = new Operand[2] { ConstF(0), ConstF(0) };
 
             int handle = imm;
-            int componentMask = _maskLut[dest2 == RegisterConsts.RegisterZeroIndex ? 0 : 1, writeMask];
+            int componentMask = _maskLut[dest2 == RegisterConsts.RegisterZeroIndex ? 0 : 1][writeMask];
 
             int componentsCount = BitOperations.PopCount((uint)componentMask);
 

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

@@ -50,20 +50,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
             InstVote op = context.GetOp<InstVote>();
 
             Operand pred = GetPredicate(context, op.SrcPred, op.SrcPredInv);
-            Operand res = null;
-
-            switch (op.VoteMode)
-            {
-                case VoteMode.All:
-                    res = context.VoteAll(pred);
-                    break;
-                case VoteMode.Any:
-                    res = context.VoteAny(pred);
-                    break;
-                case VoteMode.Eq:
-                    res = context.VoteAllEqual(pred);
-                    break;
-            }
+            Operand res = EmitVote(context, op.VoteMode, pred);
 
             if (res != null)
             {
@@ -76,7 +63,81 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             if (op.Dest != RegisterConsts.RegisterZeroIndex)
             {
-                context.Copy(GetDest(op.Dest), context.Ballot(pred));
+                context.Copy(GetDest(op.Dest), EmitBallot(context, pred));
+            }
+        }
+
+        private static Operand EmitVote(EmitterContext context, VoteMode voteMode, Operand pred)
+        {
+            int subgroupSize = context.TranslatorContext.GpuAccessor.QueryHostSubgroupSize();
+
+            if (subgroupSize <= 32)
+            {
+                return voteMode switch
+                {
+                    VoteMode.All => context.VoteAll(pred),
+                    VoteMode.Any => context.VoteAny(pred),
+                    VoteMode.Eq => context.VoteAllEqual(pred),
+                    _ => null,
+                };
+            }
+
+            // Emulate vote with ballot masks.
+            // We do that when the GPU thread count is not 32,
+            // since the shader code assumes it is 32.
+            // allInvocations => ballot(pred) == ballot(true),
+            // anyInvocation => ballot(pred) != 0,
+            // allInvocationsEqual => ballot(pred) == balot(true) || ballot(pred) == 0
+            Operand ballotMask = EmitBallot(context, pred);
+
+            Operand AllTrue() => context.ICompareEqual(ballotMask, EmitBallot(context, Const(IrConsts.True)));
+
+            return voteMode switch
+            {
+                VoteMode.All => AllTrue(),
+                VoteMode.Any => context.ICompareNotEqual(ballotMask, Const(0)),
+                VoteMode.Eq => context.BitwiseOr(AllTrue(), context.ICompareEqual(ballotMask, Const(0))),
+                _ => null,
+            };
+        }
+
+        private static Operand EmitBallot(EmitterContext context, Operand pred)
+        {
+            int subgroupSize = context.TranslatorContext.GpuAccessor.QueryHostSubgroupSize();
+
+            if (subgroupSize <= 32)
+            {
+                return context.Ballot(pred, 0);
+            }
+            else if (subgroupSize == 64)
+            {
+                // TODO: Add support for vector destination and do that with a single operation.
+
+                Operand laneId = context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+                Operand low = context.Ballot(pred, 0);
+                Operand high = context.Ballot(pred, 1);
+
+                return context.ConditionalSelect(context.BitwiseAnd(laneId, Const(32)), high, low);
+            }
+            else
+            {
+                // TODO: Add support for vector destination and do that with a single operation.
+
+                Operand laneId = context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+                Operand element = context.ShiftRightU32(laneId, Const(5));
+
+                Operand res = context.Ballot(pred, 0);
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(1)),
+                    context.Ballot(pred, 1), res);
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(2)),
+                    context.Ballot(pred, 2), res);
+                res = context.ConditionalSelect(
+                    context.ICompareEqual(element, Const(3)),
+                    context.Ballot(pred, 3), res);
+
+                return res;
             }
         }
     }

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

@@ -12,10 +12,6 @@
   <ItemGroup>
     <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\SwizzleAdd.glsl" />
   </ItemGroup>
 

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

@@ -7,10 +7,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
     {
         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

@@ -109,14 +109,15 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             Add(Instruction.PackDouble2x32,           AggregateType.FP64,   AggregateType.U32,     AggregateType.U32);
             Add(Instruction.PackHalf2x16,             AggregateType.U32,    AggregateType.FP32,    AggregateType.FP32);
             Add(Instruction.ReciprocalSquareRoot,     AggregateType.Scalar, AggregateType.Scalar);
+            Add(Instruction.Return,                   AggregateType.Void,   AggregateType.U32);
             Add(Instruction.Round,                    AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.ShiftLeft,                AggregateType.S32,    AggregateType.S32,     AggregateType.S32);
             Add(Instruction.ShiftRightS32,            AggregateType.S32,    AggregateType.S32,     AggregateType.S32);
             Add(Instruction.ShiftRightU32,            AggregateType.U32,    AggregateType.U32,     AggregateType.S32);
-            Add(Instruction.Shuffle,                  AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32,     AggregateType.U32,     AggregateType.Bool);
-            Add(Instruction.ShuffleDown,              AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32,     AggregateType.U32,     AggregateType.Bool);
-            Add(Instruction.ShuffleUp,                AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32,     AggregateType.U32,     AggregateType.Bool);
-            Add(Instruction.ShuffleXor,               AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32,     AggregateType.U32,     AggregateType.Bool);
+            Add(Instruction.Shuffle,                  AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32);
+            Add(Instruction.ShuffleDown,              AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32);
+            Add(Instruction.ShuffleUp,                AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32);
+            Add(Instruction.ShuffleXor,               AggregateType.FP32,   AggregateType.FP32,    AggregateType.U32);
             Add(Instruction.Sine,                     AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.SquareRoot,               AggregateType.Scalar, AggregateType.Scalar);
             Add(Instruction.Store,                    AggregateType.Void);
@@ -131,7 +132,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             Add(Instruction.VoteAll,                  AggregateType.Bool,   AggregateType.Bool);
             Add(Instruction.VoteAllEqual,             AggregateType.Bool,   AggregateType.Bool);
             Add(Instruction.VoteAny,                  AggregateType.Bool,   AggregateType.Bool);
-#pragma warning restore IDE0055v
+#pragma warning restore IDE0055
         }
 
         private static void Add(Instruction inst, AggregateType destType, params AggregateType[] srcTypes)

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

@@ -282,18 +282,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
                 case Instruction.MultiplyHighU32:
                     context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighU32;
                     break;
-                case Instruction.Shuffle:
-                    context.Info.HelperFunctionsMask |= HelperFunctionsMask.Shuffle;
-                    break;
-                case Instruction.ShuffleDown:
-                    context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleDown;
-                    break;
-                case Instruction.ShuffleUp:
-                    context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleUp;
-                    break;
-                case Instruction.ShuffleXor:
-                    context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleXor;
-                    break;
                 case Instruction.SwizzleAdd:
                     context.Info.HelperFunctionsMask |= HelperFunctionsMask.SwizzleAdd;
                     break;

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

@@ -112,9 +112,13 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.AtomicXor, storageKind, Local(), Const(binding), e0, e1, value);
         }
 
-        public static Operand Ballot(this EmitterContext context, Operand a)
+        public static Operand Ballot(this EmitterContext context, Operand a, int index)
         {
-            return context.Add(Instruction.Ballot, Local(), a);
+            Operand dest = Local();
+
+            context.Add(new Operation(Instruction.Ballot, index, dest, a));
+
+            return dest;
         }
 
         public static Operand Barrier(this EmitterContext context)
@@ -782,21 +786,41 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.ShiftRightU32, Local(), a, b);
         }
 
+        public static Operand Shuffle(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.Shuffle, Local(), a, b);
+        }
+
         public static (Operand, Operand) Shuffle(this EmitterContext context, Operand a, Operand b, Operand c)
         {
             return context.Add(Instruction.Shuffle, (Local(), Local()), a, b, c);
         }
 
+        public static Operand ShuffleDown(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.ShuffleDown, Local(), a, b);
+        }
+
         public static (Operand, Operand) ShuffleDown(this EmitterContext context, Operand a, Operand b, Operand c)
         {
             return context.Add(Instruction.ShuffleDown, (Local(), Local()), a, b, c);
         }
 
+        public static Operand ShuffleUp(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.ShuffleUp, Local(), a, b);
+        }
+
         public static (Operand, Operand) ShuffleUp(this EmitterContext context, Operand a, Operand b, Operand c)
         {
             return context.Add(Instruction.ShuffleUp, (Local(), Local()), a, b, c);
         }
 
+        public static Operand ShuffleXor(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.ShuffleXor, Local(), a, b);
+        }
+
         public static (Operand, Operand) ShuffleXor(this EmitterContext context, Operand a, Operand b, Operand c)
         {
             return context.Add(Instruction.ShuffleXor, (Local(), Local()), a, b, c);

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

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         InstanceId = 1 << 3,
         DrawParameters = 1 << 4,
         RtLayer = 1 << 5,
+        Shuffle = 1 << 6,
         FixedFuncAttr = 1 << 9,
         LocalMemory = 1 << 10,
         SharedMemory = 1 << 11,

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

@@ -56,6 +56,20 @@ namespace Ryujinx.Graphics.Shader.Translation
             return functionId;
         }
 
+        public int GetOrCreateShuffleFunctionId(HelperFunctionName functionName, int subgroupSize)
+        {
+            if (_functionIds.TryGetValue((int)functionName, out int functionId))
+            {
+                return functionId;
+            }
+
+            Function function = GenerateShuffleFunction(functionName, subgroupSize);
+            functionId = AddFunction(function);
+            _functionIds.Add((int)functionName, functionId);
+
+            return functionId;
+        }
+
         private Function GenerateFunction(HelperFunctionName functionName)
         {
             return functionName switch
@@ -216,6 +230,137 @@ namespace Ryujinx.Graphics.Shader.Translation
             return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, $"SharedStore{bitSize}_{id}", false, 2, 0);
         }
 
+        private static Function GenerateShuffleFunction(HelperFunctionName functionName, int subgroupSize)
+        {
+            return functionName switch
+            {
+                HelperFunctionName.Shuffle => GenerateShuffle(subgroupSize),
+                HelperFunctionName.ShuffleDown => GenerateShuffleDown(subgroupSize),
+                HelperFunctionName.ShuffleUp => GenerateShuffleUp(subgroupSize),
+                HelperFunctionName.ShuffleXor => GenerateShuffleXor(subgroupSize),
+                _ => throw new ArgumentException($"Invalid function name {functionName}"),
+            };
+        }
+
+        private static Function GenerateShuffle(int subgroupSize)
+        {
+            EmitterContext context = new();
+
+            Operand value = Argument(0);
+            Operand index = Argument(1);
+            Operand mask = Argument(2);
+
+            Operand clamp = context.BitwiseAnd(mask, Const(0x1f));
+            Operand segMask = context.BitwiseAnd(context.ShiftRightU32(mask, Const(8)), Const(0x1f));
+            Operand minThreadId = context.BitwiseAnd(GenerateLoadSubgroupLaneId(context, subgroupSize), segMask);
+            Operand maxThreadId = context.BitwiseOr(context.BitwiseAnd(clamp, context.BitwiseNot(segMask)), minThreadId);
+            Operand srcThreadId = context.BitwiseOr(context.BitwiseAnd(index, context.BitwiseNot(segMask)), minThreadId);
+            Operand valid = context.ICompareLessOrEqualUnsigned(srcThreadId, maxThreadId);
+
+            context.Copy(Argument(3), valid);
+
+            Operand result = context.Shuffle(value, GenerateSubgroupShuffleIndex(context, srcThreadId, subgroupSize));
+
+            context.Return(context.ConditionalSelect(valid, result, value));
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "Shuffle", true, 3, 1);
+        }
+
+        private static Function GenerateShuffleDown(int subgroupSize)
+        {
+            EmitterContext context = new();
+
+            Operand value = Argument(0);
+            Operand index = Argument(1);
+            Operand mask = Argument(2);
+
+            Operand clamp = context.BitwiseAnd(mask, Const(0x1f));
+            Operand segMask = context.BitwiseAnd(context.ShiftRightU32(mask, Const(8)), Const(0x1f));
+            Operand laneId = GenerateLoadSubgroupLaneId(context, subgroupSize);
+            Operand minThreadId = context.BitwiseAnd(laneId, segMask);
+            Operand maxThreadId = context.BitwiseOr(context.BitwiseAnd(clamp, context.BitwiseNot(segMask)), minThreadId);
+            Operand srcThreadId = context.IAdd(laneId, index);
+            Operand valid = context.ICompareLessOrEqualUnsigned(srcThreadId, maxThreadId);
+
+            context.Copy(Argument(3), valid);
+
+            Operand result = context.Shuffle(value, GenerateSubgroupShuffleIndex(context, srcThreadId, subgroupSize));
+
+            context.Return(context.ConditionalSelect(valid, result, value));
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "ShuffleDown", true, 3, 1);
+        }
+
+        private static Function GenerateShuffleUp(int subgroupSize)
+        {
+            EmitterContext context = new();
+
+            Operand value = Argument(0);
+            Operand index = Argument(1);
+            Operand mask = Argument(2);
+
+            Operand segMask = context.BitwiseAnd(context.ShiftRightU32(mask, Const(8)), Const(0x1f));
+            Operand laneId = GenerateLoadSubgroupLaneId(context, subgroupSize);
+            Operand minThreadId = context.BitwiseAnd(laneId, segMask);
+            Operand srcThreadId = context.ISubtract(laneId, index);
+            Operand valid = context.ICompareGreaterOrEqual(srcThreadId, minThreadId);
+
+            context.Copy(Argument(3), valid);
+
+            Operand result = context.Shuffle(value, GenerateSubgroupShuffleIndex(context, srcThreadId, subgroupSize));
+
+            context.Return(context.ConditionalSelect(valid, result, value));
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "ShuffleUp", true, 3, 1);
+        }
+
+        private static Function GenerateShuffleXor(int subgroupSize)
+        {
+            EmitterContext context = new();
+
+            Operand value = Argument(0);
+            Operand index = Argument(1);
+            Operand mask = Argument(2);
+
+            Operand clamp = context.BitwiseAnd(mask, Const(0x1f));
+            Operand segMask = context.BitwiseAnd(context.ShiftRightU32(mask, Const(8)), Const(0x1f));
+            Operand laneId = GenerateLoadSubgroupLaneId(context, subgroupSize);
+            Operand minThreadId = context.BitwiseAnd(laneId, segMask);
+            Operand maxThreadId = context.BitwiseOr(context.BitwiseAnd(clamp, context.BitwiseNot(segMask)), minThreadId);
+            Operand srcThreadId = context.BitwiseExclusiveOr(laneId, index);
+            Operand valid = context.ICompareLessOrEqualUnsigned(srcThreadId, maxThreadId);
+
+            context.Copy(Argument(3), valid);
+
+            Operand result = context.Shuffle(value, GenerateSubgroupShuffleIndex(context, srcThreadId, subgroupSize));
+
+            context.Return(context.ConditionalSelect(valid, result, value));
+
+            return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "ShuffleXor", true, 3, 1);
+        }
+
+        private static Operand GenerateLoadSubgroupLaneId(EmitterContext context, int subgroupSize)
+        {
+            if (subgroupSize <= 32)
+            {
+                return context.Load(StorageKind.Input, IoVariable.SubgroupLaneId);
+            }
+
+            return context.BitwiseAnd(context.Load(StorageKind.Input, IoVariable.SubgroupLaneId), Const(0x1f));
+        }
+
+        private static Operand GenerateSubgroupShuffleIndex(EmitterContext context, Operand srcThreadId, int subgroupSize)
+        {
+            if (subgroupSize <= 32)
+            {
+                return srcThreadId;
+            }
+
+            return context.BitwiseOr(
+                context.BitwiseAnd(context.Load(StorageKind.Input, IoVariable.SubgroupLaneId), Const(0x60)),
+                srcThreadId);
+        }
+
         private Function GenerateTexelFetchScaleFunction()
         {
             EmitterContext context = new();

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

@@ -2,12 +2,18 @@ namespace Ryujinx.Graphics.Shader.Translation
 {
     enum HelperFunctionName
     {
+        Invalid,
+
         ConvertDoubleToFloat,
         ConvertFloatToDouble,
         SharedAtomicMaxS32,
         SharedAtomicMinS32,
         SharedStore8,
         SharedStore16,
+        Shuffle,
+        ShuffleDown,
+        ShuffleUp,
+        ShuffleXor,
         TexelFetchScale,
         TextureSizeUnscale,
     }

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

@@ -0,0 +1,52 @@
+using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+using Ryujinx.Graphics.Shader.Translation.Optimizations;
+using System.Collections.Generic;
+using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
+
+namespace Ryujinx.Graphics.Shader.Translation.Transforms
+{
+    class ShufflePass : ITransformPass
+    {
+        public static bool IsEnabled(IGpuAccessor gpuAccessor, ShaderStage stage, TargetLanguage targetLanguage, FeatureFlags usedFeatures)
+        {
+            return usedFeatures.HasFlag(FeatureFlags.Shuffle);
+        }
+
+        public static LinkedListNode<INode> RunPass(TransformContext context, LinkedListNode<INode> node)
+        {
+            Operation operation = (Operation)node.Value;
+
+            HelperFunctionName functionName = operation.Inst switch
+            {
+                Instruction.Shuffle => HelperFunctionName.Shuffle,
+                Instruction.ShuffleDown => HelperFunctionName.ShuffleDown,
+                Instruction.ShuffleUp => HelperFunctionName.ShuffleUp,
+                Instruction.ShuffleXor => HelperFunctionName.ShuffleXor,
+                _ => HelperFunctionName.Invalid,
+            };
+
+            if (functionName == HelperFunctionName.Invalid || operation.SourcesCount != 3 || operation.DestsCount != 2)
+            {
+                return node;
+            }
+
+            int functionId = context.Hfm.GetOrCreateShuffleFunctionId(functionName, context.GpuAccessor.QueryHostSubgroupSize());
+
+            Operand result = operation.GetDest(0);
+            Operand valid = operation.GetDest(1);
+            Operand value = operation.GetSource(0);
+            Operand index = operation.GetSource(1);
+            Operand mask = operation.GetSource(2);
+
+            operation.Dest = null;
+
+            Operand[] callArgs = new Operand[] { Const(functionId), value, index, mask, valid };
+
+            LinkedListNode<INode> newNode = node.List.AddBefore(node, new Operation(Instruction.Call, 0, result, callArgs));
+
+            Utils.DeleteNode(node, operation);
+
+            return newNode;
+        }
+    }
+}

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

@@ -13,6 +13,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
             RunPass<TexturePass>(context);
             RunPass<SharedStoreSmallIntCas>(context);
             RunPass<SharedAtomicSignedCas>(context);
+            RunPass<ShufflePass>(context);
         }
 
         private static void RunPass<T>(TransformContext context) where T : ITransformPass

src/Ryujinx.Graphics.Vulkan/FramebufferParams.cs

@@ -148,6 +148,16 @@ namespace Ryujinx.Graphics.Vulkan
             return _attachments[index];
         }
 
+        public Auto<DisposableImageView> GetDepthStencilAttachment()
+        {
+            if (!HasDepthStencil)
+            {
+                return null;
+            }
+
+            return _attachments[AttachmentsCount - 1];
+        }
+
         public ComponentType GetAttachmentComponentType(int index)
         {
             if (_colors != null && (uint)index < _colors.Length)

src/Ryujinx.Graphics.Vulkan/HardwareCapabilities.cs

@@ -25,7 +25,6 @@ namespace Ryujinx.Graphics.Vulkan
         public readonly bool SupportsIndirectParameters;
         public readonly bool SupportsFragmentShaderInterlock;
         public readonly bool SupportsGeometryShaderPassthrough;
-        public readonly bool SupportsSubgroupSizeControl;
         public readonly bool SupportsShaderFloat64;
         public readonly bool SupportsShaderInt8;
         public readonly bool SupportsShaderStencilExport;
@@ -45,9 +44,7 @@ namespace Ryujinx.Graphics.Vulkan
         public readonly bool SupportsViewportArray2;
         public readonly bool SupportsHostImportedMemory;
         public readonly bool SupportsDepthClipControl;
-        public readonly uint MinSubgroupSize;
-        public readonly uint MaxSubgroupSize;
-        public readonly ShaderStageFlags RequiredSubgroupSizeStages;
+        public readonly uint SubgroupSize;
         public readonly SampleCountFlags SupportedSampleCounts;
         public readonly PortabilitySubsetFlags PortabilitySubset;
         public readonly uint VertexBufferAlignment;
@@ -64,7 +61,6 @@ namespace Ryujinx.Graphics.Vulkan
             bool supportsIndirectParameters,
             bool supportsFragmentShaderInterlock,
             bool supportsGeometryShaderPassthrough,
-            bool supportsSubgroupSizeControl,
             bool supportsShaderFloat64,
             bool supportsShaderInt8,
             bool supportsShaderStencilExport,
@@ -84,9 +80,7 @@ namespace Ryujinx.Graphics.Vulkan
             bool supportsViewportArray2,
             bool supportsHostImportedMemory,
             bool supportsDepthClipControl,
-            uint minSubgroupSize,
-            uint maxSubgroupSize,
-            ShaderStageFlags requiredSubgroupSizeStages,
+            uint subgroupSize,
             SampleCountFlags supportedSampleCounts,
             PortabilitySubsetFlags portabilitySubset,
             uint vertexBufferAlignment,
@@ -102,7 +96,6 @@ namespace Ryujinx.Graphics.Vulkan
             SupportsIndirectParameters = supportsIndirectParameters;
             SupportsFragmentShaderInterlock = supportsFragmentShaderInterlock;
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
-            SupportsSubgroupSizeControl = supportsSubgroupSizeControl;
             SupportsShaderFloat64 = supportsShaderFloat64;
             SupportsShaderInt8 = supportsShaderInt8;
             SupportsShaderStencilExport = supportsShaderStencilExport;
@@ -122,9 +115,7 @@ namespace Ryujinx.Graphics.Vulkan
             SupportsViewportArray2 = supportsViewportArray2;
             SupportsHostImportedMemory = supportsHostImportedMemory;
             SupportsDepthClipControl = supportsDepthClipControl;
-            MinSubgroupSize = minSubgroupSize;
-            MaxSubgroupSize = maxSubgroupSize;
-            RequiredSubgroupSizeStages = requiredSubgroupSizeStages;
+            SubgroupSize = subgroupSize;
             SupportedSampleCounts = supportedSampleCounts;
             PortabilitySubset = portabilitySubset;
             VertexBufferAlignment = vertexBufferAlignment;

src/Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -38,6 +38,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly IProgram _programColorClearF;
         private readonly IProgram _programColorClearSI;
         private readonly IProgram _programColorClearUI;
+        private readonly IProgram _programDepthStencilClear;
         private readonly IProgram _programStrideChange;
         private readonly IProgram _programConvertD32S8ToD24S8;
         private readonly IProgram _programConvertIndexBuffer;
@@ -105,6 +106,12 @@ namespace Ryujinx.Graphics.Vulkan
                 new ShaderSource(ReadSpirv("ColorClearUIFragment.spv"), ShaderStage.Fragment, TargetLanguage.Spirv),
             }, colorClearResourceLayout);
 
+            _programDepthStencilClear = gd.CreateProgramWithMinimalLayout(new[]
+            {
+                new ShaderSource(ReadSpirv("ColorClearVertex.spv"), ShaderStage.Vertex, TargetLanguage.Spirv),
+                new ShaderSource(ReadSpirv("DepthStencilClearFragment.spv"), ShaderStage.Fragment, TargetLanguage.Spirv),
+            }, colorClearResourceLayout);
+
             var strideChangeResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
@@ -446,10 +453,6 @@ namespace Ryujinx.Graphics.Vulkan
                 0f,
                 1f);
 
-            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
-            scissors[0] = new Rectangle<int>(0, 0, dstWidth, dstHeight);
-
             if (dstIsDepthOrStencil)
             {
                 _pipeline.SetProgram(src.Info.Target.IsMultisample() ? _programDepthBlitMs : _programDepthBlit);
@@ -470,7 +473,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             _pipeline.SetRenderTarget(dst, (uint)dstWidth, (uint)dstHeight, (uint)dstSamples, dstIsDepthOrStencil, dstFormat);
             _pipeline.SetRenderTargetColorMasks(new uint[] { 0xf });
-            _pipeline.SetScissors(scissors);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { new Rectangle<int>(0, 0, dstWidth, dstHeight) });
 
             if (clearAlpha)
             {
@@ -547,12 +550,8 @@ namespace Ryujinx.Graphics.Vulkan
                 0f,
                 1f);
 
-            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
-            scissors[0] = new Rectangle<int>(0, 0, dstWidth, dstHeight);
-
             _pipeline.SetRenderTarget(dst, (uint)dstWidth, (uint)dstHeight, (uint)dstSamples, true, dstFormat);
-            _pipeline.SetScissors(scissors);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { new Rectangle<int>(0, 0, dstWidth, dstHeight) });
             _pipeline.SetViewports(viewports);
             _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
 
@@ -639,7 +638,11 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        private static StencilTestDescriptor CreateStencilTestDescriptor(bool enabled)
+        private static StencilTestDescriptor CreateStencilTestDescriptor(
+            bool enabled,
+            int refValue = 0,
+            int compareMask = 0xff,
+            int writeMask = 0xff)
         {
             return new StencilTestDescriptor(
                 enabled,
@@ -647,16 +650,16 @@ namespace Ryujinx.Graphics.Vulkan
                 StencilOp.Replace,
                 StencilOp.Replace,
                 StencilOp.Replace,
-                0,
-                0xff,
-                0xff,
+                refValue,
+                compareMask,
+                writeMask,
                 CompareOp.Always,
                 StencilOp.Replace,
                 StencilOp.Replace,
                 StencilOp.Replace,
-                0,
-                0xff,
-                0xff);
+                refValue,
+                compareMask,
+                writeMask);
         }
 
         public void Clear(
@@ -695,10 +698,6 @@ namespace Ryujinx.Graphics.Vulkan
                 0f,
                 1f);
 
-            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
-            scissors[0] = scissor;
-
             IProgram program;
 
             if (type == ComponentType.SignedInteger)
@@ -718,7 +717,7 @@ namespace Ryujinx.Graphics.Vulkan
             _pipeline.SetRenderTarget(dst, (uint)dstWidth, (uint)dstHeight, false, dstFormat);
             _pipeline.SetRenderTargetColorMasks(new[] { componentMask });
             _pipeline.SetViewports(viewports);
-            _pipeline.SetScissors(scissors);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { scissor });
             _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
             _pipeline.Draw(4, 1, 0, 0);
             _pipeline.Finish();
@@ -726,6 +725,56 @@ namespace Ryujinx.Graphics.Vulkan
             gd.BufferManager.Delete(bufferHandle);
         }
 
+        public void Clear(
+            VulkanRenderer gd,
+            Auto<DisposableImageView> dst,
+            float depthValue,
+            bool depthMask,
+            int stencilValue,
+            int stencilMask,
+            int dstWidth,
+            int dstHeight,
+            VkFormat dstFormat,
+            Rectangle<int> scissor)
+        {
+            const int ClearColorBufferSize = 16;
+
+            gd.FlushAllCommands();
+
+            using var cbs = gd.CommandBufferPool.Rent();
+
+            _pipeline.SetCommandBuffer(cbs);
+
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ClearColorBufferSize);
+
+            gd.BufferManager.SetData<float>(bufferHandle, 0, stackalloc float[] { depthValue });
+
+            _pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(1, new BufferRange(bufferHandle, 0, ClearColorBufferSize)) });
+
+            Span<Viewport> viewports = stackalloc Viewport[1];
+
+            viewports[0] = new Viewport(
+                new Rectangle<float>(0, 0, dstWidth, dstHeight),
+                ViewportSwizzle.PositiveX,
+                ViewportSwizzle.PositiveY,
+                ViewportSwizzle.PositiveZ,
+                ViewportSwizzle.PositiveW,
+                0f,
+                1f);
+
+            _pipeline.SetProgram(_programDepthStencilClear);
+            _pipeline.SetRenderTarget(dst, (uint)dstWidth, (uint)dstHeight, true, dstFormat);
+            _pipeline.SetViewports(viewports);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { scissor });
+            _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
+            _pipeline.SetDepthTest(new DepthTestDescriptor(true, depthMask, CompareOp.Always));
+            _pipeline.SetStencilTest(CreateStencilTestDescriptor(stencilMask != 0, stencilValue, 0xff, stencilMask));
+            _pipeline.Draw(4, 1, 0, 0);
+            _pipeline.Finish();
+
+            gd.BufferManager.Delete(bufferHandle);
+        }
+
         public void DrawTexture(
             VulkanRenderer gd,
             PipelineBase pipeline,
@@ -778,8 +827,6 @@ namespace Ryujinx.Graphics.Vulkan
                 0f,
                 1f);
 
-            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
             pipeline.SetProgram(_programColorBlit);
             pipeline.SetViewports(viewports);
             pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
@@ -1119,11 +1166,7 @@ namespace Ryujinx.Graphics.Vulkan
                     0f,
                     1f);
 
-                Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
-                scissors[0] = new Rectangle<int>(0, 0, dst.Width, dst.Height);
-
-                _pipeline.SetScissors(scissors);
+                _pipeline.SetScissors(stackalloc Rectangle<int>[] { new Rectangle<int>(0, 0, dst.Width, dst.Height) });
                 _pipeline.SetViewports(viewports);
                 _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
 
@@ -1251,12 +1294,8 @@ namespace Ryujinx.Graphics.Vulkan
                 0f,
                 1f);
 
-            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
-
-            scissors[0] = new Rectangle<int>(0, 0, dst.Width, dst.Height);
-
             _pipeline.SetRenderTargetColorMasks(new uint[] { 0xf });
-            _pipeline.SetScissors(scissors);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { new Rectangle<int>(0, 0, dst.Width, dst.Height) });
             _pipeline.SetViewports(viewports);
             _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
 
@@ -1731,6 +1770,7 @@ namespace Ryujinx.Graphics.Vulkan
                 _programColorClearF.Dispose();
                 _programColorClearSI.Dispose();
                 _programColorClearUI.Dispose();
+                _programDepthStencilClear.Dispose();
                 _programStrideChange.Dispose();
                 _programConvertIndexBuffer.Dispose();
                 _programConvertIndirectData.Dispose();

src/Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -243,10 +243,8 @@ namespace Ryujinx.Graphics.Vulkan
             Gd.Api.CmdClearAttachments(CommandBuffer, 1, &attachment, 1, &clearRect);
         }
 
-        public unsafe void ClearRenderTargetDepthStencil(int layer, int layerCount, float depthValue, bool depthMask, int stencilValue, int stencilMask)
+        public unsafe void ClearRenderTargetDepthStencil(int layer, int layerCount, float depthValue, bool depthMask, int stencilValue, bool stencilMask)
         {
-            // TODO: Use stencilMask (fully).
-
             if (FramebufferParams == null || !FramebufferParams.HasDepthStencil)
             {
                 return;
@@ -255,7 +253,7 @@ namespace Ryujinx.Graphics.Vulkan
             var clearValue = new ClearValue(null, new ClearDepthStencilValue(depthValue, (uint)stencilValue));
             var flags = depthMask ? ImageAspectFlags.DepthBit : 0;
 
-            if (stencilMask != 0)
+            if (stencilMask)
             {
                 flags |= ImageAspectFlags.StencilBit;
             }

src/Ryujinx.Graphics.Vulkan/PipelineFull.cs

@@ -81,6 +81,42 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        public void ClearRenderTargetDepthStencil(int layer, int layerCount, float depthValue, bool depthMask, int stencilValue, int stencilMask)
+        {
+            if (FramebufferParams == null)
+            {
+                return;
+            }
+
+            if (stencilMask != 0 && stencilMask != 0xff)
+            {
+                // We can't use CmdClearAttachments if not clearing all (mask is all ones, 0xFF) or none (mask is 0) of the stencil bits,
+                // because on Vulkan, the pipeline state does not affect clears.
+                var dstTexture = FramebufferParams.GetDepthStencilAttachment();
+                if (dstTexture == null)
+                {
+                    return;
+                }
+
+                // TODO: Clear only the specified layer.
+                Gd.HelperShader.Clear(
+                    Gd,
+                    dstTexture,
+                    depthValue,
+                    depthMask,
+                    stencilValue,
+                    stencilMask,
+                    (int)FramebufferParams.Width,
+                    (int)FramebufferParams.Height,
+                    FramebufferParams.AttachmentFormats[FramebufferParams.AttachmentsCount - 1],
+                    ClearScissor);
+            }
+            else
+            {
+                ClearRenderTargetDepthStencil(layer, layerCount, depthValue, depthMask, stencilValue, stencilMask != 0);
+            }
+        }
+
         public void EndHostConditionalRendering()
         {
             if (Gd.Capabilities.SupportsConditionalRendering)

src/Ryujinx.Graphics.Vulkan/PipelineState.cs

@@ -352,11 +352,6 @@ namespace Ryujinx.Graphics.Vulkan
                 return pipeline;
             }
 
-            if (gd.Capabilities.SupportsSubgroupSizeControl)
-            {
-                UpdateStageRequiredSubgroupSizes(gd, 1);
-            }
-
             var pipelineCreateInfo = new ComputePipelineCreateInfo
             {
                 SType = StructureType.ComputePipelineCreateInfo,
@@ -616,11 +611,6 @@ namespace Ryujinx.Graphics.Vulkan
                     PDynamicStates = dynamicStates,
                 };
 
-                if (gd.Capabilities.SupportsSubgroupSizeControl)
-                {
-                    UpdateStageRequiredSubgroupSizes(gd, (int)StagesCount);
-                }
-
                 var pipelineCreateInfo = new GraphicsPipelineCreateInfo
                 {
                     SType = StructureType.GraphicsPipelineCreateInfo,
@@ -659,19 +649,6 @@ namespace Ryujinx.Graphics.Vulkan
             return pipeline;
         }
 
-        private readonly unsafe void UpdateStageRequiredSubgroupSizes(VulkanRenderer gd, int count)
-        {
-            for (int index = 0; index < count; index++)
-            {
-                bool canUseExplicitSubgroupSize =
-                    (gd.Capabilities.RequiredSubgroupSizeStages & Stages[index].Stage) != 0 &&
-                    gd.Capabilities.MinSubgroupSize <= RequiredSubgroupSize &&
-                    gd.Capabilities.MaxSubgroupSize >= RequiredSubgroupSize;
-
-                Stages[index].PNext = canUseExplicitSubgroupSize ? StageRequiredSubgroupSizes.Pointer + index : null;
-            }
-        }
-
         private void UpdateVertexAttributeDescriptions(VulkanRenderer gd)
         {
             // Vertex attributes exceeding the stride are invalid.

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

@@ -42,6 +42,7 @@
     <EmbeddedResource Include="Shaders\SpirvBinaries\DepthBlitMsFragment.spv" />
     <EmbeddedResource Include="Shaders\SpirvBinaries\DepthDrawToMsFragment.spv" />
     <EmbeddedResource Include="Shaders\SpirvBinaries\DepthDrawToNonMsFragment.spv" />
+    <EmbeddedResource Include="Shaders\SpirvBinaries\DepthStencilClearFragment.spv" />
     <EmbeddedResource Include="Shaders\SpirvBinaries\StencilBlitFragment.spv" />
     <EmbeddedResource Include="Shaders\SpirvBinaries\StencilBlitMsFragment.spv" />
     <EmbeddedResource Include="Shaders\SpirvBinaries\StencilDrawToMsFragment.spv" />

src/Ryujinx.Graphics.Vulkan/Shaders/DepthStencilClearFragmentShaderSource.frag

@@ -0,0 +1,8 @@
+#version 450 core
+
+layout (location = 0) in vec4 clear_colour;
+
+void main()
+{
+    gl_FragDepth = clear_colour.x;
+}

src/Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -37,7 +37,6 @@ namespace Ryujinx.Graphics.Vulkan
             "VK_EXT_shader_stencil_export",
             "VK_KHR_shader_float16_int8",
             "VK_EXT_shader_subgroup_ballot",
-            "VK_EXT_subgroup_size_control",
             "VK_NV_geometry_shader_passthrough",
             "VK_NV_viewport_array2",
             "VK_EXT_depth_clip_control",

src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -151,6 +151,14 @@ namespace Ryujinx.Graphics.Vulkan
                 SType = StructureType.PhysicalDeviceProperties2,
             };
 
+            PhysicalDeviceSubgroupProperties propertiesSubgroup = new()
+            {
+                SType = StructureType.PhysicalDeviceSubgroupProperties,
+                PNext = properties2.PNext,
+            };
+
+            properties2.PNext = &propertiesSubgroup;
+
             PhysicalDeviceBlendOperationAdvancedPropertiesEXT propertiesBlendOperationAdvanced = new()
             {
                 SType = StructureType.PhysicalDeviceBlendOperationAdvancedPropertiesExt,
@@ -164,18 +172,6 @@ namespace Ryujinx.Graphics.Vulkan
                 properties2.PNext = &propertiesBlendOperationAdvanced;
             }
 
-            PhysicalDeviceSubgroupSizeControlPropertiesEXT propertiesSubgroupSizeControl = new()
-            {
-                SType = StructureType.PhysicalDeviceSubgroupSizeControlPropertiesExt,
-            };
-
-            bool supportsSubgroupSizeControl = _physicalDevice.IsDeviceExtensionPresent("VK_EXT_subgroup_size_control");
-
-            if (supportsSubgroupSizeControl)
-            {
-                properties2.PNext = &propertiesSubgroupSizeControl;
-            }
-
             bool supportsTransformFeedback = _physicalDevice.IsDeviceExtensionPresent(ExtTransformFeedback.ExtensionName);
 
             PhysicalDeviceTransformFeedbackPropertiesEXT propertiesTransformFeedback = new()
@@ -315,7 +311,6 @@ namespace Ryujinx.Graphics.Vulkan
                 _physicalDevice.IsDeviceExtensionPresent(KhrDrawIndirectCount.ExtensionName),
                 _physicalDevice.IsDeviceExtensionPresent("VK_EXT_fragment_shader_interlock"),
                 _physicalDevice.IsDeviceExtensionPresent("VK_NV_geometry_shader_passthrough"),
-                supportsSubgroupSizeControl,
                 features2.Features.ShaderFloat64,
                 featuresShaderInt8.ShaderInt8,
                 _physicalDevice.IsDeviceExtensionPresent("VK_EXT_shader_stencil_export"),
@@ -335,9 +330,7 @@ namespace Ryujinx.Graphics.Vulkan
                 _physicalDevice.IsDeviceExtensionPresent("VK_NV_viewport_array2"),
                 _physicalDevice.IsDeviceExtensionPresent(ExtExternalMemoryHost.ExtensionName),
                 supportsDepthClipControl && featuresDepthClipControl.DepthClipControl,
-                propertiesSubgroupSizeControl.MinSubgroupSize,
-                propertiesSubgroupSizeControl.MaxSubgroupSize,
-                propertiesSubgroupSizeControl.RequiredSubgroupSizeStages,
+                propertiesSubgroup.SubgroupSize,
                 supportedSampleCounts,
                 portabilityFlags,
                 vertexBufferAlignment,
@@ -623,6 +616,7 @@ namespace Ryujinx.Graphics.Vulkan
                 maximumImagesPerStage: Constants.MaxImagesPerStage,
                 maximumComputeSharedMemorySize: (int)limits.MaxComputeSharedMemorySize,
                 maximumSupportedAnisotropy: (int)limits.MaxSamplerAnisotropy,
+                shaderSubgroupSize: (int)Capabilities.SubgroupSize,
                 storageBufferOffsetAlignment: (int)limits.MinStorageBufferOffsetAlignment,
                 gatherBiasPrecision: IsIntelWindows || IsAmdWindows ? (int)Capabilities.SubTexelPrecisionBits : 0);
         }

src/Ryujinx.HLE/HOS/Services/Mm/IRequest.cs

@@ -1,196 +0,0 @@
-using Ryujinx.Common.Logging;
-using Ryujinx.HLE.HOS.Services.Mm.Types;
-using System.Collections.Generic;
-
-namespace Ryujinx.HLE.HOS.Services.Mm
-{
-    [Service("mm:u")]
-    class IRequest : IpcService
-    {
-        private readonly object _sessionListLock = new();
-        private readonly List<MultiMediaSession> _sessionList = new();
-
-        private uint _uniqueId = 1;
-
-        public IRequest(ServiceCtx context) { }
-
-        [CommandCmif(0)]
-        // InitializeOld(u32, u32, u32)
-        public ResultCode InitializeOld(ServiceCtx context)
-        {
-            MultiMediaOperationType operationType = (MultiMediaOperationType)context.RequestData.ReadUInt32();
-            int fgmId = context.RequestData.ReadInt32();
-            bool isAutoClearEvent = context.RequestData.ReadInt32() != 0;
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { operationType, fgmId, isAutoClearEvent });
-
-            Register(operationType, fgmId, isAutoClearEvent);
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(1)]
-        // FinalizeOld(u32)
-        public ResultCode FinalizeOld(ServiceCtx context)
-        {
-            MultiMediaOperationType operationType = (MultiMediaOperationType)context.RequestData.ReadUInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { operationType });
-
-            lock (_sessionListLock)
-            {
-                _sessionList.Remove(GetSessionByType(operationType));
-            }
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(2)]
-        // SetAndWaitOld(u32, u32, u32)
-        public ResultCode SetAndWaitOld(ServiceCtx context)
-        {
-            MultiMediaOperationType operationType = (MultiMediaOperationType)context.RequestData.ReadUInt32();
-            uint frequenceHz = context.RequestData.ReadUInt32();
-            int timeout = context.RequestData.ReadInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { operationType, frequenceHz, timeout });
-
-            lock (_sessionListLock)
-            {
-                GetSessionByType(operationType)?.SetAndWait(frequenceHz, timeout);
-            }
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(3)]
-        // GetOld(u32) -> u32
-        public ResultCode GetOld(ServiceCtx context)
-        {
-            MultiMediaOperationType operationType = (MultiMediaOperationType)context.RequestData.ReadUInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { operationType });
-
-            lock (_sessionListLock)
-            {
-                MultiMediaSession session = GetSessionByType(operationType);
-
-                uint currentValue = session == null ? 0 : session.CurrentValue;
-
-                context.ResponseData.Write(currentValue);
-            }
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(4)]
-        // Initialize(u32, u32, u32) -> u32
-        public ResultCode Initialize(ServiceCtx context)
-        {
-            MultiMediaOperationType operationType = (MultiMediaOperationType)context.RequestData.ReadUInt32();
-            int fgmId = context.RequestData.ReadInt32();
-            bool isAutoClearEvent = context.RequestData.ReadInt32() != 0;
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { operationType, fgmId, isAutoClearEvent });
-
-            uint id = Register(operationType, fgmId, isAutoClearEvent);
-
-            context.ResponseData.Write(id);
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(5)]
-        // Finalize(u32)
-        public ResultCode Finalize(ServiceCtx context)
-        {
-            uint id = context.RequestData.ReadUInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { id });
-
-            lock (_sessionListLock)
-            {
-                _sessionList.Remove(GetSessionById(id));
-            }
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(6)]
-        // SetAndWait(u32, u32, u32)
-        public ResultCode SetAndWait(ServiceCtx context)
-        {
-            uint id = context.RequestData.ReadUInt32();
-            uint frequenceHz = context.RequestData.ReadUInt32();
-            int timeout = context.RequestData.ReadInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { id, frequenceHz, timeout });
-
-            lock (_sessionListLock)
-            {
-                GetSessionById(id)?.SetAndWait(frequenceHz, timeout);
-            }
-
-            return ResultCode.Success;
-        }
-
-        [CommandCmif(7)]
-        // Get(u32) -> u32
-        public ResultCode Get(ServiceCtx context)
-        {
-            uint id = context.RequestData.ReadUInt32();
-
-            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { id });
-
-            lock (_sessionListLock)
-            {
-                MultiMediaSession session = GetSessionById(id);
-
-                uint currentValue = session == null ? 0 : session.CurrentValue;
-
-                context.ResponseData.Write(currentValue);
-            }
-
-            return ResultCode.Success;
-        }
-
-        private MultiMediaSession GetSessionById(uint id)
-        {
-            foreach (MultiMediaSession session in _sessionList)
-            {
-                if (session.Id == id)
-                {
-                    return session;
-                }
-            }
-
-            return null;
-        }
-
-        private MultiMediaSession GetSessionByType(MultiMediaOperationType type)
-        {
-            foreach (MultiMediaSession session in _sessionList)
-            {
-                if (session.Type == type)
-                {
-                    return session;
-                }
-            }
-
-            return null;
-        }
-
-        private uint Register(MultiMediaOperationType type, int fgmId, bool isAutoClearEvent)
-        {
-            lock (_sessionListLock)
-            {
-                // Nintendo ignore the fgm id as the other interfaces were deprecated.
-                MultiMediaSession session = new(_uniqueId++, type, isAutoClearEvent);
-
-                _sessionList.Add(session);
-
-                return session.Id;
-            }
-        }
-    }
-}

src/Ryujinx.HLE/HOS/Services/Mm/Types/MultiMediaOperationType.cs

@@ -1,10 +0,0 @@
-namespace Ryujinx.HLE.HOS.Services.Mm.Types
-{
-    enum MultiMediaOperationType : uint
-    {
-        Ram = 2,
-        NvEnc = 5,
-        NvDec = 6,
-        NvJpg = 7,
-    }
-}

src/Ryujinx.HLE/HOS/Services/Mm/Types/MultiMediaSession.cs

@@ -1,24 +0,0 @@
-namespace Ryujinx.HLE.HOS.Services.Mm.Types
-{
-    class MultiMediaSession
-    {
-        public MultiMediaOperationType Type { get; }
-
-        public bool IsAutoClearEvent { get; }
-        public uint Id { get; }
-        public uint CurrentValue { get; private set; }
-
-        public MultiMediaSession(uint id, MultiMediaOperationType type, bool isAutoClearEvent)
-        {
-            Type = type;
-            Id = id;
-            IsAutoClearEvent = isAutoClearEvent;
-            CurrentValue = 0;
-        }
-
-        public void SetAndWait(uint value, int timeout)
-        {
-            CurrentValue = value;
-        }
-    }
-}

src/Ryujinx.Horizon/MmNv/Ipc/Request.cs

@@ -0,0 +1,160 @@
+using Ryujinx.Common.Logging;
+using Ryujinx.Horizon.Common;
+using Ryujinx.Horizon.Sdk.MmNv;
+using Ryujinx.Horizon.Sdk.Sf;
+using System.Collections.Generic;
+
+namespace Ryujinx.Horizon.MmNv.Ipc
+{
+    partial class Request : IRequest
+    {
+        private readonly List<Session> _sessionList = new();
+
+        private uint _uniqueId = 1;
+
+        [CmifCommand(0)]
+        public Result InitializeOld(Module module, uint fgmPriority, uint autoClearEvent)
+        {
+            bool isAutoClearEvent = autoClearEvent != 0;
+
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { module, fgmPriority, isAutoClearEvent });
+
+            Register(module, fgmPriority, isAutoClearEvent);
+
+            return Result.Success;
+        }
+
+        [CmifCommand(1)]
+        public Result FinalizeOld(Module module)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { module });
+
+            lock (_sessionList)
+            {
+                _sessionList.Remove(GetSessionByModule(module));
+            }
+
+            return Result.Success;
+        }
+
+        [CmifCommand(2)]
+        public Result SetAndWaitOld(Module module, uint clockRateMin, int clockRateMax)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { module, clockRateMin, clockRateMax });
+
+            lock (_sessionList)
+            {
+                GetSessionByModule(module)?.SetAndWait(clockRateMin, clockRateMax);
+            }
+
+            return Result.Success;
+        }
+
+        [CmifCommand(3)]
+        public Result GetOld(out uint clockRateActual, Module module)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { module });
+
+            lock (_sessionList)
+            {
+                Session session = GetSessionByModule(module);
+
+                clockRateActual = session == null ? 0 : session.ClockRateMin;
+            }
+
+            return Result.Success;
+        }
+
+        [CmifCommand(4)]
+        public Result Initialize(out uint requestId, Module module, uint fgmPriority, uint autoClearEvent)
+        {
+            bool isAutoClearEvent = autoClearEvent != 0;
+
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { module, fgmPriority, isAutoClearEvent });
+
+            requestId = Register(module, fgmPriority, isAutoClearEvent);
+
+            return Result.Success;
+        }
+
+        [CmifCommand(5)]
+        public Result Finalize(uint requestId)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { requestId });
+
+            lock (_sessionList)
+            {
+                _sessionList.Remove(GetSessionById(requestId));
+            }
+
+            return Result.Success;
+        }
+
+        [CmifCommand(6)]
+        public Result SetAndWait(uint requestId, uint clockRateMin, int clockRateMax)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { requestId, clockRateMin, clockRateMax });
+
+            lock (_sessionList)
+            {
+                GetSessionById(requestId)?.SetAndWait(clockRateMin, clockRateMax);
+            }
+
+            return Result.Success;
+        }
+
+        [CmifCommand(7)]
+        public Result Get(out uint clockRateActual, uint requestId)
+        {
+            Logger.Stub?.PrintStub(LogClass.ServiceMm, new { requestId });
+
+            lock (_sessionList)
+            {
+                Session session = GetSessionById(requestId);
+
+                clockRateActual = session == null ? 0 : session.ClockRateMin;
+            }
+
+            return Result.Success;
+        }
+
+        private Session GetSessionById(uint id)
+        {
+            foreach (Session session in _sessionList)
+            {
+                if (session.Id == id)
+                {
+                    return session;
+                }
+            }
+
+            return null;
+        }
+
+        private Session GetSessionByModule(Module module)
+        {
+            foreach (Session session in _sessionList)
+            {
+                if (session.Module == module)
+                {
+                    return session;
+                }
+            }
+
+            return null;
+        }
+
+        private uint Register(Module module, uint fgmPriority, bool isAutoClearEvent)
+        {
+            lock (_sessionList)
+            {
+                // Nintendo ignores the fgm priority as the other services were deprecated.
+                Session session = new(_uniqueId++, module, isAutoClearEvent);
+
+                _sessionList.Add(session);
+
+                return session.Id;
+            }
+        }
+    }
+}

src/Ryujinx.Horizon/MmNv/MmNvIpcServer.cs

@@ -0,0 +1,43 @@
+using Ryujinx.Horizon.MmNv.Ipc;
+using Ryujinx.Horizon.Sdk.Sf.Hipc;
+using Ryujinx.Horizon.Sdk.Sm;
+
+namespace Ryujinx.Horizon.MmNv
+{
+    class MmNvIpcServer
+    {
+        private const int MmNvMaxSessionsCount = 9;
+
+        private const int PointerBufferSize = 0;
+        private const int MaxDomains = 0;
+        private const int MaxDomainObjects = 0;
+        private const int MaxPortsCount = 1;
+
+        private static readonly ManagerOptions _mmNvOptions = new(PointerBufferSize, MaxDomains, MaxDomainObjects, false);
+
+        private SmApi _sm;
+        private ServerManager _serverManager;
+
+        public void Initialize()
+        {
+            HeapAllocator allocator = new();
+
+            _sm = new SmApi();
+            _sm.Initialize().AbortOnFailure();
+
+            _serverManager = new ServerManager(allocator, _sm, MaxPortsCount, _mmNvOptions, MmNvMaxSessionsCount);
+
+            _serverManager.RegisterObjectForServer(new Request(), ServiceName.Encode("mm:u"), MmNvMaxSessionsCount);
+        }
+
+        public void ServiceRequests()
+        {
+            _serverManager.ServiceRequests();
+        }
+
+        public void Shutdown()
+        {
+            _serverManager.Dispose();
+        }
+    }
+}

src/Ryujinx.Horizon/MmNv/MmNvMain.cs

@@ -0,0 +1,17 @@
+namespace Ryujinx.Horizon.MmNv
+{
+    class MmNvMain : IService
+    {
+        public static void Main(ServiceTable serviceTable)
+        {
+            MmNvIpcServer ipcServer = new();
+
+            ipcServer.Initialize();
+
+            serviceTable.SignalServiceReady();
+
+            ipcServer.ServiceRequests();
+            ipcServer.Shutdown();
+        }
+    }
+}

src/Ryujinx.Horizon/Sdk/MmNv/IRequest.cs

@@ -0,0 +1,17 @@
+using Ryujinx.Horizon.Common;
+using Ryujinx.Horizon.Sdk.Sf;
+
+namespace Ryujinx.Horizon.Sdk.MmNv
+{
+    interface IRequest : IServiceObject
+    {
+        Result InitializeOld(Module module, uint fgmPriority, uint autoClearEvent);
+        Result FinalizeOld(Module module);
+        Result SetAndWaitOld(Module module, uint clockRateMin, int clockRateMax);
+        Result GetOld(out uint clockRateActual, Module module);
+        Result Initialize(out uint requestId, Module module, uint fgmPriority, uint autoClearEvent);
+        Result Finalize(uint requestId);
+        Result SetAndWait(uint requestId, uint clockRateMin, int clockRateMax);
+        Result Get(out uint clockRateActual, uint requestId);
+    }
+}

src/Ryujinx.Horizon/Sdk/MmNv/Module.cs

@@ -0,0 +1,15 @@
+namespace Ryujinx.Horizon.Sdk.MmNv
+{
+    enum Module : uint
+    {
+        Cpu,
+        Gpu,
+        Emc,
+        SysBus,
+        MSelect,
+        NvDec,
+        NvEnc,
+        NvJpg,
+        Test,
+    }
+}

src/Ryujinx.Horizon/Sdk/MmNv/Session.cs

@@ -0,0 +1,26 @@
+namespace Ryujinx.Horizon.Sdk.MmNv
+{
+    class Session
+    {
+        public Module Module { get; }
+        public uint Id { get; }
+        public bool IsAutoClearEvent { get; }
+        public uint ClockRateMin { get; private set; }
+        public int ClockRateMax { get; private set; }
+
+        public Session(uint id, Module module, bool isAutoClearEvent)
+        {
+            Module = module;
+            Id = id;
+            IsAutoClearEvent = isAutoClearEvent;
+            ClockRateMin = 0;
+            ClockRateMax = -1;
+        }
+
+        public void SetAndWait(uint clockRateMin, int clockRateMax)
+        {
+            ClockRateMin = clockRateMin;
+            ClockRateMax = clockRateMax;
+        }
+    }
+}

src/Ryujinx.Horizon/ServiceTable.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Horizon.Bcat;
 using Ryujinx.Horizon.LogManager;
+using Ryujinx.Horizon.MmNv;
 using Ryujinx.Horizon.Prepo;
 using System.Collections.Generic;
 using System.Threading;
@@ -25,6 +26,7 @@ namespace Ryujinx.Horizon
             RegisterService<LmMain>();
             RegisterService<PrepoMain>();
             RegisterService<BcatMain>();
+            RegisterService<MmNvMain>();
 
             _totalServices = entries.Count;