nuget: bump ImageSharp from 2.1.8 to 2.1.9 (#7160)

While building I got some warnings, so I updated the dependency.

`warning NU1903: Package 'SixLabors.ImageSharp' 2.1.8 has a known high severity vulnerability, https://github.com/advisories/GHSA-63p8-c4ww-9cg7`

Directory.Packages.props

@@ -13,16 +13,16 @@
     <PackageVersion Include="CommandLineParser" Version="2.9.1" />
     <PackageVersion Include="Concentus" Version="2.2.0" />
     <PackageVersion Include="DiscordRichPresence" Version="1.2.1.24" />
-    <PackageVersion Include="DynamicData" Version="8.4.1" />
+    <PackageVersion Include="DynamicData" Version="9.0.1" />
     <PackageVersion Include="FluentAvaloniaUI" Version="2.0.5" />
     <PackageVersion Include="GtkSharp.Dependencies" Version="1.1.1" />
     <PackageVersion Include="GtkSharp.Dependencies.osx" Version="0.0.5" />
     <PackageVersion Include="LibHac" Version="0.19.0" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.9.2" />
-    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="7.5.2" />
+    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="8.0.1" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.9.0" />
-    <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.0" />
+    <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.1" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />
     <PackageVersion Include="NetCoreServer" Version="8.0.7" />
     <PackageVersion Include="NUnit" Version="3.13.3" />
@@ -42,7 +42,7 @@
     <PackageVersion Include="Silk.NET.Vulkan" Version="2.16.0" />
     <PackageVersion Include="Silk.NET.Vulkan.Extensions.EXT" Version="2.16.0" />
     <PackageVersion Include="Silk.NET.Vulkan.Extensions.KHR" Version="2.16.0" />
-    <PackageVersion Include="SixLabors.ImageSharp" Version="2.1.8" />
+    <PackageVersion Include="SixLabors.ImageSharp" Version="2.1.9" />
     <PackageVersion Include="SixLabors.ImageSharp.Drawing" Version="1.0.0" />
     <PackageVersion Include="SPB" Version="0.0.4-build32" />
     <PackageVersion Include="System.IO.Hashing" Version="8.0.0" />

src/ARMeilleure/CodeGen/Arm64/CodeGenContext.cs

@@ -237,7 +237,7 @@ namespace ARMeilleure.CodeGen.Arm64
             long originalPosition = _stream.Position;
 
             _stream.Seek(0, SeekOrigin.Begin);
-            _stream.Read(code, 0, code.Length);
+            _stream.ReadExactly(code, 0, code.Length);
             _stream.Seek(originalPosition, SeekOrigin.Begin);
 
             RelocInfo relocInfo;

src/ARMeilleure/CodeGen/RegisterAllocators/LinearScanAllocator.cs

@@ -251,7 +251,20 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                 }
             }
 
-            int selectedReg = GetHighestValueIndex(freePositions);
+            // If this is a copy destination variable, we prefer the register used for the copy source.
+            // If the register is available, then the copy can be eliminated later as both source
+            // and destination will use the same register.
+            int selectedReg;
+
+            if (current.TryGetCopySourceRegister(out int preferredReg) && freePositions[preferredReg] >= current.GetEnd())
+            {
+                selectedReg = preferredReg;
+            }
+            else
+            {
+                selectedReg = GetHighestValueIndex(freePositions);
+            }
+
             int selectedNextUse = freePositions[selectedReg];
 
             // Intervals starts and ends at odd positions, unless they span an entire
@@ -431,7 +444,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             }
         }
 
-        private static int GetHighestValueIndex(Span<int> span)
+        private static int GetHighestValueIndex(ReadOnlySpan<int> span)
         {
             int highest = int.MinValue;
 
@@ -798,12 +811,12 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             // The "visited" state is stored in the MSB of the local's value.
             const ulong VisitedMask = 1ul << 63;
 
-            bool IsVisited(Operand local)
+            static bool IsVisited(Operand local)
             {
                 return (local.GetValueUnsafe() & VisitedMask) != 0;
             }
 
-            void SetVisited(Operand local)
+            static void SetVisited(Operand local)
             {
                 local.GetValueUnsafe() |= VisitedMask;
             }
@@ -826,9 +839,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                         {
                             dest.NumberLocal(_intervals.Count);
 
-                            _intervals.Add(new LiveInterval(dest));
+                            LiveInterval interval = new LiveInterval(dest);
+                            _intervals.Add(interval);
 
                             SetVisited(dest);
+
+                            // If this is a copy (or copy-like operation), set the copy source interval as well.
+                            // This is used for register preferencing later on, which allows the copy to be eliminated
+                            // in some cases.
+                            if (node.Instruction == Instruction.Copy || node.Instruction == Instruction.ZeroExtend32)
+                            {
+                                Operand source = node.GetSource(0);
+
+                                if (source.Kind == OperandKind.LocalVariable &&
+                                    source.GetLocalNumber() > 0 &&
+                                    (node.Instruction == Instruction.Copy || source.Type == OperandType.I32))
+                                {
+                                    interval.SetCopySource(_intervals[source.GetLocalNumber()]);
+                                }
+                            }
                         }
                     }
                 }

src/ARMeilleure/CodeGen/RegisterAllocators/LiveInterval.cs

@@ -19,6 +19,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             public LiveRange CurrRange;
 
             public LiveInterval Parent;
+            public LiveInterval CopySource;
 
             public UseList Uses;
             public LiveIntervalList Children;
@@ -37,6 +38,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
         private ref LiveRange CurrRange => ref _data->CurrRange;
         private ref LiveRange PrevRange => ref _data->PrevRange;
         private ref LiveInterval Parent => ref _data->Parent;
+        private ref LiveInterval CopySource => ref _data->CopySource;
         private ref UseList Uses => ref _data->Uses;
         private ref LiveIntervalList Children => ref _data->Children;
 
@@ -78,6 +80,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             Register = register;
         }
 
+        public void SetCopySource(LiveInterval copySource)
+        {
+            CopySource = copySource;
+        }
+
+        public bool TryGetCopySourceRegister(out int copySourceRegIndex)
+        {
+            if (CopySource._data != null)
+            {
+                copySourceRegIndex = CopySource.Register.Index;
+
+                return true;
+            }
+
+            copySourceRegIndex = 0;
+
+            return false;
+        }
+
         public void Reset()
         {
             PrevRange = default;

src/ARMeilleure/CodeGen/X86/Assembler.cs

@@ -1444,7 +1444,7 @@ namespace ARMeilleure.CodeGen.X86
 
                 Span<byte> buffer = new byte[jump.JumpPosition - _stream.Position];
 
-                _stream.Read(buffer);
+                _stream.ReadExactly(buffer);
                 _stream.Seek(ReservedBytesForJump, SeekOrigin.Current);
 
                 codeStream.Write(buffer);

src/ARMeilleure/Decoders/OpCodeTable.cs

@@ -822,6 +822,10 @@ namespace ARMeilleure.Decoders
             SetA32("<<<<00000100xxxxxxxxxxxx1001xxxx", InstName.Umaal,   InstEmit32.Umaal,   OpCode32AluUmull.Create);
             SetA32("<<<<0000101xxxxxxxxxxxxx1001xxxx", InstName.Umlal,   InstEmit32.Umlal,   OpCode32AluUmull.Create);
             SetA32("<<<<0000100xxxxxxxxxxxxx1001xxxx", InstName.Umull,   InstEmit32.Umull,   OpCode32AluUmull.Create);
+            SetA32("<<<<01100110xxxxxxxx11110001xxxx", InstName.Uqadd16, InstEmit32.Uqadd16, OpCode32AluReg.Create);
+            SetA32("<<<<01100110xxxxxxxx11111001xxxx", InstName.Uqadd8,  InstEmit32.Uqadd8,  OpCode32AluReg.Create);
+            SetA32("<<<<01100110xxxxxxxx11110111xxxx", InstName.Uqsub16, InstEmit32.Uqsub16, OpCode32AluReg.Create);
+            SetA32("<<<<01100110xxxxxxxx11111111xxxx", InstName.Uqsub8,  InstEmit32.Uqsub8,  OpCode32AluReg.Create);
             SetA32("<<<<0110111xxxxxxxxxxxxxxx01xxxx", InstName.Usat,    InstEmit32.Usat,    OpCode32Sat.Create);
             SetA32("<<<<01101110xxxxxxxx11110011xxxx", InstName.Usat16,  InstEmit32.Usat16,  OpCode32Sat16.Create);
             SetA32("<<<<01100101xxxxxxxx11111111xxxx", InstName.Usub8,   InstEmit32.Usub8,   OpCode32AluReg.Create);
@@ -1007,6 +1011,8 @@ namespace ARMeilleure.Decoders
             SetAsimd("111100100x10xxxxxxxx1011xxx0xxxx", InstName.Vqdmulh,     InstEmit32.Vqdmulh,      OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);
             SetAsimd("111100111x11<<10xxxx00101xx0xxx0", InstName.Vqmovn,      InstEmit32.Vqmovn,       OpCode32SimdMovn.Create,        OpCode32SimdMovn.CreateT32);
             SetAsimd("111100111x11<<10xxxx001001x0xxx0", InstName.Vqmovun,     InstEmit32.Vqmovun,      OpCode32SimdMovn.Create,        OpCode32SimdMovn.CreateT32);
+            SetAsimd("111100110x01xxxxxxxx1011xxx0xxxx", InstName.Vqrdmulh,    InstEmit32.Vqrdmulh,     OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);
+            SetAsimd("111100110x10xxxxxxxx1011xxx0xxxx", InstName.Vqrdmulh,    InstEmit32.Vqrdmulh,     OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);
             SetAsimd("1111001x1x>>>xxxxxxx100101x1xxx0", InstName.Vqrshrn,     InstEmit32.Vqrshrn,      OpCode32SimdShImmNarrow.Create, OpCode32SimdShImmNarrow.CreateT32);
             SetAsimd("111100111x>>>xxxxxxx100001x1xxx0", InstName.Vqrshrun,    InstEmit32.Vqrshrun,     OpCode32SimdShImmNarrow.Create, OpCode32SimdShImmNarrow.CreateT32);
             SetAsimd("1111001x1x>>>xxxxxxx100100x1xxx0", InstName.Vqshrn,      InstEmit32.Vqshrn,       OpCode32SimdShImmNarrow.Create, OpCode32SimdShImmNarrow.CreateT32);
@@ -1030,6 +1036,7 @@ namespace ARMeilleure.Decoders
             SetAsimd("1111001x1x>>>xxxxxxx101000x1xxxx", InstName.Vshll,       InstEmit32.Vshll,        OpCode32SimdShImmLong.Create,   OpCode32SimdShImmLong.CreateT32); // A1 encoding.
             SetAsimd("1111001x1x>>>xxxxxxx0000>xx1xxxx", InstName.Vshr,        InstEmit32.Vshr,         OpCode32SimdShImm.Create,       OpCode32SimdShImm.CreateT32);
             SetAsimd("111100101x>>>xxxxxxx100000x1xxx0", InstName.Vshrn,       InstEmit32.Vshrn,        OpCode32SimdShImmNarrow.Create, OpCode32SimdShImmNarrow.CreateT32);
+            SetAsimd("111100111x>>>xxxxxxx0101>xx1xxxx", InstName.Vsli,        InstEmit32.Vsli_I,       OpCode32SimdShImm.Create,       OpCode32SimdShImm.CreateT32);
             SetAsimd("1111001x1x>>>xxxxxxx0001>xx1xxxx", InstName.Vsra,        InstEmit32.Vsra,         OpCode32SimdShImm.Create,       OpCode32SimdShImm.CreateT32);
             SetAsimd("111101001x00xxxxxxxx0000xxx0xxxx", InstName.Vst1,        InstEmit32.Vst1,         OpCode32SimdMemSingle.Create,   OpCode32SimdMemSingle.CreateT32);
             SetAsimd("111101001x00xxxxxxxx0100xx0xxxxx", InstName.Vst1,        InstEmit32.Vst1,         OpCode32SimdMemSingle.Create,   OpCode32SimdMemSingle.CreateT32);
@@ -1054,6 +1061,7 @@ namespace ARMeilleure.Decoders
             SetAsimd("111100100x10xxxxxxxx1101xxx0xxxx", InstName.Vsub,        InstEmit32.Vsub_V,       OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);
             SetAsimd("1111001x1x<<xxxxxxx00010x0x0xxxx", InstName.Vsubl,       InstEmit32.Vsubl_I,      OpCode32SimdRegLong.Create,     OpCode32SimdRegLong.CreateT32);
             SetAsimd("1111001x1x<<xxxxxxx00011x0x0xxxx", InstName.Vsubw,       InstEmit32.Vsubw_I,      OpCode32SimdRegWide.Create,     OpCode32SimdRegWide.CreateT32);
+            SetAsimd("111100111x110010xxxx00000xx0xxxx", InstName.Vswp,        InstEmit32.Vswp,         OpCode32Simd.Create,            OpCode32Simd.CreateT32);
             SetAsimd("111100111x11xxxxxxxx10xxxxx0xxxx", InstName.Vtbl,        InstEmit32.Vtbl,         OpCode32SimdTbl.Create,         OpCode32SimdTbl.CreateT32);
             SetAsimd("111100111x11<<10xxxx00001xx0xxxx", InstName.Vtrn,        InstEmit32.Vtrn,         OpCode32SimdCmpZ.Create,        OpCode32SimdCmpZ.CreateT32);
             SetAsimd("111100100x<<xxxxxxxx1000xxx1xxxx", InstName.Vtst,        InstEmit32.Vtst,         OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);

src/ARMeilleure/Instructions/InstEmitAlu32.cs

@@ -2,6 +2,8 @@ using ARMeilleure.Decoders;
 using ARMeilleure.IntermediateRepresentation;
 using ARMeilleure.State;
 using ARMeilleure.Translation;
+using System;
+using System.Diagnostics;
 using System.Diagnostics.CodeAnalysis;
 using static ARMeilleure.Instructions.InstEmitAluHelper;
 using static ARMeilleure.Instructions.InstEmitHelper;
@@ -558,6 +560,46 @@ namespace ARMeilleure.Instructions
             EmitHsub8(context, unsigned: true);
         }
 
+        public static void Uqadd16(ArmEmitterContext context)
+        {
+            OpCode32AluReg op = (OpCode32AluReg)context.CurrOp;
+
+            SetIntA32(context, op.Rd, EmitUnsigned16BitPair(context, GetIntA32(context, op.Rn), GetIntA32(context, op.Rm), (d, n, m) =>
+            {
+                EmitSaturateUqadd(context, d, context.Add(n, m), 16);
+            }));
+        }
+
+        public static void Uqadd8(ArmEmitterContext context)
+        {
+            OpCode32AluReg op = (OpCode32AluReg)context.CurrOp;
+
+            SetIntA32(context, op.Rd, EmitUnsigned8BitPair(context, GetIntA32(context, op.Rn), GetIntA32(context, op.Rm), (d, n, m) =>
+            {
+                EmitSaturateUqadd(context, d, context.Add(n, m), 8);
+            }));
+        }
+
+        public static void Uqsub16(ArmEmitterContext context)
+        {
+            OpCode32AluReg op = (OpCode32AluReg)context.CurrOp;
+
+            SetIntA32(context, op.Rd, EmitUnsigned16BitPair(context, GetIntA32(context, op.Rn), GetIntA32(context, op.Rm), (d, n, m) =>
+            {
+                EmitSaturateUqsub(context, d, context.Subtract(n, m), 16);
+            }));
+        }
+
+        public static void Uqsub8(ArmEmitterContext context)
+        {
+            OpCode32AluReg op = (OpCode32AluReg)context.CurrOp;
+
+            SetIntA32(context, op.Rd, EmitUnsigned8BitPair(context, GetIntA32(context, op.Rn), GetIntA32(context, op.Rm), (d, n, m) =>
+            {
+                EmitSaturateUqsub(context, d, context.Subtract(n, m), 8);
+            }));
+        }
+
         public static void Usat(ArmEmitterContext context)
         {
             OpCode32Sat op = (OpCode32Sat)context.CurrOp;
@@ -934,6 +976,148 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        private static void EmitSaturateUqadd(ArmEmitterContext context, Operand result, Operand value, uint saturateTo)
+        {
+            Debug.Assert(saturateTo <= 32);
+
+            if (saturateTo == 32)
+            {
+                // No saturation possible for this case.
+
+                context.Copy(result, value);
+
+                return;
+            }
+            else if (saturateTo == 0)
+            {
+                // Result is always zero if we saturate 0 bits.
+
+                context.Copy(result, Const(0));
+
+                return;
+            }
+
+            // If the result is 0, the values are equal and we don't need saturation.
+            Operand lblNoSat = Label();
+            context.BranchIfFalse(lblNoSat, context.ShiftRightUI(value, Const((int)saturateTo)));
+
+            // Saturate.
+            context.Copy(result, Const(uint.MaxValue >> (32 - (int)saturateTo)));
+
+            Operand lblExit = Label();
+            context.Branch(lblExit);
+
+            context.MarkLabel(lblNoSat);
+
+            context.Copy(result, value);
+
+            context.MarkLabel(lblExit);
+        }
+
+        private static void EmitSaturateUqsub(ArmEmitterContext context, Operand result, Operand value, uint saturateTo)
+        {
+            Debug.Assert(saturateTo <= 32);
+
+            if (saturateTo == 32)
+            {
+                // No saturation possible for this case.
+
+                context.Copy(result, value);
+
+                return;
+            }
+            else if (saturateTo == 0)
+            {
+                // Result is always zero if we saturate 0 bits.
+
+                context.Copy(result, Const(0));
+
+                return;
+            }
+
+            // If the result is 0, the values are equal and we don't need saturation.
+            Operand lblNoSat = Label();
+            context.BranchIf(lblNoSat, value, Const(0), Comparison.GreaterOrEqual);
+
+            // Saturate.
+            // Assumes that the value can only underflow, since this is only used for unsigned subtraction.
+            context.Copy(result, Const(0));
+
+            Operand lblExit = Label();
+            context.Branch(lblExit);
+
+            context.MarkLabel(lblNoSat);
+
+            context.Copy(result, value);
+
+            context.MarkLabel(lblExit);
+        }
+
+        private static Operand EmitUnsigned16BitPair(ArmEmitterContext context, Operand rn, Operand rm, Action<Operand, Operand, Operand> elementAction)
+        {
+            Operand tempD = context.AllocateLocal(OperandType.I32);
+
+            Operand tempN = context.ZeroExtend16(OperandType.I32, rn);
+            Operand tempM = context.ZeroExtend16(OperandType.I32, rm);
+            elementAction(tempD, tempN, tempM);
+            Operand tempD2 = context.ZeroExtend16(OperandType.I32, tempD);
+
+            tempN = context.ShiftRightUI(rn, Const(16));
+            tempM = context.ShiftRightUI(rm, Const(16));
+            elementAction(tempD, tempN, tempM);
+            return context.BitwiseOr(tempD2, context.ShiftLeft(tempD, Const(16)));
+        }
+
+        private static Operand EmitSigned8BitPair(ArmEmitterContext context, Operand rn, Operand rm, Action<Operand, Operand, Operand> elementAction)
+        {
+            return Emit8BitPair(context, rn, rm, elementAction, unsigned: false);
+        }
+
+        private static Operand EmitUnsigned8BitPair(ArmEmitterContext context, Operand rn, Operand rm, Action<Operand, Operand, Operand> elementAction)
+        {
+            return Emit8BitPair(context, rn, rm, elementAction, unsigned: true);
+        }
+
+        private static Operand Emit8BitPair(ArmEmitterContext context, Operand rn, Operand rm, Action<Operand, Operand, Operand> elementAction, bool unsigned)
+        {
+            Operand tempD = context.AllocateLocal(OperandType.I32);
+            Operand result = default;
+
+            for (int b = 0; b < 4; b++)
+            {
+                Operand nByte = b != 0 ? context.ShiftRightUI(rn, Const(b * 8)) : rn;
+                Operand mByte = b != 0 ? context.ShiftRightUI(rm, Const(b * 8)) : rm;
+
+                if (unsigned)
+                {
+                    nByte = context.ZeroExtend8(OperandType.I32, nByte);
+                    mByte = context.ZeroExtend8(OperandType.I32, mByte);
+                }
+                else
+                {
+                    nByte = context.SignExtend8(OperandType.I32, nByte);
+                    mByte = context.SignExtend8(OperandType.I32, mByte);
+                }
+
+                elementAction(tempD, nByte, mByte);
+
+                if (b == 0)
+                {
+                    result = context.ZeroExtend8(OperandType.I32, tempD);
+                }
+                else if (b < 3)
+                {
+                    result = context.BitwiseOr(result, context.ShiftLeft(context.ZeroExtend8(OperandType.I32, tempD), Const(b * 8)));
+                }
+                else
+                {
+                    result = context.BitwiseOr(result, context.ShiftLeft(tempD, Const(24)));
+                }
+            }
+
+            return result;
+        }
+
         private static void EmitAluStore(ArmEmitterContext context, Operand value)
         {
             IOpCode32Alu op = (IOpCode32Alu)context.CurrOp;

src/ARMeilleure/Instructions/InstEmitSimdArithmetic32.cs

@@ -1246,6 +1246,33 @@ namespace ARMeilleure.Instructions
             EmitVectorUnaryNarrowOp32(context, (op1) => EmitSatQ(context, op1, 8 << op.Size, signedSrc: true, signedDst: false), signed: true);
         }
 
+        public static void Vqrdmulh(ArmEmitterContext context)
+        {
+            OpCode32SimdReg op = (OpCode32SimdReg)context.CurrOp;
+            int eSize = 8 << op.Size;
+
+            EmitVectorBinaryOpI32(context, (op1, op2) =>
+            {
+                if (op.Size == 2)
+                {
+                    op1 = context.SignExtend32(OperandType.I64, op1);
+                    op2 = context.SignExtend32(OperandType.I64, op2);
+                }
+
+                Operand res = context.Multiply(op1, op2);
+                res = context.Add(res, Const(res.Type, 1L << (eSize - 2)));
+                res = context.ShiftRightSI(res, Const(eSize - 1));
+                res = EmitSatQ(context, res, eSize, signedSrc: true, signedDst: true);
+
+                if (op.Size == 2)
+                {
+                    res = context.ConvertI64ToI32(res);
+                }
+
+                return res;
+            }, signed: true);
+        }
+
         public static void Vqsub(ArmEmitterContext context)
         {
             OpCode32SimdReg op = (OpCode32SimdReg)context.CurrOp;

src/ARMeilleure/Instructions/InstEmitSimdMove32.cs

@@ -191,6 +191,26 @@ namespace ARMeilleure.Instructions
             context.Copy(GetVecA32(op.Qd), res);
         }
 
+        public static void Vswp(ArmEmitterContext context)
+        {
+            OpCode32Simd op = (OpCode32Simd)context.CurrOp;
+
+            if (op.Q)
+            {
+                Operand temp = context.Copy(GetVecA32(op.Qd));
+
+                context.Copy(GetVecA32(op.Qd), GetVecA32(op.Qm));
+                context.Copy(GetVecA32(op.Qm), temp);
+            }
+            else
+            {
+                Operand temp = ExtractScalar(context, OperandType.I64, op.Vd);
+
+                InsertScalar(context, op.Vd, ExtractScalar(context, OperandType.I64, op.Vm));
+                InsertScalar(context, op.Vm, temp);
+            }
+        }
+
         public static void Vtbl(ArmEmitterContext context)
         {
             OpCode32SimdTbl op = (OpCode32SimdTbl)context.CurrOp;

src/ARMeilleure/Instructions/InstEmitSimdShift32.cs

@@ -130,6 +130,36 @@ namespace ARMeilleure.Instructions
             EmitVectorUnaryNarrowOp32(context, (op1) => context.ShiftRightUI(op1, Const(shift)));
         }
 
+        public static void Vsli_I(ArmEmitterContext context)
+        {
+            OpCode32SimdShImm op = (OpCode32SimdShImm)context.CurrOp;
+            int shift = op.Shift;
+            int eSize = 8 << op.Size;
+
+            ulong mask = shift != 0 ? ulong.MaxValue >> (64 - shift) : 0UL;
+
+            Operand res = GetVec(op.Qd);
+
+            int elems = op.GetBytesCount() >> op.Size;
+
+            for (int index = 0; index < elems; index++)
+            {
+                Operand me = EmitVectorExtractZx(context, op.Qm, op.Im + index, op.Size);
+
+                Operand neShifted = context.ShiftLeft(me, Const(shift));
+
+                Operand de = EmitVectorExtractZx(context, op.Qd, op.Id + index, op.Size);
+
+                Operand deMasked = context.BitwiseAnd(de, Const(mask));
+
+                Operand e = context.BitwiseOr(neShifted, deMasked);
+
+                res = EmitVectorInsert(context, res, e, op.Id + index, op.Size);
+            }
+
+            context.Copy(GetVec(op.Qd), res);
+        }
+
         public static void Vsra(ArmEmitterContext context)
         {
             OpCode32SimdShImm op = (OpCode32SimdShImm)context.CurrOp;

src/ARMeilleure/Instructions/InstName.cs

@@ -571,6 +571,10 @@ namespace ARMeilleure.Instructions
         Umaal,
         Umlal,
         Umull,
+        Uqadd16,
+        Uqadd8,
+        Uqsub16,
+        Uqsub8,
         Usat,
         Usat16,
         Usub8,
@@ -645,6 +649,7 @@ namespace ARMeilleure.Instructions
         Vqdmulh,
         Vqmovn,
         Vqmovun,
+        Vqrdmulh,
         Vqrshrn,
         Vqrshrun,
         Vqshrn,
@@ -666,6 +671,7 @@ namespace ARMeilleure.Instructions
         Vshll,
         Vshr,
         Vshrn,
+        Vsli,
         Vst1,
         Vst2,
         Vst3,
@@ -682,6 +688,7 @@ namespace ARMeilleure.Instructions
         Vsub,
         Vsubl,
         Vsubw,
+        Vswp,
         Vtbl,
         Vtrn,
         Vtst,

src/ARMeilleure/Translation/ControlFlowGraph.cs

@@ -11,7 +11,7 @@ namespace ARMeilleure.Translation
         private int[] _postOrderMap;
 
         public int LocalsCount { get; private set; }
-        public BasicBlock Entry { get; }
+        public BasicBlock Entry { get; private set; }
         public IntrusiveList<BasicBlock> Blocks { get; }
         public BasicBlock[] PostOrderBlocks => _postOrderBlocks;
         public int[] PostOrderMap => _postOrderMap;
@@ -34,6 +34,15 @@ namespace ARMeilleure.Translation
             return result;
         }
 
+        public void UpdateEntry(BasicBlock newEntry)
+        {
+            newEntry.AddSuccessor(Entry);
+
+            Entry = newEntry;
+            Blocks.AddFirst(newEntry);
+            Update();
+        }
+
         public void Update()
         {
             RemoveUnreachableBlocks(Blocks);

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -29,7 +29,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 6634; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 6950; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

src/ARMeilleure/Translation/RegisterUsage.cs

@@ -89,6 +89,17 @@ namespace ARMeilleure.Translation
 
         public static void RunPass(ControlFlowGraph cfg, ExecutionMode mode)
         {
+            if (cfg.Entry.Predecessors.Count != 0)
+            {
+                // We expect the entry block to have no predecessors.
+                // This is required because we have a implicit context load at the start of the function,
+                // but if there is a jump to the start of the function, the context load would trash the modified values.
+                // Here we insert a new entry block that will jump to the existing entry block.
+                BasicBlock newEntry = new BasicBlock(cfg.Blocks.Count);
+
+                cfg.UpdateEntry(newEntry);
+            }
+
             // Compute local register inputs and outputs used inside blocks.
             RegisterMask[] localInputs = new RegisterMask[cfg.Blocks.Count];
             RegisterMask[] localOutputs = new RegisterMask[cfg.Blocks.Count];
@@ -201,7 +212,7 @@ namespace ARMeilleure.Translation
 
                 // The only block without any predecessor should be the entry block.
                 // It always needs a context load as it is the first block to run.
-                if (block.Predecessors.Count == 0 || hasContextLoad)
+                if (block == cfg.Entry || hasContextLoad)
                 {
                     long vecMask = globalInputs[block.Index].VecMask;
                     long intMask = globalInputs[block.Index].IntMask;

src/Ryujinx.Audio.Backends.SDL2/SDL2HardwareDeviceSession.cs

@@ -89,9 +89,9 @@ namespace Ryujinx.Audio.Backends.SDL2
                 return;
             }
 
-            using IMemoryOwner<byte> samplesOwner = ByteMemoryPool.Rent(frameCount * _bytesPerFrame);
+            using SpanOwner<byte> samplesOwner = SpanOwner<byte>.Rent(frameCount * _bytesPerFrame);
 
-            Span<byte> samples = samplesOwner.Memory.Span;
+            Span<byte> samples = samplesOwner.Span;
 
             _ringBuffer.Read(samples, 0, samples.Length);
 

src/Ryujinx.Audio.Backends.SoundIo/SoundIoHardwareDeviceSession.cs

@@ -122,9 +122,9 @@ namespace Ryujinx.Audio.Backends.SoundIo
 
             int channelCount = areas.Length;
 
-            using IMemoryOwner<byte> samplesOwner = ByteMemoryPool.Rent(frameCount * bytesPerFrame);
+            using SpanOwner<byte> samplesOwner = SpanOwner<byte>.Rent(frameCount * bytesPerFrame);
 
-            Span<byte> samples = samplesOwner.Memory.Span;
+            Span<byte> samples = samplesOwner.Span;
 
             _ringBuffer.Read(samples, 0, samples.Length);
 

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

@@ -14,7 +14,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         private readonly object _lock = new();
 
-        private IMemoryOwner<byte> _bufferOwner;
+        private MemoryOwner<byte> _bufferOwner;
         private Memory<byte> _buffer;
         private int _size;
         private int _headOffset;
@@ -24,7 +24,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         public DynamicRingBuffer(int initialCapacity = RingBufferAlignment)
         {
-            _bufferOwner = ByteMemoryPool.RentCleared(initialCapacity);
+            _bufferOwner = MemoryOwner<byte>.RentCleared(initialCapacity);
             _buffer = _bufferOwner.Memory;
         }
 
@@ -62,7 +62,7 @@ namespace Ryujinx.Audio.Backends.Common
 
         private void SetCapacityLocked(int capacity)
         {
-            IMemoryOwner<byte> newBufferOwner = ByteMemoryPool.RentCleared(capacity);
+            MemoryOwner<byte> newBufferOwner = MemoryOwner<byte>.RentCleared(capacity);
             Memory<byte> newBuffer = newBufferOwner.Memory;
 
             if (_size > 0)

src/Ryujinx.Audio/Renderer/Server/Performance/PerformanceManager.cs

@@ -18,16 +18,12 @@ namespace Ryujinx.Audio.Renderer.Server.Performance
 
             if (version == 2)
             {
-                return (ulong)PerformanceManagerGeneric<PerformanceFrameHeaderVersion2,
-                                                 PerformanceEntryVersion2,
-                                                 PerformanceDetailVersion2>.GetRequiredBufferSizeForPerformanceMetricsPerFrame(ref parameter);
+                return (ulong)PerformanceManagerGeneric<PerformanceFrameHeaderVersion2, PerformanceEntryVersion2, PerformanceDetailVersion2>.GetRequiredBufferSizeForPerformanceMetricsPerFrame(ref parameter);
             }
 
             if (version == 1)
             {
-                return (ulong)PerformanceManagerGeneric<PerformanceFrameHeaderVersion1,
-                    PerformanceEntryVersion1,
-                    PerformanceDetailVersion1>.GetRequiredBufferSizeForPerformanceMetricsPerFrame(ref parameter);
+                return (ulong)PerformanceManagerGeneric<PerformanceFrameHeaderVersion1, PerformanceEntryVersion1, PerformanceDetailVersion1>.GetRequiredBufferSizeForPerformanceMetricsPerFrame(ref parameter);
             }
 
             throw new NotImplementedException($"Unknown Performance metrics data format version {version}");

src/Ryujinx.Audio/Renderer/Server/Performance/PerformanceManagerGeneric.cs

@@ -234,7 +234,7 @@ namespace Ryujinx.Audio.Renderer.Server.Performance
         {
             performanceEntry = null;
 
-            if (_entryDetailIndex > MaxFrameDetailCount)
+            if (_entryDetailIndex >= MaxFrameDetailCount)
             {
                 return false;
             }
@@ -245,7 +245,7 @@ namespace Ryujinx.Audio.Renderer.Server.Performance
                 EntryCountOffset = (uint)CurrentHeader.GetEntryCountOffset(),
             };
 
-            uint baseEntryOffset = (uint)(Unsafe.SizeOf<THeader>() + GetEntriesSize() + Unsafe.SizeOf<IPerformanceDetailEntry>() * _entryDetailIndex);
+            uint baseEntryOffset = (uint)(Unsafe.SizeOf<THeader>() + GetEntriesSize() + Unsafe.SizeOf<TEntryDetail>() * _entryDetailIndex);
 
             ref TEntryDetail entryDetail = ref EntriesDetail[_entryDetailIndex];
 

src/Ryujinx.Common/Memory/ByteMemoryPool.ByteMemoryPoolBuffer.cs

@@ -1,51 +0,0 @@
-using System;
-using System.Buffers;
-using System.Threading;
-
-namespace Ryujinx.Common.Memory
-{
-    public partial class ByteMemoryPool
-    {
-        /// <summary>
-        /// Represents a <see cref="IMemoryOwner{Byte}"/> that wraps an array rented from
-        /// <see cref="ArrayPool{Byte}.Shared"/> and exposes it as <see cref="Memory{Byte}"/>
-        /// with a length of the requested size.
-        /// </summary>
-        private sealed class ByteMemoryPoolBuffer : IMemoryOwner<byte>
-        {
-            private byte[] _array;
-            private readonly int _length;
-
-            public ByteMemoryPoolBuffer(int length)
-            {
-                _array = ArrayPool<byte>.Shared.Rent(length);
-                _length = length;
-            }
-
-            /// <summary>
-            /// Returns a <see cref="Memory{Byte}"/> belonging to this owner.
-            /// </summary>
-            public Memory<byte> Memory
-            {
-                get
-                {
-                    byte[] array = _array;
-
-                    ObjectDisposedException.ThrowIf(array is null, this);
-
-                    return new Memory<byte>(array, 0, _length);
-                }
-            }
-
-            public void Dispose()
-            {
-                var array = Interlocked.Exchange(ref _array, null);
-
-                if (array != null)
-                {
-                    ArrayPool<byte>.Shared.Return(array);
-                }
-            }
-        }
-    }
-}

src/Ryujinx.Common/Memory/ByteMemoryPool.cs

@@ -1,106 +0,0 @@
-using System;
-using System.Buffers;
-
-namespace Ryujinx.Common.Memory
-{
-    /// <summary>
-    /// Provides a pool of re-usable byte array instances.
-    /// </summary>
-    public static partial class ByteMemoryPool
-    {
-        /// <summary>
-        /// Returns the maximum buffer size supported by this pool.
-        /// </summary>
-        public static int MaxBufferSize => Array.MaxLength;
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer may contain data from a prior use.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> Rent(long length)
-            => RentImpl(checked((int)length));
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer may contain data from a prior use.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> Rent(ulong length)
-            => RentImpl(checked((int)length));
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer may contain data from a prior use.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> Rent(int length)
-            => RentImpl(length);
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer's contents are cleared (set to all 0s) before returning.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> RentCleared(long length)
-            => RentCleared(checked((int)length));
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer's contents are cleared (set to all 0s) before returning.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> RentCleared(ulong length)
-            => RentCleared(checked((int)length));
-
-        /// <summary>
-        /// Rents a byte memory buffer from <see cref="ArrayPool{Byte}.Shared"/>.
-        /// The buffer's contents are cleared (set to all 0s) before returning.
-        /// </summary>
-        /// <param name="length">The buffer's required length in bytes</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory</returns>
-        /// <exception cref="ArgumentOutOfRangeException"></exception>
-        public static IMemoryOwner<byte> RentCleared(int length)
-        {
-            var buffer = RentImpl(length);
-
-            buffer.Memory.Span.Clear();
-
-            return buffer;
-        }
-
-        /// <summary>
-        /// Copies <paramref name="buffer"/> into a newly rented byte memory buffer.
-        /// </summary>
-        /// <param name="buffer">The byte buffer to copy</param>
-        /// <returns>A <see cref="IMemoryOwner{Byte}"/> wrapping the rented memory with <paramref name="buffer"/> copied to it</returns>
-        public static IMemoryOwner<byte> RentCopy(ReadOnlySpan<byte> buffer)
-        {
-            var copy = RentImpl(buffer.Length);
-
-            buffer.CopyTo(copy.Memory.Span);
-
-            return copy;
-        }
-
-        private static ByteMemoryPoolBuffer RentImpl(int length)
-        {
-            if ((uint)length > Array.MaxLength)
-            {
-                throw new ArgumentOutOfRangeException(nameof(length), length, null);
-            }
-
-            return new ByteMemoryPoolBuffer(length);
-        }
-    }
-}

src/Ryujinx.Common/Memory/MemoryOwner.cs

@@ -0,0 +1,140 @@
+#nullable enable
+using System;
+using System.Buffers;
+using System.Diagnostics.CodeAnalysis;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+using System.Threading;
+
+namespace Ryujinx.Common.Memory
+{
+    /// <summary>
+    /// An <see cref="IMemoryOwner{T}"/> implementation with an embedded length and fast <see cref="Span{T}"/>
+    /// accessor, with memory allocated from <seealso cref="ArrayPool{T}.Shared"/>.
+    /// </summary>
+    /// <typeparam name="T">The type of item to store.</typeparam>
+    public sealed class MemoryOwner<T> : IMemoryOwner<T>
+    {
+        private readonly int _length;
+        private T[]? _array;
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="MemoryOwner{T}"/> class with the specified parameters.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        private MemoryOwner(int length)
+        {
+            _length = length;
+            _array = ArrayPool<T>.Shared.Rent(length);
+        }
+
+        /// <summary>
+        /// Creates a new <see cref="MemoryOwner{T}"/> instance with the specified length.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        /// <returns>A <see cref="MemoryOwner{T}"/> instance of the requested length</returns>
+        /// <exception cref="ArgumentOutOfRangeException">Thrown when <paramref name="length"/> is not valid</exception>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static MemoryOwner<T> Rent(int length) => new(length);
+
+        /// <summary>
+        /// Creates a new <see cref="MemoryOwner{T}"/> instance with the specified length and the content cleared.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        /// <returns>A <see cref="MemoryOwner{T}"/> instance of the requested length and the content cleared</returns>
+        /// <exception cref="ArgumentOutOfRangeException">Thrown when <paramref name="length"/> is not valid</exception>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static MemoryOwner<T> RentCleared(int length)
+        {
+            MemoryOwner<T> result = new(length);
+
+            result._array.AsSpan(0, length).Clear();
+
+            return result;
+        }
+
+        /// <summary>
+        /// Creates a new <see cref="MemoryOwner{T}"/> instance with the content copied from the specified buffer.
+        /// </summary>
+        /// <param name="buffer">The buffer to copy</param>
+        /// <returns>A <see cref="MemoryOwner{T}"/> instance with the same length and content as <paramref name="buffer"/></returns>
+        public static MemoryOwner<T> RentCopy(ReadOnlySpan<T> buffer)
+        {
+            MemoryOwner<T> result = new(buffer.Length);
+
+            buffer.CopyTo(result._array);
+
+            return result;
+        }
+
+        /// <summary>
+        /// Gets the number of items in the current instance.
+        /// </summary>
+        public int Length
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get => _length;
+        }
+
+        /// <inheritdoc/>
+        public Memory<T> Memory
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get
+            {
+                T[]? array = _array;
+
+                if (array is null)
+                {
+                    ThrowObjectDisposedException();
+                }
+
+                return new(array, 0, _length);
+            }
+        }
+
+        /// <summary>
+        /// Gets a <see cref="Span{T}"/> wrapping the memory belonging to the current instance.
+        /// </summary>
+        /// <remarks>
+        /// Uses a trick made possible by the .NET 6+ runtime array layout.
+        /// </remarks>
+        public Span<T> Span
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get
+            {
+                T[]? array = _array;
+
+                if (array is null)
+                {
+                    ThrowObjectDisposedException();
+                }
+
+                ref T firstElementRef = ref MemoryMarshal.GetArrayDataReference(array);
+
+                return MemoryMarshal.CreateSpan(ref firstElementRef, _length);
+            }
+        }
+
+        /// <inheritdoc/>
+        public void Dispose()
+        {
+            T[]? array = Interlocked.Exchange(ref _array, null);
+
+            if (array is not null)
+            {
+                ArrayPool<T>.Shared.Return(array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
+            }
+        }
+
+        /// <summary>
+        /// Throws an <see cref="ObjectDisposedException"/> when <see cref="_array"/> is <see langword="null"/>.
+        /// </summary>
+        [DoesNotReturn]
+        private static void ThrowObjectDisposedException()
+        {
+            throw new ObjectDisposedException(nameof(MemoryOwner<T>), "The buffer has already been disposed.");
+        }
+    }
+}

src/Ryujinx.Common/Memory/SpanOwner.cs

@@ -0,0 +1,114 @@
+using System;
+using System.Buffers;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Common.Memory
+{
+    /// <summary>
+    /// A stack-only type that rents a buffer of a specified length from <seealso cref="ArrayPool{T}.Shared"/>.
+    /// It does not implement <see cref="IDisposable"/> to avoid being boxed, but should still be disposed. This
+    /// is easy since C# 8, which allows use of C# `using` constructs on any type that has a public Dispose() method.
+    /// To keep this type simple, fast, and read-only, it does not check or guard against multiple disposals.
+    /// For all these reasons, all usage should be with a `using` block or statement.
+    /// </summary>
+    /// <typeparam name="T">The type of item to store.</typeparam>
+    public readonly ref struct SpanOwner<T>
+    {
+        private readonly int _length;
+        private readonly T[] _array;
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="SpanOwner{T}"/> struct with the specified parameters.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        private SpanOwner(int length)
+        {
+            _length = length;
+            _array = ArrayPool<T>.Shared.Rent(length);
+        }
+
+        /// <summary>
+        /// Gets an empty <see cref="SpanOwner{T}"/> instance.
+        /// </summary>
+        public static SpanOwner<T> Empty
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get => new(0);
+        }
+
+        /// <summary>
+        /// Creates a new <see cref="SpanOwner{T}"/> instance with the specified length.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        /// <returns>A <see cref="SpanOwner{T}"/> instance of the requested length</returns>
+        /// <exception cref="ArgumentOutOfRangeException">Thrown when <paramref name="length"/> is not valid</exception>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static SpanOwner<T> Rent(int length) => new(length);
+
+        /// <summary>
+        /// Creates a new <see cref="SpanOwner{T}"/> instance with the length and the content cleared.
+        /// </summary>
+        /// <param name="length">The length of the new memory buffer to use</param>
+        /// <returns>A <see cref="SpanOwner{T}"/> instance of the requested length and the content cleared</returns>
+        /// <exception cref="ArgumentOutOfRangeException">Thrown when <paramref name="length"/> is not valid</exception>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static SpanOwner<T> RentCleared(int length)
+        {
+            SpanOwner<T> result = new(length);
+
+            result._array.AsSpan(0, length).Clear();
+
+            return result;
+        }
+
+        /// <summary>
+        /// Creates a new <see cref="SpanOwner{T}"/> instance with the content copied from the specified buffer.
+        /// </summary>
+        /// <param name="buffer">The buffer to copy</param>
+        /// <returns>A <see cref="SpanOwner{T}"/> instance with the same length and content as <paramref name="buffer"/></returns>
+        public static SpanOwner<T> RentCopy(ReadOnlySpan<T> buffer)
+        {
+            SpanOwner<T> result = new(buffer.Length);
+
+            buffer.CopyTo(result._array);
+
+            return result;
+        }
+
+        /// <summary>
+        /// Gets the number of items in the current instance
+        /// </summary>
+        public int Length
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get => _length;
+        }
+
+        /// <summary>
+        /// Gets a <see cref="Span{T}"/> wrapping the memory belonging to the current instance.
+        /// </summary>
+        /// <remarks>
+        /// Uses a trick made possible by the .NET 6+ runtime array layout.
+        /// </remarks>
+        public Span<T> Span
+        {
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get
+            {
+                ref T firstElementRef = ref MemoryMarshal.GetArrayDataReference(_array);
+
+                return MemoryMarshal.CreateSpan(ref firstElementRef, _length);
+            }
+        }
+
+        /// <summary>
+        /// Implements the duck-typed <see cref="IDisposable.Dispose"/> method.
+        /// </summary>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public void Dispose()
+        {
+            ArrayPool<T>.Shared.Return(_array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
+        }
+    }
+}

src/Ryujinx.Common/Utilities/EmbeddedResources.cs

@@ -1,6 +1,6 @@
+using Ryujinx.Common.Memory;
 using Ryujinx.Common.Utilities;
 using System;
-using System.Buffers;
 using System.IO;
 using System.Linq;
 using System.Reflection;
@@ -42,14 +42,14 @@ namespace Ryujinx.Common
             return StreamUtils.StreamToBytes(stream);
         }
 
-        public static IMemoryOwner<byte> ReadFileToRentedMemory(string filename)
+        public static MemoryOwner<byte> ReadFileToRentedMemory(string filename)
         {
             var (assembly, path) = ResolveManifestPath(filename);
 
             return ReadFileToRentedMemory(assembly, path);
         }
 
-        public static IMemoryOwner<byte> ReadFileToRentedMemory(Assembly assembly, string filename)
+        public static MemoryOwner<byte> ReadFileToRentedMemory(Assembly assembly, string filename)
         {
             using var stream = GetStream(assembly, filename);
 

src/Ryujinx.Common/Utilities/StreamUtils.cs

@@ -1,6 +1,5 @@
 using Microsoft.IO;
 using Ryujinx.Common.Memory;
-using System.Buffers;
 using System.IO;
 using System.Threading;
 using System.Threading.Tasks;
@@ -16,7 +15,7 @@ namespace Ryujinx.Common.Utilities
             return output.ToArray();
         }
 
-        public static IMemoryOwner<byte> StreamToRentedMemory(Stream input)
+        public static MemoryOwner<byte> StreamToRentedMemory(Stream input)
         {
             if (input is MemoryStream inputMemoryStream)
             {
@@ -26,9 +25,9 @@ namespace Ryujinx.Common.Utilities
             {
                 long bytesExpected = input.Length;
 
-                IMemoryOwner<byte> ownedMemory = ByteMemoryPool.Rent(bytesExpected);
+                MemoryOwner<byte> ownedMemory = MemoryOwner<byte>.Rent(checked((int)bytesExpected));
 
-                var destSpan = ownedMemory.Memory.Span;
+                var destSpan = ownedMemory.Span;
 
                 int totalBytesRead = 0;
 
@@ -66,14 +65,14 @@ namespace Ryujinx.Common.Utilities
             return stream.ToArray();
         }
 
-        private static IMemoryOwner<byte> MemoryStreamToRentedMemory(MemoryStream input)
+        private static MemoryOwner<byte> MemoryStreamToRentedMemory(MemoryStream input)
         {
             input.Position = 0;
 
-            IMemoryOwner<byte> ownedMemory = ByteMemoryPool.Rent(input.Length);
+            MemoryOwner<byte> ownedMemory = MemoryOwner<byte>.Rent(checked((int)input.Length));
 
             // Discard the return value because we assume reading a MemoryStream always succeeds completely.
-            _ = input.Read(ownedMemory.Memory.Span);
+            _ = input.Read(ownedMemory.Span);
 
             return ownedMemory;
         }

src/Ryujinx.Cpu/Jit/MemoryManagerHostTracked.cs

@@ -303,9 +303,9 @@ namespace Ryujinx.Cpu.Jit
             }
             else
             {
-                IMemoryOwner<byte> memoryOwner = ByteMemoryPool.Rent(size);
+                MemoryOwner<byte> memoryOwner = MemoryOwner<byte>.Rent(size);
 
-                Read(va, memoryOwner.Memory.Span);
+                Read(va, memoryOwner.Span);
 
                 return new WritableRegion(this, va, memoryOwner);
             }

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

@@ -1,6 +1,5 @@
 using Ryujinx.Cpu.LightningJit.CodeGen;
 using Ryujinx.Cpu.LightningJit.CodeGen.Arm64;
-using System.Diagnostics;
 
 namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
 {

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

@@ -114,7 +114,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
             InstEmitCommon.EmitUnsigned16BitPair(context, rd, rn, rm, (d, n, m) =>
             {
                 context.Arm64Assembler.Add(d, n, m);
-                EmitSaturateUnsignedRange(context, d, 16);
+                EmitSaturateUqadd(context, d, 16);
             });
         }
 
@@ -123,7 +123,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
             InstEmitCommon.EmitUnsigned8BitPair(context, rd, rn, rm, (d, n, m) =>
             {
                 context.Arm64Assembler.Add(d, n, m);
-                EmitSaturateUnsignedRange(context, d, 8);
+                EmitSaturateUqadd(context, d, 8);
             });
         }
 
@@ -140,7 +140,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
                     context.Arm64Assembler.Add(d, n, m);
                 }
 
-                EmitSaturateUnsignedRange(context, d, 16);
+                EmitSaturateUq(context, d, 16, e == 0);
             });
         }
 
@@ -157,25 +157,25 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
                     context.Arm64Assembler.Sub(d, n, m);
                 }
 
-                EmitSaturateUnsignedRange(context, d, 16);
+                EmitSaturateUq(context, d, 16, e != 0);
             });
         }
 
         public static void Uqsub16(CodeGenContext context, uint rd, uint rn, uint rm)
         {
-            InstEmitCommon.EmitSigned16BitPair(context, rd, rn, rm, (d, n, m) =>
+            InstEmitCommon.EmitUnsigned16BitPair(context, rd, rn, rm, (d, n, m) =>
             {
                 context.Arm64Assembler.Sub(d, n, m);
-                EmitSaturateUnsignedRange(context, d, 16);
+                EmitSaturateUqsub(context, d, 16);
             });
         }
 
         public static void Uqsub8(CodeGenContext context, uint rd, uint rn, uint rm)
         {
-            InstEmitCommon.EmitSigned8BitPair(context, rd, rn, rm, (d, n, m) =>
+            InstEmitCommon.EmitUnsigned8BitPair(context, rd, rn, rm, (d, n, m) =>
             {
                 context.Arm64Assembler.Sub(d, n, m);
-                EmitSaturateUnsignedRange(context, d, 8);
+                EmitSaturateUqsub(context, d, 8);
             });
         }
 
@@ -358,7 +358,17 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
             }
         }
 
-        private static void EmitSaturateUnsignedRange(CodeGenContext context, Operand value, uint saturateTo)
+        private static void EmitSaturateUqadd(CodeGenContext context, Operand value, uint saturateTo)
+        {
+            EmitSaturateUq(context, value, saturateTo, isSub: false);
+        }
+
+        private static void EmitSaturateUqsub(CodeGenContext context, Operand value, uint saturateTo)
+        {
+            EmitSaturateUq(context, value, saturateTo, isSub: true);
+        }
+
+        private static void EmitSaturateUq(CodeGenContext context, Operand value, uint saturateTo, bool isSub)
         {
             Debug.Assert(saturateTo <= 32);
 
@@ -379,7 +389,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
                 return;
             }
 
-            context.Arm64Assembler.Lsr(tempRegister.Operand, value, InstEmitCommon.Const(32 - (int)saturateTo));
+            context.Arm64Assembler.Lsr(tempRegister.Operand, value, InstEmitCommon.Const((int)saturateTo));
 
             int branchIndex = context.CodeWriter.InstructionPointer;
 
@@ -387,7 +397,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
             context.Arm64Assembler.Cbz(tempRegister.Operand, 0);
 
             // Saturate.
-            context.Arm64Assembler.Mov(value, uint.MaxValue >> (32 - (int)saturateTo));
+            context.Arm64Assembler.Mov(value, isSub ? 0u : uint.MaxValue >> (32 - (int)saturateTo));
 
             int delta = context.CodeWriter.InstructionPointer - branchIndex;
             context.CodeWriter.WriteInstructionAt(branchIndex, context.CodeWriter.ReadInstructionAt(branchIndex) | (uint)((delta & 0x7ffff) << 5));

src/Ryujinx.Graphics.Device/DeviceMemoryManager.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common.Memory;
 using Ryujinx.Memory;
 using System;
-using System.Buffers;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
@@ -145,9 +144,9 @@ namespace Ryujinx.Graphics.Device
             }
             else
             {
-                IMemoryOwner<byte> memoryOwner = ByteMemoryPool.Rent(size);
+                MemoryOwner<byte> memoryOwner = MemoryOwner<byte>.Rent(size);
 
-                GetSpan(va, size).CopyTo(memoryOwner.Memory.Span);
+                ReadImpl(va, memoryOwner.Span);
 
                 return new WritableRegion(this, va, memoryOwner, tracked: true);
             }

src/Ryujinx.Graphics.GAL/IImageArray.cs

@@ -1,6 +1,8 @@
+using System;
+
 namespace Ryujinx.Graphics.GAL
 {
-    public interface IImageArray
+    public interface IImageArray : IDisposable
     {
         void SetFormats(int index, Format[] imageFormats);
         void SetImages(int index, ITexture[] images);

src/Ryujinx.Graphics.GAL/ITextureArray.cs

@@ -1,6 +1,8 @@
+using System;
+
 namespace Ryujinx.Graphics.GAL
 {
-    public interface ITextureArray
+    public interface ITextureArray : IDisposable
     {
         void SetSamplers(int index, ISampler[] samplers);
         void SetTextures(int index, ITexture[] textures);

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

@@ -66,6 +66,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             Register<CounterEventDisposeCommand>(CommandType.CounterEventDispose);
             Register<CounterEventFlushCommand>(CommandType.CounterEventFlush);
 
+            Register<ImageArrayDisposeCommand>(CommandType.ImageArrayDispose);
             Register<ImageArraySetFormatsCommand>(CommandType.ImageArraySetFormats);
             Register<ImageArraySetImagesCommand>(CommandType.ImageArraySetImages);
 
@@ -88,6 +89,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             Register<TextureSetDataSliceRegionCommand>(CommandType.TextureSetDataSliceRegion);
             Register<TextureSetStorageCommand>(CommandType.TextureSetStorage);
 
+            Register<TextureArrayDisposeCommand>(CommandType.TextureArrayDispose);
             Register<TextureArraySetSamplersCommand>(CommandType.TextureArraySetSamplers);
             Register<TextureArraySetTexturesCommand>(CommandType.TextureArraySetTextures);
 

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

@@ -26,6 +26,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
         CounterEventDispose,
         CounterEventFlush,
 
+        ImageArrayDispose,
         ImageArraySetFormats,
         ImageArraySetImages,
 
@@ -48,6 +49,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
         TextureSetDataSliceRegion,
         TextureSetStorage,
 
+        TextureArrayDispose,
         TextureArraySetSamplers,
         TextureArraySetTextures,
 

src/Ryujinx.Graphics.GAL/Multithreading/Commands/ImageArray/ImageArrayDisposeCommand.cs

@@ -0,0 +1,21 @@
+using Ryujinx.Graphics.GAL.Multithreading.Model;
+using Ryujinx.Graphics.GAL.Multithreading.Resources;
+
+namespace Ryujinx.Graphics.GAL.Multithreading.Commands.ImageArray
+{
+    struct ImageArrayDisposeCommand : IGALCommand, IGALCommand<ImageArrayDisposeCommand>
+    {
+        public readonly CommandType CommandType => CommandType.ImageArrayDispose;
+        private TableRef<ThreadedImageArray> _imageArray;
+
+        public void Set(TableRef<ThreadedImageArray> imageArray)
+        {
+            _imageArray = imageArray;
+        }
+
+        public static void Run(ref ImageArrayDisposeCommand command, ThreadedRenderer threaded, IRenderer renderer)
+        {
+            command._imageArray.Get(threaded).Base.Dispose();
+        }
+    }
+}

src/Ryujinx.Graphics.GAL/Multithreading/Commands/TextureArray/TextureArrayDisposeCommand.cs

@@ -0,0 +1,21 @@
+using Ryujinx.Graphics.GAL.Multithreading.Model;
+using Ryujinx.Graphics.GAL.Multithreading.Resources;
+
+namespace Ryujinx.Graphics.GAL.Multithreading.Commands.TextureArray
+{
+    struct TextureArrayDisposeCommand : IGALCommand, IGALCommand<TextureArrayDisposeCommand>
+    {
+        public readonly CommandType CommandType => CommandType.TextureArrayDispose;
+        private TableRef<ThreadedTextureArray> _textureArray;
+
+        public void Set(TableRef<ThreadedTextureArray> textureArray)
+        {
+            _textureArray = textureArray;
+        }
+
+        public static void Run(ref TextureArrayDisposeCommand command, ThreadedRenderer threaded, IRenderer renderer)
+        {
+            command._textureArray.Get(threaded).Base.Dispose();
+        }
+    }
+}

src/Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedImageArray.cs

@@ -21,6 +21,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             return new TableRef<T>(_renderer, reference);
         }
 
+        public void Dispose()
+        {
+            _renderer.New<ImageArrayDisposeCommand>().Set(Ref(this));
+            _renderer.QueueCommand();
+        }
+
         public void SetFormats(int index, Format[] imageFormats)
         {
             _renderer.New<ImageArraySetFormatsCommand>().Set(Ref(this), index, Ref(imageFormats));

src/Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedTextureArray.cs

@@ -22,6 +22,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             return new TableRef<T>(_renderer, reference);
         }
 
+        public void Dispose()
+        {
+            _renderer.New<TextureArrayDisposeCommand>().Set(Ref(this));
+            _renderer.QueueCommand();
+        }
+
         public void SetSamplers(int index, ISampler[] samplers)
         {
             _renderer.New<TextureArraySetSamplersCommand>().Set(Ref(this), index, Ref(samplers.ToArray()));

src/Ryujinx.Graphics.GAL/ResourceLayout.cs

@@ -74,13 +74,15 @@ namespace Ryujinx.Graphics.GAL
         public int ArrayLength { get; }
         public ResourceType Type { get; }
         public ResourceStages Stages { get; }
+        public bool Write { get; }
 
-        public ResourceUsage(int binding, int arrayLength, ResourceType type, ResourceStages stages)
+        public ResourceUsage(int binding, int arrayLength, ResourceType type, ResourceStages stages, bool write)
         {
             Binding = binding;
             ArrayLength = arrayLength;
             Type = type;
             Stages = stages;
+            Write = write;
         }
 
         public override int GetHashCode()

src/Ryujinx.Graphics.GAL/TextureCreateInfo.cs

@@ -1,6 +1,5 @@
 using Ryujinx.Common;
 using System;
-using System.Numerics;
 
 namespace Ryujinx.Graphics.GAL
 {
@@ -113,25 +112,6 @@ namespace Ryujinx.Graphics.GAL
             return 1;
         }
 
-        public int GetLevelsClamped()
-        {
-            int maxSize = Width;
-
-            if (Target != Target.Texture1D &&
-                Target != Target.Texture1DArray)
-            {
-                maxSize = Math.Max(maxSize, Height);
-            }
-
-            if (Target == Target.Texture3D)
-            {
-                maxSize = Math.Max(maxSize, Depth);
-            }
-
-            int maxLevels = BitOperations.Log2((uint)maxSize) + 1;
-            return Math.Min(Levels, maxLevels);
-        }
-
         private static int GetLevelSize(int size, int level)
         {
             return Math.Max(1, size >> level);

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

@@ -199,7 +199,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
                     if (target != null)
                     {
                         target.SynchronizeMemory();
-                        var dataCopy = ByteMemoryPool.RentCopy(data);
+                        var dataCopy = MemoryOwner<byte>.RentCopy(data);
                         target.SetData(dataCopy, 0, 0, new GAL.Rectangle<int>(_dstX, _dstY, _lineLengthIn / target.Info.FormatInfo.BytesPerPixel, _lineCount));
                         target.SignalModified();
 

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

@@ -1,4 +1,5 @@
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Texture;
@@ -805,7 +806,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                         sliceDepth,
                         levels,
                         layers,
-                        out IMemoryOwner<byte> decoded))
+                        out MemoryOwner<byte> decoded))
                     {
                         string texInfo = $"{Info.Target} {Info.FormatInfo.Format} {Info.Width}x{Info.Height}x{Info.DepthOrLayers} levels {Info.Levels}";
 

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

@@ -1113,6 +1113,15 @@ namespace Ryujinx.Graphics.Gpu.Image
                 nextNode = nextNode.Next;
                 _cacheFromBuffer.Remove(toRemove.Value.Key);
                 _lruCache.Remove(toRemove);
+
+                if (toRemove.Value.Key.IsImage)
+                {
+                    toRemove.Value.ImageArray.Dispose();
+                }
+                else
+                {
+                    toRemove.Value.TextureArray.Dispose();
+                }
             }
         }
 
@@ -1124,11 +1133,20 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             List<CacheEntryFromPoolKey> keysToRemove = null;
 
-            foreach (CacheEntryFromPoolKey key in _cacheFromPool.Keys)
+            foreach ((CacheEntryFromPoolKey key, CacheEntry entry) in _cacheFromPool)
             {
                 if (key.MatchesPool(pool))
                 {
                     (keysToRemove ??= new()).Add(key);
+
+                    if (key.IsImage)
+                    {
+                        entry.ImageArray.Dispose();
+                    }
+                    else
+                    {
+                        entry.TextureArray.Dispose();
+                    }
                 }
             }
 

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

@@ -6,6 +6,7 @@ using Ryujinx.Memory.Range;
 using System;
 using System.Collections.Concurrent;
 using System.Collections.Generic;
+using System.Numerics;
 using System.Threading;
 
 namespace Ryujinx.Graphics.Gpu.Image
@@ -490,6 +491,8 @@ namespace Ryujinx.Graphics.Gpu.Image
                 levels = (maxLod - minLod) + 1;
             }
 
+            levels = ClampLevels(target, width, height, depthOrLayers, levels);
+
             SwizzleComponent swizzleR = descriptor.UnpackSwizzleR().Convert();
             SwizzleComponent swizzleG = descriptor.UnpackSwizzleG().Convert();
             SwizzleComponent swizzleB = descriptor.UnpackSwizzleB().Convert();
@@ -540,6 +543,34 @@ namespace Ryujinx.Graphics.Gpu.Image
                 swizzleA);
         }
 
+        /// <summary>
+        /// Clamps the amount of mipmap levels to the maximum allowed for the given texture dimensions.
+        /// </summary>
+        /// <param name="target">Number of texture dimensions (1D, 2D, 3D, Cube, etc)</param>
+        /// <param name="width">Width of the texture</param>
+        /// <param name="height">Height of the texture, ignored for 1D textures</param>
+        /// <param name="depthOrLayers">Depth of the texture for 3D textures, otherwise ignored</param>
+        /// <param name="levels">Original amount of mipmap levels</param>
+        /// <returns>Clamped mipmap levels</returns>
+        private static int ClampLevels(Target target, int width, int height, int depthOrLayers, int levels)
+        {
+            int maxSize = width;
+
+            if (target != Target.Texture1D &&
+                target != Target.Texture1DArray)
+            {
+                maxSize = Math.Max(maxSize, height);
+            }
+
+            if (target == Target.Texture3D)
+            {
+                maxSize = Math.Max(maxSize, depthOrLayers);
+            }
+
+            int maxLevels = BitOperations.Log2((uint)maxSize) + 1;
+            return Math.Min(levels, maxLevels);
+        }
+
         /// <summary>
         /// Gets the texture depth-stencil mode, based on the swizzle components of each color channel.
         /// The depth-stencil mode is determined based on how the driver sets those parameters.

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

@@ -2,7 +2,6 @@ using Ryujinx.Common.Memory;
 using Ryujinx.Memory;
 using Ryujinx.Memory.Range;
 using System;
-using System.Buffers;
 using System.Collections.Generic;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
@@ -242,9 +241,9 @@ namespace Ryujinx.Graphics.Gpu.Memory
             }
             else
             {
-                IMemoryOwner<byte> memoryOwner = ByteMemoryPool.Rent(size);
+                MemoryOwner<byte> memoryOwner = MemoryOwner<byte>.Rent(size);
 
-                GetSpan(va, size).CopyTo(memoryOwner.Memory.Span);
+                ReadImpl(va, memoryOwner.Span, tracked);
 
                 return new WritableRegion(this, va, memoryOwner, tracked);
             }

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

@@ -192,9 +192,9 @@ namespace Ryujinx.Graphics.Gpu.Memory
             }
             else
             {
-                IMemoryOwner<byte> memoryOwner = ByteMemoryPool.Rent(range.GetSize());
+                MemoryOwner<byte> memoryOwner = MemoryOwner<byte>.Rent(checked((int)range.GetSize()));
 
-                Memory<byte> memory = memoryOwner.Memory;
+                Span<byte> memorySpan = memoryOwner.Span;
 
                 int offset = 0;
                 for (int i = 0; i < range.Count; i++)
@@ -203,7 +203,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
                     int size = (int)currentRange.Size;
                     if (currentRange.Address != MemoryManager.PteUnmapped)
                     {
-                        GetSpan(currentRange.Address, size).CopyTo(memory.Span.Slice(offset, size));
+                        GetSpan(currentRange.Address, size).CopyTo(memorySpan.Slice(offset, size));
                     }
                     offset += size;
                 }

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

@@ -125,9 +125,18 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             CompressionAlgorithm algorithm = CompressionAlgorithm.None;
             Read(ref algorithm);
 
-            if (algorithm == CompressionAlgorithm.Deflate)
+            switch (algorithm)
             {
-                _activeStream = new DeflateStream(_stream, CompressionMode.Decompress, true);
+                case CompressionAlgorithm.None:
+                    break;
+                case CompressionAlgorithm.Deflate:
+                    _activeStream = new DeflateStream(_stream, CompressionMode.Decompress, true);
+                    break;
+                case CompressionAlgorithm.Brotli:
+                    _activeStream = new BrotliStream(_stream, CompressionMode.Decompress, true);
+                    break;
+                default:
+                    throw new ArgumentException($"Invalid compression algorithm \"{algorithm}\"");
             }
         }
 
@@ -139,9 +148,18 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         {
             Write(ref algorithm);
 
-            if (algorithm == CompressionAlgorithm.Deflate)
+            switch (algorithm)
             {
-                _activeStream = new DeflateStream(_stream, CompressionLevel.Fastest, true);
+                case CompressionAlgorithm.None:
+                    break;
+                case CompressionAlgorithm.Deflate:
+                    _activeStream = new DeflateStream(_stream, CompressionLevel.Fastest, true);
+                    break;
+                case CompressionAlgorithm.Brotli:
+                    _activeStream = new BrotliStream(_stream, CompressionLevel.Fastest, true);
+                    break;
+                default:
+                    throw new ArgumentException($"Invalid compression algorithm \"{algorithm}\"");
             }
         }
 
@@ -177,7 +195,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             switch (algorithm)
             {
                 case CompressionAlgorithm.None:
-                    stream.Read(data);
+                    stream.ReadExactly(data);
                     break;
                 case CompressionAlgorithm.Deflate:
                     stream = new DeflateStream(stream, CompressionMode.Decompress, true);
@@ -187,6 +205,14 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                     }
                     stream.Dispose();
                     break;
+                case CompressionAlgorithm.Brotli:
+                    stream = new BrotliStream(stream, CompressionMode.Decompress, true);
+                    for (int offset = 0; offset < data.Length;)
+                    {
+                        offset += stream.Read(data[offset..]);
+                    }
+                    stream.Dispose();
+                    break;
             }
         }
 
@@ -210,6 +236,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                     stream.Write(data);
                     stream.Dispose();
                     break;
+                case CompressionAlgorithm.Brotli:
+                    stream = new BrotliStream(stream, CompressionLevel.Fastest, true);
+                    stream.Write(data);
+                    stream.Dispose();
+                    break;
             }
         }
     }

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

@@ -14,5 +14,10 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         /// Deflate compression (RFC 1951).
         /// </summary>
         Deflate,
+
+        /// <summary>
+        /// Brotli compression (RFC 7932).
+        /// </summary>
+        Brotli,
     }
 }

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

@@ -51,7 +51,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         /// <returns>Compression algorithm</returns>
         public static CompressionAlgorithm GetCompressionAlgorithm()
         {
-            return CompressionAlgorithm.Deflate;
+            return CompressionAlgorithm.Brotli;
         }
     }
 }

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

@@ -220,7 +220,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 }
 
                 dataFileStream.Seek((long)entry.Offset, SeekOrigin.Begin);
-                dataFileStream.Read(cb1Data);
+                dataFileStream.ReadExactly(cb1Data);
                 BinarySerializer.ReadCompressed(dataFileStream, guestCode);
 
                 _cache[index] = (guestCode, cb1Data);
@@ -279,7 +279,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                     dataFileStream.Seek((long)entry.Offset, SeekOrigin.Begin);
                     byte[] cachedCode = new byte[entry.CodeSize];
                     byte[] cachedCb1Data = new byte[entry.Cb1DataSize];
-                    dataFileStream.Read(cachedCb1Data);
+                    dataFileStream.ReadExactly(cachedCb1Data);
                     BinarySerializer.ReadCompressed(dataFileStream, cachedCode);
 
                     if (data.SequenceEqual(cachedCode) && cb1Data.SequenceEqual(cachedCb1Data))

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 = 6870;
+        private const uint CodeGenVersion = 7131;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

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

@@ -78,9 +78,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
             ResourceStages stages = vertexAsCompute ? ResourceStages.Compute | ResourceStages.Vertex : VtgStages;
 
             PopulateDescriptorAndUsages(stages, ResourceType.UniformBuffer, uniformSetIndex, 1, rrc.ReservedConstantBuffers - 1);
-            PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, storageSetIndex, 0, rrc.ReservedStorageBuffers);
+            PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, storageSetIndex, 0, rrc.ReservedStorageBuffers, true);
             PopulateDescriptorAndUsages(stages, ResourceType.BufferTexture, textureSetIndex, 0, rrc.ReservedTextures);
-            PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, imageSetIndex, 0, rrc.ReservedImages);
+            PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, imageSetIndex, 0, rrc.ReservedImages, true);
         }
 
         /// <summary>
@@ -91,10 +91,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="setIndex">Resource set index where the resources are used</param>
         /// <param name="start">First binding number</param>
         /// <param name="count">Amount of bindings</param>
-        private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, int setIndex, int start, int count)
+        /// <param name="write">True if the binding is written from the shader, false otherwise</param>
+        private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, int setIndex, int start, int count, bool write = false)
         {
             AddDescriptor(stages, type, setIndex, start, count);
-            AddUsage(stages, type, setIndex, start, count);
+            AddUsage(stages, type, setIndex, start, count, write);
         }
 
         /// <summary>
@@ -216,11 +217,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <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, int setIndex, int binding, int count)
+        /// <param name="write">True if the binding is written from the shader, false otherwise</param>
+        private void AddUsage(ResourceStages stages, ResourceType type, int setIndex, int binding, int count, bool write = false)
         {
             for (int index = 0; index < count; index++)
             {
-                _resourceUsages[setIndex].Add(new ResourceUsage(binding + index, 1, type, stages));
+                _resourceUsages[setIndex].Add(new ResourceUsage(binding + index, 1, type, stages, write));
             }
         }
 
@@ -238,7 +240,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     buffer.Binding,
                     1,
                     isStorage ? ResourceType.StorageBuffer : ResourceType.UniformBuffer,
-                    stages));
+                    stages,
+                    buffer.Flags.HasFlag(BufferUsageFlags.Write)));
             }
         }
 
@@ -254,7 +257,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
             {
                 ResourceType type = GetTextureResourceType(texture, isImage);
 
-                GetUsages(texture.Set).Add(new ResourceUsage(texture.Binding, texture.ArrayLength, type, stages));
+                GetUsages(texture.Set).Add(new ResourceUsage(
+                    texture.Binding,
+                    texture.ArrayLength,
+                    type,
+                    stages,
+                    texture.Flags.HasFlag(TextureUsageFlags.ImageStore)));
             }
         }
 

src/Ryujinx.Graphics.OpenGL/Image/FormatConverter.cs

@@ -1,6 +1,5 @@
 using Ryujinx.Common.Memory;
 using System;
-using System.Buffers;
 using System.Numerics;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics;
@@ -10,11 +9,11 @@ namespace Ryujinx.Graphics.OpenGL.Image
 {
     static class FormatConverter
     {
-        public unsafe static IMemoryOwner<byte> ConvertS8D24ToD24S8(ReadOnlySpan<byte> data)
+        public unsafe static MemoryOwner<byte> ConvertS8D24ToD24S8(ReadOnlySpan<byte> data)
         {
-            IMemoryOwner<byte> outputMemory = ByteMemoryPool.Rent(data.Length);
+            MemoryOwner<byte> outputMemory = MemoryOwner<byte>.Rent(data.Length);
 
-            Span<byte> output = outputMemory.Memory.Span;
+            Span<byte> output = outputMemory.Span;
 
             int start = 0;
 

src/Ryujinx.Graphics.OpenGL/Image/ImageArray.cs

@@ -63,5 +63,9 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 }
             }
         }
+
+        public void Dispose()
+        {
+        }
     }
 }

src/Ryujinx.Graphics.OpenGL/Image/TextureArray.cs

@@ -48,5 +48,9 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 }
             }
         }
+
+        public void Dispose()
+        {
+        }
     }
 }

src/Ryujinx.Graphics.OpenGL/Image/TextureStorage.cs

@@ -48,7 +48,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 internalFormat = (SizedInternalFormat)format.PixelInternalFormat;
             }
 
-            int levels = Info.GetLevelsClamped();
+            int levels = Info.Levels;
 
             switch (Info.Target)
             {

src/Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -51,7 +51,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 pixelInternalFormat = format.PixelInternalFormat;
             }
 
-            int levels = Info.GetLevelsClamped();
+            int levels = Info.Levels;
 
             GL.TextureView(
                 Handle,
@@ -267,7 +267,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
         public unsafe PinnedSpan<byte> GetData()
         {
             int size = 0;
-            int levels = Info.GetLevelsClamped();
+            int levels = Info.Levels;
 
             for (int level = 0; level < levels; level++)
             {
@@ -426,7 +426,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 faces = 6;
             }
 
-            int levels = Info.GetLevelsClamped();
+            int levels = Info.Levels;
 
             for (int level = 0; level < levels; level++)
             {
@@ -716,7 +716,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
             int width = Info.Width;
             int height = Info.Height;
             int depth = Info.Depth;
-            int levels = Info.GetLevelsClamped();
+            int levels = Info.Levels;
 
             int offset = 0;
 

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

@@ -98,11 +98,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Logger = parameters.Logger;
             TargetApi = parameters.TargetApi;
 
-            AddCapability(Capability.Shader);
-            AddCapability(Capability.Float64);
-
-            SetMemoryModel(AddressingModel.Logical, MemoryModel.GLSL450);
-
             Delegates = new SpirvDelegates(this);
         }
 

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

@@ -43,6 +43,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             CodeGenContext context = new(info, parameters, instPool, integerPool);
 
+            context.AddCapability(Capability.Shader);
+
+            context.SetMemoryModel(AddressingModel.Logical, MemoryModel.GLSL450);
+
             context.AddCapability(Capability.GroupNonUniformBallot);
             context.AddCapability(Capability.GroupNonUniformShuffle);
             context.AddCapability(Capability.GroupNonUniformVote);
@@ -51,6 +55,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             context.AddCapability(Capability.ImageQuery);
             context.AddCapability(Capability.SampledBuffer);
 
+            if (parameters.HostCapabilities.SupportsShaderFloat64)
+            {
+                context.AddCapability(Capability.Float64);
+            }
+
             if (parameters.Definitions.TransformFeedbackEnabled && parameters.Definitions.LastInVertexPipeline)
             {
                 context.AddCapability(Capability.TransformFeedback);
@@ -58,7 +67,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             if (parameters.Definitions.Stage == ShaderStage.Fragment)
             {
-                if (context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.Layer)))
+                if (context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.Layer)) ||
+                    context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Input, IoVariable.PrimitiveId)))
                 {
                     context.AddCapability(Capability.Geometry);
                 }

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

@@ -24,7 +24,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             if (op.BVal)
             {
-                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), Const(0)));
+                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), ConstF(0)));
             }
             else
             {

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

@@ -156,6 +156,26 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
             return false;
         }
 
+        public static bool IsComparison(this Instruction inst)
+        {
+            switch (inst & Instruction.Mask)
+            {
+                case Instruction.CompareEqual:
+                case Instruction.CompareGreater:
+                case Instruction.CompareGreaterOrEqual:
+                case Instruction.CompareGreaterOrEqualU32:
+                case Instruction.CompareGreaterU32:
+                case Instruction.CompareLess:
+                case Instruction.CompareLessOrEqual:
+                case Instruction.CompareLessOrEqualU32:
+                case Instruction.CompareLessU32:
+                case Instruction.CompareNotEqual:
+                    return true;
+            }
+
+            return false;
+        }
+
         public static bool IsTextureQuery(this Instruction inst)
         {
             inst &= Instruction.Mask;

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

@@ -8,6 +8,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         public readonly bool SupportsGeometryShaderPassthrough;
         public readonly bool SupportsShaderBallot;
         public readonly bool SupportsShaderBarrierDivergence;
+        public readonly bool SupportsShaderFloat64;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsViewportMask;
 
@@ -18,6 +19,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             bool supportsGeometryShaderPassthrough,
             bool supportsShaderBallot,
             bool supportsShaderBarrierDivergence,
+            bool supportsShaderFloat64,
             bool supportsTextureShadowLod,
             bool supportsViewportMask)
         {
@@ -27,6 +29,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
             SupportsShaderBallot = supportsShaderBallot;
             SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
+            SupportsShaderFloat64 = supportsShaderFloat64;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsViewportMask = supportsViewportMask;
         }

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

@@ -38,6 +38,12 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     // If we can't do bindless elimination, remove the texture operation.
                     // Set any destination variables to zero.
 
+                    string typeName = texOp.Inst.IsImage()
+                        ? texOp.Type.ToGlslImageType(texOp.Format.GetComponentType())
+                        : texOp.Type.ToGlslTextureType();
+
+                    gpuAccessor.Log($"Failed to find handle source for bindless access of type \"{typeName}\".");
+
                     for (int destIndex = 0; destIndex < texOp.DestsCount; destIndex++)
                     {
                         block.Operations.AddBefore(node, new Operation(Instruction.Copy, texOp.GetDest(destIndex), OperandHelper.Const(0)));
@@ -62,17 +68,22 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 return false;
             }
 
-            Operand nvHandle = texOp.GetSource(0);
+            Operand bindlessHandle = texOp.GetSource(0);
 
-            if (nvHandle.AsgOp is not Operation handleOp ||
-                handleOp.Inst != Instruction.Load ||
-                (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
+            if (bindlessHandle.AsgOp is PhiNode phi)
             {
-                // Right now, we only allow bindless access when the handle comes from a shader input or storage buffer.
-                // This is an artificial limitation to prevent it from being used in cases where it
-                // would have a large performance impact of loading all textures in the pool.
-                // It might be removed in the future, if we can mitigate the performance impact.
+                for (int srcIndex = 0; srcIndex < phi.SourcesCount; srcIndex++)
+                {
+                    Operand phiSource = phi.GetSource(srcIndex);
 
+                    if (phiSource.AsgOp is not PhiNode && !IsBindlessAccessAllowed(phiSource))
+                    {
+                        return false;
+                    }
+                }
+            }
+            else if (!IsBindlessAccessAllowed(bindlessHandle))
+            {
                 return false;
             }
 
@@ -80,8 +91,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             Operand samplerHandle = OperandHelper.Local();
             Operand textureIndex = OperandHelper.Local();
 
-            block.Operations.AddBefore(node, new Operation(Instruction.BitwiseAnd, textureHandle, nvHandle, OperandHelper.Const(0xfffff)));
-            block.Operations.AddBefore(node, new Operation(Instruction.ShiftRightU32, samplerHandle, nvHandle, OperandHelper.Const(20)));
+            block.Operations.AddBefore(node, new Operation(Instruction.BitwiseAnd, textureHandle, bindlessHandle, OperandHelper.Const(0xfffff)));
+            block.Operations.AddBefore(node, new Operation(Instruction.ShiftRightU32, samplerHandle, bindlessHandle, OperandHelper.Const(20)));
 
             int texturePoolLength = Math.Max(BindlessToArray.MinimumArrayLength, gpuAccessor.QueryTextureArrayLengthFromPool());
 
@@ -130,6 +141,30 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
+        private static bool IsBindlessAccessAllowed(Operand bindlessHandle)
+        {
+            if (bindlessHandle.Type == OperandType.ConstantBuffer)
+            {
+                // Bindless access with handles from constant buffer is allowed.
+
+                return true;
+            }
+
+            if (bindlessHandle.AsgOp is not Operation handleOp ||
+                handleOp.Inst != Instruction.Load ||
+                (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
+            {
+                // Right now, we only allow bindless access when the handle comes from a shader input or storage buffer.
+                // This is an artificial limitation to prevent it from being used in cases where it
+                // would have a large performance impact of loading all textures in the pool.
+                // It might be removed in the future, if we can mitigate the performance impact.
+
+                return false;
+            }
+
+            return true;
+        }
+
         private static bool TryConvertBindless(BasicBlock block, ResourceManager resourceManager, IGpuAccessor gpuAccessor, TextureOperation texOp)
         {
             if (texOp.Inst == Instruction.TextureSample || texOp.Inst.IsTextureQuery())
@@ -265,7 +300,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                         resourceManager,
                         gpuAccessor,
                         texOp,
-                        TextureHandle.PackOffsets(src0.GetCbufOffset(), ((src1.Value >> 20) & 0xfff), handleType),
+                        TextureHandle.PackOffsets(src0.GetCbufOffset(), (src1.Value >> 20) & 0xfff, handleType),
                         TextureHandle.PackSlots(src0.GetCbufSlot(), 0),
                         rewriteSamplerType,
                         isImage: false);

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

@@ -126,7 +126,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     continue;
                 }
 
-                if (texOp.GetSource(0).AsgOp is not Operation handleAsgOp)
+                Operand bindlessHandle = Utils.FindLastOperation(texOp.GetSource(0), block);
+
+                if (bindlessHandle.AsgOp is not Operation handleAsgOp)
                 {
                     continue;
                 }
@@ -137,8 +139,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
                 if (handleAsgOp.Inst == Instruction.BitwiseOr)
                 {
-                    Operand src0 = handleAsgOp.GetSource(0);
-                    Operand src1 = handleAsgOp.GetSource(1);
+                    Operand src0 = Utils.FindLastOperation(handleAsgOp.GetSource(0), block);
+                    Operand src1 = Utils.FindLastOperation(handleAsgOp.GetSource(1), block);
 
                     if (src0.Type == OperandType.ConstantBuffer && src1.AsgOp is Operation)
                     {

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

@@ -152,18 +152,14 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
         {
             // If all phi sources are the same, we can propagate it and remove the phi.
 
-            Operand firstSrc = phi.GetSource(0);
-
-            for (int index = 1; index < phi.SourcesCount; index++)
+            if (!Utils.AreAllSourcesTheSameOperand(phi))
             {
-                if (!IsSameOperand(firstSrc, phi.GetSource(index)))
-                {
-                    return false;
-                }
+                return false;
             }
 
             // All sources are equal, we can propagate the value.
 
+            Operand firstSrc = phi.GetSource(0);
             Operand dest = phi.Dest;
 
             INode[] uses = dest.UseOps.ToArray();
@@ -182,17 +178,6 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static bool IsSameOperand(Operand x, Operand y)
-        {
-            if (x.Type != y.Type || x.Value != y.Value)
-            {
-                return false;
-            }
-
-            // TODO: Handle Load operations with the same storage and the same constant parameters.
-            return x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
-        }
-
         private static bool PropagatePack(Operation packOp)
         {
             // Propagate pack source operands to uses by unpack

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

@@ -31,6 +31,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     TryEliminateBitwiseOr(operation);
                     break;
 
+                case Instruction.CompareNotEqual:
+                    TryEliminateCompareNotEqual(operation);
+                    break;
+
                 case Instruction.ConditionalSelect:
                     TryEliminateConditionalSelect(operation);
                     break;
@@ -174,6 +178,32 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             }
         }
 
+        private static void TryEliminateCompareNotEqual(Operation operation)
+        {
+            // Comparison instruction returns 0 if the result is false, and -1 if true.
+            // Doing a not equal zero comparison on the result is redundant, so we can just copy the first result in this case.
+
+            Operand lhs = operation.GetSource(0);
+            Operand rhs = operation.GetSource(1);
+
+            if (lhs.Type == OperandType.Constant)
+            {
+                (lhs, rhs) = (rhs, lhs);
+            }
+
+            if (rhs.Type != OperandType.Constant || rhs.Value != 0)
+            {
+                return;
+            }
+
+            if (lhs.AsgOp is not Operation compareOp || !compareOp.Inst.IsComparison())
+            {
+                return;
+            }
+
+            operation.TurnIntoCopy(lhs);
+        }
+
         private static void TryEliminateConditionalSelect(Operation operation)
         {
             Operand cond = operation.GetSource(0);

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

@@ -34,6 +34,50 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return elemIndexSrc.Type == OperandType.Constant && elemIndexSrc.Value == elemIndex;
         }
 
+        private static bool IsSameOperand(Operand x, Operand y)
+        {
+            if (x.Type != y.Type || x.Value != y.Value)
+            {
+                return false;
+            }
+
+            // TODO: Handle Load operations with the same storage and the same constant parameters.
+            return x == y || x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
+        }
+
+        private static bool AreAllSourcesEqual(INode node, INode otherNode)
+        {
+            if (node.SourcesCount != otherNode.SourcesCount)
+            {
+                return false;
+            }
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(node.GetSource(index), otherNode.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
+        public static bool AreAllSourcesTheSameOperand(INode node)
+        {
+            Operand firstSrc = node.GetSource(0);
+
+            for (int index = 1; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(firstSrc, node.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
         private static Operation FindBranchSource(BasicBlock block)
         {
             foreach (BasicBlock sourceBlock in block.Predecessors)
@@ -55,6 +99,19 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return inst == Instruction.BranchIfFalse || inst == Instruction.BranchIfTrue;
         }
 
+        private static bool IsSameCondition(Operand currentCondition, Operand queryCondition)
+        {
+            if (currentCondition == queryCondition)
+            {
+                return true;
+            }
+
+            return currentCondition.AsgOp is Operation currentOperation &&
+                queryCondition.AsgOp is Operation queryOperation &&
+                currentOperation.Inst == queryOperation.Inst &&
+                AreAllSourcesEqual(currentOperation, queryOperation);
+        }
+
         private static bool BlockConditionsMatch(BasicBlock currentBlock, BasicBlock queryBlock)
         {
             // Check if all the conditions for the query block are satisfied by the current block.
@@ -70,10 +127,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
             return currentBranch != null && queryBranch != null &&
                    currentBranch.Inst == queryBranch.Inst &&
-                   currentCondition == queryCondition;
+                   IsSameCondition(currentCondition, queryCondition);
         }
 
-        public static Operand FindLastOperation(Operand source, BasicBlock block)
+        public static Operand FindLastOperation(Operand source, BasicBlock block, bool recurse = true)
         {
             if (source.AsgOp is PhiNode phiNode)
             {
@@ -84,10 +141,23 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 for (int i = phiNode.SourcesCount - 1; i >= 0; i--)
                 {
                     BasicBlock phiBlock = phiNode.GetBlock(i);
+                    Operand phiSource = phiNode.GetSource(i);
 
                     if (BlockConditionsMatch(block, phiBlock))
                     {
-                        return phiNode.GetSource(i);
+                        return phiSource;
+                    }
+                    else if (recurse && phiSource.AsgOp is PhiNode)
+                    {
+                        // Phi source is another phi.
+                        // Let's check if that phi has a block that matches our condition.
+
+                        Operand match = FindLastOperation(phiSource, block, false);
+
+                        if (match != phiSource)
+                        {
+                            return match;
+                        }
                     }
                 }
             }

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

@@ -155,9 +155,14 @@ namespace Ryujinx.Graphics.Shader.Translation
                         localInputs[block.Index] |= GetMask(register) & ~localOutputs[block.Index];
                     }
 
-                    if (operation.Dest != null && operation.Dest.Type == OperandType.Register)
+                    for (int dstIndex = 0; dstIndex < operation.DestsCount; dstIndex++)
                     {
-                        localOutputs[block.Index] |= GetMask(operation.Dest.GetRegister());
+                        Operand dest = operation.GetDest(dstIndex);
+
+                        if (dest != null && dest.Type == OperandType.Register)
+                        {
+                            localOutputs[block.Index] |= GetMask(dest.GetRegister());
+                        }
                     }
                 }
             }

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

@@ -363,6 +363,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 GpuAccessor.QueryHostSupportsGeometryShaderPassthrough(),
                 GpuAccessor.QueryHostSupportsShaderBallot(),
                 GpuAccessor.QueryHostSupportsShaderBarrierDivergence(),
+                GpuAccessor.QueryHostSupportsShaderFloat64(),
                 GpuAccessor.QueryHostSupportsTextureShadowLod(),
                 GpuAccessor.QueryHostSupportsViewportMask());
 

src/Ryujinx.Graphics.Texture/Astc/AstcDecoder.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common.Memory;
 using Ryujinx.Common.Utilities;
 using System;
-using System.Buffers;
 using System.Diagnostics;
 using System.Linq;
 using System.Runtime.CompilerServices;
@@ -293,9 +292,9 @@ namespace Ryujinx.Graphics.Texture.Astc
             int depth,
             int levels,
             int layers,
-            out IMemoryOwner<byte> decoded)
+            out MemoryOwner<byte> decoded)
         {
-            decoded = ByteMemoryPool.Rent(QueryDecompressedSize(width, height, depth, levels, layers));
+            decoded = MemoryOwner<byte>.Rent(QueryDecompressedSize(width, height, depth, levels, layers));
 
             AstcDecoder decoder = new(data, decoded.Memory, blockWidth, blockHeight, width, height, depth, levels, layers);
 

src/Ryujinx.Graphics.Texture/BCnDecoder.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using System;
-using System.Buffers;
 using System.Buffers.Binary;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics;
@@ -14,7 +13,7 @@ namespace Ryujinx.Graphics.Texture
         private const int BlockWidth = 4;
         private const int BlockHeight = 4;
 
-        public static IMemoryOwner<byte> DecodeBC1(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeBC1(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             int size = 0;
 
@@ -23,12 +22,12 @@ namespace Ryujinx.Graphics.Texture
                 size += Math.Max(1, width >> l) * Math.Max(1, height >> l) * Math.Max(1, depth >> l) * layers * 4;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
 
             Span<byte> tile = stackalloc byte[BlockWidth * BlockHeight * 4];
 
             Span<uint> tileAsUint = MemoryMarshal.Cast<byte, uint>(tile);
-            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             Span<Vector128<byte>> tileAsVector128 = MemoryMarshal.Cast<byte, Vector128<byte>>(tile);
 
@@ -102,7 +101,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC2(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeBC2(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             int size = 0;
 
@@ -111,12 +110,12 @@ namespace Ryujinx.Graphics.Texture
                 size += Math.Max(1, width >> l) * Math.Max(1, height >> l) * Math.Max(1, depth >> l) * layers * 4;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
 
             Span<byte> tile = stackalloc byte[BlockWidth * BlockHeight * 4];
 
             Span<uint> tileAsUint = MemoryMarshal.Cast<byte, uint>(tile);
-            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             Span<Vector128<byte>> tileAsVector128 = MemoryMarshal.Cast<byte, Vector128<byte>>(tile);
 
@@ -197,7 +196,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC3(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeBC3(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             int size = 0;
 
@@ -206,13 +205,13 @@ namespace Ryujinx.Graphics.Texture
                 size += Math.Max(1, width >> l) * Math.Max(1, height >> l) * Math.Max(1, depth >> l) * layers * 4;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
 
             Span<byte> tile = stackalloc byte[BlockWidth * BlockHeight * 4];
             Span<byte> rPal = stackalloc byte[8];
 
             Span<uint> tileAsUint = MemoryMarshal.Cast<byte, uint>(tile);
-            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputAsUint = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             Span<Vector128<byte>> tileAsVector128 = MemoryMarshal.Cast<byte, Vector128<byte>>(tile);
 
@@ -294,7 +293,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC4(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
+        public static MemoryOwner<byte> DecodeBC4(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
         {
             int size = 0;
 
@@ -306,8 +305,8 @@ namespace Ryujinx.Graphics.Texture
             // Backends currently expect a stride alignment of 4 bytes, so output width must be aligned.
             int alignedWidth = BitUtils.AlignUp(width, 4);
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
-            Span<byte> outputSpan = output.Memory.Span;
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
+            Span<byte> outputSpan = output.Span;
 
             ReadOnlySpan<ulong> data64 = MemoryMarshal.Cast<byte, ulong>(data);
 
@@ -402,7 +401,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC5(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
+        public static MemoryOwner<byte> DecodeBC5(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
         {
             int size = 0;
 
@@ -414,7 +413,7 @@ namespace Ryujinx.Graphics.Texture
             // Backends currently expect a stride alignment of 4 bytes, so output width must be aligned.
             int alignedWidth = BitUtils.AlignUp(width, 2);
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
 
             ReadOnlySpan<ulong> data64 = MemoryMarshal.Cast<byte, ulong>(data);
 
@@ -423,7 +422,7 @@ namespace Ryujinx.Graphics.Texture
             Span<byte> rPal = stackalloc byte[8];
             Span<byte> gPal = stackalloc byte[8];
 
-            Span<ushort> outputAsUshort = MemoryMarshal.Cast<byte, ushort>(output.Memory.Span);
+            Span<ushort> outputAsUshort = MemoryMarshal.Cast<byte, ushort>(output.Span);
 
             Span<uint> rTileAsUint = MemoryMarshal.Cast<byte, uint>(rTile);
             Span<uint> gTileAsUint = MemoryMarshal.Cast<byte, uint>(gTile);
@@ -527,7 +526,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC6(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
+        public static MemoryOwner<byte> DecodeBC6(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers, bool signed)
         {
             int size = 0;
 
@@ -536,8 +535,8 @@ namespace Ryujinx.Graphics.Texture
                 size += Math.Max(1, width >> l) * Math.Max(1, height >> l) * Math.Max(1, depth >> l) * layers * 8;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
-            Span<byte> outputSpan = output.Memory.Span;
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
+            Span<byte> outputSpan = output.Span;
 
             int inputOffset = 0;
             int outputOffset = 0;
@@ -566,7 +565,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeBC7(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeBC7(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             int size = 0;
 
@@ -575,8 +574,8 @@ namespace Ryujinx.Graphics.Texture
                 size += Math.Max(1, width >> l) * Math.Max(1, height >> l) * Math.Max(1, depth >> l) * layers * 4;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
-            Span<byte> outputSpan = output.Memory.Span;
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
+            Span<byte> outputSpan = output.Span;
 
             int inputOffset = 0;
             int outputOffset = 0;

src/Ryujinx.Graphics.Texture/BCnEncoder.cs

@@ -2,7 +2,6 @@ using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.Texture.Encoders;
 using System;
-using System.Buffers;
 
 namespace Ryujinx.Graphics.Texture
 {
@@ -11,7 +10,7 @@ namespace Ryujinx.Graphics.Texture
         private const int BlockWidth = 4;
         private const int BlockHeight = 4;
 
-        public static IMemoryOwner<byte> EncodeBC7(Memory<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> EncodeBC7(Memory<byte> data, int width, int height, int depth, int levels, int layers)
         {
             int size = 0;
 
@@ -23,7 +22,7 @@ namespace Ryujinx.Graphics.Texture
                 size += w * h * 16 * Math.Max(1, depth >> l) * layers;
             }
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(size);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(size);
             Memory<byte> outputMemory = output.Memory;
 
             int imageBaseIOffs = 0;

src/Ryujinx.Graphics.Texture/ETC2Decoder.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using System;
-using System.Buffers;
 using System.Buffers.Binary;
 using System.Runtime.InteropServices;
 
@@ -51,15 +50,15 @@ namespace Ryujinx.Graphics.Texture
             new int[] { -3, -5, -7, -9, 2, 4, 6, 8 },
         };
 
-        public static IMemoryOwner<byte> DecodeRgb(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeRgb(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             ReadOnlySpan<ulong> dataUlong = MemoryMarshal.Cast<byte, ulong>(data);
 
             int inputOffset = 0;
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(CalculateOutputSize(width, height, depth, levels, layers));
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(CalculateOutputSize(width, height, depth, levels, layers));
 
-            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Span);
             Span<uint> tile = stackalloc uint[BlockWidth * BlockHeight];
 
             int imageBaseOOffs = 0;
@@ -113,15 +112,15 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodePta(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodePta(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             ReadOnlySpan<ulong> dataUlong = MemoryMarshal.Cast<byte, ulong>(data);
 
             int inputOffset = 0;
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(CalculateOutputSize(width, height, depth, levels, layers));
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(CalculateOutputSize(width, height, depth, levels, layers));
 
-            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Span);
             Span<uint> tile = stackalloc uint[BlockWidth * BlockHeight];
 
             int imageBaseOOffs = 0;
@@ -170,15 +169,15 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> DecodeRgba(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
+        public static MemoryOwner<byte> DecodeRgba(ReadOnlySpan<byte> data, int width, int height, int depth, int levels, int layers)
         {
             ReadOnlySpan<ulong> dataUlong = MemoryMarshal.Cast<byte, ulong>(data);
 
             int inputOffset = 0;
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(CalculateOutputSize(width, height, depth, levels, layers));
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(CalculateOutputSize(width, height, depth, levels, layers));
 
-            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputUint = MemoryMarshal.Cast<byte, uint>(output.Span);
             Span<uint> tile = stackalloc uint[BlockWidth * BlockHeight];
 
             int imageBaseOOffs = 0;

src/Ryujinx.Graphics.Texture/LayoutConverter.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using System;
-using System.Buffers;
 using System.Runtime.Intrinsics;
 using static Ryujinx.Graphics.Texture.BlockLinearConstants;
 
@@ -95,7 +94,7 @@ namespace Ryujinx.Graphics.Texture
             };
         }
 
-        public static IMemoryOwner<byte> ConvertBlockLinearToLinear(
+        public static MemoryOwner<byte> ConvertBlockLinearToLinear(
             int width,
             int height,
             int depth,
@@ -121,8 +120,8 @@ namespace Ryujinx.Graphics.Texture
                 blockHeight,
                 bytesPerPixel);
 
-            IMemoryOwner<byte> outputOwner = ByteMemoryPool.Rent(outSize);
-            Span<byte> output = outputOwner.Memory.Span;
+            MemoryOwner<byte> outputOwner = MemoryOwner<byte>.Rent(outSize);
+            Span<byte> output = outputOwner.Span;
 
             int outOffs = 0;
 
@@ -249,7 +248,7 @@ namespace Ryujinx.Graphics.Texture
             return outputOwner;
         }
 
-        public static IMemoryOwner<byte> ConvertLinearStridedToLinear(
+        public static MemoryOwner<byte> ConvertLinearStridedToLinear(
             int width,
             int height,
             int blockWidth,
@@ -265,8 +264,8 @@ namespace Ryujinx.Graphics.Texture
             int outStride = BitUtils.AlignUp(w * bytesPerPixel, HostStrideAlignment);
             lineSize = Math.Min(lineSize, outStride);
 
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(h * outStride);
-            Span<byte> outSpan = output.Memory.Span;
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(h * outStride);
+            Span<byte> outSpan = output.Span;
 
             int outOffs = 0;
             int inOffs = 0;

src/Ryujinx.Graphics.Texture/PixelConverter.cs

@@ -1,7 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using System;
-using System.Buffers;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics;
 using System.Runtime.Intrinsics.X86;
@@ -21,13 +20,14 @@ namespace Ryujinx.Graphics.Texture
             return (remainder, outRemainder, length / stride);
         }
 
-        public unsafe static IMemoryOwner<byte> ConvertR4G4ToR4G4B4A4(ReadOnlySpan<byte> data, int width)
+        public unsafe static MemoryOwner<byte> ConvertR4G4ToR4G4B4A4(ReadOnlySpan<byte> data, int width)
         {
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(data.Length * 2);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(data.Length * 2);
+            Span<byte> outputSpan = output.Span;
 
             (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 1, 2);
 
-            Span<ushort> outputSpan = MemoryMarshal.Cast<byte, ushort>(output.Memory.Span);
+            Span<ushort> outputSpanUInt16 = MemoryMarshal.Cast<byte, ushort>(outputSpan);
 
             if (remainder == 0)
             {
@@ -38,7 +38,7 @@ namespace Ryujinx.Graphics.Texture
                     int sizeTrunc = data.Length & ~7;
                     start = sizeTrunc;
 
-                    fixed (byte* inputPtr = data, outputPtr = output.Memory.Span)
+                    fixed (byte* inputPtr = data, outputPtr = outputSpan)
                     {
                         for (ulong offset = 0; offset < (ulong)sizeTrunc; offset += 8)
                         {
@@ -49,7 +49,7 @@ namespace Ryujinx.Graphics.Texture
 
                 for (int i = start; i < data.Length; i++)
                 {
-                    outputSpan[i] = data[i];
+                    outputSpanUInt16[i] = data[i];
                 }
             }
             else
@@ -61,7 +61,7 @@ namespace Ryujinx.Graphics.Texture
                 {
                     for (int x = 0; x < width; x++)
                     {
-                        outputSpan[outOffset++] = data[offset++];
+                        outputSpanUInt16[outOffset++] = data[offset++];
                     }
 
                     offset += remainder;
@@ -72,16 +72,16 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> ConvertR5G6B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        public static MemoryOwner<byte> ConvertR5G6B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
         {
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(data.Length * 2);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(data.Length * 2);
             int offset = 0;
             int outOffset = 0;
 
             (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
 
             ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
-            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             for (int y = 0; y < height; y++)
             {
@@ -109,16 +109,16 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> ConvertR5G5B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width, bool forceAlpha)
+        public static MemoryOwner<byte> ConvertR5G5B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width, bool forceAlpha)
         {
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(data.Length * 2);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(data.Length * 2);
             int offset = 0;
             int outOffset = 0;
 
             (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
 
             ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
-            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             for (int y = 0; y < height; y++)
             {
@@ -146,16 +146,16 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> ConvertA1B5G5R5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        public static MemoryOwner<byte> ConvertA1B5G5R5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
         {
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(data.Length * 2);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(data.Length * 2);
             int offset = 0;
             int outOffset = 0;
 
             (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
 
             ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
-            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             for (int y = 0; y < height; y++)
             {
@@ -183,16 +183,16 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static IMemoryOwner<byte> ConvertR4G4B4A4ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        public static MemoryOwner<byte> ConvertR4G4B4A4ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
         {
-            IMemoryOwner<byte> output = ByteMemoryPool.Rent(data.Length * 2);
+            MemoryOwner<byte> output = MemoryOwner<byte>.Rent(data.Length * 2);
             int offset = 0;
             int outOffset = 0;
 
             (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
 
             ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
-            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Memory.Span);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output.Span);
 
             for (int y = 0; y < height; y++)
             {

src/Ryujinx.Graphics.Vulkan/BackgroundResources.cs

@@ -29,7 +29,14 @@ namespace Ryujinx.Graphics.Vulkan
 
                 lock (queueLock)
                 {
-                    _pool = new CommandBufferPool(_gd.Api, _device, queue, queueLock, _gd.QueueFamilyIndex, isLight: true);
+                    _pool = new CommandBufferPool(
+                        _gd.Api,
+                        _device,
+                        queue,
+                        queueLock,
+                        _gd.QueueFamilyIndex,
+                        _gd.IsQualcommProprietary,
+                        isLight: true);
                 }
             }
 

src/Ryujinx.Graphics.Vulkan/BarrierBatch.cs

@@ -1,6 +1,7 @@
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
+using System.Runtime.CompilerServices;
 
 namespace Ryujinx.Graphics.Vulkan
 {
@@ -8,22 +9,64 @@ namespace Ryujinx.Graphics.Vulkan
     {
         private const int MaxBarriersPerCall = 16;
 
+        private const AccessFlags BaseAccess = AccessFlags.ShaderReadBit | AccessFlags.ShaderWriteBit;
+        private const AccessFlags BufferAccess = AccessFlags.IndexReadBit | AccessFlags.VertexAttributeReadBit | AccessFlags.UniformReadBit;
+        private const AccessFlags CommandBufferAccess = AccessFlags.IndirectCommandReadBit;
+
         private readonly VulkanRenderer _gd;
 
         private readonly NativeArray<MemoryBarrier> _memoryBarrierBatch = new(MaxBarriersPerCall);
         private readonly NativeArray<BufferMemoryBarrier> _bufferBarrierBatch = new(MaxBarriersPerCall);
         private readonly NativeArray<ImageMemoryBarrier> _imageBarrierBatch = new(MaxBarriersPerCall);
 
-        private readonly List<BarrierWithStageFlags<MemoryBarrier>> _memoryBarriers = new();
-        private readonly List<BarrierWithStageFlags<BufferMemoryBarrier>> _bufferBarriers = new();
-        private readonly List<BarrierWithStageFlags<ImageMemoryBarrier>> _imageBarriers = new();
+        private readonly List<BarrierWithStageFlags<MemoryBarrier, int>> _memoryBarriers = new();
+        private readonly List<BarrierWithStageFlags<BufferMemoryBarrier, int>> _bufferBarriers = new();
+        private readonly List<BarrierWithStageFlags<ImageMemoryBarrier, TextureStorage>> _imageBarriers = new();
         private int _queuedBarrierCount;
 
+        private enum IncoherentBarrierType
+        {
+            None,
+            Texture,
+            All,
+            CommandBuffer
+        }
+
+        private PipelineStageFlags _incoherentBufferWriteStages;
+        private PipelineStageFlags _incoherentTextureWriteStages;
+        private PipelineStageFlags _extraStages;
+        private IncoherentBarrierType _queuedIncoherentBarrier;
+
         public BarrierBatch(VulkanRenderer gd)
         {
             _gd = gd;
         }
 
+        public static (AccessFlags Access, PipelineStageFlags Stages) GetSubpassAccessSuperset(VulkanRenderer gd)
+        {
+            AccessFlags access = BufferAccess;
+            PipelineStageFlags stages = PipelineStageFlags.AllGraphicsBit;
+
+            if (gd.TransformFeedbackApi != null)
+            {
+                access |= AccessFlags.TransformFeedbackWriteBitExt;
+                stages |= PipelineStageFlags.TransformFeedbackBitExt;
+            }
+
+            if (!gd.IsTBDR)
+            {
+                // Desktop GPUs can transform image barriers into memory barriers.
+
+                access |= AccessFlags.DepthStencilAttachmentWriteBit | AccessFlags.ColorAttachmentWriteBit;
+                access |= AccessFlags.DepthStencilAttachmentReadBit | AccessFlags.ColorAttachmentReadBit;
+
+                stages |= PipelineStageFlags.EarlyFragmentTestsBit | PipelineStageFlags.LateFragmentTestsBit;
+                stages |= PipelineStageFlags.ColorAttachmentOutputBit;
+            }
+
+            return (access, stages);
+        }
+
         private readonly record struct StageFlags : IEquatable<StageFlags>
         {
             public readonly PipelineStageFlags Source;
@@ -36,47 +79,130 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        private readonly struct BarrierWithStageFlags<T> where T : unmanaged
+        private readonly struct BarrierWithStageFlags<T, T2> where T : unmanaged
         {
             public readonly StageFlags Flags;
             public readonly T Barrier;
+            public readonly T2 Resource;
 
             public BarrierWithStageFlags(StageFlags flags, T barrier)
             {
                 Flags = flags;
                 Barrier = barrier;
+                Resource = default;
             }
 
-            public BarrierWithStageFlags(PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags, T barrier)
+            public BarrierWithStageFlags(PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags, T barrier, T2 resource)
             {
                 Flags = new StageFlags(srcStageFlags, dstStageFlags);
                 Barrier = barrier;
+                Resource = resource;
             }
         }
 
-        private void QueueBarrier<T>(List<BarrierWithStageFlags<T>> list, T barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags) where T : unmanaged
+        private void QueueBarrier<T, T2>(List<BarrierWithStageFlags<T, T2>> list, T barrier, T2 resource, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags) where T : unmanaged
         {
-            list.Add(new BarrierWithStageFlags<T>(srcStageFlags, dstStageFlags, barrier));
+            list.Add(new BarrierWithStageFlags<T, T2>(srcStageFlags, dstStageFlags, barrier, resource));
             _queuedBarrierCount++;
         }
 
         public void QueueBarrier(MemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
         {
-            QueueBarrier(_memoryBarriers, barrier, srcStageFlags, dstStageFlags);
+            QueueBarrier(_memoryBarriers, barrier, default, srcStageFlags, dstStageFlags);
         }
 
         public void QueueBarrier(BufferMemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
         {
-            QueueBarrier(_bufferBarriers, barrier, srcStageFlags, dstStageFlags);
+            QueueBarrier(_bufferBarriers, barrier, default, srcStageFlags, dstStageFlags);
         }
 
-        public void QueueBarrier(ImageMemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
+        public void QueueBarrier(ImageMemoryBarrier barrier, TextureStorage resource, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
         {
-            QueueBarrier(_imageBarriers, barrier, srcStageFlags, dstStageFlags);
+            QueueBarrier(_imageBarriers, barrier, resource, srcStageFlags, dstStageFlags);
         }
 
-        public unsafe void Flush(CommandBuffer cb, bool insideRenderPass, Action endRenderPass)
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public unsafe void FlushMemoryBarrier(ShaderCollection program, bool inRenderPass)
         {
+            if (_queuedIncoherentBarrier > IncoherentBarrierType.None)
+            {
+                // We should emit a memory barrier if there's a write access in the program (current program, or program since last barrier)
+                bool hasTextureWrite = _incoherentTextureWriteStages != PipelineStageFlags.None;
+                bool hasBufferWrite = _incoherentBufferWriteStages != PipelineStageFlags.None;
+                bool hasBufferBarrier = _queuedIncoherentBarrier > IncoherentBarrierType.Texture;
+
+                if (hasTextureWrite || (hasBufferBarrier && hasBufferWrite))
+                {
+                    AccessFlags access = BaseAccess;
+
+                    PipelineStageFlags stages = inRenderPass ? PipelineStageFlags.AllGraphicsBit : PipelineStageFlags.AllCommandsBit;
+
+                    if (hasBufferBarrier && hasBufferWrite)
+                    {
+                        access |= BufferAccess;
+
+                        if (_gd.TransformFeedbackApi != null)
+                        {
+                            access |= AccessFlags.TransformFeedbackWriteBitExt;
+                            stages |= PipelineStageFlags.TransformFeedbackBitExt;
+                        }
+                    }
+
+                    if (_queuedIncoherentBarrier == IncoherentBarrierType.CommandBuffer)
+                    {
+                        access |= CommandBufferAccess;
+                        stages |= PipelineStageFlags.DrawIndirectBit;
+                    }
+
+                    MemoryBarrier barrier = new MemoryBarrier()
+                    {
+                        SType = StructureType.MemoryBarrier,
+                        SrcAccessMask = access,
+                        DstAccessMask = access
+                    };
+
+                    QueueBarrier(barrier, stages, stages);
+
+                    _incoherentTextureWriteStages = program?.IncoherentTextureWriteStages ?? PipelineStageFlags.None;
+
+                    if (_queuedIncoherentBarrier > IncoherentBarrierType.Texture)
+                    {
+                        if (program != null)
+                        {
+                            _incoherentBufferWriteStages = program.IncoherentBufferWriteStages | _extraStages;
+                        }
+                        else
+                        {
+                            _incoherentBufferWriteStages = PipelineStageFlags.None;
+                        }
+                    }
+
+                    _queuedIncoherentBarrier = IncoherentBarrierType.None;
+                }
+            }
+        }
+
+        public unsafe void Flush(CommandBufferScoped cbs, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
+        {
+            Flush(cbs, null, inRenderPass, rpHolder, endRenderPass);
+        }
+
+        public unsafe void Flush(CommandBufferScoped cbs, ShaderCollection program, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
+        {
+            if (program != null)
+            {
+                _incoherentBufferWriteStages |= program.IncoherentBufferWriteStages | _extraStages;
+                _incoherentTextureWriteStages |= program.IncoherentTextureWriteStages;
+            }
+
+            FlushMemoryBarrier(program, inRenderPass);
+
+            if (!inRenderPass && rpHolder != null)
+            {
+                // Render pass is about to begin. Queue any fences that normally interrupt the pass.
+                rpHolder.InsertForcedFences(cbs);
+            }
+
             while (_queuedBarrierCount > 0)
             {
                 int memoryCount = 0;
@@ -86,20 +212,20 @@ namespace Ryujinx.Graphics.Vulkan
                 bool hasBarrier = false;
                 StageFlags flags = default;
 
-                static void AddBarriers<T>(
+                static void AddBarriers<T, T2>(
                     Span<T> target,
                     ref int queuedBarrierCount,
                     ref bool hasBarrier,
                     ref StageFlags flags,
                     ref int count,
-                    List<BarrierWithStageFlags<T>> list) where T : unmanaged
+                    List<BarrierWithStageFlags<T, T2>> list) where T : unmanaged
                 {
                     int firstMatch = -1;
                     int end = list.Count;
 
                     for (int i = 0; i < list.Count; i++)
                     {
-                        BarrierWithStageFlags<T> barrier = list[i];
+                        BarrierWithStageFlags<T, T2> barrier = list[i];
 
                         if (!hasBarrier)
                         {
@@ -162,21 +288,60 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                 }
 
-                if (insideRenderPass)
+                if (inRenderPass && _imageBarriers.Count > 0)
                 {
                     // Image barriers queued in the batch are meant to be globally scoped,
                     // but inside a render pass they're scoped to just the range of the render pass.
 
                     // On MoltenVK, we just break the rules and always use image barrier.
                     // On desktop GPUs, all barriers are globally scoped, so we just replace it with a generic memory barrier.
-                    // TODO: On certain GPUs, we need to split render pass so the barrier scope is global. When this is done,
-                    //       notify the resource that it should add a barrier as soon as a render pass ends to avoid this in future.
+                    // Generally, we want to avoid this from happening in the future, so flag the texture to immediately
+                    // emit a barrier whenever the current render pass is bound again.
 
-                    if (!_gd.IsMoltenVk)
+                    bool anyIsNonAttachment = false;
+
+                    foreach (BarrierWithStageFlags<ImageMemoryBarrier, TextureStorage> barrier in _imageBarriers)
                     {
+                        // If the binding is an attachment, don't add it as a forced fence.
+                        bool isAttachment = rpHolder.ContainsAttachment(barrier.Resource);
+
+                        if (!isAttachment)
+                        {
+                            rpHolder.AddForcedFence(barrier.Resource, barrier.Flags.Dest);
+                            anyIsNonAttachment = true;
+                        }
+                    }
+
+                    if (_gd.IsTBDR)
+                    {
+                        if (!_gd.IsMoltenVk)
+                        {
+                            if (!anyIsNonAttachment)
+                            {
+                                // This case is a feedback loop. To prevent this from causing an absolute performance disaster,
+                                // remove the barriers entirely.
+                                // If this is not here, there will be a lot of single draw render passes.
+                                // TODO: explicit handling for feedback loops, likely outside this class.
+
+                                _queuedBarrierCount -= _imageBarriers.Count;
+                                _imageBarriers.Clear();
+                            }
+                            else
+                            {
+                                // TBDR GPUs are sensitive to barriers, so we need to end the pass to ensure the data is available.
+                                // Metal already has hazard tracking so MVK doesn't need this.
+                                endRenderPass();
+                                inRenderPass = false;
+                            }
+                        }
+                    }
+                    else
+                    {
+                        // Generic pipeline memory barriers will work for desktop GPUs.
+                        // They do require a few more access flags on the subpass dependency, though.
                         foreach (var barrier in _imageBarriers)
                         {
-                            _memoryBarriers.Add(new BarrierWithStageFlags<MemoryBarrier>(
+                            _memoryBarriers.Add(new BarrierWithStageFlags<MemoryBarrier, int>(
                                 barrier.Flags,
                                 new MemoryBarrier()
                                 {
@@ -190,6 +355,22 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                 }
 
+                if (inRenderPass && _memoryBarriers.Count > 0)
+                {
+                    PipelineStageFlags allFlags = PipelineStageFlags.None;
+
+                    foreach (var barrier in _memoryBarriers)
+                    {
+                        allFlags |= barrier.Flags.Dest;
+                    }
+
+                    if (allFlags.HasFlag(PipelineStageFlags.DrawIndirectBit) || !_gd.SupportsRenderPassBarrier(allFlags))
+                    {
+                        endRenderPass();
+                        inRenderPass = false;
+                    }
+                }
+
                 AddBarriers(_memoryBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref memoryCount, _memoryBarriers);
                 AddBarriers(_bufferBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref bufferCount, _bufferBarriers);
                 AddBarriers(_imageBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref imageCount, _imageBarriers);
@@ -198,14 +379,14 @@ namespace Ryujinx.Graphics.Vulkan
                 {
                     PipelineStageFlags srcStageFlags = flags.Source;
 
-                    if (insideRenderPass)
+                    if (inRenderPass)
                     {
                         // Inside a render pass, barrier stages can only be from rasterization.
                         srcStageFlags &= ~PipelineStageFlags.ComputeShaderBit;
                     }
 
                     _gd.Api.CmdPipelineBarrier(
-                        cb,
+                        cbs.CommandBuffer,
                         srcStageFlags,
                         flags.Dest,
                         0,
@@ -219,6 +400,41 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private void QueueIncoherentBarrier(IncoherentBarrierType type)
+        {
+            if (type > _queuedIncoherentBarrier)
+            {
+                _queuedIncoherentBarrier = type;
+            }
+        }
+
+        public void QueueTextureBarrier()
+        {
+            QueueIncoherentBarrier(IncoherentBarrierType.Texture);
+        }
+
+        public void QueueMemoryBarrier()
+        {
+            QueueIncoherentBarrier(IncoherentBarrierType.All);
+        }
+
+        public void QueueCommandBufferBarrier()
+        {
+            QueueIncoherentBarrier(IncoherentBarrierType.CommandBuffer);
+        }
+
+        public void EnableTfbBarriers(bool enable)
+        {
+            if (enable)
+            {
+                _extraStages |= PipelineStageFlags.TransformFeedbackBitExt;
+            }
+            else
+            {
+                _extraStages &= ~PipelineStageFlags.TransformFeedbackBitExt;
+            }
+        }
+
         public void Dispose()
         {
             _memoryBarrierBatch.Dispose();

src/Ryujinx.Graphics.Vulkan/BufferManager.cs

@@ -103,12 +103,19 @@ namespace Ryujinx.Graphics.Vulkan
                 usage |= BufferUsageFlags.IndirectBufferBit;
             }
 
+            var externalMemoryBuffer = new ExternalMemoryBufferCreateInfo
+            {
+                SType = StructureType.ExternalMemoryBufferCreateInfo,
+                HandleTypes = ExternalMemoryHandleTypeFlags.HostAllocationBitExt,
+            };
+
             var bufferCreateInfo = new BufferCreateInfo
             {
                 SType = StructureType.BufferCreateInfo,
                 Size = (ulong)size,
                 Usage = usage,
                 SharingMode = SharingMode.Exclusive,
+                PNext = &externalMemoryBuffer,
             };
 
             gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();

src/Ryujinx.Graphics.Vulkan/CommandBufferPool.cs

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
         private readonly Queue _queue;
         private readonly object _queueLock;
+        private readonly bool _concurrentFenceWaitUnsupported;
         private readonly CommandPool _pool;
         private readonly Thread _owner;
 
@@ -30,11 +31,9 @@ namespace Ryujinx.Graphics.Vulkan
             public int SubmissionCount;
             public CommandBuffer CommandBuffer;
             public FenceHolder Fence;
-            public SemaphoreHolder Semaphore;
 
             public List<IAuto> Dependants;
             public List<MultiFenceHolder> Waitables;
-            public HashSet<SemaphoreHolder> Dependencies;
 
             public void Initialize(Vk api, Device device, CommandPool pool)
             {
@@ -50,7 +49,6 @@ namespace Ryujinx.Graphics.Vulkan
 
                 Dependants = new List<IAuto>();
                 Waitables = new List<MultiFenceHolder>();
-                Dependencies = new HashSet<SemaphoreHolder>();
             }
         }
 
@@ -61,12 +59,20 @@ namespace Ryujinx.Graphics.Vulkan
         private int _queuedCount;
         private int _inUseCount;
 
-        public unsafe CommandBufferPool(Vk api, Device device, Queue queue, object queueLock, uint queueFamilyIndex, bool isLight = false)
+        public unsafe CommandBufferPool(
+            Vk api,
+            Device device,
+            Queue queue,
+            object queueLock,
+            uint queueFamilyIndex,
+            bool concurrentFenceWaitUnsupported,
+            bool isLight = false)
         {
             _api = api;
             _device = device;
             _queue = queue;
             _queueLock = queueLock;
+            _concurrentFenceWaitUnsupported = concurrentFenceWaitUnsupported;
             _owner = Thread.CurrentThread;
 
             var commandPoolCreateInfo = new CommandPoolCreateInfo
@@ -134,14 +140,6 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        public void AddDependency(int cbIndex, CommandBufferScoped dependencyCbs)
-        {
-            Debug.Assert(_commandBuffers[cbIndex].InUse);
-            var semaphoreHolder = _commandBuffers[dependencyCbs.CommandBufferIndex].Semaphore;
-            semaphoreHolder.Get();
-            _commandBuffers[cbIndex].Dependencies.Add(semaphoreHolder);
-        }
-
         public void AddWaitable(int cbIndex, MultiFenceHolder waitable)
         {
             ref var entry = ref _commandBuffers[cbIndex];
@@ -345,19 +343,13 @@ namespace Ryujinx.Graphics.Vulkan
                 waitable.RemoveBufferUses(cbIndex);
             }
 
-            foreach (var dependency in entry.Dependencies)
-            {
-                dependency.Put();
-            }
-
             entry.Dependants.Clear();
             entry.Waitables.Clear();
-            entry.Dependencies.Clear();
             entry.Fence?.Dispose();
 
             if (refreshFence)
             {
-                entry.Fence = new FenceHolder(_api, _device);
+                entry.Fence = new FenceHolder(_api, _device, _concurrentFenceWaitUnsupported);
             }
             else
             {

src/Ryujinx.Graphics.Vulkan/CommandBufferScoped.cs

@@ -26,11 +26,6 @@ namespace Ryujinx.Graphics.Vulkan
             _pool.AddWaitable(CommandBufferIndex, waitable);
         }
 
-        public void AddDependency(CommandBufferScoped dependencyCbs)
-        {
-            _pool.AddDependency(CommandBufferIndex, dependencyCbs);
-        }
-
         public FenceHolder GetFence()
         {
             return _pool.GetFence(CommandBufferIndex);

src/Ryujinx.Graphics.Vulkan/DescriptorSetUpdater.cs

@@ -73,7 +73,6 @@ namespace Ryujinx.Graphics.Vulkan
 
         private readonly VulkanRenderer _gd;
         private readonly Device _device;
-        private readonly PipelineBase _pipeline;
         private ShaderCollection _program;
 
         private readonly BufferRef[] _uniformBufferRefs;
@@ -125,11 +124,10 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly TextureView _dummyTexture;
         private readonly SamplerHolder _dummySampler;
 
-        public DescriptorSetUpdater(VulkanRenderer gd, Device device, PipelineBase pipeline)
+        public DescriptorSetUpdater(VulkanRenderer gd, Device device)
         {
             _gd = gd;
             _device = device;
-            _pipeline = pipeline;
 
             // Some of the bindings counts needs to be multiplied by 2 because we have buffer and
             // regular textures/images interleaved on the same descriptor set.
@@ -213,7 +211,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public void Initialize()
         {
-            IMemoryOwner<byte> dummyTextureData = ByteMemoryPool.RentCleared(4);
+            MemoryOwner<byte> dummyTextureData = MemoryOwner<byte>.RentCleared(4);
             _dummyTexture.SetData(dummyTextureData);
         }
 
@@ -291,8 +289,9 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                     else
                     {
-                        PipelineStageFlags stageFlags = _textureArrayRefs[segment.Binding].Stage.ConvertToPipelineStageFlags();
-                        _textureArrayRefs[segment.Binding].Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                        ref var arrayRef = ref _textureArrayRefs[segment.Binding];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
                     }
                 }
             }
@@ -311,8 +310,40 @@ namespace Ryujinx.Graphics.Vulkan
                     }
                     else
                     {
-                        PipelineStageFlags stageFlags = _imageArrayRefs[segment.Binding].Stage.ConvertToPipelineStageFlags();
-                        _imageArrayRefs[segment.Binding].Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                        ref var arrayRef = ref _imageArrayRefs[segment.Binding];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                    }
+                }
+            }
+
+            for (int setIndex = PipelineBase.DescriptorSetLayouts; setIndex < _program.BindingSegments.Length; setIndex++)
+            {
+                var bindingSegments = _program.BindingSegments[setIndex];
+
+                if (bindingSegments.Length == 0)
+                {
+                    continue;
+                }
+
+                ResourceBindingSegment segment = bindingSegments[0];
+
+                if (segment.IsArray)
+                {
+                    if (segment.Type == ResourceType.Texture ||
+                        segment.Type == ResourceType.Sampler ||
+                        segment.Type == ResourceType.TextureAndSampler ||
+                        segment.Type == ResourceType.BufferTexture)
+                    {
+                        ref var arrayRef = ref _textureArrayExtraRefs[setIndex - PipelineBase.DescriptorSetLayouts];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
+                    }
+                    else if (segment.Type == ResourceType.Image || segment.Type == ResourceType.BufferImage)
+                    {
+                        ref var arrayRef = ref _imageArrayExtraRefs[setIndex - PipelineBase.DescriptorSetLayouts];
+                        PipelineStageFlags stageFlags = arrayRef.Stage.ConvertToPipelineStageFlags();
+                        arrayRef.Array?.QueueWriteToReadBarriers(cbs, stageFlags);
                     }
                 }
             }
@@ -651,7 +682,14 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (_dirty.HasFlag(DirtyFlags.Texture))
             {
-                UpdateAndBind(cbs, program, PipelineBase.TextureSetIndex, pbp);
+                if (program.UpdateTexturesWithoutTemplate)
+                {
+                    UpdateAndBindTexturesWithoutTemplate(cbs, program, pbp);
+                }
+                else
+                {
+                    UpdateAndBind(cbs, program, PipelineBase.TextureSetIndex, pbp);
+                }
             }
 
             if (_dirty.HasFlag(DirtyFlags.Image))
@@ -885,31 +923,84 @@ namespace Ryujinx.Graphics.Vulkan
             _gd.Api.CmdBindDescriptorSets(cbs.CommandBuffer, pbp, _program.PipelineLayout, (uint)setIndex, 1, sets, 0, ReadOnlySpan<uint>.Empty);
         }
 
-        private unsafe void UpdateBuffers(
-            CommandBufferScoped cbs,
-            PipelineBindPoint pbp,
-            int baseBinding,
-            ReadOnlySpan<DescriptorBufferInfo> bufferInfo,
-            DescriptorType type)
+        private void UpdateAndBindTexturesWithoutTemplate(CommandBufferScoped cbs, ShaderCollection program, PipelineBindPoint pbp)
         {
-            if (bufferInfo.Length == 0)
+            int setIndex = PipelineBase.TextureSetIndex;
+            var bindingSegments = program.BindingSegments[setIndex];
+
+            if (bindingSegments.Length == 0)
             {
                 return;
             }
 
-            fixed (DescriptorBufferInfo* pBufferInfo = bufferInfo)
+            if (_updateDescriptorCacheCbIndex)
             {
-                var writeDescriptorSet = new WriteDescriptorSet
-                {
-                    SType = StructureType.WriteDescriptorSet,
-                    DstBinding = (uint)baseBinding,
-                    DescriptorType = type,
-                    DescriptorCount = (uint)bufferInfo.Length,
-                    PBufferInfo = pBufferInfo,
-                };
-
-                _gd.PushDescriptorApi.CmdPushDescriptorSet(cbs.CommandBuffer, pbp, _program.PipelineLayout, 0, 1, &writeDescriptorSet);
+                _updateDescriptorCacheCbIndex = false;
+                program.UpdateDescriptorCacheCommandBufferIndex(cbs.CommandBufferIndex);
             }
+
+            var dsc = program.GetNewDescriptorSetCollection(setIndex, out _).Get(cbs);
+
+            foreach (ResourceBindingSegment segment in bindingSegments)
+            {
+                int binding = segment.Binding;
+                int count = segment.Count;
+
+                if (!segment.IsArray)
+                {
+                    if (segment.Type != ResourceType.BufferTexture)
+                    {
+                        Span<DescriptorImageInfo> textures = _textures;
+
+                        for (int i = 0; i < count; i++)
+                        {
+                            ref var texture = ref textures[i];
+                            ref var refs = ref _textureRefs[binding + i];
+
+                            texture.ImageView = refs.View?.Get(cbs).Value ?? default;
+                            texture.Sampler = refs.Sampler?.Get(cbs).Value ?? default;
+
+                            if (texture.ImageView.Handle == 0)
+                            {
+                                texture.ImageView = _dummyTexture.GetImageView().Get(cbs).Value;
+                            }
+
+                            if (texture.Sampler.Handle == 0)
+                            {
+                                texture.Sampler = _dummySampler.GetSampler().Get(cbs).Value;
+                            }
+                        }
+
+                        dsc.UpdateImages(0, binding, textures[..count], DescriptorType.CombinedImageSampler);
+                    }
+                    else
+                    {
+                        Span<BufferView> bufferTextures = _bufferTextures;
+
+                        for (int i = 0; i < count; i++)
+                        {
+                            bufferTextures[i] = _bufferTextureRefs[binding + i]?.GetBufferView(cbs, false) ?? default;
+                        }
+
+                        dsc.UpdateBufferImages(0, binding, bufferTextures[..count], DescriptorType.UniformTexelBuffer);
+                    }
+                }
+                else
+                {
+                    if (segment.Type != ResourceType.BufferTexture)
+                    {
+                        dsc.UpdateImages(0, binding, _textureArrayRefs[binding].Array.GetImageInfos(_gd, cbs, _dummyTexture, _dummySampler), DescriptorType.CombinedImageSampler);
+                    }
+                    else
+                    {
+                        dsc.UpdateBufferImages(0, binding, _textureArrayRefs[binding].Array.GetBufferViews(cbs), DescriptorType.UniformTexelBuffer);
+                    }
+                }
+            }
+
+            var sets = dsc.GetSets();
+
+            _gd.Api.CmdBindDescriptorSets(cbs.CommandBuffer, pbp, _program.PipelineLayout, (uint)setIndex, 1, sets, 0, ReadOnlySpan<uint>.Empty);
         }
 
         [MethodImpl(MethodImplOptions.AggressiveInlining)]

src/Ryujinx.Graphics.Vulkan/Effects/FsrScalingFilter.cs

@@ -59,14 +59,14 @@ namespace Ryujinx.Graphics.Vulkan.Effects
             var scalingResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             var sharpeningResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 3)
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 4)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             _sampler = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));
 

src/Ryujinx.Graphics.Vulkan/Effects/FxaaPostProcessingEffect.cs

@@ -42,7 +42,7 @@ namespace Ryujinx.Graphics.Vulkan.Effects
             var resourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             _samplerLinear = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));
 

src/Ryujinx.Graphics.Vulkan/Effects/SmaaPostProcessingEffect.cs

@@ -81,20 +81,20 @@ namespace Ryujinx.Graphics.Vulkan.Effects
             var edgeResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             var blendResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 3)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 4)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             var neighbourResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 3)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             _samplerLinear = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));
 

src/Ryujinx.Graphics.Vulkan/FenceHolder.cs

@@ -10,12 +10,15 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
         private Fence _fence;
         private int _referenceCount;
+        private int _lock;
+        private readonly bool _concurrentWaitUnsupported;
         private bool _disposed;
 
-        public unsafe FenceHolder(Vk api, Device device)
+        public unsafe FenceHolder(Vk api, Device device, bool concurrentWaitUnsupported)
         {
             _api = api;
             _device = device;
+            _concurrentWaitUnsupported = concurrentWaitUnsupported;
 
             var fenceCreateInfo = new FenceCreateInfo
             {
@@ -47,6 +50,11 @@ namespace Ryujinx.Graphics.Vulkan
             }
             while (Interlocked.CompareExchange(ref _referenceCount, lastValue + 1, lastValue) != lastValue);
 
+            if (_concurrentWaitUnsupported)
+            {
+                AcquireLock();
+            }
+
             fence = _fence;
             return true;
         }
@@ -57,6 +65,16 @@ namespace Ryujinx.Graphics.Vulkan
             return _fence;
         }
 
+        public void PutLock()
+        {
+            Put();
+
+            if (_concurrentWaitUnsupported)
+            {
+                ReleaseLock();
+            }
+        }
+
         public void Put()
         {
             if (Interlocked.Decrement(ref _referenceCount) == 0)
@@ -66,24 +84,67 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private void AcquireLock()
+        {
+            while (!TryAcquireLock())
+            {
+                Thread.SpinWait(32);
+            }
+        }
+
+        private bool TryAcquireLock()
+        {
+            return Interlocked.Exchange(ref _lock, 1) == 0;
+        }
+
+        private void ReleaseLock()
+        {
+            Interlocked.Exchange(ref _lock, 0);
+        }
+
         public void Wait()
         {
-            Span<Fence> fences = stackalloc Fence[]
+            if (_concurrentWaitUnsupported)
             {
-                _fence,
-            };
+                AcquireLock();
 
-            FenceHelper.WaitAllIndefinitely(_api, _device, fences);
+                try
+                {
+                    FenceHelper.WaitAllIndefinitely(_api, _device, stackalloc Fence[] { _fence });
+                }
+                finally
+                {
+                    ReleaseLock();
+                }
+            }
+            else
+            {
+                FenceHelper.WaitAllIndefinitely(_api, _device, stackalloc Fence[] { _fence });
+            }
         }
 
         public bool IsSignaled()
         {
-            Span<Fence> fences = stackalloc Fence[]
+            if (_concurrentWaitUnsupported)
             {
-                _fence,
-            };
+                if (!TryAcquireLock())
+                {
+                    return false;
+                }
 
-            return FenceHelper.AllSignaled(_api, _device, fences);
+                try
+                {
+                    return FenceHelper.AllSignaled(_api, _device, stackalloc Fence[] { _fence });
+                }
+                finally
+                {
+                    ReleaseLock();
+                }
+            }
+            else
+            {
+                return FenceHelper.AllSignaled(_api, _device, stackalloc Fence[] { _fence });
+            }
         }
 
         public void Dispose()

src/Ryujinx.Graphics.Vulkan/FramebufferParams.cs

@@ -286,10 +286,23 @@ namespace Ryujinx.Graphics.Vulkan
 
             _depthStencil?.Storage?.QueueLoadOpBarrier(cbs, true);
 
-            gd.Barriers.Flush(cbs.CommandBuffer, false, null);
+            gd.Barriers.Flush(cbs, false, null, null);
         }
 
-        public (Auto<DisposableRenderPass> renderPass, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
+        public void AddStoreOpUsage()
+        {
+            if (_colors != null)
+            {
+                foreach (var color in _colors)
+                {
+                    color.Storage?.AddStoreOpUsage(false);
+                }
+            }
+
+            _depthStencil?.Storage?.AddStoreOpUsage(true);
+        }
+
+        public (RenderPassHolder rpHolder, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
             VulkanRenderer gd,
             Device device,
             CommandBufferScoped cbs)

src/Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -115,7 +115,7 @@ namespace Ryujinx.Graphics.Vulkan
             var strideChangeResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
-                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
+                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
 
             _programStrideChange = gd.CreateProgramWithMinimalLayout(new[]
             {
@@ -125,7 +125,7 @@ namespace Ryujinx.Graphics.Vulkan
             var colorCopyResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 0)
-                .Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
+                .Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
 
             _programColorCopyShortening = gd.CreateProgramWithMinimalLayout(new[]
             {
@@ -155,7 +155,7 @@ namespace Ryujinx.Graphics.Vulkan
             var convertD32S8ToD24S8ResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
-                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
+                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
 
             _programConvertD32S8ToD24S8 = gd.CreateProgramWithMinimalLayout(new[]
             {
@@ -165,7 +165,7 @@ namespace Ryujinx.Graphics.Vulkan
             var convertIndexBufferResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
-                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
+                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
 
             _programConvertIndexBuffer = gd.CreateProgramWithMinimalLayout(new[]
             {
@@ -175,7 +175,7 @@ namespace Ryujinx.Graphics.Vulkan
             var convertIndirectDataResourceLayout = new ResourceLayoutBuilder()
                 .Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
-                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2)
+                .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true)
                 .Add(ResourceStages.Compute, ResourceType.StorageBuffer, 3).Build();
 
             _programConvertIndirectData = gd.CreateProgramWithMinimalLayout(new[]

src/Ryujinx.Graphics.Vulkan/ImageArray.cs

@@ -2,11 +2,10 @@ using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
-using System.Diagnostics;
 
 namespace Ryujinx.Graphics.Vulkan
 {
-    class ImageArray : IImageArray
+    class ImageArray : ResourceArray, IImageArray
     {
         private readonly VulkanRenderer _gd;
 
@@ -25,19 +24,11 @@ namespace Ryujinx.Graphics.Vulkan
 
         private HashSet<TextureStorage> _storages;
 
-        private DescriptorSet[] _cachedDescriptorSets;
-
         private int _cachedCommandBufferIndex;
         private int _cachedSubmissionCount;
 
-        private ShaderCollection _cachedDscProgram;
-        private int _cachedDscSetIndex;
-        private int _cachedDscIndex;
-
         private readonly bool _isBuffer;
 
-        private int _bindCount;
-
         public ImageArray(VulkanRenderer gd, int size, bool isBuffer)
         {
             _gd = gd;
@@ -104,12 +95,7 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _cachedCommandBufferIndex = -1;
             _storages = null;
-            _cachedDescriptorSets = null;
-
-            if (_bindCount != 0)
-            {
-                _gd.PipelineInternal.ForceImageDirty();
-            }
+            SetDirty(_gd);
         }
 
         public void QueueWriteToReadBarriers(CommandBufferScoped cbs, PipelineStageFlags stageFlags)
@@ -195,7 +181,7 @@ namespace Ryujinx.Graphics.Vulkan
             int setIndex,
             TextureView dummyTexture)
         {
-            if (_cachedDescriptorSets != null)
+            if (TryGetCachedDescriptorSets(cbs, program, setIndex, out DescriptorSet[] sets))
             {
                 // We still need to ensure the current command buffer holds a reference to all used textures.
 
@@ -208,12 +194,9 @@ namespace Ryujinx.Graphics.Vulkan
                     GetBufferViews(cbs);
                 }
 
-                return _cachedDescriptorSets;
+                return sets;
             }
 
-            _cachedDscProgram?.ReleaseManualDescriptorSetCollection(_cachedDscSetIndex, _cachedDscIndex);
-            var dsc = program.GetNewManualDescriptorSetCollection(cbs.CommandBufferIndex, setIndex, out _cachedDscIndex).Get(cbs);
-
             DescriptorSetTemplate template = program.Templates[setIndex];
 
             DescriptorSetTemplateWriter tu = templateUpdater.Begin(template);
@@ -227,24 +210,9 @@ namespace Ryujinx.Graphics.Vulkan
                 tu.Push(GetBufferViews(cbs));
             }
 
-            var sets = dsc.GetSets();
             templateUpdater.Commit(_gd, device, sets[0]);
-            _cachedDescriptorSets = sets;
-            _cachedDscProgram = program;
-            _cachedDscSetIndex = setIndex;
 
             return sets;
         }
-
-        public void IncrementBindCount()
-        {
-            _bindCount++;
-        }
-
-        public void DecrementBindCount()
-        {
-            int newBindCount = --_bindCount;
-            Debug.Assert(newBindCount >= 0);
-        }
     }
 }

src/Ryujinx.Graphics.Vulkan/MultiFenceHolder.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Common.Memory;
 using Silk.NET.Vulkan;
 using System;
 
@@ -165,14 +166,15 @@ namespace Ryujinx.Graphics.Vulkan
         /// <returns>True if all fences were signaled before the timeout expired, false otherwise</returns>
         private bool WaitForFencesImpl(Vk api, Device device, int offset, int size, bool hasTimeout, ulong timeout)
         {
-            Span<FenceHolder> fenceHolders = new FenceHolder[CommandBufferPool.MaxCommandBuffers];
+            using SpanOwner<FenceHolder> fenceHoldersOwner = SpanOwner<FenceHolder>.Rent(CommandBufferPool.MaxCommandBuffers);
+            Span<FenceHolder> fenceHolders = fenceHoldersOwner.Span;
 
             int count = size != 0 ? GetOverlappingFences(fenceHolders, offset, size) : GetFences(fenceHolders);
             Span<Fence> fences = stackalloc Fence[count];
 
             int fenceCount = 0;
 
-            for (int i = 0; i < count; i++)
+            for (int i = 0; i < fences.Length; i++)
             {
                 if (fenceHolders[i].TryGet(out Fence fence))
                 {
@@ -194,18 +196,23 @@ namespace Ryujinx.Graphics.Vulkan
 
             bool signaled = true;
 
-            if (hasTimeout)
+            try
             {
-                signaled = FenceHelper.AllSignaled(api, device, fences[..fenceCount], timeout);
+                if (hasTimeout)
+                {
+                    signaled = FenceHelper.AllSignaled(api, device, fences[..fenceCount], timeout);
+                }
+                else
+                {
+                    FenceHelper.WaitAllIndefinitely(api, device, fences[..fenceCount]);
+                }
             }
-            else
+            finally
             {
-                FenceHelper.WaitAllIndefinitely(api, device, fences[..fenceCount]);
-            }
-
-            for (int i = 0; i < fenceCount; i++)
-            {
-                fenceHolders[i].Put();
+                for (int i = 0; i < fenceCount; i++)
+                {
+                    fenceHolders[i].PutLock();
+                }
             }
 
             return signaled;

src/Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -55,6 +55,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         protected FramebufferParams FramebufferParams;
         private Auto<DisposableFramebuffer> _framebuffer;
+        private RenderPassHolder _rpHolder;
         private Auto<DisposableRenderPass> _renderPass;
         private RenderPassHolder _nullRenderPass;
         private int _writtenAttachmentCount;
@@ -85,8 +86,6 @@ namespace Ryujinx.Graphics.Vulkan
         private bool _tfActive;
 
         private readonly PipelineColorBlendAttachmentState[] _storedBlend;
-
-        private ulong _drawCountSinceBarrier;
         public ulong DrawCount { get; private set; }
         public bool RenderPassActive { get; private set; }
 
@@ -105,7 +104,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.Api.CreatePipelineCache(device, pipelineCacheCreateInfo, null, out PipelineCache).ThrowOnError();
 
-            _descriptorSetUpdater = new DescriptorSetUpdater(gd, device, this);
+            _descriptorSetUpdater = new DescriptorSetUpdater(gd, device);
             _vertexBufferUpdater = new VertexBufferUpdater(gd);
 
             _transformFeedbackBuffers = new BufferState[Constants.MaxTransformFeedbackBuffers];
@@ -135,48 +134,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public unsafe void Barrier()
         {
-            if (_drawCountSinceBarrier != DrawCount)
-            {
-                _drawCountSinceBarrier = DrawCount;
-
-                // Barriers are not supported inside a render pass on Apple GPUs.
-                // As a workaround, end the render pass.
-                if (Gd.Vendor == Vendor.Apple)
-                {
-                    EndRenderPass();
-                }
-            }
-
-            MemoryBarrier memoryBarrier = new()
-            {
-                SType = StructureType.MemoryBarrier,
-                SrcAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
-                DstAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
-            };
-
-            PipelineStageFlags pipelineStageFlags = PipelineStageFlags.VertexShaderBit | PipelineStageFlags.FragmentShaderBit;
-
-            if (Gd.Capabilities.SupportsGeometryShader)
-            {
-                pipelineStageFlags |= PipelineStageFlags.GeometryShaderBit;
-            }
-
-            if (Gd.Capabilities.SupportsTessellationShader)
-            {
-                pipelineStageFlags |= PipelineStageFlags.TessellationControlShaderBit | PipelineStageFlags.TessellationEvaluationShaderBit;
-            }
-
-            Gd.Api.CmdPipelineBarrier(
-                CommandBuffer,
-                pipelineStageFlags,
-                pipelineStageFlags,
-                0,
-                1,
-                memoryBarrier,
-                0,
-                null,
-                0,
-                null);
+            Gd.Barriers.QueueMemoryBarrier();
         }
 
         public void ComputeBarrier()
@@ -203,6 +161,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public void BeginTransformFeedback(PrimitiveTopology topology)
         {
+            Gd.Barriers.EnableTfbBarriers(true);
             _tfEnabled = true;
         }
 
@@ -249,7 +208,7 @@ namespace Ryujinx.Graphics.Vulkan
                 CreateRenderPass();
             }
 
-            Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
+            Gd.Barriers.Flush(Cbs, RenderPassActive, _rpHolder, EndRenderPassDelegate);
 
             BeginRenderPass();
 
@@ -287,7 +246,7 @@ namespace Ryujinx.Graphics.Vulkan
                 CreateRenderPass();
             }
 
-            Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
+            Gd.Barriers.Flush(Cbs, RenderPassActive, _rpHolder, EndRenderPassDelegate);
 
             BeginRenderPass();
 
@@ -299,24 +258,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public unsafe void CommandBufferBarrier()
         {
-            MemoryBarrier memoryBarrier = new()
-            {
-                SType = StructureType.MemoryBarrier,
-                SrcAccessMask = BufferHolder.DefaultAccessFlags,
-                DstAccessMask = AccessFlags.IndirectCommandReadBit,
-            };
-
-            Gd.Api.CmdPipelineBarrier(
-                CommandBuffer,
-                PipelineStageFlags.AllCommandsBit,
-                PipelineStageFlags.DrawIndirectBit,
-                0,
-                1,
-                memoryBarrier,
-                0,
-                null,
-                0,
-                null);
+            Gd.Barriers.QueueCommandBufferBarrier();
         }
 
         public void CopyBuffer(BufferHandle source, BufferHandle destination, int srcOffset, int dstOffset, int size)
@@ -722,6 +664,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public void EndTransformFeedback()
         {
+            Gd.Barriers.EnableTfbBarriers(false);
             PauseTransformFeedbackInternal();
             _tfEnabled = false;
         }
@@ -1020,6 +963,13 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _newState.RasterizerDiscardEnable = discard;
             SignalStateChange();
+
+            if (!discard && Gd.IsQualcommProprietary)
+            {
+                // On Adreno, enabling rasterizer discard somehow corrupts the viewport state.
+                // Force it to be updated on next use to work around this bug.
+                DynamicState.ForceAllDirty();
+            }
         }
 
         public void SetRenderTargetColorMasks(ReadOnlySpan<uint> componentMask)
@@ -1401,24 +1351,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public unsafe void TextureBarrier()
         {
-            MemoryBarrier memoryBarrier = new()
-            {
-                SType = StructureType.MemoryBarrier,
-                SrcAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
-                DstAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
-            };
-
-            Gd.Api.CmdPipelineBarrier(
-                CommandBuffer,
-                PipelineStageFlags.FragmentShaderBit,
-                PipelineStageFlags.FragmentShaderBit,
-                0,
-                1,
-                memoryBarrier,
-                0,
-                null,
-                0,
-                null);
+            Gd.Barriers.QueueTextureBarrier();
         }
 
         public void TextureBarrierTiled()
@@ -1525,12 +1458,15 @@ namespace Ryujinx.Graphics.Vulkan
                 // Use the null framebuffer.
                 _nullRenderPass ??= new RenderPassHolder(Gd, Device, new RenderPassCacheKey(), FramebufferParams);
 
+                _rpHolder = _nullRenderPass;
                 _renderPass = _nullRenderPass.GetRenderPass();
                 _framebuffer = _nullRenderPass.GetFramebuffer(Gd, Cbs, FramebufferParams);
             }
             else
             {
-                (_renderPass, _framebuffer) = FramebufferParams.GetPassAndFramebuffer(Gd, Device, Cbs);
+                (_rpHolder, _framebuffer) = FramebufferParams.GetPassAndFramebuffer(Gd, Device, Cbs);
+
+                _renderPass = _rpHolder.GetRenderPass();
             }
         }
 
@@ -1557,7 +1493,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
+            Gd.Barriers.Flush(Cbs, _program, RenderPassActive, _rpHolder, EndRenderPassDelegate);
 
             _descriptorSetUpdater.UpdateAndBindDescriptorSets(Cbs, PipelineBindPoint.Compute);
         }
@@ -1622,7 +1558,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
+            Gd.Barriers.Flush(Cbs, _program, RenderPassActive, _rpHolder, EndRenderPassDelegate);
 
             _descriptorSetUpdater.UpdateAndBindDescriptorSets(Cbs, PipelineBindPoint.Graphics);
 
@@ -1701,6 +1637,8 @@ namespace Ryujinx.Graphics.Vulkan
         {
             if (RenderPassActive)
             {
+                FramebufferParams.AddStoreOpUsage();
+
                 PauseTransformFeedbackInternal();
                 Gd.Api.CmdEndRenderPass(CommandBuffer);
                 SignalRenderPassEnd();

src/Ryujinx.Graphics.Vulkan/PipelineConverter.cs

@@ -9,13 +9,6 @@ namespace Ryujinx.Graphics.Vulkan
 {
     static class PipelineConverter
     {
-        private const AccessFlags SubpassAccessMask =
-            AccessFlags.MemoryReadBit |
-            AccessFlags.MemoryWriteBit |
-            AccessFlags.ShaderReadBit |
-            AccessFlags.ColorAttachmentWriteBit |
-            AccessFlags.DepthStencilAttachmentWriteBit;
-
         public static unsafe DisposableRenderPass ToRenderPass(this ProgramPipelineState state, VulkanRenderer gd, Device device)
         {
             const int MaxAttachments = Constants.MaxRenderTargets + 1;
@@ -108,7 +101,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            var subpassDependency = CreateSubpassDependency();
+            var subpassDependency = CreateSubpassDependency(gd);
 
             fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
             {
@@ -129,29 +122,33 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        public static SubpassDependency CreateSubpassDependency()
+        public static SubpassDependency CreateSubpassDependency(VulkanRenderer gd)
         {
+            var (access, stages) = BarrierBatch.GetSubpassAccessSuperset(gd);
+
             return new SubpassDependency(
                 0,
                 0,
-                PipelineStageFlags.AllGraphicsBit,
-                PipelineStageFlags.AllGraphicsBit,
-                SubpassAccessMask,
-                SubpassAccessMask,
+                stages,
+                stages,
+                access,
+                access,
                 0);
         }
 
-        public unsafe static SubpassDependency2 CreateSubpassDependency2()
+        public unsafe static SubpassDependency2 CreateSubpassDependency2(VulkanRenderer gd)
         {
+            var (access, stages) = BarrierBatch.GetSubpassAccessSuperset(gd);
+
             return new SubpassDependency2(
                 StructureType.SubpassDependency2,
                 null,
                 0,
                 0,
-                PipelineStageFlags.AllGraphicsBit,
-                PipelineStageFlags.AllGraphicsBit,
-                SubpassAccessMask,
-                SubpassAccessMask,
+                stages,
+                stages,
+                access,
+                access,
                 0);
         }
 
@@ -180,9 +177,6 @@ namespace Ryujinx.Graphics.Vulkan
             pipeline.LogicOpEnable = state.LogicOpEnable;
             pipeline.LogicOp = state.LogicOp.Convert();
 
-            pipeline.MinDepthBounds = 0f; // Not implemented.
-            pipeline.MaxDepthBounds = 0f; // Not implemented.
-
             pipeline.PatchControlPoints = state.PatchControlPoints;
             pipeline.PolygonMode = PolygonMode.Fill; // Not implemented.
             pipeline.PrimitiveRestartEnable = state.PrimitiveRestartEnable;
@@ -208,17 +202,11 @@ namespace Ryujinx.Graphics.Vulkan
             pipeline.StencilFrontPassOp = state.StencilTest.FrontDpPass.Convert();
             pipeline.StencilFrontDepthFailOp = state.StencilTest.FrontDpFail.Convert();
             pipeline.StencilFrontCompareOp = state.StencilTest.FrontFunc.Convert();
-            pipeline.StencilFrontCompareMask = 0;
-            pipeline.StencilFrontWriteMask = 0;
-            pipeline.StencilFrontReference = 0;
 
             pipeline.StencilBackFailOp = state.StencilTest.BackSFail.Convert();
             pipeline.StencilBackPassOp = state.StencilTest.BackDpPass.Convert();
             pipeline.StencilBackDepthFailOp = state.StencilTest.BackDpFail.Convert();
             pipeline.StencilBackCompareOp = state.StencilTest.BackFunc.Convert();
-            pipeline.StencilBackCompareMask = 0;
-            pipeline.StencilBackWriteMask = 0;
-            pipeline.StencilBackReference = 0;
 
             pipeline.StencilTestEnable = state.StencilTest.TestEnable;
 

src/Ryujinx.Graphics.Vulkan/PipelineFull.cs

@@ -47,10 +47,11 @@ namespace Ryujinx.Graphics.Vulkan
                 return;
             }
 
-            if (componentMask != 0xf)
+            if (componentMask != 0xf || Gd.IsQualcommProprietary)
             {
                 // We can't use CmdClearAttachments if not writing all components,
                 // because on Vulkan, the pipeline state does not affect clears.
+                // On proprietary Adreno drivers, CmdClearAttachments appears to execute out of order, so it's better to not use it at all.
                 var dstTexture = FramebufferParams.GetColorView(index);
                 if (dstTexture == null)
                 {
@@ -87,10 +88,11 @@ namespace Ryujinx.Graphics.Vulkan
                 return;
             }
 
-            if (stencilMask != 0 && stencilMask != 0xff)
+            if ((stencilMask != 0 && stencilMask != 0xff) || Gd.IsQualcommProprietary)
             {
                 // 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.
+                // On proprietary Adreno drivers, CmdClearAttachments appears to execute out of order, so it's better to not use it at all.
                 var dstTexture = FramebufferParams.GetDepthStencilView();
                 if (dstTexture == null)
                 {
@@ -255,7 +257,7 @@ namespace Ryujinx.Graphics.Vulkan
                 PreloadCbs = null;
             }
 
-            Gd.Barriers.Flush(Cbs.CommandBuffer, false, null);
+            Gd.Barriers.Flush(Cbs, false, null, null);
             CommandBuffer = (Cbs = Gd.CommandBufferPool.ReturnAndRent(Cbs)).CommandBuffer;
             Gd.RegisterFlush();
 

src/Ryujinx.Graphics.Vulkan/PipelineLayoutCacheEntry.cs

@@ -3,6 +3,7 @@ using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
 using System.Collections.ObjectModel;
+using System.Diagnostics;
 using System.Runtime.InteropServices;
 
 namespace Ryujinx.Graphics.Vulkan
@@ -15,6 +16,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
 
         public DescriptorSetLayout[] DescriptorSetLayouts { get; }
+        public bool[] DescriptorSetLayoutsUpdateAfterBind { get; }
         public PipelineLayout PipelineLayout { get; }
 
         private readonly int[] _consumedDescriptorsPerSet;
@@ -31,20 +33,37 @@ namespace Ryujinx.Graphics.Vulkan
         private struct ManualDescriptorSetEntry
         {
             public Auto<DescriptorSetCollection> DescriptorSet;
-            public int CbIndex;
-            public int CbSubmissionCount;
+            public uint CbRefMask;
             public bool InUse;
 
-            public ManualDescriptorSetEntry(Auto<DescriptorSetCollection> descriptorSet, int cbIndex, int cbSubmissionCount, bool inUse)
+            public ManualDescriptorSetEntry(Auto<DescriptorSetCollection> descriptorSet, int cbIndex)
             {
                 DescriptorSet = descriptorSet;
-                CbIndex = cbIndex;
-                CbSubmissionCount = cbSubmissionCount;
-                InUse = inUse;
+                CbRefMask = 1u << cbIndex;
+                InUse = true;
+            }
+        }
+
+        private readonly struct PendingManualDsConsumption
+        {
+            public FenceHolder Fence { get; }
+            public int CommandBufferIndex { get; }
+            public int SetIndex { get; }
+            public int CacheIndex { get; }
+
+            public PendingManualDsConsumption(FenceHolder fence, int commandBufferIndex, int setIndex, int cacheIndex)
+            {
+                Fence = fence;
+                CommandBufferIndex = commandBufferIndex;
+                SetIndex = setIndex;
+                CacheIndex = cacheIndex;
+                fence.Get();
             }
         }
 
         private readonly List<ManualDescriptorSetEntry>[] _manualDsCache;
+        private readonly Queue<PendingManualDsConsumption> _pendingManualDsConsumptions;
+        private readonly Queue<int>[] _freeManualDsCacheEntries;
 
         private readonly Dictionary<long, DescriptorSetTemplate> _pdTemplates;
         private readonly ResourceDescriptorCollection _pdDescriptors;
@@ -70,6 +89,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             _dsCacheCursor = new int[setsCount];
             _manualDsCache = new List<ManualDescriptorSetEntry>[setsCount];
+            _pendingManualDsConsumptions = new Queue<PendingManualDsConsumption>();
+            _freeManualDsCacheEntries = new Queue<int>[setsCount];
         }
 
         public PipelineLayoutCacheEntry(
@@ -78,7 +99,11 @@ namespace Ryujinx.Graphics.Vulkan
             ReadOnlyCollection<ResourceDescriptorCollection> setDescriptors,
             bool usePushDescriptors) : this(gd, device, setDescriptors.Count)
         {
-            (DescriptorSetLayouts, PipelineLayout) = PipelineLayoutFactory.Create(gd, device, setDescriptors, usePushDescriptors);
+            ResourceLayouts layouts = PipelineLayoutFactory.Create(gd, device, setDescriptors, usePushDescriptors);
+
+            DescriptorSetLayouts = layouts.DescriptorSetLayouts;
+            DescriptorSetLayoutsUpdateAfterBind = layouts.DescriptorSetLayoutsUpdateAfterBind;
+            PipelineLayout = layouts.PipelineLayout;
 
             _consumedDescriptorsPerSet = new int[setDescriptors.Count];
             _poolSizes = new DescriptorPoolSize[setDescriptors.Count][];
@@ -133,7 +158,7 @@ namespace Ryujinx.Graphics.Vulkan
                     _poolSizes[setIndex],
                     setIndex,
                     _consumedDescriptorsPerSet[setIndex],
-                    false);
+                    DescriptorSetLayoutsUpdateAfterBind[setIndex]);
 
                 list.Add(dsc);
                 isNew = true;
@@ -144,49 +169,99 @@ namespace Ryujinx.Graphics.Vulkan
             return list[index];
         }
 
-        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(int commandBufferIndex, int setIndex, out int cacheIndex)
+        public Auto<DescriptorSetCollection> GetNewManualDescriptorSetCollection(CommandBufferScoped cbs, int setIndex, out int cacheIndex)
         {
-            int submissionCount = _gd.CommandBufferPool.GetSubmissionCount(commandBufferIndex);
+            FreeCompletedManualDescriptorSets();
 
             var list = _manualDsCache[setIndex] ??= new();
             var span = CollectionsMarshal.AsSpan(list);
 
-            for (int index = 0; index < span.Length; index++)
+            Queue<int> freeQueue = _freeManualDsCacheEntries[setIndex];
+
+            // Do we have at least one freed descriptor set? If so, just use that.
+            if (freeQueue != null && freeQueue.TryDequeue(out int freeIndex))
             {
-                ref ManualDescriptorSetEntry entry = ref span[index];
+                ref ManualDescriptorSetEntry entry = ref span[freeIndex];
 
-                if (!entry.InUse && (entry.CbIndex != commandBufferIndex || entry.CbSubmissionCount != submissionCount))
-                {
-                    entry.InUse = true;
-                    entry.CbIndex = commandBufferIndex;
-                    entry.CbSubmissionCount = submissionCount;
+                Debug.Assert(!entry.InUse && entry.CbRefMask == 0);
 
-                    cacheIndex = index;
+                entry.InUse = true;
+                entry.CbRefMask = 1u << cbs.CommandBufferIndex;
+                cacheIndex = freeIndex;
 
-                    return entry.DescriptorSet;
-                }
+                _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, freeIndex));
+
+                return entry.DescriptorSet;
             }
 
+            // Otherwise create a new descriptor set, and add to our pending queue for command buffer consumption tracking.
             var dsc = _descriptorSetManager.AllocateDescriptorSet(
                 _gd.Api,
                 DescriptorSetLayouts[setIndex],
                 _poolSizes[setIndex],
                 setIndex,
                 _consumedDescriptorsPerSet[setIndex],
-                false);
+                DescriptorSetLayoutsUpdateAfterBind[setIndex]);
 
             cacheIndex = list.Count;
-            list.Add(new ManualDescriptorSetEntry(dsc, commandBufferIndex, submissionCount, inUse: true));
+            list.Add(new ManualDescriptorSetEntry(dsc, cbs.CommandBufferIndex));
+            _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, cacheIndex));
 
             return dsc;
         }
 
+        public void UpdateManualDescriptorSetCollectionOwnership(CommandBufferScoped cbs, int setIndex, int cacheIndex)
+        {
+            FreeCompletedManualDescriptorSets();
+
+            var list = _manualDsCache[setIndex];
+            var span = CollectionsMarshal.AsSpan(list);
+            ref var entry = ref span[cacheIndex];
+
+            uint cbMask = 1u << cbs.CommandBufferIndex;
+
+            if ((entry.CbRefMask & cbMask) == 0)
+            {
+                entry.CbRefMask |= cbMask;
+
+                _pendingManualDsConsumptions.Enqueue(new PendingManualDsConsumption(cbs.GetFence(), cbs.CommandBufferIndex, setIndex, cacheIndex));
+            }
+        }
+
+        private void FreeCompletedManualDescriptorSets()
+        {
+            FenceHolder signalledFence = null;
+            while (_pendingManualDsConsumptions.TryPeek(out var pds) && (pds.Fence == signalledFence || pds.Fence.IsSignaled()))
+            {
+                signalledFence = pds.Fence; // Already checked - don't need to do it again.
+                var dequeued = _pendingManualDsConsumptions.Dequeue();
+                Debug.Assert(dequeued.Fence == pds.Fence);
+                pds.Fence.Put();
+
+                var span = CollectionsMarshal.AsSpan(_manualDsCache[dequeued.SetIndex]);
+                ref var entry = ref span[dequeued.CacheIndex];
+                entry.CbRefMask &= ~(1u << dequeued.CommandBufferIndex);
+
+                if (!entry.InUse && entry.CbRefMask == 0)
+                {
+                    // If not in use by any array, and not bound to any command buffer, the descriptor set can be re-used immediately.
+                    (_freeManualDsCacheEntries[dequeued.SetIndex] ??= new()).Enqueue(dequeued.CacheIndex);
+                }
+            }
+        }
+
         public void ReleaseManualDescriptorSetCollection(int setIndex, int cacheIndex)
         {
             var list = _manualDsCache[setIndex];
             var span = CollectionsMarshal.AsSpan(list);
 
             span[cacheIndex].InUse = false;
+
+            if (span[cacheIndex].CbRefMask == 0)
+            {
+                // This is no longer in use by any array, so if not bound to any command buffer, the descriptor set can be re-used immediately.
+                (_freeManualDsCacheEntries[setIndex] ??= new()).Enqueue(cacheIndex);
+            }
         }
 
         private static Span<DescriptorPoolSize> GetDescriptorPoolSizes(Span<DescriptorPoolSize> output, ResourceDescriptorCollection setDescriptor, uint multiplier)
@@ -291,6 +366,11 @@ namespace Ryujinx.Graphics.Vulkan
                     _gd.Api.DestroyDescriptorSetLayout(_device, DescriptorSetLayouts[i], null);
                 }
 
+                while (_pendingManualDsConsumptions.TryDequeue(out var pds))
+                {
+                    pds.Fence.Put();
+                }
+
                 _descriptorSetManager.Dispose();
             }
         }

src/Ryujinx.Graphics.Vulkan/PipelineLayoutFactory.cs

@@ -1,18 +1,23 @@
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
+using System;
 using System.Collections.ObjectModel;
 
 namespace Ryujinx.Graphics.Vulkan
 {
+    record struct ResourceLayouts(DescriptorSetLayout[] DescriptorSetLayouts, bool[] DescriptorSetLayoutsUpdateAfterBind, PipelineLayout PipelineLayout);
+
     static class PipelineLayoutFactory
     {
-        public static unsafe (DescriptorSetLayout[], PipelineLayout) Create(
+        public static unsafe ResourceLayouts Create(
             VulkanRenderer gd,
             Device device,
             ReadOnlyCollection<ResourceDescriptorCollection> setDescriptors,
             bool usePushDescriptors)
         {
             DescriptorSetLayout[] layouts = new DescriptorSetLayout[setDescriptors.Count];
+            bool[] updateAfterBindFlags = new bool[setDescriptors.Count];
 
             bool isMoltenVk = gd.IsMoltenVk;
 
@@ -32,10 +37,11 @@ namespace Ryujinx.Graphics.Vulkan
 
                 DescriptorSetLayoutBinding[] layoutBindings = new DescriptorSetLayoutBinding[rdc.Descriptors.Count];
 
+                bool hasArray = false;
+
                 for (int descIndex = 0; descIndex < rdc.Descriptors.Count; descIndex++)
                 {
                     ResourceDescriptor descriptor = rdc.Descriptors[descIndex];
-
                     ResourceStages stages = descriptor.Stages;
 
                     if (descriptor.Type == ResourceType.StorageBuffer && isMoltenVk)
@@ -52,16 +58,37 @@ namespace Ryujinx.Graphics.Vulkan
                         DescriptorCount = (uint)descriptor.Count,
                         StageFlags = stages.Convert(),
                     };
+
+                    if (descriptor.Count > 1)
+                    {
+                        hasArray = true;
+                    }
                 }
 
                 fixed (DescriptorSetLayoutBinding* pLayoutBindings = layoutBindings)
                 {
+                    DescriptorSetLayoutCreateFlags flags = DescriptorSetLayoutCreateFlags.None;
+
+                    if (usePushDescriptors && setIndex == 0)
+                    {
+                        flags = DescriptorSetLayoutCreateFlags.PushDescriptorBitKhr;
+                    }
+
+                    if (gd.Vendor == Vendor.Intel && hasArray)
+                    {
+                        // Some vendors (like Intel) have low per-stage limits.
+                        // We must set the flag if we exceed those limits.
+                        flags |= DescriptorSetLayoutCreateFlags.UpdateAfterBindPoolBit;
+
+                        updateAfterBindFlags[setIndex] = true;
+                    }
+
                     var descriptorSetLayoutCreateInfo = new DescriptorSetLayoutCreateInfo
                     {
                         SType = StructureType.DescriptorSetLayoutCreateInfo,
                         PBindings = pLayoutBindings,
                         BindingCount = (uint)layoutBindings.Length,
-                        Flags = usePushDescriptors && setIndex == 0 ? DescriptorSetLayoutCreateFlags.PushDescriptorBitKhr : DescriptorSetLayoutCreateFlags.None,
+                        Flags = flags,
                     };
 
                     gd.Api.CreateDescriptorSetLayout(device, descriptorSetLayoutCreateInfo, null, out layouts[setIndex]).ThrowOnError();
@@ -82,7 +109,7 @@ namespace Ryujinx.Graphics.Vulkan
                 gd.Api.CreatePipelineLayout(device, &pipelineLayoutCreateInfo, null, out layout).ThrowOnError();
             }
 
-            return (layouts, layout);
+            return new ResourceLayouts(layouts, updateAfterBindFlags, layout);
         }
     }
 }

src/Ryujinx.Graphics.Vulkan/PipelineState.cs

@@ -71,244 +71,232 @@ namespace Ryujinx.Graphics.Vulkan
             set => Internal.Id4 = (Internal.Id4 & 0xFFFFFFFF) | ((ulong)value << 32);
         }
 
-        public float MinDepthBounds
-        {
-            readonly get => BitConverter.Int32BitsToSingle((int)((Internal.Id5 >> 0) & 0xFFFFFFFF));
-            set => Internal.Id5 = (Internal.Id5 & 0xFFFFFFFF00000000) | ((ulong)(uint)BitConverter.SingleToInt32Bits(value) << 0);
-        }
-
-        public float MaxDepthBounds
-        {
-            readonly get => BitConverter.Int32BitsToSingle((int)((Internal.Id5 >> 32) & 0xFFFFFFFF));
-            set => Internal.Id5 = (Internal.Id5 & 0xFFFFFFFF) | ((ulong)(uint)BitConverter.SingleToInt32Bits(value) << 32);
-        }
-
         public PolygonMode PolygonMode
         {
-            readonly get => (PolygonMode)((Internal.Id6 >> 0) & 0x3FFFFFFF);
-            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFC0000000) | ((ulong)value << 0);
+            readonly get => (PolygonMode)((Internal.Id5 >> 0) & 0x3FFFFFFF);
+            set => Internal.Id5 = (Internal.Id5 & 0xFFFFFFFFC0000000) | ((ulong)value << 0);
         }
 
         public uint StagesCount
         {
-            readonly get => (byte)((Internal.Id6 >> 30) & 0xFF);
-            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFC03FFFFFFF) | ((ulong)value << 30);
+            readonly get => (byte)((Internal.Id5 >> 30) & 0xFF);
+            set => Internal.Id5 = (Internal.Id5 & 0xFFFFFFC03FFFFFFF) | ((ulong)value << 30);
         }
 
         public uint VertexAttributeDescriptionsCount
         {
-            readonly get => (byte)((Internal.Id6 >> 38) & 0xFF);
-            set => Internal.Id6 = (Internal.Id6 & 0xFFFFC03FFFFFFFFF) | ((ulong)value << 38);
+            readonly get => (byte)((Internal.Id5 >> 38) & 0xFF);
+            set => Internal.Id5 = (Internal.Id5 & 0xFFFFC03FFFFFFFFF) | ((ulong)value << 38);
         }
 
         public uint VertexBindingDescriptionsCount
         {
-            readonly get => (byte)((Internal.Id6 >> 46) & 0xFF);
-            set => Internal.Id6 = (Internal.Id6 & 0xFFC03FFFFFFFFFFF) | ((ulong)value << 46);
+            readonly get => (byte)((Internal.Id5 >> 46) & 0xFF);
+            set => Internal.Id5 = (Internal.Id5 & 0xFFC03FFFFFFFFFFF) | ((ulong)value << 46);
         }
 
         public uint ViewportsCount
         {
-            readonly get => (byte)((Internal.Id6 >> 54) & 0xFF);
-            set => Internal.Id6 = (Internal.Id6 & 0xC03FFFFFFFFFFFFF) | ((ulong)value << 54);
+            readonly get => (byte)((Internal.Id5 >> 54) & 0xFF);
+            set => Internal.Id5 = (Internal.Id5 & 0xC03FFFFFFFFFFFFF) | ((ulong)value << 54);
         }
 
         public uint ScissorsCount
         {
-            readonly get => (byte)((Internal.Id7 >> 0) & 0xFF);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFFFFFFF00) | ((ulong)value << 0);
+            readonly get => (byte)((Internal.Id6 >> 0) & 0xFF);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFFFFFFF00) | ((ulong)value << 0);
         }
 
         public uint ColorBlendAttachmentStateCount
         {
-            readonly get => (byte)((Internal.Id7 >> 8) & 0xFF);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFFFFF00FF) | ((ulong)value << 8);
+            readonly get => (byte)((Internal.Id6 >> 8) & 0xFF);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFFFFF00FF) | ((ulong)value << 8);
         }
 
         public PrimitiveTopology Topology
         {
-            readonly get => (PrimitiveTopology)((Internal.Id7 >> 16) & 0xF);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFFFF0FFFF) | ((ulong)value << 16);
+            readonly get => (PrimitiveTopology)((Internal.Id6 >> 16) & 0xF);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFFFF0FFFF) | ((ulong)value << 16);
         }
 
         public LogicOp LogicOp
         {
-            readonly get => (LogicOp)((Internal.Id7 >> 20) & 0xF);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFFF0FFFFF) | ((ulong)value << 20);
+            readonly get => (LogicOp)((Internal.Id6 >> 20) & 0xF);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFFF0FFFFF) | ((ulong)value << 20);
         }
 
         public CompareOp DepthCompareOp
         {
-            readonly get => (CompareOp)((Internal.Id7 >> 24) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFF8FFFFFF) | ((ulong)value << 24);
+            readonly get => (CompareOp)((Internal.Id6 >> 24) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFF8FFFFFF) | ((ulong)value << 24);
         }
 
         public StencilOp StencilFrontFailOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 27) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFFC7FFFFFF) | ((ulong)value << 27);
+            readonly get => (StencilOp)((Internal.Id6 >> 27) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFFC7FFFFFF) | ((ulong)value << 27);
         }
 
         public StencilOp StencilFrontPassOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 30) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFE3FFFFFFF) | ((ulong)value << 30);
+            readonly get => (StencilOp)((Internal.Id6 >> 30) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFFE3FFFFFFF) | ((ulong)value << 30);
         }
 
         public StencilOp StencilFrontDepthFailOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 33) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFF1FFFFFFFF) | ((ulong)value << 33);
+            readonly get => (StencilOp)((Internal.Id6 >> 33) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFFF1FFFFFFFF) | ((ulong)value << 33);
         }
 
         public CompareOp StencilFrontCompareOp
         {
-            readonly get => (CompareOp)((Internal.Id7 >> 36) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFF8FFFFFFFFF) | ((ulong)value << 36);
+            readonly get => (CompareOp)((Internal.Id6 >> 36) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFF8FFFFFFFFF) | ((ulong)value << 36);
         }
 
         public StencilOp StencilBackFailOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 39) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFC7FFFFFFFFF) | ((ulong)value << 39);
+            readonly get => (StencilOp)((Internal.Id6 >> 39) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFFC7FFFFFFFFF) | ((ulong)value << 39);
         }
 
         public StencilOp StencilBackPassOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 42) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFFE3FFFFFFFFFF) | ((ulong)value << 42);
+            readonly get => (StencilOp)((Internal.Id6 >> 42) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFFE3FFFFFFFFFF) | ((ulong)value << 42);
         }
 
         public StencilOp StencilBackDepthFailOp
         {
-            readonly get => (StencilOp)((Internal.Id7 >> 45) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFFF1FFFFFFFFFFF) | ((ulong)value << 45);
+            readonly get => (StencilOp)((Internal.Id6 >> 45) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFFF1FFFFFFFFFFF) | ((ulong)value << 45);
         }
 
         public CompareOp StencilBackCompareOp
         {
-            readonly get => (CompareOp)((Internal.Id7 >> 48) & 0x7);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFF8FFFFFFFFFFFF) | ((ulong)value << 48);
+            readonly get => (CompareOp)((Internal.Id6 >> 48) & 0x7);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFF8FFFFFFFFFFFF) | ((ulong)value << 48);
         }
 
         public CullModeFlags CullMode
         {
-            readonly get => (CullModeFlags)((Internal.Id7 >> 51) & 0x3);
-            set => Internal.Id7 = (Internal.Id7 & 0xFFE7FFFFFFFFFFFF) | ((ulong)value << 51);
+            readonly get => (CullModeFlags)((Internal.Id6 >> 51) & 0x3);
+            set => Internal.Id6 = (Internal.Id6 & 0xFFE7FFFFFFFFFFFF) | ((ulong)value << 51);
         }
 
         public bool PrimitiveRestartEnable
         {
-            readonly get => ((Internal.Id7 >> 53) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xFFDFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 53);
+            readonly get => ((Internal.Id6 >> 53) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xFFDFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 53);
         }
 
         public bool DepthClampEnable
         {
-            readonly get => ((Internal.Id7 >> 54) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xFFBFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 54);
+            readonly get => ((Internal.Id6 >> 54) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xFFBFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 54);
         }
 
         public bool RasterizerDiscardEnable
         {
-            readonly get => ((Internal.Id7 >> 55) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xFF7FFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 55);
+            readonly get => ((Internal.Id6 >> 55) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xFF7FFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 55);
         }
 
         public FrontFace FrontFace
         {
-            readonly get => (FrontFace)((Internal.Id7 >> 56) & 0x1);
-            set => Internal.Id7 = (Internal.Id7 & 0xFEFFFFFFFFFFFFFF) | ((ulong)value << 56);
+            readonly get => (FrontFace)((Internal.Id6 >> 56) & 0x1);
+            set => Internal.Id6 = (Internal.Id6 & 0xFEFFFFFFFFFFFFFF) | ((ulong)value << 56);
         }
 
         public bool DepthBiasEnable
         {
-            readonly get => ((Internal.Id7 >> 57) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xFDFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 57);
+            readonly get => ((Internal.Id6 >> 57) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xFDFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 57);
         }
 
         public bool DepthTestEnable
         {
-            readonly get => ((Internal.Id7 >> 58) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xFBFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 58);
+            readonly get => ((Internal.Id6 >> 58) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xFBFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 58);
         }
 
         public bool DepthWriteEnable
         {
-            readonly get => ((Internal.Id7 >> 59) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xF7FFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 59);
+            readonly get => ((Internal.Id6 >> 59) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xF7FFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 59);
         }
 
         public bool DepthBoundsTestEnable
         {
-            readonly get => ((Internal.Id7 >> 60) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xEFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 60);
+            readonly get => ((Internal.Id6 >> 60) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xEFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 60);
         }
 
         public bool StencilTestEnable
         {
-            readonly get => ((Internal.Id7 >> 61) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xDFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 61);
+            readonly get => ((Internal.Id6 >> 61) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xDFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 61);
         }
 
         public bool LogicOpEnable
         {
-            readonly get => ((Internal.Id7 >> 62) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0xBFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 62);
+            readonly get => ((Internal.Id6 >> 62) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0xBFFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 62);
         }
 
         public bool HasDepthStencil
         {
-            readonly get => ((Internal.Id7 >> 63) & 0x1) != 0UL;
-            set => Internal.Id7 = (Internal.Id7 & 0x7FFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 63);
+            readonly get => ((Internal.Id6 >> 63) & 0x1) != 0UL;
+            set => Internal.Id6 = (Internal.Id6 & 0x7FFFFFFFFFFFFFFF) | ((value ? 1UL : 0UL) << 63);
         }
 
         public uint PatchControlPoints
         {
-            readonly get => (uint)((Internal.Id8 >> 0) & 0xFFFFFFFF);
-            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFF00000000) | ((ulong)value << 0);
+            readonly get => (uint)((Internal.Id7 >> 0) & 0xFFFFFFFF);
+            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFF00000000) | ((ulong)value << 0);
         }
 
         public uint SamplesCount
         {
-            readonly get => (uint)((Internal.Id8 >> 32) & 0xFFFFFFFF);
-            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFF) | ((ulong)value << 32);
+            readonly get => (uint)((Internal.Id7 >> 32) & 0xFFFFFFFF);
+            set => Internal.Id7 = (Internal.Id7 & 0xFFFFFFFF) | ((ulong)value << 32);
         }
 
         public bool AlphaToCoverageEnable
         {
-            readonly get => ((Internal.Id9 >> 0) & 0x1) != 0UL;
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFFE) | ((value ? 1UL : 0UL) << 0);
+            readonly get => ((Internal.Id8 >> 0) & 0x1) != 0UL;
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFFE) | ((value ? 1UL : 0UL) << 0);
         }
 
         public bool AlphaToOneEnable
         {
-            readonly get => ((Internal.Id9 >> 1) & 0x1) != 0UL;
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFFD) | ((value ? 1UL : 0UL) << 1);
+            readonly get => ((Internal.Id8 >> 1) & 0x1) != 0UL;
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFFD) | ((value ? 1UL : 0UL) << 1);
         }
 
         public bool AdvancedBlendSrcPreMultiplied
         {
-            readonly get => ((Internal.Id9 >> 2) & 0x1) != 0UL;
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFFB) | ((value ? 1UL : 0UL) << 2);
+            readonly get => ((Internal.Id8 >> 2) & 0x1) != 0UL;
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFFB) | ((value ? 1UL : 0UL) << 2);
         }
 
         public bool AdvancedBlendDstPreMultiplied
         {
-            readonly get => ((Internal.Id9 >> 3) & 0x1) != 0UL;
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFF7) | ((value ? 1UL : 0UL) << 3);
+            readonly get => ((Internal.Id8 >> 3) & 0x1) != 0UL;
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFF7) | ((value ? 1UL : 0UL) << 3);
         }
 
         public BlendOverlapEXT AdvancedBlendOverlap
         {
-            readonly get => (BlendOverlapEXT)((Internal.Id9 >> 4) & 0x3);
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFCF) | ((ulong)value << 4);
+            readonly get => (BlendOverlapEXT)((Internal.Id8 >> 4) & 0x3);
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFCF) | ((ulong)value << 4);
         }
 
         public bool DepthMode
         {
-            readonly get => ((Internal.Id9 >> 6) & 0x1) != 0UL;
-            set => Internal.Id9 = (Internal.Id9 & 0xFFFFFFFFFFFFFFBF) | ((value ? 1UL : 0UL) << 6);
+            readonly get => ((Internal.Id8 >> 6) & 0x1) != 0UL;
+            set => Internal.Id8 = (Internal.Id8 & 0xFFFFFFFFFFFFFFBF) | ((value ? 1UL : 0UL) << 6);
         }
 
         public bool HasTessellationControlShader;
@@ -408,8 +396,6 @@ namespace Ryujinx.Graphics.Vulkan
             fixed (VertexInputAttributeDescription* pVertexAttributeDescriptions = &Internal.VertexAttributeDescriptions[0])
             fixed (VertexInputAttributeDescription* pVertexAttributeDescriptions2 = &_vertexAttributeDescriptions2[0])
             fixed (VertexInputBindingDescription* pVertexBindingDescriptions = &Internal.VertexBindingDescriptions[0])
-            fixed (Viewport* pViewports = &Internal.Viewports[0])
-            fixed (Rect2D* pScissors = &Internal.Scissors[0])
             fixed (PipelineColorBlendAttachmentState* pColorBlendAttachmentState = &Internal.ColorBlendAttachmentState[0])
             {
                 var vertexInputState = new PipelineVertexInputStateCreateInfo
@@ -453,7 +439,7 @@ namespace Ryujinx.Graphics.Vulkan
                 {
                     SType = StructureType.PipelineInputAssemblyStateCreateInfo,
                     PrimitiveRestartEnable = primitiveRestartEnable,
-                    Topology = Topology,
+                    Topology = HasTessellationControlShader ? PrimitiveTopology.PatchList : Topology,
                 };
 
                 var tessellationState = new PipelineTessellationStateCreateInfo
@@ -472,18 +458,13 @@ namespace Ryujinx.Graphics.Vulkan
                     CullMode = CullMode,
                     FrontFace = FrontFace,
                     DepthBiasEnable = DepthBiasEnable,
-                    DepthBiasClamp = DepthBiasClamp,
-                    DepthBiasConstantFactor = DepthBiasConstantFactor,
-                    DepthBiasSlopeFactor = DepthBiasSlopeFactor,
                 };
 
                 var viewportState = new PipelineViewportStateCreateInfo
                 {
                     SType = StructureType.PipelineViewportStateCreateInfo,
                     ViewportCount = ViewportsCount,
-                    PViewports = pViewports,
                     ScissorCount = ScissorsCount,
-                    PScissors = pScissors,
                 };
 
                 if (gd.Capabilities.SupportsDepthClipControl)
@@ -511,19 +492,13 @@ namespace Ryujinx.Graphics.Vulkan
                     StencilFrontFailOp,
                     StencilFrontPassOp,
                     StencilFrontDepthFailOp,
-                    StencilFrontCompareOp,
-                    StencilFrontCompareMask,
-                    StencilFrontWriteMask,
-                    StencilFrontReference);
+                    StencilFrontCompareOp);
 
                 var stencilBack = new StencilOpState(
                     StencilBackFailOp,
                     StencilBackPassOp,
                     StencilBackDepthFailOp,
-                    StencilBackCompareOp,
-                    StencilBackCompareMask,
-                    StencilBackWriteMask,
-                    StencilBackReference);
+                    StencilBackCompareOp);
 
                 var depthStencilState = new PipelineDepthStencilStateCreateInfo
                 {
@@ -531,12 +506,10 @@ namespace Ryujinx.Graphics.Vulkan
                     DepthTestEnable = DepthTestEnable,
                     DepthWriteEnable = DepthWriteEnable,
                     DepthCompareOp = DepthCompareOp,
-                    DepthBoundsTestEnable = DepthBoundsTestEnable,
+                    DepthBoundsTestEnable = false,
                     StencilTestEnable = StencilTestEnable,
                     Front = stencilFront,
                     Back = stencilBack,
-                    MinDepthBounds = MinDepthBounds,
-                    MaxDepthBounds = MaxDepthBounds,
                 };
 
                 uint blendEnables = 0;
@@ -591,22 +564,21 @@ namespace Ryujinx.Graphics.Vulkan
                 }
 
                 bool supportsExtDynamicState = gd.Capabilities.SupportsExtendedDynamicState;
-                int dynamicStatesCount = supportsExtDynamicState ? 9 : 8;
+                int dynamicStatesCount = supportsExtDynamicState ? 8 : 7;
 
                 DynamicState* dynamicStates = stackalloc DynamicState[dynamicStatesCount];
 
                 dynamicStates[0] = DynamicState.Viewport;
                 dynamicStates[1] = DynamicState.Scissor;
                 dynamicStates[2] = DynamicState.DepthBias;
-                dynamicStates[3] = DynamicState.DepthBounds;
-                dynamicStates[4] = DynamicState.StencilCompareMask;
-                dynamicStates[5] = DynamicState.StencilWriteMask;
-                dynamicStates[6] = DynamicState.StencilReference;
-                dynamicStates[7] = DynamicState.BlendConstants;
+                dynamicStates[3] = DynamicState.StencilCompareMask;
+                dynamicStates[4] = DynamicState.StencilWriteMask;
+                dynamicStates[5] = DynamicState.StencilReference;
+                dynamicStates[6] = DynamicState.BlendConstants;
 
                 if (supportsExtDynamicState)
                 {
-                    dynamicStates[8] = DynamicState.VertexInputBindingStrideExt;
+                    dynamicStates[7] = DynamicState.VertexInputBindingStrideExt;
                 }
 
                 var pipelineDynamicStateCreateInfo = new PipelineDynamicStateCreateInfo
@@ -632,7 +604,6 @@ namespace Ryujinx.Graphics.Vulkan
                     PDynamicState = &pipelineDynamicStateCreateInfo,
                     Layout = PipelineLayout,
                     RenderPass = renderPass,
-                    BasePipelineIndex = -1,
                 };
 
                 Result result = gd.Api.CreateGraphicsPipelines(device, cache, 1, &pipelineCreateInfo, null, &pipelineHandle);

src/Ryujinx.Graphics.Vulkan/PipelineUid.cs

@@ -17,20 +17,17 @@ namespace Ryujinx.Graphics.Vulkan
         public ulong Id4;
         public ulong Id5;
         public ulong Id6;
+
         public ulong Id7;
-
         public ulong Id8;
-        public ulong Id9;
 
-        private readonly uint VertexAttributeDescriptionsCount => (byte)((Id6 >> 38) & 0xFF);
-        private readonly uint VertexBindingDescriptionsCount => (byte)((Id6 >> 46) & 0xFF);
-        private readonly uint ColorBlendAttachmentStateCount => (byte)((Id7 >> 8) & 0xFF);
-        private readonly bool HasDepthStencil => ((Id7 >> 63) & 0x1) != 0UL;
+        private readonly uint VertexAttributeDescriptionsCount => (byte)((Id5 >> 38) & 0xFF);
+        private readonly uint VertexBindingDescriptionsCount => (byte)((Id5 >> 46) & 0xFF);
+        private readonly uint ColorBlendAttachmentStateCount => (byte)((Id6 >> 8) & 0xFF);
+        private readonly bool HasDepthStencil => ((Id6 >> 63) & 0x1) != 0UL;
 
         public Array32<VertexInputAttributeDescription> VertexAttributeDescriptions;
         public Array33<VertexInputBindingDescription> VertexBindingDescriptions;
-        public Array16<Viewport> Viewports;
-        public Array16<Rect2D> Scissors;
         public Array8<PipelineColorBlendAttachmentState> ColorBlendAttachmentState;
         public Array9<Format> AttachmentFormats;
         public uint AttachmentIntegerFormatMask;
@@ -45,7 +42,7 @@ namespace Ryujinx.Graphics.Vulkan
         {
             if (!Unsafe.As<ulong, Vector256<byte>>(ref Id0).Equals(Unsafe.As<ulong, Vector256<byte>>(ref other.Id0)) ||
                 !Unsafe.As<ulong, Vector256<byte>>(ref Id4).Equals(Unsafe.As<ulong, Vector256<byte>>(ref other.Id4)) ||
-                !Unsafe.As<ulong, Vector128<byte>>(ref Id8).Equals(Unsafe.As<ulong, Vector128<byte>>(ref other.Id8)))
+                !Unsafe.As<ulong, Vector128<byte>>(ref Id7).Equals(Unsafe.As<ulong, Vector128<byte>>(ref other.Id7)))
             {
                 return false;
             }
@@ -88,8 +85,7 @@ namespace Ryujinx.Graphics.Vulkan
                            Id5 * 23 ^
                            Id6 * 23 ^
                            Id7 * 23 ^
-                           Id8 * 23 ^
-                           Id9 * 23;
+                           Id8 * 23;
 
             for (int i = 0; i < (int)VertexAttributeDescriptionsCount; i++)
             {

src/Ryujinx.Graphics.Vulkan/RenderPassHolder.cs

@@ -1,5 +1,7 @@
 using Silk.NET.Vulkan;
 using System;
+using System.Collections.Generic;
+using System.Linq;
 
 namespace Ryujinx.Graphics.Vulkan
 {
@@ -29,10 +31,13 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private readonly record struct ForcedFence(TextureStorage Texture, PipelineStageFlags StageFlags);
+
         private readonly TextureView[] _textures;
         private readonly Auto<DisposableRenderPass> _renderPass;
         private readonly HashTableSlim<FramebufferCacheKey, Auto<DisposableFramebuffer>> _framebuffers;
         private readonly RenderPassCacheKey _key;
+        private readonly List<ForcedFence> _forcedFences;
 
         public unsafe RenderPassHolder(VulkanRenderer gd, Device device, RenderPassCacheKey key, FramebufferParams fb)
         {
@@ -105,7 +110,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            var subpassDependency = PipelineConverter.CreateSubpassDependency();
+            var subpassDependency = PipelineConverter.CreateSubpassDependency(gd);
 
             fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
             {
@@ -138,6 +143,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             _textures = textures;
             _key = key;
+
+            _forcedFences = new List<ForcedFence>();
         }
 
         public Auto<DisposableFramebuffer> GetFramebuffer(VulkanRenderer gd, CommandBufferScoped cbs, FramebufferParams fb)
@@ -159,6 +166,37 @@ namespace Ryujinx.Graphics.Vulkan
             return _renderPass;
         }
 
+        public void AddForcedFence(TextureStorage storage, PipelineStageFlags stageFlags)
+        {
+            if (!_forcedFences.Any(fence => fence.Texture == storage))
+            {
+                _forcedFences.Add(new ForcedFence(storage, stageFlags));
+            }
+        }
+
+        public void InsertForcedFences(CommandBufferScoped cbs)
+        {
+            if (_forcedFences.Count > 0)
+            {
+                _forcedFences.RemoveAll((entry) =>
+                {
+                    if (entry.Texture.Disposed)
+                    {
+                        return true;
+                    }
+
+                    entry.Texture.QueueWriteToReadBarrier(cbs, AccessFlags.ShaderReadBit, entry.StageFlags);
+
+                    return false;
+                });
+            }
+        }
+
+        public bool ContainsAttachment(TextureStorage storage)
+        {
+            return _textures.Any(view => view.Storage == storage);
+        }
+
         public void Dispose()
         {
             // Dispose all framebuffers.

src/Ryujinx.Graphics.Vulkan/ResourceArray.cs

@@ -0,0 +1,74 @@
+using Silk.NET.Vulkan;
+using System;
+using System.Diagnostics;
+
+namespace Ryujinx.Graphics.Vulkan
+{
+    class ResourceArray : IDisposable
+    {
+        private DescriptorSet[] _cachedDescriptorSets;
+
+        private ShaderCollection _cachedDscProgram;
+        private int _cachedDscSetIndex;
+        private int _cachedDscIndex;
+
+        private int _bindCount;
+
+        protected void SetDirty(VulkanRenderer gd)
+        {
+            ReleaseDescriptorSet();
+
+            if (_bindCount != 0)
+            {
+                gd.PipelineInternal.ForceTextureDirty();
+            }
+        }
+
+        public bool TryGetCachedDescriptorSets(CommandBufferScoped cbs, ShaderCollection program, int setIndex, out DescriptorSet[] sets)
+        {
+            if (_cachedDescriptorSets != null)
+            {
+                _cachedDscProgram.UpdateManualDescriptorSetCollectionOwnership(cbs, _cachedDscSetIndex, _cachedDscIndex);
+
+                sets = _cachedDescriptorSets;
+
+                return true;
+            }
+
+            var dsc = program.GetNewManualDescriptorSetCollection(cbs, setIndex, out _cachedDscIndex).Get(cbs);
+
+            sets = dsc.GetSets();
+
+            _cachedDescriptorSets = sets;
+            _cachedDscProgram = program;
+            _cachedDscSetIndex = setIndex;
+
+            return false;
+        }
+
+        public void IncrementBindCount()
+        {
+            _bindCount++;
+        }
+
+        public void DecrementBindCount()
+        {
+            int newBindCount = --_bindCount;
+            Debug.Assert(newBindCount >= 0);
+        }
+
+        private void ReleaseDescriptorSet()
+        {
+            if (_cachedDescriptorSets != null)
+            {
+                _cachedDscProgram.ReleaseManualDescriptorSetCollection(_cachedDscSetIndex, _cachedDscIndex);
+                _cachedDescriptorSets = null;
+            }
+        }
+
+        public void Dispose()
+        {
+            ReleaseDescriptorSet();
+        }
+    }
+}

src/Ryujinx.Graphics.Vulkan/ResourceLayoutBuilder.cs

@@ -23,7 +23,7 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
-        public ResourceLayoutBuilder Add(ResourceStages stages, ResourceType type, int binding)
+        public ResourceLayoutBuilder Add(ResourceStages stages, ResourceType type, int binding, bool write = false)
         {
             int setIndex = type switch
             {
@@ -35,7 +35,7 @@ namespace Ryujinx.Graphics.Vulkan
             };
 
             _resourceDescriptors[setIndex].Add(new ResourceDescriptor(binding, 1, type, stages));
-            _resourceUsages[setIndex].Add(new ResourceUsage(binding, 1, type, stages));
+            _resourceUsages[setIndex].Add(new ResourceUsage(binding, 1, type, stages, write));
 
             return this;
         }