Add area sampling scaler to allow for super-sampled anti-aliasing. (#7304)

* Add area sampling scaler to allow for super-sampled anti-aliasing.

* Area scaling filter doesn't have a scaling level.

* Add further clarification to the tooltip on how to achieve supersampling.

* ShaderHelper: Merge the two CompileProgram functions.

* Convert tabs to spaces in area scaling shaders

* Fixup Vulkan and OpenGL project files.

* AreaScaling: Replace texture() by texelFetch() and use integer vectors.

No functional difference, but it cleans up the code a bit.

* AreaScaling: Delete unused sharpening level member.

Also rename _scale to _sharpeningLevel for clarity and consistency.

* AreaScaling: Delete unused scaleX/scaleY uniforms.

* AreaScaling: Force the alpha to 1 when storing the pixel.

* AreaScaling: Remove left-over sharpening buffer.

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="9.0.4" />
     <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" />
@@ -39,11 +39,11 @@
     <PackageVersion Include="securifybv.ShellLink" Version="0.1.0" />
     <PackageVersion Include="shaderc.net" Version="0.1.0" />
     <PackageVersion Include="SharpZipLib" Version="1.4.2" />
-    <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.Drawing" Version="1.0.0" />
+    <PackageVersion Include="Silk.NET.Vulkan" Version="2.21.0" />
+    <PackageVersion Include="Silk.NET.Vulkan.Extensions.EXT" Version="2.21.0" />
+    <PackageVersion Include="Silk.NET.Vulkan.Extensions.KHR" Version="2.21.0" />
+    <PackageVersion Include="SkiaSharp" Version="2.88.7" />
+    <PackageVersion Include="SkiaSharp.NativeAssets.Linux" Version="2.88.7" />
     <PackageVersion Include="SPB" Version="0.0.4-build32" />
     <PackageVersion Include="System.IO.Hashing" Version="8.0.0" />
     <PackageVersion Include="System.Management" Version="8.0.0" />

Ryujinx.sln

@@ -87,6 +87,8 @@ Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "Ryujinx.Horizon", "src\Ryuj
 EndProject
 Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "Ryujinx.Horizon.Kernel.Generators", "src\Ryujinx.Horizon.Kernel.Generators\Ryujinx.Horizon.Kernel.Generators.csproj", "{7F55A45D-4E1D-4A36-ADD3-87F29A285AA2}"
 EndProject
+Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "Ryujinx.HLE.Generators", "src\Ryujinx.HLE.Generators\Ryujinx.HLE.Generators.csproj", "{B575BCDE-2FD8-4A5D-8756-31CDD7FE81F0}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Any CPU = Debug|Any CPU
@@ -249,6 +251,10 @@ Global
 		{7F55A45D-4E1D-4A36-ADD3-87F29A285AA2}.Debug|Any CPU.Build.0 = Debug|Any CPU
 		{7F55A45D-4E1D-4A36-ADD3-87F29A285AA2}.Release|Any CPU.ActiveCfg = Release|Any CPU
 		{7F55A45D-4E1D-4A36-ADD3-87F29A285AA2}.Release|Any CPU.Build.0 = Release|Any CPU
+		{B575BCDE-2FD8-4A5D-8756-31CDD7FE81F0}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{B575BCDE-2FD8-4A5D-8756-31CDD7FE81F0}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{B575BCDE-2FD8-4A5D-8756-31CDD7FE81F0}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{B575BCDE-2FD8-4A5D-8756-31CDD7FE81F0}.Release|Any CPU.Build.0 = Release|Any CPU
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE

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

@@ -746,6 +746,7 @@ namespace ARMeilleure.Decoders
             SetA32("<<<<01101000xxxxxxxxxxxxxx01xxxx", InstName.Pkh,     InstEmit32.Pkh,     OpCode32AluRsImm.Create);
             SetA32("11110101xx01xxxx1111xxxxxxxxxxxx", InstName.Pld,     InstEmit32.Nop,     OpCode32.Create);
             SetA32("11110111xx01xxxx1111xxxxxxx0xxxx", InstName.Pld,     InstEmit32.Nop,     OpCode32.Create);
+            SetA32("<<<<01100010xxxxxxxx11110001xxxx", InstName.Qadd16,  InstEmit32.Qadd16,  OpCode32AluReg.Create);
             SetA32("<<<<011011111111xxxx11110011xxxx", InstName.Rbit,    InstEmit32.Rbit,    OpCode32AluReg.Create);
             SetA32("<<<<011010111111xxxx11110011xxxx", InstName.Rev,     InstEmit32.Rev,     OpCode32AluReg.Create);
             SetA32("<<<<011010111111xxxx11111011xxxx", InstName.Rev16,   InstEmit32.Rev16,   OpCode32AluReg.Create);
@@ -822,6 +823,10 @@ 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 +1012,8 @@ 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);
@@ -1028,8 +1035,10 @@ namespace ARMeilleure.Decoders
             SetAsimd("111100101x>>>xxxxxxx0101>xx1xxxx", InstName.Vshl,        InstEmit32.Vshl,         OpCode32SimdShImm.Create,       OpCode32SimdShImm.CreateT32);
             SetAsimd("1111001x0xxxxxxxxxxx0100xxx0xxxx", InstName.Vshl,        InstEmit32.Vshl_I,       OpCode32SimdReg.Create,         OpCode32SimdReg.CreateT32);
             SetAsimd("1111001x1x>>>xxxxxxx101000x1xxxx", InstName.Vshll,       InstEmit32.Vshll,        OpCode32SimdShImmLong.Create,   OpCode32SimdShImmLong.CreateT32); // A1 encoding.
+            SetAsimd("111100111x11<<10xxxx001100x0xxxx", InstName.Vshll,       InstEmit32.Vshll2,       OpCode32SimdMovn.Create,        OpCode32SimdMovn.CreateT32); // A2 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 +1063,7 @@ 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;
@@ -290,6 +292,16 @@ namespace ARMeilleure.Instructions
             EmitAluStore(context, res);
         }
 
+        public static void Qadd16(ArmEmitterContext context)
+        {
+            OpCode32AluReg op = (OpCode32AluReg)context.CurrOp;
+
+            SetIntA32(context, op.Rd, EmitSigned16BitPair(context, GetIntA32(context, op.Rn), GetIntA32(context, op.Rm), (d, n, m) =>
+            {
+                EmitSaturateRange(context, d, context.Add(n, m), 16, unsigned: false, setQ: false);
+            }));
+        }
+
         public static void Rbit(ArmEmitterContext context)
         {
             Operand m = GetAluM(context);
@@ -558,6 +570,46 @@ 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 +986,251 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        private static void EmitSaturateRange(ArmEmitterContext context, Operand result, Operand value, uint saturateTo, bool unsigned, bool setQ = true)
+        {
+            Debug.Assert(saturateTo <= 32);
+            Debug.Assert(!unsigned || saturateTo < 32);
+
+            if (!unsigned && 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;
+            }
+
+            Operand satValue;
+
+            if (unsigned)
+            {
+                // Negative values always saturate (to zero).
+                // So we must always ignore the sign bit when masking, so that the truncated value will differ from the original one.
+
+                satValue = context.BitwiseAnd(value, Const((int)(uint.MaxValue >> (32 - (int)saturateTo))));
+            }
+            else
+            {
+                satValue = context.ShiftLeft(value, Const(32 - (int)saturateTo));
+                satValue = context.ShiftRightSI(satValue, Const(32 - (int)saturateTo));
+            }
+
+            // If the result is 0, the values are equal and we don't need saturation.
+            Operand lblNoSat = Label();
+            context.BranchIfFalse(lblNoSat, context.Subtract(value, satValue));
+
+            // Saturate and set Q flag.
+            if (unsigned)
+            {
+                if (saturateTo == 31)
+                {
+                    // Only saturation case possible when going from 32 bits signed to 32 or 31 bits unsigned
+                    // is when the signed input is negative, as all positive values are representable on a 31 bits range.
+
+                    satValue = Const(0);
+                }
+                else
+                {
+                    satValue = context.ShiftRightSI(value, Const(31));
+                    satValue = context.BitwiseNot(satValue);
+                    satValue = context.ShiftRightUI(satValue, Const(32 - (int)saturateTo));
+                }
+            }
+            else
+            {
+                if (saturateTo == 1)
+                {
+                    satValue = context.ShiftRightSI(value, Const(31));
+                }
+                else
+                {
+                    satValue = Const(uint.MaxValue >> (33 - (int)saturateTo));
+                    satValue = context.BitwiseExclusiveOr(satValue, context.ShiftRightSI(value, Const(31)));
+                }
+            }
+
+            if (setQ)
+            {
+                SetFlag(context, PState.QFlag, Const(1));
+            }
+
+            context.Copy(result, satValue);
+
+            Operand lblExit = Label();
+            context.Branch(lblExit);
+
+            context.MarkLabel(lblNoSat);
+
+            context.Copy(result, value);
+
+            context.MarkLabel(lblExit);
+        }
+
+        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 EmitSigned16BitPair(ArmEmitterContext context, Operand rn, Operand rm, Action<Operand, Operand, Operand> elementAction)
+        {
+            Operand tempD = context.AllocateLocal(OperandType.I32);
+
+            Operand tempN = context.SignExtend16(OperandType.I32, rn);
+            Operand tempM = context.SignExtend16(OperandType.I32, rm);
+            elementAction(tempD, tempN, tempM);
+            Operand tempD2 = context.ZeroExtend16(OperandType.I32, tempD);
+
+            tempN = context.ShiftRightSI(rn, Const(16));
+            tempM = context.ShiftRightSI(rm, Const(16));
+            elementAction(tempD, tempN, tempM);
+            return context.BitwiseOr(tempD2, context.ShiftLeft(tempD, Const(16)));
+        }
+
+        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

@@ -106,6 +106,38 @@ namespace ARMeilleure.Instructions
             context.Copy(GetVecA32(op.Qd), res);
         }
 
+        public static void Vshll2(ArmEmitterContext context)
+        {
+            OpCode32Simd op = (OpCode32Simd)context.CurrOp;
+
+            Operand res = context.VectorZero();
+
+            int elems = op.GetBytesCount() >> op.Size;
+
+            for (int index = 0; index < elems; index++)
+            {
+                Operand me = EmitVectorExtract32(context, op.Qm, op.Im + index, op.Size, !op.U);
+
+                if (op.Size == 2)
+                {
+                    if (op.U)
+                    {
+                        me = context.ZeroExtend32(OperandType.I64, me);
+                    }
+                    else
+                    {
+                        me = context.SignExtend32(OperandType.I64, me);
+                    }
+                }
+
+                me = context.ShiftLeft(me, Const(8 << op.Size));
+
+                res = EmitVectorInsert(context, res, me, index, op.Size + 1);
+            }
+
+            context.Copy(GetVecA32(op.Qd), res);
+        }
+
         public static void Vshr(ArmEmitterContext context)
         {
             OpCode32SimdShImm op = (OpCode32SimdShImm)context.CurrOp;
@@ -130,6 +162,36 @@ 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

@@ -527,6 +527,7 @@ namespace ARMeilleure.Instructions
         Pld,
         Pop,
         Push,
+        Qadd16,
         Rev,
         Revsh,
         Rsb,
@@ -571,6 +572,10 @@ namespace ARMeilleure.Instructions
         Umaal,
         Umlal,
         Umull,
+        Uqadd16,
+        Uqadd8,
+        Uqsub16,
+        Uqsub8,
         Usat,
         Usat16,
         Usub8,
@@ -645,6 +650,7 @@ namespace ARMeilleure.Instructions
         Vqdmulh,
         Vqmovn,
         Vqmovun,
+        Vqrdmulh,
         Vqrshrn,
         Vqrshrun,
         Vqshrn,
@@ -666,6 +672,7 @@ namespace ARMeilleure.Instructions
         Vshll,
         Vshr,
         Vshrn,
+        Vsli,
         Vst1,
         Vst2,
         Vst3,
@@ -682,6 +689,7 @@ 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/DelegateInfo.cs

@@ -1,5 +1,4 @@
 using System;
-using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Translation
 {
@@ -11,11 +10,10 @@ namespace ARMeilleure.Translation
 
         public IntPtr FuncPtr { get; }
 
-        public DelegateInfo(Delegate dlg)
+        public DelegateInfo(Delegate dlg, IntPtr funcPtr)
         {
             _dlg = dlg;
-
-            FuncPtr = Marshal.GetFunctionPointerForDelegate<Delegate>(dlg);
+            FuncPtr = funcPtr;
         }
     }
 }

src/ARMeilleure/Translation/Delegates.cs

@@ -3,6 +3,7 @@ using ARMeilleure.State;
 using System;
 using System.Collections.Generic;
 using System.Reflection;
+using System.Runtime.InteropServices;
 
 namespace ARMeilleure.Translation
 {
@@ -64,11 +65,11 @@ namespace ARMeilleure.Translation
             return index;
         }
 
-        private static void SetDelegateInfo(Delegate dlg)
+        private static void SetDelegateInfo(Delegate dlg, IntPtr funcPtr)
         {
             string key = GetKey(dlg.Method);
 
-            _delegates.Add(key, new DelegateInfo(dlg)); // ArgumentException (key).
+            _delegates.Add(key, new DelegateInfo(dlg, funcPtr)); // ArgumentException (key).
         }
 
         private static string GetKey(MethodInfo info)
@@ -82,179 +83,353 @@ namespace ARMeilleure.Translation
         {
             _delegates = new SortedList<string, DelegateInfo>();
 
-            SetDelegateInfo(new MathAbs(Math.Abs));
-            SetDelegateInfo(new MathCeiling(Math.Ceiling));
-            SetDelegateInfo(new MathFloor(Math.Floor));
-            SetDelegateInfo(new MathRound(Math.Round));
-            SetDelegateInfo(new MathTruncate(Math.Truncate));
+            var dlgMathAbs = new MathAbs(Math.Abs);
+            var dlgMathCeiling = new MathCeiling(Math.Ceiling);
+            var dlgMathFloor = new MathFloor(Math.Floor);
+            var dlgMathRound = new MathRound(Math.Round);
+            var dlgMathTruncate = new MathTruncate(Math.Truncate);
 
-            SetDelegateInfo(new MathFAbs(MathF.Abs));
-            SetDelegateInfo(new MathFCeiling(MathF.Ceiling));
-            SetDelegateInfo(new MathFFloor(MathF.Floor));
-            SetDelegateInfo(new MathFRound(MathF.Round));
-            SetDelegateInfo(new MathFTruncate(MathF.Truncate));
+            var dlgMathFAbs = new MathFAbs(MathF.Abs);
+            var dlgMathFCeiling = new MathFCeiling(MathF.Ceiling);
+            var dlgMathFFloor = new MathFFloor(MathF.Floor);
+            var dlgMathFRound = new MathFRound(MathF.Round);
+            var dlgMathFTruncate = new MathFTruncate(MathF.Truncate);
 
-            SetDelegateInfo(new NativeInterfaceBreak(NativeInterface.Break));
-            SetDelegateInfo(new NativeInterfaceCheckSynchronization(NativeInterface.CheckSynchronization));
-            SetDelegateInfo(new NativeInterfaceEnqueueForRejit(NativeInterface.EnqueueForRejit));
-            SetDelegateInfo(new NativeInterfaceGetCntfrqEl0(NativeInterface.GetCntfrqEl0));
-            SetDelegateInfo(new NativeInterfaceGetCntpctEl0(NativeInterface.GetCntpctEl0));
-            SetDelegateInfo(new NativeInterfaceGetCntvctEl0(NativeInterface.GetCntvctEl0));
-            SetDelegateInfo(new NativeInterfaceGetCtrEl0(NativeInterface.GetCtrEl0));
-            SetDelegateInfo(new NativeInterfaceGetDczidEl0(NativeInterface.GetDczidEl0));
-            SetDelegateInfo(new NativeInterfaceGetFunctionAddress(NativeInterface.GetFunctionAddress));
-            SetDelegateInfo(new NativeInterfaceInvalidateCacheLine(NativeInterface.InvalidateCacheLine));
-            SetDelegateInfo(new NativeInterfaceReadByte(NativeInterface.ReadByte));
-            SetDelegateInfo(new NativeInterfaceReadUInt16(NativeInterface.ReadUInt16));
-            SetDelegateInfo(new NativeInterfaceReadUInt32(NativeInterface.ReadUInt32));
-            SetDelegateInfo(new NativeInterfaceReadUInt64(NativeInterface.ReadUInt64));
-            SetDelegateInfo(new NativeInterfaceReadVector128(NativeInterface.ReadVector128));
-            SetDelegateInfo(new NativeInterfaceSignalMemoryTracking(NativeInterface.SignalMemoryTracking));
-            SetDelegateInfo(new NativeInterfaceSupervisorCall(NativeInterface.SupervisorCall));
-            SetDelegateInfo(new NativeInterfaceThrowInvalidMemoryAccess(NativeInterface.ThrowInvalidMemoryAccess));
-            SetDelegateInfo(new NativeInterfaceUndefined(NativeInterface.Undefined));
-            SetDelegateInfo(new NativeInterfaceWriteByte(NativeInterface.WriteByte));
-            SetDelegateInfo(new NativeInterfaceWriteUInt16(NativeInterface.WriteUInt16));
-            SetDelegateInfo(new NativeInterfaceWriteUInt32(NativeInterface.WriteUInt32));
-            SetDelegateInfo(new NativeInterfaceWriteUInt64(NativeInterface.WriteUInt64));
-            SetDelegateInfo(new NativeInterfaceWriteVector128(NativeInterface.WriteVector128));
+            var dlgNativeInterfaceBreak = new NativeInterfaceBreak(NativeInterface.Break);
+            var dlgNativeInterfaceCheckSynchronization = new NativeInterfaceCheckSynchronization(NativeInterface.CheckSynchronization);
+            var dlgNativeInterfaceEnqueueForRejit = new NativeInterfaceEnqueueForRejit(NativeInterface.EnqueueForRejit);
+            var dlgNativeInterfaceGetCntfrqEl0 = new NativeInterfaceGetCntfrqEl0(NativeInterface.GetCntfrqEl0);
+            var dlgNativeInterfaceGetCntpctEl0 = new NativeInterfaceGetCntpctEl0(NativeInterface.GetCntpctEl0);
+            var dlgNativeInterfaceGetCntvctEl0 = new NativeInterfaceGetCntvctEl0(NativeInterface.GetCntvctEl0);
+            var dlgNativeInterfaceGetCtrEl0 = new NativeInterfaceGetCtrEl0(NativeInterface.GetCtrEl0);
+            var dlgNativeInterfaceGetDczidEl0 = new NativeInterfaceGetDczidEl0(NativeInterface.GetDczidEl0);
+            var dlgNativeInterfaceGetFunctionAddress = new NativeInterfaceGetFunctionAddress(NativeInterface.GetFunctionAddress);
+            var dlgNativeInterfaceInvalidateCacheLine = new NativeInterfaceInvalidateCacheLine(NativeInterface.InvalidateCacheLine);
+            var dlgNativeInterfaceReadByte = new NativeInterfaceReadByte(NativeInterface.ReadByte);
+            var dlgNativeInterfaceReadUInt16 = new NativeInterfaceReadUInt16(NativeInterface.ReadUInt16);
+            var dlgNativeInterfaceReadUInt32 = new NativeInterfaceReadUInt32(NativeInterface.ReadUInt32);
+            var dlgNativeInterfaceReadUInt64 = new NativeInterfaceReadUInt64(NativeInterface.ReadUInt64);
+            var dlgNativeInterfaceReadVector128 = new NativeInterfaceReadVector128(NativeInterface.ReadVector128);
+            var dlgNativeInterfaceSignalMemoryTracking = new NativeInterfaceSignalMemoryTracking(NativeInterface.SignalMemoryTracking);
+            var dlgNativeInterfaceSupervisorCall = new NativeInterfaceSupervisorCall(NativeInterface.SupervisorCall);
+            var dlgNativeInterfaceThrowInvalidMemoryAccess = new NativeInterfaceThrowInvalidMemoryAccess(NativeInterface.ThrowInvalidMemoryAccess);
+            var dlgNativeInterfaceUndefined = new NativeInterfaceUndefined(NativeInterface.Undefined);
+            var dlgNativeInterfaceWriteByte = new NativeInterfaceWriteByte(NativeInterface.WriteByte);
+            var dlgNativeInterfaceWriteUInt16 = new NativeInterfaceWriteUInt16(NativeInterface.WriteUInt16);
+            var dlgNativeInterfaceWriteUInt32 = new NativeInterfaceWriteUInt32(NativeInterface.WriteUInt32);
+            var dlgNativeInterfaceWriteUInt64 = new NativeInterfaceWriteUInt64(NativeInterface.WriteUInt64);
+            var dlgNativeInterfaceWriteVector128 = new NativeInterfaceWriteVector128(NativeInterface.WriteVector128);
 
-            SetDelegateInfo(new SoftFallbackCountLeadingSigns(SoftFallback.CountLeadingSigns));
-            SetDelegateInfo(new SoftFallbackCountLeadingZeros(SoftFallback.CountLeadingZeros));
-            SetDelegateInfo(new SoftFallbackCrc32b(SoftFallback.Crc32b));
-            SetDelegateInfo(new SoftFallbackCrc32cb(SoftFallback.Crc32cb));
-            SetDelegateInfo(new SoftFallbackCrc32ch(SoftFallback.Crc32ch));
-            SetDelegateInfo(new SoftFallbackCrc32cw(SoftFallback.Crc32cw));
-            SetDelegateInfo(new SoftFallbackCrc32cx(SoftFallback.Crc32cx));
-            SetDelegateInfo(new SoftFallbackCrc32h(SoftFallback.Crc32h));
-            SetDelegateInfo(new SoftFallbackCrc32w(SoftFallback.Crc32w));
-            SetDelegateInfo(new SoftFallbackCrc32x(SoftFallback.Crc32x));
-            SetDelegateInfo(new SoftFallbackDecrypt(SoftFallback.Decrypt));
-            SetDelegateInfo(new SoftFallbackEncrypt(SoftFallback.Encrypt));
-            SetDelegateInfo(new SoftFallbackFixedRotate(SoftFallback.FixedRotate));
-            SetDelegateInfo(new SoftFallbackHashChoose(SoftFallback.HashChoose));
-            SetDelegateInfo(new SoftFallbackHashLower(SoftFallback.HashLower));
-            SetDelegateInfo(new SoftFallbackHashMajority(SoftFallback.HashMajority));
-            SetDelegateInfo(new SoftFallbackHashParity(SoftFallback.HashParity));
-            SetDelegateInfo(new SoftFallbackHashUpper(SoftFallback.HashUpper));
-            SetDelegateInfo(new SoftFallbackInverseMixColumns(SoftFallback.InverseMixColumns));
-            SetDelegateInfo(new SoftFallbackMixColumns(SoftFallback.MixColumns));
-            SetDelegateInfo(new SoftFallbackPolynomialMult64_128(SoftFallback.PolynomialMult64_128));
-            SetDelegateInfo(new SoftFallbackSatF32ToS32(SoftFallback.SatF32ToS32));
-            SetDelegateInfo(new SoftFallbackSatF32ToS64(SoftFallback.SatF32ToS64));
-            SetDelegateInfo(new SoftFallbackSatF32ToU32(SoftFallback.SatF32ToU32));
-            SetDelegateInfo(new SoftFallbackSatF32ToU64(SoftFallback.SatF32ToU64));
-            SetDelegateInfo(new SoftFallbackSatF64ToS32(SoftFallback.SatF64ToS32));
-            SetDelegateInfo(new SoftFallbackSatF64ToS64(SoftFallback.SatF64ToS64));
-            SetDelegateInfo(new SoftFallbackSatF64ToU32(SoftFallback.SatF64ToU32));
-            SetDelegateInfo(new SoftFallbackSatF64ToU64(SoftFallback.SatF64ToU64));
-            SetDelegateInfo(new SoftFallbackSha1SchedulePart1(SoftFallback.Sha1SchedulePart1));
-            SetDelegateInfo(new SoftFallbackSha1SchedulePart2(SoftFallback.Sha1SchedulePart2));
-            SetDelegateInfo(new SoftFallbackSha256SchedulePart1(SoftFallback.Sha256SchedulePart1));
-            SetDelegateInfo(new SoftFallbackSha256SchedulePart2(SoftFallback.Sha256SchedulePart2));
-            SetDelegateInfo(new SoftFallbackSignedShrImm64(SoftFallback.SignedShrImm64));
-            SetDelegateInfo(new SoftFallbackTbl1(SoftFallback.Tbl1));
-            SetDelegateInfo(new SoftFallbackTbl2(SoftFallback.Tbl2));
-            SetDelegateInfo(new SoftFallbackTbl3(SoftFallback.Tbl3));
-            SetDelegateInfo(new SoftFallbackTbl4(SoftFallback.Tbl4));
-            SetDelegateInfo(new SoftFallbackTbx1(SoftFallback.Tbx1));
-            SetDelegateInfo(new SoftFallbackTbx2(SoftFallback.Tbx2));
-            SetDelegateInfo(new SoftFallbackTbx3(SoftFallback.Tbx3));
-            SetDelegateInfo(new SoftFallbackTbx4(SoftFallback.Tbx4));
-            SetDelegateInfo(new SoftFallbackUnsignedShrImm64(SoftFallback.UnsignedShrImm64));
+            var dlgSoftFallbackCountLeadingSigns = new SoftFallbackCountLeadingSigns(SoftFallback.CountLeadingSigns);
+            var dlgSoftFallbackCountLeadingZeros = new SoftFallbackCountLeadingZeros(SoftFallback.CountLeadingZeros);
+            var dlgSoftFallbackCrc32b = new SoftFallbackCrc32b(SoftFallback.Crc32b);
+            var dlgSoftFallbackCrc32cb = new SoftFallbackCrc32cb(SoftFallback.Crc32cb);
+            var dlgSoftFallbackCrc32ch = new SoftFallbackCrc32ch(SoftFallback.Crc32ch);
+            var dlgSoftFallbackCrc32cw = new SoftFallbackCrc32cw(SoftFallback.Crc32cw);
+            var dlgSoftFallbackCrc32cx = new SoftFallbackCrc32cx(SoftFallback.Crc32cx);
+            var dlgSoftFallbackCrc32h = new SoftFallbackCrc32h(SoftFallback.Crc32h);
+            var dlgSoftFallbackCrc32w = new SoftFallbackCrc32w(SoftFallback.Crc32w);
+            var dlgSoftFallbackCrc32x = new SoftFallbackCrc32x(SoftFallback.Crc32x);
+            var dlgSoftFallbackDecrypt = new SoftFallbackDecrypt(SoftFallback.Decrypt);
+            var dlgSoftFallbackEncrypt = new SoftFallbackEncrypt(SoftFallback.Encrypt);
+            var dlgSoftFallbackFixedRotate = new SoftFallbackFixedRotate(SoftFallback.FixedRotate);
+            var dlgSoftFallbackHashChoose = new SoftFallbackHashChoose(SoftFallback.HashChoose);
+            var dlgSoftFallbackHashLower = new SoftFallbackHashLower(SoftFallback.HashLower);
+            var dlgSoftFallbackHashMajority = new SoftFallbackHashMajority(SoftFallback.HashMajority);
+            var dlgSoftFallbackHashParity = new SoftFallbackHashParity(SoftFallback.HashParity);
+            var dlgSoftFallbackHashUpper = new SoftFallbackHashUpper(SoftFallback.HashUpper);
+            var dlgSoftFallbackInverseMixColumns = new SoftFallbackInverseMixColumns(SoftFallback.InverseMixColumns);
+            var dlgSoftFallbackMixColumns = new SoftFallbackMixColumns(SoftFallback.MixColumns);
+            var dlgSoftFallbackPolynomialMult64_128 = new SoftFallbackPolynomialMult64_128(SoftFallback.PolynomialMult64_128);
+            var dlgSoftFallbackSatF32ToS32 = new SoftFallbackSatF32ToS32(SoftFallback.SatF32ToS32);
+            var dlgSoftFallbackSatF32ToS64 = new SoftFallbackSatF32ToS64(SoftFallback.SatF32ToS64);
+            var dlgSoftFallbackSatF32ToU32 = new SoftFallbackSatF32ToU32(SoftFallback.SatF32ToU32);
+            var dlgSoftFallbackSatF32ToU64 = new SoftFallbackSatF32ToU64(SoftFallback.SatF32ToU64);
+            var dlgSoftFallbackSatF64ToS32 = new SoftFallbackSatF64ToS32(SoftFallback.SatF64ToS32);
+            var dlgSoftFallbackSatF64ToS64 = new SoftFallbackSatF64ToS64(SoftFallback.SatF64ToS64);
+            var dlgSoftFallbackSatF64ToU32 = new SoftFallbackSatF64ToU32(SoftFallback.SatF64ToU32);
+            var dlgSoftFallbackSatF64ToU64 = new SoftFallbackSatF64ToU64(SoftFallback.SatF64ToU64);
+            var dlgSoftFallbackSha1SchedulePart1 = new SoftFallbackSha1SchedulePart1(SoftFallback.Sha1SchedulePart1);
+            var dlgSoftFallbackSha1SchedulePart2 = new SoftFallbackSha1SchedulePart2(SoftFallback.Sha1SchedulePart2);
+            var dlgSoftFallbackSha256SchedulePart1 = new SoftFallbackSha256SchedulePart1(SoftFallback.Sha256SchedulePart1);
+            var dlgSoftFallbackSha256SchedulePart2 = new SoftFallbackSha256SchedulePart2(SoftFallback.Sha256SchedulePart2);
+            var dlgSoftFallbackSignedShrImm64 = new SoftFallbackSignedShrImm64(SoftFallback.SignedShrImm64);
+            var dlgSoftFallbackTbl1 = new SoftFallbackTbl1(SoftFallback.Tbl1);
+            var dlgSoftFallbackTbl2 = new SoftFallbackTbl2(SoftFallback.Tbl2);
+            var dlgSoftFallbackTbl3 = new SoftFallbackTbl3(SoftFallback.Tbl3);
+            var dlgSoftFallbackTbl4 = new SoftFallbackTbl4(SoftFallback.Tbl4);
+            var dlgSoftFallbackTbx1 = new SoftFallbackTbx1(SoftFallback.Tbx1);
+            var dlgSoftFallbackTbx2 = new SoftFallbackTbx2(SoftFallback.Tbx2);
+            var dlgSoftFallbackTbx3 = new SoftFallbackTbx3(SoftFallback.Tbx3);
+            var dlgSoftFallbackTbx4 = new SoftFallbackTbx4(SoftFallback.Tbx4);
+            var dlgSoftFallbackUnsignedShrImm64 = new SoftFallbackUnsignedShrImm64(SoftFallback.UnsignedShrImm64);
 
-            SetDelegateInfo(new SoftFloat16_32FPConvert(SoftFloat16_32.FPConvert));
-            SetDelegateInfo(new SoftFloat16_64FPConvert(SoftFloat16_64.FPConvert));
+            var dlgSoftFloat16_32FPConvert = new SoftFloat16_32FPConvert(SoftFloat16_32.FPConvert);
+            var dlgSoftFloat16_64FPConvert = new SoftFloat16_64FPConvert(SoftFloat16_64.FPConvert);
 
-            SetDelegateInfo(new SoftFloat32FPAdd(SoftFloat32.FPAdd));
-            SetDelegateInfo(new SoftFloat32FPAddFpscr(SoftFloat32.FPAddFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPCompare(SoftFloat32.FPCompare));
-            SetDelegateInfo(new SoftFloat32FPCompareEQ(SoftFloat32.FPCompareEQ));
-            SetDelegateInfo(new SoftFloat32FPCompareEQFpscr(SoftFloat32.FPCompareEQFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPCompareGE(SoftFloat32.FPCompareGE));
-            SetDelegateInfo(new SoftFloat32FPCompareGEFpscr(SoftFloat32.FPCompareGEFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPCompareGT(SoftFloat32.FPCompareGT));
-            SetDelegateInfo(new SoftFloat32FPCompareGTFpscr(SoftFloat32.FPCompareGTFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPCompareLE(SoftFloat32.FPCompareLE));
-            SetDelegateInfo(new SoftFloat32FPCompareLEFpscr(SoftFloat32.FPCompareLEFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPCompareLT(SoftFloat32.FPCompareLT));
-            SetDelegateInfo(new SoftFloat32FPCompareLTFpscr(SoftFloat32.FPCompareLTFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPDiv(SoftFloat32.FPDiv));
-            SetDelegateInfo(new SoftFloat32FPMax(SoftFloat32.FPMax));
-            SetDelegateInfo(new SoftFloat32FPMaxFpscr(SoftFloat32.FPMaxFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMaxNum(SoftFloat32.FPMaxNum));
-            SetDelegateInfo(new SoftFloat32FPMaxNumFpscr(SoftFloat32.FPMaxNumFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMin(SoftFloat32.FPMin));
-            SetDelegateInfo(new SoftFloat32FPMinFpscr(SoftFloat32.FPMinFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMinNum(SoftFloat32.FPMinNum));
-            SetDelegateInfo(new SoftFloat32FPMinNumFpscr(SoftFloat32.FPMinNumFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMul(SoftFloat32.FPMul));
-            SetDelegateInfo(new SoftFloat32FPMulFpscr(SoftFloat32.FPMulFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMulAdd(SoftFloat32.FPMulAdd));
-            SetDelegateInfo(new SoftFloat32FPMulAddFpscr(SoftFloat32.FPMulAddFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMulSub(SoftFloat32.FPMulSub));
-            SetDelegateInfo(new SoftFloat32FPMulSubFpscr(SoftFloat32.FPMulSubFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPMulX(SoftFloat32.FPMulX));
-            SetDelegateInfo(new SoftFloat32FPNegMulAdd(SoftFloat32.FPNegMulAdd));
-            SetDelegateInfo(new SoftFloat32FPNegMulSub(SoftFloat32.FPNegMulSub));
-            SetDelegateInfo(new SoftFloat32FPRecipEstimate(SoftFloat32.FPRecipEstimate));
-            SetDelegateInfo(new SoftFloat32FPRecipEstimateFpscr(SoftFloat32.FPRecipEstimateFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPRecipStep(SoftFloat32.FPRecipStep)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPRecipStepFused(SoftFloat32.FPRecipStepFused));
-            SetDelegateInfo(new SoftFloat32FPRecpX(SoftFloat32.FPRecpX));
-            SetDelegateInfo(new SoftFloat32FPRSqrtEstimate(SoftFloat32.FPRSqrtEstimate));
-            SetDelegateInfo(new SoftFloat32FPRSqrtEstimateFpscr(SoftFloat32.FPRSqrtEstimateFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPRSqrtStep(SoftFloat32.FPRSqrtStep)); // A32 only.
-            SetDelegateInfo(new SoftFloat32FPRSqrtStepFused(SoftFloat32.FPRSqrtStepFused));
-            SetDelegateInfo(new SoftFloat32FPSqrt(SoftFloat32.FPSqrt));
-            SetDelegateInfo(new SoftFloat32FPSub(SoftFloat32.FPSub));
+            var dlgSoftFloat32FPAdd = new SoftFloat32FPAdd(SoftFloat32.FPAdd);
+            var dlgSoftFloat32FPAddFpscr = new SoftFloat32FPAddFpscr(SoftFloat32.FPAddFpscr); // A32 only.
+            var dlgSoftFloat32FPCompare = new SoftFloat32FPCompare(SoftFloat32.FPCompare);
+            var dlgSoftFloat32FPCompareEQ = new SoftFloat32FPCompareEQ(SoftFloat32.FPCompareEQ);
+            var dlgSoftFloat32FPCompareEQFpscr = new SoftFloat32FPCompareEQFpscr(SoftFloat32.FPCompareEQFpscr); // A32 only.
+            var dlgSoftFloat32FPCompareGE = new SoftFloat32FPCompareGE(SoftFloat32.FPCompareGE);
+            var dlgSoftFloat32FPCompareGEFpscr = new SoftFloat32FPCompareGEFpscr(SoftFloat32.FPCompareGEFpscr); // A32 only.
+            var dlgSoftFloat32FPCompareGT = new SoftFloat32FPCompareGT(SoftFloat32.FPCompareGT);
+            var dlgSoftFloat32FPCompareGTFpscr = new SoftFloat32FPCompareGTFpscr(SoftFloat32.FPCompareGTFpscr); // A32 only.
+            var dlgSoftFloat32FPCompareLE = new SoftFloat32FPCompareLE(SoftFloat32.FPCompareLE);
+            var dlgSoftFloat32FPCompareLEFpscr = new SoftFloat32FPCompareLEFpscr(SoftFloat32.FPCompareLEFpscr); // A32 only.
+            var dlgSoftFloat32FPCompareLT = new SoftFloat32FPCompareLT(SoftFloat32.FPCompareLT);
+            var dlgSoftFloat32FPCompareLTFpscr = new SoftFloat32FPCompareLTFpscr(SoftFloat32.FPCompareLTFpscr); // A32 only.
+            var dlgSoftFloat32FPDiv = new SoftFloat32FPDiv(SoftFloat32.FPDiv);
+            var dlgSoftFloat32FPMax = new SoftFloat32FPMax(SoftFloat32.FPMax);
+            var dlgSoftFloat32FPMaxFpscr = new SoftFloat32FPMaxFpscr(SoftFloat32.FPMaxFpscr); // A32 only.
+            var dlgSoftFloat32FPMaxNum = new SoftFloat32FPMaxNum(SoftFloat32.FPMaxNum);
+            var dlgSoftFloat32FPMaxNumFpscr = new SoftFloat32FPMaxNumFpscr(SoftFloat32.FPMaxNumFpscr); // A32 only.
+            var dlgSoftFloat32FPMin = new SoftFloat32FPMin(SoftFloat32.FPMin);
+            var dlgSoftFloat32FPMinFpscr = new SoftFloat32FPMinFpscr(SoftFloat32.FPMinFpscr); // A32 only.
+            var dlgSoftFloat32FPMinNum = new SoftFloat32FPMinNum(SoftFloat32.FPMinNum);
+            var dlgSoftFloat32FPMinNumFpscr = new SoftFloat32FPMinNumFpscr(SoftFloat32.FPMinNumFpscr); // A32 only.
+            var dlgSoftFloat32FPMul = new SoftFloat32FPMul(SoftFloat32.FPMul);
+            var dlgSoftFloat32FPMulFpscr = new SoftFloat32FPMulFpscr(SoftFloat32.FPMulFpscr); // A32 only.
+            var dlgSoftFloat32FPMulAdd = new SoftFloat32FPMulAdd(SoftFloat32.FPMulAdd);
+            var dlgSoftFloat32FPMulAddFpscr = new SoftFloat32FPMulAddFpscr(SoftFloat32.FPMulAddFpscr); // A32 only.
+            var dlgSoftFloat32FPMulSub = new SoftFloat32FPMulSub(SoftFloat32.FPMulSub);
+            var dlgSoftFloat32FPMulSubFpscr = new SoftFloat32FPMulSubFpscr(SoftFloat32.FPMulSubFpscr); // A32 only.
+            var dlgSoftFloat32FPMulX = new SoftFloat32FPMulX(SoftFloat32.FPMulX);
+            var dlgSoftFloat32FPNegMulAdd = new SoftFloat32FPNegMulAdd(SoftFloat32.FPNegMulAdd);
+            var dlgSoftFloat32FPNegMulSub = new SoftFloat32FPNegMulSub(SoftFloat32.FPNegMulSub);
+            var dlgSoftFloat32FPRecipEstimate = new SoftFloat32FPRecipEstimate(SoftFloat32.FPRecipEstimate);
+            var dlgSoftFloat32FPRecipEstimateFpscr = new SoftFloat32FPRecipEstimateFpscr(SoftFloat32.FPRecipEstimateFpscr); // A32 only.
+            var dlgSoftFloat32FPRecipStep = new SoftFloat32FPRecipStep(SoftFloat32.FPRecipStep); // A32 only.
+            var dlgSoftFloat32FPRecipStepFused = new SoftFloat32FPRecipStepFused(SoftFloat32.FPRecipStepFused);
+            var dlgSoftFloat32FPRecpX = new SoftFloat32FPRecpX(SoftFloat32.FPRecpX);
+            var dlgSoftFloat32FPRSqrtEstimate = new SoftFloat32FPRSqrtEstimate(SoftFloat32.FPRSqrtEstimate);
+            var dlgSoftFloat32FPRSqrtEstimateFpscr = new SoftFloat32FPRSqrtEstimateFpscr(SoftFloat32.FPRSqrtEstimateFpscr); // A32 only.
+            var dlgSoftFloat32FPRSqrtStep = new SoftFloat32FPRSqrtStep(SoftFloat32.FPRSqrtStep); // A32 only.
+            var dlgSoftFloat32FPRSqrtStepFused = new SoftFloat32FPRSqrtStepFused(SoftFloat32.FPRSqrtStepFused);
+            var dlgSoftFloat32FPSqrt = new SoftFloat32FPSqrt(SoftFloat32.FPSqrt);
+            var dlgSoftFloat32FPSub = new SoftFloat32FPSub(SoftFloat32.FPSub);
 
-            SetDelegateInfo(new SoftFloat32_16FPConvert(SoftFloat32_16.FPConvert));
+            var dlgSoftFloat32_16FPConvert = new SoftFloat32_16FPConvert(SoftFloat32_16.FPConvert);
 
-            SetDelegateInfo(new SoftFloat64FPAdd(SoftFloat64.FPAdd));
-            SetDelegateInfo(new SoftFloat64FPAddFpscr(SoftFloat64.FPAddFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPCompare(SoftFloat64.FPCompare));
-            SetDelegateInfo(new SoftFloat64FPCompareEQ(SoftFloat64.FPCompareEQ));
-            SetDelegateInfo(new SoftFloat64FPCompareEQFpscr(SoftFloat64.FPCompareEQFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPCompareGE(SoftFloat64.FPCompareGE));
-            SetDelegateInfo(new SoftFloat64FPCompareGEFpscr(SoftFloat64.FPCompareGEFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPCompareGT(SoftFloat64.FPCompareGT));
-            SetDelegateInfo(new SoftFloat64FPCompareGTFpscr(SoftFloat64.FPCompareGTFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPCompareLE(SoftFloat64.FPCompareLE));
-            SetDelegateInfo(new SoftFloat64FPCompareLEFpscr(SoftFloat64.FPCompareLEFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPCompareLT(SoftFloat64.FPCompareLT));
-            SetDelegateInfo(new SoftFloat64FPCompareLTFpscr(SoftFloat64.FPCompareLTFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPDiv(SoftFloat64.FPDiv));
-            SetDelegateInfo(new SoftFloat64FPMax(SoftFloat64.FPMax));
-            SetDelegateInfo(new SoftFloat64FPMaxFpscr(SoftFloat64.FPMaxFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMaxNum(SoftFloat64.FPMaxNum));
-            SetDelegateInfo(new SoftFloat64FPMaxNumFpscr(SoftFloat64.FPMaxNumFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMin(SoftFloat64.FPMin));
-            SetDelegateInfo(new SoftFloat64FPMinFpscr(SoftFloat64.FPMinFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMinNum(SoftFloat64.FPMinNum));
-            SetDelegateInfo(new SoftFloat64FPMinNumFpscr(SoftFloat64.FPMinNumFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMul(SoftFloat64.FPMul));
-            SetDelegateInfo(new SoftFloat64FPMulFpscr(SoftFloat64.FPMulFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMulAdd(SoftFloat64.FPMulAdd));
-            SetDelegateInfo(new SoftFloat64FPMulAddFpscr(SoftFloat64.FPMulAddFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMulSub(SoftFloat64.FPMulSub));
-            SetDelegateInfo(new SoftFloat64FPMulSubFpscr(SoftFloat64.FPMulSubFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPMulX(SoftFloat64.FPMulX));
-            SetDelegateInfo(new SoftFloat64FPNegMulAdd(SoftFloat64.FPNegMulAdd));
-            SetDelegateInfo(new SoftFloat64FPNegMulSub(SoftFloat64.FPNegMulSub));
-            SetDelegateInfo(new SoftFloat64FPRecipEstimate(SoftFloat64.FPRecipEstimate));
-            SetDelegateInfo(new SoftFloat64FPRecipEstimateFpscr(SoftFloat64.FPRecipEstimateFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPRecipStep(SoftFloat64.FPRecipStep)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPRecipStepFused(SoftFloat64.FPRecipStepFused));
-            SetDelegateInfo(new SoftFloat64FPRecpX(SoftFloat64.FPRecpX));
-            SetDelegateInfo(new SoftFloat64FPRSqrtEstimate(SoftFloat64.FPRSqrtEstimate));
-            SetDelegateInfo(new SoftFloat64FPRSqrtEstimateFpscr(SoftFloat64.FPRSqrtEstimateFpscr)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPRSqrtStep(SoftFloat64.FPRSqrtStep)); // A32 only.
-            SetDelegateInfo(new SoftFloat64FPRSqrtStepFused(SoftFloat64.FPRSqrtStepFused));
-            SetDelegateInfo(new SoftFloat64FPSqrt(SoftFloat64.FPSqrt));
-            SetDelegateInfo(new SoftFloat64FPSub(SoftFloat64.FPSub));
+            var dlgSoftFloat64FPAdd = new SoftFloat64FPAdd(SoftFloat64.FPAdd);
+            var dlgSoftFloat64FPAddFpscr = new SoftFloat64FPAddFpscr(SoftFloat64.FPAddFpscr); // A32 only.
+            var dlgSoftFloat64FPCompare = new SoftFloat64FPCompare(SoftFloat64.FPCompare);
+            var dlgSoftFloat64FPCompareEQ = new SoftFloat64FPCompareEQ(SoftFloat64.FPCompareEQ);
+            var dlgSoftFloat64FPCompareEQFpscr = new SoftFloat64FPCompareEQFpscr(SoftFloat64.FPCompareEQFpscr); // A32 only.
+            var dlgSoftFloat64FPCompareGE = new SoftFloat64FPCompareGE(SoftFloat64.FPCompareGE);
+            var dlgSoftFloat64FPCompareGEFpscr = new SoftFloat64FPCompareGEFpscr(SoftFloat64.FPCompareGEFpscr); // A32 only.
+            var dlgSoftFloat64FPCompareGT = new SoftFloat64FPCompareGT(SoftFloat64.FPCompareGT);
+            var dlgSoftFloat64FPCompareGTFpscr = new SoftFloat64FPCompareGTFpscr(SoftFloat64.FPCompareGTFpscr); // A32 only.
+            var dlgSoftFloat64FPCompareLE = new SoftFloat64FPCompareLE(SoftFloat64.FPCompareLE);
+            var dlgSoftFloat64FPCompareLEFpscr = new SoftFloat64FPCompareLEFpscr(SoftFloat64.FPCompareLEFpscr); // A32 only.
+            var dlgSoftFloat64FPCompareLT = new SoftFloat64FPCompareLT(SoftFloat64.FPCompareLT);
+            var dlgSoftFloat64FPCompareLTFpscr = new SoftFloat64FPCompareLTFpscr(SoftFloat64.FPCompareLTFpscr); // A32 only.
+            var dlgSoftFloat64FPDiv = new SoftFloat64FPDiv(SoftFloat64.FPDiv);
+            var dlgSoftFloat64FPMax = new SoftFloat64FPMax(SoftFloat64.FPMax);
+            var dlgSoftFloat64FPMaxFpscr = new SoftFloat64FPMaxFpscr(SoftFloat64.FPMaxFpscr); // A32 only.
+            var dlgSoftFloat64FPMaxNum = new SoftFloat64FPMaxNum(SoftFloat64.FPMaxNum);
+            var dlgSoftFloat64FPMaxNumFpscr = new SoftFloat64FPMaxNumFpscr(SoftFloat64.FPMaxNumFpscr); // A32 only.
+            var dlgSoftFloat64FPMin = new SoftFloat64FPMin(SoftFloat64.FPMin);
+            var dlgSoftFloat64FPMinFpscr = new SoftFloat64FPMinFpscr(SoftFloat64.FPMinFpscr); // A32 only.
+            var dlgSoftFloat64FPMinNum = new SoftFloat64FPMinNum(SoftFloat64.FPMinNum);
+            var dlgSoftFloat64FPMinNumFpscr = new SoftFloat64FPMinNumFpscr(SoftFloat64.FPMinNumFpscr); // A32 only.
+            var dlgSoftFloat64FPMul = new SoftFloat64FPMul(SoftFloat64.FPMul);
+            var dlgSoftFloat64FPMulFpscr = new SoftFloat64FPMulFpscr(SoftFloat64.FPMulFpscr); // A32 only.
+            var dlgSoftFloat64FPMulAdd = new SoftFloat64FPMulAdd(SoftFloat64.FPMulAdd);
+            var dlgSoftFloat64FPMulAddFpscr = new SoftFloat64FPMulAddFpscr(SoftFloat64.FPMulAddFpscr); // A32 only.
+            var dlgSoftFloat64FPMulSub = new SoftFloat64FPMulSub(SoftFloat64.FPMulSub);
+            var dlgSoftFloat64FPMulSubFpscr = new SoftFloat64FPMulSubFpscr(SoftFloat64.FPMulSubFpscr); // A32 only.
+            var dlgSoftFloat64FPMulX = new SoftFloat64FPMulX(SoftFloat64.FPMulX);
+            var dlgSoftFloat64FPNegMulAdd = new SoftFloat64FPNegMulAdd(SoftFloat64.FPNegMulAdd);
+            var dlgSoftFloat64FPNegMulSub = new SoftFloat64FPNegMulSub(SoftFloat64.FPNegMulSub);
+            var dlgSoftFloat64FPRecipEstimate = new SoftFloat64FPRecipEstimate(SoftFloat64.FPRecipEstimate);
+            var dlgSoftFloat64FPRecipEstimateFpscr = new SoftFloat64FPRecipEstimateFpscr(SoftFloat64.FPRecipEstimateFpscr); // A32 only.
+            var dlgSoftFloat64FPRecipStep = new SoftFloat64FPRecipStep(SoftFloat64.FPRecipStep); // A32 only.
+            var dlgSoftFloat64FPRecipStepFused = new SoftFloat64FPRecipStepFused(SoftFloat64.FPRecipStepFused);
+            var dlgSoftFloat64FPRecpX = new SoftFloat64FPRecpX(SoftFloat64.FPRecpX);
+            var dlgSoftFloat64FPRSqrtEstimate = new SoftFloat64FPRSqrtEstimate(SoftFloat64.FPRSqrtEstimate);
+            var dlgSoftFloat64FPRSqrtEstimateFpscr = new SoftFloat64FPRSqrtEstimateFpscr(SoftFloat64.FPRSqrtEstimateFpscr); // A32 only.
+            var dlgSoftFloat64FPRSqrtStep = new SoftFloat64FPRSqrtStep(SoftFloat64.FPRSqrtStep); // A32 only.
+            var dlgSoftFloat64FPRSqrtStepFused = new SoftFloat64FPRSqrtStepFused(SoftFloat64.FPRSqrtStepFused);
+            var dlgSoftFloat64FPSqrt = new SoftFloat64FPSqrt(SoftFloat64.FPSqrt);
+            var dlgSoftFloat64FPSub = new SoftFloat64FPSub(SoftFloat64.FPSub);
 
-            SetDelegateInfo(new SoftFloat64_16FPConvert(SoftFloat64_16.FPConvert));
+            var dlgSoftFloat64_16FPConvert = new SoftFloat64_16FPConvert(SoftFloat64_16.FPConvert);
+
+            SetDelegateInfo(dlgMathAbs, Marshal.GetFunctionPointerForDelegate<MathAbs>(dlgMathAbs));
+            SetDelegateInfo(dlgMathCeiling, Marshal.GetFunctionPointerForDelegate<MathCeiling>(dlgMathCeiling));
+            SetDelegateInfo(dlgMathFloor, Marshal.GetFunctionPointerForDelegate<MathFloor>(dlgMathFloor));
+            SetDelegateInfo(dlgMathRound, Marshal.GetFunctionPointerForDelegate<MathRound>(dlgMathRound));
+            SetDelegateInfo(dlgMathTruncate, Marshal.GetFunctionPointerForDelegate<MathTruncate>(dlgMathTruncate));
+
+            SetDelegateInfo(dlgMathFAbs, Marshal.GetFunctionPointerForDelegate<MathFAbs>(dlgMathFAbs));
+            SetDelegateInfo(dlgMathFCeiling, Marshal.GetFunctionPointerForDelegate<MathFCeiling>(dlgMathFCeiling));
+            SetDelegateInfo(dlgMathFFloor, Marshal.GetFunctionPointerForDelegate<MathFFloor>(dlgMathFFloor));
+            SetDelegateInfo(dlgMathFRound, Marshal.GetFunctionPointerForDelegate<MathFRound>(dlgMathFRound));
+            SetDelegateInfo(dlgMathFTruncate, Marshal.GetFunctionPointerForDelegate<MathFTruncate>(dlgMathFTruncate));
+
+            SetDelegateInfo(dlgNativeInterfaceBreak, Marshal.GetFunctionPointerForDelegate<NativeInterfaceBreak>(dlgNativeInterfaceBreak));
+            SetDelegateInfo(dlgNativeInterfaceCheckSynchronization, Marshal.GetFunctionPointerForDelegate<NativeInterfaceCheckSynchronization>(dlgNativeInterfaceCheckSynchronization));
+            SetDelegateInfo(dlgNativeInterfaceEnqueueForRejit, Marshal.GetFunctionPointerForDelegate<NativeInterfaceEnqueueForRejit>(dlgNativeInterfaceEnqueueForRejit));
+            SetDelegateInfo(dlgNativeInterfaceGetCntfrqEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntfrqEl0>(dlgNativeInterfaceGetCntfrqEl0));
+            SetDelegateInfo(dlgNativeInterfaceGetCntpctEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntpctEl0>(dlgNativeInterfaceGetCntpctEl0));
+            SetDelegateInfo(dlgNativeInterfaceGetCntvctEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCntvctEl0>(dlgNativeInterfaceGetCntvctEl0));
+            SetDelegateInfo(dlgNativeInterfaceGetCtrEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetCtrEl0>(dlgNativeInterfaceGetCtrEl0));
+            SetDelegateInfo(dlgNativeInterfaceGetDczidEl0, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetDczidEl0>(dlgNativeInterfaceGetDczidEl0));
+            SetDelegateInfo(dlgNativeInterfaceGetFunctionAddress, Marshal.GetFunctionPointerForDelegate<NativeInterfaceGetFunctionAddress>(dlgNativeInterfaceGetFunctionAddress));
+            SetDelegateInfo(dlgNativeInterfaceInvalidateCacheLine, Marshal.GetFunctionPointerForDelegate<NativeInterfaceInvalidateCacheLine>(dlgNativeInterfaceInvalidateCacheLine));
+            SetDelegateInfo(dlgNativeInterfaceReadByte, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadByte>(dlgNativeInterfaceReadByte));
+            SetDelegateInfo(dlgNativeInterfaceReadUInt16, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt16>(dlgNativeInterfaceReadUInt16));
+            SetDelegateInfo(dlgNativeInterfaceReadUInt32, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt32>(dlgNativeInterfaceReadUInt32));
+            SetDelegateInfo(dlgNativeInterfaceReadUInt64, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadUInt64>(dlgNativeInterfaceReadUInt64));
+            SetDelegateInfo(dlgNativeInterfaceReadVector128, Marshal.GetFunctionPointerForDelegate<NativeInterfaceReadVector128>(dlgNativeInterfaceReadVector128));
+            SetDelegateInfo(dlgNativeInterfaceSignalMemoryTracking, Marshal.GetFunctionPointerForDelegate<NativeInterfaceSignalMemoryTracking>(dlgNativeInterfaceSignalMemoryTracking));
+            SetDelegateInfo(dlgNativeInterfaceSupervisorCall, Marshal.GetFunctionPointerForDelegate<NativeInterfaceSupervisorCall>(dlgNativeInterfaceSupervisorCall));
+            SetDelegateInfo(dlgNativeInterfaceThrowInvalidMemoryAccess, Marshal.GetFunctionPointerForDelegate<NativeInterfaceThrowInvalidMemoryAccess>(dlgNativeInterfaceThrowInvalidMemoryAccess));
+            SetDelegateInfo(dlgNativeInterfaceUndefined, Marshal.GetFunctionPointerForDelegate<NativeInterfaceUndefined>(dlgNativeInterfaceUndefined));
+            SetDelegateInfo(dlgNativeInterfaceWriteByte, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteByte>(dlgNativeInterfaceWriteByte));
+            SetDelegateInfo(dlgNativeInterfaceWriteUInt16, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt16>(dlgNativeInterfaceWriteUInt16));
+            SetDelegateInfo(dlgNativeInterfaceWriteUInt32, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt32>(dlgNativeInterfaceWriteUInt32));
+            SetDelegateInfo(dlgNativeInterfaceWriteUInt64, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteUInt64>(dlgNativeInterfaceWriteUInt64));
+            SetDelegateInfo(dlgNativeInterfaceWriteVector128, Marshal.GetFunctionPointerForDelegate<NativeInterfaceWriteVector128>(dlgNativeInterfaceWriteVector128));
+
+            SetDelegateInfo(dlgSoftFallbackCountLeadingSigns, Marshal.GetFunctionPointerForDelegate<SoftFallbackCountLeadingSigns>(dlgSoftFallbackCountLeadingSigns));
+            SetDelegateInfo(dlgSoftFallbackCountLeadingZeros, Marshal.GetFunctionPointerForDelegate<SoftFallbackCountLeadingZeros>(dlgSoftFallbackCountLeadingZeros));
+            SetDelegateInfo(dlgSoftFallbackCrc32b, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32b>(dlgSoftFallbackCrc32b));
+            SetDelegateInfo(dlgSoftFallbackCrc32cb, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cb>(dlgSoftFallbackCrc32cb));
+            SetDelegateInfo(dlgSoftFallbackCrc32ch, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32ch>(dlgSoftFallbackCrc32ch));
+            SetDelegateInfo(dlgSoftFallbackCrc32cw, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cw>(dlgSoftFallbackCrc32cw));
+            SetDelegateInfo(dlgSoftFallbackCrc32cx, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32cx>(dlgSoftFallbackCrc32cx));
+            SetDelegateInfo(dlgSoftFallbackCrc32h, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32h>(dlgSoftFallbackCrc32h));
+            SetDelegateInfo(dlgSoftFallbackCrc32w, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32w>(dlgSoftFallbackCrc32w));
+            SetDelegateInfo(dlgSoftFallbackCrc32x, Marshal.GetFunctionPointerForDelegate<SoftFallbackCrc32x>(dlgSoftFallbackCrc32x));
+            SetDelegateInfo(dlgSoftFallbackDecrypt, Marshal.GetFunctionPointerForDelegate<SoftFallbackDecrypt>(dlgSoftFallbackDecrypt));
+            SetDelegateInfo(dlgSoftFallbackEncrypt, Marshal.GetFunctionPointerForDelegate<SoftFallbackEncrypt>(dlgSoftFallbackEncrypt));
+            SetDelegateInfo(dlgSoftFallbackFixedRotate, Marshal.GetFunctionPointerForDelegate<SoftFallbackFixedRotate>(dlgSoftFallbackFixedRotate));
+            SetDelegateInfo(dlgSoftFallbackHashChoose, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashChoose>(dlgSoftFallbackHashChoose));
+            SetDelegateInfo(dlgSoftFallbackHashLower, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashLower>(dlgSoftFallbackHashLower));
+            SetDelegateInfo(dlgSoftFallbackHashMajority, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashMajority>(dlgSoftFallbackHashMajority));
+            SetDelegateInfo(dlgSoftFallbackHashParity, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashParity>(dlgSoftFallbackHashParity));
+            SetDelegateInfo(dlgSoftFallbackHashUpper, Marshal.GetFunctionPointerForDelegate<SoftFallbackHashUpper>(dlgSoftFallbackHashUpper));
+            SetDelegateInfo(dlgSoftFallbackInverseMixColumns, Marshal.GetFunctionPointerForDelegate<SoftFallbackInverseMixColumns>(dlgSoftFallbackInverseMixColumns));
+            SetDelegateInfo(dlgSoftFallbackMixColumns, Marshal.GetFunctionPointerForDelegate<SoftFallbackMixColumns>(dlgSoftFallbackMixColumns));
+            SetDelegateInfo(dlgSoftFallbackPolynomialMult64_128, Marshal.GetFunctionPointerForDelegate<SoftFallbackPolynomialMult64_128>(dlgSoftFallbackPolynomialMult64_128));
+            SetDelegateInfo(dlgSoftFallbackSatF32ToS32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToS32>(dlgSoftFallbackSatF32ToS32));
+            SetDelegateInfo(dlgSoftFallbackSatF32ToS64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToS64>(dlgSoftFallbackSatF32ToS64));
+            SetDelegateInfo(dlgSoftFallbackSatF32ToU32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToU32>(dlgSoftFallbackSatF32ToU32));
+            SetDelegateInfo(dlgSoftFallbackSatF32ToU64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF32ToU64>(dlgSoftFallbackSatF32ToU64));
+            SetDelegateInfo(dlgSoftFallbackSatF64ToS32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToS32>(dlgSoftFallbackSatF64ToS32));
+            SetDelegateInfo(dlgSoftFallbackSatF64ToS64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToS64>(dlgSoftFallbackSatF64ToS64));
+            SetDelegateInfo(dlgSoftFallbackSatF64ToU32, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToU32>(dlgSoftFallbackSatF64ToU32));
+            SetDelegateInfo(dlgSoftFallbackSatF64ToU64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSatF64ToU64>(dlgSoftFallbackSatF64ToU64));
+            SetDelegateInfo(dlgSoftFallbackSha1SchedulePart1, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha1SchedulePart1>(dlgSoftFallbackSha1SchedulePart1));
+            SetDelegateInfo(dlgSoftFallbackSha1SchedulePart2, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha1SchedulePart2>(dlgSoftFallbackSha1SchedulePart2));
+            SetDelegateInfo(dlgSoftFallbackSha256SchedulePart1, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha256SchedulePart1>(dlgSoftFallbackSha256SchedulePart1));
+            SetDelegateInfo(dlgSoftFallbackSha256SchedulePart2, Marshal.GetFunctionPointerForDelegate<SoftFallbackSha256SchedulePart2>(dlgSoftFallbackSha256SchedulePart2));
+            SetDelegateInfo(dlgSoftFallbackSignedShrImm64, Marshal.GetFunctionPointerForDelegate<SoftFallbackSignedShrImm64>(dlgSoftFallbackSignedShrImm64));
+            SetDelegateInfo(dlgSoftFallbackTbl1, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl1>(dlgSoftFallbackTbl1));
+            SetDelegateInfo(dlgSoftFallbackTbl2, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl2>(dlgSoftFallbackTbl2));
+            SetDelegateInfo(dlgSoftFallbackTbl3, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl3>(dlgSoftFallbackTbl3));
+            SetDelegateInfo(dlgSoftFallbackTbl4, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbl4>(dlgSoftFallbackTbl4));
+            SetDelegateInfo(dlgSoftFallbackTbx1, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx1>(dlgSoftFallbackTbx1));
+            SetDelegateInfo(dlgSoftFallbackTbx2, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx2>(dlgSoftFallbackTbx2));
+            SetDelegateInfo(dlgSoftFallbackTbx3, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx3>(dlgSoftFallbackTbx3));
+            SetDelegateInfo(dlgSoftFallbackTbx4, Marshal.GetFunctionPointerForDelegate<SoftFallbackTbx4>(dlgSoftFallbackTbx4));
+            SetDelegateInfo(dlgSoftFallbackUnsignedShrImm64, Marshal.GetFunctionPointerForDelegate<SoftFallbackUnsignedShrImm64>(dlgSoftFallbackUnsignedShrImm64));
+
+            SetDelegateInfo(dlgSoftFloat16_32FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat16_32FPConvert>(dlgSoftFloat16_32FPConvert));
+            SetDelegateInfo(dlgSoftFloat16_64FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat16_64FPConvert>(dlgSoftFloat16_64FPConvert));
+
+            SetDelegateInfo(dlgSoftFloat32FPAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPAdd>(dlgSoftFloat32FPAdd));
+            SetDelegateInfo(dlgSoftFloat32FPAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPAddFpscr>(dlgSoftFloat32FPAddFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPCompare, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompare>(dlgSoftFloat32FPCompare));
+            SetDelegateInfo(dlgSoftFloat32FPCompareEQ, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareEQ>(dlgSoftFloat32FPCompareEQ));
+            SetDelegateInfo(dlgSoftFloat32FPCompareEQFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareEQFpscr>(dlgSoftFloat32FPCompareEQFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPCompareGE, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGE>(dlgSoftFloat32FPCompareGE));
+            SetDelegateInfo(dlgSoftFloat32FPCompareGEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGEFpscr>(dlgSoftFloat32FPCompareGEFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPCompareGT, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGT>(dlgSoftFloat32FPCompareGT));
+            SetDelegateInfo(dlgSoftFloat32FPCompareGTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareGTFpscr>(dlgSoftFloat32FPCompareGTFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPCompareLE, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLE>(dlgSoftFloat32FPCompareLE));
+            SetDelegateInfo(dlgSoftFloat32FPCompareLEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLEFpscr>(dlgSoftFloat32FPCompareLEFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPCompareLT, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLT>(dlgSoftFloat32FPCompareLT));
+            SetDelegateInfo(dlgSoftFloat32FPCompareLTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPCompareLTFpscr>(dlgSoftFloat32FPCompareLTFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPDiv, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPDiv>(dlgSoftFloat32FPDiv));
+            SetDelegateInfo(dlgSoftFloat32FPMax, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMax>(dlgSoftFloat32FPMax));
+            SetDelegateInfo(dlgSoftFloat32FPMaxFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxFpscr>(dlgSoftFloat32FPMaxFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMaxNum, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxNum>(dlgSoftFloat32FPMaxNum));
+            SetDelegateInfo(dlgSoftFloat32FPMaxNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMaxNumFpscr>(dlgSoftFloat32FPMaxNumFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMin, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMin>(dlgSoftFloat32FPMin));
+            SetDelegateInfo(dlgSoftFloat32FPMinFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinFpscr>(dlgSoftFloat32FPMinFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMinNum, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinNum>(dlgSoftFloat32FPMinNum));
+            SetDelegateInfo(dlgSoftFloat32FPMinNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMinNumFpscr>(dlgSoftFloat32FPMinNumFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMul, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMul>(dlgSoftFloat32FPMul));
+            SetDelegateInfo(dlgSoftFloat32FPMulFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulFpscr>(dlgSoftFloat32FPMulFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulAdd>(dlgSoftFloat32FPMulAdd));
+            SetDelegateInfo(dlgSoftFloat32FPMulAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulAddFpscr>(dlgSoftFloat32FPMulAddFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulSub>(dlgSoftFloat32FPMulSub));
+            SetDelegateInfo(dlgSoftFloat32FPMulSubFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulSubFpscr>(dlgSoftFloat32FPMulSubFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPMulX, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPMulX>(dlgSoftFloat32FPMulX));
+            SetDelegateInfo(dlgSoftFloat32FPNegMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPNegMulAdd>(dlgSoftFloat32FPNegMulAdd));
+            SetDelegateInfo(dlgSoftFloat32FPNegMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPNegMulSub>(dlgSoftFloat32FPNegMulSub));
+            SetDelegateInfo(dlgSoftFloat32FPRecipEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipEstimate>(dlgSoftFloat32FPRecipEstimate));
+            SetDelegateInfo(dlgSoftFloat32FPRecipEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipEstimateFpscr>(dlgSoftFloat32FPRecipEstimateFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPRecipStep, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipStep>(dlgSoftFloat32FPRecipStep)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPRecipStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecipStepFused>(dlgSoftFloat32FPRecipStepFused));
+            SetDelegateInfo(dlgSoftFloat32FPRecpX, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRecpX>(dlgSoftFloat32FPRecpX));
+            SetDelegateInfo(dlgSoftFloat32FPRSqrtEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtEstimate>(dlgSoftFloat32FPRSqrtEstimate));
+            SetDelegateInfo(dlgSoftFloat32FPRSqrtEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtEstimateFpscr>(dlgSoftFloat32FPRSqrtEstimateFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPRSqrtStep, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtStep>(dlgSoftFloat32FPRSqrtStep)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat32FPRSqrtStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPRSqrtStepFused>(dlgSoftFloat32FPRSqrtStepFused));
+            SetDelegateInfo(dlgSoftFloat32FPSqrt, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPSqrt>(dlgSoftFloat32FPSqrt));
+            SetDelegateInfo(dlgSoftFloat32FPSub, Marshal.GetFunctionPointerForDelegate<SoftFloat32FPSub>(dlgSoftFloat32FPSub));
+
+            SetDelegateInfo(dlgSoftFloat32_16FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat32_16FPConvert>(dlgSoftFloat32_16FPConvert));
+
+            SetDelegateInfo(dlgSoftFloat64FPAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPAdd>(dlgSoftFloat64FPAdd));
+            SetDelegateInfo(dlgSoftFloat64FPAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPAddFpscr>(dlgSoftFloat64FPAddFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPCompare, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompare>(dlgSoftFloat64FPCompare));
+            SetDelegateInfo(dlgSoftFloat64FPCompareEQ, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareEQ>(dlgSoftFloat64FPCompareEQ));
+            SetDelegateInfo(dlgSoftFloat64FPCompareEQFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareEQFpscr>(dlgSoftFloat64FPCompareEQFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPCompareGE, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGE>(dlgSoftFloat64FPCompareGE));
+            SetDelegateInfo(dlgSoftFloat64FPCompareGEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGEFpscr>(dlgSoftFloat64FPCompareGEFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPCompareGT, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGT>(dlgSoftFloat64FPCompareGT));
+            SetDelegateInfo(dlgSoftFloat64FPCompareGTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareGTFpscr>(dlgSoftFloat64FPCompareGTFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPCompareLE, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLE>(dlgSoftFloat64FPCompareLE));
+            SetDelegateInfo(dlgSoftFloat64FPCompareLEFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLEFpscr>(dlgSoftFloat64FPCompareLEFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPCompareLT, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLT>(dlgSoftFloat64FPCompareLT));
+            SetDelegateInfo(dlgSoftFloat64FPCompareLTFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPCompareLTFpscr>(dlgSoftFloat64FPCompareLTFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPDiv, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPDiv>(dlgSoftFloat64FPDiv));
+            SetDelegateInfo(dlgSoftFloat64FPMax, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMax>(dlgSoftFloat64FPMax));
+            SetDelegateInfo(dlgSoftFloat64FPMaxFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxFpscr>(dlgSoftFloat64FPMaxFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMaxNum, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxNum>(dlgSoftFloat64FPMaxNum));
+            SetDelegateInfo(dlgSoftFloat64FPMaxNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMaxNumFpscr>(dlgSoftFloat64FPMaxNumFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMin, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMin>(dlgSoftFloat64FPMin));
+            SetDelegateInfo(dlgSoftFloat64FPMinFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinFpscr>(dlgSoftFloat64FPMinFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMinNum, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinNum>(dlgSoftFloat64FPMinNum));
+            SetDelegateInfo(dlgSoftFloat64FPMinNumFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMinNumFpscr>(dlgSoftFloat64FPMinNumFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMul, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMul>(dlgSoftFloat64FPMul));
+            SetDelegateInfo(dlgSoftFloat64FPMulFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulFpscr>(dlgSoftFloat64FPMulFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulAdd>(dlgSoftFloat64FPMulAdd));
+            SetDelegateInfo(dlgSoftFloat64FPMulAddFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulAddFpscr>(dlgSoftFloat64FPMulAddFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulSub>(dlgSoftFloat64FPMulSub));
+            SetDelegateInfo(dlgSoftFloat64FPMulSubFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulSubFpscr>(dlgSoftFloat64FPMulSubFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPMulX, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPMulX>(dlgSoftFloat64FPMulX));
+            SetDelegateInfo(dlgSoftFloat64FPNegMulAdd, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPNegMulAdd>(dlgSoftFloat64FPNegMulAdd));
+            SetDelegateInfo(dlgSoftFloat64FPNegMulSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPNegMulSub>(dlgSoftFloat64FPNegMulSub));
+            SetDelegateInfo(dlgSoftFloat64FPRecipEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipEstimate>(dlgSoftFloat64FPRecipEstimate));
+            SetDelegateInfo(dlgSoftFloat64FPRecipEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipEstimateFpscr>(dlgSoftFloat64FPRecipEstimateFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPRecipStep, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipStep>(dlgSoftFloat64FPRecipStep)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPRecipStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecipStepFused>(dlgSoftFloat64FPRecipStepFused));
+            SetDelegateInfo(dlgSoftFloat64FPRecpX, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRecpX>(dlgSoftFloat64FPRecpX));
+            SetDelegateInfo(dlgSoftFloat64FPRSqrtEstimate, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtEstimate>(dlgSoftFloat64FPRSqrtEstimate));
+            SetDelegateInfo(dlgSoftFloat64FPRSqrtEstimateFpscr, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtEstimateFpscr>(dlgSoftFloat64FPRSqrtEstimateFpscr)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPRSqrtStep, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtStep>(dlgSoftFloat64FPRSqrtStep)); // A32 only.
+            SetDelegateInfo(dlgSoftFloat64FPRSqrtStepFused, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPRSqrtStepFused>(dlgSoftFloat64FPRSqrtStepFused));
+            SetDelegateInfo(dlgSoftFloat64FPSqrt, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPSqrt>(dlgSoftFloat64FPSqrt));
+            SetDelegateInfo(dlgSoftFloat64FPSub, Marshal.GetFunctionPointerForDelegate<SoftFloat64FPSub>(dlgSoftFloat64FPSub));
+
+            SetDelegateInfo(dlgSoftFloat64_16FPConvert, Marshal.GetFunctionPointerForDelegate<SoftFloat64_16FPConvert>(dlgSoftFloat64_16FPConvert));
         }
 
         private delegate double MathAbs(double value);

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/ARMeilleure/Translation/TranslatorQueue.cs

@@ -80,7 +80,10 @@ namespace ARMeilleure.Translation
                         return true;
                     }
 
-                    Monitor.Wait(Sync);
+                    if (!_disposed)
+                    {
+                        Monitor.Wait(Sync);
+                    }
                 }
             }
 

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/GraphicsDriver/DriverUtilities.cs

@@ -1,13 +1,33 @@
+using Ryujinx.Common.Utilities;
 using System;
 
 namespace Ryujinx.Common.GraphicsDriver
 {
     public static class DriverUtilities
     {
+        private static void AddMesaFlags(string envVar, string newFlags)
+        {
+            string existingFlags = Environment.GetEnvironmentVariable(envVar);
+
+            string flags = existingFlags == null ? newFlags : $"{existingFlags},{newFlags}";
+
+            OsUtils.SetEnvironmentVariableNoCaching(envVar, flags);
+        }
+
+        public static void InitDriverConfig(bool oglThreading)
+        {
+            if (OperatingSystem.IsLinux())
+            {
+                AddMesaFlags("RADV_DEBUG", "nodcc");
+            }
+
+            ToggleOGLThreading(oglThreading);
+        }
+
         public static void ToggleOGLThreading(bool enabled)
         {
-            Environment.SetEnvironmentVariable("mesa_glthread", enabled.ToString().ToLower());
-            Environment.SetEnvironmentVariable("__GL_THREADED_OPTIMIZATIONS", enabled ? "1" : "0");
+            OsUtils.SetEnvironmentVariableNoCaching("mesa_glthread", enabled.ToString().ToLower());
+            OsUtils.SetEnvironmentVariableNoCaching("__GL_THREADED_OPTIMIZATIONS", enabled ? "1" : "0");
 
             try
             {

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/OsUtils.cs

@@ -0,0 +1,24 @@
+using System;
+using System.Diagnostics;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Common.Utilities
+{
+    public partial class OsUtils
+    {
+        [LibraryImport("libc", SetLastError = true)]
+        private static partial int setenv([MarshalAs(UnmanagedType.LPStr)] string name, [MarshalAs(UnmanagedType.LPStr)] string value, int overwrite);
+
+        public static void SetEnvironmentVariableNoCaching(string key, string value)
+        {
+            // Set the value in the cached environment variables, too.
+            Environment.SetEnvironmentVariable(key, value);
+
+            if (!OperatingSystem.IsWindows())
+            {
+                int res = setenv(key, value, 1);
+                Debug.Assert(res != -1);
+            }
+        }
+    }
+}

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.Device/DeviceState.cs

@@ -39,7 +39,10 @@ namespace Ryujinx.Graphics.Device
             {
                 var field = fields[fieldIndex];
 
-                int sizeOfField = SizeCalculator.SizeOf(field.FieldType);
+                var currentFieldOffset = (int)Marshal.OffsetOf<TState>(field.Name);
+                var nextFieldOffset = fieldIndex + 1 == fields.Length ? Unsafe.SizeOf<TState>() : (int)Marshal.OffsetOf<TState>(fields[fieldIndex + 1].Name);
+
+                int sizeOfField = nextFieldOffset - currentFieldOffset;
 
                 for (int i = 0; i < ((sizeOfField + 3) & ~3); i += 4)
                 {

src/Ryujinx.Graphics.Device/SizeCalculator.cs

@@ -1,63 +0,0 @@
-using System;
-using System.Reflection;
-
-namespace Ryujinx.Graphics.Device
-{
-    public static class SizeCalculator
-    {
-        public static int SizeOf(Type type)
-        {
-            // Is type a enum type?
-            if (type.IsEnum)
-            {
-                type = type.GetEnumUnderlyingType();
-            }
-
-            // Is type a pointer type?
-            if (type.IsPointer || type == typeof(IntPtr) || type == typeof(UIntPtr))
-            {
-                return IntPtr.Size;
-            }
-
-            // Is type a struct type?
-            if (type.IsValueType && !type.IsPrimitive)
-            {
-                // Check if the struct has a explicit size, if so, return that.
-                if (type.StructLayoutAttribute.Size != 0)
-                {
-                    return type.StructLayoutAttribute.Size;
-                }
-
-                // Otherwise we calculate the sum of the sizes of all fields.
-                int size = 0;
-                var fields = type.GetFields(BindingFlags.Public | BindingFlags.NonPublic | BindingFlags.Instance);
-
-                for (int fieldIndex = 0; fieldIndex < fields.Length; fieldIndex++)
-                {
-                    size += SizeOf(fields[fieldIndex].FieldType);
-                }
-
-                return size;
-            }
-
-            // Primitive types.
-            return (Type.GetTypeCode(type)) switch
-            {
-                TypeCode.SByte => sizeof(sbyte),
-                TypeCode.Byte => sizeof(byte),
-                TypeCode.Int16 => sizeof(short),
-                TypeCode.UInt16 => sizeof(ushort),
-                TypeCode.Int32 => sizeof(int),
-                TypeCode.UInt32 => sizeof(uint),
-                TypeCode.Int64 => sizeof(long),
-                TypeCode.UInt64 => sizeof(ulong),
-                TypeCode.Char => sizeof(char),
-                TypeCode.Single => sizeof(float),
-                TypeCode.Double => sizeof(double),
-                TypeCode.Decimal => sizeof(decimal),
-                TypeCode.Boolean => sizeof(bool),
-                _ => throw new ArgumentException($"Length for type \"{type.Name}\" is unknown."),
-            };
-        }
-    }
-}

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.GAL/UpscaleType.cs

@@ -5,5 +5,6 @@ namespace Ryujinx.Graphics.GAL
         Bilinear,
         Nearest,
         Fsr,
+        Area,
     }
 }

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

@@ -276,8 +276,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
                     dstBaseOffset += dstStride * (yCount - 1);
                 }
 
-                ReadOnlySpan<byte> srcSpan = memoryManager.GetSpan(srcGpuVa + (ulong)srcBaseOffset, srcSize, true);
-
                 // If remapping is disabled, we always copy the components directly, in order.
                 // If it's enabled, but the mapping is just XYZW, we also copy them in order.
                 bool isIdentityRemap = !remap ||
@@ -289,6 +287,52 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
                 bool completeSource = IsTextureCopyComplete(src, srcLinear, srcBpp, srcStride, xCount, yCount);
                 bool completeDest = IsTextureCopyComplete(dst, dstLinear, dstBpp, dstStride, xCount, yCount);
 
+                // Check if the source texture exists on the GPU, if it does, do a GPU side copy.
+                // Otherwise, we would need to flush the source texture which is costly.
+                // We don't expect the source to be linear in such cases, as linear source usually indicates buffer or CPU written data.
+
+                if (completeSource && completeDest && !srcLinear && isIdentityRemap)
+                {
+                    var source = memoryManager.Physical.TextureCache.FindTexture(
+                        memoryManager,
+                        srcGpuVa,
+                        srcBpp,
+                        srcStride,
+                        src.Height,
+                        xCount,
+                        yCount,
+                        srcLinear,
+                        src.MemoryLayout.UnpackGobBlocksInY(),
+                        src.MemoryLayout.UnpackGobBlocksInZ());
+
+                    if (source != null && source.Height == yCount)
+                    {
+                        source.SynchronizeMemory();
+
+                        var target = memoryManager.Physical.TextureCache.FindOrCreateTexture(
+                            memoryManager,
+                            source.Info.FormatInfo,
+                            dstGpuVa,
+                            xCount,
+                            yCount,
+                            dstStride,
+                            dstLinear,
+                            dst.MemoryLayout.UnpackGobBlocksInY(),
+                            dst.MemoryLayout.UnpackGobBlocksInZ());
+
+                        if (source.ScaleFactor != target.ScaleFactor)
+                        {
+                            target.PropagateScale(source);
+                        }
+
+                        source.HostTexture.CopyTo(target.HostTexture, 0, 0);
+                        target.SignalModified();
+                        return;
+                    }
+                }
+
+                ReadOnlySpan<byte> srcSpan = memoryManager.GetSpan(srcGpuVa + (ulong)srcBaseOffset, srcSize, true);
+
                 // Try to set the texture data directly,
                 // but only if we are doing a complete copy,
                 // and not for block linear to linear copies, since those are typically accessed from the CPU.

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/Engine/Threed/StateUpdateTracker.cs

@@ -79,7 +79,10 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             {
                 var field = fields[fieldIndex];
 
-                int sizeOfField = SizeCalculator.SizeOf(field.FieldType);
+                var currentFieldOffset = (int)Marshal.OffsetOf<TState>(field.Name);
+                var nextFieldOffset = fieldIndex + 1 == fields.Length ? Unsafe.SizeOf<TState>() : (int)Marshal.OffsetOf<TState>(fields[fieldIndex + 1].Name);
+
+                int sizeOfField = nextFieldOffset - currentFieldOffset;
 
                 if (fieldToDelegate.TryGetValue(field.Name, out int entryIndex))
                 {

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

@@ -415,7 +415,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 #pragma warning disable CS0649 // Field is never assigned to
         public int Width;
         public int Height;
-        public int Depth;
+        public ushort Depth;
+        public ushort Flags;
+
+        public readonly bool UnpackIsLayered()
+        {
+            return (Flags & 1) == 0;
+        }
 #pragma warning restore CS0649
     }
 

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

@@ -340,7 +340,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             /// <returns>True if any used entries of the pool might have been modified, false otherwise</returns>
             public bool SamplerPoolModified()
             {
-                return SamplerPool.WasModified(ref _samplerPoolSequence);
+                return SamplerPool != null && SamplerPool.WasModified(ref _samplerPoolSequence);
             }
         }
 
@@ -516,12 +516,15 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
 
                 // Check if any of our cached samplers changed on the pool.
-                foreach ((int samplerId, (Sampler sampler, SamplerDescriptor descriptor)) in SamplerIds)
+                if (SamplerPool != null)
                 {
-                    if (SamplerPool.GetCachedItem(samplerId) != sampler ||
-                        (sampler == null && SamplerPool.IsValidId(samplerId) && !SamplerPool.GetDescriptorRef(samplerId).Equals(descriptor)))
+                    foreach ((int samplerId, (Sampler sampler, SamplerDescriptor descriptor)) in SamplerIds)
                     {
-                        return true;
+                        if (SamplerPool.GetCachedItem(samplerId) != sampler ||
+                            (sampler == null && SamplerPool.IsValidId(samplerId) && !SamplerPool.GetDescriptorRef(samplerId).Equals(descriptor)))
+                        {
+                            return true;
+                        }
                     }
                 }
 
@@ -899,13 +902,19 @@ namespace Ryujinx.Graphics.Gpu.Image
                     }
                 }
 
-                Sampler sampler = samplerPool?.Get(samplerId);
-
                 entry.TextureIds[textureId] = (texture, descriptor);
-                entry.SamplerIds[samplerId] = (sampler, samplerPool?.GetDescriptorRef(samplerId) ?? default);
 
                 ITexture hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
-                ISampler hostSampler = sampler?.GetHostSampler(texture);
+                ISampler hostSampler = null;
+
+                if (!isImage && bindingInfo.Target != Target.TextureBuffer)
+                {
+                    Sampler sampler = samplerPool?.Get(samplerId);
+
+                    entry.SamplerIds[samplerId] = (sampler, samplerPool?.GetDescriptorRef(samplerId) ?? default);
+
+                    hostSampler = sampler?.GetHostSampler(texture);
+                }
 
                 Format format = bindingInfo.Format;
 
@@ -1113,6 +1122,15 @@ 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 +1142,20 @@ 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/TextureCache.cs

@@ -347,6 +347,53 @@ namespace Ryujinx.Graphics.Gpu.Image
             return texture;
         }
 
+        /// <summary>
+        /// Tries to find an existing texture, or create a new one if not found.
+        /// </summary>
+        /// <param name="memoryManager">GPU memory manager where the texture is mapped</param>
+        /// <param name="formatInfo">Format of the texture</param>
+        /// <param name="gpuAddress">GPU virtual address of the texture</param>
+        /// <param name="xCount">Texture width in bytes</param>
+        /// <param name="yCount">Texture height</param>
+        /// <param name="stride">Texture stride if linear, otherwise ignored</param>
+        /// <param name="isLinear">Indicates if the texture is linear or block linear</param>
+        /// <param name="gobBlocksInY">GOB blocks in Y for block linear textures</param>
+        /// <param name="gobBlocksInZ">GOB blocks in Z for 3D block linear textures</param>
+        /// <returns>The texture</returns>
+        public Texture FindOrCreateTexture(
+            MemoryManager memoryManager,
+            FormatInfo formatInfo,
+            ulong gpuAddress,
+            int xCount,
+            int yCount,
+            int stride,
+            bool isLinear,
+            int gobBlocksInY,
+            int gobBlocksInZ)
+        {
+            TextureInfo info = new(
+                gpuAddress,
+                xCount / formatInfo.BytesPerPixel,
+                yCount,
+                1,
+                1,
+                1,
+                1,
+                stride,
+                isLinear,
+                gobBlocksInY,
+                gobBlocksInZ,
+                1,
+                Target.Texture2D,
+                formatInfo);
+
+            Texture texture = FindOrCreateTexture(memoryManager, TextureSearchFlags.ForCopy, info, 0, sizeHint: new Size(xCount, yCount, 1));
+
+            texture?.SynchronizeMemory();
+
+            return texture;
+        }
+
         /// <summary>
         /// Tries to find an existing texture, or create a new one if not found.
         /// </summary>
@@ -468,13 +515,11 @@ namespace Ryujinx.Graphics.Gpu.Image
             int gobBlocksInY = dsState.MemoryLayout.UnpackGobBlocksInY();
             int gobBlocksInZ = dsState.MemoryLayout.UnpackGobBlocksInZ();
 
+            layered &= size.UnpackIsLayered();
+
             Target target;
 
-            if (dsState.MemoryLayout.UnpackIsTarget3D())
-            {
-                target = Target.Texture3D;
-            }
-            else if ((samplesInX | samplesInY) != 1)
+            if ((samplesInX | samplesInY) != 1)
             {
                 target = size.Depth > 1 && layered
                     ? Target.Texture2DMultisampleArray

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/Effects/AreaScalingFilter.cs

@@ -0,0 +1,106 @@
+using OpenTK.Graphics.OpenGL;
+using Ryujinx.Common;
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.OpenGL.Image;
+using System;
+using static Ryujinx.Graphics.OpenGL.Effects.ShaderHelper;
+
+namespace Ryujinx.Graphics.OpenGL.Effects
+{
+    internal class AreaScalingFilter : IScalingFilter
+    {
+        private readonly OpenGLRenderer _renderer;
+        private int _inputUniform;
+        private int _outputUniform;
+        private int _srcX0Uniform;
+        private int _srcX1Uniform;
+        private int _srcY0Uniform;
+        private int _scalingShaderProgram;
+        private int _srcY1Uniform;
+        private int _dstX0Uniform;
+        private int _dstX1Uniform;
+        private int _dstY0Uniform;
+        private int _dstY1Uniform;
+
+        public float Level { get; set; }
+
+        public AreaScalingFilter(OpenGLRenderer renderer)
+        {
+            Initialize();
+
+            _renderer = renderer;
+        }
+
+        public void Dispose()
+        {
+            if (_scalingShaderProgram != 0)
+            {
+                GL.DeleteProgram(_scalingShaderProgram);
+            }
+        }
+
+        private void Initialize()
+        {
+            var scalingShader = EmbeddedResources.ReadAllText("Ryujinx.Graphics.OpenGL/Effects/Shaders/area_scaling.glsl");
+
+            _scalingShaderProgram = CompileProgram(scalingShader, ShaderType.ComputeShader);
+
+            _inputUniform = GL.GetUniformLocation(_scalingShaderProgram, "Source");
+            _outputUniform = GL.GetUniformLocation(_scalingShaderProgram, "imgOutput");
+
+            _srcX0Uniform = GL.GetUniformLocation(_scalingShaderProgram, "srcX0");
+            _srcX1Uniform = GL.GetUniformLocation(_scalingShaderProgram, "srcX1");
+            _srcY0Uniform = GL.GetUniformLocation(_scalingShaderProgram, "srcY0");
+            _srcY1Uniform = GL.GetUniformLocation(_scalingShaderProgram, "srcY1");
+            _dstX0Uniform = GL.GetUniformLocation(_scalingShaderProgram, "dstX0");
+            _dstX1Uniform = GL.GetUniformLocation(_scalingShaderProgram, "dstX1");
+            _dstY0Uniform = GL.GetUniformLocation(_scalingShaderProgram, "dstY0");
+            _dstY1Uniform = GL.GetUniformLocation(_scalingShaderProgram, "dstY1");
+        }
+
+        public void Run(
+            TextureView view,
+            TextureView destinationTexture,
+            int width,
+            int height,
+            Extents2D source,
+            Extents2D destination)
+        {
+            int previousProgram = GL.GetInteger(GetPName.CurrentProgram);
+            int previousUnit = GL.GetInteger(GetPName.ActiveTexture);
+            GL.ActiveTexture(TextureUnit.Texture0);
+            int previousTextureBinding = GL.GetInteger(GetPName.TextureBinding2D);
+
+            GL.BindImageTexture(0, destinationTexture.Handle, 0, false, 0, TextureAccess.ReadWrite, SizedInternalFormat.Rgba8);
+
+            int threadGroupWorkRegionDim = 16;
+            int dispatchX = (width + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
+            int dispatchY = (height + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
+
+            // Scaling pass
+            GL.UseProgram(_scalingShaderProgram);
+            view.Bind(0);
+            GL.Uniform1(_inputUniform, 0);
+            GL.Uniform1(_outputUniform, 0);
+            GL.Uniform1(_srcX0Uniform, (float)source.X1);
+            GL.Uniform1(_srcX1Uniform, (float)source.X2);
+            GL.Uniform1(_srcY0Uniform, (float)source.Y1);
+            GL.Uniform1(_srcY1Uniform, (float)source.Y2);
+            GL.Uniform1(_dstX0Uniform, (float)destination.X1);
+            GL.Uniform1(_dstX1Uniform, (float)destination.X2);
+            GL.Uniform1(_dstY0Uniform, (float)destination.Y1);
+            GL.Uniform1(_dstY1Uniform, (float)destination.Y2);
+            GL.DispatchCompute(dispatchX, dispatchY, 1);
+
+            GL.UseProgram(previousProgram);
+            GL.MemoryBarrier(MemoryBarrierFlags.ShaderImageAccessBarrierBit);
+
+            (_renderer.Pipeline as Pipeline).RestoreImages1And2();
+
+            GL.ActiveTexture(TextureUnit.Texture0);
+            GL.BindTexture(TextureTarget.Texture2D, previousTextureBinding);
+
+            GL.ActiveTexture((TextureUnit)previousUnit);
+        }
+    }
+}

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

@@ -18,7 +18,7 @@ namespace Ryujinx.Graphics.OpenGL.Effects
         private int _srcY0Uniform;
         private int _scalingShaderProgram;
         private int _sharpeningShaderProgram;
-        private float _scale = 1;
+        private float _sharpeningLevel = 1;
         private int _srcY1Uniform;
         private int _dstX0Uniform;
         private int _dstX1Uniform;
@@ -30,10 +30,10 @@ namespace Ryujinx.Graphics.OpenGL.Effects
 
         public float Level
         {
-            get => _scale;
+            get => _sharpeningLevel;
             set
             {
-                _scale = MathF.Max(0.01f, value);
+                _sharpeningLevel = MathF.Max(0.01f, value);
             }
         }
 

src/Ryujinx.Graphics.OpenGL/Effects/ShaderHelper.cs

@@ -1,4 +1,5 @@
 using OpenTK.Graphics.OpenGL;
+using Ryujinx.Common.Logging;
 
 namespace Ryujinx.Graphics.OpenGL.Effects
 {
@@ -6,18 +7,7 @@ namespace Ryujinx.Graphics.OpenGL.Effects
     {
         public static int CompileProgram(string shaderCode, ShaderType shaderType)
         {
-            var shader = GL.CreateShader(shaderType);
-            GL.ShaderSource(shader, shaderCode);
-            GL.CompileShader(shader);
-
-            var program = GL.CreateProgram();
-            GL.AttachShader(program, shader);
-            GL.LinkProgram(program);
-
-            GL.DetachShader(program, shader);
-            GL.DeleteShader(shader);
-
-            return program;
+            return CompileProgram(new string[] { shaderCode }, shaderType);
         }
 
         public static int CompileProgram(string[] shaders, ShaderType shaderType)
@@ -26,6 +16,15 @@ namespace Ryujinx.Graphics.OpenGL.Effects
             GL.ShaderSource(shader, shaders.Length, shaders, (int[])null);
             GL.CompileShader(shader);
 
+            GL.GetShader(shader, ShaderParameter.CompileStatus, out int isCompiled);
+            if (isCompiled == 0)
+            {
+                string log = GL.GetShaderInfoLog(shader);
+                Logger.Error?.Print(LogClass.Gpu, $"Failed to compile effect shader:\n\n{log}\n");
+                GL.DeleteShader(shader);
+                return 0;
+            }
+
             var program = GL.CreateProgram();
             GL.AttachShader(program, shader);
             GL.LinkProgram(program);

src/Ryujinx.Graphics.OpenGL/Effects/Shaders/area_scaling.glsl

@@ -0,0 +1,119 @@
+#version 430 core
+precision mediump float;
+layout (local_size_x = 16, local_size_y = 16) in;
+layout(rgba8, binding = 0, location=0) uniform image2D imgOutput;
+layout( location=1 ) uniform sampler2D Source;
+layout( location=2 ) uniform float srcX0;
+layout( location=3 ) uniform float srcX1;
+layout( location=4 ) uniform float srcY0;
+layout( location=5 ) uniform float srcY1;
+layout( location=6 ) uniform float dstX0;
+layout( location=7 ) uniform float dstX1;
+layout( location=8 ) uniform float dstY0;
+layout( location=9 ) uniform float dstY1;
+
+/***** Area Sampling *****/
+
+// By Sam Belliveau and Filippo Tarpini. Public Domain license.
+// Effectively a more accurate sharp bilinear filter when upscaling,
+// that also works as a mathematically perfect downscale filter.
+// https://entropymine.com/imageworsener/pixelmixing/
+// https://github.com/obsproject/obs-studio/pull/1715
+// https://legacy.imagemagick.org/Usage/filter/
+vec4 AreaSampling(vec2 xy)
+{
+    // Determine the sizes of the source and target images.
+    vec2 source_size = vec2(abs(srcX1 - srcX0), abs(srcY1 - srcY0));
+    vec2 target_size = vec2(abs(dstX1 - dstX0), abs(dstY1 - dstY0));
+    vec2 inverted_target_size = vec2(1.0) / target_size;
+
+    // Compute the top-left and bottom-right corners of the target pixel box.
+    vec2 t_beg = floor(xy - vec2(dstX0 < dstX1 ? dstX0 : dstX1, dstY0 < dstY1 ? dstY0 : dstY1));
+    vec2 t_end = t_beg + vec2(1.0, 1.0);
+
+    // Convert the target pixel box to source pixel box.
+    vec2 beg = t_beg * inverted_target_size * source_size;
+    vec2 end = t_end * inverted_target_size * source_size;
+
+    // Compute the top-left and bottom-right corners of the pixel box.
+    ivec2 f_beg = ivec2(beg);
+    ivec2 f_end = ivec2(end);
+
+    // Compute how much of the start and end pixels are covered horizontally & vertically.
+    float area_w = 1.0 - fract(beg.x);
+    float area_n = 1.0 - fract(beg.y);
+    float area_e = fract(end.x);
+    float area_s = fract(end.y);
+
+    // Compute the areas of the corner pixels in the pixel box.
+    float area_nw = area_n * area_w;
+    float area_ne = area_n * area_e;
+    float area_sw = area_s * area_w;
+    float area_se = area_s * area_e;
+
+    // Initialize the color accumulator.
+    vec4 avg_color = vec4(0.0, 0.0, 0.0, 0.0);
+
+    // Accumulate corner pixels.
+    avg_color += area_nw * texelFetch(Source, ivec2(f_beg.x, f_beg.y), 0);
+    avg_color += area_ne * texelFetch(Source, ivec2(f_end.x, f_beg.y), 0);
+    avg_color += area_sw * texelFetch(Source, ivec2(f_beg.x, f_end.y), 0);
+    avg_color += area_se * texelFetch(Source, ivec2(f_end.x, f_end.y), 0);
+
+    // Determine the size of the pixel box.
+    int x_range = int(f_end.x - f_beg.x - 0.5);
+    int y_range = int(f_end.y - f_beg.y - 0.5);
+
+    // Accumulate top and bottom edge pixels.
+    for (int x = f_beg.x + 1; x <= f_beg.x + x_range; ++x)
+    {
+        avg_color += area_n * texelFetch(Source, ivec2(x, f_beg.y), 0);
+        avg_color += area_s * texelFetch(Source, ivec2(x, f_end.y), 0);
+    }
+
+    // Accumulate left and right edge pixels and all the pixels in between.
+    for (int y = f_beg.y + 1; y <= f_beg.y + y_range; ++y)
+    {
+        avg_color += area_w * texelFetch(Source, ivec2(f_beg.x, y), 0);
+        avg_color += area_e * texelFetch(Source, ivec2(f_end.x, y), 0);
+
+        for (int x = f_beg.x + 1; x <= f_beg.x + x_range; ++x)
+        {
+            avg_color += texelFetch(Source, ivec2(x, y), 0);
+        }
+    }
+
+    // Compute the area of the pixel box that was sampled.
+    float area_corners = area_nw + area_ne + area_sw + area_se;
+    float area_edges = float(x_range) * (area_n + area_s) + float(y_range) * (area_w + area_e);
+    float area_center = float(x_range) * float(y_range);
+
+    // Return the normalized average color.
+    return avg_color / (area_corners + area_edges + area_center);
+}
+
+float insideBox(vec2 v, vec2 bLeft, vec2 tRight) {
+    vec2 s = step(bLeft, v) - step(tRight, v);
+    return s.x * s.y;
+}
+
+vec2 translateDest(vec2 pos) {
+    vec2 translatedPos = vec2(pos.x, pos.y);
+    translatedPos.x = dstX1 < dstX0 ? dstX1 - translatedPos.x : translatedPos.x;
+    translatedPos.y = dstY0 > dstY1 ? dstY0 + dstY1 - translatedPos.y - 1 : translatedPos.y;
+    return translatedPos;
+}
+
+void main()
+{
+    vec2 bLeft = vec2(dstX0 < dstX1 ? dstX0 : dstX1, dstY0 < dstY1 ? dstY0 : dstY1);
+    vec2 tRight = vec2(dstX1 > dstX0 ? dstX1 : dstX0, dstY1 > dstY0 ? dstY1 : dstY0);
+    ivec2 loc = ivec2(gl_GlobalInvocationID.x, gl_GlobalInvocationID.y);
+    if (insideBox(loc, bLeft, tRight) == 0) {
+        imageStore(imgOutput, loc, vec4(0, 0, 0, 1));
+        return;
+    }
+
+    vec4 outColor = AreaSampling(loc);
+    imageStore(imgOutput, ivec2(translateDest(loc)), vec4(outColor.rgb, 1));
+}

src/Ryujinx.Graphics.OpenGL/Effects/Shaders/fsr_scaling.glsl

@@ -85,4 +85,4 @@ void main() {
     CurrFilter(gxy);
     gxy.x -= 8u;
     CurrFilter(gxy);
-}
+}

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.OpenGL/Ryujinx.Graphics.OpenGL.csproj

@@ -21,6 +21,7 @@
     <EmbeddedResource Include="Effects\Shaders\ffx_fsr1.h" />
     <EmbeddedResource Include="Effects\Shaders\ffx_a.h" />
     <EmbeddedResource Include="Effects\Shaders\fsr_scaling.glsl" />
+    <EmbeddedResource Include="Effects\Shaders\area_scaling.glsl" />
   </ItemGroup>
 
   <ItemGroup>

src/Ryujinx.Graphics.OpenGL/Window.cs

@@ -373,6 +373,16 @@ namespace Ryujinx.Graphics.OpenGL
                         _isLinear = false;
                         _scalingFilter.Level = _scalingFilterLevel;
 
+                        RecreateUpscalingTexture();
+                        break;
+                    case ScalingFilter.Area:
+                        if (_scalingFilter is not AreaScalingFilter)
+                        {
+                            _scalingFilter?.Dispose();
+                            _scalingFilter = new AreaScalingFilter(_renderer);
+                        }
+                        _isLinear = false;
+
                         RecreateUpscalingTexture();
                         break;
                 }

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 +9,55 @@ 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 bool _feedbackLoopActive;
+        private PipelineStageFlags _incoherentBufferWriteStages;
+        private PipelineStageFlags _incoherentTextureWriteStages;
+        private PipelineStageFlags _extraStages;
+        private IncoherentBarrierType _queuedIncoherentBarrier;
+        private bool _queuedFeedbackLoopBarrier;
+
         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;
+            }
+
+            return (access, stages);
+        }
+
         private readonly record struct StageFlags : IEquatable<StageFlags>
         {
             public readonly PipelineStageFlags Source;
@@ -36,47 +70,150 @@ 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;
+                    _queuedFeedbackLoopBarrier = false;
+                }
+                else if (_feedbackLoopActive && _queuedFeedbackLoopBarrier)
+                {
+                    // Feedback loop barrier.
+
+                    MemoryBarrier barrier = new MemoryBarrier()
+                    {
+                        SType = StructureType.MemoryBarrier,
+                        SrcAccessMask = AccessFlags.ShaderWriteBit,
+                        DstAccessMask = AccessFlags.ShaderReadBit
+                    };
+
+                    QueueBarrier(barrier, PipelineStageFlags.FragmentShaderBit, PipelineStageFlags.AllGraphicsBit);
+
+                    _queuedFeedbackLoopBarrier = false;
+                }
+
+                _feedbackLoopActive = false;
+            }
+        }
+
+        public unsafe void Flush(CommandBufferScoped cbs, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
+        {
+            Flush(cbs, null, false, inRenderPass, rpHolder, endRenderPass);
+        }
+
+        public unsafe void Flush(CommandBufferScoped cbs, ShaderCollection program, bool feedbackLoopActive, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
+        {
+            if (program != null)
+            {
+                _incoherentBufferWriteStages |= program.IncoherentBufferWriteStages | _extraStages;
+                _incoherentTextureWriteStages |= program.IncoherentTextureWriteStages;
+            }
+
+            _feedbackLoopActive |= feedbackLoopActive;
+
+            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 +223,20 @@ 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 +299,60 @@ 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 +366,22 @@ 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 +390,14 @@ 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 +411,43 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private void QueueIncoherentBarrier(IncoherentBarrierType type)
+        {
+            if (type > _queuedIncoherentBarrier)
+            {
+                _queuedIncoherentBarrier = type;
+            }
+
+            _queuedFeedbackLoopBarrier = true;
+        }
+
+        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/BufferHolder.cs

@@ -122,7 +122,7 @@ namespace Ryujinx.Graphics.Vulkan
                 Range = (uint)size,
             };
 
-            _gd.Api.CreateBufferView(_device, bufferViewCreateInfo, null, out var bufferView).ThrowOnError();
+            _gd.Api.CreateBufferView(_device, in bufferViewCreateInfo, null, out var bufferView).ThrowOnError();
 
             return new Auto<DisposableBufferView>(new DisposableBufferView(_gd.Api, _device, bufferView), this, _waitable, _buffer);
         }
@@ -153,7 +153,7 @@ namespace Ryujinx.Graphics.Vulkan
                     PipelineStageFlags.AllCommandsBit,
                     DependencyFlags.DeviceGroupBit,
                     1,
-                    memoryBarrier,
+                    in memoryBarrier,
                     0,
                     null,
                     0,
@@ -770,7 +770,7 @@ namespace Ryujinx.Graphics.Vulkan
                 0,
                 null,
                 1,
-                memoryBarrier,
+                in memoryBarrier,
                 0,
                 null);
         }

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();
@@ -214,7 +221,7 @@ namespace Ryujinx.Graphics.Vulkan
                     PBufferBinds = &bufferBind
                 };
 
-                gd.Api.QueueBindSparse(gd.Queue, 1, bindSparseInfo, default).ThrowOnError();
+                gd.Api.QueueBindSparse(gd.Queue, 1, in bindSparseInfo, default).ThrowOnError();
             }
 
             var holder = new BufferHolder(gd, _device, buffer, (int)size, storageAllocations);