vulkan: Fix access level of extensions fields and make them readonly (#4608)

It was only called in one place, that can be simplified.

.github/workflows/release.yml

@@ -112,6 +112,17 @@ jobs:
           repo: ${{ env.RYUJINX_TARGET_RELEASE_CHANNEL_REPO }}
           token: ${{ secrets.RELEASE_TOKEN }}
 
+      - name: Create tag
+        uses: actions/github-script@v5
+        with:
+          script: |
+            github.rest.git.createRef({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              ref: 'refs/tags/${{ steps.version_info.outputs.build_version }}',
+              sha: context.sha
+            })
+
   flatpak_release:
     uses: ./.github/workflows/flatpak.yml
     needs: release

ARMeilleure/ARMeilleure.csproj

@@ -7,6 +7,7 @@
 
   <ItemGroup>
     <ProjectReference Include="..\Ryujinx.Common\Ryujinx.Common.csproj" />
+    <ProjectReference Include="..\Ryujinx.Memory\Ryujinx.Memory.csproj" />
   </ItemGroup>
 
   <ItemGroup>

ARMeilleure/CodeGen/Arm64/CodeGenContext.cs

@@ -1,6 +1,7 @@
 using ARMeilleure.CodeGen.Linking;
 using ARMeilleure.CodeGen.RegisterAllocators;
 using ARMeilleure.IntermediateRepresentation;
+using Ryujinx.Common.Memory;
 using System;
 using System.Collections.Generic;
 using System.IO;
@@ -59,7 +60,7 @@ namespace ARMeilleure.CodeGen.Arm64
 
         public CodeGenContext(AllocationResult allocResult, int maxCallArgs, int blocksCount, bool relocatable)
         {
-            _stream = new MemoryStream();
+            _stream = MemoryStreamManager.Shared.GetStream();
 
             AllocResult = allocResult;
 

ARMeilleure/CodeGen/X86/Assembler.cs

@@ -1,5 +1,6 @@
 using ARMeilleure.CodeGen.Linking;
 using ARMeilleure.IntermediateRepresentation;
+using Ryujinx.Common.Memory;
 using System;
 using System.Collections.Generic;
 using System.Diagnostics;
@@ -1033,7 +1034,13 @@ namespace ARMeilleure.CodeGen.X86
 
             Debug.Assert(opCode != BadOp, "Invalid opcode value.");
 
-            if ((flags & InstructionFlags.Vex) != 0 && HardwareCapabilities.SupportsVexEncoding)
+            if ((flags & InstructionFlags.Evex) != 0 && HardwareCapabilities.SupportsEvexEncoding)
+            {
+                WriteEvexInst(dest, src1, src2, type, flags, opCode);
+
+                opCode &= 0xff;
+            }
+            else if ((flags & InstructionFlags.Vex) != 0 && HardwareCapabilities.SupportsVexEncoding)
             {
                 // In a vex encoding, only one prefix can be active at a time. The active prefix is encoded in the second byte using two bits.
 
@@ -1152,6 +1159,103 @@ namespace ARMeilleure.CodeGen.X86
             }
         }
 
+        private void WriteEvexInst(
+            Operand dest,
+            Operand src1,
+            Operand src2,
+            OperandType type,
+            InstructionFlags flags,
+            int opCode,
+            bool broadcast = false,
+            int registerWidth = 128,
+            int maskRegisterIdx = 0,
+            bool zeroElements = false)
+        {
+            int op1Idx = dest.GetRegister().Index;
+            int op2Idx = src1.GetRegister().Index;
+            int op3Idx = src2.GetRegister().Index;
+
+            WriteByte(0x62);
+
+            // P0
+            // Extend operand 1 register
+            bool r = (op1Idx & 8) == 0;
+            // Extend operand 3 register
+            bool x = (op3Idx & 16) == 0;
+            // Extend operand 3 register
+            bool b = (op3Idx & 8) == 0;
+            // Extend operand 1 register
+            bool rp = (op1Idx & 16) == 0;
+            // Escape code index
+            byte mm = 0b00;
+
+            switch ((ushort)(opCode >> 8))
+            {
+                case 0xf00: mm = 0b01; break;
+                case 0xf38: mm = 0b10; break;
+                case 0xf3a: mm = 0b11; break;
+
+                default: Debug.Fail($"Failed to EVEX encode opcode 0x{opCode:X}."); break;
+            }
+
+            WriteByte(
+                (byte)(
+                    (r ? 0x80 : 0) |
+                    (x ? 0x40 : 0) |
+                    (b ? 0x20 : 0) |
+                    (rp ? 0x10 : 0) |
+                    mm));
+
+            // P1
+            // Specify 64-bit lane mode
+            bool w = Is64Bits(type);
+            // Operand 2 register index
+            byte vvvv = (byte)(~op2Idx & 0b1111);
+            // Opcode prefix
+            byte pp = (flags & InstructionFlags.PrefixMask) switch
+            {
+                InstructionFlags.Prefix66 => 0b01,
+                InstructionFlags.PrefixF3 => 0b10,
+                InstructionFlags.PrefixF2 => 0b11,
+                _ => 0
+            };
+            WriteByte(
+                (byte)(
+                    (w ? 0x80 : 0) |
+                    (vvvv << 3) |
+                    0b100 |
+                    pp));
+
+            // P2
+            // Mask register determines what elements to zero, rather than what elements to merge
+            bool z = zeroElements;
+            // Specifies register-width
+            byte ll = 0b00;
+            switch (registerWidth)
+            {
+                case 128: ll = 0b00; break;
+                case 256: ll = 0b01; break;
+                case 512: ll = 0b10; break;
+
+                default: Debug.Fail($"Invalid EVEX vector register width {registerWidth}."); break;
+            }
+            // Embedded broadcast in the case of a memory operand
+            bool bcast = broadcast;
+            // Extend operand 2 register
+            bool vp = (op2Idx & 16) == 0;
+            // Mask register index
+            Debug.Assert(maskRegisterIdx < 8, $"Invalid mask register index {maskRegisterIdx}.");
+            byte aaa = (byte)(maskRegisterIdx & 0b111);
+
+            WriteByte(
+                (byte)(
+                    (z ? 0x80 : 0) |
+                    (ll << 5) |
+                    (bcast ? 0x10 : 0) |
+                    (vp ? 8 : 0) |
+                    aaa));
+        }
+
         private void WriteCompactInst(Operand operand, int opCode)
         {
             int regIndex = operand.GetRegister().Index;
@@ -1285,7 +1389,7 @@ namespace ARMeilleure.CodeGen.X86
             // Write the code, ignoring the dummy bytes after jumps, into a new stream.
             _stream.Seek(0, SeekOrigin.Begin);
 
-            using var codeStream = new MemoryStream();
+            using var codeStream = MemoryStreamManager.Shared.GetStream();
             var assembler = new Assembler(codeStream, HasRelocs);
 
             bool hasRelocs = HasRelocs;

ARMeilleure/CodeGen/X86/AssemblerTable.cs

@@ -20,6 +20,7 @@ namespace ARMeilleure.CodeGen.X86
             Reg8Dest = 1 << 2,
             RexW     = 1 << 3,
             Vex      = 1 << 4,
+            Evex     = 1 << 5,
 
             PrefixBit  = 16,
             PrefixMask = 7 << PrefixBit,
@@ -278,6 +279,7 @@ namespace ARMeilleure.CodeGen.X86
             Add(X86Instruction.Vfnmsub231sd,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bf, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
             Add(X86Instruction.Vfnmsub231ss,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bf, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Vpblendvb,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a4c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vpternlogd,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a25, InstructionFlags.Evex | InstructionFlags.Prefix66));
             Add(X86Instruction.Xor,           new InstructionInfo(0x00000031, 0x06000083, 0x06000081, BadOp,      0x00000033, InstructionFlags.None));
             Add(X86Instruction.Xorpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f57, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Xorps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f57, InstructionFlags.Vex));

ARMeilleure/CodeGen/X86/CodeGenContext.cs

@@ -1,5 +1,6 @@
 using ARMeilleure.CodeGen.RegisterAllocators;
 using ARMeilleure.IntermediateRepresentation;
+using Ryujinx.Common.Memory;
 using System.IO;
 using System.Numerics;
 
@@ -22,7 +23,7 @@ namespace ARMeilleure.CodeGen.X86
 
         public CodeGenContext(AllocationResult allocResult, int maxCallArgs, int blocksCount, bool relocatable)
         {
-            _stream = new MemoryStream();
+            _stream = MemoryStreamManager.Shared.GetStream();
             _blockLabels = new Operand[blocksCount];
 
             AllocResult = allocResult;

ARMeilleure/CodeGen/X86/HardwareCapabilities.cs

@@ -1,10 +1,14 @@
+using Ryujinx.Memory;
 using System;
+using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics.X86;
 
 namespace ARMeilleure.CodeGen.X86
 {
     static class HardwareCapabilities
     {
+        private delegate uint GetXcr0();
+
         static HardwareCapabilities()
         {
             if (!X86Base.IsSupported)
@@ -24,6 +28,34 @@ namespace ARMeilleure.CodeGen.X86
                 FeatureInfo7Ebx = (FeatureFlags7Ebx)ebx7;
                 FeatureInfo7Ecx = (FeatureFlags7Ecx)ecx7;
             }
+
+            Xcr0InfoEax = (Xcr0FlagsEax)GetXcr0Eax();
+        }
+
+        private static uint GetXcr0Eax()
+        {
+            if (!FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Xsave))
+            {
+                // XSAVE feature required for xgetbv
+                return 0;
+            }
+
+            ReadOnlySpan<byte> asmGetXcr0 = new byte[]
+            {
+                0x31, 0xc9, // xor ecx, ecx
+                0xf, 0x01, 0xd0, // xgetbv
+                0xc3, // ret
+            };
+
+            using MemoryBlock memGetXcr0 = new MemoryBlock((ulong)asmGetXcr0.Length);
+
+            memGetXcr0.Write(0, asmGetXcr0);
+
+            memGetXcr0.Reprotect(0, (ulong)asmGetXcr0.Length, MemoryPermission.ReadAndExecute);
+
+            var fGetXcr0 = Marshal.GetDelegateForFunctionPointer<GetXcr0>(memGetXcr0.Pointer);
+
+            return fGetXcr0();
         }
 
         [Flags]
@@ -44,6 +76,8 @@ namespace ARMeilleure.CodeGen.X86
             Sse42 = 1 << 20,
             Popcnt = 1 << 23,
             Aes = 1 << 25,
+            Xsave = 1 << 26,
+            Osxsave = 1 << 27,
             Avx = 1 << 28,
             F16c = 1 << 29
         }
@@ -52,7 +86,11 @@ namespace ARMeilleure.CodeGen.X86
         public enum FeatureFlags7Ebx
         {
             Avx2 = 1 << 5,
-            Sha = 1 << 29
+            Avx512f = 1 << 16,
+            Avx512dq = 1 << 17,
+            Sha = 1 << 29,
+            Avx512bw = 1 << 30,
+            Avx512vl = 1 << 31
         }
 
         [Flags]
@@ -61,10 +99,21 @@ namespace ARMeilleure.CodeGen.X86
             Gfni = 1 << 8,
         }
 
+        [Flags]
+        public enum Xcr0FlagsEax
+        {
+            Sse = 1 << 1,
+            YmmHi128 = 1 << 2,
+            Opmask = 1 << 5,
+            ZmmHi256 = 1 << 6,
+            Hi16Zmm = 1 << 7
+        }
+
         public static FeatureFlags1Edx FeatureInfo1Edx { get; }
         public static FeatureFlags1Ecx FeatureInfo1Ecx { get; }
         public static FeatureFlags7Ebx FeatureInfo7Ebx { get; } = 0;
         public static FeatureFlags7Ecx FeatureInfo7Ecx { get; } = 0;
+        public static Xcr0FlagsEax Xcr0InfoEax { get; } = 0;
 
         public static bool SupportsSse => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse);
         public static bool SupportsSse2 => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse2);
@@ -76,8 +125,13 @@ namespace ARMeilleure.CodeGen.X86
         public static bool SupportsSse42 => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Sse42);
         public static bool SupportsPopcnt => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Popcnt);
         public static bool SupportsAesni => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Aes);
-        public static bool SupportsAvx => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Avx);
+        public static bool SupportsAvx => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Avx | FeatureFlags1Ecx.Xsave | FeatureFlags1Ecx.Osxsave) && Xcr0InfoEax.HasFlag(Xcr0FlagsEax.Sse | Xcr0FlagsEax.YmmHi128);
         public static bool SupportsAvx2 => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx2) && SupportsAvx;
+        public static bool SupportsAvx512F => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512f) && FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Xsave | FeatureFlags1Ecx.Osxsave)
+            && Xcr0InfoEax.HasFlag(Xcr0FlagsEax.Sse | Xcr0FlagsEax.YmmHi128 | Xcr0FlagsEax.Opmask | Xcr0FlagsEax.ZmmHi256 | Xcr0FlagsEax.Hi16Zmm);
+        public static bool SupportsAvx512Vl => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512vl) && SupportsAvx512F;
+        public static bool SupportsAvx512Bw => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512bw) && SupportsAvx512F;
+        public static bool SupportsAvx512Dq => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512dq) && SupportsAvx512F;
         public static bool SupportsF16c => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.F16c);
         public static bool SupportsSha => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Sha);
         public static bool SupportsGfni => FeatureInfo7Ecx.HasFlag(FeatureFlags7Ecx.Gfni);
@@ -85,5 +139,6 @@ namespace ARMeilleure.CodeGen.X86
         public static bool ForceLegacySse { get; set; }
 
         public static bool SupportsVexEncoding => SupportsAvx && !ForceLegacySse;
+        public static bool SupportsEvexEncoding => SupportsAvx512F && !ForceLegacySse;
     }
 }

ARMeilleure/CodeGen/X86/IntrinsicTable.cs

@@ -180,6 +180,7 @@ namespace ARMeilleure.CodeGen.X86
             Add(Intrinsic.X86Vfnmadd231ss,  new IntrinsicInfo(X86Instruction.Vfnmadd231ss,  IntrinsicType.Fma));
             Add(Intrinsic.X86Vfnmsub231sd,  new IntrinsicInfo(X86Instruction.Vfnmsub231sd,  IntrinsicType.Fma));
             Add(Intrinsic.X86Vfnmsub231ss,  new IntrinsicInfo(X86Instruction.Vfnmsub231ss,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vpternlogd,    new IntrinsicInfo(X86Instruction.Vpternlogd,    IntrinsicType.TernaryImm));
             Add(Intrinsic.X86Xorpd,         new IntrinsicInfo(X86Instruction.Xorpd,         IntrinsicType.Binary));
             Add(Intrinsic.X86Xorps,         new IntrinsicInfo(X86Instruction.Xorps,         IntrinsicType.Binary));
         }

ARMeilleure/CodeGen/X86/X86Instruction.cs

@@ -219,6 +219,7 @@ namespace ARMeilleure.CodeGen.X86
         Vfnmsub231sd,
         Vfnmsub231ss,
         Vpblendvb,
+        Vpternlogd,
         Xor,
         Xorpd,
         Xorps,

ARMeilleure/Instructions/InstEmitSimdLogical.cs

@@ -254,7 +254,22 @@ namespace ARMeilleure.Instructions
 
         public static void Not_V(ArmEmitterContext context)
         {
-            if (Optimizations.UseSse2)
+            if (Optimizations.UseAvx512Ortho)
+            {
+                OpCodeSimd op = (OpCodeSimd)context.CurrOp;
+
+                Operand n = GetVec(op.Rn);
+
+                Operand res = context.AddIntrinsic(Intrinsic.X86Vpternlogd, n, n, Const(~0b10101010));
+
+                if (op.RegisterSize == RegisterSize.Simd64)
+                {
+                    res = context.VectorZeroUpper64(res);
+                }
+
+                context.Copy(GetVec(op.Rd), res);
+            }
+            else if (Optimizations.UseSse2)
             {
                 OpCodeSimd op = (OpCodeSimd)context.CurrOp;
 
@@ -283,6 +298,22 @@ namespace ARMeilleure.Instructions
             {
                 InstEmitSimdHelperArm64.EmitVectorBinaryOp(context, Intrinsic.Arm64OrnV);
             }
+            else if (Optimizations.UseAvx512Ortho)
+            {
+                OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;
+
+                Operand n = GetVec(op.Rn);
+                Operand m = GetVec(op.Rm);
+
+                Operand res = context.AddIntrinsic(Intrinsic.X86Vpternlogd, n, m, Const(0b11001100 | ~0b10101010));
+
+                if (op.RegisterSize == RegisterSize.Simd64)
+                {
+                    res = context.VectorZeroUpper64(res);
+                }
+
+                context.Copy(GetVec(op.Rd), res);
+            }
             else if (Optimizations.UseSse2)
             {
                 OpCodeSimdReg op = (OpCodeSimdReg)context.CurrOp;

ARMeilleure/Instructions/InstEmitSimdLogical32.cs

@@ -151,6 +151,13 @@ namespace ARMeilleure.Instructions
             {
                 InstEmitSimdHelper32Arm64.EmitVectorBinaryOpSimd32(context, (n, m) => context.AddIntrinsic(Intrinsic.Arm64OrnV | Intrinsic.Arm64V128, n, m));
             }
+            else if (Optimizations.UseAvx512Ortho)
+            {
+                EmitVectorBinaryOpSimd32(context, (n, m) =>
+                {
+                    return context.AddIntrinsic(Intrinsic.X86Vpternlogd, n, m, Const(0b11001100 | ~0b10101010));
+                });
+            }
             else if (Optimizations.UseSse2)
             {
                 Operand mask = context.VectorOne();

ARMeilleure/Instructions/InstEmitSimdMove32.cs

@@ -34,7 +34,14 @@ namespace ARMeilleure.Instructions
 
         public static void Vmvn_I(ArmEmitterContext context)
         {
-            if (Optimizations.UseSse2)
+            if (Optimizations.UseAvx512Ortho)
+            {
+                EmitVectorUnaryOpSimd32(context, (op1) =>
+                {
+                    return context.AddIntrinsic(Intrinsic.X86Vpternlogd, op1, op1, Const(0b01010101));
+                });
+            }
+            else if (Optimizations.UseSse2)
             {
                 EmitVectorUnaryOpSimd32(context, (op1) =>
                 {

ARMeilleure/IntermediateRepresentation/Intrinsic.cs

@@ -173,6 +173,7 @@ namespace ARMeilleure.IntermediateRepresentation
         X86Vfnmadd231ss,
         X86Vfnmsub231sd,
         X86Vfnmsub231ss,
+        X86Vpternlogd,
         X86Xorpd,
         X86Xorps,
 

ARMeilleure/Optimizations.cs

@@ -23,6 +23,10 @@ namespace ARMeilleure
         public static bool UseSse42IfAvailable     { get; set; } = true;
         public static bool UsePopCntIfAvailable    { get; set; } = true;
         public static bool UseAvxIfAvailable       { get; set; } = true;
+        public static bool UseAvx512FIfAvailable   { get; set; } = true;
+        public static bool UseAvx512VlIfAvailable  { get; set; } = true;
+        public static bool UseAvx512BwIfAvailable  { get; set; } = true;
+        public static bool UseAvx512DqIfAvailable  { get; set; } = true;
         public static bool UseF16cIfAvailable      { get; set; } = true;
         public static bool UseFmaIfAvailable       { get; set; } = true;
         public static bool UseAesniIfAvailable     { get; set; } = true;
@@ -47,11 +51,18 @@ namespace ARMeilleure
         internal static bool UseSse42     => UseSse42IfAvailable     && X86HardwareCapabilities.SupportsSse42;
         internal static bool UsePopCnt    => UsePopCntIfAvailable    && X86HardwareCapabilities.SupportsPopcnt;
         internal static bool UseAvx       => UseAvxIfAvailable       && X86HardwareCapabilities.SupportsAvx && !ForceLegacySse;
+        internal static bool UseAvx512F   => UseAvx512FIfAvailable   && X86HardwareCapabilities.SupportsAvx512F && !ForceLegacySse;
+        internal static bool UseAvx512Vl  => UseAvx512VlIfAvailable  && X86HardwareCapabilities.SupportsAvx512Vl && !ForceLegacySse;
+        internal static bool UseAvx512Bw  => UseAvx512BwIfAvailable  && X86HardwareCapabilities.SupportsAvx512Bw && !ForceLegacySse;
+        internal static bool UseAvx512Dq  => UseAvx512DqIfAvailable  && X86HardwareCapabilities.SupportsAvx512Dq && !ForceLegacySse;
         internal static bool UseF16c      => UseF16cIfAvailable      && X86HardwareCapabilities.SupportsF16c;
         internal static bool UseFma       => UseFmaIfAvailable       && X86HardwareCapabilities.SupportsFma;
         internal static bool UseAesni     => UseAesniIfAvailable     && X86HardwareCapabilities.SupportsAesni;
         internal static bool UsePclmulqdq => UsePclmulqdqIfAvailable && X86HardwareCapabilities.SupportsPclmulqdq;
         internal static bool UseSha       => UseShaIfAvailable       && X86HardwareCapabilities.SupportsSha;
         internal static bool UseGfni      => UseGfniIfAvailable      && X86HardwareCapabilities.SupportsGfni;
+
+        internal static bool UseAvx512Ortho => UseAvx512F && UseAvx512Vl;
+        internal static bool UseAvx512OrthoFloat => UseAvx512Ortho && UseAvx512Dq;
     }
 }

ARMeilleure/Translation/PTC/Ptc.cs

@@ -6,6 +6,7 @@ using ARMeilleure.Memory;
 using Ryujinx.Common;
 using Ryujinx.Common.Configuration;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using System;
 using System.Buffers.Binary;
 using System.Collections.Generic;
@@ -29,7 +30,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 4484; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 4485; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";
@@ -150,10 +151,10 @@ namespace ARMeilleure.Translation.PTC
 
         private void InitializeCarriers()
         {
-            _infosStream = new MemoryStream();
+            _infosStream = MemoryStreamManager.Shared.GetStream();
             _codesList = new List<byte[]>();
-            _relocsStream = new MemoryStream();
-            _unwindInfosStream = new MemoryStream();
+            _relocsStream = MemoryStreamManager.Shared.GetStream();
+            _unwindInfosStream = MemoryStreamManager.Shared.GetStream();
         }
 
         private void DisposeCarriers()
@@ -968,6 +969,7 @@ namespace ARMeilleure.Translation.PTC
                     (ulong)Arm64HardwareCapabilities.LinuxFeatureInfoHwCap,
                     (ulong)Arm64HardwareCapabilities.LinuxFeatureInfoHwCap2,
                     (ulong)Arm64HardwareCapabilities.MacOsFeatureInfo,
+                    0,
                     0);
             }
             else if (RuntimeInformation.ProcessArchitecture == Architecture.X64)
@@ -976,11 +978,12 @@ namespace ARMeilleure.Translation.PTC
                     (ulong)X86HardwareCapabilities.FeatureInfo1Ecx,
                     (ulong)X86HardwareCapabilities.FeatureInfo1Edx,
                     (ulong)X86HardwareCapabilities.FeatureInfo7Ebx,
-                    (ulong)X86HardwareCapabilities.FeatureInfo7Ecx);
+                    (ulong)X86HardwareCapabilities.FeatureInfo7Ecx,
+                    (ulong)X86HardwareCapabilities.Xcr0InfoEax);
             }
             else
             {
-                return new FeatureInfo(0, 0, 0, 0);
+                return new FeatureInfo(0, 0, 0, 0, 0);
             }
         }
 
@@ -1001,7 +1004,7 @@ namespace ARMeilleure.Translation.PTC
             return osPlatform;
         }
 
-        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 78*/)]
+        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 86*/)]
         private struct OuterHeader
         {
             public ulong Magic;
@@ -1033,8 +1036,8 @@ namespace ARMeilleure.Translation.PTC
             }
         }
 
-        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 32*/)]
-        private record struct FeatureInfo(ulong FeatureInfo0, ulong FeatureInfo1, ulong FeatureInfo2, ulong FeatureInfo3);
+        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 40*/)]
+        private record struct FeatureInfo(ulong FeatureInfo0, ulong FeatureInfo1, ulong FeatureInfo2, ulong FeatureInfo3, ulong FeatureInfo4);
 
         [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 128*/)]
         private struct InnerHeader

ARMeilleure/Translation/PTC/PtcProfiler.cs

@@ -1,6 +1,7 @@
 using ARMeilleure.State;
 using Ryujinx.Common;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using System;
 using System.Buffers.Binary;
 using System.Collections.Concurrent;
@@ -182,7 +183,7 @@ namespace ARMeilleure.Translation.PTC
                     return false;
                 }
 
-                using (MemoryStream stream = new MemoryStream())
+                using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
                 {
                     Debug.Assert(stream.Seek(0L, SeekOrigin.Begin) == 0L && stream.Length == 0L);
 
@@ -274,7 +275,7 @@ namespace ARMeilleure.Translation.PTC
 
             outerHeader.SetHeaderHash();
 
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             {
                 Debug.Assert(stream.Seek(0L, SeekOrigin.Begin) == 0L && stream.Length == 0L);
 

Directory.Packages.props

@@ -22,6 +22,7 @@
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.5.0" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.5.0" />
+    <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="2.3.2" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />
     <PackageVersion Include="NUnit" Version="3.13.3" />
     <PackageVersion Include="NUnit3TestAdapter" Version="4.1.0" />
@@ -33,7 +34,7 @@
     <PackageVersion Include="Ryujinx.Graphics.Nvdec.Dependencies" Version="5.0.1-build13" />
     <PackageVersion Include="Ryujinx.Graphics.Vulkan.Dependencies.MoltenVK" Version="1.2.0" />
     <PackageVersion Include="Ryujinx.GtkSharp" Version="3.24.24.59-ryujinx" />
-    <PackageVersion Include="Ryujinx.SDL2-CS" Version="2.26.1-build23" />
+    <PackageVersion Include="Ryujinx.SDL2-CS" Version="2.26.3-build25" />
     <PackageVersion Include="shaderc.net" Version="0.1.0" />
     <PackageVersion Include="SharpZipLib" Version="1.4.2" />
     <PackageVersion Include="Silk.NET.Vulkan" Version="2.16.0" />
@@ -46,7 +47,7 @@
     <PackageVersion Include="System.IdentityModel.Tokens.Jwt" Version="6.27.0" />
     <PackageVersion Include="System.IO.Hashing" Version="7.0.0" />
     <PackageVersion Include="System.Management" Version="7.0.0" />
-    <PackageVersion Include="UnicornEngine.Unicorn" Version="2.0.2-rc1-f7c841d" />
+    <PackageVersion Include="UnicornEngine.Unicorn" Version="2.0.2-rc1-fb78016" />
     <PackageVersion Include="XamlNameReferenceGenerator" Version="1.5.1" />
   </ItemGroup>
 </Project>

Ryujinx.Common/Extensions/BinaryReaderExtensions.cs

@@ -12,19 +12,5 @@ namespace Ryujinx.Common
         {
             return MemoryMarshal.Cast<byte, T>(reader.ReadBytes(Unsafe.SizeOf<T>()))[0];
         }
-
-        public unsafe static void WriteStruct<T>(this BinaryWriter writer, T value)
-            where T : unmanaged
-        {
-            ReadOnlySpan<byte> data = MemoryMarshal.Cast<T, byte>(MemoryMarshal.CreateReadOnlySpan(ref value, 1));
-
-            writer.Write(data);
-        }
-
-        public static void Write(this BinaryWriter writer, UInt128 value)
-        {
-            writer.Write((ulong)value);
-            writer.Write((ulong)(value >> 64));
-        }
     }
 }

Ryujinx.Common/Extensions/BinaryWriterExtensions.cs

@@ -0,0 +1,28 @@
+using System;
+using System.IO;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Common
+{
+    public static class BinaryWriterExtensions
+    {
+        public unsafe static void WriteStruct<T>(this BinaryWriter writer, T value)
+            where T : unmanaged
+        {
+            ReadOnlySpan<byte> data = MemoryMarshal.Cast<T, byte>(MemoryMarshal.CreateReadOnlySpan(ref value, 1));
+
+            writer.Write(data);
+        }
+
+        public static void Write(this BinaryWriter writer, UInt128 value)
+        {
+            writer.Write((ulong)value);
+            writer.Write((ulong)(value >> 64));
+        }
+
+        public static void Write(this BinaryWriter writer, MemoryStream stream)
+        {
+            stream.CopyTo(writer.BaseStream);
+        }
+    }
+}

Ryujinx.Common/Extensions/StreamExtensions.cs

@@ -0,0 +1,138 @@
+using System;
+using System.Buffers.Binary;
+using System.IO;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Common
+{
+    public static class StreamExtensions
+    {
+        /// <summary>
+        /// Writes a <cref="ReadOnlySpan<int>" /> to this stream.
+        ///
+        /// This default implementation converts each buffer value to a stack-allocated
+        /// byte array, then writes it to the Stream using <cref="System.Stream.Write(byte[])" />.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="buffer">The buffer of values to be written</param>
+        public static void Write(this Stream stream, ReadOnlySpan<int> buffer)
+        {
+            if (buffer.Length == 0)
+            {
+                return;
+            }
+
+            if (BitConverter.IsLittleEndian)
+            {
+                ReadOnlySpan<byte> byteBuffer = MemoryMarshal.Cast<int, byte>(buffer);
+                stream.Write(byteBuffer);
+            }
+            else
+            {
+                Span<byte> byteBuffer = stackalloc byte[sizeof(int)];
+
+                foreach (int value in buffer)
+                {
+                    BinaryPrimitives.WriteInt32LittleEndian(byteBuffer, value);
+                    stream.Write(byteBuffer);
+                }
+            }
+        }
+
+        /// <summary>
+        /// Writes a four-byte signed integer to this stream. The current position
+        /// of the stream is advanced by four.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="value">The value to be written</param>
+        public static void Write(this Stream stream, int value)
+        {
+            Span<byte> buffer = stackalloc byte[sizeof(int)];
+            BinaryPrimitives.WriteInt32LittleEndian(buffer, value);
+            stream.Write(buffer);
+        }
+
+        /// <summary>
+        /// Writes an eight-byte signed integer to this stream. The current position
+        /// of the stream is advanced by eight.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="value">The value to be written</param>
+        public static void Write(this Stream stream, long value)
+        {
+            Span<byte> buffer = stackalloc byte[sizeof(long)];
+            BinaryPrimitives.WriteInt64LittleEndian(buffer, value);
+            stream.Write(buffer);
+        }
+
+        /// <summary>
+        // Writes a four-byte unsigned integer to this stream. The current position
+        // of the stream is advanced by four.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="value">The value to be written</param>
+        public static void Write(this Stream stream, uint value)
+        {
+            Span<byte> buffer = stackalloc byte[sizeof(uint)];
+            BinaryPrimitives.WriteUInt32LittleEndian(buffer, value);
+            stream.Write(buffer);
+        }
+
+        /// <summary>
+        /// Writes an eight-byte unsigned integer to this stream. The current
+        /// position of the stream is advanced by eight.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="value">The value to be written</param>
+        public static void Write(this Stream stream, ulong value)
+        {
+            Span<byte> buffer = stackalloc byte[sizeof(ulong)];
+            BinaryPrimitives.WriteUInt64LittleEndian(buffer, value);
+            stream.Write(buffer);
+        }
+
+        /// <summary>
+        /// Writes the contents of source to stream by calling source.CopyTo(stream).
+        /// Provides consistency with other Stream.Write methods.
+        /// </summary>
+        /// <param name="stream">The stream to be written to</param>
+        /// <param name="source">The stream to be read from</param>
+        public static void Write(this Stream stream, Stream source)
+        {
+            source.CopyTo(stream);
+        }
+
+        /// <summary>
+        /// Writes a sequence of bytes to the Stream.
+        /// </summary>
+        /// <param name="stream">The stream to be written to.</param>
+        /// <param name="value">The byte to be written</param>
+        /// <param name="count">The number of times the value should be written</param>
+        public static void WriteByte(this Stream stream, byte value, int count)
+        {
+            if (count <= 0)
+            {
+                return;
+            }
+
+            const int BlockSize = 16;
+
+            int blockCount = count / BlockSize;
+            if (blockCount > 0)
+            {
+                Span<byte> span = stackalloc byte[BlockSize];
+                span.Fill(value);
+                for (int x = 0; x < blockCount; x++)
+                {
+                    stream.Write(span);
+                }
+            }
+
+            int nonBlockBytes = count % BlockSize;
+            for (int x = 0; x < nonBlockBytes; x++)
+            {
+                stream.WriteByte(value);
+            }
+        }
+    }
+}

Ryujinx.Common/Memory/MemoryStreamManager.cs

@@ -0,0 +1,99 @@
+using Microsoft.IO;
+using System;
+
+namespace Ryujinx.Common.Memory
+{
+    public static class MemoryStreamManager
+    {
+        private static readonly RecyclableMemoryStreamManager _shared = new RecyclableMemoryStreamManager();
+
+        /// <summary>
+        /// We don't expose the <c>RecyclableMemoryStreamManager</c> directly because version 2.x
+        /// returns them as <c>MemoryStream</c>. This Shared class is here to a) offer only the GetStream() versions we use
+        /// and b) return them as <c>RecyclableMemoryStream</c> so we don't have to cast.
+        /// </summary>
+        public static class Shared
+        {
+            /// <summary>
+            /// Retrieve a new <c>MemoryStream</c> object with no tag and a default initial capacity.
+            /// </summary>
+            /// <returns>A <c>RecyclableMemoryStream</c></returns>
+            public static RecyclableMemoryStream GetStream()
+                => new RecyclableMemoryStream(_shared);
+
+            /// <summary>
+            /// Retrieve a new <c>MemoryStream</c> object with the contents copied from the provided
+            /// buffer. The provided buffer is not wrapped or used after construction.
+            /// </summary>
+            /// <remarks>The new stream's position is set to the beginning of the stream when returned.</remarks>
+            /// <param name="buffer">The byte buffer to copy data from</param>
+            /// <returns>A <c>RecyclableMemoryStream</c></returns>
+            public static RecyclableMemoryStream GetStream(byte[] buffer)
+                => GetStream(Guid.NewGuid(), null, buffer, 0, buffer.Length);
+
+            /// <summary>
+            /// Retrieve a new <c>MemoryStream</c> object with the given tag and with contents copied from the provided
+            /// buffer. The provided buffer is not wrapped or used after construction.
+            /// </summary>
+            /// <remarks>The new stream's position is set to the beginning of the stream when returned.</remarks>
+            /// <param name="buffer">The byte buffer to copy data from</param>
+            /// <returns>A <c>RecyclableMemoryStream</c></returns>
+            public static RecyclableMemoryStream GetStream(ReadOnlySpan<byte> buffer)
+                => GetStream(Guid.NewGuid(), null, buffer);
+
+            /// <summary>
+            /// Retrieve a new <c>RecyclableMemoryStream</c> object with the given tag and with contents copied from the provided
+            /// buffer. The provided buffer is not wrapped or used after construction.
+            /// </summary>
+            /// <remarks>The new stream's position is set to the beginning of the stream when returned.</remarks>
+            /// <param name="id">A unique identifier which can be used to trace usages of the stream</param>
+            /// <param name="tag">A tag which can be used to track the source of the stream</param>
+            /// <param name="buffer">The byte buffer to copy data from</param>
+            /// <returns>A <c>RecyclableMemoryStream</c></returns>
+            public static RecyclableMemoryStream GetStream(Guid id, string tag, ReadOnlySpan<byte> buffer)
+            {
+                RecyclableMemoryStream stream = null;
+                try
+                {
+                    stream = new RecyclableMemoryStream(_shared, id, tag, buffer.Length);
+                    stream.Write(buffer);
+                    stream.Position = 0;
+                    return stream;
+                }
+                catch
+                {
+                    stream?.Dispose();
+                    throw;
+                }
+            }
+
+            /// <summary>
+            /// Retrieve a new <c>RecyclableMemoryStream</c> object with the given tag and with contents copied from the provided
+            /// buffer. The provided buffer is not wrapped or used after construction.
+            /// </summary>
+            /// <remarks>The new stream's position is set to the beginning of the stream when returned</remarks>
+            /// <param name="id">A unique identifier which can be used to trace usages of the stream</param>
+            /// <param name="tag">A tag which can be used to track the source of the stream</param>
+            /// <param name="buffer">The byte buffer to copy data from</param>
+            /// <param name="offset">The offset from the start of the buffer to copy from</param>
+            /// <param name="count">The number of bytes to copy from the buffer</param>
+            /// <returns>A <c>RecyclableMemoryStream</c></returns>
+            public static RecyclableMemoryStream GetStream(Guid id, string tag, byte[] buffer, int offset, int count)
+            {
+                RecyclableMemoryStream stream = null;
+                try
+                {
+                    stream = new RecyclableMemoryStream(_shared, id, tag, count);
+                    stream.Write(buffer, offset, count);
+                    stream.Position = 0;
+                    return stream;
+                }
+                catch
+                {
+                    stream?.Dispose();
+                    throw;
+                }
+            }
+        }
+    }
+}

Ryujinx.Common/Ryujinx.Common.csproj

@@ -7,6 +7,7 @@
   </PropertyGroup>
 
   <ItemGroup>
+    <PackageReference Include="Microsoft.IO.RecyclableMemoryStream" />
     <PackageReference Include="MsgPack.Cli" />
     <PackageReference Include="System.Management" />
   </ItemGroup>

Ryujinx.Common/Utilities/EmbeddedResources.cs

@@ -1,3 +1,5 @@
+using Ryujinx.Common.Memory;
+using Ryujinx.Common.Utilities;
 using System;
 using System.IO;
 using System.Linq;
@@ -38,12 +40,7 @@ namespace Ryujinx.Common
                     return null;
                 }
 
-                using (var mem = new MemoryStream())
-                {
-                    stream.CopyTo(mem);
-
-                    return mem.ToArray();
-                }
+                return StreamUtils.StreamToBytes(stream);
             }
         }
 
@@ -56,12 +53,7 @@ namespace Ryujinx.Common
                     return null;
                 }
 
-                using (var mem = new MemoryStream())
-                {
-                    await stream.CopyToAsync(mem);
-
-                    return mem.ToArray();
-                }
+                return await StreamUtils.StreamToBytesAsync(stream);
             }
         }
 

Ryujinx.Common/Utilities/StreamUtils.cs

@@ -1,4 +1,8 @@
-using System.IO;
+using Microsoft.IO;
+using Ryujinx.Common.Memory;
+using System.IO;
+using System.Threading;
+using System.Threading.Tasks;
 
 namespace Ryujinx.Common.Utilities
 {
@@ -6,12 +10,22 @@ namespace Ryujinx.Common.Utilities
     {
         public static byte[] StreamToBytes(Stream input)
         {
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             {
                 input.CopyTo(stream);
 
                 return stream.ToArray();
             }
         }
+
+        public static async Task<byte[]> StreamToBytesAsync(Stream input, CancellationToken cancellationToken = default)
+        {
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
+            {
+                await input.CopyToAsync(stream, cancellationToken);
+
+                return stream.ToArray();
+            }
+        }
     }
 }

Ryujinx.Cpu/MemoryHelper.cs

@@ -1,4 +1,6 @@
-using Ryujinx.Memory;
+using Microsoft.IO;
+using Ryujinx.Common.Memory;
+using Ryujinx.Memory;
 using System;
 using System.IO;
 using System.Runtime.CompilerServices;
@@ -40,7 +42,7 @@ namespace Ryujinx.Cpu
 
         public static string ReadAsciiString(IVirtualMemoryManager memory, ulong position, long maxSize = -1)
         {
-            using (MemoryStream ms = new MemoryStream())
+            using (RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream())
             {
                 for (long offs = 0; offs < maxSize || maxSize == -1; offs++)
                 {
@@ -54,7 +56,7 @@ namespace Ryujinx.Cpu
                     ms.WriteByte(value);
                 }
 
-                return Encoding.ASCII.GetString(ms.ToArray());
+                return Encoding.ASCII.GetString(ms.GetReadOnlySequence());
             }
         }
     }

Ryujinx.Graphics.GAL/IRenderer.cs

@@ -15,7 +15,12 @@ namespace Ryujinx.Graphics.GAL
 
         void BackgroundContextAction(Action action, bool alwaysBackground = false);
 
-        BufferHandle CreateBuffer(int size);
+        BufferHandle CreateBuffer(int size, BufferHandle storageHint);
+
+        BufferHandle CreateBuffer(int size)
+        {
+            return CreateBuffer(size, BufferHandle.Null);
+        }
 
         IProgram CreateProgram(ShaderSource[] shaders, ShaderInfo info);
 
@@ -26,7 +31,7 @@ namespace Ryujinx.Graphics.GAL
 
         void DeleteBuffer(BufferHandle buffer);
 
-        ReadOnlySpan<byte> GetBufferData(BufferHandle buffer, int offset, int size);
+        PinnedSpan<byte> GetBufferData(BufferHandle buffer, int offset, int size);
 
         Capabilities GetCapabilities();
         ulong GetCurrentSync();

Ryujinx.Graphics.GAL/ITexture.cs

@@ -15,8 +15,8 @@ namespace Ryujinx.Graphics.GAL
 
         ITexture CreateView(TextureCreateInfo info, int firstLayer, int firstLevel);
 
-        ReadOnlySpan<byte> GetData();
-        ReadOnlySpan<byte> GetData(int layer, int level);
+        PinnedSpan<byte> GetData();
+        PinnedSpan<byte> GetData(int layer, int level);
 
         void SetData(SpanOrArray<byte> data);
         void SetData(SpanOrArray<byte> data, int layer, int level);

Ryujinx.Graphics.GAL/Multithreading/Commands/Buffer/BufferGetDataCommand.cs

@@ -21,9 +21,9 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Buffer
 
         public static void Run(ref BufferGetDataCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            ReadOnlySpan<byte> result = renderer.GetBufferData(threaded.Buffers.MapBuffer(command._buffer), command._offset, command._size);
+            PinnedSpan<byte> result = renderer.GetBufferData(threaded.Buffers.MapBuffer(command._buffer), command._offset, command._size);
 
-            command._result.Get(threaded).Result = new PinnedSpan<byte>(result);
+            command._result.Get(threaded).Result = result;
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/Commands/Renderer/CreateBufferCommand.cs

@@ -5,16 +5,25 @@
         public CommandType CommandType => CommandType.CreateBuffer;
         private BufferHandle _threadedHandle;
         private int _size;
+        private BufferHandle _storageHint;
 
-        public void Set(BufferHandle threadedHandle, int size)
+        public void Set(BufferHandle threadedHandle, int size, BufferHandle storageHint)
         {
             _threadedHandle = threadedHandle;
             _size = size;
+            _storageHint = storageHint;
         }
 
         public static void Run(ref CreateBufferCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            threaded.Buffers.AssignBuffer(command._threadedHandle, renderer.CreateBuffer(command._size));
+            BufferHandle hint = BufferHandle.Null;
+
+            if (command._storageHint != BufferHandle.Null)
+            {
+                hint = threaded.Buffers.MapBuffer(command._storageHint);
+            }
+
+            threaded.Buffers.AssignBuffer(command._threadedHandle, renderer.CreateBuffer(command._size, hint));
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/Commands/Texture/TextureGetDataCommand.cs

@@ -18,9 +18,9 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Texture
 
         public static void Run(ref TextureGetDataCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            ReadOnlySpan<byte> result = command._texture.Get(threaded).Base.GetData();
+            PinnedSpan<byte> result = command._texture.Get(threaded).Base.GetData();
 
-            command._result.Get(threaded).Result = new PinnedSpan<byte>(result);
+            command._result.Get(threaded).Result = result;
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/Commands/Texture/TextureGetDataSliceCommand.cs

@@ -22,9 +22,9 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Texture
 
         public static void Run(ref TextureGetDataSliceCommand command, ThreadedRenderer threaded, IRenderer renderer)
         {
-            ReadOnlySpan<byte> result = command._texture.Get(threaded).Base.GetData(command._layer, command._level);
+            PinnedSpan<byte> result = command._texture.Get(threaded).Base.GetData(command._layer, command._level);
 
-            command._result.Get(threaded).Result = new PinnedSpan<byte>(result);
+            command._result.Get(threaded).Result = result;
         }
     }
 }

Ryujinx.Graphics.GAL/Multithreading/Model/PinnedSpan.cs

@@ -1,23 +0,0 @@
-using System;
-using System.Runtime.CompilerServices;
-using System.Runtime.InteropServices;
-
-namespace Ryujinx.Graphics.GAL.Multithreading.Model
-{
-    unsafe struct PinnedSpan<T> where T : unmanaged
-    {
-        private void* _ptr;
-        private int _size;
-
-        public PinnedSpan(ReadOnlySpan<T> span)
-        {
-            _ptr = Unsafe.AsPointer(ref MemoryMarshal.GetReference(span));
-            _size = span.Length;
-        }
-
-        public ReadOnlySpan<T> Get()
-        {
-            return new ReadOnlySpan<T>(_ptr, _size * Unsafe.SizeOf<T>());
-        }
-    }
-}

Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedTexture.cs

@@ -72,7 +72,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             return newTex;
         }
 
-        public ReadOnlySpan<byte> GetData()
+        public PinnedSpan<byte> GetData()
         {
             if (_renderer.IsGpuThread())
             {
@@ -80,7 +80,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
                 _renderer.New<TextureGetDataCommand>().Set(Ref(this), Ref(box));
                 _renderer.InvokeCommand();
 
-                return box.Result.Get();
+                return box.Result;
             }
             else
             {
@@ -90,7 +90,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             }
         }
 
-        public ReadOnlySpan<byte> GetData(int layer, int level)
+        public PinnedSpan<byte> GetData(int layer, int level)
         {
             if (_renderer.IsGpuThread())
             {
@@ -98,7 +98,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
                 _renderer.New<TextureGetDataSliceCommand>().Set(Ref(this), Ref(box), layer, level);
                 _renderer.InvokeCommand();
 
-                return box.Result.Get();
+                return box.Result;
             }
             else
             {

Ryujinx.Graphics.GAL/Multithreading/ThreadedRenderer.cs

@@ -265,10 +265,10 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             }
         }
 
-        public BufferHandle CreateBuffer(int size)
+        public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
         {
             BufferHandle handle = Buffers.CreateBufferHandle();
-            New<CreateBufferCommand>().Set(handle, size);
+            New<CreateBufferCommand>().Set(handle, size, storageHint);
             QueueCommand();
 
             return handle;
@@ -329,7 +329,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
             QueueCommand();
         }
 
-        public ReadOnlySpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
+        public PinnedSpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
         {
             if (IsGpuThread())
             {
@@ -337,7 +337,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
                 New<BufferGetDataCommand>().Set(buffer, offset, size, Ref(box));
                 InvokeCommand();
 
-                return box.Result.Get();
+                return box.Result;
             }
             else
             {

Ryujinx.Graphics.GAL/PinnedSpan.cs

@@ -0,0 +1,53 @@
+using System;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Graphics.GAL
+{
+    public unsafe struct PinnedSpan<T> : IDisposable where T : unmanaged
+    {
+        private void* _ptr;
+        private int _size;
+        private Action _disposeAction;
+
+        /// <summary>
+        /// Creates a new PinnedSpan from an existing ReadOnlySpan. The span *must* be pinned in memory.
+        /// The data must be guaranteed to live until disposeAction is called.
+        /// </summary>
+        /// <param name="span">Existing span</param>
+        /// <param name="disposeAction">Action to call on dispose</param>
+        /// <remarks>
+        /// If a dispose action is not provided, it is safe to assume the resource will be available until the next call.
+        /// </remarks>
+        public static PinnedSpan<T> UnsafeFromSpan(ReadOnlySpan<T> span, Action disposeAction = null)
+        {
+            return new PinnedSpan<T>(Unsafe.AsPointer(ref MemoryMarshal.GetReference(span)), span.Length, disposeAction);
+        }
+
+        /// <summary>
+        /// Creates a new PinnedSpan from an existing unsafe region. The data must be guaranteed to live until disposeAction is called.
+        /// </summary>
+        /// <param name="ptr">Pointer to the region</param>
+        /// <param name="size">The total items of T the region contains</param>
+        /// <param name="disposeAction">Action to call on dispose</param>
+        /// <remarks>
+        /// If a dispose action is not provided, it is safe to assume the resource will be available until the next call.
+        /// </remarks>
+        public PinnedSpan(void* ptr, int size, Action disposeAction = null)
+        {
+            _ptr = ptr;
+            _size = size;
+            _disposeAction = disposeAction;
+        }
+
+        public ReadOnlySpan<T> Get()
+        {
+            return new ReadOnlySpan<T>(_ptr, _size * Unsafe.SizeOf<T>());
+        }
+
+        public void Dispose()
+        {
+            _disposeAction?.Invoke();
+        }
+    }
+}

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

@@ -180,7 +180,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             int firstInstance = (int)_state.State.FirstInstance;
 
-            int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount();
+            int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount(_context.Renderer);
 
             if (inlineIndexCount != 0)
             {
@@ -670,7 +670,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 {
                     if (indexedInline)
                     {
-                        int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount();
+                        int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount(_context.Renderer);
                         BufferRange br = new BufferRange(_drawState.IbStreamer.GetInlineIndexBuffer(), 0, inlineIndexCount * 4);
 
                         _channel.BufferManager.SetIndexBuffer(br, IndexType.UInt);

Ryujinx.Graphics.Gpu/Engine/Threed/IbStreamer.cs

@@ -11,9 +11,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
     /// </summary>
     struct IbStreamer
     {
+        private const int BufferCapacity = 256; // Must be a power of 2.
+
         private BufferHandle _inlineIndexBuffer;
         private int _inlineIndexBufferSize;
         private int _inlineIndexCount;
+        private uint[] _buffer;
+        private int _bufferOffset;
 
         /// <summary>
         /// Indicates if any index buffer data has been pushed.
@@ -38,9 +42,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// Gets the number of elements on the current inline index buffer,
         /// while also reseting it to zero for the next draw.
         /// </summary>
+        /// <param name="renderer">Host renderer</param>
         /// <returns>Inline index bufffer count</returns>
-        public int GetAndResetInlineIndexCount()
+        public int GetAndResetInlineIndexCount(IRenderer renderer)
         {
+            UpdateRemaining(renderer);
             int temp = _inlineIndexCount;
             _inlineIndexCount = 0;
             return temp;
@@ -58,16 +64,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             byte i2 = (byte)(argument >> 16);
             byte i3 = (byte)(argument >> 24);
 
-            Span<uint> data = stackalloc uint[4];
+            int offset = _inlineIndexCount;
 
-            data[0] = i0;
-            data[1] = i1;
-            data[2] = i2;
-            data[3] = i3;
-
-            int offset = _inlineIndexCount * 4;
-
-            renderer.SetBufferData(GetInlineIndexBuffer(renderer, offset), offset, MemoryMarshal.Cast<uint, byte>(data));
+            PushData(renderer, offset, i0);
+            PushData(renderer, offset + 1, i1);
+            PushData(renderer, offset + 2, i2);
+            PushData(renderer, offset + 3, i3);
 
             _inlineIndexCount += 4;
         }
@@ -82,14 +84,10 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             ushort i0 = (ushort)argument;
             ushort i1 = (ushort)(argument >> 16);
 
-            Span<uint> data = stackalloc uint[2];
+            int offset = _inlineIndexCount;
 
-            data[0] = i0;
-            data[1] = i1;
-
-            int offset = _inlineIndexCount * 4;
-
-            renderer.SetBufferData(GetInlineIndexBuffer(renderer, offset), offset, MemoryMarshal.Cast<uint, byte>(data));
+            PushData(renderer, offset, i0);
+            PushData(renderer, offset + 1, i1);
 
             _inlineIndexCount += 2;
         }
@@ -103,13 +101,61 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         {
             uint i0 = (uint)argument;
 
-            Span<uint> data = stackalloc uint[1];
+            int offset = _inlineIndexCount++;
 
-            data[0] = i0;
+            PushData(renderer, offset, i0);
+        }
 
-            int offset = _inlineIndexCount++ * 4;
+        /// <summary>
+        /// Pushes a 32-bit value to the index buffer.
+        /// </summary>
+        /// <param name="renderer">Host renderer</param>
+        /// <param name="offset">Offset where the data should be written, in 32-bit words</param>
+        /// <param name="value">Index value to be written</param>
+        private void PushData(IRenderer renderer, int offset, uint value)
+        {
+            if (_buffer == null)
+            {
+                _buffer = new uint[BufferCapacity];
+            }
 
-            renderer.SetBufferData(GetInlineIndexBuffer(renderer, offset), offset, MemoryMarshal.Cast<uint, byte>(data));
+            // We upload data in chunks.
+            // If we are at the start of a chunk, then the buffer might be full,
+            // in that case we need to submit any existing data before overwriting the buffer.
+            int subOffset = offset & (BufferCapacity - 1);
+
+            if (subOffset == 0 && offset != 0)
+            {
+                int baseOffset = (offset - BufferCapacity) * sizeof(uint);
+                BufferHandle buffer = GetInlineIndexBuffer(renderer, baseOffset, BufferCapacity * sizeof(uint));
+                renderer.SetBufferData(buffer, baseOffset, MemoryMarshal.Cast<uint, byte>(_buffer));
+            }
+
+            _buffer[subOffset] = value;
+        }
+
+        /// <summary>
+        /// Makes sure that any pending data is submitted to the GPU before the index buffer is used.
+        /// </summary>
+        /// <param name="renderer">Host renderer</param>
+        private void UpdateRemaining(IRenderer renderer)
+        {
+            int offset = _inlineIndexCount;
+            if (offset == 0)
+            {
+                return;
+            }
+
+            int count = offset & (BufferCapacity - 1);
+            if (count == 0)
+            {
+                count = BufferCapacity;
+            }
+
+            int baseOffset = (offset - count) * sizeof(uint);
+            int length = count * sizeof(uint);
+            BufferHandle buffer = GetInlineIndexBuffer(renderer, baseOffset, length);
+            renderer.SetBufferData(buffer, baseOffset, MemoryMarshal.Cast<uint, byte>(_buffer).Slice(0, length));
         }
 
         /// <summary>
@@ -117,12 +163,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         /// <param name="renderer">Host renderer</param>
         /// <param name="offset">Offset where the data will be written</param>
+        /// <param name="length">Number of bytes that will be written</param>
         /// <returns>Buffer handle</returns>
-        private BufferHandle GetInlineIndexBuffer(IRenderer renderer, int offset)
+        private BufferHandle GetInlineIndexBuffer(IRenderer renderer, int offset, int length)
         {
             // Calculate a reasonable size for the buffer that can fit all the data,
             // and that also won't require frequent resizes if we need to push more data.
-            int size = BitUtils.AlignUp(offset + 0x10, 0x200);
+            int size = BitUtils.AlignUp(offset + length + 0x10, 0x200);
 
             if (_inlineIndexBuffer == BufferHandle.Null)
             {

Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -360,7 +360,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             texture._viewStorage = this;
 
-            Group.UpdateViews(_views);
+            Group.UpdateViews(_views, texture);
 
             if (texture.Group != null && texture.Group != Group)
             {
@@ -384,6 +384,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             _views.Remove(texture);
 
+            Group.RemoveView(texture);
+
             texture._viewStorage = texture;
 
             DecrementReferenceCount();
@@ -1020,13 +1022,12 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// This method should be used to retrieve data that was modified by the host GPU.
         /// This is not cheap, avoid doing that unless strictly needed.
         /// </remarks>
-        /// <param name="output">An output span to place the texture data into. If empty, one is generated</param>
+        /// <param name="output">An output span to place the texture data into</param>
         /// <param name="blacklist">True if the texture should be blacklisted, false otherwise</param>
         /// <param name="texture">The specific host texture to flush. Defaults to this texture</param>
-        /// <returns>The span containing the texture data</returns>
-        private ReadOnlySpan<byte> GetTextureDataFromGpu(Span<byte> output, bool blacklist, ITexture texture = null)
+        private void GetTextureDataFromGpu(Span<byte> output, bool blacklist, ITexture texture = null)
         {
-            ReadOnlySpan<byte> data;
+            PinnedSpan<byte> data;
 
             if (texture != null)
             {
@@ -1052,9 +1053,9 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
             }
 
-            data = ConvertFromHostCompatibleFormat(output, data);
+            ConvertFromHostCompatibleFormat(output, data.Get());
 
-            return data;
+            data.Dispose();
         }
 
         /// <summary>
@@ -1069,10 +1070,9 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="level">The level of the texture to flush</param>
         /// <param name="blacklist">True if the texture should be blacklisted, false otherwise</param>
         /// <param name="texture">The specific host texture to flush. Defaults to this texture</param>
-        /// <returns>The span containing the texture data</returns>
-        public ReadOnlySpan<byte> GetTextureDataSliceFromGpu(Span<byte> output, int layer, int level, bool blacklist, ITexture texture = null)
+        public void GetTextureDataSliceFromGpu(Span<byte> output, int layer, int level, bool blacklist, ITexture texture = null)
         {
-            ReadOnlySpan<byte> data;
+            PinnedSpan<byte> data;
 
             if (texture != null)
             {
@@ -1098,9 +1098,9 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
             }
 
-            data = ConvertFromHostCompatibleFormat(output, data, level, true);
+            ConvertFromHostCompatibleFormat(output, data.Get(), level, true);
 
-            return data;
+            data.Dispose();
         }
 
         /// <summary>
@@ -1473,8 +1473,8 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             MultiRange otherRange = texture.Range;
 
-            IEnumerable<MultiRange> regions = _sizeInfo.AllRegions().Select((region) => Range.GetSlice((ulong)region.Offset, (ulong)region.Size));
-            IEnumerable<MultiRange> otherRegions = texture._sizeInfo.AllRegions().Select((region) => otherRange.GetSlice((ulong)region.Offset, (ulong)region.Size));
+            IEnumerable<MultiRange> regions = _sizeInfo.AllRegions().Select((region) => Range.Slice((ulong)region.Offset, (ulong)region.Size));
+            IEnumerable<MultiRange> otherRegions = texture._sizeInfo.AllRegions().Select((region) => otherRange.Slice((ulong)region.Offset, (ulong)region.Size));
 
             foreach (MultiRange region in regions)
             {

Ryujinx.Graphics.Gpu/Image/TextureGroup.cs

@@ -397,7 +397,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                         int endOffset = Math.Min(spanLast + _sliceSizes[endInfo.BaseLevel + endInfo.Levels - 1], (int)Storage.Size);
                         int size = endOffset - spanBase;
 
-                        dataSpan = _physicalMemory.GetSpan(Storage.Range.GetSlice((ulong)spanBase, (ulong)size));
+                        dataSpan = _physicalMemory.GetSpan(Storage.Range.Slice((ulong)spanBase, (ulong)size));
                     }
 
                     // Only one of these will be greater than 1, as partial sync is only called when there are sub-image views.
@@ -473,7 +473,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             int endOffset = Math.Min(offset + _sliceSizes[level], (int)Storage.Size);
             int size = endOffset - offset;
 
-            using WritableRegion region = _physicalMemory.GetWritableRegion(Storage.Range.GetSlice((ulong)offset, (ulong)size), tracked);
+            using WritableRegion region = _physicalMemory.GetWritableRegion(Storage.Range.Slice((ulong)offset, (ulong)size), tracked);
 
             Storage.GetTextureDataSliceFromGpu(region.Memory.Span, layer, level, tracked, texture);
         }
@@ -989,7 +989,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Update the views in this texture group, rebuilding the memory tracking if required.
         /// </summary>
         /// <param name="views">The views list of the storage texture</param>
-        public void UpdateViews(List<Texture> views)
+        /// <param name="texture">The texture that has been added, if that is the only change, otherwise null</param>
+        public void UpdateViews(List<Texture> views, Texture texture)
         {
             // This is saved to calculate overlapping views for each handle.
             _views = views;
@@ -1027,17 +1028,44 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             if (!regionsRebuilt)
             {
-                // Must update the overlapping views on all handles, but only if they were not just recreated.
-
-                foreach (TextureGroupHandle handle in _handles)
+                if (texture != null)
                 {
-                    handle.RecalculateOverlaps(this, views);
+                    int offset = FindOffset(texture);
+
+                    foreach (TextureGroupHandle handle in _handles)
+                    {
+                        handle.AddOverlap(offset, texture);
+                    }
+                }
+                else
+                {
+                    // Must update the overlapping views on all handles, but only if they were not just recreated.
+
+                    foreach (TextureGroupHandle handle in _handles)
+                    {
+                        handle.RecalculateOverlaps(this, views);
+                    }
                 }
             }
 
             SignalAllDirty();
         }
 
+
+        /// <summary>
+        /// Removes a view from the group, removing it from all overlap lists.
+        /// </summary>
+        /// <param name="view">View to remove from the group</param>
+        public void RemoveView(Texture view)
+        {
+            int offset = FindOffset(view);
+
+            foreach (TextureGroupHandle handle in _handles)
+            {
+                handle.RemoveOverlap(offset, view);
+            }
+        }
+
         /// <summary>
         /// Inherit handle state from an old set of handles, such as modified and dirty flags.
         /// </summary>
@@ -1391,13 +1419,13 @@ namespace Ryujinx.Graphics.Gpu.Image
             for (int i = 0; i < _allOffsets.Length; i++)
             {
                 (int layer, int level) = GetLayerLevelForView(i);
-                MultiRange handleRange = Storage.Range.GetSlice((ulong)_allOffsets[i], 1);
+                MultiRange handleRange = Storage.Range.Slice((ulong)_allOffsets[i], 1);
                 ulong handleBase = handleRange.GetSubRange(0).Address;
 
                 for (int j = 0; j < other._handles.Length; j++)
                 {
                     (int otherLayer, int otherLevel) = other.GetLayerLevelForView(j);
-                    MultiRange otherHandleRange = other.Storage.Range.GetSlice((ulong)other._allOffsets[j], 1);
+                    MultiRange otherHandleRange = other.Storage.Range.Slice((ulong)other._allOffsets[j], 1);
                     ulong otherHandleBase = otherHandleRange.GetSubRange(0).Address;
 
                     if (handleBase == otherHandleBase)
@@ -1474,7 +1502,7 @@ namespace Ryujinx.Graphics.Gpu.Image
                 // Handles list is not modified by another thread, only replaced, so this is thread safe.
                 // Remove modified flags from all overlapping handles, so that the textures don't flush to unmapped/remapped GPU memory.
 
-                MultiRange subRange = Storage.Range.GetSlice((ulong)handle.Offset, (ulong)handle.Size);
+                MultiRange subRange = Storage.Range.Slice((ulong)handle.Offset, (ulong)handle.Size);
 
                 if (range.OverlapsWith(subRange))
                 {

Ryujinx.Graphics.Gpu/Image/TextureGroupHandle.cs

@@ -159,6 +159,42 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
         }
 
+        /// <summary>
+        /// Adds a single texture view as an overlap if its range overlaps.
+        /// </summary>
+        /// <param name="offset">The offset of the view in the group</param>
+        /// <param name="view">The texture to add as an overlap</param>
+        public void AddOverlap(int offset, Texture view)
+        {
+            // Overlaps can be accessed from the memory tracking signal handler, so access must be atomic.
+
+            if (OverlapsWith(offset, (int)view.Size))
+            {
+                lock (Overlaps)
+                {
+                    Overlaps.Add(view);
+                }
+            }
+        }
+
+        /// <summary>
+        /// Removes a single texture view as an overlap if its range overlaps.
+        /// </summary>
+        /// <param name="offset">The offset of the view in the group</param>
+        /// <param name="view">The texture to add as an overlap</param>
+        public void RemoveOverlap(int offset, Texture view)
+        {
+            // Overlaps can be accessed from the memory tracking signal handler, so access must be atomic.
+
+            if (OverlapsWith(offset, (int)view.Size))
+            {
+                lock (Overlaps)
+                {
+                    Overlaps.Remove(view);
+                }
+            }
+        }
+
         /// <summary>
         /// Registers a sync action to happen for this handle, and an interim flush action on the tracking handle.
         /// </summary>

Ryujinx.Graphics.Gpu/Image/TexturePool.cs

@@ -130,6 +130,10 @@ namespace Ryujinx.Graphics.Gpu.Image
                         return ref descriptor;
                     }
                 }
+                else
+                {
+                    texture.SynchronizeMemory();
+                }
 
                 Items[id] = texture;
 
@@ -233,7 +237,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         }
 
         /// <summary>
-        /// Queues a request to update a texture's mapping. 
+        /// Queues a request to update a texture's mapping.
         /// Mapping is updated later to avoid deleting the texture if it is still sparsely mapped.
         /// </summary>
         /// <param name="texture">Texture with potential mapping change</param>

Ryujinx.Graphics.Gpu/Memory/Buffer.cs

@@ -82,7 +82,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
             Address         = address;
             Size            = size;
 
-            Handle = context.Renderer.CreateBuffer((int)size);
+            Handle = context.Renderer.CreateBuffer((int)size, baseBuffers?.MaxBy(x => x.Size).Handle ?? BufferHandle.Null);
 
             _useGranular = size > GranularBufferThreshold;
 
@@ -415,10 +415,10 @@ namespace Ryujinx.Graphics.Gpu.Memory
         {
             int offset = (int)(address - Address);
 
-            ReadOnlySpan<byte> data = _context.Renderer.GetBufferData(Handle, offset, (int)size);
+            using PinnedSpan<byte> data = _context.Renderer.GetBufferData(Handle, offset, (int)size);
 
             // TODO: When write tracking shaders, they will need to be aware of changes in overlapping buffers.
-            _physicalMemory.WriteUntracked(address, data);
+            _physicalMemory.WriteUntracked(address, data.Get());
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Shader/DiskCache/ShaderBinarySerializer.cs

@@ -1,3 +1,5 @@
+using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Shader;
 using Ryujinx.Graphics.Shader.Translation;
@@ -11,16 +13,15 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
     {
         public static byte[] Pack(ShaderSource[] sources)
         {
-            using MemoryStream output = new MemoryStream();
-            using BinaryWriter writer = new BinaryWriter(output);
+            using MemoryStream output = MemoryStreamManager.Shared.GetStream();
 
-            writer.Write(sources.Length);
+            output.Write(sources.Length);
 
-            for (int i = 0; i < sources.Length; i++)
+            foreach (ShaderSource source in sources)
             {
-                writer.Write((int)sources[i].Stage);
-                writer.Write(sources[i].BinaryCode.Length);
-                writer.Write(sources[i].BinaryCode);
+                output.Write((int)source.Stage);
+                output.Write(source.BinaryCode.Length);
+                output.Write(source.BinaryCode);
             }
 
             return output.ToArray();

Ryujinx.Graphics.OpenGL/Buffer.cs

@@ -55,11 +55,14 @@ namespace Ryujinx.Graphics.OpenGL
                 (IntPtr)size);
         }
 
-        public static unsafe ReadOnlySpan<byte> GetData(OpenGLRenderer renderer, BufferHandle buffer, int offset, int size)
+        public static unsafe PinnedSpan<byte> GetData(OpenGLRenderer renderer, BufferHandle buffer, int offset, int size)
         {
+            // Data in the persistent buffer and host array is guaranteed to be available
+            // until the next time the host thread requests data.
+
             if (HwCapabilities.UsePersistentBufferForFlush)
             {
-                return renderer.PersistentBuffers.Default.GetBufferData(buffer, offset, size);
+                return PinnedSpan<byte>.UnsafeFromSpan(renderer.PersistentBuffers.Default.GetBufferData(buffer, offset, size));
             }
             else
             {
@@ -69,7 +72,7 @@ namespace Ryujinx.Graphics.OpenGL
 
                 GL.GetBufferSubData(BufferTarget.CopyReadBuffer, (IntPtr)offset, size, target);
 
-                return new ReadOnlySpan<byte>(target.ToPointer(), size);
+                return new PinnedSpan<byte>(target.ToPointer(), size);
             }
         }
 

Ryujinx.Graphics.OpenGL/Image/TextureBuffer.cs

@@ -39,12 +39,12 @@ namespace Ryujinx.Graphics.OpenGL.Image
             throw new NotSupportedException();
         }
 
-        public ReadOnlySpan<byte> GetData()
+        public PinnedSpan<byte> GetData()
         {
             return Buffer.GetData(_renderer, _buffer, _bufferOffset, _bufferSize);
         }
 
-        public ReadOnlySpan<byte> GetData(int layer, int level)
+        public PinnedSpan<byte> GetData(int layer, int level)
         {
             return GetData();
         }

Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -166,7 +166,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
             _renderer.TextureCopy.Copy(this, (TextureView)destination, srcRegion, dstRegion, linearFilter);
         }
 
-        public unsafe ReadOnlySpan<byte> GetData()
+        public unsafe PinnedSpan<byte> GetData()
         {
             int size = 0;
             int levels = Info.GetLevelsClamped();
@@ -196,16 +196,16 @@ namespace Ryujinx.Graphics.OpenGL.Image
                 data = FormatConverter.ConvertD24S8ToS8D24(data);
             }
 
-            return data;
+            return PinnedSpan<byte>.UnsafeFromSpan(data);
         }
 
-        public unsafe ReadOnlySpan<byte> GetData(int layer, int level)
+        public unsafe PinnedSpan<byte> GetData(int layer, int level)
         {
             int size = Info.GetMipSize(level);
 
             if (HwCapabilities.UsePersistentBufferForFlush)
             {
-                return _renderer.PersistentBuffers.Default.GetTextureData(this, size, layer, level);
+                return PinnedSpan<byte>.UnsafeFromSpan(_renderer.PersistentBuffers.Default.GetTextureData(this, size, layer, level));
             }
             else
             {
@@ -213,7 +213,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
 
                 int offset = WriteTo2D(target, layer, level);
 
-                return new ReadOnlySpan<byte>(target.ToPointer(), size).Slice(offset);
+                return new PinnedSpan<byte>((byte*)target.ToPointer() + offset, size);
             }
         }
 

Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -57,7 +57,7 @@ namespace Ryujinx.Graphics.OpenGL
             ResourcePool = new ResourcePool();
         }
 
-        public BufferHandle CreateBuffer(int size)
+        public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
         {
             BufferCount++;
 
@@ -96,7 +96,7 @@ namespace Ryujinx.Graphics.OpenGL
             return new HardwareInfo(GpuVendor, GpuRenderer);
         }
 
-        public ReadOnlySpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
+        public PinnedSpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
         {
             return Buffer.GetData(this, buffer, offset, size);
         }

Ryujinx.Graphics.OpenGL/Queries/BufferedQuery.cs

@@ -74,7 +74,7 @@ namespace Ryujinx.Graphics.OpenGL.Queries
         {
             result = Marshal.ReadInt64(_bufferMap);
 
-            return WaitingForValue(result);
+            return !WaitingForValue(result);
         }
 
         public long AwaitResult(AutoResetEvent wakeSignal = null)

Ryujinx.Graphics.Vulkan/BufferAllocationType.cs

@@ -0,0 +1,12 @@
+namespace Ryujinx.Graphics.Vulkan
+{
+    internal enum BufferAllocationType
+    {
+        Auto = 0,
+
+        HostMappedNoCache,
+        HostMapped,
+        DeviceLocal,
+        DeviceLocalMapped
+    }
+}

Ryujinx.Graphics.Vulkan/BufferHolder.cs

@@ -1,7 +1,10 @@
-using Ryujinx.Graphics.GAL;
+using Ryujinx.Common.Logging;
+using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
+using System.Collections.Generic;
 using System.Runtime.CompilerServices;
+using System.Threading;
 using VkBuffer = Silk.NET.Vulkan.Buffer;
 using VkFormat = Silk.NET.Vulkan.Format;
 
@@ -11,6 +14,12 @@ namespace Ryujinx.Graphics.Vulkan
     {
         private const int MaxUpdateBufferSize = 0x10000;
 
+        private const int SetCountThreshold = 100;
+        private const int WriteCountThreshold = 50;
+        private const int FlushCountThreshold = 5;
+
+        public const int DeviceLocalSizeThreshold = 256 * 1024; // 256kb
+
         public const AccessFlags DefaultAccessFlags =
             AccessFlags.IndirectCommandReadBit |
             AccessFlags.ShaderReadBit |
@@ -21,10 +30,10 @@ namespace Ryujinx.Graphics.Vulkan
 
         private readonly VulkanRenderer _gd;
         private readonly Device _device;
-        private readonly MemoryAllocation _allocation;
-        private readonly Auto<DisposableBuffer> _buffer;
-        private readonly Auto<MemoryAllocation> _allocationAuto;
-        private readonly ulong _bufferHandle;
+        private MemoryAllocation _allocation;
+        private Auto<DisposableBuffer> _buffer;
+        private Auto<MemoryAllocation> _allocationAuto;
+        private ulong _bufferHandle;
 
         private CacheByRange<BufferHolder> _cachedConvertedBuffers;
 
@@ -32,11 +41,28 @@ namespace Ryujinx.Graphics.Vulkan
 
         private IntPtr _map;
 
-        private readonly MultiFenceHolder _waitable;
+        private MultiFenceHolder _waitable;
 
         private bool _lastAccessIsWrite;
 
-        public BufferHolder(VulkanRenderer gd, Device device, VkBuffer buffer, MemoryAllocation allocation, int size)
+        private BufferAllocationType _baseType;
+        private BufferAllocationType _currentType;
+        private bool _swapQueued;
+
+        public BufferAllocationType DesiredType { get; private set; }
+
+        private int _setCount;
+        private int _writeCount;
+        private int _flushCount;
+        private int _flushTemp;
+
+        private ReaderWriterLock _flushLock;
+        private FenceHolder _flushFence;
+        private int _flushWaiting;
+
+        private List<Action> _swapActions;
+
+        public BufferHolder(VulkanRenderer gd, Device device, VkBuffer buffer, MemoryAllocation allocation, int size, BufferAllocationType type, BufferAllocationType currentType)
         {
             _gd = gd;
             _device = device;
@@ -47,9 +73,153 @@ namespace Ryujinx.Graphics.Vulkan
             _bufferHandle = buffer.Handle;
             Size = size;
             _map = allocation.HostPointer;
+
+            _baseType = type;
+            _currentType = currentType;
+            DesiredType = currentType;
+
+            _flushLock = new ReaderWriterLock();
         }
 
-        public unsafe Auto<DisposableBufferView> CreateView(VkFormat format, int offset, int size)
+        public bool TryBackingSwap(ref CommandBufferScoped? cbs)
+        {
+            if (_swapQueued && DesiredType != _currentType)
+            {
+                // Only swap if the buffer is not used in any queued command buffer.
+                bool isRented = _buffer.HasRentedCommandBufferDependency(_gd.CommandBufferPool);
+
+                if (!isRented && _gd.CommandBufferPool.OwnedByCurrentThread && !_flushLock.IsReaderLockHeld)
+                {
+                    var currentAllocation = _allocationAuto;
+                    var currentBuffer = _buffer;
+                    IntPtr currentMap = _map;
+
+                    (VkBuffer buffer, MemoryAllocation allocation, BufferAllocationType resultType) = _gd.BufferManager.CreateBacking(_gd, Size, DesiredType, false, _currentType);
+
+                    if (buffer.Handle != 0)
+                    {
+                        _flushLock.AcquireWriterLock(Timeout.Infinite);
+
+                        ClearFlushFence();
+
+                        _waitable = new MultiFenceHolder(Size);
+
+                        _allocation = allocation;
+                        _allocationAuto = new Auto<MemoryAllocation>(allocation);
+                        _buffer = new Auto<DisposableBuffer>(new DisposableBuffer(_gd.Api, _device, buffer), _waitable, _allocationAuto);
+                        _bufferHandle = buffer.Handle;
+                        _map = allocation.HostPointer;
+
+                        if (_map != IntPtr.Zero && currentMap != IntPtr.Zero)
+                        {
+                            // Copy data directly. Readbacks don't have to wait if this is done.
+
+                            unsafe
+                            {
+                                new Span<byte>((void*)currentMap, Size).CopyTo(new Span<byte>((void*)_map, Size));
+                            }
+                        }
+                        else
+                        {
+                            if (cbs == null)
+                            {
+                                cbs = _gd.CommandBufferPool.Rent();
+                            }
+
+                            CommandBufferScoped cbsV = cbs.Value;
+
+                            Copy(_gd, cbsV, currentBuffer, _buffer, 0, 0, Size);
+
+                            // Need to wait for the data to reach the new buffer before data can be flushed.
+
+                            _flushFence = _gd.CommandBufferPool.GetFence(cbsV.CommandBufferIndex);
+                            _flushFence.Get();
+                        }
+
+                        Logger.Debug?.PrintMsg(LogClass.Gpu, $"Converted {Size} buffer {_currentType} to {resultType}");
+
+                        _currentType = resultType;
+
+                        if (_swapActions != null)
+                        {
+                            foreach (var action in _swapActions)
+                            {
+                                action();
+                            }
+
+                            _swapActions.Clear();
+                        }
+
+                        currentBuffer.Dispose();
+                        currentAllocation.Dispose();
+
+                        _gd.PipelineInternal.SwapBuffer(currentBuffer, _buffer);
+
+                        _flushLock.ReleaseWriterLock();
+                    }
+
+                    _swapQueued = false;
+
+                    return true;
+                }
+                else
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                _swapQueued = false;
+
+                return true;
+            }
+        }
+
+        private void ConsiderBackingSwap()
+        {
+            if (_baseType == BufferAllocationType.Auto)
+            {
+                if (_writeCount >= WriteCountThreshold || _setCount >= SetCountThreshold || _flushCount >= FlushCountThreshold)
+                {
+                    if (_flushCount > 0 || _flushTemp-- > 0)
+                    {
+                        // Buffers that flush should ideally be mapped in host address space for easy copies.
+                        // If the buffer is large it will do better on GPU memory, as there will be more writes than data flushes (typically individual pages).
+                        // If it is small, then it's likely most of the buffer will be flushed so we want it on host memory, as access is cached.
+                        DesiredType = Size > DeviceLocalSizeThreshold ? BufferAllocationType.DeviceLocalMapped : BufferAllocationType.HostMapped;
+
+                        // It's harder for a buffer that is flushed to revert to another type of mapping.
+                        if (_flushCount > 0)
+                        {
+                            _flushTemp = 1000;
+                        }
+                    }
+                    else if (_writeCount >= WriteCountThreshold)
+                    {
+                        // Buffers that are written often should ideally be in the device local heap. (Storage buffers)
+                        DesiredType = BufferAllocationType.DeviceLocal;
+                    }
+                    else if (_setCount > SetCountThreshold)
+                    {
+                        // Buffers that have their data set often should ideally be host mapped. (Constant buffers)
+                        DesiredType = BufferAllocationType.HostMapped;
+                    }
+
+                    _flushCount = 0;
+                    _writeCount = 0;
+                    _setCount = 0;
+                }
+
+                if (!_swapQueued && DesiredType != _currentType)
+                {
+                    _swapQueued = true;
+
+                    _gd.PipelineInternal.AddBackingSwap(this);
+                }
+            }
+        }
+
+        public unsafe Auto<DisposableBufferView> CreateView(VkFormat format, int offset, int size, Action invalidateView)
         {
             var bufferViewCreateInfo = new BufferViewCreateInfo()
             {
@@ -62,9 +232,19 @@ namespace Ryujinx.Graphics.Vulkan
 
             _gd.Api.CreateBufferView(_device, bufferViewCreateInfo, null, out var bufferView).ThrowOnError();
 
+            (_swapActions ??= new List<Action>()).Add(invalidateView);
+
             return new Auto<DisposableBufferView>(new DisposableBufferView(_gd.Api, _device, bufferView), _waitable, _buffer);
         }
 
+        public void InheritMetrics(BufferHolder other)
+        {
+            _setCount = other._setCount;
+            _writeCount = other._writeCount;
+            _flushCount = other._flushCount;
+            _flushTemp = other._flushTemp;
+        }
+
         public unsafe void InsertBarrier(CommandBuffer commandBuffer, bool isWrite)
         {
             // If the last access is write, we always need a barrier to be sure we will read or modify
@@ -104,12 +284,22 @@ namespace Ryujinx.Graphics.Vulkan
             return _buffer;
         }
 
-        public Auto<DisposableBuffer> GetBuffer(CommandBuffer commandBuffer, bool isWrite = false)
+        public Auto<DisposableBuffer> GetBuffer(CommandBuffer commandBuffer, bool isWrite = false, bool isSSBO = false)
         {
             if (isWrite)
             {
+                _writeCount++;
+
                 SignalWrite(0, Size);
             }
+            else if (isSSBO)
+            {
+                // Always consider SSBO access for swapping to device local memory.
+
+                _writeCount++;
+
+                ConsiderBackingSwap();
+            }
 
             return _buffer;
         }
@@ -118,6 +308,8 @@ namespace Ryujinx.Graphics.Vulkan
         {
             if (isWrite)
             {
+                _writeCount++;
+
                 SignalWrite(offset, size);
             }
 
@@ -126,6 +318,8 @@ namespace Ryujinx.Graphics.Vulkan
 
         public void SignalWrite(int offset, int size)
         {
+            ConsiderBackingSwap();
+
             if (offset == 0 && size == Size)
             {
                 _cachedConvertedBuffers.Clear();
@@ -147,11 +341,76 @@ namespace Ryujinx.Graphics.Vulkan
             return _map;
         }
 
-        public unsafe ReadOnlySpan<byte> GetData(int offset, int size)
+        private void ClearFlushFence()
         {
+            // Asusmes _flushLock is held as writer.
+
+            if (_flushFence != null)
+            {
+                if (_flushWaiting == 0)
+                {
+                    _flushFence.Put();
+                }
+
+                _flushFence = null;
+            }
+        }
+
+        private void WaitForFlushFence()
+        {
+            // Assumes the _flushLock is held as reader, returns in same state.
+
+            if (_flushFence != null)
+            {
+                // If storage has changed, make sure the fence has been reached so that the data is in place.
+
+                var cookie = _flushLock.UpgradeToWriterLock(Timeout.Infinite);
+
+                if (_flushFence != null)
+                {
+                    var fence = _flushFence;
+                    Interlocked.Increment(ref _flushWaiting);
+
+                    // Don't wait in the lock.
+
+                    var restoreCookie = _flushLock.ReleaseLock();
+
+                    fence.Wait();
+
+                    _flushLock.RestoreLock(ref restoreCookie);
+
+                    if (Interlocked.Decrement(ref _flushWaiting) == 0)
+                    {
+                        fence.Put();
+                    }
+
+                    _flushFence = null;
+                }
+
+                _flushLock.DowngradeFromWriterLock(ref cookie);
+            }
+        }
+
+        public unsafe PinnedSpan<byte> GetData(int offset, int size)
+        {
+            _flushLock.AcquireReaderLock(Timeout.Infinite);
+
+            WaitForFlushFence();
+
+            _flushCount++;
+
+            Span<byte> result;
+
             if (_map != IntPtr.Zero)
             {
-                return GetDataStorage(offset, size);
+                result = GetDataStorage(offset, size);
+
+                // Need to be careful here, the buffer can't be unmapped while the data is being used.
+                _buffer.IncrementReferenceCount();
+
+                _flushLock.ReleaseReaderLock();
+
+                return PinnedSpan<byte>.UnsafeFromSpan(result, _buffer.DecrementReferenceCount);
             }
             else
             {
@@ -161,12 +420,17 @@ namespace Ryujinx.Graphics.Vulkan
                 {
                     _gd.FlushAllCommands();
 
-                    return resource.GetFlushBuffer().GetBufferData(_gd.CommandBufferPool, this, offset, size);
+                    result = resource.GetFlushBuffer().GetBufferData(_gd.CommandBufferPool, this, offset, size);
                 }
                 else
                 {
-                    return resource.GetFlushBuffer().GetBufferData(resource.GetPool(), this, offset, size);
+                    result = resource.GetFlushBuffer().GetBufferData(resource.GetPool(), this, offset, size);
                 }
+
+                _flushLock.ReleaseReaderLock();
+
+                // Flush buffer is pinned until the next GetBufferData on the thread, which is fine for current uses.
+                return PinnedSpan<byte>.UnsafeFromSpan(result);
             }
         }
 
@@ -190,6 +454,8 @@ namespace Ryujinx.Graphics.Vulkan
                 return;
             }
 
+            _setCount++;
+
             if (_map != IntPtr.Zero)
             {
                 // If persistently mapped, set the data directly if the buffer is not currently in use.
@@ -268,6 +534,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             var dstBuffer = GetBuffer(cbs.CommandBuffer, dstOffset, data.Length, true).Get(cbs, dstOffset, data.Length).Value;
 
+            _writeCount--;
+
             InsertBufferBarrier(
                 _gd,
                 cbs.CommandBuffer,
@@ -502,11 +770,19 @@ namespace Ryujinx.Graphics.Vulkan
 
         public void Dispose()
         {
+            _swapQueued = false;
+
             _gd.PipelineInternal?.FlushCommandsIfWeightExceeding(_buffer, (ulong)Size);
 
             _buffer.Dispose();
             _allocationAuto.Dispose();
             _cachedConvertedBuffers.Dispose();
+
+            _flushLock.AcquireWriterLock(Timeout.Infinite);
+
+            ClearFlushFence();
+
+            _flushLock.ReleaseWriterLock();
         }
     }
 }

Ryujinx.Graphics.Vulkan/BufferManager.cs

@@ -4,6 +4,7 @@ using System;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using VkFormat = Silk.NET.Vulkan.Format;
+using VkBuffer = Silk.NET.Vulkan.Buffer;
 
 namespace Ryujinx.Graphics.Vulkan
 {
@@ -16,17 +17,17 @@ namespace Ryujinx.Graphics.Vulkan
 
         // Some drivers don't expose a "HostCached" memory type,
         // so we need those alternative flags for the allocation to succeed there.
-        private const MemoryPropertyFlags DefaultBufferMemoryAltFlags =
+        private const MemoryPropertyFlags DefaultBufferMemoryNoCacheFlags =
             MemoryPropertyFlags.HostVisibleBit |
             MemoryPropertyFlags.HostCoherentBit;
 
         private const MemoryPropertyFlags DeviceLocalBufferMemoryFlags =
             MemoryPropertyFlags.DeviceLocalBit;
 
-        private const MemoryPropertyFlags FlushableDeviceLocalBufferMemoryFlags =
+        private const MemoryPropertyFlags DeviceLocalMappedBufferMemoryFlags =
+            MemoryPropertyFlags.DeviceLocalBit |
             MemoryPropertyFlags.HostVisibleBit |
-            MemoryPropertyFlags.HostCoherentBit |
-            MemoryPropertyFlags.DeviceLocalBit;
+            MemoryPropertyFlags.HostCoherentBit;
 
         private const BufferUsageFlags DefaultBufferUsageFlags =
             BufferUsageFlags.TransferSrcBit |
@@ -54,14 +55,14 @@ namespace Ryujinx.Graphics.Vulkan
             StagingBuffer = new StagingBuffer(gd, this);
         }
 
-        public BufferHandle CreateWithHandle(VulkanRenderer gd, int size, bool deviceLocal)
+        public BufferHandle CreateWithHandle(VulkanRenderer gd, int size, BufferAllocationType baseType = BufferAllocationType.HostMapped, BufferHandle storageHint = default)
         {
-            return CreateWithHandle(gd, size, deviceLocal, out _);
+            return CreateWithHandle(gd, size, out _, baseType, storageHint);
         }
 
-        public BufferHandle CreateWithHandle(VulkanRenderer gd, int size, bool deviceLocal, out BufferHolder holder)
+        public BufferHandle CreateWithHandle(VulkanRenderer gd, int size, out BufferHolder holder, BufferAllocationType baseType = BufferAllocationType.HostMapped, BufferHandle storageHint = default)
         {
-            holder = Create(gd, size, deviceLocal: deviceLocal);
+            holder = Create(gd, size, baseType: baseType, storageHint: storageHint);
             if (holder == null)
             {
                 return BufferHandle.Null;
@@ -74,7 +75,12 @@ namespace Ryujinx.Graphics.Vulkan
             return Unsafe.As<ulong, BufferHandle>(ref handle64);
         }
 
-        public unsafe BufferHolder Create(VulkanRenderer gd, int size, bool forConditionalRendering = false, bool deviceLocal = false)
+        public unsafe (VkBuffer buffer, MemoryAllocation allocation, BufferAllocationType resultType) CreateBacking(
+            VulkanRenderer gd,
+            int size,
+            BufferAllocationType type,
+            bool forConditionalRendering = false,
+            BufferAllocationType fallbackType = BufferAllocationType.Auto)
         {
             var usage = DefaultBufferUsageFlags;
 
@@ -98,48 +104,106 @@ namespace Ryujinx.Graphics.Vulkan
             gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();
             gd.Api.GetBufferMemoryRequirements(_device, buffer, out var requirements);
 
-            MemoryPropertyFlags allocateFlags;
-            MemoryPropertyFlags allocateFlagsAlt;
+            MemoryAllocation allocation;
 
-            if (deviceLocal)
+            do
             {
-                allocateFlags = DeviceLocalBufferMemoryFlags;
-                allocateFlagsAlt = DeviceLocalBufferMemoryFlags;
-            }
-            else
-            {
-                allocateFlags = DefaultBufferMemoryFlags;
-                allocateFlagsAlt = DefaultBufferMemoryAltFlags;
-            }
+                var allocateFlags = type switch
+                {
+                    BufferAllocationType.HostMappedNoCache => DefaultBufferMemoryNoCacheFlags,
+                    BufferAllocationType.HostMapped => DefaultBufferMemoryFlags,
+                    BufferAllocationType.DeviceLocal => DeviceLocalBufferMemoryFlags,
+                    BufferAllocationType.DeviceLocalMapped => DeviceLocalMappedBufferMemoryFlags,
+                    _ => DefaultBufferMemoryFlags
+                };
 
-            var allocation = gd.MemoryAllocator.AllocateDeviceMemory(requirements, allocateFlags, allocateFlagsAlt);
+                // If an allocation with this memory type fails, fall back to the previous one.
+                try
+                {
+                    allocation = gd.MemoryAllocator.AllocateDeviceMemory(requirements, allocateFlags, true);
+                }
+                catch (VulkanException)
+                {
+                    allocation = default;
+                }
+            }
+            while (allocation.Memory.Handle == 0 && (--type != fallbackType));
 
             if (allocation.Memory.Handle == 0UL)
             {
                 gd.Api.DestroyBuffer(_device, buffer, null);
-                return null;
+                return default;
             }
 
             gd.Api.BindBufferMemory(_device, buffer, allocation.Memory, allocation.Offset);
 
-            return new BufferHolder(gd, _device, buffer, allocation, size);
+            return (buffer, allocation, type);
         }
 
-        public Auto<DisposableBufferView> CreateView(BufferHandle handle, VkFormat format, int offset, int size)
+        public unsafe BufferHolder Create(
+            VulkanRenderer gd,
+            int size,
+            bool forConditionalRendering = false,
+            BufferAllocationType baseType = BufferAllocationType.HostMapped,
+            BufferHandle storageHint = default)
         {
-            if (TryGetBuffer(handle, out var holder))
+            BufferAllocationType type = baseType;
+            BufferHolder storageHintHolder = null;
+
+            if (baseType == BufferAllocationType.Auto)
             {
-                return holder.CreateView(format, offset, size);
+                if (gd.IsSharedMemory)
+                {
+                    baseType = BufferAllocationType.HostMapped;
+                    type = baseType;
+                }
+                else
+                {
+                    type = size >= BufferHolder.DeviceLocalSizeThreshold ? BufferAllocationType.DeviceLocal : BufferAllocationType.HostMapped;
+                }
+
+                if (storageHint != BufferHandle.Null)
+                {
+                    if (TryGetBuffer(storageHint, out storageHintHolder))
+                    {
+                        type = storageHintHolder.DesiredType;
+                    }
+                }
+            }
+
+            (VkBuffer buffer, MemoryAllocation allocation, BufferAllocationType resultType) =
+                CreateBacking(gd, size, type, forConditionalRendering);
+
+            if (buffer.Handle != 0)
+            {
+                var holder = new BufferHolder(gd, _device, buffer, allocation, size, baseType, resultType);
+
+                if (storageHintHolder != null)
+                {
+                    holder.InheritMetrics(storageHintHolder);
+                }
+
+                return holder;
             }
 
             return null;
         }
 
-        public Auto<DisposableBuffer> GetBuffer(CommandBuffer commandBuffer, BufferHandle handle, bool isWrite)
+        public Auto<DisposableBufferView> CreateView(BufferHandle handle, VkFormat format, int offset, int size, Action invalidateView)
         {
             if (TryGetBuffer(handle, out var holder))
             {
-                return holder.GetBuffer(commandBuffer, isWrite);
+                return holder.CreateView(format, offset, size, invalidateView);
+            }
+
+            return null;
+        }
+
+        public Auto<DisposableBuffer> GetBuffer(CommandBuffer commandBuffer, BufferHandle handle, bool isWrite, bool isSSBO = false)
+        {
+            if (TryGetBuffer(handle, out var holder))
+            {
+                return holder.GetBuffer(commandBuffer, isWrite, isSSBO);
             }
 
             return null;
@@ -332,14 +396,14 @@ namespace Ryujinx.Graphics.Vulkan
             return null;
         }
 
-        public ReadOnlySpan<byte> GetData(BufferHandle handle, int offset, int size)
+        public PinnedSpan<byte> GetData(BufferHandle handle, int offset, int size)
         {
             if (TryGetBuffer(handle, out var holder))
             {
                 return holder.GetData(offset, size);
             }
 
-            return ReadOnlySpan<byte>.Empty;
+            return new PinnedSpan<byte>();
         }
 
         public void SetData<T>(BufferHandle handle, int offset, ReadOnlySpan<T> data) where T : unmanaged

Ryujinx.Graphics.Vulkan/BufferState.cs

@@ -2,14 +2,14 @@
 
 namespace Ryujinx.Graphics.Vulkan
 {
-    readonly struct BufferState : IDisposable
+    struct BufferState : IDisposable
     {
         public static BufferState Null => new BufferState(null, 0, 0);
 
         private readonly int _offset;
         private readonly int _size;
 
-        private readonly Auto<DisposableBuffer> _buffer;
+        private Auto<DisposableBuffer> _buffer;
 
         public BufferState(Auto<DisposableBuffer> buffer, int offset, int size)
         {
@@ -29,6 +29,17 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        public void Swap(Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            if (_buffer == from)
+            {
+                _buffer.DecrementReferenceCount();
+                to.IncrementReferenceCount();
+
+                _buffer = to;
+            }
+        }
+
         public void Dispose()
         {
             _buffer?.DecrementReferenceCount();

Ryujinx.Graphics.Vulkan/DescriptorSetUpdater.cs

@@ -94,7 +94,7 @@ namespace Ryujinx.Graphics.Vulkan
             else
             {
                 // If null descriptors are not supported, we need to pass the handle of a dummy buffer on unused bindings.
-                _dummyBuffer = gd.BufferManager.Create(gd, 0x10000, forConditionalRendering: false, deviceLocal: true);
+                _dummyBuffer = gd.BufferManager.Create(gd, 0x10000, forConditionalRendering: false, baseType: BufferAllocationType.DeviceLocal);
             }
 
             _dummyTexture = gd.CreateTextureView(new TextureCreateInfo(
@@ -178,7 +178,7 @@ namespace Ryujinx.Graphics.Vulkan
                 var buffer = assignment.Range;
                 int index = assignment.Binding;
 
-                Auto<DisposableBuffer> vkBuffer = _gd.BufferManager.GetBuffer(commandBuffer, buffer.Handle, false);
+                Auto<DisposableBuffer> vkBuffer = _gd.BufferManager.GetBuffer(commandBuffer, buffer.Handle, false, isSSBO: true);
                 ref Auto<DisposableBuffer> currentVkBuffer = ref _storageBufferRefs[index];
 
                 DescriptorBufferInfo info = new DescriptorBufferInfo()
@@ -236,7 +236,7 @@ namespace Ryujinx.Graphics.Vulkan
             }
             else if (texture is TextureView view)
             {
-                view.Storage.InsertBarrier(cbs, AccessFlags.ShaderReadBit, stage.ConvertToPipelineStageFlags());
+                view.Storage.InsertWriteToReadBarrier(cbs, AccessFlags.ShaderReadBit, stage.ConvertToPipelineStageFlags());
 
                 _textureRefs[binding] = view.GetImageView();
                 _samplerRefs[binding] = ((SamplerHolder)sampler)?.GetSampler();
@@ -640,6 +640,23 @@ namespace Ryujinx.Graphics.Vulkan
             Array.Clear(_storageSet);
         }
 
+        private void SwapBuffer(Auto<DisposableBuffer>[] list, Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            for (int i = 0; i < list.Length; i++)
+            {
+                if (list[i] == from)
+                {
+                    list[i] = to;
+                }
+            }
+        }
+
+        public void SwapBuffer(Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            SwapBuffer(_uniformBufferRefs, from, to);
+            SwapBuffer(_storageBufferRefs, from, to);
+        }
+
         protected virtual void Dispose(bool disposing)
         {
             if (disposing)

Ryujinx.Graphics.Vulkan/Effects/FsrScalingFilter.cs

@@ -156,11 +156,11 @@ namespace Ryujinx.Graphics.Vulkan.Effects
             };
 
             int rangeSize = dimensionsBuffer.Length * sizeof(float);
-            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize, false);
+            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize);
             _renderer.BufferManager.SetData(bufferHandle, 0, dimensionsBuffer);
 
             ReadOnlySpan<float> sharpeningBuffer = stackalloc float[] { 1.5f - (Level * 0.01f * 1.5f)};
-            var sharpeningBufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, sizeof(float), false);
+            var sharpeningBufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, sizeof(float));
             _renderer.BufferManager.SetData(sharpeningBufferHandle, 0, sharpeningBuffer);
 
             int threadGroupWorkRegionDim = 16;

Ryujinx.Graphics.Vulkan/Effects/FxaaPostProcessingEffect.cs

@@ -87,7 +87,7 @@ namespace Ryujinx.Graphics.Vulkan.Effects
 
             ReadOnlySpan<float> resolutionBuffer = stackalloc float[] { view.Width, view.Height };
             int rangeSize = resolutionBuffer.Length * sizeof(float);
-            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize, false);
+            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize);
 
             _renderer.BufferManager.SetData(bufferHandle, 0, resolutionBuffer);
 

Ryujinx.Graphics.Vulkan/Effects/SmaaPostProcessingEffect.cs

@@ -266,7 +266,7 @@ namespace Ryujinx.Graphics.Vulkan.Effects
 
             ReadOnlySpan<float> resolutionBuffer = stackalloc float[] { view.Width, view.Height };
             int rangeSize = resolutionBuffer.Length * sizeof(float);
-            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize, false);
+            var bufferHandle = _renderer.BufferManager.CreateWithHandle(_renderer, rangeSize);
 
             _renderer.BufferManager.SetData(bufferHandle, 0, resolutionBuffer);
             var bufferRanges = new BufferRange(bufferHandle, 0, rangeSize);

Ryujinx.Graphics.Vulkan/EnumConversion.cs

@@ -322,7 +322,7 @@ namespace Ryujinx.Graphics.Vulkan
                 GAL.Format.S8Uint => ImageAspectFlags.StencilBit,
                 GAL.Format.D24UnormS8Uint or
                 GAL.Format.D32FloatS8Uint or
-                GAL.Format.S8UintD24Unorm    => ImageAspectFlags.DepthBit | ImageAspectFlags.StencilBit,
+                GAL.Format.S8UintD24Unorm => ImageAspectFlags.DepthBit | ImageAspectFlags.StencilBit,
                 _ => ImageAspectFlags.ColorBit
             };
         }

Ryujinx.Graphics.Vulkan/FramebufferParams.cs

@@ -218,5 +218,23 @@ namespace Ryujinx.Graphics.Vulkan
                 AccessFlags.DepthStencilAttachmentWriteBit,
                 PipelineStageFlags.ColorAttachmentOutputBit);
         }
+
+        public void InsertClearBarrier(CommandBufferScoped cbs, int index)
+        {
+            if (_colors != null)
+            {
+                int realIndex = Array.IndexOf(AttachmentIndices, index);
+
+                if (realIndex != -1)
+                {
+                    _colors[realIndex].Storage?.InsertReadToWriteBarrier(cbs, AccessFlags.ColorAttachmentWriteBit, PipelineStageFlags.ColorAttachmentOutputBit);
+                }
+            }
+        }
+
+        public void InsertClearBarrierDS(CommandBufferScoped cbs)
+        {
+            _depthStencil?.Storage?.InsertReadToWriteBarrier(cbs, AccessFlags.DepthStencilAttachmentWriteBit, PipelineStageFlags.EarlyFragmentTestsBit);
+        }
     }
 }

Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -394,7 +394,7 @@ namespace Ryujinx.Graphics.Vulkan
                 (region[2], region[3]) = (region[3], region[2]);
             }
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize);
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, region);
 
@@ -495,7 +495,7 @@ namespace Ryujinx.Graphics.Vulkan
                 (region[2], region[3]) = (region[3], region[2]);
             }
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize);
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, region);
 
@@ -649,7 +649,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             _pipeline.SetCommandBuffer(cbs);
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ClearColorBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ClearColorBufferSize);
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, clearColor);
 
@@ -726,7 +726,7 @@ namespace Ryujinx.Graphics.Vulkan
                 (region[2], region[3]) = (region[3], region[2]);
             }
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, RegionBufferSize);
 
             gd.BufferManager.SetData<float>(bufferHandle, 0, region);
 
@@ -802,7 +802,7 @@ namespace Ryujinx.Graphics.Vulkan
                 shaderParams[2] = size;
                 shaderParams[3] = srcOffset;
 
-                var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false);
+                var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize);
 
                 gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
 
@@ -958,7 +958,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             shaderParams[0] = BitOperations.Log2((uint)ratio);
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize);
 
             gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
 
@@ -1050,7 +1050,7 @@ namespace Ryujinx.Graphics.Vulkan
             (shaderParams[0], shaderParams[1]) = GetSampleCountXYLog2(samples);
             (shaderParams[2], shaderParams[3]) = GetSampleCountXYLog2((int)TextureStorage.ConvertToSampleCountFlags(gd.Capabilities.SupportedSampleCounts, (uint)samples));
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize);
 
             gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
 
@@ -1133,7 +1133,7 @@ namespace Ryujinx.Graphics.Vulkan
             (shaderParams[0], shaderParams[1]) = GetSampleCountXYLog2(samples);
             (shaderParams[2], shaderParams[3]) = GetSampleCountXYLog2((int)TextureStorage.ConvertToSampleCountFlags(gd.Capabilities.SupportedSampleCounts, (uint)samples));
 
-            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false);
+            var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize);
 
             gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
 
@@ -1407,7 +1407,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             pattern.OffsetIndex.CopyTo(shaderParams.Slice(0, pattern.OffsetIndex.Length));
 
-            var patternBufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, false, out var patternBuffer);
+            var patternBufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize, out var patternBuffer);
             var patternBufferAuto = patternBuffer.GetBuffer();
 
             gd.BufferManager.SetData<int>(patternBufferHandle, 0, shaderParams);

Ryujinx.Graphics.Vulkan/IndexBufferPattern.cs

@@ -82,7 +82,7 @@ namespace Ryujinx.Graphics.Vulkan
             }
 
             // Expand the repeating pattern to the number of requested primitives.
-            BufferHandle newBuffer = _gd.CreateBuffer(expectedSize * sizeof(int));
+            BufferHandle newBuffer = _gd.BufferManager.CreateWithHandle(_gd, expectedSize * sizeof(int));
 
             // Copy the old data to the new one.
             if (_repeatingBuffer != BufferHandle.Null)

Ryujinx.Graphics.Vulkan/IndexBufferState.cs

@@ -146,5 +146,16 @@ namespace Ryujinx.Graphics.Vulkan
         {
             return _buffer == buffer;
         }
+
+        public void Swap(Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            if (_buffer == from)
+            {
+                _buffer.DecrementReferenceCount();
+                to.IncrementReferenceCount();
+
+                _buffer = to;
+            }
+        }
     }
 }

Ryujinx.Graphics.Vulkan/MemoryAllocator.cs

@@ -28,32 +28,25 @@ namespace Ryujinx.Graphics.Vulkan
 
         public MemoryAllocation AllocateDeviceMemory(
             MemoryRequirements requirements,
-            MemoryPropertyFlags flags = 0)
+            MemoryPropertyFlags flags = 0,
+            bool isBuffer = false)
         {
-            return AllocateDeviceMemory(requirements, flags, flags);
-        }
-
-        public MemoryAllocation AllocateDeviceMemory(
-            MemoryRequirements requirements,
-            MemoryPropertyFlags flags,
-            MemoryPropertyFlags alternativeFlags)
-        {
-            int memoryTypeIndex = FindSuitableMemoryTypeIndex(requirements.MemoryTypeBits, flags, alternativeFlags);
+            int memoryTypeIndex = FindSuitableMemoryTypeIndex(requirements.MemoryTypeBits, flags);
             if (memoryTypeIndex < 0)
             {
                 return default;
             }
 
             bool map = flags.HasFlag(MemoryPropertyFlags.HostVisibleBit);
-            return Allocate(memoryTypeIndex, requirements.Size, requirements.Alignment, map);
+            return Allocate(memoryTypeIndex, requirements.Size, requirements.Alignment, map, isBuffer);
         }
 
-        private MemoryAllocation Allocate(int memoryTypeIndex, ulong size, ulong alignment, bool map)
+        private MemoryAllocation Allocate(int memoryTypeIndex, ulong size, ulong alignment, bool map, bool isBuffer)
         {
             for (int i = 0; i < _blockLists.Count; i++)
             {
                 var bl = _blockLists[i];
-                if (bl.MemoryTypeIndex == memoryTypeIndex)
+                if (bl.MemoryTypeIndex == memoryTypeIndex && bl.ForBuffer == isBuffer)
                 {
                     lock (bl)
                     {
@@ -62,18 +55,15 @@ namespace Ryujinx.Graphics.Vulkan
                 }
             }
 
-            var newBl = new MemoryAllocatorBlockList(_api, _device, memoryTypeIndex, _blockAlignment);
+            var newBl = new MemoryAllocatorBlockList(_api, _device, memoryTypeIndex, _blockAlignment, isBuffer);
             _blockLists.Add(newBl);
             return newBl.Allocate(size, alignment, map);
         }
 
         private int FindSuitableMemoryTypeIndex(
             uint memoryTypeBits,
-            MemoryPropertyFlags flags,
-            MemoryPropertyFlags alternativeFlags)
+            MemoryPropertyFlags flags)
         {
-            int bestCandidateIndex = -1;
-
             for (int i = 0; i < _physicalDeviceMemoryProperties.MemoryTypeCount; i++)
             {
                 var type = _physicalDeviceMemoryProperties.MemoryTypes[i];
@@ -84,14 +74,27 @@ namespace Ryujinx.Graphics.Vulkan
                     {
                         return i;
                     }
-                    else if (type.PropertyFlags.HasFlag(alternativeFlags))
-                    {
-                        bestCandidateIndex = i;
-                    }
                 }
             }
 
-            return bestCandidateIndex;
+            return -1;
+        }
+
+        public static bool IsDeviceMemoryShared(Vk api, PhysicalDevice physicalDevice)
+        {
+            // The device is regarded as having shared memory if all heaps have the device local bit.
+
+            api.GetPhysicalDeviceMemoryProperties(physicalDevice, out var properties);
+
+            for (int i = 0; i < properties.MemoryHeapCount; i++)
+            {
+                if (!properties.MemoryHeaps[i].Flags.HasFlag(MemoryHeapFlags.DeviceLocalBit))
+                {
+                    return false;
+                }
+            }
+
+            return true;
         }
 
         public void Dispose()

Ryujinx.Graphics.Vulkan/MemoryAllocatorBlockList.cs

@@ -162,15 +162,17 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
 
         public int MemoryTypeIndex { get; }
+        public bool ForBuffer { get; }
 
         private readonly int _blockAlignment;
 
-        public MemoryAllocatorBlockList(Vk api, Device device, int memoryTypeIndex, int blockAlignment)
+        public MemoryAllocatorBlockList(Vk api, Device device, int memoryTypeIndex, int blockAlignment, bool forBuffer)
         {
             _blocks = new List<Block>();
             _api = api;
             _device = device;
             MemoryTypeIndex = memoryTypeIndex;
+            ForBuffer = forBuffer;
             _blockAlignment = blockAlignment;
         }
 

Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -226,6 +226,8 @@ namespace Ryujinx.Graphics.Vulkan
             var attachment = new ClearAttachment(ImageAspectFlags.ColorBit, (uint)index, clearValue);
             var clearRect = FramebufferParams.GetClearRect(ClearScissor, layer, layerCount);
 
+            FramebufferParams.InsertClearBarrier(Cbs, index);
+
             Gd.Api.CmdClearAttachments(CommandBuffer, 1, &attachment, 1, &clearRect);
         }
 
@@ -256,6 +258,8 @@ namespace Ryujinx.Graphics.Vulkan
             var attachment = new ClearAttachment(flags, 0, clearValue);
             var clearRect = FramebufferParams.GetClearRect(ClearScissor, layer, layerCount);
 
+            FramebufferParams.InsertClearBarrierDS(Cbs);
+
             Gd.Api.CmdClearAttachments(CommandBuffer, 1, &attachment, 1, &clearRect);
         }
 
@@ -1297,6 +1301,25 @@ namespace Ryujinx.Graphics.Vulkan
             SignalStateChange();
         }
 
+        public void SwapBuffer(Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            _indexBuffer.Swap(from, to);
+
+            for (int i = 0; i < _vertexBuffers.Length; i++)
+            {
+                _vertexBuffers[i].Swap(from, to);
+            }
+
+            for (int i = 0; i < _transformFeedbackBuffers.Length; i++)
+            {
+                _transformFeedbackBuffers[i].Swap(from, to);
+            }
+
+            _descriptorSetUpdater.SwapBuffer(from, to);
+
+            SignalCommandBufferChange();
+        }
+
         public unsafe void TextureBarrier()
         {
             MemoryBarrier memoryBarrier = new MemoryBarrier()

Ryujinx.Graphics.Vulkan/PipelineFull.cs

@@ -17,10 +17,13 @@ namespace Ryujinx.Graphics.Vulkan
 
         private ulong _byteWeight;
 
+        private List<BufferHolder> _backingSwaps;
+
         public PipelineFull(VulkanRenderer gd, Device device) : base(gd, device)
         {
             _activeQueries = new List<(QueryPool, bool)>();
             _pendingQueryCopies = new();
+            _backingSwaps = new();
 
             CommandBuffer = (Cbs = gd.CommandBufferPool.Rent()).CommandBuffer;
         }
@@ -185,6 +188,20 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private void TryBackingSwaps()
+        {
+            CommandBufferScoped? cbs = null;
+
+            _backingSwaps.RemoveAll((holder) => holder.TryBackingSwap(ref cbs));
+
+            cbs?.Dispose();
+        }
+
+        public void AddBackingSwap(BufferHolder holder)
+        {
+            _backingSwaps.Add(holder);
+        }
+
         public void Restore()
         {
             if (Pipeline != null)
@@ -230,6 +247,8 @@ namespace Ryujinx.Graphics.Vulkan
 
             Gd.ResetCounterPool();
 
+            TryBackingSwaps();
+
             Restore();
         }
 

Ryujinx.Graphics.Vulkan/TextureBuffer.cs

@@ -57,12 +57,12 @@ namespace Ryujinx.Graphics.Vulkan
             throw new NotSupportedException();
         }
 
-        public ReadOnlySpan<byte> GetData()
+        public PinnedSpan<byte> GetData()
         {
             return _gd.GetBufferData(_bufferHandle, _offset, _size);
         }
 
-        public ReadOnlySpan<byte> GetData(int layer, int level)
+        public PinnedSpan<byte> GetData(int layer, int level)
         {
             return GetData();
         }
@@ -128,7 +128,7 @@ namespace Ryujinx.Graphics.Vulkan
         {
             if (_bufferView == null)
             {
-                _bufferView = _gd.BufferManager.CreateView(_bufferHandle, VkFormat, _offset, _size);
+                _bufferView = _gd.BufferManager.CreateView(_bufferHandle, VkFormat, _offset, _size, ReleaseImpl);
             }
 
             return _bufferView?.Get(cbs, _offset, _size).Value ?? default;
@@ -147,7 +147,7 @@ namespace Ryujinx.Graphics.Vulkan
                 return bufferView.Get(cbs, _offset, _size).Value;
             }
 
-            bufferView = _gd.BufferManager.CreateView(_bufferHandle, vkFormat, _offset, _size);
+            bufferView = _gd.BufferManager.CreateView(_bufferHandle, vkFormat, _offset, _size, ReleaseImpl);
 
             if (bufferView != null)
             {

Ryujinx.Graphics.Vulkan/TextureStorage.cs

@@ -46,6 +46,8 @@ namespace Ryujinx.Graphics.Vulkan
 
         private AccessFlags _lastModificationAccess;
         private PipelineStageFlags _lastModificationStage;
+        private AccessFlags _lastReadAccess;
+        private PipelineStageFlags _lastReadStage;
 
         private int _viewsCount;
         private ulong _size;
@@ -440,31 +442,39 @@ namespace Ryujinx.Graphics.Vulkan
             _lastModificationStage = stage;
         }
 
-        public void InsertBarrier(CommandBufferScoped cbs, AccessFlags dstAccessFlags, PipelineStageFlags dstStageFlags)
+        public void InsertReadToWriteBarrier(CommandBufferScoped cbs, AccessFlags dstAccessFlags, PipelineStageFlags dstStageFlags)
         {
+            if (_lastReadAccess != AccessFlags.NoneKhr)
+            {
+                ImageAspectFlags aspectFlags = Info.Format.ConvertAspectFlags();
+
+                TextureView.InsertImageBarrier(
+                    _gd.Api,
+                    cbs.CommandBuffer,
+                    _imageAuto.Get(cbs).Value,
+                    _lastReadAccess,
+                    dstAccessFlags,
+                    _lastReadStage,
+                    dstStageFlags,
+                    aspectFlags,
+                    0,
+                    0,
+                    _info.GetLayers(),
+                    _info.Levels);
+
+                _lastReadAccess = AccessFlags.NoneKhr;
+                _lastReadStage = PipelineStageFlags.None;
+            }
+        }
+
+        public void InsertWriteToReadBarrier(CommandBufferScoped cbs, AccessFlags dstAccessFlags, PipelineStageFlags dstStageFlags)
+        {
+            _lastReadAccess |= dstAccessFlags;
+            _lastReadStage |= dstStageFlags;
+
             if (_lastModificationAccess != AccessFlags.NoneKhr)
             {
-                ImageAspectFlags aspectFlags;
-
-                if (_info.Format.IsDepthOrStencil())
-                {
-                    if (_info.Format == GAL.Format.S8Uint)
-                    {
-                        aspectFlags = ImageAspectFlags.StencilBit;
-                    }
-                    else if (_info.Format == GAL.Format.D16Unorm || _info.Format == GAL.Format.D32Float)
-                    {
-                        aspectFlags = ImageAspectFlags.DepthBit;
-                    }
-                    else
-                    {
-                        aspectFlags = ImageAspectFlags.DepthBit | ImageAspectFlags.StencilBit;
-                    }
-                }
-                else
-                {
-                    aspectFlags = ImageAspectFlags.ColorBit;
-                }
+                ImageAspectFlags aspectFlags = Info.Format.ConvertAspectFlags();
 
                 TextureView.InsertImageBarrier(
                     _gd.Api,

Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -531,7 +531,7 @@ namespace Ryujinx.Graphics.Vulkan
             return bitmap;
         }
 
-        public ReadOnlySpan<byte> GetData()
+        public PinnedSpan<byte> GetData()
         {
             BackgroundResource resources = _gd.BackgroundResources.Get();
 
@@ -539,15 +539,15 @@ namespace Ryujinx.Graphics.Vulkan
             {
                 _gd.FlushAllCommands();
 
-                return GetData(_gd.CommandBufferPool, resources.GetFlushBuffer());
+                return PinnedSpan<byte>.UnsafeFromSpan(GetData(_gd.CommandBufferPool, resources.GetFlushBuffer()));
             }
             else
             {
-                return GetData(resources.GetPool(), resources.GetFlushBuffer());
+                return PinnedSpan<byte>.UnsafeFromSpan(GetData(resources.GetPool(), resources.GetFlushBuffer()));
             }
         }
 
-        public ReadOnlySpan<byte> GetData(int layer, int level)
+        public PinnedSpan<byte> GetData(int layer, int level)
         {
             BackgroundResource resources = _gd.BackgroundResources.Get();
 
@@ -555,11 +555,11 @@ namespace Ryujinx.Graphics.Vulkan
             {
                 _gd.FlushAllCommands();
 
-                return GetData(_gd.CommandBufferPool, resources.GetFlushBuffer(), layer, level);
+                return PinnedSpan<byte>.UnsafeFromSpan(GetData(_gd.CommandBufferPool, resources.GetFlushBuffer(), layer, level));
             }
             else
             {
-                return GetData(resources.GetPool(), resources.GetFlushBuffer(), layer, level);
+                return PinnedSpan<byte>.UnsafeFromSpan(GetData(resources.GetPool(), resources.GetFlushBuffer(), layer, level));
             }
         }
 

Ryujinx.Graphics.Vulkan/VertexBufferState.cs

@@ -129,6 +129,17 @@ namespace Ryujinx.Graphics.Vulkan
             return _buffer == buffer;
         }
 
+        public void Swap(Auto<DisposableBuffer> from, Auto<DisposableBuffer> to)
+        {
+            if (_buffer == from)
+            {
+                _buffer.DecrementReferenceCount();
+                to.IncrementReferenceCount();
+
+                _buffer = to;
+            }
+        }
+
         public void Dispose()
         {
             // Only dispose if this buffer is not refetched on each bind.

Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -20,7 +20,7 @@ namespace Ryujinx.Graphics.Vulkan
         private const string AppName = "Ryujinx.Graphics.Vulkan";
         private const int QueuesCount = 2;
 
-        public static string[] DesirableExtensions { get; } = new string[]
+        private static readonly string[] _desirableExtensions = new string[]
         {
             ExtConditionalRendering.ExtensionName,
             ExtExtendedDynamicState.ExtensionName,
@@ -42,7 +42,7 @@ namespace Ryujinx.Graphics.Vulkan
             "VK_KHR_portability_subset", // By spec, we should enable this if present.
         };
 
-        public static string[] RequiredExtensions { get; } = new string[]
+        private static readonly string[] _requiredExtensions = new string[]
         {
             KhrSwapchain.ExtensionName
         };
@@ -337,14 +337,14 @@ namespace Ryujinx.Graphics.Vulkan
                 {
                     string extensionName = Marshal.PtrToStringAnsi((IntPtr)pExtensionProperties[i].ExtensionName);
 
-                    if (RequiredExtensions.Contains(extensionName))
+                    if (_requiredExtensions.Contains(extensionName))
                     {
                         extensionMatches++;
                     }
                 }
             }
 
-            return extensionMatches == RequiredExtensions.Length && FindSuitableQueueFamily(api, physicalDevice, surface, out _) != InvalidIndex;
+            return extensionMatches == _requiredExtensions.Length && FindSuitableQueueFamily(api, physicalDevice, surface, out _) != InvalidIndex;
         }
 
         internal static uint FindSuitableQueueFamily(Vk api, PhysicalDevice physicalDevice, SurfaceKHR surface, out uint queueCount)
@@ -626,7 +626,7 @@ namespace Ryujinx.Graphics.Vulkan
                 pExtendedFeatures = &featuresCustomBorderColor;
             }
 
-            var enabledExtensions = RequiredExtensions.Union(DesirableExtensions.Intersect(supportedExtensions)).ToArray();
+            var enabledExtensions = _requiredExtensions.Union(_desirableExtensions.Intersect(supportedExtensions)).ToArray();
 
             IntPtr* ppEnabledExtensions = stackalloc IntPtr[enabledExtensions.Length];
 
@@ -672,11 +672,6 @@ namespace Ryujinx.Graphics.Vulkan
             return extensionProperties.Select(x => Marshal.PtrToStringAnsi((IntPtr)x.ExtensionName)).ToArray();
         }
 
-        internal static CommandBufferPool CreateCommandBufferPool(Vk api, Device device, Queue queue, object queueLock, uint queueFamilyIndex)
-        {
-            return new CommandBufferPool(api, device, queue, queueLock, queueFamilyIndex);
-        }
-
         internal unsafe static void CreateDebugMessenger(
             Vk api,
             GraphicsDebugLevel logLevel,

Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -80,6 +80,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal bool IsAmdGcn { get; private set; }
         internal bool IsMoltenVk { get; private set; }
         internal bool IsTBDR { get; private set; }
+        internal bool IsSharedMemory { get; private set; }
         public string GpuVendor { get; private set; }
         public string GpuRenderer { get; private set; }
         public string GpuVersion { get; private set; }
@@ -167,7 +168,9 @@ namespace Ryujinx.Graphics.Vulkan
                 SType = StructureType.PhysicalDeviceSubgroupSizeControlPropertiesExt
             };
 
-            if (Capabilities.SupportsSubgroupSizeControl)
+            bool supportsSubgroupSizeControl = supportedExtensions.Contains("VK_EXT_subgroup_size_control");
+
+            if (supportsSubgroupSizeControl)
             {
                 properties2.PNext = &propertiesSubgroupSizeControl;
             }
@@ -291,7 +294,7 @@ namespace Ryujinx.Graphics.Vulkan
                 supportedExtensions.Contains(KhrDrawIndirectCount.ExtensionName),
                 supportedExtensions.Contains("VK_EXT_fragment_shader_interlock"),
                 supportedExtensions.Contains("VK_NV_geometry_shader_passthrough"),
-                supportedExtensions.Contains("VK_EXT_subgroup_size_control"),
+                supportsSubgroupSizeControl,
                 featuresShaderInt8.ShaderInt8,
                 supportedExtensions.Contains("VK_EXT_shader_stencil_export"),
                 supportedExtensions.Contains(ExtConditionalRendering.ExtensionName),
@@ -313,9 +316,11 @@ namespace Ryujinx.Graphics.Vulkan
                 portabilityFlags,
                 vertexBufferAlignment);
 
+            IsSharedMemory = MemoryAllocator.IsDeviceMemoryShared(Api, _physicalDevice);
+
             MemoryAllocator = new MemoryAllocator(Api, _physicalDevice, _device, properties.Limits.MaxMemoryAllocationCount);
 
-            CommandBufferPool = VulkanInitialization.CreateCommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex);
+            CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex);
 
             DescriptorSetManager = new DescriptorSetManager(_device);
 
@@ -373,9 +378,9 @@ namespace Ryujinx.Graphics.Vulkan
             _initialized = true;
         }
 
-        public BufferHandle CreateBuffer(int size)
+        public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
         {
-            return BufferManager.CreateWithHandle(this, size, false);
+            return BufferManager.CreateWithHandle(this, size, BufferAllocationType.Auto, storageHint);
         }
 
         public IProgram CreateProgram(ShaderSource[] sources, ShaderInfo info)
@@ -439,7 +444,7 @@ namespace Ryujinx.Graphics.Vulkan
             _syncManager.RegisterFlush();
         }
 
-        public ReadOnlySpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
+        public PinnedSpan<byte> GetBufferData(BufferHandle buffer, int offset, int size)
         {
             return BufferManager.GetData(buffer, offset, size);
         }

Ryujinx.HLE/FileSystem/ContentManager.cs

@@ -9,6 +9,7 @@ using LibHac.Tools.FsSystem;
 using LibHac.Tools.FsSystem.NcaUtils;
 using LibHac.Tools.Ncm;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.Common.Utilities;
 using Ryujinx.HLE.Exceptions;
 using Ryujinx.HLE.HOS.Services.Ssl;
@@ -637,12 +638,12 @@ namespace Ryujinx.HLE.FileSystem
 
         private Stream GetZipStream(ZipArchiveEntry entry)
         {
-            MemoryStream dest = new MemoryStream();
+            MemoryStream dest = MemoryStreamManager.Shared.GetStream();
 
-            Stream src = entry.Open();
-
-            src.CopyTo(dest);
-            src.Dispose();
+            using (Stream src = entry.Open())
+            {
+                src.CopyTo(dest);
+            }
 
             return dest;
         }

Ryujinx.HLE/HOS/Applets/Browser/BrowserApplet.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.HLE.HOS.Services.Am.AppletAE;
 using System;
 using System.Collections.Generic;
@@ -70,7 +71,7 @@ namespace Ryujinx.HLE.HOS.Applets.Browser
 
         private byte[] BuildResponseOld(WebCommonReturnValue result)
         {
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.WriteStruct(result);
@@ -80,7 +81,7 @@ namespace Ryujinx.HLE.HOS.Applets.Browser
         }
         private byte[] BuildResponseNew(List<BrowserOutput> outputArguments)
         {
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.WriteStruct(new WebArgHeader

Ryujinx.HLE/HOS/Applets/Controller/ControllerApplet.cs

@@ -1,4 +1,5 @@
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.HLE.HOS.Services.Am.AppletAE;
 using Ryujinx.HLE.HOS.Services.Hid;
 using Ryujinx.HLE.HOS.Services.Hid.Types;
@@ -123,7 +124,7 @@ namespace Ryujinx.HLE.HOS.Applets
 
         private byte[] BuildResponse(ControllerSupportResultInfo result)
         {
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.Write(MemoryMarshal.AsBytes(MemoryMarshal.CreateReadOnlySpan(ref result, Unsafe.SizeOf<ControllerSupportResultInfo>())));
@@ -134,7 +135,7 @@ namespace Ryujinx.HLE.HOS.Applets
 
         private byte[] BuildResponse()
         {
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.Write((ulong)ResultCode.Success);

Ryujinx.HLE/HOS/Applets/PlayerSelect/PlayerSelectApplet.cs

@@ -1,4 +1,5 @@
-using Ryujinx.HLE.HOS.Services.Account.Acc;
+using Ryujinx.Common.Memory;
+using Ryujinx.HLE.HOS.Services.Account.Acc;
 using Ryujinx.HLE.HOS.Services.Am.AppletAE;
 using System;
 using System.IO;
@@ -43,7 +44,7 @@ namespace Ryujinx.HLE.HOS.Applets
         {
             UserProfile currentUser = _system.AccountManager.LastOpenedUser;
 
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.Write((ulong)PlayerSelectResult.Success);

Ryujinx.HLE/HOS/Horizon.cs

@@ -338,7 +338,7 @@ namespace Ryujinx.HLE.HOS
 
                 ProcessCreationInfo creationInfo = new ProcessCreationInfo("Service", 1, 0, 0x8000000, 1, flags, 0, 0);
 
-                int[] defaultCapabilities = new int[]
+                uint[] defaultCapabilities = new uint[]
                 {
                     0x030363F7,
                     0x1FFFFFCF,
@@ -552,4 +552,4 @@ namespace Ryujinx.HLE.HOS
             IsPaused = pause;
         }
     }
-}
+}

Ryujinx.HLE/HOS/Ipc/IpcHandleDesc.cs

@@ -1,3 +1,6 @@
+using Microsoft.IO;
+using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using System;
 using System.IO;
 
@@ -18,20 +21,27 @@ namespace Ryujinx.HLE.HOS.Ipc
 
             HasPId = (word & 1) != 0;
 
-            ToCopy = new int[(word >> 1) & 0xf];
-            ToMove = new int[(word >> 5) & 0xf];
-
             PId = HasPId ? reader.ReadUInt64() : 0;
 
-            for (int index = 0; index < ToCopy.Length; index++)
+            int toCopySize = (word >> 1) & 0xf;
+            int[] toCopy = toCopySize == 0 ? Array.Empty<int>() : new int[toCopySize];
+
+            for (int index = 0; index < toCopy.Length; index++)
             {
-                ToCopy[index] = reader.ReadInt32();
+                toCopy[index] = reader.ReadInt32();
             }
 
-            for (int index = 0; index < ToMove.Length; index++)
+            ToCopy = toCopy;
+
+            int toMoveSize = (word >> 5) & 0xf;
+            int[] toMove = toMoveSize == 0 ? Array.Empty<int>() : new int[toMoveSize];
+
+            for (int index = 0; index < toMove.Length; index++)
             {
-                ToMove[index] = reader.ReadInt32();
+                toMove[index] = reader.ReadInt32();
             }
+
+            ToMove = toMove;
         }
 
         public IpcHandleDesc(int[] copy, int[] move)
@@ -57,36 +67,27 @@ namespace Ryujinx.HLE.HOS.Ipc
             return new IpcHandleDesc(Array.Empty<int>(), handles);
         }
 
-        public byte[] GetBytes()
+        public RecyclableMemoryStream GetStream()
         {
-            using (MemoryStream ms = new MemoryStream())
+            RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream();
+
+            int word = HasPId ? 1 : 0;
+
+            word |= (ToCopy.Length & 0xf) << 1;
+            word |= (ToMove.Length & 0xf) << 5;
+
+            ms.Write(word);
+
+            if (HasPId)
             {
-                BinaryWriter writer = new BinaryWriter(ms);
-
-                int word = HasPId ? 1 : 0;
-
-                word |= (ToCopy.Length & 0xf) << 1;
-                word |= (ToMove.Length & 0xf) << 5;
-
-                writer.Write(word);
-
-                if (HasPId)
-                {
-                    writer.Write(PId);
-                }
-
-                foreach (int handle in ToCopy)
-                {
-                    writer.Write(handle);
-                }
-
-                foreach (int handle in ToMove)
-                {
-                    writer.Write(handle);
-                }
-
-                return ms.ToArray();
+                ms.Write(PId);
             }
+
+            ms.Write(ToCopy);
+            ms.Write(ToMove);
+
+            ms.Position = 0;
+            return ms;
         }
     }
 }

Ryujinx.HLE/HOS/Ipc/IpcMessage.cs

@@ -1,4 +1,8 @@
+using Microsoft.IO;
+using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using System;
+using System.Buffers;
 using System.Collections.Generic;
 using System.Diagnostics;
 using System.IO;
@@ -32,9 +36,9 @@ namespace Ryujinx.HLE.HOS.Ipc
             ObjectIds = new List<int>();
         }
 
-        public IpcMessage(byte[] data, long cmdPtr) : this()
+        public IpcMessage(ReadOnlySpan<byte> data, long cmdPtr) : this()
         {
-            using (MemoryStream ms = new MemoryStream(data))
+            using (RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream(data))
             {
                 BinaryReader reader = new BinaryReader(ms);
 
@@ -114,124 +118,119 @@ namespace Ryujinx.HLE.HOS.Ipc
 
             for (int index = 0; index < recvListCount; index++)
             {
-                RecvListBuff.Add(new IpcRecvListBuffDesc(reader));
+                RecvListBuff.Add(new IpcRecvListBuffDesc(reader.ReadUInt64()));
             }
         }
 
-        public byte[] GetBytes(long cmdPtr, ulong recvListAddr)
+        public RecyclableMemoryStream GetStream(long cmdPtr, ulong recvListAddr)
         {
-            using (MemoryStream ms = new MemoryStream())
+            RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream();
+
+            int word0;
+            int word1;
+
+            word0  = (int)Type;
+            word0 |= (PtrBuff.Count      & 0xf) << 16;
+            word0 |= (SendBuff.Count     & 0xf) << 20;
+            word0 |= (ReceiveBuff.Count  & 0xf) << 24;
+            word0 |= (ExchangeBuff.Count & 0xf) << 28;
+
+            using RecyclableMemoryStream handleDataStream = HandleDesc?.GetStream();
+
+            int dataLength = RawData?.Length ?? 0;
+
+            dataLength = (dataLength + 3) & ~3;
+
+            int rawLength = dataLength;
+
+            int pad0 = (int)GetPadSize16(cmdPtr + 8 + (handleDataStream?.Length ?? 0) + PtrBuff.Count * 8);
+
+            // Apparently, padding after Raw Data is 16 bytes, however when there is
+            // padding before Raw Data too, we need to subtract the size of this padding.
+            // This is the weirdest padding I've seen so far...
+            int pad1 = 0x10 - pad0;
+
+            dataLength = (dataLength + pad0 + pad1) / 4;
+
+            word1 = (dataLength & 0x3ff) | (2 << 10);
+
+            if (HandleDesc != null)
             {
-                BinaryWriter writer = new BinaryWriter(ms);
-
-                int word0;
-                int word1;
-
-                word0  = (int)Type;
-                word0 |= (PtrBuff.Count      & 0xf) << 16;
-                word0 |= (SendBuff.Count     & 0xf) << 20;
-                word0 |= (ReceiveBuff.Count  & 0xf) << 24;
-                word0 |= (ExchangeBuff.Count & 0xf) << 28;
-
-                byte[] handleData = Array.Empty<byte>();
-
-                if (HandleDesc != null)
-                {
-                    handleData = HandleDesc.GetBytes();
-                }
-
-                int dataLength = RawData?.Length ?? 0;
-
-                dataLength = (dataLength + 3) & ~3;
-
-                int rawLength = dataLength;
-
-                int pad0 = (int)GetPadSize16(cmdPtr + 8 + handleData.Length + PtrBuff.Count * 8);
-
-                // Apparently, padding after Raw Data is 16 bytes, however when there is
-                // padding before Raw Data too, we need to subtract the size of this padding.
-                // This is the weirdest padding I've seen so far...
-                int pad1 = 0x10 - pad0;
-
-                dataLength = (dataLength + pad0 + pad1) / 4;
-
-                word1 = (dataLength & 0x3ff) | (2 << 10);
-
-                if (HandleDesc != null)
-                {
-                    word1 |= 1 << 31;
-                }
-
-                writer.Write(word0);
-                writer.Write(word1);
-                writer.Write(handleData);
-
-                for (int index = 0; index < PtrBuff.Count; index++)
-                {
-                    writer.Write(PtrBuff[index].GetWord0());
-                    writer.Write(PtrBuff[index].GetWord1());
-                }
-
-                ms.Seek(pad0, SeekOrigin.Current);
-
-                if (RawData != null)
-                {
-                    writer.Write(RawData);
-                    ms.Seek(rawLength - RawData.Length, SeekOrigin.Current);
-                }
-
-                writer.Write(new byte[pad1]);
-                writer.Write(recvListAddr);
-
-                return ms.ToArray();
+                word1 |= 1 << 31;
             }
+
+            ms.Write(word0);
+            ms.Write(word1);
+
+            if (handleDataStream != null)
+            {
+                ms.Write(handleDataStream);
+            }
+
+            foreach (IpcPtrBuffDesc ptrBuffDesc in PtrBuff)
+            {
+                ms.Write(ptrBuffDesc.GetWord0());
+                ms.Write(ptrBuffDesc.GetWord1());
+            }
+
+            ms.WriteByte(0, pad0);
+
+            if (RawData != null)
+            {
+                ms.Write(RawData);
+                ms.WriteByte(0, rawLength - RawData.Length);
+            }
+
+            ms.WriteByte(0, pad1);
+
+            ms.Write(recvListAddr);
+
+            ms.Position = 0;
+
+            return ms;
         }
 
-        public byte[] GetBytesTipc()
+        public RecyclableMemoryStream GetStreamTipc()
         {
             Debug.Assert(PtrBuff.Count == 0);
 
-            using (MemoryStream ms = new MemoryStream())
+            RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream();
+
+            int word0;
+            int word1;
+
+            word0 = (int)Type;
+            word0 |= (SendBuff.Count & 0xf) << 20;
+            word0 |= (ReceiveBuff.Count & 0xf) << 24;
+            word0 |= (ExchangeBuff.Count & 0xf) << 28;
+
+            using RecyclableMemoryStream handleDataStream = HandleDesc?.GetStream();
+
+            int dataLength = RawData?.Length ?? 0;
+
+            dataLength = ((dataLength + 3) & ~3) / 4;
+
+            word1 = (dataLength & 0x3ff);
+
+            if (HandleDesc != null)
             {
-                BinaryWriter writer = new BinaryWriter(ms);
-
-                int word0;
-                int word1;
-
-                word0 = (int)Type;
-                word0 |= (SendBuff.Count & 0xf) << 20;
-                word0 |= (ReceiveBuff.Count & 0xf) << 24;
-                word0 |= (ExchangeBuff.Count & 0xf) << 28;
-
-                byte[] handleData = Array.Empty<byte>();
-
-                if (HandleDesc != null)
-                {
-                    handleData = HandleDesc.GetBytes();
-                }
-
-                int dataLength = RawData?.Length ?? 0;
-
-                dataLength = ((dataLength + 3) & ~3) / 4;
-
-                word1 = (dataLength & 0x3ff);
-
-                if (HandleDesc != null)
-                {
-                    word1 |= 1 << 31;
-                }
-
-                writer.Write(word0);
-                writer.Write(word1);
-                writer.Write(handleData);
-
-                if (RawData != null)
-                {
-                    writer.Write(RawData);
-                }
-
-                return ms.ToArray();
+                word1 |= 1 << 31;
             }
+
+            ms.Write(word0);
+            ms.Write(word1);
+
+            if (handleDataStream != null)
+            {
+                ms.Write(handleDataStream);
+            }
+
+            if (RawData != null)
+            {
+                ms.Write(RawData);
+            }
+
+            return ms;
         }
 
         private long GetPadSize16(long position)

Ryujinx.HLE/HOS/Ipc/IpcRecvListBuffDesc.cs

@@ -13,13 +13,11 @@ namespace Ryujinx.HLE.HOS.Ipc
             Size = size;
         }
 
-        public IpcRecvListBuffDesc(BinaryReader reader)
+        public IpcRecvListBuffDesc(ulong packedValue)
         {
-            ulong value = reader.ReadUInt64();
+            Position = packedValue & 0xffffffffffff;
 
-            Position = value & 0xffffffffffff;
-
-            Size = (ushort)(value >> 48);
+            Size = (ushort)(packedValue >> 48);
         }
     }
 }

Ryujinx.HLE/HOS/Kernel/Common/KTimeManager.cs

@@ -16,7 +16,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
 
             public WaitingObject(IKFutureSchedulerObject schedulerObj, long timePoint)
             {
-                Object    = schedulerObj;
+                Object = schedulerObj;
                 TimePoint = timePoint;
             }
         }
@@ -27,6 +27,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
         private bool _keepRunning;
         private long _enforceWakeupFromSpinWait;
 
+        private const long NanosecondsPerSecond = 1000000000L;
+        private const long NanosecondsPerMillisecond = 1000000L;
+
         public KTimeManager(KernelContext context)
         {
             _context = context;
@@ -55,7 +58,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
             {
                 _waitingObjects.Add(new WaitingObject(schedulerObj, timePoint));
 
-                if (timeout < 1000000)
+                if (timeout < NanosecondsPerMillisecond)
                 {
                     Interlocked.Exchange(ref _enforceWakeupFromSpinWait, 1);
                 }
@@ -142,7 +145,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
         private WaitingObject GetNextWaitingObject()
         {
             WaitingObject selected = null;
-            
+
             long lowestTimePoint = long.MaxValue;
 
             for (int index = _waitingObjects.Count - 1; index >= 0; index--)
@@ -161,7 +164,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
 
         public static long ConvertNanosecondsToMilliseconds(long time)
         {
-            time /= 1000000;
+            time /= NanosecondsPerMillisecond;
 
             if ((ulong)time > int.MaxValue)
             {
@@ -173,18 +176,18 @@ namespace Ryujinx.HLE.HOS.Kernel.Common
 
         public static long ConvertMillisecondsToNanoseconds(long time)
         {
-            return time * 1000000;
+            return time * NanosecondsPerMillisecond;
         }
 
         public static long ConvertNanosecondsToHostTicks(long ns)
         {
-            long nsDiv = ns / 1000000000;
-            long nsMod = ns % 1000000000;
-            long tickDiv = PerformanceCounter.TicksPerSecond / 1000000000;
-            long tickMod = PerformanceCounter.TicksPerSecond % 1000000000;
+            long nsDiv = ns / NanosecondsPerSecond;
+            long nsMod = ns % NanosecondsPerSecond;
+            long tickDiv = PerformanceCounter.TicksPerSecond / NanosecondsPerSecond;
+            long tickMod = PerformanceCounter.TicksPerSecond % NanosecondsPerSecond;
 
-            long baseTicks = (nsMod * tickMod + PerformanceCounter.TicksPerSecond - 1) / 1000000000;
-            return (nsDiv * tickDiv) * 1000000000 + nsDiv * tickMod + nsMod * tickDiv + baseTicks;
+            long baseTicks = (nsMod * tickMod + PerformanceCounter.TicksPerSecond - 1) / NanosecondsPerSecond;
+            return (nsDiv * tickDiv) * NanosecondsPerSecond + nsDiv * tickMod + nsMod * tickDiv + baseTicks;
         }
 
         public static long ConvertGuestTicksToNanoseconds(long ticks)

Ryujinx.HLE/HOS/Kernel/KernelConstants.cs

@@ -7,6 +7,8 @@ namespace Ryujinx.HLE.HOS.Kernel
         public const int InitialKipId = 1;
         public const int InitialProcessId = 0x51;
 
+        public const int SupervisorCallCount = 0xC0;
+
         public const int MemoryBlockAllocatorSize = 0x2710;
 
         public const ulong UserSlabHeapBase = DramMemoryMap.SlabHeapBase;
@@ -15,4 +17,4 @@ namespace Ryujinx.HLE.HOS.Kernel
 
         public const ulong CounterFrequency = 19200000;
     }
-}
+}

Ryujinx.HLE/HOS/Kernel/KernelStatic.cs

@@ -18,7 +18,7 @@ namespace Ryujinx.HLE.HOS.Kernel
         public static Result StartInitialProcess(
             KernelContext context,
             ProcessCreationInfo creationInfo,
-            ReadOnlySpan<int> capabilities,
+            ReadOnlySpan<uint> capabilities,
             int mainThreadPriority,
             ThreadStart customThreadStart)
         {

Ryujinx.HLE/HOS/Kernel/Process/CapabilityExtensions.cs

@@ -0,0 +1,22 @@
+using System.Numerics;
+
+namespace Ryujinx.HLE.HOS.Kernel.Process
+{
+    static class CapabilityExtensions
+    {
+        public static CapabilityType GetCapabilityType(this uint cap)
+        {
+            return (CapabilityType)(((cap + 1) & ~cap) - 1);
+        }
+
+        public static uint GetFlag(this CapabilityType type)
+        {
+            return (uint)type + 1;
+        }
+
+        public static uint GetId(this CapabilityType type)
+        {
+            return (uint)BitOperations.TrailingZeroCount(type.GetFlag());
+        }
+    }
+}

Ryujinx.HLE/HOS/Kernel/Process/CapabilityType.cs

@@ -0,0 +1,19 @@
+namespace Ryujinx.HLE.HOS.Kernel.Process
+{
+    enum CapabilityType : uint
+    {
+        CorePriority  = (1u <<  3) - 1,
+        SyscallMask   = (1u <<  4) - 1,
+        MapRange      = (1u <<  6) - 1,
+        MapIoPage     = (1u <<  7) - 1,
+        MapRegion     = (1u << 10) - 1,
+        InterruptPair = (1u << 11) - 1,
+        ProgramType   = (1u << 13) - 1,
+        KernelVersion = (1u << 14) - 1,
+        HandleTable   = (1u << 15) - 1,
+        DebugFlags    = (1u << 16) - 1,
+
+        Invalid       = 0u,
+        Padding       = ~0u
+    }
+}

Ryujinx.HLE/HOS/Kernel/Process/KHandleTable.cs

@@ -19,7 +19,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
         private int _activeSlotsCount;
 
-        private int _size;
+        private uint _size;
 
         private ushort _idCounter;
 
@@ -28,9 +28,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
             _context = context;
         }
 
-        public Result Initialize(int size)
+        public Result Initialize(uint size)
         {
-            if ((uint)size > 1024)
+            if (size > 1024)
             {
                 return KernelResult.OutOfMemory;
             }

Ryujinx.HLE/HOS/Kernel/Process/KProcess.cs

@@ -16,11 +16,11 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 {
     class KProcess : KSynchronizationObject
     {
-        public const int KernelVersionMajor = 10;
-        public const int KernelVersionMinor = 4;
-        public const int KernelVersionRevision = 0;
+        public const uint KernelVersionMajor = 10;
+        public const uint KernelVersionMinor = 4;
+        public const uint KernelVersionRevision = 0;
 
-        public const int KernelVersionPacked =
+        public const uint KernelVersionPacked =
             (KernelVersionMajor << 19) |
             (KernelVersionMinor << 15) |
             (KernelVersionRevision << 0);
@@ -119,7 +119,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
         public Result InitializeKip(
             ProcessCreationInfo creationInfo,
-            ReadOnlySpan<int> capabilities,
+            ReadOnlySpan<uint> capabilities,
             KPageList pageList,
             KResourceLimit resourceLimit,
             MemoryRegion memRegion,
@@ -190,7 +190,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
         public Result Initialize(
             ProcessCreationInfo creationInfo,
-            ReadOnlySpan<int> capabilities,
+            ReadOnlySpan<uint> capabilities,
             KResourceLimit resourceLimit,
             MemoryRegion memRegion,
             IProcessContextFactory contextFactory,

Ryujinx.HLE/HOS/Kernel/Process/KProcessCapabilities.cs

@@ -1,4 +1,3 @@
-using Ryujinx.HLE.HOS.Kernel.Common;
 using Ryujinx.HLE.HOS.Kernel.Memory;
 using Ryujinx.HLE.HOS.Kernel.Threading;
 using Ryujinx.Horizon.Common;
@@ -9,48 +8,49 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 {
     class KProcessCapabilities
     {
-        public byte[] SvcAccessMask { get; private set; }
-        public byte[] IrqAccessMask { get; private set; }
+        public byte[] SvcAccessMask { get; }
+        public byte[] IrqAccessMask { get; }
 
         public ulong AllowedCpuCoresMask    { get; private set; }
         public ulong AllowedThreadPriosMask { get; private set; }
 
-        public int DebuggingFlags       { get; private set; }
-        public int HandleTableSize      { get; private set; }
-        public int KernelReleaseVersion { get; private set; }
-        public int ApplicationType      { get; private set; }
+        public uint DebuggingFlags       { get; private set; }
+        public uint HandleTableSize      { get; private set; }
+        public uint KernelReleaseVersion { get; private set; }
+        public uint ApplicationType      { get; private set; }
 
         public KProcessCapabilities()
         {
-            SvcAccessMask = new byte[0x10];
+            // length / number of bits of the underlying type
+            SvcAccessMask = new byte[KernelConstants.SupervisorCallCount / 8];
             IrqAccessMask = new byte[0x80];
         }
 
-        public Result InitializeForKernel(ReadOnlySpan<int> capabilities, KPageTableBase memoryManager)
+        public Result InitializeForKernel(ReadOnlySpan<uint> capabilities, KPageTableBase memoryManager)
         {
             AllowedCpuCoresMask    = 0xf;
             AllowedThreadPriosMask = ulong.MaxValue;
-            DebuggingFlags        &= ~3;
+            DebuggingFlags        &= ~3u;
             KernelReleaseVersion   = KProcess.KernelVersionPacked;
 
             return Parse(capabilities, memoryManager);
         }
 
-        public Result InitializeForUser(ReadOnlySpan<int> capabilities, KPageTableBase memoryManager)
+        public Result InitializeForUser(ReadOnlySpan<uint> capabilities, KPageTableBase memoryManager)
         {
             return Parse(capabilities, memoryManager);
         }
 
-        private Result Parse(ReadOnlySpan<int> capabilities, KPageTableBase memoryManager)
+        private Result Parse(ReadOnlySpan<uint> capabilities, KPageTableBase memoryManager)
         {
             int mask0 = 0;
             int mask1 = 0;
 
             for (int index = 0; index < capabilities.Length; index++)
             {
-                int cap = capabilities[index];
+                uint cap = capabilities[index];
 
-                if (((cap + 1) & ~cap) != 0x40)
+                if (cap.GetCapabilityType() != CapabilityType.MapRange)
                 {
                     Result result = ParseCapability(cap, ref mask0, ref mask1, memoryManager);
 
@@ -66,7 +66,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                         return KernelResult.InvalidCombination;
                     }
 
-                    int prevCap = cap;
+                    uint prevCap = cap;
 
                     cap = capabilities[++index];
 
@@ -85,8 +85,8 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                         return KernelResult.InvalidSize;
                     }
 
-                    long address = ((long)(uint)prevCap << 5) & 0xffffff000;
-                    long size    = ((long)(uint)cap     << 5) & 0xfffff000;
+                    long address = ((long)prevCap << 5) & 0xffffff000;
+                    long size    = ((long)cap     << 5) & 0xfffff000;
 
                     if (((ulong)(address + size - 1) >> 36) != 0)
                     {
@@ -118,20 +118,20 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
             return Result.Success;
         }
 
-        private Result ParseCapability(int cap, ref int mask0, ref int mask1, KPageTableBase memoryManager)
+        private Result ParseCapability(uint cap, ref int mask0, ref int mask1, KPageTableBase memoryManager)
         {
-            int code = (cap + 1) & ~cap;
+            CapabilityType code = cap.GetCapabilityType();
 
-            if (code == 1)
+            if (code == CapabilityType.Invalid)
             {
                 return KernelResult.InvalidCapability;
             }
-            else if (code == 0)
+            else if (code == CapabilityType.Padding)
             {
                 return Result.Success;
             }
 
-            int codeMask = 1 << (32 - BitOperations.LeadingZeroCount((uint)code + 1));
+            int codeMask = 1 << (32 - BitOperations.LeadingZeroCount(code.GetFlag() + 1));
 
             // Check if the property was already set.
             if (((mask0 & codeMask) & 0x1e008) != 0)
@@ -143,23 +143,23 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
             switch (code)
             {
-                case 8:
+                case CapabilityType.CorePriority:
                 {
                     if (AllowedCpuCoresMask != 0 || AllowedThreadPriosMask != 0)
                     {
                         return KernelResult.InvalidCapability;
                     }
 
-                    int lowestCpuCore  = (cap >> 16) & 0xff;
-                    int highestCpuCore = (cap >> 24) & 0xff;
+                    uint lowestCpuCore  = (cap >> 16) & 0xff;
+                    uint highestCpuCore = (cap >> 24) & 0xff;
 
                     if (lowestCpuCore > highestCpuCore)
                     {
                         return KernelResult.InvalidCombination;
                     }
 
-                    int highestThreadPrio = (cap >>  4) & 0x3f;
-                    int lowestThreadPrio  = (cap >> 10) & 0x3f;
+                    uint highestThreadPrio = (cap >>  4) & 0x3f;
+                    uint lowestThreadPrio  = (cap >> 10) & 0x3f;
 
                     if (lowestThreadPrio > highestThreadPrio)
                     {
@@ -177,9 +177,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x10:
+                case CapabilityType.SyscallMask:
                 {
-                    int slot = (cap >> 29) & 7;
+                    int slot = ((int)cap >> 29) & 7;
 
                     int svcSlotMask = 1 << slot;
 
@@ -190,7 +190,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
                     mask1 |= svcSlotMask;
 
-                    int svcMask = (cap >> 5) & 0xffffff;
+                    uint svcMask = (cap >> 5) & 0xffffff;
 
                     int baseSvc = slot * 24;
 
@@ -203,7 +203,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
 
                         int svcId = baseSvc + index;
 
-                        if (svcId > 0x7f)
+                        if (svcId >= KernelConstants.SupervisorCallCount)
                         {
                             return KernelResult.MaximumExceeded;
                         }
@@ -214,20 +214,27 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x80:
+                case CapabilityType.MapIoPage:
                 {
-                    long address = ((long)(uint)cap << 4) & 0xffffff000;
+                    long address = ((long)cap << 4) & 0xffffff000;
 
                     memoryManager.MapIoMemory(address, KPageTableBase.PageSize, KMemoryPermission.ReadAndWrite);
 
                     break;
                 }
 
-                case 0x800:
+                case CapabilityType.MapRegion:
+                {
+                    // TODO: Implement capabilities for MapRegion
+
+                    break;
+                }
+
+                case CapabilityType.InterruptPair:
                 {
                     // TODO: GIC distributor check.
-                    int irq0 = (cap >> 12) & 0x3ff;
-                    int irq1 = (cap >> 22) & 0x3ff;
+                    int irq0 = ((int)cap >> 12) & 0x3ff;
+                    int irq1 = ((int)cap >> 22) & 0x3ff;
 
                     if (irq0 != 0x3ff)
                     {
@@ -242,11 +249,11 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x2000:
+                case CapabilityType.ProgramType:
                 {
-                    int applicationType = cap >> 14;
+                    uint applicationType = (cap >> 14);
 
-                    if ((uint)applicationType > 7)
+                    if (applicationType > 7)
                     {
                         return KernelResult.ReservedValue;
                     }
@@ -256,7 +263,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x4000:
+                case CapabilityType.KernelVersion:
                 {
                     // Note: This check is bugged on kernel too, we are just replicating the bug here.
                     if ((KernelReleaseVersion >> 17) != 0 || cap < 0x80000)
@@ -269,11 +276,11 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x8000:
+                case CapabilityType.HandleTable:
                 {
-                    int handleTableSize = cap >> 26;
+                    uint handleTableSize = cap >> 26;
 
-                    if ((uint)handleTableSize > 0x3ff)
+                    if (handleTableSize > 0x3ff)
                     {
                         return KernelResult.ReservedValue;
                     }
@@ -283,16 +290,16 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
                     break;
                 }
 
-                case 0x10000:
+                case CapabilityType.DebugFlags:
                 {
-                    int debuggingFlags = cap >> 19;
+                    uint debuggingFlags = cap >> 19;
 
-                    if ((uint)debuggingFlags > 3)
+                    if (debuggingFlags > 3)
                     {
                         return KernelResult.ReservedValue;
                     }
 
-                    DebuggingFlags &= ~3;
+                    DebuggingFlags &= ~3u;
                     DebuggingFlags |= debuggingFlags;
 
                     break;
@@ -304,18 +311,18 @@ namespace Ryujinx.HLE.HOS.Kernel.Process
             return Result.Success;
         }
 
-        private static ulong GetMaskFromMinMax(int min, int max)
+        private static ulong GetMaskFromMinMax(uint min, uint max)
         {
-            int range = max - min + 1;
+            uint range = max - min + 1;
 
             if (range == 64)
             {
                 return ulong.MaxValue;
             }
 
-            ulong mask = (1UL << range) - 1;
+            ulong mask = (1UL << (int)range) - 1;
 
-            return mask << min;
+            return mask << (int)min;
         }
     }
 }

Ryujinx.HLE/HOS/Kernel/Process/ProcessCreationFlags.cs

@@ -1,5 +1,8 @@
-namespace Ryujinx.HLE.HOS.Kernel.Process
+using System;
+
+namespace Ryujinx.HLE.HOS.Kernel.Process
 {
+    [Flags]
     enum ProcessCreationFlags
     {
         Is64Bit = 1 << 0,

Ryujinx.HLE/HOS/Kernel/SupervisorCall/Syscall.cs

@@ -54,7 +54,7 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
         public Result CreateProcess(
             out int handle,
             ProcessCreationInfo info,
-            ReadOnlySpan<int> capabilities,
+            ReadOnlySpan<uint> capabilities,
             IProcessContextFactory contextFactory,
             ThreadStart customThreadStart = null)
         {
@@ -553,7 +553,7 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
 
             KProcess currentProcess = KernelStatic.GetCurrentProcess();
 
-            KSynchronizationObject[] syncObjs = new KSynchronizationObject[handles.Length];
+            KSynchronizationObject[] syncObjs = handles.Length == 0 ? Array.Empty<KSynchronizationObject>() : new KSynchronizationObject[handles.Length];
 
             for (int index = 0; index < handles.Length; index++)
             {
@@ -3002,4 +3002,4 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
             return (address & 3) != 0;
         }
     }
-}
+}

Ryujinx.HLE/HOS/Kernel/Threading/KPriorityQueue.cs

@@ -5,11 +5,11 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 {
     class KPriorityQueue
     {
-        private LinkedList<KThread>[][] _scheduledThreadsPerPrioPerCore;
-        private LinkedList<KThread>[][] _suggestedThreadsPerPrioPerCore;
+        private readonly LinkedList<KThread>[][] _scheduledThreadsPerPrioPerCore;
+        private readonly LinkedList<KThread>[][] _suggestedThreadsPerPrioPerCore;
 
-        private long[] _scheduledPrioritiesPerCore;
-        private long[] _suggestedPrioritiesPerCore;
+        private readonly long[] _scheduledPrioritiesPerCore;
+        private readonly long[] _suggestedPrioritiesPerCore;
 
         public KPriorityQueue()
         {
@@ -32,43 +32,134 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
             _suggestedPrioritiesPerCore = new long[KScheduler.CpuCoresCount];
         }
 
-        public IEnumerable<KThread> SuggestedThreads(int core)
+        public readonly ref struct KThreadEnumerable
         {
-            return Iterate(_suggestedThreadsPerPrioPerCore, _suggestedPrioritiesPerCore, core);
-        }
+            readonly LinkedList<KThread>[][] _listPerPrioPerCore;
+            readonly long[] _prios;
+            readonly int _core;
 
-        public IEnumerable<KThread> ScheduledThreads(int core)
-        {
-            return Iterate(_scheduledThreadsPerPrioPerCore, _scheduledPrioritiesPerCore, core);
-        }
-
-        private IEnumerable<KThread> Iterate(LinkedList<KThread>[][] listPerPrioPerCore, long[] prios, int core)
-        {
-            long prioMask = prios[core];
-
-            int prio = BitOperations.TrailingZeroCount(prioMask);
-
-            prioMask &= ~(1L << prio);
-
-            while (prio < KScheduler.PrioritiesCount)
+            public KThreadEnumerable(LinkedList<KThread>[][] listPerPrioPerCore, long[] prios, int core)
             {
-                LinkedList<KThread> list = listPerPrioPerCore[prio][core];
+                _listPerPrioPerCore = listPerPrioPerCore;
+                _prios = prios;
+                _core = core;
+            }
 
-                LinkedListNode<KThread> node = list.First;
+            public Enumerator GetEnumerator()
+            {
+                return new Enumerator(_listPerPrioPerCore, _prios, _core);
+            }
 
-                while (node != null)
+            public ref struct Enumerator
+            {
+                private readonly LinkedList<KThread>[][] _listPerPrioPerCore;
+                private readonly int _core;
+                private long _prioMask;
+                private int _prio;
+                private LinkedList<KThread> _list;
+                private LinkedListNode<KThread> _node;
+
+                public Enumerator(LinkedList<KThread>[][] listPerPrioPerCore, long[] prios, int core)
                 {
-                    yield return node.Value;
-
-                    node = node.Next;
+                    _listPerPrioPerCore = listPerPrioPerCore;
+                    _core = core;
+                    _prioMask = prios[core];
+                    _prio = BitOperations.TrailingZeroCount(_prioMask);
+                    _prioMask &= ~(1L << _prio);
                 }
 
-                prio = BitOperations.TrailingZeroCount(prioMask);
+                public KThread Current => _node?.Value;
 
-                prioMask &= ~(1L << prio);
+                public bool MoveNext()
+                {
+                    _node = _node?.Next;
+
+                    if (_node == null)
+                    {
+                        if (!MoveNextListAndFirstNode())
+                        {
+                            return false;
+                        }
+                    }
+
+                    return _node != null;
+                }
+
+                private bool MoveNextListAndFirstNode()
+                {
+                    if (_prio < KScheduler.PrioritiesCount)
+                    {
+                        _list = _listPerPrioPerCore[_prio][_core];
+
+                        _node = _list.First;
+
+                        _prio = BitOperations.TrailingZeroCount(_prioMask);
+
+                        _prioMask &= ~(1L << _prio);
+
+                        return true;
+                    }
+                    else
+                    {
+                        _list = null;
+                        _node = null;
+                        return false;
+                    }
+                }
             }
         }
 
+        public KThreadEnumerable ScheduledThreads(int core)
+        {
+            return new KThreadEnumerable(_scheduledThreadsPerPrioPerCore, _scheduledPrioritiesPerCore, core);
+        }
+
+        public KThreadEnumerable SuggestedThreads(int core)
+        {
+            return new KThreadEnumerable(_suggestedThreadsPerPrioPerCore, _suggestedPrioritiesPerCore, core);
+        }
+
+        public KThread ScheduledThreadsFirstOrDefault(int core)
+        {
+            return ScheduledThreadsElementAtOrDefault(core, 0);
+        }
+
+        public KThread ScheduledThreadsElementAtOrDefault(int core, int index)
+        {
+            int currentIndex = 0;
+            foreach (var scheduledThread in ScheduledThreads(core))
+            {
+                if (currentIndex == index)
+                {
+                    return scheduledThread;
+                }
+                else
+                {
+                    currentIndex++;
+                }
+            }
+
+            return null;
+        }
+
+        public KThread ScheduledThreadsWithDynamicPriorityFirstOrDefault(int core, int dynamicPriority)
+        {
+            foreach (var scheduledThread in ScheduledThreads(core))
+            {
+                if (scheduledThread.DynamicPriority == dynamicPriority)
+                {
+                    return scheduledThread;
+                }
+            }
+
+            return null;
+        }
+
+        public bool HasScheduledThreads(int core)
+        {
+            return ScheduledThreadsFirstOrDefault(core) != null;
+        }
+
         public void TransferToCore(int prio, int dstCore, KThread thread)
         {
             int srcCore = thread.ActiveCore;

Ryujinx.HLE/HOS/Kernel/Threading/KScheduler.cs

@@ -1,8 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.HLE.HOS.Kernel.Process;
 using System;
-using System.Collections.Generic;
-using System.Linq;
 using System.Numerics;
 using System.Threading;
 
@@ -17,6 +15,8 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
         private static readonly int[] PreemptionPriorities = new int[] { 59, 59, 59, 63 };
 
+        private static readonly int[] _srcCoresHighestPrioThreads = new int[CpuCoresCount];
+
         private readonly KernelContext _context;
         private readonly int _coreId;
 
@@ -86,7 +86,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
             for (int core = 0; core < CpuCoresCount; core++)
             {
-                KThread thread = context.PriorityQueue.ScheduledThreads(core).FirstOrDefault();
+                KThread thread = context.PriorityQueue.ScheduledThreadsFirstOrDefault(core);
 
                 if (thread != null &&
                     thread.Owner != null &&
@@ -115,12 +115,12 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
             {
                 // If the core is not idle (there's already a thread running on it),
                 // then we don't need to attempt load balancing.
-                if (context.PriorityQueue.ScheduledThreads(core).Any())
+                if (context.PriorityQueue.HasScheduledThreads(core))
                 {
                     continue;
                 }
 
-                int[] srcCoresHighestPrioThreads = new int[CpuCoresCount];
+                Array.Fill(_srcCoresHighestPrioThreads, 0);
 
                 int srcCoresHighestPrioThreadsCount = 0;
 
@@ -136,7 +136,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                         break;
                     }
 
-                    srcCoresHighestPrioThreads[srcCoresHighestPrioThreadsCount++] = suggested.ActiveCore;
+                    _srcCoresHighestPrioThreads[srcCoresHighestPrioThreadsCount++] = suggested.ActiveCore;
                 }
 
                 // Not yet selected candidate found.
@@ -158,9 +158,9 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                 // (the first one that doesn't make the source core idle if moved).
                 for (int index = 0; index < srcCoresHighestPrioThreadsCount; index++)
                 {
-                    int srcCore = srcCoresHighestPrioThreads[index];
+                    int srcCore = _srcCoresHighestPrioThreads[index];
 
-                    KThread src = context.PriorityQueue.ScheduledThreads(srcCore).ElementAtOrDefault(1);
+                    KThread src = context.PriorityQueue.ScheduledThreadsElementAtOrDefault(srcCore, 1);
 
                     if (src != null)
                     {
@@ -422,9 +422,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
         private static void RotateScheduledQueue(KernelContext context, int core, int prio)
         {
-            IEnumerable<KThread> scheduledThreads = context.PriorityQueue.ScheduledThreads(core);
-
-            KThread selectedThread = scheduledThreads.FirstOrDefault(x => x.DynamicPriority == prio);
+            KThread selectedThread = context.PriorityQueue.ScheduledThreadsWithDynamicPriorityFirstOrDefault(core, prio);
             KThread nextThread = null;
 
             // Yield priority queue.
@@ -433,14 +431,14 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                 nextThread = context.PriorityQueue.Reschedule(prio, core, selectedThread);
             }
 
-            IEnumerable<KThread> SuitableCandidates()
+            static KThread FirstSuitableCandidateOrDefault(KernelContext context, int core, KThread selectedThread, KThread nextThread, Predicate< KThread> predicate)
             {
                 foreach (KThread suggested in context.PriorityQueue.SuggestedThreads(core))
                 {
                     int suggestedCore = suggested.ActiveCore;
                     if (suggestedCore >= 0)
                     {
-                        KThread selectedSuggestedCore = context.PriorityQueue.ScheduledThreads(suggestedCore).FirstOrDefault();
+                        KThread selectedSuggestedCore = context.PriorityQueue.ScheduledThreadsFirstOrDefault(suggestedCore);
 
                         if (selectedSuggestedCore == suggested || (selectedSuggestedCore != null && selectedSuggestedCore.DynamicPriority < 2))
                         {
@@ -453,14 +451,19 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                         nextThread == null ||
                         nextThread.LastScheduledTime >= suggested.LastScheduledTime)
                     {
-                        yield return suggested;
+                        if (predicate(suggested))
+                        {
+                            return suggested;
+                        }
                     }
                 }
+
+                return null;
             }
 
             // Select candidate threads that could run on this core.
             // Only take into account threads that are not yet selected.
-            KThread dst = SuitableCandidates().FirstOrDefault(x => x.DynamicPriority == prio);
+            KThread dst = FirstSuitableCandidateOrDefault(context, core, selectedThread, nextThread, x => x.DynamicPriority == prio);
 
             if (dst != null)
             {
@@ -469,11 +472,11 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
             // If the priority of the currently selected thread is lower or same as the preemption priority,
             // then try to migrate a thread with lower priority.
-            KThread bestCandidate = context.PriorityQueue.ScheduledThreads(core).FirstOrDefault();
+            KThread bestCandidate = context.PriorityQueue.ScheduledThreadsFirstOrDefault(core);
 
             if (bestCandidate != null && bestCandidate.DynamicPriority >= prio)
             {
-                dst = SuitableCandidates().FirstOrDefault(x => x.DynamicPriority < bestCandidate.DynamicPriority);
+                dst = FirstSuitableCandidateOrDefault(context, core, selectedThread, nextThread, x => x.DynamicPriority < bestCandidate.DynamicPriority);
 
                 if (dst != null)
                 {
@@ -534,7 +537,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
             // Move current thread to the end of the queue.
             KThread nextThread = context.PriorityQueue.Reschedule(prio, core, currentThread);
 
-            IEnumerable<KThread> SuitableCandidates()
+            static KThread FirstSuitableCandidateOrDefault(KernelContext context, int core, KThread nextThread, int lessThanOrEqualPriority)
             {
                 foreach (KThread suggested in context.PriorityQueue.SuggestedThreads(core))
                 {
@@ -554,12 +557,17 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                     if (suggested.LastScheduledTime <= nextThread.LastScheduledTime ||
                         suggested.DynamicPriority < nextThread.DynamicPriority)
                     {
-                        yield return suggested;
+                        if (suggested.DynamicPriority <= lessThanOrEqualPriority)
+                        {
+                            return suggested;
+                        }
                     }
                 }
+
+                return null;
             }
 
-            KThread dst = SuitableCandidates().FirstOrDefault(x => x.DynamicPriority <= prio);
+            KThread dst = FirstSuitableCandidateOrDefault(context, core, nextThread, prio);
 
             if (dst != null)
             {
@@ -596,7 +604,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
 
             context.PriorityQueue.TransferToCore(currentThread.DynamicPriority, -1, currentThread);
 
-            if (!context.PriorityQueue.ScheduledThreads(core).Any())
+            if (!context.PriorityQueue.HasScheduledThreads(core))
             {
                 KThread selectedThread = null;
 
@@ -609,7 +617,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                         continue;
                     }
 
-                    KThread firstCandidate = context.PriorityQueue.ScheduledThreads(suggestedCore).FirstOrDefault();
+                    KThread firstCandidate = context.PriorityQueue.ScheduledThreadsFirstOrDefault(suggestedCore);
 
                     if (firstCandidate == suggested)
                     {

Ryujinx.HLE/HOS/Kernel/Threading/KSynchronization.cs

@@ -59,7 +59,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
             }
             else
             {
-                LinkedListNode<KThread>[] syncNodes = new LinkedListNode<KThread>[syncObjs.Length];
+                LinkedListNode<KThread>[] syncNodes = syncObjs.Length == 0 ? Array.Empty<LinkedListNode<KThread>>() : new LinkedListNode<KThread>[syncObjs.Length];
 
                 for (int index = 0; index < syncObjs.Length; index++)
                 {

Ryujinx.HLE/HOS/ProgramLoader.cs

@@ -80,7 +80,7 @@ namespace Ryujinx.HLE.HOS
 
             ulong codeBaseAddress = kip.Is64BitAddressSpace ? 0x8000000UL : 0x200000UL;
 
-            ulong codeAddress = codeBaseAddress + (ulong)kip.TextOffset;
+            ulong codeAddress = codeBaseAddress + kip.TextOffset;
 
             ProcessCreationFlags flags = 0;
 
@@ -231,13 +231,13 @@ namespace Ryujinx.HLE.HOS
 
                 nsoSize = BitUtils.AlignUp<uint>(nsoSize, KPageTableBase.PageSize);
 
-                nsoBase[index] = codeStart + (ulong)codeSize;
+                nsoBase[index] = codeStart + codeSize;
 
                 codeSize += nsoSize;
 
                 if (arguments != null && argsSize == 0)
                 {
-                    argsStart = (ulong)codeSize;
+                    argsStart = codeSize;
 
                     argsSize = (uint)BitUtils.AlignDown(arguments.Length * 2 + ArgsTotalSize - 1, KPageTableBase.PageSize);
 
@@ -318,7 +318,7 @@ namespace Ryujinx.HLE.HOS
 
             result = process.Initialize(
                 creationInfo,
-                MemoryMarshal.Cast<byte, int>(npdm.KernelCapabilityData).ToArray(),
+                MemoryMarshal.Cast<byte, uint>(npdm.KernelCapabilityData),
                 resourceLimit,
                 memoryRegion,
                 processContextFactory);

Ryujinx.HLE/HOS/Services/Am/AppletAE/Storage/StorageHelper.cs

@@ -1,4 +1,5 @@
-using Ryujinx.HLE.HOS.Services.Account.Acc;
+using Ryujinx.Common.Memory;
+using Ryujinx.HLE.HOS.Services.Account.Acc;
 using System.IO;
 
 namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.Storage
@@ -10,7 +11,7 @@ namespace Ryujinx.HLE.HOS.Services.Am.AppletAE.Storage
         public static byte[] MakeLaunchParams(UserProfile userProfile)
         {
             // Size needs to be at least 0x88 bytes otherwise application errors.
-            using (MemoryStream ms = new MemoryStream())
+            using (MemoryStream ms = MemoryStreamManager.Shared.GetStream())
             {
                 BinaryWriter writer = new BinaryWriter(ms);
 

Ryujinx.HLE/HOS/Services/Nim/IShopServiceAccessServerInterface.cs

@@ -14,6 +14,9 @@ namespace Ryujinx.HLE.HOS.Services.Nim
         // CreateServerInterface(pid, handle<unknown>, u64) -> object<nn::ec::IShopServiceAccessServer>
         public ResultCode CreateServerInterface(ServiceCtx context)
         {
+            // Close transfer memory immediately as we don't use it.
+            context.Device.System.KernelContext.Syscall.CloseHandle(context.Request.HandleDesc.ToCopy[0]);
+
             MakeObject(context, new IShopServiceAccessServer());
 
             Logger.Stub?.PrintStub(LogClass.ServiceNim);

Ryujinx.HLE/HOS/Services/Sdb/Pl/SharedFontManager.cs

@@ -5,6 +5,7 @@ using LibHac.FsSystem;
 using LibHac.Ncm;
 using LibHac.Tools.FsSystem;
 using LibHac.Tools.FsSystem.NcaUtils;
+using Ryujinx.Common.Memory;
 using Ryujinx.HLE.Exceptions;
 using Ryujinx.HLE.FileSystem;
 using Ryujinx.HLE.HOS.Kernel.Memory;
@@ -160,7 +161,7 @@ namespace Ryujinx.HLE.HOS.Services.Sdb.Pl
             static uint KXor(uint data) => data ^ FontKey;
 
             using (BinaryReader reader    = new BinaryReader(bfttfStream))
-            using (MemoryStream ttfStream = new MemoryStream())
+            using (MemoryStream ttfStream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter output    = new BinaryWriter(ttfStream))
             {
                 if (KXor(reader.ReadUInt32()) != BFTTFMagic)

Ryujinx.HLE/HOS/Services/ServerBase.cs

@@ -1,3 +1,5 @@
+using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using Ryujinx.HLE.HOS.Ipc;
 using Ryujinx.HLE.HOS.Kernel;
 using Ryujinx.HLE.HOS.Kernel.Ipc;
@@ -5,6 +7,7 @@ using Ryujinx.HLE.HOS.Kernel.Process;
 using Ryujinx.HLE.HOS.Kernel.Threading;
 using Ryujinx.Horizon.Common;
 using System;
+using System.Buffers;
 using System.Buffers.Binary;
 using System.Collections.Generic;
 using System.IO;
@@ -19,7 +22,7 @@ namespace Ryujinx.HLE.HOS.Services
         // not large enough.
         private const int PointerBufferSize = 0x8000;
 
-        private readonly static int[] DefaultCapabilities = new int[]
+        private readonly static uint[] DefaultCapabilities = new uint[]
         {
             0x030363F7,
             0x1FFFFFCF,
@@ -29,6 +32,8 @@ namespace Ryujinx.HLE.HOS.Services
             0x01007FFF
         };
 
+        private readonly object _handleLock = new();
+
         private readonly KernelContext _context;
         private KProcess _selfProcess;
 
@@ -37,14 +42,27 @@ namespace Ryujinx.HLE.HOS.Services
         private readonly Dictionary<int, IpcService> _sessions = new Dictionary<int, IpcService>();
         private readonly Dictionary<int, Func<IpcService>> _ports = new Dictionary<int, Func<IpcService>>();
 
+        private readonly MemoryStream _requestDataStream;
+        private readonly BinaryReader _requestDataReader;
+
+        private readonly MemoryStream _responseDataStream;
+        private readonly BinaryWriter _responseDataWriter;
+
         public ManualResetEvent InitDone { get; }
         public string Name { get; }
         public Func<IpcService> SmObjectFactory { get; }
 
         public ServerBase(KernelContext context, string name, Func<IpcService> smObjectFactory = null)
         {
-            InitDone = new ManualResetEvent(false);
             _context = context;
+
+            _requestDataStream = MemoryStreamManager.Shared.GetStream();
+            _requestDataReader = new BinaryReader(_requestDataStream);
+
+            _responseDataStream = MemoryStreamManager.Shared.GetStream();
+            _responseDataWriter = new BinaryWriter(_responseDataStream);
+
+            InitDone = new ManualResetEvent(false);
             Name = name;
             SmObjectFactory = smObjectFactory;
 
@@ -61,7 +79,10 @@ namespace Ryujinx.HLE.HOS.Services
 
         private void AddPort(int serverPortHandle, Func<IpcService> objectFactory)
         {
-            _portHandles.Add(serverPortHandle);
+            lock (_handleLock)
+            {
+                _portHandles.Add(serverPortHandle);
+            }
             _ports.Add(serverPortHandle, objectFactory);
         }
 
@@ -76,7 +97,10 @@ namespace Ryujinx.HLE.HOS.Services
 
         public void AddSessionObj(int serverSessionHandle, IpcService obj)
         {
-            _sessionHandles.Add(serverSessionHandle);
+            lock (_handleLock)
+            {
+                _sessionHandles.Add(serverSessionHandle);
+            }
             _sessions.Add(serverSessionHandle, obj);
         }
 
@@ -110,15 +134,23 @@ namespace Ryujinx.HLE.HOS.Services
 
             while (true)
             {
-                int[] portHandles = _portHandles.ToArray();
-                int[] sessionHandles = _sessionHandles.ToArray();
-                int[] handles = new int[portHandles.Length + sessionHandles.Length];
+                int handleCount;
+                int portHandleCount;
+                int[] handles;
 
-                portHandles.CopyTo(handles, 0);
-                sessionHandles.CopyTo(handles, portHandles.Length);
+                lock (_handleLock)
+                {
+                    portHandleCount = _portHandles.Count;
+                    handleCount = portHandleCount + _sessionHandles.Count;
+
+                    handles = ArrayPool<int>.Shared.Rent(handleCount);
+
+                    _portHandles.CopyTo(handles, 0);
+                    _sessionHandles.CopyTo(handles, portHandleCount);
+                }
 
                 // We still need a timeout here to allow the service to pick up and listen new sessions...
-                var rc = _context.Syscall.ReplyAndReceive(out int signaledIndex, handles, replyTargetHandle, 1000000L);
+                var rc = _context.Syscall.ReplyAndReceive(out int signaledIndex, handles.AsSpan(0, handleCount), replyTargetHandle, 1000000L);
 
                 thread.HandlePostSyscall();
 
@@ -129,7 +161,7 @@ namespace Ryujinx.HLE.HOS.Services
 
                 replyTargetHandle = 0;
 
-                if (rc == Result.Success && signaledIndex >= portHandles.Length)
+                if (rc == Result.Success && signaledIndex >= portHandleCount)
                 {
                     // We got a IPC request, process it, pass to the appropriate service if needed.
                     int signaledHandle = handles[signaledIndex];
@@ -156,6 +188,8 @@ namespace Ryujinx.HLE.HOS.Services
                     _selfProcess.CpuMemory.Write(messagePtr + 0x4, 2 << 10);
                     _selfProcess.CpuMemory.Write(messagePtr + 0x8, heapAddr | ((ulong)PointerBufferSize << 48));
                 }
+
+                ArrayPool<int>.Shared.Return(handles);
             }
 
             Dispose();
@@ -166,13 +200,9 @@ namespace Ryujinx.HLE.HOS.Services
             KProcess process = KernelStatic.GetCurrentProcess();
             KThread thread = KernelStatic.GetCurrentThread();
             ulong messagePtr = thread.TlsAddress;
-            ulong messageSize = 0x100;
 
-            byte[] reqData = new byte[messageSize];
+            IpcMessage request = ReadRequest(process, messagePtr);
 
-            process.CpuMemory.Read(messagePtr, reqData);
-
-            IpcMessage request = new IpcMessage(reqData, (long)messagePtr);
             IpcMessage response = new IpcMessage();
 
             ulong tempAddr = recvListAddr;
@@ -202,158 +232,160 @@ namespace Ryujinx.HLE.HOS.Services
             bool shouldReply = true;
             bool isTipcCommunication = false;
 
-            using (MemoryStream raw = new MemoryStream(request.RawData))
+            _requestDataStream.SetLength(0);
+            _requestDataStream.Write(request.RawData);
+            _requestDataStream.Position = 0;
+
+            if (request.Type == IpcMessageType.HipcRequest ||
+                request.Type == IpcMessageType.HipcRequestWithContext)
             {
-                BinaryReader reqReader = new BinaryReader(raw);
+                response.Type = IpcMessageType.HipcResponse;
 
-                if (request.Type == IpcMessageType.HipcRequest ||
-                    request.Type == IpcMessageType.HipcRequestWithContext)
+                _responseDataStream.SetLength(0);
+
+                ServiceCtx context = new ServiceCtx(
+                    _context.Device,
+                    process,
+                    process.CpuMemory,
+                    thread,
+                    request,
+                    response,
+                    _requestDataReader,
+                    _responseDataWriter);
+
+                _sessions[serverSessionHandle].CallHipcMethod(context);
+
+                response.RawData = _responseDataStream.ToArray();
+            }
+            else if (request.Type == IpcMessageType.HipcControl ||
+                        request.Type == IpcMessageType.HipcControlWithContext)
+            {
+                uint magic = (uint)_requestDataReader.ReadUInt64();
+                uint cmdId = (uint)_requestDataReader.ReadUInt64();
+
+                switch (cmdId)
                 {
-                    response.Type = IpcMessageType.HipcResponse;
+                    case 0:
+                        FillHipcResponse(response, 0, _sessions[serverSessionHandle].ConvertToDomain());
+                        break;
 
-                    using (MemoryStream resMs = new MemoryStream())
-                    {
-                        BinaryWriter resWriter = new BinaryWriter(resMs);
+                    case 3:
+                        FillHipcResponse(response, 0, PointerBufferSize);
+                        break;
 
-                        ServiceCtx context = new ServiceCtx(
-                            _context.Device,
-                            process,
-                            process.CpuMemory,
-                            thread,
-                            request,
-                            response,
-                            reqReader,
-                            resWriter);
+                    // TODO: Whats the difference between IpcDuplicateSession/Ex?
+                    case 2:
+                    case 4:
+                        int unknown = _requestDataReader.ReadInt32();
 
-                        _sessions[serverSessionHandle].CallHipcMethod(context);
+                        _context.Syscall.CreateSession(out int dupServerSessionHandle, out int dupClientSessionHandle, false, 0);
 
-                        response.RawData = resMs.ToArray();
-                    }
+                        AddSessionObj(dupServerSessionHandle, _sessions[serverSessionHandle]);
+
+                        response.HandleDesc = IpcHandleDesc.MakeMove(dupClientSessionHandle);
+
+                        FillHipcResponse(response, 0);
+
+                        break;
+
+                    default: throw new NotImplementedException(cmdId.ToString());
                 }
-                else if (request.Type == IpcMessageType.HipcControl ||
-                         request.Type == IpcMessageType.HipcControlWithContext)
+            }
+            else if (request.Type == IpcMessageType.HipcCloseSession || request.Type == IpcMessageType.TipcCloseSession)
+            {
+                _context.Syscall.CloseHandle(serverSessionHandle);
+                lock (_handleLock)
                 {
-                    uint magic = (uint)reqReader.ReadUInt64();
-                    uint cmdId = (uint)reqReader.ReadUInt64();
-
-                    switch (cmdId)
-                    {
-                        case 0:
-                            request = FillResponse(response, 0, _sessions[serverSessionHandle].ConvertToDomain());
-                            break;
-
-                        case 3:
-                            request = FillResponse(response, 0, PointerBufferSize);
-                            break;
-
-                        // TODO: Whats the difference between IpcDuplicateSession/Ex?
-                        case 2:
-                        case 4:
-                            int unknown = reqReader.ReadInt32();
-
-                            _context.Syscall.CreateSession(out int dupServerSessionHandle, out int dupClientSessionHandle, false, 0);
-
-                            AddSessionObj(dupServerSessionHandle, _sessions[serverSessionHandle]);
-
-                            response.HandleDesc = IpcHandleDesc.MakeMove(dupClientSessionHandle);
-
-                            request = FillResponse(response, 0);
-
-                            break;
-
-                        default: throw new NotImplementedException(cmdId.ToString());
-                    }
-                }
-                else if (request.Type == IpcMessageType.HipcCloseSession || request.Type == IpcMessageType.TipcCloseSession)
-                {
-                    _context.Syscall.CloseHandle(serverSessionHandle);
                     _sessionHandles.Remove(serverSessionHandle);
-                    IpcService service = _sessions[serverSessionHandle];
-                    if (service is IDisposable disposableObj)
-                    {
-                        disposableObj.Dispose();
-                    }
-                    _sessions.Remove(serverSessionHandle);
-                    shouldReply = false;
                 }
-                // If the type is past 0xF, we are using TIPC
-                else if (request.Type > IpcMessageType.TipcCloseSession)
-                {
-                    isTipcCommunication = true;
-
-                    // Response type is always the same as request on TIPC.
-                    response.Type = request.Type;
-
-                    using (MemoryStream resMs = new MemoryStream())
-                    {
-                        BinaryWriter resWriter = new BinaryWriter(resMs);
-
-                        ServiceCtx context = new ServiceCtx(
-                            _context.Device,
-                            process,
-                            process.CpuMemory,
-                            thread,
-                            request,
-                            response,
-                            reqReader,
-                            resWriter);
-
-                        _sessions[serverSessionHandle].CallTipcMethod(context);
-
-                        response.RawData = resMs.ToArray();
-                    }
-
-                    process.CpuMemory.Write(messagePtr, response.GetBytesTipc());
-                }
-                else
-                {
-                    throw new NotImplementedException(request.Type.ToString());
-                }
-
-                if (!isTipcCommunication)
-                {
-                    process.CpuMemory.Write(messagePtr, response.GetBytes((long)messagePtr, recvListAddr | ((ulong)PointerBufferSize << 48)));
-                }
-
-                return shouldReply;
+                IpcService service = _sessions[serverSessionHandle];
+                (service as IDisposable)?.Dispose();
+                _sessions.Remove(serverSessionHandle);
+                shouldReply = false;
             }
-        }
-
-        private static IpcMessage FillResponse(IpcMessage response, long result, params int[] values)
-        {
-            using (MemoryStream ms = new MemoryStream())
+            // If the type is past 0xF, we are using TIPC
+            else if (request.Type > IpcMessageType.TipcCloseSession)
             {
-                BinaryWriter writer = new BinaryWriter(ms);
+                isTipcCommunication = true;
 
-                foreach (int value in values)
-                {
-                    writer.Write(value);
-                }
+                // Response type is always the same as request on TIPC.
+                response.Type = request.Type;
 
-                return FillResponse(response, result, ms.ToArray());
+                _responseDataStream.SetLength(0);
+
+                ServiceCtx context = new ServiceCtx(
+                    _context.Device,
+                    process,
+                    process.CpuMemory,
+                    thread,
+                    request,
+                    response,
+                    _requestDataReader,
+                    _responseDataWriter);
+
+                _sessions[serverSessionHandle].CallTipcMethod(context);
+
+                response.RawData = _responseDataStream.ToArray();
+
+                using var responseStream = response.GetStreamTipc();
+                process.CpuMemory.Write(messagePtr, responseStream.GetReadOnlySequence());
             }
+            else
+            {
+                throw new NotImplementedException(request.Type.ToString());
+            }
+
+            if (!isTipcCommunication)
+            {
+                using var responseStream = response.GetStream((long)messagePtr, recvListAddr | ((ulong)PointerBufferSize << 48));
+                process.CpuMemory.Write(messagePtr, responseStream.GetReadOnlySequence());
+            }
+
+            return shouldReply;
         }
 
-        private static IpcMessage FillResponse(IpcMessage response, long result, byte[] data = null)
+        private static IpcMessage ReadRequest(KProcess process, ulong messagePtr)
+        {
+            const int messageSize = 0x100;
+
+            byte[] reqData = ArrayPool<byte>.Shared.Rent(messageSize);
+
+            Span<byte> reqDataSpan = reqData.AsSpan(0, messageSize);
+            reqDataSpan.Clear();
+
+            process.CpuMemory.Read(messagePtr, reqDataSpan);
+
+            IpcMessage request = new IpcMessage(reqDataSpan, (long)messagePtr);
+
+            ArrayPool<byte>.Shared.Return(reqData);
+
+            return request;
+        }
+
+        private void FillHipcResponse(IpcMessage response, long result)
+        {
+            FillHipcResponse(response, result, ReadOnlySpan<byte>.Empty);
+        }
+
+        private void FillHipcResponse(IpcMessage response, long result, int value)
+        {
+            Span<byte> span = stackalloc byte[sizeof(int)];
+            BinaryPrimitives.WriteInt32LittleEndian(span, value);
+            FillHipcResponse(response, result, span);
+        }
+
+        private void FillHipcResponse(IpcMessage response, long result, ReadOnlySpan<byte> data)
         {
             response.Type = IpcMessageType.HipcResponse;
 
-            using (MemoryStream ms = new MemoryStream())
-            {
-                BinaryWriter writer = new BinaryWriter(ms);
+            _responseDataStream.SetLength(0);
 
-                writer.Write(IpcMagic.Sfco);
-                writer.Write(result);
+            _responseDataStream.Write(IpcMagic.Sfco);
+            _responseDataStream.Write(result);
 
-                if (data != null)
-                {
-                    writer.Write(data);
-                }
+            _responseDataStream.Write(data);
 
-                response.RawData = ms.ToArray();
-            }
-
-            return response;
+            response.RawData = _responseDataStream.ToArray();
         }
 
         protected virtual void Dispose(bool disposing)
@@ -372,6 +404,11 @@ namespace Ryujinx.HLE.HOS.Services
 
                 _sessions.Clear();
 
+                _requestDataReader.Dispose();
+                _requestDataStream.Dispose();
+                _responseDataWriter.Dispose();
+                _responseDataStream.Dispose();
+
                 InitDone.Dispose();
             }
         }

Ryujinx.HLE/Loaders/Executables/KipExecutable.cs

@@ -22,7 +22,7 @@ namespace Ryujinx.HLE.Loaders.Executables
         public uint DataSize { get; }
         public uint BssSize  { get; }
 
-        public int[] Capabilities       { get; }
+        public uint[] Capabilities      { get; }
         public bool UsesSecureMemory    { get; }
         public bool Is64BitAddressSpace { get; }
         public bool Is64Bit             { get; }
@@ -57,11 +57,11 @@ namespace Ryujinx.HLE.Loaders.Executables
             Version     = reader.Version;
             Name        = reader.Name.ToString();
 
-            Capabilities = new int[32];
+            Capabilities = new uint[32];
 
             for (int index = 0; index < Capabilities.Length; index++)
             {
-                Capabilities[index] = (int)reader.Capabilities[index];
+                Capabilities[index] = reader.Capabilities[index];
             }
 
             reader.GetSegmentSize(KipReader.SegmentType.Data, out int uncompressedSize).ThrowIfFailure();

Ryujinx.HLE/Utilities/StringUtils.cs

@@ -1,4 +1,6 @@
 using LibHac.Common;
+using Microsoft.IO;
+using Ryujinx.Common.Memory;
 using Ryujinx.HLE.HOS;
 using System;
 using System.Globalization;
@@ -77,7 +79,7 @@ namespace Ryujinx.HLE.Utilities
             ulong position = context.Request.PtrBuff[index].Position;
             ulong size     = context.Request.PtrBuff[index].Size;
 
-            using (MemoryStream ms = new MemoryStream())
+            using (RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream())
             {
                 while (size-- > 0)
                 {
@@ -91,7 +93,7 @@ namespace Ryujinx.HLE.Utilities
                     ms.WriteByte(value);
                 }
 
-                return Encoding.UTF8.GetString(ms.ToArray());
+                return Encoding.UTF8.GetString(ms.GetReadOnlySequence());
             }
         }
 
@@ -110,7 +112,7 @@ namespace Ryujinx.HLE.Utilities
             ulong position = context.Request.SendBuff[index].Position;
             ulong size     = context.Request.SendBuff[index].Size;
 
-            using (MemoryStream ms = new MemoryStream())
+            using (RecyclableMemoryStream ms = MemoryStreamManager.Shared.GetStream())
             {
                 while (size-- > 0)
                 {
@@ -124,7 +126,7 @@ namespace Ryujinx.HLE.Utilities
                     ms.WriteByte(value);
                 }
 
-                return Encoding.UTF8.GetString(ms.ToArray());
+                return Encoding.UTF8.GetString(ms.GetReadOnlySequence());
             }
         }
 

Ryujinx.Input/Motion/CemuHook/Client.cs

@@ -3,6 +3,7 @@ using Ryujinx.Common.Configuration.Hid;
 using Ryujinx.Common.Configuration.Hid.Controller;
 using Ryujinx.Common.Configuration.Hid.Controller.Motion;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.Input.HLE;
 using Ryujinx.Input.Motion.CemuHook.Protocol;
 using System;
@@ -381,7 +382,7 @@ namespace Ryujinx.Input.Motion.CemuHook
 
             Header header = GenerateHeader(clientId);
 
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.WriteStruct(header);
@@ -421,7 +422,7 @@ namespace Ryujinx.Input.Motion.CemuHook
 
             Header header = GenerateHeader(clientId);
 
-            using (MemoryStream stream = new MemoryStream())
+            using (MemoryStream stream = MemoryStreamManager.Shared.GetStream())
             using (BinaryWriter writer = new BinaryWriter(stream))
             {
                 writer.WriteStruct(header);