Batch inline index buffer update (#4587)

Protection for the `xgetbv` instruction for systems that do not support
`xcr0` such as nehalem processors.

The `XSAVE` cpuid indicates support for `XSAVE`, `XRESTOR`, `XSETBV`,
`XGETBV` while `OSXSAVE` indicates if the operating system itself has
`XSAVE` turned on. Both must be checked at the same time.

.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/X86/Assembler.cs

@@ -1034,7 +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.
 
@@ -1153,6 +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;

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/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

@@ -30,7 +30,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 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";
@@ -969,6 +969,7 @@ namespace ARMeilleure.Translation.PTC
                     (ulong)Arm64HardwareCapabilities.LinuxFeatureInfoHwCap,
                     (ulong)Arm64HardwareCapabilities.LinuxFeatureInfoHwCap2,
                     (ulong)Arm64HardwareCapabilities.MacOsFeatureInfo,
+                    0,
                     0);
             }
             else if (RuntimeInformation.ProcessArchitecture == Architecture.X64)
@@ -977,11 +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);
             }
         }
 
@@ -1002,7 +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;
@@ -1034,8 +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

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/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.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);