Do not clear the rejit queue when overlaps count is equal to 0. (#3721)

* Do not clear the rejit queue when overlaps count is equal to 0.

* Ptc and PtcProfiler must be invalidated.

* Revert "Ptc and PtcProfiler must be invalidated."

This reverts commit f5b0ad9d7d.

* Fix #3710 slow path due to #3701.

ARMeilleure/Decoders/OpCode32SimdCvtTB.cs

@@ -0,0 +1,44 @@
+namespace ARMeilleure.Decoders
+{
+    class OpCode32SimdCvtTB : OpCode32, IOpCode32Simd
+    {
+        public int Vd { get; }
+        public int Vm { get; }
+        public bool Op { get; } // Convert to Half / Convert from Half
+        public bool T { get; } // Top / Bottom
+        public int Size { get; } // Double / Single
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCode32SimdCvtTB(inst, address, opCode, false);
+        public static OpCode CreateT32(InstDescriptor inst, ulong address, int opCode) => new OpCode32SimdCvtTB(inst, address, opCode, true);
+
+        public OpCode32SimdCvtTB(InstDescriptor inst, ulong address, int opCode, bool isThumb) : base(inst, address, opCode)
+        {
+            IsThumb = isThumb;
+
+            Op   = ((opCode >> 16) & 0x1) != 0;
+            T    = ((opCode >> 7) & 0x1) != 0;
+            Size = ((opCode >> 8) & 0x1);
+
+            RegisterSize = Size == 1 ? RegisterSize.Int64 : RegisterSize.Int32;
+
+            if (Size == 1)
+            {
+                if (Op)
+                {
+                    Vm = ((opCode >> 1) & 0x10) | ((opCode >> 0) & 0xf);
+                    Vd = ((opCode >> 22) & 0x1) | ((opCode >> 11) & 0x1e);
+                }
+                else
+                {
+                    Vm = ((opCode >> 5) & 0x1) | ((opCode << 1) & 0x1e);
+                    Vd = ((opCode >> 18) & 0x10) | ((opCode >> 12) & 0xf);
+                }
+            }
+            else
+            {
+                Vm = ((opCode >> 5) & 0x1) | ((opCode << 1) & 0x1e);
+                Vd = ((opCode >> 22) & 0x1) | ((opCode >> 11) & 0x1e);
+            }
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeTable.cs

@@ -828,6 +828,7 @@ namespace ARMeilleure.Decoders
             SetVfp("<<<<11101x11110xxxxx101x11x0xxxx", InstName.Vcvt,   InstEmit32.Vcvt_FI,  OpCode32SimdCvtFI.Create,       OpCode32SimdCvtFI.CreateT32); // FP32 to int.
             SetVfp("<<<<11101x111000xxxx101xx1x0xxxx", InstName.Vcvt,   InstEmit32.Vcvt_FI,  OpCode32SimdCvtFI.Create,       OpCode32SimdCvtFI.CreateT32); // Int to FP32.
             SetVfp("111111101x1111xxxxxx101xx1x0xxxx", InstName.Vcvt,   InstEmit32.Vcvt_RM,  OpCode32SimdCvtFI.Create,       OpCode32SimdCvtFI.CreateT32); // The many FP32 to int encodings (fp).
+            SetVfp("<<<<11101x11001xxxxx101xx1x0xxxx", InstName.Vcvt,   InstEmit32.Vcvt_TB,  OpCode32SimdCvtTB.Create,       OpCode32SimdCvtTB.CreateT32);
             SetVfp("<<<<11101x00xxxxxxxx101xx0x0xxxx", InstName.Vdiv,   InstEmit32.Vdiv_S,   OpCode32SimdRegS.Create,        OpCode32SimdRegS.CreateT32);
             SetVfp("<<<<11101xx0xxxxxxxx1011x0x10000", InstName.Vdup,   InstEmit32.Vdup,     OpCode32SimdDupGP.Create,       OpCode32SimdDupGP.CreateT32);
             SetVfp("<<<<11101x10xxxxxxxx101xx0x0xxxx", InstName.Vfma,   InstEmit32.Vfma_S,   OpCode32SimdRegS.Create,        OpCode32SimdRegS.CreateT32);

ARMeilleure/Instructions/InstEmitSimdArithmetic.cs

@@ -1616,20 +1616,34 @@ namespace ARMeilleure.Instructions
         }
 
         public static void Frinta_S(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41ScalarRoundOpF(context, FPRoundingMode.ToNearestAway);
+            }
+            else
             {
                 EmitScalarUnaryOpF(context, (op1) =>
                 {
                     return EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1);
                 });
             }
+        }
 
         public static void Frinta_V(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41VectorRoundOpF(context, FPRoundingMode.ToNearestAway);
+            }
+            else
             {
                 EmitVectorUnaryOpF(context, (op1) =>
                 {
                     return EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1);
                 });
             }
+        }
 
         public static void Frinti_S(ArmEmitterContext context)
         {
@@ -3516,9 +3530,18 @@ namespace ARMeilleure.Instructions
 
             Operand n = GetVec(op.Rn);
 
+            Operand res;
+
+            if (roundMode != FPRoundingMode.ToNearestAway)
+            {
                 Intrinsic inst = (op.Size & 1) != 0 ? Intrinsic.X86Roundsd : Intrinsic.X86Roundss;
 
-            Operand res = context.AddIntrinsic(inst, n, Const(X86GetRoundControl(roundMode)));
+                res = context.AddIntrinsic(inst, n, Const(X86GetRoundControl(roundMode)));
+            }
+            else
+            {
+                res = EmitSse41RoundToNearestWithTiesToAwayOpF(context, n, scalar: true);
+            }
 
             if ((op.Size & 1) != 0)
             {
@@ -3538,9 +3561,18 @@ namespace ARMeilleure.Instructions
 
             Operand n = GetVec(op.Rn);
 
+            Operand res;
+
+            if (roundMode != FPRoundingMode.ToNearestAway)
+            {
                 Intrinsic inst = (op.Size & 1) != 0 ? Intrinsic.X86Roundpd : Intrinsic.X86Roundps;
 
-            Operand res = context.AddIntrinsic(inst, n, Const(X86GetRoundControl(roundMode)));
+                res = context.AddIntrinsic(inst, n, Const(X86GetRoundControl(roundMode)));
+            }
+            else
+            {
+                res = EmitSse41RoundToNearestWithTiesToAwayOpF(context, n, scalar: false);
+            }
 
             if (op.RegisterSize == RegisterSize.Simd64)
             {

ARMeilleure/Instructions/InstEmitSimdCvt.cs

@@ -163,34 +163,76 @@ namespace ARMeilleure.Instructions
         }
 
         public static void Fcvtas_Gp(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41Fcvts_Gp(context, FPRoundingMode.ToNearestAway, isFixed: false);
+            }
+            else
             {
                 EmitFcvt_s_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
             }
+        }
 
         public static void Fcvtas_S(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41FcvtsOpF(context, FPRoundingMode.ToNearestAway, scalar: true);
+            }
+            else
             {
                 EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: true);
             }
+        }
 
         public static void Fcvtas_V(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41FcvtsOpF(context, FPRoundingMode.ToNearestAway, scalar: false);
+            }
+            else
             {
                 EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: true, scalar: false);
             }
+        }
 
         public static void Fcvtau_Gp(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41Fcvtu_Gp(context, FPRoundingMode.ToNearestAway, isFixed: false);
+            }
+            else
             {
                 EmitFcvt_u_Gp(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1));
             }
+        }
 
         public static void Fcvtau_S(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41FcvtuOpF(context, FPRoundingMode.ToNearestAway, scalar: true);
+            }
+            else
             {
                 EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: true);
             }
+        }
 
         public static void Fcvtau_V(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitSse41FcvtuOpF(context, FPRoundingMode.ToNearestAway, scalar: false);
+            }
+            else
             {
                 EmitFcvt(context, (op1) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, op1), signed: false, scalar: false);
             }
+        }
 
         public static void Fcvtl_V(ArmEmitterContext context)
         {
@@ -1223,7 +1265,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulps, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
+                }
 
                 Operand nInt = context.AddIntrinsic(Intrinsic.X86Cvtps2dq, nRes);
 
@@ -1265,7 +1314,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulpd, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
+                }
 
                 Operand nLong = EmitSse2CvtDoubleToInt64OpF(context, nRes, scalar);
 
@@ -1314,7 +1370,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulps, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundps, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
+                }
 
                 Operand zero = context.VectorZero();
 
@@ -1369,7 +1432,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulpd, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundpd, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar);
+                }
 
                 Operand zero = context.VectorZero();
 
@@ -1424,7 +1494,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulss, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand nIntOrLong = op.RegisterSize == RegisterSize.Int32
                     ? context.AddIntrinsicInt (Intrinsic.X86Cvtss2si, nRes)
@@ -1464,7 +1541,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulsd, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand nIntOrLong = op.RegisterSize == RegisterSize.Int32
                     ? context.AddIntrinsicInt (Intrinsic.X86Cvtsd2si, nRes)
@@ -1512,7 +1596,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulss, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand zero = context.VectorZero();
 
@@ -1567,7 +1658,14 @@ namespace ARMeilleure.Instructions
                     nRes = context.AddIntrinsic(Intrinsic.X86Mulsd, nRes, fpScaledMask);
                 }
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand zero = context.VectorZero();
 

ARMeilleure/Instructions/InstEmitSimdCvt32.cs

@@ -203,6 +203,9 @@ namespace ARMeilleure.Instructions
             FPRoundingMode roundMode;
             switch (rm)
             {
+                case 0b00:
+                    roundMode = FPRoundingMode.ToNearestAway;
+                    break;
                 case 0b01:
                     roundMode = FPRoundingMode.ToNearest;
                     break;
@@ -228,7 +231,7 @@ namespace ARMeilleure.Instructions
             bool unsigned = op.Opc == 0;
             int rm = op.Opc2 & 3;
 
-            if (Optimizations.UseSse41 && rm != 0b00)
+            if (Optimizations.UseSse41)
             {
                 EmitSse41ConvertInt32(context, RMToRoundMode(rm), !unsigned);
             }
@@ -258,6 +261,74 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        public static void Vcvt_TB(ArmEmitterContext context)
+        {
+            OpCode32SimdCvtTB op = (OpCode32SimdCvtTB)context.CurrOp;
+
+            if (Optimizations.UseF16c)
+            {
+                Debug.Assert(!Optimizations.ForceLegacySse);
+
+                if (op.Op)
+                {
+                    Operand res = ExtractScalar(context, op.Size == 1 ? OperandType.FP64 : OperandType.FP32, op.Vm);
+                    if (op.Size == 1)
+                    {
+                        res = context.AddIntrinsic(Intrinsic.X86Cvtsd2ss, context.VectorZero(), res);
+                    }
+                    res = context.AddIntrinsic(Intrinsic.X86Vcvtps2ph, res, Const(X86GetRoundControl(FPRoundingMode.ToNearest)));
+                    res = context.VectorExtract16(res, 0);
+                    InsertScalar16(context, op.Vd, op.T, res);
+                }
+                else
+                {
+                    Operand res = context.VectorCreateScalar(ExtractScalar16(context, op.Vm, op.T));
+                    res = context.AddIntrinsic(Intrinsic.X86Vcvtph2ps, res);
+                    if (op.Size == 1)
+                    {
+                        res = context.AddIntrinsic(Intrinsic.X86Cvtss2sd, context.VectorZero(), res);
+                    }
+                    res = context.VectorExtract(op.Size == 1 ? OperandType.I64 : OperandType.I32, res, 0);
+                    InsertScalar(context, op.Vd, res);
+                }
+            }
+            else
+            {
+                if (op.Op)
+                {
+                    // Convert to half.
+
+                    Operand src = ExtractScalar(context, op.Size == 1 ? OperandType.FP64 : OperandType.FP32, op.Vm);
+
+                    MethodInfo method = op.Size == 1
+                        ? typeof(SoftFloat64_16).GetMethod(nameof(SoftFloat64_16.FPConvert))
+                        : typeof(SoftFloat32_16).GetMethod(nameof(SoftFloat32_16.FPConvert));
+
+                    context.StoreToContext();
+                    Operand res = context.Call(method, src);
+                    context.LoadFromContext();
+
+                    InsertScalar16(context, op.Vd, op.T, res);
+                }
+                else
+                {
+                    // Convert from half.
+
+                    Operand src = ExtractScalar16(context, op.Vm, op.T);
+
+                    MethodInfo method = op.Size == 1
+                        ? typeof(SoftFloat16_64).GetMethod(nameof(SoftFloat16_64.FPConvert))
+                        : typeof(SoftFloat16_32).GetMethod(nameof(SoftFloat16_32.FPConvert));
+
+                    context.StoreToContext();
+                    Operand res = context.Call(method, src);
+                    context.LoadFromContext();
+
+                    InsertScalar(context, op.Vd, res);
+                }
+            }
+        }
+
         // VRINTA/M/N/P (floating-point).
         public static void Vrint_RM(ArmEmitterContext context)
         {
@@ -267,15 +338,21 @@ namespace ARMeilleure.Instructions
 
             int rm = op.Opc2 & 3;
 
-            if (Optimizations.UseSse2 && rm != 0b00)
+            if (Optimizations.UseSse41)
             {
                 EmitScalarUnaryOpSimd32(context, (m) =>
                 {
-                    Intrinsic inst = (op.Size & 1) == 0 ? Intrinsic.X86Roundss : Intrinsic.X86Roundsd;
-
                     FPRoundingMode roundMode = RMToRoundMode(rm);
 
+                    if (roundMode != FPRoundingMode.ToNearestAway)
+                    {
+                        Intrinsic inst = (op.Size & 1) == 0 ? Intrinsic.X86Roundss : Intrinsic.X86Roundsd;
                         return context.AddIntrinsic(inst, m, Const(X86GetRoundControl(roundMode)));
+                    }
+                    else
+                    {
+                        return EmitSse41RoundToNearestWithTiesToAwayOpF(context, m, scalar: true);
+                    }
                 });
             }
             else
@@ -304,9 +381,19 @@ namespace ARMeilleure.Instructions
 
         // VRINTA (vector).
         public static void Vrinta_V(ArmEmitterContext context)
+        {
+            if (Optimizations.UseSse41)
+            {
+                EmitVectorUnaryOpSimd32(context, (m) =>
+                {
+                    return EmitSse41RoundToNearestWithTiesToAwayOpF(context, m, scalar: false);
+                });
+            }
+            else
             {
                 EmitVectorUnaryOpF32(context, (m) => EmitRoundMathCall(context, MidpointRounding.AwayFromZero, m));
             }
+        }
 
         // VRINTM (vector).
         public static void Vrintm_V(ArmEmitterContext context)
@@ -413,7 +500,14 @@ namespace ARMeilleure.Instructions
                 Operand nRes = context.AddIntrinsic(Intrinsic.X86Cmpss, n, n, Const((int)CmpCondition.OrderedQ));
                 nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, n);
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand zero = context.VectorZero();
 
@@ -464,7 +558,14 @@ namespace ARMeilleure.Instructions
                 Operand nRes = context.AddIntrinsic(Intrinsic.X86Cmpsd, n, n, Const((int)CmpCondition.OrderedQ));
                 nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, n);
 
+                if (roundMode != FPRoundingMode.ToNearestAway)
+                {
                     nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
+                }
+                else
+                {
+                    nRes = EmitSse41RoundToNearestWithTiesToAwayOpF(context, nRes, scalar: true);
+                }
 
                 Operand zero = context.VectorZero();
 

ARMeilleure/Instructions/InstEmitSimdHelper.cs

@@ -33,6 +33,14 @@ namespace ARMeilleure.Instructions
         };
 
         public static readonly long ZeroMask = 128L << 56 | 128L << 48 | 128L << 40 | 128L << 32 | 128L << 24 | 128L << 16 | 128L << 8 | 128L << 0;
+
+        public static ulong X86GetGf2p8LogicalShiftLeft(int shift)
+        {
+            ulong identity = (0b00000001UL << 56) | (0b00000010UL << 48) | (0b00000100UL << 40) | (0b00001000UL << 32) |
+                             (0b00010000UL << 24) | (0b00100000UL << 16) | (0b01000000UL <<  8) | (0b10000000UL <<  0);
+
+            return shift >= 0 ? identity >> (shift * 8) : identity << (-shift * 8);
+        }
 #endregion
 
 #region "X86 SSE Intrinsics"
@@ -243,19 +251,44 @@ namespace ARMeilleure.Instructions
             throw new ArgumentException($"Invalid rounding mode \"{roundMode}\".");
         }
 
-        public static ulong X86GetGf2p8LogicalShiftLeft(int shift)
+        public static Operand EmitSse41RoundToNearestWithTiesToAwayOpF(ArmEmitterContext context, Operand n, bool scalar)
         {
-            ulong identity = 
+            Debug.Assert(n.Type == OperandType.V128);
-                (0b00000001UL << 56) |
-                (0b00000010UL << 48) |
-                (0b00000100UL << 40) |
-                (0b00001000UL << 32) |
-                (0b00010000UL << 24) |
-                (0b00100000UL << 16) |
-                (0b01000000UL <<  8) |
-                (0b10000000UL <<  0);
 
-            return shift >= 0 ? identity >> (shift * 8) : identity << (-shift * 8);
+            Operand nCopy = context.Copy(n);
+
+            Operand rC = Const(X86GetRoundControl(FPRoundingMode.TowardsZero));
+
+            IOpCodeSimd op = (IOpCodeSimd)context.CurrOp;
+
+            if ((op.Size & 1) == 0)
+            {
+                Operand signMask = scalar ? X86GetScalar(context, int.MinValue) : X86GetAllElements(context, int.MinValue);
+                        signMask = context.AddIntrinsic(Intrinsic.X86Pand, signMask, nCopy);
+
+                // 0x3EFFFFFF == BitConverter.SingleToInt32Bits(0.5f) - 1
+                Operand valueMask = scalar ? X86GetScalar(context, 0x3EFFFFFF) : X86GetAllElements(context, 0x3EFFFFFF);
+                        valueMask = context.AddIntrinsic(Intrinsic.X86Por, valueMask, signMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Addss : Intrinsic.X86Addps, nCopy, valueMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Roundss : Intrinsic.X86Roundps, nCopy, rC);
+            }
+            else
+            {
+                Operand signMask = scalar ? X86GetScalar(context, long.MinValue) : X86GetAllElements(context, long.MinValue);
+                        signMask = context.AddIntrinsic(Intrinsic.X86Pand, signMask, nCopy);
+
+                // 0x3FDFFFFFFFFFFFFFL == BitConverter.DoubleToInt64Bits(0.5d) - 1L
+                Operand valueMask = scalar ? X86GetScalar(context, 0x3FDFFFFFFFFFFFFFL) : X86GetAllElements(context, 0x3FDFFFFFFFFFFFFFL);
+                        valueMask = context.AddIntrinsic(Intrinsic.X86Por, valueMask, signMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Addsd : Intrinsic.X86Addpd, nCopy, valueMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Roundsd : Intrinsic.X86Roundpd, nCopy, rC);
+            }
+
+            return nCopy;
         }
 
         public static Operand EmitCountSetBits8(ArmEmitterContext context, Operand op) // "size" is 8 (SIMD&FP Inst.).

ARMeilleure/Instructions/InstEmitSimdHelper32.cs

@@ -70,6 +70,22 @@ namespace ARMeilleure.Instructions
             context.Copy(vec, insert);
         }
 
+        public static Operand ExtractScalar16(ArmEmitterContext context, int reg, bool top)
+        {
+            return context.VectorExtract16(GetVecA32(reg >> 2), ((reg & 3) << 1) | (top ? 1 : 0));
+        }
+
+        public static void InsertScalar16(ArmEmitterContext context, int reg, bool top, Operand value)
+        {
+            Debug.Assert(value.Type == OperandType.FP32 || value.Type == OperandType.I32);
+
+            Operand vec, insert;
+            vec = GetVecA32(reg >> 2);
+            insert = context.VectorInsert16(vec, value, ((reg & 3) << 1) | (top ? 1 : 0));
+
+            context.Copy(vec, insert);
+        }
+
         public static Operand ExtractElement(ArmEmitterContext context, int reg, int size, bool signed)
         {
             return EmitVectorExtract32(context, reg >> (4 - size), reg & ((16 >> size) - 1), size, signed);

ARMeilleure/State/FPRoundingMode.cs

@@ -2,9 +2,10 @@ namespace ARMeilleure.State
 {
     public enum FPRoundingMode
     {
-        ToNearest            = 0,
+        ToNearest            = 0, // With ties to even.
         TowardsPlusInfinity  = 1,
         TowardsMinusInfinity = 2,
-        TowardsZero          = 3
+        TowardsZero          = 3,
+        ToNearestAway        = 4 // With ties to away.
     }
 }

ARMeilleure/Translation/PTC/Ptc.cs

@@ -27,7 +27,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 3710; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 3713; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

ARMeilleure/Translation/Translator.cs

@@ -455,13 +455,16 @@ namespace ARMeilleure.Translation
 
         public void InvalidateJitCacheRegion(ulong address, ulong size)
         {
-            // If rejit is running, stop it as it may be trying to rejit a function on the invalidated region.
-            ClearRejitQueue(allowRequeue: true);
-
             ulong[] overlapAddresses = Array.Empty<ulong>();
 
             int overlapsCount = Functions.GetOverlaps(address, size, ref overlapAddresses);
 
+            if (overlapsCount != 0)
+            {
+                // If rejit is running, stop it as it may be trying to rejit a function on the invalidated region.
+                ClearRejitQueue(allowRequeue: true);
+            }
+
             for (int index = 0; index < overlapsCount; index++)
             {
                 ulong overlapAddress = overlapAddresses[index];

README.md

@@ -36,7 +36,7 @@
 
 ## Compatibility
 
-As of September 2022, Ryujinx has been tested on approximately 3,600 titles; over 3,400 boot past menus and into gameplay, with roughly 2,700 of those being considered playable. You can check out the compatibility list [here](https://github.com/Ryujinx/Ryujinx-Games-List/issues).  
+As of October 2022, Ryujinx has been tested on approximately 3,600 titles; over 3,500 boot past menus and into gameplay, with roughly 3,000 of those being considered playable. You can check out the compatibility list [here](https://github.com/Ryujinx/Ryujinx-Games-List/issues).  
 Anyone is free to submit a new game test or update an existing game test entry; simply follow the new issue template and testing guidelines, or post as a reply to the applicable game issue. Use the search function to see if a game has been tested already!
 
 ## Usage

Ryujinx.Ava/Assets/Locales/pl_PL.json

@@ -588,5 +588,9 @@
   "SettingsTabGraphicsPreferredGpuTooltip": "Wybierz kartę graficzną, która będzie używana z backendem graficznym Vulkan.\n\nNie wpływa na GPU używane przez OpenGL.\n\nW razie wątpliwości ustaw flagę GPU jako \"dGPU\". Jeśli żadnej nie ma, pozostaw nietknięte.",
   "SettingsAppRequiredRestartMessage": "Wymagane Zrestartowanie Ryujinx",
   "SettingsGpuBackendRestartMessage": "Zmieniono ustawienia Backendu Graficznego lub GPU. Będzie to wymagało ponownego uruchomienia",
-  "SettingsGpuBackendRestartSubMessage": "Czy chcesz zrestartować teraz?"
+  "SettingsGpuBackendRestartSubMessage": "Czy chcesz zrestartować teraz?",
+  "RyujinxUpdaterMessage": "Czy chcesz zaktualizować Ryujinx do najnowszej wersji?",
+  "SettingsTabHotkeysVolumeUpHotkey": "Zwiększ Głośność:",
+  "SettingsTabHotkeysVolumeDownHotkey": "Zmniejsz Głośność:",
+  "VolumeShort": "Głoś"
 }

Ryujinx.Ava/Modules/Updater/Updater.cs

@@ -278,7 +278,7 @@ namespace Ryujinx.Modules
                 {
                     string ryuName = Path.GetFileName(Environment.ProcessPath);
                     string ryuExe = Path.Combine(AppDomain.CurrentDomain.BaseDirectory, ryuName);
-                    string ryuArg = string.Join(" ", Environment.GetCommandLineArgs().Skip(1).ToArray());
+                    var ryuArg = Environment.GetCommandLineArgs().Skip(1);
 
                     if (!OperatingSystem.IsWindows())
                     {

Ryujinx.Ava/Ui/ViewModels/SettingsViewModel.cs

@@ -295,8 +295,6 @@ namespace Ryujinx.Ava.Ui.ViewModels
             if (_validTzRegions.Contains(location))
             {
                 TimeZone = location;
-
-                OnPropertyChanged(nameof(TimeZone));
             }
         }
 

Ryujinx.Ava/Ui/Windows/MainWindow.axaml

@@ -164,7 +164,7 @@
                                 <MenuItem
                                     Command="{ReflectionBinding ChangeLanguage}"
                                     CommandParameter="pl_PL"
-                                    Header="Polish" />
+                                    Header="Polski" />
                                 <MenuItem
                                     Command="{ReflectionBinding ChangeLanguage}"
                                     CommandParameter="ru_RU"

Ryujinx.Ava/Ui/Windows/SettingsWindow.axaml.cs

@@ -37,7 +37,7 @@ namespace Ryujinx.Ava.Ui.Windows
             InitializeComponent();
             Load();
 
-            FuncMultiValueConverter<string, string> converter = new(parts => string.Format("{0}  {1}   {2}", parts.ToArray()));
+            FuncMultiValueConverter<string, string> converter = new(parts => string.Format("{0}  {1}   {2}", parts.ToArray()).Trim());
             MultiBinding tzMultiBinding = new() { Converter = converter };
             tzMultiBinding.Bindings.Add(new Binding("UtcDifference"));
             tzMultiBinding.Bindings.Add(new Binding("Location"));

Ryujinx.Common/Memory/SpanOrArray.cs

@@ -0,0 +1,89 @@
+using System;
+
+namespace Ryujinx.Common.Memory
+{
+    /// <summary>
+    /// A struct that can represent both a Span and Array.
+    /// This is useful to keep the Array representation when possible to avoid copies.
+    /// </summary>
+    /// <typeparam name="T">Element Type</typeparam>
+    public ref struct SpanOrArray<T> where T : unmanaged
+    {
+        public readonly T[] Array;
+        public readonly ReadOnlySpan<T> Span;
+
+        /// <summary>
+        /// Create a new SpanOrArray from an array.
+        /// </summary>
+        /// <param name="array">Array to store</param>
+        public SpanOrArray(T[] array)
+        {
+            Array = array;
+            Span = ReadOnlySpan<T>.Empty;
+        }
+
+        /// <summary>
+        /// Create a new SpanOrArray from a readonly span.
+        /// </summary>
+        /// <param name="array">Span to store</param>
+        public SpanOrArray(ReadOnlySpan<T> span)
+        {
+            Array = null;
+            Span = span;
+        }
+
+        /// <summary>
+        /// Return the contained array, or convert the span if necessary.
+        /// </summary>
+        /// <returns>An array containing the data</returns>
+        public T[] ToArray()
+        {
+            return Array ?? Span.ToArray();
+        }
+
+        /// <summary>
+        /// Return a ReadOnlySpan from either the array or ReadOnlySpan.
+        /// </summary>
+        /// <returns>A ReadOnlySpan containing the data</returns>
+        public ReadOnlySpan<T> AsSpan()
+        {
+            return Array ?? Span;
+        }
+
+        /// <summary>
+        /// Cast an array to a SpanOrArray.
+        /// </summary>
+        /// <param name="array">Source array</param>
+        public static implicit operator SpanOrArray<T>(T[] array)
+        {
+            return new SpanOrArray<T>(array);
+        }
+
+        /// <summary>
+        /// Cast a ReadOnlySpan to a SpanOrArray.
+        /// </summary>
+        /// <param name="span">Source ReadOnlySpan</param>
+        public static implicit operator SpanOrArray<T>(ReadOnlySpan<T> span)
+        {
+            return new SpanOrArray<T>(span);
+        }
+
+        /// <summary>
+        /// Cast a Span to a SpanOrArray.
+        /// </summary>
+        /// <param name="span">Source Span</param>
+        public static implicit operator SpanOrArray<T>(Span<T> span)
+        {
+            return new SpanOrArray<T>(span);
+        }
+
+        /// <summary>
+        /// Cast a SpanOrArray to a ReadOnlySpan
+        /// </summary>
+        /// <param name="spanOrArray">Source SpanOrArray</param>
+        public static implicit operator ReadOnlySpan<T>(SpanOrArray<T> spanOrArray)
+        {
+            return spanOrArray.AsSpan();
+        }
+    }
+}

Ryujinx.Common/Utilities/EmbeddedResources.cs

@@ -129,7 +129,7 @@ namespace Ryujinx.Common
 
         private static (Assembly, string) ResolveManifestPath(string filename)
         {
-            var segments = filename.Split(new[] { '/' }, 2, StringSplitOptions.RemoveEmptyEntries);
+            var segments = filename.Split('/', 2, StringSplitOptions.RemoveEmptyEntries);
 
             if (segments.Length >= 2)
             {

Ryujinx.Graphics.GAL/ITexture.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Common.Memory;
 using System;
 
 namespace Ryujinx.Graphics.GAL
@@ -17,9 +18,9 @@ namespace Ryujinx.Graphics.GAL
         ReadOnlySpan<byte> GetData();
         ReadOnlySpan<byte> GetData(int layer, int level);
 
-        void SetData(ReadOnlySpan<byte> data);
+        void SetData(SpanOrArray<byte> data);
-        void SetData(ReadOnlySpan<byte> data, int layer, int level);
+        void SetData(SpanOrArray<byte> data, int layer, int level);
-        void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region);
+        void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region);
         void SetStorage(BufferRange buffer);
         void Release();
     }

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

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL.Multithreading.Commands.Texture;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL.Multithreading.Commands.Texture;
 using Ryujinx.Graphics.GAL.Multithreading.Model;
 using System;
 
@@ -107,19 +108,19 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             _renderer.New<TextureSetDataCommand>().Set(Ref(this), Ref(data.ToArray()));
             _renderer.QueueCommand();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             _renderer.New<TextureSetDataSliceCommand>().Set(Ref(this), Ref(data.ToArray()), layer, level);
             _renderer.QueueCommand();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             _renderer.New<TextureSetDataSliceRegionCommand>().Set(Ref(this), Ref(data.ToArray()), layer, level, region);
             _renderer.QueueCommand();

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

@@ -229,7 +229,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
 
                     if (target != null)
                     {
-                        ReadOnlySpan<byte> data;
+                        byte[] data;
                         if (srcLinear)
                         {
                             data = LayoutConverter.ConvertLinearStridedToLinear(

Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Texture;
@@ -136,11 +137,6 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         public LinkedListNode<Texture> CacheNode { get; set; }
 
-        /// <summary>
-        /// Event to fire when texture data is disposed.
-        /// </summary>
-        public event Action<Texture> Disposed;
-
         /// <summary>
         /// Physical memory ranges where the texture data is located.
         /// </summary>
@@ -720,9 +716,9 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
             }
 
-            data = ConvertToHostCompatibleFormat(data);
+            SpanOrArray<byte> result = ConvertToHostCompatibleFormat(data);
 
-            HostTexture.SetData(data);
+            HostTexture.SetData(result);
 
             _hasData = true;
         }
@@ -731,7 +727,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Uploads new texture data to the host GPU.
         /// </summary>
         /// <param name="data">New data</param>
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             BlacklistScale();
 
@@ -750,7 +746,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="data">New data</param>
         /// <param name="layer">Target layer</param>
         /// <param name="level">Target level</param>
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             BlacklistScale();
 
@@ -786,7 +782,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="level">Mip level to convert</param>
         /// <param name="single">True to convert a single slice</param>
         /// <returns>Converted data</returns>
-        public ReadOnlySpan<byte> ConvertToHostCompatibleFormat(ReadOnlySpan<byte> data, int level = 0, bool single = false)
+        public SpanOrArray<byte> ConvertToHostCompatibleFormat(ReadOnlySpan<byte> data, int level = 0, bool single = false)
         {
             int width = Info.Width;
             int height = Info.Height;
@@ -799,9 +795,11 @@ namespace Ryujinx.Graphics.Gpu.Image
             height = Math.Max(height >> level, 1);
             depth = Math.Max(depth >> level, 1);
 
+            SpanOrArray<byte> result;
+
             if (Info.IsLinear)
             {
-                data = LayoutConverter.ConvertLinearStridedToLinear(
+                result = LayoutConverter.ConvertLinearStridedToLinear(
                     width,
                     height,
                     Info.FormatInfo.BlockWidth,
@@ -813,7 +811,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
             else
             {
-                data = LayoutConverter.ConvertBlockLinearToLinear(
+                result = LayoutConverter.ConvertBlockLinearToLinear(
                     width,
                     height,
                     depth,
@@ -836,7 +834,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             if (!_context.Capabilities.SupportsAstcCompression && Format.IsAstc())
             {
                 if (!AstcDecoder.TryDecodeToRgba8P(
-                    data.ToArray(),
+                    result.ToArray(),
                     Info.FormatInfo.BlockWidth,
                     Info.FormatInfo.BlockHeight,
                     width,
@@ -856,11 +854,11 @@ namespace Ryujinx.Graphics.Gpu.Image
                     decoded = BCnEncoder.EncodeBC7(decoded, width, height, depth, levels, layers);
                 }
 
-                data = decoded;
+                result = decoded;
             }
             else if (!_context.Capabilities.SupportsR4G4Format && Format == Format.R4G4Unorm)
             {
-                data = PixelConverter.ConvertR4G4ToR4G4B4A4(data);
+                result = PixelConverter.ConvertR4G4ToR4G4B4A4(result);
             }
             else if (!TextureCompatibility.HostSupportsBcFormat(Format, Target, _context.Capabilities))
             {
@@ -868,36 +866,36 @@ namespace Ryujinx.Graphics.Gpu.Image
                 {
                     case Format.Bc1RgbaSrgb:
                     case Format.Bc1RgbaUnorm:
-                        data = BCnDecoder.DecodeBC1(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC1(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc2Srgb:
                     case Format.Bc2Unorm:
-                        data = BCnDecoder.DecodeBC2(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC2(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc3Srgb:
                     case Format.Bc3Unorm:
-                        data = BCnDecoder.DecodeBC3(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC3(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc4Snorm:
                     case Format.Bc4Unorm:
-                        data = BCnDecoder.DecodeBC4(data, width, height, depth, levels, layers, Format == Format.Bc4Snorm);
+                        result = BCnDecoder.DecodeBC4(result, width, height, depth, levels, layers, Format == Format.Bc4Snorm);
                         break;
                     case Format.Bc5Snorm:
                     case Format.Bc5Unorm:
-                        data = BCnDecoder.DecodeBC5(data, width, height, depth, levels, layers, Format == Format.Bc5Snorm);
+                        result = BCnDecoder.DecodeBC5(result, width, height, depth, levels, layers, Format == Format.Bc5Snorm);
                         break;
                     case Format.Bc6HSfloat:
                     case Format.Bc6HUfloat:
-                        data = BCnDecoder.DecodeBC6(data, width, height, depth, levels, layers, Format == Format.Bc6HSfloat);
+                        result = BCnDecoder.DecodeBC6(result, width, height, depth, levels, layers, Format == Format.Bc6HSfloat);
                         break;
                     case Format.Bc7Srgb:
                     case Format.Bc7Unorm:
-                        data = BCnDecoder.DecodeBC7(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC7(result, width, height, depth, levels, layers);
                         break;
                 }
             }
 
-            return data;
+            return result;
         }
 
         /// <summary>
@@ -1445,7 +1443,6 @@ namespace Ryujinx.Graphics.Gpu.Image
             DisposeTextures();
 
             HostTexture = hostTexture;
-            InvalidatedSequence++;
         }
 
         /// <summary>
@@ -1600,6 +1597,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// </summary>
         private void DisposeTextures()
         {
+            InvalidatedSequence++;
+
             _currentData = null;
             HostTexture.Release();
 
@@ -1634,8 +1633,6 @@ namespace Ryujinx.Graphics.Gpu.Image
         {
             DisposeTextures();
 
-            Disposed?.Invoke(this);
-
             if (Group.Storage == this)
             {
                 Group.Dispose();

Ryujinx.Graphics.Gpu/Image/TextureGroup.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Cpu.Tracking;
+using Ryujinx.Common.Memory;
+using Ryujinx.Cpu.Tracking;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Texture;
@@ -348,9 +349,9 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                             ReadOnlySpan<byte> data = _physicalMemory.GetSpan(Storage.Range.GetSlice((ulong)offset, (ulong)size));
 
-                            data = Storage.ConvertToHostCompatibleFormat(data, info.BaseLevel, true);
+                            SpanOrArray<byte> result = Storage.ConvertToHostCompatibleFormat(data, info.BaseLevel, true);
 
-                            Storage.SetData(data, info.BaseLayer, info.BaseLevel);
+                            Storage.SetData(result, info.BaseLayer, info.BaseLevel);
 
                             offsetIndex++;
                         }

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

@@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         private const ushort FileFormatVersionMajor = 1;
         private const ushort FileFormatVersionMinor = 2;
         private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
-        private const uint CodeGenVersion = 3732;
+        private const uint CodeGenVersion = 3759;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

Ryujinx.Graphics.OpenGL/Image/TextureBuffer.cs

@@ -1,4 +1,5 @@
 using OpenTK.Graphics.OpenGL;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using System;
 
@@ -48,17 +49,19 @@ namespace Ryujinx.Graphics.OpenGL.Image
             return GetData();
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
-            Buffer.SetData(_buffer, _bufferOffset, data.Slice(0, Math.Min(data.Length, _bufferSize)));
+            var dataSpan = data.AsSpan();
+
+            Buffer.SetData(_buffer, _bufferOffset, dataSpan.Slice(0, Math.Min(dataSpan.Length, _bufferSize)));
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             throw new NotSupportedException();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             throw new NotSupportedException();
         }

Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -1,5 +1,6 @@
 using OpenTK.Graphics.OpenGL;
 using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using System;
 
@@ -317,32 +318,36 @@ namespace Ryujinx.Graphics.OpenGL.Image
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
-                    ReadFrom((IntPtr)ptr, data.Length);
+                    ReadFrom((IntPtr)ptr, dataSpan.Length);
                 }
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
                     int width = Math.Max(Info.Width >> level, 1);
                     int height = Math.Max(Info.Height >> level, 1);
@@ -352,11 +357,13 @@ namespace Ryujinx.Graphics.OpenGL.Image
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             int wInBlocks = BitUtils.DivRoundUp(region.Width, Info.BlockWidth);
@@ -364,7 +371,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
                     ReadFrom2D(
                         (IntPtr)ptr,

Ryujinx.Graphics.Shader/Decoders/Decoder.cs

@@ -377,6 +377,8 @@ namespace Ryujinx.Graphics.Shader.Decoders
 
                 if (lastOp.Name == InstName.Brx && block.Successors.Count == (hasNext ? 1 : 0))
                 {
+                    HashSet<ulong> visited = new HashSet<ulong>();
+
                     InstBrx opBrx = new InstBrx(lastOp.RawOpCode);
                     ulong baseOffset = lastOp.GetAbsoluteAddress();
 
@@ -392,12 +394,17 @@ namespace Ryujinx.Graphics.Shader.Decoders
                     for (int i = 0; i < cbOffsetsCount; i++)
                     {
                         uint targetOffset = config.ConstantBuffer1Read(cbBaseOffset + i * 4);
-                        Block target = getBlock(baseOffset + targetOffset);
+                        ulong targetAddress = baseOffset + targetOffset;
+
+                        if (visited.Add(targetAddress))
+                        {
+                            Block target = getBlock(targetAddress);
                             target.Predecessors.Add(block);
                             block.Successors.Add(target);
                         }
                     }
                 }
+            }
 
             return hasNewTarget;
         }

Ryujinx.Graphics.Shader/Instructions/InstEmitFlowControl.cs

@@ -41,20 +41,77 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             Operand address = context.IAdd(Register(op.SrcA, RegisterType.Gpr), Const(offset));
 
-            // Sorting the target addresses in descending order improves the code,
+            var targets = context.CurrBlock.Successors.Skip(startIndex);
-            // since it will always check the most distant targets first, then the
-            // near ones. This can be easily transformed into if/else statements.
-            var sortedTargets = context.CurrBlock.Successors.Skip(startIndex).OrderByDescending(x => x.Address);
 
-            Block lastTarget = sortedTargets.LastOrDefault();
+            bool allTargetsSinglePred = true;
+            int total = context.CurrBlock.Successors.Count - startIndex;
+            int count = 0;
 
-            foreach (Block possibleTarget in sortedTargets)
+            foreach (var target in targets.OrderBy(x => x.Address))
             {
-                Operand label = context.GetLabel(possibleTarget.Address);
+                if (++count < total && (target.Predecessors.Count > 1 || target.Address <= context.CurrBlock.Address))
-
-                if (possibleTarget != lastTarget)
                 {
-                    context.BranchIfTrue(label, context.ICompareEqual(address, Const((int)possibleTarget.Address)));
+                    allTargetsSinglePred = false;
+                    break;
+                }
+            }
+
+            if (allTargetsSinglePred)
+            {
+                // Chain blocks, each target block will check if the BRX target address
+                // matches its own address, if not, it jumps to the next target which will do the same check,
+                // until it reaches the last possible target, which executed unconditionally.
+                // We can only do this if the BRX block is the only predecessor of all target blocks.
+                // Additionally, this is not supported for blocks located before the current block,
+                // since it will be too late to insert a label, but this is something that can be improved
+                // in the future if necessary.
+
+                var sortedTargets = targets.OrderBy(x => x.Address);
+
+                Block currentTarget = null;
+                ulong firstTargetAddress = 0;
+
+                foreach (Block nextTarget in sortedTargets)
+                {
+                    if (currentTarget != null)
+                    {
+                        if (currentTarget.Address != nextTarget.Address)
+                        {
+                            context.SetBrxTarget(currentTarget.Address, address, (int)currentTarget.Address, nextTarget.Address);
+                        }
+                    }
+                    else
+                    {
+                        firstTargetAddress = nextTarget.Address;
+                    }
+
+                    currentTarget = nextTarget;
+                }
+
+                context.Branch(context.GetLabel(firstTargetAddress));
+            }
+            else
+            {
+                // Emit the branches sequentially.
+                // This generates slightly worse code, but should work for all cases.
+
+                var sortedTargets = targets.OrderByDescending(x => x.Address);
+                ulong lastTargetAddress = ulong.MaxValue;
+
+                count = 0;
+
+                foreach (Block target in sortedTargets)
+                {
+                    Operand label = context.GetLabel(target.Address);
+
+                    if (++count < total)
+                    {
+                        if (target.Address != lastTargetAddress)
+                        {
+                            context.BranchIfTrue(label, context.ICompareEqual(address, Const((int)target.Address)));
+                        }
+
+                        lastTargetAddress = target.Address;
                     }
                     else
                     {
@@ -62,6 +119,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                     }
                 }
             }
+        }
 
         public static void Cal(EmitterContext context)
         {

Ryujinx.Graphics.Shader/Translation/EmitterContext.cs

@@ -21,8 +21,33 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public int OperationsCount => _operations.Count;
 
+        private struct BrxTarget
+        {
+            public readonly Operand Selector;
+            public readonly int ExpectedValue;
+            public readonly ulong NextTargetAddress;
+
+            public BrxTarget(Operand selector, int expectedValue, ulong nextTargetAddress)
+            {
+                Selector = selector;
+                ExpectedValue = expectedValue;
+                NextTargetAddress = nextTargetAddress;
+            }
+        }
+
+        private class BlockLabel
+        {
+            public readonly Operand Label;
+            public BrxTarget BrxTarget;
+
+            public BlockLabel(Operand label)
+            {
+                Label = label;
+            }
+        }
+
         private readonly List<Operation> _operations;
-        private readonly Dictionary<ulong, Operand> _labels;
+        private readonly Dictionary<ulong, BlockLabel> _labels;
 
         public EmitterContext(DecodedProgram program, ShaderConfig config, bool isNonMain)
         {
@@ -30,7 +55,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             Config = config;
             IsNonMain = isNonMain;
             _operations = new List<Operation>();
-            _labels = new Dictionary<ulong, Operand>();
+            _labels = new Dictionary<ulong, BlockLabel>();
 
             EmitStart();
         }
@@ -158,14 +183,40 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public Operand GetLabel(ulong address)
         {
-            if (!_labels.TryGetValue(address, out Operand label))
+            return EnsureBlockLabel(address).Label;
-            {
-                label = Label();
-
-                _labels.Add(address, label);
         }
 
-            return label;
+        public void SetBrxTarget(ulong address, Operand selector, int targetValue, ulong nextTargetAddress)
+        {
+            BlockLabel blockLabel = EnsureBlockLabel(address);
+            Debug.Assert(blockLabel.BrxTarget.Selector == null);
+            blockLabel.BrxTarget = new BrxTarget(selector, targetValue, nextTargetAddress);
+        }
+
+        public void EnterBlock(ulong address)
+        {
+            BlockLabel blockLabel = EnsureBlockLabel(address);
+
+            MarkLabel(blockLabel.Label);
+
+            BrxTarget brxTarget = blockLabel.BrxTarget;
+
+            if (brxTarget.Selector != null)
+            {
+                this.BranchIfFalse(GetLabel(brxTarget.NextTargetAddress), this.ICompareEqual(brxTarget.Selector, Const(brxTarget.ExpectedValue)));
+            }
+        }
+
+        private BlockLabel EnsureBlockLabel(ulong address)
+        {
+            if (!_labels.TryGetValue(address, out BlockLabel blockLabel))
+            {
+                blockLabel = new BlockLabel(Label());
+
+                _labels.Add(address, blockLabel);
+            }
+
+            return blockLabel;
         }
 
         public void PrepareForVertexReturn()

Ryujinx.Graphics.Shader/Translation/Translator.cs

@@ -162,7 +162,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 {
                     context.CurrBlock = block;
 
-                    context.MarkLabel(context.GetLabel(block.Address));
+                    context.EnterBlock(block.Address);
 
                     EmitOps(context, block);
                 }

Ryujinx.Graphics.Texture/LayoutConverter.cs

@@ -93,7 +93,7 @@ namespace Ryujinx.Graphics.Texture
             };
         }
 
-        public static Span<byte> ConvertBlockLinearToLinear(
+        public static byte[] ConvertBlockLinearToLinear(
             int width,
             int height,
             int depth,
@@ -119,7 +119,7 @@ namespace Ryujinx.Graphics.Texture
                 blockHeight,
                 bytesPerPixel);
 
-            Span<byte> output = new byte[outSize];
+            byte[] output = new byte[outSize];
 
             int outOffs = 0;
 
@@ -246,7 +246,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static Span<byte> ConvertLinearStridedToLinear(
+        public static byte[] ConvertLinearStridedToLinear(
             int width,
             int height,
             int blockWidth,
@@ -262,14 +262,15 @@ namespace Ryujinx.Graphics.Texture
             int outStride = BitUtils.AlignUp(w * bytesPerPixel, HostStrideAlignment);
             lineSize = Math.Min(lineSize, outStride);
 
-            Span<byte> output = new byte[h * outStride];
+            byte[] output = new byte[h * outStride];
+            Span<byte> outSpan = output;
 
             int outOffs = 0;
             int inOffs = 0;
 
             for (int y = 0; y < h; y++)
             {
-                data.Slice(inOffs, lineSize).CopyTo(output.Slice(outOffs, lineSize));
+                data.Slice(inOffs, lineSize).CopyTo(outSpan.Slice(outOffs, lineSize));
 
                 inOffs += stride;
                 outOffs += outStride;

Ryujinx.Graphics.Vulkan/BufferHolder.cs

@@ -386,8 +386,25 @@ namespace Ryujinx.Graphics.Vulkan
             _waitable.WaitForFences(_gd.Api, _device, offset, size);
         }
 
+        private bool BoundToRange(int offset, ref int size)
+        {
+            if (offset >= Size)
+            {
+                return false;
+            }
+
+            size = Math.Min(Size - offset, size);
+
+            return true;
+        }
+
         public Auto<DisposableBuffer> GetBufferI8ToI16(CommandBufferScoped cbs, int offset, int size)
         {
+            if (!BoundToRange(offset, ref size))
+            {
+                return null;
+            }
+
             var key = new I8ToI16CacheKey(_gd);
 
             if (!_cachedConvertedBuffers.TryGetValue(offset, size, key, out var holder))
@@ -407,6 +424,11 @@ namespace Ryujinx.Graphics.Vulkan
 
         public Auto<DisposableBuffer> GetAlignedVertexBuffer(CommandBufferScoped cbs, int offset, int size, int stride, int alignment)
         {
+            if (!BoundToRange(offset, ref size))
+            {
+                return null;
+            }
+
             var key = new AlignedVertexBufferCacheKey(_gd, stride, alignment);
 
             if (!_cachedConvertedBuffers.TryGetValue(offset, size, key, out var holder))
@@ -428,6 +450,11 @@ namespace Ryujinx.Graphics.Vulkan
 
         public Auto<DisposableBuffer> GetBufferTopologyConversion(CommandBufferScoped cbs, int offset, int size, IndexBufferPattern pattern, int indexSize)
         {
+            if (!BoundToRange(offset, ref size))
+            {
+                return null;
+            }
+
             var key = new TopologyConversionCacheKey(_gd, pattern, indexSize);
 
             if (!_cachedConvertedBuffers.TryGetValue(offset, size, key, out var holder))

Ryujinx.Graphics.Vulkan/IndexBufferPattern.cs

@@ -38,7 +38,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         public int GetPrimitiveCount(int vertexCount)
         {
-            return Math.Max(0, ((vertexCount - BaseIndex) + IndexStride - 1) / IndexStride);
+            return Math.Max(0, (vertexCount - BaseIndex) / IndexStride);
         }
 
         public int GetConvertedCount(int indexCount)

Ryujinx.Graphics.Vulkan/IndexBufferState.cs

@@ -49,7 +49,12 @@ namespace Ryujinx.Graphics.Vulkan
             }
             else
             {
-                autoBuffer = gd.BufferManager.GetBuffer(cbs.CommandBuffer, _handle, false, out int _);
+                autoBuffer = gd.BufferManager.GetBuffer(cbs.CommandBuffer, _handle, false, out int bufferSize);
+
+                if (_offset >= bufferSize)
+                {
+                    autoBuffer = null;
+                }
 
                 offset = _offset;
                 size = _size;

Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -95,7 +95,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             using var emptyVb = gd.BufferManager.Create(gd, EmptyVbSize);
             emptyVb.SetData(0, new byte[EmptyVbSize]);
-            _vertexBuffers[0] = new VertexBufferState(emptyVb.GetBuffer(), 0, EmptyVbSize, 0);
+            _vertexBuffers[0] = new VertexBufferState(emptyVb.GetBuffer(), 0, 0, EmptyVbSize, 0);
             _vertexBuffersDirty = ulong.MaxValue >> (64 - _vertexBuffers.Length);
 
             ClearScissor = new Rectangle<int>(0, 0, 0xffff, 0xffff);
@@ -1243,7 +1243,7 @@ namespace Ryujinx.Graphics.Vulkan
 
                     _vertexBuffers[i].BindVertexBuffer(Gd, Cbs, (uint)i, ref _newState);
 
-                    _vertexBuffersDirty &= ~(1u << i);
+                    _vertexBuffersDirty &= ~(1UL << i);
                 }
             }
 

Ryujinx.Graphics.Vulkan/SamplerHolder.cs

@@ -64,6 +64,7 @@ namespace Ryujinx.Graphics.Vulkan
                 };
 
                 samplerCreateInfo.PNext = &customBorderColor;
+                samplerCreateInfo.BorderColor = BorderColor.FloatCustomExt;
             }
 
             gd.Api.CreateSampler(device, samplerCreateInfo, null, out var sampler).ThrowOnError();

Ryujinx.Graphics.Vulkan/TextureBuffer.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -90,17 +91,17 @@ namespace Ryujinx.Graphics.Vulkan
             _bufferView = null;
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             _gd.SetBufferData(_bufferHandle, _offset, data);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             throw new NotSupportedException();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             throw new NotSupportedException();
         }

Ryujinx.Graphics.Vulkan/TextureStorage.cs

@@ -480,6 +480,8 @@ namespace Ryujinx.Graphics.Vulkan
             if (--_viewsCount == 0)
             {
                 _gd.PipelineInternal?.FlushCommandsIfWeightExceeding(_imageAuto, _size);
+
+                Dispose();
             }
         }
 

Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -514,10 +515,10 @@ namespace Ryujinx.Graphics.Vulkan
                 dst.Info,
                 srcRegion,
                 dstRegion,
-                src.FirstLevel,
-                dst.FirstLevel,
                 src.FirstLayer,
                 dst.FirstLayer,
+                src.FirstLevel,
+                dst.FirstLevel,
                 layers,
                 levels,
                 linearFilter,
@@ -873,17 +874,17 @@ namespace Ryujinx.Graphics.Vulkan
             return GetDataFromBuffer(result, size, result);
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             SetData(data, 0, 0, Info.GetLayers(), Info.Levels, singleSlice: false);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             SetData(data, layer, level, 1, 1, singleSlice: true);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             SetData(data, layer, level, 1, 1, singleSlice: true, region);
         }

Ryujinx.Graphics.Vulkan/VertexBufferState.cs

@@ -57,9 +57,13 @@ namespace Ryujinx.Graphics.Vulkan
                 if (gd.NeedsVertexBufferAlignment(AttributeScalarAlignment, out int alignment) && (_stride % alignment) != 0)
                 {
                     autoBuffer = gd.BufferManager.GetAlignedVertexBuffer(cbs, _handle, _offset, _size, _stride, alignment);
-                    int stride = (_stride + (alignment - 1)) & -alignment;
 
-                    var buffer = autoBuffer.Get(cbs, _offset, _size).Value;
+                    if (autoBuffer != null)
+                    {
+                        int stride = (_stride + (alignment - 1)) & -alignment;
+                        int newSize = (_size / _stride) * stride;
+
+                        var buffer = autoBuffer.Get(cbs, 0, newSize).Value;
 
                         if (gd.Capabilities.SupportsExtendedDynamicState)
                         {
@@ -69,7 +73,7 @@ namespace Ryujinx.Graphics.Vulkan
                                 1,
                                 buffer,
                                 0,
-                            (ulong)(_size / _stride) * (ulong)stride,
+                                (ulong)newSize,
                                 (ulong)stride);
                         }
                         else
@@ -78,16 +82,23 @@ namespace Ryujinx.Graphics.Vulkan
                         }
 
                         _buffer = autoBuffer;
-                    state.Internal.VertexBindingDescriptions[DescriptorIndex].Stride = (uint)stride;
+                    }
+
+                    state.Internal.VertexBindingDescriptions[DescriptorIndex].Stride = (uint)_stride;
 
                     return;
                 }
                 else
                 {
-                    autoBuffer = gd.BufferManager.GetBuffer(cbs.CommandBuffer, _handle, false, out int _);
+                    autoBuffer = gd.BufferManager.GetBuffer(cbs.CommandBuffer, _handle, false, out int size);
 
                     // The original stride must be reapplied in case it was rewritten.
                     state.Internal.VertexBindingDescriptions[DescriptorIndex].Stride = (uint)_stride;
+
+                    if (_offset >= size)
+                    {
+                        autoBuffer = null;
+                    }
                 }
             }
 

Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -310,7 +310,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal TextureView CreateTextureView(TextureCreateInfo info, float scale)
         {
             // This should be disposed when all views are destroyed.
-            using var storage = CreateTextureStorage(info, scale);
+            var storage = CreateTextureStorage(info, scale);
             return storage.CreateView(info, 0, 0);
         }
 

Ryujinx.HLE/HOS/Diagnostics/Demangler/Demangler.cs

@@ -918,7 +918,7 @@ namespace Ryujinx.HLE.HOS.Diagnostics.Demangler
                 return -1;
             }
 
-            return int.Parse(part.Substring(0, numberLength));
+            return int.Parse(part.AsSpan(0, numberLength));
         }
 
         private string ParseNumber(bool isSigned = false)

Ryujinx.HLE/HOS/Kernel/Memory/KPageTableBase.cs

@@ -2540,11 +2540,10 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
 
                 for (int attempt = 0; attempt < 8; attempt++)
                 {
-                    address = BitUtils.AlignDown(regionStart + GetRandomValue(0, aslrMaxOffset) * (ulong)alignment, alignment);
+                    ulong aslrAddress = BitUtils.AlignDown(regionStart + GetRandomValue(0, aslrMaxOffset) * (ulong)alignment, alignment);
+                    ulong aslrEndAddr = aslrAddress + totalNeededSize;
 
-                    ulong endAddr = address + totalNeededSize;
+                    KMemoryInfo info = _blockManager.FindBlock(aslrAddress).GetInfo();
-
-                    KMemoryInfo info = _blockManager.FindBlock(address).GetInfo();
 
                     if (info.State != MemoryState.Unmapped)
                     {
@@ -2554,11 +2553,12 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
                     ulong currBaseAddr = info.Address + reservedPagesCount * PageSize;
                     ulong currEndAddr = info.Address + info.Size;
 
-                    if (address >= regionStart &&
+                    if (aslrAddress >= regionStart &&
-                        address >= currBaseAddr &&
+                        aslrAddress >= currBaseAddr &&
-                        endAddr - 1 <= regionEndAddr - 1 &&
+                        aslrEndAddr - 1 <= regionEndAddr - 1 &&
-                        endAddr - 1 <= currEndAddr - 1)
+                        aslrEndAddr - 1 <= currEndAddr - 1)
                     {
+                        address = aslrAddress;
                         break;
                     }
                 }
@@ -2603,7 +2603,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Memory
 
             ulong totalNeededSize = reservedSize + neededPagesCount * PageSize;
 
-            ulong regionEndAddr = regionStart + regionPagesCount * PageSize;
+            ulong regionEndAddr = (regionStart + regionPagesCount * PageSize) - 1;
 
             KMemoryBlock currBlock = _blockManager.FindBlock(regionStart);
 

Ryujinx.HLE/HOS/Services/Fs/IFileSystemProxy.cs

@@ -12,6 +12,7 @@ using LibHac.Tools.FsSystem.NcaUtils;
 using Ryujinx.Common;
 using Ryujinx.Common.Logging;
 using Ryujinx.HLE.HOS.Services.Fs.FileSystemProxy;
+using System;
 using System.IO;
 
 using static Ryujinx.HLE.Utilities.StringUtils;
@@ -787,6 +788,26 @@ namespace Ryujinx.HLE.HOS.Services.Fs
             return ResultCode.Success;
         }
 
+        [CommandHipc(205)]
+        // OpenDataStorageWithProgramIndex(u8 program_index) -> object<nn::fssrv::sf::IStorage>
+        public ResultCode OpenDataStorageWithProgramIndex(ServiceCtx context)
+        {
+            byte programIndex = context.RequestData.ReadByte();
+
+            if ((context.Device.Application.TitleId & 0xf) != programIndex)
+            {
+                throw new NotImplementedException($"Accessing storage from other programs is not supported (program index = {programIndex}).");
+            }
+
+            var storage = context.Device.FileSystem.RomFs.AsStorage(true);
+            using var sharedStorage = new SharedRef<LibHac.Fs.IStorage>(storage);
+            using var sfStorage = new SharedRef<IStorage>(new StorageInterfaceAdapter(ref sharedStorage.Ref()));
+
+            MakeObject(context, new FileSystemProxy.IStorage(ref sfStorage.Ref()));
+
+            return ResultCode.Success;
+        }
+
         [CommandHipc(400)]
         // OpenDataStorageByCurrentProcess() -> object<nn::fssrv::sf::IStorage> dataStorage
         public ResultCode OpenDeviceOperator(ServiceCtx context)

Ryujinx.HLE/HOS/Services/Nfc/NfcManager/INfc.cs

@@ -5,22 +5,48 @@ namespace Ryujinx.HLE.HOS.Services.Nfc.NfcManager
     class INfc : IpcService
     {
         private NfcPermissionLevel _permissionLevel;
+        private State _state;
 
         public INfc(NfcPermissionLevel permissionLevel)
         {
             _permissionLevel = permissionLevel;
+            _state = State.NonInitialized;
         }
 
         [CommandHipc(0)]
         [CommandHipc(400)] // 4.0.0+
-        // Initialize()
+        // Initialize(u64, u64, pid, buffer<unknown, 5>)
         public ResultCode Initialize(ServiceCtx context)
         {
+            _state = State.Initialized;
+
             Logger.Stub?.PrintStub(LogClass.ServiceNfc, new { _permissionLevel });
 
             return ResultCode.Success;
         }
 
+        [CommandHipc(1)]
+        [CommandHipc(401)] // 4.0.0+
+        // Finalize()
+        public ResultCode Finalize(ServiceCtx context)
+        {
+            _state = State.NonInitialized;
+
+            Logger.Stub?.PrintStub(LogClass.ServiceNfc, new { _permissionLevel });
+
+            return ResultCode.Success;
+        }
+
+        [CommandHipc(2)]
+        [CommandHipc(402)] // 4.0.0+
+        // GetState() -> u32
+        public ResultCode GetState(ServiceCtx context) 
+        {
+            context.ResponseData.Write((int)_state);
+
+            return ResultCode.Success;
+        }
+
         [CommandHipc(3)]
         [CommandHipc(403)] // 4.0.0+
         // IsNfcEnabled() -> b8

Ryujinx.HLE/HOS/Services/Nfc/NfcManager/Types/State.cs

@@ -0,0 +1,8 @@
+namespace Ryujinx.HLE.HOS.Services.Nfc.NfcManager
+{
+    enum State
+    {
+        NonInitialized,
+        Initialized
+    }
+}

Ryujinx.HLE/HOS/Services/Nfc/Nfp/NfpManager/INfp.cs

@@ -816,11 +816,11 @@ namespace Ryujinx.HLE.HOS.Services.Nfc.Nfp
                                 Reserved = new Array57<byte>()
                             };
 
-                            modelInfo.CharacterId      = BinaryPrimitives.ReverseEndianness(ushort.Parse(context.Device.System.NfpDevices[i].AmiiboId.Substring(0, 4), NumberStyles.HexNumber));
+                            modelInfo.CharacterId      = BinaryPrimitives.ReverseEndianness(ushort.Parse(context.Device.System.NfpDevices[i].AmiiboId.AsSpan(0, 4), NumberStyles.HexNumber));
-                            modelInfo.CharacterVariant = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.Substring(4, 2), NumberStyles.HexNumber);
+                            modelInfo.CharacterVariant = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.AsSpan(4, 2), NumberStyles.HexNumber);
-                            modelInfo.Series           = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.Substring(12, 2), NumberStyles.HexNumber);
+                            modelInfo.Series           = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.AsSpan(12, 2), NumberStyles.HexNumber);
-                            modelInfo.ModelNumber      = ushort.Parse(context.Device.System.NfpDevices[i].AmiiboId.Substring(8, 4), NumberStyles.HexNumber);
+                            modelInfo.ModelNumber      = ushort.Parse(context.Device.System.NfpDevices[i].AmiiboId.AsSpan(8, 4), NumberStyles.HexNumber);
-                            modelInfo.Type             = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.Substring(6, 2), NumberStyles.HexNumber);
+                            modelInfo.Type             = byte.Parse(context.Device.System.NfpDevices[i].AmiiboId.AsSpan(6, 2), NumberStyles.HexNumber);
 
                             context.Memory.Write(outputPosition, modelInfo);
 

Ryujinx.HLE/HOS/Services/Sockets/Bsd/Impl/ManagedSocket.cs

@@ -235,6 +235,13 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
                     shouldBlockAfterOperation = true;
                 }
 
+                if (!Socket.IsBound)
+                {
+                    receiveSize = -1;
+
+                    return LinuxError.EOPNOTSUPP;
+                }
+
                 receiveSize = Socket.ReceiveFrom(buffer[..size], ConvertBsdSocketFlags(flags), ref temp);
 
                 remoteEndPoint = (IPEndPoint)temp;

Ryujinx.HLE/HOS/Services/Ssl/SslService/ISslConnection.cs

@@ -349,7 +349,11 @@ namespace Ryujinx.HLE.HOS.Services.Ssl.SslService
         // GetSessionCacheMode() -> nn::ssl::sf::SessionCacheMode
         public ResultCode GetSessionCacheMode(ServiceCtx context)
         {
-            throw new ServiceNotImplementedException(this, context);
+            context.ResponseData.Write((uint)_sessionCacheMode);
+
+            Logger.Stub?.PrintStub(LogClass.ServiceSsl, new { _sessionCacheMode });
+
+            return ResultCode.Success;
         }
 
         [CommandHipc(19)]

Ryujinx.HLE/HOS/Tamper/Conditions/InputMask.cs

@@ -13,7 +13,7 @@ namespace Ryujinx.HLE.HOS.Tamper.Conditions
 
         public bool Evaluate()
         {
-            return (_input.Value & _mask) != 0;
+            return (_input.Value & _mask) == _mask;
         }
     }
 }

Ryujinx.HLE/HOS/Tamper/InstructionHelper.cs

@@ -122,9 +122,8 @@ namespace Ryujinx.HLE.HOS.Tamper
                 for (int nybbleIndex = 0; nybbleIndex < wordSize; nybbleIndex++)
                 {
                     int index = wordIndex * wordSize + nybbleIndex;
-                    string byteData = word.Substring(nybbleIndex, 1);
 
-                    instruction[index] = byte.Parse(byteData, NumberStyles.HexNumber, CultureInfo.InvariantCulture);
+                    instruction[index] = byte.Parse(word.AsSpan(nybbleIndex, 1), NumberStyles.HexNumber, CultureInfo.InvariantCulture);
                 }
             }
 

Ryujinx.HLE/Loaders/Mods/IPSwitchPatcher.cs

@@ -132,7 +132,7 @@ namespace Ryujinx.HLE.Loaders.Mods
         {
             if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
             {
-                return int.TryParse(str.Substring(2), System.Globalization.NumberStyles.HexNumber, null, out value);
+                return int.TryParse(str.AsSpan(2), System.Globalization.NumberStyles.HexNumber, null, out value);
             }
             else
             {

Ryujinx.HLE/Utilities/StringUtils.cs

@@ -54,7 +54,7 @@ namespace Ryujinx.HLE.Utilities
 
             for (int index = 0; index < bytesInHex; index++)
             {
-                output[index] = byte.Parse(hexString.Substring(index * 2, 2), NumberStyles.HexNumber);
+                output[index] = byte.Parse(hexString.AsSpan(index * 2, 2), NumberStyles.HexNumber);
             }
 
             return output;

Ryujinx.Input/HLE/NpadManager.cs

@@ -51,7 +51,16 @@ namespace Ryujinx.Input.HLE
         {
             lock (_lock)
             {
-                _device.Hid.RefreshInputConfig(_inputConfig);
+                List<InputConfig> validInputs = new List<InputConfig>();
+                foreach (var inputConfigEntry in _inputConfig)
+                {
+                    if (_controllers[(int)inputConfigEntry.PlayerIndex] != null)
+                    {
+                        validInputs.Add(inputConfigEntry);
+                    }
+                }
+
+                _device.Hid.RefreshInputConfig(validInputs);
             }
         }
 
@@ -103,6 +112,8 @@ namespace Ryujinx.Input.HLE
                     _controllers[i] = null;
                 }
 
+                List<InputConfig> validInputs = new List<InputConfig>();
+
                 foreach (InputConfig inputConfigEntry in inputConfig)
                 {
                     NpadController controller = new NpadController(_cemuHookClient);
@@ -116,6 +127,7 @@ namespace Ryujinx.Input.HLE
                     else
                     {
                         _controllers[(int)inputConfigEntry.PlayerIndex] = controller;
+                        validInputs.Add(inputConfigEntry);
                     }
                 }
 
@@ -123,7 +135,7 @@ namespace Ryujinx.Input.HLE
                 _enableKeyboard = enableKeyboard;
                 _enableMouse    = enableMouse;
 
-                _device.Hid.RefreshInputConfig(inputConfig);
+                _device.Hid.RefreshInputConfig(validInputs);
             }
         }
 

Ryujinx.Memory.Tests/Tests.cs

@@ -92,5 +92,31 @@ namespace Ryujinx.Memory.Tests
                 }
             }
         }
+
+        [Test]
+        public void Test_AliasMapLeak()
+        {
+            if (OperatingSystem.IsMacOS())
+            {
+                // Memory aliasing tests fail on CI at the moment.
+                return;
+            }
+
+            ulong pageSize = 4096;
+            ulong size = 100000 * pageSize; // The mappings limit on Linux is usually around 65K, so let's make sure we are above that.
+
+            using MemoryBlock backing = new MemoryBlock(pageSize, MemoryAllocationFlags.Mirrorable);
+            using MemoryBlock toAlias = new MemoryBlock(size, MemoryAllocationFlags.Reserve | MemoryAllocationFlags.ViewCompatible);
+
+            for (ulong offset = 0; offset < size; offset += pageSize)
+            {
+                toAlias.MapView(backing, 0, offset, pageSize);
+
+                toAlias.Write(offset, 0xbadc0de);
+                Assert.AreEqual(0xbadc0de, backing.Read<int>(0));
+
+                toAlias.UnmapView(backing, offset, pageSize);
+            }
+        }
     }
 }

Ryujinx.Memory/MemoryManagementUnix.cs

@@ -177,7 +177,7 @@ namespace Ryujinx.Memory
 
         public static void UnmapView(IntPtr location, ulong size)
         {
-            mmap(location, size, MmapProts.PROT_NONE, MmapFlags.MAP_FIXED, -1, 0);
+            mmap(location, size, MmapProts.PROT_NONE, MmapFlags.MAP_FIXED | MmapFlags.MAP_PRIVATE | MmapFlags.MAP_ANONYMOUS | MmapFlags.MAP_NORESERVE, -1, 0);
         }
     }
 }

Ryujinx.Tests/Cpu/CpuTestSimdCvt32.cs

@@ -339,6 +339,93 @@ namespace Ryujinx.Tests.Cpu
 
             CompareAgainstUnicorn();
         }
+
+        [Explicit]
+        [Test, Pairwise, Description("VCVT<top>.F16.F32 <Sd>, <Dm>")]
+        public void Vcvt_F32_F16([Values(0u, 1u, 2u, 3u)] uint rd,
+                                 [Values(0u, 1u, 2u, 3u)] uint rm,
+                                 [ValueSource(nameof(_1S_))] [Random(RndCnt)] uint s0,
+                                 [ValueSource(nameof(_1S_))] [Random(RndCnt)] uint s1,
+                                 [ValueSource(nameof(_1S_))] [Random(RndCnt)] uint s2,
+                                 [ValueSource(nameof(_1S_))] [Random(RndCnt)] uint s3,
+                                 [Values] bool top)
+        {
+            uint opcode = 0xeeb30a40; // VCVTB.F16.F32 S0, D0
+
+            if (top)
+            {
+                opcode |= 1 << 7;
+            }
+
+            opcode |= ((rd & 0x1e) << 11) | ((rd & 0x1) << 22);
+            opcode |= ((rm & 0x1e) >> 1) | ((rm & 0x1) << 5);
+
+            V128 v0 = MakeVectorE0E1E2E3(s0, s1, s2, s3);
+
+            SingleOpcode(opcode, v0: v0);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Explicit]
+        [Test, Pairwise, Description("VCVT<top>.F16.F64 <Sd>, <Dm>")]
+        public void Vcvt_F64_F16([Values(0u, 1u, 2u, 3u)] uint rd,
+                                 [Values(0u, 1u)] uint rm,
+                                 [ValueSource(nameof(_1D_F_))] ulong d0,
+                                 [ValueSource(nameof(_1D_F_))] ulong d1,
+                                 [Values] bool top)
+        {
+            uint opcode = 0xeeb30b40; // VCVTB.F16.F64 S0, D0
+
+            if (top)
+            {
+                opcode |= 1 << 7;
+            }
+
+            opcode |= ((rd & 0x1e) << 11) | ((rd & 0x1) << 22);
+            opcode |= ((rm & 0xf) << 0) | ((rm & 0x10) << 1);
+
+            V128 v0 = MakeVectorE0E1(d0, d1);
+
+            SingleOpcode(opcode, v0: v0);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Explicit]
+        [Test, Pairwise, Description("VCVT<top>.F<size>.F16 <Vd>, <Sm>")]
+        public void Vcvt_F16_Fx([Values(0u, 1u, 2u, 3u)] uint rd,
+                                 [Values(0u, 1u, 2u, 3u)] uint rm,
+                                 [ValueSource(nameof(_1D_F_))] ulong d0,
+                                 [ValueSource(nameof(_1D_F_))] ulong d1,
+                                 [Values] bool top,
+                                 [Values] bool sz)
+        {
+            uint opcode = 0xeeb20a40; // VCVTB.F32.F16 S0, S0
+
+            if (top)
+            {
+                opcode |= 1 << 7;
+            }
+
+            if (sz)
+            {
+                opcode |= 1 << 8;
+                opcode |= ((rd & 0xf) << 12) | ((rd & 0x10) << 18);
+            }
+            else
+            {
+                opcode |= ((rd & 0x1e) << 11) | ((rd & 0x1) << 22);
+            }
+
+            opcode |= ((rm & 0xf) << 0) | ((rm & 0x10) << 1);
+
+            V128 v0 = MakeVectorE0E1(d0, d1);
+
+            SingleOpcode(opcode, v0: v0);
+
+            CompareAgainstUnicorn();
+        }
 #endif
     }
 }

Ryujinx/Modules/Updater/UpdateDialog.cs

@@ -48,7 +48,7 @@ namespace Ryujinx.Modules
             {
                 string ryuName = OperatingSystem.IsWindows() ? "Ryujinx.exe" : "Ryujinx";
                 string ryuExe  = System.IO.Path.Combine(AppDomain.CurrentDomain.BaseDirectory, ryuName);
-                string ryuArg  = string.Join(" ", Environment.GetCommandLineArgs().AsEnumerable().Skip(1).ToArray());
+                var ryuArg  = Environment.GetCommandLineArgs().AsEnumerable().Skip(1);
 
                 Process.Start(ryuExe, ryuArg);