bsd: Fix Poll writting in input buffer (#3630)

This is a very old oversight on our Poll implementation.
This worked so far reliably because games and homebrews pass the same
buffer as input and output.

ARMeilleure/CodeGen/X86/AssemblerTable.cs

@@ -157,6 +157,7 @@ namespace ARMeilleure.CodeGen.X86
             Add(X86Instruction.Paddd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ffe, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Paddq,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fd4, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Paddw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ffd, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Palignr,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a0f, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Pand,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fdb, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Pandn,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fdf, InstructionFlags.Vex | InstructionFlags.Prefix66));
             Add(X86Instruction.Pavgb,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fe0, InstructionFlags.Vex | InstructionFlags.Prefix66));
@@ -239,6 +240,9 @@ namespace ARMeilleure.CodeGen.X86
             Add(X86Instruction.Rsqrtss,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f52, InstructionFlags.Vex | InstructionFlags.PrefixF3));
             Add(X86Instruction.Sar,          new InstructionInfo(0x070000d3, 0x070000c1, BadOp,      BadOp,      BadOp,      InstructionFlags.None));
             Add(X86Instruction.Setcc,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f90, InstructionFlags.Reg8Dest));
+            Add(X86Instruction.Sha256Msg1,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38cc, InstructionFlags.None));
+            Add(X86Instruction.Sha256Msg2,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38cd, InstructionFlags.None));
+            Add(X86Instruction.Sha256Rnds2,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38cb, InstructionFlags.None));
             Add(X86Instruction.Shl,          new InstructionInfo(0x040000d3, 0x040000c1, BadOp,      BadOp,      BadOp,      InstructionFlags.None));
             Add(X86Instruction.Shr,          new InstructionInfo(0x050000d3, 0x050000c1, BadOp,      BadOp,      BadOp,      InstructionFlags.None));
             Add(X86Instruction.Shufpd,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc6, InstructionFlags.Vex | InstructionFlags.Prefix66));

ARMeilleure/CodeGen/X86/HardwareCapabilities.cs

@@ -12,21 +12,28 @@ namespace ARMeilleure.CodeGen.X86
                 return;
             }
 
-            (_, _, int ecx, int edx) = X86Base.CpuId(0x00000001, 0x00000000);
+            (int maxNum, _, _, _) = X86Base.CpuId(0x00000000, 0x00000000);
 
-            FeatureInfoEdx = (FeatureFlagsEdx)edx;
-            FeatureInfoEcx = (FeatureFlagsEcx)ecx;
+            (_, _, int ecx1, int edx1) = X86Base.CpuId(0x00000001, 0x00000000);
+            FeatureInfo1Edx = (FeatureFlags1Edx)edx1;
+            FeatureInfo1Ecx = (FeatureFlags1Ecx)ecx1;
+
+            if (maxNum >= 7)
+            {
+                (_, int ebx7, _, _) = X86Base.CpuId(0x00000007, 0x00000000);
+                FeatureInfo7Ebx = (FeatureFlags7Ebx)ebx7;
+            }
         }
 
         [Flags]
-        public enum FeatureFlagsEdx
+        public enum FeatureFlags1Edx
         {
             Sse = 1 << 25,
             Sse2 = 1 << 26
         }
 
         [Flags]
-        public enum FeatureFlagsEcx
+        public enum FeatureFlags1Ecx
         {
             Sse3 = 1 << 0,
             Pclmulqdq = 1 << 1,
@@ -40,21 +47,31 @@ namespace ARMeilleure.CodeGen.X86
             F16c = 1 << 29
         }
 
-        public static FeatureFlagsEdx FeatureInfoEdx { get; }
-        public static FeatureFlagsEcx FeatureInfoEcx { get; }
+        [Flags]
+        public enum FeatureFlags7Ebx
+        {
+            Avx2 = 1 << 5,
+            Sha = 1 << 29
+        }
 
-        public static bool SupportsSse => FeatureInfoEdx.HasFlag(FeatureFlagsEdx.Sse);
-        public static bool SupportsSse2 => FeatureInfoEdx.HasFlag(FeatureFlagsEdx.Sse2);
-        public static bool SupportsSse3 => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Sse3);
-        public static bool SupportsPclmulqdq => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Pclmulqdq);
-        public static bool SupportsSsse3 => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Ssse3);
-        public static bool SupportsFma => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Fma);
-        public static bool SupportsSse41 => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Sse41);
-        public static bool SupportsSse42 => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Sse42);
-        public static bool SupportsPopcnt => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Popcnt);
-        public static bool SupportsAesni => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Aes);
-        public static bool SupportsAvx => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.Avx);
-        public static bool SupportsF16c => FeatureInfoEcx.HasFlag(FeatureFlagsEcx.F16c);
+        public static FeatureFlags1Edx FeatureInfo1Edx { get; }
+        public static FeatureFlags1Ecx FeatureInfo1Ecx { get; }
+        public static FeatureFlags7Ebx FeatureInfo7Ebx { get; } = 0;
+
+        public static bool SupportsSse => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse);
+        public static bool SupportsSse2 => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse2);
+        public static bool SupportsSse3 => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Sse3);
+        public static bool SupportsPclmulqdq => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Pclmulqdq);
+        public static bool SupportsSsse3 => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Ssse3);
+        public static bool SupportsFma => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Fma);
+        public static bool SupportsSse41 => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Sse41);
+        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 SupportsAvx2 => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx2) && SupportsAvx;
+        public static bool SupportsF16c => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.F16c);
+        public static bool SupportsSha => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Sha);
 
         public static bool ForceLegacySse { get; set; }
 

ARMeilleure/CodeGen/X86/IntrinsicTable.cs

@@ -82,6 +82,7 @@ namespace ARMeilleure.CodeGen.X86
             Add(Intrinsic.X86Paddd,        new IntrinsicInfo(X86Instruction.Paddd,        IntrinsicType.Binary));
             Add(Intrinsic.X86Paddq,        new IntrinsicInfo(X86Instruction.Paddq,        IntrinsicType.Binary));
             Add(Intrinsic.X86Paddw,        new IntrinsicInfo(X86Instruction.Paddw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Palignr,      new IntrinsicInfo(X86Instruction.Palignr,      IntrinsicType.TernaryImm));
             Add(Intrinsic.X86Pand,         new IntrinsicInfo(X86Instruction.Pand,         IntrinsicType.Binary));
             Add(Intrinsic.X86Pandn,        new IntrinsicInfo(X86Instruction.Pandn,        IntrinsicType.Binary));
             Add(Intrinsic.X86Pavgb,        new IntrinsicInfo(X86Instruction.Pavgb,        IntrinsicType.Binary));
@@ -151,6 +152,9 @@ namespace ARMeilleure.CodeGen.X86
             Add(Intrinsic.X86Roundss,      new IntrinsicInfo(X86Instruction.Roundss,      IntrinsicType.BinaryImm));
             Add(Intrinsic.X86Rsqrtps,      new IntrinsicInfo(X86Instruction.Rsqrtps,      IntrinsicType.Unary));
             Add(Intrinsic.X86Rsqrtss,      new IntrinsicInfo(X86Instruction.Rsqrtss,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Sha256Msg1,   new IntrinsicInfo(X86Instruction.Sha256Msg1,   IntrinsicType.Binary));
+            Add(Intrinsic.X86Sha256Msg2,   new IntrinsicInfo(X86Instruction.Sha256Msg2,   IntrinsicType.Binary));
+            Add(Intrinsic.X86Sha256Rnds2,  new IntrinsicInfo(X86Instruction.Sha256Rnds2,  IntrinsicType.Ternary));
             Add(Intrinsic.X86Shufpd,       new IntrinsicInfo(X86Instruction.Shufpd,       IntrinsicType.TernaryImm));
             Add(Intrinsic.X86Shufps,       new IntrinsicInfo(X86Instruction.Shufps,       IntrinsicType.TernaryImm));
             Add(Intrinsic.X86Sqrtpd,       new IntrinsicInfo(X86Instruction.Sqrtpd,       IntrinsicType.Unary));

ARMeilleure/CodeGen/X86/PreAllocator.cs

@@ -308,11 +308,13 @@ namespace ARMeilleure.CodeGen.X86
 
                 case Instruction.Extended:
                 {
+                    bool isBlend = node.Intrinsic == Intrinsic.X86Blendvpd ||
+                                   node.Intrinsic == Intrinsic.X86Blendvps ||
+                                   node.Intrinsic == Intrinsic.X86Pblendvb;
+
                     // BLENDVPD, BLENDVPS, PBLENDVB last operand is always implied to be XMM0 when VEX is not supported.
-                    if ((node.Intrinsic == Intrinsic.X86Blendvpd ||
-                         node.Intrinsic == Intrinsic.X86Blendvps ||
-                         node.Intrinsic == Intrinsic.X86Pblendvb) &&
-                         !HardwareCapabilities.SupportsVexEncoding)
+                    // SHA256RNDS2 always has an implied XMM0 as a last operand.
+                    if ((isBlend && !HardwareCapabilities.SupportsVexEncoding) || node.Intrinsic == Intrinsic.X86Sha256Rnds2)
                     {
                         Operand xmm0 = Xmm(X86Register.Xmm0, OperandType.V128);
 

ARMeilleure/CodeGen/X86/X86Instruction.cs

@@ -98,6 +98,7 @@ namespace ARMeilleure.CodeGen.X86
         Paddd,
         Paddq,
         Paddw,
+        Palignr,
         Pand,
         Pandn,
         Pavgb,
@@ -180,6 +181,9 @@ namespace ARMeilleure.CodeGen.X86
         Rsqrtss,
         Sar,
         Setcc,
+        Sha256Msg1,
+        Sha256Msg2,
+        Sha256Rnds2,
         Shl,
         Shr,
         Shufpd,

ARMeilleure/Decoders/IOpCode32MemRsImm.cs

@@ -0,0 +1,8 @@
+namespace ARMeilleure.Decoders
+{
+    interface IOpCode32MemRsImm : IOpCode32Mem
+    {
+        int Rm { get; }
+        ShiftType ShiftType { get; }
+    }
+}

ARMeilleure/Decoders/OpCode32MemMult.cs

@@ -1,3 +1,5 @@
+using System.Numerics;
+
 namespace ARMeilleure.Decoders
 {
     class OpCode32MemMult : OpCode32, IOpCode32MemMult
@@ -23,14 +25,7 @@ namespace ARMeilleure.Decoders
 
             RegisterMask = opCode & 0xffff;
 
-            int regsSize = 0;
-
-            for (int index = 0; index < 16; index++)
-            {
-                regsSize += (RegisterMask >> index) & 1;
-            }
-
-            regsSize *= 4;
+            int regsSize = BitOperations.PopCount((uint)RegisterMask) * 4;
 
             if (!u)
             {

ARMeilleure/Decoders/OpCode32MemRsImm.cs

@@ -1,6 +1,6 @@
 namespace ARMeilleure.Decoders
 {
-    class OpCode32MemRsImm : OpCode32Mem
+    class OpCode32MemRsImm : OpCode32Mem, IOpCode32MemRsImm
     {
         public int Rm { get; }
         public ShiftType ShiftType { get; }

ARMeilleure/Decoders/OpCodeT16BImmCmp.cs

@@ -1,10 +1,10 @@
 namespace ARMeilleure.Decoders
 {
-    class OpCodeT16BImmCmp : OpCodeT16
+    class OpCodeT16BImmCmp : OpCodeT16, IOpCode32BImm
     {
         public int Rn { get; }
 
-        public int Immediate { get; }
+        public long Immediate { get; }
 
         public static new OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT16BImmCmp(inst, address, opCode);
 

ARMeilleure/Decoders/OpCodeT32MemImm8D.cs

@@ -0,0 +1,31 @@
+namespace ARMeilleure.Decoders
+{
+    class OpCodeT32MemImm8D : OpCodeT32, IOpCode32Mem
+    {
+        public int Rt { get; }
+        public int Rt2 { get; }
+        public int Rn { get; }
+        public bool WBack { get; }
+        public bool IsLoad { get; }
+        public bool Index { get; }
+        public bool Add { get; }
+        public int Immediate { get; }
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT32MemImm8D(inst, address, opCode);
+
+        public OpCodeT32MemImm8D(InstDescriptor inst, ulong address, int opCode) : base(inst, address, opCode)
+        {
+            Rt2 = (opCode >> 8) & 0xf;
+            Rt = (opCode >> 12) & 0xf;
+            Rn = (opCode >> 16) & 0xf;
+
+            Index = ((opCode >> 24) & 1) != 0;
+            Add   = ((opCode >> 23) & 1) != 0;
+            WBack = ((opCode >> 21) & 1) != 0;
+
+            Immediate = opCode & 0xff;
+
+            IsLoad = ((opCode >> 20) & 1) != 0;
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeT32MemLdEx.cs

@@ -0,0 +1,24 @@
+namespace ARMeilleure.Decoders
+{
+    class OpCodeT32MemLdEx : OpCodeT32, IOpCode32MemEx
+    {
+        public int Rd => 0;
+        public int Rt { get; }
+        public int Rn { get; }
+
+        public bool WBack => false;
+        public bool IsLoad => true;
+        public bool Index => false;
+        public bool Add => false;
+
+        public int Immediate => 0;
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT32MemLdEx(inst, address, opCode);
+
+        public OpCodeT32MemLdEx(InstDescriptor inst, ulong address, int opCode) : base(inst, address, opCode)
+        {
+            Rt = (opCode >> 12) & 0xf;
+            Rn = (opCode >> 16) & 0xf;
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeT32MemMult.cs

@@ -0,0 +1,52 @@
+using System.Numerics;
+
+namespace ARMeilleure.Decoders
+{
+    class OpCodeT32MemMult : OpCodeT32, IOpCode32MemMult
+    {
+        public int Rn { get; }
+
+        public int RegisterMask { get; }
+        public int Offset       { get; }
+        public int PostOffset   { get; }
+
+        public bool IsLoad { get; }
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT32MemMult(inst, address, opCode);
+
+        public OpCodeT32MemMult(InstDescriptor inst, ulong address, int opCode) : base(inst, address, opCode)
+        {
+            Rn = (opCode >> 16) & 0xf;
+
+            bool isLoad = (opCode & (1 << 20)) != 0;
+            bool w      = (opCode & (1 << 21)) != 0;
+            bool u      = (opCode & (1 << 23)) != 0;
+            bool p      = (opCode & (1 << 24)) != 0;
+
+            RegisterMask = opCode & 0xffff;
+
+            int regsSize = BitOperations.PopCount((uint)RegisterMask) * 4;
+
+            if (!u)
+            {
+                Offset -= regsSize;
+            }
+
+            if (u == p)
+            {
+                Offset += 4;
+            }
+
+            if (w)
+            {
+                PostOffset = u ? regsSize : -regsSize;
+            }
+            else
+            {
+                PostOffset = 0;
+            }
+
+            IsLoad = isLoad;
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeT32MemRsImm.cs

@@ -0,0 +1,30 @@
+namespace ARMeilleure.Decoders
+{
+    class OpCodeT32MemRsImm : OpCodeT32, IOpCode32MemRsImm
+    {
+        public int Rt { get; }
+        public int Rn { get; }
+        public int Rm { get; }
+        public ShiftType ShiftType => ShiftType.Lsl;
+
+        public bool WBack => false;
+        public bool IsLoad { get; }
+        public bool Index => true;
+        public bool Add => true;
+
+        public int Immediate { get; }
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT32MemRsImm(inst, address, opCode);
+
+        public OpCodeT32MemRsImm(InstDescriptor inst, ulong address, int opCode) : base(inst, address, opCode)
+        {
+            Rm = (opCode >> 0) & 0xf;
+            Rt = (opCode >> 12) & 0xf;
+            Rn = (opCode >> 16) & 0xf;
+
+            IsLoad = (opCode & (1 << 20)) != 0;
+
+            Immediate = (opCode >> 4) & 3;
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeT32MemStEx.cs

@@ -0,0 +1,25 @@
+namespace ARMeilleure.Decoders
+{
+    class OpCodeT32MemStEx : OpCodeT32, IOpCode32MemEx
+    {
+        public int Rd { get; }
+        public int Rt { get; }
+        public int Rn { get; }
+
+        public bool WBack => false;
+        public bool IsLoad => false;
+        public bool Index => false;
+        public bool Add => false;
+
+        public int Immediate => 0;
+
+        public new static OpCode Create(InstDescriptor inst, ulong address, int opCode) => new OpCodeT32MemStEx(inst, address, opCode);
+
+        public OpCodeT32MemStEx(InstDescriptor inst, ulong address, int opCode) : base(inst, address, opCode)
+        {
+            Rd = (opCode >> 0) & 0xf;
+            Rt = (opCode >> 12) & 0xf;
+            Rn = (opCode >> 16) & 0xf;
+        }
+    }
+}

ARMeilleure/Decoders/OpCodeTable.cs

@@ -1070,14 +1070,19 @@ namespace ARMeilleure.Decoders
             SetT32("11110x011011xxxx0xxx1111xxxxxxxx", InstName.Cmp,      InstEmit32.Cmp,      OpCodeT32AluImm.Create);
             SetT32("11101010100<xxxx0xxx<<<<xxxxxxxx", InstName.Eor,      InstEmit32.Eor,      OpCodeT32AluRsImm.Create);
             SetT32("11110x00100<xxxx0xxx<<<<xxxxxxxx", InstName.Eor,      InstEmit32.Eor,      OpCodeT32AluImm.Create);
+            SetT32("111010001101xxxxxxxx111111101111", InstName.Ldaex,    InstEmit32.Ldaex,    OpCodeT32MemLdEx.Create);
+            SetT32("1110100010x1xxxxxxxxxxxxxxxxxxxx", InstName.Ldm,      InstEmit32.Ldm,      OpCodeT32MemMult.Create);
+            SetT32("1110100100x1xxxxxxxxxxxxxxxxxxxx", InstName.Ldm,      InstEmit32.Ldm,      OpCodeT32MemMult.Create);
             SetT32("111110000101xxxx<<<<10x1xxxxxxxx", InstName.Ldr,      InstEmit32.Ldr,      OpCodeT32MemImm8.Create);
             SetT32("111110000101xxxx<<<<1100xxxxxxxx", InstName.Ldr,      InstEmit32.Ldr,      OpCodeT32MemImm8.Create);
             SetT32("111110000101xxxx<<<<11x1xxxxxxxx", InstName.Ldr,      InstEmit32.Ldr,      OpCodeT32MemImm8.Create);
             SetT32("111110001101xxxxxxxxxxxxxxxxxxxx", InstName.Ldr,      InstEmit32.Ldr,      OpCodeT32MemImm12.Create);
+            SetT32("111110000101<<<<xxxx000000xxxxxx", InstName.Ldr,      InstEmit32.Ldr,      OpCodeT32MemRsImm.Create);
             SetT32("111110000001xxxx<<<<10x1xxxxxxxx", InstName.Ldrb,     InstEmit32.Ldrb,     OpCodeT32MemImm8.Create);
             SetT32("111110000001xxxx<<<<1100xxxxxxxx", InstName.Ldrb,     InstEmit32.Ldrb,     OpCodeT32MemImm8.Create);
             SetT32("111110000001xxxx<<<<11x1xxxxxxxx", InstName.Ldrb,     InstEmit32.Ldrb,     OpCodeT32MemImm8.Create);
             SetT32("111110001001xxxxxxxxxxxxxxxxxxxx", InstName.Ldrb,     InstEmit32.Ldrb,     OpCodeT32MemImm12.Create);
+            SetT32("1110100>x1>1<<<<xxxxxxxxxxxxxxxx", InstName.Ldrd,     InstEmit32.Ldrd,     OpCodeT32MemImm8D.Create);
             SetT32("111110000011xxxx<<<<10x1xxxxxxxx", InstName.Ldrh,     InstEmit32.Ldrh,     OpCodeT32MemImm8.Create);
             SetT32("111110000011xxxx<<<<1100xxxxxxxx", InstName.Ldrh,     InstEmit32.Ldrh,     OpCodeT32MemImm8.Create);
             SetT32("111110000011xxxx<<<<11x1xxxxxxxx", InstName.Ldrh,     InstEmit32.Ldrh,     OpCodeT32MemImm8.Create);
@@ -1102,10 +1107,15 @@ namespace ARMeilleure.Decoders
             SetT32("11110x01110xxxxx0xxxxxxxxxxxxxxx", InstName.Rsb,      InstEmit32.Rsb,      OpCodeT32AluImm.Create);
             SetT32("11101011011xxxxx0xxxxxxxxxxxxxxx", InstName.Sbc,      InstEmit32.Sbc,      OpCodeT32AluRsImm.Create);
             SetT32("11110x01011xxxxx0xxxxxxxxxxxxxxx", InstName.Sbc,      InstEmit32.Sbc,      OpCodeT32AluImm.Create);
+            SetT32("111010001100xxxxxxxx11111110xxxx", InstName.Stlex,    InstEmit32.Stlex,    OpCodeT32MemStEx.Create);
+            SetT32("1110100010x0xxxx0xxxxxxxxxxxxxxx", InstName.Stm,      InstEmit32.Stm,      OpCodeT32MemMult.Create);
+            SetT32("1110100100x0xxxx0xxxxxxxxxxxxxxx", InstName.Stm,      InstEmit32.Stm,      OpCodeT32MemMult.Create);
             SetT32("111110000100xxxxxxxx1<<>xxxxxxxx", InstName.Str,      InstEmit32.Str,      OpCodeT32MemImm8.Create);
             SetT32("111110001100xxxxxxxxxxxxxxxxxxxx", InstName.Str,      InstEmit32.Str,      OpCodeT32MemImm12.Create);
+            SetT32("111110000100<<<<xxxx000000xxxxxx", InstName.Str,      InstEmit32.Str,      OpCodeT32MemRsImm.Create);
             SetT32("111110000000xxxxxxxx1<<>xxxxxxxx", InstName.Strb,     InstEmit32.Strb,     OpCodeT32MemImm8.Create);
             SetT32("111110001000xxxxxxxxxxxxxxxxxxxx", InstName.Strb,     InstEmit32.Strb,     OpCodeT32MemImm12.Create);
+            SetT32("1110100>x1>0<<<<xxxxxxxxxxxxxxxx", InstName.Strd,     InstEmit32.Strd,     OpCodeT32MemImm8D.Create);
             SetT32("111110000010xxxxxxxx1<<>xxxxxxxx", InstName.Strh,     InstEmit32.Strh,     OpCodeT32MemImm8.Create);
             SetT32("111110001010xxxxxxxxxxxxxxxxxxxx", InstName.Strh,     InstEmit32.Strh,     OpCodeT32MemImm12.Create);
             SetT32("11101011101<xxxx0xxx<<<<xxxxxxxx", InstName.Sub,      InstEmit32.Sub,      OpCodeT32AluRsImm.Create);

ARMeilleure/Instructions/InstEmitFlow32.cs

@@ -88,7 +88,7 @@ namespace ARMeilleure.Instructions
         {
             OpCodeT16BImmCmp op = (OpCodeT16BImmCmp)context.CurrOp;
 
-            Operand value = GetIntOrZR(context, op.Rn);
+            Operand value = GetIntA32(context, op.Rn);
             Operand lblTarget = context.GetLabel((ulong)op.Immediate);
 
             if (onNotZero)

ARMeilleure/Instructions/InstEmitMemoryHelper.cs

@@ -547,7 +547,7 @@ namespace ARMeilleure.Instructions
         {
             switch (context.CurrOp)
             {
-                case OpCode32MemRsImm op: return GetMShiftedByImmediate(context, op, setCarry);
+                case IOpCode32MemRsImm op: return GetMShiftedByImmediate(context, op, setCarry);
 
                 case IOpCode32MemReg op: return GetIntA32(context, op.Rm);
 
@@ -564,7 +564,7 @@ namespace ARMeilleure.Instructions
             return new InvalidOperationException($"Invalid OpCode type \"{opCode?.GetType().Name ?? "null"}\".");
         }
 
-        public static Operand GetMShiftedByImmediate(ArmEmitterContext context, OpCode32MemRsImm op, bool setCarry)
+        public static Operand GetMShiftedByImmediate(ArmEmitterContext context, IOpCode32MemRsImm op, bool setCarry)
         {
             Operand m = GetIntA32(context, op.Rm);
 

ARMeilleure/Instructions/InstEmitSimdHash.cs

@@ -100,7 +100,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVec(op.Rn);
             Operand m = GetVec(op.Rm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashLower)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256h(context, d, n, m, part2: false);
 
             context.Copy(GetVec(op.Rd), res);
         }
@@ -113,7 +113,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVec(op.Rn);
             Operand m = GetVec(op.Rm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashUpper)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256h(context, n, d, m, part2: true);
 
             context.Copy(GetVec(op.Rd), res);
         }
@@ -125,7 +125,7 @@ namespace ARMeilleure.Instructions
             Operand d = GetVec(op.Rd);
             Operand n = GetVec(op.Rn);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart1)), d, n);
+            Operand res = InstEmitSimdHashHelper.EmitSha256su0(context, d, n);
 
             context.Copy(GetVec(op.Rd), res);
         }
@@ -138,7 +138,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVec(op.Rn);
             Operand m = GetVec(op.Rm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart2)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256su1(context, d, n, m);
 
             context.Copy(GetVec(op.Rd), res);
         }

ARMeilleure/Instructions/InstEmitSimdHash32.cs

@@ -17,7 +17,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVecA32(op.Qn);
             Operand m = GetVecA32(op.Qm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashLower)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256h(context, d, n, m, part2: false);
 
             context.Copy(GetVecA32(op.Qd), res);
         }
@@ -30,7 +30,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVecA32(op.Qn);
             Operand m = GetVecA32(op.Qm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.HashUpper)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256h(context, n, d, m, part2: true);
 
             context.Copy(GetVecA32(op.Qd), res);
         }
@@ -42,7 +42,7 @@ namespace ARMeilleure.Instructions
             Operand d = GetVecA32(op.Qd);
             Operand m = GetVecA32(op.Qm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart1)), d, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256su0(context, d, m);
 
             context.Copy(GetVecA32(op.Qd), res);
         }
@@ -55,7 +55,7 @@ namespace ARMeilleure.Instructions
             Operand n = GetVecA32(op.Qn);
             Operand m = GetVecA32(op.Qm);
 
-            Operand res = context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart2)), d, n, m);
+            Operand res = InstEmitSimdHashHelper.EmitSha256su1(context, d, n, m);
 
             context.Copy(GetVecA32(op.Qd), res);
         }

ARMeilleure/Instructions/InstEmitSimdHashHelper.cs

@@ -0,0 +1,56 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.Instructions
+{
+    static class InstEmitSimdHashHelper
+    {
+        public static Operand EmitSha256h(ArmEmitterContext context, Operand x, Operand y, Operand w, bool part2)
+        {
+            if (Optimizations.UseSha)
+            {
+                Operand src1 = context.AddIntrinsic(Intrinsic.X86Shufps, y, x, Const(0xbb));
+                Operand src2 = context.AddIntrinsic(Intrinsic.X86Shufps, y, x, Const(0x11));
+                Operand w2 = context.AddIntrinsic(Intrinsic.X86Punpckhqdq, w, w);
+
+                Operand round2 = context.AddIntrinsic(Intrinsic.X86Sha256Rnds2, src1, src2, w);
+                Operand round4 = context.AddIntrinsic(Intrinsic.X86Sha256Rnds2, src2, round2, w2);
+                
+                Operand res = context.AddIntrinsic(Intrinsic.X86Shufps, round4, round2, Const(part2 ? 0x11 : 0xbb));
+                
+                return res;
+            }
+
+            String method = part2 ? nameof(SoftFallback.HashUpper) : nameof(SoftFallback.HashLower);
+            return context.Call(typeof(SoftFallback).GetMethod(method), x, y, w);
+        }
+
+        public static Operand EmitSha256su0(ArmEmitterContext context, Operand x, Operand y)
+        {
+            if (Optimizations.UseSha)
+            {
+                return context.AddIntrinsic(Intrinsic.X86Sha256Msg1, x, y);
+            }
+
+            return context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart1)), x, y);
+        }
+
+        public static Operand EmitSha256su1(ArmEmitterContext context, Operand x, Operand y, Operand z)
+        {
+            if (Optimizations.UseSha && Optimizations.UseSsse3)
+            {
+                Operand extr = context.AddIntrinsic(Intrinsic.X86Palignr, z, y, Const(4));
+                Operand tmp = context.AddIntrinsic(Intrinsic.X86Paddd, extr, x);
+
+                Operand res = context.AddIntrinsic(Intrinsic.X86Sha256Msg2, tmp, z);
+
+                return res;
+            }
+
+            return context.Call(typeof(SoftFallback).GetMethod(nameof(SoftFallback.Sha256SchedulePart2)), x, y, z);
+        }
+    }
+}

ARMeilleure/Instructions/SoftFallback.cs

@@ -1129,7 +1129,7 @@ namespace ARMeilleure.Instructions
             return Sha256Hash(hash_abcd, hash_efgh, wk, part1: true);
         }
 
-        public static V128 HashUpper(V128 hash_efgh, V128 hash_abcd, V128 wk)
+        public static V128 HashUpper(V128 hash_abcd, V128 hash_efgh, V128 wk)
         {
             return Sha256Hash(hash_abcd, hash_efgh, wk, part1: false);
         }

ARMeilleure/IntermediateRepresentation/Intrinsic.cs

@@ -71,6 +71,7 @@ namespace ARMeilleure.IntermediateRepresentation
         X86Paddd,
         X86Paddq,
         X86Paddw,
+        X86Palignr,
         X86Pand,
         X86Pandn,
         X86Pavgb,
@@ -140,6 +141,9 @@ namespace ARMeilleure.IntermediateRepresentation
         X86Roundss,
         X86Rsqrtps,
         X86Rsqrtss,
+        X86Sha256Msg1,
+        X86Sha256Msg2,
+        X86Sha256Rnds2,
         X86Shufpd,
         X86Shufps,
         X86Sqrtpd,

ARMeilleure/Optimizations.cs

@@ -21,6 +21,7 @@ namespace ARMeilleure
         public static bool UseFmaIfAvailable       { get; set; } = true;
         public static bool UseAesniIfAvailable     { get; set; } = true;
         public static bool UsePclmulqdqIfAvailable { get; set; } = true;
+        public static bool UseShaIfAvailable       { get; set; } = true;
 
         public static bool ForceLegacySse
         {
@@ -40,5 +41,6 @@ namespace ARMeilleure
         internal static bool UseFma       => UseFmaIfAvailable       && HardwareCapabilities.SupportsFma;
         internal static bool UseAesni     => UseAesniIfAvailable     && HardwareCapabilities.SupportsAesni;
         internal static bool UsePclmulqdq => UsePclmulqdqIfAvailable && HardwareCapabilities.SupportsPclmulqdq;
+        internal static bool UseSha       => UseShaIfAvailable       && HardwareCapabilities.SupportsSha;
     }
 }

ARMeilleure/Translation/ControlFlowGraph.cs

@@ -14,7 +14,7 @@ namespace ARMeilleure.Translation
         public BasicBlock Entry { get; }
         public IntrusiveList<BasicBlock> Blocks { get; }
         public BasicBlock[] PostOrderBlocks => _postOrderBlocks;
-        public int[] PostOrderMap => _postOrderMap; 
+        public int[] PostOrderMap => _postOrderMap;
 
         public ControlFlowGraph(BasicBlock entry, IntrusiveList<BasicBlock> blocks, int localsCount)
         {

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 = 3439; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 3585; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";
@@ -946,9 +946,12 @@ namespace ARMeilleure.Translation.PTC
             return BitConverter.IsLittleEndian;
         }
 
-        private static ulong GetFeatureInfo()
+        private static FeatureInfo GetFeatureInfo()
         {
-            return (ulong)HardwareCapabilities.FeatureInfoEdx << 32 | (uint)HardwareCapabilities.FeatureInfoEcx;
+            return new FeatureInfo(
+                (uint)HardwareCapabilities.FeatureInfo1Ecx,
+                (uint)HardwareCapabilities.FeatureInfo1Edx,
+                (uint)HardwareCapabilities.FeatureInfo7Ebx);
         }
 
         private static byte GetMemoryManagerMode()
@@ -968,7 +971,7 @@ namespace ARMeilleure.Translation.PTC
             return osPlatform;
         }
 
-        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 50*/)]
+        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 54*/)]
         private struct OuterHeader
         {
             public ulong Magic;
@@ -976,7 +979,7 @@ namespace ARMeilleure.Translation.PTC
             public uint CacheFileVersion;
 
             public bool Endianness;
-            public ulong FeatureInfo;
+            public FeatureInfo FeatureInfo;
             public byte MemoryManagerMode;
             public uint OSPlatform;
 
@@ -999,6 +1002,9 @@ namespace ARMeilleure.Translation.PTC
             }
         }
 
+        [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 12*/)]
+        private record struct FeatureInfo(uint FeatureInfo0, uint FeatureInfo1, uint FeatureInfo2);
+
         [StructLayout(LayoutKind.Sequential, Pack = 1/*, Size = 128*/)]
         private struct InnerHeader
         {

Ryujinx.Ava/Assets/Locales/en_US.json

@@ -350,7 +350,7 @@
   "DialogProfileDeleteProfileTitle": "Deleting Profile",
   "DialogProfileDeleteProfileMessage": "This action is irreversible, are you sure you want to continue?",
   "DialogWarning": "Warning",
-  "DialogPPTCDeletionMessage": "You are about to delete the PPTC cache for :\n\n{0}\n\nAre you sure you want to proceed?",
+  "DialogPPTCDeletionMessage": "You are about to queue a PPTC rebuild on the next boot of:\n\n{0}\n\nAre you sure you want to proceed?",
   "DialogPPTCDeletionErrorMessage": "Error purging PPTC cache at {0}: {1}",
   "DialogShaderDeletionMessage": "You are about to delete the Shader cache for :\n\n{0}\n\nAre you sure you want to proceed?",
   "DialogShaderDeletionErrorMessage": "Error purging Shader cache at {0}: {1}",

Ryujinx.Ava/Ui/ViewModels/SettingsViewModel.cs

@@ -49,6 +49,7 @@ namespace Ryujinx.Ava.Ui.ViewModels
         private float _previousVolumeLevel;
         private float _volume;
         private bool _isVulkanAvailable = true;
+        private bool _directoryChanged = false;
         private List<string> _gpuIds = new List<string>();
         private KeyboardHotkeys _keyboardHotkeys;
         private int _graphicsBackendIndex;
@@ -112,6 +113,17 @@ namespace Ryujinx.Ava.Ui.ViewModels
             }
         }
 
+        public bool DirectoryChanged
+        {
+            get => _directoryChanged;
+            set
+            {
+                _directoryChanged = value;
+
+                OnPropertyChanged();
+            }
+        }
+
         public bool EnableDiscordIntegration { get; set; }
         public bool CheckUpdatesOnStart { get; set; }
         public bool ShowConfirmExit { get; set; }
@@ -397,10 +409,14 @@ namespace Ryujinx.Ava.Ui.ViewModels
 
         public async Task SaveSettings()
         {
-            List<string> gameDirs = new List<string>(GameDirectories);
-
             ConfigurationState config = ConfigurationState.Instance;
 
+            if (_directoryChanged)
+            {
+                List<string> gameDirs = new List<string>(GameDirectories);
+                config.Ui.GameDirs.Value = gameDirs;
+            }
+
             if (_validTzRegions.Contains(TimeZone))
             {
                 config.System.TimeZone.Value = TimeZone;
@@ -465,7 +481,6 @@ namespace Ryujinx.Ava.Ui.ViewModels
 
             config.System.SystemTimeOffset.Value = systemTimeOffset.Seconds;
             config.Graphics.ShadersDumpPath.Value = ShaderDumpPath;
-            config.Ui.GameDirs.Value = gameDirs;
             config.System.FsGlobalAccessLogMode.Value = FsGlobalAccessLogMode;
             config.System.MemoryManagerMode.Value = (MemoryManagerMode)MemoryMode;
 

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

@@ -162,6 +162,7 @@ namespace Ryujinx.Ava.Ui.Windows
             if (!string.IsNullOrWhiteSpace(path) && Directory.Exists(path) && !ViewModel.GameDirectories.Contains(path))
             {
                 ViewModel.GameDirectories.Add(path);
+                ViewModel.DirectoryChanged = true;
             }
             else
             {
@@ -170,6 +171,7 @@ namespace Ryujinx.Ava.Ui.Windows
                 if (!string.IsNullOrWhiteSpace(path))
                 {
                     ViewModel.GameDirectories.Add(path);
+                    ViewModel.DirectoryChanged = true;
                 }
             }
         }
@@ -181,6 +183,7 @@ namespace Ryujinx.Ava.Ui.Windows
             foreach (string path in selected)
             {
                 ViewModel.GameDirectories.Remove(path);
+                ViewModel.DirectoryChanged = true;
             }
         }
 
@@ -232,10 +235,12 @@ namespace Ryujinx.Ava.Ui.Windows
 
             ControllerSettings?.SaveCurrentProfile();
 
-            if (Owner is MainWindow window)
+            if (Owner is MainWindow window && ViewModel.DirectoryChanged)
             {
                 window.ViewModel.LoadApplications();
             }
+
+            ViewModel.DirectoryChanged = false;
         }
 
         protected override void OnClosed(EventArgs e)

Ryujinx.Graphics.GAL/ITexture.cs

@@ -19,6 +19,7 @@ namespace Ryujinx.Graphics.GAL
 
         void SetData(ReadOnlySpan<byte> data);
         void SetData(ReadOnlySpan<byte> data, int layer, int level);
+        void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region);
         void SetStorage(BufferRange buffer);
         void Release();
     }

Ryujinx.Graphics.GAL/Multithreading/CommandHelper.cs

@@ -113,6 +113,8 @@ namespace Ryujinx.Graphics.GAL.Multithreading
                 TextureSetDataCommand.Run(ref GetCommand<TextureSetDataCommand>(memory), threaded, renderer);
             _lookup[(int)CommandType.TextureSetDataSlice] = (Span<byte> memory, ThreadedRenderer threaded, IRenderer renderer) =>
                 TextureSetDataSliceCommand.Run(ref GetCommand<TextureSetDataSliceCommand>(memory), threaded, renderer);
+            _lookup[(int)CommandType.TextureSetDataSliceRegion] = (Span<byte> memory, ThreadedRenderer threaded, IRenderer renderer) =>
+                TextureSetDataSliceRegionCommand.Run(ref GetCommand<TextureSetDataSliceRegionCommand>(memory), threaded, renderer);
             _lookup[(int)CommandType.TextureSetStorage] = (Span<byte> memory, ThreadedRenderer threaded, IRenderer renderer) =>
                 TextureSetStorageCommand.Run(ref GetCommand<TextureSetStorageCommand>(memory), threaded, renderer);
 

Ryujinx.Graphics.GAL/Multithreading/CommandType.cs

@@ -37,6 +37,7 @@
         TextureRelease,
         TextureSetData,
         TextureSetDataSlice,
+        TextureSetDataSliceRegion,
         TextureSetStorage,
 
         WindowPresent,

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

@@ -0,0 +1,31 @@
+using Ryujinx.Graphics.GAL.Multithreading.Model;
+using Ryujinx.Graphics.GAL.Multithreading.Resources;
+using System;
+
+namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Texture
+{
+    struct TextureSetDataSliceRegionCommand : IGALCommand
+    {
+        public CommandType CommandType => CommandType.TextureSetDataSliceRegion;
+        private TableRef<ThreadedTexture> _texture;
+        private TableRef<byte[]> _data;
+        private int _layer;
+        private int _level;
+        private Rectangle<int> _region;
+
+        public void Set(TableRef<ThreadedTexture> texture, TableRef<byte[]> data, int layer, int level, Rectangle<int> region)
+        {
+            _texture = texture;
+            _data = data;
+            _layer = layer;
+            _level = level;
+            _region = region;
+        }
+
+        public static void Run(ref TextureSetDataSliceRegionCommand command, ThreadedRenderer threaded, IRenderer renderer)
+        {
+            ThreadedTexture texture = command._texture.Get(threaded);
+            texture.Base.SetData(new ReadOnlySpan<byte>(command._data.Get(threaded)), command._layer, command._level, command._region);
+        }
+    }
+}

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

@@ -119,6 +119,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             _renderer.QueueCommand();
         }
 
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            _renderer.New<TextureSetDataSliceRegionCommand>().Set(Ref(this), Ref(data.ToArray()), layer, level, region);
+            _renderer.QueueCommand();
+        }
+
         public void SetStorage(BufferRange buffer)
         {
             _renderer.New<TextureSetStorageCommand>().Set(Ref(this), buffer);

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

@@ -224,7 +224,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
                         xCount,
                         yCount,
                         dstLinear,
-                        dst.MemoryLayout);
+                        dst.MemoryLayout.UnpackGobBlocksInY(),
+                        dst.MemoryLayout.UnpackGobBlocksInZ());
 
                     if (target != null)
                     {

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

@@ -29,6 +29,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
         private int _dstHeight;
         private int _dstStride;
         private int _dstGobBlocksInY;
+        private int _dstGobBlocksInZ;
         private int _lineLengthIn;
         private int _lineCount;
 
@@ -117,6 +118,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
             _dstHeight = (int)state.SetDstHeight;
             _dstStride = (int)state.PitchOut;
             _dstGobBlocksInY = 1 << (int)state.SetDstBlockSizeHeight;
+            _dstGobBlocksInZ = 1 << (int)state.SetDstBlockSizeDepth;
             _lineLengthIn = (int)state.LineLengthIn;
             _lineCount = (int)state.LineCount;
 
@@ -176,6 +178,31 @@ namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory
             }
             else
             {
+                // TODO: Verify if the destination X/Y and width/height are taken into account
+                // for linear texture transfers. If not, we can use the fast path for that aswell.
+                // Right now the copy code at the bottom assumes that it is used on both which might be incorrect.
+                if (!_isLinear)
+                {
+                    var target = memoryManager.Physical.TextureCache.FindTexture(
+                        memoryManager,
+                        _dstGpuVa,
+                        1,
+                        _dstStride,
+                        _dstHeight,
+                        _lineLengthIn,
+                        _lineCount,
+                        _isLinear,
+                        _dstGobBlocksInY,
+                        _dstGobBlocksInZ);
+
+                    if (target != null)
+                    {
+                        target.SetData(data, 0, 0, new GAL.Rectangle<int>(_dstX, _dstY, _lineLengthIn / target.Info.FormatInfo.BytesPerPixel, _lineCount));
+
+                        return;
+                    }
+                }
+
                 var dstCalculator = new OffsetCalculator(
                     _dstWidth,
                     _dstHeight,

Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -761,6 +761,24 @@ namespace Ryujinx.Graphics.Gpu.Image
             _hasData = true;
         }
 
+        /// <summary>
+        /// Uploads new texture data to the host GPU for a specific layer/level and 2D sub-region.
+        /// </summary>
+        /// <param name="data">New data</param>
+        /// <param name="layer">Target layer</param>
+        /// <param name="level">Target level</param>
+        /// <param name="region">Target sub-region of the texture to update</param>
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            BlacklistScale();
+
+            HostTexture.SetData(data, layer, level, region);
+
+            _currentData = null;
+
+            _hasData = true;
+        }
+
         /// <summary>
         /// Converts texture data to a format and layout that is supported by the host GPU.
         /// </summary>

Ryujinx.Graphics.Gpu/Image/TextureCache.cs

@@ -905,7 +905,8 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="xCount">Number of pixels to be copied per line</param>
         /// <param name="yCount">Number of lines to be copied</param>
         /// <param name="linear">True if the texture has a linear layout, false otherwise</param>
-        /// <param name="memoryLayout">If <paramref name="linear"/> is false, should have the memory layout, otherwise ignored</param>
+        /// <param name="gobBlocksInY">If <paramref name="linear"/> is false, the amount of GOB blocks in the Y axis</param>
+        /// <param name="gobBlocksInZ">If <paramref name="linear"/> is false, the amount of GOB blocks in the Z axis</param>
         /// <returns>A matching texture, or null if there is no match</returns>
         public Texture FindTexture(
             MemoryManager memoryManager,
@@ -916,7 +917,8 @@ namespace Ryujinx.Graphics.Gpu.Image
             int xCount,
             int yCount,
             bool linear,
-            MemoryLayout memoryLayout)
+            int gobBlocksInY,
+            int gobBlocksInZ)
         {
             ulong address = memoryManager.Translate(gpuVa);
 
@@ -955,8 +957,8 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                     bool sizeMatch = xCount * bpp == texture.Info.Width * format.BytesPerPixel && height == texture.Info.Height;
                     bool formatMatch = !texture.Info.IsLinear &&
-                                        texture.Info.GobBlocksInY == memoryLayout.UnpackGobBlocksInY() &&
-                                        texture.Info.GobBlocksInZ == memoryLayout.UnpackGobBlocksInZ();
+                                        texture.Info.GobBlocksInY == gobBlocksInY &&
+                                        texture.Info.GobBlocksInZ == gobBlocksInZ;
 
                     match = sizeMatch && formatMatch;
                 }

Ryujinx.Graphics.OpenGL/Image/TextureBuffer.cs

@@ -58,6 +58,11 @@ namespace Ryujinx.Graphics.OpenGL.Image
             throw new NotSupportedException();
         }
 
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            throw new NotSupportedException();
+        }
+
         public void SetStorage(BufferRange buffer)
         {
             if (_buffer != BufferHandle.Null &&

Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -1,4 +1,5 @@
 using OpenTK.Graphics.OpenGL;
+using Ryujinx.Common;
 using Ryujinx.Graphics.GAL;
 using System;
 
@@ -385,7 +386,34 @@ namespace Ryujinx.Graphics.OpenGL.Image
                     int width = Math.Max(Info.Width >> level, 1);
                     int height = Math.Max(Info.Height >> level, 1);
 
-                    ReadFrom2D((IntPtr)ptr, layer, level, width, height);
+                    ReadFrom2D((IntPtr)ptr, layer, level, 0, 0, width, height);
+                }
+            }
+        }
+
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            if (Format == Format.S8UintD24Unorm)
+            {
+                data = FormatConverter.ConvertS8D24ToD24S8(data);
+            }
+
+            int wInBlocks = BitUtils.DivRoundUp(region.Width, Info.BlockWidth);
+            int hInBlocks = BitUtils.DivRoundUp(region.Height, Info.BlockHeight);
+
+            unsafe
+            {
+                fixed (byte* ptr = data)
+                {
+                    ReadFrom2D(
+                        (IntPtr)ptr,
+                        layer,
+                        level,
+                        region.X,
+                        region.Y,
+                        region.Width,
+                        region.Height,
+                        BitUtils.AlignUp(wInBlocks * Info.BytesPerPixel, 4) * hInBlocks);
                 }
             }
         }
@@ -397,15 +425,20 @@ namespace Ryujinx.Graphics.OpenGL.Image
 
         public void ReadFromPbo2D(int offset, int layer, int level, int width, int height)
         {
-            ReadFrom2D(IntPtr.Zero + offset, layer, level, width, height);
+            ReadFrom2D(IntPtr.Zero + offset, layer, level, 0, 0, width, height);
         }
 
-        private void ReadFrom2D(IntPtr data, int layer, int level, int width, int height)
+        private void ReadFrom2D(IntPtr data, int layer, int level, int x, int y, int width, int height)
+        {
+            int mipSize = Info.GetMipSize2D(level);
+
+            ReadFrom2D(data, layer, level, x, y, width, height, mipSize);
+        }
+
+        private void ReadFrom2D(IntPtr data, int layer, int level, int x, int y, int width, int height, int mipSize)
         {
             TextureTarget target = Target.Convert();
 
-            int mipSize = Info.GetMipSize2D(level);
-
             Bind(target, 0);
 
             FormatInfo format = FormatTable.GetFormatInfo(Info.Format);
@@ -418,7 +451,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.CompressedTexSubImage1D(
                             target,
                             level,
-                            0,
+                            x,
                             width,
                             format.PixelFormat,
                             mipSize,
@@ -429,7 +462,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.TexSubImage1D(
                             target,
                             level,
-                            0,
+                            x,
                             width,
                             format.PixelFormat,
                             format.PixelType,
@@ -443,7 +476,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.CompressedTexSubImage2D(
                             target,
                             level,
-                            0,
+                            x,
                             layer,
                             width,
                             1,
@@ -456,7 +489,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.TexSubImage2D(
                             target,
                             level,
-                            0,
+                            x,
                             layer,
                             width,
                             1,
@@ -472,8 +505,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.CompressedTexSubImage2D(
                             target,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             width,
                             height,
                             format.PixelFormat,
@@ -485,8 +518,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.TexSubImage2D(
                             target,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             width,
                             height,
                             format.PixelFormat,
@@ -503,8 +536,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.CompressedTexSubImage3D(
                             target,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             layer,
                             width,
                             height,
@@ -518,8 +551,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.TexSubImage3D(
                             target,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             layer,
                             width,
                             height,
@@ -536,8 +569,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.CompressedTexSubImage2D(
                             TextureTarget.TextureCubeMapPositiveX + layer,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             width,
                             height,
                             format.PixelFormat,
@@ -549,8 +582,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
                         GL.TexSubImage2D(
                             TextureTarget.TextureCubeMapPositiveX + layer,
                             level,
-                            0,
-                            0,
+                            x,
+                            y,
                             width,
                             height,
                             format.PixelFormat,

Ryujinx.Graphics.Vulkan/TextureBuffer.cs

@@ -98,6 +98,11 @@ namespace Ryujinx.Graphics.Vulkan
             throw new NotSupportedException();
         }
 
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            throw new NotSupportedException();
+        }
+
         public void SetStorage(BufferRange buffer)
         {
             if (_bufferHandle == buffer.Handle &&

Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -819,7 +819,7 @@ namespace Ryujinx.Graphics.Vulkan
                 var buffer = bufferHolder.GetBuffer(cbs.CommandBuffer).Get(cbs).Value;
                 var image = GetImage().Get(cbs).Value;
 
-                CopyFromOrToBuffer(cbs.CommandBuffer, buffer, image, size, true, x, y, width, height);
+                CopyFromOrToBuffer(cbs.CommandBuffer, buffer, image, size, true, 0, 0, x, y, width, height);
             }
 
             bufferHolder.WaitForFences();
@@ -893,7 +893,12 @@ namespace Ryujinx.Graphics.Vulkan
             SetData(data, layer, level, 1, 1, singleSlice: true);
         }
 
-        private void SetData(ReadOnlySpan<byte> data, int layer, int level, int layers, int levels, bool singleSlice)
+        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        {
+            SetData(data, layer, level, 1, 1, singleSlice: true, region);
+        }
+
+        private void SetData(ReadOnlySpan<byte> data, int layer, int level, int layers, int levels, bool singleSlice, Rectangle<int>? region = null)
         {
             int bufferDataLength = GetBufferDataLength(data.Length);
 
@@ -917,7 +922,25 @@ namespace Ryujinx.Graphics.Vulkan
             var buffer = bufferHolder.GetBuffer(cbs.CommandBuffer).Get(cbs).Value;
             var image = imageAuto.Get(cbs).Value;
 
-            CopyFromOrToBuffer(cbs.CommandBuffer, buffer, image, bufferDataLength, false, layer, level, layers, levels, singleSlice);
+            if (region.HasValue)
+            {
+                CopyFromOrToBuffer(
+                    cbs.CommandBuffer,
+                    buffer,
+                    image,
+                    bufferDataLength,
+                    false,
+                    layer,
+                    level,
+                    region.Value.X,
+                    region.Value.Y,
+                    region.Value.Width,
+                    region.Value.Height);
+            }
+            else
+            {
+                CopyFromOrToBuffer(cbs.CommandBuffer, buffer, image, bufferDataLength, false, layer, level, layers, levels, singleSlice);
+            }
         }
 
         private int GetBufferDataLength(int length)
@@ -1059,6 +1082,8 @@ namespace Ryujinx.Graphics.Vulkan
             Image image,
             int size,
             bool to,
+            int dstLayer,
+            int dstLevel,
             int x,
             int y,
             int width,
@@ -1071,13 +1096,21 @@ namespace Ryujinx.Graphics.Vulkan
                 aspectFlags = ImageAspectFlags.ImageAspectDepthBit;
             }
 
-            var sl = new ImageSubresourceLayers(aspectFlags, (uint)FirstLevel, (uint)FirstLayer, 1);
+            var sl = new ImageSubresourceLayers(aspectFlags, (uint)(FirstLevel + dstLevel), (uint)(FirstLayer + dstLayer), 1);
 
             var extent = new Extent3D((uint)width, (uint)height, 1);
 
+            int rowLengthAlignment = Info.BlockWidth;
+
+            // We expect all data being written into the texture to have a stride aligned by 4.
+            if (!to && Info.BytesPerPixel < 4)
+            {
+                rowLengthAlignment = 4 / Info.BytesPerPixel;
+            }
+
             var region = new BufferImageCopy(
                 0,
-                (uint)AlignUpNpot(width, Info.BlockWidth),
+                (uint)AlignUpNpot(width, rowLengthAlignment),
                 (uint)AlignUpNpot(height, Info.BlockHeight),
                 sl,
                 new Offset3D(x, y, 0),

Ryujinx.HLE/HOS/Services/Pctl/ParentalControlServiceFactory/IParentalControlService.cs

@@ -14,11 +14,14 @@ namespace Ryujinx.HLE.HOS.Services.Pctl.ParentalControlServiceFactory
         private ParentalControlFlagValue _parentalControlFlag;
         private int[]                    _ratingAge;
 
+#pragma warning disable CS0414
         // TODO: Find where they are set.
         private bool _restrictionEnabled                  = false;
         private bool _featuresRestriction                 = false;
+        private bool _freeCommunicationEnabled            = false;
         private bool _stereoVisionRestrictionConfigurable = true;
         private bool _stereoVisionRestriction             = false;
+#pragma warning restore CS0414
 
         public IParentalControlService(ServiceCtx context, ulong pid, bool withInitialize, int permissionFlag)
         {
@@ -88,13 +91,22 @@ namespace Ryujinx.HLE.HOS.Services.Pctl.ParentalControlServiceFactory
                 return ResultCode.FreeCommunicationDisabled;
             }
 
-            // NOTE: This sets an internal field to true. Usage have to be determined.
+            _freeCommunicationEnabled = true;
 
             Logger.Stub?.PrintStub(LogClass.ServicePctl);
 
             return ResultCode.Success;
         }
 
+        [CommandHipc(1017)] // 10.0.0+
+        // EndFreeCommunication()
+        public ResultCode EndFreeCommunication(ServiceCtx context)
+        {
+            _freeCommunicationEnabled = false;
+
+            return ResultCode.Success;
+        }
+
         [CommandHipc(1013)] // 4.0.0+
         // ConfirmStereoVisionPermission()
         public ResultCode ConfirmStereoVisionPermission(ServiceCtx context)

Ryujinx.HLE/HOS/Services/Sockets/Bsd/IClient.cs

@@ -219,9 +219,10 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
             int fdsCount = context.RequestData.ReadInt32();
             int timeout  = context.RequestData.ReadInt32();
 
-            (ulong bufferPosition, ulong bufferSize) = context.Request.GetBufferType0x21();
+            (ulong inputBufferPosition, ulong inputBufferSize) = context.Request.GetBufferType0x21();
+            (ulong outputBufferPosition, ulong outputBufferSize) = context.Request.GetBufferType0x22();
 
-            if (timeout < -1 || fdsCount < 0 || (ulong)(fdsCount * 8) > bufferSize)
+            if (timeout < -1 || fdsCount < 0 || (ulong)(fdsCount * 8) > inputBufferSize)
             {
                 return WriteBsdResult(context, -1, LinuxError.EINVAL);
             }
@@ -230,7 +231,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
 
             for (int i = 0; i < fdsCount; i++)
             {
-                PollEventData pollEventData = context.Memory.Read<PollEventData>(bufferPosition + (ulong)(i * Unsafe.SizeOf<PollEventData>()));
+                PollEventData pollEventData = context.Memory.Read<PollEventData>(inputBufferPosition + (ulong)(i * Unsafe.SizeOf<PollEventData>()));
 
                 IFileDescriptor fileDescriptor = _context.RetrieveFileDescriptor(pollEventData.SocketFd);
 
@@ -277,7 +278,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
             {
                 bool IsUnexpectedLinuxError(LinuxError error)
                 {
-                    return errno != LinuxError.SUCCESS && errno != LinuxError.ETIMEDOUT;
+                    return error != LinuxError.SUCCESS && error != LinuxError.ETIMEDOUT;
                 }
 
                 // Hybrid approach
@@ -332,7 +333,7 @@ namespace Ryujinx.HLE.HOS.Services.Sockets.Bsd
             // TODO: Spanify
             for (int i = 0; i < fdsCount; i++)
             {
-                context.Memory.Write(bufferPosition + (ulong)(i * Unsafe.SizeOf<PollEventData>()), events[i].Data);
+                context.Memory.Write(outputBufferPosition + (ulong)(i * Unsafe.SizeOf<PollEventData>()), events[i].Data);
             }
 
             return WriteBsdResult(context, updateCount, errno);

Ryujinx.HLE/Loaders/Mods/IPSwitchPatcher.cs

@@ -253,7 +253,7 @@ namespace Ryujinx.HLE.Loaders.Mods
 
                     if (tokens[1][0] == '"')
                     {
-                        var patch = Encoding.ASCII.GetBytes(tokens[1].Trim('"'));
+                        var patch = Encoding.ASCII.GetBytes(tokens[1].Trim('"') + "\0");
                         patches.Add((uint)offset, patch);
                     }
                     else

Ryujinx/Ui/Widgets/GameTableContextMenu.cs

@@ -522,7 +522,7 @@ namespace Ryujinx.Ui.Widgets
             DirectoryInfo mainDir   = new DirectoryInfo(System.IO.Path.Combine(AppDataManager.GamesDirPath, _titleIdText, "cache", "cpu", "0"));
             DirectoryInfo backupDir = new DirectoryInfo(System.IO.Path.Combine(AppDataManager.GamesDirPath, _titleIdText, "cache", "cpu", "1"));
 
-            MessageDialog warningDialog = GtkDialog.CreateConfirmationDialog("Warning", $"You are about to delete the PPTC cache for :\n\n<b>{_titleName}</b>\n\nAre you sure you want to proceed?");
+            MessageDialog warningDialog = GtkDialog.CreateConfirmationDialog("Warning", $"You are about to queue a PPTC rebuild on the next boot of:\n\n<b>{_titleName}</b>\n\nAre you sure you want to proceed?");
 
             List<FileInfo> cacheFiles = new List<FileInfo>();
 

Ryujinx/Ui/Windows/SettingsWindow.cs

@@ -34,6 +34,7 @@ namespace Ryujinx.Ui.Windows
 
         private long  _systemTimeOffset;
         private float _previousVolumeLevel;
+        private bool _directoryChanged = false;
 
 #pragma warning disable CS0649, IDE0044
         [GUI] CheckButton     _traceLogToggle;
@@ -501,14 +502,22 @@ namespace Ryujinx.Ui.Windows
 
         private void SaveSettings()
         {
-            List<string> gameDirs = new List<string>();
-
-            _gameDirsBoxStore.GetIterFirst(out TreeIter treeIter);
-            for (int i = 0; i < _gameDirsBoxStore.IterNChildren(); i++)
+            if (_directoryChanged)
             {
-                gameDirs.Add((string)_gameDirsBoxStore.GetValue(treeIter, 0));
+                List<string> gameDirs = new List<string>();
 
-                _gameDirsBoxStore.IterNext(ref treeIter);
+                _gameDirsBoxStore.GetIterFirst(out TreeIter treeIter);
+
+                for (int i = 0; i < _gameDirsBoxStore.IterNChildren(); i++)
+                {
+                    gameDirs.Add((string)_gameDirsBoxStore.GetValue(treeIter, 0));
+
+                    _gameDirsBoxStore.IterNext(ref treeIter);
+                }
+
+                ConfigurationState.Instance.Ui.GameDirs.Value = gameDirs;
+
+                _directoryChanged = false;
             }
 
             if (!float.TryParse(_resScaleText.Buffer.Text, out float resScaleCustom) || resScaleCustom <= 0.0f)
@@ -571,7 +580,6 @@ namespace Ryujinx.Ui.Windows
             ConfigurationState.Instance.System.SystemTimeOffset.Value             = _systemTimeOffset;
             ConfigurationState.Instance.Ui.CustomThemePath.Value                  = _custThemePath.Buffer.Text;
             ConfigurationState.Instance.Graphics.ShadersDumpPath.Value            = _graphicsShadersDumpPath.Buffer.Text;
-            ConfigurationState.Instance.Ui.GameDirs.Value                         = gameDirs;
             ConfigurationState.Instance.System.FsGlobalAccessLogMode.Value        = (int)_fsLogSpinAdjustment.Value;
             ConfigurationState.Instance.Graphics.MaxAnisotropy.Value              = float.Parse(_anisotropy.ActiveId, CultureInfo.InvariantCulture);
             ConfigurationState.Instance.Graphics.AspectRatio.Value                = Enum.Parse<AspectRatio>(_aspectRatio.ActiveId);
@@ -655,27 +663,27 @@ namespace Ryujinx.Ui.Windows
 
                 if (fileChooser.Run() == (int)ResponseType.Accept)
                 {
+                    _directoryChanged = false;
                     foreach (string directory in fileChooser.Filenames)
                     {
-                        bool directoryAdded = false;
-
                         if (_gameDirsBoxStore.GetIterFirst(out TreeIter treeIter))
                         {
                             do
                             {
                                 if (directory.Equals((string)_gameDirsBoxStore.GetValue(treeIter, 0)))
                                 {
-                                    directoryAdded = true;
                                     break;
                                 }
                             } while(_gameDirsBoxStore.IterNext(ref treeIter));
                         }
 
-                        if (!directoryAdded)
+                        if (!_directoryChanged)
                         {
                             _gameDirsBoxStore.AppendValues(directory);
                         }
                     }
+
+                    _directoryChanged = true;
                 }
 
                 fileChooser.Dispose();
@@ -693,6 +701,8 @@ namespace Ryujinx.Ui.Windows
             if (selection.GetSelected(out TreeIter treeIter))
             {
                 _gameDirsBoxStore.Remove(ref treeIter);
+
+                _directoryChanged = true;
             }
 
             ((ToggleButton)sender).SetStateFlags(StateFlags.Normal, true);