Move solution and projects to src

src/ARMeilleure/CodeGen/Arm64/Arm64Optimizer.cs

@@ -0,0 +1,270 @@
+using ARMeilleure.CodeGen.Optimizations;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System.Collections.Generic;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class Arm64Optimizer
+    {
+        private const int MaxConstantUses = 10000;
+
+        public static void RunPass(ControlFlowGraph cfg)
+        {
+            var constants = new Dictionary<ulong, Operand>();
+
+            Operand GetConstantCopy(BasicBlock block, Operation operation, Operand source)
+            {
+                // If the constant has many uses, we also force a new constant mov to be added, in order
+                // to avoid overflow of the counts field (that is limited to 16 bits).
+                if (!constants.TryGetValue(source.Value, out var constant) || constant.UsesCount > MaxConstantUses)
+                {
+                    constant = Local(source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, constant, source);
+
+                    block.Operations.AddBefore(operation, copyOp);
+
+                    constants[source.Value] = constant;
+                }
+
+                return constant;
+            }
+
+            for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
+            {
+                constants.Clear();
+
+                Operation nextNode;
+
+                for (Operation node = block.Operations.First; node != default; node = nextNode)
+                {
+                    nextNode = node.ListNext;
+
+                    // Insert copies for constants that can't fit on a 32-bit immediate.
+                    // Doing this early unblocks a few optimizations.
+                    if (node.Instruction == Instruction.Add)
+                    {
+                        Operand src1 = node.GetSource(0);
+                        Operand src2 = node.GetSource(1);
+
+                        if (src1.Kind == OperandKind.Constant && (src1.Relocatable || ConstTooLong(src1, OperandType.I32)))
+                        {
+                            node.SetSource(0, GetConstantCopy(block, node, src1));
+                        }
+
+                        if (src2.Kind == OperandKind.Constant && (src2.Relocatable || ConstTooLong(src2, OperandType.I32)))
+                        {
+                            node.SetSource(1, GetConstantCopy(block, node, src2));
+                        }
+                    }
+
+                    // Try to fold something like:
+                    //  lsl x1, x1, #2
+                    //  add x0, x0, x1
+                    //  ldr x0, [x0]
+                    //  add x2, x2, #16
+                    //  ldr x2, [x2]
+                    // Into:
+                    //  ldr x0, [x0, x1, lsl #2]
+                    //  ldr x2, [x2, #16]
+                    if (IsMemoryLoadOrStore(node.Instruction))
+                    {
+                        OperandType type;
+
+                        if (node.Destination != default)
+                        {
+                            type = node.Destination.Type;
+                        }
+                        else
+                        {
+                            type = node.GetSource(1).Type;
+                        }
+
+                        Operand memOp = GetMemoryOperandOrNull(node.GetSource(0), type);
+
+                        if (memOp != default)
+                        {
+                            node.SetSource(0, memOp);
+                        }
+                    }
+                }
+            }
+
+            Optimizer.RemoveUnusedNodes(cfg);
+        }
+
+        private static Operand GetMemoryOperandOrNull(Operand addr, OperandType type)
+        {
+            Operand baseOp = addr;
+
+            // First we check if the address is the result of a local X with immediate
+            // addition. If that is the case, then the baseOp is X, and the memory operand immediate
+            // becomes the addition immediate. Otherwise baseOp keeps being the address.
+            int imm = GetConstOp(ref baseOp, type);
+            if (imm != 0)
+            {
+                return MemoryOp(type, baseOp, default, Multiplier.x1, imm);
+            }
+
+            // Now we check if the baseOp is the result of a local Y with a local Z addition.
+            // If that is the case, we now set baseOp to Y and indexOp to Z. We further check
+            // if Z is the result of a left shift of local W by a value == 0 or == Log2(AccessSize),
+            // if that is the case, we set indexOp to W and adjust the scale value of the memory operand
+            // to match that of the left shift.
+            // There is one missed case, which is the address being a shift result, but this is
+            // probably not worth optimizing as it should never happen.
+            (Operand indexOp, Multiplier scale) = GetIndexOp(ref baseOp, type);
+
+            // If baseOp is still equal to address, then there's nothing that can be optimized.
+            if (baseOp == addr)
+            {
+                return default;
+            }
+
+            return MemoryOp(type, baseOp, indexOp, scale, 0);
+        }
+
+        private static int GetConstOp(ref Operand baseOp, OperandType accessType)
+        {
+            Operation operation = GetAsgOpWithInst(baseOp, Instruction.Add);
+
+            if (operation == default)
+            {
+                return 0;
+            }
+
+            Operand src1 = operation.GetSource(0);
+            Operand src2 = operation.GetSource(1);
+
+            Operand constOp;
+            Operand otherOp;
+
+            if (src1.Kind == OperandKind.Constant && src2.Kind == OperandKind.LocalVariable)
+            {
+                constOp = src1;
+                otherOp = src2;
+            }
+            else if (src1.Kind == OperandKind.LocalVariable && src2.Kind == OperandKind.Constant)
+            {
+                constOp = src2;
+                otherOp = src1;
+            }
+            else
+            {
+                return 0;
+            }
+
+            // If we have addition by a constant that we can't encode on the instruction,
+            // then we can't optimize it further.
+            if (ConstTooLong(constOp, accessType))
+            {
+                return 0;
+            }
+
+            baseOp = otherOp;
+
+            return constOp.AsInt32();
+        }
+
+        private static (Operand, Multiplier) GetIndexOp(ref Operand baseOp, OperandType accessType)
+        {
+            Operand indexOp = default;
+
+            Multiplier scale = Multiplier.x1;
+
+            Operation addOp = GetAsgOpWithInst(baseOp, Instruction.Add);
+
+            if (addOp == default)
+            {
+                return (indexOp, scale);
+            }
+
+            Operand src1 = addOp.GetSource(0);
+            Operand src2 = addOp.GetSource(1);
+
+            if (src1.Kind != OperandKind.LocalVariable || src2.Kind != OperandKind.LocalVariable)
+            {
+                return (indexOp, scale);
+            }
+
+            baseOp = src1;
+            indexOp = src2;
+
+            Operation shlOp = GetAsgOpWithInst(src1, Instruction.ShiftLeft);
+
+            bool indexOnSrc2 = false;
+
+            if (shlOp == default)
+            {
+                shlOp = GetAsgOpWithInst(src2, Instruction.ShiftLeft);
+
+                indexOnSrc2 = true;
+            }
+
+            if (shlOp != default)
+            {
+                Operand shSrc = shlOp.GetSource(0);
+                Operand shift = shlOp.GetSource(1);
+
+                int maxShift = Assembler.GetScaleForType(accessType);
+
+                if (shSrc.Kind == OperandKind.LocalVariable &&
+                    shift.Kind == OperandKind.Constant &&
+                    (shift.Value == 0 || shift.Value == (ulong)maxShift))
+                {
+                    scale = shift.Value switch
+                    {
+                        1 => Multiplier.x2,
+                        2 => Multiplier.x4,
+                        3 => Multiplier.x8,
+                        4 => Multiplier.x16,
+                        _ => Multiplier.x1
+                    };
+
+                    baseOp = indexOnSrc2 ? src1 : src2;
+                    indexOp = shSrc;
+                }
+            }
+
+            return (indexOp, scale);
+        }
+
+        private static Operation GetAsgOpWithInst(Operand op, Instruction inst)
+        {
+            // If we have multiple assignments, folding is not safe
+            // as the value may be different depending on the
+            // control flow path.
+            if (op.AssignmentsCount != 1)
+            {
+                return default;
+            }
+
+            Operation asgOp = op.Assignments[0];
+
+            if (asgOp.Instruction != inst)
+            {
+                return default;
+            }
+
+            return asgOp;
+        }
+
+        private static bool IsMemoryLoadOrStore(Instruction inst)
+        {
+            return inst == Instruction.Load || inst == Instruction.Store;
+        }
+
+        private static bool ConstTooLong(Operand constOp, OperandType accessType)
+        {
+            if ((uint)constOp.Value != constOp.Value)
+            {
+                return true;
+            }
+
+            return !CodeGenCommon.ConstFitsOnUImm12(constOp.AsInt32(), accessType);
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/ArmCondition.cs

@@ -0,0 +1,47 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    enum ArmCondition
+    {
+        Eq   = 0,
+        Ne   = 1,
+        GeUn = 2,
+        LtUn = 3,
+        Mi   = 4,
+        Pl   = 5,
+        Vs   = 6,
+        Vc   = 7,
+        GtUn = 8,
+        LeUn = 9,
+        Ge   = 10,
+        Lt   = 11,
+        Gt   = 12,
+        Le   = 13,
+        Al   = 14,
+        Nv   = 15
+    }
+
+    static class ComparisonArm64Extensions
+    {
+        public static ArmCondition ToArmCondition(this Comparison comp)
+        {
+            return comp switch
+            {
+                Comparison.Equal            => ArmCondition.Eq,
+                Comparison.NotEqual         => ArmCondition.Ne,
+                Comparison.Greater          => ArmCondition.Gt,
+                Comparison.LessOrEqual      => ArmCondition.Le,
+                Comparison.GreaterUI        => ArmCondition.GtUn,
+                Comparison.LessOrEqualUI    => ArmCondition.LeUn,
+                Comparison.GreaterOrEqual   => ArmCondition.Ge,
+                Comparison.Less             => ArmCondition.Lt,
+                Comparison.GreaterOrEqualUI => ArmCondition.GeUn,
+                Comparison.LessUI           => ArmCondition.LtUn,
+
+                _ => throw new ArgumentException(null, nameof(comp))
+            };
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/ArmExtensionType.cs

@@ -0,0 +1,14 @@
+namespace ARMeilleure.CodeGen.Arm64
+{
+    enum ArmExtensionType
+    {
+        Uxtb = 0,
+        Uxth = 1,
+        Uxtw = 2,
+        Uxtx = 3,
+        Sxtb = 4,
+        Sxth = 5,
+        Sxtw = 6,
+        Sxtx = 7
+    }
+}

src/ARMeilleure/CodeGen/Arm64/ArmShiftType.cs

@@ -0,0 +1,11 @@
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    enum ArmShiftType
+    {
+        Lsl = 0,
+        Lsr = 1,
+        Asr = 2,
+        Ror = 3
+    }
+}

src/ARMeilleure/CodeGen/Arm64/CallingConvention.cs

@@ -0,0 +1,96 @@
+using System;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class CallingConvention
+    {
+        private const int RegistersMask = unchecked((int)0xffffffff);
+
+        // Some of those register have specific roles and can't be used as general purpose registers.
+        // X18 - Reserved for platform specific usage.
+        // X29 - Frame pointer.
+        // X30 - Return address.
+        // X31 - Not an actual register, in some cases maps to SP, and in others to ZR.
+        private const int ReservedRegsMask = (1 << CodeGenCommon.ReservedRegister) | (1 << 18) | (1 << 29) | (1 << 30) | (1 << 31);
+
+        public static int GetIntAvailableRegisters()
+        {
+            return RegistersMask & ~ReservedRegsMask;
+        }
+
+        public static int GetVecAvailableRegisters()
+        {
+            return RegistersMask;
+        }
+
+        public static int GetIntCallerSavedRegisters()
+        {
+            return (GetIntCalleeSavedRegisters() ^ RegistersMask) & ~ReservedRegsMask;
+        }
+
+        public static int GetFpCallerSavedRegisters()
+        {
+            return GetFpCalleeSavedRegisters() ^ RegistersMask;
+        }
+
+        public static int GetVecCallerSavedRegisters()
+        {
+            return GetVecCalleeSavedRegisters() ^ RegistersMask;
+        }
+
+        public static int GetIntCalleeSavedRegisters()
+        {
+            return 0x1ff80000; // X19 to X28
+        }
+
+        public static int GetFpCalleeSavedRegisters()
+        {
+            return 0xff00; // D8 to D15
+        }
+
+        public static int GetVecCalleeSavedRegisters()
+        {
+            return 0;
+        }
+
+        public static int GetArgumentsOnRegsCount()
+        {
+            return 8;
+        }
+
+        public static int GetIntArgumentRegister(int index)
+        {
+            if ((uint)index < (uint)GetArgumentsOnRegsCount())
+            {
+                return index;
+            }
+
+            throw new ArgumentOutOfRangeException(nameof(index));
+        }
+
+        public static int GetVecArgumentRegister(int index)
+        {
+            if ((uint)index < (uint)GetArgumentsOnRegsCount())
+            {
+                return index;
+            }
+
+            throw new ArgumentOutOfRangeException(nameof(index));
+        }
+
+        public static int GetIntReturnRegister()
+        {
+            return 0;
+        }
+
+        public static int GetIntReturnRegisterHigh()
+        {
+            return 1;
+        }
+
+        public static int GetVecReturnRegister()
+        {
+            return 0;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/CodeGenCommon.cs

@@ -0,0 +1,91 @@
+using ARMeilleure.IntermediateRepresentation;
+using System.Numerics;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class CodeGenCommon
+    {
+        public const int TcAddressRegister = 8;
+        public const int ReservedRegister = 17;
+
+        public static bool ConstFitsOnSImm7(int value, int scale)
+        {
+            return (((value >> scale) << 25) >> (25 - scale)) == value;
+        }
+
+        public static bool ConstFitsOnSImm9(int value)
+        {
+            return ((value << 23) >> 23) == value;
+        }
+
+        public static bool ConstFitsOnUImm12(int value)
+        {
+            return (value & 0xfff) == value;
+        }
+
+        public static bool ConstFitsOnUImm12(int value, OperandType type)
+        {
+            int scale = Assembler.GetScaleForType(type);
+            return (((value >> scale) & 0xfff) << scale) == value;
+        }
+
+        public static bool TryEncodeBitMask(Operand operand, out int immN, out int immS, out int immR)
+        {
+            return TryEncodeBitMask(operand.Type, operand.Value, out immN, out immS, out immR);
+        }
+
+        public static bool TryEncodeBitMask(OperandType type, ulong value, out int immN, out int immS, out int immR)
+        {
+            if (type == OperandType.I32)
+            {
+                value |= value << 32;
+            }
+
+            return TryEncodeBitMask(value, out immN, out immS, out immR);
+        }
+
+        public static bool TryEncodeBitMask(ulong value, out int immN, out int immS, out int immR)
+        {
+            // Some special values also can't be encoded:
+            // 0 can't be encoded because we need to subtract 1 from onesCount (which would became negative if 0).
+            // A value with all bits set can't be encoded because it is reserved according to the spec, because:
+            // Any value AND all ones will be equal itself, so it's effectively a no-op.
+            // Any value OR all ones will be equal all ones, so one can just use MOV.
+            // Any value XOR all ones will be equal its inverse, so one can just use MVN.
+            if (value == 0 || value == ulong.MaxValue)
+            {
+                immN = 0;
+                immS = 0;
+                immR = 0;
+
+                return false;
+            }
+
+            // Normalize value, rotating it such that the LSB is 1: Ensures we get a complete element that has not
+            // been cut-in-half across the word boundary.
+            int rotation = BitOperations.TrailingZeroCount(value & (value + 1));
+            ulong rotatedValue = ulong.RotateRight(value, rotation);
+
+            // Now that we have a complete element in the LSB with the LSB = 1, determine size and number of ones
+            // in element.
+            int elementSize = BitOperations.TrailingZeroCount(rotatedValue & (rotatedValue + 1));
+            int onesInElement = BitOperations.TrailingZeroCount(~rotatedValue);
+
+            // Check the value is repeating; also ensures element size is a power of two.
+            if (ulong.RotateRight(value, elementSize) != value)
+            {
+                immN = 0;
+                immS = 0;
+                immR = 0;
+
+                return false;
+            }
+
+            immN = (elementSize >> 6) & 1;
+            immS = (((~elementSize + 1) << 1) | (onesInElement - 1)) & 0x3f;
+            immR = (elementSize - rotation) & (elementSize - 1);
+
+            return true;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/CodeGenContext.cs

@@ -0,0 +1,287 @@
+using ARMeilleure.CodeGen.Linking;
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using Ryujinx.Common.Memory;
+using System;
+using System.Collections.Generic;
+using System.IO;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    class CodeGenContext
+    {
+        private const int BccInstLength = 4;
+        private const int CbnzInstLength = 4;
+        private const int LdrLitInstLength = 4;
+
+        private Stream _stream;
+
+        public int StreamOffset => (int)_stream.Length;
+
+        public AllocationResult AllocResult { get; }
+
+        public Assembler Assembler { get; }
+
+        public BasicBlock CurrBlock { get; private set; }
+
+        public bool HasCall { get; }
+
+        public int CallArgsRegionSize { get; }
+        public int FpLrSaveRegionSize { get; }
+
+        private readonly Dictionary<BasicBlock, long> _visitedBlocks;
+        private readonly Dictionary<BasicBlock, List<(ArmCondition Condition, long BranchPos)>> _pendingBranches;
+
+        private struct ConstantPoolEntry
+        {
+            public readonly int Offset;
+            public readonly Symbol Symbol;
+            public readonly List<(Operand, int)> LdrOffsets;
+
+            public ConstantPoolEntry(int offset, Symbol symbol)
+            {
+                Offset = offset;
+                Symbol = symbol;
+                LdrOffsets = new List<(Operand, int)>();
+            }
+        }
+
+        private readonly Dictionary<ulong, ConstantPoolEntry> _constantPool;
+
+        private bool _constantPoolWritten;
+        private long _constantPoolOffset;
+
+        private ArmCondition _jNearCondition;
+        private Operand _jNearValue;
+
+        private long _jNearPosition;
+
+        private readonly bool _relocatable;
+
+        public CodeGenContext(AllocationResult allocResult, int maxCallArgs, int blocksCount, bool relocatable)
+        {
+            _stream = MemoryStreamManager.Shared.GetStream();
+
+            AllocResult = allocResult;
+
+            Assembler = new Assembler(_stream);
+
+            bool hasCall = maxCallArgs >= 0;
+
+            HasCall = hasCall;
+
+            if (maxCallArgs < 0)
+            {
+                maxCallArgs = 0;
+            }
+
+            CallArgsRegionSize = maxCallArgs * 16;
+            FpLrSaveRegionSize = hasCall ? 16 : 0;
+
+            _visitedBlocks = new Dictionary<BasicBlock, long>();
+            _pendingBranches = new Dictionary<BasicBlock, List<(ArmCondition, long)>>();
+            _constantPool = new Dictionary<ulong, ConstantPoolEntry>();
+
+            _relocatable = relocatable;
+        }
+
+        public void EnterBlock(BasicBlock block)
+        {
+            CurrBlock = block;
+
+            long target = _stream.Position;
+
+            if (_pendingBranches.TryGetValue(block, out var list))
+            {
+                foreach (var tuple in list)
+                {
+                    _stream.Seek(tuple.BranchPos, SeekOrigin.Begin);
+                    WriteBranch(tuple.Condition, target);
+                }
+
+                _stream.Seek(target, SeekOrigin.Begin);
+                _pendingBranches.Remove(block);
+            }
+
+            _visitedBlocks.Add(block, target);
+        }
+
+        public void JumpTo(BasicBlock target)
+        {
+            JumpTo(ArmCondition.Al, target);
+        }
+
+        public void JumpTo(ArmCondition condition, BasicBlock target)
+        {
+            if (_visitedBlocks.TryGetValue(target, out long offset))
+            {
+                WriteBranch(condition, offset);
+            }
+            else
+            {
+                if (!_pendingBranches.TryGetValue(target, out var list))
+                {
+                    list = new List<(ArmCondition, long)>();
+                    _pendingBranches.Add(target, list);
+                }
+
+                list.Add((condition, _stream.Position));
+
+                _stream.Seek(BccInstLength, SeekOrigin.Current);
+            }
+        }
+
+        private void WriteBranch(ArmCondition condition, long to)
+        {
+            int imm = checked((int)(to - _stream.Position));
+
+            if (condition != ArmCondition.Al)
+            {
+                Assembler.B(condition, imm);
+            }
+            else
+            {
+                Assembler.B(imm);
+            }
+        }
+
+        public void JumpToNear(ArmCondition condition)
+        {
+            _jNearCondition = condition;
+            _jNearPosition = _stream.Position;
+
+            _stream.Seek(BccInstLength, SeekOrigin.Current);
+        }
+
+        public void JumpToNearIfNotZero(Operand value)
+        {
+            _jNearValue = value;
+            _jNearPosition = _stream.Position;
+
+            _stream.Seek(CbnzInstLength, SeekOrigin.Current);
+        }
+
+        public void JumpHere()
+        {
+            long currentPosition = _stream.Position;
+            long offset = currentPosition - _jNearPosition;
+
+            _stream.Seek(_jNearPosition, SeekOrigin.Begin);
+
+            if (_jNearValue != default)
+            {
+                Assembler.Cbnz(_jNearValue, checked((int)offset));
+                _jNearValue = default;
+            }
+            else
+            {
+                Assembler.B(_jNearCondition, checked((int)offset));
+            }
+
+            _stream.Seek(currentPosition, SeekOrigin.Begin);
+        }
+
+        public void ReserveRelocatableConstant(Operand rt, Symbol symbol, ulong value)
+        {
+            if (!_constantPool.TryGetValue(value, out ConstantPoolEntry cpe))
+            {
+                cpe = new ConstantPoolEntry(_constantPool.Count * sizeof(ulong), symbol);
+                _constantPool.Add(value, cpe);
+            }
+
+            cpe.LdrOffsets.Add((rt, (int)_stream.Position));
+            _stream.Seek(LdrLitInstLength, SeekOrigin.Current);
+        }
+
+        private long WriteConstantPool()
+        {
+            if (_constantPoolWritten)
+            {
+                return _constantPoolOffset;
+            }
+
+            long constantPoolBaseOffset = _stream.Position;
+
+            foreach (ulong value in _constantPool.Keys)
+            {
+                WriteUInt64(value);
+            }
+
+            foreach (ConstantPoolEntry cpe in _constantPool.Values)
+            {
+                foreach ((Operand rt, int ldrOffset) in cpe.LdrOffsets)
+                {
+                    _stream.Seek(ldrOffset, SeekOrigin.Begin);
+
+                    int absoluteOffset = checked((int)(constantPoolBaseOffset + cpe.Offset));
+                    int pcRelativeOffset = absoluteOffset - ldrOffset;
+
+                    Assembler.LdrLit(rt, pcRelativeOffset);
+                }
+            }
+
+            _stream.Seek(constantPoolBaseOffset + _constantPool.Count * sizeof(ulong), SeekOrigin.Begin);
+
+            _constantPoolOffset = constantPoolBaseOffset;
+            _constantPoolWritten = true;
+
+            return constantPoolBaseOffset;
+        }
+
+        public (byte[], RelocInfo) GetCode()
+        {
+            long constantPoolBaseOffset = WriteConstantPool();
+
+            byte[] code = new byte[_stream.Length];
+
+            long originalPosition = _stream.Position;
+
+            _stream.Seek(0, SeekOrigin.Begin);
+            _stream.Read(code, 0, code.Length);
+            _stream.Seek(originalPosition, SeekOrigin.Begin);
+
+            RelocInfo relocInfo;
+
+            if (_relocatable)
+            {
+                RelocEntry[] relocs = new RelocEntry[_constantPool.Count];
+
+                int index = 0;
+
+                foreach (ConstantPoolEntry cpe in _constantPool.Values)
+                {
+                    if (cpe.Symbol.Type != SymbolType.None)
+                    {
+                        int absoluteOffset = checked((int)(constantPoolBaseOffset + cpe.Offset));
+                        relocs[index++] = new RelocEntry(absoluteOffset, cpe.Symbol);
+                    }
+                }
+
+                if (index != relocs.Length)
+                {
+                    Array.Resize(ref relocs, index);
+                }
+
+                relocInfo = new RelocInfo(relocs);
+            }
+            else
+            {
+                relocInfo = new RelocInfo(Array.Empty<RelocEntry>());
+            }
+
+            return (code, relocInfo);
+        }
+
+        private void WriteUInt64(ulong value)
+        {
+            _stream.WriteByte((byte)(value >> 0));
+            _stream.WriteByte((byte)(value >> 8));
+            _stream.WriteByte((byte)(value >> 16));
+            _stream.WriteByte((byte)(value >> 24));
+            _stream.WriteByte((byte)(value >> 32));
+            _stream.WriteByte((byte)(value >> 40));
+            _stream.WriteByte((byte)(value >> 48));
+            _stream.WriteByte((byte)(value >> 56));
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/CodeGeneratorIntrinsic.cs

@@ -0,0 +1,662 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using System.Diagnostics;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class CodeGeneratorIntrinsic
+    {
+        public static void GenerateOperation(CodeGenContext context, Operation operation)
+        {
+            Intrinsic intrin = operation.Intrinsic;
+
+            IntrinsicInfo info = IntrinsicTable.GetInfo(intrin & ~(Intrinsic.Arm64VTypeMask | Intrinsic.Arm64VSizeMask));
+
+            switch (info.Type)
+            {
+                case IntrinsicType.ScalarUnary:
+                    GenerateVectorUnary(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.ScalarUnaryByElem:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorUnaryByElem(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(1).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.ScalarBinary:
+                    GenerateVectorBinary(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.ScalarBinaryFPByElem:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryFPByElem(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(2).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.ScalarBinaryRd:
+                    GenerateVectorUnary(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.ScalarBinaryShl:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShlImm(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.ScalarBinaryShr:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.ScalarFPCompare:
+                    GenerateScalarFPCompare(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.ScalarFPConvFixed:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        0,
+                        ((uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift) + 2u,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.ScalarFPConvFixedGpr:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateScalarFPConvGpr(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.ScalarFPConvGpr:
+                    GenerateScalarFPConvGpr(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.ScalarTernary:
+                    GenerateScalarTernary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2),
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.ScalarTernaryFPRdByElem:
+                    Debug.Assert(operation.GetSource(3).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryFPByElem(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(3).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.ScalarTernaryShlRd:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShlImm(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        (uint)operation.GetSource(2).AsInt32());
+                    break;
+                case IntrinsicType.ScalarTernaryShrRd:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        0,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        (uint)operation.GetSource(2).AsInt32());
+                    break;
+
+                case IntrinsicType.VectorUnary:
+                    GenerateVectorUnary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.VectorUnaryByElem:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorUnaryByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(1).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(0));
+                    break;
+                case IntrinsicType.VectorBinary:
+                    GenerateVectorBinary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.VectorBinaryBitwise:
+                    GenerateVectorBinary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.VectorBinaryByElem:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(2).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.VectorBinaryFPByElem:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryFPByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(2).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.VectorBinaryRd:
+                    GenerateVectorUnary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1));
+                    break;
+                case IntrinsicType.VectorBinaryShl:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShlImm(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.VectorBinaryShr:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.VectorFPConvFixed:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        ((uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift) + 2u,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(0),
+                        (uint)operation.GetSource(1).AsInt32());
+                    break;
+                case IntrinsicType.VectorInsertByElem:
+                    Debug.Assert(operation.GetSource(1).Kind == OperandKind.Constant);
+                    Debug.Assert(operation.GetSource(3).Kind == OperandKind.Constant);
+
+                    GenerateVectorInsertByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(3).AsInt32(),
+                        (uint)operation.GetSource(1).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.VectorLookupTable:
+                    Debug.Assert((uint)(operation.SourcesCount - 2) <= 3);
+
+                    for (int i = 1; i < operation.SourcesCount - 1; i++)
+                    {
+                        Register currReg = operation.GetSource(i).GetRegister();
+                        Register prevReg = operation.GetSource(i - 1).GetRegister();
+
+                        Debug.Assert(prevReg.Index + 1 == currReg.Index && currReg.Type == RegisterType.Vector);
+                    }
+
+                    GenerateVectorBinary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        info.Inst | ((uint)(operation.SourcesCount - 2) << 13),
+                        operation.Destination,
+                        operation.GetSource(0),
+                        operation.GetSource(operation.SourcesCount - 1));
+                    break;
+                case IntrinsicType.VectorTernaryFPRdByElem:
+                    Debug.Assert(operation.GetSource(3).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryFPByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(3).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.VectorTernaryRd:
+                    GenerateVectorBinary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.VectorTernaryRdBitwise:
+                    GenerateVectorBinary(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.VectorTernaryRdByElem:
+                    Debug.Assert(operation.GetSource(3).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryByElem(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        (uint)operation.GetSource(3).AsInt32(),
+                        operation.Destination,
+                        operation.GetSource(1),
+                        operation.GetSource(2));
+                    break;
+                case IntrinsicType.VectorTernaryShlRd:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShlImm(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        (uint)operation.GetSource(2).AsInt32());
+                    break;
+                case IntrinsicType.VectorTernaryShrRd:
+                    Debug.Assert(operation.GetSource(2).Kind == OperandKind.Constant);
+
+                    GenerateVectorBinaryShrImm(
+                        context,
+                        (uint)(intrin & Intrinsic.Arm64VTypeMask) >> (int)Intrinsic.Arm64VTypeShift,
+                        (uint)(intrin & Intrinsic.Arm64VSizeMask) >> (int)Intrinsic.Arm64VSizeShift,
+                        info.Inst,
+                        operation.Destination,
+                        operation.GetSource(1),
+                        (uint)operation.GetSource(2).AsInt32());
+                    break;
+
+                case IntrinsicType.GetRegister:
+                    context.Assembler.WriteInstruction(info.Inst, operation.Destination);
+                    break;
+                case IntrinsicType.SetRegister:
+                    context.Assembler.WriteInstruction(info.Inst, operation.GetSource(0));
+                    break;
+
+                default:
+                    throw new NotImplementedException(info.Type.ToString());
+            }
+        }
+
+        private static void GenerateScalarFPCompare(
+            CodeGenContext context,
+            uint sz,
+            uint instruction,
+            Operand dest,
+            Operand rn,
+            Operand rm)
+        {
+            instruction |= (sz << 22);
+
+            if (rm.Kind == OperandKind.Constant && rm.Value == 0)
+            {
+                instruction |= 0b1000;
+                rm = rn;
+            }
+
+            context.Assembler.WriteInstructionRm16NoRet(instruction, rn, rm);
+            context.Assembler.Mrs(dest, 1, 3, 4, 2, 0);
+        }
+
+        private static void GenerateScalarFPConvGpr(
+            CodeGenContext context,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn)
+        {
+            instruction |= (sz << 22);
+
+            if (rd.Type.IsInteger())
+            {
+                context.Assembler.WriteInstructionAuto(instruction, rd, rn);
+            }
+            else
+            {
+                if (rn.Type == OperandType.I64)
+                {
+                    instruction |= Assembler.SfFlag;
+                }
+
+                context.Assembler.WriteInstruction(instruction, rd, rn);
+            }
+        }
+
+        private static void GenerateScalarFPConvGpr(
+            CodeGenContext context,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            uint fBits)
+        {
+            Debug.Assert(fBits <= 64);
+
+            instruction |= (sz << 22);
+            instruction |= (64 - fBits) << 10;
+
+            if (rd.Type.IsInteger())
+            {
+                Debug.Assert(rd.Type != OperandType.I32 || fBits <= 32);
+
+                context.Assembler.WriteInstructionAuto(instruction, rd, rn);
+            }
+            else
+            {
+                if (rn.Type == OperandType.I64)
+                {
+                    instruction |= Assembler.SfFlag;
+                }
+                else
+                {
+                    Debug.Assert(fBits <= 32);
+                }
+
+                context.Assembler.WriteInstruction(instruction, rd, rn);
+            }
+
+        }
+
+        private static void GenerateScalarTernary(
+            CodeGenContext context,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            Operand rm,
+            Operand ra)
+        {
+            instruction |= (sz << 22);
+
+            context.Assembler.WriteInstruction(instruction, rd, rn, rm, ra);
+        }
+
+        private static void GenerateVectorUnary(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn)
+        {
+            instruction |= (q << 30) | (sz << 22);
+
+            context.Assembler.WriteInstruction(instruction, rd, rn);
+        }
+
+        private static void GenerateVectorUnaryByElem(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            uint srcIndex,
+            Operand rd,
+            Operand rn)
+        {
+            uint imm5 = (srcIndex << ((int)sz + 1)) | (1u << (int)sz);
+
+            instruction |= (q << 30) | (imm5 << 16);
+
+            context.Assembler.WriteInstruction(instruction, rd, rn);
+        }
+
+        private static void GenerateVectorBinary(
+            CodeGenContext context,
+            uint q,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            Operand rm)
+        {
+            instruction |= (q << 30);
+
+            context.Assembler.WriteInstructionRm16(instruction, rd, rn, rm);
+        }
+
+        private static void GenerateVectorBinary(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            Operand rm)
+        {
+            instruction |= (q << 30) | (sz << 22);
+
+            context.Assembler.WriteInstructionRm16(instruction, rd, rn, rm);
+        }
+
+        private static void GenerateVectorBinaryByElem(
+            CodeGenContext context,
+            uint q,
+            uint size,
+            uint instruction,
+            uint srcIndex,
+            Operand rd,
+            Operand rn,
+            Operand rm)
+        {
+            instruction |= (q << 30) | (size << 22);
+
+            if (size == 2)
+            {
+                instruction |= ((srcIndex & 1) << 21) | ((srcIndex & 2) << 10);
+            }
+            else
+            {
+                instruction |= ((srcIndex & 3) << 20) | ((srcIndex & 4) << 9);
+            }
+
+            context.Assembler.WriteInstructionRm16(instruction, rd, rn, rm);
+        }
+
+        private static void GenerateVectorBinaryFPByElem(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            uint srcIndex,
+            Operand rd,
+            Operand rn,
+            Operand rm)
+        {
+            instruction |= (q << 30) | (sz << 22);
+
+            if (sz != 0)
+            {
+                instruction |= (srcIndex & 1) << 11;
+            }
+            else
+            {
+                instruction |= ((srcIndex & 1) << 21) | ((srcIndex & 2) << 10);
+            }
+
+            context.Assembler.WriteInstructionRm16(instruction, rd, rn, rm);
+        }
+
+        private static void GenerateVectorBinaryShlImm(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            uint shift)
+        {
+            instruction |= (q << 30);
+
+            Debug.Assert(shift >= 0 && shift < (8u << (int)sz));
+
+            uint imm = (8u << (int)sz) | (shift & (0x3fu >> (int)(3 - sz)));
+
+            instruction |= (imm << 16);
+
+            context.Assembler.WriteInstruction(instruction, rd, rn);
+        }
+
+        private static void GenerateVectorBinaryShrImm(
+            CodeGenContext context,
+            uint q,
+            uint sz,
+            uint instruction,
+            Operand rd,
+            Operand rn,
+            uint shift)
+        {
+            instruction |= (q << 30);
+
+            Debug.Assert(shift > 0 && shift <= (8u << (int)sz));
+
+            uint imm = (8u << (int)sz) | ((8u << (int)sz) - shift);
+
+            instruction |= (imm << 16);
+
+            context.Assembler.WriteInstruction(instruction, rd, rn);
+        }
+
+        private static void GenerateVectorInsertByElem(
+            CodeGenContext context,
+            uint sz,
+            uint instruction,
+            uint srcIndex,
+            uint dstIndex,
+            Operand rd,
+            Operand rn)
+        {
+            uint imm4 = srcIndex << (int)sz;
+            uint imm5 = (dstIndex << ((int)sz + 1)) | (1u << (int)sz);
+
+            instruction |= imm4 << 11;
+            instruction |= imm5 << 16;
+
+            context.Assembler.WriteInstruction(instruction, rd, rn);
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/HardwareCapabilities.cs

@@ -0,0 +1,185 @@
+using System;
+using System.Linq;
+using System.Reflection;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+using System.Runtime.Intrinsics.Arm;
+using System.Runtime.Versioning;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static partial class HardwareCapabilities
+    {
+        static HardwareCapabilities()
+        {
+            if (!ArmBase.Arm64.IsSupported)
+            {
+                return;
+            }
+
+            if (OperatingSystem.IsLinux())
+            {
+                LinuxFeatureInfoHwCap = (LinuxFeatureFlagsHwCap)getauxval(AT_HWCAP);
+                LinuxFeatureInfoHwCap2 = (LinuxFeatureFlagsHwCap2)getauxval(AT_HWCAP2);
+            }
+
+            if (OperatingSystem.IsMacOS())
+            {
+                for (int i = 0; i < _sysctlNames.Length; i++)
+                {
+                    if (CheckSysctlName(_sysctlNames[i]))
+                    {
+                        MacOsFeatureInfo |= (MacOsFeatureFlags)(1 << i);
+                    }
+                }
+            }
+        }
+
+#region Linux
+
+        private const ulong AT_HWCAP = 16;
+        private const ulong AT_HWCAP2 = 26;
+
+        [LibraryImport("libc", SetLastError = true)]
+        private static partial ulong getauxval(ulong type);
+
+        [Flags]
+        public enum LinuxFeatureFlagsHwCap : ulong
+        {
+            Fp        = 1 << 0,
+            Asimd     = 1 << 1,
+            Evtstrm   = 1 << 2,
+            Aes       = 1 << 3,
+            Pmull     = 1 << 4,
+            Sha1      = 1 << 5,
+            Sha2      = 1 << 6,
+            Crc32     = 1 << 7,
+            Atomics   = 1 << 8,
+            FpHp      = 1 << 9,
+            AsimdHp   = 1 << 10,
+            CpuId     = 1 << 11,
+            AsimdRdm  = 1 << 12,
+            Jscvt     = 1 << 13,
+            Fcma      = 1 << 14,
+            Lrcpc     = 1 << 15,
+            DcpOp     = 1 << 16,
+            Sha3      = 1 << 17,
+            Sm3       = 1 << 18,
+            Sm4       = 1 << 19,
+            AsimdDp   = 1 << 20,
+            Sha512    = 1 << 21,
+            Sve       = 1 << 22,
+            AsimdFhm  = 1 << 23,
+            Dit       = 1 << 24,
+            Uscat     = 1 << 25,
+            Ilrcpc    = 1 << 26,
+            FlagM     = 1 << 27,
+            Ssbs      = 1 << 28,
+            Sb        = 1 << 29,
+            Paca      = 1 << 30,
+            Pacg      = 1UL << 31
+        }
+
+        [Flags]
+        public enum LinuxFeatureFlagsHwCap2 : ulong
+        {
+            Dcpodp      = 1 << 0,
+            Sve2        = 1 << 1,
+            SveAes      = 1 << 2,
+            SvePmull    = 1 << 3,
+            SveBitperm  = 1 << 4,
+            SveSha3     = 1 << 5,
+            SveSm4      = 1 << 6,
+            FlagM2      = 1 << 7,
+            Frint       = 1 << 8,
+            SveI8mm     = 1 << 9,
+            SveF32mm    = 1 << 10,
+            SveF64mm    = 1 << 11,
+            SveBf16     = 1 << 12,
+            I8mm        = 1 << 13,
+            Bf16        = 1 << 14,
+            Dgh         = 1 << 15,
+            Rng         = 1 << 16,
+            Bti         = 1 << 17,
+            Mte         = 1 << 18,
+            Ecv         = 1 << 19,
+            Afp         = 1 << 20,
+            Rpres       = 1 << 21,
+            Mte3        = 1 << 22,
+            Sme         = 1 << 23,
+            Sme_i16i64  = 1 << 24,
+            Sme_f64f64  = 1 << 25,
+            Sme_i8i32   = 1 << 26,
+            Sme_f16f32  = 1 << 27,
+            Sme_b16f32  = 1 << 28,
+            Sme_f32f32  = 1 << 29,
+            Sme_fa64    = 1 << 30,
+            Wfxt        = 1UL << 31,
+            Ebf16       = 1UL << 32,
+            Sve_Ebf16   = 1UL << 33,
+            Cssc        = 1UL << 34,
+            Rprfm       = 1UL << 35,
+            Sve2p1      = 1UL << 36
+        }
+
+        public static LinuxFeatureFlagsHwCap LinuxFeatureInfoHwCap { get; } = 0;
+        public static LinuxFeatureFlagsHwCap2 LinuxFeatureInfoHwCap2 { get; } = 0;
+
+#endregion
+
+#region macOS
+
+        [LibraryImport("libSystem.dylib", SetLastError = true)]
+        private static unsafe partial int sysctlbyname([MarshalAs(UnmanagedType.LPStr)] string name, out int oldValue, ref ulong oldSize, IntPtr newValue, ulong newValueSize);
+
+        [SupportedOSPlatform("macos")]
+        private static bool CheckSysctlName(string name)
+        {
+            ulong size = sizeof(int);
+            if (sysctlbyname(name, out int val, ref size, IntPtr.Zero, 0) == 0 && size == sizeof(int))
+            {
+                return val != 0;
+            }
+            return false;
+        }
+
+        private static string[] _sysctlNames = new string[]
+        {
+            "hw.optional.floatingpoint",
+            "hw.optional.AdvSIMD",
+            "hw.optional.arm.FEAT_FP16",
+            "hw.optional.arm.FEAT_AES",
+            "hw.optional.arm.FEAT_PMULL",
+            "hw.optional.arm.FEAT_LSE",
+            "hw.optional.armv8_crc32",
+            "hw.optional.arm.FEAT_SHA1",
+            "hw.optional.arm.FEAT_SHA256"
+        };
+
+        [Flags]
+        public enum MacOsFeatureFlags
+        {
+            Fp      = 1 << 0,
+            AdvSimd = 1 << 1,
+            Fp16    = 1 << 2,
+            Aes     = 1 << 3,
+            Pmull   = 1 << 4,
+            Lse     = 1 << 5,
+            Crc32   = 1 << 6,
+            Sha1    = 1 << 7,
+            Sha256  = 1 << 8
+        }
+
+        public static MacOsFeatureFlags MacOsFeatureInfo { get; } = 0;
+
+#endregion
+
+        public static bool SupportsAdvSimd => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Asimd) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.AdvSimd);
+        public static bool SupportsAes => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Aes) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Aes);
+        public static bool SupportsPmull => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Pmull) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Pmull);
+        public static bool SupportsLse => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Atomics) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Lse);
+        public static bool SupportsCrc32 => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Crc32) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Crc32);
+        public static bool SupportsSha1 => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Sha1) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Sha1);
+        public static bool SupportsSha256 => LinuxFeatureInfoHwCap.HasFlag(LinuxFeatureFlagsHwCap.Sha2) || MacOsFeatureInfo.HasFlag(MacOsFeatureFlags.Sha256);
+    }
+}

src/ARMeilleure/CodeGen/Arm64/IntrinsicInfo.cs

@@ -0,0 +1,14 @@
+namespace ARMeilleure.CodeGen.Arm64
+{
+    struct IntrinsicInfo
+    {
+        public uint          Inst { get; }
+        public IntrinsicType Type { get; }
+
+        public IntrinsicInfo(uint inst, IntrinsicType type)
+        {
+            Inst = inst;
+            Type = type;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/IntrinsicTable.cs

@@ -0,0 +1,463 @@
+using ARMeilleure.Common;
+using ARMeilleure.IntermediateRepresentation;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class IntrinsicTable
+    {
+        private static IntrinsicInfo[] _intrinTable;
+
+        static IntrinsicTable()
+        {
+            _intrinTable = new IntrinsicInfo[EnumUtils.GetCount(typeof(Intrinsic))];
+
+            Add(Intrinsic.Arm64AbsS,          new IntrinsicInfo(0x5e20b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64AbsV,          new IntrinsicInfo(0x0e20b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64AddhnV,        new IntrinsicInfo(0x0e204000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64AddpS,         new IntrinsicInfo(0x5e31b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64AddpV,         new IntrinsicInfo(0x0e20bc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64AddvV,         new IntrinsicInfo(0x0e31b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64AddS,          new IntrinsicInfo(0x5e208400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64AddV,          new IntrinsicInfo(0x0e208400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64AesdV,         new IntrinsicInfo(0x4e285800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AeseV,         new IntrinsicInfo(0x4e284800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AesimcV,       new IntrinsicInfo(0x4e287800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AesmcV,        new IntrinsicInfo(0x4e286800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64AndV,          new IntrinsicInfo(0x0e201c00u, IntrinsicType.VectorBinaryBitwise));
+            Add(Intrinsic.Arm64BicVi,         new IntrinsicInfo(0x2f001400u, IntrinsicType.VectorBinaryBitwiseImm));
+            Add(Intrinsic.Arm64BicV,          new IntrinsicInfo(0x0e601c00u, IntrinsicType.VectorBinaryBitwise));
+            Add(Intrinsic.Arm64BifV,          new IntrinsicInfo(0x2ee01c00u, IntrinsicType.VectorTernaryRdBitwise));
+            Add(Intrinsic.Arm64BitV,          new IntrinsicInfo(0x2ea01c00u, IntrinsicType.VectorTernaryRdBitwise));
+            Add(Intrinsic.Arm64BslV,          new IntrinsicInfo(0x2e601c00u, IntrinsicType.VectorTernaryRdBitwise));
+            Add(Intrinsic.Arm64ClsV,          new IntrinsicInfo(0x0e204800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64ClzV,          new IntrinsicInfo(0x2e204800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmeqS,         new IntrinsicInfo(0x7e208c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmeqV,         new IntrinsicInfo(0x2e208c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CmeqSz,        new IntrinsicInfo(0x5e209800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64CmeqVz,        new IntrinsicInfo(0x0e209800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmgeS,         new IntrinsicInfo(0x5e203c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmgeV,         new IntrinsicInfo(0x0e203c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CmgeSz,        new IntrinsicInfo(0x7e208800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64CmgeVz,        new IntrinsicInfo(0x2e208800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmgtS,         new IntrinsicInfo(0x5e203400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmgtV,         new IntrinsicInfo(0x0e203400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CmgtSz,        new IntrinsicInfo(0x5e208800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64CmgtVz,        new IntrinsicInfo(0x0e208800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmhiS,         new IntrinsicInfo(0x7e203400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmhiV,         new IntrinsicInfo(0x2e203400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CmhsS,         new IntrinsicInfo(0x7e203c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmhsV,         new IntrinsicInfo(0x2e203c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CmleSz,        new IntrinsicInfo(0x7e209800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64CmleVz,        new IntrinsicInfo(0x2e209800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmltSz,        new IntrinsicInfo(0x5e20a800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64CmltVz,        new IntrinsicInfo(0x0e20a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64CmtstS,        new IntrinsicInfo(0x5e208c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64CmtstV,        new IntrinsicInfo(0x0e208c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64CntV,          new IntrinsicInfo(0x0e205800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64DupSe,         new IntrinsicInfo(0x5e000400u, IntrinsicType.ScalarUnaryByElem));
+            Add(Intrinsic.Arm64DupVe,         new IntrinsicInfo(0x0e000400u, IntrinsicType.VectorUnaryByElem));
+            Add(Intrinsic.Arm64DupGp,         new IntrinsicInfo(0x0e000c00u, IntrinsicType.VectorUnaryByElem));
+            Add(Intrinsic.Arm64EorV,          new IntrinsicInfo(0x2e201c00u, IntrinsicType.VectorBinaryBitwise));
+            Add(Intrinsic.Arm64ExtV,          new IntrinsicInfo(0x2e000000u, IntrinsicType.VectorExt));
+            Add(Intrinsic.Arm64FabdS,         new IntrinsicInfo(0x7ea0d400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FabdV,         new IntrinsicInfo(0x2ea0d400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FabsV,         new IntrinsicInfo(0x0ea0f800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FabsS,         new IntrinsicInfo(0x1e20c000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FacgeS,        new IntrinsicInfo(0x7e20ec00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FacgeV,        new IntrinsicInfo(0x2e20ec00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FacgtS,        new IntrinsicInfo(0x7ea0ec00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FacgtV,        new IntrinsicInfo(0x2ea0ec00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FaddpS,        new IntrinsicInfo(0x7e30d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FaddpV,        new IntrinsicInfo(0x2e20d400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FaddV,         new IntrinsicInfo(0x0e20d400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FaddS,         new IntrinsicInfo(0x1e202800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FccmpeS,       new IntrinsicInfo(0x1e200410u, IntrinsicType.ScalarFPCompareCond));
+            Add(Intrinsic.Arm64FccmpS,        new IntrinsicInfo(0x1e200400u, IntrinsicType.ScalarFPCompareCond));
+            Add(Intrinsic.Arm64FcmeqS,        new IntrinsicInfo(0x5e20e400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FcmeqV,        new IntrinsicInfo(0x0e20e400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FcmeqSz,       new IntrinsicInfo(0x5ea0d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcmeqVz,       new IntrinsicInfo(0x0ea0d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcmgeS,        new IntrinsicInfo(0x7e20e400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FcmgeV,        new IntrinsicInfo(0x2e20e400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FcmgeSz,       new IntrinsicInfo(0x7ea0c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcmgeVz,       new IntrinsicInfo(0x2ea0c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcmgtS,        new IntrinsicInfo(0x7ea0e400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FcmgtV,        new IntrinsicInfo(0x2ea0e400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FcmgtSz,       new IntrinsicInfo(0x5ea0c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcmgtVz,       new IntrinsicInfo(0x0ea0c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcmleSz,       new IntrinsicInfo(0x7ea0d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcmleVz,       new IntrinsicInfo(0x2ea0d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcmltSz,       new IntrinsicInfo(0x5ea0e800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcmltVz,       new IntrinsicInfo(0x0ea0e800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcmpeS,        new IntrinsicInfo(0x1e202010u, IntrinsicType.ScalarFPCompare));
+            Add(Intrinsic.Arm64FcmpS,         new IntrinsicInfo(0x1e202000u, IntrinsicType.ScalarFPCompare));
+            Add(Intrinsic.Arm64FcselS,        new IntrinsicInfo(0x1e200c00u, IntrinsicType.ScalarFcsel));
+            Add(Intrinsic.Arm64FcvtasS,       new IntrinsicInfo(0x5e21c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtasV,       new IntrinsicInfo(0x0e21c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtasGp,      new IntrinsicInfo(0x1e240000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtauS,       new IntrinsicInfo(0x7e21c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtauV,       new IntrinsicInfo(0x2e21c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtauGp,      new IntrinsicInfo(0x1e250000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtlV,        new IntrinsicInfo(0x0e217800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtmsS,       new IntrinsicInfo(0x5e21b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtmsV,       new IntrinsicInfo(0x0e21b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtmsGp,      new IntrinsicInfo(0x1e300000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtmuS,       new IntrinsicInfo(0x7e21b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtmuV,       new IntrinsicInfo(0x2e21b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtmuGp,      new IntrinsicInfo(0x1e310000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtnsS,       new IntrinsicInfo(0x5e21a800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtnsV,       new IntrinsicInfo(0x0e21a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtnsGp,      new IntrinsicInfo(0x1e200000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtnuS,       new IntrinsicInfo(0x7e21a800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtnuV,       new IntrinsicInfo(0x2e21a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtnuGp,      new IntrinsicInfo(0x1e210000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtnV,        new IntrinsicInfo(0x0e216800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64FcvtpsS,       new IntrinsicInfo(0x5ea1a800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtpsV,       new IntrinsicInfo(0x0ea1a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtpsGp,      new IntrinsicInfo(0x1e280000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtpuS,       new IntrinsicInfo(0x7ea1a800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtpuV,       new IntrinsicInfo(0x2ea1a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtpuGp,      new IntrinsicInfo(0x1e290000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtxnS,       new IntrinsicInfo(0x7e216800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtxnV,       new IntrinsicInfo(0x2e216800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtzsSFixed,  new IntrinsicInfo(0x5f00fc00u, IntrinsicType.ScalarFPConvFixed));
+            Add(Intrinsic.Arm64FcvtzsVFixed,  new IntrinsicInfo(0x0f00fc00u, IntrinsicType.VectorFPConvFixed));
+            Add(Intrinsic.Arm64FcvtzsS,       new IntrinsicInfo(0x5ea1b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtzsV,       new IntrinsicInfo(0x0ea1b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtzsGpFixed, new IntrinsicInfo(0x1e180000u, IntrinsicType.ScalarFPConvFixedGpr));
+            Add(Intrinsic.Arm64FcvtzsGp,      new IntrinsicInfo(0x1e380000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtzuSFixed,  new IntrinsicInfo(0x7f00fc00u, IntrinsicType.ScalarFPConvFixed));
+            Add(Intrinsic.Arm64FcvtzuVFixed,  new IntrinsicInfo(0x2f00fc00u, IntrinsicType.VectorFPConvFixed));
+            Add(Intrinsic.Arm64FcvtzuS,       new IntrinsicInfo(0x7ea1b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FcvtzuV,       new IntrinsicInfo(0x2ea1b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FcvtzuGpFixed, new IntrinsicInfo(0x1e190000u, IntrinsicType.ScalarFPConvFixedGpr));
+            Add(Intrinsic.Arm64FcvtzuGp,      new IntrinsicInfo(0x1e390000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FcvtS,         new IntrinsicInfo(0x1e224000u, IntrinsicType.ScalarFPConv));
+            Add(Intrinsic.Arm64FdivV,         new IntrinsicInfo(0x2e20fc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FdivS,         new IntrinsicInfo(0x1e201800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FmaddS,        new IntrinsicInfo(0x1f000000u, IntrinsicType.ScalarTernary));
+            Add(Intrinsic.Arm64FmaxnmpS,      new IntrinsicInfo(0x7e30c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FmaxnmpV,      new IntrinsicInfo(0x2e20c400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmaxnmvV,      new IntrinsicInfo(0x2e30c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FmaxnmV,       new IntrinsicInfo(0x0e20c400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmaxnmS,       new IntrinsicInfo(0x1e206800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FmaxpS,        new IntrinsicInfo(0x7e30f800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FmaxpV,        new IntrinsicInfo(0x2e20f400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmaxvV,        new IntrinsicInfo(0x2e30f800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FmaxV,         new IntrinsicInfo(0x0e20f400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmaxS,         new IntrinsicInfo(0x1e204800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FminnmpS,      new IntrinsicInfo(0x7eb0c800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FminnmpV,      new IntrinsicInfo(0x2ea0c400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FminnmvV,      new IntrinsicInfo(0x2eb0c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FminnmV,       new IntrinsicInfo(0x0ea0c400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FminnmS,       new IntrinsicInfo(0x1e207800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FminpS,        new IntrinsicInfo(0x7eb0f800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FminpV,        new IntrinsicInfo(0x2ea0f400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FminvV,        new IntrinsicInfo(0x2eb0f800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FminV,         new IntrinsicInfo(0x0ea0f400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FminS,         new IntrinsicInfo(0x1e205800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FmlaSe,        new IntrinsicInfo(0x5f801000u, IntrinsicType.ScalarTernaryFPRdByElem));
+            Add(Intrinsic.Arm64FmlaVe,        new IntrinsicInfo(0x0f801000u, IntrinsicType.VectorTernaryFPRdByElem));
+            Add(Intrinsic.Arm64FmlaV,         new IntrinsicInfo(0x0e20cc00u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64FmlsSe,        new IntrinsicInfo(0x5f805000u, IntrinsicType.ScalarTernaryFPRdByElem));
+            Add(Intrinsic.Arm64FmlsVe,        new IntrinsicInfo(0x0f805000u, IntrinsicType.VectorTernaryFPRdByElem));
+            Add(Intrinsic.Arm64FmlsV,         new IntrinsicInfo(0x0ea0cc00u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64FmovVi,        new IntrinsicInfo(0x0f00f400u, IntrinsicType.VectorFmovi));
+            Add(Intrinsic.Arm64FmovS,         new IntrinsicInfo(0x1e204000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FmovGp,        new IntrinsicInfo(0x1e260000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64FmovSi,        new IntrinsicInfo(0x1e201000u, IntrinsicType.ScalarFmovi));
+            Add(Intrinsic.Arm64FmsubS,        new IntrinsicInfo(0x1f008000u, IntrinsicType.ScalarTernary));
+            Add(Intrinsic.Arm64FmulxSe,       new IntrinsicInfo(0x7f809000u, IntrinsicType.ScalarBinaryFPByElem));
+            Add(Intrinsic.Arm64FmulxVe,       new IntrinsicInfo(0x2f809000u, IntrinsicType.VectorBinaryFPByElem));
+            Add(Intrinsic.Arm64FmulxS,        new IntrinsicInfo(0x5e20dc00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FmulxV,        new IntrinsicInfo(0x0e20dc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmulSe,        new IntrinsicInfo(0x5f809000u, IntrinsicType.ScalarBinaryFPByElem));
+            Add(Intrinsic.Arm64FmulVe,        new IntrinsicInfo(0x0f809000u, IntrinsicType.VectorBinaryFPByElem));
+            Add(Intrinsic.Arm64FmulV,         new IntrinsicInfo(0x2e20dc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FmulS,         new IntrinsicInfo(0x1e200800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FnegV,         new IntrinsicInfo(0x2ea0f800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FnegS,         new IntrinsicInfo(0x1e214000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FnmaddS,       new IntrinsicInfo(0x1f200000u, IntrinsicType.ScalarTernary));
+            Add(Intrinsic.Arm64FnmsubS,       new IntrinsicInfo(0x1f208000u, IntrinsicType.ScalarTernary));
+            Add(Intrinsic.Arm64FnmulS,        new IntrinsicInfo(0x1e208800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FrecpeS,       new IntrinsicInfo(0x5ea1d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrecpeV,       new IntrinsicInfo(0x0ea1d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrecpsS,       new IntrinsicInfo(0x5e20fc00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FrecpsV,       new IntrinsicInfo(0x0e20fc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FrecpxS,       new IntrinsicInfo(0x5ea1f800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintaV,       new IntrinsicInfo(0x2e218800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintaS,       new IntrinsicInfo(0x1e264000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintiV,       new IntrinsicInfo(0x2ea19800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintiS,       new IntrinsicInfo(0x1e27c000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintmV,       new IntrinsicInfo(0x0e219800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintmS,       new IntrinsicInfo(0x1e254000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintnV,       new IntrinsicInfo(0x0e218800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintnS,       new IntrinsicInfo(0x1e244000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintpV,       new IntrinsicInfo(0x0ea18800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintpS,       new IntrinsicInfo(0x1e24c000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintxV,       new IntrinsicInfo(0x2e219800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintxS,       new IntrinsicInfo(0x1e274000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrintzV,       new IntrinsicInfo(0x0ea19800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrintzS,       new IntrinsicInfo(0x1e25c000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrsqrteS,      new IntrinsicInfo(0x7ea1d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FrsqrteV,      new IntrinsicInfo(0x2ea1d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FrsqrtsS,      new IntrinsicInfo(0x5ea0fc00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64FrsqrtsV,      new IntrinsicInfo(0x0ea0fc00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FsqrtV,        new IntrinsicInfo(0x2ea1f800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64FsqrtS,        new IntrinsicInfo(0x1e21c000u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64FsubV,         new IntrinsicInfo(0x0ea0d400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64FsubS,         new IntrinsicInfo(0x1e203800u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64InsVe,         new IntrinsicInfo(0x6e000400u, IntrinsicType.VectorInsertByElem));
+            Add(Intrinsic.Arm64InsGp,         new IntrinsicInfo(0x4e001c00u, IntrinsicType.ScalarUnaryByElem));
+            Add(Intrinsic.Arm64Ld1rV,         new IntrinsicInfo(0x0d40c000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld1Vms,        new IntrinsicInfo(0x0c402000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld1Vss,        new IntrinsicInfo(0x0d400000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64Ld2rV,         new IntrinsicInfo(0x0d60c000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld2Vms,        new IntrinsicInfo(0x0c408000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld2Vss,        new IntrinsicInfo(0x0d600000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64Ld3rV,         new IntrinsicInfo(0x0d40e000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld3Vms,        new IntrinsicInfo(0x0c404000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld3Vss,        new IntrinsicInfo(0x0d402000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64Ld4rV,         new IntrinsicInfo(0x0d60e000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld4Vms,        new IntrinsicInfo(0x0c400000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64Ld4Vss,        new IntrinsicInfo(0x0d602000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64MlaVe,         new IntrinsicInfo(0x2f000000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64MlaV,          new IntrinsicInfo(0x0e209400u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64MlsVe,         new IntrinsicInfo(0x2f004000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64MlsV,          new IntrinsicInfo(0x2e209400u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64MoviV,         new IntrinsicInfo(0x0f000400u, IntrinsicType.VectorMovi));
+            Add(Intrinsic.Arm64MrsFpcr,       new IntrinsicInfo(0xd53b4400u, IntrinsicType.GetRegister));
+            Add(Intrinsic.Arm64MsrFpcr,       new IntrinsicInfo(0xd51b4400u, IntrinsicType.SetRegister));
+            Add(Intrinsic.Arm64MrsFpsr,       new IntrinsicInfo(0xd53b4420u, IntrinsicType.GetRegister));
+            Add(Intrinsic.Arm64MsrFpsr,       new IntrinsicInfo(0xd51b4420u, IntrinsicType.SetRegister));
+            Add(Intrinsic.Arm64MulVe,         new IntrinsicInfo(0x0f008000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64MulV,          new IntrinsicInfo(0x0e209c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64MvniV,         new IntrinsicInfo(0x2f000400u, IntrinsicType.VectorMvni));
+            Add(Intrinsic.Arm64NegS,          new IntrinsicInfo(0x7e20b800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64NegV,          new IntrinsicInfo(0x2e20b800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64NotV,          new IntrinsicInfo(0x2e205800u, IntrinsicType.VectorUnaryBitwise));
+            Add(Intrinsic.Arm64OrnV,          new IntrinsicInfo(0x0ee01c00u, IntrinsicType.VectorBinaryBitwise));
+            Add(Intrinsic.Arm64OrrVi,         new IntrinsicInfo(0x0f001400u, IntrinsicType.VectorBinaryBitwiseImm));
+            Add(Intrinsic.Arm64OrrV,          new IntrinsicInfo(0x0ea01c00u, IntrinsicType.VectorBinaryBitwise));
+            Add(Intrinsic.Arm64PmullV,        new IntrinsicInfo(0x0e20e000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64PmulV,         new IntrinsicInfo(0x2e209c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64RaddhnV,       new IntrinsicInfo(0x2e204000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64RbitV,         new IntrinsicInfo(0x2e605800u, IntrinsicType.VectorUnaryBitwise));
+            Add(Intrinsic.Arm64Rev16V,        new IntrinsicInfo(0x0e201800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64Rev32V,        new IntrinsicInfo(0x2e200800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64Rev64V,        new IntrinsicInfo(0x0e200800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64RshrnV,        new IntrinsicInfo(0x0f008c00u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64RsubhnV,       new IntrinsicInfo(0x2e206000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SabalV,        new IntrinsicInfo(0x0e205000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SabaV,         new IntrinsicInfo(0x0e207c00u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SabdlV,        new IntrinsicInfo(0x0e207000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SabdV,         new IntrinsicInfo(0x0e207400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SadalpV,       new IntrinsicInfo(0x0e206800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64SaddlpV,       new IntrinsicInfo(0x0e202800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SaddlvV,       new IntrinsicInfo(0x0e303800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SaddlV,        new IntrinsicInfo(0x0e200000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SaddwV,        new IntrinsicInfo(0x0e201000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64ScvtfSFixed,   new IntrinsicInfo(0x5f00e400u, IntrinsicType.ScalarFPConvFixed));
+            Add(Intrinsic.Arm64ScvtfVFixed,   new IntrinsicInfo(0x0f00e400u, IntrinsicType.VectorFPConvFixed));
+            Add(Intrinsic.Arm64ScvtfS,        new IntrinsicInfo(0x5e21d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64ScvtfV,        new IntrinsicInfo(0x0e21d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64ScvtfGpFixed,  new IntrinsicInfo(0x1e020000u, IntrinsicType.ScalarFPConvFixedGpr));
+            Add(Intrinsic.Arm64ScvtfGp,       new IntrinsicInfo(0x1e220000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64Sha1cV,        new IntrinsicInfo(0x5e000000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha1hV,        new IntrinsicInfo(0x5e280800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64Sha1mV,        new IntrinsicInfo(0x5e002000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha1pV,        new IntrinsicInfo(0x5e001000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha1su0V,      new IntrinsicInfo(0x5e003000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha1su1V,      new IntrinsicInfo(0x5e281800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64Sha256h2V,     new IntrinsicInfo(0x5e005000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha256hV,      new IntrinsicInfo(0x5e004000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64Sha256su0V,    new IntrinsicInfo(0x5e282800u, IntrinsicType.Vector128Unary));
+            Add(Intrinsic.Arm64Sha256su1V,    new IntrinsicInfo(0x5e006000u, IntrinsicType.Vector128Binary));
+            Add(Intrinsic.Arm64ShaddV,        new IntrinsicInfo(0x0e200400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64ShllV,         new IntrinsicInfo(0x2e213800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64ShlS,          new IntrinsicInfo(0x5f005400u, IntrinsicType.ScalarBinaryShl));
+            Add(Intrinsic.Arm64ShlV,          new IntrinsicInfo(0x0f005400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64ShrnV,         new IntrinsicInfo(0x0f008400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64ShsubV,        new IntrinsicInfo(0x0e202400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SliS,          new IntrinsicInfo(0x7f005400u, IntrinsicType.ScalarTernaryShlRd));
+            Add(Intrinsic.Arm64SliV,          new IntrinsicInfo(0x2f005400u, IntrinsicType.VectorTernaryShlRd));
+            Add(Intrinsic.Arm64SmaxpV,        new IntrinsicInfo(0x0e20a400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SmaxvV,        new IntrinsicInfo(0x0e30a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SmaxV,         new IntrinsicInfo(0x0e206400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SminpV,        new IntrinsicInfo(0x0e20ac00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SminvV,        new IntrinsicInfo(0x0e31a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SminV,         new IntrinsicInfo(0x0e206c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SmlalVe,       new IntrinsicInfo(0x0f002000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64SmlalV,        new IntrinsicInfo(0x0e208000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SmlslVe,       new IntrinsicInfo(0x0f006000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64SmlslV,        new IntrinsicInfo(0x0e20a000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SmovV,         new IntrinsicInfo(0x0e002c00u, IntrinsicType.VectorUnaryByElem));
+            Add(Intrinsic.Arm64SmullVe,       new IntrinsicInfo(0x0f00a000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SmullV,        new IntrinsicInfo(0x0e20c000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqabsS,        new IntrinsicInfo(0x5e207800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64SqabsV,        new IntrinsicInfo(0x0e207800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SqaddS,        new IntrinsicInfo(0x5e200c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqaddV,        new IntrinsicInfo(0x0e200c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqdmlalSe,     new IntrinsicInfo(0x5f003000u, IntrinsicType.ScalarBinaryByElem));
+            Add(Intrinsic.Arm64SqdmlalVe,     new IntrinsicInfo(0x0f003000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SqdmlalS,      new IntrinsicInfo(0x5e209000u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqdmlalV,      new IntrinsicInfo(0x0e209000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqdmlslSe,     new IntrinsicInfo(0x5f007000u, IntrinsicType.ScalarBinaryByElem));
+            Add(Intrinsic.Arm64SqdmlslVe,     new IntrinsicInfo(0x0f007000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SqdmlslS,      new IntrinsicInfo(0x5e20b000u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqdmlslV,      new IntrinsicInfo(0x0e20b000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqdmulhSe,     new IntrinsicInfo(0x5f00c000u, IntrinsicType.ScalarBinaryByElem));
+            Add(Intrinsic.Arm64SqdmulhVe,     new IntrinsicInfo(0x0f00c000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SqdmulhS,      new IntrinsicInfo(0x5e20b400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqdmulhV,      new IntrinsicInfo(0x0e20b400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqdmullSe,     new IntrinsicInfo(0x5f00b000u, IntrinsicType.ScalarBinaryByElem));
+            Add(Intrinsic.Arm64SqdmullVe,     new IntrinsicInfo(0x0f00b000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SqdmullS,      new IntrinsicInfo(0x5e20d000u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqdmullV,      new IntrinsicInfo(0x0e20d000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqnegS,        new IntrinsicInfo(0x7e207800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64SqnegV,        new IntrinsicInfo(0x2e207800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64SqrdmulhSe,    new IntrinsicInfo(0x5f00d000u, IntrinsicType.ScalarBinaryByElem));
+            Add(Intrinsic.Arm64SqrdmulhVe,    new IntrinsicInfo(0x0f00d000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64SqrdmulhS,     new IntrinsicInfo(0x7e20b400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqrdmulhV,     new IntrinsicInfo(0x2e20b400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqrshlS,       new IntrinsicInfo(0x5e205c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqrshlV,       new IntrinsicInfo(0x0e205c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqrshrnS,      new IntrinsicInfo(0x5f009c00u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SqrshrnV,      new IntrinsicInfo(0x0f009c00u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SqrshrunS,     new IntrinsicInfo(0x7f008c00u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SqrshrunV,     new IntrinsicInfo(0x2f008c00u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SqshluS,       new IntrinsicInfo(0x7f006400u, IntrinsicType.ScalarBinaryShl));
+            Add(Intrinsic.Arm64SqshluV,       new IntrinsicInfo(0x2f006400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64SqshlSi,       new IntrinsicInfo(0x5f007400u, IntrinsicType.ScalarBinaryShl));
+            Add(Intrinsic.Arm64SqshlVi,       new IntrinsicInfo(0x0f007400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64SqshlS,        new IntrinsicInfo(0x5e204c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqshlV,        new IntrinsicInfo(0x0e204c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqshrnS,       new IntrinsicInfo(0x5f009400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SqshrnV,       new IntrinsicInfo(0x0f009400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SqshrunS,      new IntrinsicInfo(0x7f008400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SqshrunV,      new IntrinsicInfo(0x2f008400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SqsubS,        new IntrinsicInfo(0x5e202c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SqsubV,        new IntrinsicInfo(0x0e202c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SqxtnS,        new IntrinsicInfo(0x5e214800u, IntrinsicType.ScalarBinaryRd));
+            Add(Intrinsic.Arm64SqxtnV,        new IntrinsicInfo(0x0e214800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64SqxtunS,       new IntrinsicInfo(0x7e212800u, IntrinsicType.ScalarBinaryRd));
+            Add(Intrinsic.Arm64SqxtunV,       new IntrinsicInfo(0x2e212800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64SrhaddV,       new IntrinsicInfo(0x0e201400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SriS,          new IntrinsicInfo(0x7f004400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SriV,          new IntrinsicInfo(0x2f004400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SrshlS,        new IntrinsicInfo(0x5e205400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SrshlV,        new IntrinsicInfo(0x0e205400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SrshrS,        new IntrinsicInfo(0x5f002400u, IntrinsicType.ScalarBinaryShr));
+            Add(Intrinsic.Arm64SrshrV,        new IntrinsicInfo(0x0f002400u, IntrinsicType.VectorBinaryShr));
+            Add(Intrinsic.Arm64SrsraS,        new IntrinsicInfo(0x5f003400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SrsraV,        new IntrinsicInfo(0x0f003400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SshllV,        new IntrinsicInfo(0x0f00a400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64SshlS,         new IntrinsicInfo(0x5e204400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SshlV,         new IntrinsicInfo(0x0e204400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SshrS,         new IntrinsicInfo(0x5f000400u, IntrinsicType.ScalarBinaryShr));
+            Add(Intrinsic.Arm64SshrV,         new IntrinsicInfo(0x0f000400u, IntrinsicType.VectorBinaryShr));
+            Add(Intrinsic.Arm64SsraS,         new IntrinsicInfo(0x5f001400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64SsraV,         new IntrinsicInfo(0x0f001400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64SsublV,        new IntrinsicInfo(0x0e202000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SsubwV,        new IntrinsicInfo(0x0e203000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64St1Vms,        new IntrinsicInfo(0x0c002000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64St1Vss,        new IntrinsicInfo(0x0d000000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64St2Vms,        new IntrinsicInfo(0x0c008000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64St2Vss,        new IntrinsicInfo(0x0d200000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64St3Vms,        new IntrinsicInfo(0x0c004000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64St3Vss,        new IntrinsicInfo(0x0d002000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64St4Vms,        new IntrinsicInfo(0x0c000000u, IntrinsicType.VectorLdSt));
+            Add(Intrinsic.Arm64St4Vss,        new IntrinsicInfo(0x0d202000u, IntrinsicType.VectorLdStSs));
+            Add(Intrinsic.Arm64SubhnV,        new IntrinsicInfo(0x0e206000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64SubS,          new IntrinsicInfo(0x7e208400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64SubV,          new IntrinsicInfo(0x2e208400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64SuqaddS,       new IntrinsicInfo(0x5e203800u, IntrinsicType.ScalarBinaryRd));
+            Add(Intrinsic.Arm64SuqaddV,       new IntrinsicInfo(0x0e203800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64TblV,          new IntrinsicInfo(0x0e000000u, IntrinsicType.VectorLookupTable));
+            Add(Intrinsic.Arm64TbxV,          new IntrinsicInfo(0x0e001000u, IntrinsicType.VectorLookupTable));
+            Add(Intrinsic.Arm64Trn1V,         new IntrinsicInfo(0x0e002800u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64Trn2V,         new IntrinsicInfo(0x0e006800u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UabalV,        new IntrinsicInfo(0x2e205000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64UabaV,         new IntrinsicInfo(0x2e207c00u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64UabdlV,        new IntrinsicInfo(0x2e207000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UabdV,         new IntrinsicInfo(0x2e207400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UadalpV,       new IntrinsicInfo(0x2e206800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64UaddlpV,       new IntrinsicInfo(0x2e202800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UaddlvV,       new IntrinsicInfo(0x2e303800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UaddlV,        new IntrinsicInfo(0x2e200000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UaddwV,        new IntrinsicInfo(0x2e201000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UcvtfSFixed,   new IntrinsicInfo(0x7f00e400u, IntrinsicType.ScalarFPConvFixed));
+            Add(Intrinsic.Arm64UcvtfVFixed,   new IntrinsicInfo(0x2f00e400u, IntrinsicType.VectorFPConvFixed));
+            Add(Intrinsic.Arm64UcvtfS,        new IntrinsicInfo(0x7e21d800u, IntrinsicType.ScalarUnary));
+            Add(Intrinsic.Arm64UcvtfV,        new IntrinsicInfo(0x2e21d800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UcvtfGpFixed,  new IntrinsicInfo(0x1e030000u, IntrinsicType.ScalarFPConvFixedGpr));
+            Add(Intrinsic.Arm64UcvtfGp,       new IntrinsicInfo(0x1e230000u, IntrinsicType.ScalarFPConvGpr));
+            Add(Intrinsic.Arm64UhaddV,        new IntrinsicInfo(0x2e200400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UhsubV,        new IntrinsicInfo(0x2e202400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UmaxpV,        new IntrinsicInfo(0x2e20a400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UmaxvV,        new IntrinsicInfo(0x2e30a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UmaxV,         new IntrinsicInfo(0x2e206400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UminpV,        new IntrinsicInfo(0x2e20ac00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UminvV,        new IntrinsicInfo(0x2e31a800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UminV,         new IntrinsicInfo(0x2e206c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UmlalVe,       new IntrinsicInfo(0x2f002000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64UmlalV,        new IntrinsicInfo(0x2e208000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64UmlslVe,       new IntrinsicInfo(0x2f006000u, IntrinsicType.VectorTernaryRdByElem));
+            Add(Intrinsic.Arm64UmlslV,        new IntrinsicInfo(0x2e20a000u, IntrinsicType.VectorTernaryRd));
+            Add(Intrinsic.Arm64UmovV,         new IntrinsicInfo(0x0e003c00u, IntrinsicType.VectorUnaryByElem));
+            Add(Intrinsic.Arm64UmullVe,       new IntrinsicInfo(0x2f00a000u, IntrinsicType.VectorBinaryByElem));
+            Add(Intrinsic.Arm64UmullV,        new IntrinsicInfo(0x2e20c000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UqaddS,        new IntrinsicInfo(0x7e200c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UqaddV,        new IntrinsicInfo(0x2e200c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UqrshlS,       new IntrinsicInfo(0x7e205c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UqrshlV,       new IntrinsicInfo(0x2e205c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UqrshrnS,      new IntrinsicInfo(0x7f009c00u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64UqrshrnV,      new IntrinsicInfo(0x2f009c00u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64UqshlSi,       new IntrinsicInfo(0x7f007400u, IntrinsicType.ScalarBinaryShl));
+            Add(Intrinsic.Arm64UqshlVi,       new IntrinsicInfo(0x2f007400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64UqshlS,        new IntrinsicInfo(0x7e204c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UqshlV,        new IntrinsicInfo(0x2e204c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UqshrnS,       new IntrinsicInfo(0x7f009400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64UqshrnV,       new IntrinsicInfo(0x2f009400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64UqsubS,        new IntrinsicInfo(0x7e202c00u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UqsubV,        new IntrinsicInfo(0x2e202c00u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UqxtnS,        new IntrinsicInfo(0x7e214800u, IntrinsicType.ScalarBinaryRd));
+            Add(Intrinsic.Arm64UqxtnV,        new IntrinsicInfo(0x2e214800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64UrecpeV,       new IntrinsicInfo(0x0ea1c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UrhaddV,       new IntrinsicInfo(0x2e201400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UrshlS,        new IntrinsicInfo(0x7e205400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UrshlV,        new IntrinsicInfo(0x2e205400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UrshrS,        new IntrinsicInfo(0x7f002400u, IntrinsicType.ScalarBinaryShr));
+            Add(Intrinsic.Arm64UrshrV,        new IntrinsicInfo(0x2f002400u, IntrinsicType.VectorBinaryShr));
+            Add(Intrinsic.Arm64UrsqrteV,      new IntrinsicInfo(0x2ea1c800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64UrsraS,        new IntrinsicInfo(0x7f003400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64UrsraV,        new IntrinsicInfo(0x2f003400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64UshllV,        new IntrinsicInfo(0x2f00a400u, IntrinsicType.VectorBinaryShl));
+            Add(Intrinsic.Arm64UshlS,         new IntrinsicInfo(0x7e204400u, IntrinsicType.ScalarBinary));
+            Add(Intrinsic.Arm64UshlV,         new IntrinsicInfo(0x2e204400u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UshrS,         new IntrinsicInfo(0x7f000400u, IntrinsicType.ScalarBinaryShr));
+            Add(Intrinsic.Arm64UshrV,         new IntrinsicInfo(0x2f000400u, IntrinsicType.VectorBinaryShr));
+            Add(Intrinsic.Arm64UsqaddS,       new IntrinsicInfo(0x7e203800u, IntrinsicType.ScalarBinaryRd));
+            Add(Intrinsic.Arm64UsqaddV,       new IntrinsicInfo(0x2e203800u, IntrinsicType.VectorBinaryRd));
+            Add(Intrinsic.Arm64UsraS,         new IntrinsicInfo(0x7f001400u, IntrinsicType.ScalarTernaryShrRd));
+            Add(Intrinsic.Arm64UsraV,         new IntrinsicInfo(0x2f001400u, IntrinsicType.VectorTernaryShrRd));
+            Add(Intrinsic.Arm64UsublV,        new IntrinsicInfo(0x2e202000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64UsubwV,        new IntrinsicInfo(0x2e203000u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64Uzp1V,         new IntrinsicInfo(0x0e001800u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64Uzp2V,         new IntrinsicInfo(0x0e005800u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64XtnV,          new IntrinsicInfo(0x0e212800u, IntrinsicType.VectorUnary));
+            Add(Intrinsic.Arm64Zip1V,         new IntrinsicInfo(0x0e003800u, IntrinsicType.VectorBinary));
+            Add(Intrinsic.Arm64Zip2V,         new IntrinsicInfo(0x0e007800u, IntrinsicType.VectorBinary));
+        }
+
+        private static void Add(Intrinsic intrin, IntrinsicInfo info)
+        {
+            _intrinTable[(int)intrin] = info;
+        }
+
+        public static IntrinsicInfo GetInfo(Intrinsic intrin)
+        {
+            return _intrinTable[(int)intrin];
+        }
+    }
+}

src/ARMeilleure/CodeGen/Arm64/IntrinsicType.cs

@@ -0,0 +1,59 @@
+namespace ARMeilleure.CodeGen.Arm64
+{
+    enum IntrinsicType
+    {
+        ScalarUnary,
+        ScalarUnaryByElem,
+        ScalarBinary,
+        ScalarBinaryByElem,
+        ScalarBinaryFPByElem,
+        ScalarBinaryRd,
+        ScalarBinaryShl,
+        ScalarBinaryShr,
+        ScalarFcsel,
+        ScalarFmovi,
+        ScalarFPCompare,
+        ScalarFPCompareCond,
+        ScalarFPConv,
+        ScalarFPConvFixed,
+        ScalarFPConvFixedGpr,
+        ScalarFPConvGpr,
+        ScalarTernary,
+        ScalarTernaryFPRdByElem,
+        ScalarTernaryShlRd,
+        ScalarTernaryShrRd,
+
+        VectorUnary,
+        VectorUnaryBitwise,
+        VectorUnaryByElem,
+        VectorBinary,
+        VectorBinaryBitwise,
+        VectorBinaryBitwiseImm,
+        VectorBinaryByElem,
+        VectorBinaryFPByElem,
+        VectorBinaryRd,
+        VectorBinaryShl,
+        VectorBinaryShr,
+        VectorExt,
+        VectorFmovi,
+        VectorFPConvFixed,
+        VectorInsertByElem,
+        VectorLdSt,
+        VectorLdStSs,
+        VectorLookupTable,
+        VectorMovi,
+        VectorMvni,
+        VectorTernaryFPRdByElem,
+        VectorTernaryRd,
+        VectorTernaryRdBitwise,
+        VectorTernaryRdByElem,
+        VectorTernaryShlRd,
+        VectorTernaryShrRd,
+
+        Vector128Unary,
+        Vector128Binary,
+
+        GetRegister,
+        SetRegister
+    }
+}

src/ARMeilleure/CodeGen/Arm64/PreAllocator.cs

@@ -0,0 +1,892 @@
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Collections.Generic;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.Arm64
+{
+    static class PreAllocator
+    {
+        private class ConstantDict
+        {
+            private readonly Dictionary<(ulong, OperandType), Operand> _constants;
+
+            public ConstantDict()
+            {
+                _constants = new Dictionary<(ulong, OperandType), Operand>();
+            }
+
+            public void Add(ulong value, OperandType type, Operand local)
+            {
+                _constants.Add((value, type), local);
+            }
+
+            public bool TryGetValue(ulong value, OperandType type, out Operand local)
+            {
+                return _constants.TryGetValue((value, type), out local);
+            }
+        }
+
+        public static void RunPass(CompilerContext cctx, StackAllocator stackAlloc, out int maxCallArgs)
+        {
+            maxCallArgs = -1;
+
+            Span<Operation> buffer = default;
+
+            Operand[] preservedArgs = new Operand[CallingConvention.GetArgumentsOnRegsCount()];
+
+            for (BasicBlock block = cctx.Cfg.Blocks.First; block != null; block = block.ListNext)
+            {
+                ConstantDict constants = new ConstantDict();
+
+                Operation nextNode;
+
+                for (Operation node = block.Operations.First; node != default; node = nextNode)
+                {
+                    nextNode = node.ListNext;
+
+                    if (node.Instruction == Instruction.Phi)
+                    {
+                        continue;
+                    }
+
+                    InsertConstantRegCopies(constants, block.Operations, node);
+                    InsertDestructiveRegCopies(block.Operations, node);
+
+                    switch (node.Instruction)
+                    {
+                        case Instruction.Call:
+                            // Get the maximum number of arguments used on a call.
+                            // On windows, when a struct is returned from the call,
+                            // we also need to pass the pointer where the struct
+                            // should be written on the first argument.
+                            int argsCount = node.SourcesCount - 1;
+
+                            if (node.Destination != default && node.Destination.Type == OperandType.V128)
+                            {
+                                argsCount++;
+                            }
+
+                            if (maxCallArgs < argsCount)
+                            {
+                                maxCallArgs = argsCount;
+                            }
+
+                            // Copy values to registers expected by the function
+                            // being called, as mandated by the ABI.
+                            InsertCallCopies(constants, block.Operations, node);
+                            break;
+                        case Instruction.CompareAndSwap:
+                        case Instruction.CompareAndSwap16:
+                        case Instruction.CompareAndSwap8:
+                            nextNode = GenerateCompareAndSwap(block.Operations, node);
+                            break;
+                        case Instruction.LoadArgument:
+                            nextNode = InsertLoadArgumentCopy(cctx, ref buffer, block.Operations, preservedArgs, node);
+                            break;
+                        case Instruction.Return:
+                            InsertReturnCopy(block.Operations, node);
+                            break;
+                        case Instruction.Tailcall:
+                            InsertTailcallCopies(constants, block.Operations, stackAlloc, node, node);
+                            break;
+                    }
+                }
+            }
+        }
+
+        private static void InsertConstantRegCopies(ConstantDict constants, IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.SourcesCount == 0 || IsIntrinsicWithConst(node))
+            {
+                return;
+            }
+
+            Instruction inst = node.Instruction;
+
+            Operand src1 = node.GetSource(0);
+            Operand src2;
+
+            if (src1.Kind == OperandKind.Constant)
+            {
+                if (!src1.Type.IsInteger())
+                {
+                    // Handle non-integer types (FP32, FP64 and V128).
+                    // For instructions without an immediate operand, we do the following:
+                    // - Insert a copy with the constant value (as integer) to a GPR.
+                    // - Insert a copy from the GPR to a XMM register.
+                    // - Replace the constant use with the XMM register.
+                    src1 = AddFloatConstantCopy(constants, nodes, node, src1);
+
+                    node.SetSource(0, src1);
+                }
+                else if (!HasConstSrc1(node, src1.Value))
+                {
+                    // Handle integer types.
+                    // Most ALU instructions accepts a 32-bits immediate on the second operand.
+                    // We need to ensure the following:
+                    // - If the constant is on operand 1, we need to move it.
+                    // -- But first, we try to swap operand 1 and 2 if the instruction is commutative.
+                    // -- Doing so may allow us to encode the constant as operand 2 and avoid a copy.
+                    // - If the constant is on operand 2, we check if the instruction supports it,
+                    // if not, we also add a copy. 64-bits constants are usually not supported.
+                    if (IsCommutative(node))
+                    {
+                        src2 = node.GetSource(1);
+
+                        Operand temp = src1;
+
+                        src1 = src2;
+                        src2 = temp;
+
+                        node.SetSource(0, src1);
+                        node.SetSource(1, src2);
+                    }
+
+                    if (src1.Kind == OperandKind.Constant)
+                    {
+                        src1 = AddIntConstantCopy(constants, nodes, node, src1);
+
+                        node.SetSource(0, src1);
+                    }
+                }
+            }
+
+            if (node.SourcesCount < 2)
+            {
+                return;
+            }
+
+            src2 = node.GetSource(1);
+
+            if (src2.Kind == OperandKind.Constant)
+            {
+                if (!src2.Type.IsInteger())
+                {
+                    src2 = AddFloatConstantCopy(constants, nodes, node, src2);
+
+                    node.SetSource(1, src2);
+                }
+                else if (!HasConstSrc2(inst, src2))
+                {
+                    src2 = AddIntConstantCopy(constants, nodes, node, src2);
+
+                    node.SetSource(1, src2);
+                }
+            }
+
+            if (node.SourcesCount < 3 ||
+                node.Instruction == Instruction.BranchIf ||
+                node.Instruction == Instruction.Compare ||
+                node.Instruction == Instruction.VectorInsert ||
+                node.Instruction == Instruction.VectorInsert16 ||
+                node.Instruction == Instruction.VectorInsert8)
+            {
+                return;
+            }
+
+            for (int srcIndex = 2; srcIndex < node.SourcesCount; srcIndex++)
+            {
+                Operand src = node.GetSource(srcIndex);
+
+                if (src.Kind == OperandKind.Constant)
+                {
+                    if (!src.Type.IsInteger())
+                    {
+                        src = AddFloatConstantCopy(constants, nodes, node, src);
+
+                        node.SetSource(srcIndex, src);
+                    }
+                    else
+                    {
+                        src = AddIntConstantCopy(constants, nodes, node, src);
+
+                        node.SetSource(srcIndex, src);
+                    }
+                }
+            }
+        }
+
+        private static void InsertDestructiveRegCopies(IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.Destination == default || node.SourcesCount == 0)
+            {
+                return;
+            }
+
+            Operand dest = node.Destination;
+            Operand src1 = node.GetSource(0);
+
+            if (IsSameOperandDestSrc1(node) && src1.Kind == OperandKind.LocalVariable)
+            {
+                bool useNewLocal = false;
+
+                for (int srcIndex = 1; srcIndex < node.SourcesCount; srcIndex++)
+                {
+                    if (node.GetSource(srcIndex) == dest)
+                    {
+                        useNewLocal = true;
+
+                        break;
+                    }
+                }
+
+                if (useNewLocal)
+                {
+                    // Dest is being used as some source already, we need to use a new
+                    // local to store the temporary value, otherwise the value on dest
+                    // local would be overwritten.
+                    Operand temp = Local(dest.Type);
+
+                    nodes.AddBefore(node, Operation(Instruction.Copy, temp, src1));
+
+                    node.SetSource(0, temp);
+
+                    nodes.AddAfter(node, Operation(Instruction.Copy, dest, temp));
+
+                    node.Destination = temp;
+                }
+                else
+                {
+                    nodes.AddBefore(node, Operation(Instruction.Copy, dest, src1));
+
+                    node.SetSource(0, dest);
+                }
+            }
+        }
+
+        private static void InsertCallCopies(ConstantDict constants, IntrusiveList<Operation> nodes, Operation node)
+        {
+            Operation operation = node;
+
+            Operand dest = operation.Destination;
+
+            List<Operand> sources = new List<Operand>
+            {
+                operation.GetSource(0)
+            };
+
+            int argsCount = operation.SourcesCount - 1;
+
+            int intMax = CallingConvention.GetArgumentsOnRegsCount();
+            int vecMax = CallingConvention.GetArgumentsOnRegsCount();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            int stackOffset = 0;
+
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = operation.GetSource(index + 1);
+
+                bool passOnReg;
+
+                if (source.Type.IsInteger())
+                {
+                    passOnReg = intCount < intMax;
+                }
+                else if (source.Type == OperandType.V128)
+                {
+                    passOnReg = intCount + 1 < intMax;
+                }
+                else
+                {
+                    passOnReg = vecCount < vecMax;
+                }
+
+                if (source.Type == OperandType.V128 && passOnReg)
+                {
+                    // V128 is a struct, we pass each half on a GPR if possible.
+                    Operand argReg  = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+                    Operand argReg2 = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg,  source, Const(0)));
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg2, source, Const(1)));
+
+                    continue;
+                }
+
+                if (passOnReg)
+                {
+                    Operand argReg = source.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntArgumentRegister(intCount++), source.Type)
+                        : Xmm(CallingConvention.GetVecArgumentRegister(vecCount++), source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                    InsertConstantRegCopies(constants, nodes, nodes.AddBefore(node, copyOp));
+
+                    sources.Add(argReg);
+                }
+                else
+                {
+                    Operand offset = Const(stackOffset);
+
+                    Operation spillOp = Operation(Instruction.SpillArg, default, offset, source);
+
+                    InsertConstantRegCopies(constants, nodes, nodes.AddBefore(node, spillOp));
+
+                    stackOffset += source.Type.GetSizeInBytes();
+                }
+            }
+
+            if (dest != default)
+            {
+                if (dest.Type == OperandType.V128)
+                {
+                    Operand retLReg = Gpr(CallingConvention.GetIntReturnRegister(),     OperandType.I64);
+                    Operand retHReg = Gpr(CallingConvention.GetIntReturnRegisterHigh(), OperandType.I64);
+
+                    node = nodes.AddAfter(node, Operation(Instruction.VectorCreateScalar, dest, retLReg));
+                    nodes.AddAfter(node, Operation(Instruction.VectorInsert, dest, dest, retHReg, Const(1)));
+
+                    operation.Destination = default;
+                }
+                else
+                {
+                    Operand retReg = dest.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntReturnRegister(), dest.Type)
+                        : Xmm(CallingConvention.GetVecReturnRegister(), dest.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, dest, retReg);
+
+                    nodes.AddAfter(node, copyOp);
+
+                    operation.Destination = retReg;
+                }
+            }
+
+            operation.SetSources(sources.ToArray());
+        }
+
+        private static void InsertTailcallCopies(
+            ConstantDict constants,
+            IntrusiveList<Operation> nodes,
+            StackAllocator stackAlloc,
+            Operation node,
+            Operation operation)
+        {
+            List<Operand> sources = new List<Operand>
+            {
+                operation.GetSource(0)
+            };
+
+            int argsCount = operation.SourcesCount - 1;
+
+            int intMax = CallingConvention.GetArgumentsOnRegsCount();
+            int vecMax = CallingConvention.GetArgumentsOnRegsCount();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            // Handle arguments passed on registers.
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = operation.GetSource(1 + index);
+
+                bool passOnReg;
+
+                if (source.Type.IsInteger())
+                {
+                    passOnReg = intCount + 1 < intMax;
+                }
+                else
+                {
+                    passOnReg = vecCount < vecMax;
+                }
+
+                if (source.Type == OperandType.V128 && passOnReg)
+                {
+                    // V128 is a struct, we pass each half on a GPR if possible.
+                    Operand argReg  = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+                    Operand argReg2 = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg, source, Const(0)));
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg2, source, Const(1)));
+
+                    continue;
+                }
+
+                if (passOnReg)
+                {
+                    Operand argReg = source.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntArgumentRegister(intCount++), source.Type)
+                        : Xmm(CallingConvention.GetVecArgumentRegister(vecCount++), source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                    InsertConstantRegCopies(constants, nodes, nodes.AddBefore(node, copyOp));
+
+                    sources.Add(argReg);
+                }
+                else
+                {
+                    throw new NotImplementedException("Spilling is not currently supported for tail calls. (too many arguments)");
+                }
+            }
+
+            // The target address must be on the return registers, since we
+            // don't return anything and it is guaranteed to not be a
+            // callee saved register (which would be trashed on the epilogue).
+            Operand tcAddress = Gpr(CodeGenCommon.TcAddressRegister, OperandType.I64);
+
+            Operation addrCopyOp = Operation(Instruction.Copy, tcAddress, operation.GetSource(0));
+
+            nodes.AddBefore(node, addrCopyOp);
+
+            sources[0] = tcAddress;
+
+            operation.SetSources(sources.ToArray());
+        }
+
+        private static Operation GenerateCompareAndSwap(IntrusiveList<Operation> nodes, Operation node)
+        {
+            Operand expected = node.GetSource(1);
+
+            if (expected.Type == OperandType.V128)
+            {
+                Operand dest = node.Destination;
+                Operand expectedLow = Local(OperandType.I64);
+                Operand expectedHigh = Local(OperandType.I64);
+                Operand desiredLow = Local(OperandType.I64);
+                Operand desiredHigh = Local(OperandType.I64);
+                Operand actualLow = Local(OperandType.I64);
+                Operand actualHigh = Local(OperandType.I64);
+
+                Operand address = node.GetSource(0);
+                Operand desired = node.GetSource(2);
+
+                void SplitOperand(Operand source, Operand low, Operand high)
+                {
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, low, source, Const(0)));
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, high, source, Const(1)));
+                }
+
+                SplitOperand(expected, expectedLow, expectedHigh);
+                SplitOperand(desired, desiredLow, desiredHigh);
+
+                Operation operation = node;
+
+                // Update the sources and destinations with split 64-bit halfs of the whole 128-bit values.
+                // We also need a additional registers that will be used to store temporary information.
+                operation.SetDestinations(new[] { actualLow, actualHigh, Local(OperandType.I64), Local(OperandType.I64) });
+                operation.SetSources(new[] { address, expectedLow, expectedHigh, desiredLow, desiredHigh });
+
+                // Add some dummy uses of the input operands, as the CAS operation will be a loop,
+                // so they can't be used as destination operand.
+                for (int i = 0; i < operation.SourcesCount; i++)
+                {
+                    Operand src = operation.GetSource(i);
+                    node = nodes.AddAfter(node, Operation(Instruction.Copy, src, src));
+                }
+
+                // Assemble the vector with the 64-bit values at the given memory location.
+                node = nodes.AddAfter(node, Operation(Instruction.VectorCreateScalar, dest, actualLow));
+                node = nodes.AddAfter(node, Operation(Instruction.VectorInsert, dest, dest, actualHigh, Const(1)));
+            }
+            else
+            {
+                // We need a additional register where the store result will be written to.
+                node.SetDestinations(new[] { node.Destination, Local(OperandType.I32) });
+
+                // Add some dummy uses of the input operands, as the CAS operation will be a loop,
+                // so they can't be used as destination operand.
+                Operation operation = node;
+
+                for (int i = 0; i < operation.SourcesCount; i++)
+                {
+                    Operand src = operation.GetSource(i);
+                    node = nodes.AddAfter(node, Operation(Instruction.Copy, src, src));
+                }
+            }
+
+            return node.ListNext;
+        }
+
+        private static void InsertReturnCopy(IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.SourcesCount == 0)
+            {
+                return;
+            }
+
+            Operand source = node.GetSource(0);
+
+            if (source.Type == OperandType.V128)
+            {
+                Operand retLReg = Gpr(CallingConvention.GetIntReturnRegister(),     OperandType.I64);
+                Operand retHReg = Gpr(CallingConvention.GetIntReturnRegisterHigh(), OperandType.I64);
+
+                nodes.AddBefore(node, Operation(Instruction.VectorExtract, retLReg, source, Const(0)));
+                nodes.AddBefore(node, Operation(Instruction.VectorExtract, retHReg, source, Const(1)));
+            }
+            else
+            {
+                Operand retReg = source.Type.IsInteger()
+                    ? Gpr(CallingConvention.GetIntReturnRegister(), source.Type)
+                    : Xmm(CallingConvention.GetVecReturnRegister(), source.Type);
+
+                Operation retCopyOp = Operation(Instruction.Copy, retReg, source);
+
+                nodes.AddBefore(node, retCopyOp);
+            }
+        }
+
+        private static Operation InsertLoadArgumentCopy(
+            CompilerContext cctx,
+            ref Span<Operation> buffer,
+            IntrusiveList<Operation> nodes,
+            Operand[] preservedArgs,
+            Operation node)
+        {
+            Operand source = node.GetSource(0);
+
+            Debug.Assert(source.Kind == OperandKind.Constant, "Non-constant LoadArgument source kind.");
+
+            int index = source.AsInt32();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            for (int cIndex = 0; cIndex < index; cIndex++)
+            {
+                OperandType argType = cctx.FuncArgTypes[cIndex];
+
+                if (argType.IsInteger())
+                {
+                    intCount++;
+                }
+                else if (argType == OperandType.V128)
+                {
+                    intCount += 2;
+                }
+                else
+                {
+                    vecCount++;
+                }
+            }
+
+            bool passOnReg;
+
+            if (source.Type.IsInteger())
+            {
+                passOnReg = intCount < CallingConvention.GetArgumentsOnRegsCount();
+            }
+            else if (source.Type == OperandType.V128)
+            {
+                passOnReg = intCount + 1 < CallingConvention.GetArgumentsOnRegsCount();
+            }
+            else
+            {
+                passOnReg = vecCount < CallingConvention.GetArgumentsOnRegsCount();
+            }
+
+            if (passOnReg)
+            {
+                Operand dest = node.Destination;
+
+                if (preservedArgs[index] == default)
+                {
+                    if (dest.Type == OperandType.V128)
+                    {
+                        // V128 is a struct, we pass each half on a GPR if possible.
+                        Operand pArg = Local(OperandType.V128);
+
+                        Operand argLReg = Gpr(CallingConvention.GetIntArgumentRegister(intCount), OperandType.I64);
+                        Operand argHReg = Gpr(CallingConvention.GetIntArgumentRegister(intCount + 1), OperandType.I64);
+
+                        Operation copyL = Operation(Instruction.VectorCreateScalar, pArg, argLReg);
+                        Operation copyH = Operation(Instruction.VectorInsert, pArg, pArg, argHReg, Const(1));
+
+                        cctx.Cfg.Entry.Operations.AddFirst(copyH);
+                        cctx.Cfg.Entry.Operations.AddFirst(copyL);
+
+                        preservedArgs[index] = pArg;
+                    }
+                    else
+                    {
+                        Operand pArg = Local(dest.Type);
+
+                        Operand argReg = dest.Type.IsInteger()
+                            ? Gpr(CallingConvention.GetIntArgumentRegister(intCount), dest.Type)
+                            : Xmm(CallingConvention.GetVecArgumentRegister(vecCount), dest.Type);
+
+                        Operation copyOp = Operation(Instruction.Copy, pArg, argReg);
+
+                        cctx.Cfg.Entry.Operations.AddFirst(copyOp);
+
+                        preservedArgs[index] = pArg;
+                    }
+                }
+
+                Operation nextNode;
+
+                if (dest.AssignmentsCount == 1)
+                {
+                    // Let's propagate the argument if we can to avoid copies.
+                    PreAllocatorCommon.Propagate(ref buffer, dest, preservedArgs[index]);
+                    nextNode = node.ListNext;
+                }
+                else
+                {
+                    Operation argCopyOp = Operation(Instruction.Copy, dest, preservedArgs[index]);
+                    nextNode = nodes.AddBefore(node, argCopyOp);
+                }
+
+                Delete(nodes, node);
+                return nextNode;
+            }
+            else
+            {
+                // TODO: Pass on stack.
+                return node;
+            }
+        }
+
+        private static Operand AddFloatConstantCopy(
+            ConstantDict constants,
+            IntrusiveList<Operation> nodes,
+            Operation node,
+            Operand source)
+        {
+            Operand temp = Local(source.Type);
+
+            Operand intConst = AddIntConstantCopy(constants, nodes, node, GetIntConst(source));
+
+            Operation copyOp = Operation(Instruction.VectorCreateScalar, temp, intConst);
+
+            nodes.AddBefore(node, copyOp);
+
+            return temp;
+        }
+
+        private static Operand AddIntConstantCopy(
+            ConstantDict constants,
+            IntrusiveList<Operation> nodes,
+            Operation node,
+            Operand source)
+        {
+            if (constants.TryGetValue(source.Value, source.Type, out Operand temp))
+            {
+                return temp;
+            }
+
+            temp = Local(source.Type);
+
+            Operation copyOp = Operation(Instruction.Copy, temp, source);
+
+            nodes.AddBefore(node, copyOp);
+
+            constants.Add(source.Value, source.Type, temp);
+
+            return temp;
+        }
+
+        private static Operand GetIntConst(Operand value)
+        {
+            if (value.Type == OperandType.FP32)
+            {
+                return Const(value.AsInt32());
+            }
+            else if (value.Type == OperandType.FP64)
+            {
+                return Const(value.AsInt64());
+            }
+
+            return value;
+        }
+
+        private static void Delete(IntrusiveList<Operation> nodes, Operation node)
+        {
+            node.Destination = default;
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                node.SetSource(index, default);
+            }
+
+            nodes.Remove(node);
+        }
+
+        private static Operand Gpr(int register, OperandType type)
+        {
+            return Register(register, RegisterType.Integer, type);
+        }
+
+        private static Operand Xmm(int register, OperandType type)
+        {
+            return Register(register, RegisterType.Vector, type);
+        }
+
+        private static bool IsSameOperandDestSrc1(Operation operation)
+        {
+            switch (operation.Instruction)
+            {
+                case Instruction.Extended:
+                    return IsSameOperandDestSrc1(operation.Intrinsic);
+                case Instruction.VectorInsert:
+                case Instruction.VectorInsert16:
+                case Instruction.VectorInsert8:
+                    return true;
+            }
+
+            return false;
+        }
+
+        private static bool IsSameOperandDestSrc1(Intrinsic intrinsic)
+        {
+            IntrinsicInfo info = IntrinsicTable.GetInfo(intrinsic & ~(Intrinsic.Arm64VTypeMask | Intrinsic.Arm64VSizeMask));
+
+            return info.Type == IntrinsicType.ScalarBinaryRd ||
+                   info.Type == IntrinsicType.ScalarTernaryFPRdByElem ||
+                   info.Type == IntrinsicType.ScalarTernaryShlRd ||
+                   info.Type == IntrinsicType.ScalarTernaryShrRd ||
+                   info.Type == IntrinsicType.VectorBinaryRd ||
+                   info.Type == IntrinsicType.VectorInsertByElem ||
+                   info.Type == IntrinsicType.VectorTernaryRd ||
+                   info.Type == IntrinsicType.VectorTernaryRdBitwise ||
+                   info.Type == IntrinsicType.VectorTernaryFPRdByElem ||
+                   info.Type == IntrinsicType.VectorTernaryRdByElem ||
+                   info.Type == IntrinsicType.VectorTernaryShlRd ||
+                   info.Type == IntrinsicType.VectorTernaryShrRd;
+        }
+
+        private static bool HasConstSrc1(Operation node, ulong value)
+        {
+            switch (node.Instruction)
+            {
+                case Instruction.Add:
+                case Instruction.BranchIf:
+                case Instruction.Compare:
+                case Instruction.Subtract:
+                    // The immediate encoding of those instructions does not allow Rn to be
+                    // XZR (it will be SP instead), so we can't allow a Rn constant in this case.
+                    return value == 0 && NotConstOrConst0(node.GetSource(1));
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseNot:
+                case Instruction.BitwiseOr:
+                case Instruction.ByteSwap:
+                case Instruction.CountLeadingZeros:
+                case Instruction.Multiply:
+                case Instruction.Negate:
+                case Instruction.RotateRight:
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                    return value == 0;
+                case Instruction.Copy:
+                case Instruction.LoadArgument:
+                case Instruction.Spill:
+                case Instruction.SpillArg:
+                    return true;
+                case Instruction.Extended:
+                    return value == 0;
+            }
+
+            return false;
+        }
+
+        private static bool NotConstOrConst0(Operand operand)
+        {
+            return operand.Kind != OperandKind.Constant || operand.Value == 0;
+        }
+
+        private static bool HasConstSrc2(Instruction inst, Operand operand)
+        {
+            ulong value = operand.Value;
+
+            switch (inst)
+            {
+                case Instruction.Add:
+                case Instruction.BranchIf:
+                case Instruction.Compare:
+                case Instruction.Subtract:
+                    return ConstFitsOnUImm12Sh(value);
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseOr:
+                    return value == 0 || CodeGenCommon.TryEncodeBitMask(operand, out _, out _, out _);
+                case Instruction.Multiply:
+                case Instruction.Store:
+                case Instruction.Store16:
+                case Instruction.Store8:
+                    return value == 0;
+                case Instruction.RotateRight:
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                case Instruction.VectorExtract:
+                case Instruction.VectorExtract16:
+                case Instruction.VectorExtract8:
+                    return true;
+                case Instruction.Extended:
+                    // TODO: Check if actual intrinsic is supposed to have consts here?
+                    // Right now we only hit this case for fixed-point int <-> FP conversion instructions.
+                    return true;
+            }
+
+            return false;
+        }
+
+        private static bool IsCommutative(Operation operation)
+        {
+            switch (operation.Instruction)
+            {
+                case Instruction.Add:
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseOr:
+                case Instruction.Multiply:
+                    return true;
+
+                case Instruction.BranchIf:
+                case Instruction.Compare:
+                    {
+                        Operand comp = operation.GetSource(2);
+
+                        Debug.Assert(comp.Kind == OperandKind.Constant);
+
+                        var compType = (Comparison)comp.AsInt32();
+
+                        return compType == Comparison.Equal || compType == Comparison.NotEqual;
+                    }
+            }
+
+            return false;
+        }
+
+        private static bool ConstFitsOnUImm12Sh(ulong value)
+        {
+            return (value & ~0xfffUL) == 0 || (value & ~0xfff000UL) == 0;
+        }
+
+        private static bool IsIntrinsicWithConst(Operation operation)
+        {
+            bool isIntrinsic = IsIntrinsic(operation.Instruction);
+
+            if (isIntrinsic)
+            {
+                Intrinsic intrinsic = operation.Intrinsic;
+                IntrinsicInfo info = IntrinsicTable.GetInfo(intrinsic & ~(Intrinsic.Arm64VTypeMask | Intrinsic.Arm64VSizeMask));
+
+                // Those have integer inputs that don't support consts.
+                return info.Type != IntrinsicType.ScalarFPConvGpr &&
+                       info.Type != IntrinsicType.ScalarFPConvFixedGpr &&
+                       info.Type != IntrinsicType.SetRegister;
+            }
+
+            return false;
+        }
+
+        private static bool IsIntrinsic(Instruction inst)
+        {
+            return inst == Instruction.Extended;
+        }
+    }
+}

src/ARMeilleure/CodeGen/CompiledFunction.cs

@@ -0,0 +1,68 @@
+using ARMeilleure.CodeGen.Linking;
+using ARMeilleure.CodeGen.Unwinding;
+using ARMeilleure.Translation.Cache;
+using System;
+using System.Runtime.InteropServices;
+
+namespace ARMeilleure.CodeGen
+{
+    /// <summary>
+    /// Represents a compiled function.
+    /// </summary>
+    readonly struct CompiledFunction
+    {
+        /// <summary>
+        /// Gets the machine code of the <see cref="CompiledFunction"/>.
+        /// </summary>
+        public byte[] Code { get; }
+
+        /// <summary>
+        /// Gets the <see cref="Unwinding.UnwindInfo"/> of the <see cref="CompiledFunction"/>.
+        /// </summary>
+        public UnwindInfo UnwindInfo { get; }
+
+        /// <summary>
+        /// Gets the <see cref="Linking.RelocInfo"/> of the <see cref="CompiledFunction"/>.
+        /// </summary>
+        public RelocInfo RelocInfo { get; }
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="CompiledFunction"/> struct with the specified machine code,
+        /// unwind info and relocation info.
+        /// </summary>
+        /// <param name="code">Machine code</param>
+        /// <param name="unwindInfo">Unwind info</param>
+        /// <param name="relocInfo">Relocation info</param>
+        internal CompiledFunction(byte[] code, UnwindInfo unwindInfo, RelocInfo relocInfo)
+        {
+            Code       = code;
+            UnwindInfo = unwindInfo;
+            RelocInfo  = relocInfo;
+        }
+
+        /// <summary>
+        /// Maps the <see cref="CompiledFunction"/> onto the <see cref="JitCache"/> and returns a delegate of type
+        /// <typeparamref name="T"/> pointing to the mapped function.
+        /// </summary>
+        /// <typeparam name="T">Type of delegate</typeparam>
+        /// <returns>A delegate of type <typeparamref name="T"/> pointing to the mapped function</returns>
+        public T Map<T>()
+        {
+            return MapWithPointer<T>(out _);
+        }
+
+        /// <summary>
+        /// Maps the <see cref="CompiledFunction"/> onto the <see cref="JitCache"/> and returns a delegate of type
+        /// <typeparamref name="T"/> pointing to the mapped function.
+        /// </summary>
+        /// <typeparam name="T">Type of delegate</typeparam>
+        /// <param name="codePointer">Pointer to the function code in memory</param>
+        /// <returns>A delegate of type <typeparamref name="T"/> pointing to the mapped function</returns>
+        public T MapWithPointer<T>(out IntPtr codePointer)
+        {
+            codePointer = JitCache.Map(this);
+
+            return Marshal.GetDelegateForFunctionPointer<T>(codePointer);
+        }
+    }
+}

src/ARMeilleure/CodeGen/Linking/RelocEntry.cs

@@ -0,0 +1,38 @@
+namespace ARMeilleure.CodeGen.Linking
+{
+    /// <summary>
+    /// Represents a relocation.
+    /// </summary>
+    readonly struct RelocEntry
+    {
+        public const int Stride = 13; // Bytes.
+
+        /// <summary>
+        /// Gets the position of the relocation.
+        /// </summary>
+        public int Position { get; }
+
+        /// <summary>
+        /// Gets the <see cref="Symbol"/> of the relocation.
+        /// </summary>
+        public Symbol Symbol { get; }
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="RelocEntry"/> struct with the specified position and
+        /// <see cref="Symbol"/>.
+        /// </summary>
+        /// <param name="position">Position of relocation</param>
+        /// <param name="symbol">Symbol of relocation</param>
+        public RelocEntry(int position, Symbol symbol)
+        {
+            Position = position;
+            Symbol = symbol;
+        }
+
+        /// <inheritdoc/>
+        public override string ToString()
+        {
+            return $"({nameof(Position)} = {Position}, {nameof(Symbol)} = {Symbol})";
+        }
+    }
+}

src/ARMeilleure/CodeGen/Linking/RelocInfo.cs

@@ -0,0 +1,32 @@
+using System;
+
+namespace ARMeilleure.CodeGen.Linking
+{
+    /// <summary>
+    /// Represents relocation information about a <see cref="CompiledFunction"/>.
+    /// </summary>
+    readonly struct RelocInfo
+    {
+        /// <summary>
+        /// Gets an empty <see cref="RelocInfo"/>.
+        /// </summary>
+        public static RelocInfo Empty { get; } = new RelocInfo(null);
+
+        private readonly RelocEntry[] _entries;
+
+        /// <summary>
+        /// Gets the set of <see cref="RelocEntry"/>.
+        /// </summary>
+        public ReadOnlySpan<RelocEntry> Entries => _entries;
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="RelocInfo"/> struct with the specified set of
+        /// <see cref="RelocEntry"/>.
+        /// </summary>
+        /// <param name="entries">Set of <see cref="RelocInfo"/> to use</param>
+        public RelocInfo(RelocEntry[] entries)
+        {
+            _entries = entries;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Linking/Symbol.cs

@@ -0,0 +1,99 @@
+using System;
+
+namespace ARMeilleure.CodeGen.Linking
+{
+    /// <summary>
+    /// Represents a symbol.
+    /// </summary>
+    readonly struct Symbol
+    {
+        private readonly ulong _value;
+
+        /// <summary>
+        /// Gets the <see cref="SymbolType"/> of the <see cref="Symbol"/>.
+        /// </summary>
+        public SymbolType Type { get; }
+
+        /// <summary>
+        /// Gets the value of the <see cref="Symbol"/>.
+        /// </summary>
+        /// <exception cref="InvalidOperationException"><see cref="Type"/> is <see cref="SymbolType.None"/></exception>
+        public ulong Value
+        {
+            get
+            {
+                if (Type == SymbolType.None)
+                {
+                    ThrowSymbolNone();
+                }
+
+                return _value;
+            }
+        }
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="Symbol"/> structure with the specified <see cref="SymbolType"/> and value.
+        /// </summary>
+        /// <param name="type">Type of symbol</param>
+        /// <param name="value">Value of symbol</param>
+        public Symbol(SymbolType type, ulong value)
+        {
+            (Type, _value) = (type, value);
+        }
+
+        /// <summary>
+        /// Determines if the specified <see cref="Symbol"/> instances are equal.
+        /// </summary>
+        /// <param name="a">First instance</param>
+        /// <param name="b">Second instance</param>
+        /// <returns><see langword="true"/> if equal; otherwise <see langword="false"/></returns>
+        public static bool operator ==(Symbol a, Symbol b)
+        {
+            return a.Equals(b);
+        }
+
+        /// <summary>
+        /// Determines if the specified <see cref="Symbol"/> instances are not equal.
+        /// </summary>
+        /// <param name="a">First instance</param>
+        /// <param name="b">Second instance</param>
+        /// <returns><see langword="true"/> if not equal; otherwise <see langword="false"/></returns>
+        public static bool operator !=(Symbol a, Symbol b)
+        {
+            return !(a == b);
+        }
+
+        /// <summary>
+        /// Determines if the specified <see cref="Symbol"/> is equal to this <see cref="Symbol"/> instance.
+        /// </summary>
+        /// <param name="other">Other <see cref="Symbol"/> instance</param>
+        /// <returns><see langword="true"/> if equal; otherwise <see langword="false"/></returns>
+        public bool Equals(Symbol other)
+        {
+            return other.Type == Type && other._value == _value;
+        }
+
+        /// <inheritdoc/>
+        public override bool Equals(object obj)
+        {
+            return obj is Symbol sym && Equals(sym);
+        }
+
+        /// <inheritdoc/>
+        public override int GetHashCode()
+        {
+            return HashCode.Combine(Type, _value);
+        }
+
+        /// <inheritdoc/>
+        public override string ToString()
+        {
+            return $"{Type}:{_value}";
+        }
+
+        private static void ThrowSymbolNone()
+        {
+            throw new InvalidOperationException("Symbol refers to nothing.");
+        }
+    }
+}

src/ARMeilleure/CodeGen/Linking/SymbolType.cs

@@ -0,0 +1,28 @@
+namespace ARMeilleure.CodeGen.Linking
+{
+    /// <summary>
+    /// Types of <see cref="Symbol"/>.
+    /// </summary>
+    enum SymbolType : byte
+    {
+        /// <summary>
+        /// Refers to nothing, i.e no symbol.
+        /// </summary>
+        None,
+
+        /// <summary>
+        /// Refers to an entry in <see cref="Translation.Delegates"/>.
+        /// </summary>
+        DelegateTable,
+
+        /// <summary>
+        /// Refers to an entry in <see cref="Translation.Translator.FunctionTable"/>.
+        /// </summary>
+        FunctionTable,
+
+        /// <summary>
+        /// Refers to a special symbol which is handled by <see cref="Translation.PTC.Ptc.PatchCode"/>.
+        /// </summary>
+        Special
+    }
+}

src/ARMeilleure/CodeGen/Optimizations/BlockPlacement.cs

@@ -0,0 +1,72 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class BlockPlacement
+    {
+        public static void RunPass(ControlFlowGraph cfg)
+        {
+            bool update = false;
+
+            BasicBlock block;
+            BasicBlock nextBlock;
+
+            BasicBlock lastBlock = cfg.Blocks.Last;
+
+            // Move cold blocks at the end of the list, so that they are emitted away from hot code.
+            for (block = cfg.Blocks.First; block != null; block = nextBlock)
+            {
+                nextBlock = block.ListNext;
+
+                if (block.Frequency == BasicBlockFrequency.Cold)
+                {
+                    cfg.Blocks.Remove(block);
+                    cfg.Blocks.AddLast(block);
+                }
+
+                if (block == lastBlock)
+                {
+                    break;
+                }
+            }
+
+            for (block = cfg.Blocks.First; block != null; block = nextBlock)
+            {
+                nextBlock = block.ListNext;
+
+                if (block.SuccessorsCount == 2)
+                {
+                    Operation branchOp = block.Operations.Last;
+
+                    Debug.Assert(branchOp.Instruction == Instruction.BranchIf);
+
+                    BasicBlock falseSucc = block.GetSuccessor(0);
+                    BasicBlock trueSucc = block.GetSuccessor(1);
+
+                    // If true successor is next block in list, invert the condition. We avoid extra branching by
+                    // making the true side the fallthrough (i.e, convert it to the false side).
+                    if (trueSucc == block.ListNext)
+                    {
+                        Comparison comp = (Comparison)branchOp.GetSource(2).AsInt32();
+                        Comparison compInv = comp.Invert();
+
+                        branchOp.SetSource(2, Const((int)compInv));
+
+                        block.SetSuccessor(0, trueSucc);
+                        block.SetSuccessor(1, falseSucc);
+
+                        update = true;
+                    }
+                }
+            }
+
+            if (update)
+            {
+                cfg.Update();
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/Optimizations/ConstantFolding.cs

@@ -0,0 +1,346 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class ConstantFolding
+    {
+        public static void RunPass(Operation operation)
+        {
+            if (operation.Destination == default || operation.SourcesCount == 0)
+            {
+                return;
+            }
+
+            if (!AreAllSourcesConstant(operation))
+            {
+                return;
+            }
+
+            OperandType type = operation.Destination.Type;
+
+            switch (operation.Instruction)
+            {
+                case Instruction.Add:
+                    if (operation.GetSource(0).Relocatable ||
+                        operation.GetSource(1).Relocatable)
+                    {
+                        break;
+                    }
+
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x + y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x + y);
+                    }
+                    break;
+
+                case Instruction.BitwiseAnd:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x & y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x & y);
+                    }
+                    break;
+
+                case Instruction.BitwiseExclusiveOr:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x ^ y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x ^ y);
+                    }
+                    break;
+
+                case Instruction.BitwiseNot:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => ~x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => ~x);
+                    }
+                    break;
+
+                case Instruction.BitwiseOr:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x | y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x | y);
+                    }
+                    break;
+
+                case Instruction.ConvertI64ToI32:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => x);
+                    }
+                    break;
+
+                case Instruction.Compare:
+                    if (type == OperandType.I32 &&
+                        operation.GetSource(0).Type == type &&
+                        operation.GetSource(1).Type == type)
+                    {
+                        switch ((Comparison)operation.GetSource(2).Value)
+                        {
+                            case Comparison.Equal:
+                                EvaluateBinaryI32(operation, (x, y) => x == y ? 1 : 0);
+                                break;
+                            case Comparison.NotEqual:
+                                EvaluateBinaryI32(operation, (x, y) => x != y ? 1 : 0);
+                                break;
+                            case Comparison.Greater:
+                                EvaluateBinaryI32(operation, (x, y) => x > y ? 1 : 0);
+                                break;
+                            case Comparison.LessOrEqual:
+                                EvaluateBinaryI32(operation, (x, y) => x <= y ? 1 : 0);
+                                break;
+                            case Comparison.GreaterUI:
+                                EvaluateBinaryI32(operation, (x, y) => (uint)x > (uint)y ? 1 : 0);
+                                break;
+                            case Comparison.LessOrEqualUI:
+                                EvaluateBinaryI32(operation, (x, y) => (uint)x <= (uint)y ? 1 : 0);
+                                break;
+                            case Comparison.GreaterOrEqual:
+                                EvaluateBinaryI32(operation, (x, y) => x >= y ? 1 : 0);
+                                break;
+                            case Comparison.Less:
+                                EvaluateBinaryI32(operation, (x, y) => x < y ? 1 : 0);
+                                break;
+                            case Comparison.GreaterOrEqualUI:
+                                EvaluateBinaryI32(operation, (x, y) => (uint)x >= (uint)y ? 1 : 0);
+                                break;
+                            case Comparison.LessUI:
+                                EvaluateBinaryI32(operation, (x, y) => (uint)x < (uint)y ? 1 : 0);
+                                break;
+                        }
+                    }
+                    break;
+
+                case Instruction.Copy:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => x);
+                    }
+                    break;
+
+                case Instruction.Divide:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => y != 0 ? x / y : 0);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => y != 0 ? x / y : 0);
+                    }
+                    break;
+
+                case Instruction.DivideUI:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => y != 0 ? (int)((uint)x / (uint)y) : 0);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => y != 0 ? (long)((ulong)x / (ulong)y) : 0);
+                    }
+                    break;
+
+                 case Instruction.Multiply:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x * y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x * y);
+                    }
+                    break;
+
+                case Instruction.Negate:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => -x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => -x);
+                    }
+                    break;
+
+                case Instruction.ShiftLeft:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x << y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x << (int)y);
+                    }
+                    break;
+
+                case Instruction.ShiftRightSI:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x >> y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x >> (int)y);
+                    }
+                    break;
+
+                case Instruction.ShiftRightUI:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => (int)((uint)x >> y));
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => (long)((ulong)x >> (int)y));
+                    }
+                    break;
+
+                case Instruction.SignExtend16:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => (short)x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (short)x);
+                    }
+                    break;
+
+                case Instruction.SignExtend32:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (int)x);
+                    }
+                    break;
+
+                case Instruction.SignExtend8:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => (sbyte)x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (sbyte)x);
+                    }
+                    break;
+
+                case Instruction.ZeroExtend16:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => (ushort)x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (ushort)x);
+                    }
+                    break;
+
+                case Instruction.ZeroExtend32:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (uint)x);
+                    }
+                    break;
+
+                case Instruction.ZeroExtend8:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateUnaryI32(operation, (x) => (byte)x);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateUnaryI64(operation, (x) => (byte)x);
+                    }
+                    break;
+
+                case Instruction.Subtract:
+                    if (type == OperandType.I32)
+                    {
+                        EvaluateBinaryI32(operation, (x, y) => x - y);
+                    }
+                    else if (type == OperandType.I64)
+                    {
+                        EvaluateBinaryI64(operation, (x, y) => x - y);
+                    }
+                    break;
+            }
+        }
+
+        private static bool AreAllSourcesConstant(Operation operation)
+        {
+            for (int index = 0; index < operation.SourcesCount; index++)
+            {
+                Operand srcOp = operation.GetSource(index);
+
+                if (srcOp.Kind != OperandKind.Constant)
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
+        private static void EvaluateUnaryI32(Operation operation, Func<int, int> op)
+        {
+            int x = operation.GetSource(0).AsInt32();
+
+            operation.TurnIntoCopy(Const(op(x)));
+        }
+
+        private static void EvaluateUnaryI64(Operation operation, Func<long, long> op)
+        {
+            long x = operation.GetSource(0).AsInt64();
+
+            operation.TurnIntoCopy(Const(op(x)));
+        }
+
+        private static void EvaluateBinaryI32(Operation operation, Func<int, int, int> op)
+        {
+            int x = operation.GetSource(0).AsInt32();
+            int y = operation.GetSource(1).AsInt32();
+
+            operation.TurnIntoCopy(Const(op(x, y)));
+        }
+
+        private static void EvaluateBinaryI64(Operation operation, Func<long, long, long> op)
+        {
+            long x = operation.GetSource(0).AsInt64();
+            long y = operation.GetSource(1).AsInt64();
+
+            operation.TurnIntoCopy(Const(op(x, y)));
+        }
+    }
+}

src/ARMeilleure/CodeGen/Optimizations/Optimizer.cs

@@ -0,0 +1,252 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class Optimizer
+    {
+        public static void RunPass(ControlFlowGraph cfg)
+        {
+            // Scratch buffer used to store uses.
+            Span<Operation> buffer = default;
+
+            bool modified;
+
+            do
+            {
+                modified = false;
+
+                for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
+                {
+                    Operation node;
+                    Operation prevNode;
+
+                    for (node = block.Operations.Last; node != default; node = prevNode)
+                    {
+                        prevNode = node.ListPrevious;
+
+                        if (IsUnused(node))
+                        {
+                            RemoveNode(block, node);
+
+                            modified = true;
+
+                            continue;
+                        }
+                        else if (node.Instruction == Instruction.Phi)
+                        {
+                            continue;
+                        }
+
+                        ConstantFolding.RunPass(node);
+                        Simplification.RunPass(node);
+
+                        if (DestIsSingleLocalVar(node))
+                        {
+                            if (IsPropagableCompare(node))
+                            {
+                                modified |= PropagateCompare(ref buffer, node);
+
+                                if (modified && IsUnused(node))
+                                {
+                                    RemoveNode(block, node);
+                                }
+                            }
+                            else if (IsPropagableCopy(node))
+                            {
+                                PropagateCopy(ref buffer, node);
+
+                                RemoveNode(block, node);
+
+                                modified = true;
+                            }
+                        }
+                    }
+                }
+            }
+            while (modified);
+        }
+
+        public static void RemoveUnusedNodes(ControlFlowGraph cfg)
+        {
+            bool modified;
+
+            do
+            {
+                modified = false;
+
+                for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
+                {
+                    Operation node;
+                    Operation prevNode;
+
+                    for (node = block.Operations.Last; node != default; node = prevNode)
+                    {
+                        prevNode = node.ListPrevious;
+
+                        if (IsUnused(node))
+                        {
+                            RemoveNode(block, node);
+
+                            modified = true;
+                        }
+                    }
+                }
+            }
+            while (modified);
+        }
+
+        private static bool PropagateCompare(ref Span<Operation> buffer, Operation compOp)
+        {
+            // Try to propagate Compare operations into their BranchIf uses, when these BranchIf uses are in the form
+            // of:
+            //
+            // - BranchIf %x, 0x0, Equal        ;; i.e BranchIfFalse %x
+            // - BranchIf %x, 0x0, NotEqual     ;; i.e BranchIfTrue %x
+            //
+            // The commutative property of Equal and NotEqual is taken into consideration as well.
+            //
+            // For example:
+            //
+            //  %x = Compare %a, %b, comp
+            //  BranchIf %x, 0x0, NotEqual
+            //
+            // =>
+            //
+            //  BranchIf %a, %b, comp
+
+            static bool IsZeroBranch(Operation operation, out Comparison compType)
+            {
+                compType = Comparison.Equal;
+
+                if (operation.Instruction != Instruction.BranchIf)
+                {
+                    return false;
+                }
+
+                Operand src1 = operation.GetSource(0);
+                Operand src2 = operation.GetSource(1);
+                Operand comp = operation.GetSource(2);
+
+                compType = (Comparison)comp.AsInt32();
+
+                return (src1.Kind == OperandKind.Constant && src1.Value == 0) ||
+                       (src2.Kind == OperandKind.Constant && src2.Value == 0);
+            }
+
+            bool modified = false;
+
+            Operand dest = compOp.Destination;
+            Operand src1 = compOp.GetSource(0);
+            Operand src2 = compOp.GetSource(1);
+            Operand comp = compOp.GetSource(2);
+
+            Comparison compType = (Comparison)comp.AsInt32();
+
+            Span<Operation> uses = dest.GetUses(ref buffer);
+
+            foreach (Operation use in uses)
+            {
+                // If operation is a BranchIf and has a constant value 0 in its RHS or LHS source operands.
+                if (IsZeroBranch(use, out Comparison otherCompType))
+                {
+                    Comparison propCompType;
+
+                    if (otherCompType == Comparison.NotEqual)
+                    {
+                        propCompType = compType;
+                    }
+                    else if (otherCompType == Comparison.Equal)
+                    {
+                        propCompType = compType.Invert();
+                    }
+                    else
+                    {
+                        continue;
+                    }
+
+                    use.SetSource(0, src1);
+                    use.SetSource(1, src2);
+                    use.SetSource(2, Const((int)propCompType));
+
+                    modified = true;
+                }
+            }
+
+            return modified;
+        }
+
+        private static void PropagateCopy(ref Span<Operation> buffer, Operation copyOp)
+        {
+            // Propagate copy source operand to all uses of the destination operand.
+            Operand dest   = copyOp.Destination;
+            Operand source = copyOp.GetSource(0);
+
+            Span<Operation> uses = dest.GetUses(ref buffer);
+
+            foreach (Operation use in uses)
+            {
+                for (int index = 0; index < use.SourcesCount; index++)
+                {
+                    if (use.GetSource(index) == dest)
+                    {
+                        use.SetSource(index, source);
+                    }
+                }
+            }
+        }
+
+        private static void RemoveNode(BasicBlock block, Operation node)
+        {
+            // Remove a node from the nodes list, and also remove itself
+            // from all the use lists on the operands that this node uses.
+            block.Operations.Remove(node);
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                node.SetSource(index, default);
+            }
+
+            Debug.Assert(node.Destination == default || node.Destination.UsesCount == 0);
+
+            node.Destination = default;
+        }
+
+        private static bool IsUnused(Operation node)
+        {
+            return DestIsSingleLocalVar(node) && node.Destination.UsesCount == 0 && !HasSideEffects(node);
+        }
+
+        private static bool DestIsSingleLocalVar(Operation node)
+        {
+            return node.DestinationsCount == 1 && node.Destination.Kind == OperandKind.LocalVariable;
+        }
+
+        private static bool HasSideEffects(Operation node)
+        {
+            return node.Instruction == Instruction.Call
+                || node.Instruction == Instruction.Tailcall
+                || node.Instruction == Instruction.CompareAndSwap
+                || node.Instruction == Instruction.CompareAndSwap16
+                || node.Instruction == Instruction.CompareAndSwap8;
+        }
+
+        private static bool IsPropagableCompare(Operation operation)
+        {
+            return operation.Instruction == Instruction.Compare;
+        }
+
+        private static bool IsPropagableCopy(Operation operation)
+        {
+            if (operation.Instruction != Instruction.Copy)
+            {
+                return false;
+            }
+
+            return operation.Destination.Type == operation.GetSource(0).Type;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Optimizations/Simplification.cs

@@ -0,0 +1,183 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class Simplification
+    {
+        public static void RunPass(Operation operation)
+        {
+            switch (operation.Instruction)
+            {
+                case Instruction.Add:
+                    if (operation.GetSource(0).Relocatable ||
+                        operation.GetSource(1).Relocatable)
+                    {
+                        break;
+                    }
+
+                    TryEliminateBinaryOpComutative(operation, 0);
+                    break;
+
+                case Instruction.BitwiseAnd:
+                    TryEliminateBitwiseAnd(operation);
+                    break;
+
+                case Instruction.BitwiseOr:
+                    TryEliminateBitwiseOr(operation);
+                    break;
+
+                case Instruction.BitwiseExclusiveOr:
+                    TryEliminateBitwiseExclusiveOr(operation);
+                    break;
+
+                case Instruction.ConditionalSelect:
+                    TryEliminateConditionalSelect(operation);
+                    break;
+
+                case Instruction.Divide:
+                    TryEliminateBinaryOpY(operation, 1);
+                    break;
+
+                case Instruction.Multiply:
+                    TryEliminateBinaryOpComutative(operation, 1);
+                    break;
+
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                case Instruction.Subtract:
+                    TryEliminateBinaryOpY(operation, 0);
+                    break;
+            }
+        }
+
+        private static void TryEliminateBitwiseAnd(Operation operation)
+        {
+            // Try to recognize and optimize those 3 patterns (in order):
+            // x & 0xFFFFFFFF == x,          0xFFFFFFFF & y == y,
+            // x & 0x00000000 == 0x00000000, 0x00000000 & y == 0x00000000
+            Operand x = operation.GetSource(0);
+            Operand y = operation.GetSource(1);
+
+            if (IsConstEqual(x, AllOnes(x.Type)))
+            {
+                operation.TurnIntoCopy(y);
+            }
+            else if (IsConstEqual(y, AllOnes(y.Type)))
+            {
+                operation.TurnIntoCopy(x);
+            }
+            else if (IsConstEqual(x, 0) || IsConstEqual(y, 0))
+            {
+                operation.TurnIntoCopy(Const(x.Type, 0));
+            }
+        }
+
+        private static void TryEliminateBitwiseOr(Operation operation)
+        {
+            // Try to recognize and optimize those 3 patterns (in order):
+            // x | 0x00000000 == x,          0x00000000 | y == y,
+            // x | 0xFFFFFFFF == 0xFFFFFFFF, 0xFFFFFFFF | y == 0xFFFFFFFF
+            Operand x = operation.GetSource(0);
+            Operand y = operation.GetSource(1);
+
+            if (IsConstEqual(x, 0))
+            {
+                operation.TurnIntoCopy(y);
+            }
+            else if (IsConstEqual(y, 0))
+            {
+                operation.TurnIntoCopy(x);
+            }
+            else if (IsConstEqual(x, AllOnes(x.Type)) || IsConstEqual(y, AllOnes(y.Type)))
+            {
+                operation.TurnIntoCopy(Const(AllOnes(x.Type)));
+            }
+        }
+
+        private static void TryEliminateBitwiseExclusiveOr(Operation operation)
+        {
+            // Try to recognize and optimize those 2 patterns (in order):
+            // x ^ y == 0x00000000 when x == y
+            // 0x00000000 ^ y == y, x ^ 0x00000000 == x
+            Operand x = operation.GetSource(0);
+            Operand y = operation.GetSource(1);
+
+            if (x == y && x.Type.IsInteger())
+            {
+                operation.TurnIntoCopy(Const(x.Type, 0));
+            }
+            else
+            {
+                TryEliminateBinaryOpComutative(operation, 0);
+            }
+        }
+
+        private static void TryEliminateBinaryOpY(Operation operation, ulong comparand)
+        {
+            Operand x = operation.GetSource(0);
+            Operand y = operation.GetSource(1);
+
+            if (IsConstEqual(y, comparand))
+            {
+                operation.TurnIntoCopy(x);
+            }
+        }
+
+        private static void TryEliminateBinaryOpComutative(Operation operation, ulong comparand)
+        {
+            Operand x = operation.GetSource(0);
+            Operand y = operation.GetSource(1);
+
+            if (IsConstEqual(x, comparand))
+            {
+                operation.TurnIntoCopy(y);
+            }
+            else if (IsConstEqual(y, comparand))
+            {
+                operation.TurnIntoCopy(x);
+            }
+        }
+
+        private static void TryEliminateConditionalSelect(Operation operation)
+        {
+            Operand cond = operation.GetSource(0);
+
+            if (cond.Kind != OperandKind.Constant)
+            {
+                return;
+            }
+
+            // The condition is constant, we can turn it into a copy, and select
+            // the source based on the condition value.
+            int srcIndex = cond.Value != 0 ? 1 : 2;
+
+            Operand source = operation.GetSource(srcIndex);
+
+            operation.TurnIntoCopy(source);
+        }
+
+        private static bool IsConstEqual(Operand operand, ulong comparand)
+        {
+            if (operand.Kind != OperandKind.Constant || !operand.Type.IsInteger())
+            {
+                return false;
+            }
+
+            return operand.Value == comparand;
+        }
+
+        private static ulong AllOnes(OperandType type)
+        {
+            switch (type)
+            {
+                case OperandType.I32: return ~0U;
+                case OperandType.I64: return ~0UL;
+            }
+
+            throw new ArgumentException("Invalid operand type \"" + type + "\".");
+        }
+    }
+}

src/ARMeilleure/CodeGen/Optimizations/TailMerge.cs

@@ -0,0 +1,83 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class TailMerge
+    {
+        public static void RunPass(in CompilerContext cctx)
+        {
+            ControlFlowGraph cfg = cctx.Cfg;
+
+            BasicBlock mergedReturn = new(cfg.Blocks.Count);
+
+            Operand returnValue;
+            Operation returnOp;
+
+            if (cctx.FuncReturnType == OperandType.None)
+            {
+                returnValue = default;
+                returnOp = Operation(Instruction.Return, default);
+            }
+            else
+            {
+                returnValue = cfg.AllocateLocal(cctx.FuncReturnType);
+                returnOp = Operation(Instruction.Return, default, returnValue);
+            }
+
+            mergedReturn.Frequency = BasicBlockFrequency.Cold;
+            mergedReturn.Operations.AddLast(returnOp);
+
+            for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
+            {
+                Operation op = block.Operations.Last;
+
+                if (op != default && op.Instruction == Instruction.Return)
+                {
+                    block.Operations.Remove(op);
+
+                    if (cctx.FuncReturnType == OperandType.None)
+                    {
+                        PrepareMerge(block, mergedReturn);
+                    }
+                    else
+                    {
+                        Operation copyOp = Operation(Instruction.Copy, returnValue, op.GetSource(0));
+
+                        PrepareMerge(block, mergedReturn).Append(copyOp);
+                    }
+                }
+            }
+
+            cfg.Blocks.AddLast(mergedReturn);
+            cfg.Update();
+        }
+
+        private static BasicBlock PrepareMerge(BasicBlock from, BasicBlock to)
+        {
+            BasicBlock fromPred = from.Predecessors.Count == 1 ? from.Predecessors[0] : null;
+
+            // If the block is empty, we can try to append to the predecessor and avoid unnecessary jumps.
+            if (from.Operations.Count == 0 && fromPred != null && fromPred.SuccessorsCount == 1)
+            {
+                for (int i = 0; i < fromPred.SuccessorsCount; i++)
+                {
+                    if (fromPred.GetSuccessor(i) == from)
+                    {
+                        fromPred.SetSuccessor(i, to);
+                    }
+                }
+
+                // NOTE: `from` becomes unreachable and the call to `cfg.Update()` will remove it.
+                return fromPred;
+            }
+            else
+            {
+                from.AddSuccessor(to);
+
+                return from;
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/PreAllocatorCommon.cs

@@ -0,0 +1,57 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+
+namespace ARMeilleure.CodeGen
+{
+    static class PreAllocatorCommon
+    {
+        public static void Propagate(ref Span<Operation> buffer, Operand dest, Operand value)
+        {
+            ReadOnlySpan<Operation> uses = dest.GetUses(ref buffer);
+
+            foreach (Operation use in uses)
+            {
+                for (int srcIndex = 0; srcIndex < use.SourcesCount; srcIndex++)
+                {
+                    Operand useSrc = use.GetSource(srcIndex);
+
+                    if (useSrc == dest)
+                    {
+                        use.SetSource(srcIndex, value);
+                    }
+                    else if (useSrc.Kind == OperandKind.Memory)
+                    {
+                        MemoryOperand memoryOp = useSrc.GetMemory();
+
+                        Operand baseAddr = memoryOp.BaseAddress;
+                        Operand index = memoryOp.Index;
+                        bool changed = false;
+
+                        if (baseAddr == dest)
+                        {
+                            baseAddr = value;
+                            changed = true;
+                        }
+
+                        if (index == dest)
+                        {
+                            index = value;
+                            changed = true;
+                        }
+
+                        if (changed)
+                        {
+                            use.SetSource(srcIndex, MemoryOp(
+                                useSrc.Type,
+                                baseAddr,
+                                index,
+                                memoryOp.Scale,
+                                memoryOp.Displacement));
+                        }
+                    }
+                }
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/AllocationResult.cs

@@ -0,0 +1,19 @@
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    readonly struct AllocationResult
+    {
+        public int IntUsedRegisters { get; }
+        public int VecUsedRegisters { get; }
+        public int SpillRegionSize  { get; }
+
+        public AllocationResult(
+            int intUsedRegisters,
+            int vecUsedRegisters,
+            int spillRegionSize)
+        {
+            IntUsedRegisters = intUsedRegisters;
+            VecUsedRegisters = vecUsedRegisters;
+            SpillRegionSize  = spillRegionSize;
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/CopyResolver.cs

@@ -0,0 +1,259 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using System.Collections.Generic;
+
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    class CopyResolver
+    {
+        private class ParallelCopy
+        {
+            private readonly struct Copy
+            {
+                public Register Dest   { get; }
+                public Register Source { get; }
+
+                public OperandType Type { get; }
+
+                public Copy(Register dest, Register source, OperandType type)
+                {
+                    Dest   = dest;
+                    Source = source;
+                    Type   = type;
+                }
+            }
+
+            private readonly List<Copy> _copies;
+
+            public int Count => _copies.Count;
+
+            public ParallelCopy()
+            {
+                _copies = new List<Copy>();
+            }
+
+            public void AddCopy(Register dest, Register source, OperandType type)
+            {
+                _copies.Add(new Copy(dest, source, type));
+            }
+
+            public void Sequence(List<Operation> sequence)
+            {
+                Dictionary<Register, Register> locations = new Dictionary<Register, Register>();
+                Dictionary<Register, Register> sources   = new Dictionary<Register, Register>();
+
+                Dictionary<Register, OperandType> types = new Dictionary<Register, OperandType>();
+
+                Queue<Register> pendingQueue = new Queue<Register>();
+                Queue<Register> readyQueue   = new Queue<Register>();
+
+                foreach (Copy copy in _copies)
+                {
+                    locations[copy.Source] = copy.Source;
+                    sources[copy.Dest]     = copy.Source;
+                    types[copy.Dest]       = copy.Type;
+
+                    pendingQueue.Enqueue(copy.Dest);
+                }
+
+                foreach (Copy copy in _copies)
+                {
+                    // If the destination is not used anywhere, we can assign it immediately.
+                    if (!locations.ContainsKey(copy.Dest))
+                    {
+                        readyQueue.Enqueue(copy.Dest);
+                    }
+                }
+
+                while (pendingQueue.TryDequeue(out Register current))
+                {
+                    Register copyDest;
+                    Register origSource;
+                    Register copySource;
+
+                    while (readyQueue.TryDequeue(out copyDest))
+                    {
+                        origSource = sources[copyDest];
+                        copySource = locations[origSource];
+
+                        OperandType type = types[copyDest];
+
+                        EmitCopy(sequence, GetRegister(copyDest, type), GetRegister(copySource, type));
+
+                        locations[origSource] = copyDest;
+
+                        if (origSource == copySource && sources.ContainsKey(origSource))
+                        {
+                            readyQueue.Enqueue(origSource);
+                        }
+                    }
+
+                    copyDest   = current;
+                    origSource = sources[copyDest];
+                    copySource = locations[origSource];
+
+                    if (copyDest != copySource)
+                    {
+                        OperandType type = types[copyDest];
+
+                        type = type.IsInteger() ? OperandType.I64 : OperandType.V128;
+
+                        EmitXorSwap(sequence, GetRegister(copyDest, type), GetRegister(copySource, type));
+
+                        locations[origSource] = copyDest;
+
+                        Register swapOther = copySource;
+
+                        if (copyDest != locations[sources[copySource]])
+                        {
+                            // Find the other swap destination register.
+                            // To do that, we search all the pending registers, and pick
+                            // the one where the copy source register is equal to the
+                            // current destination register being processed (copyDest).
+                            foreach (Register pending in pendingQueue)
+                            {
+                                // Is this a copy of pending <- copyDest?
+                                if (copyDest == locations[sources[pending]])
+                                {
+                                    swapOther = pending;
+
+                                    break;
+                                }
+                            }
+                        }
+
+                        // The value that was previously at "copyDest" now lives on
+                        // "copySource" thanks to the swap, now we need to update the
+                        // location for the next copy that is supposed to copy the value
+                        // that used to live on "copyDest".
+                        locations[sources[swapOther]] = copySource;
+                    }
+                }
+            }
+
+            private static void EmitCopy(List<Operation> sequence, Operand x, Operand y)
+            {
+                sequence.Add(Operation(Instruction.Copy, x, y));
+            }
+
+            private static void EmitXorSwap(List<Operation> sequence, Operand x, Operand y)
+            {
+                sequence.Add(Operation(Instruction.BitwiseExclusiveOr, x, x, y));
+                sequence.Add(Operation(Instruction.BitwiseExclusiveOr, y, y, x));
+                sequence.Add(Operation(Instruction.BitwiseExclusiveOr, x, x, y));
+            }
+        }
+
+        private Queue<Operation> _fillQueue = null;
+        private Queue<Operation> _spillQueue = null;
+        private ParallelCopy _parallelCopy = null;
+
+        public bool HasCopy { get; private set; }
+
+        public void AddSplit(LiveInterval left, LiveInterval right)
+        {
+            if (left.Local != right.Local)
+            {
+                throw new ArgumentException("Intervals of different variables are not allowed.");
+            }
+
+            OperandType type = left.Local.Type;
+
+            if (left.IsSpilled && !right.IsSpilled)
+            {
+                // Move from the stack to a register.
+                AddSplitFill(left, right, type);
+            }
+            else if (!left.IsSpilled && right.IsSpilled)
+            {
+                // Move from a register to the stack.
+                AddSplitSpill(left, right, type);
+            }
+            else if (!left.IsSpilled && !right.IsSpilled && left.Register != right.Register)
+            {
+                // Move from one register to another.
+                AddSplitCopy(left, right, type);
+            }
+            else if (left.SpillOffset != right.SpillOffset)
+            {
+                // This would be the stack-to-stack move case, but this is not supported.
+                throw new ArgumentException("Both intervals were spilled.");
+            }
+        }
+
+        private void AddSplitFill(LiveInterval left, LiveInterval right, OperandType type)
+        {
+            if (_fillQueue == null)
+            {
+                _fillQueue = new Queue<Operation>();
+            }
+
+            Operand register = GetRegister(right.Register, type);
+            Operand offset = Const(left.SpillOffset);
+
+            _fillQueue.Enqueue(Operation(Instruction.Fill, register, offset));
+
+            HasCopy = true;
+        }
+
+        private void AddSplitSpill(LiveInterval left, LiveInterval right, OperandType type)
+        {
+            if (_spillQueue == null)
+            {
+                _spillQueue = new Queue<Operation>();
+            }
+
+            Operand offset = Const(right.SpillOffset);
+            Operand register = GetRegister(left.Register, type);
+
+            _spillQueue.Enqueue(Operation(Instruction.Spill, default, offset, register));
+
+            HasCopy = true;
+        }
+
+        private void AddSplitCopy(LiveInterval left, LiveInterval right, OperandType type)
+        {
+            if (_parallelCopy == null)
+            {
+                _parallelCopy = new ParallelCopy();
+            }
+
+            _parallelCopy.AddCopy(right.Register, left.Register, type);
+
+            HasCopy = true;
+        }
+
+        public Operation[] Sequence()
+        {
+            List<Operation> sequence = new List<Operation>();
+
+            if (_spillQueue != null)
+            {
+                while (_spillQueue.TryDequeue(out Operation spillOp))
+                {
+                    sequence.Add(spillOp);
+                }
+            }
+
+            _parallelCopy?.Sequence(sequence);
+
+            if (_fillQueue != null)
+            {
+                while (_fillQueue.TryDequeue(out Operation fillOp))
+                {
+                    sequence.Add(fillOp);
+                }
+            }
+
+            return sequence.ToArray();
+        }
+
+        private static Operand GetRegister(Register reg, OperandType type)
+        {
+            return Register(reg.Index, reg.Type, type);
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/HybridAllocator.cs

@@ -0,0 +1,454 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Diagnostics;
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    class HybridAllocator : IRegisterAllocator
+    {
+        private readonly struct BlockInfo
+        {
+            public bool HasCall { get; }
+
+            public int IntFixedRegisters { get; }
+            public int VecFixedRegisters { get; }
+
+            public BlockInfo(bool hasCall, int intFixedRegisters, int vecFixedRegisters)
+            {
+                HasCall           = hasCall;
+                IntFixedRegisters = intFixedRegisters;
+                VecFixedRegisters = vecFixedRegisters;
+            }
+        }
+
+        private struct LocalInfo
+        {
+            public int Uses { get; set; }
+            public int UsesAllocated { get; set; }
+            public int Sequence { get; set; }
+            public Operand Temp { get; set; }
+            public Operand Register { get; set; }
+            public Operand SpillOffset { get; set; }
+            public OperandType Type { get; }
+
+            private int _first;
+            private int _last;
+
+            public bool IsBlockLocal => _first == _last;
+
+            public LocalInfo(OperandType type, int uses, int blkIndex)
+            {
+                Uses = uses;
+                Type = type;
+
+                UsesAllocated = 0;
+                Sequence = 0;
+                Temp = default;
+                Register = default;
+                SpillOffset = default;
+
+                _first = -1;
+                _last  = -1;
+
+                SetBlockIndex(blkIndex);
+            }
+
+            public void SetBlockIndex(int blkIndex)
+            {
+                if (_first == -1 || blkIndex < _first)
+                {
+                    _first = blkIndex;
+                }
+
+                if (_last == -1 || blkIndex > _last)
+                {
+                    _last = blkIndex;
+                }
+            }
+        }
+
+        private const int MaxIROperands = 4;
+        // The "visited" state is stored in the MSB of the local's value.
+        private const ulong VisitedMask = 1ul << 63;
+
+        private BlockInfo[] _blockInfo;
+        private LocalInfo[] _localInfo;
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static bool IsVisited(Operand local)
+        {
+            Debug.Assert(local.Kind == OperandKind.LocalVariable);
+
+            return (local.GetValueUnsafe() & VisitedMask) != 0;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void SetVisited(Operand local)
+        {
+            Debug.Assert(local.Kind == OperandKind.LocalVariable);
+
+            local.GetValueUnsafe() |= VisitedMask;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private ref LocalInfo GetLocalInfo(Operand local)
+        {
+            Debug.Assert(local.Kind == OperandKind.LocalVariable);
+            Debug.Assert(IsVisited(local), "Local variable not visited. Used before defined?");
+
+            return ref _localInfo[(uint)local.GetValueUnsafe() - 1];
+        }
+
+        public AllocationResult RunPass(ControlFlowGraph cfg, StackAllocator stackAlloc, RegisterMasks regMasks)
+        {
+            int intUsedRegisters = 0;
+            int vecUsedRegisters = 0;
+
+            int intFreeRegisters = regMasks.IntAvailableRegisters;
+            int vecFreeRegisters = regMasks.VecAvailableRegisters;
+
+            _blockInfo = new BlockInfo[cfg.Blocks.Count];
+            _localInfo = new LocalInfo[cfg.Blocks.Count * 3];
+
+            int localInfoCount = 0;
+
+            for (int index = cfg.PostOrderBlocks.Length - 1; index >= 0; index--)
+            {
+                BasicBlock block = cfg.PostOrderBlocks[index];
+
+                int intFixedRegisters = 0;
+                int vecFixedRegisters = 0;
+
+                bool hasCall = false;
+
+                for (Operation node = block.Operations.First; node != default; node = node.ListNext)
+                {
+                    if (node.Instruction == Instruction.Call)
+                    {
+                        hasCall = true;
+                    }
+
+                    foreach (Operand source in node.SourcesUnsafe)
+                    {
+                        if (source.Kind == OperandKind.LocalVariable)
+                        {
+                            GetLocalInfo(source).SetBlockIndex(block.Index);
+                        }
+                        else if (source.Kind == OperandKind.Memory)
+                        {
+                            MemoryOperand memOp = source.GetMemory();
+
+                            if (memOp.BaseAddress != default)
+                            {
+                                GetLocalInfo(memOp.BaseAddress).SetBlockIndex(block.Index);
+                            }
+
+                            if (memOp.Index != default)
+                            {
+                                GetLocalInfo(memOp.Index).SetBlockIndex(block.Index);
+                            }
+                        }
+                    }
+
+                    foreach (Operand dest in node.DestinationsUnsafe)
+                    {
+                        if (dest.Kind == OperandKind.LocalVariable)
+                        {
+                            if (IsVisited(dest))
+                            {
+                                GetLocalInfo(dest).SetBlockIndex(block.Index);
+                            }
+                            else
+                            {
+                                dest.NumberLocal(++localInfoCount);
+
+                                if (localInfoCount > _localInfo.Length)
+                                {
+                                    Array.Resize(ref _localInfo, localInfoCount * 2);
+                                }
+
+                                SetVisited(dest);
+                                GetLocalInfo(dest) = new LocalInfo(dest.Type, UsesCount(dest), block.Index);
+                            }
+                        }
+                        else if (dest.Kind == OperandKind.Register)
+                        {
+                            if (dest.Type.IsInteger())
+                            {
+                                intFixedRegisters |= 1 << dest.GetRegister().Index;
+                            }
+                            else
+                            {
+                                vecFixedRegisters |= 1 << dest.GetRegister().Index;
+                            }
+                        }
+                    }
+                }
+
+                _blockInfo[block.Index] = new BlockInfo(hasCall, intFixedRegisters, vecFixedRegisters);
+            }
+
+            int sequence = 0;
+
+            for (int index = cfg.PostOrderBlocks.Length - 1; index >= 0; index--)
+            {
+                BasicBlock block = cfg.PostOrderBlocks[index];
+
+                ref BlockInfo blkInfo = ref _blockInfo[block.Index];
+
+                int intLocalFreeRegisters = intFreeRegisters & ~blkInfo.IntFixedRegisters;
+                int vecLocalFreeRegisters = vecFreeRegisters & ~blkInfo.VecFixedRegisters;
+
+                int intCallerSavedRegisters = blkInfo.HasCall ? regMasks.IntCallerSavedRegisters : 0;
+                int vecCallerSavedRegisters = blkInfo.HasCall ? regMasks.VecCallerSavedRegisters : 0;
+
+                int intSpillTempRegisters = SelectSpillTemps(
+                    intCallerSavedRegisters & ~blkInfo.IntFixedRegisters,
+                    intLocalFreeRegisters);
+                int vecSpillTempRegisters = SelectSpillTemps(
+                    vecCallerSavedRegisters & ~blkInfo.VecFixedRegisters,
+                    vecLocalFreeRegisters);
+
+                intLocalFreeRegisters &= ~(intSpillTempRegisters | intCallerSavedRegisters);
+                vecLocalFreeRegisters &= ~(vecSpillTempRegisters | vecCallerSavedRegisters);
+
+                for (Operation node = block.Operations.First; node != default; node = node.ListNext)
+                {
+                    int intLocalUse = 0;
+                    int vecLocalUse = 0;
+
+                    Operand AllocateRegister(Operand local)
+                    {
+                        ref LocalInfo info = ref GetLocalInfo(local);
+
+                        info.UsesAllocated++;
+
+                        Debug.Assert(info.UsesAllocated <= info.Uses);
+
+                        if (info.Register != default)
+                        {
+                            if (info.UsesAllocated == info.Uses)
+                            {
+                                Register reg = info.Register.GetRegister();
+
+                                if (local.Type.IsInteger())
+                                {
+                                    intLocalFreeRegisters |= 1 << reg.Index;
+                                }
+                                else
+                                {
+                                    vecLocalFreeRegisters |= 1 << reg.Index;
+                                }
+                            }
+
+                            return info.Register;
+                        }
+                        else
+                        {
+                            Operand temp = info.Temp;
+
+                            if (temp == default || info.Sequence != sequence)
+                            {
+                                temp = local.Type.IsInteger()
+                                    ? GetSpillTemp(local, intSpillTempRegisters, ref intLocalUse)
+                                    : GetSpillTemp(local, vecSpillTempRegisters, ref vecLocalUse);
+
+                                info.Sequence = sequence;
+                                info.Temp = temp;
+                            }
+
+                            Operation fillOp = Operation(Instruction.Fill, temp, info.SpillOffset);
+
+                            block.Operations.AddBefore(node, fillOp);
+
+                            return temp;
+                        }
+                    }
+
+                    bool folded = false;
+
+                    // If operation is a copy of a local and that local is living on the stack, we turn the copy into
+                    // a fill, instead of inserting a fill before it.
+                    if (node.Instruction == Instruction.Copy)
+                    {
+                        Operand source = node.GetSource(0);
+
+                        if (source.Kind == OperandKind.LocalVariable)
+                        {
+                            ref LocalInfo info = ref GetLocalInfo(source);
+
+                            if (info.Register == default)
+                            {
+                                Operation fillOp = Operation(Instruction.Fill, node.Destination, info.SpillOffset);
+
+                                block.Operations.AddBefore(node, fillOp);
+                                block.Operations.Remove(node);
+
+                                node = fillOp;
+
+                                folded = true;
+                            }
+                        }
+                    }
+
+                    if (!folded)
+                    {
+                        foreach (ref Operand source in node.SourcesUnsafe)
+                        {
+                            if (source.Kind == OperandKind.LocalVariable)
+                            {
+                                source = AllocateRegister(source);
+                            }
+                            else if (source.Kind == OperandKind.Memory)
+                            {
+                                MemoryOperand memOp = source.GetMemory();
+
+                                if (memOp.BaseAddress != default)
+                                {
+                                    memOp.BaseAddress = AllocateRegister(memOp.BaseAddress);
+                                }
+
+                                if (memOp.Index != default)
+                                {
+                                    memOp.Index = AllocateRegister(memOp.Index);
+                                }
+                            }
+                        }
+                    }
+
+                    int intLocalAsg = 0;
+                    int vecLocalAsg = 0;
+
+                    foreach (ref Operand dest in node.DestinationsUnsafe)
+                    {
+                        if (dest.Kind != OperandKind.LocalVariable)
+                        {
+                            continue;
+                        }
+
+                        ref LocalInfo info = ref GetLocalInfo(dest);
+
+                        if (info.UsesAllocated == 0)
+                        {
+                            int mask = dest.Type.IsInteger()
+                                ? intLocalFreeRegisters
+                                : vecLocalFreeRegisters;
+
+                            if (info.IsBlockLocal && mask != 0)
+                            {
+                                int selectedReg = BitOperations.TrailingZeroCount(mask);
+
+                                info.Register = Register(selectedReg, info.Type.ToRegisterType(), info.Type);
+
+                                if (dest.Type.IsInteger())
+                                {
+                                    intLocalFreeRegisters &= ~(1 << selectedReg);
+                                    intUsedRegisters      |=   1 << selectedReg;
+                                }
+                                else
+                                {
+                                    vecLocalFreeRegisters &= ~(1 << selectedReg);
+                                    vecUsedRegisters      |=   1 << selectedReg;
+                                }
+                            }
+                            else
+                            {
+                                info.Register    = default;
+                                info.SpillOffset = Const(stackAlloc.Allocate(dest.Type.GetSizeInBytes()));
+                            }
+                        }
+
+                        info.UsesAllocated++;
+
+                        Debug.Assert(info.UsesAllocated <= info.Uses);
+
+                        if (info.Register != default)
+                        {
+                            dest = info.Register;
+                        }
+                        else
+                        {
+                            Operand temp = info.Temp;
+
+                            if (temp == default || info.Sequence != sequence)
+                            {
+                                temp = dest.Type.IsInteger()
+                                    ? GetSpillTemp(dest, intSpillTempRegisters, ref intLocalAsg)
+                                    : GetSpillTemp(dest, vecSpillTempRegisters, ref vecLocalAsg);
+
+                                info.Sequence = sequence;
+                                info.Temp     = temp;
+                            }
+
+                            dest = temp;
+
+                            Operation spillOp = Operation(Instruction.Spill, default, info.SpillOffset, temp);
+
+                            block.Operations.AddAfter(node, spillOp);
+
+                            node = spillOp;
+                        }
+                    }
+
+                    sequence++;
+
+                    intUsedRegisters |= intLocalAsg | intLocalUse;
+                    vecUsedRegisters |= vecLocalAsg | vecLocalUse;
+                }
+            }
+
+            return new AllocationResult(intUsedRegisters, vecUsedRegisters, stackAlloc.TotalSize);
+        }
+
+        private static int SelectSpillTemps(int mask0, int mask1)
+        {
+            int selection = 0;
+            int count     = 0;
+
+            while (count < MaxIROperands && mask0 != 0)
+            {
+                int mask = mask0 & -mask0;
+
+                selection |= mask;
+
+                mask0 &= ~mask;
+
+                count++;
+            }
+
+            while (count < MaxIROperands && mask1 != 0)
+            {
+                int mask = mask1 & -mask1;
+
+                selection |= mask;
+
+                mask1 &= ~mask;
+
+                count++;
+            }
+
+            Debug.Assert(count == MaxIROperands, "No enough registers for spill temps.");
+
+            return selection;
+        }
+
+        private static Operand GetSpillTemp(Operand local, int freeMask, ref int useMask)
+        {
+            int selectedReg = BitOperations.TrailingZeroCount(freeMask & ~useMask);
+
+            useMask |= 1 << selectedReg;
+
+            return Register(selectedReg, local.Type.ToRegisterType(), local.Type);
+        }
+
+        private static int UsesCount(Operand local)
+        {
+            return local.AssignmentsCount + local.UsesCount;
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/IRegisterAllocator.cs

@@ -0,0 +1,12 @@
+using ARMeilleure.Translation;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    interface IRegisterAllocator
+    {
+        AllocationResult RunPass(
+            ControlFlowGraph cfg,
+            StackAllocator stackAlloc,
+            RegisterMasks regMasks);
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/LiveInterval.cs

@@ -0,0 +1,396 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+using System.Collections.Generic;
+using System.Diagnostics;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    unsafe readonly struct LiveInterval : IComparable<LiveInterval>
+    {
+        public const int NotFound = -1;
+
+        private struct Data
+        {
+            public int End;
+            public int SpillOffset;
+
+            public LiveRange FirstRange;
+            public LiveRange PrevRange;
+            public LiveRange CurrRange;
+
+            public LiveInterval Parent;
+
+            public UseList Uses;
+            public LiveIntervalList Children;
+
+            public Operand Local;
+            public Register Register;
+
+            public bool IsFixed;
+            public bool IsFixedAndUsed;
+        }
+
+        private readonly Data* _data;
+
+        private ref int End => ref _data->End;
+        private ref LiveRange FirstRange => ref _data->FirstRange;
+        private ref LiveRange CurrRange => ref _data->CurrRange;
+        private ref LiveRange PrevRange => ref _data->PrevRange;
+        private ref LiveInterval Parent => ref _data->Parent;
+        private ref UseList Uses => ref _data->Uses;
+        private ref LiveIntervalList Children => ref _data->Children;
+
+        public Operand Local => _data->Local;
+        public ref Register Register => ref _data->Register;
+        public ref int SpillOffset => ref _data->SpillOffset;
+
+        public bool IsFixed => _data->IsFixed;
+        public ref bool IsFixedAndUsed => ref _data->IsFixedAndUsed;
+        public bool IsEmpty => FirstRange == default;
+        public bool IsSplit => Children.Count != 0;
+        public bool IsSpilled => SpillOffset != -1;
+
+        public int UsesCount => Uses.Count;
+
+        public LiveInterval(Operand local = default, LiveInterval parent = default)
+        {
+            _data = Allocators.LiveIntervals.Allocate<Data>();
+            *_data = default;
+
+            _data->IsFixed = false;
+            _data->Local = local;
+
+            Parent = parent == default ? this : parent;
+            Uses = new UseList();
+            Children = new LiveIntervalList();
+
+            FirstRange = default;
+            CurrRange = default;
+            PrevRange = default;
+
+            SpillOffset = -1;
+        }
+
+        public LiveInterval(Register register) : this(local: default, parent: default)
+        {
+            _data->IsFixed = true;
+
+            Register = register;
+        }
+
+        public void Reset()
+        {
+            PrevRange = default;
+            CurrRange = FirstRange;
+        }
+
+        public void Forward(int position)
+        {
+            LiveRange prev = PrevRange;
+            LiveRange curr = CurrRange;
+
+            while (curr != default && curr.Start < position && !curr.Overlaps(position))
+            {
+                prev = curr;
+                curr = curr.Next;
+            }
+
+            PrevRange = prev;
+            CurrRange = curr;
+        }
+
+        public int GetStart()
+        {
+            Debug.Assert(!IsEmpty, "Empty LiveInterval cannot have a start position.");
+
+            return FirstRange.Start;
+        }
+
+        public void SetStart(int position)
+        {
+            if (FirstRange != default)
+            {
+                Debug.Assert(position != FirstRange.End);
+
+                FirstRange.Start = position;
+            }
+            else
+            {
+                FirstRange = new LiveRange(position, position + 1);
+                End = position + 1;
+            }
+        }
+
+        public int GetEnd()
+        {
+            Debug.Assert(!IsEmpty, "Empty LiveInterval cannot have an end position.");
+
+            return End;
+        }
+
+        public void AddRange(int start, int end)
+        {
+            Debug.Assert(start < end, $"Invalid range start position {start}, {end}");
+
+            if (FirstRange != default)
+            {
+                // If the new range ends exactly where the first range start, then coalesce together.
+                if (end == FirstRange.Start)
+                {
+                    FirstRange.Start = start;
+
+                    return;
+                }
+                // If the new range is already contained, then coalesce together.
+                else if (FirstRange.Overlaps(start, end))
+                {
+                    FirstRange.Start = Math.Min(FirstRange.Start, start);
+                    FirstRange.End = Math.Max(FirstRange.End, end);
+                    End = Math.Max(End, end);
+
+                    Debug.Assert(FirstRange.Next == default || !FirstRange.Overlaps(FirstRange.Next));
+                    return;
+                }
+            }
+
+            FirstRange = new LiveRange(start, end, FirstRange);
+            End = Math.Max(End, end);
+
+            Debug.Assert(FirstRange.Next == default || !FirstRange.Overlaps(FirstRange.Next));
+        }
+
+        public void AddUsePosition(int position)
+        {
+            Uses.Add(position);
+        }
+
+        public bool Overlaps(int position)
+        {
+            LiveRange curr = CurrRange;
+
+            while (curr != default && curr.Start <= position)
+            {
+                if (curr.Overlaps(position))
+                {
+                    return true;
+                }
+
+                curr = curr.Next;
+            }
+
+            return false;
+        }
+
+        public bool Overlaps(LiveInterval other)
+        {
+            return GetOverlapPosition(other) != NotFound;
+        }
+
+        public int GetOverlapPosition(LiveInterval other)
+        {
+            LiveRange a = CurrRange;
+            LiveRange b = other.CurrRange;
+
+            while (a != default)
+            {
+                while (b != default && b.Start < a.Start)
+                {
+                    if (a.Overlaps(b))
+                    {
+                        return a.Start;
+                    }
+
+                    b = b.Next;
+                }
+
+                if (b == default)
+                {
+                    break;
+                }
+                else if (a.Overlaps(b))
+                {
+                    return a.Start;
+                }
+
+                a = a.Next;
+            }
+
+            return NotFound;
+        }
+
+        public ReadOnlySpan<LiveInterval> SplitChildren()
+        {
+            return Parent.Children.Span;
+        }
+
+        public ReadOnlySpan<int> UsePositions()
+        {
+            return Uses.Span;
+        }
+
+        public int FirstUse()
+        {
+            return Uses.FirstUse;
+        }
+
+        public int NextUseAfter(int position)
+        {
+            return Uses.NextUse(position);
+        }
+
+        public LiveInterval Split(int position)
+        {
+            LiveInterval result = new(Local, Parent);
+            result.End = End;
+
+            LiveRange prev = PrevRange;
+            LiveRange curr = CurrRange;
+
+            while (curr != default && curr.Start < position && !curr.Overlaps(position))
+            {
+                prev = curr;
+                curr = curr.Next;
+            }
+
+            if (curr.Start >= position)
+            {
+                prev.Next = default;
+
+                result.FirstRange = curr;
+
+                End = prev.End;
+            }
+            else
+            {
+                result.FirstRange = new LiveRange(position, curr.End, curr.Next);
+
+                curr.End = position;
+                curr.Next = default;
+
+                End = curr.End;
+            }
+
+            result.Uses = Uses.Split(position);
+
+            AddSplitChild(result);
+
+            Debug.Assert(!IsEmpty, "Left interval is empty after split.");
+            Debug.Assert(!result.IsEmpty, "Right interval is empty after split.");
+
+            // Make sure the iterator in the new split is pointing to the start.
+            result.Reset();
+
+            return result;
+        }
+
+        private void AddSplitChild(LiveInterval child)
+        {
+            Debug.Assert(!child.IsEmpty, "Trying to insert an empty interval.");
+
+            Parent.Children.Add(child);
+        }
+
+        public LiveInterval GetSplitChild(int position)
+        {
+            if (Overlaps(position))
+            {
+                return this;
+            }
+
+            foreach (LiveInterval splitChild in SplitChildren())
+            {
+                if (splitChild.Overlaps(position))
+                {
+                    return splitChild;
+                }
+                else if (splitChild.GetStart() > position)
+                {
+                    break;
+                }
+            }
+
+            return default;
+        }
+
+        public bool TrySpillWithSiblingOffset()
+        {
+            foreach (LiveInterval splitChild in SplitChildren())
+            {
+                if (splitChild.IsSpilled)
+                {
+                    Spill(splitChild.SpillOffset);
+
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        public void Spill(int offset)
+        {
+            SpillOffset = offset;
+        }
+
+        public int CompareTo(LiveInterval interval)
+        {
+            if (FirstRange == default || interval.FirstRange == default)
+            {
+                return 0;
+            }
+
+            return GetStart().CompareTo(interval.GetStart());
+        }
+
+        public bool Equals(LiveInterval interval)
+        {
+            return interval._data == _data;
+        }
+
+        public override bool Equals(object obj)
+        {
+            return obj is LiveInterval interval && Equals(interval);
+        }
+
+        public static bool operator ==(LiveInterval a, LiveInterval b)
+        {
+            return a.Equals(b);
+        }
+
+        public static bool operator !=(LiveInterval a, LiveInterval b)
+        {
+            return !a.Equals(b);
+        }
+
+        public override int GetHashCode()
+        {
+            return HashCode.Combine((IntPtr)_data);
+        }
+
+        public override string ToString()
+        {
+            LiveInterval self = this;
+
+            IEnumerable<string> GetRanges()
+            {
+                LiveRange curr = self.CurrRange;
+
+                while (curr != default)
+                {
+                    if (curr == self.CurrRange)
+                    {
+                        yield return "*" + curr;
+                    }
+                    else
+                    {
+                        yield return curr.ToString();
+                    }
+
+                    curr = curr.Next;
+                }
+            }
+
+            return string.Join(", ", GetRanges());
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/LiveIntervalList.cs

@@ -0,0 +1,40 @@
+using System;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    unsafe struct LiveIntervalList
+    {
+        private LiveInterval* _items;
+        private int _count;
+        private int _capacity;
+
+        public int Count => _count;
+        public Span<LiveInterval> Span => new(_items, _count);
+
+        public void Add(LiveInterval interval)
+        {
+            if (_count + 1 > _capacity)
+            {
+                var oldSpan = Span;
+
+                _capacity = Math.Max(4, _capacity * 2);
+                _items = Allocators.References.Allocate<LiveInterval>((uint)_capacity);
+
+                var newSpan = Span;
+
+                oldSpan.CopyTo(newSpan);
+            }
+
+            int position = interval.GetStart();
+            int i = _count - 1;
+
+            while (i >= 0 && _items[i].GetStart() > position)
+            {
+                _items[i + 1] = _items[i--];
+            }
+
+            _items[i + 1] = interval;
+            _count++;
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/LiveRange.cs

@@ -0,0 +1,74 @@
+using System;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    unsafe readonly struct LiveRange : IEquatable<LiveRange>
+    {
+        private struct Data
+        {
+            public int Start;
+            public int End;
+            public LiveRange Next;
+        }
+
+        private readonly Data* _data;
+
+        public ref int Start => ref _data->Start;
+        public ref int End => ref _data->End;
+        public ref LiveRange Next => ref _data->Next;
+
+        public LiveRange(int start, int end, LiveRange next = default)
+        {
+            _data = Allocators.LiveRanges.Allocate<Data>();
+
+            Start = start;
+            End = end;
+            Next = next;
+        }
+
+        public bool Overlaps(int start, int end)
+        {
+            return Start < end && start < End;
+        }
+
+        public bool Overlaps(LiveRange range)
+        {
+            return Start < range.End && range.Start < End;
+        }
+
+        public bool Overlaps(int position)
+        {
+            return position >= Start && position < End;
+        }
+
+        public bool Equals(LiveRange range)
+        {
+            return range._data == _data;
+        }
+
+        public override bool Equals(object obj)
+        {
+            return obj is LiveRange range && Equals(range);
+        }
+
+        public static bool operator ==(LiveRange a, LiveRange b)
+        {
+            return a.Equals(b);
+        }
+
+        public static bool operator !=(LiveRange a, LiveRange b)
+        {
+            return !a.Equals(b);
+        }
+
+        public override int GetHashCode()
+        {
+            return HashCode.Combine((IntPtr)_data);
+        }
+
+        public override string ToString()
+        {
+            return $"[{Start}, {End})";
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/RegisterMasks.cs

@@ -0,0 +1,50 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    readonly struct RegisterMasks
+    {
+        public int IntAvailableRegisters   { get; }
+        public int VecAvailableRegisters   { get; }
+        public int IntCallerSavedRegisters { get; }
+        public int VecCallerSavedRegisters { get; }
+        public int IntCalleeSavedRegisters { get; }
+        public int VecCalleeSavedRegisters { get; }
+        public int RegistersCount { get; }
+
+        public RegisterMasks(
+            int intAvailableRegisters,
+            int vecAvailableRegisters,
+            int intCallerSavedRegisters,
+            int vecCallerSavedRegisters,
+            int intCalleeSavedRegisters,
+            int vecCalleeSavedRegisters,
+            int registersCount)
+        {
+            IntAvailableRegisters   = intAvailableRegisters;
+            VecAvailableRegisters   = vecAvailableRegisters;
+            IntCallerSavedRegisters = intCallerSavedRegisters;
+            VecCallerSavedRegisters = vecCallerSavedRegisters;
+            IntCalleeSavedRegisters = intCalleeSavedRegisters;
+            VecCalleeSavedRegisters = vecCalleeSavedRegisters;
+            RegistersCount          = registersCount;
+        }
+
+        public int GetAvailableRegisters(RegisterType type)
+        {
+            if (type == RegisterType.Integer)
+            {
+                return IntAvailableRegisters;
+            }
+            else if (type == RegisterType.Vector)
+            {
+                return VecAvailableRegisters;
+            }
+            else
+            {
+                throw new ArgumentException($"Invalid register type \"{type}\".");
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/StackAllocator.cs

@@ -0,0 +1,25 @@
+using ARMeilleure.IntermediateRepresentation;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    class StackAllocator
+    {
+        private int _offset;
+
+        public int TotalSize => _offset;
+
+        public int Allocate(OperandType type)
+        {
+            return Allocate(type.GetSizeInBytes());
+        }
+
+        public int Allocate(int sizeInBytes)
+        {
+            int offset = _offset;
+
+            _offset += sizeInBytes;
+
+            return offset;
+        }
+    }
+}

src/ARMeilleure/CodeGen/RegisterAllocators/UseList.cs

@@ -0,0 +1,84 @@
+using System;
+
+namespace ARMeilleure.CodeGen.RegisterAllocators
+{
+    unsafe struct UseList
+    {
+        private int* _items;
+        private int _capacity;
+        private int _count;
+
+        public int Count => _count;
+        public int FirstUse => _count > 0 ? _items[_count - 1] : LiveInterval.NotFound;
+        public Span<int> Span => new(_items, _count);
+
+        public void Add(int position)
+        {
+            if (_count + 1 > _capacity)
+            {
+                var oldSpan = Span;
+
+                _capacity = Math.Max(4, _capacity * 2);
+                _items = Allocators.Default.Allocate<int>((uint)_capacity);
+
+                var newSpan = Span;
+
+                oldSpan.CopyTo(newSpan);
+            }
+
+            // Use positions are usually inserted in descending order, so inserting in descending order is faster,
+            // since the number of half exchanges is reduced.
+            int i = _count - 1;
+
+            while (i >= 0 && _items[i] < position)
+            {
+                _items[i + 1] = _items[i--];
+            }
+
+            _items[i + 1] = position;
+            _count++;
+        }
+
+        public int NextUse(int position)
+        {
+            int index = NextUseIndex(position);
+
+            return index != LiveInterval.NotFound ? _items[index] : LiveInterval.NotFound;
+        }
+
+        public int NextUseIndex(int position)
+        {
+            int i = _count - 1;
+
+            if (i == -1 || position > _items[0])
+            {
+                return LiveInterval.NotFound;
+            }
+
+            while (i >= 0 && _items[i] < position)
+            {
+                i--;
+            }
+
+            return i;
+        }
+
+        public UseList Split(int position)
+        {
+            int index = NextUseIndex(position);
+
+            // Since the list is in descending order, the new split list takes the front of the list and the current
+            // list takes the back of the list.
+            UseList result = new();
+            result._count = index + 1;
+            result._capacity = result._count;
+            result._items = _items;
+
+            _count = _count - result._count;
+            _capacity = _count;
+            _items = _items + result._count;
+
+            return result;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Unwinding/UnwindInfo.cs

@@ -0,0 +1,16 @@
+namespace ARMeilleure.CodeGen.Unwinding
+{
+    struct UnwindInfo
+    {
+        public const int Stride = 4; // Bytes.
+
+        public UnwindPushEntry[] PushEntries { get; }
+        public int PrologSize { get; }
+
+        public UnwindInfo(UnwindPushEntry[] pushEntries, int prologSize)
+        {
+            PushEntries = pushEntries;
+            PrologSize = prologSize;
+        }
+    }
+}

src/ARMeilleure/CodeGen/Unwinding/UnwindPseudoOp.cs

@@ -0,0 +1,11 @@
+namespace ARMeilleure.CodeGen.Unwinding
+{
+    enum UnwindPseudoOp
+    {
+        PushReg    = 0,
+        SetFrame   = 1,
+        AllocStack = 2,
+        SaveReg    = 3,
+        SaveXmm128 = 4
+    }
+}

src/ARMeilleure/CodeGen/Unwinding/UnwindPushEntry.cs

@@ -0,0 +1,20 @@
+namespace ARMeilleure.CodeGen.Unwinding
+{
+    struct UnwindPushEntry
+    {
+        public const int Stride = 16; // Bytes.
+
+        public UnwindPseudoOp PseudoOp { get; }
+        public int PrologOffset { get; }
+        public int RegIndex { get; }
+        public int StackOffsetOrAllocSize { get; }
+
+        public UnwindPushEntry(UnwindPseudoOp pseudoOp, int prologOffset, int regIndex = -1, int stackOffsetOrAllocSize = -1)
+        {
+            PseudoOp = pseudoOp;
+            PrologOffset = prologOffset;
+            RegIndex = regIndex;
+            StackOffsetOrAllocSize = stackOffsetOrAllocSize;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/AssemblerTable.cs

@@ -0,0 +1,295 @@
+using System;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    partial class Assembler
+    {
+        public static bool SupportsVexPrefix(X86Instruction inst)
+        {
+            return _instTable[(int)inst].Flags.HasFlag(InstructionFlags.Vex);
+        }
+
+        private const int BadOp = 0;
+
+        [Flags]
+        private enum InstructionFlags
+        {
+            None     = 0,
+            RegOnly  = 1 << 0,
+            Reg8Src  = 1 << 1,
+            Reg8Dest = 1 << 2,
+            RexW     = 1 << 3,
+            Vex      = 1 << 4,
+            Evex     = 1 << 5,
+
+            PrefixBit  = 16,
+            PrefixMask = 7 << PrefixBit,
+            Prefix66   = 1 << PrefixBit,
+            PrefixF3   = 2 << PrefixBit,
+            PrefixF2   = 4 << PrefixBit
+        }
+
+        private readonly struct InstructionInfo
+        {
+            public int OpRMR     { get; }
+            public int OpRMImm8  { get; }
+            public int OpRMImm32 { get; }
+            public int OpRImm64  { get; }
+            public int OpRRM     { get; }
+
+            public InstructionFlags Flags { get; }
+
+            public InstructionInfo(
+                int              opRMR,
+                int              opRMImm8,
+                int              opRMImm32,
+                int              opRImm64,
+                int              opRRM,
+                InstructionFlags flags)
+            {
+                OpRMR     = opRMR;
+                OpRMImm8  = opRMImm8;
+                OpRMImm32 = opRMImm32;
+                OpRImm64  = opRImm64;
+                OpRRM     = opRRM;
+                Flags     = flags;
+            }
+        }
+
+        private readonly static InstructionInfo[] _instTable;
+
+        static Assembler()
+        {
+            _instTable = new InstructionInfo[(int)X86Instruction.Count];
+
+            //  Name                                             RM/R        RM/I8       RM/I32      R/I64       R/RM        Flags
+            Add(X86Instruction.Add,           new InstructionInfo(0x00000001, 0x00000083, 0x00000081, BadOp,      0x00000003, InstructionFlags.None));
+            Add(X86Instruction.Addpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f58, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Addps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f58, InstructionFlags.Vex));
+            Add(X86Instruction.Addsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f58, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Addss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f58, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Aesdec,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38de, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Aesdeclast,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38df, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Aesenc,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38dc, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Aesenclast,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38dd, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Aesimc,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38db, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.And,           new InstructionInfo(0x00000021, 0x04000083, 0x04000081, BadOp,      0x00000023, InstructionFlags.None));
+            Add(X86Instruction.Andnpd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f55, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Andnps,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f55, InstructionFlags.Vex));
+            Add(X86Instruction.Andpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f54, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Andps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f54, InstructionFlags.Vex));
+            Add(X86Instruction.Blendvpd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3815, InstructionFlags.Prefix66));
+            Add(X86Instruction.Blendvps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3814, InstructionFlags.Prefix66));
+            Add(X86Instruction.Bsr,           new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fbd, InstructionFlags.None));
+            Add(X86Instruction.Bswap,         new InstructionInfo(0x00000fc8, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.RegOnly));
+            Add(X86Instruction.Call,          new InstructionInfo(0x020000ff, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Cmovcc,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f40, InstructionFlags.None));
+            Add(X86Instruction.Cmp,           new InstructionInfo(0x00000039, 0x07000083, 0x07000081, BadOp,      0x0000003b, InstructionFlags.None));
+            Add(X86Instruction.Cmppd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc2, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Cmpps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc2, InstructionFlags.Vex));
+            Add(X86Instruction.Cmpsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc2, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Cmpss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc2, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Cmpxchg,       new InstructionInfo(0x00000fb1, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Cmpxchg16b,    new InstructionInfo(0x01000fc7, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.RexW));
+            Add(X86Instruction.Cmpxchg8,      new InstructionInfo(0x00000fb0, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Reg8Src));
+            Add(X86Instruction.Comisd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2f, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Comiss,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2f, InstructionFlags.Vex));
+            Add(X86Instruction.Crc32,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38f1, InstructionFlags.PrefixF2));
+            Add(X86Instruction.Crc32_16,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38f1, InstructionFlags.PrefixF2 | InstructionFlags.Prefix66));
+            Add(X86Instruction.Crc32_8,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38f0, InstructionFlags.PrefixF2 | InstructionFlags.Reg8Src));
+            Add(X86Instruction.Cvtdq2pd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fe6, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Cvtdq2ps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5b, InstructionFlags.Vex));
+            Add(X86Instruction.Cvtpd2dq,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fe6, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Cvtpd2ps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5a, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Cvtps2dq,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5b, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Cvtps2pd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5a, InstructionFlags.Vex));
+            Add(X86Instruction.Cvtsd2si,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2d, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Cvtsd2ss,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5a, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Cvtsi2sd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2a, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Cvtsi2ss,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2a, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Cvtss2sd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5a, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Cvtss2si,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f2d, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Div,           new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x060000f7, InstructionFlags.None));
+            Add(X86Instruction.Divpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5e, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Divps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5e, InstructionFlags.Vex));
+            Add(X86Instruction.Divsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5e, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Divss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5e, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Gf2p8affineqb, new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3ace, InstructionFlags.Prefix66));
+            Add(X86Instruction.Haddpd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f7c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Haddps,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f7c, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Idiv,          new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x070000f7, InstructionFlags.None));
+            Add(X86Instruction.Imul,          new InstructionInfo(BadOp,      0x0000006b, 0x00000069, BadOp,      0x00000faf, InstructionFlags.None));
+            Add(X86Instruction.Imul128,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x050000f7, InstructionFlags.None));
+            Add(X86Instruction.Insertps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a21, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Jmp,           new InstructionInfo(0x040000ff, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Ldmxcsr,       new InstructionInfo(0x02000fae, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex));
+            Add(X86Instruction.Lea,           new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x0000008d, InstructionFlags.None));
+            Add(X86Instruction.Maxpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5f, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Maxps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5f, InstructionFlags.Vex));
+            Add(X86Instruction.Maxsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5f, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Maxss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5f, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Minpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5d, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Minps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5d, InstructionFlags.Vex));
+            Add(X86Instruction.Minsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5d, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Minss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5d, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Mov,           new InstructionInfo(0x00000089, BadOp,      0x000000c7, 0x000000b8, 0x0000008b, InstructionFlags.None));
+            Add(X86Instruction.Mov16,         new InstructionInfo(0x00000089, BadOp,      0x000000c7, BadOp,      0x0000008b, InstructionFlags.Prefix66));
+            Add(X86Instruction.Mov8,          new InstructionInfo(0x00000088, 0x000000c6, BadOp,      BadOp,      0x0000008a, InstructionFlags.Reg8Src | InstructionFlags.Reg8Dest));
+            Add(X86Instruction.Movd,          new InstructionInfo(0x00000f7e, BadOp,      BadOp,      BadOp,      0x00000f6e, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Movdqu,        new InstructionInfo(0x00000f7f, BadOp,      BadOp,      BadOp,      0x00000f6f, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Movhlps,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f12, InstructionFlags.Vex));
+            Add(X86Instruction.Movlhps,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f16, InstructionFlags.Vex));
+            Add(X86Instruction.Movq,          new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f7e, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Movsd,         new InstructionInfo(0x00000f11, BadOp,      BadOp,      BadOp,      0x00000f10, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Movss,         new InstructionInfo(0x00000f11, BadOp,      BadOp,      BadOp,      0x00000f10, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Movsx16,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fbf, InstructionFlags.None));
+            Add(X86Instruction.Movsx32,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000063, InstructionFlags.None));
+            Add(X86Instruction.Movsx8,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fbe, InstructionFlags.Reg8Src));
+            Add(X86Instruction.Movzx16,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fb7, InstructionFlags.None));
+            Add(X86Instruction.Movzx8,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fb6, InstructionFlags.Reg8Src));
+            Add(X86Instruction.Mul128,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x040000f7, InstructionFlags.None));
+            Add(X86Instruction.Mulpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f59, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Mulps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f59, InstructionFlags.Vex));
+            Add(X86Instruction.Mulsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f59, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Mulss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f59, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Neg,           new InstructionInfo(0x030000f7, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Not,           new InstructionInfo(0x020000f7, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Or,            new InstructionInfo(0x00000009, 0x01000083, 0x01000081, BadOp,      0x0000000b, InstructionFlags.None));
+            Add(X86Instruction.Paddb,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ffc, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            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));
+            Add(X86Instruction.Pavgw,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fe3, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pblendvb,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3810, InstructionFlags.Prefix66));
+            Add(X86Instruction.Pclmulqdq,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a44, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpeqb,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f74, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpeqd,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f76, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpeqq,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3829, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpeqw,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f75, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpgtb,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f64, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpgtd,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f66, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpgtq,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3837, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pcmpgtw,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f65, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pextrb,        new InstructionInfo(0x000f3a14, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pextrd,        new InstructionInfo(0x000f3a16, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pextrq,        new InstructionInfo(0x000f3a16, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.RexW | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pextrw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc5, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pinsrb,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a20, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pinsrd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a22, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pinsrq,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a22, InstructionFlags.Vex | InstructionFlags.RexW | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pinsrw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc4, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxsb,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxsd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383d, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxsw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fee, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxub,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fde, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxud,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383f, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmaxuw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383e, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminsb,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3838, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminsd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3839, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminsw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fea, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminub,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fda, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminud,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383b, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pminuw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f383a, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovsxbw,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3820, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovsxdq,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3825, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovsxwd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3823, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovzxbw,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3830, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovzxdq,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3835, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmovzxwd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3833, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmulld,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3840, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pmullw,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fd5, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pop,           new InstructionInfo(0x0000008f, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Popcnt,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fb8, InstructionFlags.PrefixF3));
+            Add(X86Instruction.Por,           new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000feb, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pshufb,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3800, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pshufd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f70, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pslld,         new InstructionInfo(BadOp,      0x06000f72, BadOp,      BadOp,      0x00000ff2, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Pslldq,        new InstructionInfo(BadOp,      0x07000f73, BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psllq,         new InstructionInfo(BadOp,      0x06000f73, BadOp,      BadOp,      0x00000ff3, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psllw,         new InstructionInfo(BadOp,      0x06000f71, BadOp,      BadOp,      0x00000ff1, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psrad,         new InstructionInfo(BadOp,      0x04000f72, BadOp,      BadOp,      0x00000fe2, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psraw,         new InstructionInfo(BadOp,      0x04000f71, BadOp,      BadOp,      0x00000fe1, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psrld,         new InstructionInfo(BadOp,      0x02000f72, BadOp,      BadOp,      0x00000fd2, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psrlq,         new InstructionInfo(BadOp,      0x02000f73, BadOp,      BadOp,      0x00000fd3, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psrldq,        new InstructionInfo(BadOp,      0x03000f73, BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psrlw,         new InstructionInfo(BadOp,      0x02000f71, BadOp,      BadOp,      0x00000fd1, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psubb,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ff8, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psubd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ffa, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psubq,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ffb, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Psubw,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000ff9, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpckhbw,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f68, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpckhdq,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f6a, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpckhqdq,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f6d, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpckhwd,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f69, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpcklbw,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f60, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpckldq,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f62, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpcklqdq,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f6c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Punpcklwd,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f61, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Push,          new InstructionInfo(BadOp,      0x0000006a, 0x00000068, BadOp,      0x060000ff, InstructionFlags.None));
+            Add(X86Instruction.Pxor,          new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fef, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Rcpps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f53, InstructionFlags.Vex));
+            Add(X86Instruction.Rcpss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f53, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Ror,           new InstructionInfo(0x010000d3, 0x010000c1, BadOp,      BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Roundpd,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a09, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Roundps,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a08, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Roundsd,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a0b, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Roundss,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a0a, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Rsqrtps,       new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f52, InstructionFlags.Vex));
+            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));
+            Add(X86Instruction.Shufps,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000fc6, InstructionFlags.Vex));
+            Add(X86Instruction.Sqrtpd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f51, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Sqrtps,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f51, InstructionFlags.Vex));
+            Add(X86Instruction.Sqrtsd,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f51, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Sqrtss,        new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f51, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Stmxcsr,       new InstructionInfo(0x03000fae, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex));
+            Add(X86Instruction.Sub,           new InstructionInfo(0x00000029, 0x05000083, 0x05000081, BadOp,      0x0000002b, InstructionFlags.None));
+            Add(X86Instruction.Subpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Subps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5c, InstructionFlags.Vex));
+            Add(X86Instruction.Subsd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5c, InstructionFlags.Vex | InstructionFlags.PrefixF2));
+            Add(X86Instruction.Subss,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f5c, InstructionFlags.Vex | InstructionFlags.PrefixF3));
+            Add(X86Instruction.Test,          new InstructionInfo(0x00000085, BadOp,      0x000000f7, BadOp,      BadOp,      InstructionFlags.None));
+            Add(X86Instruction.Unpckhpd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f15, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Unpckhps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f15, InstructionFlags.Vex));
+            Add(X86Instruction.Unpcklpd,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f14, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Unpcklps,      new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f14, InstructionFlags.Vex));
+            Add(X86Instruction.Vblendvpd,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a4b, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vblendvps,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a4a, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vcvtph2ps,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3813, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vcvtps2ph,     new InstructionInfo(0x000f3a1d, BadOp,      BadOp,      BadOp,      BadOp,      InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfmadd231pd,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38b8, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfmadd231ps,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38b8, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfmadd231sd,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38b9, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfmadd231ss,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38b9, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfmsub231sd,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bb, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfmsub231ss,   new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bb, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfnmadd231pd,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bc, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfnmadd231ps,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bc, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfnmadd231sd,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bd, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfnmadd231ss,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bd, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vfnmsub231sd,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bf, InstructionFlags.Vex | InstructionFlags.Prefix66 | InstructionFlags.RexW));
+            Add(X86Instruction.Vfnmsub231ss,  new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f38bf, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vpblendvb,     new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a4c, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Vpternlogd,    new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x000f3a25, InstructionFlags.Evex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Xor,           new InstructionInfo(0x00000031, 0x06000083, 0x06000081, BadOp,      0x00000033, InstructionFlags.None));
+            Add(X86Instruction.Xorpd,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f57, InstructionFlags.Vex | InstructionFlags.Prefix66));
+            Add(X86Instruction.Xorps,         new InstructionInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f57, InstructionFlags.Vex));
+
+            static void Add(X86Instruction inst, in InstructionInfo info)
+            {
+                _instTable[(int)inst] = info;
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/CallConvName.cs

@@ -0,0 +1,8 @@
+namespace ARMeilleure.CodeGen.X86
+{
+    enum CallConvName
+    {
+        SystemV,
+        Windows
+    }
+}

src/ARMeilleure/CodeGen/X86/CallingConvention.cs

@@ -0,0 +1,158 @@
+using System;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    static class CallingConvention
+    {
+        private const int RegistersMask = 0xffff;
+
+        public static int GetIntAvailableRegisters()
+        {
+            return RegistersMask & ~(1 << (int)X86Register.Rsp);
+        }
+
+        public static int GetVecAvailableRegisters()
+        {
+            return RegistersMask;
+        }
+
+        public static int GetIntCallerSavedRegisters()
+        {
+            if (GetCurrentCallConv() == CallConvName.Windows)
+            {
+                return (1 << (int)X86Register.Rax) |
+                       (1 << (int)X86Register.Rcx) |
+                       (1 << (int)X86Register.Rdx) |
+                       (1 << (int)X86Register.R8)  |
+                       (1 << (int)X86Register.R9)  |
+                       (1 << (int)X86Register.R10) |
+                       (1 << (int)X86Register.R11);
+            }
+            else /* if (GetCurrentCallConv() == CallConvName.SystemV) */
+            {
+                return (1 << (int)X86Register.Rax) |
+                       (1 << (int)X86Register.Rcx) |
+                       (1 << (int)X86Register.Rdx) |
+                       (1 << (int)X86Register.Rsi) |
+                       (1 << (int)X86Register.Rdi) |
+                       (1 << (int)X86Register.R8)  |
+                       (1 << (int)X86Register.R9)  |
+                       (1 << (int)X86Register.R10) |
+                       (1 << (int)X86Register.R11);
+            }
+        }
+
+        public static int GetVecCallerSavedRegisters()
+        {
+            if (GetCurrentCallConv() == CallConvName.Windows)
+            {
+                return (1 << (int)X86Register.Xmm0) |
+                       (1 << (int)X86Register.Xmm1) |
+                       (1 << (int)X86Register.Xmm2) |
+                       (1 << (int)X86Register.Xmm3) |
+                       (1 << (int)X86Register.Xmm4) |
+                       (1 << (int)X86Register.Xmm5);
+            }
+            else /* if (GetCurrentCallConv() == CallConvName.SystemV) */
+            {
+                return RegistersMask;
+            }
+        }
+
+        public static int GetIntCalleeSavedRegisters()
+        {
+            return GetIntCallerSavedRegisters() ^ RegistersMask;
+        }
+
+        public static int GetVecCalleeSavedRegisters()
+        {
+            return GetVecCallerSavedRegisters() ^ RegistersMask;
+        }
+
+        public static int GetArgumentsOnRegsCount()
+        {
+            return 4;
+        }
+
+        public static int GetIntArgumentsOnRegsCount()
+        {
+            return 6;
+        }
+
+        public static int GetVecArgumentsOnRegsCount()
+        {
+            return 8;
+        }
+
+        public static X86Register GetIntArgumentRegister(int index)
+        {
+            if (GetCurrentCallConv() == CallConvName.Windows)
+            {
+                switch (index)
+                {
+                    case 0: return X86Register.Rcx;
+                    case 1: return X86Register.Rdx;
+                    case 2: return X86Register.R8;
+                    case 3: return X86Register.R9;
+                }
+            }
+            else /* if (GetCurrentCallConv() == CallConvName.SystemV) */
+            {
+                switch (index)
+                {
+                    case 0: return X86Register.Rdi;
+                    case 1: return X86Register.Rsi;
+                    case 2: return X86Register.Rdx;
+                    case 3: return X86Register.Rcx;
+                    case 4: return X86Register.R8;
+                    case 5: return X86Register.R9;
+                }
+            }
+
+            throw new ArgumentOutOfRangeException(nameof(index));
+        }
+
+        public static X86Register GetVecArgumentRegister(int index)
+        {
+            int count;
+
+            if (GetCurrentCallConv() == CallConvName.Windows)
+            {
+                count = 4;
+            }
+            else /* if (GetCurrentCallConv() == CallConvName.SystemV) */
+            {
+                count = 8;
+            }
+
+            if ((uint)index < count)
+            {
+                return X86Register.Xmm0 + index;
+            }
+
+            throw new ArgumentOutOfRangeException(nameof(index));
+        }
+
+        public static X86Register GetIntReturnRegister()
+        {
+            return X86Register.Rax;
+        }
+
+        public static X86Register GetIntReturnRegisterHigh()
+        {
+            return X86Register.Rdx;
+        }
+
+        public static X86Register GetVecReturnRegister()
+        {
+            return X86Register.Xmm0;
+        }
+
+        public static CallConvName GetCurrentCallConv()
+        {
+            return OperatingSystem.IsWindows()
+                ? CallConvName.Windows
+                : CallConvName.SystemV;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/CodeGenCommon.cs

@@ -0,0 +1,19 @@
+using ARMeilleure.IntermediateRepresentation;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    static class CodeGenCommon
+    {
+        public static bool IsLongConst(Operand op)
+        {
+            long value = op.Type == OperandType.I32 ? op.AsInt32() : op.AsInt64();
+
+            return !ConstFitsOnS32(value);
+        }
+
+        private static bool ConstFitsOnS32(long value)
+        {
+            return value == (int)value;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/CodeGenContext.cs

@@ -0,0 +1,105 @@
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using Ryujinx.Common.Memory;
+using System.IO;
+using System.Numerics;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    class CodeGenContext
+    {
+        private readonly Stream _stream;
+        private readonly Operand[] _blockLabels;
+
+        public int StreamOffset => (int)_stream.Length;
+
+        public AllocationResult AllocResult { get; }
+
+        public Assembler Assembler { get; }
+        public BasicBlock CurrBlock { get; private set; }
+
+        public int CallArgsRegionSize { get; }
+        public int XmmSaveRegionSize { get; }
+
+        public CodeGenContext(AllocationResult allocResult, int maxCallArgs, int blocksCount, bool relocatable)
+        {
+            _stream = MemoryStreamManager.Shared.GetStream();
+            _blockLabels = new Operand[blocksCount];
+
+            AllocResult = allocResult;
+            Assembler = new Assembler(_stream, relocatable);
+
+            CallArgsRegionSize = GetCallArgsRegionSize(allocResult, maxCallArgs, out int xmmSaveRegionSize);
+            XmmSaveRegionSize  = xmmSaveRegionSize;
+        }
+
+        private static int GetCallArgsRegionSize(AllocationResult allocResult, int maxCallArgs, out int xmmSaveRegionSize)
+        {
+            // We need to add 8 bytes to the total size, as the call to this function already pushed 8 bytes (the
+            // return address).
+            int intMask = CallingConvention.GetIntCalleeSavedRegisters() & allocResult.IntUsedRegisters;
+            int vecMask = CallingConvention.GetVecCalleeSavedRegisters() & allocResult.VecUsedRegisters;
+
+            xmmSaveRegionSize = BitOperations.PopCount((uint)vecMask) * 16;
+
+            int calleeSaveRegionSize = BitOperations.PopCount((uint)intMask) * 8 + xmmSaveRegionSize + 8;
+
+            int argsCount = maxCallArgs;
+
+            if (argsCount < 0)
+            {
+                // When the function has no calls, argsCount is -1. In this case, we don't need to allocate the shadow
+                // space.
+                argsCount = 0;
+            }
+            else if (argsCount < 4)
+            {
+                // The ABI mandates that the space for at least 4 arguments is reserved on the stack (this is called
+                // shadow space).
+                argsCount = 4;
+            }
+
+            // TODO: Align XMM save region to 16 bytes because unwinding on Windows requires it.
+            int frameSize = calleeSaveRegionSize + allocResult.SpillRegionSize;
+
+            // TODO: Instead of always multiplying by 16 (the largest possible size of a variable, since a V128 has 16
+            // bytes), we should calculate the exact size consumed by the arguments passed to the called functions on
+            // the stack.
+            int callArgsAndFrameSize = frameSize + argsCount * 16;
+
+            // Ensure that the Stack Pointer will be aligned to 16 bytes.
+            callArgsAndFrameSize = (callArgsAndFrameSize + 0xf) & ~0xf;
+
+            return callArgsAndFrameSize - frameSize;
+        }
+
+        public void EnterBlock(BasicBlock block)
+        {
+            Assembler.MarkLabel(GetLabel(block));
+
+            CurrBlock = block;
+        }
+
+        public void JumpTo(BasicBlock target)
+        {
+            Assembler.Jmp(GetLabel(target));
+        }
+
+        public void JumpTo(X86Condition condition, BasicBlock target)
+        {
+            Assembler.Jcc(condition, GetLabel(target));
+        }
+
+        private Operand GetLabel(BasicBlock block)
+        {
+            ref Operand label = ref _blockLabels[block.Index];
+
+            if (label == default)
+            {
+                label = Operand.Factory.Label();
+            }
+
+            return label;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/HardwareCapabilities.cs

@@ -0,0 +1,144 @@
+using Ryujinx.Memory;
+using System;
+using System.Runtime.InteropServices;
+using System.Runtime.Intrinsics.X86;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    static class HardwareCapabilities
+    {
+        private delegate uint GetXcr0();
+
+        static HardwareCapabilities()
+        {
+            if (!X86Base.IsSupported)
+            {
+                return;
+            }
+
+            (int maxNum, _, _, _) = X86Base.CpuId(0x00000000, 0x00000000);
+
+            (_, _, int ecx1, int edx1) = X86Base.CpuId(0x00000001, 0x00000000);
+            FeatureInfo1Edx = (FeatureFlags1Edx)edx1;
+            FeatureInfo1Ecx = (FeatureFlags1Ecx)ecx1;
+
+            if (maxNum >= 7)
+            {
+                (_, int ebx7, int ecx7, _) = X86Base.CpuId(0x00000007, 0x00000000);
+                FeatureInfo7Ebx = (FeatureFlags7Ebx)ebx7;
+                FeatureInfo7Ecx = (FeatureFlags7Ecx)ecx7;
+            }
+
+            Xcr0InfoEax = (Xcr0FlagsEax)GetXcr0Eax();
+        }
+
+        private static uint GetXcr0Eax()
+        {
+            if (!FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Xsave))
+            {
+                // XSAVE feature required for xgetbv
+                return 0;
+            }
+
+            ReadOnlySpan<byte> asmGetXcr0 = new byte[]
+            {
+                0x31, 0xc9, // xor ecx, ecx
+                0xf, 0x01, 0xd0, // xgetbv
+                0xc3, // ret
+            };
+
+            using MemoryBlock memGetXcr0 = new MemoryBlock((ulong)asmGetXcr0.Length);
+
+            memGetXcr0.Write(0, asmGetXcr0);
+
+            memGetXcr0.Reprotect(0, (ulong)asmGetXcr0.Length, MemoryPermission.ReadAndExecute);
+
+            var fGetXcr0 = Marshal.GetDelegateForFunctionPointer<GetXcr0>(memGetXcr0.Pointer);
+
+            return fGetXcr0();
+        }
+
+        [Flags]
+        public enum FeatureFlags1Edx
+        {
+            Sse = 1 << 25,
+            Sse2 = 1 << 26
+        }
+
+        [Flags]
+        public enum FeatureFlags1Ecx
+        {
+            Sse3 = 1 << 0,
+            Pclmulqdq = 1 << 1,
+            Ssse3 = 1 << 9,
+            Fma = 1 << 12,
+            Sse41 = 1 << 19,
+            Sse42 = 1 << 20,
+            Popcnt = 1 << 23,
+            Aes = 1 << 25,
+            Xsave = 1 << 26,
+            Osxsave = 1 << 27,
+            Avx = 1 << 28,
+            F16c = 1 << 29
+        }
+
+        [Flags]
+        public enum FeatureFlags7Ebx
+        {
+            Avx2 = 1 << 5,
+            Avx512f = 1 << 16,
+            Avx512dq = 1 << 17,
+            Sha = 1 << 29,
+            Avx512bw = 1 << 30,
+            Avx512vl = 1 << 31
+        }
+
+        [Flags]
+        public enum FeatureFlags7Ecx
+        {
+            Gfni = 1 << 8,
+        }
+
+        [Flags]
+        public enum Xcr0FlagsEax
+        {
+            Sse = 1 << 1,
+            YmmHi128 = 1 << 2,
+            Opmask = 1 << 5,
+            ZmmHi256 = 1 << 6,
+            Hi16Zmm = 1 << 7
+        }
+
+        public static FeatureFlags1Edx FeatureInfo1Edx { get; }
+        public static FeatureFlags1Ecx FeatureInfo1Ecx { get; }
+        public static FeatureFlags7Ebx FeatureInfo7Ebx { get; } = 0;
+        public static FeatureFlags7Ecx FeatureInfo7Ecx { get; } = 0;
+        public static Xcr0FlagsEax Xcr0InfoEax { get; } = 0;
+
+        public static bool SupportsSse => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse);
+        public static bool SupportsSse2 => FeatureInfo1Edx.HasFlag(FeatureFlags1Edx.Sse2);
+        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 | FeatureFlags1Ecx.Xsave | FeatureFlags1Ecx.Osxsave) && Xcr0InfoEax.HasFlag(Xcr0FlagsEax.Sse | Xcr0FlagsEax.YmmHi128);
+        public static bool SupportsAvx2 => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx2) && SupportsAvx;
+        public static bool SupportsAvx512F => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512f) && FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.Xsave | FeatureFlags1Ecx.Osxsave)
+            && Xcr0InfoEax.HasFlag(Xcr0FlagsEax.Sse | Xcr0FlagsEax.YmmHi128 | Xcr0FlagsEax.Opmask | Xcr0FlagsEax.ZmmHi256 | Xcr0FlagsEax.Hi16Zmm);
+        public static bool SupportsAvx512Vl => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512vl) && SupportsAvx512F;
+        public static bool SupportsAvx512Bw => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512bw) && SupportsAvx512F;
+        public static bool SupportsAvx512Dq => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Avx512dq) && SupportsAvx512F;
+        public static bool SupportsF16c => FeatureInfo1Ecx.HasFlag(FeatureFlags1Ecx.F16c);
+        public static bool SupportsSha => FeatureInfo7Ebx.HasFlag(FeatureFlags7Ebx.Sha);
+        public static bool SupportsGfni => FeatureInfo7Ecx.HasFlag(FeatureFlags7Ecx.Gfni);
+
+        public static bool ForceLegacySse { get; set; }
+
+        public static bool SupportsVexEncoding => SupportsAvx && !ForceLegacySse;
+        public static bool SupportsEvexEncoding => SupportsAvx512F && !ForceLegacySse;
+    }
+}

src/ARMeilleure/CodeGen/X86/IntrinsicInfo.cs

@@ -0,0 +1,14 @@
+namespace ARMeilleure.CodeGen.X86
+{
+    readonly struct IntrinsicInfo
+    {
+        public X86Instruction Inst { get; }
+        public IntrinsicType  Type { get; }
+
+        public IntrinsicInfo(X86Instruction inst, IntrinsicType type)
+        {
+            Inst = inst;
+            Type = type;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/IntrinsicTable.cs

@@ -0,0 +1,200 @@
+using ARMeilleure.Common;
+using ARMeilleure.IntermediateRepresentation;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    static class IntrinsicTable
+    {
+        private static IntrinsicInfo[] _intrinTable;
+
+        static IntrinsicTable()
+        {
+            _intrinTable = new IntrinsicInfo[EnumUtils.GetCount(typeof(Intrinsic))];
+
+            Add(Intrinsic.X86Addpd,         new IntrinsicInfo(X86Instruction.Addpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Addps,         new IntrinsicInfo(X86Instruction.Addps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Addsd,         new IntrinsicInfo(X86Instruction.Addsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Addss,         new IntrinsicInfo(X86Instruction.Addss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Aesdec,        new IntrinsicInfo(X86Instruction.Aesdec,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Aesdeclast,    new IntrinsicInfo(X86Instruction.Aesdeclast,    IntrinsicType.Binary));
+            Add(Intrinsic.X86Aesenc,        new IntrinsicInfo(X86Instruction.Aesenc,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Aesenclast,    new IntrinsicInfo(X86Instruction.Aesenclast,    IntrinsicType.Binary));
+            Add(Intrinsic.X86Aesimc,        new IntrinsicInfo(X86Instruction.Aesimc,        IntrinsicType.Unary));
+            Add(Intrinsic.X86Andnpd,        new IntrinsicInfo(X86Instruction.Andnpd,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Andnps,        new IntrinsicInfo(X86Instruction.Andnps,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Andpd,         new IntrinsicInfo(X86Instruction.Andpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Andps,         new IntrinsicInfo(X86Instruction.Andps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Blendvpd,      new IntrinsicInfo(X86Instruction.Blendvpd,      IntrinsicType.Ternary));
+            Add(Intrinsic.X86Blendvps,      new IntrinsicInfo(X86Instruction.Blendvps,      IntrinsicType.Ternary));
+            Add(Intrinsic.X86Cmppd,         new IntrinsicInfo(X86Instruction.Cmppd,         IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Cmpps,         new IntrinsicInfo(X86Instruction.Cmpps,         IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Cmpsd,         new IntrinsicInfo(X86Instruction.Cmpsd,         IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Cmpss,         new IntrinsicInfo(X86Instruction.Cmpss,         IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Comisdeq,      new IntrinsicInfo(X86Instruction.Comisd,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Comisdge,      new IntrinsicInfo(X86Instruction.Comisd,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Comisdlt,      new IntrinsicInfo(X86Instruction.Comisd,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Comisseq,      new IntrinsicInfo(X86Instruction.Comiss,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Comissge,      new IntrinsicInfo(X86Instruction.Comiss,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Comisslt,      new IntrinsicInfo(X86Instruction.Comiss,        IntrinsicType.Comis_));
+            Add(Intrinsic.X86Crc32,         new IntrinsicInfo(X86Instruction.Crc32,         IntrinsicType.Crc32));
+            Add(Intrinsic.X86Crc32_16,      new IntrinsicInfo(X86Instruction.Crc32_16,      IntrinsicType.Crc32));
+            Add(Intrinsic.X86Crc32_8,       new IntrinsicInfo(X86Instruction.Crc32_8,       IntrinsicType.Crc32));
+            Add(Intrinsic.X86Cvtdq2pd,      new IntrinsicInfo(X86Instruction.Cvtdq2pd,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtdq2ps,      new IntrinsicInfo(X86Instruction.Cvtdq2ps,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtpd2dq,      new IntrinsicInfo(X86Instruction.Cvtpd2dq,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtpd2ps,      new IntrinsicInfo(X86Instruction.Cvtpd2ps,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtps2dq,      new IntrinsicInfo(X86Instruction.Cvtps2dq,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtps2pd,      new IntrinsicInfo(X86Instruction.Cvtps2pd,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Cvtsd2si,      new IntrinsicInfo(X86Instruction.Cvtsd2si,      IntrinsicType.UnaryToGpr));
+            Add(Intrinsic.X86Cvtsd2ss,      new IntrinsicInfo(X86Instruction.Cvtsd2ss,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Cvtsi2sd,      new IntrinsicInfo(X86Instruction.Cvtsi2sd,      IntrinsicType.BinaryGpr));
+            Add(Intrinsic.X86Cvtsi2si,      new IntrinsicInfo(X86Instruction.Movd,          IntrinsicType.UnaryToGpr));
+            Add(Intrinsic.X86Cvtsi2ss,      new IntrinsicInfo(X86Instruction.Cvtsi2ss,      IntrinsicType.BinaryGpr));
+            Add(Intrinsic.X86Cvtss2sd,      new IntrinsicInfo(X86Instruction.Cvtss2sd,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Cvtss2si,      new IntrinsicInfo(X86Instruction.Cvtss2si,      IntrinsicType.UnaryToGpr));
+            Add(Intrinsic.X86Divpd,         new IntrinsicInfo(X86Instruction.Divpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Divps,         new IntrinsicInfo(X86Instruction.Divps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Divsd,         new IntrinsicInfo(X86Instruction.Divsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Divss,         new IntrinsicInfo(X86Instruction.Divss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Gf2p8affineqb, new IntrinsicInfo(X86Instruction.Gf2p8affineqb, IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Haddpd,        new IntrinsicInfo(X86Instruction.Haddpd,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Haddps,        new IntrinsicInfo(X86Instruction.Haddps,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Insertps,      new IntrinsicInfo(X86Instruction.Insertps,      IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Ldmxcsr,       new IntrinsicInfo(X86Instruction.None,          IntrinsicType.Mxcsr));
+            Add(Intrinsic.X86Maxpd,         new IntrinsicInfo(X86Instruction.Maxpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Maxps,         new IntrinsicInfo(X86Instruction.Maxps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Maxsd,         new IntrinsicInfo(X86Instruction.Maxsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Maxss,         new IntrinsicInfo(X86Instruction.Maxss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Minpd,         new IntrinsicInfo(X86Instruction.Minpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Minps,         new IntrinsicInfo(X86Instruction.Minps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Minsd,         new IntrinsicInfo(X86Instruction.Minsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Minss,         new IntrinsicInfo(X86Instruction.Minss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Movhlps,       new IntrinsicInfo(X86Instruction.Movhlps,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Movlhps,       new IntrinsicInfo(X86Instruction.Movlhps,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Movss,         new IntrinsicInfo(X86Instruction.Movss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Mulpd,         new IntrinsicInfo(X86Instruction.Mulpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Mulps,         new IntrinsicInfo(X86Instruction.Mulps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Mulsd,         new IntrinsicInfo(X86Instruction.Mulsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Mulss,         new IntrinsicInfo(X86Instruction.Mulss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Paddb,         new IntrinsicInfo(X86Instruction.Paddb,         IntrinsicType.Binary));
+            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));
+            Add(Intrinsic.X86Pavgw,         new IntrinsicInfo(X86Instruction.Pavgw,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Pblendvb,      new IntrinsicInfo(X86Instruction.Pblendvb,      IntrinsicType.Ternary));
+            Add(Intrinsic.X86Pclmulqdq,     new IntrinsicInfo(X86Instruction.Pclmulqdq,     IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Pcmpeqb,       new IntrinsicInfo(X86Instruction.Pcmpeqb,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpeqd,       new IntrinsicInfo(X86Instruction.Pcmpeqd,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpeqq,       new IntrinsicInfo(X86Instruction.Pcmpeqq,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpeqw,       new IntrinsicInfo(X86Instruction.Pcmpeqw,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpgtb,       new IntrinsicInfo(X86Instruction.Pcmpgtb,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpgtd,       new IntrinsicInfo(X86Instruction.Pcmpgtd,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpgtq,       new IntrinsicInfo(X86Instruction.Pcmpgtq,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pcmpgtw,       new IntrinsicInfo(X86Instruction.Pcmpgtw,       IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxsb,        new IntrinsicInfo(X86Instruction.Pmaxsb,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxsd,        new IntrinsicInfo(X86Instruction.Pmaxsd,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxsw,        new IntrinsicInfo(X86Instruction.Pmaxsw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxub,        new IntrinsicInfo(X86Instruction.Pmaxub,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxud,        new IntrinsicInfo(X86Instruction.Pmaxud,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmaxuw,        new IntrinsicInfo(X86Instruction.Pmaxuw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminsb,        new IntrinsicInfo(X86Instruction.Pminsb,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminsd,        new IntrinsicInfo(X86Instruction.Pminsd,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminsw,        new IntrinsicInfo(X86Instruction.Pminsw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminub,        new IntrinsicInfo(X86Instruction.Pminub,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminud,        new IntrinsicInfo(X86Instruction.Pminud,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pminuw,        new IntrinsicInfo(X86Instruction.Pminuw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmovsxbw,      new IntrinsicInfo(X86Instruction.Pmovsxbw,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmovsxdq,      new IntrinsicInfo(X86Instruction.Pmovsxdq,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmovsxwd,      new IntrinsicInfo(X86Instruction.Pmovsxwd,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmovzxbw,      new IntrinsicInfo(X86Instruction.Pmovzxbw,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmovzxdq,      new IntrinsicInfo(X86Instruction.Pmovzxdq,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmovzxwd,      new IntrinsicInfo(X86Instruction.Pmovzxwd,      IntrinsicType.Unary));
+            Add(Intrinsic.X86Pmulld,        new IntrinsicInfo(X86Instruction.Pmulld,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pmullw,        new IntrinsicInfo(X86Instruction.Pmullw,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Popcnt,        new IntrinsicInfo(X86Instruction.Popcnt,        IntrinsicType.PopCount));
+            Add(Intrinsic.X86Por,           new IntrinsicInfo(X86Instruction.Por,           IntrinsicType.Binary));
+            Add(Intrinsic.X86Pshufb,        new IntrinsicInfo(X86Instruction.Pshufb,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Pshufd,        new IntrinsicInfo(X86Instruction.Pshufd,        IntrinsicType.BinaryImm));
+            Add(Intrinsic.X86Pslld,         new IntrinsicInfo(X86Instruction.Pslld,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Pslldq,        new IntrinsicInfo(X86Instruction.Pslldq,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Psllq,         new IntrinsicInfo(X86Instruction.Psllq,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psllw,         new IntrinsicInfo(X86Instruction.Psllw,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psrad,         new IntrinsicInfo(X86Instruction.Psrad,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psraw,         new IntrinsicInfo(X86Instruction.Psraw,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psrld,         new IntrinsicInfo(X86Instruction.Psrld,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psrlq,         new IntrinsicInfo(X86Instruction.Psrlq,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psrldq,        new IntrinsicInfo(X86Instruction.Psrldq,        IntrinsicType.Binary));
+            Add(Intrinsic.X86Psrlw,         new IntrinsicInfo(X86Instruction.Psrlw,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psubb,         new IntrinsicInfo(X86Instruction.Psubb,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psubd,         new IntrinsicInfo(X86Instruction.Psubd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psubq,         new IntrinsicInfo(X86Instruction.Psubq,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Psubw,         new IntrinsicInfo(X86Instruction.Psubw,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpckhbw,     new IntrinsicInfo(X86Instruction.Punpckhbw,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpckhdq,     new IntrinsicInfo(X86Instruction.Punpckhdq,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpckhqdq,    new IntrinsicInfo(X86Instruction.Punpckhqdq,    IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpckhwd,     new IntrinsicInfo(X86Instruction.Punpckhwd,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpcklbw,     new IntrinsicInfo(X86Instruction.Punpcklbw,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpckldq,     new IntrinsicInfo(X86Instruction.Punpckldq,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpcklqdq,    new IntrinsicInfo(X86Instruction.Punpcklqdq,    IntrinsicType.Binary));
+            Add(Intrinsic.X86Punpcklwd,     new IntrinsicInfo(X86Instruction.Punpcklwd,     IntrinsicType.Binary));
+            Add(Intrinsic.X86Pxor,          new IntrinsicInfo(X86Instruction.Pxor,          IntrinsicType.Binary));
+            Add(Intrinsic.X86Rcpps,         new IntrinsicInfo(X86Instruction.Rcpps,         IntrinsicType.Unary));
+            Add(Intrinsic.X86Rcpss,         new IntrinsicInfo(X86Instruction.Rcpss,         IntrinsicType.Unary));
+            Add(Intrinsic.X86Roundpd,       new IntrinsicInfo(X86Instruction.Roundpd,       IntrinsicType.BinaryImm));
+            Add(Intrinsic.X86Roundps,       new IntrinsicInfo(X86Instruction.Roundps,       IntrinsicType.BinaryImm));
+            Add(Intrinsic.X86Roundsd,       new IntrinsicInfo(X86Instruction.Roundsd,       IntrinsicType.BinaryImm));
+            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));
+            Add(Intrinsic.X86Sqrtps,        new IntrinsicInfo(X86Instruction.Sqrtps,        IntrinsicType.Unary));
+            Add(Intrinsic.X86Sqrtsd,        new IntrinsicInfo(X86Instruction.Sqrtsd,        IntrinsicType.Unary));
+            Add(Intrinsic.X86Sqrtss,        new IntrinsicInfo(X86Instruction.Sqrtss,        IntrinsicType.Unary));
+            Add(Intrinsic.X86Stmxcsr,       new IntrinsicInfo(X86Instruction.None,          IntrinsicType.Mxcsr));
+            Add(Intrinsic.X86Subpd,         new IntrinsicInfo(X86Instruction.Subpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Subps,         new IntrinsicInfo(X86Instruction.Subps,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Subsd,         new IntrinsicInfo(X86Instruction.Subsd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Subss,         new IntrinsicInfo(X86Instruction.Subss,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Unpckhpd,      new IntrinsicInfo(X86Instruction.Unpckhpd,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Unpckhps,      new IntrinsicInfo(X86Instruction.Unpckhps,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Unpcklpd,      new IntrinsicInfo(X86Instruction.Unpcklpd,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Unpcklps,      new IntrinsicInfo(X86Instruction.Unpcklps,      IntrinsicType.Binary));
+            Add(Intrinsic.X86Vcvtph2ps,     new IntrinsicInfo(X86Instruction.Vcvtph2ps,     IntrinsicType.Unary));
+            Add(Intrinsic.X86Vcvtps2ph,     new IntrinsicInfo(X86Instruction.Vcvtps2ph,     IntrinsicType.BinaryImm));
+            Add(Intrinsic.X86Vfmadd231pd,   new IntrinsicInfo(X86Instruction.Vfmadd231pd,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfmadd231ps,   new IntrinsicInfo(X86Instruction.Vfmadd231ps,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfmadd231sd,   new IntrinsicInfo(X86Instruction.Vfmadd231sd,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfmadd231ss,   new IntrinsicInfo(X86Instruction.Vfmadd231ss,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfmsub231sd,   new IntrinsicInfo(X86Instruction.Vfmsub231sd,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfmsub231ss,   new IntrinsicInfo(X86Instruction.Vfmsub231ss,   IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmadd231pd,  new IntrinsicInfo(X86Instruction.Vfnmadd231pd,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmadd231ps,  new IntrinsicInfo(X86Instruction.Vfnmadd231ps,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmadd231sd,  new IntrinsicInfo(X86Instruction.Vfnmadd231sd,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmadd231ss,  new IntrinsicInfo(X86Instruction.Vfnmadd231ss,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmsub231sd,  new IntrinsicInfo(X86Instruction.Vfnmsub231sd,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vfnmsub231ss,  new IntrinsicInfo(X86Instruction.Vfnmsub231ss,  IntrinsicType.Fma));
+            Add(Intrinsic.X86Vpternlogd,    new IntrinsicInfo(X86Instruction.Vpternlogd,    IntrinsicType.TernaryImm));
+            Add(Intrinsic.X86Xorpd,         new IntrinsicInfo(X86Instruction.Xorpd,         IntrinsicType.Binary));
+            Add(Intrinsic.X86Xorps,         new IntrinsicInfo(X86Instruction.Xorps,         IntrinsicType.Binary));
+        }
+
+        private static void Add(Intrinsic intrin, IntrinsicInfo info)
+        {
+            _intrinTable[(int)intrin] = info;
+        }
+
+        public static IntrinsicInfo GetInfo(Intrinsic intrin)
+        {
+            return _intrinTable[(int)intrin];
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/IntrinsicType.cs

@@ -0,0 +1,18 @@
+namespace ARMeilleure.CodeGen.X86
+{
+    enum IntrinsicType
+    {
+        Comis_,
+        Mxcsr,
+        PopCount,
+        Unary,
+        UnaryToGpr,
+        Binary,
+        BinaryGpr,
+        BinaryImm,
+        Crc32,
+        Ternary,
+        TernaryImm,
+        Fma
+    }
+}

src/ARMeilleure/CodeGen/X86/Mxcsr.cs

@@ -0,0 +1,15 @@
+using System;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    [Flags]
+    enum Mxcsr
+    {
+        Ftz = 1 << 15, // Flush To Zero.
+        Rhi = 1 << 14, // Round Mode high bit.
+        Rlo = 1 << 13, // Round Mode low bit.
+        Um = 1 << 11,  // Underflow Mask.
+        Dm = 1 << 8,   // Denormal Mask.
+        Daz = 1 << 6   // Denormals Are Zero.
+    }
+}

src/ARMeilleure/CodeGen/X86/PreAllocator.cs

@@ -0,0 +1,796 @@
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    class PreAllocator
+    {
+        public static void RunPass(CompilerContext cctx, StackAllocator stackAlloc, out int maxCallArgs)
+        {
+            maxCallArgs = -1;
+
+            Span<Operation> buffer = default;
+
+            CallConvName callConv = CallingConvention.GetCurrentCallConv();
+
+            Operand[] preservedArgs = new Operand[CallingConvention.GetArgumentsOnRegsCount()];
+
+            for (BasicBlock block = cctx.Cfg.Blocks.First; block != null; block = block.ListNext)
+            {
+                Operation nextNode;
+
+                for (Operation node = block.Operations.First; node != default; node = nextNode)
+                {
+                    nextNode = node.ListNext;
+
+                    if (node.Instruction == Instruction.Phi)
+                    {
+                        continue;
+                    }
+
+                    InsertConstantRegCopies(block.Operations, node);
+                    InsertDestructiveRegCopies(block.Operations, node);
+                    InsertConstrainedRegCopies(block.Operations, node);
+
+                    switch (node.Instruction)
+                    {
+                        case Instruction.Call:
+                            // Get the maximum number of arguments used on a call.
+                            // On windows, when a struct is returned from the call,
+                            // we also need to pass the pointer where the struct
+                            // should be written on the first argument.
+                            int argsCount = node.SourcesCount - 1;
+
+                            if (node.Destination != default && node.Destination.Type == OperandType.V128)
+                            {
+                                argsCount++;
+                            }
+
+                            if (maxCallArgs < argsCount)
+                            {
+                                maxCallArgs = argsCount;
+                            }
+
+                            // Copy values to registers expected by the function
+                            // being called, as mandated by the ABI.
+                            if (callConv == CallConvName.Windows)
+                            {
+                                PreAllocatorWindows.InsertCallCopies(block.Operations, stackAlloc, node);
+                            }
+                            else /* if (callConv == CallConvName.SystemV) */
+                            {
+                                PreAllocatorSystemV.InsertCallCopies(block.Operations, node);
+                            }
+                            break;
+
+                        case Instruction.ConvertToFPUI:
+                            GenerateConvertToFPUI(block.Operations, node);
+                            break;
+
+                        case Instruction.LoadArgument:
+                            if (callConv == CallConvName.Windows)
+                            {
+                                nextNode = PreAllocatorWindows.InsertLoadArgumentCopy(cctx, ref buffer, block.Operations, preservedArgs, node);
+                            }
+                            else /* if (callConv == CallConvName.SystemV) */
+                            {
+                                nextNode = PreAllocatorSystemV.InsertLoadArgumentCopy(cctx, ref buffer, block.Operations, preservedArgs, node);
+                            }
+                            break;
+
+                        case Instruction.Negate:
+                            if (!node.GetSource(0).Type.IsInteger())
+                            {
+                                GenerateNegate(block.Operations, node);
+                            }
+                            break;
+
+                        case Instruction.Return:
+                            if (callConv == CallConvName.Windows)
+                            {
+                                PreAllocatorWindows.InsertReturnCopy(cctx, block.Operations, preservedArgs, node);
+                            }
+                            else /* if (callConv == CallConvName.SystemV) */
+                            {
+                                PreAllocatorSystemV.InsertReturnCopy(block.Operations, node);
+                            }
+                            break;
+
+                        case Instruction.Tailcall:
+                            if (callConv == CallConvName.Windows)
+                            {
+                                PreAllocatorWindows.InsertTailcallCopies(block.Operations, stackAlloc, node);
+                            }
+                            else
+                            {
+                                PreAllocatorSystemV.InsertTailcallCopies(block.Operations, stackAlloc, node);
+                            }
+                            break;
+
+                        case Instruction.VectorInsert8:
+                            if (!HardwareCapabilities.SupportsSse41)
+                            {
+                                GenerateVectorInsert8(block.Operations, node);
+                            }
+                            break;
+
+                        case Instruction.Extended:
+                            if (node.Intrinsic == Intrinsic.X86Ldmxcsr)
+                            {
+                                int stackOffset = stackAlloc.Allocate(OperandType.I32);
+
+                                node.SetSources(new Operand[] { Const(stackOffset), node.GetSource(0) });
+                            }
+                            else if (node.Intrinsic == Intrinsic.X86Stmxcsr)
+                            {
+                                int stackOffset = stackAlloc.Allocate(OperandType.I32);
+
+                                node.SetSources(new Operand[] { Const(stackOffset) });
+                            }
+                            break;
+                    }
+                }
+            }
+        }
+
+        protected static void InsertConstantRegCopies(IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.SourcesCount == 0 || IsXmmIntrinsic(node))
+            {
+                return;
+            }
+
+            Instruction inst = node.Instruction;
+
+            Operand src1 = node.GetSource(0);
+            Operand src2;
+
+            if (src1.Kind == OperandKind.Constant)
+            {
+                if (!src1.Type.IsInteger())
+                {
+                    // Handle non-integer types (FP32, FP64 and V128).
+                    // For instructions without an immediate operand, we do the following:
+                    // - Insert a copy with the constant value (as integer) to a GPR.
+                    // - Insert a copy from the GPR to a XMM register.
+                    // - Replace the constant use with the XMM register.
+                    src1 = AddXmmCopy(nodes, node, src1);
+
+                    node.SetSource(0, src1);
+                }
+                else if (!HasConstSrc1(inst))
+                {
+                    // Handle integer types.
+                    // Most ALU instructions accepts a 32-bits immediate on the second operand.
+                    // We need to ensure the following:
+                    // - If the constant is on operand 1, we need to move it.
+                    // -- But first, we try to swap operand 1 and 2 if the instruction is commutative.
+                    // -- Doing so may allow us to encode the constant as operand 2 and avoid a copy.
+                    // - If the constant is on operand 2, we check if the instruction supports it,
+                    // if not, we also add a copy. 64-bits constants are usually not supported.
+                    if (IsCommutative(node))
+                    {
+                        src2 = node.GetSource(1);
+
+                        Operand temp = src1;
+
+                        src1 = src2;
+                        src2 = temp;
+
+                        node.SetSource(0, src1);
+                        node.SetSource(1, src2);
+                    }
+
+                    if (src1.Kind == OperandKind.Constant)
+                    {
+                        src1 = AddCopy(nodes, node, src1);
+
+                        node.SetSource(0, src1);
+                    }
+                }
+            }
+
+            if (node.SourcesCount < 2)
+            {
+                return;
+            }
+
+            src2 = node.GetSource(1);
+
+            if (src2.Kind == OperandKind.Constant)
+            {
+                if (!src2.Type.IsInteger())
+                {
+                    src2 = AddXmmCopy(nodes, node, src2);
+
+                    node.SetSource(1, src2);
+                }
+                else if (!HasConstSrc2(inst) || CodeGenCommon.IsLongConst(src2))
+                {
+                    src2 = AddCopy(nodes, node, src2);
+
+                    node.SetSource(1, src2);
+                }
+            }
+        }
+
+        protected static void InsertConstrainedRegCopies(IntrusiveList<Operation> nodes, Operation node)
+        {
+            Operand dest = node.Destination;
+
+            switch (node.Instruction)
+            {
+                case Instruction.CompareAndSwap:
+                case Instruction.CompareAndSwap16:
+                case Instruction.CompareAndSwap8:
+                {
+                    OperandType type = node.GetSource(1).Type;
+
+                    if (type == OperandType.V128)
+                    {
+                        // Handle the many restrictions of the compare and exchange (16 bytes) instruction:
+                        // - The expected value should be in RDX:RAX.
+                        // - The new value to be written should be in RCX:RBX.
+                        // - The value at the memory location is loaded to RDX:RAX.
+                        void SplitOperand(Operand source, Operand lr, Operand hr)
+                        {
+                            nodes.AddBefore(node, Operation(Instruction.VectorExtract, lr, source, Const(0)));
+                            nodes.AddBefore(node, Operation(Instruction.VectorExtract, hr, source, Const(1)));
+                        }
+
+                        Operand rax = Gpr(X86Register.Rax, OperandType.I64);
+                        Operand rbx = Gpr(X86Register.Rbx, OperandType.I64);
+                        Operand rcx = Gpr(X86Register.Rcx, OperandType.I64);
+                        Operand rdx = Gpr(X86Register.Rdx, OperandType.I64);
+
+                        SplitOperand(node.GetSource(1), rax, rdx);
+                        SplitOperand(node.GetSource(2), rbx, rcx);
+
+                        Operation operation = node;
+
+                        node = nodes.AddAfter(node, Operation(Instruction.VectorCreateScalar, dest, rax));
+                        nodes.AddAfter(node, Operation(Instruction.VectorInsert, dest, dest, rdx, Const(1)));
+
+                        operation.SetDestinations(new Operand[] { rdx, rax });
+                        operation.SetSources(new Operand[] { operation.GetSource(0), rdx, rax, rcx, rbx });
+                    }
+                    else
+                    {
+                        // Handle the many restrictions of the compare and exchange (32/64) instruction:
+                        // - The expected value should be in (E/R)AX.
+                        // - The value at the memory location is loaded to (E/R)AX.
+                        Operand expected = node.GetSource(1);
+                        Operand newValue = node.GetSource(2);
+
+                        Operand rax = Gpr(X86Register.Rax, expected.Type);
+
+                        nodes.AddBefore(node, Operation(Instruction.Copy, rax, expected));
+
+                        // We need to store the new value into a temp, since it may
+                        // be a constant, and this instruction does not support immediate operands.
+                        Operand temp = Local(newValue.Type);
+
+                        nodes.AddBefore(node, Operation(Instruction.Copy, temp, newValue));
+
+                        node.SetSources(new Operand[] { node.GetSource(0), rax, temp });
+
+                        nodes.AddAfter(node, Operation(Instruction.Copy, dest, rax));
+
+                        node.Destination = rax;
+                    }
+
+                    break;
+                }
+
+                case Instruction.Divide:
+                case Instruction.DivideUI:
+                {
+                    // Handle the many restrictions of the division instructions:
+                    // - The dividend is always in RDX:RAX.
+                    // - The result is always in RAX.
+                    // - Additionally it also writes the remainder in RDX.
+                    if (dest.Type.IsInteger())
+                    {
+                        Operand src1 = node.GetSource(0);
+
+                        Operand rax = Gpr(X86Register.Rax, src1.Type);
+                        Operand rdx = Gpr(X86Register.Rdx, src1.Type);
+
+                        nodes.AddBefore(node, Operation(Instruction.Copy,    rax, src1));
+                        nodes.AddBefore(node, Operation(Instruction.Clobber, rdx));
+
+                        nodes.AddAfter(node, Operation(Instruction.Copy, dest, rax));
+
+                        node.SetSources(new Operand[] { rdx, rax, node.GetSource(1) });
+                        node.Destination = rax;
+                    }
+
+                    break;
+                }
+
+                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.
+                    // 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);
+
+                        nodes.AddBefore(node, Operation(Instruction.Copy, xmm0, node.GetSource(2)));
+
+                        node.SetSource(2, xmm0);
+                    }
+
+                    break;
+                }
+
+                case Instruction.Multiply64HighSI:
+                case Instruction.Multiply64HighUI:
+                {
+                    // Handle the many restrictions of the i64 * i64 = i128 multiply instructions:
+                    // - The multiplicand is always in RAX.
+                    // - The lower 64-bits of the result is always in RAX.
+                    // - The higher 64-bits of the result is always in RDX.
+                    Operand src1 = node.GetSource(0);
+
+                    Operand rax = Gpr(X86Register.Rax, src1.Type);
+                    Operand rdx = Gpr(X86Register.Rdx, src1.Type);
+
+                    nodes.AddBefore(node, Operation(Instruction.Copy, rax, src1));
+
+                    node.SetSource(0, rax);
+
+                    nodes.AddAfter(node, Operation(Instruction.Copy, dest, rdx));
+
+                    node.SetDestinations(new Operand[] { rdx, rax });
+
+                    break;
+                }
+
+                case Instruction.RotateRight:
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                {
+                    // The shift register is always implied to be CL (low 8-bits of RCX or ECX).
+                    if (node.GetSource(1).Kind == OperandKind.LocalVariable)
+                    {
+                        Operand rcx = Gpr(X86Register.Rcx, OperandType.I32);
+
+                        nodes.AddBefore(node, Operation(Instruction.Copy, rcx, node.GetSource(1)));
+
+                        node.SetSource(1, rcx);
+                    }
+
+                    break;
+                }
+            }
+        }
+
+        protected static void InsertDestructiveRegCopies(IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.Destination == default || node.SourcesCount == 0)
+            {
+                return;
+            }
+
+            Instruction inst = node.Instruction;
+
+            Operand dest = node.Destination;
+            Operand src1 = node.GetSource(0);
+
+            // The multiply instruction (that maps to IMUL) is somewhat special, it has
+            // a three operand form where the second source is a immediate value.
+            bool threeOperandForm = inst == Instruction.Multiply && node.GetSource(1).Kind == OperandKind.Constant;
+
+            if (IsSameOperandDestSrc1(node) && src1.Kind == OperandKind.LocalVariable && !threeOperandForm)
+            {
+                bool useNewLocal = false;
+
+                for (int srcIndex = 1; srcIndex < node.SourcesCount; srcIndex++)
+                {
+                    if (node.GetSource(srcIndex) == dest)
+                    {
+                        useNewLocal = true;
+
+                        break;
+                    }
+                }
+
+                if (useNewLocal)
+                {
+                    // Dest is being used as some source already, we need to use a new
+                    // local to store the temporary value, otherwise the value on dest
+                    // local would be overwritten.
+                    Operand temp = Local(dest.Type);
+
+                    nodes.AddBefore(node, Operation(Instruction.Copy, temp, src1));
+
+                    node.SetSource(0, temp);
+
+                    nodes.AddAfter(node, Operation(Instruction.Copy, dest, temp));
+
+                    node.Destination = temp;
+                }
+                else
+                {
+                    nodes.AddBefore(node, Operation(Instruction.Copy, dest, src1));
+
+                    node.SetSource(0, dest);
+                }
+            }
+            else if (inst == Instruction.ConditionalSelect)
+            {
+                Operand src2 = node.GetSource(1);
+                Operand src3 = node.GetSource(2);
+
+                if (src1 == dest || src2 == dest)
+                {
+                    Operand temp = Local(dest.Type);
+
+                    nodes.AddBefore(node, Operation(Instruction.Copy, temp, src3));
+
+                    node.SetSource(2, temp);
+
+                    nodes.AddAfter(node, Operation(Instruction.Copy, dest, temp));
+
+                    node.Destination = temp;
+                }
+                else
+                {
+                    nodes.AddBefore(node, Operation(Instruction.Copy, dest, src3));
+
+                    node.SetSource(2, dest);
+                }
+            }
+        }
+
+        private static void GenerateConvertToFPUI(IntrusiveList<Operation> nodes, Operation node)
+        {
+            // Unsigned integer to FP conversions are not supported on X86.
+            // We need to turn them into signed integer to FP conversions, and
+            // adjust the final result.
+            Operand dest   = node.Destination;
+            Operand source = node.GetSource(0);
+
+            Debug.Assert(source.Type.IsInteger(), $"Invalid source type \"{source.Type}\".");
+
+            Operation currentNode = node;
+
+            if (source.Type == OperandType.I32)
+            {
+                // For 32-bits integers, we can just zero-extend to 64-bits,
+                // and then use the 64-bits signed conversion instructions.
+                Operand zex = Local(OperandType.I64);
+
+                node = nodes.AddAfter(node, Operation(Instruction.ZeroExtend32, zex,  source));
+                node = nodes.AddAfter(node, Operation(Instruction.ConvertToFP,  dest, zex));
+            }
+            else /* if (source.Type == OperandType.I64) */
+            {
+                // For 64-bits integers, we need to do the following:
+                // - Ensure that the integer has the most significant bit clear.
+                // -- This can be done by shifting the value right by 1, that is, dividing by 2.
+                // -- The least significant bit is lost in this case though.
+                // - We can then convert the shifted value with a signed integer instruction.
+                // - The result still needs to be corrected after that.
+                // -- First, we need to multiply the result by 2, as we divided it by 2 before.
+                // --- This can be done efficiently by adding the result to itself.
+                // -- Then, we need to add the least significant bit that was shifted out.
+                // --- We can convert the least significant bit to float, and add it to the result.
+                Operand lsb  = Local(OperandType.I64);
+                Operand half = Local(OperandType.I64);
+
+                Operand lsbF = Local(dest.Type);
+
+                node = nodes.AddAfter(node, Operation(Instruction.Copy, lsb,  source));
+                node = nodes.AddAfter(node, Operation(Instruction.Copy, half, source));
+
+                node = nodes.AddAfter(node, Operation(Instruction.BitwiseAnd,   lsb,  lsb,  Const(1L)));
+                node = nodes.AddAfter(node, Operation(Instruction.ShiftRightUI, half, half, Const(1)));
+
+                node = nodes.AddAfter(node, Operation(Instruction.ConvertToFP, lsbF, lsb));
+                node = nodes.AddAfter(node, Operation(Instruction.ConvertToFP, dest, half));
+
+                node = nodes.AddAfter(node, Operation(Instruction.Add, dest, dest, dest));
+                nodes.AddAfter(node, Operation(Instruction.Add, dest, dest, lsbF));
+            }
+
+            Delete(nodes, currentNode);
+        }
+
+        private static void GenerateNegate(IntrusiveList<Operation> nodes, Operation node)
+        {
+            // There's no SSE FP negate instruction, so we need to transform that into
+            // a XOR of the value to be negated with a mask with the highest bit set.
+            // This also produces -0 for a negation of the value 0.
+            Operand dest   = node.Destination;
+            Operand source = node.GetSource(0);
+
+            Debug.Assert(dest.Type == OperandType.FP32 ||
+                         dest.Type == OperandType.FP64, $"Invalid destination type \"{dest.Type}\".");
+
+            Operation currentNode = node;
+
+            Operand res = Local(dest.Type);
+
+            node = nodes.AddAfter(node, Operation(Instruction.VectorOne, res));
+
+            if (dest.Type == OperandType.FP32)
+            {
+                node = nodes.AddAfter(node, Operation(Intrinsic.X86Pslld, res, res, Const(31)));
+            }
+            else /* if (dest.Type == OperandType.FP64) */
+            {
+                node = nodes.AddAfter(node, Operation(Intrinsic.X86Psllq, res, res, Const(63)));
+            }
+
+            node = nodes.AddAfter(node, Operation(Intrinsic.X86Xorps, res, res, source));
+
+            nodes.AddAfter(node, Operation(Instruction.Copy, dest, res));
+
+            Delete(nodes, currentNode);
+        }
+
+        private static void GenerateVectorInsert8(IntrusiveList<Operation> nodes, Operation node)
+        {
+            // Handle vector insertion, when SSE 4.1 is not supported.
+            Operand dest = node.Destination;
+            Operand src1 = node.GetSource(0); // Vector
+            Operand src2 = node.GetSource(1); // Value
+            Operand src3 = node.GetSource(2); // Index
+
+            Debug.Assert(src3.Kind == OperandKind.Constant);
+
+            byte index = src3.AsByte();
+
+            Debug.Assert(index < 16);
+
+            Operation currentNode = node;
+
+            Operand temp1 = Local(OperandType.I32);
+            Operand temp2 = Local(OperandType.I32);
+
+            node = nodes.AddAfter(node, Operation(Instruction.Copy, temp2, src2));
+
+            Operation vextOp = Operation(Instruction.VectorExtract16, temp1, src1, Const(index >> 1));
+
+            node = nodes.AddAfter(node, vextOp);
+
+            if ((index & 1) != 0)
+            {
+                node = nodes.AddAfter(node, Operation(Instruction.ZeroExtend8, temp1, temp1));
+                node = nodes.AddAfter(node, Operation(Instruction.ShiftLeft,   temp2, temp2, Const(8)));
+                node = nodes.AddAfter(node, Operation(Instruction.BitwiseOr,   temp1, temp1, temp2));
+            }
+            else
+            {
+                node = nodes.AddAfter(node, Operation(Instruction.ZeroExtend8, temp2, temp2));
+                node = nodes.AddAfter(node, Operation(Instruction.BitwiseAnd,  temp1, temp1, Const(0xff00)));
+                node = nodes.AddAfter(node, Operation(Instruction.BitwiseOr,   temp1, temp1, temp2));
+            }
+
+            Operation vinsOp = Operation(Instruction.VectorInsert16, dest, src1, temp1, Const(index >> 1));
+
+            nodes.AddAfter(node, vinsOp);
+
+            Delete(nodes, currentNode);
+        }
+
+        protected static Operand AddXmmCopy(IntrusiveList<Operation> nodes, Operation node, Operand source)
+        {
+            Operand temp = Local(source.Type);
+            Operand intConst = AddCopy(nodes, node, GetIntConst(source));
+
+            Operation copyOp = Operation(Instruction.VectorCreateScalar, temp, intConst);
+
+            nodes.AddBefore(node, copyOp);
+
+            return temp;
+        }
+
+        protected static Operand AddCopy(IntrusiveList<Operation> nodes, Operation node, Operand source)
+        {
+            Operand temp = Local(source.Type);
+
+            Operation copyOp = Operation(Instruction.Copy, temp, source);
+
+            nodes.AddBefore(node, copyOp);
+
+            return temp;
+        }
+
+        private static Operand GetIntConst(Operand value)
+        {
+            if (value.Type == OperandType.FP32)
+            {
+                return Const(value.AsInt32());
+            }
+            else if (value.Type == OperandType.FP64)
+            {
+                return Const(value.AsInt64());
+            }
+
+            return value;
+        }
+
+        protected static void Delete(IntrusiveList<Operation> nodes, Operation node)
+        {
+            node.Destination = default;
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                node.SetSource(index, default);
+            }
+
+            nodes.Remove(node);
+        }
+
+        protected static Operand Gpr(X86Register register, OperandType type)
+        {
+            return Register((int)register, RegisterType.Integer, type);
+        }
+
+        protected static Operand Xmm(X86Register register, OperandType type)
+        {
+            return Register((int)register, RegisterType.Vector, type);
+        }
+
+        private static bool IsSameOperandDestSrc1(Operation operation)
+        {
+            switch (operation.Instruction)
+            {
+                case Instruction.Add:
+                    return !HardwareCapabilities.SupportsVexEncoding && !operation.Destination.Type.IsInteger();
+                case Instruction.Multiply:
+                case Instruction.Subtract:
+                    return !HardwareCapabilities.SupportsVexEncoding || operation.Destination.Type.IsInteger();
+
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseNot:
+                case Instruction.BitwiseOr:
+                case Instruction.ByteSwap:
+                case Instruction.Negate:
+                case Instruction.RotateRight:
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                    return true;
+
+                case Instruction.Divide:
+                    return !HardwareCapabilities.SupportsVexEncoding && !operation.Destination.Type.IsInteger();
+
+                case Instruction.VectorInsert:
+                case Instruction.VectorInsert16:
+                case Instruction.VectorInsert8:
+                    return !HardwareCapabilities.SupportsVexEncoding;
+
+                case Instruction.Extended:
+                    return IsIntrinsicSameOperandDestSrc1(operation);
+            }
+
+            return IsVexSameOperandDestSrc1(operation);
+        }
+
+        private static bool IsIntrinsicSameOperandDestSrc1(Operation operation)
+        {
+            IntrinsicInfo info = IntrinsicTable.GetInfo(operation.Intrinsic);
+
+            return info.Type == IntrinsicType.Crc32 || info.Type == IntrinsicType.Fma || IsVexSameOperandDestSrc1(operation);
+        }
+
+        private static bool IsVexSameOperandDestSrc1(Operation operation)
+        {
+            if (IsIntrinsic(operation.Instruction))
+            {
+                IntrinsicInfo info = IntrinsicTable.GetInfo(operation.Intrinsic);
+
+                bool hasVex = HardwareCapabilities.SupportsVexEncoding && Assembler.SupportsVexPrefix(info.Inst);
+
+                bool isUnary = operation.SourcesCount < 2;
+
+                bool hasVecDest = operation.Destination != default && operation.Destination.Type == OperandType.V128;
+
+                return !hasVex && !isUnary && hasVecDest;
+            }
+
+            return false;
+        }
+
+        private static bool HasConstSrc1(Instruction inst)
+        {
+            switch (inst)
+            {
+                case Instruction.Copy:
+                case Instruction.LoadArgument:
+                case Instruction.Spill:
+                case Instruction.SpillArg:
+                    return true;
+            }
+
+            return false;
+        }
+
+        private static bool HasConstSrc2(Instruction inst)
+        {
+            switch (inst)
+            {
+                case Instruction.Add:
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseOr:
+                case Instruction.BranchIf:
+                case Instruction.Compare:
+                case Instruction.Multiply:
+                case Instruction.RotateRight:
+                case Instruction.ShiftLeft:
+                case Instruction.ShiftRightSI:
+                case Instruction.ShiftRightUI:
+                case Instruction.Store:
+                case Instruction.Store16:
+                case Instruction.Store8:
+                case Instruction.Subtract:
+                case Instruction.VectorExtract:
+                case Instruction.VectorExtract16:
+                case Instruction.VectorExtract8:
+                    return true;
+            }
+
+            return false;
+        }
+
+        private static bool IsCommutative(Operation operation)
+        {
+            switch (operation.Instruction)
+            {
+                case Instruction.Add:
+                case Instruction.BitwiseAnd:
+                case Instruction.BitwiseExclusiveOr:
+                case Instruction.BitwiseOr:
+                case Instruction.Multiply:
+                    return true;
+
+                case Instruction.BranchIf:
+                case Instruction.Compare:
+                {
+                    Operand comp = operation.GetSource(2);
+
+                    Debug.Assert(comp.Kind == OperandKind.Constant);
+
+                    var compType = (Comparison)comp.AsInt32();
+
+                    return compType == Comparison.Equal || compType == Comparison.NotEqual;
+                }
+            }
+
+            return false;
+        }
+
+        private static bool IsIntrinsic(Instruction inst)
+        {
+            return inst == Instruction.Extended;
+        }
+
+        private static bool IsXmmIntrinsic(Operation operation)
+        {
+            if (operation.Instruction != Instruction.Extended)
+            {
+                return false;
+            }
+
+            IntrinsicInfo info = IntrinsicTable.GetInfo(operation.Intrinsic);
+
+            return info.Type != IntrinsicType.Crc32;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/PreAllocatorSystemV.cs

@@ -0,0 +1,334 @@
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Collections.Generic;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    class PreAllocatorSystemV : PreAllocator
+    {
+        public static void InsertCallCopies(IntrusiveList<Operation> nodes, Operation node)
+        {
+            Operand dest = node.Destination;
+
+            List<Operand> sources = new List<Operand>
+            {
+                node.GetSource(0)
+            };
+
+            int argsCount = node.SourcesCount - 1;
+
+            int intMax = CallingConvention.GetIntArgumentsOnRegsCount();
+            int vecMax = CallingConvention.GetVecArgumentsOnRegsCount();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            int stackOffset = 0;
+
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = node.GetSource(index + 1);
+
+                bool passOnReg;
+
+                if (source.Type.IsInteger())
+                {
+                    passOnReg = intCount < intMax;
+                }
+                else if (source.Type == OperandType.V128)
+                {
+                    passOnReg = intCount + 1 < intMax;
+                }
+                else
+                {
+                    passOnReg = vecCount < vecMax;
+                }
+
+                if (source.Type == OperandType.V128 && passOnReg)
+                {
+                    // V128 is a struct, we pass each half on a GPR if possible.
+                    Operand argReg  = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+                    Operand argReg2 = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg,  source, Const(0)));
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg2, source, Const(1)));
+
+                    continue;
+                }
+
+                if (passOnReg)
+                {
+                    Operand argReg = source.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntArgumentRegister(intCount++), source.Type)
+                        : Xmm(CallingConvention.GetVecArgumentRegister(vecCount++), source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                    InsertConstantRegCopies(nodes, nodes.AddBefore(node, copyOp));
+
+                    sources.Add(argReg);
+                }
+                else
+                {
+                    Operand offset = Const(stackOffset);
+
+                    Operation spillOp = Operation(Instruction.SpillArg, default, offset, source);
+
+                    InsertConstantRegCopies(nodes, nodes.AddBefore(node, spillOp));
+
+                    stackOffset += source.Type.GetSizeInBytes();
+                }
+            }
+
+            node.SetSources(sources.ToArray());
+
+            if (dest != default)
+            {
+                if (dest.Type == OperandType.V128)
+                {
+                    Operand retLReg = Gpr(CallingConvention.GetIntReturnRegister(),     OperandType.I64);
+                    Operand retHReg = Gpr(CallingConvention.GetIntReturnRegisterHigh(), OperandType.I64);
+
+                    Operation operation = node;
+
+                    node = nodes.AddAfter(node, Operation(Instruction.VectorCreateScalar, dest, retLReg));
+                    nodes.AddAfter(node, Operation(Instruction.VectorInsert, dest, dest, retHReg, Const(1)));
+
+                    operation.Destination = default;
+                }
+                else
+                {
+                    Operand retReg = dest.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntReturnRegister(), dest.Type)
+                        : Xmm(CallingConvention.GetVecReturnRegister(), dest.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, dest, retReg);
+
+                    nodes.AddAfter(node, copyOp);
+
+                    node.Destination = retReg;
+                }
+            }
+        }
+
+        public static void InsertTailcallCopies(IntrusiveList<Operation> nodes, StackAllocator stackAlloc, Operation node)
+        {
+            List<Operand> sources = new List<Operand>
+            {
+                node.GetSource(0)
+            };
+
+            int argsCount = node.SourcesCount - 1;
+
+            int intMax = CallingConvention.GetIntArgumentsOnRegsCount();
+            int vecMax = CallingConvention.GetVecArgumentsOnRegsCount();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            // Handle arguments passed on registers.
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = node.GetSource(1 + index);
+
+                bool passOnReg;
+
+                if (source.Type.IsInteger())
+                {
+                    passOnReg = intCount + 1 < intMax;
+                }
+                else
+                {
+                    passOnReg = vecCount < vecMax;
+                }
+
+                if (source.Type == OperandType.V128 && passOnReg)
+                {
+                    // V128 is a struct, we pass each half on a GPR if possible.
+                    Operand argReg = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+                    Operand argReg2 = Gpr(CallingConvention.GetIntArgumentRegister(intCount++), OperandType.I64);
+
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg, source, Const(0)));
+                    nodes.AddBefore(node, Operation(Instruction.VectorExtract, argReg2, source, Const(1)));
+
+                    continue;
+                }
+
+                if (passOnReg)
+                {
+                    Operand argReg = source.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntArgumentRegister(intCount++), source.Type)
+                        : Xmm(CallingConvention.GetVecArgumentRegister(vecCount++), source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                    InsertConstantRegCopies(nodes, nodes.AddBefore(node, copyOp));
+
+                    sources.Add(argReg);
+                }
+                else
+                {
+                    throw new NotImplementedException("Spilling is not currently supported for tail calls. (too many arguments)");
+                }
+            }
+
+            // The target address must be on the return registers, since we
+            // don't return anything and it is guaranteed to not be a
+            // callee saved register (which would be trashed on the epilogue).
+            Operand retReg = Gpr(CallingConvention.GetIntReturnRegister(), OperandType.I64);
+
+            Operation addrCopyOp = Operation(Instruction.Copy, retReg, node.GetSource(0));
+
+            nodes.AddBefore(node, addrCopyOp);
+
+            sources[0] = retReg;
+
+            node.SetSources(sources.ToArray());
+        }
+
+        public static Operation InsertLoadArgumentCopy(
+            CompilerContext cctx,
+            ref Span<Operation> buffer,
+            IntrusiveList<Operation> nodes,
+            Operand[] preservedArgs,
+            Operation node)
+        {
+            Operand source = node.GetSource(0);
+
+            Debug.Assert(source.Kind == OperandKind.Constant, "Non-constant LoadArgument source kind.");
+
+            int index = source.AsInt32();
+
+            int intCount = 0;
+            int vecCount = 0;
+
+            for (int cIndex = 0; cIndex < index; cIndex++)
+            {
+                OperandType argType = cctx.FuncArgTypes[cIndex];
+
+                if (argType.IsInteger())
+                {
+                    intCount++;
+                }
+                else if (argType == OperandType.V128)
+                {
+                    intCount += 2;
+                }
+                else
+                {
+                    vecCount++;
+                }
+            }
+
+            bool passOnReg;
+
+            if (source.Type.IsInteger())
+            {
+                passOnReg = intCount < CallingConvention.GetIntArgumentsOnRegsCount();
+            }
+            else if (source.Type == OperandType.V128)
+            {
+                passOnReg = intCount + 1 < CallingConvention.GetIntArgumentsOnRegsCount();
+            }
+            else
+            {
+                passOnReg = vecCount < CallingConvention.GetVecArgumentsOnRegsCount();
+            }
+
+            if (passOnReg)
+            {
+                Operand dest = node.Destination;
+
+                if (preservedArgs[index] == default)
+                {
+                    if (dest.Type == OperandType.V128)
+                    {
+                        // V128 is a struct, we pass each half on a GPR if possible.
+                        Operand pArg = Local(OperandType.V128);
+
+                        Operand argLReg = Gpr(CallingConvention.GetIntArgumentRegister(intCount),     OperandType.I64);
+                        Operand argHReg = Gpr(CallingConvention.GetIntArgumentRegister(intCount + 1), OperandType.I64);
+
+                        Operation copyL = Operation(Instruction.VectorCreateScalar, pArg, argLReg);
+                        Operation copyH = Operation(Instruction.VectorInsert,       pArg, pArg, argHReg, Const(1));
+
+                        cctx.Cfg.Entry.Operations.AddFirst(copyH);
+                        cctx.Cfg.Entry.Operations.AddFirst(copyL);
+
+                        preservedArgs[index] = pArg;
+                    }
+                    else
+                    {
+                        Operand pArg = Local(dest.Type);
+
+                        Operand argReg = dest.Type.IsInteger()
+                            ? Gpr(CallingConvention.GetIntArgumentRegister(intCount), dest.Type)
+                            : Xmm(CallingConvention.GetVecArgumentRegister(vecCount), dest.Type);
+
+                        Operation copyOp = Operation(Instruction.Copy, pArg, argReg);
+
+                        cctx.Cfg.Entry.Operations.AddFirst(copyOp);
+
+                        preservedArgs[index] = pArg;
+                    }
+                }
+
+                Operation nextNode;
+
+                if (dest.AssignmentsCount == 1)
+                {
+                    // Let's propagate the argument if we can to avoid copies.
+                    PreAllocatorCommon.Propagate(ref buffer, dest, preservedArgs[index]);
+                    nextNode = node.ListNext;
+                }
+                else
+                {
+                    Operation argCopyOp = Operation(Instruction.Copy, dest, preservedArgs[index]);
+                    nextNode = nodes.AddBefore(node, argCopyOp);
+                }
+
+                Delete(nodes, node);
+                return nextNode;
+            }
+            else
+            {
+                // TODO: Pass on stack.
+                return node;
+            }
+        }
+
+        public static void InsertReturnCopy(IntrusiveList<Operation> nodes, Operation node)
+        {
+            if (node.SourcesCount == 0)
+            {
+                return;
+            }
+
+            Operand source = node.GetSource(0);
+
+            if (source.Type == OperandType.V128)
+            {
+                Operand retLReg = Gpr(CallingConvention.GetIntReturnRegister(),     OperandType.I64);
+                Operand retHReg = Gpr(CallingConvention.GetIntReturnRegisterHigh(), OperandType.I64);
+
+                nodes.AddBefore(node, Operation(Instruction.VectorExtract, retLReg, source, Const(0)));
+                nodes.AddBefore(node, Operation(Instruction.VectorExtract, retHReg, source, Const(1)));
+            }
+            else
+            {
+                Operand retReg = source.Type.IsInteger()
+                    ? Gpr(CallingConvention.GetIntReturnRegister(), source.Type)
+                    : Xmm(CallingConvention.GetVecReturnRegister(), source.Type);
+
+                Operation retCopyOp = Operation(Instruction.Copy, retReg, source);
+
+                nodes.AddBefore(node, retCopyOp);
+            }
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/PreAllocatorWindows.cs

@@ -0,0 +1,327 @@
+using ARMeilleure.CodeGen.RegisterAllocators;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System;
+using System.Diagnostics;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    class PreAllocatorWindows : PreAllocator
+    {
+        public static void InsertCallCopies(IntrusiveList<Operation> nodes, StackAllocator stackAlloc, Operation node)
+        {
+            Operand dest = node.Destination;
+
+            // Handle struct arguments.
+            int retArgs = 0;
+            int stackAllocOffset = 0;
+
+            int AllocateOnStack(int size)
+            {
+                // We assume that the stack allocator is initially empty (TotalSize = 0).
+                // Taking that into account, we can reuse the space allocated for other
+                // calls by keeping track of our own allocated size (stackAllocOffset).
+                // If the space allocated is not big enough, then we just expand it.
+                int offset = stackAllocOffset;
+
+                if (stackAllocOffset + size > stackAlloc.TotalSize)
+                {
+                    stackAlloc.Allocate((stackAllocOffset + size) - stackAlloc.TotalSize);
+                }
+
+                stackAllocOffset += size;
+
+                return offset;
+            }
+
+            Operand arg0Reg = default;
+
+            if (dest != default && dest.Type == OperandType.V128)
+            {
+                int stackOffset = AllocateOnStack(dest.Type.GetSizeInBytes());
+
+                arg0Reg = Gpr(CallingConvention.GetIntArgumentRegister(0), OperandType.I64);
+
+                Operation allocOp = Operation(Instruction.StackAlloc, arg0Reg, Const(stackOffset));
+
+                nodes.AddBefore(node, allocOp);
+
+                retArgs = 1;
+            }
+
+            int argsCount = node.SourcesCount - 1;
+            int maxArgs = CallingConvention.GetArgumentsOnRegsCount() - retArgs;
+
+            if (argsCount > maxArgs)
+            {
+                argsCount = maxArgs;
+            }
+
+            Operand[] sources = new Operand[1 + retArgs + argsCount];
+
+            sources[0] = node.GetSource(0);
+
+            if (arg0Reg != default)
+            {
+                sources[1] = arg0Reg;
+            }
+
+            for (int index = 1; index < node.SourcesCount; index++)
+            {
+                Operand source = node.GetSource(index);
+
+                if (source.Type == OperandType.V128)
+                {
+                    Operand stackAddr = Local(OperandType.I64);
+
+                    int stackOffset = AllocateOnStack(source.Type.GetSizeInBytes());
+
+                    nodes.AddBefore(node, Operation(Instruction.StackAlloc, stackAddr, Const(stackOffset)));
+
+                    Operation storeOp = Operation(Instruction.Store, default, stackAddr, source);
+
+                    InsertConstantRegCopies(nodes, nodes.AddBefore(node, storeOp));
+
+                    node.SetSource(index, stackAddr);
+                }
+            }
+
+            // Handle arguments passed on registers.
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = node.GetSource(index + 1);
+                Operand argReg;
+
+                int argIndex = index + retArgs;
+
+                if (source.Type.IsInteger())
+                {
+                    argReg = Gpr(CallingConvention.GetIntArgumentRegister(argIndex), source.Type);
+                }
+                else
+                {
+                    argReg = Xmm(CallingConvention.GetVecArgumentRegister(argIndex), source.Type);
+                }
+
+                Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                InsertConstantRegCopies(nodes, nodes.AddBefore(node, copyOp));
+
+                sources[1 + retArgs + index] = argReg;
+            }
+
+            // The remaining arguments (those that are not passed on registers)
+            // should be passed on the stack, we write them to the stack with "SpillArg".
+            for (int index = argsCount; index < node.SourcesCount - 1; index++)
+            {
+                Operand source = node.GetSource(index + 1);
+                Operand offset = Const((index + retArgs) * 8);
+
+                Operation spillOp = Operation(Instruction.SpillArg, default, offset, source);
+
+                InsertConstantRegCopies(nodes, nodes.AddBefore(node, spillOp));
+            }
+
+            if (dest != default)
+            {
+                if (dest.Type == OperandType.V128)
+                {
+                    Operand retValueAddr = Local(OperandType.I64);
+
+                    nodes.AddBefore(node, Operation(Instruction.Copy, retValueAddr, arg0Reg));
+
+                    Operation loadOp = Operation(Instruction.Load, dest, retValueAddr);
+
+                    nodes.AddAfter(node, loadOp);
+
+                    node.Destination = default;
+                }
+                else
+                {
+                    Operand retReg = dest.Type.IsInteger()
+                        ? Gpr(CallingConvention.GetIntReturnRegister(), dest.Type)
+                        : Xmm(CallingConvention.GetVecReturnRegister(), dest.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, dest, retReg);
+
+                    nodes.AddAfter(node, copyOp);
+
+                    node.Destination = retReg;
+                }
+            }
+
+            node.SetSources(sources);
+        }
+
+        public static void InsertTailcallCopies(IntrusiveList<Operation> nodes, StackAllocator stackAlloc, Operation node)
+        {
+            int argsCount = node.SourcesCount - 1;
+            int maxArgs = CallingConvention.GetArgumentsOnRegsCount();
+
+            if (argsCount > maxArgs)
+            {
+                throw new NotImplementedException("Spilling is not currently supported for tail calls. (too many arguments)");
+            }
+
+            Operand[] sources = new Operand[1 + argsCount];
+
+            // Handle arguments passed on registers.
+            for (int index = 0; index < argsCount; index++)
+            {
+                Operand source = node.GetSource(1 + index);
+                Operand argReg = source.Type.IsInteger()
+                    ? Gpr(CallingConvention.GetIntArgumentRegister(index), source.Type)
+                    : Xmm(CallingConvention.GetVecArgumentRegister(index), source.Type);
+
+                Operation copyOp = Operation(Instruction.Copy, argReg, source);
+
+                InsertConstantRegCopies(nodes, nodes.AddBefore(node, copyOp));
+
+                sources[1 + index] = argReg;
+            }
+
+            // The target address must be on the return registers, since we
+            // don't return anything and it is guaranteed to not be a
+            // callee saved register (which would be trashed on the epilogue).
+            Operand retReg = Gpr(CallingConvention.GetIntReturnRegister(), OperandType.I64);
+
+            Operation addrCopyOp = Operation(Instruction.Copy, retReg, node.GetSource(0));
+
+            nodes.AddBefore(node, addrCopyOp);
+
+            sources[0] = retReg;
+
+            node.SetSources(sources);
+        }
+
+        public static Operation InsertLoadArgumentCopy(
+            CompilerContext cctx,
+            ref Span<Operation> buffer,
+            IntrusiveList<Operation> nodes,
+            Operand[] preservedArgs,
+            Operation node)
+        {
+            Operand source = node.GetSource(0);
+
+            Debug.Assert(source.Kind == OperandKind.Constant, "Non-constant LoadArgument source kind.");
+
+            int retArgs = cctx.FuncReturnType == OperandType.V128 ? 1 : 0;
+
+            int index = source.AsInt32() + retArgs;
+
+            if (index < CallingConvention.GetArgumentsOnRegsCount())
+            {
+                Operand dest = node.Destination;
+
+                if (preservedArgs[index] == default)
+                {
+                    Operand argReg, pArg;
+
+                    if (dest.Type.IsInteger())
+                    {
+                        argReg = Gpr(CallingConvention.GetIntArgumentRegister(index), dest.Type);
+                        pArg = Local(dest.Type);
+                    }
+                    else if (dest.Type == OperandType.V128)
+                    {
+                        argReg = Gpr(CallingConvention.GetIntArgumentRegister(index), OperandType.I64);
+                        pArg = Local(OperandType.I64);
+                    }
+                    else
+                    {
+                        argReg = Xmm(CallingConvention.GetVecArgumentRegister(index), dest.Type);
+                        pArg = Local(dest.Type);
+                    }
+
+                    Operation copyOp = Operation(Instruction.Copy, pArg, argReg);
+
+                    cctx.Cfg.Entry.Operations.AddFirst(copyOp);
+
+                    preservedArgs[index] = pArg;
+                }
+
+                Operation nextNode;
+
+                if (dest.Type != OperandType.V128 && dest.AssignmentsCount == 1)
+                {
+                    // Let's propagate the argument if we can to avoid copies.
+                    PreAllocatorCommon.Propagate(ref buffer, dest, preservedArgs[index]);
+                    nextNode = node.ListNext;
+                }
+                else
+                {
+                    Operation argCopyOp = Operation(dest.Type == OperandType.V128
+                        ? Instruction.Load
+                        : Instruction.Copy, dest, preservedArgs[index]);
+
+                    nextNode = nodes.AddBefore(node, argCopyOp);
+                }
+
+                Delete(nodes, node);
+                return nextNode;
+            }
+            else
+            {
+                // TODO: Pass on stack.
+                return node;
+            }
+        }
+
+        public static void InsertReturnCopy(
+            CompilerContext cctx,
+            IntrusiveList<Operation> nodes,
+            Operand[] preservedArgs,
+            Operation node)
+        {
+            if (node.SourcesCount == 0)
+            {
+                return;
+            }
+
+            Operand source = node.GetSource(0);
+            Operand retReg;
+
+            if (source.Type.IsInteger())
+            {
+                retReg = Gpr(CallingConvention.GetIntReturnRegister(), source.Type);
+            }
+            else if (source.Type == OperandType.V128)
+            {
+                if (preservedArgs[0] == default)
+                {
+                    Operand preservedArg = Local(OperandType.I64);
+                    Operand arg0 = Gpr(CallingConvention.GetIntArgumentRegister(0), OperandType.I64);
+
+                    Operation copyOp = Operation(Instruction.Copy, preservedArg, arg0);
+
+                    cctx.Cfg.Entry.Operations.AddFirst(copyOp);
+
+                    preservedArgs[0] = preservedArg;
+                }
+
+                retReg = preservedArgs[0];
+            }
+            else
+            {
+                retReg = Xmm(CallingConvention.GetVecReturnRegister(), source.Type);
+            }
+
+            if (source.Type == OperandType.V128)
+            {
+                Operation retStoreOp = Operation(Instruction.Store, default, retReg, source);
+
+                nodes.AddBefore(node, retStoreOp);
+            }
+            else
+            {
+                Operation retCopyOp = Operation(Instruction.Copy, retReg, source);
+
+                nodes.AddBefore(node, retCopyOp);
+            }
+
+            node.SetSources(Array.Empty<Operand>());
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/X86Condition.cs

@@ -0,0 +1,47 @@
+using ARMeilleure.IntermediateRepresentation;
+using System;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    enum X86Condition
+    {
+        Overflow       = 0x0,
+        NotOverflow    = 0x1,
+        Below          = 0x2,
+        AboveOrEqual   = 0x3,
+        Equal          = 0x4,
+        NotEqual       = 0x5,
+        BelowOrEqual   = 0x6,
+        Above          = 0x7,
+        Sign           = 0x8,
+        NotSign        = 0x9,
+        ParityEven     = 0xa,
+        ParityOdd      = 0xb,
+        Less           = 0xc,
+        GreaterOrEqual = 0xd,
+        LessOrEqual    = 0xe,
+        Greater        = 0xf
+    }
+
+    static class ComparisonX86Extensions
+    {
+        public static X86Condition ToX86Condition(this Comparison comp)
+        {
+            return comp switch
+            {
+                Comparison.Equal            => X86Condition.Equal,
+                Comparison.NotEqual         => X86Condition.NotEqual,
+                Comparison.Greater          => X86Condition.Greater,
+                Comparison.LessOrEqual      => X86Condition.LessOrEqual,
+                Comparison.GreaterUI        => X86Condition.Above,
+                Comparison.LessOrEqualUI    => X86Condition.BelowOrEqual,
+                Comparison.GreaterOrEqual   => X86Condition.GreaterOrEqual,
+                Comparison.Less             => X86Condition.Less,
+                Comparison.GreaterOrEqualUI => X86Condition.AboveOrEqual,
+                Comparison.LessUI           => X86Condition.Below,
+
+                _ => throw new ArgumentException(null, nameof(comp))
+            };
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/X86Instruction.cs

@@ -0,0 +1,231 @@
+namespace ARMeilleure.CodeGen.X86
+{
+    enum X86Instruction
+    {
+        None,
+        Add,
+        Addpd,
+        Addps,
+        Addsd,
+        Addss,
+        Aesdec,
+        Aesdeclast,
+        Aesenc,
+        Aesenclast,
+        Aesimc,
+        And,
+        Andnpd,
+        Andnps,
+        Andpd,
+        Andps,
+        Blendvpd,
+        Blendvps,
+        Bsr,
+        Bswap,
+        Call,
+        Cmovcc,
+        Cmp,
+        Cmppd,
+        Cmpps,
+        Cmpsd,
+        Cmpss,
+        Cmpxchg,
+        Cmpxchg16b,
+        Cmpxchg8,
+        Comisd,
+        Comiss,
+        Crc32,
+        Crc32_16,
+        Crc32_8,
+        Cvtdq2pd,
+        Cvtdq2ps,
+        Cvtpd2dq,
+        Cvtpd2ps,
+        Cvtps2dq,
+        Cvtps2pd,
+        Cvtsd2si,
+        Cvtsd2ss,
+        Cvtsi2sd,
+        Cvtsi2ss,
+        Cvtss2sd,
+        Cvtss2si,
+        Div,
+        Divpd,
+        Divps,
+        Divsd,
+        Divss,
+        Gf2p8affineqb,
+        Haddpd,
+        Haddps,
+        Idiv,
+        Imul,
+        Imul128,
+        Insertps,
+        Jmp,
+        Ldmxcsr,
+        Lea,
+        Maxpd,
+        Maxps,
+        Maxsd,
+        Maxss,
+        Minpd,
+        Minps,
+        Minsd,
+        Minss,
+        Mov,
+        Mov16,
+        Mov8,
+        Movd,
+        Movdqu,
+        Movhlps,
+        Movlhps,
+        Movq,
+        Movsd,
+        Movss,
+        Movsx16,
+        Movsx32,
+        Movsx8,
+        Movzx16,
+        Movzx8,
+        Mul128,
+        Mulpd,
+        Mulps,
+        Mulsd,
+        Mulss,
+        Neg,
+        Not,
+        Or,
+        Paddb,
+        Paddd,
+        Paddq,
+        Paddw,
+        Palignr,
+        Pand,
+        Pandn,
+        Pavgb,
+        Pavgw,
+        Pblendvb,
+        Pclmulqdq,
+        Pcmpeqb,
+        Pcmpeqd,
+        Pcmpeqq,
+        Pcmpeqw,
+        Pcmpgtb,
+        Pcmpgtd,
+        Pcmpgtq,
+        Pcmpgtw,
+        Pextrb,
+        Pextrd,
+        Pextrq,
+        Pextrw,
+        Pinsrb,
+        Pinsrd,
+        Pinsrq,
+        Pinsrw,
+        Pmaxsb,
+        Pmaxsd,
+        Pmaxsw,
+        Pmaxub,
+        Pmaxud,
+        Pmaxuw,
+        Pminsb,
+        Pminsd,
+        Pminsw,
+        Pminub,
+        Pminud,
+        Pminuw,
+        Pmovsxbw,
+        Pmovsxdq,
+        Pmovsxwd,
+        Pmovzxbw,
+        Pmovzxdq,
+        Pmovzxwd,
+        Pmulld,
+        Pmullw,
+        Pop,
+        Popcnt,
+        Por,
+        Pshufb,
+        Pshufd,
+        Pslld,
+        Pslldq,
+        Psllq,
+        Psllw,
+        Psrad,
+        Psraw,
+        Psrld,
+        Psrlq,
+        Psrldq,
+        Psrlw,
+        Psubb,
+        Psubd,
+        Psubq,
+        Psubw,
+        Punpckhbw,
+        Punpckhdq,
+        Punpckhqdq,
+        Punpckhwd,
+        Punpcklbw,
+        Punpckldq,
+        Punpcklqdq,
+        Punpcklwd,
+        Push,
+        Pxor,
+        Rcpps,
+        Rcpss,
+        Ror,
+        Roundpd,
+        Roundps,
+        Roundsd,
+        Roundss,
+        Rsqrtps,
+        Rsqrtss,
+        Sar,
+        Setcc,
+        Sha256Msg1,
+        Sha256Msg2,
+        Sha256Rnds2,
+        Shl,
+        Shr,
+        Shufpd,
+        Shufps,
+        Sqrtpd,
+        Sqrtps,
+        Sqrtsd,
+        Sqrtss,
+        Stmxcsr,
+        Sub,
+        Subpd,
+        Subps,
+        Subsd,
+        Subss,
+        Test,
+        Unpckhpd,
+        Unpckhps,
+        Unpcklpd,
+        Unpcklps,
+        Vblendvpd,
+        Vblendvps,
+        Vcvtph2ps,
+        Vcvtps2ph,
+        Vfmadd231pd,
+        Vfmadd231ps,
+        Vfmadd231sd,
+        Vfmadd231ss,
+        Vfmsub231sd,
+        Vfmsub231ss,
+        Vfnmadd231pd,
+        Vfnmadd231ps,
+        Vfnmadd231sd,
+        Vfnmadd231ss,
+        Vfnmsub231sd,
+        Vfnmsub231ss,
+        Vpblendvb,
+        Vpternlogd,
+        Xor,
+        Xorpd,
+        Xorps,
+
+        Count
+    }
+}

src/ARMeilleure/CodeGen/X86/X86Optimizer.cs

@@ -0,0 +1,259 @@
+using ARMeilleure.CodeGen.Optimizations;
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System.Collections.Generic;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
+using static ARMeilleure.IntermediateRepresentation.Operation.Factory;
+
+namespace ARMeilleure.CodeGen.X86
+{
+    static class X86Optimizer
+    {
+        private const int MaxConstantUses = 10000;
+
+        public static void RunPass(ControlFlowGraph cfg)
+        {
+            var constants = new Dictionary<ulong, Operand>();
+
+            Operand GetConstantCopy(BasicBlock block, Operation operation, Operand source)
+            {
+                // If the constant has many uses, we also force a new constant mov to be added, in order
+                // to avoid overflow of the counts field (that is limited to 16 bits).
+                if (!constants.TryGetValue(source.Value, out var constant) || constant.UsesCount > MaxConstantUses)
+                {
+                    constant = Local(source.Type);
+
+                    Operation copyOp = Operation(Instruction.Copy, constant, source);
+
+                    block.Operations.AddBefore(operation, copyOp);
+
+                    constants[source.Value] = constant;
+                }
+
+                return constant;
+            }
+
+            for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
+            {
+                constants.Clear();
+
+                Operation nextNode;
+
+                for (Operation node = block.Operations.First; node != default; node = nextNode)
+                {
+                    nextNode = node.ListNext;
+
+                    // Insert copies for constants that can't fit on a 32-bits immediate.
+                    // Doing this early unblocks a few optimizations.
+                    if (node.Instruction == Instruction.Add)
+                    {
+                        Operand src1 = node.GetSource(0);
+                        Operand src2 = node.GetSource(1);
+
+                        if (src1.Kind == OperandKind.Constant && (src1.Relocatable || CodeGenCommon.IsLongConst(src1)))
+                        {
+                            node.SetSource(0, GetConstantCopy(block, node, src1));
+                        }
+
+                        if (src2.Kind == OperandKind.Constant && (src2.Relocatable || CodeGenCommon.IsLongConst(src2)))
+                        {
+                            node.SetSource(1, GetConstantCopy(block, node, src2));
+                        }
+                    }
+
+                    // Try to fold something like:
+                    //  shl rbx, 2
+                    //  add rax, rbx
+                    //  add rax, 0xcafe
+                    //  mov rax, [rax]
+                    // Into:
+                    //  mov rax, [rax+rbx*4+0xcafe]
+                    if (IsMemoryLoadOrStore(node.Instruction))
+                    {
+                        OperandType type;
+
+                        if (node.Destination != default)
+                        {
+                            type = node.Destination.Type;
+                        }
+                        else
+                        {
+                            type = node.GetSource(1).Type;
+                        }
+
+                        Operand memOp = GetMemoryOperandOrNull(node.GetSource(0), type);
+
+                        if (memOp != default)
+                        {
+                            node.SetSource(0, memOp);
+                        }
+                    }
+                }
+            }
+
+            Optimizer.RemoveUnusedNodes(cfg);
+        }
+
+        private static Operand GetMemoryOperandOrNull(Operand addr, OperandType type)
+        {
+            Operand baseOp = addr;
+
+            // First we check if the address is the result of a local X with 32-bits immediate
+            // addition. If that is the case, then the baseOp is X, and the memory operand immediate
+            // becomes the addition immediate. Otherwise baseOp keeps being the address.
+            int imm = GetConstOp(ref baseOp);
+
+            // Now we check if the baseOp is the result of a local Y with a local Z addition.
+            // If that is the case, we now set baseOp to Y and indexOp to Z. We further check
+            // if Z is the result of a left shift of local W by a value >= 0 and <= 3, if that
+            // is the case, we set indexOp to W and adjust the scale value of the memory operand
+            // to match that of the left shift.
+            // There is one missed case, which is the address being a shift result, but this is
+            // probably not worth optimizing as it should never happen.
+            (Operand indexOp, Multiplier scale) = GetIndexOp(ref baseOp);
+
+            // If baseOp is still equal to address, then there's nothing that can be optimized.
+            if (baseOp == addr)
+            {
+                return default;
+            }
+
+            if (imm == 0 && scale == Multiplier.x1 && indexOp != default)
+            {
+                imm = GetConstOp(ref indexOp);
+            }
+
+            return MemoryOp(type, baseOp, indexOp, scale, imm);
+        }
+
+        private static int GetConstOp(ref Operand baseOp)
+        {
+            Operation operation = GetAsgOpWithInst(baseOp, Instruction.Add);
+
+            if (operation == default)
+            {
+                return 0;
+            }
+
+            Operand src1 = operation.GetSource(0);
+            Operand src2 = operation.GetSource(1);
+
+            Operand constOp;
+            Operand otherOp;
+
+            if (src1.Kind == OperandKind.Constant && src2.Kind == OperandKind.LocalVariable)
+            {
+                constOp = src1;
+                otherOp = src2;
+            }
+            else if (src1.Kind == OperandKind.LocalVariable && src2.Kind == OperandKind.Constant)
+            {
+                constOp = src2;
+                otherOp = src1;
+            }
+            else
+            {
+                return 0;
+            }
+
+            // If we have addition by 64-bits constant, then we can't optimize it further,
+            // as we can't encode a 64-bits immediate on the memory operand.
+            if (CodeGenCommon.IsLongConst(constOp))
+            {
+                return 0;
+            }
+
+            baseOp = otherOp;
+
+            return constOp.AsInt32();
+        }
+
+        private static (Operand, Multiplier) GetIndexOp(ref Operand baseOp)
+        {
+            Operand indexOp = default;
+
+            Multiplier scale = Multiplier.x1;
+
+            Operation addOp = GetAsgOpWithInst(baseOp, Instruction.Add);
+
+            if (addOp == default)
+            {
+                return (indexOp, scale);
+            }
+
+            Operand src1 = addOp.GetSource(0);
+            Operand src2 = addOp.GetSource(1);
+
+            if (src1.Kind != OperandKind.LocalVariable || src2.Kind != OperandKind.LocalVariable)
+            {
+                return (indexOp, scale);
+            }
+
+            baseOp = src1;
+            indexOp = src2;
+
+            Operation shlOp = GetAsgOpWithInst(src1, Instruction.ShiftLeft);
+
+            bool indexOnSrc2 = false;
+
+            if (shlOp == default)
+            {
+                shlOp = GetAsgOpWithInst(src2, Instruction.ShiftLeft);
+
+                indexOnSrc2 = true;
+            }
+
+            if (shlOp != default)
+            {
+                Operand shSrc = shlOp.GetSource(0);
+                Operand shift = shlOp.GetSource(1);
+
+                if (shSrc.Kind == OperandKind.LocalVariable && shift.Kind == OperandKind.Constant && shift.Value <= 3)
+                {
+                    scale = shift.Value switch
+                    {
+                        1 => Multiplier.x2,
+                        2 => Multiplier.x4,
+                        3 => Multiplier.x8,
+                        _ => Multiplier.x1
+                    };
+
+                    baseOp = indexOnSrc2 ? src1 : src2;
+                    indexOp = shSrc;
+                }
+            }
+
+            return (indexOp, scale);
+        }
+
+        private static Operation GetAsgOpWithInst(Operand op, Instruction inst)
+        {
+            // If we have multiple assignments, folding is not safe
+            // as the value may be different depending on the
+            // control flow path.
+            if (op.AssignmentsCount != 1)
+            {
+                return default;
+            }
+
+            Operation asgOp = op.Assignments[0];
+
+            if (asgOp.Instruction != inst)
+            {
+                return default;
+            }
+
+            return asgOp;
+        }
+
+        private static bool IsMemoryLoadOrStore(Instruction inst)
+        {
+            return inst == Instruction.Load ||
+                   inst == Instruction.Load16 ||
+                   inst == Instruction.Load8 ||
+                   inst == Instruction.Store ||
+                   inst == Instruction.Store16 ||
+                   inst == Instruction.Store8;
+        }
+    }
+}

src/ARMeilleure/CodeGen/X86/X86Register.cs

@@ -0,0 +1,41 @@
+namespace ARMeilleure.CodeGen.X86
+{
+    enum X86Register
+    {
+        Invalid = -1,
+
+        Rax = 0,
+        Rcx = 1,
+        Rdx = 2,
+        Rbx = 3,
+        Rsp = 4,
+        Rbp = 5,
+        Rsi = 6,
+        Rdi = 7,
+        R8  = 8,
+        R9  = 9,
+        R10 = 10,
+        R11 = 11,
+        R12 = 12,
+        R13 = 13,
+        R14 = 14,
+        R15 = 15,
+
+        Xmm0  = 0,
+        Xmm1  = 1,
+        Xmm2  = 2,
+        Xmm3  = 3,
+        Xmm4  = 4,
+        Xmm5  = 5,
+        Xmm6  = 6,
+        Xmm7  = 7,
+        Xmm8  = 8,
+        Xmm9  = 9,
+        Xmm10 = 10,
+        Xmm11 = 11,
+        Xmm12 = 12,
+        Xmm13 = 13,
+        Xmm14 = 14,
+        Xmm15 = 15
+    }
+}