mirror of
https://github.com/GreemDev/Ryujinx.git
synced 2024-12-23 19:45:47 +00:00
22b2cb39af
* Turn `MemoryOperand` into a struct * Remove `IntrinsicOperation` * Remove `PhiNode` * Remove `Node` * Turn `Operand` into a struct * Turn `Operation` into a struct * Clean up pool management methods * Add `Arena` allocator * Move `OperationHelper` to `Operation.Factory` * Move `OperandHelper` to `Operand.Factory` * Optimize `Operation` a bit * Fix `Arena` initialization * Rename `NativeList<T>` to `ArenaList<T>` * Reduce `Operand` size from 88 to 56 bytes * Reduce `Operation` size from 56 to 40 bytes * Add optimistic interning of Register & Constant operands * Optimize `RegisterUsage` pass a bit * Optimize `RemoveUnusedNodes` pass a bit Iterating in reverse-order allows killing dependency chains in a single pass. * Fix PPTC symbols * Optimize `BasicBlock` a bit Reduce allocations from `_successor` & `DominanceFrontiers` * Fix `Operation` resize * Make `Arena` expandable Change the arena allocator to be expandable by allocating in pages, with some of them being pooled. Currently 32 pages are pooled. An LRU removal mechanism should probably be added to it. Apparently MHR can allocate bitmaps large enough to exceed the 16MB limit for the type. * Move `Arena` & `ArenaList` to `Common` * Remove `ThreadStaticPool` & co * Add `PhiOperation` * Reduce `Operand` size from 56 from 48 bytes * Add linear-probing to `Operand` intern table * Optimize `HybridAllocator` a bit * Add `Allocators` class * Tune `ArenaAllocator` sizes * Add page removal mechanism to `ArenaAllocator` Remove pages which have not been used for more than 5s after each reset. I am on fence if this would be better using a Gen2 callback object like the one in System.Buffers.ArrayPool<T>, to trim the pool. Because right now if a large translation happens, the pages will be freed only after a reset. This reset may not happen for a while because no new translation is hit, but the arena base sizes are rather small. * Fix `OOM` when allocating larger than page size in `ArenaAllocator` Tweak resizing mechanism for Operand.Uses and Assignemnts. * Optimize `Optimizer` a bit * Optimize `Operand.Add<T>/Remove<T>` a bit * Clean up `PreAllocator` * Fix phi insertion order Reduce codegen diffs. * Fix code alignment * Use new heuristics for degree of parallelism * Suppress warnings * Address gdkchan's feedback Renamed `GetValue()` to `GetValueUnsafe()` to make it more clear that `Operand.Value` should usually not be modified directly. * Add fast path to `ArenaAllocator` * Assembly for `ArenaAllocator.Allocate(ulong)`: .L0: mov rax, [rcx+0x18] lea r8, [rax+rdx] cmp r8, [rcx+0x10] ja short .L2 .L1: mov rdx, [rcx+8] add rax, [rdx+8] mov [rcx+0x18], r8 ret .L2: jmp ArenaAllocator.AllocateSlow(UInt64) A few variable/field had to be changed to ulong so that RyuJIT avoids emitting zero-extends. * Implement a new heuristic to free pooled pages. If an arena is used often, it is more likely that its pages will be needed, so the pages are kept for longer (e.g: during PPTC rebuild or burst sof compilations). If is not used often, then it is more likely that its pages will not be needed (e.g: after PPTC rebuild or bursts of compilations). * Address riperiperi's feedback * Use `EqualityComparer<T>` in `IntrusiveList<T>` Avoids a potential GC hole in `Equals(T, T)`.
524 lines
No EOL
16 KiB
C#
524 lines
No EOL
16 KiB
C#
using ARMeilleure.Common;
|
|
using ARMeilleure.Translation.PTC;
|
|
using System;
|
|
using System.Collections.Generic;
|
|
using System.Diagnostics;
|
|
using System.Runtime.CompilerServices;
|
|
|
|
namespace ARMeilleure.IntermediateRepresentation
|
|
{
|
|
unsafe struct Operand : IEquatable<Operand>
|
|
{
|
|
internal struct Data
|
|
{
|
|
public byte Kind;
|
|
public byte Type;
|
|
public byte SymbolType;
|
|
public ushort AssignmentsCount;
|
|
public ushort AssignmentsCapacity;
|
|
public ushort UsesCount;
|
|
public ushort UsesCapacity;
|
|
public Operation* Assignments;
|
|
public Operation* Uses;
|
|
public ulong Value;
|
|
public ulong SymbolValue;
|
|
}
|
|
|
|
private Data* _data;
|
|
|
|
public OperandKind Kind
|
|
{
|
|
get => (OperandKind)_data->Kind;
|
|
private set => _data->Kind = (byte)value;
|
|
}
|
|
|
|
public OperandType Type
|
|
{
|
|
get => (OperandType)_data->Type;
|
|
private set => _data->Type = (byte)value;
|
|
}
|
|
|
|
public ulong Value
|
|
{
|
|
get => _data->Value;
|
|
private set => _data->Value = value;
|
|
}
|
|
|
|
public Symbol Symbol
|
|
{
|
|
get
|
|
{
|
|
Debug.Assert(Kind != OperandKind.Memory);
|
|
|
|
return new Symbol((SymbolType)_data->SymbolType, _data->SymbolValue);
|
|
}
|
|
private set
|
|
{
|
|
Debug.Assert(Kind != OperandKind.Memory);
|
|
|
|
if (value.Type == SymbolType.None)
|
|
{
|
|
_data->SymbolType = (byte)SymbolType.None;
|
|
}
|
|
else
|
|
{
|
|
_data->SymbolType = (byte)value.Type;
|
|
_data->SymbolValue = value.Value;
|
|
}
|
|
}
|
|
}
|
|
|
|
public ReadOnlySpan<Operation> Assignments
|
|
{
|
|
get
|
|
{
|
|
Debug.Assert(Kind != OperandKind.Memory);
|
|
|
|
return new ReadOnlySpan<Operation>(_data->Assignments, _data->AssignmentsCount);
|
|
}
|
|
}
|
|
|
|
public ReadOnlySpan<Operation> Uses
|
|
{
|
|
get
|
|
{
|
|
Debug.Assert(Kind != OperandKind.Memory);
|
|
|
|
return new ReadOnlySpan<Operation>(_data->Uses, _data->UsesCount);
|
|
}
|
|
}
|
|
|
|
public int UsesCount => _data->UsesCount;
|
|
public int AssignmentsCount => _data->AssignmentsCount;
|
|
|
|
public bool Relocatable => Symbol.Type != SymbolType.None;
|
|
|
|
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
|
public Register GetRegister()
|
|
{
|
|
Debug.Assert(Kind == OperandKind.Register);
|
|
|
|
return new Register((int)Value & 0xffffff, (RegisterType)(Value >> 24));
|
|
}
|
|
|
|
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
|
public MemoryOperand GetMemory()
|
|
{
|
|
Debug.Assert(Kind == OperandKind.Memory);
|
|
|
|
return new MemoryOperand(this);
|
|
}
|
|
|
|
public int GetLocalNumber()
|
|
{
|
|
Debug.Assert(Kind == OperandKind.LocalVariable);
|
|
|
|
return (int)Value;
|
|
}
|
|
|
|
public byte AsByte()
|
|
{
|
|
return (byte)Value;
|
|
}
|
|
|
|
public short AsInt16()
|
|
{
|
|
return (short)Value;
|
|
}
|
|
|
|
public int AsInt32()
|
|
{
|
|
return (int)Value;
|
|
}
|
|
|
|
public long AsInt64()
|
|
{
|
|
return (long)Value;
|
|
}
|
|
|
|
public float AsFloat()
|
|
{
|
|
return BitConverter.Int32BitsToSingle((int)Value);
|
|
}
|
|
|
|
public double AsDouble()
|
|
{
|
|
return BitConverter.Int64BitsToDouble((long)Value);
|
|
}
|
|
|
|
internal ref ulong GetValueUnsafe()
|
|
{
|
|
return ref _data->Value;
|
|
}
|
|
|
|
internal void NumberLocal(int number)
|
|
{
|
|
if (Kind != OperandKind.LocalVariable)
|
|
{
|
|
throw new InvalidOperationException("The operand is not a local variable.");
|
|
}
|
|
|
|
Value = (ulong)number;
|
|
}
|
|
|
|
public void AddAssignment(Operation operation)
|
|
{
|
|
if (Kind == OperandKind.LocalVariable)
|
|
{
|
|
Add(operation, ref _data->Assignments, ref _data->AssignmentsCount, ref _data->AssignmentsCapacity);
|
|
}
|
|
else if (Kind == OperandKind.Memory)
|
|
{
|
|
MemoryOperand memOp = GetMemory();
|
|
Operand addr = memOp.BaseAddress;
|
|
Operand index = memOp.Index;
|
|
|
|
if (addr != default)
|
|
{
|
|
Add(operation, ref addr._data->Assignments, ref addr._data->AssignmentsCount, ref addr._data->AssignmentsCapacity);
|
|
}
|
|
|
|
if (index != default)
|
|
{
|
|
Add(operation, ref index._data->Assignments, ref index._data->AssignmentsCount, ref index._data->AssignmentsCapacity);
|
|
}
|
|
}
|
|
}
|
|
|
|
public void RemoveAssignment(Operation operation)
|
|
{
|
|
if (Kind == OperandKind.LocalVariable)
|
|
{
|
|
Remove(operation, ref _data->Assignments, ref _data->AssignmentsCount);
|
|
}
|
|
else if (Kind == OperandKind.Memory)
|
|
{
|
|
MemoryOperand memOp = GetMemory();
|
|
Operand addr = memOp.BaseAddress;
|
|
Operand index = memOp.Index;
|
|
|
|
if (addr != default)
|
|
{
|
|
Remove(operation, ref addr._data->Assignments, ref addr._data->AssignmentsCount);
|
|
}
|
|
|
|
if (index != default)
|
|
{
|
|
Remove(operation, ref index._data->Assignments, ref index._data->AssignmentsCount);
|
|
}
|
|
}
|
|
}
|
|
|
|
public void AddUse(Operation operation)
|
|
{
|
|
if (Kind == OperandKind.LocalVariable)
|
|
{
|
|
Add(operation, ref _data->Uses, ref _data->UsesCount, ref _data->UsesCapacity);
|
|
}
|
|
else if (Kind == OperandKind.Memory)
|
|
{
|
|
MemoryOperand memOp = GetMemory();
|
|
Operand addr = memOp.BaseAddress;
|
|
Operand index = memOp.Index;
|
|
|
|
if (addr != default)
|
|
{
|
|
Add(operation, ref addr._data->Uses, ref addr._data->UsesCount, ref addr._data->UsesCapacity);
|
|
}
|
|
|
|
if (index != default)
|
|
{
|
|
Add(operation, ref index._data->Uses, ref index._data->UsesCount, ref index._data->UsesCapacity);
|
|
}
|
|
}
|
|
}
|
|
|
|
public void RemoveUse(Operation operation)
|
|
{
|
|
if (Kind == OperandKind.LocalVariable)
|
|
{
|
|
Remove(operation, ref _data->Uses, ref _data->UsesCount);
|
|
}
|
|
else if (Kind == OperandKind.Memory)
|
|
{
|
|
MemoryOperand memOp = GetMemory();
|
|
Operand addr = memOp.BaseAddress;
|
|
Operand index = memOp.Index;
|
|
|
|
if (addr != default)
|
|
{
|
|
Remove(operation, ref addr._data->Uses, ref addr._data->UsesCount);
|
|
}
|
|
|
|
if (index != default)
|
|
{
|
|
Remove(operation, ref index._data->Uses, ref index._data->UsesCount);
|
|
}
|
|
}
|
|
}
|
|
|
|
private static void New<T>(ref T* data, ref ushort count, ref ushort capacity, ushort initialCapacity) where T : unmanaged
|
|
{
|
|
count = 0;
|
|
capacity = initialCapacity;
|
|
data = Allocators.References.Allocate<T>(initialCapacity);
|
|
}
|
|
|
|
private static void Add<T>(T item, ref T* data, ref ushort count, ref ushort capacity) where T : unmanaged
|
|
{
|
|
if (count < capacity)
|
|
{
|
|
data[(uint)count++] = item;
|
|
|
|
return;
|
|
}
|
|
|
|
// Could not add item in the fast path, fallback onto the slow path.
|
|
ExpandAdd(item, ref data, ref count, ref capacity);
|
|
|
|
static void ExpandAdd(T item, ref T* data, ref ushort count, ref ushort capacity)
|
|
{
|
|
ushort newCount = checked((ushort)(count + 1));
|
|
ushort newCapacity = (ushort)Math.Min(capacity * 2, ushort.MaxValue);
|
|
|
|
var oldSpan = new Span<T>(data, count);
|
|
|
|
capacity = newCapacity;
|
|
data = Allocators.References.Allocate<T>(capacity);
|
|
|
|
oldSpan.CopyTo(new Span<T>(data, count));
|
|
|
|
data[count] = item;
|
|
count = newCount;
|
|
}
|
|
}
|
|
|
|
private static void Remove<T>(in T item, ref T* data, ref ushort count) where T : unmanaged
|
|
{
|
|
var span = new Span<T>(data, count);
|
|
|
|
for (int i = 0; i < span.Length; i++)
|
|
{
|
|
if (EqualityComparer<T>.Default.Equals(span[i], item))
|
|
{
|
|
if (i + 1 < count)
|
|
{
|
|
span.Slice(i + 1).CopyTo(span.Slice(i));
|
|
}
|
|
|
|
count--;
|
|
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
public override int GetHashCode()
|
|
{
|
|
if (Kind == OperandKind.LocalVariable)
|
|
{
|
|
return base.GetHashCode();
|
|
}
|
|
else
|
|
{
|
|
return (int)Value ^ ((int)Kind << 16) ^ ((int)Type << 20);
|
|
}
|
|
}
|
|
|
|
public bool Equals(Operand operand)
|
|
{
|
|
return operand._data == _data;
|
|
}
|
|
|
|
public override bool Equals(object obj)
|
|
{
|
|
return obj is Operand operand && Equals(operand);
|
|
}
|
|
|
|
public static bool operator ==(Operand a, Operand b)
|
|
{
|
|
return a.Equals(b);
|
|
}
|
|
|
|
public static bool operator !=(Operand a, Operand b)
|
|
{
|
|
return !a.Equals(b);
|
|
}
|
|
|
|
public static class Factory
|
|
{
|
|
private const int InternTableSize = 256;
|
|
private const int InternTableProbeLength = 8;
|
|
|
|
[ThreadStatic]
|
|
private static Data* _internTable;
|
|
|
|
private static Data* InternTable
|
|
{
|
|
get
|
|
{
|
|
if (_internTable == null)
|
|
{
|
|
_internTable = (Data*)NativeAllocator.Instance.Allocate((uint)sizeof(Data) * InternTableSize);
|
|
|
|
// Make sure the table is zeroed.
|
|
new Span<Data>(_internTable, InternTableSize).Clear();
|
|
}
|
|
|
|
return _internTable;
|
|
}
|
|
}
|
|
|
|
private static Operand Make(OperandKind kind, OperandType type, ulong value, Symbol symbol = default)
|
|
{
|
|
Debug.Assert(kind != OperandKind.None);
|
|
|
|
Data* data = null;
|
|
|
|
// If constant or register, then try to look up in the intern table before allocating.
|
|
if (kind == OperandKind.Constant || kind == OperandKind.Register)
|
|
{
|
|
uint hash = (uint)HashCode.Combine(kind, type, value);
|
|
|
|
// Look in the next InternTableProbeLength slots for a match.
|
|
for (uint i = 0; i < InternTableProbeLength; i++)
|
|
{
|
|
Operand interned = new();
|
|
interned._data = &InternTable[(hash + i) % InternTableSize];
|
|
|
|
// If slot matches the allocation request then return that slot.
|
|
if (interned.Kind == kind && interned.Type == type && interned.Value == value && interned.Symbol == symbol)
|
|
{
|
|
return interned;
|
|
}
|
|
// Otherwise if the slot is not occupied, we store in that slot.
|
|
else if (interned.Kind == OperandKind.None)
|
|
{
|
|
data = interned._data;
|
|
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// If we could not get a slot from the intern table, we allocate somewhere else and store there.
|
|
if (data == null)
|
|
{
|
|
data = Allocators.Operands.Allocate<Data>();
|
|
}
|
|
|
|
*data = default;
|
|
|
|
Operand result = new();
|
|
result._data = data;
|
|
result.Value = value;
|
|
result.Kind = kind;
|
|
result.Type = type;
|
|
|
|
if (kind != OperandKind.Memory)
|
|
{
|
|
result.Symbol = symbol;
|
|
}
|
|
|
|
// If local variable, then the use and def list is initialized with default sizes.
|
|
if (kind == OperandKind.LocalVariable)
|
|
{
|
|
New(ref result._data->Assignments, ref result._data->AssignmentsCount, ref result._data->AssignmentsCapacity, 1);
|
|
New(ref result._data->Uses, ref result._data->UsesCount, ref result._data->UsesCapacity, 4);
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
public static Operand Const(OperandType type, long value)
|
|
{
|
|
Debug.Assert(type is OperandType.I32 or OperandType.I64);
|
|
|
|
return type == OperandType.I32 ? Const((int)value) : Const(value);
|
|
}
|
|
|
|
public static Operand Const(bool value)
|
|
{
|
|
return Const(value ? 1 : 0);
|
|
}
|
|
|
|
public static Operand Const(int value)
|
|
{
|
|
return Const((uint)value);
|
|
}
|
|
|
|
public static Operand Const(uint value)
|
|
{
|
|
return Make(OperandKind.Constant, OperandType.I32, value);
|
|
}
|
|
|
|
public static Operand Const(long value)
|
|
{
|
|
return Const(value, symbol: default);
|
|
}
|
|
|
|
public static Operand Const<T>(ref T reference, Symbol symbol = default)
|
|
{
|
|
return Const((long)Unsafe.AsPointer(ref reference), symbol);
|
|
}
|
|
|
|
public static Operand Const(long value, Symbol symbol)
|
|
{
|
|
return Make(OperandKind.Constant, OperandType.I64, (ulong)value, symbol);
|
|
}
|
|
|
|
public static Operand Const(ulong value)
|
|
{
|
|
return Make(OperandKind.Constant, OperandType.I64, value);
|
|
}
|
|
|
|
public static Operand ConstF(float value)
|
|
{
|
|
return Make(OperandKind.Constant, OperandType.FP32, (ulong)BitConverter.SingleToInt32Bits(value));
|
|
}
|
|
|
|
public static Operand ConstF(double value)
|
|
{
|
|
return Make(OperandKind.Constant, OperandType.FP64, (ulong)BitConverter.DoubleToInt64Bits(value));
|
|
}
|
|
|
|
public static Operand Label()
|
|
{
|
|
return Make(OperandKind.Label, OperandType.None, 0);
|
|
}
|
|
|
|
public static Operand Local(OperandType type)
|
|
{
|
|
return Make(OperandKind.LocalVariable, type, 0);
|
|
}
|
|
|
|
public static Operand Register(int index, RegisterType regType, OperandType type)
|
|
{
|
|
return Make(OperandKind.Register, type, (ulong)((int)regType << 24 | index));
|
|
}
|
|
|
|
public static Operand Undef()
|
|
{
|
|
return Make(OperandKind.Undefined, OperandType.None, 0);
|
|
}
|
|
|
|
public static Operand MemoryOp(
|
|
OperandType type,
|
|
Operand baseAddress,
|
|
Operand index = default,
|
|
Multiplier scale = Multiplier.x1,
|
|
int displacement = 0)
|
|
{
|
|
Operand result = Make(OperandKind.Memory, type, 0);
|
|
|
|
MemoryOperand memory = result.GetMemory();
|
|
memory.BaseAddress = baseAddress;
|
|
memory.Index = index;
|
|
memory.Scale = scale;
|
|
memory.Displacement = displacement;
|
|
|
|
return result;
|
|
}
|
|
}
|
|
}
|
|
} |