crafterpika/ryujinx
1
0
Fork 0
mirror of https://git.ryujinx.app/ryubing/ryujinx.git synced 2025-06-28 06:46:24 +02:00
Code Issues Projects Releases Packages Wiki Activity

JIT Sparse Function Table (#250)

Browse source 
More up to date build of the JIT Sparse PR for continued development.
JIT Sparse Function Table was originally developed by riperiperi for the
original Ryujinx project, and decreased the amount of layers in the
Function Table structure, to decrease lookup times at the cost of
slightly higher RAM usage.
This PR rebalances the JIT Sparse Function Table to be a bit more RAM
intensive, but faster in workloads where the JIT Function Table is a
bottleneck. Faster RAM will see a bigger impact and slower RAM (DDR3 and
potentially slow DDR4) will see a slight performance decrease.
This PR also implements a base for a PPTC profile system that could
allow for PPTC with ExeFS mods enabled in the future.
This PR also potentially fixes a strange issue where Avalonia would time
out in some rare instances, e.g. when running ExeFS mods with TotK and a
strange controller configuration.

---------

Co-authored-by: Evan Husted <gr33m11@gmail.com>
This commit is contained in:
LotP1 2024-11-22 22:33:44 +01:00 • committed by GitHub
parent 5534001152
commit e653848a2c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
27 changed files with 990 additions and 327 deletions
Download patch file Download diff file Expand all files Collapse all files

252
src/ARMeilleure/Common/AddressTable.cs
View file

@ -1,252 +0,0 @@
using ARMeilleure.Diagnostics;
using System;
using System.Collections.Generic;
using System.Runtime.InteropServices;
namespace ARMeilleure.Common
{
/// <summary>
/// Represents a table of guest address to a value.
/// </summary>
/// <typeparam name="TEntry">Type of the value</typeparam>
public unsafe class AddressTable<TEntry> : IDisposable where TEntry : unmanaged
{
/// <summary>
/// Represents a level in an <see cref="AddressTable{TEntry}"/>.
/// </summary>
public readonly struct Level
{
/// <summary>
/// Gets the index of the <see cref="Level"/> in the guest address.
/// </summary>
public int Index { get; }
/// <summary>
/// Gets the length of the <see cref="Level"/> in the guest address.
/// </summary>
public int Length { get; }
/// <summary>
/// Gets the mask which masks the bits used by the <see cref="Level"/>.
/// </summary>
public ulong Mask => ((1ul << Length) - 1) << Index;
/// <summary>
/// Initializes a new instance of the <see cref="Level"/> structure with the specified
/// <paramref name="index"/> and <paramref name="length"/>.
/// </summary>
/// <param name="index">Index of the <see cref="Level"/></param>
/// <param name="length">Length of the <see cref="Level"/></param>
public Level(int index, int length)
{
(Index, Length) = (index, length);
}
/// <summary>
/// Gets the value of the <see cref="Level"/> from the specified guest <paramref name="address"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns>Value of the <see cref="Level"/> from the specified guest <paramref name="address"/></returns>
public int GetValue(ulong address)
{
return (int)((address & Mask) >> Index);
}
}
private bool _disposed;
private TEntry** _table;
private readonly List<nint> _pages;
/// <summary>
/// Gets the bits used by the <see cref="Levels"/> of the <see cref="AddressTable{TEntry}"/> instance.
/// </summary>
public ulong Mask { get; }
/// <summary>
/// Gets the <see cref="Level"/>s used by the <see cref="AddressTable{TEntry}"/> instance.
/// </summary>
public Level[] Levels { get; }
/// <summary>
/// Gets or sets the default fill value of newly created leaf pages.
/// </summary>
public TEntry Fill { get; set; }
/// <summary>
/// Gets the base address of the <see cref="EntryTable{TEntry}"/>.
/// </summary>
/// <exception cref="ObjectDisposedException"><see cref="EntryTable{TEntry}"/> instance was disposed</exception>
public nint Base
{
get
{
ObjectDisposedException.ThrowIf(_disposed, this);
lock (_pages)
{
return (nint)GetRootPage();
}
}
}
/// <summary>
/// Constructs a new instance of the <see cref="AddressTable{TEntry}"/> class with the specified list of
/// <see cref="Level"/>.
/// </summary>
/// <exception cref="ArgumentNullException"><paramref name="levels"/> is null</exception>
/// <exception cref="ArgumentException">Length of <paramref name="levels"/> is less than 2</exception>
public AddressTable(Level[] levels)
{
ArgumentNullException.ThrowIfNull(levels);
if (levels.Length < 2)
{
throw new ArgumentException("Table must be at least 2 levels deep.", nameof(levels));
}
_pages = new List<nint>(capacity: 16);
Levels = levels;
Mask = 0;
foreach (var level in Levels)
{
Mask |= level.Mask;
}
}
/// <summary>
/// Determines if the specified <paramref name="address"/> is in the range of the
/// <see cref="AddressTable{TEntry}"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns><see langword="true"/> if is valid; otherwise <see langword="false"/></returns>
public bool IsValid(ulong address)
{
return (address & ~Mask) == 0;
}
/// <summary>
/// Gets a reference to the value at the specified guest <paramref name="address"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns>Reference to the value at the specified guest <paramref name="address"/></returns>
/// <exception cref="ObjectDisposedException"><see cref="EntryTable{TEntry}"/> instance was disposed</exception>
/// <exception cref="ArgumentException"><paramref name="address"/> is not mapped</exception>
public ref TEntry GetValue(ulong address)
{
ObjectDisposedException.ThrowIf(_disposed, this);
if (!IsValid(address))
{
throw new ArgumentException($"Address 0x{address:X} is not mapped onto the table.", nameof(address));
}
lock (_pages)
{
return ref GetPage(address)[Levels[^1].GetValue(address)];
}
}
/// <summary>
/// Gets the leaf page for the specified guest <paramref name="address"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns>Leaf page for the specified guest <paramref name="address"/></returns>
private TEntry* GetPage(ulong address)
{
TEntry** page = GetRootPage();
for (int i = 0; i < Levels.Length - 1; i++)
{
ref Level level = ref Levels[i];
ref TEntry* nextPage = ref page[level.GetValue(address)];
if (nextPage == null)
{
ref Level nextLevel = ref Levels[i + 1];
nextPage = i == Levels.Length - 2 ?
(TEntry*)Allocate(1 << nextLevel.Length, Fill, leaf: true) :
(TEntry*)Allocate(1 << nextLevel.Length, nint.Zero, leaf: false);
}
page = (TEntry**)nextPage;
}
return (TEntry*)page;
}
/// <summary>
/// Lazily initialize and get the root page of the <see cref="AddressTable{TEntry}"/>.
/// </summary>
/// <returns>Root page of the <see cref="AddressTable{TEntry}"/></returns>
private TEntry** GetRootPage()
{
if (_table == null)
{
_table = (TEntry**)Allocate(1 << Levels[0].Length, fill: nint.Zero, leaf: false);
}
return _table;
}
/// <summary>
/// Allocates a block of memory of the specified type and length.
/// </summary>
/// <typeparam name="T">Type of elements</typeparam>
/// <param name="length">Number of elements</param>
/// <param name="fill">Fill value</param>
/// <param name="leaf"><see langword="true"/> if leaf; otherwise <see langword="false"/></param>
/// <returns>Allocated block</returns>
private nint Allocate<T>(int length, T fill, bool leaf) where T : unmanaged
{
var size = sizeof(T) * length;
var page = (nint)NativeAllocator.Instance.Allocate((uint)size);
var span = new Span<T>((void*)page, length);
span.Fill(fill);
_pages.Add(page);
TranslatorEventSource.Log.AddressTableAllocated(size, leaf);
return page;
}
/// <summary>
/// Releases all resources used by the <see cref="AddressTable{TEntry}"/> instance.
/// </summary>
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
/// <summary>
/// Releases all unmanaged and optionally managed resources used by the <see cref="AddressTable{TEntry}"/>
/// instance.
/// </summary>
/// <param name="disposing"><see langword="true"/> to dispose managed resources also; otherwise just unmanaged resouces</param>
protected virtual void Dispose(bool disposing)
{
if (!_disposed)
{
foreach (var page in _pages)
{
Marshal.FreeHGlobal(page);
}
_disposed = true;
}
}
/// <summary>
/// Frees resources used by the <see cref="AddressTable{TEntry}"/> instance.
/// </summary>
~AddressTable()
{
Dispose(false);
}
}
}

44
src/ARMeilleure/Common/AddressTableLevel.cs Normal file
View file

@ -0,0 +1,44 @@
namespace ARMeilleure.Common
{
/// <summary>
/// Represents a level in an <see cref="IAddressTable{TEntry}"/>.
/// </summary>
public readonly struct AddressTableLevel
{
/// <summary>
/// Gets the index of the <see cref="Level"/> in the guest address.
/// </summary>
public int Index { get; }
/// <summary>
/// Gets the length of the <see cref="AddressTableLevel"/> in the guest address.
/// </summary>
public int Length { get; }
/// <summary>
/// Gets the mask which masks the bits used by the <see cref="AddressTableLevel"/>.
/// </summary>
public ulong Mask => ((1ul << Length) - 1) << Index;
/// <summary>
/// Initializes a new instance of the <see cref="AddressTableLevel"/> structure with the specified
/// <paramref name="index"/> and <paramref name="length"/>.
/// </summary>
/// <param name="index">Index of the <see cref="AddressTableLevel"/></param>
/// <param name="length">Length of the <see cref="AddressTableLevel"/></param>
public AddressTableLevel(int index, int length)
{
(Index, Length) = (index, length);
}
/// <summary>
/// Gets the value of the <see cref="AddressTableLevel"/> from the specified guest <paramref name="address"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns>Value of the <see cref="AddressTableLevel"/> from the specified guest <paramref name="address"/></returns>
public int GetValue(ulong address)
{
return (int)((address & Mask) >> Index);
}
}
}

75
src/ARMeilleure/Common/AddressTablePresets.cs Normal file
View file

@ -0,0 +1,75 @@
namespace ARMeilleure.Common
{
public static class AddressTablePresets
{
private static readonly AddressTableLevel[] _levels64Bit =
new AddressTableLevel[]
{
new(31, 17),
new(23, 8),
new(15, 8),
new( 7, 8),
new( 2, 5),
};
private static readonly AddressTableLevel[] _levels32Bit =
new AddressTableLevel[]
{
new(31, 17),
new(23, 8),
new(15, 8),
new( 7, 8),
new( 1, 6),
};
private static readonly AddressTableLevel[] _levels64BitSparseTiny =
new AddressTableLevel[]
{
new( 11, 28),
new( 2, 9),
};
private static readonly AddressTableLevel[] _levels32BitSparseTiny =
new AddressTableLevel[]
{
new( 10, 22),
new( 1, 9),
};
private static readonly AddressTableLevel[] _levels64BitSparseGiant =
new AddressTableLevel[]
{
new( 38, 1),
new( 2, 36),
};
private static readonly AddressTableLevel[] _levels32BitSparseGiant =
new AddressTableLevel[]
{
new( 31, 1),
new( 1, 30),
};
//high power will run worse on DDR3 systems and some DDR4 systems due to the higher ram utilization
//low power will never run worse than non-sparse, but for most systems it won't be necessary
//high power is always used, but I've left low power in here for future reference
public static AddressTableLevel[] GetArmPreset(bool for64Bits, bool sparse, bool lowPower = false)
{
if (sparse)
{
if (lowPower)
{
return for64Bits ? _levels64BitSparseTiny : _levels32BitSparseTiny;
}
else
{
return for64Bits ? _levels64BitSparseGiant : _levels32BitSparseGiant;
}
}
else
{
return for64Bits ? _levels64Bit : _levels32Bit;
}
}
}
}

2
src/ARMeilleure/Common/Allocator.cs
View file

@ -2,7 +2,7 @@ using System;
namespace ARMeilleure.Common
{
unsafe abstract class Allocator : IDisposable
public unsafe abstract class Allocator : IDisposable
{
public T* Allocate<T>(ulong count = 1) where T : unmanaged
{

51
src/ARMeilleure/Common/IAddressTable.cs Normal file
View file

@ -0,0 +1,51 @@
using System;
namespace ARMeilleure.Common
{
public interface IAddressTable<TEntry> : IDisposable where TEntry : unmanaged
{
/// <summary>
/// True if the address table's bottom level is sparsely mapped.
/// This also ensures the second bottom level is filled with a dummy page rather than 0.
/// </summary>
bool Sparse { get; }
/// <summary>
/// Gets the bits used by the <see cref="Levels"/> of the <see cref="IAddressTable{TEntry}"/> instance.
/// </summary>
ulong Mask { get; }
/// <summary>
/// Gets the <see cref="AddressTableLevel"/>s used by the <see cref="IAddressTable{TEntry}"/> instance.
/// </summary>
AddressTableLevel[] Levels { get; }
/// <summary>
/// Gets or sets the default fill value of newly created leaf pages.
/// </summary>
TEntry Fill { get; set; }
/// <summary>
/// Gets the base address of the <see cref="EntryTable{TEntry}"/>.
/// </summary>
/// <exception cref="ObjectDisposedException"><see cref="EntryTable{TEntry}"/> instance was disposed</exception>
nint Base { get; }
/// <summary>
/// Determines if the specified <paramref name="address"/> is in the range of the
/// <see cref="IAddressTable{TEntry}"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns><see langword="true"/> if is valid; otherwise <see langword="false"/></returns>
bool IsValid(ulong address);
/// <summary>
/// Gets a reference to the value at the specified guest <paramref name="address"/>.
/// </summary>
/// <param name="address">Guest address</param>
/// <returns>Reference to the value at the specified guest <paramref name="address"/></returns>
/// <exception cref="ObjectDisposedException"><see cref="EntryTable{TEntry}"/> instance was disposed</exception>
/// <exception cref="ArgumentException"><paramref name="address"/> is not mapped</exception>
ref TEntry GetValue(ulong address);
}
}

2
src/ARMeilleure/Common/NativeAllocator.cs
View file

@ -3,7 +3,7 @@ using System.Runtime.InteropServices;
namespace ARMeilleure.Common
{
unsafe sealed class NativeAllocator : Allocator
public unsafe sealed class NativeAllocator : Allocator
{
public static NativeAllocator Instance { get; } = new();

26
src/ARMeilleure/Instructions/InstEmitFlowHelper.cs
View file

@ -193,6 +193,8 @@ namespace ARMeilleure.Instructions
Operand hostAddress;
var table = context.FunctionTable;
// If address is mapped onto the function table, we can skip the table walk. Otherwise we fallback
// onto the dispatch stub.
if (guestAddress.Kind == OperandKind.Constant && context.FunctionTable.IsValid(guestAddress.Value))
@ -203,6 +205,30 @@ namespace ARMeilleure.Instructions
hostAddress = context.Load(OperandType.I64, hostAddressAddr);
}
else if (table.Sparse)
{
// Inline table lookup. Only enabled when the sparse function table is enabled with 2 levels.
// Deliberately attempts to avoid branches.
Operand tableBase = !context.HasPtc ?
Const(table.Base) :
Const(table.Base, Ptc.FunctionTableSymbol);
hostAddress = tableBase;
for (int i = 0; i < table.Levels.Length; i++)
{
var level = table.Levels[i];
int clearBits = 64 - (level.Index + level.Length);
Operand index = context.ShiftLeft(
context.ShiftRightUI(context.ShiftLeft(guestAddress, Const(clearBits)), Const(clearBits + level.Index)),
Const(3)
);
hostAddress = context.Load(OperandType.I64, context.Add(hostAddress, index));
}
}
else
{
hostAddress = !context.HasPtc ?

2
src/ARMeilleure/Signal/NativeSignalHandlerGenerator.cs
View file

@ -8,7 +8,7 @@ namespace ARMeilleure.Signal
{
public static class NativeSignalHandlerGenerator
{
public const int MaxTrackedRanges = 8;
public const int MaxTrackedRanges = 16;
private const int StructAddressOffset = 0;
private const int StructWriteOffset = 4;

4
src/ARMeilleure/Translation/ArmEmitterContext.cs
View file

@ -46,7 +46,7 @@ namespace ARMeilleure.Translation
public IMemoryManager Memory { get; }
public EntryTable<uint> CountTable { get; }
public AddressTable<ulong> FunctionTable { get; }
public IAddressTable<ulong> FunctionTable { get; }
public TranslatorStubs Stubs { get; }
public ulong EntryAddress { get; }
@ -62,7 +62,7 @@ namespace ARMeilleure.Translation
public ArmEmitterContext(
IMemoryManager memory,
EntryTable<uint> countTable,
AddressTable<ulong> funcTable,
IAddressTable<ulong> funcTable,
TranslatorStubs stubs,
ulong entryAddress,
bool highCq,

15
src/ARMeilleure/Translation/PTC/Ptc.cs
View file

@ -30,7 +30,7 @@ namespace ARMeilleure.Translation.PTC
private const string OuterHeaderMagicString = "PTCohd\0\0";
private const string InnerHeaderMagicString = "PTCihd\0\0";
private const uint InternalVersion = 6950; //! To be incremented manually for each change to the ARMeilleure project.
private const uint InternalVersion = 6992; //! To be incremented manually for each change to the ARMeilleure project.
private const string ActualDir = "0";
private const string BackupDir = "1";
@ -41,6 +41,7 @@ namespace ARMeilleure.Translation.PTC
public static readonly Symbol PageTableSymbol = new(SymbolType.Special, 1);
public static readonly Symbol CountTableSymbol = new(SymbolType.Special, 2);
public static readonly Symbol DispatchStubSymbol = new(SymbolType.Special, 3);
public static readonly Symbol FunctionTableSymbol = new(SymbolType.Special, 4);
private const byte FillingByte = 0x00;
private const CompressionLevel SaveCompressionLevel = CompressionLevel.Fastest;
@ -101,7 +102,7 @@ namespace ARMeilleure.Translation.PTC
Disable();
}
public void Initialize(string titleIdText, string displayVersion, bool enabled, MemoryManagerType memoryMode)
public void Initialize(string titleIdText, string displayVersion, bool enabled, MemoryManagerType memoryMode, string cacheSelector)
{
Wait();
@ -127,6 +128,8 @@ namespace ARMeilleure.Translation.PTC
DisplayVersion = !string.IsNullOrEmpty(displayVersion) ? displayVersion : DisplayVersionDefault;
_memoryMode = memoryMode;
Logger.Info?.Print(LogClass.Ptc, $"PPTC (v{InternalVersion}) Profile: {DisplayVersion}-{cacheSelector}");
string workPathActual = Path.Combine(AppDataManager.GamesDirPath, TitleIdText, "cache", "cpu", ActualDir);
string workPathBackup = Path.Combine(AppDataManager.GamesDirPath, TitleIdText, "cache", "cpu", BackupDir);
@ -140,8 +143,8 @@ namespace ARMeilleure.Translation.PTC
Directory.CreateDirectory(workPathBackup);
}
CachePathActual = Path.Combine(workPathActual, DisplayVersion);
CachePathBackup = Path.Combine(workPathBackup, DisplayVersion);
CachePathActual = Path.Combine(workPathActual, DisplayVersion) + "-" + cacheSelector;
CachePathBackup = Path.Combine(workPathBackup, DisplayVersion) + "-" + cacheSelector;
PreLoad();
Profiler.PreLoad();
@ -706,6 +709,10 @@ namespace ARMeilleure.Translation.PTC
{
imm = translator.Stubs.DispatchStub;
}
else if (symbol == FunctionTableSymbol)
{
imm = translator.FunctionTable.Base;
}
if (imm == null)
{

30
src/ARMeilleure/Translation/Translator.cs
View file

@ -22,33 +22,13 @@ namespace ARMeilleure.Translation
{
public class Translator
{
private static readonly AddressTable<ulong>.Level[] _levels64Bit =
new AddressTable<ulong>.Level[]
{
new(31, 17),
new(23, 8),
new(15, 8),
new( 7, 8),
new( 2, 5),
};
private static readonly AddressTable<ulong>.Level[] _levels32Bit =
new AddressTable<ulong>.Level[]
{
new(31, 17),
new(23, 8),
new(15, 8),
new( 7, 8),
new( 1, 6),
};
private readonly IJitMemoryAllocator _allocator;
private readonly ConcurrentQueue<KeyValuePair<ulong, TranslatedFunction>> _oldFuncs;
private readonly Ptc _ptc;
internal TranslatorCache<TranslatedFunction> Functions { get; }
internal AddressTable<ulong> FunctionTable { get; }
internal IAddressTable<ulong> FunctionTable { get; }
internal EntryTable<uint> CountTable { get; }
internal TranslatorStubs Stubs { get; }
internal TranslatorQueue Queue { get; }
@ -57,7 +37,7 @@ namespace ARMeilleure.Translation
private Thread[] _backgroundTranslationThreads;
private volatile int _threadCount;
public Translator(IJitMemoryAllocator allocator, IMemoryManager memory, bool for64Bits)
public Translator(IJitMemoryAllocator allocator, IMemoryManager memory, IAddressTable<ulong> functionTable)
{
_allocator = allocator;
Memory = memory;
@ -72,15 +52,15 @@ namespace ARMeilleure.Translation
CountTable = new EntryTable<uint>();
Functions = new TranslatorCache<TranslatedFunction>();
FunctionTable = new AddressTable<ulong>(for64Bits ? _levels64Bit : _levels32Bit);
FunctionTable = functionTable;
Stubs = new TranslatorStubs(FunctionTable);
FunctionTable.Fill = (ulong)Stubs.SlowDispatchStub;
}
public IPtcLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled)
public IPtcLoadState LoadDiskCache(string titleIdText, string displayVersion, bool enabled, string cacheSelector)
{
_ptc.Initialize(titleIdText, displayVersion, enabled, Memory.Type);
_ptc.Initialize(titleIdText, displayVersion, enabled, Memory.Type, cacheSelector);
return _ptc;
}

4
src/ARMeilleure/Translation/TranslatorStubs.cs
View file

@ -19,7 +19,7 @@ namespace ARMeilleure.Translation
private bool _disposed;
private readonly AddressTable<ulong> _functionTable;
private readonly IAddressTable<ulong> _functionTable;
private readonly Lazy<nint> _dispatchStub;
private readonly Lazy<DispatcherFunction> _dispatchLoop;
private readonly Lazy<WrapperFunction> _contextWrapper;
@ -86,7 +86,7 @@ namespace ARMeilleure.Translation
/// </summary>
/// <param name="functionTable">Function table used to store pointers to the functions that the guest code will call</param>
/// <exception cref="ArgumentNullException"><paramref name="translator"/> is null</exception>
public TranslatorStubs(AddressTable<ulong> functionTable)
public TranslatorStubs(IAddressTable<ulong> functionTable)
{
ArgumentNullException.ThrowIfNull(functionTable);