mono-jit (#167)

- fix alignment issue with jitcache that could cause a crash.
- Reimplement sparse-jit to only use 1 layer.
    - the old 2 layer approach basically only ever used 1 path due to the bottom-layer size, so doing a double look-up seemed wasteful.
    - changed memory block allocation stategy to allocate based on requested code range instead of the possible address-space.

Co-authored-by: LotP1 <68976644+LotP1@users.noreply.github.com>
Reviewed-on: https://git.ryujinx.app/projects/Ryubing/pulls/167

src/ARMeilleure/Common/AddressTablePresets.cs

@@ -20,45 +20,21 @@ namespace ARMeilleure.Common
                 new( 1,  6)
         ];
         
-        private static readonly AddressTableLevel[] _levels64BitSparseTiny =
+        private static readonly AddressTableLevel[] _monoSparse64Bit =
         [
-            new( 11, 28),
+            new( 2, 37)
-                new( 2, 9)
         ];
 
-        private static readonly AddressTableLevel[] _levels32BitSparseTiny =
+        private static readonly AddressTableLevel[] _monoSparse32Bit =
         [
-            new( 10, 22),
+            new( 1, 31)
-                new( 1, 9)
         ];
         
-        private static readonly AddressTableLevel[] _levels64BitSparseGiant =
+        public static AddressTableLevel[] GetArmPreset(bool for64Bits, bool sparse)
-        [
-            new( 38, 1),
-                new( 2, 36)
-        ];
-
-        private static readonly AddressTableLevel[] _levels32BitSparseGiant =
-        [
-            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 ? _monoSparse64Bit : _monoSparse32Bit;
-                {
-                    return for64Bits ? _levels64BitSparseTiny : _levels32BitSparseTiny;
-                }
-                else
-                {
-                    return for64Bits ? _levels64BitSparseGiant : _levels32BitSparseGiant;
-                }
             }
             else
             {

src/ARMeilleure/Instructions/InstEmitFlowHelper.cs

@@ -5,7 +5,7 @@ using ARMeilleure.IntermediateRepresentation;
 using ARMeilleure.State;
 using ARMeilleure.Translation;
 using ARMeilleure.Translation.PTC;
-
+using System.Linq;
 using static ARMeilleure.Instructions.InstEmitHelper;
 using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
 
@@ -238,7 +238,7 @@ namespace ARMeilleure.Instructions
             }
             else if (table.Sparse)
             {
-                // Inline table lookup. Only enabled when the sparse function table is enabled with 2 levels.
+                // Inline table lookup. Only enabled when the sparse function table is enabled with 1 level.
                 // Deliberately attempts to avoid branches.
 
                 Operand tableBase = !context.HasPtc ?
@@ -247,18 +247,15 @@ namespace ARMeilleure.Instructions
 
                 hostAddress = tableBase;
 
-                for (int i = 0; i < table.Levels.Length; i++)
+                AddressTableLevel level = table.Levels.Last();
-                {
+                int clearBits = 64 - (level.Index + level.Length);
-                    AddressTableLevel level = table.Levels[i];
-                    int clearBits = 64 - (level.Index + level.Length);
 
-                    Operand index = context.ShiftLeft(
+                Operand index = context.ShiftLeft(
-                        context.ShiftRightUI(context.ShiftLeft(guestAddress, Const(clearBits)), Const(clearBits + level.Index)),
+                    context.ShiftRightUI(context.ShiftLeft(guestAddress, Const(clearBits)), Const(clearBits + level.Index)),
-                        Const(3)
+                    Const(3)
-                    );
+                );
 
-                    hostAddress = context.Load(OperandType.I64, context.Add(hostAddress, index));
+                hostAddress = context.Load(OperandType.I64, context.Add(hostAddress, index));
-                }
             }
             else
             {

src/ARMeilleure/Translation/Cache/JitCache.cs

@@ -129,7 +129,9 @@ namespace ARMeilleure.Translation.Cache
         {
             int endOffs = offset + size;
             int regionStart = (offset % (int)CacheSize) & ~_pageMask;
-            int regionEnd = ((endOffs % (int)CacheSize) + _pageMask) & ~_pageMask;
+            int regionEnd = endOffs % (int)CacheSize == 0
+                ? (((int)CacheSize) + _pageMask) & ~_pageMask
+                : ((endOffs % (int)CacheSize) + _pageMask) & ~_pageMask;
 
             GetRegion(offset).Block.MapAsRwx((ulong)regionStart, (ulong)(regionEnd - regionStart));
         }
@@ -138,7 +140,9 @@ namespace ARMeilleure.Translation.Cache
         {
             int endOffs = offset + size;
             int regionStart = (offset % (int)CacheSize) & ~_pageMask;
-            int regionEnd = ((endOffs % (int)CacheSize) + _pageMask) & ~_pageMask;
+            int regionEnd = endOffs % (int)CacheSize == 0
+                ? (((int)CacheSize) + _pageMask) & ~_pageMask
+                : ((endOffs % (int)CacheSize) + _pageMask) & ~_pageMask;
 
             GetRegion(offset).Block.MapAsRx((ulong)regionStart, (ulong)(regionEnd - regionStart));
         }

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -34,7 +34,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 7016; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 7018; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

src/ARMeilleure/Translation/TranslatorStubs.cs

@@ -161,7 +161,7 @@ namespace ARMeilleure.Translation
             context.BranchIfTrue(lblFallback, masked);
 
             Operand index = default;
-            Operand page = Const((long)_functionTable.Base);
+            Operand page = Const(_functionTable.Base);
 
             for (int i = 0; i < _functionTable.Levels.Length; i++)
             {

src/Ryujinx.Cpu/AddressTable.cs

@@ -81,12 +81,13 @@ namespace ARMeilleure.Common
 
         private bool _disposed;
         private TEntry** _table;
+        private TEntry* _sparseTable;
         private readonly List<AddressTablePage> _pages;
         private TEntry _fill;
 
-        private readonly MemoryBlock _sparseFill;
+        private MemoryBlock _sparseFill;
-        private readonly SparseMemoryBlock _fillBottomLevel;
+        private SparseMemoryBlock _fillBottomLevel;
-        private readonly TEntry* _fillBottomLevelPtr;
+        private TEntry* _fillBottomLevelPtr;
 
         private readonly List<TableSparseBlock> _sparseReserved;
         private readonly ReaderWriterLockSlim _sparseLock;
@@ -122,16 +123,23 @@ namespace ARMeilleure.Common
             {
                 ObjectDisposedException.ThrowIf(_disposed, this);
 
-                lock (_pages)
+                if (Sparse)
                 {
-                    return (nint)GetRootPage();
+                    return (nint)_sparseTable;
+                }
+                else
+                {
+                    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"/>.
+        /// <see cref="AddressTableLevel"/>.
         /// </summary>
         /// <param name="levels">Levels for the address table</param>
         /// <param name="sparse">True if the bottom page should be sparsely mapped</param>
@@ -156,18 +164,8 @@ namespace ARMeilleure.Common
             if (sparse)
             {
                 // If the address table is sparse, allocate a fill block
-
-                _sparseFill = new MemoryBlock(268435456ul, MemoryAllocationFlags.Mirrorable); //low Power TC uses size: 65536ul
-
-                ulong bottomLevelSize = (1ul << levels.Last().Length) * (ulong)sizeof(TEntry);
-
-                _fillBottomLevel = new SparseMemoryBlock(bottomLevelSize, null, _sparseFill);
-                _fillBottomLevelPtr = (TEntry*)_fillBottomLevel.Block.Pointer;
-
                 _sparseReserved = [];
                 _sparseLock = new ReaderWriterLockSlim();
-
-                _sparseBlockSize = bottomLevelSize;
             }
         }
 
@@ -209,13 +207,24 @@ namespace ARMeilleure.Common
         public void SignalCodeRange(ulong address, ulong size)
         {
             AddressTableLevel bottom = Levels.Last();
-            ulong bottomLevelEntries = 1ul << bottom.Length;
             
-            ulong entryIndex = address >> bottom.Index;
             ulong entries = size >> bottom.Index;
-            entries += entryIndex - BitUtils.AlignDown(entryIndex, bottomLevelEntries);
             
-            _sparseBlockSize = Math.Max(_sparseBlockSize, BitUtils.AlignUp(entries, bottomLevelEntries) * (ulong)sizeof(TEntry));
+            if (Sparse)
+            {
+                ulong bottomLevelSize = (ulong)BitUtils.Pow2RoundUp((int)entries) * (ulong)sizeof(TEntry);
+                
+                _sparseFill = new MemoryBlock(bottomLevelSize, MemoryAllocationFlags.Mirrorable);
+
+                _fillBottomLevel = new SparseMemoryBlock(bottomLevelSize, null, _sparseFill);
+                _fillBottomLevelPtr = (TEntry*)_fillBottomLevel.Block.Pointer;
+
+                _sparseBlockSize = bottomLevelSize;
+                
+                _sparseTable = (TEntry*)Allocate((int)entries, Fill, leaf: true);
+
+                _sparseTable -= Levels.Last().GetValue(address);
+            }
         }
 
         /// <inheritdoc/>
@@ -234,15 +243,26 @@ namespace ARMeilleure.Common
                 throw new ArgumentException($"Address 0x{address:X} is not mapped onto the table.", nameof(address));
             }
 
-            lock (_pages)
+            if (Sparse)
             {
-                TEntry* page = GetPage(address);
+                long index = Levels.Last().GetValue(address);
                 
-                long index = Levels[^1].GetValue(address);
+                EnsureMapped((nint)(_sparseTable + index));
                 
-                EnsureMapped((nint)(page + index));
+                return ref _sparseTable[index];
+            }
+            else
+            {
+                lock (_pages)
+                {
+                    TEntry* page = GetPage(address);
 
-                return ref page[index];
+                    long index = Levels.Last().GetValue(address);
+
+                    EnsureMapped((nint)(page + index));
+
+                    return ref page[index];
+                }
             }
         }
 

src/Ryujinx.Cpu/LightningJit/Arm32/Target/Arm64/InstEmitFlow.cs

@@ -3,6 +3,7 @@ using Ryujinx.Cpu.LightningJit.CodeGen;
 using Ryujinx.Cpu.LightningJit.CodeGen.Arm64;
 using System;
 using System.Diagnostics;
+using System.Linq;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 
@@ -189,7 +190,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
             }
             else if (inlineLookup)
             {
-                // Inline table lookup. Only enabled when the sparse function table is enabled with 2 levels.
+                // Inline table lookup. Only enabled when the sparse function table is enabled with 1 level.
 
                 Operand indexReg = Register(NextFreeRegister(tempRegister + 1, tempGuestAddress));
 
@@ -203,18 +204,15 @@ namespace Ryujinx.Cpu.LightningJit.Arm32.Target.Arm64
                 // Index into the table.
                 asm.Mov(rn, tableBase);
 
-                for (int i = 0; i < funcTable.Levels.Length; i++)
+                AddressTableLevel level = funcTable.Levels.Last();
-                {
+                asm.Ubfx(indexReg, guestAddress, level.Index, level.Length);
-                    AddressTableLevel level = funcTable.Levels[i];
+                asm.Lsl(indexReg, indexReg, Const(3));
-                    asm.Ubfx(indexReg, guestAddress, level.Index, level.Length);
-                    asm.Lsl(indexReg, indexReg, Const(3));
 
-                    // Index into the page.
+                // Index into the page.
-                    asm.Add(rn, rn, indexReg);
+                asm.Add(rn, rn, indexReg);
 
-                    // Load the page address.
+                // Load the page address.
-                    asm.LdrRiUn(rn, rn, 0);
+                asm.LdrRiUn(rn, rn, 0);
-                }
 
                 if (tempGuestAddress != -1)
                 {

src/Ryujinx.Cpu/LightningJit/Arm64/Target/Arm64/InstEmitSystem.cs

@@ -3,6 +3,7 @@ using Ryujinx.Cpu.LightningJit.CodeGen;
 using Ryujinx.Cpu.LightningJit.CodeGen.Arm64;
 using System;
 using System.Diagnostics;
+using System.Linq;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
@@ -355,7 +356,7 @@ namespace Ryujinx.Cpu.LightningJit.Arm64.Target.Arm64
             }
             else if (inlineLookup)
             {
-                // Inline table lookup. Only enabled when the sparse function table is enabled with 2 levels.
+                // Inline table lookup. Only enabled when the sparse function table is enabled with 1 level.
 
                 Operand indexReg = Register(NextFreeRegister(tempRegister + 1, tempGuestAddress));
 
@@ -369,18 +370,15 @@ namespace Ryujinx.Cpu.LightningJit.Arm64.Target.Arm64
                 // Index into the table.
                 asm.Mov(rn, tableBase);
 
-                for (int i = 0; i < funcTable.Levels.Length; i++)
+                AddressTableLevel level = funcTable.Levels.Last();
-                {
+                asm.Ubfx(indexReg, guestAddress, level.Index, level.Length);
-                    AddressTableLevel level = funcTable.Levels[i];
+                asm.Lsl(indexReg, indexReg, Const(3));
-                    asm.Ubfx(indexReg, guestAddress, level.Index, level.Length);
-                    asm.Lsl(indexReg, indexReg, Const(3));
 
-                    // Index into the page.
+                // Index into the page.
-                    asm.Add(rn, rn, indexReg);
+                asm.Add(rn, rn, indexReg);
 
-                    // Load the page address.
+                // Load the page address.
-                    asm.LdrRiUn(rn, rn, 0);
+                asm.LdrRiUn(rn, rn, 0);
-                }
 
                 if (tempGuestAddress != -1)
                 {