mirror of
https://github.com/GreemDev/Ryujinx
synced 2024-11-23 02:06:55 +01:00
a16682cfd3
* Allow _volatile to be set from MultiRegionHandle checks again Tracking handles have a `_volatile` flag which indicates that the resource being tracked is modified every time it is used under a new sequence number. This is used to reduce the time spent reprotecting memory for tracking writes to commonly modified buffers, like constant buffers. This optimisation works by detecting if a buffer is modified every time a check happens. If a buffer is checked but it is not dirty, then that data is likely not modified every sequence number, and should use memory protection for write tracking. If the opposite is the case all the time, it is faster to just assume it's dirty as we'd just be wasting time protecting the memory. The new MultiRegionBitmap could not notify handles that they had been checked as part of the fast bitmap lookup, so bindings larger than 4096 bytes wouldn't trigger it at all. This meant that they would be subject to a ton of reprotection if they were modified often. This does mean there are two separate sources for a _volatile set: VolatileOrDirty + _checkCount, and the bitmap check. These shouldn't interfere with each other, though. This fixes performance regressions from #3775 in Pokemon Sword, and hopefully Yu-Gi-Oh! RUSH DUEL: Dawn of the Battle Royale. May affect other games. * Fix stupid mistake
408 lines
14 KiB
C#
408 lines
14 KiB
C#
using System;
|
|
using System.Collections.Generic;
|
|
using System.Numerics;
|
|
using System.Runtime.CompilerServices;
|
|
using System.Threading;
|
|
|
|
namespace Ryujinx.Memory.Tracking
|
|
{
|
|
/// <summary>
|
|
/// A region handle that tracks a large region using many smaller handles, to provide
|
|
/// granular tracking that can be used to track partial updates. Backed by a bitmap
|
|
/// to improve performance when scanning large regions.
|
|
/// </summary>
|
|
public class MultiRegionHandle : IMultiRegionHandle
|
|
{
|
|
/// <summary>
|
|
/// A list of region handles for each granularity sized chunk of the whole region.
|
|
/// </summary>
|
|
private readonly RegionHandle[] _handles;
|
|
private readonly ulong Address;
|
|
private readonly ulong Granularity;
|
|
private readonly ulong Size;
|
|
|
|
private ConcurrentBitmap _dirtyBitmap;
|
|
|
|
private int _sequenceNumber;
|
|
private BitMap _sequenceNumberBitmap;
|
|
private BitMap _dirtyCheckedBitmap;
|
|
private int _uncheckedHandles;
|
|
|
|
public bool Dirty { get; private set; } = true;
|
|
|
|
internal MultiRegionHandle(MemoryTracking tracking, ulong address, ulong size, IEnumerable<IRegionHandle> handles, ulong granularity)
|
|
{
|
|
_handles = new RegionHandle[size / granularity];
|
|
Granularity = granularity;
|
|
|
|
_dirtyBitmap = new ConcurrentBitmap(_handles.Length, true);
|
|
_sequenceNumberBitmap = new BitMap(_handles.Length);
|
|
_dirtyCheckedBitmap = new BitMap(_handles.Length);
|
|
|
|
int i = 0;
|
|
|
|
if (handles != null)
|
|
{
|
|
// Inherit from the handles we were given. Any gaps must be filled with new handles,
|
|
// and old handles larger than our granularity must copy their state onto new granular handles and dispose.
|
|
// It is assumed that the provided handles do not overlap, in order, are on page boundaries,
|
|
// and don't extend past the requested range.
|
|
|
|
foreach (RegionHandle handle in handles)
|
|
{
|
|
int startIndex = (int)((handle.Address - address) / granularity);
|
|
|
|
// Fill any gap left before this handle.
|
|
while (i < startIndex)
|
|
{
|
|
RegionHandle fillHandle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
|
|
fillHandle.Parent = this;
|
|
_handles[i++] = fillHandle;
|
|
}
|
|
|
|
lock (tracking.TrackingLock)
|
|
{
|
|
if (handle is RegionHandle bitHandle && handle.Size == granularity)
|
|
{
|
|
handle.Parent = this;
|
|
|
|
bitHandle.ReplaceBitmap(_dirtyBitmap, i);
|
|
|
|
_handles[i++] = bitHandle;
|
|
}
|
|
else
|
|
{
|
|
int endIndex = (int)((handle.EndAddress - address) / granularity);
|
|
|
|
while (i < endIndex)
|
|
{
|
|
RegionHandle splitHandle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
|
|
splitHandle.Parent = this;
|
|
|
|
splitHandle.Reprotect(handle.Dirty);
|
|
|
|
RegionSignal signal = handle.PreAction;
|
|
if (signal != null)
|
|
{
|
|
splitHandle.RegisterAction(signal);
|
|
}
|
|
|
|
_handles[i++] = splitHandle;
|
|
}
|
|
|
|
handle.Dispose();
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Fill any remaining space with new handles.
|
|
while (i < _handles.Length)
|
|
{
|
|
RegionHandle handle = tracking.BeginTrackingBitmap(address + (ulong)i * granularity, granularity, _dirtyBitmap, i);
|
|
handle.Parent = this;
|
|
_handles[i++] = handle;
|
|
}
|
|
|
|
_uncheckedHandles = _handles.Length;
|
|
|
|
Address = address;
|
|
Size = size;
|
|
}
|
|
|
|
public void SignalWrite()
|
|
{
|
|
Dirty = true;
|
|
}
|
|
|
|
public IEnumerable<RegionHandle> GetHandles()
|
|
{
|
|
return _handles;
|
|
}
|
|
|
|
public void ForceDirty(ulong address, ulong size)
|
|
{
|
|
Dirty = true;
|
|
|
|
int startHandle = (int)((address - Address) / Granularity);
|
|
int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
|
|
|
|
for (int i = startHandle; i <= lastHandle; i++)
|
|
{
|
|
if (_sequenceNumberBitmap.Clear(i))
|
|
{
|
|
_uncheckedHandles++;
|
|
}
|
|
|
|
_handles[i].ForceDirty();
|
|
}
|
|
}
|
|
|
|
public void QueryModified(Action<ulong, ulong> modifiedAction)
|
|
{
|
|
if (!Dirty)
|
|
{
|
|
return;
|
|
}
|
|
|
|
Dirty = false;
|
|
|
|
QueryModified(Address, Size, modifiedAction);
|
|
}
|
|
|
|
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
|
private void ParseDirtyBits(long dirtyBits, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
|
|
{
|
|
while (dirtyBits != 0)
|
|
{
|
|
int bit = BitOperations.TrailingZeroCount(dirtyBits);
|
|
|
|
dirtyBits &= ~(1L << bit);
|
|
|
|
int handleIndex = baseBit + bit;
|
|
|
|
RegionHandle handle = _handles[handleIndex];
|
|
|
|
if (handleIndex != prevHandle + 1)
|
|
{
|
|
// Submit handles scanned until the gap as dirty
|
|
if (rgSize != 0)
|
|
{
|
|
modifiedAction(rgStart, rgSize);
|
|
rgSize = 0;
|
|
}
|
|
rgStart = handle.Address;
|
|
}
|
|
|
|
if (handle.Dirty)
|
|
{
|
|
rgSize += handle.Size;
|
|
handle.Reprotect();
|
|
}
|
|
|
|
prevHandle = handleIndex;
|
|
}
|
|
|
|
baseBit += ConcurrentBitmap.IntSize;
|
|
}
|
|
|
|
public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction)
|
|
{
|
|
int startHandle = (int)((address - Address) / Granularity);
|
|
int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
|
|
|
|
ulong rgStart = _handles[startHandle].Address;
|
|
|
|
if (startHandle == lastHandle)
|
|
{
|
|
RegionHandle handle = _handles[startHandle];
|
|
|
|
if (handle.Dirty)
|
|
{
|
|
handle.Reprotect();
|
|
modifiedAction(rgStart, handle.Size);
|
|
}
|
|
|
|
return;
|
|
}
|
|
|
|
ulong rgSize = 0;
|
|
|
|
long[] masks = _dirtyBitmap.Masks;
|
|
|
|
int startIndex = startHandle >> ConcurrentBitmap.IntShift;
|
|
int startBit = startHandle & ConcurrentBitmap.IntMask;
|
|
long startMask = -1L << startBit;
|
|
|
|
int endIndex = lastHandle >> ConcurrentBitmap.IntShift;
|
|
int endBit = lastHandle & ConcurrentBitmap.IntMask;
|
|
long endMask = (long)(ulong.MaxValue >> (ConcurrentBitmap.IntMask - endBit));
|
|
|
|
long startValue = Volatile.Read(ref masks[startIndex]);
|
|
|
|
int baseBit = startIndex << ConcurrentBitmap.IntShift;
|
|
int prevHandle = startHandle - 1;
|
|
|
|
if (startIndex == endIndex)
|
|
{
|
|
ParseDirtyBits(startValue & startMask & endMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
else
|
|
{
|
|
ParseDirtyBits(startValue & startMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
|
|
for (int i = startIndex + 1; i < endIndex; i++)
|
|
{
|
|
ParseDirtyBits(Volatile.Read(ref masks[i]), ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
|
|
long endValue = Volatile.Read(ref masks[endIndex]);
|
|
|
|
ParseDirtyBits(endValue & endMask, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
|
|
if (rgSize != 0)
|
|
{
|
|
modifiedAction(rgStart, rgSize);
|
|
}
|
|
}
|
|
|
|
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
|
private void ParseDirtyBits(long dirtyBits, long mask, int index, long[] seqMasks, long[] checkMasks, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
|
|
{
|
|
long seqMask = mask & ~seqMasks[index];
|
|
long checkMask = (~dirtyBits) & seqMask;
|
|
dirtyBits &= seqMask;
|
|
|
|
while (dirtyBits != 0)
|
|
{
|
|
int bit = BitOperations.TrailingZeroCount(dirtyBits);
|
|
long bitValue = 1L << bit;
|
|
|
|
dirtyBits &= ~bitValue;
|
|
|
|
int handleIndex = baseBit + bit;
|
|
|
|
RegionHandle handle = _handles[handleIndex];
|
|
|
|
if (handleIndex != prevHandle + 1)
|
|
{
|
|
// Submit handles scanned until the gap as dirty
|
|
if (rgSize != 0)
|
|
{
|
|
modifiedAction(rgStart, rgSize);
|
|
rgSize = 0;
|
|
}
|
|
rgStart = handle.Address;
|
|
}
|
|
|
|
rgSize += handle.Size;
|
|
handle.Reprotect(false, (checkMasks[index] & bitValue) == 0);
|
|
|
|
checkMasks[index] &= ~bitValue;
|
|
|
|
prevHandle = handleIndex;
|
|
}
|
|
|
|
checkMasks[index] |= checkMask;
|
|
seqMasks[index] |= mask;
|
|
_uncheckedHandles -= BitOperations.PopCount((ulong)seqMask);
|
|
|
|
baseBit += ConcurrentBitmap.IntSize;
|
|
}
|
|
|
|
public void QueryModified(ulong address, ulong size, Action<ulong, ulong> modifiedAction, int sequenceNumber)
|
|
{
|
|
int startHandle = (int)((address - Address) / Granularity);
|
|
int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
|
|
|
|
ulong rgStart = Address + (ulong)startHandle * Granularity;
|
|
|
|
if (sequenceNumber != _sequenceNumber)
|
|
{
|
|
if (_uncheckedHandles != _handles.Length)
|
|
{
|
|
_sequenceNumberBitmap.Clear();
|
|
_uncheckedHandles = _handles.Length;
|
|
}
|
|
|
|
_sequenceNumber = sequenceNumber;
|
|
}
|
|
|
|
if (startHandle == lastHandle)
|
|
{
|
|
var handle = _handles[startHandle];
|
|
if (_sequenceNumberBitmap.Set(startHandle))
|
|
{
|
|
_uncheckedHandles--;
|
|
|
|
if (handle.DirtyOrVolatile())
|
|
{
|
|
handle.Reprotect();
|
|
|
|
modifiedAction(rgStart, handle.Size);
|
|
}
|
|
}
|
|
|
|
return;
|
|
}
|
|
|
|
if (_uncheckedHandles == 0)
|
|
{
|
|
return;
|
|
}
|
|
|
|
ulong rgSize = 0;
|
|
|
|
long[] seqMasks = _sequenceNumberBitmap.Masks;
|
|
long[] checkedMasks = _dirtyCheckedBitmap.Masks;
|
|
long[] masks = _dirtyBitmap.Masks;
|
|
|
|
int startIndex = startHandle >> ConcurrentBitmap.IntShift;
|
|
int startBit = startHandle & ConcurrentBitmap.IntMask;
|
|
long startMask = -1L << startBit;
|
|
|
|
int endIndex = lastHandle >> ConcurrentBitmap.IntShift;
|
|
int endBit = lastHandle & ConcurrentBitmap.IntMask;
|
|
long endMask = (long)(ulong.MaxValue >> (ConcurrentBitmap.IntMask - endBit));
|
|
|
|
long startValue = Volatile.Read(ref masks[startIndex]);
|
|
|
|
int baseBit = startIndex << ConcurrentBitmap.IntShift;
|
|
int prevHandle = startHandle - 1;
|
|
|
|
if (startIndex == endIndex)
|
|
{
|
|
ParseDirtyBits(startValue, startMask & endMask, startIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
else
|
|
{
|
|
ParseDirtyBits(startValue, startMask, startIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
|
|
for (int i = startIndex + 1; i < endIndex; i++)
|
|
{
|
|
ParseDirtyBits(Volatile.Read(ref masks[i]), -1L, i, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
|
|
long endValue = Volatile.Read(ref masks[endIndex]);
|
|
|
|
ParseDirtyBits(endValue, endMask, endIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
|
|
}
|
|
|
|
if (rgSize != 0)
|
|
{
|
|
modifiedAction(rgStart, rgSize);
|
|
}
|
|
}
|
|
|
|
public void RegisterAction(ulong address, ulong size, RegionSignal action)
|
|
{
|
|
int startHandle = (int)((address - Address) / Granularity);
|
|
int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
|
|
|
|
for (int i = startHandle; i <= lastHandle; i++)
|
|
{
|
|
_handles[i].RegisterAction(action);
|
|
}
|
|
}
|
|
|
|
public void RegisterPreciseAction(ulong address, ulong size, PreciseRegionSignal action)
|
|
{
|
|
int startHandle = (int)((address - Address) / Granularity);
|
|
int lastHandle = (int)((address + (size - 1) - Address) / Granularity);
|
|
|
|
for (int i = startHandle; i <= lastHandle; i++)
|
|
{
|
|
_handles[i].RegisterPreciseAction(action);
|
|
}
|
|
}
|
|
|
|
public void Dispose()
|
|
{
|
|
foreach (var handle in _handles)
|
|
{
|
|
handle.Dispose();
|
|
}
|
|
}
|
|
}
|
|
}
|