ryujinx/Ryujinx.Graphics.Gpu/Memory/PhysicalMemory.cs
riperiperi 7c5ead1c19
Fast path for Inline2Memory buffer write that skips write tracking (#2624)
* Fast path for Inline2Memory buffer write

This PR adds a method to PhysicalMemory that attempts to write all cached resources directly, so that memory tracking can be avoided. The goal of this is both to avoid flushing buffer data, and to avoid raising the sequence number when data is written, which causes buffer and texture handles to be re-checked.

This currently only targets buffers, with a side check on textures that falls back to a tracked write if any exist within the target range. It's not expected to write textures from here - this is just a mechanism to protect us if someone does decide to do that. It's possible to add a fast path for this in future (and for ShaderCache, once that starts using tracking)

The forced read before inline2memory begins has been skipped, as the data is fully written when the transfer is completed anyways. This allows us to flush on read in emergency situations, but still write the new data over the flushed data.

Improves performance on Xenoblade 2 and DE, which was flushing buffer data on the GPU thread when trying to write compute data. May improve performance in other games that write SSBOs from compute, and update data in the same/nearby pages often.

Super Smash Bros Ultimate should probably be tested to make sure the vertex explosions haven't returned, as I think that's what this AdvanceSequence was for.

* ForceDirty before write, to make sure data does not flush over the new write
2021-09-19 15:09:53 +02:00

326 lines
13 KiB
C#

using Ryujinx.Cpu;
using Ryujinx.Cpu.Tracking;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Gpu.Shader;
using Ryujinx.Memory;
using Ryujinx.Memory.Range;
using Ryujinx.Memory.Tracking;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Threading;
namespace Ryujinx.Graphics.Gpu.Memory
{
/// <summary>
/// Represents physical memory, accessible from the GPU.
/// This is actually working CPU virtual addresses, of memory mapped on the application process.
/// </summary>
class PhysicalMemory : IDisposable
{
public const int PageSize = 0x1000;
private readonly GpuContext _context;
private IVirtualMemoryManagerTracked _cpuMemory;
private int _referenceCount;
/// <summary>
/// In-memory shader cache.
/// </summary>
public ShaderCache ShaderCache { get; }
/// <summary>
/// GPU buffer manager.
/// </summary>
public BufferCache BufferCache { get; }
/// <summary>
/// GPU texture manager.
/// </summary>
public TextureCache TextureCache { get; }
/// <summary>
/// Creates a new instance of the physical memory.
/// </summary>
/// <param name="context">GPU context that the physical memory belongs to</param>
/// <param name="cpuMemory">CPU memory manager of the application process</param>
public PhysicalMemory(GpuContext context, IVirtualMemoryManagerTracked cpuMemory)
{
_context = context;
_cpuMemory = cpuMemory;
ShaderCache = new ShaderCache(context);
BufferCache = new BufferCache(context, this);
TextureCache = new TextureCache(context, this);
if (cpuMemory is IRefCounted rc)
{
rc.IncrementReferenceCount();
}
_referenceCount = 1;
}
/// <summary>
/// Increments the memory reference count.
/// </summary>
public void IncrementReferenceCount()
{
Interlocked.Increment(ref _referenceCount);
}
/// <summary>
/// Decrements the memory reference count.
/// </summary>
public void DecrementReferenceCount()
{
if (Interlocked.Decrement(ref _referenceCount) == 0 && _cpuMemory is IRefCounted rc)
{
rc.DecrementReferenceCount();
}
}
/// <summary>
/// Write data to memory that is destined for a resource in a cache.
/// This avoids triggering write tracking when possible, which can avoid flushes and incrementing sequence number.
/// </summary>
/// <param name="memoryManager">The GPU memory manager</param>
/// <param name="gpuVa">GPU virtual address to write the data into</param>
/// <param name="data">The data to be written</param>
public void CacheResourceWrite(MemoryManager memoryManager, ulong gpuVa, ReadOnlySpan<byte> data)
{
if (TextureCache.IsTextureInRange(memoryManager, gpuVa, (ulong)data.Length))
{
// No fast path yet - copy the data back and trigger write tracking.
memoryManager.Write(gpuVa, data);
_context.AdvanceSequence();
}
else
{
BufferCache.ForceDirty(memoryManager, gpuVa, (ulong)data.Length);
memoryManager.WriteUntracked(gpuVa, data);
}
}
/// <summary>
/// Gets a span of data from the application process.
/// </summary>
/// <param name="address">Start address of the range</param>
/// <param name="size">Size in bytes to be range</param>
/// <param name="tracked">True if read tracking is triggered on the span</param>
/// <returns>A read only span of the data at the specified memory location</returns>
public ReadOnlySpan<byte> GetSpan(ulong address, int size, bool tracked = false)
{
return _cpuMemory.GetSpan(address, size, tracked);
}
/// <summary>
/// Gets a span of data from the application process.
/// </summary>
/// <param name="range">Ranges of physical memory where the data is located</param>
/// <param name="tracked">True if read tracking is triggered on the span</param>
/// <returns>A read only span of the data at the specified memory location</returns>
public ReadOnlySpan<byte> GetSpan(MultiRange range, bool tracked = false)
{
if (range.Count == 1)
{
var singleRange = range.GetSubRange(0);
return _cpuMemory.GetSpan(singleRange.Address, (int)singleRange.Size, tracked);
}
else
{
Span<byte> data = new byte[range.GetSize()];
int offset = 0;
for (int i = 0; i < range.Count; i++)
{
var currentRange = range.GetSubRange(i);
int size = (int)currentRange.Size;
_cpuMemory.GetSpan(currentRange.Address, size, tracked).CopyTo(data.Slice(offset, size));
offset += size;
}
return data;
}
}
/// <summary>
/// Gets a writable region from the application process.
/// </summary>
/// <param name="address">Start address of the range</param>
/// <param name="size">Size in bytes to be range</param>
/// <param name="tracked">True if write tracking is triggered on the span</param>
/// <returns>A writable region with the data at the specified memory location</returns>
public WritableRegion GetWritableRegion(ulong address, int size, bool tracked = false)
{
return _cpuMemory.GetWritableRegion(address, size, tracked);
}
/// <summary>
/// Reads data from the application process.
/// </summary>
/// <typeparam name="T">Type of the structure</typeparam>
/// <param name="address">Address to read from</param>
/// <returns>The data at the specified memory location</returns>
public T Read<T>(ulong address) where T : unmanaged
{
return _cpuMemory.Read<T>(address);
}
/// <summary>
/// Reads data from the application process, with write tracking.
/// </summary>
/// <typeparam name="T">Type of the structure</typeparam>
/// <param name="address">Address to read from</param>
/// <returns>The data at the specified memory location</returns>
public T ReadTracked<T>(ulong address) where T : unmanaged
{
return _cpuMemory.ReadTracked<T>(address);
}
/// <summary>
/// Writes data to the application process.
/// </summary>
/// <param name="address">Address to write into</param>
/// <param name="data">Data to be written</param>
public void Write(ulong address, ReadOnlySpan<byte> data)
{
_cpuMemory.Write(address, data);
}
/// <summary>
/// Writes data to the application process.
/// </summary>
/// <param name="range">Ranges of physical memory where the data is located</param>
/// <param name="data">Data to be written</param>
public void Write(MultiRange range, ReadOnlySpan<byte> data)
{
WriteImpl(range, data, _cpuMemory.Write);
}
/// <summary>
/// Writes data to the application process, without any tracking.
/// </summary>
/// <param name="address">Address to write into</param>
/// <param name="data">Data to be written</param>
public void WriteUntracked(ulong address, ReadOnlySpan<byte> data)
{
_cpuMemory.WriteUntracked(address, data);
}
/// <summary>
/// Writes data to the application process, without any tracking.
/// </summary>
/// <param name="range">Ranges of physical memory where the data is located</param>
/// <param name="data">Data to be written</param>
public void WriteUntracked(MultiRange range, ReadOnlySpan<byte> data)
{
WriteImpl(range, data, _cpuMemory.WriteUntracked);
}
private delegate void WriteCallback(ulong address, ReadOnlySpan<byte> data);
/// <summary>
/// Writes data to the application process, using the supplied callback method.
/// </summary>
/// <param name="range">Ranges of physical memory where the data is located</param>
/// <param name="data">Data to be written</param>
/// <param name="writeCallback">Callback method that will perform the write</param>
private static void WriteImpl(MultiRange range, ReadOnlySpan<byte> data, WriteCallback writeCallback)
{
if (range.Count == 1)
{
var singleRange = range.GetSubRange(0);
writeCallback(singleRange.Address, data);
}
else
{
int offset = 0;
for (int i = 0; i < range.Count; i++)
{
var currentRange = range.GetSubRange(i);
int size = (int)currentRange.Size;
writeCallback(currentRange.Address, data.Slice(offset, size));
offset += size;
}
}
}
/// <summary>
/// Obtains a memory tracking handle for the given virtual region. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <returns>The memory tracking handle</returns>
public CpuRegionHandle BeginTracking(ulong address, ulong size)
{
return _cpuMemory.BeginTracking(address, size);
}
/// <summary>
/// Obtains a memory tracking handle for the given virtual region. This should be disposed when finished with.
/// </summary>
/// <param name="range">Ranges of physical memory where the data is located</param>
/// <returns>The memory tracking handle</returns>
public GpuRegionHandle BeginTracking(MultiRange range)
{
var cpuRegionHandles = new CpuRegionHandle[range.Count];
for (int i = 0; i < range.Count; i++)
{
var currentRange = range.GetSubRange(i);
cpuRegionHandles[i] = _cpuMemory.BeginTracking(currentRange.Address, currentRange.Size);
}
return new GpuRegionHandle(cpuRegionHandles);
}
/// <summary>
/// Obtains a memory tracking handle for the given virtual region, with a specified granularity. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <param name="handles">Handles to inherit state from or reuse</param>
/// <param name="granularity">Desired granularity of write tracking</param>
/// <returns>The memory tracking handle</returns>
public CpuMultiRegionHandle BeginGranularTracking(ulong address, ulong size, IEnumerable<IRegionHandle> handles = null, ulong granularity = 4096)
{
return _cpuMemory.BeginGranularTracking(address, size, handles, granularity);
}
/// <summary>
/// Obtains a smart memory tracking handle for the given virtual region, with a specified granularity. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <param name="granularity">Desired granularity of write tracking</param>
/// <returns>The memory tracking handle</returns>
public CpuSmartMultiRegionHandle BeginSmartGranularTracking(ulong address, ulong size, ulong granularity = 4096)
{
return _cpuMemory.BeginSmartGranularTracking(address, size, granularity);
}
/// <summary>
/// Release our reference to the CPU memory manager.
/// </summary>
public void Dispose()
{
_context.DeferredActions.Enqueue(Destroy);
}
/// <summary>
/// Performs disposal of the host GPU caches with resources mapped on this physical memory.
/// This must only be called from the render thread.
/// </summary>
private void Destroy()
{
ShaderCache.Dispose();
BufferCache.Dispose();
TextureCache.Dispose();
DecrementReferenceCount();
}
}
}