GPU: Pre-emptively flush textures that are flushed often (to imported memory when available) (#4711)

* WIP texture pre-flush

Improve performance of TextureView GetData to buffer

Fix copy/sync ordering

Fix minor bug

Make this actually work

WIP host mapping stuff

* Fix usage flags

* message

* Cleanup 1

* Fix rebase

* Fix

* Improve pre-flush rules

* Fix pre-flush

* A lot of cleanup

* Use the host memory bits

* Select the correct memory type

* Cleanup TextureGroupHandle

* Missing comment

* Remove debugging logs

* Revert BufferHandle _value access modifier

* One interrupt action at a time.

* Support D32S8 to D24S8 conversion, safeguards

* Interrupt cannot happen in sync handle's lock

Waitable needs to be checked twice now, but this should stop it from deadlocking.

* Remove unused using

* Address some feedback

* Address feedback

* Address more feedback

* Address more feedback

* Improve sync rules

Should allow for faster sync in some cases.
This commit is contained in:
riperiperi 2023-05-01 20:05:12 +01:00 committed by GitHub
parent 36f10df775
commit e18d258fa0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
40 changed files with 1328 additions and 79 deletions

View File

@ -0,0 +1,8 @@
namespace Ryujinx.Graphics.GAL
{
public enum BufferAccess
{
Default,
FlushPersistent
}
}

View File

@ -21,11 +21,14 @@ namespace Ryujinx.Graphics.GAL
{
return CreateBuffer(size, BufferHandle.Null);
}
BufferHandle CreateBuffer(nint pointer, int size);
BufferHandle CreateBuffer(int size, BufferAccess access);
IProgram CreateProgram(ShaderSource[] shaders, ShaderInfo info);
ISampler CreateSampler(SamplerCreateInfo info);
ITexture CreateTexture(TextureCreateInfo info, float scale);
bool PrepareHostMapping(nint address, ulong size);
void CreateSync(ulong id, bool strict);

View File

@ -12,6 +12,7 @@ namespace Ryujinx.Graphics.GAL
void CopyTo(ITexture destination, int firstLayer, int firstLevel);
void CopyTo(ITexture destination, int srcLayer, int dstLayer, int srcLevel, int dstLevel);
void CopyTo(ITexture destination, Extents2D srcRegion, Extents2D dstRegion, bool linearFilter);
void CopyTo(BufferRange range, int layer, int level, int stride);
ITexture CreateView(TextureCreateInfo info, int firstLayer, int firstLevel);

View File

@ -43,6 +43,8 @@ namespace Ryujinx.Graphics.GAL.Multithreading
Register<ActionCommand>(CommandType.Action);
Register<CreateBufferCommand>(CommandType.CreateBuffer);
Register<CreateBufferAccessCommand>(CommandType.CreateBufferAccess);
Register<CreateHostBufferCommand>(CommandType.CreateHostBuffer);
Register<CreateProgramCommand>(CommandType.CreateProgram);
Register<CreateSamplerCommand>(CommandType.CreateSampler);
Register<CreateSyncCommand>(CommandType.CreateSync);
@ -69,6 +71,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
Register<TextureCopyToCommand>(CommandType.TextureCopyTo);
Register<TextureCopyToScaledCommand>(CommandType.TextureCopyToScaled);
Register<TextureCopyToSliceCommand>(CommandType.TextureCopyToSlice);
Register<TextureCopyToBufferCommand>(CommandType.TextureCopyToBuffer);
Register<TextureCreateViewCommand>(CommandType.TextureCreateView);
Register<TextureGetDataCommand>(CommandType.TextureGetData);
Register<TextureGetDataSliceCommand>(CommandType.TextureGetDataSlice);

View File

@ -4,6 +4,8 @@
{
Action,
CreateBuffer,
CreateBufferAccess,
CreateHostBuffer,
CreateProgram,
CreateSampler,
CreateSync,
@ -29,6 +31,7 @@
SamplerDispose,
TextureCopyTo,
TextureCopyToBuffer,
TextureCopyToScaled,
TextureCopyToSlice,
TextureCreateView,

View File

@ -0,0 +1,22 @@
namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Renderer
{
struct CreateBufferAccessCommand : IGALCommand, IGALCommand<CreateBufferAccessCommand>
{
public CommandType CommandType => CommandType.CreateBufferAccess;
private BufferHandle _threadedHandle;
private int _size;
private BufferAccess _access;
public void Set(BufferHandle threadedHandle, int size, BufferAccess access)
{
_threadedHandle = threadedHandle;
_size = size;
_access = access;
}
public static void Run(ref CreateBufferAccessCommand command, ThreadedRenderer threaded, IRenderer renderer)
{
threaded.Buffers.AssignBuffer(command._threadedHandle, renderer.CreateBuffer(command._size, command._access));
}
}
}

View File

@ -0,0 +1,22 @@
namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Renderer
{
struct CreateHostBufferCommand : IGALCommand, IGALCommand<CreateHostBufferCommand>
{
public CommandType CommandType => CommandType.CreateHostBuffer;
private BufferHandle _threadedHandle;
private nint _pointer;
private int _size;
public void Set(BufferHandle threadedHandle, nint pointer, int size)
{
_threadedHandle = threadedHandle;
_pointer = pointer;
_size = size;
}
public static void Run(ref CreateHostBufferCommand command, ThreadedRenderer threaded, IRenderer renderer)
{
threaded.Buffers.AssignBuffer(command._threadedHandle, renderer.CreateBuffer(command._pointer, command._size));
}
}
}

View File

@ -0,0 +1,29 @@
using Ryujinx.Graphics.GAL.Multithreading.Model;
using Ryujinx.Graphics.GAL.Multithreading.Resources;
namespace Ryujinx.Graphics.GAL.Multithreading.Commands.Texture
{
struct TextureCopyToBufferCommand : IGALCommand, IGALCommand<TextureCopyToBufferCommand>
{
public CommandType CommandType => CommandType.TextureCopyToBuffer;
private TableRef<ThreadedTexture> _texture;
private BufferRange _range;
private int _layer;
private int _level;
private int _stride;
public void Set(TableRef<ThreadedTexture> texture, BufferRange range, int layer, int level, int stride)
{
_texture = texture;
_range = range;
_layer = layer;
_level = level;
_stride = stride;
}
public static void Run(ref TextureCopyToBufferCommand command, ThreadedRenderer threaded, IRenderer renderer)
{
command._texture.Get(threaded).Base.CopyTo(threaded.Buffers.MapBufferRange(command._range), command._layer, command._level, command._stride);
}
}
}

View File

@ -108,6 +108,12 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
}
}
public void CopyTo(BufferRange range, int layer, int level, int stride)
{
_renderer.New<TextureCopyToBufferCommand>().Set(Ref(this), range, layer, level, stride);
_renderer.QueueCommand();
}
public void SetData(SpanOrArray<byte> data)
{
_renderer.New<TextureSetDataCommand>().Set(Ref(this), Ref(data.ToArray()));

View File

@ -57,6 +57,7 @@ namespace Ryujinx.Graphics.GAL.Multithreading
private int _refConsumerPtr;
private Action _interruptAction;
private object _interruptLock = new();
public event EventHandler<ScreenCaptureImageInfo> ScreenCaptured;
@ -274,6 +275,24 @@ namespace Ryujinx.Graphics.GAL.Multithreading
return handle;
}
public BufferHandle CreateBuffer(nint pointer, int size)
{
BufferHandle handle = Buffers.CreateBufferHandle();
New<CreateHostBufferCommand>().Set(handle, pointer, size);
QueueCommand();
return handle;
}
public BufferHandle CreateBuffer(int size, BufferAccess access)
{
BufferHandle handle = Buffers.CreateBufferHandle();
New<CreateBufferAccessCommand>().Set(handle, size, access);
QueueCommand();
return handle;
}
public IProgram CreateProgram(ShaderSource[] shaders, ShaderInfo info)
{
var program = new ThreadedProgram(this);
@ -448,11 +467,14 @@ namespace Ryujinx.Graphics.GAL.Multithreading
}
else
{
while (Interlocked.CompareExchange(ref _interruptAction, action, null) != null) { }
lock (_interruptLock)
{
while (Interlocked.CompareExchange(ref _interruptAction, action, null) != null) { }
_galWorkAvailable.Set();
_galWorkAvailable.Set();
_interruptRun.WaitOne();
_interruptRun.WaitOne();
}
}
}
@ -461,6 +483,11 @@ namespace Ryujinx.Graphics.GAL.Multithreading
// Threaded renderer ignores given interrupt action, as it provides its own to the child renderer.
}
public bool PrepareHostMapping(nint address, ulong size)
{
return _baseRenderer.PrepareHostMapping(address, size);
}
public void Dispose()
{
// Dispose must happen from the render thread, after all commands have completed.

View File

@ -1,5 +1,6 @@
using Ryujinx.Graphics.Device;
using Ryujinx.Graphics.Gpu.Engine.MME;
using Ryujinx.Graphics.Gpu.Synchronization;
using System;
using System.Collections.Generic;
using System.Threading;
@ -59,7 +60,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
if (_createSyncPending)
{
_createSyncPending = false;
_context.CreateHostSyncIfNeeded(false, false);
_context.CreateHostSyncIfNeeded(HostSyncFlags.None);
}
}
@ -157,7 +158,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
}
else if (operation == SyncpointbOperation.Incr)
{
_context.CreateHostSyncIfNeeded(true, true);
_context.CreateHostSyncIfNeeded(HostSyncFlags.StrictSyncpoint);
_context.Synchronization.IncrementSyncpoint(syncpointId);
}
@ -184,7 +185,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
{
_context.Renderer.Pipeline.CommandBufferBarrier();
_context.CreateHostSyncIfNeeded(false, true);
_context.CreateHostSyncIfNeeded(HostSyncFlags.Strict);
}
/// <summary>

View File

@ -3,6 +3,7 @@ using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Engine.GPFifo;
using Ryujinx.Graphics.Gpu.Engine.InlineToMemory;
using Ryujinx.Graphics.Gpu.Engine.Threed.Blender;
using Ryujinx.Graphics.Gpu.Synchronization;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
@ -254,7 +255,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
uint syncpointId = (uint)argument & 0xFFFF;
_context.AdvanceSequence();
_context.CreateHostSyncIfNeeded(true, true);
_context.CreateHostSyncIfNeeded(HostSyncFlags.StrictSyncpoint);
_context.Renderer.UpdateCounters(); // Poll the query counters, the game may want an updated result.
_context.Synchronization.IncrementSyncpoint(syncpointId);
}

View File

@ -60,14 +60,14 @@ namespace Ryujinx.Graphics.Gpu
/// If there are more than 0 items when this happens, a host sync object will be generated for the given <see cref="SyncNumber"/>,
/// and the SyncNumber will be incremented.
/// </summary>
internal List<Action> SyncActions { get; }
internal List<ISyncActionHandler> SyncActions { get; }
/// <summary>
/// Actions to be performed when a CPU waiting syncpoint is triggered.
/// If there are more than 0 items when this happens, a host sync object will be generated for the given <see cref="SyncNumber"/>,
/// and the SyncNumber will be incremented.
/// </summary>
internal List<Action> SyncpointActions { get; }
internal List<ISyncActionHandler> SyncpointActions { get; }
/// <summary>
/// Buffer migrations that are currently in-flight. These are checked whenever sync is created to determine if buffer migration
@ -114,8 +114,8 @@ namespace Ryujinx.Graphics.Gpu
HostInitalized = new ManualResetEvent(false);
SyncActions = new List<Action>();
SyncpointActions = new List<Action>();
SyncActions = new List<ISyncActionHandler>();
SyncpointActions = new List<ISyncActionHandler>();
BufferMigrations = new List<BufferMigration>();
DeferredActions = new Queue<Action>();
@ -296,9 +296,9 @@ namespace Ryujinx.Graphics.Gpu
/// Registers an action to be performed the next time a syncpoint is incremented.
/// This will also ensure a host sync object is created, and <see cref="SyncNumber"/> is incremented.
/// </summary>
/// <param name="action">The action to be performed on sync object creation</param>
/// <param name="action">The resource with action to be performed on sync object creation</param>
/// <param name="syncpointOnly">True if the sync action should only run when syncpoints are incremented</param>
public void RegisterSyncAction(Action action, bool syncpointOnly = false)
internal void RegisterSyncAction(ISyncActionHandler action, bool syncpointOnly = false)
{
if (syncpointOnly)
{
@ -315,10 +315,13 @@ namespace Ryujinx.Graphics.Gpu
/// Creates a host sync object if there are any pending sync actions. The actions will then be called.
/// If no actions are present, a host sync object is not created.
/// </summary>
/// <param name="syncpoint">True if host sync is being created by a syncpoint</param>
/// <param name="strict">True if the sync should signal as soon as possible</param>
public void CreateHostSyncIfNeeded(bool syncpoint, bool strict)
/// <param name="flags">Modifiers for how host sync should be created</param>
internal void CreateHostSyncIfNeeded(HostSyncFlags flags)
{
bool syncpoint = flags.HasFlag(HostSyncFlags.Syncpoint);
bool strict = flags.HasFlag(HostSyncFlags.Strict);
bool force = flags.HasFlag(HostSyncFlags.Force);
if (BufferMigrations.Count > 0)
{
ulong currentSyncNumber = Renderer.GetCurrentSync();
@ -336,24 +339,17 @@ namespace Ryujinx.Graphics.Gpu
}
}
if (_pendingSync || (syncpoint && SyncpointActions.Count > 0))
if (force || _pendingSync || (syncpoint && SyncpointActions.Count > 0))
{
Renderer.CreateSync(SyncNumber, strict);
SyncActions.ForEach(action => action.SyncPreAction(syncpoint));
SyncpointActions.ForEach(action => action.SyncPreAction(syncpoint));
SyncNumber++;
foreach (Action action in SyncActions)
{
action();
}
foreach (Action action in SyncpointActions)
{
action();
}
SyncActions.Clear();
SyncpointActions.Clear();
SyncActions.RemoveAll(action => action.SyncAction(syncpoint));
SyncpointActions.RemoveAll(action => action.SyncAction(syncpoint));
}
_pendingSync = false;

View File

@ -1423,7 +1423,7 @@ namespace Ryujinx.Graphics.Gpu.Image
_scaledSetScore = Math.Max(0, _scaledSetScore - 1);
}
if (_modifiedStale || Group.HasCopyDependencies)
if (_modifiedStale || Group.HasCopyDependencies || Group.HasFlushBuffer)
{
_modifiedStale = false;
Group.SignalModifying(this, bound);
@ -1685,6 +1685,8 @@ namespace Ryujinx.Graphics.Gpu.Image
if (Group.Storage == this)
{
Group.Unmapped();
Group.ClearModified(unmapRange);
}
}

View File

@ -57,6 +57,12 @@ namespace Ryujinx.Graphics.Gpu.Image
/// </summary>
public bool HasCopyDependencies { get; set; }
/// <summary>
/// Indicates if the texture group has a pre-emptive flush buffer.
/// When one is present, the group must always be notified on unbind.
/// </summary>
public bool HasFlushBuffer => _flushBuffer != BufferHandle.Null;
/// <summary>
/// Indicates if this texture has any incompatible overlaps alive.
/// </summary>
@ -89,6 +95,10 @@ namespace Ryujinx.Graphics.Gpu.Image
private bool _incompatibleOverlapsDirty = true;
private bool _flushIncompatibleOverlaps;
private BufferHandle _flushBuffer;
private bool _flushBufferImported;
private bool _flushBufferInvalid;
/// <summary>
/// Create a new texture group.
/// </summary>
@ -464,8 +474,9 @@ namespace Ryujinx.Graphics.Gpu.Image
/// </remarks>
/// <param name="tracked">True if writing the texture data is tracked, false otherwise</param>
/// <param name="sliceIndex">The index of the slice to flush</param>
/// <param name="inBuffer">Whether the flushed texture data is up to date in the flush buffer</param>
/// <param name="texture">The specific host texture to flush. Defaults to the storage texture</param>
private void FlushTextureDataSliceToGuest(bool tracked, int sliceIndex, ITexture texture = null)
private void FlushTextureDataSliceToGuest(bool tracked, int sliceIndex, bool inBuffer, ITexture texture = null)
{
(int layer, int level) = GetLayerLevelForView(sliceIndex);
@ -475,7 +486,16 @@ namespace Ryujinx.Graphics.Gpu.Image
using WritableRegion region = _physicalMemory.GetWritableRegion(Storage.Range.Slice((ulong)offset, (ulong)size), tracked);
Storage.GetTextureDataSliceFromGpu(region.Memory.Span, layer, level, tracked, texture);
if (inBuffer)
{
using PinnedSpan<byte> data = _context.Renderer.GetBufferData(_flushBuffer, offset, size);
Storage.ConvertFromHostCompatibleFormat(region.Memory.Span, data.Get(), level, true);
}
else
{
Storage.GetTextureDataSliceFromGpu(region.Memory.Span, layer, level, tracked, texture);
}
}
/// <summary>
@ -484,12 +504,13 @@ namespace Ryujinx.Graphics.Gpu.Image
/// <param name="tracked">True if writing the texture data is tracked, false otherwise</param>
/// <param name="sliceStart">The first slice to flush</param>
/// <param name="sliceEnd">The slice to finish flushing on (exclusive)</param>
/// <param name="inBuffer">Whether the flushed texture data is up to date in the flush buffer</param>
/// <param name="texture">The specific host texture to flush. Defaults to the storage texture</param>
private void FlushSliceRange(bool tracked, int sliceStart, int sliceEnd, ITexture texture = null)
private void FlushSliceRange(bool tracked, int sliceStart, int sliceEnd, bool inBuffer, ITexture texture = null)
{
for (int i = sliceStart; i < sliceEnd; i++)
{
FlushTextureDataSliceToGuest(tracked, i, texture);
FlushTextureDataSliceToGuest(tracked, i, inBuffer, texture);
}
}
@ -520,7 +541,7 @@ namespace Ryujinx.Graphics.Gpu.Image
{
if (endSlice > startSlice)
{
FlushSliceRange(tracked, startSlice, endSlice);
FlushSliceRange(tracked, startSlice, endSlice, false);
flushed = true;
}
@ -553,7 +574,7 @@ namespace Ryujinx.Graphics.Gpu.Image
}
else
{
FlushSliceRange(tracked, startSlice, endSlice);
FlushSliceRange(tracked, startSlice, endSlice, false);
}
flushed = true;
@ -565,6 +586,58 @@ namespace Ryujinx.Graphics.Gpu.Image
return flushed;
}
/// <summary>
/// Flush the texture data into a persistently mapped buffer.
/// If the buffer does not exist, this method will create it.
/// </summary>
/// <param name="handle">Handle of the texture group to flush slices of</param>
public void FlushIntoBuffer(TextureGroupHandle handle)
{
// Ensure that the buffer exists.
if (_flushBufferInvalid && _flushBuffer != BufferHandle.Null)
{
_flushBufferInvalid = false;
_context.Renderer.DeleteBuffer(_flushBuffer);
_flushBuffer = BufferHandle.Null;
}
if (_flushBuffer == BufferHandle.Null)
{
if (!TextureCompatibility.CanTextureFlush(Storage.Info, _context.Capabilities))
{
return;
}
bool canImport = Storage.Info.IsLinear && Storage.Info.Stride >= Storage.Info.Width * Storage.Info.FormatInfo.BytesPerPixel;
var hostPointer = canImport ? _physicalMemory.GetHostPointer(Storage.Range) : 0;
if (hostPointer != 0 && _context.Renderer.PrepareHostMapping(hostPointer, Storage.Size))
{
_flushBuffer = _context.Renderer.CreateBuffer(hostPointer, (int)Storage.Size);
_flushBufferImported = true;
}
else
{
_flushBuffer = _context.Renderer.CreateBuffer((int)Storage.Size, BufferAccess.FlushPersistent);
_flushBufferImported = false;
}
Storage.BlacklistScale();
}
int sliceStart = handle.BaseSlice;
int sliceEnd = sliceStart + handle.SliceCount;
for (int i = sliceStart; i < sliceEnd; i++)
{
(int layer, int level) = GetLayerLevelForView(i);
Storage.GetFlushTexture().CopyTo(new BufferRange(_flushBuffer, _allOffsets[i], _sliceSizes[level]), layer, level, _flushBufferImported ? Storage.Info.Stride : 0);
}
}
/// <summary>
/// Clears competing modified flags for all incompatible ranges, if they have possibly been modified.
/// </summary>
@ -1570,10 +1643,7 @@ namespace Ryujinx.Graphics.Gpu.Image
_context.Renderer.BackgroundContextAction(() =>
{
if (!isGpuThread)
{
handle.Sync(_context);
}
bool inBuffer = !isGpuThread && handle.Sync(_context);
Storage.SignalModifiedDirty();
@ -1585,13 +1655,24 @@ namespace Ryujinx.Graphics.Gpu.Image
}
}
if (TextureCompatibility.CanTextureFlush(Storage.Info, _context.Capabilities))
if (TextureCompatibility.CanTextureFlush(Storage.Info, _context.Capabilities) && !(inBuffer && _flushBufferImported))
{
FlushSliceRange(false, handle.BaseSlice, handle.BaseSlice + handle.SliceCount, Storage.GetFlushTexture());
FlushSliceRange(false, handle.BaseSlice, handle.BaseSlice + handle.SliceCount, inBuffer, Storage.GetFlushTexture());
}
});
}
/// <summary>
/// Called if any part of the storage texture is unmapped.
/// </summary>
public void Unmapped()
{
if (_flushBufferImported)
{
_flushBufferInvalid = true;
}
}
/// <summary>
/// Dispose this texture group, disposing all related memory tracking handles.
/// </summary>
@ -1606,6 +1687,11 @@ namespace Ryujinx.Graphics.Gpu.Image
{
incompatible.Group._incompatibleOverlaps.RemoveAll(overlap => overlap.Group == this);
}
if (_flushBuffer != BufferHandle.Null)
{
_context.Renderer.DeleteBuffer(_flushBuffer);
}
}
}
}

View File

@ -1,4 +1,5 @@
using Ryujinx.Cpu.Tracking;
using Ryujinx.Graphics.Gpu.Synchronization;
using System;
using System.Collections.Generic;
using System.Linq;
@ -13,8 +14,14 @@ namespace Ryujinx.Graphics.Gpu.Image
/// Also tracks copy dependencies for the handle - references to other handles that must be kept
/// in sync with this one before use.
/// </summary>
class TextureGroupHandle : IDisposable
class TextureGroupHandle : ISyncActionHandler, IDisposable
{
private const int FlushBalanceIncrement = 6;
private const int FlushBalanceWriteCost = 1;
private const int FlushBalanceThreshold = 7;
private const int FlushBalanceMax = 60;
private const int FlushBalanceMin = -10;
private TextureGroup _group;
private int _bindCount;
private int _firstLevel;
@ -26,6 +33,8 @@ namespace Ryujinx.Graphics.Gpu.Image
/// The sync number last registered.
/// </summary>
private ulong _registeredSync;
private ulong _registeredBufferSync = ulong.MaxValue;
private ulong _registeredBufferGuestSync = ulong.MaxValue;
/// <summary>
/// The sync number when the texture was last modified by GPU.
@ -42,6 +51,12 @@ namespace Ryujinx.Graphics.Gpu.Image
/// </summary>
private bool _syncActionRegistered;
/// <summary>
/// Determines the balance of synced writes to flushes.
/// Used to determine if the texture should always write data to a persistent buffer for flush.
/// </summary>
private int _flushBalance;
/// <summary>
/// The byte offset from the start of the storage of this handle.
/// </summary>
@ -132,6 +147,12 @@ namespace Ryujinx.Graphics.Gpu.Image
}
Handles = handles;
if (group.Storage.Info.IsLinear)
{
// Linear textures are presumed to be used for readback initially.
_flushBalance = FlushBalanceThreshold + FlushBalanceIncrement;
}
}
/// <summary>
@ -159,6 +180,35 @@ namespace Ryujinx.Graphics.Gpu.Image
}
}
/// <summary>
/// Determine if the next sync will copy into the flush buffer.
/// </summary>
/// <returns>True if it will copy, false otherwise</returns>
private bool NextSyncCopies()
{
return _flushBalance - FlushBalanceWriteCost > FlushBalanceThreshold;
}
/// <summary>
/// Alters the flush balance by the given value. Should increase significantly with each sync, decrease with each write.
/// A flush balance higher than the threshold will cause a texture to repeatedly copy to a flush buffer on each use.
/// </summary>
/// <param name="modifier">Value to add to the existing flush balance</param>
/// <returns>True if the new balance is over the threshold, false otherwise</returns>
private bool ModifyFlushBalance(int modifier)
{
int result;
int existingBalance;
do
{
existingBalance = _flushBalance;
result = Math.Max(FlushBalanceMin, Math.Min(FlushBalanceMax, existingBalance + modifier));
}
while (Interlocked.CompareExchange(ref _flushBalance, result, existingBalance) != existingBalance);
return result > FlushBalanceThreshold;
}
/// <summary>
/// Adds a single texture view as an overlap if its range overlaps.
/// </summary>
@ -204,7 +254,7 @@ namespace Ryujinx.Graphics.Gpu.Image
if (!_syncActionRegistered)
{
_modifiedSync = context.SyncNumber;
context.RegisterSyncAction(SyncAction, true);
context.RegisterSyncAction(this, true);
_syncActionRegistered = true;
}
@ -241,6 +291,13 @@ namespace Ryujinx.Graphics.Gpu.Image
{
SignalModified(context);
if (!bound && _syncActionRegistered && NextSyncCopies())
{
// On unbind, textures that flush often should immediately create sync so their result can be obtained as soon as possible.
context.CreateHostSyncIfNeeded(HostSyncFlags.Force);
}
// Note: Bind count currently resets to 0 on inherit for safety, as the handle <-> view relationship can change.
_bindCount = Math.Max(0, _bindCount + (bound ? 1 : -1));
}
@ -266,25 +323,35 @@ namespace Ryujinx.Graphics.Gpu.Image
/// removing the modified flag if it was reached, or leaving it set if it has not yet been created.
/// </summary>
/// <param name="context">The GPU context used to wait for sync</param>
public void Sync(GpuContext context)
/// <returns>True if the texture data can be read from the flush buffer</returns>
public bool Sync(GpuContext context)
{
ulong registeredSync = _registeredSync;
long diff = (long)(context.SyncNumber - registeredSync);
// Currently assumes the calling thread is a guest thread.
bool inBuffer = _registeredBufferGuestSync != ulong.MaxValue;
ulong sync = inBuffer ? _registeredBufferGuestSync : _registeredSync;
long diff = (long)(context.SyncNumber - sync);
ModifyFlushBalance(FlushBalanceIncrement);
if (diff > 0)
{
context.Renderer.WaitSync(registeredSync);
context.Renderer.WaitSync(sync);
if ((long)(_modifiedSync - registeredSync) > 0)
if ((long)(_modifiedSync - sync) > 0)
{
// Flush the data in a previous state. Do not remove the modified flag - it will be removed to ignore following writes.
return;
return inBuffer;
}
Modified = false;
return inBuffer;
}
// If the difference is <= 0, no data is not ready yet. Flush any data we can without waiting or removing modified flag.
return false;
}
/// <summary>
@ -296,15 +363,41 @@ namespace Ryujinx.Graphics.Gpu.Image
Interlocked.Exchange(ref _actionRegistered, 0);
}
/// <summary>
/// Action to perform before a sync number is registered after modification.
/// This action will copy the texture data to the flush buffer if this texture
/// flushes often enough, which is determined by the flush balance.
/// </summary>
/// <inheritdoc/>
public void SyncPreAction(bool syncpoint)
{
if (syncpoint || NextSyncCopies())
{
if (ModifyFlushBalance(0) && _registeredBufferSync != _modifiedSync)
{
_group.FlushIntoBuffer(this);
_registeredBufferSync = _modifiedSync;
}
}
}
/// <summary>
/// Action to perform when a sync number is registered after modification.
/// This action will register a read tracking action on the memory tracking handle so that a flush from CPU can happen.
/// </summary>
private void SyncAction()
/// <inheritdoc/>
public bool SyncAction(bool syncpoint)
{
// The storage will need to signal modified again to update the sync number in future.
_group.Storage.SignalModifiedDirty();
bool lastInBuffer = _registeredBufferSync == _modifiedSync;
if (!lastInBuffer)
{
_registeredBufferSync = ulong.MaxValue;
}
lock (Overlaps)
{
foreach (Texture texture in Overlaps)
@ -314,6 +407,7 @@ namespace Ryujinx.Graphics.Gpu.Image
}
// Register region tracking for CPU? (again)
_registeredSync = _modifiedSync;
_syncActionRegistered = false;
@ -321,6 +415,14 @@ namespace Ryujinx.Graphics.Gpu.Image
{
_group.RegisterAction(this);
}
if (syncpoint)
{
_registeredBufferGuestSync = _registeredBufferSync;
}
// If the last modification is in the buffer, keep this sync action alive until it sees a syncpoint.
return syncpoint || !lastInBuffer;
}
/// <summary>

View File

@ -1,5 +1,6 @@
using Ryujinx.Cpu.Tracking;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Synchronization;
using Ryujinx.Memory.Range;
using Ryujinx.Memory.Tracking;
using System;
@ -11,7 +12,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// <summary>
/// Buffer, used to store vertex and index data, uniform and storage buffers, and others.
/// </summary>
class Buffer : IRange, IDisposable
class Buffer : IRange, ISyncActionHandler, IDisposable
{
private const ulong GranularBufferThreshold = 4096;
@ -248,7 +249,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
if (!_syncActionRegistered)
{
_context.RegisterSyncAction(SyncAction);
_context.RegisterSyncAction(this);
_syncActionRegistered = true;
}
}
@ -267,7 +268,8 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// Action to be performed when a syncpoint is reached after modification.
/// This will register read/write tracking to flush the buffer from GPU when its memory is used.
/// </summary>
private void SyncAction()
/// <inheritdoc/>
public bool SyncAction(bool syncpoint)
{
_syncActionRegistered = false;
@ -284,6 +286,8 @@ namespace Ryujinx.Graphics.Gpu.Memory
_memoryTracking.RegisterAction(_externalFlushDelegate);
SynchronizeMemory(Address, Size);
}
return true;
}
/// <summary>
@ -296,7 +300,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
{
if (from._syncActionRegistered && !_syncActionRegistered)
{
_context.RegisterSyncAction(SyncAction);
_context.RegisterSyncAction(this);
_syncActionRegistered = true;
}

View File

@ -8,6 +8,7 @@ using Ryujinx.Memory.Tracking;
using System;
using System.Collections.Generic;
using System.Runtime.InteropServices;
using System.Linq;
using System.Threading;
namespace Ryujinx.Graphics.Gpu.Memory
@ -82,6 +83,34 @@ namespace Ryujinx.Graphics.Gpu.Memory
}
}
/// <summary>
/// Gets a host pointer for a given range of application memory.
/// If the memory region is not a single contiguous block, this method returns 0.
/// </summary>
/// <remarks>
/// Getting a host pointer is unsafe. It should be considered invalid immediately if the GPU memory is unmapped.
/// </remarks>
/// <param name="range">Ranges of physical memory where the target data is located</param>
/// <returns>Pointer to the range of memory</returns>
public nint GetHostPointer(MultiRange range)
{
if (range.Count == 1)
{
var singleRange = range.GetSubRange(0);
if (singleRange.Address != MemoryManager.PteUnmapped)
{
var regions = _cpuMemory.GetHostRegions(singleRange.Address, singleRange.Size);
if (regions != null && regions.Count() == 1)
{
return (nint)regions.First().Address;
}
}
}
return 0;
}
/// <summary>
/// Gets a span of data from the application process.
/// </summary>

View File

@ -0,0 +1,30 @@
using System;
namespace Ryujinx.Graphics.Gpu.Synchronization
{
/// <summary>
/// Modifier flags for creating host sync.
/// </summary>
[Flags]
internal enum HostSyncFlags
{
None = 0,
/// <summary>
/// Present if host sync is being created by a syncpoint.
/// </summary>
Syncpoint = 1 << 0,
/// <summary>
/// Present if the sync should signal as soon as possible.
/// </summary>
Strict = 1 << 1,
/// <summary>
/// Present will force the sync to be created, even if no actions are eligible.
/// </summary>
Force = 1 << 2,
StrictSyncpoint = Strict | Syncpoint
}
}

View File

@ -0,0 +1,22 @@
namespace Ryujinx.Graphics.Gpu.Synchronization
{
/// <summary>
/// This interface indicates that a class can be registered for a sync action.
/// </summary>
interface ISyncActionHandler
{
/// <summary>
/// Action to be performed when some synchronizing action is reached after modification.
/// Generally used to register read/write tracking to flush resources from GPU when their memory is used.
/// </summary>
/// <param name="syncpoint">True if the action is a guest syncpoint</param>
/// <returns>True if the action is to be removed, false otherwise</returns>
bool SyncAction(bool syncpoint);
/// <summary>
/// Action to be performed immediately before sync is created.
/// </summary>
/// <param name="syncpoint">True if the action is a guest syncpoint</param>
void SyncPreAction(bool syncpoint) { }
}
}

View File

@ -42,6 +42,20 @@ namespace Ryujinx.Graphics.OpenGL
return Handle.FromInt32<BufferHandle>(handle);
}
public static BufferHandle CreatePersistent(int size)
{
int handle = GL.GenBuffer();
GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle);
GL.BufferStorage(BufferTarget.CopyWriteBuffer, size, IntPtr.Zero,
BufferStorageFlags.MapPersistentBit |
BufferStorageFlags.MapCoherentBit |
BufferStorageFlags.ClientStorageBit |
BufferStorageFlags.MapReadBit);
return Handle.FromInt32<BufferHandle>(handle);
}
public static void Copy(BufferHandle source, BufferHandle destination, int srcOffset, int dstOffset, int size)
{
GL.BindBuffer(BufferTarget.CopyReadBuffer, source.ToInt32());
@ -60,7 +74,11 @@ namespace Ryujinx.Graphics.OpenGL
// Data in the persistent buffer and host array is guaranteed to be available
// until the next time the host thread requests data.
if (HwCapabilities.UsePersistentBufferForFlush)
if (renderer.PersistentBuffers.TryGet(buffer, out IntPtr ptr))
{
return new PinnedSpan<byte>(IntPtr.Add(ptr, offset).ToPointer(), size);
}
else if (HwCapabilities.UsePersistentBufferForFlush)
{
return PinnedSpan<byte>.UnsafeFromSpan(renderer.PersistentBuffers.Default.GetBufferData(buffer, offset, size));
}

View File

@ -49,6 +49,11 @@ namespace Ryujinx.Graphics.OpenGL.Image
return GetData();
}
public void CopyTo(BufferRange range, int layer, int level, int stride)
{
throw new NotImplementedException();
}
public void SetData(SpanOrArray<byte> data)
{
var dataSpan = data.AsSpan();

View File

@ -3,6 +3,7 @@ using Ryujinx.Common;
using Ryujinx.Common.Memory;
using Ryujinx.Graphics.GAL;
using System;
using System.Diagnostics;
namespace Ryujinx.Graphics.OpenGL.Image
{
@ -287,6 +288,26 @@ namespace Ryujinx.Graphics.OpenGL.Image
}
}
public void CopyTo(BufferRange range, int layer, int level, int stride)
{
if (stride != 0 && stride != BitUtils.AlignUp(Info.Width * Info.BytesPerPixel, 4))
{
throw new NotSupportedException("Stride conversion for texture copy to buffer not supported.");
}
GL.BindBuffer(BufferTarget.PixelPackBuffer, range.Handle.ToInt32());
FormatInfo format = FormatTable.GetFormatInfo(Info.Format);
if (format.PixelFormat == PixelFormat.DepthStencil)
{
throw new InvalidOperationException("DepthStencil copy to buffer is not supported for layer/level > 0.");
}
int offset = WriteToPbo2D(range.Offset, layer, level);
Debug.Assert(offset == 0);
}
public void WriteToPbo(int offset, bool forceBgra)
{
WriteTo(IntPtr.Zero + offset, forceBgra);

View File

@ -58,10 +58,31 @@ namespace Ryujinx.Graphics.OpenGL
}
public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
{
return CreateBuffer(size, GAL.BufferAccess.Default);
}
public BufferHandle CreateBuffer(int size, GAL.BufferAccess access)
{
BufferCount++;
return Buffer.Create(size);
if (access == GAL.BufferAccess.FlushPersistent)
{
BufferHandle handle = Buffer.CreatePersistent(size);
PersistentBuffers.Map(handle, size);
return handle;
}
else
{
return Buffer.Create(size);
}
}
public BufferHandle CreateBuffer(nint pointer, int size)
{
throw new NotSupportedException();
}
public IProgram CreateProgram(ShaderSource[] shaders, ShaderInfo info)
@ -88,6 +109,8 @@ namespace Ryujinx.Graphics.OpenGL
public void DeleteBuffer(BufferHandle buffer)
{
PersistentBuffers.Unmap(buffer);
Buffer.Delete(buffer);
}
@ -272,5 +295,10 @@ namespace Ryujinx.Graphics.OpenGL
{
ScreenCaptured?.Invoke(this, bitmap);
}
public bool PrepareHostMapping(nint address, ulong size)
{
return false;
}
}
}

View File

@ -1,8 +1,9 @@
using OpenTK.Graphics.OpenGL;
using OpenTK.Graphics.OpenGL;
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.OpenGL.Image;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
@ -13,6 +14,8 @@ namespace Ryujinx.Graphics.OpenGL
private PersistentBuffer _main = new PersistentBuffer();
private PersistentBuffer _background = new PersistentBuffer();
private Dictionary<BufferHandle, IntPtr> _maps = new Dictionary<BufferHandle, IntPtr>();
public PersistentBuffer Default => BackgroundContextWorker.InBackground ? _background : _main;
public void Dispose()
@ -20,6 +23,30 @@ namespace Ryujinx.Graphics.OpenGL
_main?.Dispose();
_background?.Dispose();
}
public void Map(BufferHandle handle, int size)
{
GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle.ToInt32());
IntPtr ptr = GL.MapBufferRange(BufferTarget.CopyWriteBuffer, IntPtr.Zero, size, BufferAccessMask.MapReadBit | BufferAccessMask.MapPersistentBit);
_maps[handle] = ptr;
}
public void Unmap(BufferHandle handle)
{
if (_maps.ContainsKey(handle))
{
GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle.ToInt32());
GL.UnmapBuffer(BufferTarget.CopyWriteBuffer);
_maps.Remove(handle);
}
}
public bool TryGet(BufferHandle handle, out IntPtr ptr)
{
return _maps.TryGetValue(handle, out ptr);
}
}
class PersistentBuffer : IDisposable

View File

@ -33,6 +33,7 @@ namespace Ryujinx.Graphics.Vulkan
private MemoryAllocation _allocation;
private Auto<DisposableBuffer> _buffer;
private Auto<MemoryAllocation> _allocationAuto;
private bool _allocationImported;
private ulong _bufferHandle;
private CacheByRange<BufferHolder> _cachedConvertedBuffers;
@ -81,6 +82,26 @@ namespace Ryujinx.Graphics.Vulkan
_flushLock = new ReaderWriterLock();
}
public BufferHolder(VulkanRenderer gd, Device device, VkBuffer buffer, Auto<MemoryAllocation> allocation, int size, BufferAllocationType type, BufferAllocationType currentType, int offset)
{
_gd = gd;
_device = device;
_allocation = allocation.GetUnsafe();
_allocationAuto = allocation;
_allocationImported = true;
_waitable = new MultiFenceHolder(size);
_buffer = new Auto<DisposableBuffer>(new DisposableBuffer(gd.Api, device, buffer), _waitable, _allocationAuto);
_bufferHandle = buffer.Handle;
Size = size;
_map = _allocation.HostPointer + offset;
_baseType = type;
_currentType = currentType;
DesiredType = currentType;
_flushLock = new ReaderWriterLock();
}
public bool TryBackingSwap(ref CommandBufferScoped? cbs)
{
if (_swapQueued && DesiredType != _currentType)
@ -775,8 +796,15 @@ namespace Ryujinx.Graphics.Vulkan
_gd.PipelineInternal?.FlushCommandsIfWeightExceeding(_buffer, (ulong)Size);
_buffer.Dispose();
_allocationAuto.Dispose();
_cachedConvertedBuffers.Dispose();
if (_allocationImported)
{
_allocationAuto.DecrementReferenceCount();
}
else
{
_allocationAuto.Dispose();
}
_flushLock.AcquireWriterLock(Timeout.Infinite);

View File

@ -10,7 +10,7 @@ namespace Ryujinx.Graphics.Vulkan
{
class BufferManager : IDisposable
{
private const MemoryPropertyFlags DefaultBufferMemoryFlags =
public const MemoryPropertyFlags DefaultBufferMemoryFlags =
MemoryPropertyFlags.HostVisibleBit |
MemoryPropertyFlags.HostCoherentBit |
MemoryPropertyFlags.HostCachedBit;
@ -40,6 +40,10 @@ namespace Ryujinx.Graphics.Vulkan
BufferUsageFlags.VertexBufferBit |
BufferUsageFlags.TransformFeedbackBufferBitExt;
private const BufferUsageFlags HostImportedBufferUsageFlags =
BufferUsageFlags.TransferSrcBit |
BufferUsageFlags.TransferDstBit;
private readonly Device _device;
private readonly IdList<BufferHolder> _buffers;
@ -48,11 +52,47 @@ namespace Ryujinx.Graphics.Vulkan
public StagingBuffer StagingBuffer { get; }
public MemoryRequirements HostImportedBufferMemoryRequirements { get; }
public BufferManager(VulkanRenderer gd, Device device)
{
_device = device;
_buffers = new IdList<BufferHolder>();
StagingBuffer = new StagingBuffer(gd, this);
HostImportedBufferMemoryRequirements = GetHostImportedUsageRequirements(gd);
}
public unsafe BufferHandle CreateHostImported(VulkanRenderer gd, nint pointer, int size)
{
var usage = HostImportedBufferUsageFlags;
if (gd.Capabilities.SupportsIndirectParameters)
{
usage |= BufferUsageFlags.IndirectBufferBit;
}
var bufferCreateInfo = new BufferCreateInfo()
{
SType = StructureType.BufferCreateInfo,
Size = (ulong)size,
Usage = usage,
SharingMode = SharingMode.Exclusive
};
gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();
(Auto<MemoryAllocation> allocation, ulong offset) = gd.HostMemoryAllocator.GetExistingAllocation(pointer, (ulong)size);
gd.Api.BindBufferMemory(_device, buffer, allocation.GetUnsafe().Memory, allocation.GetUnsafe().Offset + offset);
var holder = new BufferHolder(gd, _device, buffer, allocation, size, BufferAllocationType.HostMapped, BufferAllocationType.HostMapped, (int)offset);
BufferCount++;
ulong handle64 = (uint)_buffers.Add(holder);
return Unsafe.As<ulong, BufferHandle>(ref handle64);
}
public BufferHandle CreateWithHandle(VulkanRenderer gd, int size, BufferAllocationType baseType = BufferAllocationType.HostMapped, BufferHandle storageHint = default)
@ -75,6 +115,32 @@ namespace Ryujinx.Graphics.Vulkan
return Unsafe.As<ulong, BufferHandle>(ref handle64);
}
public unsafe MemoryRequirements GetHostImportedUsageRequirements(VulkanRenderer gd)
{
var usage = HostImportedBufferUsageFlags;
if (gd.Capabilities.SupportsIndirectParameters)
{
usage |= BufferUsageFlags.IndirectBufferBit;
}
var bufferCreateInfo = new BufferCreateInfo()
{
SType = StructureType.BufferCreateInfo,
Size = (ulong)Environment.SystemPageSize,
Usage = usage,
SharingMode = SharingMode.Exclusive
};
gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();
gd.Api.GetBufferMemoryRequirements(_device, buffer, out var requirements);
gd.Api.DestroyBuffer(_device, buffer, null);
return requirements;
}
public unsafe (VkBuffer buffer, MemoryAllocation allocation, BufferAllocationType resultType) CreateBacking(
VulkanRenderer gd,
int size,

View File

@ -364,6 +364,15 @@ namespace Ryujinx.Graphics.Vulkan
};
}
public static BufferAllocationType Convert(this BufferAccess access)
{
return access switch
{
BufferAccess.FlushPersistent => BufferAllocationType.HostMapped,
_ => BufferAllocationType.Auto
};
}
private static T2 LogInvalidAndReturn<T1, T2>(T1 value, string name, T2 defaultValue = default)
{
Logger.Debug?.Print(LogClass.Gpu, $"Invalid {name} enum value: {value}.");

View File

@ -41,6 +41,7 @@ namespace Ryujinx.Graphics.Vulkan
public readonly bool SupportsPipelineStatisticsQuery;
public readonly bool SupportsGeometryShader;
public readonly bool SupportsViewportArray2;
public readonly bool SupportsHostImportedMemory;
public readonly uint MinSubgroupSize;
public readonly uint MaxSubgroupSize;
public readonly ShaderStageFlags RequiredSubgroupSizeStages;
@ -75,6 +76,7 @@ namespace Ryujinx.Graphics.Vulkan
bool supportsPipelineStatisticsQuery,
bool supportsGeometryShader,
bool supportsViewportArray2,
bool supportsHostImportedMemory,
uint minSubgroupSize,
uint maxSubgroupSize,
ShaderStageFlags requiredSubgroupSizeStages,
@ -108,6 +110,7 @@ namespace Ryujinx.Graphics.Vulkan
SupportsPipelineStatisticsQuery = supportsPipelineStatisticsQuery;
SupportsGeometryShader = supportsGeometryShader;
SupportsViewportArray2 = supportsViewportArray2;
SupportsHostImportedMemory = supportsHostImportedMemory;
MinSubgroupSize = minSubgroupSize;
MaxSubgroupSize = maxSubgroupSize;
RequiredSubgroupSizeStages = requiredSubgroupSizeStages;

View File

@ -31,6 +31,7 @@ namespace Ryujinx.Graphics.Vulkan
private readonly IProgram _programColorClearSI;
private readonly IProgram _programColorClearUI;
private readonly IProgram _programStrideChange;
private readonly IProgram _programConvertD32S8ToD24S8;
private readonly IProgram _programConvertIndexBuffer;
private readonly IProgram _programConvertIndirectData;
private readonly IProgram _programColorCopyShortening;
@ -158,6 +159,17 @@ namespace Ryujinx.Graphics.Vulkan
new ShaderSource(ShaderBinaries.ColorDrawToMsFragmentShaderSource, colorDrawToMsFragmentBindings, ShaderStage.Fragment, TargetLanguage.Spirv),
});
var convertD32S8ToD24S8Bindings = new ShaderBindings(
new[] { 0 },
new[] { 1, 2 },
Array.Empty<int>(),
Array.Empty<int>());
_programConvertD32S8ToD24S8 = gd.CreateProgramWithMinimalLayout(new[]
{
new ShaderSource(ShaderBinaries.ConvertD32S8ToD24S8ShaderSource, convertD32S8ToD24S8Bindings, ShaderStage.Compute, TargetLanguage.Spirv),
});
var convertIndexBufferBindings = new ShaderBindings(
new[] { 0 },
new[] { 1, 2 },
@ -1644,6 +1656,82 @@ namespace Ryujinx.Graphics.Vulkan
_pipeline.Finish(gd, cbs);
}
public unsafe void ConvertD32S8ToD24S8(VulkanRenderer gd, CommandBufferScoped cbs, BufferHolder src, Auto<DisposableBuffer> dstBufferAuto, int pixelCount, int dstOffset)
{
int inSize = pixelCount * 2 * sizeof(int);
int outSize = pixelCount * sizeof(int);
var srcBufferAuto = src.GetBuffer();
var srcBuffer = srcBufferAuto.Get(cbs, 0, inSize).Value;
var dstBuffer = dstBufferAuto.Get(cbs, dstOffset, outSize).Value;
var access = AccessFlags.ShaderWriteBit;
var stage = PipelineStageFlags.ComputeShaderBit;
BufferHolder.InsertBufferBarrier(
gd,
cbs.CommandBuffer,
srcBuffer,
BufferHolder.DefaultAccessFlags,
AccessFlags.ShaderReadBit,
PipelineStageFlags.AllCommandsBit,
stage,
0,
outSize);
BufferHolder.InsertBufferBarrier(
gd,
cbs.CommandBuffer,
dstBuffer,
BufferHolder.DefaultAccessFlags,
access,
PipelineStageFlags.AllCommandsBit,
stage,
0,
outSize);
const int ParamsBufferSize = sizeof(int) * 2;
Span<int> shaderParams = stackalloc int[2];
shaderParams[0] = pixelCount;
shaderParams[1] = dstOffset;
var bufferHandle = gd.BufferManager.CreateWithHandle(gd, ParamsBufferSize);
gd.BufferManager.SetData<int>(bufferHandle, 0, shaderParams);
_pipeline.SetCommandBuffer(cbs);
_pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(0, new BufferRange(bufferHandle, 0, ParamsBufferSize)) });
Span<Auto<DisposableBuffer>> sbRanges = new Auto<DisposableBuffer>[2];
sbRanges[0] = srcBufferAuto;
sbRanges[1] = dstBufferAuto;
_pipeline.SetStorageBuffers(1, sbRanges);
_pipeline.SetProgram(_programConvertD32S8ToD24S8);
_pipeline.DispatchCompute(1, 1, 1);
gd.BufferManager.Delete(bufferHandle);
_pipeline.Finish(gd, cbs);
BufferHolder.InsertBufferBarrier(
gd,
cbs.CommandBuffer,
dstBuffer,
access,
BufferHolder.DefaultAccessFlags,
stage,
PipelineStageFlags.AllCommandsBit,
0,
outSize);
}
protected virtual void Dispose(bool disposing)
{
if (disposing)

View File

@ -0,0 +1,188 @@
using Ryujinx.Common;
using Ryujinx.Common.Collections;
using Ryujinx.Common.Logging;
using Silk.NET.Vulkan;
using Silk.NET.Vulkan.Extensions.EXT;
using System;
using System.Collections.Generic;
namespace Ryujinx.Graphics.Vulkan
{
internal class HostMemoryAllocator
{
private struct HostMemoryAllocation
{
public readonly Auto<MemoryAllocation> Allocation;
public readonly IntPtr Pointer;
public readonly ulong Size;
public ulong Start => (ulong)Pointer;
public ulong End => (ulong)Pointer + Size;
public HostMemoryAllocation(Auto<MemoryAllocation> allocation, IntPtr pointer, ulong size)
{
Allocation = allocation;
Pointer = pointer;
Size = size;
}
}
private readonly MemoryAllocator _allocator;
private readonly Vk _api;
private readonly ExtExternalMemoryHost _hostMemoryApi;
private readonly Device _device;
private readonly object _lock = new();
private List<HostMemoryAllocation> _allocations;
private IntervalTree<ulong, HostMemoryAllocation> _allocationTree;
public HostMemoryAllocator(MemoryAllocator allocator, Vk api, ExtExternalMemoryHost hostMemoryApi, Device device)
{
_allocator = allocator;
_api = api;
_hostMemoryApi = hostMemoryApi;
_device = device;
_allocations = new List<HostMemoryAllocation>();
_allocationTree = new IntervalTree<ulong, HostMemoryAllocation>();
}
public unsafe bool TryImport(
MemoryRequirements requirements,
MemoryPropertyFlags flags,
IntPtr pointer,
ulong size)
{
lock (_lock)
{
// Does a compatible allocation exist in the tree?
var allocations = new HostMemoryAllocation[10];
ulong start = (ulong)pointer;
ulong end = start + size;
int count = _allocationTree.Get(start, end, ref allocations);
// A compatible range is one that where the start and end completely cover the requested range.
for (int i = 0; i < count; i++)
{
HostMemoryAllocation existing = allocations[i];
if (start >= existing.Start && end <= existing.End)
{
try
{
existing.Allocation.IncrementReferenceCount();
return true;
}
catch (InvalidOperationException)
{
// Can throw if the allocation has been disposed.
// Just continue the search if this happens.
}
}
}
nint pageAlignedPointer = BitUtils.AlignDown(pointer, Environment.SystemPageSize);
nint pageAlignedEnd = BitUtils.AlignUp((nint)((ulong)pointer + size), Environment.SystemPageSize);
ulong pageAlignedSize = (ulong)(pageAlignedEnd - pageAlignedPointer);
Result getResult = _hostMemoryApi.GetMemoryHostPointerProperties(_device, ExternalMemoryHandleTypeFlags.HostAllocationBitExt, (void*)pageAlignedPointer, out MemoryHostPointerPropertiesEXT properties);
if (getResult < Result.Success)
{
return false;
}
int memoryTypeIndex = _allocator.FindSuitableMemoryTypeIndex(properties.MemoryTypeBits & requirements.MemoryTypeBits, flags);
if (memoryTypeIndex < 0)
{
return false;
}
ImportMemoryHostPointerInfoEXT importInfo = new ImportMemoryHostPointerInfoEXT()
{
SType = StructureType.ImportMemoryHostPointerInfoExt,
HandleType = ExternalMemoryHandleTypeFlags.HostAllocationBitExt,
PHostPointer = (void*)pageAlignedPointer
};
var memoryAllocateInfo = new MemoryAllocateInfo()
{
SType = StructureType.MemoryAllocateInfo,
AllocationSize = pageAlignedSize,
MemoryTypeIndex = (uint)memoryTypeIndex,
PNext = &importInfo
};
Result result = _api.AllocateMemory(_device, memoryAllocateInfo, null, out var deviceMemory);
if (result < Result.Success)
{
Logger.Debug?.PrintMsg(LogClass.Gpu, $"Host mapping import 0x{pageAlignedPointer:x16} 0x{pageAlignedSize:x8} failed.");
return false;
}
var allocation = new MemoryAllocation(this, deviceMemory, pageAlignedPointer, 0, pageAlignedSize);
var allocAuto = new Auto<MemoryAllocation>(allocation);
var hostAlloc = new HostMemoryAllocation(allocAuto, pageAlignedPointer, pageAlignedSize);
allocAuto.IncrementReferenceCount();
allocAuto.Dispose(); // Kept alive by ref count only.
// Register this mapping for future use.
_allocationTree.Add(hostAlloc.Start, hostAlloc.End, hostAlloc);
_allocations.Add(hostAlloc);
}
return true;
}
public (Auto<MemoryAllocation>, ulong) GetExistingAllocation(IntPtr pointer, ulong size)
{
lock (_lock)
{
// Does a compatible allocation exist in the tree?
var allocations = new HostMemoryAllocation[10];
ulong start = (ulong)pointer;
ulong end = start + size;
int count = _allocationTree.Get(start, end, ref allocations);
// A compatible range is one that where the start and end completely cover the requested range.
for (int i = 0; i < count; i++)
{
HostMemoryAllocation existing = allocations[i];
if (start >= existing.Start && end <= existing.End)
{
return (existing.Allocation, start - existing.Start);
}
}
throw new InvalidOperationException($"No host allocation was prepared for requested range 0x{pointer:x16}:0x{size:x16}.");
}
}
public void Free(DeviceMemory memory, ulong offset, ulong size)
{
lock (_lock)
{
_allocations.RemoveAll(allocation =>
{
if (allocation.Allocation.GetUnsafe().Memory.Handle == memory.Handle)
{
_allocationTree.Remove(allocation.Start, allocation);
return true;
}
return false;
});
}
_api.FreeMemory(_device, memory, ReadOnlySpan<AllocationCallbacks>.Empty);
}
}
}

View File

@ -7,6 +7,7 @@ namespace Ryujinx.Graphics.Vulkan
{
private readonly MemoryAllocatorBlockList _owner;
private readonly MemoryAllocatorBlockList.Block _block;
private readonly HostMemoryAllocator _hostMemory;
public DeviceMemory Memory { get; }
public IntPtr HostPointer { get;}
@ -29,9 +30,30 @@ namespace Ryujinx.Graphics.Vulkan
Size = size;
}
public MemoryAllocation(
HostMemoryAllocator hostMemory,
DeviceMemory memory,
IntPtr hostPointer,
ulong offset,
ulong size)
{
_hostMemory = hostMemory;
Memory = memory;
HostPointer = hostPointer;
Offset = offset;
Size = size;
}
public void Dispose()
{
_owner.Free(_block, Offset, Size);
if (_hostMemory != null)
{
_hostMemory.Free(Memory, Offset, Size);
}
else
{
_owner.Free(_block, Offset, Size);
}
}
}
}

View File

@ -57,7 +57,7 @@ namespace Ryujinx.Graphics.Vulkan
return newBl.Allocate(size, alignment, map);
}
private int FindSuitableMemoryTypeIndex(
internal int FindSuitableMemoryTypeIndex(
uint memoryTypeBits,
MemoryPropertyFlags flags)
{

View File

@ -0,0 +1,58 @@
#version 450 core
#extension GL_EXT_scalar_block_layout : require
layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout (std430, set = 0, binding = 0) uniform stride_arguments
{
int pixelCount;
int dstStartOffset;
};
layout (std430, set = 1, binding = 1) buffer in_s
{
uint[] in_data;
};
layout (std430, set = 1, binding = 2) buffer out_s
{
uint[] out_data;
};
void main()
{
// Determine what slice of the stride copies this invocation will perform.
int invocations = int(gl_WorkGroupSize.x);
int copiesRequired = pixelCount;
// Find the copies that this invocation should perform.
// - Copies that all invocations perform.
int allInvocationCopies = copiesRequired / invocations;
// - Extra remainder copy that this invocation performs.
int index = int(gl_LocalInvocationID.x);
int extra = (index < (copiesRequired % invocations)) ? 1 : 0;
int copyCount = allInvocationCopies + extra;
// Finally, get the starting offset. Make sure to count extra copies.
int startCopy = allInvocationCopies * index + min(copiesRequired % invocations, index);
int srcOffset = startCopy * 2;
int dstOffset = dstStartOffset + startCopy;
// Perform the conversion for this region.
for (int i = 0; i < copyCount; i++)
{
float depth = uintBitsToFloat(in_data[srcOffset++]);
uint stencil = in_data[srcOffset++];
uint rescaledDepth = uint(clamp(depth, 0.0, 1.0) * 16777215.0);
out_data[dstOffset++] = (rescaledDepth << 8) | (stencil & 0xff);
}
}

View File

@ -1236,6 +1236,213 @@ namespace Ryujinx.Graphics.Vulkan.Shaders
0x38, 0x00, 0x01, 0x00,
};
public static readonly byte[] ConvertD32S8ToD24S8ShaderSource = new byte[]
{
0x03, 0x02, 0x23, 0x07, 0x00, 0x05, 0x01, 0x00, 0x0A, 0x00, 0x0D, 0x00, 0x77, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0B, 0x00, 0x06, 0x00,
0x01, 0x00, 0x00, 0x00, 0x47, 0x4C, 0x53, 0x4C, 0x2E, 0x73, 0x74, 0x64, 0x2E, 0x34, 0x35, 0x30,
0x00, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x0F, 0x00, 0x09, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6D, 0x61, 0x69, 0x6E,
0x00, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x00, 0x00,
0x67, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
0x40, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x03, 0x00, 0x03, 0x00,
0x02, 0x00, 0x00, 0x00, 0xC2, 0x01, 0x00, 0x00, 0x04, 0x00, 0x08, 0x00, 0x47, 0x4C, 0x5F, 0x45,
0x58, 0x54, 0x5F, 0x73, 0x63, 0x61, 0x6C, 0x61, 0x72, 0x5F, 0x62, 0x6C, 0x6F, 0x63, 0x6B, 0x5F,
0x6C, 0x61, 0x79, 0x6F, 0x75, 0x74, 0x00, 0x00, 0x04, 0x00, 0x0A, 0x00, 0x47, 0x4C, 0x5F, 0x47,
0x4F, 0x4F, 0x47, 0x4C, 0x45, 0x5F, 0x63, 0x70, 0x70, 0x5F, 0x73, 0x74, 0x79, 0x6C, 0x65, 0x5F,
0x6C, 0x69, 0x6E, 0x65, 0x5F, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x69, 0x76, 0x65, 0x00, 0x00,
0x04, 0x00, 0x08, 0x00, 0x47, 0x4C, 0x5F, 0x47, 0x4F, 0x4F, 0x47, 0x4C, 0x45, 0x5F, 0x69, 0x6E,
0x63, 0x6C, 0x75, 0x64, 0x65, 0x5F, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x69, 0x76, 0x65, 0x00,
0x05, 0x00, 0x04, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6D, 0x61, 0x69, 0x6E, 0x00, 0x00, 0x00, 0x00,
0x05, 0x00, 0x05, 0x00, 0x08, 0x00, 0x00, 0x00, 0x69, 0x6E, 0x76, 0x6F, 0x63, 0x61, 0x74, 0x69,
0x6F, 0x6E, 0x73, 0x00, 0x05, 0x00, 0x06, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x63, 0x6F, 0x70, 0x69,
0x65, 0x73, 0x52, 0x65, 0x71, 0x75, 0x69, 0x72, 0x65, 0x64, 0x00, 0x00, 0x05, 0x00, 0x07, 0x00,
0x0B, 0x00, 0x00, 0x00, 0x73, 0x74, 0x72, 0x69, 0x64, 0x65, 0x5F, 0x61, 0x72, 0x67, 0x75, 0x6D,
0x65, 0x6E, 0x74, 0x73, 0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00, 0x0B, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x43, 0x6F, 0x75, 0x6E, 0x74, 0x00, 0x00,
0x06, 0x00, 0x07, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x64, 0x73, 0x74, 0x53,
0x74, 0x61, 0x72, 0x74, 0x4F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
0x0D, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x07, 0x00, 0x12, 0x00, 0x00, 0x00,
0x61, 0x6C, 0x6C, 0x49, 0x6E, 0x76, 0x6F, 0x63, 0x61, 0x74, 0x69, 0x6F, 0x6E, 0x43, 0x6F, 0x70,
0x69, 0x65, 0x73, 0x00, 0x05, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x69, 0x6E, 0x64, 0x65,
0x78, 0x00, 0x00, 0x00, 0x05, 0x00, 0x08, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x67, 0x6C, 0x5F, 0x4C,
0x6F, 0x63, 0x61, 0x6C, 0x49, 0x6E, 0x76, 0x6F, 0x63, 0x61, 0x74, 0x69, 0x6F, 0x6E, 0x49, 0x44,
0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00, 0x65, 0x78, 0x74, 0x72,
0x61, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00, 0x29, 0x00, 0x00, 0x00, 0x63, 0x6F, 0x70, 0x79,
0x43, 0x6F, 0x75, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00, 0x2D, 0x00, 0x00, 0x00,
0x73, 0x74, 0x61, 0x72, 0x74, 0x43, 0x6F, 0x70, 0x79, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00,
0x37, 0x00, 0x00, 0x00, 0x73, 0x72, 0x63, 0x4F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x00, 0x00,
0x05, 0x00, 0x05, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x64, 0x73, 0x74, 0x4F, 0x66, 0x66, 0x73, 0x65,
0x74, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00, 0x40, 0x00, 0x00, 0x00, 0x69, 0x00, 0x00, 0x00,
0x05, 0x00, 0x04, 0x00, 0x4B, 0x00, 0x00, 0x00, 0x64, 0x65, 0x70, 0x74, 0x68, 0x00, 0x00, 0x00,
0x05, 0x00, 0x04, 0x00, 0x4D, 0x00, 0x00, 0x00, 0x69, 0x6E, 0x5F, 0x73, 0x00, 0x00, 0x00, 0x00,
0x06, 0x00, 0x05, 0x00, 0x4D, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x69, 0x6E, 0x5F, 0x64,
0x61, 0x74, 0x61, 0x00, 0x05, 0x00, 0x03, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x05, 0x00, 0x04, 0x00, 0x57, 0x00, 0x00, 0x00, 0x73, 0x74, 0x65, 0x6E, 0x63, 0x69, 0x6C, 0x00,
0x05, 0x00, 0x06, 0x00, 0x5C, 0x00, 0x00, 0x00, 0x72, 0x65, 0x73, 0x63, 0x61, 0x6C, 0x65, 0x64,
0x44, 0x65, 0x70, 0x74, 0x68, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00, 0x65, 0x00, 0x00, 0x00,
0x6F, 0x75, 0x74, 0x5F, 0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00, 0x65, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x6F, 0x75, 0x74, 0x5F, 0x64, 0x61, 0x74, 0x61, 0x00, 0x00, 0x00, 0x00,
0x05, 0x00, 0x03, 0x00, 0x67, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00,
0x0B, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x4C, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x4D, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x4D, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x64, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x48, 0x00, 0x05, 0x00, 0x65, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x65, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x67, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x67, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x47, 0x00, 0x04, 0x00, 0x76, 0x00, 0x00, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
0x40, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x04, 0x00, 0x0B, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x0B, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x18, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x17, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x39, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x03, 0x00,
0x49, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x4A, 0x00, 0x00, 0x00,
0x07, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00, 0x1D, 0x00, 0x03, 0x00, 0x4C, 0x00, 0x00, 0x00,
0x17, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x03, 0x00, 0x4D, 0x00, 0x00, 0x00, 0x4C, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x4E, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x4D, 0x00, 0x00, 0x00,
0x3B, 0x00, 0x04, 0x00, 0x4E, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x52, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00,
0x20, 0x00, 0x04, 0x00, 0x56, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00,
0x2B, 0x00, 0x04, 0x00, 0x49, 0x00, 0x00, 0x00, 0x5E, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x2B, 0x00, 0x04, 0x00, 0x49, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3F,
0x2B, 0x00, 0x04, 0x00, 0x49, 0x00, 0x00, 0x00, 0x61, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0x7F, 0x4B,
0x1D, 0x00, 0x03, 0x00, 0x64, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x03, 0x00,
0x65, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x66, 0x00, 0x00, 0x00,
0x0C, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00, 0x66, 0x00, 0x00, 0x00,
0x67, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x6B, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
0x6E, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
0x74, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
0x75, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x06, 0x00, 0x18, 0x00, 0x00, 0x00,
0x76, 0x00, 0x00, 0x00, 0x74, 0x00, 0x00, 0x00, 0x75, 0x00, 0x00, 0x00, 0x75, 0x00, 0x00, 0x00,
0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x07, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x4A, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x56, 0x00, 0x00, 0x00, 0x57, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x04, 0x00,
0x56, 0x00, 0x00, 0x00, 0x5C, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x08, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00,
0x10, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x0A, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x13, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x14, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x87, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x15, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x12, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x1C, 0x00, 0x00, 0x00,
0x1D, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x1B, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x17, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x00, 0x00, 0x1D, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x1F, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x16, 0x00, 0x00, 0x00, 0x1F, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x21, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x22, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x23, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x8B, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x24, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0xB1, 0x00, 0x05, 0x00,
0x25, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
0xA9, 0x00, 0x06, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00,
0x27, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
0x28, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2A, 0x00, 0x00, 0x00,
0x12, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x00, 0x00,
0x20, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00,
0x2A, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00,
0x2C, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2E, 0x00, 0x00, 0x00,
0x12, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2F, 0x00, 0x00, 0x00,
0x16, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00,
0x2E, 0x00, 0x00, 0x00, 0x2F, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x31, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x32, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x8B, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x33, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x07, 0x00,
0x06, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
0x33, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x36, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x2D, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x38, 0x00, 0x00, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x3A, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x39, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x37, 0x00, 0x00, 0x00, 0x3A, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0F, 0x00, 0x00, 0x00,
0x3C, 0x00, 0x00, 0x00, 0x0D, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x00, 0x00, 0x3C, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x00, 0x00, 0x2D, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x3F, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x3F, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x40, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0x41, 0x00, 0x00, 0x00,
0xF8, 0x00, 0x02, 0x00, 0x41, 0x00, 0x00, 0x00, 0xF6, 0x00, 0x04, 0x00, 0x43, 0x00, 0x00, 0x00,
0x44, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00, 0x45, 0x00, 0x00, 0x00,
0xF8, 0x00, 0x02, 0x00, 0x45, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x46, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x47, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0xB1, 0x00, 0x05, 0x00, 0x25, 0x00, 0x00, 0x00,
0x48, 0x00, 0x00, 0x00, 0x46, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00, 0xFA, 0x00, 0x04, 0x00,
0x48, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00,
0x42, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x50, 0x00, 0x00, 0x00,
0x37, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00,
0x50, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00, 0x37, 0x00, 0x00, 0x00,
0x51, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x52, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00,
0x4F, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x50, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x17, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x7C, 0x00, 0x04, 0x00,
0x49, 0x00, 0x00, 0x00, 0x55, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x4B, 0x00, 0x00, 0x00, 0x55, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
0x58, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
0x59, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x03, 0x00,
0x37, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x52, 0x00, 0x00, 0x00,
0x5A, 0x00, 0x00, 0x00, 0x4F, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00,
0x3D, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x5B, 0x00, 0x00, 0x00, 0x5A, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00, 0x5B, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x49, 0x00, 0x00, 0x00, 0x5D, 0x00, 0x00, 0x00, 0x4B, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x08, 0x00,
0x49, 0x00, 0x00, 0x00, 0x60, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2B, 0x00, 0x00, 0x00,
0x5D, 0x00, 0x00, 0x00, 0x5E, 0x00, 0x00, 0x00, 0x5F, 0x00, 0x00, 0x00, 0x85, 0x00, 0x05, 0x00,
0x49, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00, 0x60, 0x00, 0x00, 0x00, 0x61, 0x00, 0x00, 0x00,
0x6D, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x63, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x5C, 0x00, 0x00, 0x00, 0x63, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x69, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x3B, 0x00, 0x00, 0x00, 0x69, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x17, 0x00, 0x00, 0x00, 0x6A, 0x00, 0x00, 0x00, 0x5C, 0x00, 0x00, 0x00, 0xC4, 0x00, 0x05, 0x00,
0x17, 0x00, 0x00, 0x00, 0x6C, 0x00, 0x00, 0x00, 0x6A, 0x00, 0x00, 0x00, 0x6B, 0x00, 0x00, 0x00,
0x3D, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x6D, 0x00, 0x00, 0x00, 0x57, 0x00, 0x00, 0x00,
0xC7, 0x00, 0x05, 0x00, 0x17, 0x00, 0x00, 0x00, 0x6F, 0x00, 0x00, 0x00, 0x6D, 0x00, 0x00, 0x00,
0x6E, 0x00, 0x00, 0x00, 0xC5, 0x00, 0x05, 0x00, 0x17, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
0x6C, 0x00, 0x00, 0x00, 0x6F, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x52, 0x00, 0x00, 0x00,
0x71, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x71, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
0x44, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x44, 0x00, 0x00, 0x00, 0x3D, 0x00, 0x04, 0x00,
0x06, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
0x06, 0x00, 0x00, 0x00, 0x73, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
0x3E, 0x00, 0x03, 0x00, 0x40, 0x00, 0x00, 0x00, 0x73, 0x00, 0x00, 0x00, 0xF9, 0x00, 0x02, 0x00,
0x41, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x02, 0x00, 0x43, 0x00, 0x00, 0x00, 0xFD, 0x00, 0x01, 0x00,
0x38, 0x00, 0x01, 0x00
};
public static readonly byte[] ConvertIndexBufferShaderSource = new byte[]
{
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x0A, 0x00, 0x08, 0x00, 0x91, 0x00, 0x00, 0x00,

View File

@ -125,6 +125,24 @@ namespace Ryujinx.Graphics.Vulkan
if (result != null)
{
if (result.Waitable == null)
{
return;
}
long beforeTicks = Stopwatch.GetTimestamp();
if (result.NeedsFlush(FlushId))
{
_gd.InterruptAction(() =>
{
if (result.NeedsFlush(FlushId))
{
_gd.FlushAllCommands();
}
});
}
lock (result)
{
if (result.Waitable == null)
@ -132,19 +150,6 @@ namespace Ryujinx.Graphics.Vulkan
return;
}
long beforeTicks = Stopwatch.GetTimestamp();
if (result.NeedsFlush(FlushId))
{
_gd.InterruptAction(() =>
{
if (result.NeedsFlush(FlushId))
{
_gd.FlushAllCommands();
}
});
}
bool signaled = result.Signalled || result.Waitable.WaitForFences(_gd.Api, _device, 1000000000);
if (!signaled)

View File

@ -67,6 +67,11 @@ namespace Ryujinx.Graphics.Vulkan
return GetData();
}
public void CopyTo(BufferRange range, int layer, int level, int stride)
{
throw new NotImplementedException();
}
public void Release()
{
if (_gd.Textures.Remove(this))

View File

@ -563,6 +563,34 @@ namespace Ryujinx.Graphics.Vulkan
}
}
public void CopyTo(BufferRange range, int layer, int level, int stride)
{
_gd.PipelineInternal.EndRenderPass();
var cbs = _gd.PipelineInternal.CurrentCommandBuffer;
int outSize = Info.GetMipSize(level);
int hostSize = GetBufferDataLength(outSize);
var image = GetImage().Get(cbs).Value;
int offset = range.Offset;
Auto<DisposableBuffer> autoBuffer = _gd.BufferManager.GetBuffer(cbs.CommandBuffer, range.Handle, true);
VkBuffer buffer = autoBuffer.Get(cbs, range.Offset, outSize).Value;
if (PrepareOutputBuffer(cbs, hostSize, buffer, out VkBuffer copyToBuffer, out BufferHolder tempCopyHolder))
{
offset = 0;
}
CopyFromOrToBuffer(cbs.CommandBuffer, copyToBuffer, image, hostSize, true, layer, level, 1, 1, singleSlice: true, offset, stride);
if (tempCopyHolder != null)
{
CopyDataToOutputBuffer(cbs, tempCopyHolder, autoBuffer, hostSize, range.Offset);
tempCopyHolder.Dispose();
}
}
private ReadOnlySpan<byte> GetData(CommandBufferPool cbp, PersistentFlushBuffer flushBuffer)
{
int size = 0;
@ -693,6 +721,30 @@ namespace Ryujinx.Graphics.Vulkan
return storage;
}
private bool PrepareOutputBuffer(CommandBufferScoped cbs, int hostSize, VkBuffer target, out VkBuffer copyTarget, out BufferHolder copyTargetHolder)
{
if (NeedsD24S8Conversion())
{
copyTargetHolder = _gd.BufferManager.Create(_gd, hostSize);
copyTarget = copyTargetHolder.GetBuffer().Get(cbs, 0, hostSize).Value;
return true;
}
copyTarget = target;
copyTargetHolder = null;
return false;
}
private void CopyDataToOutputBuffer(CommandBufferScoped cbs, BufferHolder hostData, Auto<DisposableBuffer> copyTarget, int hostSize, int dstOffset)
{
if (NeedsD24S8Conversion())
{
_gd.HelperShader.ConvertD32S8ToD24S8(_gd, cbs, hostData, copyTarget, hostSize / (2 * sizeof(int)), dstOffset);
}
}
private bool NeedsD24S8Conversion()
{
return FormatCapabilities.IsD24S8(Info.Format) && VkFormat == VkFormat.D32SfloatS8Uint;
@ -708,7 +760,9 @@ namespace Ryujinx.Graphics.Vulkan
int dstLevel,
int dstLayers,
int dstLevels,
bool singleSlice)
bool singleSlice,
int offset = 0,
int stride = 0)
{
bool is3D = Info.Target == Target.Texture3D;
int width = Math.Max(1, Info.Width >> dstLevel);
@ -718,8 +772,6 @@ namespace Ryujinx.Graphics.Vulkan
int layers = dstLayers;
int levels = dstLevels;
int offset = 0;
for (int level = 0; level < levels; level++)
{
int mipSize = GetBufferDataLength(Info.GetMipSize2D(dstLevel + level) * dstLayers);
@ -731,7 +783,7 @@ namespace Ryujinx.Graphics.Vulkan
break;
}
int rowLength = (Info.GetMipStride(dstLevel + level) / Info.BytesPerPixel) * Info.BlockWidth;
int rowLength = ((stride == 0 ? Info.GetMipStride(dstLevel + level) : stride) / Info.BytesPerPixel) * Info.BlockWidth;
var aspectFlags = Info.Format.ConvertAspectFlags();

View File

@ -27,6 +27,7 @@ namespace Ryujinx.Graphics.Vulkan
ExtTransformFeedback.ExtensionName,
KhrDrawIndirectCount.ExtensionName,
KhrPushDescriptor.ExtensionName,
ExtExternalMemoryHost.ExtensionName,
"VK_EXT_blend_operation_advanced",
"VK_EXT_custom_border_color",
"VK_EXT_descriptor_indexing", // Enabling this works around an issue with disposed buffer bindings on RADV.

View File

@ -44,6 +44,7 @@ namespace Ryujinx.Graphics.Vulkan
internal object QueueLock { get; private set; }
internal MemoryAllocator MemoryAllocator { get; private set; }
internal HostMemoryAllocator HostMemoryAllocator { get; private set; }
internal CommandBufferPool CommandBufferPool { get; private set; }
internal DescriptorSetManager DescriptorSetManager { get; private set; }
internal PipelineLayoutCache PipelineLayoutCache { get; private set; }
@ -307,6 +308,7 @@ namespace Ryujinx.Graphics.Vulkan
_physicalDevice.PhysicalDeviceFeatures.PipelineStatisticsQuery,
_physicalDevice.PhysicalDeviceFeatures.GeometryShader,
_physicalDevice.IsDeviceExtensionPresent("VK_NV_viewport_array2"),
_physicalDevice.IsDeviceExtensionPresent(ExtExternalMemoryHost.ExtensionName),
propertiesSubgroupSizeControl.MinSubgroupSize,
propertiesSubgroupSizeControl.MaxSubgroupSize,
propertiesSubgroupSizeControl.RequiredSubgroupSizeStages,
@ -319,6 +321,9 @@ namespace Ryujinx.Graphics.Vulkan
MemoryAllocator = new MemoryAllocator(Api, _physicalDevice, _device);
Api.TryGetDeviceExtension(_instance.Instance, _device, out ExtExternalMemoryHost hostMemoryApi);
HostMemoryAllocator = new HostMemoryAllocator(MemoryAllocator, Api, hostMemoryApi, _device);
CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex);
DescriptorSetManager = new DescriptorSetManager(_device);
@ -375,11 +380,21 @@ namespace Ryujinx.Graphics.Vulkan
_initialized = true;
}
public BufferHandle CreateBuffer(int size, BufferAccess access)
{
return BufferManager.CreateWithHandle(this, size, access.Convert());
}
public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
{
return BufferManager.CreateWithHandle(this, size, BufferAllocationType.Auto, storageHint);
}
public BufferHandle CreateBuffer(nint pointer, int size)
{
return BufferManager.CreateHostImported(this, pointer, size);
}
public IProgram CreateProgram(ShaderSource[] sources, ShaderInfo info)
{
bool isCompute = sources.Length == 1 && sources[0].Stage == ShaderStage.Compute;
@ -816,5 +831,11 @@ namespace Ryujinx.Graphics.Vulkan
// Last step destroy the instance
_instance.Dispose();
}
public bool PrepareHostMapping(nint address, ulong size)
{
return Capabilities.SupportsHostImportedMemory &&
HostMemoryAllocator.TryImport(BufferManager.HostImportedBufferMemoryRequirements, BufferManager.DefaultBufferMemoryFlags, address, size);
}
}
}