epic/OpenRA
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ThreadedGraphicsContext improvements.

Browse Source 
- VertexBuffer interface redefined to remove an IntPtr overload for SetData. This removes some unsafe code in TerrainSpriteLayer. This also allows the ThreadedVertexBuffer to use a buffer and post these calls, meaning the SetData call can now be non-blocking.
- ThreadedTexture SetData now checks the incoming array size. As the arrays sent here are usually large (megabytes) this allows us to avoid creating temp arrays in the LOH and skip Array.Copy calls on large arrays. This means the call is now blocking more often, but significantly reduces memory churn and GC Gen2 collections.
This commit is contained in:
RoosterDragon
2020-10-11 09:46:41 +01:00
committed by abcdefg30
parent 5eadd26f66
commit b2b639434c
4 changed files with 44 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

61
OpenRA.Platforms.Default/ThreadedGraphicsContext.cs
View File

@@ -29,9 +29,9 @@ namespace OpenRA.Platforms.Default
readonly Stack<Message> messagePool = new Stack<Message>();
readonly Queue<Message> messages = new Queue<Message>();
public readonly int BatchSize;
readonly object syncObject = new object();
readonly Thread renderThread;
readonly int batchSize;
volatile ExceptionDispatchInfo messageException;
// Delegates that perform actions on the real device.
@@ -53,7 +53,7 @@ namespace OpenRA.Platforms.Default
public ThreadedGraphicsContext(Sdl2GraphicsContext context, int batchSize)
{
this.batchSize = batchSize;
BatchSize = batchSize;
renderThread = new Thread(RenderThread)
{
Name = "ThreadedGraphicsContext RenderThread",
@@ -143,15 +143,15 @@ namespace OpenRA.Platforms.Default
internal Vertex[] GetVertices(int size)
{
lock (verticesPool)
if (size <= batchSize && verticesPool.Count > 0)
if (size <= BatchSize && verticesPool.Count > 0)
return verticesPool.Pop();
return new Vertex[size < batchSize ? batchSize : size];
return new Vertex[size < BatchSize ? BatchSize : size];
}
internal void ReturnVertices(Vertex[] vertices)
{
if (vertices.Length == batchSize)
if (vertices.Length == BatchSize)
lock (verticesPool)
verticesPool.Push(vertices);
}
@@ -513,7 +513,8 @@ namespace OpenRA.Platforms.Default
readonly ThreadedGraphicsContext device;
readonly Action bind;
readonly Action<object> setData1;
readonly Func<object, object> setData2;
readonly Action<object> setData2;
readonly Func<object, object> setData3;
readonly Action dispose;
public ThreadedVertexBuffer(ThreadedGraphicsContext device, IVertexBuffer<Vertex> vertexBuffer)
@@ -521,7 +522,8 @@ namespace OpenRA.Platforms.Default
this.device = device;
bind = vertexBuffer.Bind;
setData1 = tuple => { var t = (ValueTuple<Vertex[], int>)tuple; vertexBuffer.SetData(t.Item1, t.Item2); device.ReturnVertices(t.Item1); };
setData2 = tuple => { var t = (ValueTuple<IntPtr, int, int>)tuple; vertexBuffer.SetData(t.Item1, t.Item2, t.Item3); return null; };
setData2 = tuple => { var t = (ValueTuple<Vertex[], int, int, int>)tuple; vertexBuffer.SetData(t.Item1, t.Item2, t.Item3, t.Item4); device.ReturnVertices(t.Item1); };
setData3 = tuple => { setData2(tuple); return null; };
dispose = vertexBuffer.Dispose;
}
@@ -537,17 +539,20 @@ namespace OpenRA.Platforms.Default
device.Post(setData1, (buffer, length));
}
public void SetData(IntPtr data, int start, int length)
public void SetData(Vertex[] vertices, int offset, int start, int length)
{
// We can't return until we are finished with the data, so we must Send here.
device.Send(setData2, (data, start, length));
}
public void SetData(Vertex[] vertices, int start, int length)
{
var buffer = device.GetVertices(length);
Array.Copy(vertices, start, buffer, 0, length);
device.Post(setData1, (buffer, length));
if (length <= device.BatchSize)
{
// If we are able to use a buffer without allocation, post a message to avoid blocking.
var buffer = device.GetVertices(length);
Array.Copy(vertices, offset, buffer, 0, length);
device.Post(setData2, (buffer, 0, start, length));
}
else
{
// If the length is too large for a buffer, send a message and block to avoid allocations.
device.Send(setData3, (vertices, offset, start, length));
}
}
public void Dispose()
@@ -567,6 +572,7 @@ namespace OpenRA.Platforms.Default
readonly Func<byte[]> getData;
readonly Func<object, object> setData1;
readonly Action<object> setData2;
readonly Func<object, object> setData3;
readonly Action dispose;
public ThreadedTexture(ThreadedGraphicsContext device, ITextureInternal texture)
@@ -580,6 +586,7 @@ namespace OpenRA.Platforms.Default
getData = () => texture.GetData();
setData1 = colors => { texture.SetData((uint[,])colors); return null; };
setData2 = tuple => { var t = (ValueTuple<byte[], int, int>)tuple; texture.SetData(t.Item1, t.Item2, t.Item3); };
setData3 = tuple => { setData2(tuple); return null; };
dispose = texture.Dispose;
}
@@ -630,11 +637,21 @@ namespace OpenRA.Platforms.Default
public void SetData(byte[] colors, int width, int height)
{
// This creates some garbage for the GC to clean up,
// but allows us post a message instead of blocking the message queue by sending it.
var temp = new byte[colors.Length];
Array.Copy(colors, temp, temp.Length);
device.Post(setData2, (temp, width, height));
// Objects 85000 bytes or more will be directly allocated in the Large Object Heap (LOH).
// https://docs.microsoft.com/en-us/dotnet/standard/garbage-collection/large-object-heap
if (colors.Length < 85000)
{
// If we are able to create a small array the GC can collect easily, post a message to avoid blocking.
var temp = new byte[colors.Length];
Array.Copy(colors, temp, temp.Length);
device.Post(setData2, (temp, width, height));
}
else
{
// If the length is large and would result in an array on the Large Object Heap (LOH),
// send a message and block to avoid LOH allocation as this requires a Gen2 collection.
device.Send(setData3, (colors, width, height));
}
}
public void Dispose()

16
OpenRA.Platforms.Default/VertexBuffer.cs
View File

@@ -57,10 +57,10 @@ namespace OpenRA.Platforms.Default
public void SetData(T[] data, int length)
{
SetData(data, 0, length);
SetData(data, 0, 0, length);
}
public void SetData(T[] data, int start, int length)
public void SetData(T[] data, int offset, int start, int length)
{
Bind();
@@ -70,7 +70,7 @@ namespace OpenRA.Platforms.Default
OpenGL.glBufferSubData(OpenGL.GL_ARRAY_BUFFER,
new IntPtr(VertexSize * start),
new IntPtr(VertexSize * length),
ptr.AddrOfPinnedObject());
ptr.AddrOfPinnedObject() + VertexSize * offset);
}
finally
{
@@ -80,16 +80,6 @@ namespace OpenRA.Platforms.Default
OpenGL.CheckGLError();
}
public void SetData(IntPtr data, int start, int length)
{
Bind();
OpenGL.glBufferSubData(OpenGL.GL_ARRAY_BUFFER,
new IntPtr(VertexSize * start),
new IntPtr(VertexSize * length),
data);
OpenGL.CheckGLError();
}
public void Bind()
{
VerifyThreadAffinity();