UE4.25 Slate源码解读

概述

Slate系统是UE的一套UI解决方案，UMG系统也是依赖Slate系统实现的。

问题：

• Slate系统是如何组织的？
  • 控件树的父子关系是如何绑定的？
• Slate系统是如何渲染的？
  • slate渲染结构和流程是如何组织的？
  • 如何进行合批？

结构

SWidget控件类型

SWidget是Slate系统中所有控件的父类。

控件有三种类型。
叶控件 - 不带子槽的控件。如显示一块文本的 STextBlock。其原生便了解如何绘制文本。
面板 - 子槽数量为动态的控件。如垂直排列任意数量子项，形成一些布局规则的 SVerticalBox。
合成控件 - 子槽显式命名、数量固定的控件。如拥有一个名为 Content 的槽（包含按钮中所有控件）的 SButton。
– 官方文档

也有一些其他控件直接继承自SWidget，情况比较特殊，暂时忽略。

SWidget 控件树实现

上述控件三种类型中，其中SPanel、SCompoundWidget可以作为父节点，控件之间的父子关系是依赖Slot实现的。父控件引用Slot,Slot引用子控件并且保留子控件相对于父控件的布局信息。UMG的控件树的实现方式类似，以UCanvasPanel为例：

UCanvasPanel 控件树相关源码分析

相关类图

• UCanvasPanel有一个SConsntraintCanvas的引用，UCanvasPanel功能依赖SConsntraintCanvas实现。（组合关系）

Class UMG_API UCanvasPanel : public UPanelWidget
{
	// ...
protected:
	TSharedPtr<class SConstraintCanvas> MyCanvas;
	// ...
}

• UCanvasPanel有一个Slot容器，AddChild会生成Slot并与Child互相绑定引用，然后把Slot放入Slot容器。

UCanvasPanelSlot* UCanvasPanel::AddChildToCanvas(UWidget* Content)
{
	return Cast<UCanvasPanelSlot>( Super::AddChild(Content) );
}

class UMG_API UPanelWidget : public UWidget
{
	// ...
protected:
	TArray<UPanelSlot*> Slots;
	// ...
}

UPanelSlot* UPanelWidget::AddChild(UWidget* Content)
{
	// ...
	UPanelSlot* PanelSlot = NewObject<UPanelSlot>(this, GetSlotClass(), NAME_None, NewObjectFlags);
	PanelSlot->Content = Content;
	PanelSlot->Parent = this;

	Content->Slot = PanelSlot;

	Slots.Add(PanelSlot);

	OnSlotAdded(PanelSlot);

	InvalidateLayoutAndVolatility();

	return PanelSlot;
}

• 当UCanvasPanel增加一个UCanvasPanelSlot，其SConstraintCanvas引用也响应的添加一个FSlot(SConstraintCanvas::FSlot),且UCanvasPanelSlot保存FSlot的引用。

void UCanvasPanel::OnSlotAdded(UPanelSlot* InSlot)
{
	// Add the child to the live canvas if it already exists
	if ( MyCanvas.IsValid() )
	{
		CastChecked<UCanvasPanelSlot>(InSlot)->BuildSlot(MyCanvas.ToSharedRef());
	}
}

class UMG_API UCanvasPanelSlot : public UPanelSlot
{
// ...
private:
	SConstraintCanvas::FSlot* Slot;
// ...
}

void UCanvasPanelSlot::BuildSlot(TSharedRef<SConstraintCanvas> Canvas)
{
	Slot = &Canvas->AddSlot()
		[
			Content == nullptr ? SNullWidget::NullWidget : Content->TakeWidget()
		];

	SynchronizeProperties();
}

class SLATE_API SConstraintCanvas : public SPanel
{
public:
	class FSlot : public TSlotBase<FSlot> { /* Offset,Anchors,Alignment 等布局数据... */ }
	// ...
protected:
	TPanelChildren< FSlot > Children;
	// ...
public:
	FSlot& AddSlot()
	{
		Invalidate(EInvalidateWidget::Layout);

		SConstraintCanvas::FSlot& NewSlot = *(new FSlot());
		this->Children.Add( &NewSlot );
		return NewSlot;
	}
	// ...
}

• 当修改UCanvasPanelSlot的属性时，通用引用也修改了SConstraintCanvas::FSlot对应的属性。

void UCanvasPanelSlot::SetOffsets(FMargin InOffset)
{
	LayoutData.Offsets = InOffset;
	if ( Slot )
	{
		Slot->Offset(InOffset);
	}
}

渲染

Slate渲染由Game线程驱动，收集渲染单元并转换成渲染参数打包推送到渲染线程，渲染线程依据渲染参数分批生成RHICommand,RHIConmand调用图形库API设置渲染状态和绘制。

• RHICommand是多态的，提供了OpenGL,D3D,Vulkan等多个图像库对应的子类。

渲染流程图

渲染相关类图

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-RIZ0dKUr-1671417317007)(https://img2022.cnblogs.com/blog/2905902/202207/2905902-20220728154212154-773346408.jpg)]

FSlateApplication::PrivateDrawWindows

遍历所有Window，收集渲染图元信息。

FSlateApplication::DrawPrepass

对控件树进行中序遍历，缓存每个控件的DesiredSize，给后面DrawWindowAndChildren遍历时使用。ComputeDesiredSize行为是多态的，例如：

• SImage 依据ImageBrush->ImageSize计算。
• SConstraintCanvas 依据子控件布局计算。

FSlateApplication::DrawWindowAndChildren

从树根开始，依据每个节点的遍历策略遍历，调用Paint函数收集图元信息保存在上下文中。OnPaint行为是多态的，例如：

• SConstraintCanvas 先遍历计算孩子的布局信息，再遍历孩子的Paint方法。
• SImage 会调用FSlateDrawElement::MakeBox等方法计算计算自身的图元信息保存在上下文中。

FDrawWindowArgs

• FSlateDrawBuffer 负载所有Window的图元信息。
• FSlateWindowElementList 负载Window内所有图元信息。
• FSlateDrawElement 负载一个元素的图元信息

以SImage的OnPaint为例：

void FSlateApplication::DrawWindowAndChildren( const TSharedRef<SWindow>& WindowToDraw, FDrawWindowArgs& DrawWindowArgs )
{
	// ...
	FSlateWindowElementList& WindowElementList = DrawWindowArgs.OutDrawBuffer.AddWindowElementList(WindowToDraw);
	// ...
	MaxLayerId = WindowToDraw->PaintWindow(
					GetCurrentTime(),
					GetDeltaTime(),
					WindowElementList,
					FWidgetStyle(),
					WindowToDraw->IsEnabled());
	// ...
}

int32 SImage::OnPaint( const FPaintArgs& Args, const FGeometry& AllottedGeometry, const FSlateRect& MyCullingRect, FSlateWindowElementList& OutDrawElements, int32 LayerId, const FWidgetStyle& InWidgetStyle, bool bParentEnabled ) const
{
	// ...
	FSlateDrawElement::MakeBox(OutDrawElements, LayerId, AllottedGeometry.ToPaintGeometry(), ImageBrush, DrawEffects, FinalColorAndOpacity);
	// ...
	return LayerId;
}

FSlateDrawElement& FSlateDrawElement::MakeBoxInternal(
	FSlateWindowElementList& ElementList,
	uint32 InLayer,
	const FPaintGeometry& PaintGeometry,
	const FSlateBrush* InBrush,
	ESlateDrawEffect InDrawEffects,
	const FLinearColor& InTint
)
{
	EElementType ElementType = (InBrush->DrawAs == ESlateBrushDrawType::Border) ? EElementType::ET_Border : EElementType::ET_Box;

	FSlateDrawElement& Element = ElementList.AddUninitialized();

	const FMargin& Margin = InBrush->GetMargin();
	FSlateBoxPayload& BoxPayload = ElementList.CreatePayload<FSlateBoxPayload>(Element);

	Element.Init(ElementList, ElementType, InLayer, PaintGeometry, InDrawEffects);

	BoxPayload.SetTint(InTint);
	BoxPayload.SetBrush(InBrush);

	return Element;
}

SImage调用了FSlateDrawElement::MakeBox令FSlateWindowElementList增加一个FSlateDrawElement并将自身的图元信息保存其中。

FSlateRHIRenderer::DrawWindows_Private

• 调用FSlateElementBatcher::AddElements生成渲染参数（顶点数组，索引数组，shader相关参数…）
• 生成渲染命令闭包放到RHI渲染命令队列中，供渲染线程取出调用。

void FSlateRHIRenderer::DrawWindows_Private(FSlateDrawBuffer& WindowDrawBuffer)
{
	// ...
	for (int32 ListIndex = 0; ListIndex < WindowElementLists.Num(); ++ListIndex)
	{
		// ...
		ElementBatcher->AddElements(ElementList);
		// ...

		// ...
		if (GIsClient && !IsRunningCommandlet() && !GUsingNullRHI)
		{
			ENQUEUE_RENDER_COMMAND(SlateDrawWindowsCommand)(
				[Params, ViewInfo](FRHICommandListImmediate& RHICmdList)
				{
					Params.Renderer->DrawWindow_RenderThread(RHICmdList, *ViewInfo, *Params.WindowElementList, Params);
				}
			);
		}
	// ...
}

FSlateElementBatcher::AddElements

将 FSlateApplication::PrivateDrawWindows 阶段生成的 FSlateDrawElement 所负载的图元信息，转换成渲染所需的参数封装到FSlateRenderBatch中，放入FSlateWindowElementList的FSlateBatchData成员中，对于缓存/未缓存的数据有不同的处理策略：

void FSlateElementBatcher::AddElements(FSlateWindowElementList& WindowElementList)
{
	// ...
	AddElementsInternal(WindowElementList.GetUncachedDrawElements(), ViewportSize);

	// ...
	const TArrayView<FSlateCachedElementData* const> CachedElementDataList = WindowElementList.GetCachedElementDataList();

	if(CachedElementDataList.Num())
	{
		for (FSlateCachedElementData* CachedElementData : CachedElementDataList)
		{
			AddCachedElements(*CachedElementData, ViewportSize);
		}
	}
	// ...
}

• 未缓存的调用AddElements，AddElements调用AddElementsInternal生成和封装渲染参数，放入FSlateWindowElementList的FSlateBatchData成员中。

void FSlateElementBatcher::AddElementsInternal(const FSlateDrawElementArray& DrawElements, const FVector2D& ViewportSize)
{
	for (const FSlateDrawElement& DrawElement : DrawElements)
	{
		switch ( DrawElement.GetElementType() )
		{
		case EElementType::ET_Box:
		{
			SCOPED_NAMED_EVENT_TEXT("Slate::AddBoxElement", FColor::Magenta);
			STAT(ElementStat_Boxes++);
			DrawElement.IsPixelSnapped() ? AddBoxElement<ESlateVertexRounding::Enabled>(DrawElement) : AddBoxElement<ESlateVertexRounding::Disabled>(DrawElement);
		}
		// ...
	}
}

template<ESlateVertexRounding Rounding>
void FSlateElementBatcher::AddBoxElement(const FSlateDrawElement& DrawElement)
{
	const FSlateBoxPayload& DrawElementPayload = DrawElement.GetDataPayload<FSlateBoxPayload>();
	const FColor Tint = PackVertexColor(DrawElementPayload.GetTint());
	const FSlateRenderTransform& ElementRenderTransform = DrawElement.GetRenderTransform();
	// ...

	RenderBatch.AddVertex( FSlateVertex::Make<Rounding>( RenderTransform, FVector2D( Position.X, Position.Y ),		LocalSize, DrawScale, FVector4(StartUV,										Tiling),	Tint ) ); //0
	RenderBatch.AddVertex( FSlateVertex::Make<Rounding>( RenderTransform, FVector2D( Position.X, TopMarginY ),		LocalSize, DrawScale, FVector4(FVector2D( StartUV.X, TopMarginV ),			Tiling),	Tint ) ); //1
	// ...

	RenderBatch.AddIndex( IndexStart + 0 );
	RenderBatch.AddIndex( IndexStart + 1 );
	// ...
}

• 已缓存的调用AddCachedElements：
  • 遍历 ListsWithNewData 中的FSlateDrawElement，调用AddElementsInternal生成和封装渲染参数，放入FSlateWindowElementList的FSlateBatchData成员中。
  • 直接将 CachedElementData 中所有FSlateRenderBatch放入FSlateWindowElementList的FSlateBatchData成员中。

void FSlateElementBatcher::AddCachedElements(FSlateCachedElementData& CachedElementData, const FVector2D& ViewportSize)
{
	// ...
	for (FSlateCachedElementList* List : CachedElementData.ListsWithNewData)
	{
		// ...
		AddElementsInternal(List->DrawElements, ViewportSize);
		// ...
	}
	// ...
	BatchData->AddCachedBatches(CachedElementData.GetCachedBatches());
	// ...
}

DrawWindow_RenderThread

合并和处理批次，提交渲染参数，调用渲染相关API进行绘制。

void FSlateRHIRenderer::DrawWindow_RenderThread(FRHICommandListImmediate& RHICmdList, FViewportInfo& ViewportInfo, FSlateWindowElementList& WindowElementList, const struct FSlateDrawWindowCommandParams& DrawCommandParams)
{
	// ...
	RenderingPolicy->BuildRenderingBuffers(RHICmdList, BatchData);

	// ...
	RenderingPolicy->DrawElements
			(
				RHICmdList,
				BackBufferTarget,
				BackBuffer,
				PostProcessBuffer,
				ViewportInfo.bRequiresStencilTest ? ViewportInfo.DepthStencil : EmptyTarget,
				BatchData.GetFirstRenderBatchIndex(),
				BatchData.GetRenderBatches(),
				RenderParams
			);

	// ...
	RHICmdList.EndDrawingViewport(ViewportInfo.ViewportRHI, true, DrawCommandParams.bLockToVsync);
	// ...
}

FSlateRHIRenderingPolicy::BuildRenderingBuffers

合并批次并收集所有batch的顶点/索引数据分别填充到数组中（方便后面一次性提交给GPU）。

void FSlateRHIRenderingPolicy::BuildRenderingBuffers(FRHICommandListImmediate& RHICmdList, FSlateBatchData& InBatchData)
{
	// ...
	InBatchData.MergeRenderBatches();

	// ...
	uint32 RequiredVertexBufferSize = NumBatchedVertices * sizeof(FSlateVertex);
	uint8* VertexBufferData = (uint8*)InRHICmdList.LockVertexBuffer(VertexBuffer, 0, RequiredVertexBufferSize, RLM_WriteOnly);

	uint32 RequiredIndexBufferSize = NumBatchedIndices * sizeof(SlateIndex);
	uint8* IndexBufferData = (uint8*)InRHICmdList.LockIndexBuffer(IndexBuffer, 0, RequiredIndexBufferSize, RLM_WriteOnly);

	FMemory::Memcpy(VertexBufferData, LambdaFinalVertexData.GetData(), RequiredVertexBufferSize);
	FMemory::Memcpy(IndexBufferData, LambdaFinalIndexData.GetData(), RequiredIndexBufferSize);
	// ...
}

• 调用FSlateBatchData::MergeRenderBatches设置批次顶点/索引偏移（每次绘制时按照偏移读取一段数据进行绘制）并进行合批，注意合批条件：
  • TestBatch.GetLayer() == CurBatch.GetLayer()
  • CurBatch.IsBatchableWith(TestBatch)

void FSlateBatchData::MergeRenderBatches()
{
	// ...
	FillBuffersFromNewBatch(CurBatch, FinalVertexData, FinalIndexData);
	// ...
	if (CurBatch.bIsMergable)
	{
		for (int32 TestIndex = BatchIndex + 1; TestIndex < BatchIndices.Num(); ++TestIndex)
		{
			const TPair<int32, int32>& NextBatchIndexPair = BatchIndices[TestIndex];
			FSlateRenderBatch& TestBatch = RenderBatches[NextBatchIndexPair.Key];
			if (TestBatch.GetLayer() != CurBatch.GetLayer())
			{
				// none of the batches will be compatible since we encountered an incompatible layer
				break;
			}
			else if (!TestBatch.bIsMerged && CurBatch.IsBatchableWith(TestBatch))
			{
				CombineBatches(CurBatch, TestBatch, FinalVertexData, FinalIndexData);

				check(TestBatch.NextBatchIndex == INDEX_NONE);

			}
		}
	}
	// ...
}

void FSlateBatchData::FillBuffersFromNewBatch(FSlateRenderBatch& Batch, FSlateVertexArray& FinalVertices, FSlateIndexArray& FinalIndices)
{
	if(Batch.HasVertexData())
	{
		const int32 SourceVertexOffset = Batch.VertexOffset;
		const int32 SourceIndexOffset = Batch.IndexOffset;

		// At the start of a new batch, just direct copy the verts
		// todo: May need to change this to use absolute indices
		Batch.VertexOffset = FinalVertices.Num();
		Batch.IndexOffset = FinalIndices.Num();
		
		FinalVertices.Append(&(*Batch.SourceVertices)[SourceVertexOffset], Batch.NumVertices);
		FinalIndices.Append(&(*Batch.SourceIndices)[SourceIndexOffset], Batch.NumIndices);
	}
}

bool IsBatchableWith(const FSlateRenderBatch& Other) const
{
	return
		ShaderResource == Other.ShaderResource
		&& DrawFlags == Other.DrawFlags
		&& ShaderType == Other.ShaderType
		&& DrawPrimitiveType == Other.DrawPrimitiveType
		&& DrawEffects == Other.DrawEffects
		&& ShaderParams == Other.ShaderParams
		&& InstanceData == Other.InstanceData
		&& InstanceCount == Other.InstanceCount
		&& InstanceOffset == Other.InstanceOffset
		&& DynamicOffset == Other.DynamicOffset
		&& CustomDrawer == Other.CustomDrawer
		&& SceneIndex == Other.SceneIndex
		&& ClippingState == Other.ClippingState;
}

FRHICommandList::BeginDrawingViewport

调用FRHICommandListImmediate::ImmediateFlush提交上文提到的所有顶点/索引数组等渲染状态信息。

void FRHICommandList::BeginDrawingViewport(FRHIViewport* Viewport, FRHITexture* RenderTargetRHI)
{
	// ...
	FRHICommandListExecutor::GetImmediateCommandList().ImmediateFlush(EImmediateFlushType::FlushRHIThread);
	// ...
}

FORCEINLINE_DEBUGGABLE void FRHICommandListImmediate::ImmediateFlush(EImmediateFlushType::Type FlushType)
{
	// ...
	GRHICommandList.ExecuteList(*this); // 执行并销毁所有命令
	// ...
}

FSlateRHIRenderingPolicy::DrawElements

为每一个批次生成渲染状态信息和绘制相关RHI命令。

void FSlateRHIRenderingPolicy::DrawElements(
	FRHICommandListImmediate& RHICmdList,
	FSlateBackBuffer& BackBuffer,
	FTexture2DRHIRef& ColorTarget,
	FTexture2DRHIRef& PostProcessTexture,
	FTexture2DRHIRef& DepthStencilTarget,
	int32 FirstBatchIndex,
	const TArray<FSlateRenderBatch>& RenderBatches,
	const FSlateRenderingParams& Params)
{
	// ...
	while (NextRenderBatchIndex != INDEX_NONE)
	{
		// ...
		RHICmdList.SetStreamSource(0, VertexBufferPtr->VertexBufferRHI, RenderBatch.VertexOffset * sizeof(FSlateVertex));
		RHICmdList.DrawIndexedPrimitive(IndexBufferPtr->IndexBufferRHI, 0, 0, RenderBatch.NumVertices, RenderBatch.IndexOffset, PrimitiveCount, RenderBatch.InstanceCount);
		// ...
	}
	// ...
}

FRHICommandList::EndDrawingViewport

再次调用FRHICommandListImmediate::ImmediateFlush执行并销毁所有命令，调用图形库API提交所有渲染状态和绘制命令。

FD3D11DynamicRHI::RHIDrawIndexedPrimitive

绘制命令调用FD3D11DynamicRHI::RHIDrawIndexedPrimitive最终调到ID3D11DeviceContext::DrawIndexed调用图形库API进行绘制。

拓展阅读

• 跳转链接：D3D11 Hello Triangle
• 跳转链接：知乎-南山搬砖道人-UE4 Slate合批流程流程
• 跳转链接：博客园-0向往0-剖析虚幻渲染体系