Unity SRP渲染流程源码分析

准备一般cull参数cullResults.sceneCullParameters

ExecuteCommandBuffer

是

从Umbra获取一些数据用于阴影裁减

          

Cull？

instancing?

1

从源码中发现，Hybrid V1的扩展必须要在C++层去做，而且会非常麻烦，比如我想尝试支持rendering layer，就必须做大量改动参见（GetHybridV1InstancingVarMask、AllocAndFillInstancedData、FillInstancedDataT）。而且Unity也不打算继续支持V1了，所以我们的目标是支持V2。

ScriptableRenderLoopDrawDispatch

准备cull参数（light for shadow）SetupShadowCullData

执行CullingGroupCullingGroupManager::Get().CullAndSendEvents

ScriptableRenderLoopJob

HybridV2?

Cull结束

准备要渲染的物体（一个node就是一个renderer）PrepareSceneNodes

这俩压根不走Culling，直接DrawBuffers

WinEditorMain:::WinMainwinutils::UninstallInternalCrashHandler();PlayerLoopController::UpdateSceneIfNeeded()PlayerLoopController::UpdateScene

Umbra OCC?

C#RenderPipeline::InternalRender

CopyActiveReflectionProbes

kRenderCommand_DrawRenderer

主动暴露一些cull结果参数给C#端的SRP使用CullResultsToScriptingData

Editor渲染

填充deviceBatchParams参数    deviceBatchParams.items = batch;    deviceBatchParams.itemCount = batchWrite - batch;    deviceBatchParams.drawBufferRangesStream = bufferRangesBuffer;    deviceBatchParams.builtInSysmemCBuffer = builtInSysmemCBuffer;    deviceBatchParams.builtInCBSize = CBSize;    deviceBatchParams.realCBSize = m_CurrentBuiltInInfo->constantBufferSize;

Draw Renderers

否

提取裁减结果render，填充到sharedRendererScene中的RenderNodeQueue中ExtractSceneRenderNodeQueue

0

ScriptableRenderContext::Cull(根据需要可以多次调用)

ScriptableBatchRenderer::Flush ->ScriptableBatchRenderer::RenderMultipleMeshes

CullShadowCastersWithoutUmbraThreaded

前后排序在哪？

每一级shadowmap 的cullingBatchRendererGroup::OnBeforeCulling

？？CullAllPerObjectLights

一些打断Batch的判断

CullLocalLightsLocalLightShadowCasterCullingCullLocalLightShadowCasters

RenderManager::RenderOffscreenCameras

itemCount>0?

TerrainManager::CullAllTerrains

内置渲染管线

一些初始化、内存分配等deviceBatchParams.instanceItemsPStream = builtinInstanceDatasPtrs;(BRG instance数据)

foreach(BatchInstanceData)-填充batchWrite

DrawShadows

1.Editor下离屏渲染（Camera设置了RT）

使用SRP管线？RenderCamerasWithScriptableRenderLoop

GameView.OnGUI

foreach(renderer in rendererArrays){//将场景中的物体数组分组，每组交由一个job处理，job中处理每一个物体的包围盒和相机frustum的culling。             CullDynamicObjectsJob()        {               1，通过一些mask参数快速剔除一些               2，包围盒测试裁减               3，一些比较慢的culling操作放到最后        }}

kRenderCommand_DrawRendererList（对应C#的cmd.DrawRendererList）

RenderManager::RenderCameras

执行SRP渲染管线ScriptableRenderContext::ExtractAndExecuteRenderPipeline

PlayModeView.OnGUI

     

执行Renders BRG CullBatchRendererGroup::OnBeforeCulling结果存储在results.sceneCullingOutput.batchRendererCullingOutputs

BatchInstanceData

要支持多线程渲染，得保证线程安全，SharedXXX是线程安全的这里就是要把数据填充到线程安全的数据结构中（sharedRendererScene）去，这个数据还是在cull results之中CullResults::GetOrCreateSharedRendererScene()

计算drawBufferRangeCountdrawBufferRangeCount = PrepareDrawRangesbatchWrite->drawBufferRangesToConsume = drawBufferRangeCount;

设置RTkRenderCommand_SetRT

存在terrain？

V2设置HybridV2AllocAndFillInstancedData拷贝hybridMetadataBuffer

foreach(command) =>

ScheduleRenderJobs

ExecuteDrawRenderersCommand

CullDirectionalShadowCastersforeach(dirShadowLightCount)

非方向光BatchRendererGroup::OnBeforeCulling

SrpBatch

    

设置Textures

RepaintController::RenderPlayModeViewCameras

C#调用C++ Cull接口::CullScriptable（ScriptableCulling.cpp）

如果有必要cull lights && shadow castersCullLights && CullDirectionalShadowCasters

Commands

Container

C#实现具体的Render例如HDRenderPipeline::Render

SRP Batch渲染流程：ScriptableRenderLoopDrawSRPBatcher（ScriptableDrawRenderers.cpp）foreach(ScriptableLoopObjectData)

真正的cull开始了void CullScene(CullResults& results)

EditorGUIUtility.RenderPlayModeViewCamerasInternal

回到C++进行渲染ScriptableRenderContext::ExecuteScriptableRenderLoop()

     RenderQueue在哪排序的？

BatchRendererGroup::OnBeforeCulling 是相机culling还是shadow culling取决于被调用的位置，相应的culling matrix和culling planes会被传给用户。

2

ProcessShadowCasterNodeVisibilityAndCullWithoutUmbra

2.Editor下Game窗口刷新

通知如ParticleSystem/LineRenderer/TrailRenderer之类的执行一些回调函数DispatchGeometryJobs

广播裁减完毕通知CullSendEvents-》InvokeOnWillRenderObject

CullStaticSceneWithUmbra

ScriptableLoopObjectData

各种渲染逻辑cmd.DrawXXXcmd.RenderXXX

......

CullDynamicScene

如果有必要，CullReflectionProbesCullReflectionProbes

V1设置AllocAndFillInstancedData

CommandBuffers

目前的instancing基本上就是BRG。

for m_DrawShadowCommands

SetupBatchRendererGroupContext

打包到各平台下的渲染

ScriptableLoopObjectData打包到BatchInstanceData

调用C#端的SRP管线invocation(GetCoreScriptingClasses().doRenderLoop_Internal)

Shader兼容SRP Batch?

填充active lightsCopyActiveLights

同级别排列

？？CullPerObjectReflectionProbes

先OffScreen再OnScreenPlayerRender（Player.cpp）

普通渲染流程：ScriptableRenderLoopDraw

阴影投射体裁减foreach(sceneCullParameters.totalRendererListsCount)

Culling

只为为光源执行shadow castersBatchRendererGroup::OnBeforeCulling

PrepareDrawShadowsCommandStep1

Draw Shadows

CullDirectionalShadowCastersJob

for drawShadowData.size()PrepareDrawShadowsCommandStep2

BatchRenderGroup依然是DrawRenderList来驱动。

device.DrawBuffersBatchMode(deviceBatchParams);

C#端Submit例如RenderPipeline::Submit

阴影：Runtime\\Camera\\Shadows.cpp 这里大都是内置管线的代码Runtime\\Graphics\\ScriptableRenderLoop\\ScriptableDrawShadows.cpp 这是SRP的代码

ScriptableBatchRenderer::RenderMultipleMeshes（填充GfxBatchItem，并提交DrawBuffers）

计算staticBatchCountstaticBatchCount = GetNumberOfConsecutiveStaticBatches